Seabed Biodiversity
on the
Continental Shelf of the
Great Barrier Reef World Heritage Area
•RolandPitcher
•PeterDoherty•PeterArnold
•JohnHooper•NeilGribble
FINALREPORTtothe
CooperativeResearchCentre
fortheGreatBarrierReefWorldHeritageArea
JULY 2007
National Library of Australia Cataloguing-in-Publication entry:
Pitcher, C. R. (Clifford Roland).
Seabed biodiversity on the continental shelf of the Great
Barrier Reef World Heritage Area.
Bibliography.
Includes index.
ISBN 978-1-921232-87-9 (pbk.).
ISBN 978-1-921232-88-6 (web).
1. Marine biodiversity - Queensland - Great Barrier Reef.
2. Marine biodiversity - Research - Queensland - Great
Barrier Reef. 3. Great Barrier Reef (Qld.) - Environmental
aspects. I. CRC Reef Research Centre. II. Title.
333.95616
Citation:
Pitcher, C.R., Doherty, P., Arnold, P., Hooper, J., Gribble, N., Bartlett, C., Browne, M., Campbell, N.,
Cannard, T., Cappo, M., Carini, G., Chalmers, S., Cheers, S., Chetwynd, D., Colefax, A., Coles, R.,
Cook, S., Davie, P., De'ath, G., Devereux, D., Done, B., Donovan, T., Ehrke, B., Ellis, N., Ericson, G.,
Fellegara, I., Forcey, K., Furey, M., Gledhill, D., Good, N., Gordon, S., Haywood, M., Hendriks, P.,
Jacobsen, I., Johnson, J., Jones, M., Kinninmoth, S., Kistle, S., Last, P., Leite, A., Marks, S., McLeod,
I., Oczkowicz, S., Robinson, M., Rose, C., Seabright, D., Sheils, J., Sherlock, M., Skelton, P., Smith,
D., Smith, G., Speare, P., Stowar, M., Strickland, C., Van der Geest, C., Venables, W., Walsh, C.,
Wassenberg, T., Welna, A., Yearsley, G. (2007). Seabed Biodiversity on the Continental Shelf of the
Great Barrier Reef World Heritage Area. AIMS/CSIRO/QM/QDPI CRC Reef Research Task Final
Report. 320 pp.
Published: July 2007 by CSIRO Marine and Atmospheric Research
© Australian Institute of Marine Science, CSIRO Marine and Atmospheric Research, Queensland
Museum, Queensland Department of Primary Industries, CRC Reef Research Centre, Fisheries
Research and Development Corporation, and the National Oceans Office, 2007.
This work is copyright. Except as permitted under the Copyright Act 1968 (Cth), no part of this
publication may be reproduced by any process, electronic or otherwise, without the specific written
permission of the copyright owners. Neither may information be stored electronically in any form
whatsoever without such permission.
DISCLAIMER
The authors have taken all reasonable steps to ensure that the information contents in this publication
are accurate at the time of publication. Readers should ensure that they make appropriate inquiries to
determine whether new information is available on the particular subject matter.
GBR Seabed Biodiversity
i
July 2007
Seabed Biodiversity on the Continental Shelf of
the Great Barrier Reef World Heritage Area
CRC-REEF Task Number: C1.1.2
FRDC Project Number: 2003/021
NOO Contract Number: 2004/015
Roland Pitcher², Peter Doherty¹, Peter Arnold³, John Hooper³, Neil Gribble4,
Chris Bartlett³, Matthew Browne², Norm Campbell², Toni Cannard², Mike Cappo¹,
Giovannella Carini³, Susan Chalmers4, Sue Cheers², Doug Chetwynd², Andrew
Colefax³, Rob Coles4, Stephen Cook³, Peter Davie³, Glenn De'ath¹, Drew Devereux²,
Barbara Done³, Tim Donovan¹, Barry Ehrke4, Nick Ellis², Gavin Ericson¹, Ida
Fellegara³, Karl Forcey², Melodyrose Furey², Dan Gledhill², Norm Good4, Scott
Gordon², Mick Haywood², Patricia Hendriks³, Ian Jacobsen, Jeff Johnson³, Michelle
Jones³, Stuart Kinninmoth¹, Sarah Kistle4, Peter Last², Anita Leite³, Shona Marks²,
Ian McLeod², Sybilla Oczkowicz4, Melissa Robinson³, Cassanda Rose4, Denise
Seabright³, Jacquie Sheils², Matt Sherlock², Posa Skelton4, David H Smith², Greg
Smith², Peter Speare¹, Marcus Stowar¹, Colleen Strickland³, Claire Van der Geest4,
Bill Venables², Cath Walsh4, Ted Wassenberg², Andrzej Welna², Gus Yearsley²
¹Australian Institute of Marine Science
Cape Ferguson, TOWNSVILLE, Qld. 4810, Australia
²Commonwealth Scientific & Industrial Research Organisation
Marine & Atmospheric Research
Mathematics & Information Sciences
233 Middle Street, CLEVELAND, Qld. 4163 Australia
³Queensland Museum
MTQ, TOWNSVILLE, Qld. 4810, Australia
South Bank, SOUTH BRISBANE, Qld. 4101, Australia
Queensland Department of Primary Industries
Northern Fisheries Centre, Tingara Street, CAIRNS, Qld. 4870, Australia
4
AUSTRALIAN INSTITUTE
OF MARINE SCIENCE
ISBN 978-1-921232-87-9.
CRC Reef Research Task Final Report
GBR Seabed Biodiversity
ii
ACKNOWLEDGEMENTS
This document is the final report of the Great Barrier Reef Seabed Biodiversity Project, a
collaboration between the Australian Institute of Marine Science (AIMS), the Commonwealth
Scientific and Industrial Research Organisation (CSIRO), Queensland Department of Primary
Industries & Fisheries (QDPI&F), and the Queensland Museum (QM); funded by the CRC Reef
Research Centre (CRC-Reef), the Fisheries Research and Development Corporation (FRDC), and the
National Oceans Office (NOO) of the Department of Environment and Water Resources. We
gratefully acknowledge the support of end-user agencies Great Barrier Reef Marine Park Authority
(GBRMPA), QDPI & Fisheries, Queensland Seafood Industry Association (QSIA) and the NOO, and
the contributions of the project's Steering Committee members: David Williams (CRC-Reef),
Dorothea Huber & Phil Cadwallader (GBRMPA), Brigid Kerrigan & Malcolm Dunning (QDPI&F),
Duncan Souter, Barry Ehrke & Martin Hicks (QSIA), Vicki Nelson (NOO) and Vern Veitch (Sunfish).
For the provision of physical environmental data, we thank Chris Jenkins (Ocean Sciences Institute –
OSI), Lance Bode (James Cook University – JCU), Scott Condie (CSIRO), the GBRMPA, the RAN
Hydrographers Office (RAN HO) — and Peter Harris, Andrew Heap, Emma Mathews, Alison
Hancock (Geoscience Australia – GA) for processing the sediment samples. We also wish to thank the
multi-agency teams and the crews of the RV Lady Basten (AIMS) and FRV Gwendoline May
(QDPI&F) that contributed to the success of the fieldwork; the research agencies AIMS, CSIRO, QM,
QDPI&F for providing support to the project; and all the project's team members without whose
valuable efforts this project would not have been possible. A large number of other people also helped
with aspects of the project and we are grateful for their assistance, including: Tony Reese, Steve
Edgar, Mark Tonks, Quinton Dell, Gary Fry, William White, Al Graham, Louise Conboy, Spikey
Riddoch, Bob Ward, Bronwyn Holmes, Tom Munro, Hiro Motomura, Mike Sugden, Henok Goitom,
Barry Russell, Barry Hutchins, Martin Gomon and Di Bray. Shane Griffiths, John Kirkwood and Piers
Dunstan provided valuable comments that improved the document. Louise Bell designed the cover.
This Report is dedicated to the memory of our friend, colleague and expert extraordinaire on “all
matters seabed”, Dr Peter William Arnold, Senior Curator Biodiversity, Museum of Tropical
Queensland, Townsville (1949-2006)
GBR Seabed Biodiversity
iii
TABLE OF CONTENTS
ACKNOWLEDGEMENTS .................................................................................................................................... ii
TABLE OF CONTENTS....................................................................................................................................... iii
FIGURES............................................................................................................................................................ v
TABLES ............................................................................................................................................................ xi
NON-TECHNICAL SUMMARY ........................................................................................................................ xv
PROJECT:......................................................................................................................................................... xv
PRINCIPAL INVESTIGATORS: .................................................................................................................... xv
ADDRESS: ....................................................................................................................................................... xv
OBJECTIVES: .................................................................................................................................................. xv
NON-TECHNICAL SUMMARY: .................................................................................................................. xvi
Outcomes Achieved ................................................................................................................................... xviii
1. INTRODUCTION ....................................................................................................................................... 1-1
1.1. BACKGROUND ................................................................................................................................. 1-1
1.2. NEED................................................................................................................................................... 1-2
1.3. OBJECTIVES ...................................................................................................................................... 1-3
2. METHODS .................................................................................................................................................. 2-5
2.1. SAMPLING DESIGN.......................................................................................................................... 2-5
2.1.1.
Physical environmental data (I McLeod & R Pitcher) ................................................................. 2-5
2.1.2.
Study Area Stratification (N Ellis) ............................................................................................. 2-15
2.1.3.
Site Selection.............................................................................................................................. 2-29
2.2. FIELD SAMPLING ........................................................................................................................... 2-31
2.2.1.
Research Vessels (T Wassenberg & N Gribble) ........................................................................ 2-31
2.2.2.
Towed Video Camera (G Smith, K Forcey, M Haywood)......................................................... 2-32
2.2.3.
Baited Remote Underwater Video Stations (M Cappo) ............................................................. 2-34
2.2.4.
Single-beam Acoustics............................................................................................................... 2-35
2.2.5.
Epibenthic Sled (T Wassenberg & M Stowar) ........................................................................... 2-36
2.2.6.
Scientific Trawl (T Wassenberg, D Gledhill & N Gribble)........................................................ 2-37
2.2.7.
Sample Processing at Sea (T Wassenberg, M Stowar, C Bartlett) ............................................. 2-38
2.3. LABORATORY PROCESSING AND IDENTIFICATION............................................................. 2-43
2.3.1.
Towed Video (T Wassenberg, J Sheils) ..................................................................................... 2-43
2.3.2.
BRUVS Video (M Cappo) ......................................................................................................... 2-46
2.3.3.
Sample Processing (T Hendriks, M Stowar, C Bartlett, T Wassenberg, D Gledhill) ................. 2-49
2.4. DATA ANALYSES........................................................................................................................... 2-53
2.4.1.
BRUVS Species Models, Characterization & Prediction (M Cappo, G De’Ath) ...................... 2-53
2.4.2.
Single Species Biophysical Models and Prediction (M Browne, W Venables) ......................... 2-54
2.4.3.
Species Groups Characterization and Prediction (M Browne)................................................... 2-58
2.4.4.
Site Groups Characterization and Prediction (W Venables) ...................................................... 2-59
2.4.5.
Video Habitat Characterization and Prediction (W Venables)................................................... 2-64
2.4.6.
Acoustics Discrimination and Classification.............................................................................. 2-68
2.4.7.
Ecological Risk Indicators ......................................................................................................... 2-75
2.4.8.
Trawl Management Scenario Model (N Ellis, A Welna, R Pitcher) .......................................... 2-77
3. RESULTS .................................................................................................................................................. 3-86
3.1. BRUVS SPECIES MODELS, CHARACTERIZATION & PREDICTION (M Cappo, G De’Ath) . 3-86
3.1.1.
BRUVS Species richness ........................................................................................................... 3-86
3.1.2.
BRUVS Species presence/absence Biophysical Models and Prediction.................................... 3-87
3.1.3.
BRUVS Site-groups Characterization and Prediction.............................................................. 3-105
GBR Seabed Biodiversity
iv
3.2. SINGLE SPECIES, BIOPHYSICAL MODELS AND PREDICTION ........................................... 3-112
3.2.1.
Sled and Trawl samples species richness ................................................................................. 3-113
3.2.2.
Single species models (M Browne & R Pitcher)...................................................................... 3-116
3.2.3.
Selected single species distribution maps................................................................................. 3-125
3.3. SPECIES GROUPS CHARACTERIZATION AND PREDICTION (M Browne & R Pitcher) ..... 3-133
3.3.1.
Characterization and Prediction Model performance ............................................................... 3-133
3.3.2.
Selected species group distribution maps................................................................................. 3-135
3.4. SITE GROUPS CHARACTERIZATION AND PREDICTIONS (W Venables & R Pitcher)........ 3-138
3.4.1.
Decision tree results ................................................................................................................. 3-138
3.4.2.
Species affinity groups ............................................................................................................. 3-141
3.4.3.
Description of site-group assemblages..................................................................................... 3-142
3.5. VIDEO HABITAT CHARACTERIZATION AND PREDICTION ............................................... 3-145
3.5.1.
Seabed substratum.................................................................................................................... 3-145
3.5.2.
Seabed biological habitat ......................................................................................................... 3-145
3.5.3.
Statistical characterization and prediction (W Venables & R Pitcher)..................................... 3-151
3.6. ACOUSTICS DISCRIMINATION AND CLASSIFICATION....................................................... 3-155
3.6.1.
Wavelet Packet-Based Techniques Applied to Data in the Angular Domain (D H Smith)...... 3-155
3.6.2.
Canonical Variate Analysis of Acoustic Data (N Campbell & D Devereux)........................... 3-166
3.6.3.
Linear Discriminant Analyses of QTC View data (I McLeod) ................................................ 3-186
3.7. ECOLOGICAL RISK INDICATORS ............................................................................................. 3-197
3.7.1.
Indicators for species-groups biomass...................................................................................... 3-198
3.7.2.
Indicators for individual species biomass................................................................................. 3-202
3.7.3.
Assemblage indicators.............................................................................................................. 3-238
3.7.4.
Habitat indicators ..................................................................................................................... 3-239
3.8. TRAWL MANAGEMENT SCENARIO MODEL (N Ellis, R Pitcher) .......................................... 3-243
4. DISCUSSION .......................................................................................................................................... 4-250
4.1. BRUVS SPECIES MODELS, CHARACTERIZATION & PREDICTION (M Cappo, G De’Ath) 4-251
4.1.1.
BRUVS Fish species ................................................................................................................ 4-251
4.1.2.
BRUVS Fish Assemblages....................................................................................................... 4-251
4.2. SINGLE SPECIES, BIOPHYSICAL MODELS AND PREDICTION ........................................... 4-253
4.3. SPECIES GROUPS CHARACTERIZATION AND PREDICTION .............................................. 4-255
4.4. SITE GROUPS CHARACTERIZATION AND PREDICTION ..................................................... 4-255
4.5. VIDEO HABITAT CHARACTERIZATION AND PREDICTION ............................................... 4-256
4.6. ACOUSTICS DISCRIMINATION AND CLASSIFICATION....................................................... 4-257
4.6.1.
Wavelet Packet-Based Techniques Applied to Data in the Angular Domain (D H Smith)...... 4-257
4.6.2.
Canonical Variate Analysis of Acoustic Data (N Campbell & D Devereux)........................... 4-258
4.6.3.
Linear Discriminant Analyses of QTC View data.................................................................... 4-259
4.6.4.
Acoustics summary .................................................................................................................. 4-260
4.7. ECOLOGICAL RISK INDICATORS ............................................................................................. 4-261
4.8. TRAWL MANAGEMENT SCENARIO MODEL.......................................................................... 4-264
5. BENEFITS............................................................................................................................................... 5-266
6. FURTHER DEVELOPMENT ................................................................................................................. 6-267
7. ACHIEVEMENT OF OUTCOMES........................................................................................................ 7-269
8. CONCLUSIONS...................................................................................................................................... 8-271
9. RECOMMENDATIONS ......................................................................................................................... 9-273
10.
REFERENCES................................................................................................................................... 10-276
11.
ABBREVIATIONS............................................................................................................................ 11-281
12.
APPENDIX 1: INTELLECTUAL PROPERTY ................................................................................ 12-282
13.
APPENDIX 2: STAFF....................................................................................................................... 13-283
14.
APPENDIX 3: PROJECT STEERING COMMITTEE MEMBERS ............................................................ 14-285
15.
APPENDIX 4: SINGLE SPECIES TRAWL EXPOSURE................................................................ 15-286
GBR Seabed Biodiversity
v
FIGURES
Figure 2-1: DEM of the bathymetry, slope and aspect of the GBR continental shelf, on a 0.01º grid, from various
sources including soundings in uncharted areas recorded by the Project; map of modeled seabed current
shear stress (RMS N/m²) (sources, see Section 2.1.1.1 for). ........................................................................ 2-9
Figure 2-2: Maps of sediment attributes for the GBR continental shelf: percent mud/sand/gravel grain size
fractions and percent carbonate (source, Geoscience Australia. Includes samples collected by the project
and processed by GA). ............................................................................................................................... 2-10
Figure 2-3: Maps of CARS bottom water physical attributes for the GBR continental shelf: temperature (mean &
SD ºC), salinity (mean & SD ‰), dissolved oxygen (mean & SD ml/l) (source, see Section 2.1.1.1). ..... 2-11
Figure 2-4: Maps of CARS bottom water nutrient attributes: silicate (mean & SD μM), nitrate (mean & SD μM),
and phosphate (mean & SD μM), (source, see section 2.1.1.1). ................................................................ 2-12
Figure 2-5. Maps of SeaWiFS predicted chlorophyll-A (mean & SD mg/m³), light absorption (attenuation
coefficient K) at 490 nm (mean & SD m⎯¹), benthic irradiance (relative to sea surface at equator estimated
from latitude, K490 and Depth), and weighted average annual trawl effort (hrs/0.01º grid) for the GBR
continental shelf (sources, see section 2.1.1.1). ......................................................................................... 2-13
Figure 2-6. Partitioning covariate space in two dimensions: (a) 1,000 points randomly sampled from the square
covariate space. (b) a partitioning into 20 clusters using PAM; (c) a preferred partitioning that accounts for
the relative importance of the variables; (d) the partitioning in (c) is achieved using PAM on the scaled
covariate space. .......................................................................................................................................... 2-16
Figure 2-7. Variable importance computed by (a) cross-validated trees and (b) random forests....................... 2-17
Figure 2-8. Importance measures without reliability (IbioQ), with reliability (IbioQR), and tuned reliability
(IbioQR)0.74 to match the shape without reliability. Each version is normalized to sum to 1. The orders of the
variables with and without reliability are different. ................................................................................... 2-20
Figure 2-9. (a) Bivariate normal distribution of 1,000 points. (b) Partitioning into 20 clusters using PAM. Each
cluster is labeled by the number of points in the cluster. The more populous clusters tend to be tighter and
so more homogeneous................................................................................................................................ 2-23
Figure 2-10. Density of bottom stress estimated by a Gaussian kernel of width 0.01 calculated using biased
cross-validation. Also shown is a ‘rug’ of values for 200 randomly selected sites. ................................... 2-24
Figure 2-11. Number of subclusters vs primary cluster size for 3 different values of the exponent a. The sloping
line corresponds to Nsub ∝ S, the curve to Nsub ∝ √S, and the horizontal line to Nsub = constant. ............... 2-24
Figure 2-12. The 200 primary clusters in geographical space. Sixty of the clusters have been separated into six
panels in order to make them distinct and assess the degree of fragmentation. ......................................... 2-25
Figure 2-13. Distribution of the most important physical covariates on the full the GBR data (orange). The thin
curves are 90% confidence intervals for the density sampled from the clusters. For clarity we show
covariates on a log scale for bottom stress, a logit scale for mud and an inverse scale for chlA. Also shown
is a rug of 200 sample values (jittered for mud)......................................................................................... 2-26
Figure 2-14: Map of the biophysical stratification of the Great Barrier Reef continental shelf. Inset: colour key
showing distribution of seabed grids on the first two principal components (which explain 65% of the
variation) of the biologically weighted physical covariates; biplot vectors indicate direction and magnitude
of the major physical factors. ..................................................................................................................... 2-29
Figure 2-15: Map of the sites selected for sampling the seabed on the continental shelf in the GBR. Ä: sites for
benthic and trawl sampling; «: sites for benthic sampling only................................................................ 2-30
Figure 2-16: The 27 m Australian Institute Marine Science research vessel RV Lady Basten........................... 2-31
Figure 2-17: The 18 m Queensland Department of Primary Industries & Fisheries research trawler FRV
Gwendoline May. ....................................................................................................................................... 2-32
Figure 2-18: The Drop-Cam system being recovered after completion of a 500 m video transect and the surface
real-time monitoring, control and data acquisition system......................................................................... 2-33
Figure 2-19: Diagram of single BRUVS frame and housing. ............................................................................ 2-35
Figure 2-20: Applying camera and bait arm to BRUVS before deployment. Note ballast on frame and 8 mm
hauling rope................................................................................................................................................ 2-35
Figure 2-21: The epibenthic sled being deployed through the A-frame for a 200 m tow along the seabed; note the
weak link at the top of the bridle and retrieve chain leading to the rear of the sled. .................................. 2-36
Figure 2-22: Sediment pipe dredge, showing sister-clip for attachment behind the sled ................................... 2-37
Figure 2-23: Net plan for the eight fathom Florida Flyer net used for scientific trawl sampling and the net
suspended from the A-frame on the stern of the Gwendoline May. ........................................................... 2-37
Figure 2-24: Drop chain links and trawl boards................................................................................................. 2-38
Figure 2-25: Sorting the catch from the epibenthic sled on the 10 mm square mesh sieve drawer into major
taxonomic classes....................................................................................................................................... 2-39
GBR Seabed Biodiversity
vi
Figure 2-26: Samples of sorted dredge catch organisms with bar code labels ready to be photographed and data
recorded in the vessel data base. ................................................................................................................ 2-40
Figure 2-27: Data and images from each sled site were entered into the vessel database entry form that also
included a photo of the entire site sample (left) and of the sample (in this case, of echinoderms). ........... 2-40
Figure 2-28: A photograph of an entire trawl catch (site photo) showing the diversity of organisms. .............. 2-41
Figure 2-29: A sample of crustaceans showing the barcode label. The label number, weight and class were
entered into a data base at sea. ................................................................................................................... 2-42
Figure 2-30: Data and images from each trawl site were entered into the vessel database entry form that also
contained a photo of the entire catch (left) and of the sample (in this case, of an elasmobranch). ............ 2-42
Figure 2-31: Data entry screens of the Delphi video analysis software showing the trapezoid overlaid on the
paused video image. ................................................................................................................................... 2-44
Figure 2-32: Drop down lists of physical and biological attributes to be used in analyzing the video image.... 2-45
Figure 2-33: Image reference form in BRUVS2.1.mdb ..................................................................................... 2-48
Figure 2-34: Reference image for Pristipomoides multidens, with Lutjanus sebae, L. adetii and Epinephelus
undulatostriatus and E. areolatus in the background................................................................................. 2-48
Figure 2-35: An example of the form used in the laboratory to enter data obtained from the field samples. The
sample barcode number is entered and the database retrieves the site details including the sample
photograph from onboard the vessel. Individual species or OTU were then entered into genus or species
boxes (middle fields) and a pick list of names appears. By selecting the appropriate name the species
numbers and weights were then able to be recorded into the bottom RHS field........................................ 2-50
Figure 2-36: Identifying and processing invertebrate samples at the Queensland Museum. ............................. 2-50
Figure 2-37: Colour scheme used for species distribution mapping .................................................................. 2-57
Figure 38: Spans, medoids, deviance and deviance reduction due to partition. (a) shows a group of ten sites
using two-dimensional Euclidean distances to represent Bray-Curtis distances. (b) shows the span of the
group from an arbitrary reference site. (c) shows the minimum span, which defines the medoid; the sum of
squared distances from the medoid is then the deviance of the group. (d) shows a partition of the original
group into two subgroups, and the spans defining the deviance component of each. The partition is
achieved by a split on the x-coordinate. The reduction in deviance achieved by partition is then 32.83-14.28
= 18.55. ...................................................................................................................................................... 2-61
Figure 2-39: An acoustic data sample from site 1505, with indicated depth of 23.97 m, shown in the original
distance/time domain, and after transformation to the angular domain. .................................................... 2-68
Figure 2-40: Measured depth and pressure signals derived from the drop camera and sonar transducer for site
1505, highlighting the segment of matched sonar data for which the calculated delay result is 0.502 minutes.
................................................................................................................................................................... 2-69
Figure 2-41: Three dimensional plot of the first three Local Discriminant Basis coordinates for data representing
(sand, no biohabitat) and (sand, seagrass) seabed types, generated with Daubechies 2 wavelet filter
coefficients, showing visible separation between the two classes.............................................................. 2-70
Figure 2-42. Total effort in the study area for the period 1993–2005. Also shown is the projected mean effort for
4 scenarios.................................................................................................................................................. 2-79
Figure 2-43. Pitcher et al. (2004) models (points) and fitted Schaefer model response (lines) for 0 to 10 initial
trawl tows for two OTUs: (left) Ianthella flabelliformis and (right) Junceella juncea. The vertical scale is
biomass relative to initial unimpacted biomass. The horizontal scale is years since impact...................... 2-81
Figure 2-44. Models obtained from sled video observations for 18 OTUs (Pitcher et al. 2004). The vertical (b)
axis represents biomass relative to initial unimpacted biomass, t is time since trawl impact (ranging from 0
to 5 years), and i is the number of trawl tows (ranging from 0 to 10)........................................................ 2-82
Figure 2-45. Recovery and depletion parameter estimates using sled (green) and ROV (blue) measurements.
Where both sled and ROV measurements are available the points are joined by a dashed line. Also shown
(red triangles) are the values from Poiner et al. (1998) and Hill et al. (2002) (not all labeled)................. 2-83
Figure 3-1: Patterns of species prevalence and richness at BRUVS stations. .................................................... 3-86
Figure 3-2: Species richness from BRUVS data by location in the GBRMP. ................................................... 3-87
Figure 3-3: "Heatmap" showing relationships amongst and between the top 20 predictors and 25 species
responses (presence/absence). The dendrogram along the side of the heatmap shows which species are
similar in having a relationship with a set of predictor variables. It does not imply these species have the
same relationship. The dendrogram along the top shows which explanatory variables cluster together, and
the coloured bars along the top show the percentage of the variation in the explanatory variables explained
by a particular variable. Red indicates higher influence. The “redness” of the individual cells in the figure
show the relative influence of the particular explanatory variable on the presence/absence of the particular
species, and the heaviness of the blue line shows the degree and shape of the relationship. ..................... 3-90
Figure 3-4: Species occurrence as a function of location across the shelf. Plots are ranked in descending order of
relative influence of the predictor variable for the species. The “rugs” on the X axes are 10 percentiles in the
distribution of the predictor variable. The Yaxes (log-odds) are Log(base 2) (1-Probability of occurrence)
and the plots are centred on the mean of Y. ............................................................................................... 3-91
Figure 3-5: Species occurrence as a function of mud content of the sediments. Details as for Figure 3-4. ....... 3-92
GBR Seabed Biodiversity
vii
Figure 3-6: Species occurrence as a function of carbonate content of the sediments. Details as for Figure 3-4.3-93
Figure 3-7: Species occurrence as a function of gravel content of the sediments. Details as for Figure 3-4. .... 3-94
Figure 3-8: Species occurrence as a function of average seawater temperature. Details as for Figure 3-4........ 3-95
Figure 3-9: Species occurrence as a function of average salinity. Details as for Figure 3-4.............................. 3-96
Figure 3-10: Species occurrence as a function of trawl effort index. Details as for Figure 3-4. ........................ 3-97
Figure 3-11 Predicted occurrence of 3 species of Nemipterus recorded by BRUVS. Circles represent observed
abundance (untransformed) and influential covariates are listed in the inset panels. “%XVar” describes the
percentage of the variation in the presence/absence of the species accounted for by the gbm model. “yres” is
(1-%prediction error). ................................................................................................................................ 3-99
Figure 3-12 Predicted occurrence of small benthic microcarnivores in the genera Pentapodus, Lethrinus and
Upeneus. Conventions as for Figure 3-11. ............................................................................................... 3-100
Figure 3-13 Predicted occurrence of small carangids in the genera Alepes, Decapterus, Selaroides and Seriolina.
Conventions as for Figure 3-11................................................................................................................ 3-101
Figure 3-14 Predicted occurrence of predators in the genera Scomberomorus, Echeneis, Saurida and Parapercis.
Conventions as for Figure 3-11................................................................................................................ 3-102
Figure 3-15 Predicted occurrence of demersal omnivores and predators in the genera Abalistes, Lagocephalus,
Paramonacanthus and Gymnothorax. Conventions as for Figure 3-11. .................................................. 3-103
Figure 3-16 Predicted occurrence of the large predatory carangids in the genera Carangoides and Gnathanodon.
Conventions as for Figure 3-11................................................................................................................ 3-104
Figure 3-17 Predicted occurrence of the large benthic macrocarnivore Choerodon venustus. Conventions as for
Figure 3-11............................................................................................................................................... 3-105
Figure 3-18: Multivariate regression tree analysis defining abundance (transformed by 4th root) of vertebrate
assemblages (top 25 species) in terms of location across and along the GBRMP (366 sites). The terminal
nodes represent 12 assemblages (see Table 3-4 for definitions of nodes), corresponding with different
regions of the GBRMP, and the higher level nodes represent the 11 assemblages at higher spatial scales.
The indicator species are shown with the DLI value for nodes where maxima in DLI occurred............. 3-106
Figure 3-19 Predicted distribution of 12 fish assemblages (terminal nodes from Table 3-4) as defined by the
explanatory variables “across” and “along” the shelf. ............................................................................. 3-108
Figure 3-20 Multivariate regression tree analysis defining abundance (transformed by 4th root) of vertebrate
assemblages (top 25 species) in terms of the top 20 environmental covariates in the GBRMP (366 sites).
The terminal nodes represent 12 assemblages (see Table 3-5 for definitions of nodes), corresponding with
various levels of mud, sand, gravel and silica and different regions of the GBRMP. The indicator species
are shown with the DLI value for nodes where maxima in DLI occurred. .............................................. 3-109
Figure 3-21 Predicted distribution of 12 fish assemblages (terminal nodes from Table 3-5) as defined by the top
20 explanatory environmental variables and location. ............................................................................. 3-111
Figure 3-22: Patterns of prevalence and richness of 4,723 species at 1,190 Sled stations. .............................. 3-114
Figure 3-23: Patterns of prevalence and richness of 3,510 species at 457 Trawl stations................................ 3-114
Figure 3-24: Species richness from epibenthic Sled data by location in the GBRMP. .................................... 3-115
Figure 3-25: Species richness from research Trawl data by location in the GBRMP. ..................................... 3-115
Figure 3-26: Example of a single species distribution map, for the Platycephalid fish, Elates ransonnetii..... 3-116
Figure 3-27: ROC curve for presence-absence estimation of Actinopterygii: Elates ransonnetii. As noted in the
previous figure, this ROC has an AUC of 0.97........................................................................................ 3-118
Figure 3-28: Scatter plot of the weighted versus unweighted AUCs for all species. ....................................... 3-119
Figure 3-29: Frequency distributions of species biomass distribution model performance diagnostics for the
presence model weighted AUC (P_AUCW) and for the biomass model relative deviance explained
(Deviance ratio). The median is indicated by the dashed vertical lines. .................................................. 3-120
Figure 3-30: Relationship between species biomass distribution model performance diagnostics for presence
model weighted AUC (P_AUCW) and for biomass model Deviance ratio. The medians are indicated by the
dashed lines. Symbol colour indicates frequency: least frequent=dark blue to most frequent=red.......... 3-120
Figure 3-31: Model distribution maps of selected species with higher performing diagnostics. ..................... 3-122
Figure 3-32: Model distribution maps of selected species with among the poorest performing diagnostics. .. 3-123
Figure 3-33: Model distribution maps of selected species with median performing diagnostics..................... 3-124
Figure 3-34: Model distribution maps of selected species with positive and negative affinities for mud........ 3-126
Figure 3-35: Model distribution maps of selected species with positive and negative affinities for benthic
irradiance.................................................................................................................................................. 3-127
Figure 3-36: Model distribution maps of selected species with positive and negative affinities for seabed current
stress......................................................................................................................................................... 3-128
Figure 3-37: Model distribution maps of selected species with affinities for shallow and deep bathymetry... 3-129
Figure 3-38: Model distribution maps of selected species within genera having contrasting distributions. .... 3-130
Figure 3-39: Model distribution maps of selected species within genera having contrasting distributions. .... 3-131
Figure 3-40: Model distribution maps of selected species within genera having contrasting distributions. .... 3-132
Figure 3-41: Cluster dendrogram of the single species estimates illustrating the hierarchical cluster structure
determined by Ward’s method. ................................................................................................................ 3-133
GBR Seabed Biodiversity
viii
Figure 3-42: Aggregated biomass map and model for an example species-group (“7”). The top 10 of 22 species
is tabulated with cumulative biomass....................................................................................................... 3-134
Figure 3-43: Model distribution maps of selected species groups. .................................................................. 3-136
Figure 3-44: Model distribution maps of selected species groups. .................................................................. 3-137
Figure 3-45: Recursive decision tree partitioning the sites into 16 groups, corresponding to the terminal nodes.
The labels indicate the split variable and threshold for the group corresponding to the left hand branch in
each case. The distances used were Bray-Curtis dissimilarities on 1/8th root transforms of the predicted site
species biomass data. ............................................................................................................................... 3-138
Figure 3-46: Dendrogram of biological similarities between the medoids of the site groups, as defined by the tree
Figure 3-45, based on hierarchical clustering of Bray-Curtis dissimilarities using Ward's method......... 3-139
Figure 3-47: Map of predicted distributions of 16 seabed assemblages (site groups clusters)......................... 3-140
Figure 3-48: Dendrogram for species, defining clusterings based on inter-species distances that reflect affinities
between species and site-groups. The red line shows a cut-off that defines the 12 groups used in this
analysis. The dendrogram was constructed using Ward’s method.......................................................... 3-141
Figure 3-49: Plot of relative biomass of 12 species affinity groups (labeled A–L) across the 16 site-group
assemblages mapped in Figure 3-47. ....................................................................................................... 3-142
Figure 3-50: Map of the distribution of seabed substratum types summarized as percent of transect length
observed by towed video camera. ............................................................................................................ 3-146
Figure 3-51: Map of the distribution of broad biological seabed habitat types summarized as percent of transect
length observed by towed video camera. ................................................................................................. 3-146
Figure 3-52: Photos of some example habitat types observed by towed video camera. .................................. 3-147
Figure 3-53: Map of the distribution and cover of conspicuous genera and other morpho-types of algae. ..... 3-148
Figure 3-54: Map of the distribution and cover of morpho-types of seagrasses. ............................................. 3-148
Figure 3-55: Map of the distribution and cover of conspicuous genera & other morpho-types of sponges..... 3-149
Figure 3-56: Map of the distribution and cover of conspicuous genera & other morpho-types of gorgonians.3-149
Figure 3-57: Map of the distribution and cover of conspicuous genera & other morpho-types of alcyonarian softcorals. ....................................................................................................................................................... 3-150
Figure 3-58: Map of the distribution and cover of morpho-types of bryozoans............................................... 3-150
Figure 3-59: Map of the distribution and cover of morpho-types of hard corals. ............................................ 3-151
Figure 3-60: Recursive partitioning of sites based on the grouped vessel biological cover proportions, the
Manhattan (Bray-Curtis) distance metric and the medoid partitioning algorithm.................................... 3-152
Figure 3-61: Mean profiles (centroids) of 9 site groups as defined by the recursive partitioning algorithm. .. 3-153
Figure 3-62: Map of predictions of group membership to the GBR grid. The groups are those from the medoid
algorithm with grouped vessel biological data and Manhattan distances shown in (Figure 3-60)........... 3-154
Figure 3-63: Results for a single two-class classification experiment for sand substratum with no biohabitat and
sand substratum with seagrass, calculated with a Tree classifier. Training data is from sites 1701 and 2441
and test data is from sites 1580 and 2083................................................................................................. 3-156
Figure 3-64: Classification results across 54 different test sets for training data from sites 1701 and 2441,
calculated with a Tree classifier. Circles and squares indicate test sets containing no contributions in
common with the training set. .................................................................................................................. 3-157
Figure 3-65: Mis-classification results from Figure 5 plotted against absolute depth difference between the sand
components of the training and test sets, showing a distinct depth divide near 15 m separating the good and
poor results............................................................................................................................................... 3-158
Figure 3-66: Tree classification results for (sand, no biohabitat) and (sand, seagrass) seabed types across 54 test
sets with training data from sites 2005 and 2441. .................................................................................... 3-158
Figure 3-67: Results of a single two-class classification experiment for (sand, sponge garden dense) and (sand,
seagrass) seabed types, generated with a Tree classifier. Training data is from sites 2580 and 2441 and test
data is from sites 2593 and 2084.............................................................................................................. 3-159
Figure 3-68: Tree classification results across 15 different test sets for training data from sites 2580 and 2441.
Circles indicate test sets containing no contribution in common with the training set. Seagrass is well
classified across the full test set range, while sponge garden undergoes larger variation to be the dominant
error source for those test sets with high mis-classification rates ............................................................ 3-160
Figure 3-69: Results for a single classification experiment on three seabed classes comprising (sand, no
biohabitat), (sand, sponge garden dense) and (sand, seagrass), calculated via Linear Discriminant Analysis.
Training data is from sites 1701, 2009 and 2441, and test data is from sites 1580, 2593 and 2083......... 3-161
Figure 3-70: Additional classification results for the three classes (sand, no biohabitat), (sand, sponge garden
dense) and (sand, seagrass), showing maximum confusion matrix diagonal elements against training/test
depth departure, for test sets containing contributions from all available sites........................................ 3-162
Figure 3-71: Results for a single classification experiment on four seabed classes, with training data taken from
sites 2191, 1701, 2009 and 2441, and test data from sites 2447, 1580, 2593 and 2083, calculated via a Tree
classifier. .................................................................................................................................................. 3-163
Figure 3-72: Computed four-class classification results for additional test sets, displayed as maximum confusion
matrix diagonal elements for each of the four seabed classes. ................................................................. 3-164
GBR Seabed Biodiversity
ix
Figure 3-73: Confusion matrix diagonals and overall mis-classification rates for a 5-class classification
experiment on selected substrata without biohabitat, generated with Linear Discriminant Analysis. Training
data is supplied from sites 1828, 1701, 2458, 2407 and 2163, with test data from sites 2315, 1580, 2750,
1897 and 1940.......................................................................................................................................... 3-166
Figure 3-74: (a) Plot of the original pelagic data against sample time for a shallow sandy site (depth = 12 m); and
(b) plot of depth-normalised data against sample number for the same site. ........................................... 3-167
Figure 3-75: (a) Plot of the original pelagic data against sample time for a deep sandy site (depth = 87 m); and
(b) plot of depth-normalised data against sample number for the same site. ........................................... 3-167
Figure 3-76: Plot of the depth-normalised pelagic data against sample number for sand sites for a range of
depths: (a) 12 m; (b) 20 m; (c) 50 m; and (d) 87.5 m............................................................................... 3-167
Figure 3-77: (a) Plot of the original pelagic data against sample time for a sand/algae group; and (b) plot of the
corresponding bottom data for the same group. ....................................................................................... 3-168
Figure 3-78: Plot of the depth-normalised pelagic data against sample number for (a) a shallow sandy site (depth
= 12 m); and (b) a deep sandy site (depth = 70 m)................................................................................... 3-168
Figure 3-79: (a) Plot of the response averaged over sample numbers 110 – 112 against depth for all groups of
>100 contiguous pings for all classes; and (b) plot of the difference in response for sample numbers 110 –
112 against depth for the same data. ........................................................................................................ 3-169
Figure 3-80: Plot of the response averaged over sample numbers 110 – 112 against depth for (a) class 617 (sand)
and (b) class 217 (coarse sand). ............................................................................................................... 3-169
Figure 3-81: Plots of the depth-normalised pelagic data against sample number for sand sites for depths ranging
from 12 m (a) to 85 m in (h). ................................................................................................................... 3-170
Figure 3-82: Plot of the response for the peak-aligned data against 100/depth for (a) sand class 617 for sample
number 113; (b) sand class 617 for sample number 115; (c) silt class 817 for sample number 113; (d) silt
class 817 for sample number 115; (e) mud class 917 for sample number 113; and (f) mud class 917 for
sample number 115. ................................................................................................................................. 3-171
Figure 3-83: Plot of the group means for the first two canonical variates for the canonical variate analysis of the
4519 groups from 117 classes, without regard to the class labels, for (a) the peak-aligned and depth-adjusted
data; and (b) the peak-aligned, row-corrected and depth-adjusted data. .................................................. 3-173
Figure 3-84: Plot of the first canonical vector for the canonical variate analysis of the 4519 groups from 117
classes, without regard to the class labels, for the peak-aligned and depth-adjusted data........................ 3-174
Figure 3-85: Plot of the canonical vectors for the canonical variate analysis of the 4519 groups, for (a) the peakaligned and depth-adjusted data; and (b) the peak-aligned, row-corrected and depth-adjusted data. ...... 3-174
Figure 3-86: Plot of (a) the second canonical vector for the canonical variate analysis of the 4519 groups for the
peak-aligned and depth-adjusted data; and (b) the first canonical vector the peak-aligned, row-corrected and
depth-adjusted data. ................................................................................................................................. 3-174
Figure 3-87: Plot of (a) the third canonical vector for the canonical variate analysis of the 4519 groups for the
peak-aligned and depth-adjusted data; and (b) the second canonical vector the peak-aligned, row-corrected
and depth-adjusted data............................................................................................................................ 3-174
Figure 3-88: Plot of the group means for the first two canonical variates for the canonical variate analysis of the
4519 groups from 117 classes, without regard to the class labels, for the peak-aligned, row-corrected and
depth-adjusted data for (a) sand class 617; (b) seagrass class 621; (c) silt class 817; (d) mud class 917. 3-175
Figure 3-89: Plot of the group means for the first two canonical variates for the canonical variate analysis of the
4519 groups which result from contrasts between class 617 vs classes 618 and 621, and class 618 vs class
621, for the peak-aligned, row-corrected and depth-adjusted data for (a) all groups for all classes; (b) sand
class 617; (c) sponge class 618; and (d) seagrass class 621. .................................................................... 3-176
Figure 3-90: Plot of the group means for the first canonical variate for the canonical variate analysis of the 4519
groups for the peak-aligned, row-corrected and depth-adjusted data (a) against depth; and (b) against
1/depth...................................................................................................................................................... 3-177
Figure 3-91: Plot of the group means for the first canonical variate for the canonical variate analysis of the 4519
groups which result from contrasts between class 617 vs classes 618 and 621, and class 618 vs class 621,
for the peak-aligned, row-corrected and depth-adjusted data (a) against depth; and (b) against 1/depth. 3-177
Figure 3-92: Plot of the group means for the first canonical variate for the canonical variate analysis of the 4519
groups which result from contrasts between class 617 vs classes 618 and 621, and class 618 vs class 621,
for the peak-aligned, row-corrected and depth-adjusted data against depth for (a) all groups for all classes;
(b) sand class 617; (c) sponge class 618; and (d) seagrass class 621. ...................................................... 3-178
Figure 3-93: Plot of the group means for the first canonical variate for the canonical variate analysis of the 4519
groups which result from contrasts between class 617 vs classes 618 and 621, and class 618 vs class 621,
for the peak-aligned, row-corrected and depth-adjusted data against 1/depth for (a) all groups for all classes;
(b) sand class 617; (c) sponge class 618; and (d) seagrass class 621. ...................................................... 3-179
Figure 3-94: Plot of the group means for the first two canonical variates for the canonical variate analysis of the
3358 groups in the larger CV1-CV2 cluster for the peak-aligned, row-corrected and depth-adjusted data for
(a) all groups; (b) sand class 617; (c) seagrass class 621; and (d) mud class 917. ................................... 3-180
GBR Seabed Biodiversity
x
Figure 3-95: Plot of the group means for the first canonical variate for the canonical variate analysis of the 3358
groups for the peak-aligned, row-corrected and depth-adjusted data against depth................................. 3-180
Figure 3-96: Plot of the group means for the first two canonical variates for the canonical variate analysis of the
1161 groups in the smaller CV1-CV2 cluster for the peak-aligned, row-corrected and depth-adjusted data
for (a) all groups; (b) sand class 617; and (c) mud class 917 ................................................................... 3-181
Figure 3-97: Plot of the group means for the first canonical variate for the canonical variate analysis of the 1161
groups for the peak-aligned, row-corrected and depth-adjusted data against depth................................. 3-182
Figure 3-98: Plots of (a) the canonical variate scores and (b) the group means for the first two canonical variates
for a canonical variate analysis of the depth-normalised data for 42 groups from sites 1631 and 2552,
without regard to the class labels. ............................................................................................................ 3-182
Figure 3-99: Plots of the depth-normalised pelagic data against sample number for group means for (a) 12 groups
for the seagrass site 1631 (class 621); and (b) 30 groups for the sand site 2552 (class 617). .................. 3-183
Figure 3-100: Plots of (a) the canonical variate scores and (b) the group means for the first two canonical variates
for a canonical variate analysis of the depth-normalised data for 56 groups from sites 1631, 2552 and 2224,
without regard to the class labels. ............................................................................................................ 3-183
Figure 3-101: Plots of the depth-normalised pelagic data against sample number for group means for (a) 30
groups for class 617 (sand) from site 2552, and (b) 14 groups for class 617 (sand) from site 2224. ....... 3-184
Figure 3-102: Plots of the echo responses for the depth-normalised pelagic data against sample number for
profiles 48 – 50 for group 53 (from the sand site 2552 - class 617)......................................................... 3-184
Figure 3-103: Plots of (a) the canonical variate scores and (b) the group means for the first two canonical variates
for a canonical variate analysis of the depth-normalised data for 61 groups from sites 1631, 2552, 2224 and
2441, without regard to the class labels. .................................................................................................. 3-185
Figure 3-104: Plots of the depth-normalised pelagic data against sample number for group means for (a) 12
groups for class 621 (seagrass) from site 1631, (b) 5 groups for class 621 (seagrass) from site 2441..... 3-185
Figure 3-105: Plots of the depth-normalised pelagic data against sample number for group means for (a) 14
groups for class 617 (sand) from site 2224 and (b) the group for class 617 from site 1580 superimposed on
the groups from site 2224......................................................................................................................... 3-185
Figure 3-106: Distribution maps of the most exposed species groups (a) exposed over 50 %, (b) – (d) exposed by
25-50%..................................................................................................................................................... 3-200
Figure 3-107: Distribution maps of the most exposed species groups: (a) and (b) exposed by 25-50%; and species
groups with negative trawl effort coefficients and possible population decreases in abundance as a result of
trawling of >5%; (c) -5.3% and (d) -6% respectively. ............................................................................. 3-201
Figure 3-108: Model distribution maps of selected species with higher trawl exposure indicators................. 3-209
Figure 3-109: Model distribution maps of selected species with higher trawl exposure indicators................. 3-210
Figure 3-110: Model distribution maps of selected species with higher trawl exposure indicators................. 3-211
Figure 3-111: Model distribution maps of selected species with higher trawl exposure indicators................. 3-212
Figure 3-112: Model distribution maps of selected species with higher trawl exposure indicators................. 3-213
Figure 3-113: Model distribution maps of selected species with higher trawl exposure indicators................. 3-214
Figure 3-114: Model distribution maps of selected species with higher trawl exposure indicators................. 3-215
Figure 3-115: Model distribution maps of selected species with higher trawl exposure indicators................. 3-216
Figure 3-116: Model distribution maps of selected species with larger negative trawl coefficients................ 3-220
Figure 3-117: Model distribution maps of selected species with larger negative trawl coefficients................ 3-221
Figure 3-118: Model distribution maps of selected species with larger positive trawl coefficients................. 3-225
Figure 3-119: Model distribution maps of selected species with multiple trawl coefficients. ......................... 3-227
Figure 3-120: Plot of estimated percentage of population caught against mean RSA recovery rank. Species at
greatest potential risk should plot towards the upper left quadrant. The top ranking species are labeled with
the first three letters of their genus and species name (see Table 3-57). .................................................. 3-229
Figure 3-121: Model distribution maps of selected species with higher relative risk identified from exposure and
SRA recovery attributes. .......................................................................................................................... 3-231
Figure 3-122: Model distribution maps of selected species with higher relative risk identified from exposure and
SRA recovery attributes. .......................................................................................................................... 3-232
Figure 3-123: Model distribution maps of selected species with highest sustainability risk indicators........... 3-237
Figure 3-124. Total annual effort averaged over 20 replicate simulations for the 7 scenarios......................... 3-243
Figure 3-125. Average and standard deviations over 20 replicates for scenarios SQ2001 (blue) and
CL/BB2001+P+RAP+BB2005 (green) of (a) effort and (b) relative biomass......................................... 3-244
Figure 3-126. Relative biomass histories for all scenarios for two widely different vulnerability types: (left) a
highly resilient taxon (r, d) = (0.7, 0.1); (right) a highly vulnerable taxon (r, d) = (0.1, 0.44). ............... 3-244
Figure 3-127. Average density of genus- and higher-level taxa in 2025 under each scenario. ........................ 3-245
Figure 3-128. Average density of individual species in 2025 under each scenario.......................................... 3-246
Figure 3-129. Time histories since 1990 of mean density 20 individual species under all scenarios. ............. 3-248
Figure 3-130. Time histories since 1990 of mean density of 18 genus- and higher-level taxa under all scenarios.
................................................................................................................................................................. 3-248
GBR Seabed Biodiversity
xi
TABLES
Table 2-1. Correlation matrix of physical environmental covariates. Non-significant correlations are greyed;
larger positive and negative correlations >0.05 are highlighted................................................................. 2-14
Table 2-2. Calculation of adjusted importance Iadj: derr is error distance in degrees, Ibio is the random forests biotic
importance, reliability is R = (derr)–½, and Iadj = (IbioQR)0.74........................................................................ 2-19
Table 2-3. Variable loadings for the first 7 principal components. Absolute loadings greater than 0.5 are
highlighted in yellow, and absolute loadings between 0.3 and 0.5 are highlighted in green. The variables are
ordered by adjusted importance. Relative variance is the fraction of the total variance explained by the
principal component................................................................................................................................... 2-21
Table 2-4. Voyages completed by the Lady Basten with scheduled duration, numbers of sites sampled by towed
camera, epibenthic sled and BRUVS. ........................................................................................................ 2-31
Table 2-5: Voyages completed by the Gwendoline May with scheduled duration, numbers of scientific trawl sites
sampled, sites with hookups and those too rough to trawl. ........................................................................ 2-32
Table 2-6: Substratum and Biological habitat types and animal events types entered in real time to annotate the
video transect. Numbers in parentheses show index values used in acoustics sections. ............................ 2-34
Table 2-7: Designated preservation methods on board the vessel and destinations for further processing (MTQTVL = Museum of Tropical Queensland; QMSB-BRS = Queensland Museum South Brisbane; CMR-CV =
CSIRO Cleveland, QDPI-TVL = Queensland Department of Primary Industries Townsville)................. 2-39
Table 2-8: The taxonomic groups into which samples were sorted onboard and their specific requirements for
preservation on board the vessel and destinations for further processing were provided. ......................... 2-42
Table 2-9: Sediment and group biological cover classes for analysed video tow data. Note that sediment classes
up to lage pebble could be further classified as rippled or in waves and cobble as waves......................... 2-64
Table 2-10. Habitat Events re-coding table showing mapping from the original BioHabitat code to
Habitat_Code2. .......................................................................................................................................... 2-74
Table 2-11. Habitat Events re-coding table showing mapping from the original BioHabitat code to
Habitat_Code3. .......................................................................................................................................... 2-74
Table 2-12. Substratum Events re-coding table showing mapping from the original Substratum code to
Substratum_Code2 ..................................................................................................................................... 2-75
Table 2-13. Substratum Events re-coding table showing mapping from the original Substratum code to
Substratum_Code3. .................................................................................................................................... 2-75
Table 2-14. Effort (boat days) in various regions of the East Coast Trawl Fishery, 2001–2005. ...................... 2-78
Table 2-15. Parameters ci, ci, ct, citt and ctt for the ROV recovery data from Pitcher et al. (2004); and
corresponding estimates r and d fit by non-linear least squares. *For Subergorgia suberosa and
Solenocaulon the value r = 0.22 was used. ................................................................................................ 2-80
Table 2-16. Parameters ci, ci, ct, citt and ctt for the sled recovery data from Pitcher et al. (2004); and
corresponding estimates r and d fit by non-linear least squares................................................................. 2-81
Table 2-17. Parameters r and d for coarse taxonomic groupings; d comes from Poiner et al. (1998) and r from
Hill et al. (2002)......................................................................................................................................... 2-83
Table 2-18. Relationship between modeled OTU and the source OTU for providing r and d at 3 levels of
taxonomic resolution: A) species, B) genus, and C) coarse (family or higher). The 3rd column indicates
whether a GLM model exists for the OTU................................................................................................. 2-85
Table 3-1: Twenty-five most predictable species (y) using best 20 explanatory variables. "yres" = (1 %prediction error). “%Var” is the percentage of the variation in presence/absence of the species explained
by the best gbmmv model, for production of biophysical maps. ............................................................... 3-88
Table 3-2: Top 20 explanatory variables (x) sorted by descending order of "xres" = (% of [1-% prediction error]
for each x). “xvar” is the mean percentage of the variation in the responses (species occurrence) explained
by each of the explanatory variables in the best gbmmv model, for production of biophysical maps. ...... 3-88
Table 3-3: Matrix of percentage of the variability in occurrence of 25 species responses explained by the top 20
explanatory variables. ................................................................................................................................ 3-89
Table 3-4: Hierarchy of nodes in the multivariate tree using location along and across the shelf to represent the
location of species assemblages. The number of BRUVS stations and of species with maxima in DLI values
are listed for each node. Terminal nodes are in bold font. ....................................................................... 3-107
Table 3-5 Hierarchy of nodes in the multivariate tree using spatial and environmental covariates to represent the
location of species assemblages. The number of species with maxima in DLI values are listed for each node.
Terminal nodes are in bold font. .............................................................................................................. 3-110
Table 3-6: Number of OTUs by Phyla sampled by the epibenthic sled and research trawl, and in the merged
dataset. ..................................................................................................................................................... 3-112
Table 3-7: Overall total and mean sampling rates (g per Ha) for the major Phyla sampled by the epibenthic Sled
and research Trawl, indicating overall composition and relative catchability. Ratio shows the trawl sampling
rate relative to the sled. ............................................................................................................................ 3-113
Table 3-8. List of species comprising species group 7..................................................................................... 3-135
GBR Seabed Biodiversity
xii
Table 3-9: A sample confusion matrix for the two-class case of sand and seagrass. ....................................... 3-155
Table 3-10: Sites with at least 1,000 consecutive samples of (sand, no biohabitat) seabed type . ................... 3-155
Table 3-11: Sites with at least 1,000 consecutive samples of (sand, seagrass) seabed type............................. 3-156
Table 3-12: Sites with 1,000 consecutive samples of (sand, sponge garden dense) seabed type (6,18). ......... 3-159
Table 3-13: The calculated confusion matrix at feature dimension 50 from Figure 3-69. ............................... 3-161
Table 3-14: Sites with 1,000 consecutive samples of (sand, bioturbated) seabed type (6,5). .......................... 3-163
Table 3-15: The calculated confusion matrix at feature dimension 40 from Figure 3-71. ............................... 3-163
Table 3-16: Sites with 1,000 consecutive samples of (coarse sand, no biohabitat) seabed type (2,17). .......... 3-165
Table 3-17: Sites with 1,000 consecutive samples of (sand waves/dunes, no biohabitat) seabed type (7,17). 3-165
Table 3-18: Sites with 1,000 consecutive samples of (silt, no biohabitat) seabed type (8,17). ........................ 3-165
Table 3-19: Sites with 1,000 consecutive samples of (soft mud, no biohabitat) seabed type (9,17)................ 3-165
Table 3-20: The calculated confusion matrix at feature dimension 40 from Figure 3-73. ............................... 3-166
Table 3-21: Intercept, slope and r2 values for regressions of the response at various sample numbers for the peakaligned depth-normalised pelagic data on 1/depth. .................................................................................. 3-172
Table 3-22. Sub1_Hab2: Observed (Row Totals) counts and percentage contribution, Observed versus Predicted
Diagonal counts and percentages. ............................................................................................................ 3-187
Table 3-23. Sub2_Hab2: Observed (Row Totals) counts and percentage contribution, Observed verse Predicted
Diagonal counts and percentages. ............................................................................................................ 3-188
Table 3-24. Habitat_Code2: Confusion matrix of total counts observed vs. predicted.................................... 3-189
Table 3-25. Habitat_Code2: Confusion matrix of percentage contribution as a percentage of row totals ....... 3-189
Table 3-26. Habitat_Code2: Confusion matrix of percentage contribution as a percentage of totals .............. 3-190
Table 3-27. Habitat_Code3: Confusion matrix of total counts observed vs. predicted.................................... 3-190
Table 3-28. Habitat_Code3: Confusion matrix of percentage contribution as a percentage of row totals ....... 3-191
Table 3-29. Habitat_Code3: Confusion matrix of percentage contribution as a percentage of totals .............. 3-191
Table 3-30. Substratum_Code1: Confusion matrix of total counts observed vs. predicted ............................. 3-192
Table 3-31. Substratum_Code1: Confusion matrix of percentage contribution as a percentage of row totals. 3-192
Table 3-32. Substratum_Code1: Confusion matrix of percentage contribution as a percentage of totals........ 3-192
Table 3-33. Substratum_Code2: Confusion matrix of total counts observed vs. predicted ............................. 3-193
Table 3-34. Substratum_Code2: Confusion matrix of percentage contribution as a percentage of row totals. 3-193
Table 3-35. Substratum_Code2: Confusion matrix of percentage contribution as a percentage of totals........ 3-193
Table 3-36. Substratum_Code3: Confusion matrix of total counts observed vs. predicted ............................. 3-194
Table 3-37. Substratum_Code3: Confusion matrix of percentage contribution as a percentage of row totals. 3-194
Table 3-38. Substratum_Code3: Confusion matrix of percentage contribution as a percentage of total ......... 3-194
Table 3-39. Substratum_Code3 Depth Partitioned: Confusion matrix of total counts observed vs. predicted 3-195
Table 3-40. Substratum_Code3 Depth Partitioned: Confusion matrix of percentage of rows ......................... 3-195
Table 3-41. Substratum_Code3 Depth Partitioned: Confusion matrix of percentage of total.......................... 3-195
Table 3-42. Summary of LDA classification performance of QTC View data................................................ 3-196
Table 3-43: Total area and percentage of the study area on the continental shelf of the GBRMP in various
management zones considered for estimating ecological risk indicators. ................................................ 3-197
Table 3-44: Total area of the study area on the continental shelf of the GBRMP exposed to various levels of
trawl effort, measured by VMS in 2005, considered for estimating ecological risk indicators. The total
effective area trawled and total area swept are also estimated. ................................................................ 3-197
Table 3-45: Ecological Risk Indicators with respect to trawling for estimated Biomass (tonnes) of correlated
species groups: by GBRMP Zoning indicating percent of biomass available; by areas not trawled/trawled
indicating percent biomass potentially exposed; by trawl intensity (ann_hrs/0.01º cell) indicating percent
biomass directly exposed to effort. .......................................................................................................... 3-199
Table 3-46: Ecological Risk Indicators with respect to trawling for estimated Biomass (kg) of species in
group#29: biomass available in General Use zone; biomass potentially exposed in trawled cells; and
biomass directly exposed to trawl effort. Pale orange: >25% biomass exposed; dark orange: >50% biomass
exposed; red: >50% biomass exposed...................................................................................................... 3-204
Table 3-47: Ecological Risk Indicators with respect to trawling for estimated Biomass (kg) of species in
group#9: biomass available in General Use zone; biomass potentially exposed in trawled cells; and biomass
directly exposed to trawl effort. Pale orange: >25% biomass exposed; dark orange: >50% biomass exposed;
red: >50% biomass exposed..................................................................................................................... 3-204
Table 3-48: Ecological Risk Indicators with respect to trawling for estimated Biomass (kg) of species in
group#22: biomass available in General Use zone; biomass potentially exposed in trawled cells; and
biomass directly exposed to trawl effort. Pale orange: >25% biomass exposed; dark orange: >50% biomass
exposed; red: >50% biomass exposed...................................................................................................... 3-205
Table 3-49: Ecological Risk Indicators with respect to trawling for estimated Biomass (kg) of species in
group#14: biomass available in General Use zone; biomass potentially exposed in trawled cells; and
biomass directly exposed to trawl effort. Pale orange: >25% biomass exposed; dark orange: >50% biomass
exposed; red: >50% biomass exposed...................................................................................................... 3-205
GBR Seabed Biodiversity
xiii
Table 3-50: Ecological Risk Indicators with respect to trawling for estimated Biomass (kg) of species in
group#13: biomass available in General Use zone; biomass potentially exposed in trawled cells; and
biomass directly exposed to trawl effort. Pale orange: >25% biomass exposed; dark orange: >50% biomass
exposed; red: >50% biomass exposed...................................................................................................... 3-206
Table 3-51: Ecological Risk Indicators with respect to trawling for estimated Biomass (kg) of species in
group#33: biomass available in General Use zone; biomass potentially exposed in trawled cells; and
biomass directly exposed to trawl effort. Pale orange: >25% biomass exposed; dark orange: >50% biomass
exposed; red: >50% biomass exposed...................................................................................................... 3-206
Table 3-52: Ecological Risk Indicators with respect to trawling for estimated Biomass (kg) of species in
group#27: biomass available in General Use zone; biomass potentially exposed in trawled cells; and
biomass directly exposed to trawl effort. Pale orange: >25% biomass exposed; dark orange: >50% biomass
exposed; red: >50% biomass exposed...................................................................................................... 3-206
Table 3-53: Ecological Risk Indicators with respect to trawling for estimated Biomass (kg) of top ranking
species in groups with low total exposure (<25%): biomass available in General Use zone; biomass
potentially exposed in trawled cells; and biomass directly exposed to trawl effort. Pale orange: >25%
biomass exposed; dark orange: >50% biomass exposed; red: >50% biomass exposed. .......................... 3-207
Table 3-54: Results for the Trawl Effort covariate: species with negative coefficients for presence (P) or biomass
(B), coefficients with p>0.05 are greyed, the magnitude of the coefficient in terms of overall % change in
abundance is also indicated. The group membership, total estimated biomass (kg), % available, % exposed
and effort exposed % are as above. .......................................................................................................... 3-218
Table 3-55: Results for the Trawl Effort covariate: species with positive coefficients for presence (P) or biomass
(B), coefficients with p>0.05 are greyed, the magnitude of the coefficient in terms of overall % change in
abundance is also indicated. The group membership, total estimated biomass (kg), % available, %exposed
and effort exposed are as above. .............................................................................................................. 3-223
Table 3-56: Results for the Trawl Effort covariate: species with an additional term involving the Trawl Effort
covariate, as well as coefficients for presence (P) or biomass (B), coefficients with p>0.05 are grayed, the
magnitude of the coefficient in terms of overall % change in abundance is also indicated. The group
membership, total estimated biomass (kg), % available, % exposed and effort exposed are as above. ... 3-226
Table 3-57: Summary of species exposure estimates for the top 280 of 840 species ranked by percent biomass
exposed to trawl effort intensity, showing estimated relative catchability from various sources and
indicative uncertainty, possible BRD effect for bycatch fish, leading to an estimate of potential percentage
of population caught annually. Recovery attributes from SRA and natural mortality estimates (M) are
tabulated where available. Where M was available, a sustainability indicator — estimate proportion
Caught/M — is also tabulated. Column * relates to indicator uncertainty due to catchability (see explanation
in text, page 3-230). ................................................................................................................................. 3-233
Table 3-58: Ecological Risk Indicators with respect to trawling for estimated area (km²) of predicted distributed
of species assemblages (site clusters): by GBRMP Zoning indicating percent of area available; by area not
trawled/trawled indicating percent area potentially exposed; by trawl intensity (ann_hrs/0.01º cell)
indicating percent area exposed to effort. ................................................................................................ 3-241
Table 3-59: Ecological Risk Indicators with respect to trawling for estimated area (km²) of predicted distributed
of video habitat clusters: by GBRMP Zoning indicating percent of area available; by area not
trawled/trawled indicating percent area potentially exposed; by trawl intensity (ann_hrs/0.01º cell)
indicating percent area directly exposed to effort. ................................................................................... 3-241
Table 3-60. Species with greatest affinity (top 40) for site-group assemblages identified in Section 3.4, with the
highest levels of trawl exposure. .............................................................................................................. 3-242
Table 3-61. Lowest historical (pre-2001) percentage relative biomass and final relative biomass in 2025 under
scenarios SQ2001 and SQ2006 for (left) species- and genus-level taxa and (right) coarse-level taxa. ... 3-246
Table 9-1. Top 20 ranked species for four different indicators. (a) Percent of biomass directly exposed to effort
(red = >75%, orange = >50%). (b) Estimated percent of biomass caught per annum (red = >75%, orange =
>50%, pale = >25%). (c) highest relative risk ranked species from plotting ‘Recovery’ rank from ‘SRA’
against estimated catch from b (no reference points are possible). (d) Sustainability indicator: estimated
catch b / natural mortality rate (red = exceeds limit reference point 1.0, orange = exceeds conservative
reference point 0.8, pale = exceeds conservative reference point 0.6). (e) Highest ranked species from
assemblage exposure and species affinities for assemblages (no reference points). ................................ 9-275
Table 15-1: Summary of species exposure estimates for all 840 modelled species, ranked in decending order by
percent of biomass exposed to trawl effort intensity; showing also species group membership, total
estimated biomass, percent of biomass available in General Use zone, percent of biomass potentially
exposed in trawled cells. Pale orange: >25% biomass exposed; dark orange: >50% biomass exposed; red:
>50% biomass exposed. ......................................................................................................................... 15-286
GBR Seabed Biodiversity
xv
NON-TECHNICAL SUMMARY
PROJECT:
CRC-REEF C1.1.2: Seabed Biodiversity on the Continental Shelf of the Great Barrier
Reef World Heritage Area (GBRWHA)
FRDC 2003/021: Mapping Bycatch & Seabed Benthos Assemblages in the GBR
Region for Environmental Risk Assessment & Sustainable Management of the
Queensland East Coast Trawl Fishery (QECTF).
NOO 2004/15: Assessment of the performance of acoustic remote sensing for Seabed
Mapping and as a surrogate for biodiversity on the continental shelf of the GBR
PRINCIPAL INVESTIGATORS:
Dr C. Roland Pitcher, CSIRO, Task Leader
Dr Peter Doherty, AIMS
Dr Neil Gribble, QDPI-F
Drs Peter Arnold & John Hooper, QM
ADDRESS:
CSIRO
PO Box 120,
Cleveland, Q.4164
Ph: 07 3826 7200
Fax: 07 3826 7222
AIMS
PMB 3 Townsville
MC, Q. 4810
07 4753 4211
07 4772 5852
QDPI-F/NFC
PO Box 5396,
Cairns, Q. 4870
07 4035 0100
07 4035 1401
Qld Museum
PO Box 3300, South
Brisbane, Q. 4101
07 3840 7722
07 3846 1226
roland.pitcher@csiro.au
p.doherty@aims.gov.au
neil.gribble@dpi.qld.gov.au
john.hooper@qm.qld.gov.au
OBJECTIVES:
The overall objective was to address key issues identified by the GBRMPA, QDPI&F, QSIA and their
advisory committees, in relation to biodiversity assessment and provision of information for future
Marine Park planning needs, and environmental sustainability assessments of the Qld East Coast
Trawl Fishery with respect to effects on bycatch, benthic assemblages and seabed habitat, to support
ecologically-based management of the fishery. Specifically:
1.
Develop comprehensive inventories & maps of the distribution and abundance of seabed
habitats & assemblages, as a benchmark of their current status, and provide these to GBR
Marine Park and Fisheries managers and stakeholders for future planning, management and
status reporting, with the outcome of management strategies effective in minimising fishery
effects on seabed habitats & assemblages, and achievement of Management Plan targets.
2.
Analyse bio-physical relationships and assess the utility of environmental correlates for spatial
prediction of assemblages of biodiversity.
3.
Provide information for refining the non-reef bio-regions defined for the continental shelf by
the Representative Areas Program (RAP) process
4.
Develop attributes (e.g. biomass, species richness, rarity, uniqueness, condition, potential
vulnerability to impact etc.) for bio-assemblages defined by this project, and for RAP bioregions, for future planning, management and World Heritage Area (WHA) reporting.
5.
Provide recommendations for monitoring trends in the status of bio-assemblages and highly
protected area (HPAs) selected by the RAP process.
6.
Develop & provide maps of the distribution of vulnerable seabed habitats and assemblages to
fishery managers and stakeholders, with the outcome of management strategies effective in
GBR Seabed Biodiversity
xvi
minimising fishery effects, achievement of Management Plan targets and for future planning
needs.
7.
For both bycatch & benthos, develop quantitative indicators of exposure to & effects of
trawling, and sustainability risk indicators, as required for the environmental Strategic
Assessment of the Queensland East Coast Trawl Fishery (QECTF).
8.
Provide critical input to a dynamic model of indicators for the status of seabed assemblages
and conduct ecological assessments of recent and proposed management changes using an
MSE approach, and enable capability for evaluation of future options.
9.
Contribute to quantifying the large-scale effects of trawling on bycatch species and benthos
assemblages by analysing their abundance across the range of trawl intensities within and
outside trawl grounds, while accounting for habitat variability.
10. Develop transferable scientific methods and tools to facilitate regional marine management
planning nationally, including: knowledge of bio-physical relationships between assemblages
and the physical environment (surrogates), cost-effective survey designs & techniques
(including development & performance assessment of non-invasive remote tools video &
single-beam acoustics for mapping seabed and as surrogates for patterns in seabed
assemblages), spatial-statistical classification & prediction methods, and sustainability risk
indicators for seabed species, assemblages and communities.
NON-TECHNICAL SUMMARY:
The Great Barrier Reef is a unique World Heritage Area of national and international significance. As
a multiple use Marine Park, activities such as fishing and tourism occur along with conservation goals.
Managers need information on habitats and biodiversity distribution and risks to ensure these activities
are conducted sustainably. However, while the coral reefs have been relatively well studied, less was
known about the deeper seabed in the region. From 2003 to 2006, the GBR Seabed Biodiversity
Project has mapped habitats and their associated biodiversity across the length and breadth of the
210,000 km² shelf in the Marine Park to provide information that will help managers with
conservation planning and to assess whether fisheries are ecologically sustainable, as required by
environmental protection legislation (e.g. EPBC Act 1999).
Holistic information on the biodiversity of the seabed was acquired by visiting almost 1,400 sites,
representing a full range of known environments, during 10 month-long voyages on two vessels and
deploying several types of devices such as: towed video and digital cameras, baited remote underwater
video stations (BRUVS), a digital echo-sounder, an epibenthic sled and a research trawl to collect
samples for more detailed data about plants, invertebrates and fishes on the seabed. Data were
collected and processed from >600 km of towed video and almost 100,000 photos, 1150 BRUVS
videos, ~140 GB of digital echograms, and from sorting and identification of ~14,000 benthic
samples, ~4,000 seabed fish samples, and ~1,200 sediment samples.
The project has analysed this information and produced all of the outputs as originally proposed; these
included:
•
Images and videos of seabed habitat types and fishes, including more than 150 substratum and
biological habitat component types; and >300 fishes, sharks, rays and sea snakes attracted to BRUVS.
•
An inventory of more than 5,300 species of benthos, bycatch and fishes, with catalogued
museum voucher specimens, many of which were new species, and a database of more than 140,000
records of species distribution and abundance on the seabed.
•
Identification of the key environmental variables likely to be important in structuring seabed
distributions, including: sediment grain size and carbonate composition, benthic irradiance, current
stress, bathymetry, bottom water physical attributes, nutrients and turbidity — and predictive models
of bio-physical relationships between seabed species, their assemblages and the physical environment.
•
Maps of the distribution and abundance of ~850 seabed species throughout the GBR shelf
region.
•
Estimates of the likely extent of past effects of trawling on benthos and bycatch over the entire
shelf of the GBR region, which indicated that trawl effort had a significant effect on the biomass of
GBR Seabed Biodiversity
xvii
6.5% of 850 species mapped; with negative biomass change of -1% to -36% for 4.5% of these species
and positive biomass change of +1% to +96% for 2% of these species.
•
Estimates of exposure to trawl effort showed that about 70% of the 850 species mapped had
low or very low exposure, and at the other extreme, about 33 species had high to very high exposure to
trawl effort — of these species, after taking relative catch rates into account, five had high estimates of
proportion caught annually and 28 were intermediate. The remainder (>800 species) had low or very
low estimates of proportion caught annually.
•
Indicators based on qualitative recovery ranks showed that about 15 species stood out as being
at higher relative risk with respect to trawling. An additional, quantitative absolute sustainability
indicator showed that three species (Fistularia petimba, Brachirus muelleri, Trixiphichthys weberi)
exceeded a limit reference point (analogous to maximum sustainable yield, MSY), one species
(Pomadasys maculatus) exceeded a first conservative reference point (≡0.8×MSY) and two others
(Psettodes erumei, Sillago burrus) exceeded a second conservative reference point (≡0.6×MSY) —
another 10 species were also listed due to uncertainty in parameters, though they were below the
sustainability reference points.
•
Evaluations of the environmental performance of several recent management interventions
showed that generalized depletion trends up until the late 1990s have all been arrested and reversed —
the 2001 buyback of fishing licences and subsequent penalties made the biggest positive contributions
with the 2004 rezoning of the marine park making a small positive contribution for some species.
•
Methods and tools for regional marine planning, including: representative survey design and
techniques, spatial-statistical classification and prediction methods, and biodiversity and bycatch
species risk assessment methods.
A key output from the project is the identification, by means of the trawl exposure and sustainability
indicators, of species at risk or potentially at risk from trawling. Different species were highlighted by
different indicators, though there was some overlap. The most quantitative indicator was directly
related to sustainability, with biologically based reference points — while three species appeared to be
at risk and another three species exceeded conservative reference points (as listed above), there was
uncertainty that requires a more precautionary response. Hence, the top ranked 50 species, across all
indicators developed, were listed herein and recommended to be considered for stakeholder
consultation regarding future action. Options may include clarification of the identified uncertainties,
monitoring of species at risk, management interventions that reduce risk or combinations of these
actions.
It is also recommended that long-term monitoring of trends in ecological condition of identified key
seabed habitats and constituent species be implemented to assess responses to regional pressures,
including climate change. Candidate habitats should include those that have been demonstrated to be
particularly species rich such as vegetated areas and epibenthic gardens. The seabed may be
vulnerable to climate change, as climate change modelling has indicated that the thermocline is likely
to deepen and upwellings to become weaker and less frequent, with potential consequences for
productive habitat dependent on nutrients from such sources. Such a possibility should be investigated.
Further work is needed to address the uncertainties in the risk assessments that arise from uncertainties
in estimates of catchability and natural mortality rates. Currently, the uncertainty is such that several
additional species could exceed the reference points and many species with unknown mortality might
be of concern. It is also possible that clarification of these uncertainties may show that species
currently thought to be at risk or potentially at risk may be demonstrated to be of no sustainability
concern. Thus, it is recommended that further studies of catchabilities and natural mortality rates be
conducted to address this key uncertainty for affected species. Such results are likely to have wide
application in risk assessments being conducted in multiple jurisdictions.
Many fisheries in Australia are conducting qualitative approaches to Ecological Risk Assessments.
The results of the quantitative sustainability indicators applied here raise concerns about the reliability
of the qualitative approaches, which have not been benchmarked because of the lack of a suitable test
bed. Such a test bed is now available with the Seabed Biodiversity dataset and an assessment of the
performance of the qualitative methods is recommended. The Seabed dataset also provides an
opportunity to develop condition and trend and vulnerability indicators for seabed communities and
ecosystems that are needed to meet the increasing requirement for ecosystem-based management
approaches.
GBR Seabed Biodiversity
xviii
The results of the Seabed Biodiversity Project have been adopted for marine park zoning assessment
by a follow on project supported by the Reef and Rainforest Research Centre and involving
collaboration with marine park managers. This project has contributed to ongoing marine park
planning with respect to meeting WHA obligations.
Another project supported by the CERF National Marine Biodiversity Hub will use the unique Seabed
Biodiversity dataset in comparisons with other datasets to test the inter-regional utility of physical
variables and cross-taxonomic patterns as surrogates for application in marine planning at a national
scale.
Other further opportunities include: sorting and identification of remaining samples that could not be
completed within the scope of the project, taxonomic work to properly identify the more difficult
specimens, and quantification of visible species from the available towed camera video and digital
photos to fill significant gaps in areas too rough for sampling and currently lacking species
information. These activities would provide full utilization of the samples and deliver additional value,
with expected benefits for greater understanding of the seabed ecosystem, fishery sustainability,
zoning assessment and ongoing marine park planning.
Outcomes Achieved
Preliminary outputs have been presented to management and stakeholder committees during the
course of the project and team members contributed to management/industry activities such as
bycatch risk assessments, assessments of trawl plan targets and marine park monitoring strategies.
The final results for each objective have become available only near the end of the project and, with
delivery of the final outputs and planned uptake activities, the anticipated outcomes may now be
achieved. Progress against expected industry, management and stakeholder outcomes has included:
•
Reports and presentations to various audiences and multimedia information available via a
website have contributed to raising the level of stakeholder knowledge of the status of the region’s
ecosystems, facilitating understanding of reasonable use, development and implementation of
regional ecosystem management plans to achieve sustainable and acceptable multiple use. Further
activities are planned to disseminate and adopt the outputs of the project among managers,
stakeholders and scientists. Further development of the website would be valuable.
•
Contributions by team members to management/industry assessments of the current Trawl
Plan targets and bycatch risk assessments. The 40% bycatch and 25% benthos reductions were
considered largely with respect to reductions in trawl effort; the outputs from this project have now
provided an assessment of their likely sustainability.
•
Ecological risk/sustainability indicators and biological reference points, developed with
management and industry involvement, and evaluations of recent management changes showing
positive implications for the environment, have contributed to meeting an Environmental Assessment
condition on the wildlife trade operation (WTO) for this fishery and, with management response, can
be expected to facilitate meeting of environmental sustainability objectives under Commonwealth
legislation.
•
Indicators of the level of impact under the current management arrangements and biological
reference points that will contribute to planned revisions of the Trawl Management Plan.
•
Capability to evaluate future alternative management strategies needed to meet environmental
sustainability legislation, by estimating outcomes for the environment and for the fishery in a MSE
context.
KEYWORDS: Great Barrier Reef, seabed, biodiversity, habitat, trawl, epibenthic sled, video,
BRUVS, acoustics, benthos, bycatch, fish, distribution, abundance, biophysical relationships,
surrogates, statistical models, classification, prediction, mapping, effects of trawling, ecological risk
assessment, sustainability indicators, biological reference points, management strategy evaluation,
effort reduction, zoning, survey design, stratification.
GBR Seabed Biodiversity
1-1
1. INTRODUCTION
1.1. BACKGROUND
The Key Result Areas of the CRC Reef Program C "Maintaining Ecosystem Quality" and Project C1
"Conserving Biodiversity" included “...to know the status and trends of marine ecosystems within the
GBRWHA…” and “...contribute to regional marine planning...”. This project has delivered to these
KRAs and the outputs will benefit the GBRMPA by assisting evaluation of the marine park zoning
and future planning needs for managing human uses, to optimise trade-offs in a multiple-use
environment to meet its goals of conserving habitat and species diversity, and to meet Australia's
international obligations for reporting and monitoring the status of values in the WHA.
The goals of the Queensland Trawl Management Plan included achieving environmental sustainability
with respect to the Queensland East Coast Trawl Fishery (QECTF), as a response to stakeholder and
community concerns about effects of trawling. The CSIRO/QDPI Effects of Trawling Study (Poiner et
al., 1998; GBRMPA & FRDC 93/096) clearly showed that the cumulative effects of repeated trawling
could be substantial, though the impacts differ greatly for different types of organisms and for
different habitats. The overall impacts depend not only on how much life is removed when a trawl
passes and how fast the seabed recovers between trawls, but most importantly also on where trawling
occurs in relation to where the vulnerable seabed plants and animals live. The most vulnerable types
of seabed organisms are those that are easily removed or killed and/or slow to recover — as
vulnerability is a function of mortality and recovery rates. That Study concluded that prawn trawling
has the potential to be environmentally sustainable, but there remained important assessments that
needed to be conducted. The additional information needs for these assessments include:
1. Distribution of trawling effort at scales of 10x10 km and finer (from logbooks and VMS)
2. Recovery rates of fauna after trawling (GBRMPA Trawl Recovery Project & FRDC 97/205
Megabenthos Dynamics)
3. Distribution of seabed fauna and bycatch throughout the GBR (provided by this project)
4. Impacts of trawling on soft-sediment fauna (FRDC 2002/102 Effects of Trawling in the NPF)
This information was critical because the status of vulnerable seabed organisms is greatly affected by
their spatial exposure to trawl effort, in addition to impact and recovery rates. The environmental
sustainability of the fishery would be greatly facilitated and demonstrable if the distribution of
vulnerable seabed life could be taken into account in management of the distribution of trawl effort.
The provision of this knowledge of biotic assemblage distributions by this project has overcome a
major impediment. These data have now been synthesized by this project, into a Trawl Management
Scenario Model to conduct ecological assessments of alternative management strategies (Management
Strategy Evaluation, MSE).
The Trawl fishery, like other fisheries, has been required to conduct environmental Strategic
Assessments and respond to points raised by the assessors (Department of Environment & Water
Resources, DEW) — this project has provided quantification of the broad scale effects of trawling on
benthos and bycatch, and developed sustainability risk indicators as key components of the response
required for this Strategic Assessment.
The inventories of the distribution and abundance of bycatch species and seabed fauna will also enable
development and evaluation of future strategies to minimise impacts and improve the environmental
sustainability of the fishery. This will assist managers to respond effectively to industry and
community concerns and achieve an informed objective balance between the pressures of exploitation
and needs for conservation in a multiple-use environment. The community will have available
objective and independent information on the environmental sustainability of trawling.
GBR Seabed Biodiversity
1-2
The benefits of this project’s outputs to the Queensland trawl fishery, its managers and the community
include a factual biological/ecological basis for objective management decisions and facilitating the
assessment of the QECTF Management Plan’s stated goals of reducing catch of benthos by 25% and
catch of bycatch by 40%. The Project has also developed and delivered operational ESD indicators for
use under State and Commonwealth fishery and environmental legislation, including the EPBC Act;
and for the national ESD reporting system.
The outputs from this project have national applicability to the implementation of Australia’s Oceans
Policy, Regional Marine Planning and to the National Work Programs of the National Oceans Office.
These outputs have contributed to our understanding of how acoustic remote sensing instruments can
be used in approaches to seabed habitat mapping and deliver to priorities identified at the joint
NOO/FRDC Habitat Workshop in Melbourne 23-24 September 2002. The large acoustics data-set
collected by the project has been extremely valuable for this purpose and can serve as an example for
marine assessment for conservation planning and multiple-use management (incl. the National
Representative System of marine Protected Areas – NSRMPA) elsewhere.
The objectives of the project have also delivered to the highest 'High Priority' research areas identified
by the Biological Diversity Advisory Council as well as the “areas of research of national importance”
(Biodiversity Research: Australia’s Priorities — a Discussion Paper. Environment Australia, 2000).
1.2. NEED
Information from strategic mapping and analysis of seabed biodiversity was a fundamental need to
assess the status and condition of benthic biodiversity in the large and complex ecosystems of the
GBR seabed; to establish benchmarks and performance indicators for feedback to management; and to
develop sustainability risk indicators and facilitate detection of anthropogenic impact in seabed
ecosystems (particularly trawling) among the milieu of natural environmental variability.
With respect to trawling, the CSIRO/QDPI Effects of Trawling Study (Poiner et al., 1998) concluded
that if the potentially substantial cumulative environmental effects of trawling are to be managed for
sustainability then fundamental information on the distribution and abundance of seabed assemblages
and bycatch would be needed. The “Trawl Management Scenario Model” for the QECTF indicated
that potential sustainability indicators for Management Strategies Evaluations (MSE) are highly
sensitive to current assumptions about the distribution and abundance of species vulnerable to
trawling. This project has addressed this important information gap and impediment to management
for environmental sustainability by conducting a comprehensive inventory and mapping of seabed
assemblages and species caught in bycatch throughout the GBR region, for development of
sustainability risk indicators and MSE approaches.
Bycatch sustainability has been a priority issue in the QECTF. This project has addressed the
information needs of this issue in two ways: by (1) developing bycatch sustainability risk indicators
and (2) quantification of the impacts on populations of bycatch species. To address (1), the project has
mapped the distribution and abundance of species caught in bycatch, within and beyond trawl grounds,
and estimated the proportion of their populations exposed to trawling by conducting spatial analyses of
bycatch species abundance in relation to trawl effort distribution and intensity. For (2) this analysis
was developed further, using available data on the catch-rate of bycatch species by the fishery, to
estimate the proportion of bycatch populations caught annually, as a risk indicator. The Project has
also applied the bycatch vulnerability criteria for life history traits (recovery), which were developed
by the NPF Bycatch Sustainability Project (FRDC 96/257), and are now playing a key role in the NPF
Bycatch Action Plan. Together, this information has been used to identify those species potentially at
risk in the QECTF and has delivered directly to the bycatch reporting requirements for Strategic
Assessment and subsequent accreditation outcomes. Similar information has become available for
several target and byproduct species, as well as some threatened or potentially threatened species such
as syngnathids.
The direct impacts of trawling on seabed benthic assemblages have also been a priority issue in the
QECTF. This project has addressed the information needs of this issue by mapping the distributions of
GBR Seabed Biodiversity
1-3
seabed assemblages, conducting spatial analyses and developing benthos sustainability indicators
similar to that for the bycatch. This has been done by applying the vulnerability algorithms developed
for the CSIRO/QDPI Effects of Trawling Study FRDC 93/096 (i.e. the dynamics of per trawl removal
rate × trawl-effort, plus recovery rate information from the GBRMPA follow-on project Seabed
Habitat Recovery Dynamics, as well as the FRDC 97/205 Megabenthos Dynamics Project). This
information has enabled development of benthos status indicators and evaluation of the environmental
performance of different management scenarios (MSE) that may be adopted by the fishery
management. Again, these outputs have delivered directly to the reporting requirements for Strategic
Assessment and subsequent accreditation outcomes.
Management and Industry considered the outputs from this project essential for the requirement to
provide a comprehensive assessment of the sustainability of the fishery. Information from this project
will assist stakeholders with their management of the fishery, including: assessment of performance
against Trawl Management Plan targets (40% reduction in bycatch and 25% reduction in benthos),
response to Strategic Assessment and meeting requirements of the EPBC Act, conduct of ecological
risk assessments and development of biologically meaningful reference points, evaluation of the
zoning changes in the GBRMP, review of the Trawl Management Plan (2004-06) — and reaching the
goal of achieving a sustainable fishery. The Project will deliver results progressively, so that timely
outputs will be available for these review processes.
This Project will also deliver information needed for the sustainability of the Queensland Reef Line
Fishery. A significant uncertainty regarding the sustainability of this fishery is the unknown area of
deeper reef habitat and the populations of demersal fishes therein — this Project has provided
estimates for these uncertainties and it is expected that the Effects of Line Fishing Project can
capitalise on this information. The Project will also deliver priority research needs relevant to the
development of national habitat classification and mapping methods, as identified at the FRDC/NOO
Habitat Workshop, 23-24 September 2002.
The challenging broad-scale objectives of this task have been met by multidisciplinary inputs from
collaborating specialists from several research providers. This approach will serve as a model for
marine conservation planning and multiple-use management (incl. NSRMPA) elsewhere and so is also
relevant to the needs of DEW's national objectives and the regional marine planning promoted by
Australia's Oceans Policy. In this context, the performance of so-called rapid marine assessment
methods, such as acoustic remote sensing, as a surrogate for patterns in seabed biodiversity needed to
be formally tested.
1.3. OBJECTIVES
The overall objective is to address key issues identified by the GBRMPA, QDPI&F, QSIA and their
advisory committees, in relation to biodiversity assessment and provision of information for future
Marine Park planning needs, and environmental sustainability assessments of the Qld East Coast
Trawl Fishery with respect to effects on bycatch, benthic assemblages and seabed habitat, to support
ecologically-based management of the fishery. Specifically:
1. Develop comprehensive inventories and maps of the distribution and abundance of seabed habitats
and assemblages, as a benchmark of their current status, and provide these to GBR Marine Park
and Fisheries managers and stakeholders for future planning, management and status reporting,
with the outcome of management strategies effective in minimising fishery effects on seabed
habitats and assemblages, and achievement of Management Plan targets.
2. Analyse bio-physical relationships and assess the utility of environmental correlates for spatial
prediction of assemblages of biodiversity.
3. Provide information for refining the non-reef bio-regions defined for the continental shelf by the
RAP process
GBR Seabed Biodiversity
1-4
4. Develop attributes (e.g. biomass, species richness, rarity, uniqueness, condition, potential
vulnerability to impact etc.) for bio-assemblages defined by this project, and for RAP bio-regions,
for future planning, management and WHA reporting.
5. Provide recommendations for monitoring trends in the status of bio-assemblages and HPAs
selected by the RAP process.
6. Develop and provide maps of the distribution of vulnerable seabed habitats and assemblages to
fishery managers and stakeholders, with the outcome of management strategies effective in
minimising fishery effects, achievement of Management Plan targets and for future planning
needs.
7. For both bycatch and benthos, develop quantitative indicators of exposure to and effects of
trawling, and sustainability risk indicators, as required for the environmental Strategic Assessment
of the QECTF.
8. Provide critical input to a dynamic model of indicators for the status of seabed assemblages and
conduct ecological assessments of recent and proposed management changes using an MSE
approach, and enable capability for evaluation of future options.
9. Contribute to quantifying the large-scale effects of trawling on bycatch species and benthos
assemblages by analysing their abundance across the range of trawl intensities within and outside
trawl grounds, while accounting for habitat variability.
10. Develop transferable scientific methods and tools to facilitate regional marine management
planning nationally, including: knowledge of bio-physical relationships between assemblages and
the physical environment (surrogates), cost-effective survey designs and techniques (including
development and performance assessment of non-invasive remote tools video and single-beam
acoustics for mapping seabed and as surrogates for patterns in seabed assemblages), spatialstatistical classification and prediction methods, and sustainability risk indicators for seabed
species, assemblages and communities.
GBR Seabed Biodiversity
2-5
2. METHODS
This Project characterised the major patterns in the seabed biodiversity and habitats of the Great
Barrier Reef, at spatial scales relevant to regional conservation and management needs. The
information included seabed species and habitat distribution in inter-reef areas, and physical attributes
that may drive patterns within the region.
The approach was to collate and integrate the available biological, habitat, physical and bottom-water
data; analyse bio-physical relationships to identify important environmental variables; stratify the
GBR seabed based on these variables weighted by their biological importance; design sampling and
conduct a series of seabed surveys to obtain representative inclusion of important biological
components, major habitat strata, and areas of uncertainty; sort and identify samples and analyse data
to produce predictive maps of habitats and biodiversity, which formed the basis of ecological risk
assessments for the trawl fishery in the region. Details of these approaches are described below.
2.1. SAMPLING DESIGN
The sampling design was based on a bio-physical stratification of the GBR continental shelf and the
analyses were based on the same types of broad-scale bio-physical data (updated where possible). This
required collation of available broad coverage bio-physical environmental datasets that were likely to
be important in influencing patterns of distribution of biota.
2.1.1. Physical environmental data (I McLeod & R Pitcher)
2.1.1.1. Datasets collated
Phase 1 of the project (Pitcher et al. 2002) collated 22 datasets of biological and physical data from
internal and external sources. There was some duplication of the data types available among datasets
from different sources, in which case the dataset providing the widest reliable coverage was selected.
The remaining variables useful for modelling and stratification included 21 physical factors and
measures of seasonal variability for eight of these.
The Effects of Trawling dataset for the Far Northern Section cross-shelf closure and adjacent
areas from CSIRO Marine Research. This dataset was used in the design phase and included:
•
•
•
•
•
Station information (lat/lon, mud/sand/gravel/carbonate, temperature, salinity), 311 sites
Catch of fish trawl net sampled fish species, 436 taxa from 292 stations
Catch of prawn trawl net sampled bycatch species, 926 taxa from 269 stations
Catch of epibenthic sled net sampled benthos species, 1194 taxa from 306 stations
Combined dataset from all devices (Kg/Ha), accounts for overlapping taxa, 1655 taxa from
311 stations
Bathymetry model grid, as used in RAP, collated from various sources by Adam Lewis, GBRMPA.
•
Bathymetry grid, resolution 15 arc-second (~500 m); used in design phase
Bathymetry from RAN Hydrographic Office (HO) and acquired by the project
•
•
•
Coverage extensive in GBR but not complete
Updated with soundings acquired by the project during fieldwork at sea
Used to produced a 36 arc-second resolution DEM for analysis phase
GBR Seabed Biodiversity
2-6
Seabed current-stress from Lance Bode & Lou Mason, JCU/Reef-CRC
•
•
Modelled coverage for the entire GBR region, 1 minute of arc resolution
RMS stress (Pascals (N/m²)) over period of model run (approximately 6 months)
Seabed sediment composition and related attributes, from Chris Jenkins OSI, Sydney University;
used in design phase
•
•
•
•
•
•
•
Coverage extensive, but not complete
Resolution: gridded at 0.01 degree, where samples were available
% carbonate
% gravel grainsize fraction, percent
% sand grainsize fraction, percent
% mud grainsize fraction, percent
Characteristic Grain size (phi)
Seabed sediment composition, samples collected by the project and processed by Geoscience
Australia (Mathews & Heap 2006); used in analysis phase
•
•
•
•
•
•
•
Approximately 1190 samples collected, average 13 km apart throughout the GBR shelf
% carbonate
% gravel grainsize fraction, percent
% sand grainsize fraction, percent
% mud grainsize fraction, percent
laser volume fractions
dataset supplemented by existing data from CSIRO, QDPI seagrass survey, and other sources
CSIRO Atlas of Regional Seas (CARS2000) from Scott Condie & Jeff Dunn, CSIRO Marine
Research Hobart (Ridgway et al. 2002). A weighted averaging, which takes into account bathymetry
and seasonality, of all available measurements of water column properties. Properties were evaluated
at the seabed:
•
•
•
•
•
•
•
full–coverage of the GBR region at 1/8 degree resolution (~14 km)
T – temperature – degrees C, mean and standard deviations
S – salinity – psu, mean and standard deviations
O2 – oxygen – ml/l, mean and standard deviations
Si – silicate – uM, mean and standard deviations
PO4 – phosphate – uM, mean and standard deviations
NO3 – nitrate – uM, mean and standard deviations
SeaWiFS oceancolor satellite data from Scott Condie & Jeff Dunn, CSIRO Marine & Atmospheric
Research Hobart. The data provided are chlorophyll concentration and turbidity, averaged over 36
months, based on SeaWiFS Calibration and Validation algorithms:
•
•
•
•
full-coverage of the GBR region, with 0.01° resolution ~(1.11 km)
Chlorophyll-a (mg/m³) concentration, mean and standard deviations
K490 diffuse attenuation coefficient at wavelength 490nm, m-1, mean and standard deviations
relative benthic irradiance calculated, based on K490, latitude and depth.
Queensland east coast trawl effort data from QDPI&F. The data are a combination of full-coverage
30-minute grid resolution logbook effort data and higher resolution voluntary data, mapped at 6minute resolution, and vessel monitoring system (VMS) data processed by Norm Good QDPI&F.
•
•
full-coverage of the GBR region,
logbook effort (boat-days) at 6 minute (0.1°) resolution (~11.1 km) for 1996–2001
•
•
•
GBR Seabed Biodiversity
2-7
VMS effort (hours) at (0.1°) resolution for 2002
a weighted average of the above 1996–2002 data was used for design phase
logbook data 1996–2001 and VMS data 2001–2005 updated for analysis phase
Qld State permanent spatial Trawl Closure Areas, from QDPI&F.
•
•
State permanent spatial Trawl Closure boundaries
updated in 2006 for analysis phase
GBRMPA Spatial Information, from GBRMPA,
•
•
•
•
GBR Marine Park Zoning
RAP Bioregions (Reef and Non-Reef)
Topographic coverages (Water, Shoal, Reef, Cay, Foreshore, Mangrove, Land)
2004 RAP Rezoning.
2.1.1.2. Data Processing:
After accounting for redundancy among the collated data, there were 21 covariates (+8 measures of
variability in the CARS and SeaWiFS attributes) for stratification of the GBR non-reef region and for
developing bio-physical models of biodiversity. These datasets were checked and imported into an
ArcInfo GIS.
The covariates were constrained to the continental shelf by establishing a base study area bounded by
the GBRWHA excluding those areas deeper than 80 m but including areas deeper than 80 m across the
mouth of the Capricorn Trough, and excluding those areas shallower than 7 m near the coast, or
shallower than 12 m if topographically identified as shoal, or topographically identified as reef.
A 36-arc-second grid (0.01 decimal degree, ~1.11 km) was generated for this area. Each grid cell was
assigned a unique identifier that was the key to this dataset. As the collated data were of various
spatial resolutions, we resampled those data to the 36-arc-second grid framework by a discrete thin
plate spline technique (Wahba, 1990) using the TOPOGRID module in ArcInfo, to provide a
consistent set of full-coverage covariates for the Project. As some covariates were not available for
every grid cell, a “reliability indicator” was calculated that represented the distance to the nearest
source data.
The GBR wide coverage of all of the collated covariates was thematically mapped using a colour
range appropriate to the individual covariate distribution. These were presented as a landscape map of
the study area divided up into two areas: north and south (see Section 2.1.1.3).
The full coverage interpolated physical data for each grid cell were exported out of ArcInfo for
statistical analysis. This physical data set was also geographically matched to the location of each
sampling station in the Effects of Trawling dataset and the GBR Seabed Biodiversity. These were also
exported from the ArcInfo GIS into a database suitable to provide physical covariates matching
biological sample data for statistical analyses of bio-physical relationships.
2.1.1.3. Maps of Physical Covariate Data
The Great Barrier Reef has a complex physical seabed environment. The major physical
environmental factors that appear to influence the distribution and abundance of seabed habitats and
assemblages in the GBR include: sediment grain size (particularly the amount of mud); force of water
currents (benthic stress); chlorophyll and/or turbidity, hence benthic irradiance; and to a lesser extent
depth and some nutrients (Pitcher et al. 2002). The complex multi-dimensional physical environment
of the GBR has been analysed and used to stratify the region for sampling (Section 2.1.2). The most
GBR Seabed Biodiversity
2-8
common environments (>50% of the seabed) are mostly carbonate sands (Figure 2-2) away from the
coast.
Benthic stress (Figure 2-1) is one of the strongest bio-physical forces and corresponds to the red areas
in the map. The red inshore areas of Broad Sound and Shoalwater Bay have the largest tidal range in
the GBR and are accompanied by extreme currents that scour the seabed re-suspend sediments and
leave behind only gravel and larger particle sizes (Figure 2-2). The red offshore areas show where
these strong tidal currents surge through the narrow channels of the reef matrix, scouring the seabed to
gravel and rock substratum. High tidal current areas also occur in Torres Strait in the far northern GBR
and in some other areas such as Whitsunday Passage. The scoured sediments typically are deposited in
ripples, waves and dunes on the fringes of these red areas (Figure 2-2). At the opposite extreme are
muddy seabeds (Figure 2-2).
Inshore areas along much of the length of the GBR are muddy or silty (Figure 2-2), and comprised of
terrestrial sediments (low carbonate). Typically, with distance across the shelf, the substratum
becomes sandier or even coarser (Figure 2-2), and comprised of biogenic carbonate (of biological
origin). In offshore areas, coralline outcrops, reefs and shoals may occur in deep areas between
emergent coral reefs. The Capricorn Channel, a wide area of GBR lagoon, has a very silty and muddy
seabed. The south-eastern entrance to this channel is the deepest area on the GBR shelf, at 100-130 m.
The Capricorn Region, the southernmost part of the GBR, is typically sandy right across the shelf. It
lies at the northern end of the Great Sandy Region, just beyond Fraser Island, the source of large
quantities of terrestrial sand.
Many inshore areas are also very turbid and/or have high levels of chlorophyll (Figure 2-5). These
inshore areas also tend to be very muddy (Figure 2-2). Along the outer margins of the inshore
turbid/muddy areas, the water is clear enough to allow sufficient light for photosynthesis to reach the
seabed (Figure 2-5). The deeper areas in clear waters near the outer edge of the continental shelf may
at times be influenced by nutrients upwelled from below the ocean thermocline (Figure 2-4).
Many of the physical environmental covariates are correlated (Table 2-1). In particular, SeaWiFS light
attenuation coefficient (K490) and Chlorophyll-a are very strongly correlated, and most of the CARS
bottom-water parameters are also highly correlated. The sediment fractions (% mud, sand, gravel) are
complementary, so are negatively correlated. Bathymetry is highly correlated with a large number of
other covariates. These correlations among covariates make interpretation of analysis results difficult
— when significant relationships between biota and any covariates were identified, not only does
correlation not imply causality, but further other correlated covariates may be important.
GBR Seabed Biodiversity
2-9
Figure 2-1: DEM of the bathymetry, slope and aspect of the GBR continental shelf, on a 0.01º grid, from
various sources including soundings in uncharted areas recorded by the Project; map of modeled seabed current
shear stress (RMS N/m²) (sources, see Section 2.1.1.1 for).
GBR Seabed Biodiversity
2-10
Figure 2-2: Maps of sediment attributes for the GBR continental shelf: percent mud/sand/gravel grain size
fractions and percent carbonate (source, Geoscience Australia. Includes samples collected by the project and
processed by GA).
GBR Seabed Biodiversity
2-11
Figure 2-3: Maps of CARS bottom water physical attributes for the GBR continental shelf: temperature (mean
& SD ºC), salinity (mean & SD ‰), dissolved oxygen (mean & SD ml/l) (source, see Section 2.1.1.1).
GBR Seabed Biodiversity
2-12
Figure 2-4: Maps of CARS bottom water nutrient attributes: silicate (mean & SD μM), nitrate (mean & SD μM),
and phosphate (mean & SD μM), (source, see section 2.1.1.1).
GBR Seabed Biodiversity
2-13
Figure 2-5. Maps of SeaWiFS predicted chlorophyll-A (mean & SD mg/m³), light absorption (attenuation
coefficient K) at 490 nm (mean & SD m⎯¹), benthic irradiance (relative to sea surface at equator estimated from
latitude, K490 and Depth), and weighted average annual trawl effort (hrs/0.01º grid) for the GBR continental
shelf (sources, see section 2.1.1.1).
GBR Seabed Biodiversity
2-14
Table 2-1. Correlation matrix of physical environmental covariates. Non-significant correlations are greyed; larger positive and negative correlations >0.05 are highlighted.
Variable
Bathy
Aspect
Slope
B Stress
Crbnt
Gravel
Sand
Mud
NO3 Av
NO3 Sd
O2 Av
O2 Sd
PO4 Av
PO4 Sd
Si Av
Si Sd
S Av
S Sd
T Av
T Sd
Chla Av
Chla Sd
K490 Av
K490 Sd
Ben Irr
Trwl Eff
Topo
Bathy
1.000
-0.125
-0.201
0.075
-0.479
0.059
-0.033
-0.009
-0.271
-0.617
0.590
-0.247
-0.300
-0.540
0.139
0.260
-0.541
0.547
0.387
0.234
0.513
0.381
0.412
0.318
0.609
0.214
0.138
Aspect Slope B Stress Crbnt Gravel Sand Mud
1.000
0.084
0.123
0.239
0.086
0.048
-0.106
0.054
0.068
-0.074
0.022
0.051
0.044
0.068
0.062
0.008
-0.023
-0.012
-0.067
-0.111
-0.090
-0.112
-0.107
-0.037
-0.095
0.096
1.000
0.123
0.309
0.156
0.024
-0.133
0.212
0.135
-0.224
-0.114
0.213
0.086
0.071
-0.008
-0.004
-0.193
-0.128
-0.231
-0.133
-0.057
-0.115
-0.097
-0.121
-0.129
0.150
1.000
0.158
0.522
-0.031
-0.335
-0.146
-0.125
0.205
0.017
-0.142
-0.154
-0.074
-0.090
0.110
0.021
-0.012
0.303
0.192
-0.017
0.098
-0.042
-0.159
-0.130
0.007
1.000
0.330
0.003
-0.234
0.217
0.177
-0.419
-0.205
0.230
0.125
0.139
0.134
0.032
-0.227
-0.046
-0.428
-0.362
-0.236
-0.332
-0.231
-0.318
-0.206
0.175
1.000
-0.302
-0.411
0.103
-0.075
0.015
-0.131
0.110
-0.128
0.128
0.029
-0.036
-0.023
-0.099
0.076
0.105
-0.008
0.034
-0.051
-0.093
-0.116
0.074
1.000
-0.745
-0.155
-0.102
0.215
0.113
-0.127
-0.106
-0.109
-0.095
0.273
-0.183
-0.078
0.062
-0.268
-0.232
-0.218
-0.212
0.102
-0.127
0.035
1.000
0.077
0.150
-0.216
-0.016
0.044
0.191
0.015
0.071
-0.236
0.191
0.144
-0.113
0.183
0.227
0.185
0.239
-0.032
0.203
-0.085
NO3 Av NO3 sd O2 Av O2 sd PO4 Av PO4 sd Si Av Si sd S Av
1.000
0.667
-0.682
0.050
0.984
0.506
0.754
0.312
0.012
-0.315
-0.755
-0.342
-0.198
-0.087
-0.170
-0.100
-0.104
-0.112
0.056
1.000
-0.725
0.559
0.658
0.925
0.271
0.076
0.424
-0.353
-0.661
-0.074
-0.289
-0.209
-0.237
-0.158
-0.290
-0.148
-0.041
1.000
-0.023
-0.698
-0.702
-0.317
-0.163
-0.080
0.303
0.388
0.554
0.339
0.204
0.251
0.170
0.272
0.102
-0.054
1.000
0.022
0.585
-0.095
-0.039
0.530
0.008
-0.363
0.467
-0.040
-0.073
-0.051
-0.009
-0.110
-0.009
-0.123
1.000
0.512
0.727
0.276
0.062
-0.360
-0.776
-0.351
-0.233
-0.139
-0.201
-0.143
-0.122
-0.130
0.049
1.000
0.182
0.197
0.339
-0.177
-0.457
-0.077
-0.246
-0.186
-0.204
-0.127
-0.223
-0.098
-0.012
1.000
0.681
-0.335
0.131
-0.368
-0.291
-0.036
0.024
-0.044
-0.001
0.108
-0.047
0.135
1.000
-0.594
0.556
0.167
-0.305
0.017
0.033
0.025
0.015
0.229
0.042
0.224
1.000
-0.570
-0.556
0.385
-0.314
-0.288
-0.304
-0.221
-0.316
-0.202
-0.184
S sd T Av
T sd
1.000
0.618
0.176
0.396
0.322
0.340
0.316
0.283
0.200
0.042
1.000
0.340 1.000
0.069 0.597 1.000
0.264 0.920 0.397 1.000
0.075 0.547 0.891 0.377 1.000
0.003 -0.049 0.025 -0.088 0.019 1.000
0.005 0.104 0.071 0.112 0.094 0.123 1.000
-0.166 -0.060 -0.034 -0.050 -0.037 0.195 -0.061 1.000
1.000
-0.107
0.224
0.224
0.209
0.205
0.211
0.186
0.088
Chla Av Chla sd K490 Av K490 sd Ben Irr Trwl Eff Topo
GBR Seabed Biodiversity
2-15
2.1.2. Study Area Stratification (N Ellis)
The sampling for seabed biodiversity mapping in the Great Barrier Reef required an optimal strategy
for the selection of survey sites. The primary purpose of the survey is to obtain data on the spatial
distribution of benthic biota, so that subsequent bio-physical modelling can make use of the physical
environment covariates to interpolate and map. Given that the number of sites that can be sampled is
limited, it was obviously important to place the samples in a way that would yield as much
information as possible. This required that the environment space, or multi-dimensional covariate
space, rather than the 2-dimensional space must be sampled representatively and the approach to
achieve this is stratification. Further, the stratification must be relevant to the benthic biotic, so it must
be informed by measures of the biological importance of each covariate. This approach will optimally
ensure that the biodiversity and physical attributes of as many different habitats types as possible,
given the available resources, would be characterised. The physical variables collated as part of this
project, which were known in advance of the survey, were used to guide the stratification. Biological
information will be taken into account by weighting the physical variables based on their relative
importance in bio-physical relationships — variables of greater influence on biological patterns having
a larger weighting and influence in the stratification.
From an earlier study (Pitcher et al., 2002) we have measures of the “importance” of these covariates
with respect to correlations with the abundance of many benthic species in a detailed survey of an
8,000 km² area of the far northern GBR. Conceptually, important variables are those for which benthic
composition changes significantly along a gradient of the variable. The survey should be designed to
ensure that such important variables are sampled finely, so that the expected benthic diversity is
reliably captured. That is, we should stratify our design with respect to the important variables.
Further, the sampling strategy should also consider the spatial resolution required for management
utility. A scale of several 10s km was considered appropriate for broad scale characterisation. The
implications of the spatial auto-correlation distance (the similarity of locations decreases with
distance, such that on average sites ~10 km apart are uncorrelated; Pitcher et al., 2002) and
considerations of the benefit-cost of logistics (at maximum 1 site per hour) also indicate a sampling
density of approximately 10 km average separation. In this approach, approximately 200 primary
strata with similar physical characteristics were identified from importance weighted physical
covariates of more than 170,000 0.01° grid cells covering the shelf area of the GBR. The size (area) of
strata will vary depending on the number of grid cells having particular similar physical
characteristics. About 10 ‘Replicate’ sampling sites were assigned to each primary stratum by an
importance weighted process.
The potential survey area in the GBR, after excluding reefs and other areas that were too shallow,
included 171,560 cells of size 0.01° (~1.11 km), each cell being a candidate sample site. Given the
spatial autocorrelation distances, the average distance between sites should not exceed about 0.1°
(~11.1 km) indicating that about 1,400–1,600 of these cells should be sampled. A margin was added to
this lower limit, thus the design provided for 1450 sites. At the centre of each 0.01 degree cell, the
values of 28 physical variables were collated or interpolated and represent the GBR region as a cloud
of 171,560 points within a 28-dimensional physical-variable space. Ultimately, this space was to be
partitioned into 1450 relatively homogeneous regions (or strata), such that the expected benthic
biodiversity would be homogeneous within each stratum but heterogeneous among strata. A sampling
site would then be selected from each stratum to produce a set of 1450 sites. This section describes the
methods for achieving this partitioning or stratification of physical-variable space.
2.1.2.1. Principles of Partitioning
The basic principle behind the partitioning can be illustrated with the following simple twodimensional example. Consider two physical variables x and y for which we have values at 1,000 sites,
and suppose that these sites sample the covariate space roughly uniformly (Figure 2-6(a)). We wish to
partition the covariate space into 20 strata. If the two variables were equally important, then the
partitioning in Figure 2-6(b) would be adequate, since the strata are roughly the same width in x and y.
GBR Seabed Biodiversity
2-16
This partitioning was achieved using the “partitioning around medoids” (PAM) algorithm (Kaufman
and Rousseeuw, 1990) (see below).
However, suppose the x variable is known to be 4 times more important than the y variable. Then we
would prefer a partitioning more like that in Figure 2-6(c), where the strata are roughly 4 times
narrower in the x direction than in the y. This is very simply achieved by first scaling the x variable 4fold and then applying PAM to the scaled covariates, as in Figure 2-6(d).
(a)
More important variable
Less important variable
(c)
Less important variable
Less important variable
(b)
More important variable
More important variable
Less important variable
(d)
Scaled important variable
Figure 2-6. Partitioning covariate space in two dimensions: (a) 1,000 points randomly sampled from the square
covariate space. (b) a partitioning into 20 clusters using PAM; (c) a preferred partitioning that accounts for the
relative importance of the variables; (d) the partitioning in (c) is achieved using PAM on the scaled covariate
space.
The partitioning of the GBR grid cells was an analogous procedure in 28 dimensions. Each variable
was scaled so that its ‘range’ was proportional to its importance. However, unlike in the example, the
physical variables were not uniformly distributed across their range and may have extreme outlying
values. To guard against the distorting influence of such values, the ‘range’ was taken as that of the
middle 95 percentiles. The term “I95R” is used here for this range, in acknowledgment of the interquartile range, IQR, of which this is a generalization. Formally,
I95R(v) = v(97.5%) – v(2.5%),
where v(i %) is the i-th percentile of variable v.
2.1.2.2. Variable Importance
The collated physical variables were quantified on various disparate measurement scales that were
unlikely to have any direct relevance to their biological importance. To scale the variables
appropriately to inform the stratification, it was necessary to derive an importance value for each
variable. The primary component was the biotic importance, but it was also necessary to include a
study area adjustment and a reliability adjustment. The biotic importance quantifies the link between
the biota and the physical variables and was developed from the detailed species data in the Effects of
Trawling dataset. The study area adjustment was a refinement to the biotic importance to account for
potential differences in the range of the physical variables between the Effects of Trawling study area
in the far northern GBR and the entire GBR shelf to be sampled by the Seabed Biodiversity Project.
GBR Seabed Biodiversity
2-17
The reliability adjustment was a further refinement to reduce the influence of variables that are
spatially poorly resolved. These are described in detail below.
2.1.2.2.1 Biotic importance Ibio
In a previous study, Pitcher et al. (2002) performed univariate analyses of 30 benthic statistical
assemblages (comprising ~800 species) and 90 single species analyses on 306 sites of the Effects of
Trawling dataset using the same suite of physical covariates as explanatory variables. They derived
tree models for abundance, logistic regression models for presence/absence data and lognormal
regression models for abundance conditional on presence. Their method used model selection to arrive
at parsimonious models with some explanatory power and lead to the derivation of a measure of
importance for each variable. For each species the relative amount of variation explained by each
variable was computed, i.e. the contribution of the variable to the overall R2. The average of this
quantity over all species was defined to be the importance for that variable.
Clearly, the actual dependence of biota on the physical variables is multivariate and highly complex.
Moreover, the explanatory power of the physical variables was often low, averaging about 30%.
Nevertheless, this definition of importance captured the broad pattern over a fairly diverse range of
biota. Also it allowed for variation in explanatory power, since species that had low R2 contributed less
to the importance.
The three types of models considered by Pitcher et al. (2002) were in broad agreement over the
ranking of the variables. However, as the tree model approach was most readily cross-validated, these
results are reproduced here; the importances are shown in Figure 2-7(a).
(b)
(a)
mud.pct
m.bstress
crs.o2.sd
topo.code
crs.po4.sd
sw.chla.sd
caco3.pct
crs.po4.av
sw.chla.av
depth
crs.t.sd
sw.k490.sd
sw.k490.av
osi.grnsz
crs.si.av
crs.t.av
crs.s.sd
crs.si.sd
gravel.pct
sand.pct
effort
crs.o2.av
crs.no3.sd
gbr.slope
sw.k.b.irr
crs.no3.av
crs.s.av
gbr.aspect
osi.rock
mud.pct
crs.o2.sd
topo.code
m.bstress
crs.po4.sd
caco3.pct
gravel.pct
sw.chla.sd
crs.po4.av
crs.si.av
crs.t.sd
crs.s.sd
crs.o2.av
crs.s.av
sw.k490.sd
osi.grnsz
depth
sw.k490.av
sw.chla.av
sand.pct
gbr.aspect
crs.no3.av
crs.si.sd
crs.no3.sd
gbr.slope
crs.t.av
sw.k.b.irr
osi.rock
effort
0
2
4
6
Importance (rpart)
8
0
1
2
3
4
5
6
Importance (random forest)
Figure 2-7. Variable importance computed by (a) cross-validated trees and (b) random forests.
An alternative but similar approach called random forests (Breiman, 2001) was also considered. In
this procedure a bootstrap sample (with replacement) of all 306 sites was taken and a full tree model is
fit without pruning. The method for selecting the splitting variable at each node differs from standard
trees, where all variables are considered for splitting. In contrast, for random forests, a reduced set of
GBR Seabed Biodiversity
2-18
m candidate variables, chosen at random, are considered for splitting and the candidate with the best
split is selected as usual. This bootstrap procedure is repeated 500 times to produce a ‘forest’ of tree
models. Predictions can be made from the forest by taking the average prediction from the individual
trees. For each sample, roughly 37% of sites are not selected. These ‘out-of-bag’ sites provide a test
data set for estimating (without bias) the prediction error of the forest as a whole. As m increases two
effects occur: the prediction error of individual trees improves, and the correlation among trees
increases. The first acts to reduce overall prediction error but the second acts to increase it. There is
therefore an optimal value for m, which Breiman has shown to be close to the square root of the total
number of variables. Given that 28 covariates were available for the GBR, m = 5 was chosen.
The out-of-bag sites also provide a means of defining importance. The importance of variable v is the
percent rise in the out-of-bag mean sum-of-squared errors when the values of v are randomly
permuted. This is a relative measure that can be averaged over species. The results are shown in Figure
2-7(b).
The results for random forests were qualitatively similar to those for the tree models with slight
adjustments to the rankings. The decay in importance with ranking was somewhat smoother for the
random forests. Also, because of the use of random candidate variables, the random forests procedure
tended to overcome the potential of some variables to dominate other closely correlated variables in
the fitting; each variable gets a ‘fair go’. Thus, the random forest importances were considered more
robust and were used in the stratification approach.
2.1.2.2.2 Study area adjustment Q
The raw importance values from the Effects of Trawling study area needed to be adjusted to take into
account that the full GBR study area is different. Some variables, such as bottom stress, have a larger
range elsewhere in the GBR than in the far northern GBR survey area. Such variables may therefore
have more importance in the GBR as a whole. Thus, importances were rescaled in proportion to the
ratio of I95R between the smaller and larger regions; the scale factor Q (see Table 2-2).
The derived importances were also checked by comparisons with analyses of biotic data from the
QDPI Deepwater Seagrass Survey (Pitcher et al. 2002). It was not possible to perform an importance
analysis for the Seagrass dataset in the same way as for the northern GBR study, because it largely
consisted of generalized habitat characterisation or biotic Class level presence/absence data. However,
a guide to relative covariate importance was available from F-values from stepwise discriminant
analysis on clusters defined from the Seagrass dataset. The selected variables were in broad agreement
with the adjusted importances here.
2.1.2.2.3 Reliability adjustment R
The third consideration was that the physical variables had widely differing reliability that needed to
be taken into account when using the calculated importance. All the physical variables were available
on the design grid of 0.01° cells. However, most variables were interpolated onto this grid based on
sample data at a coarser resolution. Therefore, an error distance derr was defined to quantify this spatial
imprecision (see Table 2-2).
The CARS data were interpolated from a rather limited number of CTD casts; the worst case was for
phosphate, the average density of casts with this attribute was approximately 1 in 1,400 km²,
corresponding to an average distance derr of 0.36 degrees between casts. For the effort data, which
came from logbooks reporting effort at 6-minute resolution, derr was set to be the average distance
from the design grid cell to the centre of the 6-minute effort cell. For the OSI data, derr was set to be
the average distance to a sample point from each design grid cell. The SeaWiFS data in their raw form
were already specified at the same scale as the design grid; in this case derr was set to be the average
distance to the grid cell centre within a grid cell.
The ratio of largest to smallest derr was about 270 (refer Table 2-2). It was considered that rescaling
over such a large range would be too severe an adjustment and would effectively eliminate the CARS
GBR Seabed Biodiversity
2-19
variables from influencing the stratification. Thus, the square root of derr was taken and its reciprocal
was defined as the reliability scaling factor R.
Table 2-2. Calculation of adjusted importance Iadj: derr is error distance in degrees, Ibio is the random forests
biotic importance, reliability is R = (derr)–½, and Iadj = (IbioQR)0.74.
Distance
Variable
m.bstress
osi.mud
sw.chla.av
sw.chla.sd
sw.k490.av
sw.k490.sd
crs.po4.av
crs.po4.sd
crs.o2.sd
topo.code
gbr.bathy
osi.crbnt
crs.no3.sd
osi.gravel
crs.t.sd
osi.grnsz
crs.no3.av
crs.si.sd
osi.sand
crs.si.av
gbr.slope
crs.t.av
effort
crs.s.av
crs.s.sd
crs.o2.av
sw.k.b.irr
derr(°)
0.008
0.017
0.004
0.004
0.004
0.004
0.359
0.359
0.224
0.010
0.098
0.102
0.198
0.039
0.179
0.040
0.198
0.198
0.039
0.198
0.098
0.179
0.039
0.179
0.179
0.224
0.051
Biotic
imp. Ibio
2.3
6.0
0.6
1.0
0.5
0.6
0.9
1.1
1.9
1.4
0.6
0.9
0.2
0.4
0.6
0.5
0.1
0.4
0.3
0.4
0.1
0.4
0.2
0.1
0.4
0.2
0.1
I95R
ratio Q
Reliability
2.3
1.1
3.4
1.8
3.0
2.1
8.1
4.1
1.7
0.4
2.6
1.6
7.4
1.5
2.1
1.1
10.8
2.2
1.2
1.7
3.3
1.3
0.9
4.2
0.9
1.7
0.9
11.5
7.7
16.0
16.0
16.0
16.0
1.7
1.7
2.1
10.0
3.2
3.1
2.2
5.0
2.4
5.0
2.2
2.2
5.1
2.2
3.2
2.4
5.1
2.4
2.4
2.1
4.4
R
Adjusted
imp. Iadj
20.8
18.5
13.4
12.0
10.9
8.8
6.4
4.4
4.2
3.9
3.3
3.2
2.5
2.2
2.2
2.1
2.0
1.6
1.5
1.5
1.3
1.2
1.1
0.9
0.9
0.9
0.6
2.1.2.2.4 Adjusted biotic importance Iadj
To incorporate reliability, initially the product IbioQR was considered and compared with the study
area-adjusted importance IbioQ. First, the two adjusted importances were normalized to sum to 1 and
sorted in descending importance, as in Figure 2-8. The reliability-adjusted importance has much
stronger contrast between low-ranked and high-ranked variables, a distortion which was considered
unacceptable. Therefore, the reliability-adjusted importance was ‘tuned’ by raising to a power γ. The
value of γ was chosen to make the tuned importance match the study area-adjusted importance as
closely as possible: γ = 0.74 gave the minimum sum-of-square differences (compare the blue and
green lines in Figure 3):
Iadj = (IbioQR)0.74.
Finally, for each physical variable v, the scaled version vscaled that was used in the stratification was
defined thus:
vscaled = [v ÷ I95R(v)] × Iadj(v)
This scaling ensures the I95R’s of the scaled variables are proportional to the adjusted importances.
GBR Seabed Biodiversity
2-20
0.25
W ithout reliability
W ith reliability
Tuned reliability
Normalised importance index
0.20
0.15
0.10
0.05
0.00
0
5
10
15
20
25
30
Rank
Figure 2-8. Importance measures without reliability (IbioQ), with reliability (IbioQR), and tuned reliability
(IbioQR)0.74 to match the shape without reliability. Each version is normalized to sum to 1. The orders of the
variables with and without reliability are different.
2.1.2.3. The Clustering Process
Having achieved a biologically informed scaling of the physical variables, the next step was
partitioning. However, before proceeding, it was necessary to reduce the dataset for computational
manageability and to provide an orthogonal coordinate space for clustering.
There was a certain degree of redundancy among the physical variables (see correlation matrix Table
2-1). For instance, some variables (phosphate, silicate, chlorophyll A, K490) had a high correlation
between their average value and standard deviation. There was strong correlation among all SeaWiFS
chlorophyll A and K490 measurements, and there were also some negative correlations, e.g. between
temperature and silicate standard deviations. Hence, there was an opportunity to apply data reduction
techniques to make the data set more manageable and, importantly, orthogonal prior to clustering.
2.1.2.3.1 Data reduction
Singular value decomposition (SVD) was used to separate the data into principal components, from
which we retained the most important components accounting for 99% of the variance in the data.
This was contained in the first 14 components, and in fact the first 7 components contained 95% of the
variance. SVD decomposed the 171,560 × 28 data matrix X of scaled physical variables into a product
of matrices UDVT, where U was the 171,560 × 28 score matrix, D was the 28 × 28 diagonal matrix of
singular values, and V was the 28 × 28 orthogonal loadings matrix. To project the data into a smaller
dimensional space, but retain the relative distances of the data, a new data set was defined as UD*
where D* (28 × 18) consists of the first 18 columns of D. This data is equivalent to rotating the scaled
data by V (i.e. XV) and projecting into the 18-dimensional subspace spanned by the first 18 columns.
The effect of this transformation was observed by examining the variable loadings V. The rows of V
correspond to the original variables and the columns to the principal components. Large values (on the
scale 0 to 1) indicate alignment of the variable with the principal component. The important variables
GBR Seabed Biodiversity
2-21
should be expected to have high loadings on the first few principal components, and the less important
variables to have higher loadings on the later principal components.
The loadings on the first seven principal components are shown in Table 2-3. Principal component 1
was mainly associated with mud and various SeaWiFS measurements, whereas the second and third
components were associated with bottom stress and mud. Because the 3 most important SeaWiFS
variables are highly correlated with one another, they have similar loadings. The 4th component
introduced contrasts between the SeaWiFS means and standard deviations, the 5th introduced
phosphate and oxygen, and the 6th contrasted the standard deviations of chlorophyll A and K490.
Other variables were also loaded to a lesser extent.
Table 2-3. Variable loadings for the first 7 principal components. Absolute loadings greater than 0.5 are
highlighted in yellow, and absolute loadings between 0.3 and 0.5 are highlighted in green. The variables are
ordered by adjusted importance. Relative variance is the fraction of the total variance explained by the principal
component.
Loadings
Variable
m.bstress
osi.mud
sw.chla.av
sw.chla.sd
sw.k490.av
sw.k490.sd
crs.po4.av
crs.po4.sd
crs.o2.sd
topo.code
gbr.bathy
osi.crbnt
crs.no3.sd
osi.gravel
crs.t.sd
osi.grnsz
crs.no3.av
crs.si.sd
osi.sand
crs.si.av
gbr.slope
crs.t.av
effort
crs.s.av
crs.s.sd
crs.o2.av
sw.k.b.irr
gbr.aspect
Relative variance (%)
1
–0.07
–0.55
–0.42
–0.45
–0.36
–0.42
0.07
0.04
0.02
0.00
–0.07
0.03
0.03
0.02
–0.01
–0.04
0.02
–0.01
0.02
0.00
0.00
–0.01
–0.01
0.01
–0.02
0.00
0.00
0.00
35.1
2
–0.80
0.51
–0.22
–0.10
–0.18
–0.06
0.03
0.02
–0.01
–0.01
–0.01
–0.01
0.01
–0.02
–0.02
0.03
0.01
0.00
–0.01
0.00
–0.01
0.00
0.01
0.00
0.00
–0.01
0.00
0.00
31.6
Principal Component
3
4
5
–0.55
0.23
–0.03
–0.65
0.07
–0.03
0.14
–0.46
–0.09
0.37
0.42
0.00
0.10
–0.60
–0.09
0.32
0.42
–0.01
0.02
0.07
–0.73
0.00
0.04
–0.43
–0.31
0.00
–0.01
0.01
0.01
0.02
0.02
–0.04
0.18
0.00
0.05
–0.01
0.00
0.02
–0.26
0.02
0.01
0.02
–0.01
–0.04
–0.01
–0.03
–0.01
–0.03
0.01
0.03
–0.23
0.00
0.01
–0.03
0.02
–0.01
0.01
0.00
0.01
–0.09
0.00
0.02
–0.01
–0.01
0.00
0.05
0.00
–0.01
0.02
0.00
0.00
–0.03
0.00
–0.01
0.03
0.00
–0.01
0.05
0.00
0.00
0.02
0.00
0.00
0.00
16.7
6.0
3.3
6
0.01
–0.03
–0.05
–0.54
0.03
0.62
–0.22
0.11
0.38
–0.08
–0.01
–0.23
0.07
–0.08
0.14
0.05
–0.08
–0.03
0.04
–0.06
–0.07
0.02
0.02
0.06
0.01
0.03
0.00
–0.01
2.1
7
–0.01
0.01
0.04
0.39
–0.11
–0.40
–0.33
0.15
0.59
–0.15
–0.02
–0.26
0.10
–0.13
0.20
0.07
–0.09
–0.06
0.08
–0.11
–0.05
0.00
0.01
0.08
0.00
0.05
0.00
0.00
1.6
2.1.2.3.2 Including geographic constraints
Another important consideration was whether spatial position should be included in the stratification.
In the absence of covariate information, it would be usual to stratify entirely on geographical position,
making each stratum simply connected. On the other hand, if we ignore geography completely, and
base the stratification only on physical covariates, then the strata will tend to be fragmented in
geographical space. This would not necessarily be a bad thing. However, if the fragments become very
small then the quality of the stratification may become degraded by spatial uncertainty in the
covariates themselves.
Instead of using latitude and longitude as geographical coordinates, along and across were used,
which are covariates tailored to the shape of the GBRMP region (see section 2.4.1), one varying from
GBR Seabed Biodiversity
2-22
0 to 1 along the coastline and the other going from 0 on the coast to 1 at the outer edge of the reef. To
assess the relative scaling to apply to these, we ran a simple linear fit of all the covariates to along and
across, and found the average absolute value of the coefficients; they were 1.17 and 1.21 respectively.
We therefore used the two covariates in equal scaling.
After studying the degree of fragmentation of clustering under various scalings of along and across,
we decided that the I95R of along should equal 0.25 times the I95R of the first principal component of
the rotated data. The scaled spatial variables were included as extra dimensions in the clustering, and
their effect was generally to prevent the clusters becoming too highly fragmented in space.
2.1.2.3.3 The PAM and CLARA algorithms
The clustering algorithm “partitioning around medoids” (PAM) of Kaufman and Rousseeuw (1990),
which is implemented in Splus, was used to cluster the physical dataset. The PAM algorithm is a
robust alternative to the k-means algorithm. It uses a distance matrix and the number of clusters must
be specified. Whereas K-means minimizes distances to the average for the cluster, in PAM, each
cluster contains a medoid that is the cluster member whose summed distance to all other cluster
members is a minimum. The medoid is a kind of generalized median for multiple dimensions; it is to
this that the algorithm owes its robustness. The algorithm works by searching for clusters that
minimize the total distance to cluster medoids.
PAM is not immediately useable for large data sets, because the size of the distance matrix becomes
unmanageable. Therefore Kaufman and Rousseeuw’s CLARA algorithm, which is an implementation
of PAM for large data sets, was applied. This works by first selecting a random subset of the data, then
applying PAM to generate a clustering, and finally assigning the remainder of the data to the nearest
cluster in the subset. The procedure is repeated many times to give several candidate clustering’s, from
which the candidate that minimizes the total distance to cluster medoids is chosen. The algorithm can
be tuned by adjusting the subset size and the number of repeats, both of which should be as large as
practicable.
Further, a weighted version of CLARA was developed specifically for this project. In this
implementation, each initial subset was selected with non-uniform probabilities or weights, which
enabled the clustering to be influenced to some extent to seek rarer physical environment strata, as
explained below.
2.1.2.3.4 Two-stage partitioning
The partitioning was performed in two stages. In stage 1, we generated an initial coarse partitioning of
the entire data set into 200 ‘primary clusters’, or primary strata. Then in stage 2, each primary cluster
in turn was partitioned, generating a total of 1450 subclusters.
The initial reason for having two stages was computational efficiency. For k clusters and n
observations, the computation time is of order kn2; but if √k primary clusters was computed first, and
then √k subclusters (on average), the computation time can be reduced to the order √k n2. In fact stage
1 is the most computationally intensive stage, taking of order √k times longer than stage 2. Even for
200 primary clusters, which was rather larger than √1450, the computational saving was substantial.
This was an important consideration when developing a method, particularly where many subsets of
the data must be run.
However, the main reason for using a two-stage method was that it allowed more control over the
partitioning. This was because at stage 2, it becomes possible to choose the number of subclusters
within each primary cluster, subject to a total of 1450. In particular, it was possible to raise the level
sampling effort into uncommon and rarer areas in covariate space, which may be potentially more
interesting in terms of biota, at some expense to common areas.
GBR Seabed Biodiversity
2-23
2.1.2.3.5 Choosing the number of subclusters
After stage 1, there were 200 primary clusters of various sizes ranging from 2 to 7680 cells. Then it
was important to determine how to optimally distribute the 1450 subclusters among the 200 primary
clusters.
In order to answer this question, initially the following hypothesis was adopted: clusters with large
numbers of cell members tend to be more homogeneous and represent commonness, compared with
small clusters. Support for this hypothesis can be seen in Figure 2-9 for a synthetic bivariate normal
data set. The larger clusters (in terms of numbers of cells) near the middle have smaller bivariate space
(i.e. are more homogeneous), whereas the more heterogenous clusters around the fringes tend to have
fewer points (i.e. are smaller clusters).
(a)
(b)
24
48
49
48
24
48
59
63
62
33
64
41
70
57
75
69
47
41
38
40
Figure 2-9. (a) Bivariate normal distribution of 1,000 points. (b) Partitioning into 20 clusters using PAM. Each
cluster is labeled by the number of points in the cluster. The more populous clusters tend to be tighter and so
more homogeneous.
Therefore the stratification strategy should be such that the density of sampling should be lower for
larger primary clusters, i.e. the number of subclusters Nsub depends sub-linearly on the primary cluster
size S. This issue also arises in the context of species-area curves, where the number of species
increases with area sampled, but less than linearly. In fact, for species-area curves a square-root
relationship is sometimes used. Following this principle, the initial approach could be Nsub ∝ √S.
This approach would attempt to bias the sampling away from common sites towards rarer, perhaps
more ‘interesting’, sites so that they also can be sampled adequately. Nevertheless, the square-root
approach provides a somewhat crude approximation to the amount of ‘interest’ in a primary cluster,
relating it simply to the size of the primary cluster, without regard to its contents. A better approach
would be to quantify the interest as a sum over the interest in individual sites. For this, it was
necessary to define the interest at a site.
The more common sites are those lying in high-density areas of covariate space. Since common sites
will be well sampled in any case, it was reasonable to define ‘interest’ as some inverse power of
density. However, computing the density in more than 2 dimensions is difficult; instead the onedimensional density of each physical variable was considered separately. Suppose dvi is the density of
variable v at site i, normalized so that the total density over all sites is 1. Then we define the interest wi
at site i as the variable importance-weighted sum,
wi = ∑ I adj (v)d vi− a ,
28
v =1
where a > 0 is a parameter to be chosen. Then define the interest of a primary cluster as the total
interest over sites within the primary cluster, and choose the number of subclusters to be proportional
to this quantity. That is, for the kth primary cluster C(k):
∑
GBR Seabed Biodiversity
N sub (k ) ∝
i∈C ( k )
2-24
wi .
2
0
1
Density
3
4
The density is estimated from the 171,560 values using a Gaussian kernel whose width is calculated
by biased cross-validation (Scott, 1992). As an example, Figure 2-10 shows the true density (total area
= 1) for bottom stress. The bulk of the distribution lies below 0.5; whereas previous experience has
demonstrated that sites above 0.7 were of particular interest for epibenthic fauna (Pitcher et al. 2002).
0.0
0.2
0.4
0.6
0.8
1.0
Bottom stress
Figure 2-10. Density of bottom stress estimated by a Gaussian kernel of width 0.01 calculated using biased
cross-validation. Also shown is a ‘rug’ of values for 200 randomly selected sites.
0
500
1000
1500
Supercluster size
2000
30
20
0
0
29
10
27
8
Number of subclusters
20
3
10
Number of subclusters
20
10
0
Number of subclusters
a=1
30
a = 0.5
30
a = 0.25
0
500
1000
1500
Supercluster size
2000
0
500
1000
1500
2000
Supercluster size
Figure 2-11. Number of subclusters vs primary cluster size for 3 different values of the exponent a. The sloping
line corresponds to Nsub ∝ S, the curve to Nsub ∝ √S, and the horizontal line to Nsub = constant.
Figure 2-11 shows the relationship between number of subclusters and primary cluster size for a =
(0.25, 0.5, 1). For the case a = 0.25, the relationship was almost linear; this was barely distinguishable
from the case a = 0, in which all sites had equal interest. At the other extreme, case a = 1 flattened the
relationship, making number of subclusters nearly independent of primary cluster size and too
sensitive to individual high-interest sites within a primary cluster. The intermediate case a = 0.5 was
close to the square-root proposal discussed earlier and provided the required increased sampling of
rarer sites without unacceptable under-sampling of common sites. This value for a was used as it
provided an improved stratification adjustment compared with the initial square-root proposal.
GBR Seabed Biodiversity
2-25
p
Clusters 1 to 10
Clusters 11 to 20
Clusters 21 to 30
Clusters 31 to 40
Clusters 41 to 50
Clusters 51 to 60
Figure 2-12. The 200 primary clusters in geographical space. Sixty of the clusters have been separated into six
panels in order to make them distinct and assess the degree of fragmentation.
There was a concern that, at the primary clustering stage, rarer sites might be missed in the CLARA
random subset selection stage since rare sites would be unlikely to be selected in a small random
subset and, as a consequence, the primary clusters could be too large and homogeneous. Such primary
clusters, being comprised largely of common sites, would have fewer subclusters, and so there would
be less chance of isolating the rarer sites into their own subclusters. Two steps were taken to reduce
this risk. Firstly, we computed more primary clusters than was computationally optimal (i.e. 200 >
√1450). Thus, primary clusters would be smaller, allowing for better detection of heterogeneity within
GBR Seabed Biodiversity
2-26
a primary cluster. Secondly, a weighted version of CLARA was developed; with site interest wi as the
weighting. Thus, rarer sites were more likely to have a chance at being chosen in the random sample
of the algorithm, and therefore more likely to seed a separate primary cluster.
0.4
0.2
0.0
Density
0.6
Figure 2-12 shows maps of the resulting 200 primary clusters after the first stage of clustering.
Because the clustering was in covariate space, there was no guarantee that the clusters would be
simply connected in geographical space, even though latitude and longitude were included as
covariates. Indeed some clusters were quite fragmented. Despite their geographical separation, these
clusters’ sites have similar physical characteristics. In the other hand, some clusters re fairly spatially
contiguous. Part of the reason for this is that the covariate values in these regions are based on spatial
interpolation from sparse data points, and so the covariates vary smoothly in space. The primary
clusters were further partitioned into subclusters as described above.
0.005
0.05
0.1
0.2
0.5
1
2
5
0.10
0.0
0.05
Density
0.15
0.20
Bottom stress
0.01
0.1
1
10
50
90
99
0.2
0.0
0.1
Density
0.3
0.4
Percentage mud
1
0.7
0.5
0.4
0.3
0.2
Average chlA
Figure 2-13. Distribution of the most important physical covariates on the full the GBR data (orange). The thin
curves are 90% confidence intervals for the density sampled from the clusters. For clarity we show covariates on
a log scale for bottom stress, a logit scale for mud and an inverse scale for chlA. Also shown is a rug of 200
sample values (jittered for mud).
2.1.2.3.6 Assessing the resulting stratification
There is no unequivocally optimal approach to survey design. For instance, in the two-dimensional
example of Figure 2-6, we could have used the k-means algorithm instead of PAM, and the resulting
partitioning, which would have been different, would nevertheless have been a quite reasonable
alternative. Although there is no single ‘right answer’, it is nevertheless necessary to establish that the
resulting partitioning is reasonable. There are several ways to assess the stratification.
GBR Seabed Biodiversity
2-27
First, the strata were mapped. We have already partially shown this in Figure 2-12. However, a map of
all 1450 substrata would be rather overwhelming and very difficult to interpret. Alternatively the
locations of the substratum medoids could be plotted, since each medoid was in some sense the most
typical representative of the stratum. In fact, the choice of medoids as actual survey sites would be a
quite reasonable candidate sampling strategy and could be called “medoid sampling”. This would
provide acceptable general coverage of the entire the GBR region. However, the sampling would be
finer in some areas where environmental gradients were steeper and coarser in broader more
homogeneous areas. This was consistent with expectations and a desirable property of the
stratification, which was being sought.
The second way to assess the stratification was to examine the expected distribution of the physical
covariates at the sample sites. Again, the medoid sampling can be used as a representative sampling.
Figure 2-13 shows the density of bottom stress, percentage mud and average chlA over the stratum
medoids compared to over all 171,560 sites. Transformed scales have been used, on which the
distributions were roughly symmetrical, to make the comparison clearer. For completely random
sampling, the density would be similar to that over the full data set. But in the medoid sampling, there
was relatively less sampling in the high density (common) areas, and more sampling in the tails (rarer
areas), which was the objective of the stratification. For example, in the case of bottom stress, more
sampling is put into sites with values above 0.5, at the expense of the more common sites with values
in the range 0.1–0.3.
The representativeness of the medoid sampling can be checked by comparing its density with densities
arising from many random samplings of the stratification. Figure 2-13 also shows confidence intervals
for the density, which were obtained from the 5th and 95th percentiles of the pointwise densities of 20
random samples. Although there were small biases, overall the medoid-sampling density was fairly
representative of the range of possible densities arising from stratified sampling.
2.1.2.3.7 Defining trawl substrata
The above has described how 1450 substrata were defined from which benthic sampling sites may be
chosen. However, about one third of these same sites (595) were to be selected for trawl sampling and
it was necessary to identify which would be the most representative. Although one method would be
to simply choose the 595 sites at random, an approach that took advantage of the existing stratification
was preferred, to ensure that the selection was as representative as possible. The approach taken was to
go back to the primary clusters and recompute the number of subclusters required per primary cluster
to give a total of 595, using the same methodology as before. On average the number of trawl
subclusters was about 0.41 (595/1450) the number of original subclusters. For instance, primary
cluster 92, which had 20 original strata, had 8 trawl strata. It was not feasible to try to cluster the sites
into trawl subclusters, because there was no way to prevent the original strata straddling several trawl
strata. Instead, it was necessary to cluster the sites such that all sites in an original stratum remain
together.
The simplest way to do this was to cluster the stratum medoids. It was appropriate to use the medoid to
represent its stratum as a whole because the medoid lies centrally within the stratum in co-variate
space. Since there were at most 44 medoids to cluster, the calculation was computationally simple. For
example, in primary cluster 92, substrata 1, 3, 4, 8, 11, 14 and 17 were amalgamated into one trawl
cluster, substrata 2, 9 and 16 into a second, and substrata 6, 7, 12, 13 and 18 into a third, while the
other 5 trawl clusters coincided with the original substrata.
After the medoids were clustered, each medoid’s trawl substratum number was assigned to all other
cells in its substratum. Thus, each cell now belongs to both a substratum and a trawl substratum. Thus
for any selection of 1450 benthic survey sites, the trawl sites could be selected from these by choosing
one from each trawl stratum, either at random or by other objective.
2.1.2.4. Mapping the Stratification
The biologically informed stratification developed in this section is a physical characterisation of the
GBR seabed that can be considered an a priori surrogate for patterns in seabed biodiversity
GBR Seabed Biodiversity
2-28
assemblages, to be tested and improved by the sampling to be conducted by the Project. A method of
representing this complex multi-variate data in a single map was sought.
2.1.2.4.1 The Colour Key
The objective was to produce a map of the GBR with similar colours representing similar physical
environments, which might be expected to have similar benthic biotic assemblages. The colour
mapping should encompass as much information as possible in a reduced form — this was achieved
by deriving a colour key from the first and second principal components of the biological importance
weighted covariate data used in the stratification. A biplot of the principal components and physical
variable vectors would provide a key to the environmental characteristics of the map. Particular
directions in the biplot that corresponded to important covariates should be coloured in an intuitive
manner. Red was used to denote high bottom stress and green to denote high average chlorophyll A
(which correlated with K490). Blue corresponded with depth. High density areas of the biplot
(common areas) should have a neutral colour such as grey.
A further desirable property of the colour key was that it should cover the data space compactly, to
avoid large areas of the key having no data and wasting part of the colour space. The colour key
should therefore be shaped to conform to the distribution of the data in principal components (PC-)
space. This was done by mapping a circular colour disk to a simply connected region enclosing the
data. In order to do this, it was necessary to first define a boundary of the data in PC-space. One way
to do this was to find the convex hull; however, for the GBR data, this included voids in which no data
existed. Instead, a more compact boundary was found by computing a two-dimensional kernel density
function and delineating a contour of sufficiently low density. The boundary is partly concave.
Having defined a boundary, there were two alternative methods for mapping the colour disk to the
region inside the outer density contour boundary: polynomial mapping and conformal mapping. The
polynomial mapping was found to be more flexible but because of the partly concave shape there was
not always a one-to-one mapping between PC-space and colour space, and it was non-trivial to invert
from PC-space to colour.
2.1.2.4.2 Conformal Mapping
The conformal mapping method originates from complex number theory. A mapping from the colour
disk to a simply connected polygon is expressible as a complex integral, whose parameters must be
estimated by a non-linear algorithm. Trefethen (1980) provided a FORTRAN program to compute this
integral. An interface to this code was developed that runs in R. Conformal mappings have certain
benefits (such as local preservation of angles) but most importantly they are guaranteed to map the
interior of the colour disk to the interior of the polygon (i.e. the mapping will not stray outside the
boundary).
As with the polynomial method, the point in PC-space that the centre of the disk was mapped to was
specified. The matching of points on the edge of the disk with vertices of the polygon was done by the
non-linear algorithm. In order to match intuitive colours to the desired directions in PC space, it was
necessary to impose a further transformation on the colour disk, which amounted to an angular stretch
and shift. This was done using a periodic piecewise linear function of the angle. To complete the
physical characterisation map, each grid cell must have a colour associated with it. Hence, the colour
key mapping must be inverted, so that points in PC-space become mapped to points in colour space.
This inverse mapping is available in FORTRAN code (Trefethen, 1980).
The resulting physical characterisation map of the GBR is shown in Figure 2-14. High bottom stress
areas were coloured red, high Chlorophyll/K490 areas green, and the mud direction was coloured
cyan. Sites coloured cyan have high levels of mud. Deeper areas tend to be blue.
The colouring of a map to highlight different covariates can be highly effective at illustrating similar
and different physical environments, especially when the colour space has been fully utilized. The two
colour mapping techniques investigated each had advantages and disadvantages. The main
GBR Seabed Biodiversity
2-29
disadvantage of the polynomial method was discussed above. On the other hand, the conformal
mapping method tended to cover jutting-out parts of the PC-space from fairly small regions in colour
space (e.g. the red area of the key in Figure 2-14). This would be a significant disadvantage if such an
area were densely populated with data.
Figure 2-14: Map of the biophysical stratification of the Great Barrier Reef continental shelf. Inset: colour key
showing distribution of seabed grids on the first two principal components (which explain 65% of the variation)
of the biologically weighted physical covariates; biplot vectors indicate direction and magnitude of the major
physical factors.
2.1.3. Site Selection
In the previous section, the idea of medoid sampling was used to illustrate the stratification. Medoid
sampling could be an acceptable method of selecting sites that would deliver the “most typical” cell,
with respect to physical covariates, within each of the sub-strata. A random selection of sites from
within each of the sub-strata would also be an acceptable method. However, random selection has a
relatively high risk of selecting some cells too close together and too far apart, creating clumps and
voids in the coverage of the survey area, when a representative coverage that also takes account of the
spatial autocorrelation distance was desired. Noting also that strata were often fragmented into patches
of varying numbers of cells, including single cells, there was also a high risk of selecting isolated cells
as sites — these would be less likely to be representative of their stratum due to errors in the
covariates. A site selection method that avoided these issues as much as possible was sought.
Initially, a weighted random selection was used, with weights dependent on the spatial geometry of the
patch(es) of cells within each stratum that cells belonged to. Cells with fewer neighbours of the same
stratum and on the edges of patches (i.e. geographically close to a different stratum) were given less
weight, whereas sites in the middle of patches were given more weight. This strategy was intended to
GBR Seabed Biodiversity
2-30
reduce the possibility of a site being unrepresentative of its stratum due to errors in the covariates and
to avoid selecting adjacent sites. Examination of several weighted random selection options indicated
quite a number of adjacent cells being selected and a number of excessively large voids between
selected sites. Consequently, a method that more stringently avoided selection of adjacent cells and
voids was preferred.
The method finally used did not include any deliberate random jittering of site selection. For each of
the ~1450 benthic sub-strata, first all those cells that had the maximum number of neighbours and
were the maximum distance from the edge of patches were selected. For many of the sub-strata,
several cells met these criteria and to exclude duplicates, the cell with the minimum medoid distance
was selected. In about one-sixth of cases, the actual medoid cell was selected. This strategy maximized
the co-variate representativeness and spatial regularity of the selection, within the desired constraint of
the stratification, and minimized the likelihood of clumps and voids, and adjacent, edge and isolated
cells.
As described in the previous section, fewer sites could be sampled by trawl methods, so the ~1450
benthic medoids were clustered to provide 600 most representative options. Of these, 236 were a one
to one match with their benthic strata, so no further selection was needed. However, in 364 cases, a
trawl site had to be selected from typically 2-7 benthic site options (up to 19 in an extreme case). In
these cases, the benthic site chosen to be sampled by trawl also was, to maintain spatial coverage, that
which belonged to the largest patch in its cluster.
The sites selected are mapped in Figure 2-15. Note that sites are more sparsely distributed in broader
more homogeneous areas, allowing a higher density of sites where environmental gradients are
steeper. This site selection process provided a good compromise between coverage of the range of
biologically important physical environments in the GBR and evenness of spatial coverage, given the
limited number of sites that could be sampled and the inadequacies of the data available for the
stratification. Such coverage could not be achieved with regular grid sampling or completely
randomised sampling.
ÑÑ ÑÑ Ñ Ñ
Ñr
Ñ
Ñ rr
Ñ
r
Ñr Ñ
ÑÑ
Ñ
r
Ñ
Ñ
Ñ
ÑÑ Ñ
Ñ
rÑ
Ñ r
Ñ
Ñ
Ñ
rr
rrr rrÑÑÑÑ Ñ r Ñ ÑÑÑÑÑ
r rÑ Ñ Ñ
r
r
r
ÑÑ
r rr ÑÑÑ r r Ñr Ñ
r
Ñr r Ñ r
Ñ Ñ ÑÑ
r Ñ
Ñ Ñ
r
Ñ
Ñ
Ñ Ñ
rr
r
Ñ
Ñ
rÑ r
Ñ r
rÑ Ñ
rÑ
Ñ ÑrÑ
Ñ Ñ
r
ÑÑr
ÑÑ Ñ
rr ÑÑ ÑÑÑ Ñr
Ñ ÑÑ Ñ
r
Ñr
Ñ r
r
r ÑÑ r
Ñ
r
rr
Ñ
Ñ
ÑÑ Ñ
r
r
ÑÑ
ÑÑ
Ñ
r r
Ñ
Ñ
Ñ
Ñ
Ñ
r
Ñ
Ñ Ñ
rÑÑ
Ñ r
Ñ
rÑ
r Ñ rÑ
Ñ
Ñ
r
rr
ÑÑ Ñ
rr
Ñ
Ñ Ñ Ñ
r
Ñ
r Ñ
r rr Ñ Ñ
r
Ñ
Ñ
Ñ
r Ñ
rr rr Ñ r
rÑr
Ñ ÑÑ
Ñ
rÑ Ñ
Ñr
Ñr
rr
Ñ
ÑÑr
Ñ
ÑÑ
rÑÑ
N
W
r
r
r
r
Ñ
rÑ
Ñ Ñr Ñ
r
r
Ñ
Ñ
r Ñ Ñ ÑÑ Ñ r Ñ
Ñ rÑ
Ñ Ñ ÑÑr Ñ Ñ Ñr
r
r ÑÑÑ Ñ Ñ Ñ
Ñ
rr
Ñ
r
Ñ
r r Ñ Ñ ÑÑ r Ñ r
Ñ r Ñ
Ñ
r
rÑ
r
r
Ñ
Ñ
Ñ rr
rÑr Ñ r rÑÑ Ñ Ñ
Ñ
Ñ r
Ñ Ñ
ÑrÑ Ñr
ÑÑ
Ñ
Ñ Ñ
Ñ
r rÑ
Ñ Ñ
Ñ
ÑÑÑ ÑÑ r ÑÑ
r Ñ Ñr
Ñ
r Ñ r Ñr r Ñ
r Ñ
ÑrÑ Ñ r
Ñ
Ñ
Ñ
r
r
rÑ Ñ
r Ñ
Ñ
Ñ ÑÑ ÑÑÑ
r Ñ
rÑ Ñ Ñ Ñ
r
ÑÑ
Ñ
r r
rÑ
r
Ñ
r ÑÑ
Ñ
Ñ
rÑ Ñ
rr
r
r
Ñ
r Ñ r
ÑÑ Ñ r
Ñ Ñ Ñ
r
Ñ Ñ
r
r r ÑÑr
rÑ r
Ñ
rr
Ñ Ñ Ñ r
Ñ Ñ
r
Ñ r
Ñ
Ñ
Ñ
Ñ
r rÑ Ñ Ñ
r r ÑÑ
rÑ
r Ñ r Ñ
r
Ñ
Ñ
Ñ
rÑ Ñ
r
r
r Ñ rÑ
r
Ñ
r
Ñ
Ñ
Ñ Ñ
Ñ
ÑÑ
r Ñ r
Ñ Ñ
Ñ
Ñ
Ñ Ñ
Ñr Ñ
Ñ
r
Ñ
Ñ rr
Ñ
r
Ñ r Ñ r
Ñ
rÑ
rÑ
r
Ñ
r r Ñ r ÑÑ
Ñ
Ñ
Ñ
r
Ñ
Ñ
Ñ
Ñ
Ñ rr
rÑ
Ñ
rÑ
Ñ Ñ
r rÑ Ñ ÑÑr Ñ
r r
Ñ
r
Ñ
Ñ
r
r
Ñ Ñ
ÑÑ r
rr Ñ
r
Ñr
Ñ r Ñ
Ñr Ñr
r
rrÑÑ
r
r
Ñr
ÑÑÑ rr
ÑÑ
ÑrrÑ
r
rÑ ÑÑ
r
r
r
Ñ
Ñ
r
Ñ
r
ÑrÑ
Ñ
Ñ
Ñ
Ñ
ÑÑ
Ñ
r
Ñ r Ñ
Ñ
Ñ
r Ñ Ñ
Ñ
Ñ
Ñ
r
Ñ
Ñ
r
Ñ Ñ Ñ
Ñ r
Ñ
r
r
r Ñ r
Ñ
Ñ Ñr
Ñ
Ñ r
r Ñ
Ñ
rÑ Ñ rr
Ñ
r
Ñ
Ñr
Ñ
Ñr r
Ñ
Ñ
rÑ
Ñ Ñ
r
r ÑÑ Ñ
Ñr
Ñ
rÑ Ñ r
r
Ñ Ñr
Ñ
r
r
r
Ñ
r
Ñ
r r
Ñ
Ñ
Ñ
r
r Ñ
Ñ Ñ
Ñ Ñ Ñr Ñ
Ñ Ñ rÑ Ñ Ñ
r
r r
Ñ
Ñ
Ñ
r
Ñ Ñ
Ñ
Ñ
r
r
rr
r
r Ñ
Ñ
Ñ
Ñ
Ñ
Ñ
Ñ rr r Ñ r
Ñ
Ñ
Ñ Ñ Ñ r
rÑ Ñ
Ñ
Ñ
r
Ñ
Ñ
ÑÑ
r Ñ
ÑÑ r
r
r
Ñ r r ÑÑÑ r
Ñ
Ñ
Ñ r Ñ
rÑ rr
r Ñ ÑÑ Ñ Ñ
Ñ Ñ
Ñ
Ñ r
r Ñ Ñ
Ñ r Ñ
Ñ Ñ
Ñ
rÑ
Ñ Ñ
r Ñ
r rr
Ñr
rrr r Ñ Ñ r Ñ Ñ r
rr Ñ
Ñ
Ñ Ñr
r r
r
Ñ Ñ
Ñ Ñ
r
Ñ
Ñ Ñ r
ÑÑ Ñ
Ñ
r
Ñ
Ñ
r
r
r
ÑÑ
Ñ
Ñ
Ñ
Ñ
Ñ
r
r
Ñ
r
Ñ
ÑrÑ Ñ r Ñ Ñ
r r
r
Ñ
ÑÑ
r
Ñ
Ñ
Ñ Ñr Ñ Ñ Ñ
Ñ Ñ
Ñ Ñ ÑÑ
ÑÑ
rÑ Ñ r
Ñr
r Ñ rr
r
Ñ
r ÑÑ Ñ Ñ
Ñ
Ñ rÑ
Ñ
Ñ
Ñ
ÑÑ
r Ñ ÑrÑ r
r
Ñ
Ñ
Ñ
Ñ
Ñ
r
rÑ
rr ÑÑÑ r
Ñ
r rÑ Ñ Ñr Ñ r Ñ r
r r Ñ rÑr
Ñ
r Ñ Ñ Ñ Ñ
Ñ
Ñ
rr ÑÑÑÑ
Ñ
Ñr
Ñ Ñ r ÑÑ Ñ Ñ
Ñ rr Ñ Ñr
r
Ñ ÑÑ
r ÑÑ Ñ
r
Ñ Ñ
Ñ
ÑÑ ÑÑ Ñ ÑÑ
Ñ Ñ
Ñrr
rÑÑ Ñ
Ñ
r
Ñ
r
Ñ
r
Ñ
Ñ
Ñ
r
Ñ
Ñ
r rÑ r
r
ÑrÑ r
ÑÑ
r
ÑÑ
r Ñ Ñr r Ñ
Ñ Ñ Ñ Ñr
Ñ Ñ
Ñ Ñ
ÑÑ ÑrÑ r Ñ r Ñ
r
Ñ Ñ Ñ Ñ
Ñ
Ñ
rÑ
r
Ñr Ñ
rÑ
rÑrÑ
rr
r
r r
rr
Ñ rr r
Ñ
Ñ Ñ
Ñ r Ñr Ñ Ñ Ñ
Ñ Ñr Ñ
Ñ
Ñ Ñ
r Ñr rÑ
r
r
Ñ
Ñ
Ñ
Ñ
r
rr
ÑÑ
Ñ
Ñ
Ñ Ñ
Ñ
r
Ñ
r
r
r
r rr
Ñ
Ñ r Ñ Ñ
r
Ñ Ñ
rÑÑ
Ñ
Ñ rÑ
rÑ
r
Ñ
Ñ
rÑ r
r r
Ñ
r rÑ
r r
Ñr
r r rr
Ñ
r Ñ ÑÑ
Ñ
Ñ Ñ
ÑÑ
r
r
r
Ñ Ñ
rÑ
r Ñ rÑr rÑ
Ñr
Ñ
r
Ñ
r
rÑ r
ÑÑ
ÑÑ
rr
Ñr r
Ñ
rr rÑ
r Ñ Ñ
Ñr Ñ
r
Ñ Ñ rr
rr Ñ Ñ r Ñ r r
rÑ
rr Ñ
r
r
Ñ rr Ñr rr
Ñ Ñ ÑÑ Ñ
rÑ
ÑÑ
Ñ
r
Ñ
r
Ñ
Ñ
r rr
r
Ñ
rÑ
Ñ ÑrÑ
r ÑÑÑ Ñrr
r
Ñ rr Ñ
Ñ r
r Ñ
r
r
Ñ
rÑ rr
Ñ Ñ
Ñ
rr
Ñr
Ñ
Ñ Ñ
r
Ñ
r
Ñ
r Ñ Ñr Ñ
Ñr
ÑÑÑ Ñ
Ñr
Ñrr
r
Ñ
Ñ
Ñ
r
rrr
r
r
r r
Ñ
Ñ
r
Ñ
r
r
Ñ
r
r
Ñ Ñ
r
Ñ
Ñ
Ñ r
r
Ñ
r
r
Ñ
rÑ
r
rÑ Ñ r
Ñ
rr
Ñ
r Ñ
ÑÑ
Ñ Ñ Ñ Ñr Ñ
Ñ
Ñ r
Ñ
ÑÑ
Ñ
Ñ Ñ
Ñ
Ñ Ñ
Ñ
ÑÑ Ñ
r Ñ
Ñ
Ñ ÑÑ r
r
rrrrÑ
rÑ r ÑÑ
r
r
Ñ r Ñ
Ñ r ÑÑ Ñ
r
Ñ
Ñ
r
r
r Ñ Ñ r Ñr
rÑ
rr
r Ñ ÑÑ
rÑ rr
ÑÑ
r Ñ
Ñ rÑ ÑÑ
Ñr
Ñ
r r
r
rÑ
Ñrrr rÑ
r
Ñ
Ñ
Ñ
Ñ Ñ
r rr
rr Ñ
Ñ r
r Ñr r Ñ
r r
r Ñ Ñ Ñ
Ñ
ÑÑ r
Ñ Ñr
r
Ñr
Ñ
rrr
E
S
100
r Ñ
Ñ
ÑÑ
ÑÑ Ñ
ÑÑ
Miles
0
100
Figure 2-15: Map of the sites selected for sampling the seabed on the continental shelf in the GBR. Ä: sites for
benthic and trawl sampling; «: sites for benthic sampling only.
GBR Seabed Biodiversity
2-31
2.2. FIELD SAMPLING
2.2.1. Research Vessels (T Wassenberg & N Gribble)
Two research vessels were used during the seabed mapping project: the Australian Institute of Marine
Science research vessel RV Lady Basten (Figure 2-16) and the Queensland Department of Primary
Industries and Fisheries research trawler FRV Gwendoline May (Figure 2-17).
The RV Lady Basten operates from AIMS in Townsville. It generally sails with a crew of seven,
Master, Mate, Engineers, deck crew and Cook and can accommodate up to seven scientific staff. Its
operational area is up to 2,500 nautical miles from port. The vessel is 27.4 metres in length and has a
wet and dry laboratory and can deploy numerous instruments through the stern A-frame.
Figure 2-16: The 27 m Australian Institute Marine Science research vessel RV Lady Basten
Six voyages were conducted on the RV Lady Basten between September 2003 and November 2005
(Table 2-4). Multiple operations on this vessel continued over 24 hr each day while at sea with the
crew and scientific staff working 12 hr shifts.
Table 2-4. Voyages completed by the Lady Basten with scheduled duration, numbers of sites sampled by towed
camera, epibenthic sled and BRUVS.
Voyage
Start date
End date
Days
# Camera
# Sled
# BRUVS
LB_01
17-09-03
12-10-03
26
263
215
89
LB_02
22-11-03
08-12-03
17
121
137
39
LB_03
25-04-04
30-05-04
36
124
206
62
LB_04
07-09-04
10-10-04
34
266
241
81
LB_05
10-01-05
11-02-05
33
212
181
56
LB_06
26-10-05
30-11-05
36
233
214
74
182
1219
1194
401
Total
The Gwendoline May operates from the QDPI&F Northern Fisheries Centre in Cairns. It sails with a
crew of three, Master, Mate and Cook and can accommodate up to five scientific staff. Its operational
area is up to 200 nautical miles from the coast, and from New Guinea to Southport. The vessel is 18
GBR Seabed Biodiversity
2-32
metres in length and can operate a single trawl over the stern or quad gear from the booms. It has a
Kortz nozzle fitted to increase trawl efficiency and reduce fuel costs. The electronics allows 3D
mapping of the seabed and the radar and radios are state of the art.
Figure 2-17: The 18 m Queensland Department of Primary Industries & Fisheries research trawler FRV
Gwendoline May.
Four primary voyages were conducted on the FRV Gwendoline May between November 2003 and
December 2005, with several additional days sampled during a subsequent QDPI monitoring survey
(Table 2-5), all under the command of a former commercial fisher with substantial experience in the
region. Scientific trawl sampling operations on this vessel were conducted at night time.
Table 2-5: Voyages completed by the Gwendoline May with scheduled duration, numbers of scientific trawl
sites sampled, sites with hookups and those too rough to trawl.
Voyage
Start date
End date
Days
Sampled
Hook-ups
Too rough
GM_01
17-11-03
16-12-03
30
133
6
18
GM_02
12-04-04
02-05-04
21
103
3
5
GM_03
20-09-04
20-10-04
30
107
5
27
GM_04
09-11-05
15-12-05
37
112
1
40
GM_05
03-03-06
04-03-06
3
6
121
461
15
90
TOTALS
2.2.2. Towed Video Camera (G Smith, K Forcey, M Haywood)
An underwater video camera system (Drop-Cam, Figure 2-18) was towed just above the seabed at
each site wherever possible, for a distance of ~500 m, to characterise habitats and visible biota. The
Drop-Cam system consisted of cameras, frame, fibre-optic towing cable, cable winch, hydraulic crane,
CTD instrument, Control and data logging computers, video recorders and display monitors.
Video and still cameras, and a CTD instrument, were mounted within a galvanised steel frame. The
video cameras were twin 3-chip Panasonic E300 digital video cameras fitted with 2.8 mm Fujinon
lens. The video field of view was illuminated with 2 × 500 and 2 × 250 W lights. A Canon 20D 8.2
mega pixel digital still camera fitted with a 4 GB memory card, Canon EF 17 – 40 mm auto focus lens
GBR Seabed Biodiversity
2-33
and two Speedlite 550EX strobes recorded still photographs of the seabed every 5 s during the camera
transects. Twin lasers, spaced 28.5 cm apart, were fitted in from of the field of view of the cameras —
when visible on the seabed, these enabled scaling of seabed objects. All cameras, lights and strobes
were housed in custom built housings rated to 3,000 m.
The CTD was a Seabird SBE 19plus Seacat Profiler fitted with sensors for conductivity, temperature,
pressure, oxygen, chlorophyll, turbidity and light (PAR). Data from the sensors was recorded at 0.25 s
intervals and logged onto a computer database on board. Data from a PAR sensor fitted to the vessel
was also recorded and logged to enable comparison between surface and underwater light levels.
Also mounted in the frame were two pressure housings; one for the power supply and the other for a
computing system for data and video collection. Apart from the digital still images, all data and video
were sent from the Drop-Cam to the vessel via an optic fibre link, in real time during the transect; the
digital still images where downloaded at the end of each transect.
Data and video were converted to fibre optic media through a Focal 903 multiplexer and sent to the
demultiplexer surface unit in the vessel via a single optical fibre. At the demultiplexer the signals were
separated into data and video. Control of the Drop-Cam system (cameras, lights, lasers, CTD) and all
processing and logging of data was done using in-house custom software on Pentium PCs, with the
video recorded onto Panasonic DVC-Pro digital tapes.
The general procedure upon arrival at each site was as follows. The video camera was deployed and
lowered to within approximately 0.5 m of the seabed. The CTD was set to switch on upon contact with
seawater and recorded data throughout the deployment, transect and recovery. The vessel was then
driven at approximately 1.5 knots, towing the camera frame for a distance of 500 m. Position and
distance towed was recorded by differential GPS every 0.1 s. Video was displayed in real time on a
monitor in the vessel laboratory, enabling scientific staff to raise and lower the camera, with remote
joystick control of the winch, in order to maintain altitude above the seabed during the transect (Figure
2-18).
Figure 2-18: The Drop-Cam system being recovered after completion of a 500 m video transect and the surface
real-time monitoring, control and data acquisition system.
2.2.2.1. Tasks, Events and Navigation Data Logging
Custom software modules acquired navigation, video and Drop-Cam data on several computers.
Independent modules acquired and logged date/time and position data from differential GPS, depth
data from the vessel echo sounder and heading data from the vessel gyrocompass on a navigation
computer and shared these data onto a local area network. A custom map module also ran on the
navigation computer to plot site waypoint and vessel position and constantly checked current position
GBR Seabed Biodiversity
2-34
against the designated site waypoint list and also shared these data onto the network to ensure all
modules indexed logged data against the correct site number and date/time. A second computer ran
modules for Drop-Cam system control and data (depth, pitch, roll, altitude) acquisition/logging, CTD
data acquisition/logging and surface PAR data acquisition/logging. A third computer ran modules for
control and data acquisition/logging of the DVC-Pro video tape recorders (VTRs). A fourth computer
ran the Tasks control and logging module and events logging module. The Tasks module was
manually operated, with touch screens for recording start and end operations for Drop-Cam transects,
epibenthic sled tows and BRUVS drops. Each task record included task type logged with site number,
date/time, position, and depth data as shared onto the network by other modules. The Tasks module
also measured transect and tow lengths, and the touch of the Drop-Cam start task initiated recording of
video by the VTRs and data logging by all other modules.
The events module was also manually operated — a scientific staff member used a modified keyboard
to enter a real-time summary of the seabed substratum type, biological habitats and conspicuous
individual animals. A lookup table transcribed keyboard scan codes into the seabed types listed in
Table 2-6, which were logged and prefixed with date/time, position, and depth data for each seabed
event recorded. Note that the position recorded was that of the GPS antenna on the vessel, whilst the
observed event being recorded could be ~25 m (about 30 seconds) behind that because the camera was
being towed from the stern.
Table 2-6: Substratum and Biological habitat types and animal events types entered in real time to annotate the
video transect. Numbers in parentheses show index values used in acoustics sections.
Substratum
Soft Mud
(9)
Silt (Sandy-Mud)
(8)
Sand
(6)
Coarse Sand
(2)
Sand Waves / Dunes (7)
Rubble (5-50 mm)
(5)
Stones (50-250 mm) (10)
Rocks (> 250 mm)
(4)
Bedrock / Reef
(1)
Biological habitat
Bioturbated
Bivalve Shell Beds
Alcyonarians:
Sparse
Medium
Dense
Whip Garden:
Sparse
Medium
Dense
Gorgonian Garden: Sparse
Medium
Dense
Sponge Garden:
Sparse
Medium
Dense
Hard Coral Garden: Sparse
Medium
Dense
Live Reef Corals
Flora:
Seagrass
Algae:
Caulerpa
Halimeda
No BioHabitat
(5)
(6)
(3)
(2)
(1)
(25)
(26)
(24)
(11)
(10)
(9)
(20)
(19)
(18)
(15)
(14)
(13)
(16)
(8)
(21)
(4)
(7)
(12)
(17)
Animal events
Anemone
Ascidian
Bryozoan
Commercial Fish
Crab
Crinoid
Gastropod
Holothurian
Hydroid
Non Commercial Fish
Sea Pen
Solitary Coral
Starfish
Urchin
2.2.3. Baited Remote Underwater Video Stations (M Cappo)
A fleet of Baited Remote Underwater Video Stations (BRUVS) were deployed about 350-400 metres
apart with the prevailing wind to bracket the coordinate of each sampling site. Each replicate was
considered to be sampling independently from the others at this separation (Cappo et al. 2004). At
each site, a stereo-video BRUVS was deployed first, followed by three or four BRUVS with single
cameras. Footage from the stereo-video was included for a small number of sites to make up the
minimum number of three replicates in the BRUVS data. The BRUVS consisted of a galvanized
GBR Seabed Biodiversity
2-35
trestle-shaped frame enclosing a simple camera housing made from PVC pipe with acrylic front and
rear ports (Figure 2-19). Sony™ Mini-DV HandiCams (models TRV18E, TRV19E) with wide-angle
lens adapters (0.6X) were used in the housings. Exposure was set to “Auto”, focus was set to
“Infinity/manual” and “Standard Play” mode was selected to provide at least 45 minutes of filming at
the seabed. Detachable bait arms (20 mm plastic conduit) had a 350 mm plastic mesh canister
containing one kilogram of crushed oily sardines (Sardinops or Sardinella spp) as bait (Figure 2-19,
Figure 2-20). BRUVS were deployed with 8 mm polypropylene ropes and polystyrene surface floats
bearing a marker flag, and were retrieved with a hydraulic pot-hauler wheel. A scope length of 2 times
water depth was used on the ropes.
Figure 2-19: Diagram of single BRUVS frame and housing.
Figure 2-20: Applying camera and bait arm to BRUVS before deployment. Note ballast on frame and 8 mm
hauling rope.
2.2.4. Single-beam Acoustics
Single-beam acoustic remote sensing of the seabed was conducted as continuously as possible on
board the RV Lady Basten, using a Simrad EY500 120 kHz digital echo-sounder with a hull mounted
Simrad 120-25 transducer (10° beam angle). Each ping was sampled in two blocks: (1) low resolution:
700 samples from the surface to (usually) beyond the 2nd echo (the depth was monitored and the
range setting adjusted (100, 150, or 250 m) to ensure the second echo was captured, thus sampling rate
varied with range ~15-35 cm), and (2) high resolution: 500 samples from 5 m above bottom to 10 m
below bottom (a constant sampling rate of 3 cm). Data were logged in Simrad format files, each 5Mb
(Simrad EY500 Operating Manual). The majority of the data were acquired at the same pulse length
GBR Seabed Biodiversity
2-36
(0.3 ms), but when range setting was 250 m (when depth 75-125 m) a longer pulse was used (1.0 ms),
necessitating separate post-analysis. Depths averaged about 40 m (range about 6-120 m).
A Quester Tangent Corporation QTC View Series IV signal processor was also connected to the
EY500 transducer, and the QTC proprietary 166 feature data were acquired and logged to QTC CAL
files using the QTC CAPS version 3.15 software.
Seabed ‘ground-truth’ data were collected in real-time by entry of substratum and biological habitat
during remote video camera tows (see Section 2.2.2), at about 1,200 sites. The towed camera (and so
the real time coded habitat category data) trailed about 25 m behind the echo-sounder, and this lag
back (estimated separately for each tow) was taken into account before relating ground-truth data to
acoustics pings (see Section 2.4.6).
In total, more than 20,000 km of vessel track digital acoustic data were logged from six voyages all
over the GBR shelf between 2003 and 2005.
2.2.5. Epibenthic Sled (T Wassenberg & M Stowar)
An epibenthic sled was deployed through the A-frame over the stern of the Lady Basten. The sled
(Figure 2-21) was 1.0 m long, 1.5 m wide between the skids and 0.5 m high and weighed about 250
kg. The sides were closed with steel plate and the top and bottom panels were 20 mm square steel
mesh. A heavy duty net (25 mm stretched mesh, 48 ply) was attached to the rear of the frame. Chafing
mats of old nets were attached beneath the codend to minimise the damage to the net codend due to
dragging on the seabed. The sled was attached to the main winch warp by two lengths of chain (1.5 m
long) and a weak link (2 tonne – connected by chain to the rear of the frame) set to release the front of
the frame in the event of a hook-up on the seabed that permitted the frame to flip over and be retrieved
backwards. The sled was deployed using the ship’s deck winch whilst the ship maintained a constant
bearing and speed of ~2 knots. A winch cable to water depth ratio of 3:1 was used and the towed
distance of 200 m measured by onboard differential GPS from the ships position when the full amount
of cable had been paid out.
Figure 2-21: The epibenthic sled being deployed through the A-frame for a 200 m tow along the seabed; note
the weak link at the top of the bridle and retrieve chain leading to the rear of the sled.
GBR Seabed Biodiversity
2-37
In order to collect sediment samples at the same time as the sled was deployed, a pipe dredge (Figure
2-22) was suspended from each rear corner of the sled frame by ~1 m length of chain. The pipe was
0.6 m long and 90 mm internal diameter. Of the sediment collected at each site, 500 ml was placed
into a plastic bag, labelled and retained at 4oC for particle size analysis by Geoscience Australia and a
further 500 ml sample was washed in a 1 mm square mesh sieve. The retained portion after sieving
was placed into a cloth bag and preserved in a 10% solution of Formalin containing Rose Bengal to
stain biological material in the sediment sample.
Figure 2-22: Sediment pipe dredge, showing sister-clip for attachment behind the sled
2.2.6. Scientific Trawl (T Wassenberg, D Gledhill & N Gribble)
A single high-flying Florida Flyer net having a head rope length of 8 fathoms and stretched mesh size
of 50 mm of 400D/27 ply was towed over the stern (Figure 2-23) of the Gwendoline May. Drop chains
were 5 × 8 mm stainless steel links attached to a 10 mm stainless steel link ground chain and twin
No. 3 Bison boards with “high-riser” extensions to the board area (approximately the spreading power
of a Pollards’ No 4 Bison board) plus extra weights attached giving a total weight 153 kg, were used to
spread the net (Figure 2-24).
Figure 2-23: Net plan for the eight fathom Florida Flyer net used for scientific trawl sampling and the net
suspended from the A-frame on the stern of the Gwendoline May.
GBR Seabed Biodiversity
2-38
At the sites that were able to be trawled, the net was towed in a relatively straight line for a distance of
1 km at a speed of about 2.7 knots. At the end of the tow, the catch was retrieved and emptied onto the
sorting tray.
All trawling was conducted from one hour after sunset until dawn, to correspond with the activity of
commercial prawn trawlers and also because many mobile seabed species have diurnal behaviour. The
start and end points for each trawl were logged electronically using computer logging software with a
GPS data connection.
The Gwendoline May visited 551 sites; 461 sites were sampled, and 15 were curtailed due to hook-ups
and 90 were abandoned because the seabed was too rough and therefore unsuitable for sampling by
trawl (Table 2-5).
Figure 2-24: Drop chain links and trawl boards.
2.2.7. Sample Processing at Sea (T Wassenberg, M Stowar, C Bartlett)
2.2.7.1. Epibenthic sled samples
On retrieving the epibenthic sled, the catch was placed into fish baskets. First, a photograph was taken
of the entire catch. Large animals (sponges, gorgonians, large holothurians, large starfish etc.) were
removed for immediate processing. The remaining sample was then processed entirely or subsampled.
In either case, the sample was sorted into rough phylogenetic groups (sponges, crustaceans, algae,
ascidians, seagrasses, fishes, echinoderms, molluscs and remaining invertebrates).
The fish baskets were emptied onto a sieve table that had three sieve drawers. The top drawer was 100
mm square mesh, the second had 50 mm square mesh and the lower drawer had 10 mm square mesh.
The principle being that the smaller items would fall through to the bottom drawer and the largest
items retained at the top (Figure 2-25) but the drawers were interchangeable to suit the content of the
catch to be sorted.
If the catch was very large (> 140 L), it was necessary to take a random subsample of the catch.
Subsampling was undertaken as a two stage process. Firstly, the entire sample was sorted for larger
fauna and flora, as retained by a ‘coarse’ 100 mm × 100 mm sorting table mesh. Secondly, once the
coarse fraction had been entirely sorted, a subsample of approximately 70 L of the fine fraction was
sorted over the 10 mm square mesh. The total volume of the fine component was also recorded to
enable determination of subsampling factors. The plant and animal samples retained from both coarse
and fine fractions were sorted, bagged and recorded separately. Corrections for subsampling factors
were made during the data analysis stage.
GBR Seabed Biodiversity
2-39
Figure 2-25: Sorting the catch from the epibenthic sled on the 10 mm square mesh sieve drawer into major
taxonomic classes.
Table 2-7: Designated preservation methods on board the vessel and destinations for further processing (MTQTVL = Museum of Tropical Queensland; QMSB-BRS = Queensland Museum South Brisbane; CMR-CV =
CSIRO Cleveland, QDPI-TVL = Queensland Department of Primary Industries Townsville).
CLASS
Annelida: Worms
Ascidia: Tunicata:
Biological conglomerates (no Sponge)
Biological conglomerates (+Sponge)
Brachiopoda:
Bryozoa:
Cnidaria:
Cnidaria: Anthozoan: Octocorallia
Cnidaria: Hydrozoa
Cnidaria: Zoantharia: Hexacorallia
Crustacea:
Crustacea: Penaeidae
Echinodermata:
Echinodermata: Crinoidea
Echinodermata: Holothuroidea
Fishes:
Fishes: Syngnathid
Mollusca:
Mollusca: Cephalopoda
Porifera:
Marine plants:
Marine plants: Algae:
Marine plants: Seagrass:
Sediment animals 1 mm sieved
Sediment for grain size
PRESERVATION
Formalin 10%
Formalin 10%
Formalin 10%
Frozen
Frozen
Frozen
Frozen
Ethanol 70%
Frozen
Frozen
Frozen
Frozen
Ethanol 70%
Ethanol 100%
Frozen
Formalin 10%
Formalin 10%
Frozen
Frozen
Frozen
Frozen
Frozen
Frozen
Rose-Bengal/Formalin 10%
Cool room.
DESTINATION
MTQ-TVL
MTQ-TVL
MTQ-TVL
QMSB-BRS
MTQ-TVL
MTQ-TVL
MTQ-TVL
QMSB-BRS
MTQ-TVL
MTQ-TVL
CMR-CV
CMR-CV
MTQ-TVL
MTQ-TVL
MTQ-TVL
QMSB-BRS
QMSB-BRS
MTQ-TVL
MTQ-TVL
QMSB-BRS
QDPI-TVL
QDPI-TVL
QDPI-TVL
CMR-CV
CMR-CV
GBR Seabed Biodiversity
2-40
Sorted samples were transferred to an onboard laboratory where each was allocated a water and
solvent-proof barcoded label (Figure 2-26). Each sample was then individually photographed, weighed
and transferred into a plastic bag. All details including barcode, collection details, taxonomic group,
weight, photographs, storage location, subsampling factor (if any) and any relevant comments were
entered into in a database onboard (Figure 2-27). The individually bagged samples were then
preserved onboard according to prescribed preservation techniques (Table 2-7) until later processing at
several destination laboratories onshore.
Figure 2-26: Samples of sorted dredge catch organisms with bar code labels ready to be photographed and data
recorded in the vessel data base.
Figure 2-27: Data and images from each sled site were entered into the vessel database entry form that also
included a photo of the entire site sample (left) and of the sample (in this case, of echinoderms).
GBR Seabed Biodiversity
2-41
2.2.7.2. Scientific trawl samples
The catch from the scientific trawl was spilled into a sorting tray and any large rocks were discarded.
The entire catch was photographed to give a site photograph (Figure 2-28). Large animals
(elasmobranches, turtles, sea snakes, sponges, large holothurians, large fishes etc.) were removed for
immediate processing. The remaining catch was examined for uniformity and species that were
readily-recognisable and in tractable numbers were removed and processed immediately. If the catch
was very small it was sorted in full.
Figure 2-28: A photograph of an entire trawl catch (site photo) showing the diversity of organisms.
If the catch was very large, a random subsample of the catch was taken by making a slice through the
catch. The intent was to retain at least 20% of the catch. A photograph was taken of the subsample and
the weight of the remainder that was returned to the sea was also taken. The subsample was then
sorted into fish, invertebrates and prawns etc, photographed with a bar code label and put into the
respectively labelled boxes in the freezer.
Subsamples were also taken if a large quantity of one species or any obviously abundant/common
species was caught. For example: a very large sponge would be weighed but only a smaller portion
would be retained for detailed analysis; whereas in the cases of very large catches of Leiognathids
(Ponyfish) only a few specimens were retained after weighing the total.
After the large, low incidence or rare organisms were removed from the catch and given a unique
barcode label, photographed, weighed and either retained or returned to the sea, the remainder of the
catch was sorted into rough phylogenetic groups (commercial prawns, non-commercial prawns, other
crustaceans, algae, seagrasses, syngnathids, remaining fishes, holothurians, squid and remaining
invertebrates — Table 2-8). The sorted material was allocated a barcode label, weighed and
photographed (Figure 2-29). Some reference specimens of small fishes were preserved in formalin as
these can be damaged during freezing. Other material was packed in plastic bags and placed in
cardboard cartons. Seagrasses were stored at ca 3 ºC and other groups were frozen at -20 ºC.
Data and images from each site were entered into the vessel database while on board (Figure 2-30).
The weights entered into the database were the total weight and the retained weight for each sample.
GBR Seabed Biodiversity
2-42
Figure 2-29: A sample of crustaceans showing the barcode label. The label number, weight and class were
entered into a data base at sea.
Table 2-8: The taxonomic groups into which samples were sorted onboard and their specific requirements for
preservation on board the vessel and destinations for further processing were provided.
CLASS
Crustacea:
Crustacea: Penaeidae
Invertebrates:
Fishes:
Fishes: Syngnathid
Marine plants:
Marine plants: Algae:
Marine plants: Seagrass:
PRESERVATION
Frozen
Frozen
Frozen
Frozen
Frozen
Frozen
Frozen
Frozen
DESTINATION
CMR-CV
QDPI-CNS
QMSB-BRS
CMR-CV
CMR-CV
QMSB-BRS
QMSB-BRS
QMSB-BRS
Figure 2-30: Data and images from each trawl site were entered into the vessel database entry form that also
contained a photo of the entire catch (left) and of the sample (in this case, of an elasmobranch).
GBR Seabed Biodiversity
2-43
On two voyages digital colour images were taken of many specimens of selected fish species, which
were preserved in 10% formalin. Penaeid prawn species and seagrass species were forwarded to
QDPI&F in Cairns, while the remaining samples were sent to CSIRO Cleveland for further
processing.
2.2.7.3. Vessel Sample Database (D Chetwynd)
The purpose of this database was to efficiently store and retrieve data and images for sites visited and
samples collected via either the Scientific Trawl or the Epibenthic Sled. It was designed to be portable,
user-friendly and, in the often difficult shift-work conditions at sea, minimise errors by means of
prompts and real-time checks. Microsoft Access was used for storage of data in tables and Visual
Basic for Applications (VBA) for the software development, in particular data entry Forms as well as
Reports and Queries.
All data was stored as unique samples at the site level. Sites were a predetermined position within the
GBR region and constituted the highest level for the data storage. The critical importance of a unique
sample identification system led to the use of barcode labels and barcode readers which proved
successful in minimising errors. A single barcode label was assigned to, and bagged with, each
sample. A sample constituted a coarse sorting of the Trawl/Sled into general taxonomic groups for
each site. Typically there would be one group per site, such as one sample of sponges for any given
site, unless there was subsampling. There could be many samples for each site — there was a 1 to
many relationship between the site and the sample (barcode).
Once sorted into groups, each sample was assigned a physical barcode Label, torn from a pre-printed
perforated roll. A digital image was taken of each sample with its respective barcode. Each sample
was then entered into the database against the current site number. The sample's weight, count and
depth were recorded. The image for the sample was then associated to the sample’s unique barcode by
the database within the database and the image stored within a folder for that particular site. Barcode
and site number were saved within the image filename.
In the event of subsampling, the total and retained weight was recorded for that sample. If a large
animal was brought onboard it was photographed with a barcode then released alive, and the data and
photograph were entered.
Once each sample was logged in the database it was preserved by one of several predetermined
methods, such as freezing, formalin or alcohol.
The database allowed users to backtrack through previous sites or samples and check images against
the recorded data, with updating procedures available to add/update to any stored data.
On completion of each voyage a copy of the vessel data was sent to Cleveland and both the data and
images were extracted for use in the Laboratory sample Database.
2.3. LABORATORY PROCESSING AND IDENTIFICATION
2.3.1. Towed Video (T Wassenberg, J Sheils)
2.3.1.1. Video data processing
The recordings of the seabed towed video transects at each site were transferred from the digital tapes
via firewire to removable computer hard drives for storage and archiving. Each site’s camera tow was
saved as a separate .avi video file, a process which was automated using batch files scripted from VTR
time-code data logged in the field. The site identity of each video clip was cross-checked during the
video capture process, using the audio data stream from the vessel GPS, which had been recorded onto
the audio tracks of the video tape during each camera tow. A decoder box and software translated the
audio signal on each recording into GPS data and displayed the date-time and positional information
and calculated the appropriate site number for cross reference against the data logged in the field.
GBR Seabed Biodiversity
2-44
The objective of the analysis of the towed video was to characterise the seabed habitat by describing
abiotic features and sessile biota, and visually estimating their approximate size and percentage cover.
The operators who performed the video analysis underwent training before the process began and
continued to consult regularly throughout the video processing to maintain consistency. A procedure
manual was developed to facilitate consistent decision making.
Video files were copied from the removable hard drives onto work folders on the video analysis
workstation computers, from where copies were deleted once analysed. A custom Delphi software
application was used to view the videos and enter the data. The software paused the video at random
intervals (in this case, between 10 and 30 seconds), overlaying a trapezoid outline onto the screen to
highlight the target quadrat to be scored. The operator then entered data for that video frame before
proceeding to the next. The user interface of the software is shown in Figure 2-31 and Figure 2-32.
The data entry screen included drop-down lists for the operator to choose from, and consisted of three
sections:
1: Large Scale Feature
These were defined as topological features of the seabed that of a scale larger than the target quadrat,
for example reef, sand dune or flat seabed. Estimates were made of the vertical scale of these features.
2: Biological
Sessile fauna and flora were classified into categories; for example, different growth forms or (where
identifiable) genera of sponges, corals and algae etc. The categories were developed to distinguish
between different organisms as much as the resolution of the video would permit. Signs of animal
activity such as burrows, mounds, pits and tracks were also recorded here. For each category type,
estimates were made of percentage cover, along with estimates of their vertical and horizontal scale.
3: Sediment
The abiotic component of the habitat was classified into broad categories based on the Wentworth
scale of sediment classification, modified to suit the level of discrimination possible from video
footage. The presence and scale of sand ripples and waves were also recorded here. Habitat
components larger than the trapezoid were treated in Large Scale Features.
Figure 2-31: Data entry screens of the Delphi video analysis software showing the trapezoid overlaid on the
paused video image.
GBR Seabed Biodiversity
2-45
Figure 2-32: Drop down lists of physical and biological attributes to be used in analyzing the video image.
Two lasers aimed at the seabed 28.5 cm apart are visible in the recordings (red dots just in front of the
trapezoid), providing a scale for size estimation. Once the operator has made their selections the data
was committed to an Oracle database, where it was linked to the site number and millisecond position
within the video clip. This allows future retrieval of the image and the record of the exact place in the
video archive where the original data entry was made.
2.3.1.2. Laboratory Video Software & Database (D Chetwynd)
The Laboratory processing of the DropCam video used a custom Delphi 7 application which stored
data (user selected Benthic information) and images (paused video .bmp) directly to an Oracle
database.
Delphi is a Windows Rapid Application Development (RAD) environment that uses Object Oriented
Programming, which is effectively a collection of cooperating objects. The video application is form
driven (see Section 2.3.1.1) and benthic analysis and images were captured from the video and stored
directly into the database.
Viewing the .avi files within the application was controlled by a recognised Active X component
‘MoviePro’. This was professionally developed for use within Delphi and other Object Oriented
application environments and uses the basic features of Windows Media Player to play, fast-forward,
rewind and pause the video, as well as allowing the capture of paused video as a .bmp.
Access from the Video software to the database was provided via the third party tool, Oracle Data
Access Components (ODAC). This is a set of VCL native components for Delphi, which supports
many Oracle specific features and simplifies developing of client/server applications. Connectivity to
the Oracle database works directly through TCP/IP and doesn't require Oracle's software on the client
side.
GBR Seabed Biodiversity
2-46
The Oracle database was Oracle Database 10g Enterprise Edition Release 10.2.0.2.0 and is fully
supported and backed up as per CSIRO policy.
Data and images were written to the database via a form driven set of Oracle Stored Procedures, called
by the Delphi application. Each Database Manipulation Language (DML) call was run within a
transaction and any resulting errors forced a ROLLBACK of any insert, updates and deletes within the
transaction so the problem could be rectified without affecting any other data. Images (paused video
.bmp) are stored in the database itself as a Binary Large Object (BLOB) and therefore are actually
stored within the specific row. This ensures any backups encompass the entire dataset, as opposed to
the alternative choice of image storing, which stores the image within a predefined directory and the
directory address and filename, the pointer, within the database.
2.3.2. BRUVS Video (M Cappo)
Interrogation of each tape was conducted using a custom interface (BRUVS2.1.mdb©, Australian
Institute of Marine Science 2006) to manage data from field operations and tape reading, to capture the
timing of events and reference images of the seafloor and fish in the field of view. Records were made,
for each species, of the time of first sighting, time of first feeding at the bait, the maximum number of
fish seen together in any one time on the whole tape (MaxN), time at which MaxN occurred, and the
intraspecific and interspecific behaviour in 8 categories. The use of MaxN as an estimator of relative
abundance has been reviewed in detail by Cappo et al. (2003, 2004).
Species identifications were confirmed by checking the collection of reference images with museum
taxonomists [Dr Barry Hutchins (Paramonacanthus), Barry Russell (Pentapodus)] and with other
project staff [Jeff Johnson and Dan Gledhill]. It was decided some taxa were indistinguishable on
video footage, so these were pooled at the level of taxa, genus, family or order. These taxa are hitherto
referred to as species. The MaxN data were then summed for each species over all single BRUVS
replicates at a site, and 4th root transformed. Data were analysed at the level of individual sites.
2.3.2.1. BRUVS Video Software & Database
A custom BRUVS tape analysis interface was developed by AIMS staff (Gavin Ericson, Greg
Coleman) for this project, and was improved in 8 versions (BRUVS2.1.mdb©, Australian Institute of
Marine Science 2006).
One hour of Mini-DV tape footage equates to 12 gigabytes of digital data. Capturing, digitising and
compressing the tapes would consume 1.5 hours each. Given 1585 tapes were collected, it was
impractical to digitise them.
Instead, we developed a method where the tapes were played in a tapedeck with a jog shuttle control to
a 50cm screen. The tapedeck was connected via “firewire” to the BRUVS2.1mdb, where the video
playback was also visible in small windows.
The tape was played to and fro and the timecodes [converted to decimal minutes] of important events
were captured via firewire from the tapedeck. When a new fish was seen, drop-down menus in
BRUVS2.1.mdb offered selections for family, genus and species. Once species was selected a
CAABCODE was generated with the record. When certain “events” buttons were selected, the
timecode was grabbed from the tapedeck and stored with the record. The tape deck was paused to
allow grabbing of “benthos” and “fish” images, which were named by the software and distributed to
folders. If the species was unknown, various buttons allowed the reference imagery to be searched for
a match. If the species was new, a dialogue box enabled generation of custom CAABCODES and a
description.
GBR Seabed Biodiversity
2.3.2.1.1
2-47
Reading Tapes
The term “fish” is this instance refers to any mobile marine organism, including sharks, rays, sea
snakes, sea turtles, squid, cuttlefish, spanner crabs and portunid crabs.
The data gathered from BRUVS tapes and stored in the database concern:
•
•
classification of the habitat in the field of view (topography, sediments, benthos)
•
their time of arrival
•
their maturity [adult or juvenile]
•
the identity of fish and CAABCODES
•
their behaviour [ 8 categories, including feeding on the bait]
•
their relative abundance (as MaxN = the maximum number visible at one time, or
distinguishable at different times as separate individuals – such as much larger and much
smaller, male and female)
the time elapsed before MaxN and feeding occurs
The BRUVS2.1mdb adds this data to, and calls up, “operations” data collected at sea when each
BRUVS is deployed.
The unique combination of a site and a Camera Number links all records in all tables of the relational
database. The main idea of this interface is to easily grab times that events occur in the tape, together
with reference images and reference video. Like any reference collection, this allows users to:
•
•
•
•
•
name unknown taxa
learn identification skills by comparing taxa with existing images
compile a useful, watermarked, representation of images of species from different locations,
aspects, colour phases and sizes/sexes. These can be emailed to international taxonomists for
verification of identifications.
apply “quality assurance” in updates to the parent databases by correct mis-identifications as
new information comes to hand
provide material to help interpret our results with clients.
Tape reading protocols were developed, ensuring tape readers must have:
•
•
captured image(s) for every taxa sighted on every tape, and benthos in the field of view for
every tape
saved the better images as “reference image(s)” for every taxon sighted – all new taxa were
accompanied by at least one reference image as they were named. We collected as many shots
from different angles, sizes, and colour phases as possible in the “library” (Figure 2-33, Figure
2-34).
The database now contains information on over 39,900 individual animals seen during the Seabed
Biodiversity Project, and over 17,000 images for reference by site, with 2,200 of the best reference
images in the “reference library”.
These protocols, and the design, operation, and troubleshooting for BRUVS2.1.mdb were fully
described in a manual. The software and manual can be obtained with the reference image library,
under certain terms and conditions of use, from BRUVS@aims.gov.au.
GBR Seabed Biodiversity
2-48
Figure 2-33: Image reference form in BRUVS2.1.mdb
Figure 2-34: Reference image for Pristipomoides multidens, with Lutjanus sebae, L. adetii and Epinephelus
undulatostriatus and E. areolatus in the background.
GBR Seabed Biodiversity
2-49
2.3.3. Laboratory Processing of Samples (T Hendriks, M Stowar, C Bartlett,
T Wassenberg, D Gledhill)
Following the field-trips, samples were freighted to several laboratories where detailed sorting,
identification, curation and detailed data recording were continued. Comprehensive reference
collections of voucher specimens were established and recorded into the database.
•
•
•
•
•
•
Porifera, octocorals, sled sampled fish and trawl bycatch invertebrates were sorted at the
Queensland Museum, South Bank campus (QMSB);
Molluscs, echinoderms, bryozoans and scleractinians were sorted at the Museum of Tropical
Queensland (QM, MTQ campus) by MTQ and Australian Institute of Marine Science (AIMS)
staff;
Crustaceans and trawl sampled fish were sorted at CSIRO Marine and Atmospheric Research,
Cleveland;
Trawl sampled prawns were sorted at Queensland Department of Primary Industries and
Fisheries (QDPI&F) in Cairns;
Sled sampled algae and seagrasses were sorted at Queensland Department of Primary
Industries and Fisheries (QDPI&F) in Townsville;
Particle size and carbonate composition analysis of sediment samples were completed by
Geoscience Australia (GA).
By decision of the Project’s Steering Committee, annelids, ascidians, crinoids, hydroids and trawl
sampled marine plants were not completed within the scope of the project. All other groups have been
completed.
For all samples at all agencies, the following processes were completed.
•
•
•
•
•
•
Each sample was physically separated into groups corresponding to species or species
equivalent Operational Taxonomic Unit (OTU);
Identifications were done to Operational Taxonomic Unit (OTU) level: known species where
possible, alpha species otherwise;
Total counts and weights of each OTU within a sample were recorded;
Specimen barcodes were assigned to each OTU for database purposes, as well as to facilitate
any further studies that may be completed with these specimens.
Reference material of specimens of all OTUs was retained. Specimens were kept the first
time that a new OTU was encountered at any laboratory — a voucher reference. These were
preserved and retained at the relevant laboratory.
Entry of all data into the database (Figure 2-35, Section 2.3.3.3).
2.3.3.1. Queensland Museum
The Queensland Museum, at QMSB and MTQ, was responsible for the taxonomy of many major
invertebrate Phyla, including Porifera, Echinodermata, Mollusca, Cnidaria, Ascidacea and Bryozoa
(Figure 2-36). All voucher specimens were preserved in ethanol and all other processed specimens
were stored either in ethanol, or frozen in freezers located at CSIRO Cleveland.
Identifications of Porifera, Cnidaria and Ascidacea have been especially problematic, as they would
normally require extensive preparations before genus/species combinations can be given with a high
level of confidence. Time constraints prevented histology from being completed, leaving scope for
further research to be completed in the future to allocate species names using histology or other means
as necessary.
GBR Seabed Biodiversity
2-50
Figure 2-35: An example of the form used in the laboratory to enter data obtained from the field samples. The
sample barcode number is entered and the database retrieves the site details including the sample photograph
from onboard the vessel. Individual species or OTU were then entered into genus or species boxes (middle
fields) and a pick list of names appears. By selecting the appropriate name the species numbers and weights were
then able to be recorded into the bottom RHS field.
Sponges were taken to operational taxonomic unit (OTU), with far fewer genus allocations, or even
higher taxonomy in some cases due to the difficulty of identification without histology. Some
histology has been completed and new species have been found.
Figure 2-36: Identifying and processing invertebrate samples at the Queensland Museum.
GBR Seabed Biodiversity
2-51
Octocoral sclerites could be seen on most specimens using a stereomicroscope, allowing for
taxonomic identification to the genus level. Differing specimens were then given an operational
taxonomic unit (OTU), rather than species name, in most cases.
Ascidians require treatment when collected to relax the zooids for a short period of time before being
fixed in formalin. They are subsequently transferred into ethanol before identification. As these
processes could not be followed upon collection at sea, identification became difficult and
consequently many samples were not identified.
The first year of the project included the design and implementation of the bar coding system used
across the project by different laboratories (QMSB, MTQ, AIMS, QDPI&F and CSIRO). This
allowed for a consistent tracking method across all specimens in which duplication of numbers was
eliminated. Now that the databases have been combined, this system has proven to be efficient.
The large amount of knowledge and experience gained by staff over the study period facilitated an
increase in the sorting rate. This was invaluable as deadlines were fast approaching with a high
quantity of material remaining to be sorted. The volume of samples collected for sorting and
identification was challenging throughout the project. In response, additional staff resources were
applied to the Project by the QM, CSIRO, QDPIF and AIMS. The rate of sorting, identifying, and
processing samples into the collections and databases was increased, and has been reviewed regularly,
with priorities for the various biotic groups being reviewed by the Steering Committee.
Processes for entry of data into the database were amended to improve efficiency. Originally, all data
at QM was initially being hand-written on paper datasets, with data entry only occurring at intervals
throughout the sorting process. This created difficulties in OTU allocation and any misprints on the
datasheets were difficult to rectify. This process was revised, so that hard copies of datasheets were
retained, but the data was entered immediately into the database, as the sorting was being done. This
minimised entry times, and any problems could be rectified instantly, rather than revisiting a sample
after being re-frozen or preserved and often stored at a different location.
More than 200,000 specimens have been identified in total, which has greatly enhanced the reference
collections of the Queensland Museum. This resource is now being used as a basis in taxonomy for
other studies of biodiversity and remains a resource for other institutions and researchers who are
unlikely to get access to these remote or protected areas in the future.
2.3.3.2. CSIRO (CMAR) and QDPI&F
Trawl-sampled fishes and all crustaceans, except penaeid prawns from the scientific trawl, were sorted
and identified at the Cleveland laboratory of CSIRO Marine and Atmospheric Research (CMAR).
Samples were stored at -20° C in cardboard cartons, with individual samples double-wrapped in
plastic bags to reduce freezer burn and deterioration.
Samples were thawed and sorted to OTU’s (Operational Taxonomic Units – roughly equivalent to a
species, but potentially not always aligned with currently recognised/named species) and retained on
crushed ice during processing to reduce deterioration of colour and body tissues. Each OTU was
allocated a registration barcode number and its weight and count for that site’s sample were recorded
in the laboratory database. Identified fish were re-frozen at -20°C for future work or dissemination to
collections, while crustaceans were preserved in 70% ethanol. All but the voucher specimens were
forwarded to the Queensland Museum.
To ensure accurate and repeatable identifications of specimen, a reference collection, consisting of a
voucher specimen and replicates, was established to represent all OTU’s and variants identified. When
an OTU was encountered for the first time, the sample was registered and a specimen was allocated as
the voucher. The voucher was photographed and given a score to indicate the taxonomist’s confidence
in the identification, following Williams et al. (1996).
Voucher fish were preserved by soaking them in 10% formalin for about a month, prior to transferring
them to 70% ethanol for storage. Additional specimens were also preserved to provide replicate
GBR Seabed Biodiversity
2-52
reference material. A tissue sample, for genetic analysis, was removed from some voucher and
replicate specimens prior to preservation. Tissues were stored in ethanol. Additional specimens, of
virtually all species, ranging from across the geographic range of the study area, were retained (frozen)
for the collection of tissue samples at a later date. Sub-collections of frozen and preserved material
were sent to the National Fish Collection, CSIRO Hobart, for further examination.
The fresh colour of many fish and crustacean species is a useful diagnostic tool, and for some species
colour patterns are the only field characters. A laboratory identification guide was compiled containing
images taken onboard and in the laboratory for each OTU and variant. Identification characters were
recorded to assist in accurate and consistent identifications. Images and identification characters were
also collated in an online database to allow interstate colleagues and visiting international experts to
view specimens remotely and thus assist with the identification of taxonomically difficult species.
2.3.3.3. Laboratory Sample Database (D Chetwynd)
The Laboratory Database, similar to its Vessel counterpart used MSAccess for storage of data in tables
and Visual Basic for Applications (VBA) for the software development, in particular data entry Forms
as well as Reports and Queries.
CSIRO, the Queensland Museum Brisbane and Townsville and the QDPI all had networked
installations of the Laboratory database local to their laboratory.
The database’s main purpose was to log taxonomic specimen information and images against the
samples collected in the field. The collected samples, initially sorted on the vessel to a general biotic
group level, would be further sorted to a species level or to the nearest taxonomic level possible, given
the available resources and or the state of the sample.
A list of Australian Marine species was compiled and stored in the database in taxonomic hierarchy to
act as a starting list to build on. Each of these unique entries was given an incrementing ID to denote
each Operational Taxonomic Unit (OTU); if and when a new entry was added an incremental value
was added as the new ID for the new OTU.
When sorting the samples, a sample was opened and, using the bar code scanner, the barcode label
from the sample was read. This triggered a response within the database, which retrieved the site and
sample data and images displaying them for the user to verify. The sample was then separated into its
various OTU’s. For examples, a crustacean sample could have three specimens of different crab
species and two specimens of the same species of prawn, therefore 4 OTU’s existed within this
sample. There was a 1 to many relationship between the sample (barcode) and the OTU’s within that
sample; for each sample there could be many OTU’s.
Once the OTU’s were identified the user allocated a separate barcode label to each OTU — this was
known as the jar code – and was scanned into the database and the label was placed in the OTU bag or
jar. If a digital image of the specimen was required, it was taken at this point with the jar code label
within the image.
Once the samples were sorted into OTU’s, they were individually added to the database. This was
done by selecting the lowest known taxonomic level to each OTU group via the pick lists in the entry
forms. If the identified species did not already exist in the precompiled list, then it was added via
another entry form for new OTUs, hence making it available.
Once the desired species was selected, the count and weight of specimens were entered against the
record, as well as the jar code, read by the barcode scanner. Any images previously taken were stored
against the OTU at this point.
Species data can be queried by any of the taxonomic levels, site, sample or even jar code.
GBR Seabed Biodiversity
2-53
2.4. DATA ANALYSES
2.4.1. BRUVS Species Models, Characterization & Prediction (M Cappo, G De’Ath)
A total of 40 environmental and spatial explanatory variables were used in analyses of BRUVS data.
These comprised: three spatial variables (depth, “Along”, “Across”); four sediment characteristics (%
mud, sand, gravel, carbonate); 20 physico-chemical parameters from the CARS and SeaWiFS dataset,
including measures of location (mean) and spread (Std Dev); 12 “harmonics” of polar temporal
variables (diurnal, lunar, seasonal); seabed current shear stress and shelf slope; and a trawl effort
index.
Distance “along” was set to range from 0 at the southern end to 1 at the far northern end. Distance
“across” was 0 on the coast and 1 on the 80 m isobath. The corners of the polygon formed in this way
were 142.53°E, -10.69°S and 144.06°E, -10.68°S at the northern end, and 152.49°E, -25.00°S and
152.90°E, -24.22°S at the southern end.
The BRUVS data were analysed in two ways. Firstly each species was treated as a univariate response
and, using boosted trees (GBM, Friedman 2001, De’ath 2006), its presence-absence was predicted
from the environmental data. Boosted trees are widely regarded as one of the best predictive
methodologies, and handle complex data sets and a broad range of loss functions. For the presenceabsence data analysed here, the binomial loss function was used. The collection of species was then
related to the best environmental predictors. Statistics representing the predictability of each species
and the predicting capacity of the environmental measures were then calculated. This produced a
“short-list” of 25 species and 20 explanatory variables. The 25 species were the best predicted species
that occurred on at least 7% of sites. The 20 explanatory variables were selected as the best predictors
of the 25 species. The binomial loss function was used in the boosting.
Secondly, the relative abundances on BRUVS (the sum of MaxN for each sampling station) were 4th
root transformed and analysed using multivariate regression trees (MRT; De'ath 2002). MRT used
sums of squares (Euclidean distance) for splitting. Twenty explanatory variables and twenty-five
species responses were short-listed in the same way as described above for the boosted trees. The bestsized tree was selected by five-fold cross-validation.
The tree defines a hierarchy of species communities and their spatial and environmental values that
locate them on the GBR. This hierarchical approach can be used with any clustering method
(constrained as is the case here with multivariate regression trees, or unconstrained). It also identifies
groups of species that co-occur at varying spatial scales to form communities. This contrasts with nonhierarchical methods which derive mutually exclusive clusters at a single spatial scale, thereby lacking
high-level (broad spatial scale) structure and ignoring information from highly prevalent species. The
homogeneity of the clusters formed by MRT (from 1 through to 13 nodes) was compared with similar
numbers of unconstrained cluster groups, using K-means clustering and Euclidean distance.
Indicator values (DLI; Dufrêne and Legendre 1997) were calculated for each species for each node of
the tree. For a given species and a given group of sites, the DLI is defined as the product of the mean
species abundance occurring in the group divided by the sum of the mean abundances in all other
groups (a type of specificity), times the proportion of sites within the group where the species occurs
(fidelity), multiplied by 100. DLI takes a maximum value of 100 if the species occurs at all sites in the
group and nowhere else, and 0 if it occurs at no sites within a node. Each species was associated with
the tree node (assemblage) where its maximum DLI value occurred, and the numbers of indicator
species and their values were used to characterize each node of the tree. Species with high DLIs were
used as characteristic members of each assemblage, and the spatial extent of the group indicated the
region where the species was predominantly found.
GBR Seabed Biodiversity
2-54
2.4.2. Single Species Biophysical Models and Prediction (M Browne, W Venables)
2.4.2.1. Assumptions and Challenges
The distribution of each identified species was considered separately in relation to the available
physical variables used as predictors in the study. This approach assumed that the observed geo-spatial
distribution of each species may be adequately explained by an underlying physical gradient. Each
species would be expected to have its own preferred habitat range and tolerance for various physical
parameters, and the spatial variation in physical variables is thought to drive the observed regional
scale spatial distributions of the taxa.
From the point of view of biophysical modelling, it is known that sampling in a marine environment is
very susceptible to sampling variability. That is, for fixed levels of the biophysical parameters
involved, the event of observing a species, and the biomass obtained given that the species was
observed, were still liable to be highly variable due to random influences — neither observable nor
under the control of the sampling mechanism. Thus, the modelling approach applied needed to be
conservative and to anticipate that much random variation will remain unexplained. Technically, the
biological response variables may be described as having a zero-inflated log-normal distribution. That
is, given that some biomass is observed, the samples are approximately Gaussian on a log-scale, with a
mean depending linearly on the physical predictors. However, given that the estimated probability of
observing any particular species is typically relatively low (< 10%), the very large number of zeros in
any species’ site-records strongly suggests a two-phase approach: initially modelling the probability of
observing a species at all and conditionally modelling the distribution of the log-biomass given that
the species is observed.
2.4.2.2. Model Formulation
The two-stage model used to relate the observed biomass to the underlying physical variables may be
described as follows:
•
We use a logistic regression model for the chance of observing a species in a sampling event.
For any given species, S , the chance of observation is
T
eβ x
pS = Pr (S ) =
T
1 + eβ x
•
or equivalently
logit (ps ) = βT x
where x is a collection of suitably chosen physical variables for that species and β is the
vector of coefficients multiplying them, to be estimated from the data. This first phase model
is a standard logistic regression for presence/absence. The first component of x 0 is often a
constant predictor (unity) and its coefficient β0 is called the intercept term.
The log-biomass of the species given the fact that it has been observed in a sampling event is
them modelled as a normal linear regression. In symbols using an evident notation
log BS | S ~ N (γT z, σ 2 )
i.e. E [log BS | S ] = γT z, Var [log BS | S ] = σ 2
where again z is a collection of suitably chosen physical predictors for species S and z 0 = 1
making γ 0 the intercept term.
Note that for the logistic regression stage all site records can be used in the model calibration, but for
the second stage only those site records for which there is a non-zero biomass contribute information
that can be used in this calibration process. This sometimes limits the modelling strategies possible.
Combining both stages of the model, the estimate of expected biomass at a given sampling site is then
ˆ [B ] = pˆ exp Eˆ [log B | S ] =
E
S
s
S
exp (βˆT x )
1 + exp (βˆT x )
× exp (γˆT z )
GBR Seabed Biodiversity
2-55
where the circumflex accents denote quantities estimated from sample data. Technically this is the
median rather than the mean but since for the lognormal distribution the median is proportional to the
mean it produces the same relative picture of biomass distribution.
Approximate standard errors for the quantities involved can be derived by the usual delta method. The
formulae are complex, but as the method is entirely standard and can largely be relegated to the
computational procedures, we omit them here for simplicity.
2.4.2.3. Explanatory physical variables, covariates, offsets and prediction
In the formulae above, the general-purpose explanatory variable x j (or z j ) represents any member of
the combined set of:
•
•
•
•
•
•
•
28 physical explanatory variables (e.g. percent_mud). A full list and description of the
physical variables is provided elsewhere (Section 2.1.1).
2 spatial variables (‘across’ and ‘along’) representing in relative terms the distance between
shoreline and outer reef, and between northern- and southern-most points of the GBR.
Squared terms of physical and spatial variates (e.g. percent_mud^2)
Second degree interaction terms between physical or temporal explanatory variables, or both
(e.g. percent_mud × benthic_irradiance)
Harmonic terms in the temporal covariates: (described specifically below)
The measurement method used (i.e. modified trawl net / benthic sled) as a factorial predictor,
A weighted annual average of commercial trawl fishery effort local to the sampling site.
For the purposes of generating predictions on the GBR study region, we were primarily interested in
the relationship between the physical variables and observed biomass. It was, however, recognised that
other factors were be expected to play a role in the observed biomass. For example, many species are
known to have a strong diurnal or lunar behavioural cycle, or a seasonal abundance component. Also,
when generating estimates on the physical grid for full coverage of the GBR region, it is desirable that
the estimated distributions are independent of sampling device, which we term the ‘survey type’.
These variables are included in the models in order to ensure that these effects do not interfere with
estimation of the effect of genuinely physical or spatial predictors. When predictions were later made
from these fitted models on to the entire GBR grid, the covariate predictors are set to values that
represent the sampling disposition under which the estimated biomass would be largest. This is to
promote maximum contrast in the predictions.
The temporal covariates included ‘Time of Day’, ‘Moon Phase’, and ‘Time of Year’ (or ‘season’).
Since GLMs depend on the explanatory variables through a single linear function, it is appropriate to
represent the effect of such temporal predictors through harmonic terms, that is
⎛ 2 pkt ⎞
x a( k ) (t ) = sin ⎜
⎟
⎝ T ⎠
⎛ 2 pkt ⎞
xb( k ) (t ) = cos⎜
⎟
⎝ T ⎠
k = 1,2, K
where T is the appropriate fundamental period. Including temporal covariates in this way ensures
that the predictions will obey the natural periodicity with respect to such predictors.
It is has been mentioned that ‘survey type’, whether the measurement was made with the trawl or sled,
was also included as a covariate factor in the analysis. If a species was observed a minimum number of
times in both sampling devices, then all data was included in the model, with the ‘Survey Type’ factor
included. However, if a species was observed on very few occasions, or not at all, with either one of
the sampling devices, then the usefulness of a model incorporating both devices was limited. In these
cases the models were based on data from the most productive sampling device only, obviating the
need for a ‘Survey Type’ covariate in the model.
GBR Seabed Biodiversity
2-56
The area swept by each device at each site was recorded, and although efforts were made to keep this
constant, it did vary somewhat. The effect of trawling (or ‘sledding’) over a greater area is to increase
the probability of a catch in some monotic way, and to increase the amount of biomass caught, on
average, proportionally. Thus, the recorded swept area on the log scale was included as a candidate
predictor for the logistic regression models and as a fixed offset in the biomass models. Predictions
were then standardised by setting swept area to 1 Ha.
2.4.2.4. Model construction and variable selection
Each species model included, at a minimum, harmonics in temporal covariates as well as ‘survey type’
and ‘log-swept area’, as explanatory variables (or offset). It has been noted above that potentially
included variables included 30 physical or spatial variables, as well as squared terms and interactions
between physical explanatory variables (30 + 30 × 31/2 = 495 possible candidate predictors). Not all
possible choices, however, are equally reasonable a priori. In practice we note that if a physical
predictor is important either for the presence/absence or the conditional biomass of a taxon then the
linear term will generally make this manifest and the additional contribution to the model due to
higher order terms involving these will be much smaller, though often quite useful, of course. We
adopted a two-stage variable screening method. At the first stage we allowed selections to occur only
on the linear terms. Once these were found, we considered all possible squared and interaction terms
involving the identified variables for additional inclusion in the model at the second stage. In this way
we ensured that if a second order term was included in the model, its marginal linear terms were also
included, which is generally regarded as a desirable feature of empirical statistical modelling.
The physical variables available are by their nature highly collinear (or confounded) and to counteract
this we adopted a rigorous variable inclusion policy. Such a policy also has a good chance of ensuring
some interpretability for the models as well as predictive effectiveness, although this has to be taken
with some caution. The criterion used in these analyses was the ‘Bayesian Information Criteria’ due to
Schwarz, 1978, defined as:
BIC = −2 log L + k log n
with log L denoting the log-likelihood of the optimized model, k the number of estimated parameters
used in the model and n being the number of observations. The BIC attempts to balance model
performance in the training sample with a penalized measure of model complexity to ensure that the
model will capture as much signal and as little noise as possible. The related Akaike Information
Criterion (AIC) penalizes complexity by a factor of 2 rather than log n , leading to more complex
models which in this context, from our experience seem to sacrifice interpretability for minimal gains
in predictive performance.
At each iteration of the stepwise search procedure, a term was added or removed from the working
model if the inclusion or removal resulted in the greatest reduction in the BIC criterion. As mentioned
earlier, this was done for the linear terms initially, and then for the second degree terms involving
those predictors, with the linear terms chosen fixed in the model. This was done independently for the
presence/absence and the conditional biomass models for any sampled species.
2.4.2.5. Prediction on the GBR grid
Based on the constructed models, estimates of expected biomass were generated for each species, for
each grid location in the GBR study region. This involved interpolation between the sites of the
training set and also, as noted, above, a transform back from the log scale to the natural scale. Both of
these operations can result in unrealistic estimates and in this case they will be much more likely to be
unrealistically large than small. As a heuristic, it was determined that the largest standardised per Ha
biomass actually caught for a particular species would be used to determine a ceiling on the largest
confident estimate on the grid to be used. Final estimates were truncated at the largest observed
standardised catch rate observed in the sample itself. This seems preferable to having unrealistic
estimates dominating a graphical presentation.
GBR Seabed Biodiversity
2-57
For plotting purposes, the coefficient of variation (CV = estimated SD/estimated mean), was then
calculated at each grid point. In this case the CV offers a more suitable statistic to represent
uncertainty than the SE because it does so in a proportional way. We used models for log biomass, for
example, for the same reasons.
2.4.2.6. Graphical presentation of results
14
Figure 2-37 displays the colouring scheme used to present the species model estimates over the GBR
region. Inspection of the table shows that an order 2 ‘octave’ style is used for both the mean biomass
estimates and the CV estimates. This method proved effective for illustrating the significant spatial
variation over the study area. The other advantage of this approach is that all single-species maps are
plotted using the same colour key, facilitating abundance comparisons between species. A rainbow
colour scheme is used to differentiate between different mean biomass estimates, whilst colour
intensity reduces as the CV increases. Thus, intensity may be taken to indicate the relative certainty of
our estimate at each grid point. Finally, a quantized colour scheme (i.e. constant colours within each
biomass / CV range) was found improve the readability of the generated maps, without losing any
significant information.
16384g
12
4096g
10
2048g
1024g
8
512g
256g
6
128g
64g
32g
4
log base 2 (mean estimate): grams /Ha
8192g
16g
8g
2
4g
-1
1
2
4
8
16
0
1
2
3
4
log base 2 (mean estimate / SE) ratio
Figure 2-37: Colour scheme used for species distribution mapping
GBR Seabed Biodiversity
2-58
2.4.3. Species Groups Characterization and Prediction (M Browne)
One of the goals of the statistical analyses and modelling was to develop groups of ecologically
similar species (at least with regard to their distribution in the physical environment) and to generate
predictive models for the biomass of the entire group. We note that the taxonomic hierarchy may not
provide a suitable basis for forming groups with this kind of similarity. Clustering here refers to an
empirical grouping, hierarchical or otherwise, based on the characteristic distribution of species as
observed from the survey effort.
There are several reasons why such a grouping or clustering of species is useful. For example,
clustering allows the organisation of species with similar characteristics into common categories. As
an exploratory technique it may provide the basis for more targeted analysis for establishing possible
ecological dependencies and relationships. Due to the broad scope of the current project, individual
modelling and prediction has been carried out for a very large number of distinct species (> 800).
Species clustering and aggregated biomass data within groups can be a convenient method for
summarising or quantifying the variation in species distributions through a small set of key grouped
models. Finally, in cases where distinct and relatively rare species are observed at different sites, but
share similar biophysical distributions, aggregation of the separate species into a single group can
provide a way to utilise parts of the data set not otherwise available for modelling input.
It should be acknowledged that although clustering methods are well developed with acknowledged
utility, to some degree heuristics and arbitrary decisions inevitably play some role, such as, for
example, in something as fundamental as the choice of a distance metric. In our view the purposes of
clustering have to be always kept in mind, as much biological insight and knowledge has to guide the
process at every stage and the results need to be viewed in this light afterwards before a grouping is
used.
2.4.3.1. Approach
Most clustering methods work by considering a similarity / dissimilarity matrix which describes
‘distances’ between objects. Broadly speaking, a clustering algorithm attempts to assign objects to
groups so that similar ‘objects’ (in this case species) are put together in the same groups. The most
obvious (and simplest) method for species clustering is to begin with the site / species presence /
absence matrix and define species similarity using ‘Manhattan’ or ‘city block’ distances calculated
between the distribution of species over sampled sites.
The species clustering method that has been adopted here is based on upon estimates at survey sites, as
described in section 2.4.2, and thus may be described as constrained by the physical predictors. The
modelled distributions are used as a species descriptor, rather than using the profile of raw biomass or
presence /absence, to describe the similarity between the distribution of species. There are several
important considerations that motivated this approach.
There are advantages and disadvantages to this approach. The most obvious disadvantage is that the
representation is only as good as the species models were an effective summary of the information in
the data itself. This will not always be totally satisfactory, of course. However the advantages include
the fact that this allows us to control for factors such as the temporal, sampling method and swept area
covariates which are inseparable from the remainder of the information in the raw data itself. It also
allows us at least a potential device for counteracting the effects of false negatives, though admittedly,
and inevitably, a somewhat speculative one. We note, also, that the data smoothed need not be entirely
realistic for the groupings based on it to be effective enough for our subsequent purposes. Once the
groups were decided, the subsequent analysis was based on the aggregated data; that is, on real data
once again. This provides a useful safeguard in the logical chain of events.
GBR Seabed Biodiversity
2-59
2.4.3.2. Method
There was a total of 840 species (N) that both occurred sufficiently frequently for biophysical
modelling, and also led to statistically satisfactory two-stage models by the model construction
methods as detailed in the previous sections. Model estimates were generated for the 1644 sites (M)
visited by the vessels, controlling for covariate predictions, for the reasons previously outlined.
As a measure of distributional affinities for the species, an N × N Spearman correlation matrix C was
created from the model estimates over sites. The correlation was chosen as the base for a similarity
measure because it was desired that the clustering criterion be invariant to absolute abundance.
Choosing spearman correlations also makes the estimate robust to outlying values, as are frequently
obtained under extrapolation from an empirical model as here. Two species with very different
prevalence or mean observed biomass may therefore have high correlations if the standardized pattern
of co-variation is similar.
To convert the correlation matrix, C, into a dissimilarity matrix D to be used for clustering we used the
following transformation:
Dij =
1 − Cij
1 + Cij
This guarantees that self-dissimilarities are zero, distances are otherwise non-negative and, since the
correlations themselves are between -1 and 1, the dissimilarities are in the range 0 ≤ Dij < ∞ . The
denominator chosen was motivated by the fact that so defined log Dij will be approximately normal
with equal variance, by a standard result in correlation theory. It also ensures an open-ended upper
limit for the dissimilarities themselves. (Fortunately C ij = −1 did not occur in practice.)
After extensive experimentation with different methods for clustering species, hierarchical clustering
using Ward’s method was chosen as a reliable and well established procedure for heuristic clustering.
Ward’s (1963) method is a clustering procedure seeking to form the partitions Pn, P n-1,........, P1 in a
manner that minimizes the approximation error associated with each grouping. At each step in the
analysis, the union of every possible cluster pair is considered and the two clusters whose fusion
results in minimum increase in the total sum-of-squares are combined. The ANOVA-type approach
used by Ward makes this method one of the more principled approaches to clustering.
2.4.4. Site Groups Characterization and Prediction (W Venables)
Another goal of the statistical analysis and modelling was to identify various areas of seabed in the
GBR study area in which the mix of biota was as homogeneous as possible and in some way distinct
from the mix in other areas. These different, approximately homogeneous, mixtures can be called
"assemblages". Any individual assemblage may be expressed in several disjoint geographical regions;
there was no requirement that they be spatially contiguous. A further property was that, at the broad
scale, these assemblages would be characterised using the available full-coverage physical variables.
The biotic data used were the biomasses of identified fauna and flora at sites sampled by the
epibenthic sled and research trawl.
A number of strategies are possible to achieve this result. One strategy would involve three separate
steps, each with a number of options regarding method:
•
•
•
First, partition the study sites into clusters based on the biological data alone, then
Develop models to predict these clusters from the physical variables alone, and
Use the predictive model to classify the entire GBR region.
GBR Seabed Biodiversity
2-60
The strategy finally adopted essentially combined the above partitioning, modelling and prediction
steps into a single procedure, using a tree-modelling method. In outline this procedure was as follows:
•
•
•
•
Using the biological information alone, construct a distance matrix between sites,
Define a deviance measure on any clustering of sites, based on the biological distance matrix,
Construct a decision tree using the available physical variables; the resulting terminal nodes of
the decision tree define the site clusters,
Use the decision tree to classify the entire GBR study grid and map the result.
The details of this procedure are explained in the following sections.
2.4.4.1. Biological distance matrix
The distance matrix between the sampled sites was based on the species biomass predictions from the
individual species modelling, as described in Section 2.4.2. Distance matrices based on real sample
biomass data would be preferable. However, more than one third of species had one or more
significant temporal cycles (such as season, moon phase and time of day), which strongly affected
their sampling-rates but not their abundance, and may have caused the raw data to produce assemblage
splits due to arbitrary timing of sampling rather than actual distributions. Use of the predicted values,
adjusted for the temporal variables, allowed distances to be based on biomass values for a standardized
set of environmental conditions, but optimal with respect to the temporal catchability variables for
each species. It was nevertheless acknowledged that the modelled predictions, being smoothed fitted
derived data, would not be free of other issues — to minimize these, only predictions to sampled sites
were used, rather than predictions (extrapolations) to the full GBR study grid.
The form of the distance matrix was the commonly used Bray-Curtis metric, but a number of possible
prior transformations of the predicted biomasses were considered. The final transformation used was
the 8th root, which for large biomasses behaves like a ‘weak’ log transformation, but for small
biomasses behaves approximately linearly and, unlike log(b), was not adversely affected by zeros.
Thus, the dissimilarity d between sites i and j was:
∑ bα − bα
t
dij =
∑ (b
k =1
t
k =1
ik
α
ik
+ b jk )
jk
α
, where α = 1/8 and t is the number of taxa.
and b was the biomass of each species k.
2.4.4.2. Deviance measure
Consider any group of g sites, G . These can be represented by points in a plane, where the
(Euclidean) distances between them are intended to represent distances from the biological distance
matrix. Choosing any site from the group, say Si , the span of the group from Si was defined to be the
sum of squared distances from the reference site, Si , to all other sites of the group:
span ( Si ) = ∑ dik2
g
k =1
The deviance of group G , say D ( G ) , was then defined as the minimum value of the span, and the
reference site from which this minimum was achieved was called the medoid of G, say S M , using the
terminology of Kaufmann and Rousseeuw, (1990).
D ( G ) = min span ( Si ) = span ( S M )
Si ∈G
GBR Seabed Biodiversity
2-61
If the group G was partitioned into two subgroups, G = G1 ∪ G2 , G1 ∩ G2 = ∅ , then the deviance
of the partition was defined as the sum of the two component deviances:
D ( G1 , G2 ) = D ( G1 ) + D ( G2 )
It can be seen that D ( G1 , G2 ) ≤ D ( G ) , since partitioning the group lead to one fewer non-zero
distance in the sum defining the deviance, and the remaining distances could be no larger than they
were prior to partition. The difference D ( G ) − D ( G1 , G2 ) ≥ 0 was the reduction in deviance
achieved by the partition. The geometric representation of these notions was illustrated by a simple
two-dimensional example in Figure 38 below.
Figure 38: Spans, medoids, deviance and deviance reduction due to partition. (a) shows a group of ten sites
using two-dimensional Euclidean distances to represent Bray-Curtis distances. (b) shows the span of
the group from an arbitrary reference site. (c) shows the minimum span, which defines the medoid; the
sum of squared distances from the medoid is then the deviance of the group. (d) shows a partition of
the original group into two subgroups, and the spans defining the deviance component of each. The
partition is achieved by a split on the x-coordinate. The reduction in deviance achieved by partition is
then 32.83-14.28 = 18.55.
2.4.4.3. Decision tree construction
A decision tree was constructed using the standard technique, described below. The rpart package
of R (R Development Core Team, 2005), which allows users to define their own deviance and splitting
criterion, was used for the computations, with standard defaults selected.
The method was as follows:
•
Initially all sites were considered a single cluster, with an initial total deviance.
•
•
•
•
GBR Seabed Biodiversity
2-62
Consider any potential predictor variable, X . The sites may be partitioned into two groups
by selecting those sites whose values for X are below some threshold, X ≤ a as one group
and the complementary set, those for which X > a , as the other. This was described as a split
of the group on variable X at threshold a .
Partition the group into two clusters by selecting the optimal split variable and the optimal
threshold to achieve the greatest possible reduction in deviance of the span of Bray-Curtis
distances.
Apply the same procedure recursively to both clusters, until some stopping criterion has been
achieved.
The stopping criteria were as follows:
o
The group must contain at least 20 sites for further splits to be considered.
o
The group has deviance zero, or
o
A further split of the group would not reduce the deviance by more than 1%.
In this application, the third criterion was typically encountered, the first criterion was achieved only
once and the second was never triggered — the outcome of which produced 16 groups. The result was
displayed as a decision tree in the usual graphical form.
As a check on the biological similarities between the resulting 16 site groups, as defined by the above
tree method that split recursively on physical variables, the following procedure was applied to the
biological data only:
•
•
•
the medoid site of each site-group was identified,
the Bray-Curtis dissimilarities between the medoids were extracted from the entire distance
matrix,
a hierarchical clustering and dendrogram was computed using Ward's method.
The congruence between the structure of the decision tree and that of the cluster dendrogram was then
examined.
2.4.4.4. Classification of the GBR grids
The grids of the entire GBR study region were classified into the 16 groups, based on the decision tree
splits of the physical variables, and the resulting pattern of groups was mapped.
In this application, the recursive partitioning algorithm was used for unsupervised learning, which was
somewhat unusual. More usually, tree methods are used for supervised learning and the result is used
for prediction of some à priori defined attribute (classification tree) or quantity (regression tree), i.e.
supervised learning. In such cases, the decision tree is normally pruned to some justifiable level of
complexity by cross-validation, with the aim of improving the accuracy of prediction for new data.
In the current application, the procedure was used to define the attribute (site groups) itself. Thus, the
normal cross-validated pruning process was not available as there is no à priori definition of the
attribute. Correspondingly, the claims made for the result were simply that it defined a partition of the
study sites into groups which were (a) defined by splits in the prediction variables and hence readily
extended to the entire GBR study region and (b) as homogeneous as possible within, in the sense of
Bray-Curtis distances based on transformed predicted biomasses. There was no “correct” number of
site-group assemblages and the resultant number of groups was somewhat arbitrary. While the default
criteria were generally accepted guidelines, the 16 groups identified here may be more (or less) than
was warranted.
GBR Seabed Biodiversity
2-63
2.4.4.5. Derived species groups and indicators
In addition to defining site-groups assemblages, it was also useful to examine the extent of
associations between these site groups and species biomass distributions in a simple and direct way. If
a species was strongly associated with one site group and only weakly associated with others, such a
species could be described as an indicator species for that site group assemblage. The approach was as
follows.
A site group characteristic distribution for the kth site group was defined by the vector ξ ( k ) , where the
components were defined as:
Note that
∑ξ
⎧1/ nk for sites Si in site group k , which has nk sites, and
otherwise
⎩0
ξi( k ) = ⎨
(k )
i
= 1 for all k, conforming to the definition of a distribution.
i
Using the predicted mean biomass for each species, the species distribution vector for the mth species
was defined as:
ηi( m ) = Bi / ∑ Bi
i
that is, the predicted biomass, also normalized to add to unity over all study sites.
Further, a quantity termed the affinity distance between species and a site-group was defined as:
α mk =
⎛
⎞
cos −1 ⎜ ∑ ηi( m )ξi( k ) ⎟ ,
π
⎝ i
⎠
2
which has values in the range 0 ≤ α mk ≤ 1 . The quantity was zero only if the species was entirely
within the site group and uniformly distributed, and unity if the species distribution was entirely
outside the site group. The affinity distance was also a minor variation on the standard Hellinger
distance measure between two distributions. It also has a geometric interpretation as the angle
between two unit vectors, but heuristically, the affinity measure was an indicator of the preference any
species had for any particular site-group.
Then, a distance between species was defined that represented the dissimilarity of their preference
patterns for the site-group assemblages. Since the affinities were all measured on the same bounded
scale, ordinary Euclidean distances were used:
2
= ∑ (α pk − α qk )
E pq
2
k
In turn, this distance matrix was used to generate species clusterings that grouped together species
with similar preference patterns for the site-groups. Standard hierarchical clustering, using Ward’s
method of linkage between groups, was applied. The resulting dendrogram, for all 839 taxa for which
prediction models were possible, was cut at the appropriate location — for purposes of this analysis —
to define 12 species groups.
While the species affinities provided information that helped characterise the site-group assemblages,
it was also useful to investigate the relative density of each species group within each site group as
additional characterising information. This computation was done in four stages, namely:
• sum the species biomass predictions within each species group,
• normalize these aggregate biomass distributions, as for single species, to give 12 species group
distributions,
• average these species group biomass distributions within each site group, to give “percentage
biomass per site” within each site group,
• present the result as a series of bar charts, one for each site group.
The 16 site groups were labelled numerically and the 12 species groups were labelled alphabetically.
GBR Seabed Biodiversity
2-64
2.4.5. Video Habitat Characterization and Prediction (W Venables)
The towed video camera yielded data in two forms. While the vessel was in motion an operator
recorded both the substratum type and the biota currently in view in a continuous manner (Section
2.2.2). Also, later, in the laboratory, individual frames from the video record were scored for a
substrate and biota in a much more detailed way, though not on a continuous basis but as a sample of
about 30 frames from the tow record (Section 2.3.1). This section outlines how the towed video
camera data was used to provide some indication of the broad habitat types in the study area.
Table 2-9: Sediment and group biological cover classes for analysed video tow data. Note that sediment classes
up to lage pebble could be further classified as rippled or in waves and cobble as waves.
Grouped biological
Bioturbation
Algae_Caulerpa
Algae_Coralline
Algae_Filamentous
Algae_Halimeda
Algae_Mixed
Algae_Udotea
Algae_Ulva
Seagrass_H.ovalis
Seagrass_H.spinulosa
Seagrass_strapform
Bryozoan_branching
Bryozoan_encrusting
Hydroid
Sea pen
Soft_Coral
Solenocaulon
Sea_Whip
Gorgonian
Sponge
Solitary_Coral
Hard_Coral
Sediment
Mud (0.06 mm)
Sand (0.06-2.0 mm)
Coarse sand (2-4 mm)
Small pebble (4-16 mm)
Large pebble (16-64 mm)
Cobble (64-256 mm)
Boulder (256-1024 mm)
Large boulder (1024 mm+)
Larger than field of view
2.4.5.1. Features of the data
In all four cases (Vessel or Laboratory, Substratum or Biology) the quantities characterising each site
formed a vector of proportions, i.e. they summed to unity. In the case of the Vessel data this was a
feature of the recording protocol itself. For the laboratory data, video frames were scored by the
laboratory operator on a proportional basis and these were aggregated and re-normalised to give an
estimate of the respective covering proportions for the entire transect.
The laboratory data set for the benthic biological cover had 114 different classes. These were
determined primarily by a fairly coarse ‘feature’ type (e.g. algae) with more detail provided by a
‘descriptor’ (e.g. filamentous). At this level of differentiation the analysis diagnostics indicated little
grouping structure in the data, the likely reason being that the finer morphotypical classification did
not correspond with the biophysical affinities that were ecologically important. For this reason the
biology cover classes were grouped for analysis: the feature/descriptor classes were amalgamated to
the feature level except where morphotypes had been observed to be strongly differentiated in the field
GBR Seabed Biodiversity
2-65
(e.g. filamentous mats vs other algae) or where identifiable genera dominated certain habitats (e.g.
Halimeda or Caulerpa). From an observer viewpoint, the resultant groups would appear to represent
the more visually dominant habitat components. The cover classes after grouping are shown in Table
2-9.
Some grouping of the vessel biology cover classes was also done for similar reasons. In this case,
however, the amalgamation merely grouped ‘sparse’, ‘medium’ and ‘high’ cover classes of several key
epibenthic feature types (Table 2-6). No grouping was done for the substratum.
2.4.5.2. Distance metrics
Using data that comes in the form of proportions for clustering is somewhat unusual in ecology and
several possible distance metrics were considered. The three that were considered in most detail were:
1. The standard Euclidean metric: E ij2 =
K
∑ (p
ik
k =1
− p jk )
2
2. The Hellinger distance metric between two probability distributions: H ij2 =
K
∑(
k =1
pik − p jk )
2
3. The Manhattan distance metric, (which for data in the form of proportions is equivalent to the
Bray-Curtis metric): M ij =
K
∑p
ik
− p jk
k =1
These differ in the weight they give to deviations in proportions. The Euclidean metric weights large
differences much more than small ones, the Manhattan metric weights all deviations equally and the
Hellinger metric is intermediate in the sense that it initially reduces differences before giving higher
weight to those which remain large. After clustering, by the methods to be described below, there
were noticeable differences between the results, though the same general picture emerged at the higher
levels. However the Manhattan metric appeared to best recover the known large scale habitat patterns
and this metric has been retained in the results.
2.4.5.3. Clustering methods
The primary aim of this analysis of the video data was to ‘find habitats’. That is, groupings of sites
that appeared to have a physical and biological profile that was reasonably consistent within, but as
distinct as possible between. These ‘habitats’ must be interpreted as structure at a scale allowed by the
data and methods, for management as much as scientific purposes. To be useful for management
purposes they must not only be cogent, but they also must be interpolated on to the entire GBR spatial
grid. For scientific purposes it is clearly more desirable to have groupings that can be related to the
physical variables alone, rather than by physical and spatial variables, as this would suggest that the
results to some extent might be interpretable as biological responses and may be transferable outside
the study area. The approach taken, however, was to leave the spatial coordinates in the suite of
potential predictors and to consider outcomes in which they were not needed. Spatial coordinates
were in the form of ‘Along’ and ‘Across’, measured relative to the reef geometry itself. In other
analyses, these have been the most flexible and sensitive method for incorporating spatial predictors
into the analysis.
Two general clustering strategies were considered:
1. Identify groups using the cover proportions alone and having established the number and content
of the groups, develop predictors for them from the physical variables, both to define further the
physical characteristics of the ‘habitat’ and to allow some measure of interpolation to the grid.
GBR Seabed Biodiversity
2-66
2. Use methods that partition the data according to the physical variables into groups that are as
homogeneous as possible with respect to their cover proportions.
The first strategy was successful in building groups that had a demonstrable cogency, although the
number of groups was usually much smaller than either the number of ‘real’ habitats on the GBR,
however defined, and frequently probably too broad even for management purposes. The real problem
came when trying to develop predictors for them on the basis of the physical variables. No
satisfactory prediction outcomes were obtained from application of a variety of possible techniques,
including discriminant analysis, classification trees and neural networks.
The second strategy produced the more satisfactory results. Tree-based methods have a number of
advantages in this context, in particular:
•
•
•
They partition the data based on cuts in the predictor variables and so the results are easy to
appreciate and may be both practical for management and ecologically informative,
They intrinsically address the variables selection issue, and
They simultaneously produce a prediction device to interpolate the results to the GBR grid,
(and elsewhere, if somewhat more speculatively).
Trees have a number of disadvantages as well, of course. These include the fact that trees of this kind
can be structurally unstable (even if predictively stable), and there is a need for caution in interpreting
the structure of the results. Many different tree structures can often lead to virtually equivalent
prediction results, particularly when there are strong collinearities in the predictors, as is usually the
case.
Multivariate regression trees (De’Ath, 2002) seek to construct a tree predictor for a multivariate
response based on a within-group deviance measure that may be described as the sum of the squared
Euclidean distances to the group centroid:
K
E g2 = ∑ ∑ (pij − p•(jg ) )
i ∈g
2
j =1
(g )
•j
where p is the mean value for group g of the jth proportion. De’Ath’s software is available as a
package for the R statistical computing environment. See http://www.R-project.org,
package ‘mvpart’. This software package can be used for partitioning with the Euclidean metric
directly and with the Hellinger metric using square roots of the proportions as the responses. An
advantage of this method and software is that it automatically provides cross-validation and hence
some guidance on the degree of complexity in the grouping structure warranted by the data. This is
because ‘mvpart’ develops a predictor of the proportions, and so the data may be used for crossvalidation purposes.
Adapting mvpart to accommodate a general distance metric, and hence evaluation of the Manhattan
metric, was well outside the scope of this project. However, an alternative tree strategy, the ‘rpart’
package, was readily adapted for this purpose using its ‘user-written splits’ feature.
This alternative strategy began with the matrix of pairwise distances between sites. Let Dij be the
distance between site i and site j. Given a group g, let g ∈ g be the index of, for now, an arbitrary
member of the group. The within group deviance was then defined as the sum of squared distances of
all members of the group to g :
E g2 = ∑ D ig2
i ∈g
To complete the definition, the reference member, g , was chosen so as to minimise this quantity. The
object within the group relative to which this minimum was achieved is sometimes called the ‘medoid’
of the group as opposed to the centroid.
At any stage the partitioning algorithm considered all possible cuts with respect to the predictor
variables and, for each, compared the sum of deviances for the two child nodes with that of the parent.
A cut was made using the best place within the best predictor, and the process was repeated
recursively for each child node. The process stops either when a group was too small for partition
GBR Seabed Biodiversity
2-67
according to a criterion set at the outset (by default 20 sites) or if the improvement in deviance offered
by partitioning into child nodes is less than a preset percentage of the deviance of the parent (by
default, 1%). In these data, no groups were determined by the group size criterion, implying that the
partitioning process was terminated when no further real improvement in group homogeneity could be
obtained.
The partitioning algorithm was used directly to find groups in the data, and not to find a predictor for
some underlying quantity. The process is technically known as “unsupervised, constrained learning”.
However, this method does not perform cross-validation as an objective indicator of the number of
groups in the data. While lengthy solutions could be developed, the mvpart cross validations were
used to give some information on the degree of complexity warranted, given the similarity of the
methods.
2.4.5.4. Choice of data source
As noted above, video habitat data was available for both biological structural types as well as
substratum, and sourced from real-time entry on the vessel or from more detailed post-analysis in the
laboratory. All were evaluated for the purpose of characterising and mapping seabed habitats.
The biological habitat data was seen as the primary output of the video tow, although the substratum
data was also investigated as follows:
•
•
•
Development of predictors for the substratum data provided a reality check on the physical
predictors (especially sediment type) used for analysis. This did not reveal any anomalies in the
sense that the physical predictors split substratum types as would be expected.
To check the adequacy of the physical predictors for developing tree models for the biological
profile, the substratum variables were included as predictors along with the external physical
predictors. Such variables are only known at the observed sites and so cannot be used for
predicting to the full coverage GBR grid, but if they were to be chosen ahead of the external
predictors it would have suggested that defining useful habitats required either different physical
variables from the ones we have available, or that they were required on a finer scale.
Under the first clustering strategy listed above, partitioning using the substratum variables alone
produced groups that were easily predicted by the external variables, perhaps not a surprising
result but a reality check nevertheless. Partitioning using the biological proportions alone gave
meaningful groups, as noted, but for which no satisfactory predicting device from the external
predictors alone could be developed. Combining biological and substratum profiles for inter-site
distances produced a somewhat intermediate situation, with only marginally satisfactory
predictors possible.
Clustering and predicting sites on the basis of the laboratory biological data was surprising difficult.
The reasons were not fully known, but appeared to be related to the relative scales at which the
physical variables and the benthic profiles are measured. This was the primary reason that grouping of
cover classes was investigated, which did appear to produce some improvement. Partitioning groups
with the laboratory biological data led to very few groups, mostly determined by spatial predictors and
hence probably linking to a property of the physical environment for which the only surrogates we had
in our data set were the spatial ones.
The vessel data, by contrast, was much more amenable to modelling with the external predictors. As
will become apparent the number of groups was still smaller than would have been expected if they
were to represent all ‘habitats’ that would be commonly recognised in the GBR, suggesting again, that
they have potential to form groups at quite a coarse scale, but perhaps useful at least for management
purposes.
GBR Seabed Biodiversity
2-68
2.4.6. Acoustics Discrimination and Classification
2.4.6.1. Application of Wavelet Packet-Based Feature Extraction Techniques to Acoustic
Data in the Angular Domain (D H Smith)
2.4.6.1.1
Data and data pre-processing
Acoustic data was acquired as described in Section 2.2.4 with ground truth data provided by the
underwater DropCam video system (Section 2.2.2) available from about 1,000 sites. The acoustic data
constitutes an echo from the seabed, resulting from the transmission and reflection of an acoustic pulse
generated by the hull-mounted transducer. In addition to depth, the measured return signal contains
information about the seabed, to be determined by applicable data inversion techniques.
Data from the sonar transducer occurs in two forms, surface referenced and bottom referenced,
associated with different temporal and spatial measurement intervals and resolution. The former case
includes the entire water column while the latter case involves a small interval in the seabed vicinity.
In order to partially remove bathymetry effects, the original data is transformed from the time domain
into the angular domain (Sternlicht and de Moustier, 2003), prior to feature extraction and
classification operations, based on a local tangent (flat seabed) approximation, which ignores the
effects of vessel pitch and roll motion. This allows a fixed angular domain, from zero degrees or
normal incidence, up to a specified limit determined by the transducer properties, on which all data is
compared. Figure 2-39 shows a single data sample, including the bottom and surface referenced
values, in their original time/distance domain, and after transformation to the angular domain.
0
10
Bottom Referenced
Surface Referenced
−2
Signal Strength
10
−4
10
−6
10
−8
10
−10
10
10
20
30
40
50
60
Distance in Metres
70
80
90
100
0.016
Signal Strength
0.014
0.012
0.01
0.008
0.006
0.004
0.002
0
0
5
10
15
20
25
Angle in Degrees
Figure 2-39: An acoustic data sample from site 1505, with indicated depth of 23.97 m, shown in the original
distance/time domain, and after transformation to the angular domain.
GBR Seabed Biodiversity
2-69
For each site, the drop camera towed behind the boat recorded video data (Section 2.2.2) for a patch of
seabed that was scanned by the sonar transducer roughly half a minute earlier. Calculation of the delay
is required in order to match seabed characterisation labels, as input by a keyboard operator viewing
the video onboard the vessel (Section 2.2.2.1), with acoustic data for the subsequent application of
classification procedures. Figure 2-40 shows a typical pair of signals derived from the echo sounder
and drop camera, the latter in this case comprising just under 7.5 minutes of video time, indicating that
portion of the echo sounder signal matched to the drop camera signal. Once the delay was calculated
for each transect, the seabed habitat classes specified by two attributes, substratum and biohabitat (as
detailed in Section 2.2.2.1, Table 2-6) were used to label each matching echo return signal.
Application of the matching procedure on the available site data generated an extensive library of
single beam acoustic signatures for over 250 different (substratum, biohabitat) combinations.
Video
Sonar
Match
29
28
27
Signals
26
25
24
23
22
21
20
25
30
Time in Minutes
35
40
Figure 2-40: Measured depth and pressure signals derived from the drop camera and sonar transducer for site
1505, highlighting the segment of matched sonar data for which the calculated delay result is 0.502 minutes.
2.4.6.1.2
Techniques Applied
Feature extraction and classification are complementary numerical processes aimed at identifying the
source of a certain data sample, which in this case constitutes an acoustic echo from the seabed. For
supervised classification, training of a particular feature extraction scheme is performed on data of
known types from each of several possible seabed classes prior to application on data of unknown
type. In this section, feature extraction is performed via a wavelet packet-based technique known as
the Local Discriminant Basis (Saito and Coifman, 1995) in conjunction with Daubechies filter
coefficients (Cohen et al., 1993). This approach involves selecting a data transform from a very large
library of candidate wavelet packet transforms (Jensen and la Cour-Harbo, 2001), in order to provide
maximum discrimination between a set of data items representing different classes. A demonstration
of this discrimination capacity for the two class case comprising (sand, no biohabitat) and (sand,
GBR Seabed Biodiversity
2-70
seagrass) seabed types is given graphically by the three dimensional plot in Figure 2-41, which
displays the first three local discriminant basis coordinates plotted against each other. Distinct clouds
have emerged in this “feature space” view, showing a visible separation between classes afforded by
the Local Discriminant Basis.
Class 1: (Sand, None)
Class 2: (Sand,Seagrass)
0.8
0.6
3rd
0.4
0.2
0
0.2
0.4
0
0.5
1
0.6
1.5
2
0.8
2.5
3
1
2nd
3.5
4
1st
Figure 2-41: Three dimensional plot of the first three Local Discriminant Basis coordinates for data representing
(sand, no biohabitat) and (sand, seagrass) seabed types, generated with Daubechies 2 wavelet filter coefficients,
showing visible separation between the two classes.
Application of the chosen wavelet packet transform to data in the angular domain returns features
which are ranked by discrimination power, providing an automatic truncation facility allowing
dimension reduction to seek compact, efficient feature sets. Subsequent classification operations are
performed in the transformed domain, or “feature space”, utilising both Tree and Linear Discriminant
Analysis classifiers (Duda et al. 2001). By providing a relatively small number of significant features
to the classifier, the Local Discriminant Basis offers performance enhancement, in comparison to that
derived from direct application on the original data. This feature extraction and classification approach
constitutes one particular avenue in the performance assessment of single beam acoustic remote
sensing technology.
2.4.6.2. Canonical Variate Analysis of Acoustic Data (N Campbell & D Devereux)
This section outlines methods used to analyse the large volume of single-beam sonar echo timeresponse curve data, collected as described in Section 2.2.4 from a range of seabed habitat cover types,
to determine the degree of discrimination between different habitats. The underlying assumption is
that different types of seabed exhibit reflected acoustic signals with characteristic shapes. Appropriate
mathematical descriptions of these shapes and the use of discrimination-based statistical procedures
are used to assess the separability of a range of cover classes. The 700 samples from the surface to
GBR Seabed Biodiversity
2-71
beyond the 2nd echo are here called the pelagic data, and the 500 samples from 5 m above the bottom
pick to 10 m below the bottom pick are called the bottom data.
The emphasis initially was on the complete volume of data, to assess the degree to which particular
cover types could be consistently and accurately identified over the whole of the GBR. Some analyses
of the degree of discrimination among cover classes from sites collected close together in time and
hence geographically were also carried out.
The main data analysed here are for all groups of more than 100 contiguous echo responses from the
same nominal class as identified from the ground-truth video data labels matched as per Section
2.4.6.1.1. There were 4519 such groups from 117 classes. Because of the sheer volume of data, the
individual group means and pooled within-groups covariance matrix were calculated, after appropriate
data normalisation.
2.4.6.2.1
Depth Normalisation
Prior to analysis, the acoustic data were depth-normalised to a constant depth so that signatures could
be compared over the whole range of the data. This was done by taking the depth pick selection
provided by the Simrad EY500 system, depthR, and resampling the echo time series to a constant set of
depths. Specifically, the estimated depth in metres for a given sample time is calculated as:
deptht = (ts – 0.5 ) { ( re – rb ) / ns } + rb
(1)
where ts denotes the sample time; rb is the beginning range of the echo sounder (that is, the echo
sounder is set to commence sampling at the moment when an echo from depth rb would be
received); re is the end range of the echo sounder (that is, the echo sounder is set to cease sampling
at the moment when an echo from depth re would be received); and ns is the number of samples
collected.
The data are resampled by calculating the estimated depths for a range of sample times from 0 to 500,
which give a depth of 0 metres at time 0, and a depth of depthR after 200 sample times, then using (1)
to calculate the corresponding sample time on the observed scale, and using nearest neighbour or local
smoothing such as bilinear or cubic to calculate the resampled values.
2.4.6.2.2
Data Normalisation
The profiles can be peak-aligned to remove one effect of depth.
Another correction considered is to remove the so-called “size” effect, and focus on differences in
shape. This can be done by calculating a row mean (a simple measure of the average area under the
curve), and subtracting this mean from all the values across an echo response curve.
Plots of the group means in Section 3.6.2.2 show that there was an obvious effect of depth on the
shape of the group means; an attempt to remove this effect is made by regressing the echo response
values against 1/depth.
2.4.6.2.3
Group Discrimination
An obvious approach to examine the discrimination between various seabed cover types is to base the
analyses on a number of training groups, each one reasonably homogeneous, and representing a
particular cover class. A canonical variate analysis (CVA) can be carried out to examine the degree
and nature of the differences between the groups. The means and variances / correlations are
calculated for each group. The group means are used to calculate a "between" matrix, B, and the
variances / correlations to form a "within" matrix, W. Canonical vectors, c, are then calculated which
maximise the ratio of the between-groups to within-groups sums of squares, f = ct B c / ct W c, for the
canonical variate scores, ct x; f is referred to as the canonical root.
GBR Seabed Biodiversity
2-72
The eigen-equation for the canonical variate analysis is (B – f W) c = 0; the canonical vectors are
scaled so that ct W c = nw, the within-groups degrees of freedom. That is, the average within-groups
variance is one. Successive canonical vectors are chosen so that the corresponding canonical variate
scores are uncorrelated with previously-determined sets of scores.
The only assumption made in this approach is that each group of data is reasonably homogeneous (and
that the corresponding histograms are roughly symmetric). While each training group is representative
of a known cover type, this is not used explicitly in the analysis. In essence, a supervised clustering of
the group means is being carried out, with the within covariance matrix providing the metric against
which to judge similarities and differences.
A desirable feature of the CVA approach is that it produces linear combinations of the input variables
which can then be displayed and analysed in a lower number of dimensions. The analyses can be
carried out on the actual echo time response data (after the data have been suitably depth normalised
and averaged), or on features derived from the input curves.
2.4.6.2.4
Directed Class Discrimination
An obvious approach to examine the discrimination between various seabed cover classes is to base
the analyses on groups from each cover classes. Pool the groups into a super-group for a class, but use
the usual within-groups covariance matrix.
2.4.6.2.5
Site Contrast CV Analyses for Depth-Normalised Data
The data analysed here are from sites which were collected close together in time and space,
concentrating on potential extremes of cover, such as sand and seagrass, and mud, silt and sand.
The sites included the following:
Site#
Date
Depth m Substratum Biohabitat
1631 23/11/03
~37
Sand 100% Seagrass 98%
2552 23/11/03
~30
Sand 100% Seagrass 0%
2441 24/11/03
~28
Sand 100% Seagrass 98%
1580 24/11/03
~58
Sand 100% Seagrass 0%
2224 23/09/04
~49
Sand 100% Seagrass 0%
For each site, sequences of contiguous echo responses from the same class were formed as groups, and
these groups were then ordered by class.
2.4.6.2.6
Echo Response Curves vs Features
A common approach for the analysis of single-beam echo data is to transform the echo data (after
suitable depth normalisation and averaging) to a set of derived parameters or features such as
quantiles, amplitudes, power spectrum coefficients and wavelets. A principal component analysis
(PCA) is then carried out on these derived parameters, and the first few principal components (PCs)
are used in subsequent analyses (ref QTC Manual).
The degree of discrimination provided by the various derived parameters and by the echo time
response data was examined for a few groups. The stacked-averages-of-5-pings data that QTC
produces as part of its routine processing provide the “raw” data for the data sets discussed in this
section. The derived feature parameters are those summarising the echo envelope proposed by
Tegowski and colleagues, and 166 parameters capturing the shape and spectral character of the echo
produced by QTC. The degree of separation provided by the echo envelope and QTC shape
parameters and by the averaged ping time response data are compared.
GBR Seabed Biodiversity
2-73
2.4.6.3. Linear Discriminant Analyses of QTC View data (I McLeod)
As described in Section 2.2.2.1, acoustic data was acquired during a series of cruises. In addition to the
raw digital acoustic data collected directly from the EY500 echo sounder, pre-processed data was also
recorded from a QTC View IV acoustic processor (Questar Tangent Corporation). This device
digitised the acoustic responses from the transducer and processed the wave form. After stacking up
five pings envelopes, the QTC processed the raw data and transmitted 166 extracted features to the
logging PC, where they were recorded as each observation, interspersed with raw GPS NMEA strings
as they were transmitted. Thus, the 166 QTC parameter data string was merged sequentially with the
raw GPS strings. Due to system settings, QTC data from areas deeper than 80 m was spurious and data
from these areas was excluded from analysis.
In the laboratory, the interspersed data was post-processed into a database record format of GPS
date/time/position stamped QTC parameters, which were then joined to the seabed type codes derived
from the real-time camera observations of substratum and biological habitat (Events). As noted in
Sections 2.2.2 and 2.4.6.1, there was a lag from the QTC data to the habitat code data. As the
transducer was mounted to the underside of the hull and facing directly downwards, the acoustic
signals processed by the QTC system were reflected by the seabed lying directly under the boat. In
contrast, the habitat events data came from the camera sled, which was towed behind the boat but time
stamped with the boat’s time and position. There was approximately a 30 second delay between the
seabed that the acoustics sampled and when that seabed was observed by the towed video and habitat
events recorded. The habitat events data was lagged, by a time delay determined separately for each
site as described in Section 0, to join it to the QTC data (e.g. we matched events from 10:25:17 with
QTC from 10:24:47).
The habitat events data was available only from sites where video transects were conducted, whereas
the QTC data was collected ~continuously along the vessel's track — thus, QTC data was also
available for tracks between sites. Only a small fraction of the QTC data coincided with where the
"ground-truthed" habitat events were available. After completing the merge, there were 141,032
ground-truthed observations in the training set. From the habitat events data we had observations of
twenty four classes of biological habitat and nine classes of substratum (Table 2-6). The combination
of these two events data classes gave a potential 216 classes, not all of which were represented in the
data. The intent was to use the merged data as a training set from which a classifier could be
developed, which would be applied to all the along track QTC data to produce a more extensive
habitat map predicted from classified acoustic track data.
It was anticipated from the outset that it would not be possible to reliably discriminate all observed
combinations and that considerable aggregation of the habitat events classes would be required. Our
previous experience had indicated that about 4–5 seabed types might be distinguishable with about
60% success (e.g. Skewes et al. 1996; Long et al. 1997, McLeod et al. 2007). Nevertheless, significant
improvement on 4–5 seabed types was expected, due to the greater number and detail of acoustics
features extracted in this application.
In a series of analyses, beginning with the most detailed comprising all available classes (24 biohabitat
by 9 substratum), classification performance was used to guide aggregation of ecologically similar
habitat classes. Several aggregations of the biohabitat and substratum were trialled, including
analysing the biological habitats and substratum types separately as well as combined. In each analysis
it was necessary to remap the original classes into a new aggregated class schema. These remapping
tables are presented below (Table 2-10 to Table 2-13), and the outcomes of selected analyses in
section 3.6.3. Further, classification analyses were trialled with and without depth partitioning, as it
was clear that the acoustic feature data were not independent of depth.
The statistical method applied was linear discriminant analysis, and classification performance was
assessed by cross-validated classification error rates.
GBR Seabed Biodiversity
2-74
Table 2-10. Habitat Events re-coding table showing mapping from the original BioHabitat code to
Habitat_Code2.
Habitat_Code
0
6
3
11
20
25
Habitat_Desc
No BioHabitat
Bivalve Shell Beds
Alcyonarians Sparse
Gorgonian Garden Sparse
Sponge Garden Sparse
Whip Garden Sparse
Habitat_Code2
0
Habitat_Des2
No BioHabitat
1
Sparse garden
1
2
9
10
23
24
Alcyonarians Dense
Alcyonarians Medium
Gorgonian Garden Dense
Gorgonian Garden Medium
Whip Garden Dense
Whip Garden Medium
2
Alcyonarians
3
Gorgonian
18
19
4
8
7
12
17
5
13
16
14
15
Sponge Garden Dense
Sponge Garden Medium
Algae
Flora
Caulerpa
Halimeda
Seagrass
Bioturbated
Hard Coral Garden Dense
Live Reef Corals
Hard Coral Garden Medium
Hard Coral Garden Sparse
4
Sponge
5
Algae
6
7
8
9
10
Caulerpa
Halimeda
Seagrass
Bioturbated
Coral Dense
11
Coral Sparse
Table 2-11. Habitat Events re-coding table showing mapping from the original BioHabitat code to
Habitat_Code3.
Habitat_Code
0
25
Habitat_Desc
No BioHabitat
Whip Garden Sparse
1
9
18
Alcyonarians Dense
Gorgonian Garden Dense
Sponge Garden Dense
2
10
19
23
Alcyonarians Medium
Gorgonian Garden Medium
Sponge Garden Medium
Whip Garden Dense
3
11
20
24
Alcyonarians Sparse
Gorgonian Garden Sparse
Sponge Garden Sparse
Whip Garden Medium
4
7
8
12
17
Algae
Caulerpa
Flora
Halimeda
Seagrass
5
6
13
16
14
15
Bioturbated
Bivalve Shell Beds
Hard Coral Garden Dense
Live Reef Corals
Hard Coral Garden Medium
Hard Coral Garden Sparse
Habitat_Code3
0
Habitat_Des3
No BioHabitat
1
Soft – Dense
2
Soft - Medium
3
Soft – Sparse
4
Algae and Seagrass
5
Bioturbated
6
Coral - Dense
7
Coral – Sparse
GBR Seabed Biodiversity
2-75
Table 2-12. Substratum Events re-coding table showing mapping from the original Substratum code to
Substratum_Code2
Substratum_Code
1
2
3
4
5
6
7
8
9
Substratum_Desc
Bedrock / Reef
Rocks (> 250 Mm)
Stones (50-250 Mm)
Rubble (5-50 Mm)
Substratum_Code2
1
2
3
4
Coarse Sand
Fine Sand
Sand Waves
Silt
Mud
Substratum_Des2
Reef
Boulders
Cobbles
Gravel
5
Sand
6
7
Silt
Mud
Table 2-13. Substratum Events re-coding table showing mapping from the original Substratum code to
Substratum_Code3.
Substratum_Code
1
2
3
4
5
6
8
9
7
Substratum_Desc
Bedrock / Reef
Rocks (> 250 mm)
Stones (50-250 mm)
Rubble (5-50 mm)
Coarse Sand
Fine Sand
Silt
Mud
Sand Waves
Substratum_Code3
1
2
3
4
Substratum_Des3
Reef
Boulders
Cobbles
Gravel
5
Sand Silt
6
7
Mud
Sand Waves
2.4.7. Ecological Risk Indicators
A progressive series of indicators of exposure to trawling have been estimated for habitat types,
seabed assemblages (predicted sites groups), species-groups and selected individual species. This
series includes:
1. estimates of the percentage by area of the distribution of each habitat, assemblage, speciesgroup and individual species, located in areas open to trawling under zoning or other
management — without account of the distribution of trawl effort.
2. estimates of the percentage by area of the distribution of each habitat, assemblage, speciesgroup and individual species, located in areas where trawl effort is present — without account
of the intensity of trawl effort.
3. estimates of the percentage by area of the distribution of each habitat, assemblage, speciesgroup and individual species, located in areas where trawl effort is present taking account of
the intensity of trawl effort.
For species-groups and selected individual species, predicted biomass distributions have been
estimated (Sections 2.4.2 and 2.4.3) and biomass related indices can be estimated.
4. estimates of the percentage of biomass of the distribution of each species-group and individual
species, located in areas open to trawling under zoning or other management — without
account of the distribution of trawl effort.
5. estimates of the percentage of biomass of the distribution of each species-group and individual
species, located in areas where trawl effort is present — without account of the intensity of
trawl effort.
GBR Seabed Biodiversity
2-76
6. estimates of the percentage of biomass of the distribution of each species-group and individual
species, located in areas where trawl effort is present taking account of the intensity of trawl
effort.
The intensity of trawl effort was taken into account as a coverage of the study’s 0.01 degree grid cells
as if trawling were conducted uniformly at that fine scale. Given the typical swept width of gear and
speed of trawling for prawns in the region, approximately 8 hours of trawling would be required to
have a 1× coverage of a 0.01 degree grid cell. Similarly, 4 hours would have a 0.5× coverage and 16
hours would have a 2× coverage. A given grid cell’s contribution to the overall index was the
estimated proportion by area or biomass of the respective biological attribute, multiplied by the
estimated effort coverage; these estimates for grid cells were summed to provide the overall index for
the GBR region. The effort intensity information for each grid cell was provided by spatial processing
of the 2005 fishery Vessel Monitoring System (VMS) data by Norm Good, QDPI&F (Gribble et al.
2007). The year 2005 was the first full year post the re-zoning of the GBR Marine Park, which came
into effect on 1 July 2004.
Exposure to trawl effort may present varying levels of risk for different species depending on how
effectively the trawl net catches any given species (catchability), or how much mortality is caused as a
result of contact with the net. For example, a species that lives well down in the sediment, or one that
moves up into the water column during the night, is unlikely to be directly affected by the pass of a
trawl net. On the other hand, a slow moving species that lives less a metre from the seabed may be
very effectively caught by a prawn trawl net. For species estimated to have higher levels of exposure,
information on relative catchability was sought wherever possible. This study was able to provide
relative catch rate information between the research trawl and the epibenthic sled (Section 2.4.2).
Wherever the sampling rate of the research trawl was less than the epibenthic sled, the prawn trawl
was considered to catch that fraction (0-1) of the population present during a pass of the net. Wherever
the trawl had a higher catch rate, the prawn trawl was considered to a relative catch rate of 1. Similar
information was available from the GBR Effects of Trawling Study (Poiner et al., 1998), including
prawn trawl catch rates relative to those of a fish trawl. Information on the possible impact on species
remaining on the seabed was available from the GBR Effects of Trawling Recovery Dynamics Project
(Pitcher et al., 2004) and the Northern Prawn Fishery Effects of Trawling Project (Haywood et al.,
2005). If evidence was available that could demonstrate that catchability or mortality was <1, this
information could reduce the estimated percentage of the biomass of a species exposed in indicator 6
above — i.e. an estimate of the proportion of the total population caught.
Further, the Queensland trawl fishery is required by legislation to have turtle excluder devices (TEDs)
and bycatch reduction devices (BRDs) installed in their nets. TEDs are very effective in allowing
larger animals such as turtles, rays and sharks to pass out of the trawl net, greatly reducing mortality
(Robins-Troeger et al. 1995; Robins & McGilvray 1999). BRDs provide escape opportunities for
smaller fish and reduce the catch rate of non-target species (bycatch) by varying amounts depending
on the species — over all species and different sectors of the industry, the average reduction achieved
by BRDs is about 8% (Courtney et al. FRDC 2000/170 Report 2006), though greater reductions are
possible (Courtney et al. 2006). If evidence (from these or other sources, e.g. Brewer et al. 1998) was
available that demonstrated that TEDs and/or BRDs further reduce catchability or mortality, this
information further reduced the estimated percentage of the biomass of a species exposed in indicator
6 above — again, reducing the estimate of the proportion of the total population caught.
Exposure to trawling, leading to estimates of the potential proportion of species populations caught
annually is only one axis of vulnerability to trawling. The second axis is the ability of the species to
recover from any reductions in population size. A species with a high recovery rate can sustain higher
levels of incidental catch than a species with a low recovery rate. Previous ecological risk assessment
methods that take this axis into consideration include susceptibility-recovery analysis (SRA) — a
qualitative ranking approach (Stobutzki et al. 2001, see also Griffiths et al. 2006) — and sustainability
assessment for fishing effects (SAFE) — a quantitative approach where estimated fishing mortality is
compared against reference points of estimated natural mortality (e.g. 0.8M = maximum sustainable
mortality) (Brewer et al. 2007; Zhou & Griffiths 2007). For species estimated herein to have higher
exposure to trawling, information about the recovery axis was obtained, where available, for mean
recovery attribute ranks from SRA analyses in northern Australia (e.g. additive ranks of probability of
breeding before capture, maximum size, reproductive strategy, hermaphrodism for fishes (Stobutzki et
GBR Seabed Biodiversity
2-77
al. 2001) and invertebrates (Hill et al. 2002)) and for estimated natural mortality to calculate a
sustainability indicator in a manner analogous to the reference points of Zhou & Griffiths (2007).
These reference points were based on the Schaffer surplus production model, where for a population at
maximum sustainable yield (MSY), fishing mortality (F) is equal to natural mortality (M), that is
F/M=1. This is regarded as a limit reference point and should not be exceeded. Zhou & Griffiths
(2007) consider reviews of exploited species that suggest F=0.8M (≡ F/M=0.8) is a more conservation
reference point, and Gulland (1983) suggested a conservative MSY of 0.3MB0 in data limited
situations, which (as BMSY = 0.5B0) corresponds to F/M=0.6. These three reference points are
considered in this report. Where exploitation is low, F is approximately equal to exploitation — the
estimated proportion of the total population caught — thus, the indicator calculated herein is
exploitation divided by natural mortality. Note that this method is only a ‘discrete time’
approximation, it is not an ‘instantaneous time’ stock assessment, and becomes increasingly uncertain
with higher levels of exploitation and/or natural mortality (Hilborn & Walters 1992).
A further indicator of potential ecological risk is available from the biophysical modelling (Section
2.4.2). If the trawl effort covariate was selected by the statistical modelling of any species, possibly in
addition to other environmental variables such as sediment type, the sign and significance of the
coefficient was examined. Further, in order to examine the regional implications of an included trawl
effort term, a prediction of the biomass was made with the trawl covariate set to zero throughout the
region. This estimate was then contrasted with the biomass prediction for the actual current situation to
estimate how much smaller, or larger (in the case of positive effects), the species population may be as
a result of incidental trawl catches over the history of the fishery. It is important to note here that many
of the physical covariates are correlated and it may not be possible to interpret model coefficients as
causal effects. In the case of target species in particular, a positive trawl coefficient could indicate that
the effort data — reflecting the searching ability of the fishers — are a good indicator of the sampled
distribution, of say commercial prawns, at a scale finer than that of spatial patterns in the physical
environmental data, rather than indicating that prawns are more abundant because of trawling.
2.4.8. Trawl Management Scenario Model (N Ellis, A Welna, R Pitcher)
A dynamic model was applied to assess the effects of several major management interventions, which
were implemented between the years 2000 and 2006, on benthic fauna — particularly sessile benthic
fauna that were the focus of experiments on trawl depletion rates (Poiner et al. 1998, Burridge et al.
2003) and subsequent recovery (Pitcher et al. 2004). The management interventions included two
large-scale closures comprising the 2000/2001 low-effort areas closure and the 2004 representative
areas program (RAP) re-zoning of the GBR; two effort reductions comprising a major buy-back
effective in 2001 and another RAP-associated buy-back effective in 2005; and a progressive penalty
system operating between the latter two. The dynamic model applied depletion and recovery
parameters estimated from previous experiments and annual trawl effort as provided by industry and
management data to estimate the relative status of fauna in model grid cells of 6 minute resolution.
The model was run with and without the effects on effort of each management intervention. The
relative status estimates were combined with the abundance distributions available from the current
project in order to estimate the regional absolute status of these fauna.
2.4.8.1. Specification of the management scenarios
We wanted to test the effect of the management interventions that had been applied over the period
2001–2005. We took as a baseline the ‘status quo’ scenario (SQ2001) from the view-point of the pre2001 fishery; i.e. we projected from 2001 to 2025 assuming the fishing effort remained at year-2000
levels. We then constructed a sequence of scenarios in which each intervention was included
progressively: the 2001 closure (CL2001); the 2001 buyback (BB2001); penalties on trading effort
units (P); the 2004 RAP closure (RAP); and the 2005 buyback (BB2005).
GBR Seabed Biodiversity
2-78
Over the period 2001–2005, in reality, some or all of these interventions were in effect simultaneously.
We therefore had to make some assumptions about what the separate effects of these interventions
were. Table 2-14 shows the total effort in various regions of the East Coast Trawl Fishery for the
period 2001–2005. Figure 2-42 shows the total effort in the study area over the period 1993–2005. The
reduction in 2001 is evident, followed by a more gradual reduction in subsequent years. The drop in
2005 is attributable to both the 2005 buyback and the continuing penalties. Also, there is some
variation due to other (unknown) causes.
Table 2-14. Effort (boat days) in various regions of the East Coast Trawl Fishery, 2001–2005.
Year
2001
2002
2003
2004
2005
Inside
GBRMP
46,107
47,978
44,753
40,162
34,562
Outside
GBRMP
23,395
19,808
20,363
23,447
21,813
Total
69,502
67,786
65,116
63,609
58,380
Study
Area
35,531
35,387
29,465
28,236
24,621
In restricting the modelling to the study area, which is a subregion of the GBRMP, we have assumed
that management interventions operating at the level of the entire fishery (i.e. effort capping) can be
applied proportionally to the study area. This assumption is reasonable given the common trends in the
study area, the GBRMP and the rest of the fishery (see Table 2-14). Large-scale relocation of the fleet
into or out of the study area caused by influences unrelated to area closures are not accounted for in
this analysis.
By modelling the effect of penalties as a constant annual proportional reduction, and the 2005 buyback
as a 6% reduction (Andrew Thwaites, QDPI, pers. comm.), we estimated the penalty reduction using a
linear model:
log(effort) = a + b × (year–2001) + buyback(year) + error,
where
buyback(year) = log(1–0.06)
if year = 2005
0
otherwise.
The error term is a normal variate with mean zero and standard deviation s. The estimate of b was
-0.084, indicating an annual 8% reduction due to penalties, and the estimate of s was 0.043. This
model was then used as the mean projected effort. The coefficient of variation of annual effort was
taken to be the same as in the period 1993–2000. Such variation is possible in this fishery because the
fleet does not usually fill its allocation; for example in 2005 only 66% of the allocation was used.
Figure 2-42 shows the projected mean effort for the 4 effort reduction scenarios. The effort for the
CL2001 scenario is the same as for the SQ2001 scenario; the effort for the RAP scenario is the same
as for the penalties scenario; and the projected effort (from 2006) for the SQ2006 scenario is the same
as for the BB2005 scenario.
2.4.8.2. The trawl depletion-recovery model
The dynamic biomass model is a set of Schaefer-like models operating independently in each 0.1°
spatial cell:
dBsx/dt = rsBsx(t)(1 – Bsx(t)/Ksx) – dsEx(t)Bsx(t)
where Bsx(t) is the biomass at time t of benthic species (or OTU) s in cell x, Ksx is the carrying capacity
of species s in cell x, Ex(t) is the effort rate at time t in units of swept area per unit time, rs is the
recovery rate of species s, and ds is the depletion rate per tow of species s. The equation simplifies to:
dbsx/dt = rsbsx(t)(1 – bsx(t)) – dsEx(t)bsx(t)
Bsx = Ksxbsx
GBR Seabed Biodiversity
2-79
where bsx(t) is the relative biomass. This has the practical consequence that the biomass distribution
can be split into two components, one, bsx, depending only on the vulnerability pair (rs, ds) and the
other, Ks, depending only on survey data. Each component can then be computed independently and
combined later. To provide an initial condition for bsx, we assume the pre-fishery biomass (at time t0)
in each cell was at the carrying capacity, i.e. bsx(t0) = 1.
70
projected
historical
Effort ('000s boat days)
60
50
40
30
20
Effort data
SQ2001
and with BB2001
and with penalties
and with BB2005
1995
2000
2005
2010
Year
Figure 2-42. Total effort in the study area for the period 1993–2005. Also shown is the projected mean effort for
4 scenarios.
2.4.8.3. Specification of the depletion-recovery parameters
Pitcher et al. (2004) obtained parameters describing the depletion and recovery dynamics of a set of
benthic taxa following a previous trawl depletion experiment in the Far Northern section of the Great
Barrier Reef. They traced over time, using both a video sled and a remotely operated vehicle (ROV),
the abundance of several OTUs within differentially impacted transects. Their generalized linear
models took the following form:
log(bit) = log(b0) + ci×i + cit×i×t + citt×i×t2 + ctt×t2,
(1)
where i is the number of trawl tows (impact), t is the time in years after impact, bit is the biomass at
time t after impact i, and b0 is the initial biomass. The values for the four model parameters are shown
in Table 2-16 and Table 2-15. Figure 2-43 shows the form of these models for a range of times and
impacts for the sled-based parameters.
It is evident from (1) and Figure 2-43 that these models do not follow a simple logistic form. In
particular, the models are not bound above by an asymptote (e.g. Alcyonacea), and the functions can
decrease in time after an initial increase (e.g. Sarcophyton sp). In order to allow these species to be
modelled by the trawl model, we fit the predicted form of (1) to a logistic
bit / b0 ~ expit(rt + logit[(1 – d)i ]),
where
logit(x) ≡ log x/(1 – x)
and
expit(x) ≡ ex/(1 + ex)
mutual inverses. This form has the property that bi0 / b0 = (1 – d)i and that bit → b0 as t → ∞.
(2)
GBR Seabed Biodiversity
2-80
For parameterizing the trawl model, only the initial recovery part of the Pitcher et al. model (1) was
deemed relevant. We therefore fit this over a range of t from 0 to 5 years and i from 1 to 10 using
weighted nonlinear least squares. Parts of the function that were decreasing in time or above the
asymptote were down-weighted in the fitting. The weight we used was
wit = W(bit < b0, 1, 0) W(dbit/dt > 0, 1, 0.001) i–1
where the function W(x, a, b) takes value a if x is true, otherwise b. As an example, Figure 2-44 shows
the fits for Ianthella flabelliformis and Junceella juncea.
Table 2-15. Parameters ci, ci, ct, citt and ctt for the ROV recovery data from Pitcher et al. (2004); and
corresponding estimates r and d fit by non-linear least squares. *For Subergorgia suberosa and Solenocaulon the
value r = 0.22 was used.
Taxa
Alcyonacea
Annella reticulata
Ascidiacea
Bebryce sp
Ctenocella
Cymbastella
Dichotella sp1
Echinogorgia
Ellisella sp
Hypodistoma deerratum
Ianthella basta
Ianthella flabelliformis
Iciligorgia sp1
Junceella juncea
Junceella sp2
Nephtheidae
Porifera
Sarcophyton sp
Scleractinia
Solenocaulon
Subergorgia sp
Subergorgia suberosa
Turbinaria
Xestospongia testudinaria
ci
0.0055
–0.0198
–0.0196
–0.0220
0.0181
–0.0050
–0.0068
–0.0100
–0.0185
–0.0427
–0.0117
–0.0205
–0.0107
–0.0056
–0.0072
–0.0140
–0.0299
–0.0291
–0.0454
–0.0095
–0.0671
–0.0235
–0.0722
–0.0747
ct
–0.00237
0.00031
0.00016
0.00020
–0.00301
–0.00038
–0.00012
0.00048
0.00022
0.00275
0.00006
0.00019
0.00050
0.00006
–0.00037
0.00024
0.00048
0.00002
–0.00177
–0.00110
0.00378
–0.00225
–0.00228
0.00365
citt
0.00001
–
–
–
0.00004
–
–
–
–
–0.00004
–
–
0.00001
–
–
–
–
–
–
–
–0.00005
–
–
–0.00005
ctt
0.00018
–
–
–
–
–
–
–
–
–
–
–
–0.00011
–
–
–
–
–
0.00020
0.00015
–
0.00019
0.00028
–
r
–
0.57
0.19
0.22
–
–
–
4.17
0.32
0.12
0.11
0.22
2.12
0.21
–
0.58
0.62
0.03
–0.41
–0.53
1.66
–0.62
–0.39
0.29
d
–
0.417
0.496
0.447
–
–
–
0.712
0.415
0.554
0.427
0.239
0.692
0.040
–
0.164
0.310
0.500
0.484
0.101
0.700
0.318
0.568
0.598
used?
–
yes
yes
yes
–
–
–
yes
yes
no
yes
no
yes
no
–
no
no
no
no
Yes*
yes
yes*
no
yes
The estimated parameters for all fits are shown in Table 2-16 and Table 2-15. When the main temporal
effect ct was negative, it was not usually not possible to obtain an estimate for r and d, since the two
models were so different. The exceptions were when a positive ctt term counteracted the main effect
(e.g. ROV data for Solenocaulon). One species, Junceella sp2, did not yield any estimates. All other
OTUs were estimated by one or the other of the sled and ROV data, and some by both. Figure 2-45
shows all the estimates. Where estimates were available from both sled and ROV we used the sled
estimate in preference. For Subergorgia suberosa and Solenocaulon the estimate of r was unreliable
(being negative). We used the median value of all other OTUs having slow recovery, which was 0.22.
Except for Hypodistoma deerratum, the estimates of d are fairly consistent between the two devices.
The estimates of r on the other hand are much more variable. While the two devices observed different
though overlapping populations, this is indicative of the precision with which the two parameters can
be estimated.
GBR Seabed Biodiversity
2-81
Table 2-16. Parameters ci, ci, ct, citt and ctt for the sled recovery data from Pitcher et al. (2004); and
corresponding estimates r and d fit by non-linear least squares.
Taxa
Alcyonacea
Annella reticulata
Ctenocella
Cymbastella
Dichotella sp1
Echinogorgia
Hypodistoma deerratum
Ianthella flabelliformis
Junceella juncea
Junceella sp2
Nephtheidae
Porifera
Sarcophyton sp
Scleractinia
Solenocaulon
Subergorgia suberosa
Turbinaria
Xestospongia testudinaria
ci
–0.096
–0.130
–0.103
–0.413
–0.080
0.046
–0.078
–0.230
–0.069
–0.059
–0.306
–0.148
–0.610
–0.433
0.013
0.628
–0.693
–0.096
ct
0.0143
–0.0112
0.0022
0.0210
0.0088
–0.0015
0.0056
0.0031
0.0015
–0.0020
0.0058
0.0027
0.0386
0.0294
–0.0007
–0.0170
0.0333
–0.0154
citt
–0.00009
0.00030
–
–0.00014
–
–
–
–
–
–
–
–
–0.00050
–0.00040
–
0.00038
–0.00029
0.00030
ctt
–0.0012
–
–
–0.0026
–0.0012
–
–
–
–
–
–
–
–
–
–
–0.0018
–0.0012
–
r
1.70
–
0.69
0.75
0.82
–
2.61
0.38
0.64
–
0.81
0.61
3.14
2.96
–
–
1.37
–
d
0.112
–
0.115
0.369
0.128
–
0.094
0.211
0.078
–
0.293
0.157
0.483
0.390
–
–
0.522
–
used?
yes
–
yes
yes
yes
–
yes
yes
yes
–
yes
yes
yes
yes
–
–
yes
–
Figure 2-43. Pitcher et al. (2004) models (points) and fitted Schaefer model response (lines) for 0 to 10 initial
trawl tows for two OTUs: (left) Ianthella flabelliformis and (right) Junceella juncea. The vertical scale is
biomass relative to initial unimpacted biomass. The horizontal scale is years since impact.
In addition to the fine taxonomic units from the Pitcher et al. (2004) study, we also used the coarser
taxonomic groupings reported by Poiner et al. (1998) with recovery rates obtained using the
categorical method of Hill et al. (2002). Table 2-17 summarizes the taxonomic units and their (r, d)
values.
GBR Seabed Biodiversity
Annella
reticulata
Alcyonacea
Ctenocella
i
b
i
t
b
b
i
i
Subergorgia
suberosa
Solenocaulon
b
b
t
i
t
Porifera
b
b
b
t
Scleractinia
i
Ianthella
flabelliformis
Nephtheidae
i
i
t
b
b
b
b
t
Sarcophyton
sp
t
t
Junceella
sp2
i
i
Hypodistoma
deerratum
i
t
Junceella
juncea
t
t
Echinogorgia
i
t
b
i
t
Dichotella
sp1
Cymbastella
b
b
b
i
t
2-82
t
i
t
i
Xestospongia
testudinaria
Turbinaria
b
b
t
i
t
i
Figure 2-44. Models obtained from sled video observations for 18 OTUs (Pitcher et al. 2004). The vertical (b)
axis represents biomass relative to initial unimpacted biomass, t is time since trawl impact (ranging from 0 to 5
years), and i is the number of trawl tows (ranging from 0 to 10).
4
GBR Seabed Biodiversity
2-83
Echinogorgia
ROV
Sled
Poiner/Hill
Hy podistoma deerratum
2
Iciligorgia sp1
Alcy onacea
Subergorgia sp
Turbinaria
1
Asteroidea
Nephtheidae
Cy mbastella
Annella
Porif era
reticulata
Gastropoda
Ellisella sp
Ianthella
Junceella
Bebry ce sp
Ascidiacea
f labellif ormis
juncea
Ianthella basta
Dichotella sp1
Ctenocella
0
Recovery rate (per year)
3
Sarcophy ton sp
Scleractinia
Solenocaulon
0.1
0.2
Xestospongia
testudinaria
Subergorgia suberosa
0.3
0.4
0.5
0.6
0.7
Depletion per tow
Figure 2-45. Recovery and depletion parameter estimates using sled (green) and ROV (blue) measurements.
Where both sled and ROV measurements are available the points are joined by a dashed line. Also shown (red
triangles) are the values from Poiner et al. (1998) and Hill et al. (2002) (not all labeled).
Table 2-17. Parameters r and d for coarse taxonomic groupings; d comes from Poiner et al. (1998) and r from
Hill et al. (2002).
Taxa
Asteroidea
Bivalvia
Bryozoa
Crinoidea
Crustacea
Echinoidea
Gastropoda
Gorgonacea
Holothuroidea
Hydrozoa
Ophiuroidea
r(year–1)
0.97
0.52
0.40
0.56
0.52
0.40
0.41
0.71
0.56
0.56
0.63
d
0.10
0.09
0.09
0.08
0.13
0.14
0.20
0.15
0.11
0.08
0.09
GBR Seabed Biodiversity
2-84
2.4.8.4. Specification of the pristine biomass model
In Sections 2.4.2 and 3.2, single-species (or single-OTU) models have been built and density
predictions laid on the fine scale spatial grid providing maps. The density predictions are on physical
environmental and spatial covariates, under optimal settings of temporal covariates (time of day, time
of year, phase of moon). They are also relative to the unknown catchability, which is assumed to be
constant.
These models can also be used to predict the pristine density over the area of the fishery. This is done
by predicting using the optimal covariable settings as before and setting the trawl effort predictor,
where selected, to zero. The result is then the predicted density taking into account the physical
covariates but in the absence of trawling, i.e. the pristine pre-fishery density. The prediction of pristine
density at a trawled site is based on the observed density at sites having similar physical attributes
away from trawled areas.
For OTUs having no trawl predictor in their model, the pristine biomass prediction is the same as the
current prediction of section 2.4.3. Such OTUs should either have little overlap with trawled areas (for
reasons of habitat independently of trawl distribution) or have high resilience to trawling due to high r
or low d or both.
Some OTUs had a negative trawl effort coefficient in the linear predictor of either the presenceabsence GLM or the biomass GLM. OTUs where these were either statistically significant or of large
effect were Carijoa sp1, Dendronephthya spp, Echinogorgia sp3 and sp5, Euplexaura sp6, Iciligorgia
sp1, Mopsella sp2 and several Demospongiae taxa. From the list of trawl exposed species in Section
3.7.2, Alertigorgia orientalis, Subergorgia suberosa and several Demospongiae (including Ircinia
1255 and Ircinia 2710) were identified.
For several of these species, specific r and d estimates were not available from the recovery study so
estimates were used from morphologically and taxonomically related species. The choices are listed in
section A2 of Table 2-18. Some taxa were identified only to the genus level. Where possible these
were linked to the (r, d) values of the same genus (Table 2-18, section B1); for Dendronephthya spp
the family Nephtheidae was used.
We also modelled the impact of trawling on coarse taxonomic groupings (family and higher) by
summing the available pristine biomass models of all taxa within each grouping, and using the (r, d)
values either from Pitcher et al. (2004) or from Poiner et al. (1998) and Hill et al. (2002) (Table 2-18,
section C). A GLM model of predicted distribution was not available for all OTUs; MSE results are
presented only for those that did.
GBR Seabed Biodiversity
2-85
Table 2-18. Relationship between modeled OTU and the source OTU for providing r and d at 3 levels of
taxonomic resolution: A) species, B) genus, and C) coarse (family or higher). The 3rd column indicates whether a
GLM model exists for the OTU.
OTU for r and d
Taxa
model?
A1. Taxonomy at species level: r and d measured for same species from Pitcher et al. (2004)
Annella sp2–6
Annella reticulata
no
Bebryce sp1
Bebryce sp
no
Dichotella sp1
Dichotella sp1
yes
Echinogorgia sp3
Echinogorgia
yes
Echinogorgia sp5
Echinogorgia
yes
Ellisella sp1–3
Ellisella sp
no
Ianthella basta
Ianthella basta
no
Ianthella flabelliformis
Ianthella flabelliformis
no
Iciligorgia sp1
Iciligorgia sp1
yes
Junceella juncea
Junceella juncea
yes
Subergorgia sp1–6
Subergorgia sp
no
Subergorgia suberosa
Subergorgia suberosa
yes
Xestospongia testudinaria
Xestospongia testudinaria
no
A2. Taxonomy at species level: r and d measured from different species from Pitcher et al. (2004)
Alertigorgia orientalis
Dichotella sp1
yes
Dichotella gemmacea
Dichotella sp1
yes
Carijoa sp1
Alcyonacea
yes
Euplexaura sp6
Annella reticulata
yes
Hippospongia elastica
Xestospongia testudinaria
yes
Ianthella quadrangulata
Ianthella flabelliformis
yes
Ircinia 1255
Xestospongia testudinaria
yes
Ircinia 2710
Xestospongia testudinaria
yes
Ircinia spp
Xestospongia testudinaria
yes
Junceella sp2
Junceella juncea
yes
Melithaea sp2
Annella reticulata
yes
Mopsella sp1
Annella reticulata
yes
Mopsella sp2
Annella reticulata
yes
B1. Taxonomy genus level: r and d measured at the same level from Pitcher et al. (2004)
Ctenocella
Ctenocella
no
Cymbastella
Cymbastella
no
Echinogorgia
Echinogorgia
yes
Solenocaulon
Solenocaulon
yes
Turbinaria
Turbinaria
yes
B2. Taxonomy genus level: r and d measured at different level from Pitcher et al. (2004)
Dendronephthya spp
Nephtheidae
yes
C1. Taxonomy coarse: r and d measured at the same level from Pitcher et al. (2004)
Ascidiacea
Ascidiacea
yes
Porifera
Porifera
yes
Alcyonacea
Alcyonacea
yes
Nephtheidae
Nephtheidae
yes
Scleractinia
Scleractinia
yes
C2. Taxonomy coarse: r from Hill et al. (2002) and d from Poiner et al. (1998) measured at the same level
Asteroidea
Asteroidea
yes
Bivalvia
Bivalvia
yes
Bryozoa
Bryozoa
yes
Crinoidea
Crinoidea
yes
Crustacea
Crustacea
yes
Echinoidea
Echinoidea
yes
Gastropoda
Gastropoda
yes
Gorgonacea
Gorgonacea
no
Holothuroidea
Holothuroidea
yes
Hydrozoa
Hydrozoa
yes
Ophiuroidea
Ophiuroidea
yes
GBR Seabed Biodiversity
3-86
3. RESULTS
3.1. BRUVS SPECIES MODELS, CHARACTERIZATION & PREDICTION
(M Cappo, G De’Ath)
The final BRUVS dataset comprised 39,989 individuals from 347 species of fishes, sharks, rays and
sea snakes observed at 366 sites. The bony fishes were from 10 orders, dominated by Perciformes (267
species), Tetraodontiformes (27), Anguilliformes (6), Aulopiformes (3), Scorpaeniformes,
Clupeiformes, Beryciformes with 2 species, and Siluriformes, Pleuronectiformes and
Gasterosteiformes each with a single species. The chondrichthyians were represented by
Carcharhiniformes (15 species), Rajiformes (13) and Orectolobiformes (3). There were 5 species of
sea snakes from the family Hydrophiidae.
3.1.1. BRUVS Species richness
Most of the 347 species recorded were rare or uncommon, occurring in only a very small percentage
of sites surveyed. There was an average of 13.8 ± 6 (s.d.) species per site, ranging from 2 to 43.
Ordering of the most diverse sites produced a sigmoid curve (Figure 3-1). Only ~14% of sites had
comparatively high species richness (≥20 species per site), ~41% had moderate richness (≥13 species),
and 18% had relatively low richness (≤8 species). Just over 90% of the species were recorded in less
than 10% of the sites and ~43% were recorded only 1–3 times (Figure 3-1). Only ~5% of the species
were moderately prevalent, occurring in ≥20% of the sites and, of these, only Nemipterus furcosus had
a prevalence of >50%. General patterns in species richness by latitude and longitude showed that
cross-shelf and long-shore gradients were not simple (Figure 3-2). Higher richness occurred at sites in
the outer reef matrix, particularly north of Proserpine (20.4°S), with a “hotspot” off Cape Flattery
(15°S) in the far north. Richness in the southern half of the GBRMP was higher around the CapricornBunker (23.5°S) island group, and consistently lower for the coastal bays, the deep mid-shelf waters of
the Capricorn trough (≥22.5°S), and the inter-reef waters of the outer barrier reefs (Figure 3-2).
50
Percentage of sites
Richness (species per site)
40
30
20
10
40
30
20
10
0
0
20
40
60
80
Sites rank-ordered by richness
(% of 366 sites)
100
0
20
40
60
80
100
Species rank-ordered by abundance
(% of 347 species)
Figure 3-1: Patterns of species prevalence and richness at BRUVS stations.
GBR Seabed Biodiversity
# ##
#
##
##
##
#
##
#
##
#
#
##
#
##
## ##
### # #
# #
## # #
#
##
#
#
##
#
## ##
### # #
# ###
#
E
#
## #
##
Kilometers
0
100
200
##
##
#
#
##
#
#
#
##
#
###
## # # #
# ##
#
#
#
##
#
#
###
#
# ##
#
# #
#
#
# ##
#
#
##
# #
#
# # #
#
# # #
# #
#
# ##
##
##
#
#
###
#
##
#
##
##
## ##
## #
#
S
#
##
# #
#
##
#
#
#
#
##
##
##
#
N
W
#
# #
#
#
#
# #
100
BRUVS Species Richness
#
1-7
#
8-9
#
10 - 12
#
13 - 14
#
15 - 16
#
17 - 18
#
19 - 22
#
23 - 43
#
##
## #
##
#
#
#
##
#
#
##
##
# #
#
#
#
#
#
#
# #
## #
#
##
#
#
#
# ##
# ## #
#
# #
# # ##
#
# ## #
#
#
#
# ##
# # #
###
#
# #
# #
# # #
# #
##
#
#
#
# #
#
#
##
# #
##
#
##
#
##
#
##
#
3-87
# #
## ####
##
#
# ### # #
#
#
#
#
#
#
##
# # ##
# #
#
## #
#
#
# ##
# # ## #
##
#
##
##
#
#
# #
## #
#
#
##
#
# # ##
# ##
## #
#
# #
#
## #
#
#
#
##
#
# #
# #
##
# # #
#
Figure 3-2: Species richness from BRUVS data by location in the GBRMP.
3.1.2. BRUVS Species presence/absence Biophysical Models and Prediction
A number of scenarios were run in univariate models using boosted trees.
First, the top 50 species that occurred on at least 7% (n=18) of all BRUVS sites were analysed using
(a) all 40 environmental variables, (b) using only spatial (Along, Across, Depth) variables, (c) using
only environmental variables, and (d) using all spatial and environmental variables, but dropping
“nuisance” temporal harmonics.
It was found that dropping the temporal harmonics had little effect on the models, implying that
predictive models would not have to adjust the presence/absence of species by the season, moon phase
or time of day of sampling. The top 25 species, with a predictability of ≥80%, were selected from this
analysis. Using “yres” to represent the “predictability” of each of the 25 species (yres= (1 %prediction error)), showed “all variables” (mean yres=15.15) not to be different from “all variables,
no harmonics” (mean yres=15.16).
The best 20 explanatory spatial and environmental variables (Table 3-2) were analysed with the most
predictable 25 species (Table 3-1). Models were produced to apply to the entire sampling grid, in order
to make biophysical maps of species occurrence throughout the GBRMP. The mean variance in the
species responses explained by these predictive models was 79.3%.
An ideal way to visualize the relationships amongst predictors, amongst species responses, and
between response and predictors is to plot them together on a “heatmap” (Figure 3-3). The percentage
of the variation in occurrence of each species explained by each predictor is shown in Table 3-1.
GBR Seabed Biodiversity
3-88
Table 3-1: Twenty-five most predictable species (y) using best 20 explanatory variables. "yres" = (1 %prediction error). “%Var” is the percentage of the variation in presence/absence of the species explained by the
best gbmmv model, for production of biophysical maps.
Species code
Sco.quee
Ser.nigr
Nem.theo
Nem.furc
Pen.naga
Sel.lept
Pen.para
Aba.stel
Sau.grp
Ech.nauc
Nem.hexo
Lag.scel
Car.coer
Dec.russ
Let.geni
Gym.mino
Car.fulv
Upe.trag
Par.otis
Par.nebu
Car.gymn
Cho.venu
Gna.spec
Ale.aper
Nem.pero
Species
Scomberomorus queenslandicus
Seriolina nigrofasciata
Nemipterus theodorei
Nemipterus furcosus
Pentapodus nagasakiensis
Selaroides leptolepis
Pentapodus paradiseus
Abalistes stellatus
Saurida grp
Echeneis naucrates
Nemipterus hexodon
Lagocephalus sceleratus
Carangoides coeruleopinnatus
Decapterus russelli
Lethrinus genivittatus
Gymnothorax minor
Carangoides fulvoguttatus
Upeneus tragula_grp
Paramonacanthus otisensis
Parapercis nebulosa_grp
Carangoides gymnostethus
Choerodon venustus
Gnathanodon speciosus
Alepes apercna
Nemipterus peronii
yres
19.66
18.88
18.76
17.72
17.57
17.52
17.49
17.4
16.45
16.25
15.73
15.48
15.2
15.05
14.86
14.31
14.05
13.93
13.24
12.62
12.51
12.22
10.46
10.29
9.81
%Var
79.23
76.23
83.33
72.95
86.34
79.23
77.60
72.95
77.87
65.85
91.80
68.85
70.22
77.32
77.32
83.33
72.68
78.14
87.98
82.79
80.87
83.33
85.25
85.52
86.07
Table 3-2: Top 20 explanatory variables (x) sorted by descending order of "xres" = (% of [1-% prediction error]
for each x). “xvar” is the mean percentage of the variation in the responses (species occurrence) explained by
each of the explanatory variables in the best gbmmv model, for production of biophysical maps.
Explanatory variable
across
ga.mud
ga.crbnt
ga.gravel
gbr.bathy
along
m.bstress
crs.s.av
sw.chla.sd
crs.s.sd
crs.no3.sd
sw.k.b.irr
crs.t.av
gbr.slope
trwl.eff.i
crs.o2.av
crs.si.sd
gbr.aspect
crs.si.av
ga.sand
xres
1.75
1.18
1.14
0.99
0.88
0.82
0.74
0.71
0.68
0.67
0.64
0.61
0.58
0.58
0.58
0.52
0.52
0.5
0.5
0.49
xvar
8.59
6.47
5.76
5.18
4.34
4.39
4.05
3.64
3.53
3.28
3.79
3.28
3.04
3.16
3.19
2.81
2.94
2.71
2.59
2.58
GBR Seabed Biodiversity
3-89
Table 3-3: Matrix of percentage of the variability in occurrence of 25 species responses explained by the top 20 explanatory variables.
sp.code across ga.mud ga.crbnt ga.gravel gbr.bathy along m.bstress crs.s.av sw.k.b.irr crs.s.sd crs.no3.sd gbr.slope trwl.eff.i crs.t.av ga.sand gbr.aspect crs.si.av crs.o2.av sw.chla.sd crs.si.sd
Sco.quee 28.2
2.2
6.5
3.0
2.0 2.0
1.9
1.9
1.4
5.1
2.2
2.7
3.4
1.4
1.6
3.1
1.4
2.1
5.4
0.9
Ser.nigr 16.1
2.3
6.2
3.3
8.8 4.3
1.5
3.3
3.9
3.0
1.4
1.9
0.8
2.3
1.7
1.3
1.1
3.1
7.0
4.1
Sel.lept 24.8
2.3
6.5
2.9
3.0 4.6
2.4
1.6
0.9
8.8
3.3
1.9
1.4
5.1
1.7
1.6
1.8
1.5
0.9
1.4
Nem.theo
6.2
5.2
4.0
6.3
14.4 5.8
1.5
10.7
2.9
2.0
1.9
2.5
0.7
0.7
2.8
1.3
3.6
2.9
4.1
3.0
Aba.stel
3.8
2.3
6.2
9.1
8.3 1.9
0.9
4.4
1.8
5.4
1.5
4.0
5.0
1.0
2.8
1.8
2.9
1.6
5.6
1.7
Pen.para
2.7
9.1
3.8
17.4
3.5 5.4
4.4
1.4
2.7
2.6
2.7
1.7
1.1
3.2
1.8
3.6
2.2
2.6
2.9
1.7
Nem.furc
7.6
3.4
4.9
2.7
3.2 2.7
4.8
5.4
3.4
3.4
2.4
2.2
2.0 11.4
2.3
3.0
1.3
2.7
1.4
1.4
Pen.naga 14.8
6.3
31.2
7.1
1.1 0.9
1.1
0.9
1.7
1.7
2.8
1.8
2.4
0.9
1.3
1.3
2.1
1.7
3.4
1.0
Sau.grp
7.9
3.5
4.5
5.2
4.8 6.3
2.0
7.9
5.8
3.1
3.1
3.2
3.5
1.1
1.8
2.0
2.7
3.6
3.9
1.9
Ech.nauc
4.4
2.2
3.9
5.3
1.6 1.7
2.6
2.0
4.2
2.5
3.8
4.2
3.5
3.4
4.4
2.7
2.6
2.2
5.0
2.1
Car.coer
4.2 21.1
5.0
2.2
2.2 2.7
5.1
2.6
3.7
1.0
2.9
1.9
2.6
1.1
2.9
1.8
1.6
2.1
2.5
1.8
Nem.hexo 20.4 30.9
4.1
2.4
2.6 1.2
4.1
4.5
1.1
2.1
1.6
3.8
5.0
1.0
1.1
0.4
2.3
0.8
1.9
0.6
Lag.scel
1.5
2.0
3.7
3.9
2.7 10.0
5.2
1.9
3.8
1.1
4.9
1.3
4.7
4.2
1.7
3.0
6.3
2.3
2.3
3.2
Dec.russ
5.2
6.5
5.9
2.1
7.8 3.5
5.4
5.6
2.6
7.9
2.5
4.5
1.7
1.9
2.0
2.2
2.5
2.1
5.0
1.4
Let.geni
4.4
4.3
4.2
6.1
2.8 8.7
4.4
2.8
6.4
3.1
4.0
4.0
3.0
2.2
3.1
3.5
3.5
2.5
2.7
2.0
Gym.mino
5.7
7.0
7.6
4.4
4.8 2.6
4.0
4.5
2.2
2.0
12.2
1.9
5.5
1.2
3.8
4.1
2.9
1.4
3.8
2.0
Car.fulv
6.1
2.9
3.7
6.6
6.2 6.2
1.7
2.7
2.7
4.8
1.5
3.1
2.3
2.4
2.1
2.5
3.2
2.3
3.5
4.8
Upe.trag
2.8
4.2
2.2
10.4
4.2 3.1
10.4
3.6
4.1
1.8
2.6
5.3
1.3
5.3
4.3
4.2
2.0
3.0
1.8
1.9
Par.otis
9.4
9.7
2.9
1.7
5.3 5.5
3.1
3.6
5.0
2.8
3.3
2.8
4.0
3.5
4.3
2.5
3.1
11.9
1.1
2.9
Par.nebu
4.8 11.5
2.0
2.8
2.3 2.1
8.4
2.6
4.8
1.7
2.8
5.2
6.1
1.9
4.4
4.2
2.7
6.1
3.5
2.6
Car.gymn
2.8
2.2
3.9
1.6
2.5 4.3
4.3
2.8
2.3
4.3
2.1
5.1
3.2
2.5
2.5
4.7
4.7
2.8
6.0
16.2
Cho.venu
7.0
3.7
14.2
6.8
2.1 8.4
11.4
4.3
1.9
4.7
2.9
4.5
1.5
2.0
0.9
1.4
1.9
1.1
1.8
1.5
Gna.spec
1.8
3.2
1.8
4.8
2.7 8.6
4.1
3.7
9.9
4.7
5.5
3.4
3.4
3.8
2.9
7.4
1.6
2.1
2.0
8.0
Ale.aper 12.0
5.6
2.2
7.7
3.5 3.9
2.9
4.1
0.9
3.8
3.3
3.4
4.1 11.9
3.4
2.3
4.1
2.3
2.2
1.7
Nem.pero 12.3
4.9
5.9
4.0
3.2 3.1
3.4
3.9
1.4
1.3
12.9
3.1
8.3
2.8
2.0
2.8
2.2
1.8
4.9
2.3
GBR Seabed Biodiversity
3-90
The dendrogram along the side of the heatmap shows which species were similar in having a
relationship with a set of predictor variables. It does not imply these species have the same
relationship. For example, Nemipterus hexodon, Pentapodus nagasakiensis and Scomberomorus
queenslandicus were all highly predictable [orange-red bars on left side of figure] and cluster together
in the left-hand dendrogram. However, N. hexodon and P. nagasakiensis are completely opposite in
their response to “Across” and “GA.mud”. The coloured bars along the top show the percentage of the
variation in the explanatory variables explained by a particular variable — note that “Across” and
“GA.mud” are red. The “redness” of the individual cells in the figure show the relative influence of the
particular explanatory variable on the presence/absence of the particular species, and the heaviness of
the blue line shows the degree and shape of the relationship.
Sel.lept
Sco.quee
Pen.naga
Upe.trag
Pen.para
Cho.v enu
Par.nebu
Par.otis
Car.coer
Dec.russ
Aba.stel
Car.f ulv
Sau.grp
Ech.nauc
Ale.aper
Nem.f urc
Nem.pero
Gy m.mino
Nem.theo
Ser.nigr
Let.geni
Lag.scel
Gna.spec
Car.gy mn
crs.s.av
gbr.bathy
ga.gravel
m.bstress
along
crs.no3.sd
crs.t.av
crs.o2.av
sw.k.b.irr
crs.si.av
gbr.slope
gbr.aspect
ga.sand
trwl.eff.i
sw.chla.sd
crs.s.sd
crs.si.sd
ga.crbnt
ga.mud
across
Nem.hexo
Figure 3-3: "Heatmap" showing relationships amongst and between the top 20 predictors and 25 species
responses (presence/absence). The dendrogram along the side of the heatmap shows which species are similar in
having a relationship with a set of predictor variables. It does not imply these species have the same relationship.
The dendrogram along the top shows which explanatory variables cluster together, and the coloured bars along
the top show the percentage of the variation in the explanatory variables explained by a particular variable. Red
indicates higher influence. The “redness” of the individual cells in the figure show the relative influence of the
particular explanatory variable on the presence/absence of the particular species, and the heaviness of the blue
line shows the degree and shape of the relationship.
GBR Seabed Biodiversity
3-91
The relative influence of the predictors and the shape of the relationships between species occurrence
and a selection of 7 of the 20 predictors are shown in a series of plots for: position across the shelf
(Figure 3-4); content of the sediments in terms of mud (Figure 3-5), carbonate (Figure 3-6), gravel
(Figure 3-7); water temperature (Figure 3-8) and salinity (Figure 3-9); and trawl effort (Figure 3-10).
The “rugs” on the X axes show the 10 percentiles of the distribution of the predictor variables, and for
the trawl index the data is dominated by zero effort in most of the sampling area, with high levels in
less than 10 percent of the data. This produces much leverage and complicated shapes in the functional
relationships.
Response: Sco.quee
1.0
0.5
0.0
-0.5
-1.0
Response: Sel.lept
1.5
1.0
0.5
0.0
-0.5
-1.0
0.2
0.6
1.0
0.2
0.6
1.0
Response: Ale.aper
1.0
0.2
0.6
1.0
0.0
-0.2
0.2
0.6
across
0.6
1.0
0.05
0.00
-0.05
-0.10
0.2
0.6
Response: Let.geni
1.0
0.2
0.6
1.0
0.06
0.04
0.02
0.00
-0.02
-0.04
-0.06
-0.08
0.2
0.6
across
1.0
0.2
0.6
0.6
across
1.0
0.6
across
0.2
0.6
1.0
across
Response: Car.coer
Response: Aba.stel
0.10
0.05
0.00
-0.05
-0.10
-0.15
0.2
0.6
1.0
0.2
across
1.0
1.0
Response: Dec.russ
1.0
0.6
1.0
across
Response: Gna.spec
0.03
0.02
0.01
0.00
-0.01
-0.02
-0.03
0.6
0.6
0.15
0.10
0.05
0.00
-0.05
-0.10
0.2
1.0
Response: Pen.para
0.2
0.2
across
0.15
0.10
0.05
0.00
-0.05
-0.10
-0.15
-0.20
0.10
0.05
0.00
-0.05
-0.10
-0.15
-0.20
0.2
1.0
Response: Gym.mino
across
Response: Upe.trag
0.6
across
Response: Ech.nauc
across
0.01
0.00
-0.01
-0.02
0.2
1.0
1.0
Response: Nem.furc
across
0.2
0.1
0.0
-0.1
-0.2
-0.3
-0.4
-0.5
across
Response: Car.gymn
0.6
0.6
across
0.6
0.4
0.2
0.0
-0.2
0.2
0.2
0.2
0.0
-0.2
-0.4
-0.6
across
0.2
0.1
0.0
-0.1
-0.2
-0.3
-0.4
1.0
Response: Sau.grp
1.0
Response: Car.fulv
across
Response: Par.nebu
0.6
0.2
0.0
-0.2
-0.4
0.2
0.6
0.4
0.2
0.0
-0.2
-0.4
0.2
Response: Nem.theo
1.0
0.2
across
0.2
0.1
0.0
-0.1
-0.2
-0.3
-0.4
-0.5
0.2
-0.5
across
Response: Par.otis
across
0.4
0.0
-1.0
1.0
0.6
0.4
0.2
0.0
-0.2
across
Response: Cho.venu
0.6
0.5
-0.5
across
0.8
0.6
0.4
0.2
0.0
-0.2
-0.4
0.6
0.4
0.2
0.0
-0.2
0.6
0.2
Response: Pen.naga
1.0
0.0
across
Response: Nem.pero
Response: Ser.nigr
0.5
1.5
1.0
0.5
0.0
-0.5
across
0.2
Response: Nem.hexo
Response: Lag.scel
0.02
0.00
-0.02
-0.04
-0.06
-0.08
0.2
0.6
across
1.0
0.2
0.6
1.0
across
Figure 3-4: Species occurrence as a function of location across the shelf, f(across). Plots are ranked in
descending order of relative influence of the predictor variable for the species. The “rugs” on the X axes are 10
percentiles in the distribution of the predictor variable. The Yaxes (log-odds) are Log(base 2) (1-Probability of
occurrence) and the plots are centred on the mean of Y.
GBR Seabed Biodiversity
Response: Nem.hexo
1
0
-1
-2
Response: Car.coer
-0.5
-1.0
0 20
60
0 20
ga.mud
Response: Gym.mino
0.2
0.0
-0.2
-0.4
-0.6
-0.8
60
Response: Dec.russ
60
ga.mud
0.10
0.05
0.00
-0.05
-0.10
-0.15
-0.20
60
Response: Let.geni
60
0 20
ga.mud
Response: Nem.furc
0 20
ga.mud
Response: Aba.stel
ga.mud
Response: Ech.nauc
0.2
0.1
0.0
-0.1
0.15
0.10
0.05
0.00
0 20
60
ga.mud
60
ga.mud
60
ga.mud
Response: Sel.lept
0.1
0.0
-0.1
-0.2
0 20
60
0 20
ga.mud
Response: Car.gymn
60
ga.mud
Response: Lag.scel
0.02
0.00
-0.02
-0.04
-0.06
-0.08
-0.10
-0.12
0.000
-0.005
ga.mud
0 20
Response: Ser.nigr
0.005
60
Response: Sau.grp
60
0.010
0 20
60
ga.mud
ga.mud
ga.mud
Response: Sco.quee
0 20
0.1
0.0
-0.1
-0.2
-0.3
0 20
60
0.10
0.05
0.00
-0.05
-0.10
-0.15
0 20
60
0.05
0.00
-0.05
-0.10
-0.15
0 20
Response: Nem.theo
Response: Cho.venu
Response: Car.fulv
60
ga.mud
ga.mud
ga.mud
0.15
0.10
0.05
0.00
-0.05
0 20
0 20
60
60
0.1
0.0
-0.1
-0.2
-0.3
-0.4
-0.5
-0.6
0.25
0.20
0.15
0.10
0.05
0.00
0 20
Response: Gna.spec
60
Response: Ale.aper
60
Response: Upe.trag
0 20
ga.mud
ga.mud
ga.mud
0.06
0.04
0.02
0.00
-0.02
0.05
0.00
-0.05
-0.10
-0.15
0 20
60
60
0.4
0.2
0.0
-0.2
-0.4
0.20
0.15
0.10
0.05
0.00
-0.05
0.3
0.2
0.1
0.0
-0.1
0 20
Response: Pen.naga
Response: Pen.para
1.2
1.0
0.8
0.6
0.4
0.2
0.0
-0.2
0 20
ga.mud
ga.mud
Response: Nem.pero
60
0.5
0.4
0.3
0.2
0.1
0.0
0 20
Response: Par.otis
0.4
0.3
0.2
0.1
0.0
-0.1
0 20
ga.mud
0.1
0.0
-0.1
-0.2
-0.3
-0.4
0 20
Response: Par.nebu
0.6
0.4
0.2
0.0
0.0
3-92
0 20
60
ga.mud
0 20
60
ga.mud
Figure 3-5: Species occurrence as a function of mud content of the sediments, f(ga.mud). Conventions as for
Figure 3-4.
GBR Seabed Biodiversity
Response: Pen.naga
Response: Cho.venu
1.0
0.8
0.6
0.4
0.2
0.0
1.5
1.0
0.5
0.0
20
60
-0.2
-0.4
60
Response: Aba.stel
Response: Ser.nigr
0.1
0.0
-0.1
-0.2
-0.3
-0.4
-0.5
0.5
0.4
0.3
0.2
0.1
0.0
-0.1
20
60
60
Response: Sau.grp
0.0
-0.2
-0.4
60
20
ga.crbnt
Response: Car.gymn
0.03
0.02
Response: Ech.nauc
20
60
ga.crbnt
Response: Par.otis
20
60
ga.crbnt
60
ga.crbnt
20
Response: Car.fulv
20
ga.crbnt
60
ga.crbnt
Response: Lag.scel
60
20
ga.crbnt
Response: Par.nebu
60
ga.crbnt
Response: Gna.spec
0.10
0.08
0.06
0.04
0.02
0.00
-0.02
0.04
0.02
0.00
-0.02
ga.crbnt
20
0.15
0.10
0.05
0.00
-0.05
-0.10
-0.15
-0.1
60
Response: Nem.theo
ga.crbnt
60
60
ga.crbnt
60
0.1
0.0
20
20
0.0
-0.1
-0.2
-0.3
-0.4
-0.5
0.2
Response: Upe.trag
Response: Car.coer
60
Response: Nem.hexo
Response: Pen.para
60
ga.crbnt
ga.crbnt
60
0.08
0.06
0.04
0.02
0.00
-0.02
20
20
ga.crbnt
20
20
0.0
-0.1
-0.2
-0.3
0.2
0.1
0.0
-0.1
-0.2
20
ga.crbnt
0.06
0.04
0.02
0.00
-0.02
-0.04
-0.06
0.3
0.2
0.1
0.0
Response: Let.geni
60
Response: Ale.aper
Response: Dec.russ
60
0.15
0.10
0.05
0.00
-0.05
-0.10
-0.15
20
60
ga.crbnt
ga.crbnt
ga.crbnt
0.2
0.1
0.0
-0.1
-0.2
-0.3
0.01
0.00
-0.01
20
60
-0.2
-0.4
0.10
0.05
0.00
-0.05
-0.10
-0.15
-0.20
0.3
0.2
0.1
0.0
-0.1
-0.2
-0.3
0.2
20
Response: Nem.pero
Response: Sel.lept
0.2
0.0
20
ga.crbnt
ga.crbnt
Response: Nem.furc
60
0.15
0.10
0.05
0.00
-0.05
-0.10
-0.15
20
ga.crbnt
0.2
0.1
0.0
-0.1
-0.2
20
ga.crbnt
Response: Sco.quee
0.2
0.0
-0.2
-0.4
-0.6
0.2
0.0
20
ga.crbnt
Response: Gym.mino
3-93
20
60
ga.crbnt
20
60
ga.crbnt
Figure 3-6: Species occurrence as a function of carbonate content of the sediments, f(ga.crbnt). Conventions as
for Figure 3-4.
GBR Seabed Biodiversity
Response: Pen.para
0.0
-0.5
-1.0
-1.5
Response: Upe.trag
0.0
-0.2
-0.4
-0.6
-0.8
-1.0
0.0
-0.2
-0.4
0 20
60
0 20
ga.gravel
60
Response: Car.fulv
0.2
0.0
-0.2
-0.4
-0.6
0 20
60
ga.gravel
Response: Sau.grp
0.6
0.2
0.0
0 20
Response: Gna.spec
ga.gravel
0.20
0.15
0.10
0.05
0.00
60
0 20
Response: Nem.hexo
60
ga.gravel
Response: Car.coer
0.02
0.00
-0.02
-0.04
-0.06
-0.08
0.20
0.15
0.10
0.05
0.00
0 20
60
ga.gravel
0 20
0 20
60
ga.gravel
0 20
60
ga.gravel
Response: Lag.scel
60
Response: Par.nebu
0 20
60
ga.gravel
Response: Nem.furc
0.15
0.10
0.05
0.00
-0.05
0 20
60
0 20
ga.gravel
Response: Par.otis
60
ga.gravel
Response: Car.gymn
0.002
0.000
-0.002
-0.004
-0.006
0.05
0.00
-0.05
-0.10
-0.15
0.00
ga.gravel
ga.gravel
Response: Dec.russ
60
0.25
0.20
0.15
0.10
0.05
0.00
-0.05
0 20
ga.gravel
0.05
0 20
Response: Nem.pero
60
0.10
Response: Ech.nauc
60
0.10
0.05
0.00
-0.05
-0.10
-0.15
0.04
0.03
0.02
0.01
0.00
-0.01
-0.02
0 20
ga.gravel
Response: Sel.lept
ga.gravel
ga.gravel
ga.gravel
Response: Sco.quee
60
0 20
60
60
0.3
0.2
0.1
0.0
-0.1
-0.2
-0.3
-0.4
0.06
0.05
0.04
0.03
0.02
0.01
0.00
-0.01
0 20
ga.gravel
0.00
-0.05
-0.10
-0.15
-0.20
Response: Let.geni
ga.gravel
Response: Gym.mino
0 20
ga.gravel
60
0.5
0.4
0.3
0.2
0.1
0.0
0 20
60
0.0
-0.2
-0.4
-0.6
0 20
ga.gravel
60
Response: Ser.nigr
Response: Nem.theo
Response: Pen.naga
0.0
-0.2
-0.4
-0.6
0 20
ga.gravel
60
0.1
0.0
-0.1
-0.2
-0.3
0.4
60
0.8
0.6
0.4
0.2
0.0
0 20
Response: Ale.aper
0.4
0.3
0.2
0.1
0.0
0 20
ga.gravel
Response: Cho.venu
0.1
0.0
-0.1
-0.2
-0.3
-0.4
-0.5
Response: Aba.stel
3-94
0 20
60
ga.gravel
0 20
60
ga.gravel
Figure 3-7: Species occurrence as a function of gravel content of the sediments, f(ga.gravel). Conventions as for
Figure 3-4.
GBR Seabed Biodiversity
Response: Ale.aper
1.5
Response: Nem.furc
1.0
0.8
0.6
0.4
0.2
0.0
-0.2
-0.4
1.0
0.5
0.0
20
24
Response: Gna.spec
20
Response: Par.otis
Response: Car.gymn
24
20
crs.t.av
24
24
Response: Aba.stel
0.02
0.01
0.00
-0.01
-0.02
-0.03
-0.04
24
crs.t.av
Response: Nem.hexo
20
24
crs.t.av
20
24
crs.t.av
20
24
crs.t.av
Response: Cho.venu
24
20
crs.t.av
Response: Gym.mino
24
crs.t.av
Response: Sau.grp
0.04
0.02
0.00
-0.02
-0.04
-0.06
-0.08
20
24
20
crs.t.av
Response: Pen.naga
24
crs.t.av
Response: Nem.theo
0.04
0.03
0.02
0.01
0.00
-0.01
0.02
0.01
0.00
-0.01
-0.02
-0.10
-0.01
24
0.20
0.15
0.10
0.05
0.00
-0.05
-0.10
-0.15
20
crs.t.av
-0.05
0.00
0.00
Response: Let.geni
24
0.00
0.01
20
20
crs.t.av
0.05
0.04
0.02
0.00
-0.02
-0.04
-0.06
-0.08
-0.10
0.00
-0.05
Response: Nem.pero
crs.t.av
crs.t.av
Response: Sco.quee
crs.t.av
0.10
20
24
0.10
0.05
20
crs.t.av
Response: Car.coer
20
24
0.15
0.15
0.10
0.05
0.00
-0.05
-0.05
Response: Par.nebu
Response: Pen.para
crs.t.av
Response: Ser.nigr
20
crs.t.av
24
0.00
0.04
0.02
0.00
-0.02
-0.04
-0.06
0.10
0.08
0.06
0.04
0.02
0.00
-0.02
20
24
24
0.5
0.4
0.3
0.2
0.1
0.0
-0.1
0.05
20
crs.t.av
Response: Dec.russ
Response: Ech.nauc
crs.t.av
Response: Car.fulv
20
crs.t.av
24
0.20
0.15
0.10
0.05
0.00
-0.05
20
24
0.3
0.2
0.1
0.0
-0.1
-0.2
20
crs.t.av
0.010
0.005
0.000
-0.005
-0.010
-0.015
-0.020
20
crs.t.av
24
0.0
-0.1
0.5
0.4
0.3
0.2
0.1
0.0
-0.1
0.20
0.15
0.10
0.05
0.00
0.2
0.0
Response: Lag.scel
0.3
0.2
0.1
0.0
-0.1
0.4
0.1
24
Response: Sel.lept
0.6
0.2
20
crs.t.av
Response: Upe.trag
3-95
20
24
crs.t.av
20
24
crs.t.av
Figure 3-8: Species occurrence as a function of average water temperature, f(crs.t.av). Conventions as for Figure
3-4.
GBR Seabed Biodiversity
Response: Nem.theo
1.0
Response: Sau.grp
0.8
0.6
0.4
0.2
0.0
-0.2
0.5
0.0
-0.5
35.0
35.4
Response: Nem.hexo
35.4
35.0
crs.s.av
Response: Aba.stel
35.4
Response: Cho.venu
0.2
0.0
35.0
35.4
35.0
crs.s.av
Response: Gna.spec
35.4
Response: Upe.trag
0.10
0.4
0.3
0.2
0.1
0.0
0.00
-0.05
-0.10
35.4
35.0
crs.s.av
Response: Let.geni
Response: Car.fulv
0.3
0.3
0.2
0.1
0.0
-0.1
0.1
0.0
-0.1
35.4
35.0
crs.s.av
Response: Sco.quee
0.10
0.05
0.00
-0.05
-0.10
Response: Lag.scel
35.4
crs.s.av
Response: Sel.lept
35.4
crs.s.av
35.4
crs.s.av
35.4
crs.s.av
Response: Car.gymn
35.4
35.0
crs.s.av
Response: Par.nebu
35.4
crs.s.av
Response: Ech.nauc
0.1
0.0
-0.1
-0.2
35.0
35.4
35.0
crs.s.av
Response: Pen.para
0.06
0.04
0.02
0.00
-0.02
35.0
35.0
0.015
0.010
0.005
0.000
-0.005
-0.010
-0.015
-0.020
35.0
crs.s.av
-0.15
35.0
Response: Ser.nigr
35.4
0.05
0.00
-0.05
-0.10
Response: Nem.pero
35.4
0.20
0.15
0.10
0.05
0.00
-0.05
35.0
35.4
crs.s.av
crs.s.av
35.4
Response: Car.coer
35.0
0.4
0.3
0.2
0.1
0.0
35.0
crs.s.av
crs.s.av
0.02
0.00
-0.02
-0.04
-0.06
35.0
35.0
35.4
35.4
0.1
0.0
-0.1
-0.2
-0.3
0.15
0.10
0.05
0.00
-0.05
0.2
35.0
Response: Par.otis
crs.s.av
0.0
-0.2
Response: Ale.aper
crs.s.av
35.4
0.2
crs.s.av
35.4
0.8
0.6
0.4
0.2
0.0
0.05
35.0
35.0
crs.s.av
0.4
0.5
0.4
0.3
0.2
0.1
0.0
-0.1
0.6
0.4
0.2
0.0
0.4
Response: Gym.mino
0.6
35.0
crs.s.av
0.6
1.0
0.8
0.6
0.4
0.2
0.0
-0.2
Response: Nem.furc
0.3
0.2
0.1
0.0
-0.1
-0.2
0.4
0.3
0.2
0.1
0.0
-0.1
35.0
crs.s.av
Response: Dec.russ
3-96
35.4
crs.s.av
Response: Pen.naga
0.01
0.00
-0.01
-0.02
35.0
35.4
crs.s.av
35.0
35.4
crs.s.av
Figure 3-9: Species occurrence as a function of average salinity, f(crs.s.av). Conventions as for Figure 3-4.
GBR Seabed Biodiversity
Response: Nem.pero
Response: Par.nebu
0.0
0.05
0.00
-0.05
-0.10
-0.15
-0.20
-0.1
-0.2
-0.3
0
40
80
Response: Gym.mino
0.005
0.000
-0.005
-0.010
-0.015
-0.020
-0.025
0
trwl.eff.i
40
80
0
Response: Ale.aper
0.0
0.10
-0.1
0.05
-0.2
-0.3
0.00
80
40
80
Response: Gna.spec
0.2
0.0
0
40
80
40
-0.5
-1.0
40
0
40
40
trwl.eff.i
0.05
0
80
40
80
40
Response: Pen.para
0.000
trwl.eff.i
80
0
40
trwl.eff.i
Response: Car.coer
80
0
80
40
80
trwl.eff.i
Response: Cho.venu
8e-04
6e-04
4e-04
2e-04
0e+00
0
40
80
Response: Ser.nigr
0
40
80
trwl.eff.i
Response: Nem.theo
0.00
-0.01
-0.02
-0.03
-0.04
-0.05
0.00
-0.01
-0.02
-0.03
-0.04
0.005
80
trwl.eff.i
trwl.eff.i
0.010
40
40
Response: Dec.russ
trwl.eff.i
Response: Upe.trag
0
0
80
40
0.06
0.04
0.02
0.00
0.01
0.00
-0.01
-0.02
-0.03
-0.04
-0.05
0
0
trwl.eff.i
Response: Nem.furc
80
0.004
0.003
0.002
0.001
0.000
0
-0.010
80
Response: Ech.nauc
80
0.15
trwl.eff.i
Response: Sel.lept
40
Response: Let.geni
0.10
40
trwl.eff.i
trwl.eff.i
0.00
-0.02
-0.04
-0.06
-0.08
-0.10
-0.12
trwl.eff.i
0.02
0.00
-0.02
-0.04
-0.06
-0.08
0
trwl.eff.i
Response: Car.fulv
80
80
-0.005
80
0
0.00
-0.05
-0.10
-0.15
-0.20
-0.25
-0.30
0.00
0
0.06
0.05
0.04
0.03
0.02
0.01
0.00
0.0
40
Response: Car.gymn
trwl.eff.i
Response: Pen.naga
0
0
0.000
trwl.eff.i
1.0
0.5
80
Response: Sau.grp
trwl.eff.i
0.02
0.01
0.00
-0.01
-0.02
-0.03
-0.04
-0.05
40
0.0
-0.1
-0.2
-0.3
-0.4
trwl.eff.i
0.6
0.4
0
trwl.eff.i
-0.3
0
trwl.eff.i
Response: Sco.quee
80
Response: Par.otis
0.15
40
40
Response: Nem.hexo
0.0
-0.1
-0.2
-0.3
trwl.eff.i
0.0
-0.1
-0.2
0
Response: Aba.stel
0.05
0.04
0.03
0.02
0.01
0.00
trwl.eff.i
Response: Lag.scel
3-97
0
40
80
trwl.eff.i
0
40
80
trwl.eff.i
Figure 3-10: Species occurrence as a function of trawl effort index, f(trwl.eff.i). Conventions as for Figure 3-4.
3.1.2.1. BRUVS single species distribution maps
The GBM models produced for the 25 most predictable species seen on BRUVS footage were applied
to the overall grid of cells comprising the entire GBRMP, to produce biophysical maps of the
occurrence (presence or absence only) of these species. These maps are provided here as six panels of
4 species each, including a measure of reliability of the predictions and a shortlist of the most
influential spatial and environmental covariates. Note that the influence of the covariates does not
indicate the actual shape of the relationship. For example, high influence of seabed shear stress may
influence occurrence of a species positively, negatively, modally or asymptotically. The actual
abundance [4th root transformed] recorded at each BRUVS station is featured as scaled circles. This
gives some idea of how consistent the predictions are with the actual observations.
GBR Seabed Biodiversity
3-98
In Figure 3-11 the three species in the genus Nemipterus are featured. Nemipterus furcosus is
influenced by seawater temperature and cross-shelf position, and is commonest north of the Palm
Islands in lagoonal and inter-reef waters. Nemipterus peronii shows only partial concordance with
observed concentrations of abundance – the high abundance in Princess Charlotte Bay is not matched
by a high probability of occurrence there. N.hexodon description awaiting corrected version of map.
The two small nemipterids, Pentapodus paradiseus and P. nagasakiensis, are similar in shape, but
have different biophysical distribution maps (Figure 3-12). P. paradiseus evidently prefers gravelly
sediments and is widespread in most habitats across the shelf, except the deeper lagoonal areas. P.
nagasakiensis was influenced most by high carbonate content of sediments and was found to be
prevalent on outer-shelf areas in channels and passes. Lethrinus genivittatus was influenced by high
light levels at the seabed [perhaps in association with marine plants] and was found in patches
throughout the GBRMP. Results from the far north may be less certain due to the lack of sampling off
Cape York. Predictions for abundance of Upeneus tragula_grp were more prevalent amongst the reef
matrix in high current, gravelly areas.
The deep-bodied Alepes apercna was predicted to be most common in the far north and also in the
area off Mackay (Figure 3-13). Temperature had high influence, but the temperature in these two
regions is at opposite extremes. Consistently high occurrence and abundance were predicted and
observed for Decapterus russelli in the deeper lagoon waters of the Capricorn Channel. This species is
both a planktivore and demersal microcarnivore. Selaroides leptolepis has a similar habit and diet, but
is restricted strongly to inshore waters less than half way across the shelf in the south, although this
distribution extends much further offshore north of Cape Flattery. Seriolina nigrofasciata is a pelagic,
fusiform species seen on BRUVS mainly as a juvenile. Its predicted occurrence was highest in deeper,
lagoon waters south of Cape Flattery. It may be both a piscivore and planktivore. It has been seen in
the BRUVS field of view settled on the seabed on outstretched pelvic fins.
The Queensland school mackerel, Scomberomorus queenslandicus, is an active visual predator that
hunts small planktivorous fishes, squid and pelagic crustaceans. Its distribution closely matches that of
the potential prey species Selaroides leptolepis, being confined to inshore areas in the south and
extending further offshore in the north (Figure 3-14). The suckerfish, Echeneis naucrates, is reputedly
a scavenger associated with large sharks and rays, yet we recorded this species very commonly and in
abundance as free-swimming individuals. It is ubiquitous in most habitats, with a “hotspot” of
occurrence in the central section off Townsville. The lizardfishes, Saurida_grp, were lumped into one
taxa and are demersal ambush predators most prevalent in the more saline southern waters of the
GBRMP, in the deeper, clearer waters of the lagoon. Parapercis nebulosa_grp occurred in the south
also with high probability, but more inshore than Saurida_grp, influenced by mud content of the
sediments. It is also an ambush predator, but moves more frequently than Saurida_grp about the
BRUVS bait stations.
Deeper waters with high gravel and carbonate were predicted to be the favoured habitat of Abalistes
stellatus, which extended to the outer shelf, but not nearshore waters (Figure 3-15). The silver
toadfish, Lagocephalus sceleratus, was most abundant and prevalent across the shelf in the centralnorthern section between Bowen and Cape Flattery. Highly oxygenated waters and muddy sediments
in the southern region were predicted to have the greatest occurrence of the small leatherjacket,
Paramonacanthus otisensis. Lagoonal sites with high mud content were apparently favoured by the
small predatory moray eel, Gymonthorax minor.
A variety of carangids were seen on BRUVS footage. One of the most common was the “onion
trevally”, Carangoides coeruleopinnatus. It was influenced by mud and current, but its inshore
habitats north of the Whitsundays were replaced by highest predicted occurrence in the deeper lagoon
waters offshore from the macrotidal coast and bays of the southern region (Figure 3-16). Carangoides
fulvoguttatus and C. gymnostethus are much larger predators of fish, crustaceans and molluscs
anywhere in the water column and from the seabed. C. fulvoguttatus had predicted occurrence in most
cross-shelf habitats with the exception of shallow nearshore margins. C. gymnostethus had a similar
distribution in BRUVS records, but the predictions were very weak on the biophysical map. The
golden trevally, Gnathanodon speciosus, had patchy distribution of predictions, but consistently higher
records in the far north.
GBR Seabed Biodiversity
(a) Actinopterygii: Nemipterus theodorei
(b) Actinopterygii: Nemipterus furcosus
(c) Actinopterygii: Nemipterus peronii
(d) Actinopterygii: Nemipterus hexodon
3-99
Figure 3-11 Predicted occurrence of 3 species of Nemipterus recorded by BRUVS. Circles represent observed
abundance (untransformed) and influential covariates are listed in the inset panels. “%XVar” describes the
percentage of the variation in the presence/absence of the species accounted for by the gbm model. “yres” is (1%prediction error).
GBR Seabed Biodiversity
(a) Actinopterygii: Pentapodus paradiseus
(b) Actinopterygii: Pentapodus nagasakiensis
(c) Actinopterygii: Lethrinus genivittatus
(d) Actinopterygii: Upeneus tragula_grp
3-100
Figure 3-12 Predicted occurrence of small benthic microcarnivores in the genera Pentapodus, Lethrinus and
Upeneus. Conventions as for Figure 3-11.
GBR Seabed Biodiversity
(a) Actinopterygii: Alepes apercna
(b) Actinopterygii: Decapterus russelli
(c) Actinopterygii: Selaroides leptolepis
(d) Actinopterygii: Seriolina nigrofasciata
3-101
Figure 3-13 Predicted occurrence of small carangids in the genera Alepes, Decapterus, Selaroides and Seriolina.
Conventions as for Figure 3-11.
GBR Seabed Biodiversity
(a) Actinopterygii: Scomberomorus queenslandicus
(b) Actinopterygii: Echeneis naucrates
(c) Actinopterygii: Saurida_grp
(d) Actinopterygii: Parapercis nebulosa_grp
3-102
Figure 3-14 Predicted occurrence of predators in the genera Scomberomorus, Echeneis, Saurida and Parapercis.
Conventions as for Figure 3-11.
GBR Seabed Biodiversity
(a) Actinopterygii: Abalistes stellatus
(b) Actinopterygii: Lagocephalus sceleratus
(c) Actinopterygii: Paramonacanthus otisensis
(d) Actinopterygii: Gymonthorax minor
3-103
Figure 3-15 Predicted occurrence of demersal omnivores and predators in the genera Abalistes, Lagocephalus,
Paramonacanthus and Gymnothorax. Conventions as for Figure 3-11.
GBR Seabed Biodiversity
(a) Actinopterygii: Carangoides coeruleopinnatus
(b) Actinopterygii: Carangoides fulvoguttatus
(c) Actinopterygii: Carangoides gymnostethus
(d) Actinopterygii: Gnathanodon speciosus
3-104
Figure 3-16 Predicted occurrence of the large predatory carangids in the genera Carangoides and Gnathanodon.
Conventions as for Figure 3-11.
GBR Seabed Biodiversity
3-105
High current areas with coarse, gravely sediments and high carbonate produced the highest sightings
and predicted occurrence of the Venus Tusk fish, Choerodon venustus. This species is known to
frequent reef edges and deeper shoals (Figure 3-17).
Actinopterygii: Choerodon venustus
Figure 3-17 Predicted occurrence of the large benthic macrocarnivore Choerodon venustus. Conventions as for
Figure 3-11.
3.1.3. BRUVS Site-groups Characterization and Prediction
The best 20 predictors and the best 25 responses were selected with the same process described above
for the relative abundance (4th root MaxN) data obtained from BRUVS. This subset differed slightly in
species membership, and was analysed with multivariate trees to allow the definition and biophysical
mapping of assemblages in the sampling grid. The aim was to produce maps of assemblages that could
be readily interpreted in two dimensions, but reflect also the underlying environmental correlates.
Two scenarios were examined for the top 25 species: a) assemblages defined by only distance along
and across the shelf, and b) assemblages defined by all the top 20 environmental covariates.
3.1.3.1. Assemblages defined by location across and along
The top 25 species in the first scenario, and the assemblage groups in which their Dufrêne Legendre
Indicator (DLI) values were maximised, are shown in Figure 2-26. The best representation of these
“spatial only” assemblages for prediction and mapping had 12 terminal nodes that were readily
interpreted by distance across and along the shelf, in relation to coastal landmarks at equivalent
latitudes (Table 3-4). Six assemblages had no species with DLI maxima, because species comprising
these assemblages occurred elsewhere in higher numbers. A large proportion of species were
ubiquitous, having maximum DLI in nodes at higher spatial scales. For example, the school mackerel,
GBR Seabed Biodiversity
3-106
Scomberomorus queenslandicus, occurred abundantly in most inshore assemblages along the shelf, so
had maximum DLI in the “inshore” node.
A geographical interpretation of the “spatial” assemblages is presented as Figure 3-19, split into the
“inshore” and “offshore” groups. Important faunal boundaries can be seen around Bowen and Cape
Flattery. Between these break points there are 3 inshore, mid-shelf and outer-shelf groupings evident.
This zonation is the “classic” cross-shelf pattern recognised by many authors in previous GBRMP
studies. However, north and south of this “central” section the cross-shelf pattern becomes more
complex. Offshore in the south, the eastern and western Swains reefs are distinguished from the
Pompey's and Whitsunday reefs. Off Bowen, the Gould/Cobham reefs appear to be the western
boundary of offshore assemblage “O-Wh”. The adjacent assemblage “O-SRf” is bounded to the west
by an extension of Hydrographer’s Passage near the Pompey reefs and to the east by the “TReefs” and
Herald’s Prong. Inshore to the south, the macrotidal Whitsunday, Shoalwater Bay and Broad Sound
regions have an assemblage separate from the deep Capricorn channel and Curtis Channel stations.
The Curtis Channel assemblage is distinct from the Capricorn-Bunker-eastern Swains assemblage. To
the north of Cape Flattery both shallow and deep, Cape York and far north, sites can be distinguished
in separate assemblages. The narrow channel separating Jewell-Waining reefs from Hicks / Ribbon
Reefs consistently appears to be related to this faunal break off Cape Flattery. This is somewhat
surprising given the much wider Trinity Opening lies further south.
Across< 0.53
Across>=0.53
ALL
Nemipterus furcosus 52
Echeneis naucrates 47
Abalistes stellatus 40
Pentapodus paradiseus 37
Nth of Bowen
along>=0.39
along<
0.74
Sth of Bowen
Insh
Nth of Mackay
Scomberomorus queenslandicus 64
along>=0.26
O-S
O-FN
along>=0.19
along< 0.19
O-SSC
O-S-Rf
Paramonacanthus japonicus 41
Oxycheilinus bimaculatus 36
Nth of Cape Flattery
along>=0.74
along< 0.74
Offsh
Pentapodus nagasakiensis 51
along>=0.74
along< 0.26
O-CFN
O-CN
Sth of Cape Flattery
along< 0.39
O-FS-Rf
O-Wh-Rf
Parapercis xanthozona_grp 40
Choerodon venustus 38
Selaroides leptolepis 56
Nemipterus hexodon 40
Carangoides coeruleopinnatus 34
Paramonacanthus otisensis 30
Gymnothorax minor 29
Across< 0.23
Nemipterus peronii 21 along< 0.84 along>=0.84
(Nearshore)
Across>=0.23
Terapon theraps 13
I-SC
I-FN
I-FN
along>=0.36
along< 0.36
Nrsh
I-CN
along>=0.38
along< 0.38
Lag
Seriolina nigrofasciata 46
I-S
along< 0.12
Lag-MS-CN
I-CY
Scolopsis taeniopterus 18
Atule mate 45
Alepes apercna 24
along>=0.12
Lag-CBG-S
I-Curt
Lag-Capr
Nemipterus theodorei 43
Saurida grp 32
Suezichthys devisi_grp 34
Paramonacanthus filicauda 59
Figure 3-18: Multivariate regression tree analysis defining abundance (transformed by 4th root) of vertebrate
assemblages (top 25 species) in terms of location across and along the GBRMP (366 sites). The terminal nodes
represent 12 assemblages (see Table 3-4 for definitions of nodes), corresponding with different regions of the
GBRMP, and the higher level nodes represent the 11 assemblages at higher spatial scales. The indicator species
are shown with the DLI value for nodes where maxima in DLI occurred.
GBR Seabed Biodiversity
3-107
Table 3-4: Hierarchy of nodes in the multivariate tree using location along and across the shelf to represent the
location of species assemblages. The number of BRUVS stations and of species with maxima in DLI values are
listed for each node. Terminal nodes are in bold font.
node
1
abbrvn
All
split
root
N sites
366
Description
Entire study area
2
Offsh
across>=0.52
170
Offshore
4
O-CFN
along>=0.38
68
8
O-CN
along<0.74
53
Offshore, Central to far Northern
GBRMP
Offshore, central-north
9
O-FN
along>=0.74
15
Offshore, far north
5
O-S
along<0.38
102
Offshore, Southern
between Bowen and Cape
Flattery
between Cape Flattery and
Cape Grenville
South of Bowen
10
O-SSC
along<0.26
71
Offshore, southern-south-central
South of Mackay
20
O-S-Rf
along>=0.19
20
21
O-FS-Rf
along<0.19
51
Mackay to Shaw Island,
Whitsundays
South of Mackay
11
O-Wh
along>=0.26
31
3
Insh
across<0.52
196
Offshore, mid-south, Pompey’s and
western Swains Reef channels
Offshore, eastern Swains channels
and Capricorn-Bunker shoals
Offshore, Whitsunday sector interreef
Inshore
6
I-SC
along<0.74
157
Inshore, South and Central GBRMP South of Cape Flattery
12
Nrsh
across<0.23
83
Inshore, nearshore
South of Cape Flattery
24
I-CN
along>=0.36
41
25
I-S
along<0.36
42
Inshore, nearshore, central-north
section
Nearshore, south
13
Lag
across>=0.23
74
Inshore, Lagoon
between Bowen and Cape
Flattery
between Agnes Waters and
Bowen
South of Cape Flattery
26
37
37
Lagoon and mid-shelf, central-north between Bowen and Cape
sections
Flattery
Inshore, Lagoon
South of Cape Bowling Green 2
54
Lag-MS- along>=0.38
CN
Lag-CBG- along<0.38
S
I-Curt
along<0.14
9
Curtis Channel
55
Lag-Capr along>=0.14
28
Capricorn Channel, lagoon waters
7
I-Nth
along>=0.74
39
Inshore, Far North
14
I-FN
along<0.84
21
Inshore, far north
15
I-CY
along>=0.84
18
Inshore, farthest north, Cape York
27
Boundary landmarks
DLI
4
1
North of Bowen
2
between Bowen and Mackay 2
8
1
Curtis Channel inshore of
Capricorn-Bunker Group
Capricorn Channel to
Whitsundays
North of Cape Flattery
1
between Cape Flattery and
Cape Sidmouth
North of Cape Sidmouth
1
1
2
GBR Seabed Biodiversity
Offshore
Inshore
3-108
O-CN
O-FN
O-S-Rf
O-FS-Rf
O-Wh
I-CN
I-S
Lag-MS-CN
I-Curt
Lag-Capr
I-FN
I-CY
Figure 3-19 Predicted distribution of 12 fish assemblages (terminal nodes from Table 3-4) as defined by the
explanatory variables “across” and “along” the shelf.
3.1.3.2. Assemblages defined by influential environmental covariates and location
The second scenario included the top 20 environmental covariates and transformed abundance of the
25 most predictable species. It is notable that only position across and along the shelf, depth, mud,
sand and gravel content of the sediments, and silica concentration in the water, predominated in higher
nodes of the most parsimonious tree. This tree had 12 nodes also (see Table 3-5 and Figure 3-20).
The location of these 12 fish assemblages is best visualised in inshore and offshore groupings (see
Figure 3-21). On the inshore side of the tree, Cape Flattery once again represents a faunal boundary,
with shallow assemblages in the far north and off Cape York influenced most by depth, rather than
sediment or water column characteristics. The Cape York assemblage had mobile, schooling carangid
microcarnivores (Alepes apercna, Atule mate) with high DLI. Between Cape Flattery and the Palm
Islands a widespread assemblage characterised by high mud content of the sediments comprised mid-
GBR Seabed Biodiversity
3-109
shelf regions in the north and lagoonal sites in the south, but there was no evidence from the tree of
differences in carbonate levels in the muds of these two regions. Most species in this assemblage were
found elsewhere, although the demersal microcarnivore Scolopsis taeniopterus, in the family
Nemipteridae, had highest DLI there. An assemblage characterised by low mud in the sediments and
high silica in the water comprised areas in open, lagoonal waters inside the mid-shelf reefs between
about Cardwell and Cape Upstart, and again in the Curtis Channel inside the Capricorn – Bunker
group of reefs in the far south.
across>=0.52
across< 0.52
All
Nemipterus.furcosus 52
Echeneis.naucrates 47
Pentapodus.paradiseus 37
ga.gravel< 8.983
ga.gravel>=8.983
gbr.bathy>=-49.3
Offsh
Pentapodus.nagasakiensis 51
along>=0.38
ga.sand< 89.95
along< 0.38
ga.sand>=89.95
LoGrav
LGrav-LSand
HiGrav
Rf-LGrav-HSand
gbr.bathy< -49.3
Insh
Sth of Cape Flattery
along< 0.74
HGrav-CntrlNth
Rf-HGravSthn
Scomberomorus.queenslandicus 64
Selaroides.leptolepis 56
Nemipterus.hexodon 40
Carangoides.coeruleopinnatus 34
Gymnothorax.minor 29
Nemipterus.peronii 21
along>=0.74
Shal
Deep-Capr
Suezichthys.devisi_grp 15
Oxycheilinus.bimaculatus 37
Paramonacanthus.japonicus 12
ga.mud>=16.05
ga.mud< 16.05
Sth of Cape Direction
along< 0.84
Choerodon.venustus 44
Paramonacanthus.filicauda 65
Nemipterus.theodorei 58
Saurida.grp 48
Seriolina.nigrofasciata 47
Abalistes.stellatus 41
along>=0.84
Nthn
Parapercis.xanthozona_grp 39
crs.si.av< 1.56
crs.si.av>=1.56
across>=0.28
HiMud
LoMud
ga.gravel>=10.29 ga.gravel< 10.29
HiSil
HMud-Nrshr
HMud-Lag
HSil-LMud- HSil-LMudLMud-LSil
HGrav
LGrav
Terapon.theraps 14
Shal-FN
Shal-CY
across< 0.28
Scolopsis.taeniopterus 29
Atule.mate 48
Alepes.apercna 25
Paramonacanthus.otisensis 39
Figure 3-20 Multivariate regression tree analysis defining abundance (transformed by 4th root) of vertebrate
assemblages (top 25 species) in terms of the top 20 environmental covariates in the GBRMP (366 sites). The
terminal nodes represent 12 assemblages (see Table 3-5 for definitions of nodes), corresponding with various
levels of mud, sand, gravel and silica and different regions of the GBRMP. The indicator species are shown with
the DLI value for nodes where maxima in DLI occurred.
An assemblage characterised by high mud hugged the nearshore northern coast to the vicinity Bowen,
but then showed a notable extension to deeper waters offshore from Mackay and the Shoalwater BayBroad Sound region. The small, demersal microcarnivore Terapon theraps was an indicator species
for this group. It was remarkable to find it restricted to the shallowest nearshore sites in the central
section, but then record it in deeper [>40 m] sites far offshore in the southern region outside the
macrtotidal bays. The high tidal energy in the Shoalwater Bay and Broad Sound was expected to
produce scoured demersal habitats with coarse sediments. Indeed, the fish assemblage recognised
there was influenced most by high gravel and low mud fractions in the sediments and high silica in the
water column. A mix of sites offshore from Gladstone to the Whitsundays was influenced by low mud
and low gravel in the sediments and high silica in the water column.
The offshore assemblages (see Figure 3-21) were less distinct spatially. In the eastern Swains Reefs
for example there were a mix of three assemblages – two near reefs and one in channels and passes
influenced by low gravel and low sand fractions. An assemblage near reefs in the southern region is
influenced by high gravel, with DLI indicator species the Venus Tuskfish (Choerodon venustus), a
large carnivore of benthic invertebrates, and Parapercis xanthozona_grp, a small ambush predator.
Some adjacent sites in areas of high sand and low gravel, such as channels and passes, form a separate
assemblage with the small wrasse Suezichthys devisi_grp being the sole indicator species. In the
GBR Seabed Biodiversity
3-110
central-north, north of about Bowen, an assemblage defined by high gravel offshore has two indicator
species – the small wrasse, Oxycheilinus bimaculatus, and the small monacanthid Paramonacanthus
japonicus. These species were often seen on video footage from sites with much marine plant growth.
The most distinctive assemblage, in terms of number of indicator species and spatial restriction,
occurred in the deep Capricorn Channel northward into deeper lagoonal waters below 49 m. The
indicator species there were from a variety of functional groups: the small schooling
Paramonacanthus filicauda, which was restricted almost solely to sites south of Cape Upstart in
BRUVS footage; the benthic microcarnivore Nemipterus theodorei, again mainly a southern species;
the ambush predator Saurida_grp; the mobile semi-pelagic predator Seriolina nigrofasciata; and the
demersal omnivorous triggerfish Abalistes stellatus.
Table 3-5 Hierarchy of nodes in the multivariate tree using spatial and environmental covariates to represent the
location of species assemblages. The number of species with maxima in DLI values are listed for each node.
Terminal nodes are in bold font.
node abbrvn
1
All
split
root
N sites
366
Description
2
Offsh
across>=0.52
170
4
LoGrav
ga.gravel<8.98
84
More than halfway across
the shelf
Low gravel, offshore
8
LGrav-LSand ga.sand<89.94
64
9
Rf-LGravHSand
HiGrav
5
10
Boundary Landmarks
DLI
3
1
0
ga.gravel>=8.98 86
low sand, low gravel,
offshore
High sand, low gravel,
near reefs
High gravel, offshore
along>=0.38
31
Offshore, High gravel,
Offshore, high gravel,
Near reefs offshore between Bowen and 2
southern region
Capricorn-Bunker group
Less than halfway across
6
the shelf
Under 49 metres depth
0
ga.sand>=89.94 20
0
Offshore reefs, mostly southern - to
north
1
0
along<0.38
55
3
HGravCntrlNth
Rf-HGravSthn
Insh
across<0.52
196
6
Shal
gbr.bathy>=-49.3 164
12
Sth-Cntrl
along<0.74
24
HiMud
ga.mud>=16.04 70
48
HMud-Nrshr across<0.28
48
Nearshore, high mud
49
HMud-Lag
across>=0.28
22
Lagoon region, high mud Cape Flattery to Palm Islands
25
LoMud
ga.mud<16.04
55
1
50
HiSil
crs.si.av>=1.56
29
0
11
125
Cape Upstart – Cape Melville
South of Cape Flattery
2
0
0
1
1
100 HSil-LMudHGrav
101 HSil-LMudLGrav
51 LMud-LSil
ga.gravel>=10.29 12
crs.si.av<1.56
26
13
Nthn
along>=0.74
39
0
26
Shal-FN
along<0.84
21
27
Shal-CY
along>=0.84
18
Shallow, far northern
Cape Sidmouth [Princess Charlotte Bay] 0
region
to Cape Flattery
Shallow, tip of Cape York North of Cape Direction to tip
2
7
Deep-Capr
gbr.bathy<-49.3 32
ga.gravel<10.29 17
High gravel, Low mud,
High Silica
Low gravel, Low mud,
High Silica
Low Mud, Low Silica
Cape Flattery to Gladstone
Shoalwater Bay to Broad Sound
0
Whitsundays to Gladstone
0
Cardwell to Curtis Channel
0
Capricorn Channel
5
GBR Seabed Biodiversity
Offshore
Inshore
3-111
LGrav-LSand
Rf-LGrav-HSand
HGrav-CntrlNth
Rf-HGrav-Sthn
HMud-Nrshr
HMud-Lag
HSil-LMud-HGrav
HSil-LMud-LGrav
LMud-LSil
Shal-FN
Shal-CY
Deep-Capr
Figure 3-21 Predicted distribution of 12 fish assemblages (terminal nodes from Table 3-5) as defined by the top
20 explanatory environmental variables and location.
GBR Seabed Biodiversity
3-112
3.2. SINGLE SPECIES, BIOPHYSICAL MODELS AND PREDICTION
The GBR epibenthic sled dataset comprised 70,860 site-by-species records of 4,723 species (OTUs)
from 1,190 sites. The sled biota were from more than 15 phyla of marine plants, invertebrates and
vertebrates (Table 3-6). Taking into account that sorting of hydroids, annelids, crinoids and ascidians
was not completed due to resourcing, this was a highly diverse biota dominated by sponges, molluscs,
crustaceans, fishes, echinoderms, corals, bryozoans and algae (Table 3-6).
The GBR research trawl dataset comprised 39,702 site-by-species records of 3,510 species (OTUs)
from 457 sites. The trawl biota were from more than 12 phyla of marine plants, invertebrates and
vertebrates (Table 3-6). Taking into account that sorting of hydroids, annelids, crinoids, ascidians and
marine plants was not completed due to resourcing, this was also a very diverse biota, in this case
dominated by fishes, sponges, crustaceans, molluscs, echinoderms, corals, bryozoans and algae (Table
3-6).
Table 3-6: Number of OTUs by Phyla sampled by the epibenthic sled and research trawl, and in the merged
dataset.
Phylum
Porifera
Mollusca
Arthropoda
Chordata
Echinodermata
Cnidaria
Bryozoa
Rhodophyta
Chlorophyta
Phaeophyta
Magnoliophyta
Annelida
Brachiopoda
Cyanophyta
Hemichordata
Sled OTUs
952
913
575
563
443
435
361
210
152
47
18
17
15
8
2
Phylum
Chordata
Porifera
Arthropoda
Mollusca
Echinodermata
Cnidaria
Bryozoa
Chlorophyta
Annelida
Rhodophyta
Phaeophyta
Magnoliophyta
Trawl OTUs
993
768
410
401
374
358
117
31
18
17
11
3
Phylum
Porifera
Mollusca
Chordata
Arthropoda
Echinodermata
Cnidaria
Bryozoa
Rhodophyta
Chlorophyta
Phaeophyta
Annelida
Magnoliophyta
Brachiopoda
Cyanophyta
Hemichordata
Merged OTUs
1121
1036
869
589
509
375
321
214
167
54
29
18
15
8
2
For analyses, the sled and trawl datasets were merged. This required the reconciliation of synonymous
OTUs between the two devices and taking all OTUs up to a common taxonomic level at which sorting
and identification was consistent among the different laboratories. This merged reconciled dataset
comprised 121,334 site-by-species records of 5,344 species (OTUs) from both Sled and Trawl
sampled sites (Table 3-6). Of these species, 2,435 were unique to the sled, 1,085 were unique to the
trawl and 1,824 were common to both devices.
The relative sampling rates per swept area of the two devices also differed markedly among different
biota. The swept area of the sled was ~0.03 Ha and that of the research trawl was ~1.02 Ha, but when
samples from both were each scaled to a per Ha basis, the sled had higher sampling rates for most
biota, except crustaceans for which the sampling rates were similar, fishes for which the trawl
sampling rate was >7-fold greater than the sled, and elasmobranchs, which were not well sampled by
the prawn trawl but hardly at all by the sled (Table 3-7).
GBR Seabed Biodiversity
3-113
Table 3-7: Overall total and mean sampling rates (g per Ha) for the major Phyla sampled by the epibenthic Sled
and research Trawl, indicating overall composition and relative catchability. Ratio shows the trawl sampling rate
relative to the sled.
Sled
Group
Green algae
Sponges
Echinoderms
Ascidians
Red algae
Cnidarians
Molluscs
Bryozoans
Seagrasses
Brown algae
Crustaceans
Fishes
Worms
Elasmobranchs
TOTAL
Sum_Wt (g)
35,905,360
32,617,604
24,609,197
18,986,710
16,791,467
16,039,343
12,599,413
12,193,301
5,410,059
3,426,676
2,348,077
1,710,445
1,139,923
3,629
183,781,203
N
1187
1187
1187
1187
1187
1187
1187
1187
1187
1187
1187
1187
1187
1187
Mean
30,249
27,479
20,732
15,996
14,146
13,513
10,615
10,272
4,558
2,887
1,978
1,441
960
3
%
19.54
17.75
13.39
10.33
9.14
8.73
6.86
6.63
2.94
1.86
1.28
0.93
0.62
0.00
Trawl
Group
Fishes
Sponges
Echinoderms
Crustaceans
Molluscs
Cnidarians
Ascidians
Elasmobranchs
Bryozoans
Worms
TOTAL
Sum_Wt (g)
4,789,876
2,235,381
2,040,998
937,739
891,499
507,847
441,504
181,258
79,669
2,344
12,108,115
N Mean Ratio
%
457 10,481
7.27 39.56
457 4,891
0.18 18.46
457 4,466
0.22 16.86
457 2,052
1.04 7.74
457 1,951
0.18 7.36
457 1,111
0.08 4.19
457
966
0.06 3.65
457
397 132.21 1.50
457
174
0.02 0.66
457
5
0.01 0.02
3.2.1. Sled and Trawl samples species richness
As is typical of benthic sampling, most of the species recorded were rare or uncommon, occurring in
only a very small percentage of the sites surveyed. Most of the Sled species (~95%) were recorded in
less than 5% of Sled sites; 1,347 OTUs (~29%) were recorded at only one site, 1,571 OTUs (~33%)
were recorded at only 2-5 sites (Figure 3-22a). Only <1% of the species were prevalent at more than
≥20% of the sites and, of these, only 5 species had a prevalence >50%. Similarly, most of the Trawl
species (~92%) were recorded in less than 8% of Trawl sites; 1,059 OTUs (~30%) were recorded at
only one site, 1,213 OTUs (~35%) were recorded at only 2-5 sites (Figure 3-23a). Only ~2.5% of the
species were prevalent at more than ≥20% of the sites and, of these, only 2 species had a prevalence
>50%. The implications for analysis were that a relatively small proportion of the biota were abundant
enough for analyses: only 850 species occurred at more than 25 sites, which was considered adequate
for developing biophysical models for predicting broad-scale distributions without over-fitting.
There was an average of 59.5 ± 44.2 (s.d.) species per Sled site, ranging from 1 to 268. Ordering of the
most diverse sites produced a sigmoid curve (Figure 3-22b). About 50% of the sites had high species
richness (≥50 species per site), ~43% had moderate richness and only <7% had relatively low richness
(≤10 species). Similarly, there was an average of 86.7 ± 30.9 (s.d.) species per Trawl site, ranging
from 6 to 193. Ordering of the most diverse sites produced a sigmoid curve (Figure 3-23b). More than
90% of the sites had high species richness (≥50 species per site), ~9% had moderate richness and only
<1% had relatively low richness (≤20 species). While the Sled had greater total species richness, it was
more variable than the Trawl, which consistently sampled a representative number of species though it
accumulated fewer species in total.
There were some clear spatial patterns in species richness of the Sled samples (Figure 3-24). Areas of
high richness included the offshore Halimeda beds north of Princess Charlotte Bay and near Lizard
Island, the Halophila spinulosa seagrass beds near the Turtle Island Group and in the Capricorn
Region, the mixed algal-seagrass beds in the mid-shelf off Townsville extending north almost to
Cairns, high current areas near Broad Sound/Shoalwater Bay, Torres Strait, Whitsunday Passage and
offshore passages, most outer shelf areas including the Swains. Areas of low richness included most
inshore and muddy areas, with lowest richness in the deep muddy entrance to the Capricorn Channel.
GBR Seabed Biodiversity
3-114
The spatial patterns in species richness of the Trawl samples were similar, though less clear due to the
lower density of sampled sites and difficulty of sampling in more structured habitats (Figure 3-25).
70
300
60
250
Richness (species per site)
Percentage of sites
50
40
30
20
200
150
100
50
10
0
0
0
20
40
60
80
100
0
Species rank-ordered by Abundance (% of 4723)
20
40
60
80
100
Sites rank-ordered by Richness (% of 1190)
Figure 3-22: Patterns of prevalence and richness of 4,723 species at 1,190 Sled stations.
300
70
60
250
Richness (species per site)
Percentage of sites
50
40
30
20
200
150
100
50
10
0
0
0
20
40
60
80
Species rank-ordered by Abundance (% of 3510)
100
0
20
40
60
80
Sites rank-ordered by Richness (% of 457)
Figure 3-23: Patterns of prevalence and richness of 3,510 species at 457 Trawl stations.
100
GBR Seabed Biodiversity
3-115
##
#
### #
# # #
Sled Richness
#
## # #
##
#
1 - 13
###
#####
#
#
## # #
## # ### # # #
#
14 - 23
#
#
#
#
##
#
#
24 - 35
## # #
#
# #
# # #
#
##
36 - 50
##
#
#
#
#
#
#
#
#
#
#
# ## #
## # #
#
##
#
#
#
51 - 67
#
## # # # # ##
###
#### #### #
##
# ## ### #
##
### #
#
#
#
#
#
#
#
#
# # #
##
#
##### #
#
68 - 93
#
#
# ## #
## ##
####
# ####
# # # ##
#
#
#
#
#
#
#
#
#
#
#
## #
# ##
## ## #
### ## ### # ###
#
#
94 - 118
#
#
# #### # # # # # # #
## # ### #
# #
##
### ##
# #
#
119 - 268
#
## # ##
# #
#
#
#
#
#
#
#
#
##
#
#
#
## #
## ## # # # # #
##
# # ##
# ## #
#
#
#
#
#
#
#
#
# #
# # ##
##
# ###
#
# # ## # # ## #
#
# # ###
# ###
####
#
## #
#
# ##
##
#
# ## #
# ## # #
#
#
#
# ##
#
## #
##
#
#
#
#
#
# # #
#
#
#
## #
#
### # #
#### # ##
# ## #
# # ##
### # ###
#
#
## # # # ###
#
#
#
#
#
#
# # # #
##
# ### # ## # #
## # #
#
######### #
#
## #
# # #
## # #
# ##
##
#
#
#
#
#
#
#
#
#
#
#
# ## # #
##### ### # #
## # ## # # #
#### ##
## #
#
#
#
#
#
# #
#
#
# ##
####
#
#
# # #
#
###
# # # # ##
## # #
# #
# # # # ##
####
# # # ##
# # ## ## #
# ## # # ##
# ## # # #
# # # ### #
#
#
#
#
#
#
#
# # ###
# ### ## #
# # ## #
#
#
#
#
# ##
## # #
#
# # ## # # # #
# # # ##
## ##
###
#
#
##
## #
## ### # ## # # # # # # # ### ##
####
#
#
#
#
# ## # # # # # #
# # #### # ## #
#
###
# # # ###
# # #
# #### #
#
## # #
#
#####
##
# ###### ## # ## # # #
# ## # # # ## # #
#
####
## #
# ## ##
### #
#
## ##
#
#
#
#### # # # # # # # # # # ## # ## ##
####
#
# ## #
##
##
## ## # ## #
# ##
#
### #
###
## # # # ## #### ## # #### # # # # ##
#
#
#
# # ##
#
# #
##
## # # #
##
##
#
#
##
## # # # # # ## ### # # ##
##
## #
# ## #
## # # ## # # ####
N
# ##
# ##
# #
## #
#### # #### # #
#
#### #
# ## #
#
#
#
#
# ##
##
#
#
#
#
#
#
####
#
#
##
# # ####
#
## #
W
E
# # #####
### #
##
#
#
#
#
## # # #
#
#
## # # # # #
##
####
## #
# ## #
# #### #
# # ### #
#
# #### #
#
# # # ## # ###
# #
#
# #
S
# #
## ####
#
#####
##
# ##
# # ##
#
#
#
#
#
#
# # #####
#
#
# ##
##
##### #
#
Kilometers
# #
# # # # ## ###
# # ####
#
##
#
#
0
100
200
#
#
#
## #
## # #
#
# # #
# # # ###
#
## # #
# # ## # # # # #
##
###
#####
#### # ## ###
#
### #
##
#
####
## ## # # #
#
# ##
100
Figure 3-24: Species richness from epibenthic Sled data by location in the GBRMP.
#
# #
# #
#
#
#
#
#
#
#
#
#
##
#
##
# #
#
## #
##
#
##
# # #
## # #
##
#
##
#
# ##
#
#
#
#
#
# ###
# ##
#
#
#
# #
# # #
#
#
# ##
#
#
#
#
## # #
#
# #
#
#
#
#
# ## #
# ### #
##
#
#
# #
#
###
##
#
#
##
#
###
# #
#
##
#### #
#
#
#
#
#
#
#
#
#
##
#
#
##
# #
##
# #
##### ##
#
#
###
###
#
#
#
# # #
#
# #
#
#
# #
##
#
##
#
#
#
#
# # ##
# #
#
E
S
Kilometers
0
100
#
#
#
## #
## # #
#
#
# # ## #
#
#
# ## ##
##
#
#
#
##
#
#
#
#
## #
#
#
200
# #
# #
#
#
#
###
###
#
#
#
#
##
#
#
#
#
#
# ##
#
# ##
# #
##
###
#
## ##
## ## #
#
N
#
#
# #
#
#
#
#
# #
100
#
#
#
## ##
#### # ##
#
W
#
# # #
## #
#
Trawl Richness
#
6 - 54
#
55 - 66
#
67 - 76
#
77 - 86
#
87 - 97
#
98 - 106
#
107 - 123
#
124 - 193
# #
#
#
# ## #
#
#
##
#
#
#
#
#
###
# #
#
#
#
##
#
## ##
####
# ##
#
##
#
#
#
#
#
#
# # #
##
#
#
#
#
# ##
#
#
#
#
#
#
#
#
#
# #
# #
#
#
# ## #
##
## # # # #
#
# ##
#
#
#
#
#
# #
# #
## #
### # #
# # ## ##
# ##
#
## #
#
#
#
#
# ###
#
#
# ##
#
#
#
#
Figure 3-25: Species richness from research Trawl data by location in the GBRMP.
#
##
# #
#
#
# ###
#
GBR Seabed Biodiversity
3-116
3.2.2. Single species models (M Browne & R Pitcher)
Single species distribution models and maps were generated for the 851 species that were observed at
>25 sites. Species observed at less than 25 sites were considered to possess inadequate sample power
to adequately estimate the array of measurement, temporal, physical and spatial effects considered in
this study. Figure 3-26 provides just one of many possible examples of a single-species data / model /
map summary, in this case for the relatively prevalent species Class: Actinopterygii, Family:
Platycephalidae, Elates ransonnetii. In the upper left-hand side of the plot there are notes on the basic
survey counts and weights. In this case, a total of 14.9 kg of the species was sampled with the
scientific trawl at 94 locations and 100 g with the sled at 5 sites. In this case, the data from both the
trawl and sled samples was included in the modelling.
A number of circles have been plotted on the distribution map to indicate the relative size of the raw
sample weights at each of the survey sites. The area of the circle is directly proportional to the quantity
of biomass observed, but the relative scale of circles was necessarily different for each species. The
table on the right hand side of the map provides a key to interpret the biomass catch circles. The first
column indicates the percentile of the biomass-only data set, the second indicates the corresponding
circle size, and in the final column is the corresponding biomass. Note that the biomass circle overlay
was provided to complement the model information, and does not indicate any modelling on the data,
and are not standardized in any way (e.g. in order to compensate for covariate effects).
Figure 3-26: Example of a single species distribution map, for the Platycephalid fish, Elates ransonnetii.
GBR Seabed Biodiversity
3-117
Information about the model is provided in each single species map in three ways. Firstly, model
estimates over the entire GBR region are shown. As stated previously, the model estimates reflect an
inference based on the pattern of observed presence / biomass, considering the physical and spatial
relationships, while controlling for temporal / measurement effects. Since the generated model-map is
independent of temporal effects (such as time of year) it might correctly estimate a high prevalence in
areas where nothing was actually caught, if this absence can be well explained by (for example) season
and the physical and spatial properties of that area in question are similar to areas where the species
was actually found to be abundant.
In the bottom left corner of the map is a list of variables chosen by the stepwise variable selection
procedure with BIC criterion. For both stages, biomass and presence, the list of included variables is
provided. The +/− signs indicate either a positive or negative relationship between the variable and
either the probability of presence or log-biomass. In the example shown in Figure 3-26, the variable
+GA_MUD (indicating percentage mud fraction) appears twice in the presence / absence model, with a
positive linear effect and a negative quadratic effect (indicated by -I(GA_MUD^2)). This indicates
that there is a non-linear relationship between mud fraction and species presence — the probability of
observing this species first increases and then decreases with larger mud fraction. The P-AUC and
Dev.Ratio relate to model performance and are described below.
3.2.2.1. Estimates of model performance: AUCs and ROCs
The confidence that should be placed in the modelled distribution map is naturally related to the
degree of model fit. Two measures of model performance are provided on the left hand side of each
map just below the species code. They are Presence / Absence: Area Under the Curve (P-AUC) and
Biomass: Deviance Ratio (Dev.Ratio). These measures of model performance relate to the two stages
of the modelling procedure and capture how well the model estimated whether a species is likely to be
present or not, and the biomass given that the species was present. Both measures are calculated by
comparing model estimates on the sites that were visited with the actual catch on those sites. It was
somewhat difficult to quantify robustly the quality of the fit of the biomass component of the models
because the data was distributed approximately log-normally and standard measures such as linear
correlation do not apply. The deviance ratio, or relative deviance explained, was calculated as 1 –
residual deviance / null deviance of the biomass models, which yielded a standardised variance ratio
statistic, interpretable similar to that of a correlation as the proportional reduction in deviance
explained by the biomass stage of the model. However, this statistic is intended as a rough guide to
performance only.
The P-AUC was a little more difficult to describe concisely, but is a popular and well-regarded
method for threshold-independent assessment of classifier performance. The P-AUC is calculated
from the Receiver Operating characteristic Curve (ROC) which itself is a non-parametric summary of
how a classifier responds given a list of target decisions (in this case actual observed presence-absence
or 0-1 data), and an associated estimator output (in this case, the probability of presence model
estimate p). The presence-absence data may be ranked so that estimate p is ordered from largest to
smallest. The ROC is generated by starting at the bottom left corner of a [0,1] box and proceeding
through the ordered presence-absence data. Each time a ‘presence’ is encountered the curve moves up
by 1/N, each time an absence is encountered the curve moves right by 1/M, where N and M are the
number of presences and absences, respectively. A well-performing presence-absence model will have
the ‘presences’ grouped towards the beginning of the list ordered from largest to smallest probability,
and therefore tend to move up before moving right towards [1,1]. The area under the ROC (i.e. AUC)
will therefore tend toward 1 as performance becomes perfect. A poorly performing estimator will
move approximately diagonal from [0,0] to [1,1] and will therefore tend have an area of approximately
0.5. Figure 3-27 displays the ROC for the species in the previous distribution map example. The ideal
predictor of presence-absence given the available variables is located at the intersection of the red
lines, indicating that it would correctly predict this species to be present at 91% of the sites at which it
was actually observed, and incorrectly estimate it to be present at 4% of sites at which it was not
GBR Seabed Biodiversity
3-118
actually observed. By standard interpretation guidelines of ROC and AUC, this is usually considered
rather good estimation performance.
0.6
0.4
0.0
0.2
Sensitivity
0.8
1.0
The median AUC observed for all species was 0.844, which indicates that overall, the models were
relatively successful in explaining the presence or absence of species given the explanatory variables.
The 25% and 75% quantiles of the AUC measure for all species were 0.785 and 0.894, indicating that
reasonable to good performance was obtained on at least 75% of species considered. However, it is
important to note that there was considerable variation in model performance: the presence
distributions of some species were estimated very well, while that of others could not be explained
adequately given the explanatory variables. Therefore, the model and associated map for each
individual species should be evaluated. For the estimates of biomass where present, the median
relative deviance explained was 0.315, with the lower and upper quartiles (25%, 75%) being 0.136 and
0.482 respectively, but for 132 species none of the deviance could be explained by the given models
for biomass where present. Thus, summarizing over all species, while it was possible to explain some
of the variation in biomass, a significant proportion of variability could not explained given the input
data. Although the two stages in the modelling procedure are qualitatively different and cannot be
directly compared, it may be argued that the presence / absence of a species in an area can be
estimated with more certainty than the likely biomass observed, given that it is present.
0.0
0.2
0.4
0.6
0.8
1.0
1 - Specificity
Figure 3-27: ROC curve for presence-absence estimation of Actinopterygii: Elates ransonnetii. As noted in the
previous figure, this ROC has an AUC of 0.97
If desired, it is possible to weight the vertical ROC step size by another variable, about which
information was available. In this case, weighting the ‘presence’ step sizes (going vertical on the
ROC) by the actual biomass of the species measured at that site was considered. Instead of each
vertical step being equal (1/N), the vertical steps were weighted proportional to the biomass observed
at that site. Considering only the sites at which a species was actually observed, if a presence-absence
model was more likely to correctly classify a site with relatively high biomass than one with a
relatively low biomass, then the weighted AUC (P_AUCW) would be larger than the unweighted
AUC. It may be argued that high biomass sites should be more likely to be closer to the centre of
species preferred habitat range, then the P_AUCW should be higher than the AUC for any given
species. From this point of view, comparing the P_AUCW to the AUC represents an independent
validation of model efficacy, since the biomass ‘weightings’ were not incorporated in any way to the
construction of the presence-absence models.
GBR Seabed Biodiversity
3-119
Figure 3-28 compares weighted versus unweighted AUCs for all species. Overall, the mean P_AUCW
(0.851) was significantly higher than the mean AUC (0.837), with t = 4.7. This means that over all
species, sites with high observed biomasses were relatively more likely to be correctly classified as a
‘presence’ than sites with low observed biomass. This was in accordance with expectations and was
taken as an independent indication that the presence / absence models were performing properly. This
can be explored further as noted from Figure 3-28 that the P_AUCWs tended to be relatively higher
when the AUC was relatively high (e.g. > 0.9). Of the low to moderately highly effective models
(those with an AUC < 0.9), the P_AUCW > AUC 61% of the time. Of the highly effective models
(AUC > 0.9), the P_AUCW > AUC 81% of the time. Finally, it can demonstrated explicitly that
models with high AUC will tend to have a higher P_AUCWs by transforming both of these
logistically distributed variables so to have approximately normal distributions via the logit link:
⎛ x ⎞
g ( x ) = log ⎜
⎟
⎝ 1− x ⎠
and modelling the transformed P_AUCW as a linear function of the transformed AUC. We establish
that the slope of g(P_AUCW) with respect to g(AUC) is 1.24 (SD= 0.027) which is significantly
higher than the null hypothesis (i.e. that the slope = 1), with t = 8.59. Thus, not only do P_AUCW tend
to be higher than AUC overall, but stronger models tend to have relatively higher P_AUCW. This was
again consistent with the view that high biomass sites ought to conform, on average, more strongly
with characteristics associated with species presence. Stronger models should better reflect the
underlying relationship of covariates to biomass, and this explains why the differential between AUC
and P_AUCW should be stronger for stronger models. In summary, the overall strong presence /
absence modelling results, as well as the meta-analyses conducted on the AUC and P_AUCW
statistics, show quite strong support for the modelling effectiveness.
1
0.95
0.9
0.85
Weighted
0.8
0.75
0.7
0.65
0.6
0.55
0.5
0.5
0.55
0.6
0.65
0.7
0.75
Unweighted
0.8
0.85
0.9
0.95
1
Figure 3-28: Scatter plot of the weighted versus unweighted AUCs for all species.
Having demonstrated that weighted AUC provided a more sensitive diagnostic of model performance
against sampled biomass, the performance of all species model fits was examined for both the
weighted AUC of the presence model and the deviance ratio of the biomass model at all sites (Figure
3-29). The presence model weighted AUC (P_AUCW) was >0.75 for 699 of 838 species modelled
GBR Seabed Biodiversity
3-120
(~83%) with a median of 0.87, which represents good agreement between the actual samples and the
model predictions for the majority of species. The deviance ratio was >0.3 for 439 of 838 species
modelled (~52%) with a mean of 0.32, which represents reasonable agreement between the actual
sample biomass and the model predictions for many species, though 15% of models could do no better
than the grand mean where present. There was some correspondence between the presence model
weighted AUC and the biomass model deviance ratio (p=0.001, Figure 3-30). However, there was no
indication that frequency of occurrence had a strong influence on model performance (Figure 3-30).
100
100
0.10
0.10
80
0.08
0.08
60
60
0.06
40
0.06
40
0.04
20
0
0.5
0.02
0.6
0.7
0.8
P_AUCW
0.9
0.00
1.0
0.04
20
Proportion per Bar
Frequency
80
0.02
0
0.00
0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0
Deviance ratio
Figure 3-29: Frequency distributions of species biomass distribution model performance diagnostics for the
presence model weighted AUC (P_AUCW) and for the biomass model relative deviance explained (Deviance
ratio). The median is indicated by the dashed vertical lines.
0.9
0.8
Deviance ratio
0.7
0.6
0.5
0.4
0.3
0.2
0.1
0.0
0.5
0.6
0.7
0.8
P_AUCW
0.9
1.0
Figure 3-30: Relationship between species biomass distribution model performance diagnostics for presence
model weighted AUC (P_AUCW) and for biomass model Deviance ratio. The medians are indicated by the
dashed lines. Symbol colour indicates species frequency: least frequent=dark blue to most frequent=red.
GBR Seabed Biodiversity
3-121
The fit of the biomass model was typically less that that of the presence model, largely because the
biomass model could be fit only on sites where a species was present — less than 60 sites for half the
species (but >25 sites, the lower limit for modelling). For similar reasons, the biomass stage of the
modelling often did not select any physical covariates (intercept only for 131 species, plus device only
in 205 cases, plus temporal variables only in 33 cases), leading to ‘flat’ biomass predictions given
presence (thus patterns in biomass distribution were determined solely by the probability of presence
and average biomass where present). Only rarely did this occur with the more powerful presence stage
model; physical covariates were not selected in only 11 cases leading to ‘flat’ probability of presence
predictions.
Thus, while the majority of species were modelled satisfactorily, not all species were seen to have a
good relationship with the available physical environment or spatial covariates with the consequence
that their broader distribution beyond actual presence at sampled sites could not be estimated
adequately. To illustrate, high, poor and median model performance, a selection of examples is
provided below.
Examples of distribution maps of some species with higher performing models are shown in Figure
3-31. The fish Kanekonia queenslandica has P_AUC=0.88 and dev.ratio=0.80; another four species
have diagnostics greater than these. The decapod crustacean Solenocera pectinata has P_AUC=0.92
and dev.ratio=0.73; another nine species have diagnostics greater than these. The gastropod
Xenophora cerea has P_AUC=0.91 and dev.ratio=0.68; another 25 species have diagnostics greater
than these. The fish Paramonacanthus filicauda has P_AUC=0.88 and dev.ratio=0.65; another 45
species have diagnostics greater than these.
Examples of distribution maps of some species with among the poorest performing models are shown
in Figure 3-32. The bivalve Corbula fortisulcata had P_AUC=0.69 and dev.ratio=0; eight other
species had both diagnostics lower/equal to these. The bryozoan Synnotum spp had P_AUC=0.66 and
dev.ratio=0.12; three other species had both diagnostics lower than these. The Asteroid starfish
Goniasteridae spp had P_AUC=0.64 and dev.ratio=0; one other species had both diagnostics
lower/equal to these. The model for the small ray Dasyatis leylandi selected no physical or spatial
covariates and was a flat prediction with P_AUC=0.52 and dev.ratio=0 — its pattern of occurrence did
not correspond consistently with any physical or spatial covariate; no species had diagnostics lower
than this species.
Examples of distribution maps of some species with typical model performance are shown in Figure
3-33. The crab Portunus rubromarginatus has P_AUC=0.89 and dev.ratio=0.38; the urchin
Prionocidaris bispinosa has P_AUC=0.83 and dev.ratio=0.39; the fish Priacanthus tayenus has
P_AUC=0.85 and dev.ratio=0.37; and the bryozoan Celleporaria spp has P_AUC=0.85 and
dev.ratio=0.30. In general about half the model fits were poorer than these and about half were
stronger.
GBR Seabed Biodiversity
(a) Actinopterygii: Kanekonia queenslandica
(b) Crustacea: Solenocera pectinata
(c) Gastropoda: Xenophora cerea cf
(d) Actinopterygii: Paramonacanthus filicauda
Figure 3-31: Model distribution maps of selected species with higher performing diagnostics.
3-122
GBR Seabed Biodiversity
(a) Bivalvia: Corbula fortisulcata
(b) Gymnolaemata: Synnotum spp
(c) Asteroidea: Goniasteridae spp
(d) Chondrichthyes: Dasyatis leylandi
Figure 3-32: Model distribution maps of selected species with among the poorest performing diagnostics.
3-123
GBR Seabed Biodiversity
(a) Crustacea: Portunus rubromarginatus
(b) Echinoidea: Prionocidaris bispinosa
(c) Actinopterygii: Priacanthus tayenus
(d) Gymnolaemata: Celleporaria spp
Figure 3-33: Model distribution maps of selected species with median performing diagnostics.
3-124
GBR Seabed Biodiversity
3-125
3.2.3. Selected single species distribution maps
With model distribution maps available for more than 800 species, it was not possible to present them
all in this report. When particular species become of interest due to, for example, outcomes of the risk
assessment (Section 3.7.2) then maps are presented where appropriate. In this section selected maps
are presented to illustrate a variety of contrasting species distribution patterns with respect to some of
the important physical environment covariates and between species within genera. Other distribution
maps not presented in this report, as well as information about distribution of less frequently occurring
species that were not modelled, can be made available for agreed purposes if required.
The first series of distribution maps show divergent distributions against key physical environment
covariates. Figure 3-34 shows two species with positive relationships for muddy sediments (a fish
Nemipterus hexodon and a bivalve Anadara ferruginea cf) and two species with negative relationships
for muddy sediments (a brittle star Euryale asperum and a Gastropod Conus ammiralis).
Figure 3-35 shows two species with positive relationships for benthic irradiance (green algae Caulerpa
taxifolia and Halimeda bikensis) and two species with negative relationships for benthic irradiance (a
fish Lepidotrigla calodactyla and a crustacean Myra eudactyla). It was not surprising that benthic
irradiance arises as an important covariate for marine plants, which form important vegetated
structural habitats that appear to support a high biodiversity (Figure 3-24).
Figure 3-36 shows two species with positive relationships for seabed current stress (a gorgonian
Echinogorgia sp3 and a sponge Callyspongia sp23) and two species with negative relationships for
seabed current stress (a holothurian Stichopus ocellatus and a crab Charybdis truncata). Gorgonians
and other sessile epibenthic fauna (e.g. sea whips, soft corals and sponges) often formed seabed
gardens in medium-high (but not extreme) current areas where harder substratum was often exposed
and the currents facilitate feeding by these animals. Low current areas usually had fine sediment and a
completely different biota.
Figure 3-37 shows two species with relationships for shallow seabed (a fish Pseudorhombus arsius
and a green algae Caulerpa serrulata) and two species with relationships for deep seabed (a
stomatopod Quollastria gonypetes and a shrimp Carid sp4931).
The following series of distribution maps show divergent distributions of different species within the
same genera. The threadfin bream, Nemipterus furcosus, had a strong northerly distribution extending
south to the Swains, whereas Nemipterus theodorei had strong southerly distribution centred in the
Capricorn Channel extending north past Cairns (Figure 3-38ab). The crab, Portunus tenuipes, had low
abundance in the southern GBR and was distributed primarily from off Cape Upstart north along the
midshelf into the far northern GBR — Portunus sanguinolentus occurred almost exclusively on the
innershelf south of Shoalwater Bay with a few scattered records from coastal areas further north
(Figure 3-38cd). The flounder, Pseudorhombus argus, was distributed in the northern GBR, primarily
the far north midshelf, with a few innershelf records further south — Pseudorhombus dupliciocellatus
occurred primarily in the southern GBR, primarily on the mid/outer shelf south of Cardwell, with a
few records further north (Figure 3-39ab). The gastropod, Strombus campbelli, occurred in sandy
inner-shelf areas along the length of the GBR— Strombus dilatatus occurred in sandy outer-shelf areas
throughout the GBR (Figure 3-39cd). The cardinal fish, Apogon fasciatus, occurred in gravelly areas
on the fringes of the Shoalwater Bay and Broad Sound high current area as well as muddy areas in the
deeper Capricorn Channel and innershelf elsewhere— Apogon timorensis occurred primarily in
carbonate sandy areas along the outer-shelf (Figure 3-40ab). The flatfish Cynoglossus sp4 occurred
primarily in four sandy mid/outer-shelf areas, the far north, central, Swains and Capricorn —
Cynoglossus kopsi occurred primarily on the inner-shelf with lower abundance in the central and
Capricorn sections (Figure 3-40cd). These few examples indicate how different the distribution
patterns of closely related species can be and demonstrate the importance of identifying specimens to
species level in order to understand spatial patterns of biodiversity and develop predictive models of
distributions.
GBR Seabed Biodiversity
(a) Actinopterygii: Nemipterus hexodon
(b) Bivalvia: Anadara ferruginea cf
(c) Ophiuroidea: Euryale asperum
(d) Gastropoda: Conus ammiralis
Figure 3-34: Model distribution maps of selected species with positive and negative affinities for mud.
3-126
GBR Seabed Biodiversity
(a) Chlorophyceae: Caulerpa taxifolia
(b) Chlorophyceae: Halimeda bikensis
(c) Actinopterygii: Lepidotrigla calodactyla
(d) Crustacea: Myra eudactyla
Figure 3-35: Model distribution maps of selected species with positive and negative affinities for benthic
irradiance.
3-127
GBR Seabed Biodiversity
(a) Anthozoa: Echinogorgia sp3
(b) Demospongiae: Callyspongia sp23
(c) Holothuroidea: Stichopus ocellatus
(d) Crustacea: Charybdis truncata
3-128
Figure 3-36: Model distribution maps of selected species with positive and negative affinities for seabed current
stress.
GBR Seabed Biodiversity
(a) Actinopterygii: Pseudorhombus arsius
(b) Chlorophyceae: Caulerpa serrulata
(c) Crustacea: Quollastria gonypetes
(d) Crustacea: Carid sp4931
Figure 3-37: Model distribution maps of selected species with affinities for shallow and deep bathymetry.
3-129
GBR Seabed Biodiversity
(a) Actinopterygii: Nemipterus furcosus
(b) Actinopterygii: Nemipterus theodorei
(c) Crustacea: Portunus tenuipes
(d) Crustacea: Portunus sanguinolentus
Figure 3-38: Model distribution maps of selected species within genera having contrasting distributions.
3-130
GBR Seabed Biodiversity
(a) Actinopterygii: Pseudorhombus argus
(b) Actinopterygii: Pseudorhombus dupliciocellatus
(c) Gastropoda: Strombus campbelli
(d) Gastropoda: Strombus dilatatus
Figure 3-39: Model distribution maps of selected species within genera having contrasting distributions.
3-131
GBR Seabed Biodiversity
(a) Actinopterygii: Apogon timorensis
(b) Actinopterygii: Apogon fasciatus
(c) Actinopterygii: Cynoglossus sp4
(d) Actinopterygii: Cynoglossus kopsi
Figure 3-40: Model distribution maps of selected species within genera having contrasting distributions.
3-132
GBR Seabed Biodiversity
3-133
3.3. SPECIES GROUPS CHARACTERIZATION AND PREDICTION
(M Browne & R Pitcher)
3.3.1. Characterization and Prediction Model performance
Figure 3-41 shows the overall cluster structure produced by the clustering algorithm. Note that the
hierarchical nature of the dendrogram allows for the definition of a range of possible subsets of the
data. Choosing the level of the dendrogram at which to ‘cut’ so as to agglomerate all nodes below into
a single cluster may be done with respect to the cost function, shown on the y axis. Further
information may be gained by considering silhouette plots for a range of candidate clusterings, as the
number of groups is increased as the critical ‘cut height’ (on the y axis) is lowered.
From a traditional clustering point of view (Kaufman and Rousseeuw 1990) the diagnostics for
‘objective clusters’ show relatively few strong clusters. However, the purpose here does not require a
clustering this kind of objective basis. The objective here was small groups of species with estimated
physical distributions that were strongly correlated to make modelling their aggregated distribution a
useful and insightful exercise for the purposes of the project. In this sense the clustering tools used do
give a useful basis for achieving this. It is not claimed that the resulting groups have any stronger
objective basis than this utilitarian one.
10
0
5
height
15
20
Another way of viewing this is to note that single species models form a limiting case for modelling
groups of species, that is, the case when each group consists of just one species. Our aim here is to
move away from this limiting case in a meaningful way by combining species which appear to have
correlated distributions and ‘borrowing strength’ from their combined data to obtain a clearer picture
of the properties of this approximately common distribution pattern.
Figure 3-41: Cluster dendrogram of the single species estimates illustrating the hierarchical cluster structure
determined by Ward’s method.
GBR Seabed Biodiversity
3-134
The high measurement uncertainty, the relative sparsity of data, and the large number of distinct
species seen in this study was typical of marine biological sampling in general, and especially for
places of high species diversity such as the GBR. These considerations motivate the aggregation of
species providing that the group models were sufficiently representative of their member species
models. In deciding the extent of aggregation to perform, it was decided to set a criterion that the
resulting intra-cluster species estimate correlation at sites be >.5. Aggregating into 38 groups led to an
overall average intra-cluster species correlation of .58. Averaging with respect to groups lead to an
average of r = 0.55, with a range of 0.4 < r < 0.69.
Crustacea
Crustacea
Actinopterygii
Cephalopoda
Actinopterygii
Actinopterygii
Phaeophyceae
Actinopterygii
Echinoidea
Chlorophyceae
Portunus
Thenus
Sorsogona
Sepiidae
Torquigener
Synodus
Dictyotales
Torquigener
Salmacis
Bornetella
rubromarginatus
0.41
australiensis
0.62
tuberculata
0.72
spp
0.77
sp1 (gloerfelt-tarp) 0.82
tectus group
0.87
sp
0.90
cf pallimaculatus
0.93
sphaeroides
0.95
sphaerica
0.97
10 of 22 species
Figure 3-42: Aggregated biomass map and model for an example species-group (“7”). The top 10 of 22 species
is tabulated with cumulative biomass.
The biomass data for each species was aggregated within each of the 38 clusters to create a group
biomass that was then treated in the same way as the individual species. To provide an example,
GBR Seabed Biodiversity
3-135
Figure 3-42 shows the model predictions for a relatively typical species group#7. The individual
species comprising the group are listed in Table 3-8. The mix of high and low estimated biomass
densities illustrate that despite the relatively large number of members, the model is able to specify
with some confidence the areas of high and low prevalence. Compared to individual species maps, the
overall estimated biomass is obviously higher (since the estimate is of the sum of many individual
species). The intensity of the colours is also high compared to most estimates for single species,
reflecting the relatively higher confidence in the estimates (when considered with respect to the mean
fit). The large number of predictor variables in the biomass component of the model is a result of the
data set being comprised of far fewer zero entries than an individual species model. This provides
more data to generate a more sophisticated model of the biomass. Despite the fact that a large number
of species are being modelled simultaneously, the AUC model performance measure in particular was
quite reasonable (0.87). This suggests that the constituent species possess sufficiently similar
biophysical responses so as to permit a relatively effective common model.
Table 3-8. List of species comprising species group 7.
Class
Crustacea
Crustacea
Actinopterygii
Cephalopoda
Actinopterygii
Actinopterygii
Phaeophyceae
Actinopterygii
Echinoidea
Chlorophyceae
Actinopterygii
Genus
Portunus
Thenus
Sorsogona
Sepiidae
Torquigener
Synodus
Dictyotales
Torquigener
Salmacis
Bornetella
Orbonymus
Species
rubromarginatus
australiensis
tuberculata
spp
sp1 (gloerfelt-tarp)
tectus group
sp
cf pallimaculatus
sphaeroides
sphaerica
rameus
Class
Actinopterygii
Gastropoda
Chlorophyceae
Actinopterygii
Crustacea
Cephalopoda
Gastropoda
Chlorophyceae
Actinopterygii
Crustacea
Crustacea
Genus
Calliurichthys
Strombus
Cladophora
Cynoglossus
Dardanus
Sepiadariidae
Volva
Halimeda
Kanekonia
Leucosia
Sicyonia
Species
ogilbyi
vittatus
sp
maccullochi
callichela var
sp5
volva
cuneata
queenslandica
formosensis
rectirostris
3.3.2. Selected species group distribution maps
Species grouped together in this way share approximately similar biophysical responses, and may or
may not reflect some functional association, such as between a species that provides biological
structural habitat and another species that uses that habitat. Average group membership was ~22
species and ranged from 3 to 80 species. Group 10 was one of the larger with 39 species and a
distribution (Figure 3-44) that appeared to be largely coincident with some of the most vegetated areas
in the region (see Section 3.5.2). This group included the largest biomass and number of marine plant
species (14), such as Halophila spinulosa and Halophila ovalis and a dozen other species of green, red
and brown algae. It is possible that some of the dominant fauna in this group, such as Lethrinus
genivittatus and Oreasteridae sp1 may have some functional dependencies on the habitat forming
biota, but this would need further ecological investigation. Group 32 was dominated by another
common algae Caulerpa racemosa along with 21 other species of mostly fish and bryozoans, with a
strong outer shelf distribution pattern. Group 4 species appeared to favour carbonate sand areas in
mid/outer-shelf areas and included the commercial redspot prawn, Penaeus longistylus, the coral
prawn, Trachypenaeus curvirostris, the flounder, Pseudorhombus diplospilus, and 25 other species.
Group 24 was one of the smaller groups with only 12 species dominated by just two species,
Paramonacanthus filicauda and Priacanthus tayenus, and a distribution that favoured low gravel,
intermediate mud areas, particularly in the southern GBR lagoon. Group 23 included the lumped
ascidians and hydroids and Group 18 included the lumped crinoids; three faunal classes which could
not be fully sorted and identified within the scope of the project, coincidentally having similar
distributions. Group 16 was dominated by bryozoans in terms of number of species (47 of 80). Group
37 was dominated by sponges, both in terms of biomass (~99%) and number of species (22 of 29).
GBR Seabed Biodiversity
(a) Group 10
3-136
(b) Group 32
Rhodophyceae
Phaeophyceae
Liliopsida
Actinopterygii
Liliopsida
Phaeophyceae
Bivalvia
Asteroidea
Bivalvia
Phaeophyceae
Lithophyllum
Lobophora
Halophila
Lethrinus
Halophila
Distromium
Plicatula
Oreasteridae
Amusium
Sporochnus
0.26
sp1
variegata 0.44
spinulosa 0.60
genivittatus 0.68
ovalis
0.73
flabellatum 0.77
chinensis cf 0.81
sp1
0.84
balloti
0.87
comosus
0.89
10 of 39 species
(c) Group 4
0.30
ChlorophyceaeCaulerpa
racemosa
Actinopterygii Engyprosopon maldivensis 0.58
Actinopterygii Dactyloptena orientalis
0.69
Echinoidea Echinodiscus tenuissimus 0.76
Actinopterygii Crossorhombushowensis
0.81
Actinopterygii Apogon
capricornis 0.85
Actinopterygii Apogon
9(dg)
0.89
Actinopterygii Dendrochirus brachypterus 0.92
Actinopterygii Apogon
septemstriatus0.95
ChlorophyceaeStruvea
elegans
0.97
10 of 22 species
(d) Group 24
Crustacea Penaeus
longistylus 0.43
Crustacea Trachypenaeus curvirostris0.55
ActinopterygiiPseudorhombus diplospilus 0.61
Bivalvia
Dosinia
histrio cf 0.67
Crustacea Izanami (matuta)inermis 0.72
ActinopterygiiZebrias
craticula 0.77
Bivalvia
Cardita
sp1
0.80
ActinopterygiiSamaris
cristatus 0.83
ActinopterygiiApogon
timorensis 0.86
Cephalopoda Sepiadarium
austrinum 0.88
10 of 28 species
Figure 3-43: Model distribution maps of selected species groups.
ActinopterygiiParamonacanthusfilicauda
0.74
ActinopterygiiPriacanthus
tayenus
0.87
ActinopterygiiLepidotrigla
calodactyla 0.94
ActinopterygiiElates
ransonnetii 0.98
ActinopterygiiSirembo
imberbis
0.99
ActinopterygiiLeiognathus
bindus
0.99
ActinopterygiiCynoglossus
sp kopsi
1.00
Crustacea Calappa
terraereginae1.00
Bivalvia
Paphia
undulata cf 1.00
Bivalvia
Pitar
sp2
1.00
10 of 12 species
GBR Seabed Biodiversity
(a) Group 23
3-137
(b) Group 18
Actinopterygii
Crinoidea
Anthozoa
Ophiuroidea
Ophiuroidea
Gastropoda
Asteroidea
Actinopterygii
Actinopterygii
Echinoidea
GymnolaemataCelleporaria
spp
0.45
Ascidiacea
Ascidiacea
spp
0.63
Asteroidea
Pentaceraster gracilis 0.71
Holothuroidea Pseudocolochirusviolaceus0.77
Holothuroidea Holothuroidea sp30
0.80
Echinoidea
Salmacis
belli
0.83
GymnolaemataTurbicellepora laevis 0.85
Actinopterygii Rhynchostracion nasus 0.87
GymnolaemataCelleporaria
sp1_AIM 0.89
Cephalopoda Photololligo
chinensis0.90
10 of 59 species
(c) Group 16
Pentapodus
Crinoidea
Dendronephthya
Ophiochasma
Ophiopsammus
Fusinus
Luidia
Parupeneus
Onigocia
Salmaciella
paradiseus
0.26
spp
0.50
spp
0.63
stellatum
0.73
yoldii
0.81
colus
0.86
maculata
0.90
heptacanthus 0.93
sp juv/unident 0.94
oligopora
0.95
10 of 23 species
(d) Group 37
0.39
Rhodophyceae Hydrolithon reinboldii
Crustacea
Barnacle
sp1
0.60
Actinopterygii Pentapodus nagasakiensis0.65
Phaeophyceae Sargassum sp
0.69
Ophiuroidea Euryale
asperum
0.73
Anthozoa
Umbellulifera sp1
0.76
Actinopterygii Abalistes
stellatus
0.80
GymnolaemataAmathia
spp
0.82
GymnolaemataEmballothecaspp
0.85
Demospongiae Hyattella
sp2
0.87
10 of 80 species
Figure 3-44: Model distribution maps of selected species groups.
0.38
DemospongiaeOceanapia
sp21
DemospongiaeCinachyrella
sp1
0.63
DemospongiaeClathria (thalysias)vulpina
0.70
DemospongiaeDemospongiae sp6
0.76
DemospongiaeHyattella
intestinalis 0.80
DemospongiaeDemospongiae sp53
0.83
DemospongiaeDemospongiae conglomerate0.86
DemospongiaeFascaplysinopsis sp3
0.89
DemospongiaeFascaplysinopsis sp1
0.92
DemospongiaeDysidea
arenaria
0.94
10 of 29 species
GBR Seabed Biodiversity
3-138
3.4. SITE GROUPS CHARACTERIZATION AND PREDICTIONS
(W Venables & R Pitcher)
3.4.1. Decision tree results
The recursive splitting on the physical variables, to achieve reduction in deviance of the sites BrayCurtis matrix, produced 16 groups with the given stopping criteria. The resulting graphical decision
tree is shown in Figure 3-45. The primary split was at 25% mud fraction, and mud or another sediment
attribute accounted for several other splits, suggesting the importance of sediment grain size
composition in structuring seabed assemblages. Given the correlation between variables and that other
variables may be good surrogates to split at each node, caution is necessary in interpreting the
importance of variables. Nevertheless, percent mud and other sediment attributes were often the most
frequently selected in a wide range of biophysical analyses. Other variables likely to be important
included bathymetry, oxygen variability, current stress, chlorophyll and/or light attenuation (K490),
nutrients and temperature.
Figure 3-45: Recursive decision tree partitioning the sites into 16 groups, corresponding to the terminal nodes.
The labels indicate the split variable and threshold for the group corresponding to the left hand branch in each
case. The distances used were Bray-Curtis dissimilarities on 1/8th root transforms of the predicted site species
biomass data.
GBR Seabed Biodiversity
3-139
While the decision tree did split a Bray-Curtis matrix to group together sites with similar biota,
nevertheless, the splits were on physical variables and so constrained the tree structure and may not
necessarily have represented the structure of biological similarities between the 16 site groups. The
representation of the biological similarities of the site groups, by hierarchical clustering of the BrayCurtis dissimilarities of the medoid sites using Ward's method, demonstrated a similar structure,
particularly between site-groups 1–9 and 10–16, and within site-groups 1-9 (Figure 3-46). However,
the biological similarities of the site medoids within site-groups 10–16 was rather different from the
structure of the decision tree, with groups 10–11 and 12–13 being placed biologically close and 11–12
moderately dissimilar. It is important to note that the GBR seabed assemblages are not distinct, but
have fuzzy gradients of biotic composition and different transformations, distance metrics, and
clustering methods will produce somewhat different results — sometimes transposing some sitegroups across the primary mud split. Nevertheless, low-mud and high-mud site-groups typically were
separated.
Figure 3-46: Dendrogram of biological similarities between the medoids of the site groups, as defined by the
tree Figure 3-45, based on hierarchical clustering of Bray-Curtis dissimilarities using Ward's method.
The set of decision rules for splitting physical variables, defined on data from >1,000 sites, was
straightforward to apply to the remaining >170,000 grids cells of the GBR study area and map the
result (Figure 3-47). Overall, several rather distinct regions with little if any assemblage representation
elsewhere were apparent, including: the Capricorn Section, the Capricorn Channel, Shoalwater Bay
and Broad Sound, Swains and Pompey Reefs regions, central/Townsville area, and the northern GBR
from about Hinchinbrook Is/Cairns. The composition of the site-group assemblages is discussed in
more detail below (Section 3.4.3).
GBR Seabed Biodiversity
Figure 3-47: Map of predicted distributions of 16 seabed assemblages (site groups clusters).
3-140
GBR Seabed Biodiversity
3-141
3.4.2. Species affinity groups
The dendrogram of distances between species with respect to their affinities for the site-group
assemblages is shown in Figure 3-48, for all 839 taxa for which prediction models were possible. The
distance that defines the 12 species groups for purposes of this analysis is shown by cutting the
dendrogram at the dashed red line in the diagram.
The first left branch represents those species with higher affinity for muddier site-groups. Of the right
main branch, the smaller left branch represents those species with higher affinity for high current sitegroups, and the rightmost branches represent those species with higher affinity for the remaining
(coarser, vegetated and/or offshore) site-groups.
The relative biomasses of the 12 species affinity groups across the 16 site-group assemblages are
shown in Figure 3-49. While the distribution and composition of the site-group assemblages are
described in more detail below, there appear to be about 4–7 relatively distinct mixtures over the 16
site-group assemblages on the basis of summary patterns of relative biomasses of about five sets of the
12 species affinity groups (Figure 3-49). Most distinctive was site-group #8, which was dominated by
species affinity group #L — most similar was site-group #6. The coastal/inshore muddy site-groups
11, 12, 13 were characterised primarily by species affinity groups E, G, C. Another, less distinct group
of (deeper) muddy sites comprised site-groups 10, 15, 16 and 14 and 5, which were characterised
primarily by species affinity groups D, F, H. The remaining site-groups were less distinctly structured;
nevertheless, site-groups #1–2 were most similar, characterised primarily by the low biomass of
species affinity groups G, E. Next were site-groups #4 & 7, characterised primarily by the species
affinity groups K, I, B. Site-group #6 was somewhat similar to #1, 7, 8. Site-groups #9, 3 had the
highest abundance of affinity group J.
Figure 3-48: Dendrogram for species, defining clusterings based on inter-species distances that reflect affinities
between species and site-groups. The red line shows a cut-off that defines the 12 groups used in this analysis.
The dendrogram was constructed using Ward’s method.
GBR Seabed Biodiversity
3-142
Figure 3-49: Plot of relative biomass of 12 species affinity groups (labeled A–L) across the 16 site-group
assemblages mapped in Figure 3-47.
3.4.3. Description of site-group assemblages
Site cluster, or species Assemblage#1, occurred in low mud, deep, low gravel, low current stress areas
(Figure 3-45) represented by the red areas in Figure 3-47, primarily on the outer shelf off Townsville.
No particular species group stood-out in terms of relative biomass associated with assemblage#1 areas,
except perhaps A — and lack of G, E, J. Similarly, no particular species had very strong affinities for
GBR Seabed Biodiversity
3-143
assemblage#1; those most aligned were Asteroidea: Poraster superbus (species affinity group A),
Crustacea: Portunus argentatus (group A) and Gastropoda: Atys cylindricus cf (L).
Assemblage#2, occurred in low mud, deep, low gravel, slightly higher current stress areas (Figure
3-45) represented by the dark blue areas primarily on the outer shelf of the southern GBR.
Assemblage#2 areas were also characterised by lack of species groups, but groups A, K, B, D had
some affinity and slightly higher relative biomass associated with Assemblage#2. A number of species
had moderately strong affinities for assemblage#2; those most aligned were: Gymnolaemata:
Retiflustra spp (A), Hippopetraliella magna cf (A), Tetraplaria immerse (K), Nellia tenella cf (D),
Echinoidea: Temnopleuridae sp2_QMS (A), Ophiuroidea: Euryale asperum (B), Crustacea:
Parthenope sp32091 (K), Takedana eriphioides (A), Myrine kesslerii (A), Actinopterygii: Samaris
cristatus (A), Hippocampus queenslandicus (A), Engyprosopon maldivensis (B), Kanekonia
queenslandica (B), Bivalvia: Cardita sp1 (A).
Assemblage#3, occurred in low mud, deep, high gravel areas (Figure 3-45) represented by the dark
green areas in Figure 3-47, primarily on the outer shelf offshore from the Whitsundays and Mackay,
with a patch occurring on the shelf edge offshore from Townsville. Species groups J, H, K stood-out in
having higher affinity and/or relative biomass associated with assemblage#3 areas. At the species
level, some of strongest affinities were seen for assemblage#3; those most aligned were:
Gymnolaemata: Adenifera armata (H), Hippomenella avicularis (H), Celleporaria spp (J),
Euthyrisella obtecta (J), Macropora spp (K), Sinupetraliella spp (H), Calcarea: Clathrina sp1 (H),
Demospongiae: Demospongiae sp10 (H), Demospongiae sp26 (J), Callyspongia sp26 (J),
Demospongiae sp27 (H).
Assemblage#4, occurred in very low mud, deep areas (Figure 3-45) represented by the purple areas in
Figure 3-47, primarily on the outer shelf of the Capricorn Section of the GBR. Species groups K, B, I
stood-out in having higher affinity and/or relative biomass distributed in assemblage#4 areas. A
number of species had moderate affinities for assemblage#4; those most aligned were: Gymnolaemata:
Orthoscuticella spp (K), Arachnopusia spp (K), Scuticella plagiostoma (K), Actinopterygii:
Ambiserrula jugosa (K), Demospongiae Xenospongia patelliformis (K).
Assemblage#5, occurred in intermediate low mud, deeper areas (Figure 3-45) represented by the
orange areas in Figure 3-47, on the flanks of the Capricorn Channel in the southern GBR. Species
group D had the highest affinity and relative biomass distributed in assemblage#5 areas. At the species
level, some of weakest affinities were seen for assemblage#5; those most aligned were:
Actinopterygii: Lepidotrigla calodactyla (D), Foraminifera: Discobotellina biperforata (K).
Assemblage#6, occurred in low mud, shallower, low current stress areas (Figure 3-45) represented by
the yellow areas in Figure 3-47, primarily on the inner/mid shelf off Townsville from Cape Upstart to
Fitzroy Island, the inner shelf of the Mackay Coast, the mid shelf from Lizard Is to Turtle Is, and
smaller scattered areas of the outer shelf and Swains. Although these were some of the more vegetated
areas in the GBR, no particular species group stood-out in terms of affinity with assemblage#6, but L
had elevated relative biomass. Again, at the species level, some of weakest affinities were seen for
assemblage#6; those most aligned were: Actinopterygii: Scorpaenopsis furneauxi (G), Chlorophyceae:
Caulerpa taxifolia (L), Avrainvillea sp1 (A), Udotea flabellum (A), Gastropoda: Strombus campbelli
(G), Dolabella sp1 (I), Liliopsida: Halophila spinulosa (I).
Assemblage#7, occurred in somewhat similar areas as #6 (Figure 3-45), but primarily on the inner
shelf of the Capricorn Coast indicated by the brown areas in Figure 3-47. These were also some of the
more vegetated areas in the GBR, and species groups I, K, G stood-out in terms of affinity and/or
relative biomass associated with assemblage#7. A number of species had moderate affinities for
assemblage#7; those most aligned were: Crustacea: Portunus sanguinolentus (K), Asteroidea:
Oreasteridae sp1 (I), Holothuroidea: Holothuria sp2 (I), Actinopterygii: Ambiserrula jugosa (K),
Suezichthys gracilis (K), Aploactis aspera (I), Paramonacanthus otisensis (I), Rhodophyceae:
Chondrophycus sp1 (K), Phaeophyceae: Padina sp. (K).
Assemblage#8, also occurred in low mud, shallower, low current stress areas (Figure 3-45)
represented by the pink areas in Figure 3-47, primarily on the outer shelf of the far northern GBR,
extending south inside the barrier to about Lizard Island; the coastal silica sand strip from Shelbourne
Bay north was also included. Some of the most extensive Halimeda banks occurred in some of these
areas. Species group L was a stand-out in having higher affinity and relative biomass associated with
GBR Seabed Biodiversity
3-144
assemblage#8 areas; followed by species groups H, I and C, A. A number of species had moderatehigh affinities for assemblage#8; those most aligned were: Chlorophyceae: Dictyosphaeria cavernosa
(L), Halimeda gigas (L), Halimeda opuntia (L), Caulerpa sertularioides (L), Caulerpa serrulata (L),
Actinopterygii: Pseudorhombus argus (L), Gastropoda: Terebellum terebellum (L), Atys sp1,
Echinoidea: Breynia desorii (L), Peronella lesueuri (H), Anthozoa: Heteropsammia cochlea (L).
Assemblage#9, occurred in low mud, shallower, high current stress areas (Figure 3-45) represented by
the grey areas in Figure 3-47, primarily in the vicinity of Broad Sound and Shoalwater Bay, but
including offshore narrow inter-reef channels and the approaches to Torres Strait. Some of the most
extensive epibenthic faunal gardens occurred in some of these areas. Species group J stood-out clearly
in terms of relative biomass associated with assemblage#9, followed by F, G. At the species level,
some the strongest affinities were seen for assemblage#9; those most aligned were: Actinopterygii:
Centrogenys vaigiensis (G), Inegocia harrisii (G), Crustacea: Metapenaeopsis novaeguineae (G),
Paradorippe australiensis (G), Hyastenus elatus (J), Gymnolaemata: Micropora angusta cf (J),
Bivalvia: Arca navicularis (J), avellana_MTQ (J), Asteroidea: Goniasteridae sp5 (J), Demospongiae:
Callyspongia schultzi (F), Anthozoa: Melithaea sp2 (J), Mopsella sp2 (J).
Assemblage#10, occurred in high mud, low nitrate variability, low chlorophyll areas (Figure 3-45)
represented by the aqua-blue areas in Figure 3-47, primarily in the mid-Lagoon of the Whitsunday
region and re-occurring on the outer shelf offshore from Hinchinbrook to Cape Flattery. Some of the
most barren habitats occurred in some of these areas, although the sled and trawl revealed significant
biodiversity. Species groups E, D had the highest relative biomass in this assemblage, but none had
high affinity. At the species level, weak affinities were seen for assemblage#10; those most aligned
with were: Actinopterygii: Nemipterus hexodon (E), Crustacea: Cloridina chlorida (E), Iphiculus
spongiosus (E), Demospongiae: Fascaplysinopsis sp3 (H), Bivalvia: Anadara ferruginea cf (E).
Assemblages #11, 12, 13, occurred in high mud, low nitrate variability, high chlorophyll areas (Figure
3-45), with 12 being in the muddiest habitats and 13 in the most turbid, represented respectively by the
salmon, grey-blue and pale-green areas in Figure 3-47, primarily in shallower inner shelf areas
extending from ~Whitsunday Islands to Torres Strait, and broader north of about Cairns. Again, some
of the most barren habitats occurred in some of these areas, although the sled and trawl revealed
significant biodiversity. Species groups E, G were associated with these assemblages. At the species
level, the affinities were moderate-weak (weakest for assemblage#12); those most aligned with
assemblage #11 were: Actinopterygii: Scolopsis taeniopterus (E), Terapon theraps (E), Leiognathus
leuciscus (E), Crustacea: Charybdis truncata (E), Crustacea: Metapenaeus endeavouri (E); those most
aligned with assemblage #12 were: Crustacea: Cryptolutea arafurensis (E), Actinopterygii: Saurida
argentea/tumbil (E); and those most aligned with assemblage #13 were: Actinopterygii: Leiognathus
splendens (E), Leiognathus moretoniensis (G), Gerres filamentosus (G), Tripodichthys angustifrons
(E), Terapon puta (E), Apogon poecilopterus (E), Crustacea: Trachypenaeus anchoralis (G),
Crustacea Metapenaeus ensis (E), Penaeus semisulcatus (E).
Assemblages #14, 15, 16 occurred in high mud, high nitrate variability areas (Figure 3-45)
represented by the deeper pale-khaki, pale-yellow and pale-pink areas in Figure 3-47. Assemblage #14
was located along the far outer shelf offshore from Hinchinbrook to Cairns; #15, 16 occurred near the
entrance of the Capricorn Channel. Some of the most barren habitats occurred in some of these areas,
although the sled and trawl revealed significant biodiversity. No species group or individual species
had a clear association with Assemblage #14 in terms of relative biomass or affinity. Species groups
D, F showed somewhat higher relative biomass in assemblages 15, 16, and D had some affinity with
15. Affinities were weak at the species level; those most aligned with assemblage#15 were: Crustacea:
Arcania heptacantha (D), Solenocera choprai (D), Actinopterygii: Upeneus moluccensis (D),
Lepidotrigla calodactyla (D), Elates ransonnetii (E); and those most aligned with assemblage#16
were: Crustacea: Solenocera choprai (D), Actinopterygii: Upeneus moluccensis (D).
GBR Seabed Biodiversity
3-145
3.5. VIDEO HABITAT CHARACTERIZATION AND PREDICTION
The data from the towed video camera, entered in real-time in the field ("tappity") and from postanalysis of random frames in the laboratory, was used to provide information on seabed habitats — in
the form of maps summarizing the data and as a statistical characterisation.
3.5.1. Seabed substratum
The real-time tappity data showed that inshore areas along much of the length of the GBR were
muddy or silty (Figure 3-50, Figure 3-52a), and comprised terrestrial sediments. Typically, with
distance across the shelf, the substratum becomes sandier or even coarser (Figure 3-50, Figure 3-52d),
and comprised of carbonate of biological origin. In offshore areas, coralline outcrops, reefs and shoals
may occur in deep areas between emergent coral reefs (Figure 3-52k).
The strong tidal current areas among the dense reef matrix offshore in the central-southern GBR were
rubbly or stony, with rocks and limestone bedrock often exposed. Much of the rubble in these areas is
formed by encrusting bryozoans (Figure 3-52o). The inshore strong tide areas of Broad Sound and
Shoalwater Bay are also very coarse or rocky. Between these lies the Capricorn Channel, a wide area
of GBR lagoon with a very silty and muddy seabed. The south-eastern entrance to this channel is the
deepest area on the GBR shelf, at 100-130 m.
The Capricorn Region, the southernmost part of the GBR, is typically sandy right across the shelf. It
lies at the northern end of the Great Sandy Region, just beyond Fraser Island, the source of large
quantities of terrestrial sand.
3.5.2. Seabed biological habitat
The majority of the seabed in the GBR was devoid of visible biological habitat attached to the surface
of the substratum; however, most of these areas were bioturbated indicating the activity of animals in
the sediments (Figure 3-51, white and grey areas; Figure 3-52 cf. bm). In offshore sandy areas with
medium currents, crinoid feather stars were sometimes extremely abundant on the seabed (Figure
3-52n).
Marine plants form dominant cover over large areas of the GBR shelf (Figure 3-53, e.g. Figure 3-52f).
A long band of mixed algae and patchy seagrass (primarily Halophila spinulosa) occurs along the
mid-shelf off Townsville (Figure 3-51). Dense beds of H. spinulosa also occur over much of the shelf
in the Capricorn region (Figure 3-52e) as well as around the Turtle Is Group in the central northern
GBR (Figure 3-54). Vast banks of Halimeda algae (Figure 3-52g) occur just inside the outer barrier
reef near Lizard Is, also in the central northern GBR, as well as in the far northern GBR (Figure 3-53).
These Halimeda banks may be up to 15 m thick, comprised of the deposited carbonate skeletons of
these algae. Other types of algae, including crustose corallines, are prolific along some sections of the
outer shelf in water up to 80-100 m deep (Figure 3-53, Figure 3-52eh).
Epibenthic fauna such as alcyonarian soft corals, whips & gorgonians and sponges may occur in
patchy gardens (Figure 3-51, Figure 3-52ij, Figure 3-57, Figure 3-56, Figure 3-55), particularly in
medium-strong current areas corresponding to red areas in Figure 2-14; bryozoans are important in
similar areas (Figure 3-58). Hard corals may grow on some of the hard ground areas, typically
offshore (Figure 3-52, Figure 3-59).
GBR Seabed Biodiversity
3-146
Tappity_substratum
SoftMud
Silt(Sandy-Mud)
Sand
CoarseSand
SandWaves/Dunes
Rubble(5-50mm)
Stones(50-250mm)
Rocks(>250mm)
Bedrock/Reef
N
W
E
S
Miles
0
100
100
Figure 3-50: Map of the distribution of seabed substratum types summarized as percent of transect length
observed by towed video camera.
Tappity_Biohabitat
NoBiohabitat
Bioturbated
AlcyonariansSparse
AlcyonariansMedium
AlcyonariansDense
WhipGardenSparse
WhipGardenMedium
WhipGardenDense
GorgonianGardenSparse
GorgonianGardenMedium
GorgonianGardenDense
SpongeGardenSparse
SpongeGardenMedium
SpongeGardenDense
HardCoralGardenSparse
HardCoralGardenMedium
HardCoralGardenDense
LiveReefCorals
Flora
Algae
Halimeda
Caulerpa
Seagrass
BivalveShellBeds
SquidEggs
TubePolychaeteBeds
N
W
E
S
100
Miles
0
100
Figure 3-51: Map of the distribution of broad biological seabed habitat types summarized as percent of transect
length observed by towed video camera.
GBR Seabed Biodiversity
3-147
a) Muddy inshore seabed, with filefish.
b) Bioturbated silty inner shelf seabed
c) Rippled sand in high current area
d) Coarse outer shelf sediment with soft corals
e) Seagrass (Halophila spinulosa) bed
f) Dense algal bed (Caulerpa)
g) Halimeda bank
h) Ulva growing on coralline algae at shelf edge
i) Soft corals in strong current channel
j) Gorgonian garden on hard ground
k) Shoal ground in deep water
l) Solitary coral and algae near shelf edge
m) Large bioturbation mounds in offshore sand n) Crinoids on sand in strong current area
o) Bryozoan rubble in strong current channel p) Scoured rocky seabed in extreme current area
Figure 3-52: Photos of some example habitat types observed by towed video camera.
GBR Seabed Biodiversity
3-148
90%
50%
25%
12%
6%
3%
<1%
Frame_Algae
Acetabularia
Branching
Bushy
Caulerpa
Coralline
Digitate
Encrusting
Filamentous
Foliose
Globular
Halimeda
Padina
Prostrate
Sargassum
Straplike
Tufted
Udotea
Ulva
N
W
E
S
Miles
0
100
100
Figure 3-53: Map of the distribution and cover of conspicuous genera and other morpho-types of algae.
60%
30%
15%
8%
4%
<1%
N
W
E
S
100
Miles
0
100
Figure 3-54: Map of the distribution and cover of morpho-types of seagrasses.
Frame_Seagrass
Compound
Ovoid
Strapform
Syringodium
GBR Seabed Biodiversity
3-149
50%
25%
12%
6%
3%
<1%
Frame_Sponges
Barrel
Branching
Bushy
Club
Cymbastela
Digitate
Encrusting
Fan
Foliose
Globular
Ianthella
Ircinia
Lobate
Nodules
Plate
Prostrate branching
Vase
Xestospongia
N
W
E
S
Miles
0
100
100
Figure 3-55: Map of the distribution and cover of conspicuous genera and other morpho-types of sponges.
Frame_Gorgonians
Antipatharian
Cirriphathes
Branching
12%
Bushy
Ctenocella
6%
Fan
Feathery
3%
Foliose
1.5%
Iciligorgia
<1%
Solenocaulon bare
Solenocaulon encr
Whip
25%
N
W
E
S
100
Miles
0
100
Figure 3-56: Map of the distribution and cover of conspicuous genera and other morpho-types of gorgonians.
GBR Seabed Biodiversity
3-150
30%
15%
8%
4%
2%
<1%
Frame_Alcyonarians
Branching
Bushy
Digitate
Foliose
Lobate
Lobophytum
Nephtheid
Sarcophyton
Sinularia
Xenia-like
Cavernulina
Pteroides
Virgularia
N
W
E
S
Miles
0
100
100
Figure 3-57: Map of the distribution and cover of conspicuous genera and other morpho-types of alcyonarian
soft-corals.
60%
30%
15%
8%
4%
<1%
#
N
W
E
S
100
Miles
0
100
Figure 3-58: Map of the distribution and cover of morpho-types of bryozoans.
Frame_Bryozoans
Branching
Bushy
Encrusting
Foliose
Nodules
GBR Seabed Biodiversity
3-151
20%
10%
5%
3%
2%
<1%
Frame_HardCorals
Branching
Digitate
Encrusting coral
Foliose coral
Free living coral
Massive coral
Plate coral
Submassive coral
Vase
N
W
E
S
100
Miles
0
100
Figure 3-59: Map of the distribution and cover of morpho-types of hard corals.
3.5.3. Statistical characterization and prediction (W Venables & R Pitcher)
The medoid rpart algorithm, using the Manhattan (Bray-Curtis) distance metric applied to the vessel
biological data (with the three densities of epibenthos grouped), produced the tree shown in Figure
3-60 — a result which appeared to capture more of the known habitat distributions, compared with the
other tree metrics. The improvement (proportional reduction) in deviance achieved by any node is
reflected by the height of the vertical lines descending from the node. Hence the most primary and
most substantial cut is on the sediment variable GA_MUD with sites for which this value is less than
15.51% proceeding down to the left hand node and the remainder to the right hand node. In general,
the labelling of each interior node indicated the cases going to the left hand node and the complement
to the right. The labelling of the terminal nodes is with an arbitrary group number used only for
identification purposes in the following descriptions
Experience with the mvpart algorithm using both the Euclidean and Hellinger metrics suggested that
a complexity, in those cases, of about 6 or 7 groups was justified on the basis of cross-validation. The
stopping rules of the rpart algorithm terminated the Manhattan (Bray-Curtis) tree at 9 groups, a
similar though perhaps somewhat more optimistic number compared to the others and possibly with
cross-validation of Manhattan if that was available.
Information on the biological habitat character of these 9 groups could be obtained from either the
group medoids, or nearly equivalently, the group centroids. The latter are shown as horizontal bar
graphs in Figure 3-61 and are described in more detail below. Nevertheless, it is clear from Figure
3-61 that while the biological habitat profiles of some of the groups stand out as different from others,
some are not strikingly dissimilar. For example, there are degrees of similarity between 6 and 7, 3 and
4, and 5 and 9.
GBR Seabed Biodiversity
3-152
Figure 3-60: Recursive partitioning of sites based on the grouped vessel biological cover proportions, the
Manhattan (Bray-Curtis) distance metric and the medoid partitioning algorithm.
3.5.3.1. Effectiveness of the substratum video data as predictors of biology
The check on the effectiveness of the external physical predictors, by including the substratum profile
proportions as additional predictors and noting whether any were chosen in advance of the external
physical variables. The resulting tree was mostly identical to the previous one (Figure 3-60), with the
following exceptions: the previous clusters 3 and 4 were divided into three, with splits on the video
substratum variable Bedrock.Reef and M_BSTRESS, and group 5 was further split into two on the
video substratum variable, Rocks.250mm. The added predictors are only selected at a low level in
the tree and the effect is to isolate similar subgroups within groups already present rather than to
isolate different groups.
3.5.3.2. Predictions to the GBR grid
The predictions of node membership on the entire GBR covariate grid, based on the splits of the
Manhattan (Bray-Curtis) tree, are shown in Figure 3-62, colour coded according to the scheme in the
attached legend and numbered according to the tree (Figure 3-60).
With reference to the cluster profiles (Figure 3-61), the Manhattan (Bray-Curtis) tree (Figure 3-60) and
the distribution map (Figure 3-62), the clusters were broadly characterised as follows: Cluster 1
represented the most barren seabed type, almost entirely bare and bioturbated with very little
GBR Seabed Biodiversity
3-153
biohabitat, distributed in muddy areas of the inshore and midshelf and the deep end of the Capricorn
Channel. Cluster 2 was also very barren, with some bioturbation and very little epibenthos or algae,
distributed in muddy-sand areas of the southern midshelf and far north. Cluster 3 had significant
patches of epibenthic gardens separated by tracts of bare seabed, distributed in low mud higher current
areas, primarily in the southern GBR. Cluster 4 was similar to cluster 3, but with more algae, and
distributed in similar low mud higher current areas with higher benthic irradiance, in both the southern
and far northern GBR. Cluster 5 represented mostly bioturbated and bare seabed with a little algae and
seagrass algal habitat distributed over much of the shelf in the central and northern sections of the
GBR. Cluster 6 represented seagrass and algal habitat distributed along much of the inner half of the
shelf in the southern Capricorn section of the GBR. Cluster 7 represented similar patchy seagrass and
algal habitat distributed along the mid-shelf from Cape Upstart to Innisfail. Cluster 8 represented
much of the Halimeda algal habitat, as well as some other algae and epibenthos, distributed in various
patches along the outer shelf, including the Halimeda banks inside the ribbon reefs near Lizard Is and
in the far north. Cluster 9 represented patchy algae (including some Halimeda) with some bioturbation
and a little other biohabitat, distributed primarily in the outer-shelf offshore from Townsville.
Figure 3-61: Mean profiles (centroids) of the 9 site groups as defined by the recursive partitioning algorithm.
GBR Seabed Biodiversity
3-154
Figure 3-62: Map of predictions of group membership to the entire GBR grid. The groups are those from the
medoid algorithm with grouped vessel biological data and Manhattan distances shown in (Figure 3-60).
GBR Seabed Biodiversity
3-155
3.6. ACOUSTICS DISCRIMINATION AND CLASSIFICATION
3.6.1. Wavelet Packet-Based Techniques Applied to Data in the Angular Domain (D H
Smith)
3.6.1.1. Two Class Seabed Classification
This is the simplest classification case on which basic understanding of the data behaviour and applied
techniques is sought before progressing to the more complex cases involving larger numbers of
classes.
3.6.1.1.1 Sand Substratum with and without Biohabitat
For initial study and analysis purposes, sites containing large continuous blocks of each seabed type
have been selected. Table 3-10 indicates a set of 18 sites possessing at least 1,000 consecutive samples
with sand substratum and no biohabitat, while Table 3-11 lists three available sites with a similar
number of (sand, seagrass) data samples.
A single classification experiment for the two class case refers to a given pair of sites from which
training data is extracted and another distinct pair of sites from which a similar quantity of test data is
also extracted. Such site-wise experiments can be considered as local, as opposed to global in the
sense of training and test sets comprising data from multiple sites.
Table 3-9: A sample confusion matrix for the two-class case of sand and seagrass.
Sand
Seagrass
Sand
Seagrass
98.0
5.7
2.0
94.3
Table 3-10: Sites with at least 1,000 consecutive samples of (sand, no biohabitat) seabed type with indices
(6,17).
Site
897
2583
1701
2016
1580
2018
2631
828
2380
1917
2100
1647
833
1939
744
2626
1719
2005
Mean Depth (m)
41
16
51
30
61
33
21
63
24
14
46
42
50
20
42
27
54
26
Once the feature extraction scheme has been established on the training data, feature extraction is
performed on the entire test set and subsequent classification is attempted, starting with just one
feature and progressing to the full original data size. Every test data item is classified into one of the
GBR Seabed Biodiversity
3-156
two classes under consideration, and since the actual classes are known, mis-classification rates can be
calculated to assess the overall procedure. These are derived from the calculated confusion matrix,
which indicates what proportion of each class type is correctly classified, through its diagonal
elements, with rows representing actual classes and columns representing predictions made by the
classifier. The confusion matrix displayed below shows strong diagonal behaviour for both classes,
with 98% of the sand correctly classified and 94.3% of the seagrass correctly classified.
Table 3-11: Sites with at least 1,000 consecutive samples of (sand, seagrass) seabed type with indices (6,21).
Site
2083
2084
2441
Mean Depth (m)
33
31
28
Pairwise combinations of the site data from Table 3-10 and Table 3-11 yielded a total of 54
training/test sets, on which 2916 individual classification experiments were performed, and selected
results are presented. Results for a typical two-class classification experiment are shown in Figure
3-63, where the horizontal axis indicates the number of features used and the three curves represent the
two confusion matrix diagonal elements and the associated overall mis-classification rate as
percentages. Training data for the two seabed classes (sand, no biohabitat) and (sand, seagrass) comes
from sites 1701 and 2441, while test data comes from sites 1580 and 2083, further details of which are
given in Table 3-10 and Table 3-11. Best results in this picture occur just below 40 features, where the
sand is perfectly classified and the seagrass is classified to over 90% accuracy, with overall misclassification rate below 10%. Initial steep activity visible in all three curves produced good results
below 20% mis-classification with less than 10 features, offering a significant reduction from the
original data size of 64.
100
90
Confusion Diagonals/Mis−Class Rate %
80
70
1st Diagonal(Sand)
2nd Diagonal(Seagrass)
Mis−Classification Rate
60
50
40
30
20
10
10
20
30
Feature Dimension
40
50
60
Figure 3-63: Results for a single two-class classification experiment for sand substratum with no biohabitat and
sand substratum with seagrass, calculated with a Tree classifier. Training data is from sites 1701 and 2441 and
test data is from sites 1580 and 2083.
GBR Seabed Biodiversity
3-157
Repeating the single classification experiment represented in Figure 3-63 across all 54 test sets
produced the result in Figure 3-64, which indicates the lowest mis-classification rate with respect to
feature dimension for each test set together with the confusion matrix diagonal elements. Circles and
squares on this plot denote those test sets which are completely distinct from the training set,
containing no data contribution from sites 1701 or 2441. A number of good results below 20% misclassification have been recorded along with some poor results as high as 80% mis-classification rate.
Sand being mis-classified as seagrass is responsible for the poor overall results at some sites, while
seagrass remains well classified across the test set range with strong diagonal behaviour above 90%.
90
Mis−Classification Rate %
80
70
60
50
40
30
20
10
0
3
6
9
12
15
18
21
24
27
30
Test Set Index
33
36
39
42
45
48
51
54
Confusion Matrix Diagonals
100
90
80
70
60
50
40
1st Diagonal(sand)
30
2nd Diagonal(seagrass)
20
3
6
9
12
15
18
21
24
27
30
Test Set Index
33
36
39
42
45
48
51
54
Figure 3-64: Classification results across 54 different test sets for training data from sites 1701 and
2441, calculated with a Tree classifier. Circles and squares indicate test sets containing no
contributions in common with the training set.
The variation in classification performance seen in Figure 3-64 is accompanied by a variation in mean
depth of the sand contribution from 14 to 63 m across the test sets, compared with 51 m mean depth
for the corresponding training set contribution. An alternative view of Figure 3-64, plotting the same
mis-classification rates against the actual depth departure between the training and test set sand
components appears in Figure 3-65, showing a distinct depth divide separating the good and poor
results. Inspection of the associated acoustic data for sand with no biohabitat reveals obvious structural
differences with depth, which are in fact consistent with calculated predictions based on a model
equation derived from underwater acoustic theory (Sternlicht and de Moustier, 2003). Depth variation
between training and test data demands careful attention in the classification process, with pictures
such as Figure 3-65 suggesting a depth partitioning strategy which limits allowable depth mismatches
between training and test data.
Replacing the sand component of the training set with data from site 2005, which has a lower mean
depth of 26 m, and repeating the classification calculations across all 54 test sets produced the misclassification rates displayed in Figure 3-66. This training set produced a more steady behaviour than
that previously observed in Figure 3-64, with a maximum mis-classification rate of just over 60% and
most values residing in a band between 20 and 30%. As with the previous training set, large errors in
some test sets are the result of sand being mis-classified as seagrass.
GBR Seabed Biodiversity
3-158
80
70
Mis−Classification Rate
60
50
40
30
20
10
0
5
10
15
20
Absolute Mean Depth Departure (m)
25
30
35
Figure 3-65: Mis-classification results from Figure 5 plotted against absolute depth difference between the sand
components of the training and test sets, showing a distinct depth divide near 15 m separating the good and poor
results.
Mis−Classification Rate %
60
50
40
30
20
10
3
6
9
12
15
18
21
24
27
30
Test Set Index
33
36
39
42
45
48
51
54
Mis−Classification Rate
60
50
40
30
20
10
0
5
10
15
20
Absolute Mean Depth Departure (m)
25
30
35
Figure 3-66: Tree classification results for (sand, no biohabitat) and (sand, seagrass) seabed types across 54 test
sets with training data from sites 2005 and 2441.
GBR Seabed Biodiversity
3-159
In terms of depth departure between the sand component of the training and test sets, an entirely
different behaviour is observed in Figure 3-66 whereby reasonable results occur across a wide range of
departures, in stark contrast to the previous result of Figure 3-65. The largest mis-classification rate
occurred roughly half way through the interval, and other large values were also seen near 5 m
departure, signaling caution in the future application of any depth partitioning process.
3.6.1.1.2 Different Biohabitats on Sand Substratum
A second two-class classification study involved retaining the sand substratum and attempting to
discriminate between two different biohabitats, namely sponge garden dense, which is available from
five sites as indicated in Table 3-12, and seagrass as in the previous case. A total of 15 different
training/test pairs were available, allowing 225 individual classification experiments.
Computed results for a single classification experiment with training data from sites 2580 and 2441
and test data from sites 2593 and 2084 are displayed in Figure 3-67, including both confusion matrix
diagonal elements and the overall mis-classification rate. An abrupt change were observed just after 40
features, marking a transition to steady behaviour with strong diagonals and corresponding low misclassification rates near 10%. Seagrass results, which reach well above 90%, are slightly superior to
their sponge garden counterparts which remain just below 90%.
Table 3-12: Sites with 1,000 consecutive samples of (sand, sponge garden dense) seabed type with indices
(6,18).
Site
2009
2593
2023
2584
2580
Mean Depth (m)
41
33
35
31
29
100
90
Confusion Diagonals/Mis−Class Rate %
80
70
60
50
40
30
20
10
1st Diagonal(sponge)
2nd Diagonal(seagrass)
Mis−Classification Rate
10
20
30
Feature Dimension
40
50
60
Figure 3-67: Results of a single two-class classification experiment for (sand, sponge garden dense) and (sand,
seagrass) seabed types, generated with a Tree classifier. Training data is from sites 2580 and 2441 and test data
is from sites 2593 and 2084.
GBR Seabed Biodiversity
3-160
Combining the three (sand, seagrass) sites in Table 3-11 with the five (sand, sponge garden) sites from
Table 3-12 yielded a total of 15 training/test pairs for consideration. Repeating the classification
calculations across all 15 test sets with the same training set produced the results summary in Figure
3-68. This indicates minimum mis-classification rates and maximum individual confusion diagonal
elements with respect to feature dimension for each test set. Strong seagrass performance of over 90%
is maintained across all test sets, as in the previous two-class case, with overall mis-classification rates
residing between approximately 10 and 50%. For those test sets that were not well classified, where
sponge garden being mis-classified as seagrass was clearly the dominant source of error.
55
Mis−Classification Rate %
50
45
40
35
30
25
20
15
10
5
3
6
9
12
15
9
12
15
Test Set Index
Confusion Matrix Diagonals
100
90
80
70
60
1st Diagonal(sponge)
2nd Diagonal(seagrass)
50
3
6
Test Set Index
Figure 3-68: Tree classification results across 15 different test sets for training data from sites 2580 and 2441.
Circles indicate test sets containing no contribution in common with the training set. Seagrass is well classified
across the full test set range, while sponge garden undergoes larger variation to be the dominant error source for
those test sets with high mis-classification rates
3.6.1.2. More than Two Seabed Classes
Taking all three of the previously considered seabed classes together now offered an extra dimension
of complexity, to which the feature extraction methods must be subjected as part of the study and
development process. Constructing a training set with data from sites 1701, 2009 and 244, and
performing classification on a test set comprising data from sites 1580, 2593 and 2083 produced the
results in Figure 3-69. This shows all three confusion matrix diagonal elements and the associated misclassification rate against feature dimension. All three seabed types were classified to higher than 80%
accuracy provided enough features are used, with sand returning the best result followed by sponge
garden and seagrass. Abrupt jumps present in the sponge garden and seagrass curves between 30 and
40 features were not seen in the sand curve, which underwent a steady climb to reach above 95%
accuracy. With 50 features the overall mis-classification rate dropped to approximately 15%,
providing an effective dimension reduction of smaller magnitude to that previously observed in the
two-class cases.
GBR Seabed Biodiversity
3-161
Table 3-13: The calculated confusion matrix at feature dimension 50 from Figure 3-69.
Sand
Sponge Garden
Seagrass
Sand
99.1
0.0
13.5
Sponge Garden
0.4
89.4
4.1
Seagrass
0.5
10.6
82.4
The actual computed confusion matrix at 50 features shown above indicates precisely how the sponge
garden and seagrass were mis-classified via the corresponding off-diagonal elements, which do not
appear in Figure 3-69. Inspection of the second row shows the sponge garden error of approximately
10% to result entirely from mis-classification of this biohabitat as seagrass. Total seagrass errors of
just under 20% are divided between sand and sponge garden, with sand taking the major share of
almost 15% and sponge garden taking the remaining 5%. For sand, the very small total error of
approximately 1% is essentially equally split between sponge garden and seagrass.
90
Confusion Diagonals/Mis−Class Rate %
80
70
1st Diagonal(sand)
2nd(sponge)
3rd(seagrass)
Mis−Classification Rate
60
50
40
30
20
10
20
30
Feature Dimension
40
50
60
Figure 3-69: Results for a single classification experiment on three seabed classes comprising (sand, no
biohabitat), (sand, sponge garden dense) and (sand, seagrass), calculated via Linear Discriminant Analysis.
Training data is from sites 1701, 2009 and 2441, and test data is from sites 1580, 2593 and 2083.
Retaining the same training set, additional calculations were carried out on test sets containing all
available data contributions from the same three seabed classes indicated in Table 3-10, Table 3-11
and Table 3-12, with maximum diagonal confusion matrix elements chosen as performance indicators
for each test set. Viewing these values against their respective absolute depth departures between
training and test set contributions under the application of Tree and Linear Discriminant Analysis
classifiers produced the results shown in Figure 3-70, also offering a useful performance comparison.
As in the initial two class case of Figure 3-65, a distinct depth divide has emerged in the (sand, no
biohabitat) results, with a cluster of strong diagonal behaviour occurring below 15 m depth departure,
and poor results below 50% essentially confined to test sites with large depth departures. Recovery of
the training set is also evident from the high values at zero depth departure, when the two sets
coincide, and little performance difference was observed between the two classifiers. In sharp contrast,
very noticeable differences have appeared for the remaining two seabed classes, with Linear
GBR Seabed Biodiversity
3-162
Max Sand Diagonal
Discriminant Analysis proving clearly superior in both cases for all test sets excluding the training set.
Seagrass has been well classified to over 80% accuracy, as seen previously in the two class case, with
similar success for sponge garden on all but one test set where a moderate 60% is achieved. Depth
departures were also considerably lower for these two seabed types, reaching up to approximately 12
m for sponge garden and 5 m for seagrass.
100
50
Tree
LDA
Max Seagrass Diagonal Max Sponge Garden Diagonal
0
0
5
10
15
20
Depth Departure
25
30
35
80
60
Tree
LDA
40
0
2
4
6
Depth Departure
8
10
90
80
Tree
LDA
70
0
0.5
1
1.5
2
2.5
Depth Departure
3
3.5
4
4.5
Figure 3-70: Additional classification results for the three classes (sand, no biohabitat), (sand, sponge garden
dense) and (sand, seagrass), showing maximum confusion matrix diagonal elements against training/test depth
departure, for test sets containing contributions from all available sites.
Continuing with the sand substratum, there was one more biohabitat type with at least 1,000
consecutive data samples available, namely bioturbated, which is indicated in Table 3-14. Adding data
from sites 2191 and 2447 to the previous training and test sets and performing a single classification
experiment gave the four confusion matrix diagonals and associated mis-classification rate curves
displayed in Figure 3-71. The first two classes in this experiment, namely bioturbated and sand, had
immediately distinguished themselves with exceptional performance of above 95% accuracy at feature
dimension of 40, where a clear transition was observed. Bioturbated has further distinguished itself by
its behaviour at low feature dimension, achieving above 90% accuracy with less than 10 features in
comparison to 70% for sand. For the remaining two classes, seagrass reached approximately 65% at
40 features, while sponge garden produced the only poor result of approximately 30% to dominate an
overall mis-classification rate of just above 50%.
To provide a closer look at the mis-classification behaviour and identify how the actual errors were
distributed, the actual confusion matrix at feature dimension 40 from Figure 3-71 is displayed in Table
3-15. Small errors present in the bioturbated and sand results were essentially due to mis-classification
as seagrass and bioturbated respectively, while the relatively large sponge garden error of
approximately 70% was entirely due to mis-classification as seagrass. Errors for seagrass are
dominated by mis-classification as sponge garden, with small, almost equal contributions of less than
10% attributed to both bioturbated and sand.
GBR Seabed Biodiversity
3-163
Table 3-14: Sites with 1,000 consecutive samples of (sand, bioturbated) seabed type (6,5).
Site
2191
2447
2563
735
1847
1594
1743
1600
739
2322
1139
150
181
139
653
Mean Depth (m)
31
30
23
43
24
31
48
28
40
52
42
84
41
70
46
Training 2191/1701/2009/2441, Test 2447/1580/2593/2083
100
90
Confusion Diagonals/Mis−Class Rate %
80
70
60
50
40
30
1st Diagonal(bioturbation)
2nd(sand)
3rd(sponge garden)
4th(seagrass)
Mis−Classification
20
10
10
20
30
Feature Dimension
40
50
60
Figure 3-71: Results for a single classification experiment on four seabed classes, with training data taken from
sites 2191, 1701, 2009 and 2441, and test data from sites 2447, 1580, 2593 and 2083, calculated via a Tree
classifier.
Table 3-15: The calculated confusion matrix at feature dimension 40 from Figure 3-71.
Bioturbated
Sand
Sponge Garden
Seagrass
Bioturbated
97.5
2.2
0.0
7.7
Sand
0.1
97.8
0.0
6.0
Sponge Garden
0.0
0.0
30.7
21.8
Seagrass
2.4
0.0
69.3
64.5
GBR Seabed Biodiversity
3-164
As in the previous three-class case, additional calculations have been performed with the same training
set on a collection of test sets containing all available contributions from each of the four classes under
consideration, with results for both classifiers summarised in Figure 3-72. Bioturbated results have
shown an exceptional performance from Linear Discriminant Analysis, which produced a cluster of
strong diagonal elements above 95% for depth departures up to almost 20 m, and never dropped below
50% across the entire range. By contrast, corresponding Tree results reach as low as 20% at large
depth departures, and as low as 40% below 20 m departure where Linear Discriminant Analysis
excelled. For sand with no biohabitat, a familiar decline in performance with depth departure was
observed, with the Tree classifier showing superiority, particularly at low departures where Linear
Discriminant Analysis gave some poor results below 20%. Performance declines with depth departure
were also recorded for the sponge garden and seagrass, with Linear Discriminant Analysis superior in
the former case and clearly inferior to Tree classification in the latter case.
Tree
LDA
Max Sand Diagonal
Max Bioturbated Diagonal
100
80
60
40
20
10
20
30
40
Depth Departure
Tree
LDA
90
80
70
60
50
40
60
40
20
0
50
Max Seagrass Diagonal
Max Sponge Garden Diagonal
0
Tree
LDA
80
0
0
2
4
6
8
Depth Departure
10
20
Depth Departure
30
90
80
70
60
50
Tree
LDA
40
30
10
0
1
2
3
Depth Departure
4
Figure 3-72: Computed four-class classification results for additional test sets, displayed as maximum confusion
matrix diagonal elements for each of the four seabed classes.
3.6.1.2.1 Different Substrata in the Absence of Biohabitat
With over 250 possible seabed classes present in the original specification, merging of classes will be
necessary in order to produce a smaller, computationally practical set of working seabed classes. A
useful starting point in this direction involves the consideration of selected substrata in the absence of
biohabitat. Accompanying the previously considered sand without biohabitat are the four substrata
types of coarse sand, sand waves/dunes, silt and soft mud, for which sites are available with at least
1,000 consecutive data samples as indicated in Table 3-16, Table 3-17, Table 3-18 and Table 3-19. An
initial classification experiment on these five classes, using training data from sites 1828, 1701, 2458,
2407 and 2163, and test data from sites 2315, 1580, 2750, 1897 and 1940, has yielded the confusion
diagonal elements and mis-classification rate curves shown in Figure 3-73.
GBR Seabed Biodiversity
3-165
Table 3-16: Sites with 1,000 consecutive samples of (coarse sand, no biohabitat) seabed type (2,17).
Site
1828
2315
1644
Mean Depth (m)
55
55
59
Table 3-17: Sites with 1,000 consecutive samples of (sand waves/dunes, no biohabitat) seabed type (7,17).
Site
2458
2750
1758
Mean Depth (m)
20
14
13
Table 3-18: Sites with 1,000 consecutive samples of (silt, no biohabitat) seabed type (8,17).
Site
2407
1897
1728
Mean Depth (m)
34
20
35
Table 3-19: Sites with 1,000 consecutive samples of (soft mud, no biohabitat) seabed type (9,17).
Site
2163
1940
2564
Mean Depth (m)
15
29
17
While the overall mis-classification rates were high (>60%), some noteworthy discrimination
behaviour has taken place, dominated by strong performance of the sand waves/dunes diagonal
element, which clearly distinguished itself by reaching above 90% with just over 40 features. The next
best result belongs to coarse sand, which reached above 70% with less than 25 features, while silt
started poorly at low feature dimension then underwent a very slow climb to finally reach over 60%
with 50 features. Sand did not rise above 50% and soft mud reached just above 40% with less than 10
features before undergoing a decline. Of particular importance here is the actual mis-classification
structure, as indicated by the calculated confusion matrix at feature dimension of 40 (Table 3-20).
In the first row, approximately two thirds of the coarse sand was correctly classified, with misclassification as sand being the principal error contributor, and silt responsible for the remaining
smaller contribution. Confusion between sand and coarse sand persists to a larger extent in the second
row, showing almost equal amounts of the sand being classified as sand and coarse sand, with silt
making up most of the remaining small error contribution. These two rows immediately suggest a
possible merging of the sand and coarse sand substrata types, subject to further testing. Strong
diagonal performance near 90% in the third row belongs to sand waves/dunes, as observed in Figure
3-73, none of which was mis-classified as either sand or coarse sand, with soft mud responsible for
most of the remaining error. In the final two rows, almost 60% of the silt was mis-classified as soft
mud and approximately 70% of the soft mud was mis-classified as silt, indicating strong confusion
between this substrata pair, which can also be seen as potential candidates for merging, subject to
additional testing. Additional tests on these five classes without biohabitat need to be carried out,
followed by further calculations on the same substrata with various biohabitats present.
GBR Seabed Biodiversity
3-166
100
90
Confusion Diagonals/Mis−Class Rate %
80
70
60
50
40
30
20
Coarse Sand
Sand
Sand Waves/Dunes
Silt
Soft Mud
Mis−Classification
10
0
10
20
30
Feature Dimension
40
50
60
Figure 3-73: Confusion matrix diagonals and overall mis-classification rates for a 5-class classification
experiment on selected substrata without biohabitat, generated with Linear Discriminant Analysis. Training data
is supplied from sites 1828, 1701, 2458, 2407 and 2163, with test data from sites 2315, 1580, 2750, 1897 and
1940.
Table 3-20: The calculated confusion matrix at feature dimension 40 from Figure 3-73.
Coarse Sand
Sand
Sand Waves
Silt
Soft Mud
Coarse Sand
67.1
45.0
0.0
0.0
0.0
Sand
25.6
43.8
0.0
0.0
3.5
Sand Waves
0.0
0.0
88.1
7.4
1.1
Silt
6.6
9.9
1.0
33.7
70.7
Soft Mud
0.7
1.3
10.9
58.9
24.7
3.6.2. Canonical Variate Analysis of Acoustic Data (N Campbell & D Devereux)
3.6.2.1. Depth Normalisation
Prior to analysis, the acoustic data were normalised to a constant depth so that signatures could be
compared over the whole range of the data. Figure 3-74 and Figure 3-75 show plots of the original
pelagic data and the depth-normalised data for a shallow sand site and a deeper sand site. The plot
scales for the abscissa have been chosen to provide a good visual match. Since the depth normalisation
involves only a linear transformation followed by nearest neighbour or linear or cubic resampling,
there is strong agreement between the shapes for the original and depth-normalised plots, which is to
be expected.
Figure 3-76 shows that after the linear scaling which arises during the depth normalisation, the main
echo and second echo are essentially aligned for the depth-normalised pelagic data.
GBR Seabed Biodiversity
3-167
Plots of the pelagic data and the bottom data for a shallow site for a sand / algae group in Figure 3-77
show essentially the same shape; there was no apparent loss of features arising from the lower
sampling rate for the pelagic data. There was no change in the shape of the ping response with depth
as a result of the depth normalisation.
a)
b)
Figure 3-74: (a) Plot of the original pelagic data against sample time for a shallow sandy site (depth = 12 m);
and (b) plot of depth-normalised data against sample number for the same site.
a)
b)
Figure 3-75: (a) Plot of the original pelagic data against sample time for a deep sandy site (depth = 87 m); and
(b) plot of depth-normalised data against sample number for the same site.
a)
b)
c)
d)
Figure 3-76: Plot of the depth-normalised pelagic data against sample number for sand sites for a range of
depths: (a) 12 m; (b) 20 m; (c) 50 m; and (d) 87.5 m.
GBR Seabed Biodiversity
a)
3-168
b)
Figure 3-77: (a) Plot of the original pelagic data against sample time for a sand/algae group; and (b) plot of the
corresponding bottom data for the same group.
Initial discriminant analyses were conducted on data for all groups of more than 100 contiguous echo
responses from the same nominal substrate and biohabitat classes. The initial discriminant analyses of
the 4519 such groups from 117 classes, without regard to the class labels, showed groupings into
clusters. However, closer examination of the resulting plots showed that the differences between
groups for the same habitat cover class were often as large as the differences between the cover
classes. Extensive examination of the data suggested that even after depth normalisation there were
depth-related differences in the shapes of the profiles, particularly at shallower depths (< 25 m), which
required additional adjustment.
3.6.2.2. Depth Adjustment
Figure 3-74 and Figure 3-75 for the depth-normalised data for a shallow sand site and a deeper sand
site show obvious shape differences related to depth. There are also obvious and marked differences in
the profile responses for a shallow site and a deep site for two groups of >100 contiguous pings for
class index 617 (sand:no biohabitat) in Figure 3-78 (see also the plots in Figure 3-81).
a)
b)
Figure 3-78: Plot of the depth-normalised pelagic data against sample number for (a) a shallow sandy site (depth
= 12 m); and (b) a deep sandy site (depth = 70 m).
Plots of the response averaged over depth normalised sample numbers 110 – 112 against depth in (a)
of Figure 3-79 and Figure 3-80 show a marked initial decrease until a depth of 20 – 25 m, followed by
a much more gradual decrease; there appeared to be a systematic decrease in the width of the first echo
with increasing depth.
Figure 3-74 and Figure 3-75 suggest that there was no obvious bias in the depth normalisation
procedure; curves at shallower depths are not scaled differently to those at greater depths.
GBR Seabed Biodiversity
3-169
Figure 3-76 shows that after the linear scaling which arises during the depth normalisation, the main
echo and second echo are essentially aligned for the depth-normalised pelagic data.
a)
b)
Figure 3-79: (a) Plot of the response averaged over sample numbers 110 – 112 against depth for all groups of
>100 contiguous pings for all classes; and (b) plot of the difference in response for sample numbers 110 – 112
against depth for the same data.
Figure 3-77 shows that the pelagic data and the bottom data show essentially the same shape, and that
there was no apparent loss of features arising from the lower sampling rate for the pelagic data.
Hence it seems reasonable to conclude that there is an obvious change in the shape of the ping
response with depth; for shallower depths, the response for the first echo is much broader than it is at
greater depths. This does not appear to be an artefact of the depth normalisation procedure, or of the
use of the lower-resolution pelagic data.
The plots in Figure 3-82 show reasonably linear relationships of the response at sample numbers 113
and 115 with inverse depth. Note that the slopes of the relationships for the sand class are greater than
those for the mud class.
a)
b)
Figure 3-80: Plot of the response averaged over sample numbers 110 – 112 against depth for (a) class 617 (sand)
and (b) class 217 (coarse sand).
GBR Seabed Biodiversity
a)
b)
c)
d)
e)
f)
h)
i)
3-170
Figure 3-81: Plots of the depth-normalised pelagic data against sample number for sand sites for depths ranging
from 12 m (a) to 85 m in (h).
GBR Seabed Biodiversity
a)
b)
c)
d)
e)
f)
3-171
Figure 3-82: Plot of the response for the peak-aligned data against 100/depth for (a) sand class 617 for sample
number 113; (b) sand class 617 for sample number 115; (c) silt class 817 for sample number 113; (d) silt class
817 for sample number 115; (e) mud class 917 for sample number 113; and (f) mud class 917 for sample number
115.
GBR Seabed Biodiversity
3-172
Table 3-21: Intercept, slope and r2 values for regressions of the response at various sample numbers for the
peak-aligned depth-normalised pelagic data on 1/depth.
Time
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
intercept
-73.1729
-74.6114
-79.4014
-86.2605
-78.9563
-31.8452
-17.1195
-23.0639
-31.6964
-40.4079
-49.1043
-56.4091
-60.0027
-59.0950
-56.6323
-54.5358
-53.3320
-52.8377
-52.8877
-53.3637
-54.1319
-55.2195
-56.5394
-57.8636
-59.3030
-60.7373
-62.1345
-63.3999
-64.5797
-65.5028
-66.2810
-67.0064
-67.4714
-67.8956
-68.2446
-68.4714
-68.7402
-69.0259
-69.2736
-69.6341
slope
-0.2186
0.3777
2.4815
6.4673
8.8755
2.5684
0.2755
1.0738
2.0394
2.8949
3.7135
4.2980
4.3289
3.7213
3.0226
2.4534
2.0885
1.8351
1.6525
1.5095
1.3700
1.2790
1.2252
1.1609
1.1419
1.1353
1.1370
1.1364
1.1588
1.1619
1.1759
1.2074
1.1793
1.1448
1.0849
0.9746
0.8539
0.7077
0.5300
0.3543
r²
0.0057
0.0131
0.2387
0.5882
0.6630
0.1629
0.0111
0.1408
0.3753
0.5301
0.6329
0.7250
0.7817
0.7837
0.7570
0.7020
0.6375
0.5619
0.4833
0.4132
0.3375
0.2816
0.2494
0.2238
0.2247
0.2349
0.2540
0.2710
0.2928
0.3019
0.3082
0.3199
0.3056
0.2902
0.2668
0.2290
0.1880
0.1403
0.0846
0.0382
3.6.2.3. Discriminant Analyses of the Depth-adjusted Data
The initial discriminant analyses of the 4500+ groups, without regard to the class labels, showed that
the differences between groups for the same cover class were as large as the differences between the
cover classes, and that there was no consistent discrimination between any of the cover classes.
Closer examination of the data indicated that there were depth-related differences in the shapes of the
profiles, particularly at the shallower depths (< 25 m). There is also a small but reasonably consistent
change in the position of the peak with depth.
GBR Seabed Biodiversity
3-173
In this section, the mean profiles are peak-aligned to remove one effect of depth. An attempt is made
to remove the obvious effect of depth on the shape of the group means by regressing the echo response
values against 1/depth.
Another correction considered was to remove the so-called “size” effect, and focus on differences in
shape. This is done by calculating a row mean (a simple measure of the average area under the curve),
and subtracting this mean from all the values across an echo response curve; dividing by this mean is
also advocated to equalise the area under the echo response curve.
3.6.2.3.1 Regressions of Peak-Aligned Responses against 1/ Depth
Table 3-21 lists the intercept, slope and r2 values for the regressions of the peak-aligned depthnormalised pelagic data on 1/depth. There is a reasonably strong relationship for sample numbers 110
– 120, but other parts of the ping have little relationship with 1/depth.
3.6.2.3.2 Canonical Variate Analyses of the Peak-Aligned and Depth-Adjusted Data
The first five canonical roots for the canonical variate analysis of the 4519 groups from 117 classes,
without regard to the class labels, for the peak-aligned and depth-adjusted data are 6.329, 4.225, 1.129,
0.6273 and 0.3591.
The first five canonical roots for the peak-aligned, depth-adjusted and row-corrected data are 4.332,
1.787, 0.6470, 0.3938 and 0.2862.
Both plots show two clusters, in one case along the second canonical variate for the peak-aligned and
depth-adjusted data, and along the first canonical variate for the peak-aligned, row-corrected and
depth-adjusted data.
a)
b)
Figure 3-83: Plot of the group means for the first two canonical variates for the canonical variate analysis of the
4519 groups from 117 classes, without regard to the class labels, for (a) the peak-aligned and depth-adjusted
data; and (b) the peak-aligned, row-corrected and depth-adjusted data.
A plot of the first canonical vector for the peak-aligned and depth-adjusted data in Figure 3-84 shows
a clear summing or “size” effect – all the values are positive, and very roughly the same.
Plots of the subsequent canonical vectors for the peak-aligned and depth-adjusted data, and the
canonical vectors for the peak-aligned, row-corrected and depth-adjusted data — Figure 3-85, Figure
3-86 and Figure 3-87 — show obvious visual similarities (remember that reversing the sign of a vector
has no effect on the relative positions of the groups along an axis).
GBR Seabed Biodiversity
3-174
Figure 3-84: Plot of the first canonical vector for the canonical variate analysis of the 4519 groups from 117
classes, without regard to the class labels, for the peak-aligned and depth-adjusted data.
a)
b)
Figure 3-85: Plot of the canonical vectors for the canonical variate analysis of the 4519 groups, for (a) the peakaligned and depth-adjusted data; and (b) the peak-aligned, row-corrected and depth-adjusted data.
a)
b)
Figure 3-86: Plot of (a) the second canonical vector for the canonical variate analysis of the 4519 groups for the
peak-aligned and depth-adjusted data; and (b) the first canonical vector the peak-aligned, row-corrected and
depth-adjusted data.
a)
b)
Figure 3-87: Plot of (a) the third canonical vector for the canonical variate analysis of the 4519 groups for the
peak-aligned and depth-adjusted data; and (b) the second canonical vector the peak-aligned, row-corrected and
depth-adjusted data.
GBR Seabed Biodiversity
3-175
For the usual analysis of the peak-aligned and depth-adjusted data, it seems clear that the first
canonical vector is reflecting mainly size differences between the echo response profiles. Shape
differences are evident in the successive canonical vectors for the peak-aligned and depth-adjusted
data, and for all the canonical vectors for the peak-aligned, row-corrected and depth-adjusted data.
The plot of the group means for the first two canonical variates for the canonical variate analysis of the
4519 groups from 117 classes, without regard to the class labels, for the peak-aligned, row-corrected
and depth-adjusted data in Figure 3-83 (b) showed two obvious clusters separated along the first
canonical variate. However, plots for the group means from the individual classes in Figure 3-88 show
that for all of the 117 classes, there are groups which fall within the two main clusters, and that there is
no obvious separation between any of the classes.
a)
b)
c)
d)
Figure 3-88: Plot of the group means for the first two canonical variates for the canonical variate analysis of the
4519 groups from 117 classes, without regard to the class labels, for the peak-aligned, row-corrected and depthadjusted data for (a) sand class 617; (b) seagrass class 621; (c) silt class 817; and (d) mud class 917.
GBR Seabed Biodiversity
3-176
3.6.2.3.3 Directed Contrasts for the Peak-Aligned, Depth-Adjusted and Row-Corrected Data
Directed contrasts can be defined between classes of interest to calculate those canonical variates
which best separate the specified classes. Figure 3-89 shows a plot (top-left) of the group means for
the two canonical variates which result from contrasts between class 617 vs classes 618 and 621, and
class 618 vs class 621.
Canonical variate plots for the groups for the classes involved in the contrast calculations — 617 (topright), 618 (bottom-left) and 621 (bottom-right) — show that, for two of the three classes, there are
groups which fall within the two main clusters. There is considerable overlap between class 617 and
classes 618 and 621, and some overlap between classes 618 and 621. There is evidence of two clusters
in the canonical variate group plots for most classes.
a)
b)
c)
d)
Figure 3-89: Plot of the group means for the first two canonical variates for the canonical variate analysis of the
4519 groups which result from contrasts between class 617 vs classes 618 and 621, and class 618 vs class 621,
for the peak-aligned, row-corrected and depth-adjusted data for (a) all groups for all classes; (b) sand class 617;
(c) sponge class 618; and (d) seagrass class 621.
The regressions of the echo response values on 1/depth were designed to remove or minimise depthrelated effects. However, plots of the scores for the first canonical variate against depth and 100/depth
in Figure 3-90 and Figure 3-91 show an unusual pattern. Clearly for the groups which have canonical
GBR Seabed Biodiversity
3-177
variate scores which fall within the top cluster in the overall plot of the first two canonical variates,
there is a relationship between the scores for the first canonical variate and depth. Moreover, this
pattern is really only clearly evident when the data for all the groups are plotted (see Figure 3-92). The
pattern is again evident when the data for all the groups are plotted against 100/depth in Figure 3-93.
a)
b)
Figure 3-90: Plot of the group means for the first canonical variate for the canonical variate analysis of the 4519
groups for the peak-aligned, row-corrected and depth-adjusted data (a) against depth; and (b) against 1/depth.
a)
b)
Figure 3-91: Plot of the group means for the first canonical variate for the canonical variate analysis of the 4519
groups which result from contrasts between class 617 vs classes 618 and 621, and class 618 vs class 621, for the
peak-aligned, row-corrected and depth-adjusted data (a) against depth; and (b) against 1/depth.
GBR Seabed Biodiversity
a)
b)
c)
d)
3-178
Figure 3-92: Plot of the group means for the first canonical variate for the canonical variate analysis of the 4519
groups which result from contrasts between class 617 vs classes 618 and 621, and class 618 vs class 621, for the
peak-aligned, row-corrected and depth-adjusted data against depth for (a) all groups for all classes; (b) sand class
617; (c) sponge class 618; and (d) seagrass class 621.
GBR Seabed Biodiversity
a)
b)
c)
d)
3-179
Figure 3-93: Plot of the group means for the first canonical variate for the canonical variate analysis of the 4519
groups which result from contrasts between class 617 vs classes 618 and 621, and class 618 vs class 621, for the
peak-aligned, row-corrected and depth-adjusted data against 1/depth for (a) all groups for all classes; (b) sand
class 617; (c) sponge class 618; and (d) seagrass class 621.
3.6.2.3.4 Canonical Variate Analyses for the Two Clusters
Clearly for the groups which have canonical variate scores which fall within the smaller cluster in the
overall CV1-CV2 plots, there is a relationship between their CV1 scores and depth.
The CV1 vs CV2 plot was used to define two clusters. Groups were assigned to the larger cluster if
CV1 ≥ 11.5. If CV1 < 10.5, then groups were assigned to the smaller cluster. If 10.5 < CV1 < 11 and
CV2 ≥ 32, then groups were assigned to the smaller cluster. Other groups with CV1 scores < 11.5
were assigned to the smaller cluster.
Figure 3-94(a) shows a plot of the canonical variate scores for the 3358 groups in the larger cluster.
Canonical variate plots for the individual classes again show no obvious separation between any of the
classes. Examples for sand (class 617), seagrass (class 621) and mud (class 917) are shown in Figure
3-94.
Figure 3-95 shows that there is no obvious pattern of the scores for the first canonical variate for the
larger cluster with depth.
GBR Seabed Biodiversity
a)
b)
c)
d)
3-180
Figure 3-94: Plot of the group means for the first two canonical variates for the canonical variate analysis of the
3358 groups in the larger CV1-CV2 cluster for the peak-aligned, row-corrected and depth-adjusted data for (a)
all groups; (b) sand class 617; (c) seagrass class 621; and (d) mud class 917.
Figure 3-95: Plot of the group means for the first canonical variate for the canonical variate analysis of the 3358
groups for the peak-aligned, row-corrected and depth-adjusted data against depth.
GBR Seabed Biodiversity
3-181
Figure 3-96 (a) shows a plot of the canonical variate scores for the 1161 groups in the smaller cluster.
Canonical variate plots for the individual classes again show no obvious separation between any of the
classes. Examples for sand (class 617) and mud (class 917) are shown in Figure 3-96.
Figure 3-97 shows that there is no obvious pattern of the scores for the first canonical variate for the
smaller cluster with depth.
a)
b)
c)
Figure 3-96: Plot of the group means for the first two canonical variates for the canonical variate analysis of the
1161 groups in the smaller CV1-CV2 cluster for the peak-aligned, row-corrected and depth-adjusted data for (a)
all groups; (b) sand class 617; and (c) mud class 917
GBR Seabed Biodiversity
3-182
Figure 3-97: Plot of the group means for the first canonical variate for the canonical variate analysis of the 1161
groups for the peak-aligned, row-corrected and depth-adjusted data against depth
3.6.2.3.5 Site Contrast Canonical Variate Analyses
The data analysed here were from sites which were collected close together in time and hence
geographically, concentrating on potential extremes of cover, such as sand and seagrass, and mud, silt
and sand.
Sites 1631 vs 2552 – Seagrass vs Sand
There were 12 groups for class 621 (seagrass) from site 1631, and 30 groups for class 617 (sand) from
site 2552. The first four canonical roots were 7.04, 0.383, 0.163 and 0.071. Figure 3-98 shows two
obvious clusters along the first canonical variate, corresponding to groups from the seagrass site (on
the left) and from the sand site (on the right).
a)
b)
Figure 3-98: Plots of (a) the canonical variate scores and (b) the group means for the first two canonical variates
for a canonical variate analysis of the depth-normalised data for 42 groups from sites 1631 and 2552, without
regard to the class labels.
Aligning the peaks reduces the first canonical root from 7.04 to 6.57. The two sites are at average
depths of 37 m (1631 - seagrass) and 30 m (2552 - sand). Adjusting the responses for the regressions
GBR Seabed Biodiversity
3-183
on inverse depth further reduces the first canonical root to 5.89. Aligning the peaks and correcting for
size by subtracting the row means removes much of the class separation, reducing the first canonical
root to 1.13; adjusting area under the curve by dividing by the row means reduces the first canonical
root to 0.79. These analyses and the plots in Figure 3-99 suggest that the responses for seagrass and
sand are quite similar in their shapes, and that the discrimination in Figure 3-98 results from
differences in the magnitudes of the responses between the two classes.
b)
a)
Figure 3-99: Plots of the depth-normalised pelagic data against sample number for group means for (a) 12
groups for the seagrass site 1631 (class 621); and (b) 30 groups for the sand site 2552 (class 617).
Sites 1631 vs 2552 vs 2224 – Seagrass vs Sand vs Sand
There were 12 groups for class 621 (seagrass) from site 1631, 30 groups for class 617 (sand) from site
2552, and 14 groups for class 617 (sand) from site 2224. The first four canonical roots were 9.26,
1.402, 0.253 and 0.154. Figure 3-100 shows that when sand groups from site 2224 collected on
23/09/2004 are included in an analysis of the depth-normalised data, there is marked separation
between the two sand sites along the first canonical variate, and separation of the sand and seagrass
sites along the second canonical variate. Figure 3-101 shows plots of group means for the two sand
sites; the obvious differences in the shapes are reflected in the canonical variate scores in Figure
3-100. These plots of mean profiles for the two sand sites show greater differences in shape than do
plots of the first sand site and the seagrass site (Figure 3-99). Note also the cluster of scores with low
CV2 and high CV1 values in Figure 3-100, which do not correspond to defined sand groups. The top
profile in Figure 3-102 is from sand group 53, but its canonical variate score plots in the low CV2 –
high CV1 cluster, whereas the canonical variate scores for the other two profiles plot in the nominally
correct high CV1 – high CV2 cluster.
a)
b)
Figure 3-100: Plots of (a) the canonical variate scores and (b) the group means for the first two canonical
variates for a canonical variate analysis of the depth-normalised data for 56 groups from sites 1631, 2552 and
2224, without regard to the class labels.
GBR Seabed Biodiversity
a)
3-184
b)
Figure 3-101: Plots of the depth-normalised pelagic data against sample number for group means for (a) 30
groups for class 617 (sand) from site 2552, and (b) 14 groups for class 617 (sand) from site 2224.
Figure 3-102: Plots of the echo responses for the depth-normalised pelagic data against sample number for
profiles 48 – 50 for group 53 (from the sand site 2552 - class 617).
Sites 1631 vs 2552 vs 2224 vs 2441 – Seagrass vs Sand vs Sand vs Seagrass
In addition to the seagrass groups from site 1631, the sand groups from site 2552, and the sand groups
site 2224, there are 5 groups for class 621 (seagrass) from site 2441 (the fifth group nominally consists
of 59 contiguous pings from class 617 sand). The first four canonical roots are 9.38, 1.696, 0.488 and
0.188. Figure 3-103 shows that the new site 2441 seagrass groups 57 – 61 cluster with the initial site
1631 seagrass groups 1 – 12 along both CV1 and CV2. Note that group 61 (sand) is not separated from
groups 57 – 60 along any of the canonical variates. There are obvious similarities in the shapes of the
group mean ping profiles for the two seagrass sites in Figure 3-104.
As can be seen from Figure 3-104 (b), there are no obvious differences in shape between the four
nominal seagrass groups and the nominal sand group from site 2441 — this sand group from the same
predominantly seagrass site also clusters with the seagrass groups from the same site (Figure 3-103).
This pattern was also observed for the analyses of the first pair of sites (1631 vs 2552) above, where
there were a number of groups of contiguous pings, typically less than 20, from other classes. For
these analyses, the CV scores for these groups always clustered with the respective groups for the
dominant class for that site.
GBR Seabed Biodiversity
a)
3-185
b)
Figure 3-103: Plots of (a) the canonical variate scores and (b) the group means for the first two canonical
variates for a canonical variate analysis of the depth-normalised data for 61 groups from sites 1631, 2552, 2224
and 2441, without regard to the class labels.
a)
b)
Figure 3-104: Plots of the depth-normalised pelagic data against sample number for group means for (a) 12
groups for class 621 (seagrass) from site 1631, and (b) 5 groups for class 621 (seagrass) from site 2441.
Variability of sand sites
Plots of the group mean ping profiles for the sand sites are shown in Figure 3-101 and Figure 3-105.
Figure 3-105(b) shows that the group for nominal class 617 from site 1580 is obviously different from
the groups from site 2224. The canonical variate scores for the group from site 1580 plot with the
cluster of scores with low CV2 and high CV1 values noted earlier.
a)
b)
Figure 3-105: Plots of the depth-normalised pelagic data against sample number for group means for (a) 14
groups for class 617 (sand) from site 2224 and (b) the group for class 617 from site 1580 superimposed on the
groups from site 2224.
GBR Seabed Biodiversity
3-186
3.6.3. Linear Discriminant Analyses of QTC View data (I McLeod)
3.6.3.1. Substratum and BioHabitat combinations v2
The full combination of the original 24 BioHabitat by 9 substratum event classes (v1) did not
converge, so the combination of the original substratum 9 class coding (Table 2-6) and the first 12
class biohabitat re-coding (Table 2-10) schema was investigated. This potentially had 108 classes of
which 102 classes were expressed.
The completed linear discriminant analysis (LDA) produced a confusion matrix (as described in
Section 3.6.1) of 103 rows by 103 columns, which was too large to tabulate within this report. Instead
an abridged version is presented (Table 3-22), that shows the relevant information from the diagonals
of the relevant confusion matrices for all but the most infrequent habitat event class combinations.
The “Observed” are analogous to the row totals from the full confusion matrices. “Observed Counts”
are original number of observations in each class and the “Observed % of Total” are the observed
counts as a percentage of the total observations. The three right hand columns indicate the
classification results. The “Observed vs Predicted Counts” are the number of observations of a given
class correctly classified as that class. The “OvsP % of total” are the correctly classified counts as a
percentage of the total observations. The “OvsP % of Row” are the correctly classified counts as a
percentage of the observed counts, giving the percentage within class prediction skill.
Superficially, it seems that a number of the classes are reasonably accurately classified. However,
these are relatively infrequent and the overall classification success is only ~3.4%. The most
frequently occurring classes have very low classification success and were incorrectly classified to
other (off-diagonal) habitat classes (not tabulated in the above abridged version). In particular, the
more common classes, such as the no biohabitat classes, were almost always incorrectly classified as a
range of other biohabitat combinations.
This poor performance with this number of classes was not unexpected, and further aggregation was
necessary, but nevertheless clearly demonstrates that acoustics data are not able to discriminate a
modest number of basic habitat types that are readily recognised by observers
3.6.3.2. Substratum and BioHabitat combinations v3
The next combination of the re-coding classes investigated was the combination of the second
substratum re-coding (7 classes, Table 2-13) and the first biohabitat re-coding (12 classes, Table 2-10).
This potentially had 84 classes of which 79 were expressed.
Again, the full confusion matrix from the cross-validated LDA was too large to present within this
report, and an abridged table (Table 3-23) is again presented, which shows the relevant information
from the diagonals of the relevant confusion matrices for all but the most infrequent habitat event class
combinations. The table columns are as described for the previous table.
Again, a number of the relatively infrequent classes appear to have been reasonably accurately
classified, but the overall classification showed little improvement at approximately 6%. Again, the
most frequently occurring classes, such as the no biohabitat and bioturbated classes, had very low
classification success and were incorrectly classified to other (off-diagonal) habitat class combinations
(not tabulated in the abridged Table 3-23). Clearly, a much greater level of aggregation was required.
GBR Seabed Biodiversity
3-187
Table 3-22. Sub1_Hab2: Observed (Row Totals) counts and percentage contribution, Observed versus Predicted
Diagonal counts and percentages.
Sub1_Hab2 Description
Reef : No BioHabitat
Reef : Sparse garden
Reef : Gorgonian
Reef : Sponge
Reef : Algae
Reef : Coral Dense
Boulders : No BioHabitat
Boulders : Sparse garden
Boulders : Gorgonian
Boulders : Sponge
Boulders : Algae
Boulders : Coral Sparse
Cobbles : No BioHabitat
Cobbles : Sparse garden
Cobbles : Gorgonian
Cobbles : Algae
Gravel : No BioHabitat
Gravel : Sparse garden
Gravel : Gorgonian
Gravel : Sponge
Gravel : Algae
Gravel : Caulerpa
Gravel : Halimeda
Gravel : Seagrass
Gravel : Bioturbated
Coarse Sand : No BioHabitat
Coarse Sand : Sparse garden
Coarse Sand : Alcyonarians
Coarse Sand : Gorgonian
Coarse Sand : Algae
Coarse Sand : Caulerpa
Coarse Sand : Halimeda
Coarse Sand : Seagrass
Coarse Sand : Bioturbated
Fine Sand : No BioHabitat
Fine Sand : Sparse garden
Fine Sand : Alcyonarians
Fine Sand : Algae
Fine Sand : Caulerpa
Fine Sand : Halimeda
Fine Sand : Seagrass
Fine Sand : Bioturbated
Sand Waves : No BioHabitat
Sand Waves : Sparse garden
Sand Waves : Gorgonian
Sand Waves : Caulerpa
Silt : No BioHabitat
Silt : Sparse garden
Silt : Alcyonarians
Silt : Algae
Silt : Caulerpa
Silt : Halimeda
Silt : Seagrass
Silt : Bioturbated
Mud : No BioHabitat
Mud : Sparse garden
Mud : Alcyonarians
Mud : Sponge
Mud : Seagrass
Mud : Bioturbated
Total
Observed
Counts
% of total
140
0.10
195
0.13
94
0.06
74
0.05
195
0.13
891
0.61
624
0.43
626
0.43
170
0.12
56
0.04
218
0.15
193
0.13
1007
0.69
950
0.65
92
0.06
323
0.22
5834
3.98
2186
1.49
197
0.13
134
0.09
2207
1.51
120
0.08
216
0.15
344
0.23
236
0.16
10567
7.21
1214
0.83
1618
1.10
198
0.14
2864
1.95
226
0.15
2584
1.76
1579
1.08
1643
1.12
20055
13.69
1735
1.18
394
0.27
3719
2.54
500
0.34
1613
1.10
3802
2.59
10825
7.39
4069
2.78
227
0.15
1021
0.70
108
0.07
15201
10.37
1287
0.88
666
0.45
1011
0.69
165
0.11
222
0.15
1456
0.99
18398
12.56
8141
5.56
1089
0.74
341
0.23
118
0.08
480
0.33
8689
5.93
146533
Observed vs Predicted
Counts
% of total
% of Row
4
0.00
2.86
22
0.02
11.28
10
0.01
10.64
64
0.04
86.49
17
0.01
8.72
125
0.09
14.03
85
0.06
13.62
23
0.02
3.67
11
0.01
6.47
2
0.00
3.57
55
0.04
25.23
50
0.03
25.91
201
0.14
19.96
305
0.21
32.11
28
0.02
30.43
23
0.02
7.12
3
0.00
0.05
7
0.00
0.32
24
0.02
12.18
13
0.01
9.70
19
0.01
0.86
33
0.02
27.50
13
0.01
6.02
28
0.02
8.14
24
0.02
10.17
0
0.00
0.00
2
0.00
0.16
511
0.35
31.58
4
0.00
2.02
12
0.01
0.42
73
0.05
32.30
28
0.02
1.08
16
0.01
1.01
28
0.02
1.70
2
0.00
0.01
8
0.01
0.46
207
0.14
52.54
117
0.08
3.15
1
0.00
0.20
54
0.04
3.35
62
0.04
1.63
6
0.00
0.06
14
0.01
0.34
60
0.04
26.43
283
0.19
27.72
50
0.03
46.30
14
0.01
0.09
3
0.00
0.23
169
0.12
25.38
31
0.02
3.07
21
0.01
12.73
20
0.01
9.01
297
0.20
20.40
299
0.20
1.63
44
0.03
0.54
52
0.04
4.78
265
0.18
77.71
115
0.08
97.46
81
0.06
16.88
374
0.26
4.30
3.4
GBR Seabed Biodiversity
3-188
Table 3-23. Sub2_Hab2: Observed (Row Totals) counts and percentage contribution, Observed verse Predicted
Diagonal counts and percentages.
Sub2_Hab2 Description
Reef : No BioHabitat
Reef : Sparse garden
Reef : Gorgonian
Reef : Sponge
Reef : Algae
Reef : Coral Dense
Reef : Coral Sparse
Boulders : No BioHabitat
Boulders : Sparse garden
Boulders : Gorgonian
Boulders : Sponge
Boulders : Algae
Boulders : Coral Dense
Boulders : Coral Sparse
Cobbles : No BioHabitat
Cobbles : Sparse garden
Cobbles : Gorgonian
Cobbles : Algae
Gravel : No BioHabitat
Gravel : Sparse garden
Gravel : Gorgonian
Gravel : Sponge
Gravel : Algae
Gravel : Caulerpa
Gravel : Halimeda
Gravel : Sea grass
Gravel : Bioturbated
Sand : No BioHabitat
Sand : Sparse garden
Sand : Alcyonarians
Sand : Gorgonian
Sand : Sponge
Sand : Algae
Sand : Caulerpa
Sand : Halimeda
Sand : Sea grass
Sand : Bioturbated
Silt : No BioHabitat
Silt : Sparse garden
Silt : Alcyonarians
Silt : Gorgonian
Silt : Sponge
Silt : Algae
Silt : Caulerpa
Silt : Halimeda
Silt : Sea grass
Silt : Bioturbated
Mud : No BioHabitat
Mud : Sparse garden
Mud : Alcyonarians
Mud : Sponge
Mud : Algae
Mud : Sea grass
Mud : Bioturbated
TOTAL
Observed
Counts % of total
140
0.10
194
0.13
94
0.06
74
0.05
193
0.13
887
0.61
80
0.05
621
0.43
621
0.43
170
0.12
56
0.04
218
0.15
96
0.07
193
0.13
1005
0.69
940
0.64
92
0.06
321
0.22
5813
3.98
2176
1.49
195
0.13
134
0.09
2201
1.51
120
0.08
216
0.15
343
0.24
232
0.16
34543
23.68
3165
2.17
2003
1.37
1293
0.89
76
0.05
6595
4.52
829
0.57
4168
2.86
5368
3.68
12408
8.51
15137
10.38
1279
0.88
658
0.45
70
0.05
62
0.04
1005
0.69
165
0.11
219
0.15
1447
0.99
18339
12.57
8109
5.56
1086
0.74
340
0.23
118
0.08
90
0.06
475
0.33
8674
5.95
145886
Observed vs. Predicted
Counts % of total % of Row
51
0.03
36.4
39
0.03
20.1
50
0.03
53.2
44
0.03
59.5
61
0.04
31.6
56
0.04
6.3
43
0.03
53.8
125
0.09
20.1
54
0.04
8.7
61
0.04
35.9
35
0.02
62.5
34
0.02
15.6
56
0.04
58.3
36
0.02
18.7
87
0.06
8.7
219
0.15
23.3
39
0.03
42.4
70
0.05
21.8
218
0.15
3.8
22
0.02
1.0
41
0.03
21.0
39
0.03
29.1
274
0.19
12.4
64
0.04
53.3
62
0.04
28.7
145
0.10
42.3
58
0.04
25.0
452
0.31
1.3
36
0.02
1.1
186
0.13
9.3
184
0.13
14.2
39
0.03
51.3
245
0.17
3.7
98
0.07
11.8
130
0.09
3.1
243
0.17
4.5
250
0.17
2.0
203
0.14
1.3
59
0.04
4.6
181
0.12
27.5
45
0.03
64.3
35
0.02
56.5
87
0.06
8.7
99
0.07
60.0
84
0.06
38.4
263
0.18
18.2
880
0.60
4.8
336
0.23
4.1
138
0.09
12.7
184
0.13
54.1
105
0.07
89.0
59
0.04
65.6
151
0.10
31.8
1420
0.97
16.4
6.02
GBR Seabed Biodiversity
3-189
3.6.3.3. BioHabitat v2
The re-coding of habitat events from the original 26 class biohabitat code to the reduced 12 class set
(Habitat_Code2, Table 2-10) was at least partly based on taxonomy. Due to the much smaller number
of classes, it is possible to present here the confusion matrix, which is the cross-tabulation of the
observed versus LDA classified results (Table 3-24). The diagonal of this matrix (left to right down
the page) reports the observations that were correctly classified by the LDA. The row totals (far right
column) show the total number of observations in each habitat class. The column totals (bottom row)
show the total number of cases that were classified to each habitat class by LDA. The grand total
(bottom right cell) shows the total number of observations that were classified. The other cells (offdiagonals) report the misclassification or confusion.
The next matrix (Table 3-25) is similar except that the numbers are shown as percentages of the row
totals from Table 3-24. The diagonal now shows the relative classification success for each biohabitat
class. A number of the biohabitat classes appear to have good classification performance; however, as
was the case for the combinations examined above, these are the relatively infrequent types and
performance is poor for the common seabed class types, such as No BioHabitat and Bioturbated,
which tend to get classified as seagrass or algae or various types of epibenthos.
Table 3-24. Habitat_Code2: Confusion matrix of total counts observed vs. predicted
Predicted
Observed
0
1
2
3
4
5
6
7
8
9
10 11 Total
No BioHabitat 0 10259 5884 6290 7011 2837 5395 4140 6052 7615 5452 3506 1166 65607
Sparse garden 1
484 2809 872 1200 466 487 540 545 876 365 641 222 9507
Alcyonarians 2
9 2402
50 48
339
91
24
130 30 3123
Gorgonian 3
1
32 1567 132
6 72
6
1
38 61 1916
Sponge 4
5 528
21
554
Algae 5
395 638 579 1113 397 3632 503 727 963 360 1048 312 10667
Caulerpa 6
23
31 34
1015
1
33 13 1150
Halimeda 7
174 694 281 129 117 523 2256 219
14 268 91 4766
Seagrass 8
105 292 535 569 217 421 573 422 3534 236 622 124 7650
Bioturbated 9 1901 2156 4084 2743 956 2665 2252 4127 4426 10991 2904 674 39879
Coral Dense 10
6
20 24
4
1043 35 1132
Coral Sparse 11
4
482
486
Total
13144 11963 15517 14590 5772 12723 9961 14226 17659 17418 10233 3231 146437
Table 3-25. Habitat_Code2: Confusion matrix of percentage contribution as a percentage of row totals
Observed
No BioHabitat 0
Sparse garden 1
Alcyonarians 2
Gorgonian 3
Sponge 4
Algae 5
Caulerpa 6
Halimeda 7
Seagrass 8
Bioturbated 9
Coral Dense 10
Coral Sparse 11
0
1
2
3
4
15.6 9.0 9.6 10.7 4.3
5.1 29.5 9.2 12.6 4.9
0.3 76.9 1.6 1.5
0.1 1.7 81.8 6.9
0.9 95.3
3.7 6.0 5.4 10.4 3.7
2.0 2.7 3.0
3.7 14.6 5.9 2.7
1.4 3.8 7.0 7.4 2.8
4.8 5.4 10.2 6.9 2.4
0.5 1.8 2.1
0.8
Predicted
5
6
8.2
6.3
5.1
5.7
10.9
0.3
3.8
34.0
2.5
5.5
6.7
4.7
88.3
11.0
7.5
5.6
0.4
7
9.2
5.7
2.9
0.3
8
11.6
9.2
0.8
0.1
9
8.3
3.8
10
5.3
6.7
4.2
2.0
6.8
9.0
0.1
4.6
46.2
11.1
3.4
9.8
2.9
5.6
8.1
7.3
92.1
47.3
5.5
10.3
0.3
3.1
27.6
11
1.8
2.3
1.0
3.2
3.8
2.9
1.1
1.9
1.6
1.7
3.1
99.2
Total
100
100
100
100
100
100
100
100
100
100
100
100
GBR Seabed Biodiversity
3-190
The third matrix (Table 3-26) shows the classification results as percentages of the total number of
observations. A comparison of the row and column total percentages, in particular, highlight the
overall confusion — those biohabitats that were infrequent in the original observations are far too
frequent in the classified results (by up to 10 times). In the case of seagrass, for example, the observed
frequency was 5.2% whereas the classified frequency was 12.1% — clearly a large number of No
biohabitat and bioturbated Observations were being classified as Seagrass. Conversely, those Habitats
that were very frequent in the original observations are far too infrequent in the classified results. For
example, the observed frequency of bioturbated was 27.2% whereas the classified frequency was
11.9% — clearly a large number of bioturbated observations were being classified as other types of
biohabitats. The overall classification success was only 27.7%.
Table 3-26. Habitat_Code2: Confusion matrix of percentage contribution as a percentage of totals
Observed
No biohabitat 0
Sparse garden 1
Alcyonarians 2
Gorgonian 3
Sponge 4
Algae 5
Caulerpa 6
Halimeda 7
Seagrass 8
Bioturbated 9
Coral Dense 10
Coral Sparse 11
Total
0
1
2
7.01 4.02 4.30
0.33 1.92 0.60
0.01 1.64
0.00 0.02
0.27 0.44 0.40
0.02
0.12 0.47
0.07 0.20 0.37
1.30 1.47 2.79
0.00
9.0
8.2 10.6
3
4.79
0.82
0.03
1.07
0.00
0.76
0.02
0.19
0.39
1.87
0.01
4
1.94
0.32
0.03
0.09
0.36
0.27
0.02
0.09
0.15
0.65
0.02
0.00
10.0 3.9
Predicted
5
6
3.68 2.83
0.33 0.37
0.23
0.00 0.05
7
4.13
0.37
0.06
0.00
8
5.20
0.60
0.02
0.00
9
3.72
0.25
10
2.39
0.44
0.09
0.03
2.48
0.34
0.69
0.36
0.39
1.54
0.00
0.50
0.25
1.54
0.29
2.82
0.66
0.00
0.15
2.41
3.02
0.72
0.02
0.18
0.42
1.98
0.71
6.8
9.7
12.1
11.9
0.08
0.29
1.82
8.7
0.01
0.16
7.51
7.0
11
0.80
0.15
0.02
0.04
0.01
0.21
0.01
0.06
0.08
0.46
0.02
0.33
2.2
Total
44.8
6.5
2.1
1.3
0.4
7.3
0.8
3.3
5.2
27.2
0.8
0.3
27.7
3.6.3.4. Biohabitat v3
An alternative, and more severe, aggregation of the original habitat event codes reduced the number of
habitat classes to eight by lumping all epibenthos by density and all marine plants (Table 2-11). Very
sparse epibenthos was considered to differ little from “No biohabitat”.
Again, cross-tabulations of observed versus classified results are presented as a series of confusion
matrices. Table 3-27 shows the counts, Table 3-28 shows the percentages of row total observations
and Table 3-29 shows the percentages of the total observations.
Table 3-27. Habitat_Code3: Confusion matrix of total counts observed vs. predicted
Observed
No biohabitat
Soft – Dense
Soft – Medium
Soft – Sparse
Algae and Seagrass
Bioturbated
Coral – Dense
Coral – Sparse
Total
0
1
2
3
4
5
6
7
0
15716
1657
2251
19624
1
1631
444
143
221
461
574
14
21
3509
Predicted
2
3
4
5
10763 16027 9282 10795
4347
994
3610
5363
157
6046
4470
6516
6
2454
1
130
72
9831 2545
4526 18288
146
357
1380
2030
1094
25077 33216 23770 31700
7461
7 Total
769 67437
1
445
60 4854
110 7930
326 24280
421 39969
24 1132
465
486
2176 146533
GBR Seabed Biodiversity
3-191
Again, the less frequent biohabitat types appeared to have been reasonably well classified whereas the
more frequent types were poorly classified. For example, unaccepted proportions of No BioHabitat
and Bioturbated were classified as epibenthos or Seagrass. The overall classification success was
somewhat improved with fewer classes, at 38.4%.
Table 3-28. Habitat_Code3: Confusion matrix of percentage contribution as a percentage of row totals
Observed
No BioHabitat
Soft – Dense
Soft – Medium
Soft – Sparse
Algae and Seagrass
Bioturbated
Coral - Dense
Coral – Sparse
0
1
2
3
4
5
6
7
0
23.3
6.8
5.6
1
2.4
99.8
2.9
2.8
1.9
1.4
1.2
4.3
2
16.0
Predicted
3
4
23.8
13.8
89.6
12.5
14.9
13.4
3.2
76.2
18.4
16.3
0.0
1.6
40.5
11.3
5
16.0
6
3.6
0.9
10.5
45.8
3.0
4.5
5.7
5.1
96.6
7
1.1
0.2
1.2
1.4
1.3
1.1
2.1
95.7
Total
100
100
100
100
100
100
100
100
Table 3-29. Habitat_Code3: Confusion matrix of percentage contribution as a percentage of totals
Observed
No BioHabitat
Soft – Dense
Soft – Medium
Soft – Sparse
Algae and Seagrass
Bioturbated
Coral - Dense
Coral – Sparse
Total
0
1
2
3
4
5
6
7
0
10.73
1.13
1.54
13.39
1
1.11
0.30
0.10
0.15
0.31
0.39
0.01
0.01
2.39
2
7.35
Predicted
3
4
10.94
6.33
5
7.37
6
1.67
2.97
0.68
2.46
3.66
0.11
4.13
3.05
4.45
0.00
0.09
6.71
3.09
0.05
1.74
12.48
0.10
0.24
0.94
1.39
0.75
17.11
22.67
16.22
21.63
5.09
7
0.52
0.00
0.04
0.08
0.22
0.29
0.02
0.32
1.48
Total
46.02
0.30
3.31
5.41
16.57
27.28
0.77
0.33
38.4
3.6.3.5. Substratum v1
The first analysis of substratum alone attempted to classify the original 9 sediment substratum event
classes (Table 2-6) at the observed sites. As above, cross-tabulations of observed versus classified
results are presented as a series of confusion matrices. Table 3-30 shows the counts, Table 3-31 shows
the percentages of row total observations and Table 3-32 shows the percentages of the total
observations.
Results similar to those for biohabitat types were obtained, in that the less frequent substratum types
appeared to have been reasonably well classified whereas the more frequent types were poorly
classified. The best result was ~99% success in predicting reef where it was observed; the residual 1%
was spread over less-rough substrata and sand waves. However, Reef was only a very small
component of the dataset (<1.2%), so in reality the high accuracy of the classification Reef is lost amid
the mass of smooth or soft sediments — and incorrectly classified sands and silt inflates the predicted
reef by more than 5-fold. The smooth and soft sediments displayed high levels of confusion with, for
example, observed sand being classified correctly only 25% of the time and incorrectly being labelled
in significant proportions in all other classes. Further, sand was observed 29% of the time but was
predicted correctly only 7.5% the time.
GBR Seabed Biodiversity
3-192
Overall, the rough substratum types had >80% classification success whereas the more common
smoother substratum types had <30% classification success and unacceptable proportions of smoother
classes were classified as rougher classes. The overall classification success was 36.6%, that is, in
~37% of cases the classified result matched the observed, whereas in 63% of cases the classified result
did not match the observed (i.e. the classification was confused).
Table 3-30. Substratum_Code1: Confusion matrix of total counts observed vs. predicted
Predicted
1
2
3
4
5
6
7
8
9
1682
6
11
3
193 1904
44
1
53
146
171 2141
8
53
983 1012 1265 3907
641
494 2361
408
516
1781 1699 1525 1655 5842 1475 4434 1548 2614
3640 3047 2954 2955 3059 10926 7014 3286 5942
196
348
385
110
47
4 4356
23
103
2424 2470 2186 1877 2543 2794 6617 10390 7256
967
763
460
358
591
810 1786
731 12539
12012 11420 10971 10871 12723 16503 26677 16386 28970
Observed
Reef
Rocks
Stones
Rubble
Coarse Sand
Sand
Sand Waves
Silt
Mud
Total
1
2
3
4
5
6
7
8
9
Total
1702
2195
2519
11587
22573
42823
5572
38557
19005
146533
Table 3-31. Substratum_Code1: Confusion matrix of percentage contribution as a percentage of row totals
Observed
Reef
Rocks
Stones
Rubble
Coarse Sand
Sand
Sand Waves
Silt
Mud
1
2
3
4
5
6
7
8
9
1
98.8
8.8
5.8
8.5
7.9
8.5
3.5
6.3
5.1
2
0.4
86.7
6.8
8.7
7.5
7.1
6.2
6.4
4.0
3
0.6
2.0
85.0
10.9
6.8
6.9
6.9
5.7
2.4
Predicted
5
6
4
0.0
0.3
33.7
7.3
6.9
2.0
4.9
1.9
5.5
25.9
7.1
0.8
6.6
3.1
4.3
6.5
25.5
0.1
7.2
4.3
7
0.2
2.4
2.1
20.4
19.6
16.4
78.2
17.2
9.4
8
9
3.5
6.9
7.7
0.4
26.9
3.8
4.5
11.6
13.9
1.8
18.8
66.0
Total
100
100
100
100
100
100
100
100
100
Table 3-32. Substratum_Code1: Confusion matrix of percentage contribution as a percentage of totals
Observed
Reef
Rocks
Stones
Rubble
Coarse Sand
Sand
Sand Waves
Silt
Mud
Total
1
2
3
4
5
6
7
8
9
1
1.15
0.13
0.10
0.67
1.22
2.48
0.13
1.65
0.66
8.20
2
0.00
1.30
0.12
0.69
1.16
2.08
0.24
1.69
0.52
7.79
3
0.01
0.03
1.46
0.86
1.04
2.02
0.26
1.49
0.31
7.49
4
0.00
0.01
2.67
1.13
2.02
0.08
1.28
0.24
7.42
Predicted
5
6
0.44
3.99
2.09
0.03
1.74
0.40
8.68
0.34
1.01
7.46
0.00
1.91
0.55
11.26
7
0.00
0.04
0.04
1.61
3.03
4.79
2.97
4.52
1.22
18.21
8
9
0.28
1.06
2.24
0.02
7.09
0.50
11.18
0.35
1.78
4.06
0.07
4.95
8.56
19.77
Total
1.16
1.50
1.72
7.91
15.40
29.22
3.80
26.31
12.97
36.6
GBR Seabed Biodiversity
3-193
3.6.3.6. Substratum v2
In the previous analysis, there was considerable confusion among the several classes of sand, causing
poor overall performance. Consequently, the three separate classes of sand were aggregated into a
single class “Sand”. This yielded seven classes, with separate classes for gravel, silt and mud (Table
2-12).
Again, cross-tabulations of observed versus classified results are presented as a series of confusion
matrices. Table 3-33 shows the counts, Table 3-34 shows the percentages of row total observations
and Table 3-35 shows the percentages of the total observations.
The sands were correctly classified in 30.4% of cases, but large proportions of observed sand were
misclassified into gravel, silt or mud substratum classes. Observed Silt was also classified into
anything from Gravel to Mud. Again, the more frequent observed classes (Sand 48.4% and Silt 26.3%)
had the poorest classification success and were markedly under-classified in the results (18.2% and
18.8% respectively). On the other hand, while the less frequently observed classes appeared to have
good classification success (75% to 99%), they were seriously over-classified in the results —
erroneously indicating a much higher occurrence of harder/rougher substratum types than was
observed. Overall classification success was ~45%.
Table 3-33. Substratum_Code2: Confusion matrix of total counts observed vs. predicted
Observed
Reef
Boulders
Cobbles
Gravel
Sand
Silt
Mud
Total
1
1690
158
109
766
4374
1850
752
9699
1
2
3
4
5
6
7
2
4
2006
149
828
4215
2003
601
9806
Predicted
4
5
3
8
31
2260
1090
4139
1805
377
9710
1
7923
16540
5156
818
30438
21543
4015
1126
26684
6
7
390
9721
16353
1101
27565
590
10436
7375
14230
32631
Total
1702
2195
2519
11587
70968
38557
19005
146533
Table 3-34. Substratum_Code2: Confusion matrix of percentage contribution as a percentage of row totals
Observed
Reef
Boulders
Cobbles
Gravel
Sand
Silt
Mud
1
2
3
4
5
6
7
1
99.3
7.2
4.3
6.6
6.2
4.8
4.0
2
0.2
91.4
5.9
7.1
5.9
5.2
3.2
3
0.5
1.4
89.7
9.4
5.8
4.7
2.0
Predicted
4
5
0.0
68.4
23.3
13.4
4.3
30.4
10.4
5.9
6
7
3.4
13.7
42.4
5.8
5.1
14.7
19.1
74.9
Total
100
100
100
100
100
100
100
Table 3-35. Substratum_Code2: Confusion matrix of percentage contribution as a percentage of totals
Observed
Reef
Boulders
Cobbles
Gravel
Sand
Silt
Mud
Total
1
2
3
4
5
6
7
1
1.15
0.11
0.07
0.52
2.98
1.26
0.51
6.62
2
0.00
1.37
0.10
0.57
2.88
1.37
0.41
6.69
3
0.01
0.02
1.54
0.74
2.82
1.23
0.26
6.63
Predicted
4
5
0.00
5.41
11.29
3.52
0.56
20.77
14.70
2.74
0.77
18.21
6
7
0.27
6.63
11.16
0.75
18.81
0.40
7.12
5.03
9.71
22.27
Total
1.16
1.50
1.72
7.91
48.43
26.31
12.97
45.04
GBR Seabed Biodiversity
3-194
3.6.3.7. Substratum v3
In the ‘Substratum v1’ analysis, the sand waves class had reasonable classification success, so as an
alternative aggregation, “Sand Waves” were retained as a separate class but coarse sand, fine sand and
silt were aggregated in a third substratum-only recoding schema (Table 2-13) of seven classes. Again,
cross-tabulations of observed versus classified results are presented as a series of confusion matrices.
Table 3-36 shows the counts, Table 3-37 shows the percentages of row total observations and Table
3-38 shows the percentages of the total observations.
In this aggregation, the Sand-Silt class represented about 71% of the observations and the
classification success for this class remained a rather poor 35.5%, as well as being grossly underrepresented in the classification results due large number of observations being classified as other
substrata. Given the values of the diagonals, the other observed classes appeared to have much greater
classification success and the overall classification success was 45.8%. However, as before, these
classes were seriously over-classified in the results — erroneously indicating a much higher
occurrence of harder/rougher or muddy substratum types than was observed.
Table 3-36. Substratum_Code3: Confusion matrix of total counts observed vs. predicted
Observed
Reef
Boulders
Cobbles
Gravel
Sand Silt
Mud
Sand Waves
Total
1
1689
158
109
767
6059
752
159
9693
1
2
3
4
5
6
7
2
4
1979
148
817
5757
603
325
9633
Predicted
4
5
3
8
31
2235
1075
5467
375
353
9544
1
6108
16789
702
74
23674
37181
1595
38776
6
7
1
27
26
2325
16754
1691
4603
25427
495
15946
13287
58
29786
Total
1702
2195
2519
11587
103953
19005
5572
146533
Table 3-37. Substratum_Code3: Confusion matrix of percentage contribution as a percentage of row totals
Observed
Reef
Boulders
Cobbles
Gravel
Sand Silt
Mud
Sand Waves
1
2
3
4
5
6
7
1
99.2
7.2
4.3
6.6
5.8
4.0
2.9
2
0.2
90.2
5.9
7.1
5.5
3.2
5.8
3
0.5
1.4
88.7
9.3
5.3
2.0
6.3
Predicted
4
5
0.0
52.7
16.2
3.7
1.3
35.8
8.4
6
4.3
15.3
69.9
1.0
7
0.1
1.2
1.0
20.1
16.1
8.9
82.6
Total
100
100
100
100
100
100
100
Table 3-38. Substratum_Code3: Confusion matrix of percentage contribution as a percentage of total
Observed
Reef
Boulders
Cobbles
Gravel
Sand Silt
Mud
Sand Waves
Total
1
2
3
4
5
6
7
1
1.15
0.11
0.07
0.52
4.13
0.51
0.11
6.61
2
0.00
1.35
0.10
0.56
3.93
0.41
0.22
6.57
3
0.01
0.02
1.53
0.73
3.73
0.26
0.24
6.51
Predicted
4
5
0.00
4.17
11.46
0.48
0.05
16.16
25.37
1.09
26.46
6
0.34
10.88
9.07
0.04
20.33
7
0.00
0.02
0.02
1.59
11.43
1.15
3.14
17.35
Total
1.16
1.50
1.72
7.91
70.94
12.97
3.80
45.78
GBR Seabed Biodiversity
3-195
3.6.3.8. Substratum v3 with depth partitioning
The raw acoustics data was observed to have a substantial depth dependency (Sections 3.6.1, 3.6.2),
and though the QTC View has a patented proprietary algorithm intended to compensate for the depth
effect, most of the 166 QTC features had a positive or negative correlation (r) with depth of 0.2–0.5 —
thus, the QTC data also had a strong depth dependency. Consequently, the data was partitioned into
six depth classes, 5-10 m, 10-15 m, 15-20 m, 20-30 m, 30-40 m, and 40-60 m, for separate training
and classification analyses of the third substratum recoding schema (Table 2-13) of seven classes. As
above, cross-tabulations of observed versus classified results are presented as a series of confusion
matrices. Table 3-39 shows the counts, Table 3-40 shows the percentages of row total observations
and Table 3-41 shows the percentages of the total observations.
Table 3-39. Substratum_Code3 Depth Partitioned: Confusion matrix of total counts observed vs. predicted
Observed
Reef
Boulders
Cobbles
Gravel
Sand Silt
Mud
Sand Waves
Total
1
1595
62
55
706
5257
667
150
8492
1
2
3
4
5
6
7
2
60
1995
139
930
5477
607
360
9568
Predicted
4
5
3
15
12
1964
914
4613
317
212
8047
28
6376
15298
1320
144
23166
341
40207
1758
42306
6
2
4
293
13549
12685
37
26570
7
2
24
172
1690
14743
1585
4667
22883
Total
1672
2095
2362
11250
99144
18939
5570
141032
Table 3-40. Substratum_Code3 Depth Partitioned: Confusion matrix of contribution as a percentage of rows
Observed
Reef
Boulders
Cobbles
Gravel
Sand Silt
Mud
Sand Waves
1
95.4
3.0
2.3
6.3
5.3
3.5
2.7
1
2
3
4
5
6
7
2
3.6
95.2
5.9
8.3
5.5
3.2
6.5
3
0.9
0.6
83.1
8.1
4.7
1.7
3.8
Predicted
4
5
0.0
0.0
0.0
0.0
1.2
0.0
56.7
3.0
15.4
40.6
7.0
9.3
2.6
0.0
6
0.0
0.1
0.2
2.6
13.7
67.0
0.7
7
0.1
1.1
7.3
15.0
14.9
8.4
83.8
Total
100
100
100
100
100
100
100
Table 3-41. Substratum_Code3 Depth Partitioned: Confusion matrix of contribution as a percentage of total
Observed
Reef
Boulders
Cobbles
Gravel
Sand Silt
Mud
Sand Waves
Total
1
2
3
4
5
6
7
1
1.1
0.0
0.0
0.5
3.7
0.5
0.1
6.0
2
0.0
1.4
0.1
0.7
3.9
0.4
0.3
6.8
3
0.0
0.0
1.4
0.6
3.3
0.2
0.2
5.7
Predicted
4
5
0.0
0.0
0.0
0.0
0.0
0.0
4.5
0.2
10.8
28.5
0.9
1.2
0.1
0.0
16.4
30.0
6
0.0
0.0
0.0
0.2
9.6
9.0
0.0
18.8
7
0.0
0.0
0.1
1.2
10.5
1.1
3.3
16.2
Total
1.2
1.5
1.7
8.0
70.3
13.4
3.9
49.3
As in the previous section, the Sands-Silt class represented about 71% of the observations and the
classification success for this class improved from 35.5% to a still rather poor 40.6%, as well as
remaining grossly under-represented in the classification results due large number of observations
being classified as other substrata. Again, given the values of the diagonals, the other observed classes
GBR Seabed Biodiversity
3-196
appeared to have much greater classification success and the overall classification success improved to
49.3%. However, despite the small improvement, as before these classes were seriously over-classified
in the results — again erroneously indicating a much higher occurrence of harder/rougher or muddy
substratum types than was observed.
3.6.3.9. QTC results summary
The depth partitioned analyses were conducted on all of the re-coding schemas presented above,
though details of all have not been discussed. However, a condensed version of the overall
classification success results from the separate analyses yields a summary performance table (Table
3-42). Spatial partitioning of the data into 1 by 1 degree blocks was also attempted, but did not yield
perceptible improvements and has not been presented.
It is clear that the depth partitioning generally leads to some improvement in classification
performance and performance decreases inversely with the number of seabed class types. It is also
clear that substantial further reduction in the number of classes (<5 to 3) would be necessary in an
attempt to raise the level of classification success to a satisfactory level. However, so few
distinguishable classes would have little information content of value in terms of broad scale seabed
habitat mapping as only a few classes of substratum (i.e. mud, sand, rocks, reef) intersecting with only
very basic biological habitats (i.e. none, bioturbation, vegetation, epibenthos, hard-coral) would realize
at least 8 of 16 possible combinations.
Table 3-42. Summary of LDA classification performance of QTC View data.
Analysis
Substrate v1, Habitat v2
Substrate v2, Habitat v2
Substrate v3, Habitat v2
BioHabitat v2
BioHabitat v3
Substrate v1
Substrate v2
Substrate v3
Name
sub_hab_cod2
sub_hab_cod3
sub_hab_cod4
habitat_cod3
habitat_cod2
sbstrt_code
sbstrt_cod2
sbstrt_cod3
Potential
classes
108
84
84
12
8
9
7
7
Actual
classes
102
79
78
12
8
9
7
7
No
partitioning
3.4%
6.0%
27.7%
38.4%
36.6%
45.0%
45.8%
Depth
partitioning
4.6%
5.8%
5.9%
29.5%
40.5%
39.0%
49.1%
49.3%
GBR Seabed Biodiversity
3-197
3.7. ECOLOGICAL RISK INDICATORS
The basic approach to establishing the ecological risk indicators involved estimating the proportion of
area or biomass of an assemblage, species group or species in various zones of the GBRMP or
exposed to trawling and at various intensities of effort. The study area on the continental shelf of the
GBRMP (excluding islands, coral reefs and shallow shoals < ~12 m, and coastal shallows < ~7 m) was
almost 200,000 km², of which 44% was zoned General Use, 28% was Habitat Protection, 28% was
Marine National Park (and Conservation Park), and <1% was Preservation (Table 3-43).
Table 3-43: Total area and percentage of the study area on the continental shelf of the GBRMP in various
management zones considered for estimating ecological risk indicators.
ZONE
Area km²
Area %
General Use
87,016
44
Habitat Protection
56,709
28
Marine National Park
55,535
28
Preservation
383
<1
TOTAL
199,644
100
Of the almost 200,000 km² study area, just over 47,000 km² of 0.01º study grid cells had trawl effort
recorded by VMS in 2005. For most of this area, the level of effort was very low and only fractions of
these grid cells were swept by trawl gear. At the other extreme, about 10,000 km² of 0.01º grid cells
were trawled with ≥8 hours of effort, which if distributed uniformly was roughly sufficient to cover a
0.01º cell one or more times. In these areas, the effective area trawled was approximately the same as
the grid cell area even though the total swept area was greater (Table 3-44). Thus, while the total area
of seabed swept in 2005 was approximately 38,500 km², the actual area of seabed potentially affected
was much less (at approximately 17,200 km²) because of aggregation of the majority of effort (~80%)
into a small area (~5%), with consequent environmental benefit to the vast remainder of the seabed
environment. Further, because of the assumption here of uniformly distributed effort within grid cells,
the estimate of 17,200 km² is likely to be an upper estimate. In reality, trawling at this scale would be
random or even aggregated with the consequence that slightly less seabed would be potentially
affected (Ellis & Pantus 2001); possibly only ~13,000 km² in the case of random distribution within
cells. On the other hand, such very small scale randomness or aggregations are unlikely to be
consistent from year to year, even though the larger scale pattern is very consistent. Thus, over time
periods of years the total area of seabed affected would be greater than that in any single year, but the
true area is uncertain. The estimates provided here indicate a likely range for the area of seabed
affected by trawling.
Table 3-44: Total area of the study area on the continental shelf of the
GBRMP exposed to various levels of trawl effort, measured by VMS in
2005, considered for estimating ecological risk indicators. The total effective
area trawled and total area swept are also estimated.
Effort interval
(hrs / 0.01º cell)
0
0.125
0.25
0.5
1
2
4
8
16
32
64
128
256
Total
Area
km²
152,419
10,554
6,054
5,718
5,139
4,804
4,839
4,746
3,198
1,533
593
44
5
199,644
Effort
hrs
0
1,175
1,840
3,486
6,302
11,870
24,059
46,680
61,670
57,168
43,576
5,900
1,266
264,991
Effective
area km²
0
171
268
507
917
1,727
3,500
4,746
3,198
1,533
593
44
5
17,207
Swept
area km²
0
171
268
507
917
1,727
3,500
6,790
8,971
8,316
6,339
858
184
38,546
GBR Seabed Biodiversity
3-198
3.7.1. Indicators for species-groups biomass
On the basis of the biophysical model predictions of species-group biomass distributions, the first
indicator considered was the amount of biomass of each group located in various marine park zones,
in particular the percentage of the total located in General Use (GU) zones was available to trawling
and potentially at risk (Table 3-45). Thirty six of the 38 species groups had more than 25% of their
biomass in GU zones (Table 3-45, pale orange) and 12 groups had more than 50% of their biomass in
GU zones (Table 3-45, dark orange). The lowest level of availability was 23% and the highest level
was 65%.
The next indicator considered was the percentage of biomass of each species-group located in grid
cells where trawl effort was present — regardless of the intensity of effort in the grid cells (Table
3-45). Fifteen of the 38 species groups had more than 25% of their biomass in grid cells with trawl
effort (Table 3-45, pale orange) and none had more than 50% of their biomass in grid cells with trawl
effort. The lowest level of exposure was 10% and the highest level was 41%. This indicator is more
specific and more sensitive than the previous.
The third indicator for species-groups was the percentage of biomass of each group directly exposed to
trawl effort taking into account the intensity of trawl effort (Table 3-45). The table shows the amount
of biomass exposed at several different levels of effort intensity, as well as the final total exposure as a
percentage. Recall that as ~8 hrs of effort is sufficient to cover a 0.01º grid cell once, all biomass in
cells with 8 hrs was considered exposed, whereas in cells with say 4 hrs only half the biomass was
exposed and in cells with say 16 hrs the biomass was exposed 2 times. This method was an
approximation and leads to an upper limit. That is, the level of exposure was unlikely to exceed the
estimates provided. Only seven of the 38 species groups had more than 25% of their biomass directly
exposed to trawl effort in 2005 (Table 3-45, pale orange) and only one had more than 50% of biomass
directly exposed (Table 3-45, dark orange) (Figure 3-106, Figure 3-107). The lowest level of exposure
to effort was 7% and the highest level was 60%. This most specific and sensitive indicator suggested
that 713 of 840 OTUs (85%) represented by the 38 species-groups have very low risk of exposure, 94
(11%) have moderately low risk of exposure, and 22 (3%) have moderately high risk of exposure. For
each of the higher exposure groups, the potential risk for each member species was examined in
further detail (Section 3.7.2).
Finally, the trawl effort coefficient of the biophysical models for species-groups was considered where
included. Trawl effort was selected in 12 of the 38 group models and was significant in 11 cases. For
the highest exposure-ranked group#29, the trawl effort coefficient was positive 0.009 for the biomass
component of the model, suggesting that this group was ~0.9% more abundant per annual hour of
effort per 0.01º grid cell, with a possible implication that this group was almost 5% more abundant as
a result of trawling. For the 8th exposure-ranked group#10 (with an exposure of 24%), the trawl effort
coefficient was positive for the presence component of the model, but there was a negative interactive
with mud, with a possible implication that this group had + ~0.1% overall change in abundance as a
result of trawling. For the 12th exposure-ranked group#1 (with an exposure of 21%), the trawl effort
coefficient was -0.019 for the presence component of the model, suggesting that this group was ~1.9%
less likely to be present per annual hour of effort per 0.01º grid cell, with a possible implication that
this group was ~0.2% less abundant as a result of trawling. For the 15th exposure-ranked group#8
(Figure 3-107, with an exposure of 20%), the trawl effort coefficient was -0.08 for the presence model,
but there was a negative interactive with carbonate, with a possible implication that this group had
~1.2% overall decrease in abundance as a result of trawling. The remaining eight groups with
significant trawl effort terms had coefficients ranging from -0.01 to -0.04 and possible decreases in
abundance as a result of trawling of -0.1% to -6%; their exposures to trawl effort ranged from 14% to
7%. Those groups with greatest potential negative change in biomass are shown in Figure 3-107. For
each of the groups with significant trawl coefficients, the potential risk for each member species was
examined in further detail below (Section 3.7.2).
GBR Seabed Biodiversity
3-199
Table 3-45: Ecological Risk Indicators with respect to trawling for estimated Biomass (tonnes) of correlated species groups: by GBRMP Zoning indicating percent of biomass
available; by areas not trawled/trawled indicating percent biomass potentially exposed; by trawl intensity (ann_hrs/0.01º cell) indicating percent biomass directly exposed to effort.
Group
29
29
29
29
29
29
29
29
29
29
29
29
29
29
29
29
29
29
29
29
29
29
29
29
29
29
29
29
29
29
29
29
29
29
29
29
29
29
Number
of OTUs
22
23
23
16
19
8
5
39
12
22
23
39
16
11
20
3
17
18
12
28
8
4
29
8
39
20
11
25
11
29
26
21
60
10
22
80
23
27
General
Habitat
Marine
PreserTotal
%
Use
Protection Nat Park
vation biomass Available
1716
559
790
5
3069
56
2469
756
1225
5
4454
55
2637
643
993
5
4278
62
736
387
408
4
1534
48
570
165
298
1
1034
55
1078
203
388
1
1670
65
316
126
156
1
599
53
19805
5473
8779
161
34219
58
1822
816
1234
2
3875
47
5737
2270
2485
17
10510
55
4992
2356
2495
45
9888
50
5900
3632
3369
44
12946
46
112
34
45
0
191
59
2040
854
1029
7
3929
52
427
192
231
2
852
50
83
60
58
0
201
41
602
304
368
3
1277
47
1484
962
862
10
3318
45
1675
989
1137
4
3806
44
1344
1140
826
9
3319
41
768
665
523
8
1965
39
241
173
149
1
564
43
352
387
358
2
1100
32
27
34
25
0
87
31
9350
7451
7386
100
24287
38
1206
1059
1055
5
3325
36
194
107
125
0
426
46
1638
1071
1287
7
4003
41
640
550
451
4
1646
39
10531
18911
11332
329
41104
26
181
207
153
2
543
33
316
317
248
5
886
36
7137
10786
8616
88
26626
27
66
102
88
1
257
26
217
360
258
4
838
26
2439
4204
3214
33
9890
25
2749
3097
2650
26
8521
32
254
475
398
3
1131
23
Not
%
Trawled
trawled
Exposed
1970
1099
36
2925
1529
34
2672
1606
38
1098
437
28
694
340
33
1028
642
38
428
170
28
20170
14050
41
3003
872
22
6915
3595
34
6786
3103
31
8864
4082
32
122
69
36
2764
1166
30
572
280
33
157
44
22
870
407
32
2495
823
25
2877
929
24
2553
766
23
1546
419
21
443
121
21
939
161
15
72
15
18
18922
5365
22
2750
575
17
344
82
19
3295
707
18
1326
320
19
35009
6094
15
453
89
16
716
171
19
22812
3814
14
223
34
13
722
117
14
8450
1440
15
7297
1223
14
1012
119
10
0 0.125 0.25 0.5
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
3
5
6
1
1
3
1
52
3
12
12
12
0
4
1
0
2
3
3
3
1
0
1
0
20
2
0
3
1
21
0
1
15
0
0
6
5
1
1
2
4
8
16
32
64
128 256
6 11 19 34
72 155 287 431 598 130
8 16 28 54 112 227 361 481 506 81
9 17 30 56 113 228 347 386 329 51
2
4
8 15
33
67 110 142 152 22
2
4
6 12
22
45
70
83 70 12
4
7 12 23
43
90 119
90 39
5
1
2
3
6
14
27
36
34 28
4
80 146 277 547 1157 2301 2365 1147 174
4
5
9 15 30
60 128 204 223 193 28
20 38 72 144 309 592 652 393 112 14
19 34 61 112 214 429 544 444 244 27
21 42 84 174 382 753 808 387 61
1
0
1
1
3
5
10
11
6
1
0
7 13 23 45
93 180 209 146 64 11
2
3
5 11
26
46
45
26
6
1
0
0
1
2
3
6
8
9
7
1
2
5
8 16
31
60
63
36
6
0
5
9 16 32
64 126 149 106 60
9
5 10 19 36
76 149 177 106 35
4
4
8 15 31
66 129 139
70 22
3
2
5
9 17
35
67
75
44 21
2
1
1
2
5
9
16
19
15
7
1
1
2
3
5
10
20
29
36 34
5
0
0
0
1
1
2
3
2
2
0
32 59 110 204 416 791 848 522 171 14
3
6 11 21
43
80
94
68 33
3
1
1
2
3
5
9
12
8
4
0
4
9 14 26
48
87 105
85 37
4
2
4
7 12
26
41
42
27 10
1
34 67 124 239 499 972 1053 672 360 47
1
1
2
3
7
13
14
8
3
0
1
2
4
7
15
22
18
8
1
0
23 43 77 138 272 523 535 358 173 29
0
0
1
1
2
4
4
4
2
0
1
1
3
4
8
14
15
11
2
0
9 17 30 52 101 195 185
89 16
1
8 14 26 46
91 160 153
77 19
2
1
1
2
4
8
14
17
15 11
2
87
15
14
6
3
2
1
0
8
4
6
0
0
2
0
0
0
2
1
0
0
0
1
0
2
0
0
0
0
1
0
0
7
0
0
0
0
0
Effort
Exp%
60
42
37
37
32
26
26
24
23
22
22
21
20
20
20
19
18
18
16
15
14
14
13
13
13
11
11
11
10
10
10
9
8
8
7
7
7
7
GBR Seabed Biodiversity
(a) Group 29
(b) Group 9
(c) Group 22
(d) Group 14
3-200
Figure 3-106: Distribution maps of the most exposed species groups (a) exposed over 50 %, (b) – (d) exposed
by 25-50%
GBR Seabed Biodiversity
(a) Group 13
(b) Group 33
(c) Group 35
(d) Group 6
3-201
Figure 3-107: Distribution maps of the most exposed species groups: (a) and (b) exposed by 25-50%; and
species groups with negative trawl effort coefficients and possible population decreases in abundance as a result
of trawling of >5%; (c) -5.3% and (d) -6% respectively.
GBR Seabed Biodiversity
3-202
3.7.2. Indicators for individual species biomass
Risk indicators were estimated for all individual species that were members of the species groups that
appeared to be at higher levels of potential risk. The results for group #29 species are shown in Table
3-46, in a similar though abbreviated form as for species groups in Table 3-45. All but one of the 22
species in this group had high proportions of their exposed biomass in areas of high effort, so that their
total effort exposed biomass was greater than their trawled biomass. Four species appear to have very
high levels of exposure — greater than 100% of their standing biomass. Given that group 29 had the
greatest overlap with the highest levels of trawl effort, it is not surprising a key prawn target species,
Penaeus semisulcatus (grooved tiger prawn), was a member of this group and had a very high level of
exposure — even though 26% of its biomass was protected by zoning and 36% of its biomass was not
exposed to any effort, the exposed 64% of biomass was trawled an average of more than 2.7 times in
2005, contributing to a total indicator of 174% of standing biomass exposed to trawl effort (Figure
3-108a). The second most exposed species in group 29 was Cryptolutea arafurensis (a Pilumnid
Crab), with 57% of its biomass available in GU zones, 41% in cells with recorded effort that was
trawled an average of ~3.1 times giving a total of ~128% total direct exposure to trawling (Figure
3-108b). The third most exposed species in group 29 was Brachirus muelleri (a Pleuronectiform
flatfish), with 69% of its biomass available in GU zones, 59% in cells with recorded effort that was
trawled an average of ~2 times giving a total of 119% exposure to trawling (Figure 3-108c). The
fourth most exposed species in group 29 was Pentaprion longimanus (a Gerreid fish), with 62% of its
biomass available in GU zones, 48% in cells with recorded effort that was trawled an average of ~2.4
times giving a total of 117% exposure to trawling (Figure 3-108d). Another two species had high
levels of exposure — between 75% and 100% of their standing biomass. Terapon puta (a Terapontid
fish), had 56% of its biomass available in GU zones, 47% in cells with recorded effort that was
trawled an average of ~1.6 times giving a total of 78% exposure to trawling (Figure 3-109a). The
Bivalve mollusc Enisiculus cultellus had 61% of its biomass available in GU zones, 46% in cells with
recorded effort that was trawled an average of ~1.6 times giving a total of 75% exposure to trawling
(Figure 3-109b). Four species had moderate-high levels of exposure — between 50% and 75% of their
standing biomass. The prawn Metapenaeus ensis had 67% of its biomass available in GU zones, 49%
in cells with recorded effort that was trawled an average of ~1.4 times giving a total of 67% exposure
to trawling (Figure 3-109c). The lizardfish Saurida argentea/tumbil had 58% of its biomass available
in GU zones, 38% in cells with recorded effort that was trawled an average of ~1.6 times giving a total
of 63% exposure to trawling (Figure 3-109d). The Bivalve mollusc Placamen tiara had 55% of its
biomass available in GU zones, 35% in cells with recorded effort that was trawled an average of ~1.6
times giving a total of 55% exposure to trawling (Figure 3-110a). Euristhmus nudiceps (a Plotosid
fish) had 56% of its biomass available in GU zones, 33% in cells with recorded effort that was trawled
an average of ~1.6 times giving a total of 55% exposure to trawling (Figure 3-110a). Ten species had
moderate-low exposures of 25-50% and one species had low exposure <25% (Table 3-46).
The results for group #9 species are shown in Table 3-47. All of the 23 species in this group had high
proportions of their exposed biomass in areas of high effort, so that their total effort exposed biomass
was greater than their trawled biomass. While as a correlated species group, group#9 had moderatelow exposure, three individual species appeared to have high levels of exposure: Leiognathus
leuciscus (a Leiognathid ponyfish), with 59% of its biomass available in GU zones, 43% in cells with
recorded effort that was trawled an average of ~2.2 times giving a total of 95% exposure to trawling
(Figure 3-110b); Upeneus sundaicus (a Mullid fish), with 63% of its biomass available in GU zones,
50% in cells with recorded effort that was trawled an average of ~1.9 times giving a total of 93%
exposure to trawling (Figure 3-110c); and Portunus gracilimanus (a crab), with 59% of its biomass
available in GU zones, 38% in cells with recorded effort that was trawled an average of ~2.2 times
giving a total of 86% exposure to trawling (Figure 3-110d). Five species from group 9 had a moderatehigh exposure, including fishes Calliurichthys grossi and Cynoglossus maculipennis, bivalves
Amusium pleuronectes cf and Melaxinaea vitrea, and the bug lobster Thenus parindicus — with 54–
60% of their biomass available in GU zones, 36–39% in cells with recorded effort that was trawled an
average of ~1.4–1.7 times giving totals of 52–59% exposure to trawling (see Figure 3-111a-d and
Figure 3-112a).
GBR Seabed Biodiversity
3-203
The results for group #22 species are shown in Table 3-48. Again, all of the species in this group had
high proportions of their exposed biomass in areas of high effort, so that their total effort exposed
biomass was greater than their trawled biomass. While as a correlated group, group#22 had moderatelow exposure, seven individual species appeared to have moderate-high levels of exposure, including
fishes: Leiognathus splendens, Psettodes erumei, Terapon theraps, Upeneus sulphureus, crustaceans:
Erugosquilla woodmasoni, Myra tumidospina, and a gastropod: Nassarius cremmatus cf — with 54–
70% of their biomass available in GU zones, 38–49% in cells with recorded effort that was trawled an
average of ~1.2–1.6 times giving totals of 54–65% exposure to trawling (see Figure 3-112b-d and
Figure 3-113a-d). Sixteen group #22 species had moderate-low exposures of 30–50% and none had
low exposure <25%.
The results for group #14 species are shown in Table 3-49. All but two of the 16 species in this group
had high proportions of their exposed biomass in areas of high effort, so that their total effort exposed
biomass was greater than their trawled biomass. One species, Scolopsis taeniopterus (a Nemipterid
fish), appeared to have a moderate-high level of exposure with 51% of its biomass available in GU
zones, 33% in cells with recorded effort that was trawled an average of ~1.7 times giving a total of
54% exposure to trawling (Figure 3-114a). Ten species had moderate-low exposures between 25-50%
and five species had low exposure <25%.
The results for group #13 species are shown in Table 3-50. Twelve of the 19 species in this group had
high proportions of their exposed biomass in areas of high effort, so that their total effort exposed
biomass was greater than their trawled biomass. Two species from group 13 had a moderate-high
exposures including: Repomucenus belcheri (a Callionymid fish) and Trachypenaeus anchoralis (a
prawn) — with 64% of their biomass available in GU zones, 42–44% in cells with recorded effort that
was trawled an average of ~1.3–1.5 times giving totals of 53–67% exposure to trawling (see Figure
3-114 bc). Ten species had moderate-low exposures between 25-50% and seven species had low
exposure <25%.
The results for group #33 species are shown in Table 3-51. All eight species appeared to have
moderate-low levels of exposure, with 49–67% of their biomass available in GU zones, 28–40% in
cells with recorded effort that was trawled an average of about 0.9 times giving a total of 27–35%
exposure to trawling.
Group #27 overall had a low level of exposure (Table 3-45) and the results for species in this group
are shown in Table 3-52. Two of the five species in this group had high proportions of their exposed
biomass in areas of high effort, so that their total effort exposed biomass was somewhat greater than
their trawled biomass. No species had high exposure; four species appeared to have moderate-low
levels of exposure, with 50–59% of their biomass available in GU zones, 25–33% in cells with
recorded effort that was trawled an average of about 1.1 times giving a total of 25–37% exposure to
trawling. Only one species had low exposure <25%.
While the remaining 31 of the 38 groups, representing 713 (85%) of the 840 species examined, as
groups had low levels of exposure (<25%) (Table 3-45), the species members of each were examined
in more detail and key summary results are presented (Table 3-53). Of these remaining species, 91 had
50-69% of their biomass available in General Use, 568 had 25-50% of their biomass available and 52
had <25%. Only three species had more than 50% (51%, 54% and 60%) of their biomass in cells with
recorded effort, 235 species had 25-50% of their biomass in trawled cells and 474 species had <25%.
One additional species had high total direct exposure to trawl effort: Pelates quadrilineatus (a
Terapontid fish), with 69% of its biomass available in GU zones, 47% in cells with recorded effort that
was trawled an average of ~2.2 times giving a total of 103% exposure to trawling (Figure 3-114d).
Four species had moderate-high levels of direct exposure (Table 3-53): Brachaluteres taylori (a
Monocanthid filefish), with 71% of its biomass available in GU zones, 60% in cells with recorded
effort that was trawled an average of ~1.2 times giving a total of 72% exposure to trawling (Figure
3-115a), Leiognathus bindus (a Leiognathid pony fish), with 42% of its biomass available in GU
zones, 28% in cells with recorded effort that was trawled an average of ~2.3 times giving a total of
63% exposure to trawling (Figure 3-115b), Yongeichthys nebulosus (a Gobiid fish), with 42% of its
biomass available in GU zones, 25% in cells with recorded effort that was trawled an average of ~2.1
times giving a total of 51% exposure to trawling (Figure 3-115c), and Apogon poecilopterus (a
Cardinal fish), with 50% of its biomass available in GU zones, 34% in cells with recorded effort that
GBR Seabed Biodiversity
3-204
was trawled an average of ~1.5 times giving a total of 51% exposure to trawling (Figure 3-115d). 144
species had moderate-low exposures of 25-38% (Table 3-53) and 563 species had low exposure <25%.
Table 3-46: Ecological Risk Indicators with respect to trawling for estimated Biomass (kg) of species in
group#29: biomass available in General Use zone; biomass potentially exposed in trawled cells; and biomass
directly exposed to trawl effort. Pale orange: >25% biomass exposed; dark orange: >50% biomass exposed; red:
>50% biomass exposed.
Class
Actinopterygii
Actinopterygii
Actinopterygii
Actinopterygii
Actinopterygii
Actinopterygii
Actinopterygii
Actinopterygii
Actinopterygii
Bivalvia
Bivalvia
Bivalvia
Bivalvia
Bivalvia
Cephalopoda
Crustacea
Crustacea
Crustacea
Crustacea
Crustacea
Crustacea
Genus
Apogon
Brachirus
Epinephelus
Euristhmus
Nemipterus
Pentaprion
Saurida
Terapon
Tripodichthys
Corbula
Corbula
Dosinia
Enisiculus
Placamen
Sepia
Ceratoplax
Cryptolutea
Iphiculus
Liagore
Metapenaeus
Penaeus
Species
fasciatus
muelleri
sexfasciatus
nudiceps
hexodon
longimanus
argentea/tumbil
puta
angustifrons
fortisulcata
sp2
altenai
cultellus
tiara
elliptica
ciliata
arafurensis
spongiosus
rubromaculata
ensis
semisulcatus
Biomass
223485
80330
285546
1374323
1421345
61963
1109937
60300
43969
6462
828447
191530
984
3225
158747
562
480
364
49419
31126
301314
Gen
Use
%
Available
96920
55595
136199
766789
733021
38453
648706
33734
19930
2765
396212
96271
596
1787
81668
234
273
156
23704
20849
222786
43
69
48
56
52
62
58
56
45
43
48
50
61
55
51
42
57
43
48
67
74
Not
%
Trawled
trawled
Exposed
Effort Effort
Exp Exp%
170194
33330
224032
926088
1126693
31941
687390
31663
28301
4996
605599
139017
535
2084
111826
436
282
298
37024
15901
109085
69893
95910
86811
697347
456490
72566
696889
47047
22028
1709
241721
71166
744
1788
60233
172
616
68
21489
20893
524946
53291
47000
61514
448235
294652
30022
422547
28638
15668
1466
222848
52514
449
1141
46921
126
198
66
12395
15225
192229
24
59
22
33
21
48
38
47
36
23
27
27
46
35
30
22
41
18
25
49
64
31
119
30
51
32
117
63
78
50
26
29
37
75
55
38
30
128
18
43
67
174
Table 3-47: Ecological Risk Indicators with respect to trawling for estimated Biomass (kg) of species in
group#9: biomass available in General Use zone; biomass potentially exposed in trawled cells; and biomass
directly exposed to trawl effort. Pale orange: >25% biomass exposed; dark orange: >50% biomass exposed; red:
>50% biomass exposed.
Class
Actinopterygii
Actinopterygii
Actinopterygii
Actinopterygii
Actinopterygii
Actinopterygii
Actinopterygii
Asteroidea
Asteroidea
Bivalvia
Bivalvia
Bivalvia
Bivalvia
Bivalvia
Bivalvia
Crustacea
Crustacea
Crustacea
Crustacea
Crustacea
Echinoidea
Gastropoda
Ophiuroidea
Genus
Calliurichthys
Cynoglossus
Cynoglossus
Inegocia
Leiognathus
Selaroides
Upeneus
Astropecten
Astropecten
Amusium
Antigona
Corbula
Melaxinaea
Modiolus
Trisidos
Calappa
Dorippe
Penaeus
Portunus
Thenus
Brissopsis
Lamellaria
Dougaloplus
Species
grossi
maculipinnis
sp 1 punctate
japonica
leuciscus
leptolepis
sundaicus
granulatus cf
sp4_AIM
pleuronectes cf
lamellaris
macgillvrayi
vitrea
elongatus
semitortata
sp44
sp7142-12
esculentus
gracilimanus
parindicus
luzonica
sp1
echinata
Biomass
171819
78915
80719
1096930
171753
586810
370945
16683
11187
824663
23273
205900
171979
39291
402307
10969
601019
1031505
204641
518607
1377669
5697
2513
Gen
Use
%
Available
93340
47483
44939
659059
101800
326375
232738
7633
5025
494743
10562
96200
102275
21864
189840
6240
267505
637143
121469
284865
632373
2604
1187
54
60
56
60
59
56
63
46
45
60
45
47
59
56
47
57
45
62
59
55
46
46
47
Not
%
Trawled
trawled
Exposed
Effort Effort
Exp Exp%
105561
49006
53525
701655
97296
378385
187180
12560
8242
517346
17522
153974
107391
25442
296079
7269
455947
663465
126373
332219
1006889
4175
1845
101488
41438
38139
485552
163452
291238
346289
5134
3703
428647
6953
58988
108090
18670
137145
4746
159314
489438
175640
294405
525901
2139
939
66258
29909
27193
395275
74458
208425
183765
4123
2946
307317
5752
51925
64588
13849
106229
3700
145072
368040
78268
186388
370780
1522
668
39
38
34
36
43
36
50
25
26
37
25
25
38
35
26
34
24
36
38
36
27
27
27
59
52
47
44
95
49
93
31
33
52
30
29
63
47
34
43
26
47
86
57
38
37
37
GBR Seabed Biodiversity
3-205
Table 3-48: Ecological Risk Indicators with respect to trawling for estimated Biomass (kg) of species in
group#22: biomass available in General Use zone; biomass potentially exposed in trawled cells; and biomass
directly exposed to trawl effort. Pale orange: >25% biomass exposed; dark orange: >50% biomass exposed; red:
>50% biomass exposed.
Gen
%
Not
%
Effort Effort
Class
Genus
Species
Biomass
Trawled
Use Available trawled
Exposed
Exp Exp%
Actinopterygii
Actinopterygii
Actinopterygii
Actinopterygii
Actinopterygii
Actinopterygii
Actinopterygii
Actinopterygii
Actinopterygii
Actinopterygii
Actinopterygii
Anthozoa
Crustacea
Crustacea
Crustacea
Crustacea
Crustacea
Echinoidea
Gastropoda
Gastropoda
Gastropoda
Gastropoda
Holothuroidea
Caranx
Gerres
Leiognathus
Nemipterus
Nemipterus
Psettodes
Sillago
Suggrundus
Terapon
Torquigener
Upeneus
Sea pen
Erugosquilla
Myra
Paguristes
Portunus
Portunus
Chaetodiadema
Aplysia
Bufonaria
Gemmula
Nassarius
Holothuria
bucculentus
filamentosus
splendens
peronii
sp juv/unident
erumei
burrus
macracanthus
theraps
whitleyi
sulphureus
sp1
woodmasoni
tumidospina
sp2358-2
hastatoides
tuberculatus
granulatum
sp1_QMS
rana
sp2
cremmatus cf
ocellata
1236784
84315
270168
1355758
6496
361247
307944
559472
359964
150537
723274
507
19542
14791
30865
5197
1226
80329
450338
19213
7259
35852
858968
797276
47619
145446
865424
3035
221081
140790
330620
227353
78812
504361
287
12829
8393
15957
2846
570
38799
227820
10701
3372
19832
445403
64
56
54
64
47
61
46
59
63
52
70
57
66
57
52
55
46
48
51
56
46
55
52
754020
49621
151522
851889
4705
215737
215146
375946
206648
99749
390634
318
10017
9237
21256
3264
882
58367
307454
12811
5226
22024
587871
482764
34695
118646
503869
1791
145511
92797
183525
153316
50788
332639
189
9525
5555
9609
1933
344
21962
142885
6402
2033
13827
271098
39
41
44
37
28
40
30
33
43
34
46
37
49
38
31
37
28
27
32
33
28
39
32
580618
41978
145528
651640
2009
204453
115526
253496
224795
57201
423359
252
12819
8930
11286
2261
391
24544
170472
7564
2314
20331
307892
47
50
54
48
31
56
37
45
62
38
58
50
65
60
36
43
32
30
38
39
32
57
36
Table 3-49: Ecological Risk Indicators with respect to trawling for estimated Biomass (kg) of species in
group#14: biomass available in General Use zone; biomass potentially exposed in trawled cells; and biomass
directly exposed to trawl effort. Pale orange: >25% biomass exposed; dark orange: >50% biomass exposed; red:
>50% biomass exposed.
Gen
%
Not
%
Effort Effort
Class
Genus
Species
Biomass
Trawled
Use Available trawled
Exposed
Exp Exp%
Actinopterygii
Actinopterygii
Actinopterygii
Bivalvia
Bivalvia
Cephalopoda
Crustacea
Crustacea
Crustacea
Crustacea
Crustacea
Crustacea
Crustacea
Gastropoda
Gastropoda
Liliopsida
Fistularia
Nemipterus
Scolopsis
Anadara
Fulvia
Sepia
Charybdis
Cloridina
Diogenidae
Metapenaeus
Portunus
Portunus
Scyllarus
Lophiotoma
Vexillum
Halophila
petimba
nematopus
taeniopterus
ferruginea cf
scalata
pharaonis
truncata
chlorida
sp356-1
endeavouri
spinipes
tuberculosus
sp3418
acuta
obeliscus cf
tricostata
135435
693470
1016419
11718
3799
139386
437520
374
442
534272
5017
394
29264
4385
2302
911642
59846
255198
517873
5388
1503
71540
210733
110
200
275629
1764
187
11114
2383
1007
413948
44
37
51
46
40
51
48
29
45
52
35
47
38
54
44
45
100778
546854
685276
9089
3075
92518
303449
362
331
367939
4436
270
23714
2897
1730
682405
34657
146616
331143
2630
723
46868
134071
12
111
166333
582
123
5550
1488
572
229238
26
21
33
22
19
34
31
3
25
31
12
31
19
34
25
25
43146
250954
549892
2563
896
67431
201102
9
162
246691
678
181
5861
1944
910
270360
32
36
54
22
23
48
46
2
36
46
13
46
20
44
39
30
GBR Seabed Biodiversity
3-206
Table 3-50: Ecological Risk Indicators with respect to trawling for estimated Biomass (kg) of species in
group#13: biomass available in General Use zone; biomass potentially exposed in trawled cells; and biomass
directly exposed to trawl effort. Pale orange: >25% biomass exposed; dark orange: >50% biomass exposed; red:
>50% biomass exposed.
Gen
%
Not
%
Effort Effort
Class
Genus
Species
Biomass
Trawled
Use Available trawled
Exposed
Exp Exp%
Actinopterygii
Actinopterygii
Actinopterygii
Actinopterygii
Actinopterygii
Actinopterygii
Actinopterygii
Actinopterygii
Actinopterygii
Actinopterygii
Bivalvia
Bivalvia
Crustacea
Crustacea
Crustacea
Crustacea
Crustacea
Gastropoda
Gastropoda
Apogon
Cynoglossus
Leiognathus
Leiognathus
Parapercis
Polydactylus
Pomadasys
Repomucenus
Siganus
Trixiphichthys
Solen
Solen
Cryptopodia
Eucrate
Leucosia
Pagurid
Trachypenaeus
Murex
Natica
cavitiensis
sp juv/unident
cf bindus
moretoniensis
diplospilus
multiradiatus
maculatus
belcheri
canaliculatus
weberi
siphons only
sp3
queenslandi
affinis
ocellata
sp2358-1
anchoralis
brevispina
vitellus
7972
14818
22870
47237
3855
418667
1542585
98260
377618
59106
69535
35340
5162
2137
13523
15945
45119
4747
8642
3738
7957
13528
24755
2275
127018
1000058
63114
160420
33297
34171
15962
2767
1022
8034
6996
29038
2906
4609
47
54
59
52
59
30
65
64
42
56
49
45
54
48
59
44
64
61
53
6186
10208
14551
31263
2477
335464
1007063
57470
304851
39951
53272
29504
3530
1745
8713
12791
25097
3087
6049
1786
4610
8319
15974
1378
83203
535523
40790
72767
19155
16263
5836
1632
392
4811
3154
20022
1660
2593
22
31
36
34
36
20
35
42
19
32
23
17
32
18
36
20
44
35
30
1164
5389
10329
19323
1733
93241
549160
52407
45039
23536
11424
3313
1971
235
5981
2867
30501
1691
2560
14
36
45
41
45
22
35
53
12
40
16
9
38
11
44
18
67
35
29
Table 3-51: Ecological Risk Indicators with respect to trawling for estimated Biomass (kg) of species in
group#33: biomass available in General Use zone; biomass potentially exposed in trawled cells; and biomass
directly exposed to trawl effort. Pale orange: >25% biomass exposed; dark orange: >50% biomass exposed; red:
>50% biomass exposed.
Gen
%
Not
%
Effort Effort
Class
Genus
Species
Biomass
Trawled
Use Available trawled
Exposed
Exp Exp%
Actinopterygii
Actinopterygii
Actinopterygii
Asteroidea
Crustacea
Foraminifera
Gastropoda
Polychaeta
Apistus
carinatus
1073477
Minous
versicolor
92283
Pseudorhombus elevatus
775731
Luidia
hardwicki
26743
Portunus
sanguinolentus 1018755
Discobotellina
biperforata
151281
Nassarius
conoidalis cf
2625
Chloeia
flava
15742
604069
53586
522838
15522
666215
88000
1378
7784
56
58
67
58
65
58
52
49
708259
60500
496676
17560
611257
97759
1869
11292
365218
31783
279055
9183
407498
53522
756
4450
34
34
36
34
40
35
29
28
359383
30116
270665
8964
349207
47083
702
4432
33
32
35
33
34
31
27
28
Table 3-52: Ecological Risk Indicators with respect to trawling for estimated Biomass (kg) of species in
group#27: biomass available in General Use zone; biomass potentially exposed in trawled cells; and biomass
directly exposed to trawl effort. Pale orange: >25% biomass exposed; dark orange: >50% biomass exposed; red:
>50% biomass exposed.
Gen
%
Not
%
Effort Effort
Class
Genus
Species
Biomass
Trawled
Use Available trawled
Exposed Exp Exp%
Actinopterygii
Actinopterygii
Bivalvia
Crustacea
Crustacea
Grammatobothus
Pseudochromis
Leionucula
Nursilia
Trachypenaeus
polyophthalmus
quinquedentatus
superba
sp nov
granulosus
358459 211058
1907 1024
5499 3078
3789 1909
424353 211608
59
54
56
50
50
239915
1404
3738
2800
316595
118544
503
1761
989
107758
33
26
32
26
25
133790
471
2069
984
108573
37
25
37
26
25
GBR Seabed Biodiversity
3-207
Table 3-53: Ecological Risk Indicators with respect to trawling for estimated Biomass (kg) of top ranking
species in groups with low total exposure (<25%): biomass available in General Use zone; biomass potentially
exposed in trawled cells; and biomass directly exposed to trawl effort. Pale orange: >25% biomass exposed; dark
orange: >50% biomass exposed; red: >50% biomass exposed.
Class
Genus
Species
Biomass
Actinopterygii
Actinopterygii
Actinopterygii
Actinopterygii
Actinopterygii
Crustacea
Actinopterygii
Bivalvia
Holothuroidea
Actinopterygii
Crustacea
Actinopterygii
Crustacea
Crustacea
Actinopterygii
Actinopterygii
Actinopterygii
Crustacea
Actinopterygii
Gastropoda
Actinopterygii
Gymnolaemata
Gastropoda
Actinopterygii
Crustacea
Actinopterygii
Chlorophyceae
Holothuroidea
Phaeophyceae
Crustacea
Anthozoa
Demospongiae
Actinopterygii
Gastropoda
Actinopterygii
Gymnolaemata
Demospongiae
Actinopterygii
Actinopterygii
Actinopterygii
Crustacea
Crustacea
Phaeophyceae
Actinopterygii
Echinoidea
Liliopsida
Bivalvia
Echinoidea
Gymnolaemata
Actinopterygii
Actinopterygii
Actinopterygii
Holothuroidea
Rhodophyceae
Actinopterygii
Crustacea
Holothuroidea
Bivalvia
Crustacea
Crustacea
Bivalvia
Liliopsida
Liliopsida
Rhodophyceae
Chlorophyceae
Phaeophyceae
Rhodophyceae
Actinopterygii
Actinopterygii
Actinopterygii
Gastropoda
Gastropoda
Pelates
Brachaluteres
Leiognathus
Yongeichthys
Apogon
Penaeus
Saurida
Amusium
Bohadschia
Pseudorhombus
Ixa
Ambiserrula
Oratosquillina
Portunus
Aploactis
Inimicus
Upeneus
Dorippe
Suezichthys
Strombus
Apogon
Iodictyum
Strombus
Paramonacanthus
Calappa
Scorpaenopsis
Chaetomorpha
Holothuria
Sporochnus
Paradorippe
Virgularia
Xenospongia
Paracentropogon
Strombus
Pseudorhombus
Selenaria
Ircinia
Asterorhombus
Paraploactis
Nemipterus
Scyllarus
Actumnus
Padina
Dactylopus
Salmacis
Halophila
Ctenocardia
Ova
Hippothoa
Torquigener
Chaetodermis
Cynoglossus
Holothuroidea
Chondrophycus
Centriscus
Ebalia
Stichopus
Lomopsis
Austrolibinia
Penaeus
Barbatia
Halophila
Halophila
Dasya
Udotea
Lobophora
Osmundaria
Nemipteridae
Calliurichthys
Trachinocephalus
Xenophora
Philine
quadrilineatus
taylori
bindus
nebulosus
poecilopterus
latisulcatus
grandi/undo
balloti
marmorata cf
arsius
inermis
jugosa
gravieri
pelagicus
aspera
caledonicus
asymmetricus
quadridens
gracilis
vittatus
nigripinnis
spp
campbelli
otisensis
terraereginae
furneauxi
crassa
sp2
comosus
australiensis
sp1
patelliformis
longispinus
dilatatus
spinosus
maculata cf
1255
intermedius
kagoshimensis
furcosus
demani
squamosus
sp.
dactylopus
sphaeroides
decipiens
virgo cf
lacunosus
distans
sp1(gloerfelt-tarp)
penicilligera
sp kopsi group
sp2
sp1
scutatus
lambriformis
ocellatus
sp1
gracilipes
plebejus
parvillosa cf
ovalis
spinulosa
sp1
argentea
variegata
fimbriata
sp juv/unident
ogilbyi
myops
indica
sp1
129842
62129
76017
66438
121050
235627
8331858
2355308
270670
329560
2544
501376
48611
2172862
21363
711097
367368
3584
14695
56120
59272
17890
22441
402900
10382
2174
360585
110155
1515084
2758
2422
599
89968
92276
969118
288844
7482318
154382
18985
4012361
135376
915
658602
63493
342726
3925942
6808
136339
404
734504
119061
58222
44967
29227
19885
1021
2416172
61039
1570
129674
1283
4093618
13547972
60829
785198
14640448
2542368
7775
123272
1028380
25049
8236
Gen
%
Use
Available
89539
69
44181
71
32175
42
27575
42
60196
50
139435
59
4916897
59
1307128
55
186045
69
225655
68
1577
62
342014
68
23338
48
1306892
60
14019
66
459435
65
220133
60
2274
63
8997
61
31951
57
38751
65
9941
56
14456
64
233514
58
4714
45
1303
60
188016
52
70252
64
865602
57
1346
49
984
41
350
59
43681
49
48784
53
559815
58
170877
59
3450630
46
93736
61
11709
62
1986269
50
74608
55
490
54
370925
56
34105
54
193408
56
1964586
50
3706
54
59747
44
224
55
417300
57
69733
59
30773
53
22398
50
15711
54
11253
57
535
52
1188600
49
31169
51
766
49
70628
54
631
49
2096080
51
7234543
53
33266
55
403519
51
7923598
54
1215220
48
4025
52
65083
53
545675
53
11102
44
4432
54
Not
trawled
68895
24955
54928
49795
80291
143343
5264999
1481100
125407
193022
1529
243531
34043
1359976
11222
378816
229876
2238
8207
34728
35138
9814
13451
254992
7518
1285
238093
65074
941720
2025
1811
370
62866
57238
657360
181454
5442057
97646
11732
3018423
91137
583
397645
42492
211011
2735851
4514
104618
258
457019
74249
41652
32119
18081
12852
748
1732647
43409
1164
83784
922
2785392
9056257
40433
517823
9307668
1827412
5444
80429
658876
18242
5559
%
Trawled Exposed
60947
47
37174
60
21089
28
16643
25
40759
34
92283
39
3066860
37
874208
37
145263
54
136538
41
1016
40
257845
51
14568
30
812887
37
10141
47
332282
47
137492
37
1346
38
6488
44
21392
38
24134
41
8076
45
8991
40
147907
37
2864
28
889
41
122492
34
45081
41
573364
38
732
27
611
25
228
38
27102
30
35037
38
311758
32
107390
37
2040262
27
56736
37
7253
38
993938
25
44239
33
332
36
260958
40
21001
33
131716
38
1190092
30
2294
34
31721
23
146
36
277485
38
44813
38
16571
28
12848
29
11146
38
7033
35
273
27
683525
28
17631
29
406
26
45889
35
361
28
1308226
32
4491716
33
20396
34
267375
34
5332780
36
714956
28
2331
30
42843
35
369504
36
6806
27
2677
33
Effort Effort
Exp Exp%
133695 103
45038 72
48103 63
34243 51
62361 51
115925 49
3842936 46
1069696 45
120797 44
146163 44
1126 44
216540 43
20299 42
883971 40
8606 40
279259 39
138716 38
1341 37
5270 36
20071 36
21157 35
6330 35
7935 35
141465 35
3597 34
754 34
123114 34
37578 34
515425 34
928 33
811 33
199 33
29703 33
30403 33
317591 33
94694 33
2446072 33
50228 32
6168 32
1299389 32
43874 32
296 32
213230 32
20420 32
109830 32
1248025 32
2162 32
43065 31
127 31
231619 31
37413 31
18238 31
14018 31
9072 31
6157 31
315 31
744082 31
18758 31
480 30
39633 30
391 30
1247363 30
4115195 30
18456 30
238064 30
4438275 30
768109 30
2340 30
36986 30
308200 30
7464 30
2443 29
GBR Seabed Biodiversity
Class
Bivalvia
Rhodophyceae
Anthozoa
Asteroidea
Rhodophyceae
Crustacea
Bivalvia
Crustacea
Bivalvia
Crustacea
Actinopterygii
Chlorophyceae
Actinopterygii
Gymnolaemata
Actinopterygii
Phaeophyceae
Rhodophyceae
Crustacea
Cephalopoda
Crustacea
Asteroidea
Crustacea
Chlorophyceae
Phaeophyceae
Chlorophyceae
Cephalopoda
Actinopterygii
Echinoidea
Gymnolaemata
Gymnolaemata
Rhodophyceae
Cephalopoda
Crustacea
Demospongiae
Actinopterygii
Anthozoa
Chlorophyceae
Crustacea
Gastropoda
Holothuroidea
Gymnolaemata
Crustacea
Gastropoda
Actinopterygii
Crustacea
Rhodophyceae
Phaeophyceae
Actinopterygii
Chlorophyceae
Actinopterygii
Cephalopoda
Gastropoda
Actinopterygii
Asteroidea
Demospongiae
Bivalvia
Echinoidea
Gymnolaemata
Demospongiae
Gastropoda
Actinopterygii
Rhodophyceae
Echinoidea
Bivalvia
Crustacea
Rhodophyceae
Actinopterygii
Brachiopoda
Rhodophyceae
Phaeophyceae
Actinopterygii
Actinopterygii
Actinopterygii
Actinopterygii
Crustacea
Crustacea
Rhodophyceae
Bivalvia
Genus
Annachlamys
Polysiphonia
Alertigorgia
Astropecten
Laurencia
Scyllarus
Paphia
Neopalicus
Annachlamys
Oratosquillina
Cynoglossus
Codium
Carangidae
Schizomavella
Repomucenus
Dictyotales
Griffithsia
Leucosia
Sepiidae
Portunus
Oreasteridae
Sicyonia
Udotea
Sporochnus
Cladophora
Sepiadariidae
Torquigener
Laganum
Thalamoporella
Retelepralia
Lithophyllum
Cephalopoda
Dromidiopsis
Reniochalina
Cynoglossus
Trachyphyllia
Halimeda
Pagurid
Conus
Holothuroidea
Robertsonidra
Porcellanid
Atys
Sorsogona
Arcania
Heterosiphonia
Dictyopteris
Adventor
Halimeda
Upeneus
Sepia
Biplex
Cynoglossus
Stellaster
Mycale
Spondylus
Peronella
Orthoscuticella
Disyringa
Xenophora
Engyprosopon
Amansia
Laganidae
Chama
Portunus
Gracilaria
Minous
Brachiopoda
Gracilaria
Lobophora
Erosa
Tragulichthys
Choerodon
Zebrias
Pilumnus
Dardanus
Dasya
Chama
Species
kuhnholtzi
sp1
orientalis
spp
sp2
martensii
undulata cf
jukesii
flabellata
quinquedentata
maccullochi
geppii
sp juv/unident
spp
limiceps
sp
sp
formosensis
spp
rubromarginatus
sp1
rectirostris
glaucescens
moorei
sp
sp5
cf pallimaculatus
depressum
spp
mosaica
sp1
spp
edwardsi
stalagmitis
sp4
geoffroyi
sp2
sp17
ammiralis
sp22
spp
sp4154
cylindricus cf
tuberculata
elongata
muelleri
sp2
elongatus
borneenses
sp juv/unident
whitleyana
pulchellum
ogilbyi
equestris cf
mirabilis
wrightianus
orbicularis cf
spp
sp1
solarioides
grandisquama
glomerata
sp3
spp
tenuipes
sp1
trachycephalus
sp1_MTQ
sp2
sp
erosa
jaculiferus
cephalotes
craticula
longicornis
callichela var
sp
pulchella
Biomass
365468
36529
33318
76778
259365
6856
7989
12459
289876
62852
32047
224230
19532
2378
267345
433762
10773
1341
792725
5914213
2814889
259
1402555
2797072
39457
17032
358774
2276625
56643
62
21086914
750874
27212
500459
466488
3171803
477419
2067
17677
1113121
12430
693
4963
1332608
11264
1083464
1307425
11470
10124447
5365
493757
59197
28505
2055943
401414
103088
32469
26462
9021
37081
1306624
1902406
190493
13590
1911035
1475393
649193
79799
975404
6615675
175940
637207
421829
206568
2472
29106
365331
874718
Gen
%
Use
Available
176904
48
19546
54
16607
50
34373
45
129910
50
3462
50
3624
45
5878
47
148742
51
30691
49
18970
59
116326
52
7325
38
937
39
140261
52
227955
53
5585
52
714
53
388586
49
3247328
55
1566347
56
146
56
590506
42
1487058
53
20704
52
8933
52
185425
52
1146355
50
30301
53
28
45
10622744
50
355413
47
12482
46
205995
41
227355
49
1472396
46
218989
46
974
47
8977
51
573036
51
6155
50
341
49
2417
49
705084
53
5306
47
558467
52
674710
52
5029
44
4577094
45
2576
48
248847
50
27910
47
12554
44
1015647
49
181402
45
51044
50
16229
50
12601
48
4141
46
17499
47
743836
57
785753
41
83134
44
6738
50
810926
42
704426
48
307734
47
32419
41
467726
48
3504082
53
83524
47
297871
47
192515
46
100189
49
1140
46
15082
52
179967
49
406176
46
3-208
Not
trawled
237633
24535
24464
56827
179458
4722
5975
8834
196576
46808
20691
144453
15723
1851
177614
282900
7231
883
557790
3891220
1783079
171
1077411
1902751
26394
11396
246085
1587606
38444
41
14096325
547455
20034
383144
322128
2305860
334031
1517
11631
825491
8309
490
3575
904387
8200
748223
905791
8918
7452302
3980
336763
43686
21468
1455060
275030
72614
23427
18595
6907
27249
911451
1499495
145365
9291
1494442
1045925
482204
55366
687744
4641125
120405
476315
309332
145326
1827
20198
258000
636851
Trawled
127834
11993
8854
19951
79908
2134
2014
3625
93301
16044
11356
79777
3809
527
89730
150863
3542
458
234935
2022993
1031810
88
325144
894321
13063
5636
112688
689019
18199
20
6990590
203419
7177
117315
144359
865944
143388
550
6045
287630
4121
203
1388
428221
3064
335241
401634
2552
2672145
1385
156994
15511
7037
600883
126384
30474
9042
7867
2114
9833
395173
402910
45128
4300
416593
429468
166989
24433
287659
1974551
55535
160892
112497
61243
645
8908
107332
237867
%
Exposed
35
33
27
26
31
31
25
29
32
26
35
36
20
22
34
35
33
34
30
34
37
34
23
32
33
33
31
30
32
33
33
27
26
23
31
27
30
27
34
26
33
29
28
32
27
31
31
22
26
26
32
26
25
29
31
30
28
30
23
27
30
21
24
32
22
29
26
31
29
30
32
25
27
30
26
31
29
27
Effort Effort
Exp Exp%
107790 29
10748 29
9784 29
22521 29
76123 29
2012 29
2334 29
3636 29
84411 29
18237 29
9306 29
65045 29
5639 29
685 29
77094 29
124879 29
3096 29
383 28
226105 28
1682218 28
801374 28
73 28
395614 28
789868 28
11088 28
4782 28
100299 28
636115 28
15836 28
17 28
5849839 28
206977 27
7439 27
136672 27
127468 27
865076 27
130238 27
563 27
4824 27
302511 27
3368 27
187 27
1336 27
358959 27
3030 27
290998 27
350262 27
3049 26
2693135 26
1427 26
130762 26
15648 26
7525 26
542964 26
105927 26
27150 26
8540 26
6958 26
2367 26
9726 26
341335 26
495916 26
49382 26
3519 26
492225 26
380336 26
167060 26
20467 25
249200 25
1684980 25
44776 25
161446 25
106853 25
52224 25
624 25
7346 25
92190 25
219457 25
GBR Seabed Biodiversity
(a) Crustacea: Penaeus semisulcatus
(b) Crustacea: Cryptolutea arafurensis
(c) Actinopterygii: Brachirus muelleri
(d) Actinopterygii: Pentaprion longimanus
Figure 3-108: Model distribution maps of selected species with higher trawl exposure indicators.
3-209
GBR Seabed Biodiversity
(a) Actinopterygii: Terapon puta
(b) Bivalvia: Enisiculus cultellus
(c) Crustacea: Metapenaeus ensis
(d) Actinopterygii: Saurida argentea/tumbil
Figure 3-109: Model distribution maps of selected species with higher trawl exposure indicators.
3-210
GBR Seabed Biodiversity
(a) Actinopterygii: Euristhmus nudiceps
(b) Actinopterygii: Leiognathus leuciscus
(c) Actinopterygii: Upeneus sundaicus
(d) Crustacea: Portunus gracilimanus
Figure 3-110: Model distribution maps of selected species with higher trawl exposure indicators.
3-211
GBR Seabed Biodiversity
(a) Actinopterygii: Calliurichthys grossi
(b) Actinopterygii Cynoglossus maculipinnis
(c) Bivalvia: Amusium pleuronectes cf
(d) Bivalvia: Melaxinaea vitrea
Figure 3-111: Model distribution maps of selected species with higher trawl exposure indicators.
3-212
GBR Seabed Biodiversity
(a) Crustacea: Thenus parindicus
(b) Actinopterygii: Leiognathus splendens
(c) Actinopterygii: Psettodes erumei,
(d) Actinopterygii: Terapon theraps
Figure 3-112: Model distribution maps of selected species with higher trawl exposure indicators.
3-213
GBR Seabed Biodiversity
(a) Actinopterygii: Upeneus sulphureus
(b) Crustacea: Erugosquilla woodmasoni
(c) Crustacea: Myra tumidospina
(d) Gastropoda: Nassarius cremmatus cf
Figure 3-113: Model distribution maps of selected species with higher trawl exposure indicators.
3-214
GBR Seabed Biodiversity
(a) Actinopterygii: Scolopsis taeniopterus
(b) Actinopterygii: Repomucenus belcheri
(c) Crustacea: Trachypenaeus anchoralis
(d) Actinopterygii: Pelates quadrilineatus
Figure 3-114: Model distribution maps of selected species with higher trawl exposure indicators.
3-215
GBR Seabed Biodiversity
(a) Actinopterygii: Brachaluteres taylori
(b) Actinopterygii: Leiognathus bindus
(c) Actinopterygii: Yongeichthys nebulosus
(d) Actinopterygii: Apogon poecilopterus
Figure 3-115: Model distribution maps of selected species with higher trawl exposure indicators.
3-216
GBR Seabed Biodiversity
3-217
Modelling of the presences of BRUVS fishes provided distribution maps for a total of 25 species
(Section 3.1.2), including 6 species that were infrequent in trawl data and not modelled, and 4 species
were not sampled at all by the research trawl. The proportion of populations in GU zones for these 10
species ranged from 38–54%, the proportions in cells with recorded effort ranged from 20–33%, and
proportions exposed to trawl effort ranged from 13–28%. These species in order of total effort
exposure (%) included: Gymnothorax minor (28), Alepes apercna (27), Scomberomorus
queenslandicus (27), Echeneis naucrates (26), Carangoides coeruleopinnatus (25), Gnathanodon
speciosus (22), Seriolina nigrofasciata (19), Carangoides gymnostethus (19), Decapterus russelli (16)
and Carangoides fulvoguttatus (13). As these species were rare or absent from the prawn trawl
samples, it follows that their catchability was low or very low (as corroborated by relative catch rates
of Fish Trawls from the Effects of Trawling Study, Poiner et al. 1998) and the proportions of their
populations caught is likely to be considerably less than the proportions exposed.
In the case of those BRUVS fish species also analysed from research trawl data sources, the exposure
assessments were similar and did not change the level of risk for any species — even though BRUVS
were able to be deployed in areas too rugged for the trawl.
3.7.2.1. Trawl effort coefficients
The species modelling process selected the Trawl Effort Index covariate for 81 of 840 species
analysed (9.6%), and was significant in 55 cases (6.5%). This frequency is little more than expected by
chance and suggests that trawling does not have a strong influence on overall seabed distribution
patterns. Nevertheless, taking each species model as an independent test, but recalling the caution
expressed in Section 2.4.7, the probability of presence was negatively affected for 43 species,
significant in 24 cases; biomass was negatively affected for 7 species, significant in each case (Table
3-54). Probability of presence was positively affected for 11 species, significant in all cases, and
biomass was positively affected for 1 species, also significant (Table 3-55). Thirteen species had
models with a second term involving the Trawl Effort covariate — in addition to a linear Trawl Effort
term for presence or biomass — such as biomass, a quadratic or an interaction (Table 3-56).
The possible magnitude of the Trawl Effort coefficient where selected, in terms of predicted percent
change in overall biomass, is also indicated in the Tables. Of the 50 negative Trawl Effort responses,
five species (3 significant) were estimated to have moderate negative change in biomass of >25%–
33%, compared with a model prediction with trawl effort set to zero over the entire region (Table
3-54). Another 15 species (9 significant) were estimated to have negative change in biomass of 15%–
25%. The remaining 30 negative responses (19 significant) were between –1% and –15%. There was
considerable uncertainty in these estimates. In the case of non-significant coefficients, the uncertainty
was greater than the estimate and so includes the possibility of no (or even positive) change due to
trawl effort. In the case of significant coefficients, the typical uncertainty was about 75% of the
estimate. Not surprisingly, all species with negative trawl coefficients have low to very low exposure
to current effort.
Species for which the negative trawl effect was larger and significant are examined in more detail
below. The first point to note is that many of these species tended to be infrequent and low in
abundance, which can present a challenge for the modelling. Further, while many of the AUC
diagnostics were reasonable, the presence models for these species typically selected very few
environmental covariates and the biomass models often did not select any environmental covariates —
this tended to lead to rather smooth distribution predictions. On the other hand, those species with
significant small negative coefficients (% change < -10%) were almost always more frequently
sampled and more abundant, with more specific biophysical models.
The species with the largest significant predicted negative change (-33%) was the Majid crab
Thacanophrys sp165 (Figure 3-116a). This species appears to have a widely scattered distribution in
non-muddy environments, as indicated by the negative mud coefficient. Given the significant negative
trawl coefficient, there was a statistical expectation that this species would have greater presence on
GBR Seabed Biodiversity
3-218
non-muddy seabeds in the absence of trawling. About three quarters of the estimated distribution of
Thacanophrys sp165 was protected by the zoning and only ~3% annually was directly exposed to
current effort.
Table 3-54: Results for the Trawl Effort covariate: species with negative coefficients for presence (P) or
biomass (B), coefficients with p>0.05 are greyed, the magnitude of the coefficient in terms of overall % change
in abundance is also indicated. The group membership, total estimated biomass (kg), % available, % exposed
and effort exposed % are as above.
Class
Genus
Species
Crustacea
Anthozoa
Demospongiae
Crustacea
Ophiuroidea
Bivalvia
Demospongiae
Crustacea
Anthozoa
Anthozoa
Crustacea
Demospongiae
Bivalvia
Gymnolaemata
Actinopterygii
Demospongiae
Actinopterygii
Actinopterygii
Anthozoa
Gymnolaemata
Asteroidea
Cephalopoda
Actinopterygii
Demospongiae
Cephalopoda
Anthozoa
Actinopterygii
Crustacea
Crustacea
Actinopterygii
Anthozoa
Demospongiae
Crustacea
Bivalvia
Actinopterygii
Crustacea
Gymnolaemata
Asteroidea
Actinopterygii
Actinopterygii
Crustacea
Chlorophyceae
Actinopterygii
Crustacea
Actinopterygii
Actinopterygii
Actinopterygii
Annelida
Anthozoa
Asteroidea
Thacanophrys
Euplexaura
Cinachyrella
Eucrate
Ophionereis
Solen
Demospongiae
Pilumnus
Echinogorgia
Iciligorgia
Barnacle
Oceanapia
Globivenus
Crassimarginatella
Lutjanus
Demospongiae
Siganus
Apogon
Junceella
Telopora
Tamaria cf
Loligo
Sillago
Cinachyrella
Photololligo
Echinogorgia
Centrogenys
Urnalana
Thalamita
Paramonacanthus
Mopsella
Spirastrella
Gonodactylaceus
Solen
Upeneus
Hyastenus
Adeonella
Astropecten
Paramonacanthus
Choerodon
Pandalidae
Udotea
Lethrinus
Lupocyclus
Lagocephalus
Apogon
Rogadius
Annelida
Dendronephthya
Astropecten
sp165
sp6
sp1
affinis
semoni cf
sp3
sp146
spinicarpus
sp5
sp1
sp1
tubes only
embrithes cf
spp
vitta
sp16
canaliculatus
cavitiensis
sp2
spp
sp3
sp1
ingenuua
australiensis
sp1
sp3
vaigiensis
whitei
intermedia
oblongus
sp2
sp2
graphurus
siphons only
luzonius
elatus
lichenoides cf
zebra
filicauda
monostigma
sp916
orientalis
genivittatus
rotundatus
sceleratus
truncatus
pristiger
spp
spp
spp
%
%
Effort
%
Model Coefficient p
Available Exposed Exp%
Change
885
25
6
3
P
-1.5918 0.021 -33
483859
36
15
9
P
-0.7738 0.045 -28
3327419
25
5
3
P
-1.6453 0.084 -27
2137
48
18
11
P
-0.2340 0.029 -27
7324
30
8
4
P
-0.7199 0.064 -26
35340
45
17
9
P
-0.3686 0.066 -25
6870
31
9
5
P
-0.5931 0.047 -25
1620
26
7
4
P
-0.8290 0.042 -24
11065
26
8
4
P
-0.6230 0.088 -22
129051
31
13
7
P
-0.4120 0.039 -20
5009088
23
12
6
P
-0.6110 0.035 -20
3664
33
13
7
P
-0.2857 0.088 -19
113931
42
15
9
P
-0.2447 0.036 -18
222
27
14
8
P
-0.5266 0.104 -18
316853
37
13
7
P
-0.2158 0.066 -18
43858
34
12
7
B
-0.2684 0.007 -18
377618
42
19
12
P
-0.1625 0.002 -17
7972
47
22
14
P
-0.1169 0.013 -16
137591
28
12
7
P
-0.3584 0.066 -16
442
28
14
8
P
-0.3682 0.049 -16
66687
43
20
12
P
-0.1315 0.047 -15
87486
30
11
6
P
-0.2538 0.047 -14
400298
48
24
16
P
-0.1590 0.038 -13
125583
20
5
3
P
-0.5908 0.123 -13
126860
32
14
8
P
-0.1871 0.066 -12
110904
18
5
3
P
-0.5701 0.196 -12
20426
23
5
3
P
-0.4504 0.245 -11
41726
47
22
15
P
-0.0797 0.033 -11
3456
22
8
4
P
-0.3043 0.196 -10
139092
41
20
13
P
-0.0844 0.001 -10
26395
24
9
5
P
-0.2236 0.089 -10
528299
35
14
9
P
-0.1009 0.044 -9
6851
49
26
18
P
-0.0555 0.018 -8
69535
49
23
16
P
-0.0462 0.052 -7
584420
36
15
10
P
-0.0705 0.058 -7
18772
33
13
8
P
-0.0858 0.076 -7
84065
40
23
16
P
-0.0554 0.008 -6
12709
52
26
20
B
-0.0274 0.000 -6
8764207
39
11
7
B
-0.0858 0.002 -6
285266
36
9
7
P
-0.0711 0.157 -6
11205
38
17
12
P
-0.0580 0.022 -5
842312
34
16
10
P
-0.0660 0.008 -5
6198005
45
24
17
P
-0.0687 0.004 -5
73747
44
21
17
P
-0.0297 0.020 -5
210666
44
22
17
P
-0.0298 0.042 -4
736274
42
20
17
B
-0.0175 0.000 -3
447881
34
14
14
P
-0.0225 0.078 -3
5213418
42
21
19
B
-0.0183 0.008 -2
1302022
29
13
10
B
-0.0196 0.003 -2
76778
45
26
29
B
-0.0414 0.009 -1
Grp Biomass
17
3
37
13
15
13
15
15
23
17
16
3
12
31
20
15
13
13
16
31
25
37
21
37
35
23
3
12
23
2
23
15
25
13
20
3
17
20
24
12
15
2
10
20
36
36
35
20
18
20
GBR Seabed Biodiversity
3-219
The species with the second largest significant predicted negative change (-28%) was the gorgonian
Euplexaura sp6 (Figure 3-116b). Like all gorgonians, this species tended to be associated with harder
seabed; also indicated by the positive gravel coefficient. Given the significant negative trawl
coefficient, there is a statistical expectation that this species would have greater presence on gravel
seabeds in the absence of trawling, probably on patches of biogenic hard substratum. Almost two
thirds of the estimated distribution of Euplexaura sp6 was protected by the zoning and only ~9%
annually was directly exposed to current effort.
The species with the third largest significant predicted negative change (-27%) was the decapod
crustacean Eucrate affinis (Figure 3-116c). This species appeared to have a widely scattered inner
shelf but not inshore distribution. Again, given the significant negative trawl coefficient, there was a
statistical expectation that this species would have greater presence on inner shelf seabeds in the
absence of trawling. About 52% of the estimated distribution of Eucrate affinis was protected by the
zoning and only 11% annually was directly exposed to current effort.
The species with the next largest significant predicted negative change (-25%) was Demospongiae
sp146 (Figure 3-116d). This species appeared to have a very low predicted biomass (lowest colour
from key everywhere) and, with no spatial variables other than effort, a very broad predicted
distribution. Again, given the significant negative trawl coefficient, there was a statistical expectation
that this species would have greater presence on the seabed in the absence of trawling. About 69% of
the estimated distribution of Demospongiae sp146 was protected by the zoning and only ~5% annually
was directly exposed to current effort.
The species with the next largest predicted negative change (-24%) was the Xanthid crab Pilumnus
spinicarpus (Figure 3-117a). This species appeared to have a very low predicted biomass (lowest
colour from key ~everywhere). Again, given the significant negative trawl coefficient, there was a
statistical expectation that this species would have greater presence on the seabed in the absence of
trawling. About 74% of the estimated distribution of Pilumnus spinicarpus was protected by the
zoning and only ~4% annually was directly exposed to current effort.
The species with the next largest predicted negative change (-20%) was the gorgonian Iciligorgia sp1
(Figure 3-117b). Like other gorgonians, this species was expected to be associated with harder seabed;
and was predicted to be widely scattered in non-muddy seabeds. Given the significant negative trawl
coefficient, there is a statistical expectation that this species would have greater presence on such
seabeds in the absence of trawling, probably on patches of biogenic hard substratum. About 69% of
the estimated distribution of Iciligorgia sp1 was protected by the zoning and only ~7% annually was
directly exposed to current effort.
The species with the next largest predicted negative change (-21%) was an unidentified Barnacle sp1
(Figure 3-117c). This species appeared to have a more specific distribution in non-muddy carbonate
gravel in the southern GBR and with the significant negative trawl coefficient, a statistical expectation
of greater presence on these types of seabeds in the absence of trawling. About 77% of the estimated
distribution of Barnacle sp1 was protected by the zoning and only ~6% annually was directly exposed
to current effort.
The species with the next largest predicted negative change (-18%) was the bivalve Globivenus
embrithes cf (Figure 3-117d). This species appeared to be associated with inshore low effort areas
seabeds primarily in the southern GBR. Again, given the significant negative trawl coefficient, there is
a statistical expectation that this species would have greater presence on such seabeds in the absence
of trawling. About 58% of the estimated distribution of Globivenus embrithes cf was protected by the
zoning and only ~9% annually was directly exposed to current effort.
GBR Seabed Biodiversity
(a) Crustacea: Thacanophrys sp165
(b) Anthozoa: Euplexaura sp6.
(c) Crustacea: Eucrate affinis
(d) Demospongiae: Demospongiae sp146
3-220
Figure 3-116: Model distribution maps of selected species with significant larger negative trawl coefficients.
GBR Seabed Biodiversity
(a) Crustacea: Pilumnus spinicarpus
(b) Anthozoa: Iciligorgia sp1
(c) Crustacea: Barnacle sp1
(d) Bivalvia: Globivenus embrithes cf
3-221
Figure 3-117: Model distribution maps of selected species with significant larger negative trawl coefficients.
The 12 positive Trawl Effort responses were relatively smaller, four species (all significant) were
estimated to have positive change in biomass of >12%–19%, compared with a model prediction with
trawl effort set to zero over the entire region (Table 3-55). The remaining 8 positive responses (all
GBR Seabed Biodiversity
3-222
significant) were between >0% and +10%. The typical uncertainty in these estimates was about 70%
of the estimate. Not surprisingly, many species with positive trawl coefficients have high to very high
exposure to current effort and most were highlighted in the previous section. Species for which the
positive trawl effect was larger are examined in more detail below. The first point to note is that most
of these species were sampled relatively frequently (though not necessarily abundant) and had
relatively strong presence and biomass models.
The species with the largest predicted positive change (+19%) was the Pilumnid crab Cryptolutea
arafurensis (Figure 3-108b above). This species was moderately frequent in sled samples and
appeared to be common on muddy (but not extremely) seabeds. Given the significant positive trawl
coefficient, there was a statistical expectation that this species would have lower presence on such
seabeds in the absence of trawling. About 43% of the estimated distribution of Cryptolutea arafurensis
was protected by the zoning but a high proportion of its biomass in GU zones was in high effort areas
leading to an annual direct exposure to current effort of 128%.
The species with the second largest predicted positive change (+18%) was the Terapontid fish Pelates
quadrilineatus (Figure 3-114d). This species was the least frequent of this group and appeared to be
widely scattered on inshore substratums of terrestrial origin (low carbonate). Given the significant
positive trawl coefficient, there was a statistical expectation that this species would have lower
presence on such seabeds in the absence of trawling. About 31% of the estimated distribution of
Pelates quadrilineatus was protected by the zoning but a high proportion of its biomass in GU zones
was in high effort areas leading to an annual direct exposure to current effort of 103%.
The species with the third largest predicted positive change (+13%) was the Gerreid fish Pentaprion
longimanus (Figure 3-108d). This species was moderately frequent in trawl samples and appeared to
be common on intermediate carbonate muddy seabeds, mostly inner-shelf. Given the significant
positive trawl coefficient, there was a statistical expectation that this species would have lower
presence on such seabeds in the absence of trawling. About 38% of the estimated distribution of
Pentaprion longimanus was protected by the zoning but a high proportion of its biomass in GU zones
was in high effort areas leading to an annual direct exposure to current effort of 117%.
The species with the next largest predicted positive change (+12%) was the bivalve scallop Amusium
balloti (Figure 3-118a). This species was very frequent in sled and trawl samples and was often
abundant on sandy seabed in the southern GBR. Given the significant positive trawl coefficient, there
was a statistical expectation that this species would have lower presence in these areas in the absence
of trawling; however, given also that this scallop is also a target species, a plausible alternative
explanation is that the searching ability of fisherman — as represented by the effort data — are a
better indicator of the sampled distribution of scallops at a scale finer than that of spatial patterns in
the physical environmental data, rather than indicating that scallops are more abundant because of
trawling. About 45% of the estimated distribution of Amusium balloti was protected by the zoning and
a somewhat high proportion of the biomass in GU zones lies in higher effort areas leading to an annual
direct exposure to current effort of 45%.
The species with the next largest predicted positive change (+10%) was the Leiognathid ponyfish
Leiognathus leuciscus (Figure 3-110b). This species was moderately frequent in trawl samples and
appeared to be common on inshore substratums of terrestrial origin (low carbonate) on the northern
two-thirds of the GBR. Given the significant positive trawl coefficient, there was a statistical
expectation that this species would have lower presence on such seabeds in the absence of trawling.
About 41% of the estimated distribution of Leiognathus leuciscus was protected by the zoning but a
high proportion of its biomass in GU zones was in high effort areas leading to an annual direct
exposure to current effort of 95%.
The species with the next largest predicted positive change (+9%) was another Leiognathid ponyfish
Leiognathus bindus (Figure 3-115b). This species was moderately infrequent in trawl samples and
appeared to be common on muddy substratums mostly inshore but extending offshore in the Capricorn
Channel. Given the significant positive trawl coefficient, there was a statistical expectation that this
species would have lower presence on such seabeds in the absence of trawling. About 58% of the
estimated distribution of Leiognathus bindus was protected by the zoning but a somewhat higher
proportion of its biomass in GU zones was in high effort areas leading to an annual direct exposure to
current effort of 63%.
GBR Seabed Biodiversity
3-223
The Gastropod, Xenophora indica had a predicted positive change of +8% and was moderately
frequent in sled and trawl samples and appeared to be common on clearer sandy seabeds, distributed
offshore in the southern half of the GBR (Figure 3-118b). Again, given the significant positive trawl
coefficient, there was a statistical expectation that this species would have lower presence on such
seabeds in the absence of trawling. About 56% of the estimated distribution of Xenophora indica was
protected by the zoning and its biomass in trawled GU zones had an annual direct exposure to current
effort of 30%.
Another notable species in this group is the commercial western king prawn, Penaeus latisulcatus,
which was moderately frequent in trawl samples at sites in clear shallow sandy areas particularly in the
southernmost GBR (Figure 3-118c). With a significant positive trawl coefficient corresponding to
predicted positive change of 6%, there was a statistical expectation that this species would have lower
presence in these areas in the absence of trawling. However, like scallops, this prawn is also a target
species and the same alternative explanation cannot be excluded. About 41% of the estimated
distribution of Penaeus latisulcatus was protected by the zoning and a somewhat high proportion of
the biomass in GU zones lies in higher effort areas leading to an annual direct exposure to current
effort of 49%.
The Leucosiid crab, Myra tumidospina had a predicted positive change of +6% and was relatively
frequent in sled and trawl samples and appeared to be common in muddy high-chlorophyll areas,
primarily in inshore region along much of the GBR but also in some muddy offshore areas (Figure
3-113c). Again, given the significant positive trawl coefficient, there was a statistical expectation that
this species would have lower presence on such seabeds in the absence of trawling. About 43% of the
estimated distribution of Myra tumidospina was protected by the zoning but a somewhat higher
proportion of its biomass in GU zones was in high effort areas leading to an annual direct exposure to
current effort of 60%.
The Xanthid crab, Liagore rubromaculata had a predicted positive change of +5% and was
moderately frequent in trawl and sled samples and appeared to be common in muddy high-chlorophyll
areas, primarily on deeper muddy substratums mostly inshore but extending offshore in the Capricorn
Channel (Figure 3-118d). Again, given the significant positive trawl coefficient, there was a statistical
expectation that this species would have lower presence on such seabeds in the absence of trawling.
About 52% of the estimated distribution of Liagore rubromaculata was protected by the zoning but a
somewhat higher proportion of its biomass in GU zones was in high effort areas leading to an annual
direct exposure to current effort of 43%.
Table 3-55: Results for the Trawl Effort covariate: species with positive coefficients for presence (P) or biomass
(B), coefficients with p>0.05 are greyed, the magnitude of the coefficient in terms of overall % change in
abundance is also indicated. The group membership, total estimated biomass (kg), % available, %exposed and
effort exposed are as above.
Class
Genus
Species
Crustacea
Actinopterygii
Actinopterygii
Bivalvia
Actinopterygii
Actinopterygii
Gastropoda
Crustacea
Crustacea
Crustacea
Actinopterygii
Rhodophyceae
Cryptolutea
Pelates
Pentaprion
Amusium
Leiognathus
Leiognathus
Xenophora
Penaeus
Myra
Liagore
Elates
Haloplegma
arafurensis
quadrilineatus
longimanus
balloti
leuciscus
bindus
indica
latisulcatus
tumidospina
rubromaculata
ransonnetii
duperreyi
Group Biomass
29
21
29
10
9
24
8
21
22
29
24
1
480
129842
61963
2355308
171753
76017
25049
235627
14791
49419
431203
3474678
%
%
Effort
%
Model Coefficient p
Available Exposed Exp%
Change
57
41
128
P
0.0213 0.001 19
69
47
103
P
0.0299 0.001 18
62
48
117
P
0.0189 0.048 13
55
37
45
B
0.0424 0.003 12
59
43
95
P
0.0235 0.026 10
42
28
63
P
0.0279 0.003
9
44
27
30
P
0.0451 0.000
7
59
39
49
P
0.0265 0.004
6
57
38
60
P
0.0169 0.006
6
48
25
43
P
0.0275 0.001
5
30
12
17
P
0.0213 0.038
1
44
24
20
P
0.0392 0.004
0
In the case of the 15 species with more complex responses to the Trawl Effort covariate involving an
additional term, the magnitude and direction varied widely and could not be inferred simply from the
GBR Seabed Biodiversity
3-224
coefficients (Table 3-56). Seven species were estimated to have significant negative change in biomass
of -5% to -36% and five species were estimated to have significant positive change in biomass of +4%
to +96%, compared with a model prediction with trawl effort set to zero over the entire region. The
typical uncertainty in these estimates was about 70% of the estimate. As before, species predicted to
had a positive change with trawling had moderately high to high exposure to current effort, while
species with a negative change had low to very low exposure. Six species with the largest effects are
examined in more detail below.
The species with the largest predicted positive change (+96%) was the Monacanthid fish
Brachaluteres taylori (Figure 3-115a). This species was not uncommon in trawl samples at sites in
shallow clear non-muddy areas, particularly in the southern half of the GBR, with intermediate
intensity of trawl effort. With a positive trawl effort term and a negative squared term for the biomass
model, giving an overall large positive predicted trawl effect, there was a statistical expectation that
this species would have lower presence on those types of seabeds in the absence of trawling. About
29% of the estimated distribution of Brachaluteres taylori is protected by the zoning and 72%
annually is directly exposed to current effort.
The species with the second largest predicted positive change (+50%) was the commercial grooved
tiger prawn, Penaeus semisulcatus (Figure 3-108a). This species was common in trawl samples at sites
in shallow muddy (but not extreme) low-light inshore areas, mostly in the northern two-thirds of the
GBR, with intermediate intensity of trawl effort. With a positive trawl effort term and a negative
squared term for the biomass model, giving an overall large positive predicted trawl effect, there was a
statistical expectation that this species would have lower presence on those types of seabeds in the
absence of trawling. About 26% of the estimated distribution of Penaeus semisulcatus was protected
by the zoning and a high proportion of its biomass in GU zones was in high effort areas leading to an
annual direct exposure to current effort of 174%. Like Penaeus latisulcatus and Amusium balloti
discussed above, Penaeus semisulcatus is also a target species and the same alternative explanation
cannot be excluded. The negative Biomass:Trawl_Eff^2 term was indicative of reduced standing
biomass at very high levels of effort.
The species with the largest predicted negative change (-36%) was a gorgonian soft coral Carijoa sp1
(Figure 3-119a). This species appeared to have a widely scattered patchy distribution and with a
negative trawl effort term for both the presence and the biomass model giving a strong negative
predicted trawl effect, there is a statistical expectation that this species would have higher abundance
in the absence of trawling. Three-quarters of the estimated distribution of Carijoa sp1 was protected
by the zoning and only ~3% annually is directly exposed to current effort.
The species with the next largest predicted change (-27%) was Inegocia harrisii (a Platycephalid fish)
(Figure 3-119b). This species was common at sites in low-light inshore areas, particularly in the
vicinity of the very high tidal range areas of Shoalwater Bay, Broad Sound and the Whitsunday
Islands. With a negative trawl effort term for both the presence and the biomass model, giving an
overall negative predicted trawl effect, there was a statistical expectation that this species would have
higher abundance on those types of seabeds in the absence of trawling. About 78% of the estimated
distribution of Inegocia harrisii was protected by the zoning and only ~3% annually was directly
exposed to current effort.
The species with the next largest predicted change (-26%) was the sediment infaunal sponge
Oceanapia sp21 (Figure 3-119c). This species was not uncommon in sled samples at widely scattered
sites in clear intermediate-gravel areas (mostly mid-to-outer shelf). With a negative trawl effort term
and a positive effort*gravel interaction term in the presence model, giving an overall negative
predicted trawl effect, there was a statistical expectation that this species would have higher abundance
on those types of seabeds in the absence of trawling. It is possible that the interaction indicates that the
negative trawl effect is less as gravel increases. About 75% of the estimated distribution of Oceanapia
sp21 was protected by the zoning and only 3% annually was directly exposed to current effort.
The species with the next largest predicted change (-20%) was the Majid crab Austrolabidia gracilipes
(Figure 3-119d). This species was not uncommon in sled samples at particularly off Mackay and the
Whitsunday Islands, with scattered records elsewhere. With a positive trawl effort term and a negative
effort*temp_SD interaction term in the presence model, giving an overall negative predicted trawl
effect, there was a statistical expectation that this species would have higher abundance in the region
GBR Seabed Biodiversity
3-225
in the absence of trawling. It was possible that the interaction indicates that the negative trawl effect is
greater where temperature is more variable. About two-thirds of the estimated distribution of
Austrolabidia gracilipes was protected by the zoning and only 7% annually was directly exposed to
current effort.
(a) Bivalvia: Amusium balloti
(b) Gastropoda: Xenophora indica
(c) Crustacea: Penaeus latisulcatus
(d) Crustacea: Liagore rubromaculata
Figure 3-118: Model distribution maps of selected species with significant larger positive trawl coefficients.
GBR Seabed Biodiversity
3-226
Table 3-56: Results for the Trawl Effort covariate: species with an additional term involving the Trawl Effort covariate, as well as coefficients for presence (P) or biomass (B),
coefficients with p>0.05 are grayed, the magnitude of the coefficient in terms of overall % change in abundance is also indicated. The group membership, total estimated biomass (kg),
% available, % exposed and effort exposed are as above.
Class
Actinopterygii
Crustacea
Crustacea
Actinopterygii
Bivalvia
Crustacea
Actinopterygii
Demospongiae
Crustacea
Echinoidea
Crustacea
Crustacea
Demospongiae
Actinopterygii
Anthozoa
Genus
Brachaluteres
Penaeus
Portunus
Saurida
Melaxinaea
Trachypenaeus
Pentapodus
Demospongiae
Pagurid
Mespilia
Austrolabidia
Cloridina
Oceanapia
Inegocia
Carijoa
Species
taylori
semisulcatus
gracilimanus
grandi/undo
vitrea
anchoralis
paradiseus
sp109
sp2358-1
globulus
gracilipes
chlorida
sp21
harrisii
sp1
Group
8
29
9
21
9
13
18
37
13
30
3
14
37
3
3
Biomass
%
%
Effort
Trawl Effort
Model
kg
Available Exposed Exp %
Coefficient
62129
71
60
72
B
0.2436
301314
74
64
174
B
0.0513
204641
59
38
86
P
0.0154
8331858
59
37
46
P
0.0348
171979
59
38
63
P
-0.0002
45119
64
44
67
P
0.0773
2615371
34
16
11
P
0.1900
119911
25
6
4
P
0.3294
15945
44
20
18
P
0.0693
22836
12
3
2
P
0.1490
12992
33
12
7
P
0.7857
374
29
3
2
P
0.0578
5236461
25
5
3
P
-2.3602
217420
22
5
3
P
-0.5655
78340
25
5
3
P
-0.6098
p
Second term
Coefficient
p
0.000
0.000
0.009
0.031
0.980
0.001
0.007
0.001
0.024
0.464
0.006
0.569
0.010
0.010
0.004
I(TRWL_EFF_I^2)
I(TRWL_EFF_I^2)
biomass: TRWL_EFF_I
biomass: TRWL_EFF_I
TRWL_EFF_I:GA_GRAVEL
SW_K_B_IRR:TRWL_EFF_I
TRWL_EFF_I:SW_CHLA_AV
TRWL_EFF_I:GA_CRBNT
Across:TRWL_EFF_I
GA_GRAVEL:TRWL_EFF_I
CRS_T_SD:TRWL_EFF_I
M_BSTRESS:TRWL_EFF_I
TRWL_EFF_I:GA_GRAVEL
biomass: TRWL_EFF_I
biomass: TRWL_EFF_I
-0.0047
-0.0004
0.0148
0.0116
0.0022
-0.4881
-0.5093
-0.0106
-1.1173
-0.1363
-0.7849
-20.8768
0.0539
-0.4681
-1.4558
0.001
0.004
0.001
0.009
0.008
0.008
0.001
0.001
0.002
0.008
0.006
0.124
0.011
0.001
0.006
%
Change
96
50
11
7
6
4
-5
-10
-16
-17
-20
-22
-26
-27
-36
GBR Seabed Biodiversity
(a) Anthozoa: Carijoa sp1
(b) Actinopterygii: Inegocia harrisii
(c) Demospongiae: Oceanapia sp21
(d) Crustacea: Austrolabidia gracilipes
Figure 3-119: Model distribution maps of selected species with multiple trawl coefficients.
3-227
GBR Seabed Biodiversity
3-228
The trawl effort covariate was included in all 25 BRUVS fish species presence models, although
significance tests were not possible, and the potential population change due to the influence of
trawling was estimated for all species. Most (16) of the estimated changes were small (≤ ±5%) and
most (18) were positive. The largest negative estimates were for Carangoides fulvoguttatus (-9%) and
Abalistes stellatus (-8). The largest estimated changes were all positive: Nemipterus peronii (17%),
Gymnothorax minor (17%), Parapercis nebulosa_grp (16%), Paramonacanthus otisensis (11%),
Echeneis naucrates (11%), Lagocephalus sceleratus (10%). Of the above named species, only
Carangoides fulvoguttatus, Echeneis naucrates and Gymnothorax minor were not assessed from the
research trawl data; however, the uncertainty in these estimates from BRUVS is unknown.
3.7.2.2. Species exposure rank, catchability, and recovery indicators
The species exposure estimates from the previous sections were tabulated and ranked in order of most
exposed to trawl effort intensity (Table 3-57). This exposure ranking is the primary output of the
Seabed Project with respect to risk indicators for the trawl fishery in the GBR region. All of those
species with effort exposure >50% have been discussed above, as well as several with exposure <50%.
In this section, these exposure rankings were developed further with additional information from the
Project and/or from external sources. By multiplying the effort exposure by the relative catch rate and
BRD effect (if appropriate), an estimate of the percentage of population caught annually is tabulated
(Table 3-57). Note that this estimate is not a true ‘catchability’ and there has not been a formal
analysis of catchability as part of this Project. Where data from both sled and trawl was included in the
modelling for a given species the coefficient of the device term was taken (indicated by “Model” in the
table) and the range of ±SE was taken as an indicator of the uncertainty. Where model coefficients
were not available at the species level, the mean of genus level coefficients was taken (indicated by
“MdlGen” in the table) and similarly for the uncertainty. If model coefficients were not available, a
simple ratio of average catch rates between devices, at the species or genus level (“Mean” or “MnGen”
respectively), was used. If there was evidence that a fish trawl net had a higher catch rate in the earlier
“Green Zone Effects of Trawling Study” then a simple ratio of average catch rates from that source,
again at the species or genus level (“GZFsh” or “GZMn” respectively), was used. Often, there was
considerable variability in relative catch rate among sources, which was taken as an indicator of
uncertainty where the model result was not available. The estimated relative catch rate usually was
less than 1 and the estimated percentage of populations caught annually was usually less than the
estimated percentage exposed to trawl effort. The estimated effect of relative catch rate varied widely
among species and substantially altered the ranking of species potentially at risk in terms of estimated
percentage caught. At this point, the highest ranked species was the Pleuronectiform flatfish Brachirus
muelleri (~110% caught), followed by other small fishes Terapon puta (59%), Saurida
argentea/tumbil (58%), Psettodes erumei (52%) and the commercial prawn Penaeus semisulcatus
(55%). Of the 33 species with effort exposure >50%, only 5 had >50% caught and 19 had <25%
caught (Table 3-57). Of the 218 species with effort exposure between 25% and 50%, only 19 had
>25% caught and 199 had <25% caught (Table 3-57). While these estimates of potential relative
incidental (or in some cases target) catch make a critical contribution to understanding potential
environmental risk, they do not provide a definitive indication of sustainability risk. For this, some
indication of population recovery (the propensity for the population to replenish) is required.
Information on potential population recovery that was available were the “recovery” rankings for
fishes from the NPF Bycatch Sustainability Project approach (Stobutzki et al. 2001) and for
invertebrates from the NPF Ecological Surrogates Project (Hill et al. 2002). Where recovery ranks
were available, weighted mean added ranks were tabulated (Table 3-57) for matching fish species
genera. In the case of invertebrates, recovery ranks were available at the family level only. Low ranks
(1.5–1.875, shaded red) indicate lower relative potential recovery and high ranks (2.65–3.0, not
shaded) indicate higher relative potential recovery (low-moderate 1.875–2.25 and moderate 2.25–2.65
ranks are shaded orange and pale respectively). The available mean recovery rank for 422 species was
plotted against the estimated percentage of population caught (Figure 3-120). The species at greatest
relative risk should plot towards the upper left quadrant of the graph. The top ranking species were:
Brachirus muelleri, Sepia pharaonis, Terapon puta, Saurida argentea/tumbil, Penaeus semisulcatus,
Euristhmus nudiceps, Apogon poecilopterus, Sepia elliptica, Scolopsis taeniopterus, Psettodes erumei,
GBR Seabed Biodiversity
3-229
Amusium pleuronectes cf, Tripodichthys angustifrons, Saurida grandi/undo, Yongeichthys nebulosus,
Sepia whitleyana, Upeneus sundaicus, Leiognathus leuciscus, Sepia smithi, Portunus gracilimanus,
Chaetodermis penicilligera and Sepia plangon (see species distribution maps Figure 3-108 to Figure
3-115 above and Figure 3-121, Figure 3-122 below). While these species have higher relative risk on
the basis of their recovery attributes, such that management attention as to their future status is
warranted, it is nevertheless unclear whether these species are currently at sustainability risk or not.
120
Estimated percent caught
Bra.mue
100
80
60
Sep.pha
40 Sep.ell
Sep.whi
Sep.smi
Eur.nud
Apo.poe
Amu.ple
Cha.pen
Ter.put
Sau.arg
Pen.sem
Pse.eru
Sco.tae
Tri.ang
Sau.gra
Yon.neb
Upe.sun
Lei.leu
Por.gra
20
0
1.5
2.0
2.5
3.0
PSA mean
rank
SRA
meanproductivity
recovery rank
Figure 3-120: Plot of estimated percentage of population caught against mean RSA recovery rank. Species at
greatest potential risk should plot towards the upper left quadrant. The top ranking species are labeled with the
first three letters of their genus and species name (see Table 3-57).
In this report, a sustainability indicator — analogous to that of Zhou & Griffiths (2007) was estimated,
where natural mortality rates have been collated in Brewer et al. (2007) or were available from other
sources — as the proportion of the total population caught/natural mortality (Table 3-57). Where this
indicator exceeded the reference points 0.6 and 0.8, and the limit reference point 1.0, the indicator was
highlighted (pale, orange, and red, respectively). Three species exceeded the limit reference point:
Fistularia petimba, the Rough Flutemouth (at 1.12); Brachirus muelleri, the Tufted Sole (at 1.11) and
Trixiphichthys weberi, the Blacktip Tripodfish (at 1.09). Fistularia petimba was moderately frequent
in trawl samples and was distributed along the length of the GBR, though more to the north, in low
current stress, low light areas (Figure 3-123a). Most individuals caught were small (average: 15.5 g,
range: 2–58 g) compared to adults, which would usually be considered a reef associated species.
Brachirus muelleri was moderately infrequent in trawl samples and was distributed on intermediate
carbonate mud sediments innershelf from Shelbourne Bay to the Whitsundays, and near the mouth of
the Capricorn Channel (Figure 3-108c). Individuals caught ranged in size from 4–189 g (average: 61
g). Brachirus muelleri was also listed among the highest risk SRA species. Trixiphichthys weberi was
infrequent in trawl samples and was distributed along most of the length of the GBR in inner-shelf but
not inshore areas (Figure 3-123b). Most individuals caught were small (average: 25 g, range: 6–79 g).
One species exceeded the first conservative reference point: Pomadasys maculatus, the Blotched
Javelin (a grunter bream) (at 0.96). Pomadasys maculatus was moderately frequent in trawl samples
and was distributed along most of the length of the GBR in inshore areas with low-light levels on the
seabed (Figure 3-123c). Most individuals caught were relatively small (average: 35 g, range: 4–104 g).
GBR Seabed Biodiversity
3-230
Two species exceeded the second conservative reference point: Psettodes erumei, the Australian
Halibut or Spiny Turbot (at 0.75) and Sillago burrus, the Western Trumpeter Whiting (at 0.60).
Psettodes erumei occurred in 51 trawl samples and was distributed along most of the length of the
GBR, from about Mackay northwards, in inshore muddy areas with low current stress (Figure 3-112c).
Individuals caught averaged 220 g, range: 21–1960 g. Psettodes erumei was also listed among the
highest risk RSA species. Sillago burrus was relatively infrequent in trawl samples and was
distributed along most of the length of the GBR in shallow areas with low turbidity, primarily inshore
(Figure 3-123d). Individuals caught ranged in weight from 14–129 g (average: 50 g).
The next 10 ranked species were below the natural mortality based sustainability reference points, but
included seven species ranked highly by the SRA method (indicated by *): Dasyatis leylandi (0.59),
Nemipterus furcosus (0.56), Tripodichthys angustifrons* (0.53), Terapon puta* (0.53), Euristhmus
nudiceps* (0.52), Saurida argentea/tumbil* (0.52), Nemipterus peronii (0.41), Sepia pharaonis*
(0.39), Saurida grandi/undosquamis* (0.38) and Amusium pleuronectes cf* (0.35). The highest ranked
target species were Thenus parindicus (at 0.31), Penaeus semisulcatus (at 0.24) and Amusium balloti
(at 0.16).
The final column (*) in Table 3-57 indicates the importance of the catchability parameter in altering
the sustainability indicator outcome. It assumed that catchability was 1 and showed “!!!” if the 1×M
limit reference point would be exceeded and “!!” and “!” if the 0.8×M and 0.6×M reference points
would be exceeded; if M was unknown, then the importance of catchability was also unknown (“u”) in
this respect. The uncertainty in catchability was particularly critical for Pomadasys maculatus, which
would exceed the limit reference point at the higher end of the catchability uncertainty range.
Psettodes erumei would be above the first conservative reference point; and Nemipterus peronii,
Terapon puta and Nemipterus furcosus would be above the second conservative reference point at the
higher end of the catchability uncertainty range. On the other hand, at the lower end of the catchability
uncertainty range, Trixiphichthys weberi would drop from exceeding the limit reference point to the
lowest reference point. Further investigation of catchability is required to have greater confidence in
the sustainability risk indicators for these species, as well as others flagged in Table 3-57. The
estimates of natural mortality from other sources are also accompanied by unspecified uncertainty.
Of the species for which M was unknown, only three had an estimated % caught >25% and these were
considered further because of the uncertainty regarding their sustainability in the absence of M. The
highest ranking was the anaspid gastropod Aplysia sp1_QMS with an estimated catch of 38%. While
uncertain, the relative catch rate of 1 for this species may be erroneous due identification issues
between sled and trawl samples as, at the family level, the relative catch rate was 0.35 and if realistic,
this suggests an actual % caught of ~13%. No information on mortality was found. Next was the
sorbeoconch gastropod Lamellaria sp1 with an estimated catch of 37%. Again, while uncertain, the
relative catch rate of 1 for this species may be erroneous due identification issues between sled and
trawl samples as, at the order level, the relative catch rate was 0.08 and if realistic, this suggests an
actual % caught of ~3%. No information on mortality was found. Finally, the polychaete bristle worm
Chloeia flava had an estimated catch of 21% and again, the relative catch rate of 1 for this species may
be erroneous as the sled worm samples were not sorted. Further, this group is known to respond
positively to trawl disturbance by feeding on carrion (Engel & Kvitek 1998), has regeneration capacity
and multiple reproductive modes, and is very likely to have a short lifespan and corresponding high
natural mortality rate, so is unlikely to be at risk.
Indicators for all other species with modelled distributions were tabulated in APPENDIX 4: SINGLE
SPECIES TRAWL EXPOSURE — ranked by species level exposure, if known.
GBR Seabed Biodiversity
(a) Cephalopoda: Sepia pharanonis
(b) Cephalopoda: Sepia elliptica
(c) Actinopterygii: Tripodichythys angustifrons
(d) Actinopterygii: Saurida grandi/undosquamis
3-231
Figure 3-121: Model distribution maps of selected species with higher relative risk identified from exposure and
SRA recovery attributes.
GBR Seabed Biodiversity
(a) Cephalopoda: Sepia whitleyana
(b) Cephalopoda: Sepia smithi
(c) Actinopterygii: Chaetodermis penicilligera
(d) Cephalopoda: Sepia plangon
3-232
Figure 3-122: Model distribution maps of selected species with higher relative risk identified from exposure and
SRA recovery attributes.
GBR Seabed Biodiversity
3-233
Table 3-57: Summary of species exposure estimates for the top 280 of 840 species ranked by percent biomass
exposed to trawl effort intensity, showing estimated relative catchability from various sources and indicative
uncertainty, possible BRD effect for bycatch fish, leading to an estimate of potential percentage of population
caught annually. Recovery attributes from SRA and natural mortality estimates (M) are tabulated where
available. Where M was available, a sustainability indicator — estimate proportion Caught/M — is also
tabulated. Column * relates to indicator uncertainty due to catchability (see explanation in text, page 3-230).
Class
Crustacea
Crustacea
Actinopterygii
Actinopterygii
Actinopterygii
Actinopterygii
Actinopterygii
Crustacea
Actinopterygii
Bivalvia
Actinopterygii
Crustacea
Crustacea
Crustacea
Actinopterygii
Bivalvia
Actinopterygii
Actinopterygii
Crustacea
Actinopterygii
Actinopterygii
Crustacea
Gastropoda
Actinopterygii
Bivalvia
Actinopterygii
Actinopterygii
Actinopterygii
Actinopterygii
Bivalvia
Actinopterygii
Actinopterygii
Actinopterygii
Actinopterygii
Anthozoa
Actinopterygii
Actinopterygii
Crustacea
Cephalopoda
Actinopterygii
Bivalvia
Crustacea
Actinopterygii
Actinopterygii
Crustacea
Actinopterygii
Crustacea
Crustacea
Bivalvia
Actinopterygii
Actinopterygii
Actinopterygii
Holothuroidea
Gastropoda
Actinopterygii
Actinopterygii
Crustacea
Crustacea
Crustacea
Crustacea
Crustacea
Actinopterygii
Crustacea
Actinopterygii
Crustacea
Genus
Penaeus
Cryptolutea
Brachirus
Pentaprion
Pelates
Leiognathus
Upeneus
Portunus
Terapon
Enisiculus
Brachaluteres
Trachypenaeus
Metapenaeus
Erugosquilla
Leiognathus
Melaxinaea
Saurida
Terapon
Myra
Calliurichthys
Upeneus
Thenus
Nassarius
Psettodes
Placamen
Scolopsis
Leiognathus
Repomucenus
Cynoglossus
Amusium
Yongeichthys
Apogon
Euristhmus
Tripodichthys
Sea pen
Gerres
Selaroides
Penaeus
Sepia
Nemipterus
Modiolus
Penaeus
Cynoglossus
Caranx
Metapenaeus
Saurida
Charybdis
Portunus
Amusium
Suggrundus
Leiognathus
Parapercis
Bohadschia
Lophiotoma
Pseudorhombus
Inegocia
Ixa
Leucosia
Liagore
Portunus
Calappa
Ambiserrula
Oratosquillina
Leiognathus
Portunus
Species
semisulcatus
arafurensis
muelleri
longimanus
quadrilineatus
leuciscus
sundaicus
gracilimanus
puta
cultellus
taylori
anchoralis
ensis
woodmasoni
bindus
vitrea
argentea/tumbil
theraps
tumidospina
grossi
sulphureus
parindicus
cremmatus cf
erumei
tiara
taeniopterus
splendens
belcheri
maculipinnis
pleuronectes cf
nebulosus
poecilopterus
nudiceps
angustifrons
sp1
filamentosus
leptolepis
latisulcatus
pharaonis
peronii
elongatus
esculentus
sp 1 punctate
bucculentus
endeavouri
grandi/undo
truncata
tuberculosus
balloti
macracanthus
cf bindus
diplospilus
marmorata cf
acuta
arsius
japonica
inermis
ocellata
rubromaculata
hastatoides
sp44
jugosa
gravieri
moretoniensis
pelagicus
Biomass
%
%
%Effort
Rel
Source
kg
Available Exposed Exposed Catch
301314
74
64
174
0.32 Model
480
57
41
128
0.03 Model
80330
69
59
119
1.00 Mean
61963
62
48
117
0.11 GZFsh
129842
69
47
103
0.15 GZFsh
171753
59
43
95
0.43 MdlGen
370945
63
50
93
0.45 MdlGen
204641
59
38
86
0.39 Model
60300
56
47
78
0.82 GZFsh
984
61
46
75
0.07 Model
62129
71
60
72
0.13 Model
45119
64
44
67
0.26 Model
31126
67
49
67
0.19 Model
19542
66
49
65
0.18 GZFsh
76017
42
28
63
0.01 GZFsh
171979
59
38
63
0.07 Model
1109937
58
38
63
1.00 Mean
359964
63
43
62
0.11 GZFsh
14791
57
38
60
0.13 Model
171819
54
39
59
0.43 MdlGen
723274
70
46
58
0.45 GZFsh
518607
55
36
57
0.49 Model
35852
55
39
57
0.03 Model
361247
61
40
56
1.00 Mean
3225
55
35
55
0.04 Model
1016419
51
33
54
1.00 Mean
270168
54
44
54
0.07 GZMn
98260
64
42
53
0.42 Model
78915
60
38
52
0.14 Model
824663
60
37
52
0.73 Model
66438
42
25
51
1.00 Mean
121050
50
34
51
0.95 Model
1374323
56
33
51
1.00 Mean
43969
45
36
50
1.00 Mean
507
57
37
50
0.16 Model
84315
56
41
50
0.28 GZFsh
586810
56
36
49
0.02 GZFsh
235627
59
39
49
0.20 MdlGen
139386
51
34
48
1.00 Mean
1355758
64
37
48
0.62 GZFsh
39291
56
35
47
0.07 Model
1031505
62
36
47
0.16 Model
80719
56
34
47
0.18 Model
1236784
64
39
47
0.05 GZFsh
534272
52
31
46
0.21 Model
8331858
59
37
46
1.00 Mean
437520
48
31
46
0.39 Model
394
47
31
46
0.06 Model
2355308
55
37
45
0.39 Model
559472
59
33
45
0.24 Model
22870
59
36
45
0.07 GZMn
3855
59
36
45
0.05 Model
270670
69
54
44
0.08 Model
4385
54
34
44
0.05 Model
329560
68
41
44
0.35 MdlGen
1096930
60
36
44
0.42 Model
2544
62
40
44
0.04 Model
13523
59
36
44
0.03 Model
49419
48
25
43
0.20 Model
5197
55
37
43
0.10 Model
10969
57
34
43
0.14 Model
501376
68
51
43
0.25 Model
48611
48
30
42
0.24 Model
47237
52
34
41
0.43 MdlGen
2172862
60
37
40
0.20 Model
Uncert BRD
0.18
0.01
0.00
0.00
0.87
0.61
0.13
1.26
0.05
0.08
0.13
0.27
0.40
0.12
0.04
0.92
0.92
0.92
0.92
0.92
0.92
0.92
0.92
0.92
0.17 0.92
0.05
0.26 0.92
1.10 0.92
0.42
0.01
0.92
0.02
0.92
0.92
0.26 0.92
0.07 0.92
0.28
0.92
0.39 0.92
0.92
0.92
0.13
1.05 0.92
0.00 0.92
0.16
0.75 0.92
0.12
0.10
0.13 0.92
0.00 0.92
0.11
0.92
0.14
0.03
0.18
0.13 0.92
0.92
0.03 0.92
0.06
0.02
0.25 0.92
0.15 0.92
0.02
0.02
0.12
0.04
0.08
0.14 0.92
0.12
0.87 0.92
0.15
%
Caught
55
4
110
12
14
37
39
33
59
5
9
18
13
12
1
5
58
6
8
23
24
28
1
52
2
50
4
21
7
38
47
45
47
46
8
13
1
10
48
27
3
7
8
2
10
42
18
3
18
10
3
2
3
2
14
17
2
2
9
5
6
10
10
16
8
Est
C/M *
M
2.3 2.35 0.24 !
2.0
u
2.4 0.98 1.11 !!!
2.4 1.79 0.07 !
3.0 1.11 0.13 !!
2.4 2.41 0.15
2.4 2.23 0.17
2.3 1.73 0.19
2.4 1.11 0.53 !
2.0
u
2.4 2.33 0.04
2.3 2.35 0.07
2.3 2.35 0.06
1.5 ²0.87 0.14 !
3.0 1.72 0.00
1.8
u
2.4 1.10 0.52
3.0 1.11 0.05
2.0
u
2.5 1.11 0.21
2.4 2.23 0.11
2.3 0.90 0.31 !
1.8
u
3.0 0.69 0.75 !!
2.0
u
2.8 2.27 0.22
3.0 2.03 0.02
2.5 1.11 0.18
2.4 0.59 0.12 !!
2.0 1.08 0.35
2.8 4.15 0.11
2.0 1.73 0.26
2.0 0.89 0.52
2.6 0.86 0.53
1.5
u
2.3 2.78 0.05
2.4 1.96 0.01
2.3 ³1.82 0.05
1.5 1.25 0.39
3.0 0.66 0.41 !
u
2.3 2.35 0.03
2.4 0.70 0.11 !
2.4 0.59 0.03 !
2.3 2.35 0.04
2.4 1.10 0.38
2.3
u
2.3
u
2.0 ¹1.08 0.16
2.4 0.62 0.16 !
2.5 2.35 0.01
2.1 1.24 0.02
u
u
2.7 0.67 0.21 !
2.4 0.62 0.27 !
2.0
u
2.0
u
2.3
u
2.3
u
2.0
u
2.4 0.65 0.15 !
1.5
u
2.4 2.52 0.06
2.3
u
SRA
GBR Seabed Biodiversity
Biomass
%
%
%Effort
Rel
Source
kg
Available Exposed Exposed Catch
Actinopterygii Aploactis
aspera
21363
66
47
40
0.30 Model
Actinopterygii Trixiphichthys
weberi
59106
56
32
40
1.00 Model
Gastropoda
Vexillum
obeliscus cf
2302
44
25
39
0.00 Model
Gastropoda
Bufonaria
rana
19213
56
33
39
0.05 Model
Actinopterygii Inimicus
caledonicus
711097
65
47
39
0.16 Model
Echinoidea
Brissopsis
luzonica
1377669
46
27
38
0.02 Model
Crustacea
Cryptopodia
queenslandi
5162
54
32
38
0.14 Model
Actinopterygii Torquigener
whitleyi
150537
52
34
38
0.63 Model
Cephalopoda Sepia
elliptica
158747
51
30
38
1.00 Mean
Gastropoda
Aplysia
sp1_QMS
450338
51
32
38
1.00 Mean
Actinopterygii Upeneus
asymmetricus
367368
60
37
38
0.45 MdlGen
Bivalvia
Leionucula
superba
5499
56
32
37
0.00 Mean
Gastropoda
Lamellaria
sp1
5697
46
27
37
1.00 Mean
Actinopterygii Sillago
burrus
307944
46
30
37
1.00 Mean
Ophiuroidea
Dougaloplus
echinata
2513
47
27
37
0.00 Model
Crustacea
Dorippe
quadridens
3584
63
38
37
0.23 Model
Actinopterygii Grammatobothus polyophthalmus
358459
59
33
37
0.64 GZFsh
Bivalvia
Dosinia
altenai
191530
50
27
37
0.00 Model
Crustacea
Diogenidae
sp356-1
442
45
25
36
0.07 Model
Crustacea
Paguristes
sp2358-2
30865
52
31
36
0.10 Model
Actinopterygii Cynoglossus
sp juv/unident
14818
54
31
36
0.11 Model
Actinopterygii Nemipterus
nematopus
693470
37
21
36
0.61 GZFsh
Holothuroidea Holothuria
ocellata
858968
52
32
36
0.10 Model
Actinopterygii Suezichthys
gracilis
14695
61
44
36
0.13 Model
Gastropoda
Strombus
vittatus
56120
57
38
36
0.12 Model
Actinopterygii Apogon
nigripinnis
59272
65
41
35
0.07 Model
Gastropoda
Murex
brevispina
4747
61
35
35
0.04 Model
Actinopterygii Pomadasys
maculatus
1542585
65
35
35
1.00 Mean
Gymnolaemata Iodictyum
spp
17890
56
45
35
0.23 Model
Gastropoda
Strombus
campbelli
22441
64
40
35
0.05 Model
Actinopterygii Paramonacanthus otisensis
402900
58
37
35
0.23 Model
Actinopterygii Pseudorhombus elevatus
775731
67
36
35
0.38 Model
Crustacea
Calappa
terraereginae
10382
45
28
34
0.04 Model
Actinopterygii Scorpaenopsis
furneauxi
2174
60
41
34
0.16 Model
Crustacea
Portunus
sanguinolentus 1018755
65
40
34
0.21 Model
Chlorophyceae Chaetomorpha
crassa
360585
52
34
34
0.00 Model
Bivalvia
Trisidos
semitortata
402307
47
26
34
0.01 Model
Holothuroidea Holothuria
sp2
110155
64
41
34
0.06 Model
Phaeophyceae Sporochnus
comosus
1515084
57
38
34
0.00 Model
Crustacea
Paradorippe
australiensis
2758
49
27
33
0.23 Model
Anthozoa
Virgularia
sp1
2422
41
25
33
0.12 Model
Asteroidea
Luidia
hardwicki
26743
58
34
33
0.03 Model
Actinopterygii Apistus
carinatus
1073477
56
34
33
0.34 Model
Demospongiae Xenospongia
patelliformis
599
59
38
33
0.03 Model
Crustacea
Pronotonyx
leavis
329
46
24
33
0.00 Model
Asteroidea
Astropecten
sp4_AIM
11187
45
26
33
0.24 Model
Actinopterygii Paracentropogon longispinus
89968
49
30
33
0.23 Model
Gastropoda
Strombus
dilatatus
92276
53
38
33
0.06 Model
Actinopterygii Pseudorhombus spinosus
969118
58
32
33
0.35 MdlGen
Gymnolaemata Selenaria
maculata cf
288844
59
37
33
0.00 Mean
Demospongiae Ircinia
1255
7482318
46
27
33
0.23 Model
Actinopterygii Minous
versicolor
92283
58
34
32
0.16 Model
Actinopterygii Asterorhombus intermedius
154382
61
37
32
0.23 Model
Actinopterygii Paraploactis
kagoshimensis
18985
62
38
32
0.11 Model
Actinopterygii Nemipterus
furcosus
4012361
50
25
32
1.00 Mean
Crustacea
Scyllarus
demani
135376
55
33
32
0.15 Model
Crustacea
Actumnus
squamosus
915
54
36
32
0.06 Model
Phaeophyceae Padina
sp.
658602
56
40
32
0.00 Model
Actinopterygii Nemipterus
hexodon
1421345
52
21
32
1.00 Mean
Actinopterygii Dactylopus
dactylopus
63493
54
33
32
0.26 Model
Echinoidea
Salmacis
sphaeroides
342726
56
38
32
0.14 Model
Crustacea
Portunus
tuberculatus
1226
46
28
32
0.09 Model
Actinopterygii Fistularia
petimba
135435
44
26
32
1.00 Mean
Gastropoda
Gemmula
sp2
7259
46
28
32
0.00 Model
Liliopsida
Halophila
decipiens
3925942
50
30
32
0.00 Model
Bivalvia
Ctenocardia
virgo cf
6808
54
34
32
0.08 Model
Echinoidea
Ova
lacunosus
136339
44
23
31
0.00 Model
Gymnolaemata Hippothoa
distans
404
55
36
31
0.00 Model
Actinopterygii Torquigener
Sp1gloerfelt-tarp 734504
57
38
31
0.63 MdlGen
Actinopterygii Chaetodermis
penicilligera
119061
59
38
31
1.00 Mean
Actinopterygii Cynoglossus
sp kopsi group
58222
53
28
31
0.11 Model
Actinopterygii Apogon
fasciatus
223485
43
24
31
0.22 Model
Class
Genus
Species
3-234
Uncert BRD
0.21
0.36
0.00
0.01
0.06
0.01
0.07
0.41
0.92
0.92
0.92
0.92
0.61 0.92
0.92
0.00
0.20
0.00
0.00
0.04
0.05
0.06
0.77
0.04
0.08
0.06
0.04
0.03
0.92
0.92
0.92
0.92
0.92
0.92
0.31
0.02
0.10
0.16
0.03
0.25
0.15
0.00
0.01
0.04
0.00
0.15
0.09
0.01
0.26
0.03
0.00
0.22
0.23
0.02
0.25
0.92
0.92
0.92
0.92
0.92
0.92
0.16
0.10 0.92
0.09 0.92
0.07 0.92
0.92
0.05
0.02
0.00
0.92
0.31 0.92
0.12
0.06
0.92
0.00
0.00
0.03
0.00
0.00
0.41 0.92
0.92
0.04 0.92
0.09 0.92
%
Caught
11
36
0
2
6
1
5
22
38
38
16
0
37
34
0
9
22
0
3
4
4
20
4
4
4
2
1
33
8
2
7
12
2
5
7
0
0
2
0
8
4
1
10
1
0
8
7
2
10
0
7
5
7
3
30
5
2
0
29
8
4
3
29
0
0
2
0
0
18
29
3
6
SRA
2.7
3.0
2.0
2.7
1.8
2.4
1.5
2.7
3.0
2.0
2.4
2.0
2.3
2.3
2.4
3.0
2.5
1.9
1.5
3.0
2.4
2.4
2.0
2.2
2.3
2.0
2.0
2.4
2.3
2.2
2.4
2.3
2.4
2.4
2.7
2.5
2.3
2.3
3.0
3.0
2.3
2.4
2.0
2.5
2.0
2.4
2.3
Est
C/M *
M
u
0.33 1.09 !!!
u
u
u
u
u
0.88 0.25
1.25 0.30
u
2.23 0.07
u
u
0.57 0.60 !
u
u
1.19 0.18
u
u
u
0.70 0.05
1.07 0.19
u
1.05 0.04
u
1.73 0.01
u
0.34 0.96 !!!
u
u
2.53 0.03
0.62 0.19
u
0.40 0.13 !!
41.73 0.04
u
u
u
u
u
u
u
1.35 0.08
u
u
u
0.33 0.21 !!
u
0.62 0.17
u
u
u
1.19 0.06
u
0.53 0.56 !
u
u
u
0.96 0.31
1.00 0.08
u
u
0.26 1.12 !!!
u
u
u
u
u
1.08 0.17
2.53 0.11
0.70 0.04
1.73 0.04
GBR Seabed Biodiversity
Biomass
%
%
%Effort
Rel
Source
kg
Available Exposed Exposed Catch
Holothuroidea Holothuroidea
sp2
44967
50
29
31
0.07 Model
Foraminifera Discobotellina
biperforata
151281
58
35
31
0.00 Model
Rhodophyceae Chondrophycus sp1
29227
54
38
31
0.00 Model
Actinopterygii Nemipterus
sp juv/unident
6496
47
28
31
1.00 Mean
Actinopterygii Centriscus
scutatus
19885
57
35
31
0.14 Model
Crustacea
Ebalia
lambriformis
1021
52
27
31
0.00 Model
Holothuroidea Stichopus
ocellatus
2416172
49
28
31
0.11 Model
Asteroidea
Astropecten
granulatus cf
16683
46
25
31
0.05 Model
Bivalvia
Lomopsis
sp1
61039
51
29
31
0.00 Mean
Crustacea
Austrolibinia
gracilipes
1570
49
26
30
0.03 Model
Echinoidea
Chaetodiadema granulatum
80329
48
27
30
0.02 Model
Crustacea
Ceratoplax
ciliata
562
42
22
30
0.05 Model
Crustacea
Penaeus
plebejus
129674
54
35
30
0.17 Model
Bivalvia
Barbatia
parvillosa cf
1283
49
28
30
0.13 Model
Liliopsida
Halophila
ovalis
4093618
51
32
30
0.00 Model
Actinopterygii Epinephelus
sexfasciatus
285546
48
22
30
0.23 Model
Liliopsida
Halophila
spinulosa
13547972
53
33
30
0.00 Model
Rhodophyceae Dasya
sp1
60829
55
34
30
0.00 Model
Chlorophyceae Udotea
argentea
785198
51
34
30
0.00 Model
Phaeophyceae Lobophora
variegata
14640448
54
36
30
0.00 Model
Rhodophyceae Osmundaria
fimbriata
2542368
48
28
30
0.00 Model
Actinopterygii Nemipteridae
sp juv/unident
7775
52
30
30
1.00 Mean
Actinopterygii Calliurichthys
ogilbyi
123272
53
35
30
0.43 Model
Actinopterygii Trachinocephalus myops
1028380
53
36
30
0.76 Model
Bivalvia
Antigona
lamellaris
23273
45
25
30
0.10 Model
Gastropoda
Xenophora
indica
25049
44
27
30
0.08 Model
Liliopsida
Halophila
tricostata
911642
45
25
30
0.00 Model
Gastropoda
Philine
sp1
8236
54
33
29
0.04 Model
Gastropoda
Natica
vitellus
8642
53
30
29
0.04 Model
Bivalvia
Annachlamys
kuhnholtzi
365468
48
35
29
0.07 MdlGen
Rhodophyceae Polysiphonia
sp1
36529
54
33
29
0.00 Model
Anthozoa
Alertigorgia
orientalis
33318
50
27
29
0.09 Model
Asteroidea
Astropecten
spp
76778
45
26
29
0.00 Model
Rhodophyceae Laurencia
sp2
259365
50
31
29
0.00 Model
Crustacea
Scyllarus
martensii
6856
50
31
29
0.12 Model
Bivalvia
Paphia
undulata cf
7989
45
25
29
0.00 Model
Bivalvia
Corbula
sp2
828447
48
27
29
0.13 MdlGen
Crustacea
Neopalicus
jukesii
12459
47
29
29
0.00 Model
Bivalvia
Annachlamys
flabellata
289876
51
32
29
0.07 Model
Crustacea
Oratosquillina
quinquedentata
62852
49
26
29
0.40 Model
Actinopterygii Cynoglossus
maccullochi
32047
59
35
29
0.18 Model
Chlorophyceae Codium
geppii
224230
52
36
29
0.00 Model
Actinopterygii Carangidae
sp juv/unident
19532
38
20
29
1.00 Mean
Gymnolaemata Schizomavella
spp
2378
39
22
29
0.41 MdlGen
Actinopterygii Repomucenus
limiceps
267345
52
34
29
0.86 Model
Phaeophyceae Dictyotales
sp
433762
53
35
29
0.00 Model
Rhodophyceae Griffithsia
sp
10773
52
33
29
0.00 Model
Bivalvia
Corbula
macgillvrayi
205900
47
25
29
0.13 Model
Crustacea
Leucosia
formosensis
1341
53
34
28
0.06 Model
Cephalopoda Sepiidae
spp
792725
49
30
28
0.00 Model
Crustacea
Portunus
rubromarginatus 5914213
55
34
28
0.49 Model
Asteroidea
Oreasteridae
sp1
2814889
56
37
28
0.10 Model
Crustacea
Sicyonia
rectirostris
259
56
34
28
0.10 Model
Chlorophyceae Udotea
glaucescens
1402555
42
23
28
0.00 Model
Phaeophyceae Sporochnus
moorei
2797072
53
32
28
0.00 Model
Polychaeta
Chloeia
flava
15742
49
28
28
1.00 Mean
Chlorophyceae Cladophora
sp
39457
52
33
28
0.00 Model
Cephalopoda Sepiadariidae
sp5
17032
52
33
28
0.10 MdlGen
Actinopterygii Torquigener
cf pallimaculatus 358774
52
31
28
0.63 MdlGen
Echinoidea
Laganum
depressum
2276625
50
30
28
0.04 Model
Gymnolaemata Thalamoporella spp
56643
53
32
28
0.00 Model
Gymnolaemata Retelepralia
mosaica
62
45
33
28
0.00 Model
Rhodophyceae Lithophyllum
sp1
21086914
50
33
28
0.00 Model
Cephalopoda Cephalopoda
spp
750874
47
27
27
0.00 Model
Crustacea
Dromidiopsis
edwardsi
27212
46
26
27
0.16 Model
Demospongiae Reniochalina
stalagmitis
500459
41
23
27
0.04 Model
Actinopterygii Cynoglossus
sp4
466488
49
31
27
0.16 Model
Anthozoa
Trachyphyllia
geoffroyi
3171803
46
27
27
0.00 Model
Chlorophyceae Halimeda
sp2
477419
46
30
27
0.00 Model
Crustacea
Pagurid
sp17
2067
47
27
27
0.06 Model
Gastropoda
Conus
ammiralis
17677
51
34
27
0.08 Model
Holothuroidea Holothuroidea
sp22
1113121
51
26
27
0.00 Model
Class
Genus
Species
3-235
Uncert BRD
0.05
0.00
0.00
0.92
0.19 0.92
0.00
0.07
0.05
0.02
0.02
0.03
0.28
0.09
0.00
0.15 0.92
0.00
0.00
0.00
0.00
0.00
0.92
0.26 0.92
0.48 0.92
0.09
0.05
0.00
0.02
0.02
0.03
0.00
0.08
0.00
0.00
0.04
0.00
0.12
0.00
0.03
0.22
0.12 0.92
0.00
0.92
0.58
0.43 0.92
0.00
0.00
0.12
0.05
0.00
0.16
0.04
0.12
0.00
0.00
0.00
0.06
0.41 0.92
0.03
0.00
0.00
0.00
0.00
0.12
0.04
0.06 0.92
0.00
0.00
0.02
0.04
0.00
%
Caught
2
0
0
28
4
0
3
1
0
1
1
2
5
4
0
6
0
0
0
0
0
28
12
21
3
2
0
1
1
2
0
3
0
0
3
0
4
0
2
12
5
0
26
12
23
0
0
4
2
0
14
3
3
0
0
28
0
3
16
1
0
0
0
0
4
1
4
0
0
2
2
0
SRA
2.0
2.7
2.4
2.0
1.8
2.3
2.3
2.0
2.7
2.6
2.5
3.0
2.0
2.0
2.0
2.3
2.0
2.0
1.5
2.4
2.5
2.5
2.0
1.5
2.3
2.3
2.5
1.8
2.3
2.4
2.3
2.0
2.0
Est
C/M *
M
u
u
u
0.86 0.33
0.83 0.05
u
u
u
u
u
u
u
u
u
u
0.98 0.07
u
u
u
u
u
1.33 0.21
1.05 0.11
0.97 0.22
u
u
u
u
u
u
u
0.21 0.12 !!!
u
u
u
u
u
u
u
u
0.70 0.07
u
0.98 0.27
u
1.08 0.21
u
u
u
u
u
u
u
u
u
u
u
u
u
1.08 0.15
u
u
u
u
u
u
u
0.70 0.06
u
u
u
u
u
GBR Seabed Biodiversity
Biomass
%
%
%Effort
Rel
Source
kg
Available Exposed Exposed Catch
Gymnolaemata Robertsonidra
spp
12430
50
33
27
0.00 Mean
Crustacea
Porcellanid
sp4154
693
49
29
27
0.09 Model
Gastropoda
Atys
cylindricus cf
4963
49
28
27
0.16 Model
Actinopterygii Sorsogona
tuberculata
1332608
53
32
27
0.65 Model
Crustacea
Arcania
elongata
11264
47
27
27
0.18 Model
Rhodophyceae Heterosiphonia muelleri
1083464
52
31
27
0.00 Model
Phaeophyceae Dictyopteris
sp2
1307425
52
31
27
0.00 Model
Gastropoda
Nassarius
conoidalis cf
2625
52
29
27
0.05 Model
Actinopterygii Adventor
elongatus
11470
44
22
26
0.20 Model
Chlorophyceae Halimeda
borneenses
10124447
45
26
26
0.00 Model
Actinopterygii Upeneus
sp juv/unident
5365
48
26
26
0.06 Model
Crustacea
Dorippe
sp7142-12
601019
45
24
26
0.09 Model
Bivalvia
Corbula
fortisulcata
6462
43
23
26
0.13 MdlGen
Cephalopoda Sepia
whitleyana
493757
50
32
26
1.00 Mean
Gastropoda
Biplex
pulchellum
59197
47
26
26
0.00 Mean
Actinopterygii Cynoglossus
ogilbyi
28505
44
25
26
0.15 Model
Asteroidea
Stellaster
equestris cf
2055943
49
29
26
0.16 Model
Demospongiae Mycale
mirabilis
401414
45
31
26
0.07 Model
Bivalvia
Spondylus
wrightianus
103088
50
30
26
0.09 Model
Echinoidea
Peronella
orbicularis cf
32469
50
28
26
0.07 Model
Gymnolaemata Orthoscuticella
spp
26462
48
30
26
0.00 Model
Demospongiae Disyringa
sp1
9021
46
23
26
0.00 Mean
Gastropoda
Xenophora
solarioides
37081
47
27
26
0.07 Model
Actinopterygii Engyprosopon
grandisquama
1306624
57
30
26
0.50 Model
Rhodophyceae Amansia
glomerata
1902406
41
21
26
0.00 Model
Crustacea
Nursilia
sp nov
3789
50
26
26
0.00 Mean
Echinoidea
Laganidae
sp3
190493
44
24
26
0.00 Model
Bivalvia
Chama
spp
13590
50
32
26
0.02 MdlGen
Crustacea
Portunus
tenuipes
1911035
42
22
26
0.68 Model
Rhodophyceae Gracilaria
sp1
1475393
48
29
26
0.00 Model
Actinopterygii Minous
trachycephalus
649193
47
26
26
0.00 Model
Brachiopoda Brachiopoda
sp1_MTQ
79799
41
31
25
0.00 Model
Crustacea
Trachypenaeus granulosus
424353
50
25
25
0.29 Model
Rhodophyceae Gracilaria
sp2
975404
48
29
25
0.00 Model
Phaeophyceae Lobophora
sp
6615675
53
30
25
0.00 Model
Actinopterygii Erosa
erosa
175940
47
32
25
0.21 Model
Actinopterygii Tragulichthys
jaculiferus
637207
47
25
25
0.11 Model
Actinopterygii Choerodon
cephalotes
421829
46
27
25
0.23 Model
Actinopterygii Zebrias
craticula
206568
49
30
25
0.12 Model
Crustacea
Pilumnus
longicornis
2472
46
26
25
0.06 Model
Crustacea
Dardanus
callichela var
29106
52
31
25
0.11 Model
Rhodophyceae Dasya
sp
365331
49
29
25
0.00 Model
Bivalvia
Chama
pulchella
874718
46
27
25
0.02 Model
Phaeophyceae Dictyota
sp1
82767
46
31
25
0.00 Model
Actinopterygii Upeneus
tragula
843838
47
25
25
0.45 MdlGen
Crustacea
Allogalathea
elegans
2905
49
29
25
0.04 Model
Gymnolaemata Scuticella
plagiostoma
233332
48
30
25
0.00 Mean
Crustacea
Carinosquilla
redacta
44868
41
23
25
0.24 Model
Actinopterygii Pseudochromis quinquedentatus
1907
54
26
25
0.05 Model
Crustacea
Actaea
jacquelinae
958
46
30
25
0.07 Model
Crustacea
Phalangipus
filiformis
30500
44
23
25
0.16 Model
Actinopterygii Synodus
tectus group
707843
49
29
25
0.43 Model
Demospongiae Demospongiae sp11
48060
40
28
25
0.12 Model
Ophiuroidea
Ophiothrix
sp14
4806
46
27
25
0.09 Model
Demospongiae Mycale
sp9
1633320
47
30
25
0.05 Model
Anthozoa
Cycloseris
cyclolites
112329
55
32
24
0.11 Model
Bivalvia
Mimachlamys
gloriosa
24876
44
24
24
0.08 Model
Crustacea
Myra
mammillaris
25080
46
27
24
0.02 Model
Demospongiae Demospongiae sp89
63728
50
27
24
0.06 Model
Actinopterygii Priacanthus
tayenus
1576906
45
19
24
1.00 Mean
Crustacea
Parapenaeopsis venusta
10034
51
32
24
0.18 Model
Crustacea
Penaeid unknown unknown
2674
44
24
24
0.10 Model
Gastropoda
Philine
angasi
5859
44
24
24
0.04 Model
Chondrichthyes Dasyatis
leylandi
176702
44
24
24
1.00 Mean
Crustacea
Thalamita
hanseni
54500
44
24
24
0.13 Model
Mollusca
Mollusca
eggs
18397
44
24
24
0.09 Model
Crustacea
Alpheidae
sp2434
164
44
24
24
0.03 Model
Anthozoa
Pteroides
sp1
22889
44
24
24
0.13 Model
Anthozoa
Pteroides
sp2
9105
44
24
24
0.12 Model
Class
Genus
Species
3-236
Uncert BRD
0.04
0.14
0.19 0.92
0.08
0.00
0.00
0.02
0.14 0.92
0.00
0.09 0.92
0.09
0.12
0.11 0.92
0.04
0.07
0.04
0.02
0.00
0.02
0.13 0.92
0.00
0.00
0.03
0.25
0.00
0.00
0.00
0.08
0.00
0.00
0.08
0.08
0.22
0.05
0.03
0.04
0.00
0.03
0.00
0.61
0.02
0.92
0.92
0.92
0.92
0.92
0.92
0.17
0.03 0.92
0.06
0.07
0.27 0.92
0.08
0.05
0.03
0.09
0.03
0.03
0.05
0.92
0.16
0.12
0.03
0.11
0.06
0.02
0.11
0.18
¹Dredge (1985); ²estimated from Kodama et al. (2006); ³Penn (1976); 4Lee and Hsu (2003)
%
Caught
0
3
4
16
5
0
0
1
5
0
1
2
3
26
0
4
4
2
2
2
0
0
2
12
0
0
0
1
17
0
0
0
7
0
0
5
3
5
3
1
3
0
0
0
10
1
0
6
1
2
4
10
3
2
1
3
2
1
2
22
4
2
1
24
3
2
1
3
3
SRA
2.0
2.4
2.0
2.7
2.6
2.0
1.5
2.0
2.4
2.3
2.0
2.3
2.0
2.4
2.3
2.4
2.3
2.2
2.7
2.7
2.4
2.3
2.3
2.6
1.5
2.0
2.3
1.8
2.4
2.3
2.0
2.0
2.7
2.3
2.3
2.3
2.3
2.0
Est
C/M *
M
u
u
u
0.62 0.26
u
u
u
u
u
u
2.18 0.01
u
u
1.25 0.21 u
u
0.94 0.04
u
u
u
u
u
u
u
1.19 0.10
u
u
u
u
u
u
u
u
u
u
u
u
u
1.05 0.05
0.84 0.03
u
u
u
u
u
2.18 0.05
u
u
u
u
u
u
0.97 0.10
u
u
u
u
u
u
u
1.20 0.19
u
u
u
0.41 0.59
u
u
u
u
u
GBR Seabed Biodiversity
(a) Actinopterygii: Fistularia petimba
(b) Actinopterygii: Trixiphichthys weberi
(c) Actinopterygii: Pomadasys maculatus
(d) Actinopterygii: Sillago burrus
Figure 3-123: Model distribution maps of selected species with highest sustainability risk indicators.
3-237
GBR Seabed Biodiversity
3-238
3.7.3. Assemblage indicators
On the basis of the sites-groups characterisation and biophysical predictions of the species assemblage
data in Section 3.4, area-based (number of grid cells) exposure indicators were estimated similar to
those considered for species-group biomass (Section 3.7.1). First was the amount of area of each
assemblage type located in various marine park zones, in particular the percentage of the total area
located in GU zones was available to trawling and potentially at risk (Table 3-58).
Eleven of the 16 assemblages had more than 25% of their area in GU zones (Table 3-58, % Available,
pale orange), 9 of those had more than 50% of their area in GU zones (dark orange) and none had
more than 75% of its area in GU zones. The lowest level of availability was 5% and the highest level
was 73%.
The next indicator was the percentage of area of each assemblage located in grid cells where trawl
effort was present — regardless of the intensity of effort in the grid cells (Table 3-58, % Exposed).
Seven assemblages had more than 25% of their area in grid cells with trawl effort (pale orange) and
three of those had more than 50% of their area in grid cells with trawl effort. The lowest level of
exposure was 0% and the highest level was 58%. This indicator is more specific and more sensitive
than the previous.
The third indicator was the percentage of area of each assemblage directly exposed to trawl effort
taking into account the intensity of trawl effort (Table 3-58, Effort Exposed %). The table shows the
amount of area exposed at several different levels of effort intensity, as well as the final total exposure
as a percentage. Five of the 16 assemblages had more than 25% of their area directly exposed to trawl
effort in 2005 (Table 3-58, pale orange), two had more than 50% of area directly exposed (dark
orange) and one had more than 100% of area directly exposed due to being trawled multiple times
(red). The lowest level of exposure was 0% and the highest level was 108%. The exposures between
32% and 41% indicate moderate-low risk, exposure of 58% indicates moderate-high risk, and
exposure of 108% indicates high potential risk.
The highest exposure, at 108%, was assemblage 12 representing primarily group E species (Figure
3-49) — including Cryptolutea arafurensis, Saurida argentea/tumbil, Enisiculus cultellus and
Placamen tiara (see Table 3-60 for list of 40 species with greatest affinity for assemblage 14) —
followed by groups G and C species. Assemblage 12 was distributed in patches along the
coastal/inner-shelf from the Whitsundays to Cape Upstart and from Cairns north (Figure 3-47).
The next most exposed, at 58%, was assemblage 11 representing primarily group E species (Figure
3-49) — including Scolopsis taeniopterus, Charybdis truncata, Terapon theraps, Leiognathus
leuciscus, Metapenaeus endeavouri and Calliurichthys grossi (see Table 3-60 for top 40 species) —
— followed by groups G and C species. Assemblage 11 was distributed along coastal areas north of
Mackay and more broadly across the inner/mid-shelf from Cairns north (Figure 3-47).
Next was assemblage 4, at 41%, representing primarily group K species (Figure 3-49) — including
Orthoscuticella spp, Ambiserrula jugosa, Arachnopusia spp, Xenospongia patelliformis and Scuticella
plagiostoma (see Table 3-60 for top 40 species) — followed by groups G and C species. Assemblage 4
was distributed over much of the mid/outer shelf in the Capricorn section of the GBR (Figure 3-47).
Next was assemblage 13, at 41%, representing primarily groups E and G species (Figure 3-49) —
including Leiognathus splendens, Leiognathus moretoniensis, Trachypenaeus anchoralis, Gerres
filamentosus and Metapenaeus ensis (see Table 3-60 for top 40 species) — and was scattered patchily
close inshore from the Whitsundays north (Figure 3-47).
Assemblage 1 had 32% exposure and represented primarily group A species (Figure 3-49) —
including Poraster superbus, Portunus argentatus, Atys cylindricus cf, Richardsonichthys leucogaster
and Caulerpa brachypus (see Table 3-60 for top 40 species) — and was distributed in the mid/outer
shelf in the central section offshore from about Cape Upstart to Hinchinbrook Is (Figure 3-47).
The remaining clusters had low to zero levels of exposure and included a number of species affinity
groups, particularly K, J, but also L, H, B, D (Figure 3-49).
GBR Seabed Biodiversity
3-239
A number of species occurred repeatedly across the more highly exposed assemblages. As an
indication of their cumulative exposure, species were ranked by the sum of the products of their
assemblage exposure by assemblage affinity — so that species with higher affinities for several more
exposed assemblages would have a higher ranking. The top 40 species with highest exposure are also
listed in Table 3-60 and primarily includes group E species, with some G and C. Many of these species
are the same as those ranked with higher exposure in the singles species assessment (Section 3.7.2.2,
Table 3-57).
3.7.4. Habitat indicators
On the basis of the characterisation and biophysical predictions of the video habitat data in Section
3.5.3, area-based (number of grid cells) exposure indicators were estimated similar to those considered
for species-group biomass (Section 3.7.1) and assemblage distribution (Section 3.7.3). First was the
amount of area of each habitat cluster type located in various marine park zones, in particular the
percentage of the total located in GU zones was available to trawling and potentially at risk (Table
3-59).
Eight of the 9 habitat clusters had more than 25% of their area in GU zones (Table 3-59, % Available,
pale orange), 2 of those groups had more than 50% of their area in GU zones (dark orange) and one
group (cluster 6) had more than 75% of its area in GU zones (red). The lowest level of availability was
12% and the highest level was 80%.
The next indicator was the percentage of area of each cluster located in grid cells where trawl effort
was present — regardless of the intensity of effort in the grid cells (Table 3-59, % Exposed). Four
clusters had more than 25% of their area in grid cells with trawl effort (pale orange) and two of those
had more than 50% of their area in grid cells with trawl effort. The lowest level of exposure was 6%
and the highest level was 64%. This indicator is more specific and more sensitive than the previous.
The third indicator was the percentage of area of each cluster directly exposed to trawl effort taking
into account the intensity of trawl effort (Table 3-59, % Effort Exposed). The table shows the amount
of area exposed at several different levels of effort intensity, as well as the final total exposure as a
percentage. Five of the 9 clusters had more than 25% of their area directly exposed to trawl effort in
2005 (Table 3-59, pale orange) and none had more than 50% of area directly exposed. The lowest
level of exposure was 3% and the highest level was 39%. Exposures between 25% and 39% indicate
moderate-low risk. The highest, at 39%, was cluster 7 representing patchy seagrass and algal habitat
(Figure 3-61) distributed along the mid-shelf from Cape Upstart to Innisfail (Figure 3-62). The next
most exposed, at 34%, was cluster 6 also representing patchy seagrass and algal habitat (Figure 3-61)
distributed along much of the inner-shelf in the southern Capricorn section of the GBR (Figure 3-62).
Next was cluster 5, also at 34%, representing mostly bioturbated and bare seabed with a little algae
and seagrass algal habitat (Figure 3-61) distributed over much of the shelf in the central and northern
sections of the GBR (Figure 3-62). Next was cluster 1, at 26%, representing the most barren seabed
type — almost entirely bare and bioturbated with very little biohabitat (Figure 3-61) — distributed in
inshore muddy areas and the Capricorn Channel (Figure 3-62). Cluster 9 had 25% exposure and
represented patchy algae (including limited Halimeda) with some bioturbation and a little other
biohabitat distributed offshore from Townsville (Figure 3-62). The remaining clusters included most
of the Halimeda banks and epibenthic garden biohabitats and had low levels of exposure.
The exposure of habitat components identified from the frame-level video post-analysis (Section
3.5.2) was examined from the point data (i.e. not from models of predicted distributions), which
carries some risk of bias due to the purposeful stratification of the sampling design. Overall, 62% of
the observed seagrass occurred in GU zones and 47% was observed in grid cells where trawl effort
was present; but given the intensity of trawl effort the total exposure was 21%. The majority of
observed seagrass was Halophila spinulosa and like-species, with indicators or 67%, 54% and 24%
respectively, which is a very similar exposure outcome as the modelled sample distribution for this
species (22%, Table 3-49). Ovoid leaf Halophila’s were ranked next at 54%, 33%, and 15%, which is
also a very similar exposure outcome as the modelled sample distribution for Halophila ovalis (18%,
Table 3-49). The concordance between these two completely independent sources of data, and raw
GBR Seabed Biodiversity
3-240
versus modelled distributions, gives confidence to the estimates of exposure. Other observed
morphotypes of seagrass had very low exposure.
Thirty four percent of observed algae, overall growth forms combined, occurred in GU zones and 18%
was observed in grid cells where trawl effort was present; given the intensity of trawl effort the total
exposure was 14%. The exposure of the diversity of different morphotypes of algae varied
considerably. The most exposed form was crustose coralline algae (total exposure 44%); primarily in
the trawl grounds off Gladstone. Crustose coralline algae nodules can be considered a robust growth
form. The next most exposed form was filamentous blue-green algae (total exposure 25%). All other
growth forms were ≤17% and most (including Halimeda’s) were <5%.
Thirty eight percent of observed gorgonians, overall growth forms combined, occurred in GU zones
and 15% was observed in grid cells where trawl effort was present; given the intensity of trawl effort
the total exposure was 3%. The exposure of the diversity of different morphotypes of gorgonians was
≤2% for all but Solenocaulon, with forms covered with epifauna having 20% exposure and those
without having 14% exposure.
Twenty two percent of observed soft corals, overall growth forms combined, occurred in GU zones
and 9% was observed in grid cells where trawl effort was present; given the intensity of trawl effort
the total exposure was 4%. The exposure of the diversity of different morphotypes of soft corals was
≤4% for all but Pteroides, which had 15% exposure. This is also a very similar exposure outcome as
the modelled sample distribution for this genus of sea pen (16%). Sea pens appear to have a low
catchability (~0.06) with narrow uncertainty (~0.05), so would appear to be at low risk.
Seventeen percent of observed sponges, overall growth forms combined, occurred in GU zones and
10% was observed in grid cells where trawl effort was present; given the intensity of trawl effort the
total exposure was 3%. The exposure of the diversity of different morphotypes of sponges was ≤6%
for all but barrel forms, which had 14% exposure (excluding Ircinia and Xestospongia) and foliose
forms, which had 10% exposure.
Twenty nine percent of observed bryozoans, overall growth forms combined, occurred in GU zones
and 14% was observed in grid cells where trawl effort was present; given the intensity of trawl effort
the total exposure was 7%. The exposure of the diversity of different morphotypes of bryozoans varied
between 0–11%.
Six percent of observed hard corals, overall growth forms combined, occurred in GU zones and 6%
was observed in grid cells where trawl effort was present; given the intensity of trawl effort the total
exposure was ~1%. The exposure of the diversity of different morphotypes of hard corals was mostly
close to zero (but ≤3%) for all forms except solitary corals, which had 12% exposure. The most
common sampled solitary corals were of the genus Cycloseris with a total exposure of ~17% and low
catchability (~0.07) with narrow uncertainty (~0.06), so would appear to be at low risk.
GBR Seabed Biodiversity
3-241
Table 3-58: Ecological Risk Indicators with respect to trawling for estimated area (km²) of predicted distributed of species assemblages (site clusters): by GBRMP Zoning indicating
percent of area available; by area not trawled/trawled indicating percent area potentially exposed; by trawl intensity (ann_hrs/0.01º cell) indicating percent area exposed to effort.
Assemblage
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
General
Habitat
Marine Preser
%
TOTAL
Use
Protection Nat Park -vation
Available
7521
3960
3212
16 14709
51
10736
11496
6287
45 28565
38
2336
6308
5717
13 14374
16
6670
573
2240
23
9506
70
2900
548
1844
17
5310
55
8369
4409
3718
49 16545
51
8934
1015
2955
37 12940
69
4082
4990
5061
38 14172
29
1523
4511
3513
29
9576
16
12004
4344
5154
35 21537
56
13934
4501
5912
20 24367
57
2668
1284
712
42
4706
57
2827
92
929
3
3851
73
208
1258
630
14
2110
10
2124
6889
4677
0 13690
16
179
532
2973
0
3684
5
Not
%
Trawled
trawled
Exposed
8556
6154
42
23098
5467
19
13157
1217
8
4175
5331
56
4285
1025
19
11870
4676
28
7329
5611
43
13656
516
4
9369
207
2
19475
2062
10
14161
10206
42
2247
2459
52
1634
2217
58
2110
0
0
13620
70
1
3676
8
0
0 0.125 0.25 0.5
1
2
4
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
134
110
28
109
21
90
105
11
3
35
194
38
40
0
0
0
303
210
32
231
41
172
178
14
1
47
356
67
75
0
0
0
812 1337 1229 640 132
0
0
365 610 663 279
27
0
0
41
84
58
59
0
0
0
496 1048 1326 538
51
0
0
61 157 159
66
0
0
0
286 542 548 374
42
0
0
317 709 773 529
29
0
0
16
31
41
29 210 59
0
3
0
0
0
0
0
0
69 115
84
31
13 21
0
690 1378 2543 3968 4041 699 184
191 449 1119 1406 1696 80
0
151 331 428 396
96
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
12
22
7
15
4
23
25
3
2
15
30
5
8
0
1
0
25 61
38 68
9 16
25 54
7
9
30 48
39 65
3
9
2
1
14 23
52 109
9 17
14 25
0
0
1
1
0
0
8
16
32
64
128 256
Effort
Exp%
32
8
2
41
10
13
21
3
0
2
58
108
41
0
0
0
Table 3-59: Ecological Risk Indicators with respect to trawling for estimated area (km²) of predicted distributed of video habitat clusters: by GBRMP Zoning indicating percent of
area available; by area not trawled/trawled indicating percent area potentially exposed; by trawl intensity (ann_hrs/0.01º cell) indicating percent area directly exposed to effort.
Cluster
1
2
3
4
5
6
7
8
9
General
Use
6669
15208
9639
10426
25857
10737
4541
779
3158
Habitat
Marine
ProtecNat Park
tion
2979
5586
9397
9023
11538
8994
11239
9328
14870
13874
385
2234
165
1469
2706
2892
3431
2134
Preser
-vation
10
41
42
97
113
3
0
38
38
TOTAL
15244
33670
30213
31089
54714
13360
6175
6414
8761
%
Available
44
45
32
34
47
80
74
12
36
Not
%
Trawled
trawled
Exposed
11606
30004
24461
26720
39217
5729
2239
6035
6407
3639
3667
5752
4370
15497
7632
3936
379
2354
24
11
19
14
28
57
64
6
27
0
0.125 0.25 0.5
0
0
0
0
0
0
0
0
0
13
20
22
19
54
28
10
2
5
1
2
4
8
16
32
64
17 36 56 114 279 590 1229 1161 351
27 52 71 107 140 150 246 280 206
36 64 120 223 424 805 884 297
20
28 47 95 154 268 511 642 417
67
84 161 291 525 1012 2076 3335 4785 5521
45 84 144 299 576 1339 1423 605
41
17 36 79 187 510 904 439 177
27
2
3
6 16
33
29
29
64
0
10 24 55 100 261 386 744 531 105
%
128 256 Effort
Exp
55 47 26
0
0 4
0
0 10
0
0 7
803 137 34
0
0 34
0
0 39
0
0 3
0
0 25
GBR Seabed Biodiversity
3-242
Table 3-60. Species with greatest affinity (top 40) for site-group assemblages identified in Section 3.4, with the highest levels of trawl exposure.
Assemblage 12
Cryptolutea
arafurensis
Saurida
argentea/tumbil
Enisiculus
cultellus
Pentaprion
longimanus
Placamen
tiara
Tripodichthys
angustifrons
Upeneus
sundaicus
Brachirus
muelleri
Metapenaeus
ensis
Nassarius
cremmatus cf
Apogon
poecilopterus
Penaeus
semisulcatus
Erugosquilla
woodmasoni
Nemipterus
nematopus
Diogenidae
sp356-1
Psettodes
erumei
Leiognathus
leuciscus
Thenus
parindicus
Charybdis
truncata
Yongeichthys
nebulosus
Upeneus
sulphureus
Portunus
tuberculosus
Leiognathus
splendens
Euristhmus
nudiceps
Scolopsis
taeniopterus
Calliurichthys
grossi
Gerres
filamentosus
Leiognathus
bindus
Terapon
puta
Portunus
hastatoides
Apogon
fasciatus
Nemipterus
hexodon
Terapon
theraps
Metapenaeus
endeavouri
Dosinia
altenai
Vexillum
obeliscus cf
Corbula
fortisulcata
Sepia
elliptica
Epinephelus
sexfasciatus
Sea pen
sp1
Assemblage 11
Scolopsis
taeniopterus
Charybdis
truncata
Terapon
theraps
Leiognathus
leuciscus
Metapenaeus
endeavouri
Calliurichthys
grossi
Thenus
parindicus
Upeneus
sundaicus
Euristhmus
nudiceps
Dougaloplus
echinata
Vexillum
obeliscus cf
Penaeus
semisulcatus
Sepia
pharaonis
Melaxinaea
vitrea
Selaroides
leptolepis
Amusium
pleuronectes cf
Penaeus
esculentus
Lophiotoma
acuta
Cynoglossus
maculipinnis
Nemipterus
peronii
Psettodes
erumei
Pronotonyx
leavis
Diogenidae
sp356-1
Nemipterus
furcosus
Nassarius
cremmatus cf
Portunus
tuberculosus
Ova
lacunosus
Inegocia
japonica
Modiolus
elongatus
Brissopsis
luzonica
Pentaprion
longimanus
Dosinia
altenai
Trisidos
semitortata
Yongeichthys
nebulosus
Astropecten
spp
Sea pen
sp1
Carangidae
sp juv/unident
Astropecten
zebra
Cynoglossus
sp 1 punctate
Nemipterus
nematopus
Assemblage 4
Orthoscuticella
spp
Ambiserrula
jugosa
Arachnopusia
spp
Xenospongia
patelliformis
Scuticella
plagiostoma
Iodictyum
spp
Annachlamys
kuhnholtzi
Exochella
conjuncta cf
Junceella
juncea
Codium
geppii
Emballotheca
spp
Suezichthys
gracilis
Paguridae
sp213
Penaeus
plebejus
Robertsonidra
spp
Trachinocephalus
myops
Retelepralia
mosaica
Choerodon
venustus
Bohadschia
marmorata cf
Inimicus
caledonicus
Batrachomoeus
dubius/trispinosus
Chama
spp
Aploactis
aspera
Annachlamys
flabellata
Brachiopoda
sp1_MTQ
Beania
discodermiae cf
Struvea
elegans
Plicatula
chinensis cf
Upeneus
filifer
Telopora
spp
Macropora
spp
Conus
ammiralis
Lepralia
elimata
Beania
plurispinosa cf
Padina
sp
Amusium
balloti
Figularia
clithridiata cf
Erosa
erosa
Crepidacantha
spp
Onigocia
cf macrolepis
Assemblage 13
Leiognathus
splendens
Leiognathus
moretoniensis
Trachypenaeus
anchoralis
Gerres
filamentosus
Metapenaeus
ensis
Penaeus
semisulcatus
Tripodichthys
angustifrons
Terapon
puta
Apogon
poecilopterus
Leiognathus
cf bindus
Enisiculus
cultellus
Nassarius
cremmatus cf
Chaetodiadema
granulatum
Erugosquilla
woodmasoni
Oratosquillina
gravieri
Terapon
theraps
Portunus
hastatoides
Repomucenus
belcheri
Pseudorhombus
arsius
Saurida
argentea/tumbil
Brachirus
muelleri
Upeneus
sundaicus
Torquigener
whitleyi
Myra
tumidospina
Caranx
bucculentus
Polydactylus
multiradiatus
Pentaprion
longimanus
Calliurichthys
grossi
Leucosia
ocellata
Psettodes
erumei
Sillago
burrus
Upeneus
sulphureus
Parapercis
diplospilus
Thenus
parindicus
Leiognathus
leuciscus
Sea pen
sp1
Placamen
tiara
Euristhmus
nudiceps
Leiognathus
bindus
Cryptopodia
queenslandi
Assemblage 1
Poraster
superbus
Portunus
argentatus
Atys
cylindricus cf
Richardsonichthys
leucogaster
Caulerpa
brachypus
Takedana
eriphioides
Palicoides
whitei
Xenophora
indica
Demospongiae
sp11
Solenocera
pectinata
Sicyonia
lancifer
Atys
sp1
Phyllodictyon
sp1
Actumnus
squamosus
Mycale
mirabilis
Trachypenaeus
curvirostris
Didymozoum
spp
Strombus
dilatatus
Xenophora
cerea cf
Parthenope
turriger
Parapercis
snyderi
Udotea
flabellum
Udotea
argentea
Scyllarus
sp3418
Conescharellina
spp
Naxoides
taurus
Choerodon
frenatus
Calappa
sp 1984
Calliurichthys
ogilbyi
Apogon
timorensis
Demospongiae
sp13
Myrine
kesslerii
Microdictyon
umbilicatum
Avrainvillea
sp1
Portunus
tenuipes
Rogadius
patriciae
Scyllarus
martensii
Callyspongia
sp6
Temnopleuridae
sp2_QMS
Crella
1525
Exposure across Assemblages
Enisiculus
cultellus
Cryptolutea
arafurensis
Saurida
argentea/tumbil
Pentaprion
longimanus
Thenus
parindicus
Nassarius
cremmatus cf
Placamen
tiara
Tripodichthys
angustifrons
Upeneus
sundaicus
Erugosquilla
woodmasoni
Penaeus
semisulcatus
Psettodes
erumei
Charybdis
truncata
Aplysia
sp1_QMS
Metapenaeus
ensis
Calliurichthys
grossi
Melaxinaea
vitrea
Diogenidae
sp356-1
Portunus
tuberculosus
Sea pen
sp1
Portunus
hastatoides
Terapon
theraps
Dosinia
altenai
Leiognathus
leuciscus
Holothuria
ocellata
Myra
tumidospina
Sepia
elliptica
Sepia
pharaonis
Euristhmus
nudiceps
Vexillum
obeliscus cf
Apogon
poecilopterus
Selaroides
leptolepis
Nemipterus
sp juv/unident
Metapenaeus
endeavouri
Portunus
tuberculatus
Gemmula
sp2
Apogon
fasciatus
Scolopsis
taeniopterus
Torquigener
whitleyi
Dougaloplus
echinata
GBR Seabed Biodiversity
3-243
3.8. TRAWL MANAGEMENT SCENARIO MODEL (N Ellis, R Pitcher)
Total trawl effort in the region increased gradually after the fishery initially commenced, but grew
rapidly in the early 1990s, before peaking in 1996/1997 and falling rapidly in the late 1990s (by
~25%) (Figure 2-42) — even before implementation of the management scenarios evaluated here. The
average simulated effort is shown in Figure 3-124. The status quo 2001 model scenario (SQ2001)
maintained these effort levels through until 2025. The first intervention was the 2001 spatial closure
(CL2001) with the same effort levels. The second intervention was the latter closure plus the 2001
buy-back (CL/BB2001), which reduced effort by a further ~30% (down ~45% from the 1990s peak).
The third intervention was the latter plus the progressive penalty (CL/BB2001+P), which reduced
effort by a further ~30% again (down ~60% from the 1990s peak). The fourth intervention added the
RAP re-zoning CL/BB2001+P+RAP) at the same effort levels. The fifth intervention added the RAPassociated buy-back (CL/BB2001+P+RAP+BB2005), which reduced effort again by almost ~10%
(down ~65% from the 1990s peak). The final scenario was the actual effort observed through this
period, including all management interventions — the status quo 2006 (SQ2006).
Figure 3-124. Total annual effort averaged over 20 replicate simulations for the 7 scenarios.
Each scenario was replicated 20 times to encompass a range of realized behaviours of the fleet and the
results reported here are averages over the 20 replicates. The variation of the trawl model response
within scenario is shown in Figure 3-125. This variation arises entirely from random variation in the
realized effort. The variation in biomass is larger when the value approaches 50% and would be
smaller as the value approached 0% or 100%. The variation in relative biomass is fairly small because
it is an average over all model cells.
All the MSE results presented herein are subject to uncertainty, which arises from different sources.
First, given values of r and d, the MSE simulations are subject to process error due to variation in the
effort allocation. According to Figure 3-125, this is a fairly small effect. The uncertainty in the r and d
values themselves is likely to be more important. The error in r is likely to be greater than that in d, as
replication from Figure 2-45 suggests. The effect of variation in these parameters can be assessed from
Figure 3-126, which encompass the extremes of behaviour. The largest uncertainty is in the pristine
GBR Seabed Biodiversity
3-244
biomass values; the standard errors for individual cell predictions from the GLM models are typically
of the same order of magnitude as the estimates themselves.
b)
60
Relative biomass (%)
Annual Effort (000's boat days)
a)
50
40
30
65
60
55
20
1990
2000
2010
1990
2020
2000
2010
2020
Year
Year
Figure 3-125. Average and standard deviations over 20 replicates for scenarios SQ2001 (blue) and
CL/BB2001+P+RAP+BB2005 (green) of (a) effort and (b) relative biomass.
Although the absolute results from the MSEs are subject to large errors, the comparison across
scenarios is much more robust. For instance, the ranking of the scenarios by impact on the benthos is
largely unaffected by these errors. The ratios of the indicators across scenarios depended to some
extent on r and d, mainly through r/d.
While keeping in mind the uncertainty, the general pattern of relative population status (equivalent to a
hypothetical uniform pristine distribution across the region) across a range of observed depletionrecovery parameters was slow decline until ~1990, then more dramatic decline through the high effort
period of the 1990s. The falling effort in the late 1990s arrested or reversed the decline for all except
the most vulnerable depletion-recovery combinations, which would have continued to decline under
status quo 2001 (Figure 3-126). Given all of the actual management interventions that were
implemented over the period, the status quo 2006 indicates recovery trends for the most vulnerable
fauna, while the least vulnerable recovered almost completely. Individually, each intervention made
varying contributions to the overall response: the 2001 low-effort areas closure made almost no
contribution; the 2001 buy-back contributed about half of the recovery response; the progressive
penalty contributed about half to most of the remainder, depending on vulnerability (high to low,
respectively); the RAP re-zoning made some contribution, particularly in the case of higher
vulnerability fauna, and the 2005 buy-back lead to a slight additional improvement.
a)
b)
Figure 3-126. Relative biomass histories for all scenarios for two widely different vulnerability types: (left) a
highly resilient taxon (r, d) = (0.7, 0.1); (right) a highly vulnerable taxon (r, d) = (0.1, 0.44).
GBR Seabed Biodiversity
3-245
After combining these relative status results with the newly available absolute abundance distribution
predictions, estimates of regional absolute population status were possible. Thirty-eight cases were
examined, including: all mapped species that could be matched with the previous trawl recovery
project (Pitcher et al. 2004); all mapped species whose individual trawl exposure was examined and/or
had a significant trawl effort covariate and could be matched with a morpho-typically similar
recovery-project species; and all major invertebrate classes for which impact rate and recovery had
been estimated. The overall patterns of individual responses were similar to the general case outlined
above. That is, by 2000 almost all taxa had arrested or reversed the declines of the early-mid 1990s, all
taxa responded positively to the management interventions of 2001–2005 with the 2001 buy-back
contributing about half of the recovery response and the progressive penalty contributing most of the
remainder (see Figure 3-127 and Figure 3-128).
However, differences were apparent because different species were distributed differently in relation
to trawl effort and closed areas, as well as differing in estimates of their depletion and recovery
parameters. The average lowest population status for sessile species, prior to these management
interventions, was about 83% of pristine (range ~50% to 96%) and the average projection for 2025
with all current interventions in place was about 89% of pristine (range ~57% to 98%). Low points for
mobile species ranged from ~83% to 96% (average ~88%) and projections for 2025 ranged from
~93% to 98% (average ~95%) (see Table 3-61).
Figure 3-127. Average density of genus- and higher-level taxa in 2025 under each scenario.
GBR Seabed Biodiversity
3-246
Figure 3-128. Average density of individual species in 2025 under each scenario.
Table 3-61. Lowest historical (pre-2001) percentage relative biomass and final relative biomass in 2025 under
scenarios SQ2001 and SQ2006 for (left) species- and genus-level taxa and (right) coarse-level taxa.
Species/Genus
Alertigorgia orientalis
Carijoa sp1
Dendronephthya spp
Dichotella gemmacea
Dichotella sp1
Echinogorgia sp3
Echinogorgia sp5
Euplexaura sp6
Hippospongia elastica
Ianthella quadrangulata
Iciligorgia sp1
Ircinia 1255
Ircinia 2710
Ircinia spp
Junceella juncea
Junceella sp2
Melithaea sp2
Mopsella sp1
Mopsella sp2
Subergorgia suberosa
Echinogorgia
Solenocaulon
Turbinaria
lowest
88.5
74.2
95.8
95.8
93.8
93.3
86.9
65.0
60.7
87.2
78.1
49.6
59.8
66.8
93.6
94.0
90.6
83.2
83.1
74.1
89.8
89.4
73.3
SQ’01
90.8
77.3
96.6
96.7
95.1
94.8
89.5
67.3
59.1
89.1
82.0
48.6
58.2
65.7
94.9
95.3
91.7
84.8
84.5
69.2
91.9
90.9
77.3
SQ’06
95.8
85.6
98.4
98.3
97.5
97.5
94.7
77.0
67.9
94.4
89.6
57.1
67.1
73.4
97.3
97.6
95.2
90.2
90.1
76.6
96.0
95.5
87.7
OTU
Alcyonacea
Ascidiacea
Asteroidea
Bivalvia
Bryozoa
Crinoidea
Crustacea
Echinoidea
Gastropoda
Holothuroidea
Hydrozoa
Nephtheidae
Ophiuroidea
Porifera
Scleractinia
lowest
79.7
71.8
90.7
83.6
92.1
92.0
83.0
89.2
87.1
85.3
95.1
93.8
95.6
93.6
90.3
SQ’01
82.3
70.3
92.5
85.7
93.4
93.2
85.1
90.8
89.2
87.1
95.9
95.1
96.5
94.7
92.3
SQ’06
89.4
77.3
96.5
92.9
96.8
96.6
92.5
95.5
94.6
93.7
98.1
97.6
98.3
97.6
96.5
GBR Seabed Biodiversity
3-247
For six of the 38 taxa examined, the reducing effort in the late 1990s was not sufficient to arrest or
reverse the population decline projected for the preceding period, and the status quo 2001 would have
seen these decline even further (Figure 3-129 and Figure 3-130). These taxa included two species of
the sponge genus Ircinia and other unidentified Ircinia, the sponge Hippospongia elastica, the
gorgonian Subergorgia suberosa, and lumped ascidians. While this result for ascidians may be an
artefact of the pattern of their recruitment into the Trawl Recovery experiment, which substantially
under-estimated their recovery potential (Pitcher et al. 2004), the result for the others was likely to be
realistic. The sponges responded positively to the 2001 buy-back, and to the penalties by a similar
amount, and to the re-zoning by a similar amount again; the reponse to the 2005 buy-back was
imperceptible. Ircinia sp.1255 had 46% of its biomass in GU, 27% in trawled grids and 33% exposed
to effort; having an estimated catchability of 0.23, its annual incidental bycatch would be about 7%
(Section 3.7.2). Ircinia sp.2710 had 43% of its biomass in GU, 23% in trawled grids and 24% exposed
to effort; having an estimated catchability of 0.10, its annual incidental bycatch would be about 2%.
Hippospongia elastica was less abundant, and had 42% of its biomass in GU, 22% in trawled grids
and 20% exposed to effort; having an estimated catchability of 0.14, its annual incidental bycatch
would be about 3%. Subergorgia suberosa was projected to remain approximately static under the
2001 buy-back, but the subsequent penalties improved that with a projected positive response, and the
re-zoning and the response to the 2005 buy-back was imperceptible. Subergorgia suberosa had 32%
of its biomass in GU, 16% in trawled grids and 11% exposed to effort, and, with an estimated
catchability of 0.10, its annual incidental bycatch would be about 1%.
The most exposed gorgonian modelled was Alertigorgia orientalis, with 50% of its biomass in GU,
27% in trawled grids and 29% exposed to effort; with an estimated catchability of 0.09, its annual
incidental bycatch would be about 3% (Section 3.7.2). This species showed the same general pattern
of positive response to the series of management interventions, and under status quo 2006
management was projected to reach close to pre-WHA abundance by 2025 (Figure 3-129).
Several species examined by the trawl scenario model had negative trawl effort terms in the
biophysical modelling. The most negative trawl effect for a species modelled (–36%) was for the
gorgonian soft coral Carijoa sp1; nevertheless, this species responded positively to the series of
management interventions and under status quo 2006 management was projected to reach >90% of
pre-WHA (~85% of pristine) abundance by 2025 (Figure 3-129). The current exposure of this species
was 25% of its biomass in GU, 5% in trawled grids and 3% exposed to effort, and with an estimated
catchability of 0.15, its annual incidental bycatch would be <1% (Section 3.7.2). The next most
negative trawl effect for a species modelled (–28%) was for the gorgonian Euplexaura sp6; this
species also responded positively to the series of management interventions and under status quo 2006
management was projected to reach ~90% of pre-WHA (~77% of pristine) abundance by 2025 (Figure
3-129). The current exposure of this species was 36% of its biomass in GU, 15% in trawled grids and
9% exposed to effort, and with an estimated catchability of 0.14, its annual incidental bycatch would
be about 1% (Section 3.7.2). Other modelled species having negative (or possible) trawl effects
included Echinogorgia sp5 (–22%, ns), Iciligorgia sp1 (–20%), Junceella sp2 (-16% ns),
Echinogorgia sp3 (–12%, ns), Mopsella sp2 (–10%, ns) and Dendronephthya spp (–2%). Again, each
of these species responded positively to the series of management interventions, and under status quo
2006 management were projected to reach 90%–98% of pristine abundance by 2025 (Figure 3-129,
Table 3-61) — and each had low to very low exposure to current effort.
For a few species, e.g. the sea whip Junceella juncea, the sponge Ianthella quadrangulata and the
coral Turbinaria, the positive effect of the 2001 buy-back and subsequent penalties was slightly
greater than the additional re-zoning (Figure 3-129). This could be a result of displacement of effort
out of newly closed areas, increasing effort in areas where these species were distributed. The effects,
however, were very slight (<1%) and appeared to be rectified by the additional buyback.
Nine other species examined by the Effects of Trawling Recovery Project were too infrequent in the
GBR Seabed Biodiversity samples for biophysical distribution modelling, though they were observed
during towed video transects typically on hard ground not likely to be exposed to trawling.
GBR Seabed Biodiversity
3-248
Figure 3-129. Time histories since 1990 of mean density 20 individual species under all scenarios.
Figure 3-130. Time histories since 1990 of mean density of 18 genus- and higher-level taxa under all scenarios.
GBR Seabed Biodiversity
3-249
The common consistent result from the trawl management scenario modelling was that the
omnipresent depletion trends of structural epibenthos up until the late 1990s all appear to have been
arrested and reversed by the series of management interventions implemented between 2000 and 2005.
The 2001 buyback and the subsequent progressive penalties appeared to make the biggest positive
contributions. For ~7 of 38 representative taxa modelled, the rezoning made an observable additional
positive contribution, and for a similar number of species a slight contribution of the 2005 buy-back
was also observable.
GBR Seabed Biodiversity
4-250
4. DISCUSSION
The GBR seabed is a complex mix of physical environments. The component species and biological
assemblages have been observed to respond significantly, though in different ways, to the multiple
interacting physical gradients and few of these gradients have simple trends in 2-dimensional space.
Much of the prior knowledge about seabed biodiversity has been sourced from the Central
(Townsville) Section of the GBR where there are a series of relatively linear cross-shelf coastal to
offshore gradients. Along with a gradient of progressively increasing depth, there is an inner shelf
prism of terrigenous sandy-mud sediments that extends 15–20 km offshore to water depths of 20–22
m. Further offshore, the GBR lagoon in depths ~22–40 m has little terrigenous sediment but a thin
veneer of mixed shelly, muddy-sand overlying weathered Pleistocene clay, which can be exposed by
cyclones to form outcrops on the seabed. Amongst the mid-shelf and outer-shelf reefs, the seabed (40–
80 m) has virtually no terrigenous sediment, but is covered thinly by shelly biogenic carbonate sand;
here, old Pleistocene reef platforms are the foundations on which modern coral reefs emerge
(Larcombe & Carter 2004). With these geological patterns, there is a gradient from high turbidity to
clear water. The mid/outer shelf reef matrix off Townsville is relatively open and this permeability
allows influx of the EAC oceanic water and induced upwellings, which penetrate as far as the midshelf (Wolanski 1994). The south-easterly trade winds drive a northward flowing coastal boundary
layer that limits mixing of nearshore and offshore waters (Brinkman et al. 2002). As a result, there are
cross shelf gradients in bottom water attributes coastal to offshore: warm to cooler temperature (high
to low variability), low to high salinity (high to low variability), high to low oxygen (low to high
variability), and low to high nutrients (low to high variability). Further, tidal currents are weak though
most of the area. This has lead to an established view of fixed across shelf zoning of the biota
(discussed further in sections below).
However, these simple cross-shelf physical environment patterns largely apply only to the Townsville
vicinity from about Cape Upstart to about Hinchinbrook Is. The physical covariates collated by the
project showed that elsewhere, the physical environment does generally change more quickly in the
cross-shelf direction than the along-shelf direction, but in ways that may be completely different to
that in the Townsville vicinity. For example, about a third of the coast from about Mackay south is
sandy not muddy, as is the far northern coast from Shelbourne Bay north (as well as a number of other
locations). Conversely, much of the mid-shelf from about Mackay south is muddy not sandy, as is the
far northern inner/mid-shelf from about Shelbourne Bay north; the outer shelf from about Innisfail to
Cooktown has significant areas of high (carbonate) mud fraction. The Capricornia section is almost
entirely sand; terrestrial silica sand along the coast and across the shelf in the south; carbonate to the
northeast. While coastal areas are always shallow, much of the outer shelf north of about Cooktown is
about as shallow as the inner and mid shelf, as is the Swains even though it is the most offshore area in
the region. In the southern GBR, the mid-shelf is the deepest (Capricorn Channel). In much of the
southern GBR, and far northern/Torres Strait, extreme tidal currents create forces on the seabed in
both inshore and offshore areas that lead to sediment scoured and epibenthic habitats that rarely occur
in the Townsville and Cairns Sections. The ribbon reefs extending north from about Cooktown to
Torres Strait limit exchange and upwelling of cooler, saline, nutrient rich water onto the outer shelf; in
the southern GBR, these occur well into the Capricorn Channel. These different physical environment
patterns elsewhere contribute to the complexity in patterns of seabed biota observed by this project.
It has been noted previously that there are a wide range of inter-reef habitats dominating in different
regions of the GBRMP due to varying riverine inputs, tides, currents and upwellings, seasonal winds,
waves, and cyclonic events, with different combinations of these forces governing the topography,
grain size and composition of sediments, the chemical properties of overlying waters (Larcombe &
Carter 2004; Porter-Smith et al. 2004) and therefore, the nature of seabed assemblages and their
dynamics. Local influences, such as tidal jetting of nutrients, facilitate the development of substantial
Halimeda algal banks 15-20 m thick inside the ribbon reef passages of the northern GBR (Drew
2001). Porter-Smith et al. (2004) noted that in the macrotidal areas of Broad Sound and Shoalwater
Bay, tidal currents were the dominant force influencing the mobility and grain size properties and
contrasted with the rest of the GBRMP. The bifurcation of the South Equatorial Current (SEC) against
GBR Seabed Biodiversity
4-251
the outer reef (Wolanski 1994) between about Lizard Is and Cairns, which produces the EAC to the
south and the NW Coral Sea gyre to the north, can influence the dispersal of species with oceanic
larvae (Pitcher et al. 2005).
There are also latitudinal differences in flushing rates and the amplitude of seasonal variation in sea
surface temperature (SST) and salinity. Hancock et al. (in Press) found that inner lagoon diffusivity
was about 2.5 times higher in the central section compared to the northern, so that water within 20 km
of coast is flushed with outer lagoon water on a time scale of 18-45 days, with greater flushing times
in the north. Salinities in the southern lagoon are significantly higher than those in the central and
northern sections, and seasonal variation is lower. Summer SST are ~2-3°C lower in the region south
of Bowen compared to the far north, and in winter a relatively cold coastal water body forms there
(Condie and Dunn 2006).
4.1. BRUVS SPECIES MODELS, CHARACTERIZATION & PREDICTION
(M Cappo, G De’Ath)
4.1.1. BRUVS Fish species
Only 50 of the 347 species recorded by BRUVS occurred at 7% or more of the sampling sites.
Univariate biophysical models of this subset produced many erroneous (negative) predictions of
abundance. It was decided then to produce biophysical models and maps of the presence/absence
(occurrence) of single species in relation to the environmental covariates. The occurrence of only 25
fish species could be predicted with errors of 20% or less, and a shortlist of 20 environmental
covariates were useful as predictors.
The “nuisance” temporal variables such as moon phase, season and time of day had no effect on
average rates of prediction, so they were dropped from the best models. This implied the biophysical
maps of occurrence in the BRUVS dataset do not need to be adjusted for these temporal variables.
The spatial location (across, along, depth) and mud, gravel and carbonate content of sediments were
the top six predictors of species occurrence. There were a variety of relationships between these
physical parameters and the occurrence of particular species. The presence of some species was best
explained by by a single variable, including Scomberomorus queenslandicus found inshore,
Pentapodus paradiseus influenced by gravely sediments, and Nemipterus furcosus and Alepes apercna
found in warmer temperatures. The presence of other species was influenced most most by a
combination of spatial and environmental variables. Nemipterus hexodon was one of the most
predictable species, occurring nearshore in sediments with high mud content. In contrast, Pentapodus
nagasakiensis was found offshore in sediments with high carbonate. Within the same family N.
theodorei was influenced most by deep water and high salinity. This species was not seen in the far
north. A number of species were influenced most by environmental variables. The economically
important Choerodon venustus occurred most in areas of higher current with sediments containing
high carbonate. Parapercis nebulosa also occurred in higher current, over sediments with low mud
content.
The trawl effort index explained a moderate proportion of the variability in occurrence of only a single
species (Nemipterus peronii), but interpreting such relationships was very difficult because trawling
occurs infrequently, or not at all, over most of the GBRMP and in high levels in some areas, such as
Cape Flattery. Thus the 10th percentile in the trawl effort index had much leverage on the response by
N. peronii. The relationship between the species responses and the environmental gradients in the
GBRMP are discussed below with the characterisation of species groups.
4.1.2. BRUVS Fish Assemblages
Significant differences have been reported in the distribution and abundance of a range of faunal and
floral groups in the GBRMP along the strong cross-shelf gradients readily measurable in salinity,
nutrient input, water clarity and exposure to prevailing wind and waves with increasing distance from
GBR Seabed Biodiversity
4-252
the coast (Drew 2001, Wilkinson & Cheshire 1988, Newman et al. 1997, Gust et al. 2001). However,
studies have incorporated the latitudinal gradient along the shelf (DeVantier et al. 2006, Fabricius &
De’ath 2001, Williams 1991), and most have been restricted to the depth limits of SCUBA diving
observations on shallow reefs. The results reported here are the first attempt at describing the patterns
in fish communities in terms of both location in the GBRMP and critical environmental covariates.
The use of boosted and multivariate regression trees provided compelling results concerning the crossshelf rise in species richness to “hotspots” about the shallow banks and shoals amongst the offshore
reef matrix, the existence of spatially-contiguous fish communities along and across the shelf of the
GBRMP, and the existence of major community boundaries near Bowen (20°S) in the south and
Princess Charlotte Bay (13.3°S) in the north. These robust patterns were detected amongst a
functionally diverse cross-section of the fish fauna by analysis of data collected with a simple,
efficient baited video technique.
The majority of species recorded by the BRUVS occurred rarely and this pattern also seems
characteristic of tropical fish faunas sampled by trawl. The widespread sampling with BRUVS
recorded a similar number of species (347) to those recorded by trawling (300 – 350) in similar
latitudes by Watson et al. (1990), Blaber et al. (1994), Wassenberg et al. (1997), and Stobutzki
(2001b). Those trawl inventories were also dominated by species that occurred rarely and in low
abundance. Stobutzki (2001b) found that 75% of species occurred in less than 10% of prawn trawls,
and Blaber et al. (1994) found that 75% of the biomass in fish trawls comprised only 8% of the species
caught. Like estuarine fish faunas (Magurran & Henderson, 2003), the vertebrates in the “inter-reef”
waters of the GBRMP probably comprise “core species”, which are persistent, abundant and
biologically associated with particular habitats, and “occasional species”, which occur infrequently in
surveys, are typically low in abundance and have different habitat requirements.
Only 50 species occurred at 7% or more of the sampling sites and the abundance of only 25 of these
species could be predicted, in terms of the 40 environmental covariates, with errors of 20% or less.
This shortlist was chosen for preparation of biophysical maps relating communities to the
environment. Multivariate trees were used within this shortlist of species to define a hierarchy of
communities constrained by their spatial and environmental values that locate them in the GBRMP.
This hierarchical approach identified groups of the 25 predictable species that co-occur at varying
spatial scales to form communities.
A number of multivariate trees were assembled using different combinations of spatial and
environmental variables. The simplest, most easily interpreted model chosen to produce biophysical
maps used only position across and along the shelf as explanatory variables for the 25 most predictable
species. The broad patterns identified by this tree largely coincide with the analysis of the entire
species list by Cappo et al. (subm.) in a study that aimed to explain (not predict or map) communities
in terms of location and depth.
The across and along tree produced spatially contiguous communities occurring around major faunal
breaks in the nearshore half and middle of the lagoon, and alongshore. Latitudinal variation was
greatest in the inner half of the shelf, where Bowen and Princess Charlotte Bay separated inner and
outer shelf groups. The offshore communities were latitudinally more extensive showing that outershelf deep communities were more similar amongst latitudes than to inner-shelf communities at the
same latitude and vice versa. These trends were also reported by Williams (1991) for reef fish
communities on outer slopes of outer-shelf and mid-shelf coral reefs, with mid-shelf reefs in the far
north being more similar to nearshore reefs elsewhere than to mid-shelf reefs at more southerly
latitudes.
Cross-shelf gradients in demersal fish communities have been reported elsewhere in the southern
Indo-Pacific. A “nearshore” group of sites (<24 m depth) was distinguished from a “mid-shelf” (outer
lagoon 35-42 m depth) and an “inter-reef” group (mid-shelf reef matrix 43-56 m) in catches by prawn
trawl in the central GBRMP (Watson et al. 1990). That study concluded the composition of the 'interreef' fauna remained strikingly uniform regardless of proximity (~0.5 to 10 km) to coral reef
formations. In the Gulf of Carpentaria, Blaber et al. (1994) distinguished six main site groups and 15
fish community groups in fish trawls, correlated with depth but not with sediment type, salinity,
temperature or turbidity. Letourneur et al. (1998) concluded that the effect of sediment type in shaping
reef fish communities in lagoon waters was confounded by cross-shelf position, which acted as an
GBR Seabed Biodiversity
4-253
easily measured surrogate for the gradient from terrigenous to oceanic influences in New Caledonia.
At broader scales, Ramm et al. (1990) found that a bycatch fauna from prawn trawls samples clustered
along geographic and bathymetric gradients, forming distinct western and eastern groups separated
near132°E, either side of the 30 metre isobath. This longitudinal boundary separates the faunistic
provinces of the Timor Sea to the west and the Arafura Sea to the east.
Reviews have concluded that sediment type, water clarity, seabed topography, the nature of epibenthic
communities and thermal stratification all shape the composition of tropical, demersal fish
communities (Longhurst & Pauly 1987, Sainsbury et al. 1997, Lowe-McConnell 1987). In the tropical
western Atlantic, Lowe-McConnell (1987) characterised a cross-shelf gradient from a “brown water”
zone (ariid catfishes, dasyatid rays) over mud, to a “golden fish” zone (sciaenid croakers) over
sandy/mud, to a “silver fish” zone (carangid jacks, haemulid grunts) in “green water” (40-60 m deep),
above a “red fish” zone (lutjanid snappers) over hard sand/rock in “blue” oceanic waters (~100 m).
Similar patterns were reported in the eastern Atlantic, with the additional influence of a strong
thermocline causing sub-tropical sparids to dominate in the cooler waters under the thermocline over
sand, rock and Holocene reef edges (Fager & Longhurst 1968, Lowe-McConnell 1987).
We found that there were not strict cross-shelf differences in the occurrence of different families and
that single families often contained a number of species that characterised different communities.
Ubiquitous families such as the nemipterid threadfin breams, monacanthid file fishes, carcharhinid
requiem sharks and tetraodontid pufferfish had representatives in both inshore and offshore, deep and
shallow communities. The inshore community included many indicator species from the “small
pelagic” functional groups, such as the piscivorous Scomberomorus queenslandicus and deep-bodied
micro-invertebrate carnivores (Carangidae: Atule mate, Selaroidesleptolepis and Carangoides
coeruleopinnatus). Demersal teraponid grunters and bathysaurid lizardfishes were also characteristic
of inshore groups. These are known to inhabit soft sediments in the Indo-Pacific (Blaber et al. 1994;
Sainsbury et al. 1997). Indicator species offshore included pinguipedid grubfishes (Parapercis
xanthozona_grp 40) and labrid wrasses (Choerodon venustus, Oxycheilinus bimaculatus) thought to be
associated with more complex seafloor topography, such as reefs, rocks and rubble.
The distributions of fishes and their assemblages are likely to be shaped by variation in sedimentary
and oceanic processes and other influences that determine the wide range of seabed physical and
biological habitats dominating in different regions of the GBRMP, as described above. In turn, these
habitats may influence the recruitment, feeding success and mortality of fish communities inhabiting
them. The nearshore and mid-shelf boundaries separating the fish assemblages appear related to
sediment carbonate and grain-size composition particularly in the Central Section. The observed fish
assemblage patterns fit well with knowledge of gradients and boundaries in sedimentary processes,
water movement, and seafloor fish habitats such as erosional features, depositional banks and
vegetated meadows.
Major latitudinal boundaries in the fish assemblages may be related to circulation patterns and their
consequences. The latitudinal variation amongst the communities along the shelf, with boundaries near
Mackay, Bowen, Cape Bowling Green, Princess Charlotte Bay and Cape Direction, may be explained
by circulation patterns in a cooler, macrotidal southern region, a well-flushed central region with
deepwater seagrass beds, and a warmer, constricted northern region where Halimeda algal banks
thrive behind a dense reef matrix. The next step is to analyse these spatial groupings within a
comprehensive suite of biotic measurements from other sampling gears to determine if predictions and
explanations of the shelf-scale patterns in fish communities can be improved with knowledge of
epibenthic communities now available from the Seabed Biodiversity dataset.
4.2. SINGLE SPECIES, BIOPHYSICAL MODELS AND PREDICTION
The epibenthic sled and research trawl both sampled a highly diverse seabed biota of more than 14
phyla and >5,300 species, of which about a third were sampled by both devices, almost half were
unique to the sled, and just less than a quarter were unique to the trawl. While the Sled samples were
rich with almost 60 taxa per site on average and had greater total species richness, the Trawl samples
averaged about 87 taxa and were less variable (<1% sampled fewer than 20 species). Thus, the trawl
GBR Seabed Biodiversity
4-254
more consistently sampled local populations representatively (particularly fishes and crustaceans),
whereas the Sled sampled all other biota better, though with greater variability, perhaps due to the
greater number of sites and habitats sampled by the sled. These devices provided specimens that could
be properly identified, showed that otherwise inseparable taxa could have strikingly different
distribution patterns and revealed the enormous biodiversity of the region (even at sites where other
devices observed no biota at all). They also provided excellent benefit:cost in terms of enormous
information yield per unit investment and given the very small area sampled (~1/50000).
As is typical of biological sampling, a large proportion of these taxa occurred in only one or a very
few sites. This, and patterns of species accumulation curves, indicated that many more seabed species
remain to be discovered — despite the regional extent and number of sites sampled by the project —
and confirms the significant biodiversity of the deeper lagoon and inter-reef seabed in the GBR.
There were some clear basic patterns of species richness: structured habitats were more diverse, and
structure was provided by marine plants, epibenthic fauna, and rugose hard substrata. Key vegetated
habitats occurred in a midshelf band in the central GBR, the inner/mid-shelf in the Capricorn region,
the outer shelf in the far northern region and near the Turtle and Howick Island groups. Key epibenthic
habitats occurred in the vicinity of Broad Sound and Shoalwater Bay, approaches to Torres Strait,
inshore and offshore passages and hard ground. Topographically complex hard ground occurred
primarily among the outer shelf reef matrix as relic reef growth from eras of lower sea levels. Muddy
habitats, on the other hand, had relatively low biodiversity and tended to be dominated by smaller
fishes. The structured areas also had steeper species accumulation curves, i.e. additional sites were
more likely to yield additional species compared with muddy areas, which consequently tended to be
more homogeneous.
Compared with the total number of species sampled, relatively fewer species were considered frequent
enough for analyses (at >25 sites). Nevertheless, there were about 840 species that met this criterion.
This presented considerable computational challenges for biophysical modelling, necessitating a
consistent mechanistic approach and meant that in the available timeframe models could not be
manually customized for individual species. The analyses also produced an enormous output, of which
only summary distributional highlights can be presented, even in a large report, and future
examination of the dataset and results would provide valuable specific information.
The two-stage presence-biomass GLMs provided a robust and flexible method of selecting physical
covariates having statistical relationships with the biological data suitable for modelling biophysical
responses and predicting distributions. More than 65% of the ~840 species analysed had good models;
however, not all species could be modelled well. Up to ~10% had poor models though many of the
latter were for species that could not be weighed readily, and 11 species had no statistical relationship
with any of the physical or spatial variables. For the majority of species, i.e. those occurring at <25
sites, no analyses or modelling has been reported at this time, though point data will be available.
The biophysical modelling provided an indication of covariate importance in relation to patterns of
species abundances. The most frequent covariates were sediment grain size and carbonate
composition, followed by space, benthic irradiance, current stress, bathymetry, then bottom water
physical attributes, seasonal effects, nutrients and turbidity, and other temporal effects. Trawl effort
was selected infrequently, at about 10% and was significant in just over half these cases. Aspect was
selected least. However, frequency of covariate selection is not necessarily a direct test of performance
among the various individual physical or spatial covariates due to the high correlations between many
of them. That is, a given covariate may be parsimoniously selected though it is merely a surrogate for
a number of others with which it is correlated.
The environmental covariates have demonstrated utility for spatial prediction of the broad scale
patterns of presence for the majority of species with occurrence >25 sites, although they do not often
account for the majority of observed variation in local biomass. Other factors, including stochastic
processes such as recruitment and mortality, biological interactions, and random sampling effects
typically outweigh deterministic environmental relationships at the local site biomass scale.
The variety of biophysical responses, and hence distributions, expressed was almost as large as the
number of species analysed, and while there were numerous highly contrasting responses there was a
continuum of responses between the extremes. Some of the contrasting responses observed followed
the more frequently occurring covariates, such as ± mud or sand or gravel, shallow vs deep, low vs
GBR Seabed Biodiversity
4-255
high benthic irradiance, inshore vs offshore, northern vs southern, strong vs weak current stress, cool
vs warm temperature etc. It is important to note that taxonomy provides little guide to patterns of
distribution. For example, many similar con-generic species were observed to have highly contrasting
distributions, demonstrating the importance of species-level identifications wherever possible, when
assessing biophysical responses and physical surrogate performance, and when addressing
management applications such as fisheries risk assessments and conservation planning.
4.3. SPECIES GROUPS CHARACTERIZATION AND PREDICTION
The ~840 modelled species were clustered into 38 groups largely to facilitate manageable processing
of the trawl risk assessments rather than any particular attempt at this time to identify an objective
number of functional “communities” of interacting species; though the methods used may be useful to
facilitate initial identification of potential functional relationships between species, significant future
ecological investigation would be required. As noted in the previous section, there was almost a
continuum of distributional patterns with few strong clusters. The groups were constructed to
comprise species with highly correlated distributions, given the stated purpose, leading to a greater
number of groups than conventional applications of the method. The basic patterns of groups were
similar to those of single species and though the models were more complex, covariate importance
ranks were also similar to those of single species.
Species groups distributions were also not clearly related to taxonomic groupings, as noted above,
with closely related species occurring in different groups and most groups comprising numerous
unrelated taxa. Possible exceptions may have included some marine plants, sponges and bryozoans.
Marine plants appeared to dominate in a number of rather specific areas determined at least in part by
light availability; this and other shared constraints lead to a number of related species sharing similar
distributions. Similarly, a number of encrusting bryozoan taxa were distributed in high current
passages of the Pompey Reefs complex and a number of sponge taxa had similar distributions. Even in
these constrained taxa, however, members of the same genus, family or class may have markedly
different distributions.
4.4. SITE GROUPS CHARACTERIZATION AND PREDICTION
Due to the complex mix of physical environments described above, as well as additional variability
remaining unexplained by the available covariates, the biodiversity assemblages of the GBR seabed
are difficult to represent adequately in any single characterisation. The multitude of species respond in
different, overlapping and varying ways to the multiple interacting physical gradients, which as noted
above do not have simple trends in geographic space. Much of the prior biological sampling has been
conducted in the Central (Townsville) Section of the GBR where the gradients are among the simplest
in the region.
Seabed biological assemblages off Townsville were first sampled by Birtles and Arnold (1983 &
1988) using an epibenthic sled during a series of studies between 1977 and 1983. The shallower (<20
m) inshore muddy zone, to about 30 km offshore was characterised by low species richness of
carnivorous and deposit feeding echinoderms, molluscs, crustaceans, fishes, bryozoans and algae, and
low species evenness — i.e. a relatively low number of species was dominated by even fewer. Further
offshore, from ~30 km to the mid-shelf reef-matrix at ~80 km, the deeper (20–50 m) carbonate sand
lagoon zone was characterised by higher species richness of all faunal groups, due to increased habitat
heterogeneity with patches of harder substratum that allow a wide variety of suspension feeders, such
as sponges, ascidians, crinoids, holothurians, and bryozoans, a foothold in addition to the deposit
feeders in the sediments between the patches. On the deeper gravely outer shelf (> ~80 km) offshore
inter-reef zone the fauna was dominated by the suspension feeders (Birtles and Arnold, 1983).
Temporal sampling showed that patterns of distribution remained essentially stable over the period.
Greatest variability was apparent in the nearshore sites, due to physical instability of the sediments
GBR Seabed Biodiversity
4-256
caused by wind generated waves. Later, Watson and Geoden (1989) sampled around the Birtles and
Arnold study area with somewhat greater alongshore coverage, using commercial prawn trawl gear,
and found similar parallel zones that were stable over time. With few benthic studies conducted
elsewhere, there has tended to exist a rather fixed view of simple cross-shelf zonation in the biota.
However, as already noted, the physical environment in other areas of the GBR has different
combinations of physical environments and geographic gradients that may be completely different to
that offshore from Townsville. These different physical environment patterns elsewhere explain why
there have been reports of what had been regarded as nearshore faunas occurring in mid-shelf areas
north of Cairns and even outer shelf areas (i.e. inner edge of Swains) (e.g. Williams, 1991). The
Effects of Trawling Study in the far northern section (Poiner et al. 1998) noted that the coastal silica
sand strip had fauna more similar to mid/outer-shelf areas than to the inner/mid-shelf high mud area.
These "aberrant" patterns may now be understood more clearly with the extensive bio-physical
information provided by the Seabed Biodiversity Project — the patterns are not aberrant at all but
reflect the complex variety of environments manifest and distributed in different spatial patterns
throughout the GBR region. The biotas are located largely (but not entirely) by environment than by
cross-shelf position per se, or by latitude per se.
The biological assemblages observed by the Seabed Biodiversity Project are largely in line with the
physical covariates found to be important in the single-species modelling and hence with the complex
multidimensional physical environment patterns collated and mapped by the project and outlined at
the beginning of the discussion. That is, the patterns are consistent with gradients between high and
low mud areas, shallow and deep areas, high and low current areas, with further separations on
sediments, water chemistry and turbidity. Note that each of these variables have surrogates that could
separate assemblages, given the numerous correlations between the covariates. These broad patterns
were largely unrelated to higher level taxonomic groupings; with members of most groups of biota
occurring in most areas.
However, not all known patterns were captured within the stopping/cross-validation rules used by
most statistical splitting algorithms, largely due to the variability in the data. For example, the inshore
and offshore high current stress areas were grouped together and while they were similar in some
respects there were nevertheless differences in species occurrences, depth and turbidity; the
topographic shoal strata was not separated but was shown to differ from surrounding deeper seabed in
a previous study (Poiner et al. 1998); and vegetated habitats were not separated precisely in the
biophysical characterisation compared with the point species data.
4.5. VIDEO HABITAT CHARACTERIZATION AND PREDICTION
The physical habitats observed during towed video transect largely followed the physical environment
patterns already discussed, not surprisingly given the similarity of the attributes quantified. Substratum
was closely related to sediment grain size and carbonate, with exposed hard substratum occurring in
high current areas, and deeper rocky areas amongst the outer shelf reefs. The biological habitats
observed by video in part followed the physical substratum patterns, with bioturbated habitats
occurring in many softer sediment areas and sessile epibenthic fauna and bryozoans common in high
current hard ground areas. Vegetated habitats occurred on a variety of substrata and have constraints
related more to irradiance and bottom water attributes. While some of the densest vegetated habitats
were captured by the biophysical characterisation of video habitats, including much of the Townsville
midshelf band, the Capricorn inner shelf area and far northern Halimeda banks, they were not
separated very precisely. Furthermore, the biophysical characterisation confused intermediate density
vegetated areas with some high density areas as well as with some very low density areas. Statistical
stopping rules in conjunction with data variance contribute to this imprecise separation, and better
broad scale data on relevant bottom water attributes would likely improve the biophysical
characterisation.
A key issue with video, however, is the inability to identify most of the biota adequately. For example,
hundreds of species of algae that comprise algal habitats can be identified to only a very few distinct
genera and most to a handful of morpho-types, which blurs biophysical relationships that were
GBR Seabed Biodiversity
4-257
apparent at the species level and leads to inconsistent biophysical responses that are difficult to model
with statistical methods, and thus difficult to map with biophysical spatial modelling.
A further issue was indicated by results of the first habitat characterisation strategy (first cluster the
video biological data alone, then attempt to model their physical relationships), which had some
difficulty representing the broader distribution of habitat clusters. Whereas, the second strategy found
‘habitats’ by using the biological profile together with and at the same time as the physical
environment in which it resides. In this case, very similar biological habitats appeared in different tree
nodes suggestive of different regions of environmental space. As an example from the results, two
very similar seagrass dominated habitats (tree nodes 6 and 7, Figure 3-60, Figure 3-61,) were four
nodes apart in the tree diagram suggesting different regions in environmental space. The mapped
predictions (Figure 3-62) indicate a spatial divergence as well, with group 6 largely confined to the
southern region of the GBR but with group 7 distributed in the mid-shelf off Townsville. Note that
spatial predictors, however, played no direct part in the definition of these groups. However, the risk
this brings to the second strategy is that essentially the same habitat, split in this way, may be
interpreted as a different habitat — unless the composition is examined carefully. It is possible that if
the habitat groups are initially defined in terms of biology only (the first strategy), sites such as those
present in groups 6 and 7 may be clustered together, though this may make their prediction using the
external physical variables more difficult because they are in different regions of environmental space.
Note that the GLM method used in the single species modelling successfully modelled H. spinulosa
distribution, the key species present in these two example seagrass dominated habitat cluster types and
apparently occupying different regions of environmental space.
4.6. ACOUSTICS DISCRIMINATION AND CLASSIFICATION
Three approaches were taken to examine the performance of remotely sensed acoustic data, from a
normal-incidence 120 kHz single beam digital echo sounder, for discriminating different seabed
habitats and hence, as a surrogate for patterns in habitats and their constituent biodiversity.
(1) wavelet-based methods on angular transformed and re-sampled EY500 digital data, initially on
simple pair-wise habitat class comparisons between few sites then progressively adding more habitat
classes and sites.
(2) canonical variate-based methods on dilation-translation transformed and re-sampled EY500 digital
data, initially on the entire ground-truthed dataset including all available sites and classes, then
directed contrasts between fewer sites and fewer habitat classes.
(3) linear discriminant analysis methods on the QTC View proprietary 166 feature data, initially on the
entire ground-truthed dataset including all available sites and classes, amalgamating classes, and by a
moving window of restricted spatial and depth dimensions.
4.6.1. Wavelet Packet-Based Techniques Applied to Data in the Angular Domain (D H
Smith)
Supervised classification experiments were performed on data in the angular domain via the Local
Discriminant Basis in conjunction with Daubechies-2 filter coefficients and two different standard
classifiers for up to five seabed habitat classes of long contiguous blocks of >1,000 pings, which
tended to exclude certain habitats that were typically patchy in nature. Two-class testing on (sand,no
biohabitat) and (sand,seagrass) seabeds with a single deep (sand,no biohabitat) training set produced
very satisfactory classification results in a few pairwise site comparisons; however, addition of more
sites demonstrated a strong sensitivity to the depth of the corresponding test set component, despite
angular transformation of the data. This can be partially explained by the physics of acoustics,
resulting in a distinct depth divide separating good and poor classification results. Substitution of a
much shallower alternative training set for the (sand,no biohabitat) component produced no clear
divide with less-variable classification accuracy lying between the extremes of the about experiment.
In terms of overall mis-classification rates, the best results reached below 10% while the worst cases
GBR Seabed Biodiversity
4-258
reached just over 80% for these two-class cases. High mis-classification rates, when they did occur,
were essentially the result of sand being mis-classified as seagrass.
Highly variable performance was also recorded in a second series of two-class experiments involving
(sand,sponge garden dense) and (sand,seagrass) seabed types. Mis-classification rates for a selected
training set ranged approximately between 10 and 55%, the upper limit caused by sponge garden being
mis-classified as seagrass on many test sets. Little depth variation was present in these training and
test sets, and the variable performance has little to with the depth effect noted above.
A three-class classification test involving (sand, no biohabitat), (sand, sponge garden dense) and (sand,
seagrass) seabed types produced satisfactory classification results on a particular single training/test
set pair, with confusion matrix diagonals all exceeding 80% and a best overall mis-classification rate
of approximately 15% achieved with Linear Discriminant Analysis. Sand produced the best
classification result, followed by sponge garden and seagrass. Feature dimension reduction was also
achieved, but to a smaller extent than that obtained on the two class cases. Additional calculations with
the same training data on a collection of test sets, including all available data of >1,000 contiguous
pings for each class, indicated some significant performance differences between the Linear
Discriminant Analysis and Tree classifiers. For sponge garden and seagrass biohabitats the Linear
Discriminant Analysis was clearly superior, while little difference was observed for sand without
biohabitat. In each case a general decline in performance with depth departure between the training
and test sets was observed, however only one single training set was applied and performance with
multiple training sets would be needed to fully assess variability in performance.
Incorporating a fourth seabed habitat class, namely (sand, bioturbated), gave mixed results on a
selected single training/test set combination, with good performance recorded for the (sand, no
biohabitat) and (sand, bioturbated) classes accompanied by moderate performance for (sand, seagrass).
Principal error contributions for the latter case were due to mis-classification as (sand, sponge garden),
which in turn was poorly classified and largely mistaken as seagrass. Further calculations on a range
of different test sets showed good results from Linear Discriminant Analysis on the bioturbated class
for depth departures up to almost 20 m, beyond which above 50% accuracy was retained up to 50 m
departure. On the remaining three classes the two classifiers demonstrated clear differences, with
Linear Discriminant Analysis superior for sponge garden and inferior for sand and seagrass; whereas
as, in the experiment above, LDA had performed better than Tree on the same seagrass datasets.
Again, classification performance was observed to decline with depth differences.
For a single five-class experiment involving different substrata in the absence of biohabitat, evidence
for the concept of class merging appeared, in which different nominal classes are merged to produce a
workable reduced set of seabed classes. Specifically, this applied to coarse sand and sand substrata
types, and also to silt and soft mud, which were strongly confused by Linear Discriminant Analysis.
The fifth substrata type, namely sand waves/dunes was almost 90% correctly classified in this
experiment, with the error primarily due to mis-classification as soft mud.
In these tests with training and test set class contributions from individual sites, the Local Discriminant
Basis operating on angular transformed data was able in selected localised cases to extract features that
provided a good basis for classification of a limited number of classes by methods such as Trees or
LDA. However, as the number of sites involved in tests was increased, or as the depth departure
between sites increased, or as the number of classes of biological habitat on a fixed substratum was
increased, or as the number of substratum types was increased, classification performance declined to
unsatisfactory levels. The increased ambiguity means that relatively few bottom types can be
consistently classified and so included in a workable set of seabed classes, and then only within a
limited range of depth variation.
4.6.2. Canonical Variate Analysis of Acoustic Data (N Campbell & D Devereux)
Canonical Variate Analyses were performed on digitised ping data, depth-normalised by the dilation
method, and tested primarily with all available habitat classes and depths, but restricted to data
representing classes with >100 contiguous pings.
GBR Seabed Biodiversity
4-259
A depth problem was again apparent and was examined in some detail. Depth-related differences in
the shapes of normalised ping profiles were observed, at shallower depths (especially <25 m) the
response for the first echo is much broader than it is at greater depths. This effect is due to the physics
of echo-sounder operation and was seen to have a 1/depth relationship. Several methods to standardise
for the effect were attempted, including an empirical regression of the echo response values on
1/depth. While a physically-based correction would be useful, none exist in the literature and the issue
remains unresolved.
Analyses of data from sites that were collected close together in time, space and depth, suggested that
it may be possible to provide local separation of extremes of cover in some circumstances. However,
analyses of the larger data set over an extensive geographic area showed that the differences among
pings from the same cover class were often as large as the differences between the cover classes. Even
the site-contrast analyses for localised areas show that differences between sand sites can be greater
than differences between sand and seagrass sites. Sand classes were the most common, were highly
variable and their range of variability included almost all other habitat classes to a greater or lesser
extent. Data from almost all classes were seen to plot almost throughout the full range of variability of
the principal CVs. This demonstrated very considerable ambiguity between the acoustic signatures of
the same and different classes, even at the same depth.
The digital echo sounder system used to collect the data analysed here had a transducer with a nominal
beam angle of 10° — with greatest sensitivity < ±10° and little sensitivity between 10°–50° — as is
fairly typical of single beam echo sounders. Relatively recent analyses of multibeam data show that
much of the discrimination between cover classes occurs at angles 25°–35°, where the backscatter
response for this single-beam instrument was much less than at narrower angles. For the narrower
angles, there was some discrimination, though context from neighbouring pings is needed to improve
the reliability. Thus, in the data from this echo-sounder, while some local scale discrimination in shape
was observed, most discrimination would seem to depend on differences in the magnitudes of the
responses at narrow angles, and not from consistent differences in the broader shape between the
various cover classes.
4.6.3. Linear Discriminant Analyses of QTC View data
Linear Discriminant Analyses were performed on the QTC View proprietary 166 feature data, and
initially on the entire set of ground-truth seabed type classes, including all available sites and
substratum and biological habitat types. Subsequently, similar types of habitat classes were
amalgamated on the basis of ambiguity observed in LDA confusion matrices. The depth problem
found in other types of data and analyses was also present in the QTC data and was addressed by
partitioning the data by depth (and restricted spatial dimensions) to limit data selected for iterative
training and testing trials, with some small improvement in amalgamated classification success.
Overall, the QTC data also had low levels of success in classifying the observed seabed classes, based
upon its 166 acoustic parameters, and that success increased inversely as the number of classes was
reduced. The fewest number of substratum classes analysed was seven, yet it was clear that a reduction
to as few as 3–5 would be necessary in an attempt to raise classification success.
The best result achieved (~49%) was with a simple set of seven substratum classes (after partitioning
into 6 depth strata). The majority of the confusion involved the most frequent substratum class (the
sands), which accounted for ~70% of the observed seabed types, yet sands could be classified
correctly in significantly less than half of observed cases, with the result that most sand was classified
as either a more structured class type (gravel, rock, reef) or as mud.
It is acknowledged that the ground truth (EVENTS) data was at times also subject to misclassification
by human observers of the towed video; for example, on occasion it could be difficult to accurately
determine the sub types of soft sediments (sand, silt, mud) from video. Further, sediments can vary in
ways that affect their acoustic properties, which cannot be observed on the sediment surface. While the
reliability of the EVENTS data with respect to visually similar substrata can at times be questioned,
there was little doubt that extreme substrata (mud, sand, rocks, reef) were identified with certainty and
yet these distinct classes are substantially confused in the analyses of the QTC data.
GBR Seabed Biodiversity
4-260
It appears that the signal to noise ratio of the acoustics data is not adequate to consistently distinguish
readily observable basic habitat types. Any given class (especially sand) is so variable that its feature
attributes overlap strongly with neighbouring categories. This is exacerbated by vessel pitch and roll,
weather effects, and the relatively narrow angle (10 deg) of single beam acoustics.
4.6.4. Acoustics summary
Three different approaches were taken to examine the performance of normal-incidence single beam
echo sounder data, for discriminating different seabed habitats and hence as a surrogate for patterns in
habitats and their constituent biodiversity. Within each approach, several different techniques of
analysing the data were attempted. Typically, localised contrasts among two or a few classes would
sometimes yield satisfactory results, but inclusion of greater numbers of classes of interest over greater
spatial and depth scales increased ambiguity, and hence mis-classification rates, to unsatisfactory
levels.
This confusion can be explained at least partially by the physics of echo-sounder operation, whereby
in shallow water the length of the ping pulse is large in relation to the difference between the slant
range of the side lobes and the normal-incident range to the seabed — whereas in deep water, the slant
range difference is much larger than the pulse length. This causes a continuous change of the shape of
the returned pulse, in proportion to the factor 1/depth, that cannot be removed by the angular or
dilation transformations that normalise for different depths.
Attempts to remove the depth affect had limited success, and even restricting contrasts to similar
depths, the acoustics data showed great variability within any habitat class across the broad range of
available data. That is, habitats of the same type, as identified by video, do not consistently have the
same shape of echo-return, as characterised by a range of types of features. Further, the range of
variability in any acoustic features extracted for any given habitat overlap broadly with those extracted
for other habitats. While some merging of habitat classes is reasonable where they are ecologically
similar, such as merging mud and silt to say soft or fine substratum, or sand and coarse-sand to say
coarse substratum, it is not reasonable to merge the biohabitat seagrass with sponge gardens.
After merging and with partitioning by space and depth, only a few extreme bottom types such as
mud, rocks, reef could be consistently separated but overall errors were still in the vicinity of 50%.
This was largely as a result of sand being erroneously classified as other more extreme substratum
types. These results contrast with the claimed success of some other studies, usually of small areas,
with more limited depth range and fewer habitat types. In this study, as the scale of coverage was
increased in terms of area, number of habitat classes and depth, classification success quickly declined
to unsatisfactory levels. This means that as a general guide, ground-truthing needs to be conducted
regularly in space, by depth and by habitat type, and the number and types of classes that can be
expected to be separated are few and simple, with limited information content.
The success rate here was similar to our previous results with normal-incidence single beam echo
sounders, in studies on scales of a few 10s of sq km to a few 1,000s of sq km that used only two
simple features — the Hardness and Roughness (E1 and E2) indices acquired by a RoxAnn™ system
(e.g. Skewes et al. 1996; Long et al. 1997) or their equivalents implemented for digital data (e.g.
McLeod et al. 2007) — to classify about 4–5 seabed types with about 60% success. Even recent
analyses of swathe acoustics data have been able to separate reliably only about three seabed classes
such as soft, hard, and rough (R. Kloser pers. comm., N. Campbell pers. comm.). With the greater
range and detail of features analysed in this study, significant improvements in classification success
were expected, particularly for biological habitats. However, the greater spatial scales, depth range,
number of habitats, and variability conspired to limit the number of classes that can be separated with
any consistency in this broad-scale mapping of seabed habitat.
GBR Seabed Biodiversity
4-261
4.7. ECOLOGICAL RISK INDICATORS
Trawl exposure indicators were estimated for 38 species groups representing ~840 species that were
sufficiently abundant in sled and trawl samples for analyses. For the more exposed groups and those
groups (and species) having significant trawl effort terms, their constituent species were assessed
individually for estimated availability in GU zones, abundance in trawled areas and total exposure to
trawl effort. For the more exposed species, further information was sought on their relative catch rates
and possible BRD effects to estimate likely proportions of populations caught. Wherever available,
Susceptibility-Recovery Analysis (SRA) “recovery” rank scores (from Stobutzki et al. 2001a) and
natural mortality estimates (primarily from Zhou and Griffiths 2007) were sought to estimate,
respectively, relative sustainability risk and a quantitative sustainability indicator against reference
points — particularly for species with higher estimates of proportion caught. This represents the most
extensive and detailed quantitative species-level risk assessment conducted for any fishery in Australia
to date.
Of the ~840 species analysed, ~586 had total exposure to trawl effort of less 25%, ~218 species had
exposure to trawl effort of between 25%-50%, 23 species had exposure between 50%-75%, and 10
species had exposure >75%. Of the latter 10 species, five had estimated exposure greater than their
estimated standing stock and the most exposed of these was a key prawn target species, Penaeus
semisulcatus. The majority of highly exposed species were smaller fishes typical of tropical trawl
bycatch.
After taking into account available relative catch rate information, the potential risks were much
reduced for the majority of species. Of the ~840 species analysed, ~804 had estimates of annual catch
of less 25%, ~28 species had annual catch estimates of between 25%-50%, 4 species had catch
estimates between 50%-75%, and 1 species had catch of ~110%. These estimates apply further focus
on those species that may be at risk; however, understanding the potential for sustainability requires
information on the propensity of populations to recover — two approaches were considered.
For the first approach, SRA recovery rank scores were available for a large number of bycatch species
from previous ERA assessments in northern Australia (Stobutzki et al. 2001a). These provided a
“recovery” axis orthogonal to the catch axis already discussed and allowed further differentiation of
species risk in relation to their recovery attributes — those species orientated towards the higher
catch:lower recovery quadrant are at greater relative risk. The top 20 ranking species were listed as
having higher relative risk on the basis of their SRA recovery attributes, and while the SRA method
does not confirm whether those species are actually at sustainability risk, it would be prudent that
attention be given to their future status.
The northern Australia SRA risk assessment (Stobutzki et al. 2001a, 2002) also used a ranking method
for susceptibility, unlike the quantitative exposures estimated for the GBR; nevertheless, it is of
interest to consider any similarities between the northern Australia overall SRA risk ranking and those
for the GBR. It is possible that similar fishes have similar habitat preferences in the two regions and
those species that favour habitats also favoured by prawns are likely to be similar and hence, are
exposed to trawling. However, only two species were among the top 20 relative risk ranks for both
regions, Saurida undosqamis and Chaetodermis penicilligera, which may reflect the different
assessment methods (for the susceptibility axis) or different relative distributions of bycatch species
relative to prawns and trawling.
For the second approach, estimates of natural mortality rates had been collated for a large number of
bycatch species from a recent bycatch risk assessment in Northern Prawn Fishery (Brewer et al. 2007).
These, and estimates of natural mortality from other sources, enabled calculation of a sustainability
indicator (catch/mortality), which could be compared against reference points, based on the Schaffer
surplus production model, and allowed further differentiation of species risk in relation to their
recovery potential — those species with higher catch/mortality are at greater relative risk. This was
analogous to the Zhou & Griffiths (2007) approach — “sustainability assessment for fishing effects”
(SAFE). The benefit of this method was that it provided an absolute estimate of sustainability rather
GBR Seabed Biodiversity
4-262
than a relative rank. It was nevertheless, only an approximation and not as robust as a thorough stock
assessment.
Three species exceeded the limit reference point (set at C/M=1, ≡MSY). The Rough Flutemouth,
Fistularia petimba, with 44% of its biomass available in GU zones, an estimate of 29% caught and a
low estimated natural mortality rate M of 0.26, the sustainability indicator for this species was 1.12.
The Tufted Sole, Brachirus muelleri, with 69% of its biomass available in General Use, an estimate of
110% caught and a medium estimated natural mortality rate M of 0.98, the sustainability indicator for
this species was 1.11. Brachirus muelleri was ranked highest by SRA. The Blacktip Tripodfish,
Trixiphichthys weberi, with 56% of its biomass available in General Use, an estimate of 36% caught
and a low estimated natural mortality rate M of 0.33, the sustainability indicator for this species was
1.09. One species exceeded the first conservative reference point (set at C/M=0.8), Pomadasys
maculatus (a grunter bream) had 65% of its biomass available in General Use, an estimate of 33%
caught and a low estimated natural mortality rate M of 0.34, giving a sustainability indicator of 0.96.
Two species exceeded the second conservative reference point (set at C/M=0.6): Psettodes erumei and
Sillago burrus. Psettodes erumei (Australian Halibut) had 61% of its biomass available in General
Use, an estimate of 52% caught and a medium estimated natural mortality rate M of 0.69, giving a
sustainability indicator of 0.75. Psettodes erumei was ranked ~10 by SRA. Sillago burrus (Western
Trumpeter Whiting) had 46% of its biomass available in General Use, an estimate of 34% caught and
a medium estimated natural mortality rate M of 0.57, giving a sustainability indicator of 0.60. Future
attention should be directed at these species to clarify uncertainties and take actions to ensure their
sustainability. A further 10 species of next highest risk rank below the reference points were also listed
and included seven species also ranked highly by the SRA method — again, it would be prudent that
attention be given to the future status of these species also.
There are uncertainties to consider in these estimates of sustainability risk, not only in the modelling
of biomass distributions, but in the estimates of relative catch rates and natural mortality rates. For this
reason, a larger number of top ranking species were also listed for future attention even though they
were below conservative reference points. Where estimates of uncertainty in catchability were
available, the implications were assessed. At the higher end of the catchability uncertainty range,
Pomadasys maculatus and Psettodes erumei would step up one reference point and three additional
species Nemipterus peronii, Terapon puta and Nemipterus furcosus fell above the first C/M reference
point — T. puta was also listed by the SRA recovery rank approach.
It is notable that most of the species identified as most at risk by the quantitative C/M method did not
coincide with the top ranked species by the qualitative RSA method. Further, most of species ranked
highest by the RSA method did not appear to be at risk by the C/M method. Note also that in this
application, only the recovery rank axis of the RSA method was used against the quantitative Catch
axis. In its usual application, the RSA method uses a qualitative susceptibility rank axis instead of a
quantitative Catch axis and, given the now widespread use of the RSA and similar methods, it would
be timely to test the method against a fully quantitative approach to assess its reliability. Such an
assessment is now possible with the availability of the Seabed Biodiversity dataset.
Natural mortality estimates were not available for all species. Nevertheless, those species for which
natural mortality was unavailable had low estimates of annual catch. The highest of these were
examined and on the basis of known life history of related species, or comparative catchability, were
considered unlikely to have natural mortality rates low enough, or actual catch rates high enough, to
put them at risk.
The selection and statistical significance of the trawl effort covariate was another potential risk
indicator examined. The trawl covariate was selected for 77 of ~840 species analysed and was
significant in 55 cases (~6.4%). Of the significant effects, 17 were overall positive and 38 were overall
negative. Of the negative effects, 11 were indicative of changes of about -20% to -36% and 13 were
indicative of changes of about -10% to -20% — these included several gorgonians and sponges that
were addressed by the trawl scenario modelling, as well as a range of other fauna. It is possible that
these negative effects are indicative of the magnitude of historical impacts of trawling in the region.
Nevertheless, all species with negative coefficients had low to very low exposure to current
distribution of effort and were unlikely to be at ongoing risk. Of the 17 positive effects, 2 were
indicative of changes of >50% and 7 were indicative of changes of about 10% to 19%. These included
three target species, and in these cases the positive trawl coefficient could reflect fishers ability to
GBR Seabed Biodiversity
4-263
focus their effort rather than a real positive response to trawling (however, there is evidence for the
latter, Gribble 2003, 2004). All species showing trawl effects and having high exposure to trawl effort
had positive trawl coefficients.
The sled and trawl species data at sites were statistically separated into 16 relatively homogeneous
groups that were mapped to the GBR study area on the basis of their biophysical relationships. Areabased trawl exposure indicators were estimated for these assemblages similar to those considered for
species biomass. Most of these assemblages had very low to low exposures to trawl effort; three had
exposures between 32% and 41%, one had an exposure of 58%, and the highest was 108%. Species
having highest affinity for these assemblages were identified and a number of species were seen to
occur repeatedly across the more exposed assemblages. Their cumulative exposure was considered so
that species with higher affinities for more exposed assemblages would have a higher ranking. The top
ranking 40 species were listed and included many of the same species ranked with higher effort
exposure in the single species assessment. While it has not been established whether there are any
functional relationships among these species, consideration should be given to the potential need to
monitor the ongoing status of these species.
The data from post-processing of video transects at sites were statistically separated into 9 somewhat
homogeneous habitat mixture groups that were mapped to the region on the basis of their biophysical
relationships. Area-based trawl exposure indicators were estimated for these habitat groups, similar to
those considered for assemblages. Five of these habitats had medium exposures between 25% and
39%, the other four had low exposures between 3% and 10%. The highest exposures were for two
relatively dense but patchy seagrass and algal habitat groups: one distributed in the mid-shelf of the
central region with 39% exposure to effort; the other along much of the inner-shelf in the Capricorn
section with 34% exposure. Another (sparsely) vegetated habitat group distributed well offshore from
Townsville included patchy algae (with a little Halimeda) some bioturbation and occasional
epibenthos had 25% exposure. The most wide-spread habitat group included mostly bioturbated and
bare seabed with a little sparse algae and seagrass distributed over much of the shelf in the central and
northern sections had 34% exposure. The most barren seabed type was mostly bioturbated and
distributed in muddy areas of the inshore, midshelf and Capricorn Channel had 26% exposure. The
remaining habitat mixture groups included those with most of the Halimeda banks and epibenthic
garden biohabitats, as well as very extensive groups with mostly bare and bioturbated seabed, had low
levels of exposure. While important seagrass and algal habitats were moderately exposed, such that
their level of risk needs to be considered, there was no habitat group particularly associated with
trawling areas and trawl effort was not selected by the statistical methods as a splitting variable for
habitat groups. This suggests that trawling has not been a dramatic modifier of habitat state in the
region.
The habitat components from the video data comprising the habitat groups were examined
individually, particularly marine plants, due to their higher level of exposure. Halophila spinulosa and
like-species had 24% exposure and ovoid leaf Halophila’s had 15%, which were very similar exposure
outcomes as the modelled sample distributions for these species. The different morphotypes of algae
varied in their exposure. Crustose coralline algae was most exposed (44%), primarily off Gladstone,
though these nodules should be robust. Filamentous blue-green algae was the most extensive and had
exposure of 25%. All other morphotypes were ≤17% and most (including Halimeda’s) were <5%. The
available information suggested that catchability of marine plants is low and the single species
assessments indicated that the exposure risk of marine plant species was low. Seagrasses have
persisted in these exposed areas since earlier surveys (Rob Coles pers comm.) and it has previously
been found that trawling has not reduced the probability of seagrass occurrence and suggested that
trawling may even facilitate seagrass (Coles et al. 2006). Of the other 57 habitat components
examined, the majority had very low exposures to trawling. The exceptions included Solenocaulon (a
gorgonian), with 14% to 20% exposure, Pteroides (a sea pen), which had 15% exposure and solitary
corals, which had 12% exposure. Again, the available information suggested that catchability of these
fauna was very low. The final exception was unidentified barrel sponges (excluding Ircinia and
Xestospongia), which had 14% exposure; such sponges may have moderate catchability (based on
known species) which may place their overall risk to trawl at about 3%-10%.
GBR Seabed Biodiversity
4-264
4.8. TRAWL MANAGEMENT SCENARIO MODEL
A dynamic model was applied to assess the effects of several major management interventions,
including large scale closures and effort reductions, which were implemented between the years 2000
and 2006, on benthic fauna — particularly sessile benthic fauna that were the focus of experiments on
trawl depletion rates (Poiner et al. 1998, Burridge et al. 2003) and subsequent recovery (Pitcher et al.
2004). The dynamic model applied depletion and recovery parameters estimated from previous
experiments and annual trawl effort as provided by industry and management data, and estimated the
relative status of fauna in the region. The model was run with and without the effects on effort of each
management intervention. The relative status estimates were combined with the abundance
distributions available from this project in order to estimate the regional absolute status of these fauna.
Total trawl effort in the region grew gradually after the fishery initially commenced, but increased
rapidly in the early 1990s, before peaking in 1996/1997 and falling rapidly in the late 1990s (by
~25%) — even before implementation of the management scenarios evaluated here. The status quo
2001 model scenario maintained these effort levels through until 2525. The first intervention was the
2001 low effort areas spatial closure with the same effort levels. The second intervention was the latter
closure plus the 2001 major effort reduction buy-back, which reduced effort by a further ~30% (down
~45% from the 1990s peak). The third intervention was the latter plus an effort trading penalty system
operating over several years, which progressively reduced effort by a further ~30% again (down ~60%
from the 1990s peak). The fourth intervention added the 2004 representative areas program (RAP) rezoning of the GBR at the same effort levels. The fifth intervention added the 2005 RAP associated
buy-back, which reduced effort again by almost ~10% (down ~65% from the 1990s peak). The final
scenario was the actual effort observed throughout this period, including all management interventions
— the status quo 2006.
The general pattern of relative (to a uniform pristine distribution) population status across a range of
observed depletion-recovery parameters was slow decline until ~1990, then more dramatic decline
through the high effort period of the 1990s. The decreasing effort in the late 1990s arrested or reversed
the decline for all except the most vulnerable depletion-recovery combinations, which would have
continued to decline under status quo 2001. With all of the management interventions actually
implemented over the period, the status quo 2006 indicated recovery trends for the most vulnerable
fauna while the least vulnerable recovered. Each intervention by itself made varying contributions to
the overall response. The 2001 low effort areas closure made almost no contribution as only areas with
very little or no recorded effort were closed; nevertheless, that action would have had the effect of
preventing any possible expansion of effort into such areas. The 2001 buy-back contributed about half
of the recovery response; and the progressive penalty contributed about half to most of the remainder
depending on faunal vulnerability (high to low, respectively). The RAP re-zoning made some
contribution (particularly in the case of higher vulnerability fauna), though more limited because the
re-zoning policy was to minimize disruption to current activities, hence only a relatively small
proportion of effort was affected spatially. The 2005 buy-back lead to a slight additional improvement.
Estimates of regional absolute population status were possible by combining the relative status results
for the observed range of faunal vulnerabilities, with this project's predicted absolute abundance
distributions. The patterns of individual responses were similar to the general case outlined above: by
2000 almost all taxa had arrested or reversed the declines of the early-mid 1990s, all taxa responded
positively to the management interventions of 2001–2005 with the 2001 buy-back contributing about
half of the recovery response and the progressive penalty contributing most of the remainder.
However, differences were apparent because different species were distributed differently in relation
to trawl effort and closed areas, as well as differing in estimates of their depletion and recovery
parameters. Prior to the management interventions evaluated, the average population low-points for
sessile species were about 83% of pristine (range ~50% to 96%) and the average projection for 2025
with all current interventions in place was about 89% of pristine (range ~57% to 98%). Low points for
mobile species ranged from ~83% to 96% (average ~88%) and projections for 2025 ranged from
~93% to 98% (average ~95%). Species with lower low points tended to benefit most from the
interventions, typically by 5-15%.
GBR Seabed Biodiversity
4-265
For six of the 38 taxa examined, the falling effort in the late 1990s was not sufficient to arrest or
reverse the population decline projected for the preceding period, and the status quo 2001 would have
seen these decline even further. These taxa included two species of the sponge genus Ircinia,
unidentified Ircinia, the sponge Hippospongia elastica, and the gorgonian Subergorgia suberosa.
Ircinia sp1255 was the most exposed to trawl effort of the species modelled. Nevertheless, the
management measures implemented were predicted to produce a positive recovery response for each
— among the largest predicted. Each species had <46–32% of their biomass located in General Use
and <33–11% exposed to current effort, and estimated incidental mortality of 7–1%.
The most exposed gorgonian was Alertigorgia orientalis, with 50% of its biomass in GU, 27% in
trawled grids and 29% exposed to effort, and having an estimated catchability of 0.09 its annual
incidental bycatch would be about 3%. This species showed the same general pattern of positive
response to the series of management interventions and under status quo 2006 management was
projected to reach close to pre-WHA abundance by 2025.
Several species examined by the trawl scenario model had negative trawl effort terms in the
biophysical modelling. The most negative trawl effect (-36%) was for the gorgonian soft coral Carijoa
sp1, nevertheless, this species was predicted to respond positively to the series of management
interventions and under status quo 2006 management was projected to reach >90% of pre-WHA
abundance by 2025. The current exposure of this species was 25% of its biomass in GU, 5% in trawled
grids and 3% exposed to effort, and with an estimated catchability of 0.15, its current annual incidental
bycatch would be <1. The next most negative trawl effect (-28%) was for the gorgonian Euplexaura
sp6, nevertheless, this species also responded positively to the series of management interventions and
under status quo 2006 management was projected to reach ~90% of pre-WHA abundance by 2025.
The current exposure of this species was 36% of its biomass in GU, 19% in trawled grids and 9%
exposed to effort, and with an estimated catchability of 0.14, its current annual incidental bycatch
would be <1. Other modelled species having negative (or possible) trawl effects also responded
positively (again, among the largest predicted) to the series of management interventions and under
status quo 2006 management were projected to reach 90%–98% of pre-WHA abundance by 2025 —
and each had low to very low exposure to current effort, suggesting little future risk to their current
status.
For a few species, e.g. the sea whip Junceella juncea, the sponge Ianthella quadrangulata and the
coral Turbinaria, the positive effect of the 2001 buy-back and subsequent penalties was slightly
greater than the additional RAP re-zoning. This may be a possible result of displacement of effort out
of newly closed areas, increasing effort in areas where these species were distributed. Though the
effects were very slight (<1%), such consequences have been reported previously in the effects of
trawling literature (e.g. Fogarty and Murawski 1998; Brendan Ball, Ireland, pers comm.) and are a
consideration when planning and implementing closed-area management and/or marine protected
areas. The 2005 buyback appeared to neutralize this predicted impact of displaced effort and highlights
the importance of removing effort that is affected by closed area management.
The consistent prediction from the trawl management scenario modelling was that the ubiquitous
depletion trends of structural epibenthos up until the late 1990s have all been arrested and reversed by
the series of management interventions implemented between 2000 and 2005. The 2001 buyback and
the subsequent progressive penalties appeared to make the biggest positive contributions. For ~7 of 38
representative taxa modelled, the rezoning made a small additional positive contribution, and for a
similar number of species the contribution of the 2005 buy-back was also discernable.
GBR Seabed Biodiversity
5-266
5. BENEFITS
This project has produced comprehensive new scientific knowledge of seabed habitats, biodiversity,
and bycatch — including species new to science — in the GBR region, and delivered to the highest
'High Priority' research areas identified by the Biological Diversity Advisory Council as well as the
“areas of research of national importance” (Biodiversity Research: Australia’s Priorities — a
Discussion Paper. Environment Australia 2000).
This knowledge-base has benchmarked the current status of the region’s assemblages and will raise
the level of stakeholder knowledge of the nature and status of the region’s ecosystems. The outputs are
already facilitating assessment of spatial management in the region and will benefit future planning
and regional ecosystem management, including a basis for assessing issues such as climate change.
The outputs have enabled managers and stakeholders to identify the likely extent of past and current
impacts of trawl fishery effort, as well as the environmental benefits of recent measures introduced in
the region. If needed, the project has also provided the basis for evaluation of any future strategies to
minimise identified impacts and further improve the environmental sustainability of the fishery. These
assessments have provided a quantitative regional context that will benefit managers needing to
respond effectively to industry and community concerns and achieve an objective balance between the
pressures of exploitation and needs for conservation in a multiple-use environment. The community
will be informed and benefit from independent information on the environmental sustainability of
trawling.
Further benefits of this project’s outputs to the Queensland trawl fishery, its managers and the
community include delivery of quantitative ecological risk and sustainability indicators, and lists of
species potentially at risk, for responding to environmental assessment under the EPBC Act, as well as
other State and Commonwealth fishery and environmental legislation, and national ESD reporting. It
can be expected that these outputs, in combination with appropriate management and industry
responses, will likely lead to positive assessment and exemption from export controls status.
GBR Seabed Biodiversity
6-267
6. FURTHER DEVELOPMENT
In order to fully adopt the outputs of the research, it will be necessary for project staff to present the
results to appropriate fisheries/management committees, collaborate with fisheries/management staff
and contribute as requested to ongoing revision of management arrangements regarding meeting the
requirements of EPBC assessment, including strategies for the species identified to be at sustainability
risk in this project. Some aspects of these activities may require additional support.
Presentations to marine park managers and committees, and collaboration with marine park
managers/staff would contribute to adoption of results for zoning assessment and ongoing marine park
planning arrangements with respect to meeting WHA obligations. Some aspects of these activities are
being supported by the M&TSRF Project 1.1.1.
Further dissemination of results to other research providers, engaged in providing similar kinds of
fishery management and marine planning outputs, could be achieved by presentations/contributions to
relevant policy and/or scientific workshop/forums.
Broader dissemination of results will be achieved by seminars at scientific conferences, by scientific
publications, through availability of data via OBIS, and potentially by relocation and further
development of the project's former CRC Reef website with the addition of site data, images, video,
and maps.
Further research is warranted to address key uncertainties in the risk assessments, such as catchability
and natural mortality rates, particularly for higher risk species and those with higher exposures. The
benefits of such research are likely to be widely applicable because similar suites of species are likely
to be present in the trawled grounds of most tropical prawn fisheries due to their habitat preferences.
Ecological Risk Assessments for fisheries of type "Likelihood and Consequence" and SRA (or
"Productivity Susceptibility Analysis") (Level 1 & 2 respectively, in Hobday et al. 2006) are now
being conducted widely in many fisheries in Australia. However, the hybrid SRA method examined
herein produced results that had little in common with the much more quantitative C/M approach also
applied, which raises concerns about the reliability of the qualitative approaches. Similar concerns
were raised by Griffiths et al. (2006), which lead them to develop the quantitative "SAFE" approach
(Zhou and Griffiths 2007). The reliability of the Level 1 and 2 methods has never been fully
benchmarked, largely because of the lack of a suitable test bed, but now the GBR Seabed Biodiversity
species distribution dataset provides a powerful opportunity to assess of the performance of these
methods.
It is widely held that physical environmental variables, or distribution patterns of a broadly sampled
taxonomic group (e.g. fishes), are useful surrogates for the distribution patterns of biodiversity more
generally for the purposes of regional marine planning. While environmental variables had utility for
spatial prediction of many species herein, the more general inter-regional utility of physical variables
has not been tested and neither have cross-taxonomic patterns, or the spatial scales at which they may
or may not be effective. Again, the GBR Seabed Biodiversity dataset provides a powerful opportunity
to assess of the performance of these as surrogates for application in marine planning in other regions.
Aspects of such an assessment are being supported by the CERF National Marine Biodiversity Hub.
The project was unable to complete sorting and identification of all samples, with the resources and
timeframe available to the project, even though considerable extra resources were applied. The
samples that remain unsorted include: annelid, ascidian, crinoid, and hydroid samples from both the
epibenthic sled and scientific trawl, and all marine plants sampled by the trawl. Completion of sorting
these samples and further taxonomic work to move beyond the macroscopic OTU identifications
possible within the scope of the project would provide full utilization of the samples and specimens
and deliver additional value to science and end-users.
Similarly, 140 sites were videoed by towed camera but were too rough for the epibenthic sled or trawl.
While the habitat for these sites was characterised from the video, there is essentially no species
information for these sites comparable with that available from the sled or trawl, which has limited the
GBR Seabed Biodiversity
6-268
project's ability to model distributions of structural habitat species that were the subject of the Trawl
Scenario Modelling completed herein, and possibly underestimated their population sizes in areas
inaccessible to industry. Quantification of species from the available video and digital stills photos
would provide information on the abundance of visible species in areas not sampled by other methods,
as well as further develop the non-extractive methods.
This project has developed and applied population level sustainability indicators, which have
identified species potentially at risk. Nevertheless, with the increasing requirement for ecosystembased management, there is a need to develop condition and trend and vulnerability indicators for
seabed communities and ecosystems — not only in relation to fisheries but also other issues such as
climate change. A wide range of potential indicators have been suggested (e.g. production/biomass,
trophic indices, functional redundancy, diversity, size ratios, size spectra, dominance, habitat
complexity and fragmentation, susceptibility and productivity. Fulton et al. 2004) — the extensive
Seabed Biodiversity sample collections and dataset provide an opportunity to research and examine
such indicators for the region.
GBR Seabed Biodiversity
7-269
7. ACHIEVEMENT OF OUTCOMES
The project has produced all of the outputs as originally proposed. Preliminary outputs were presented
during the course of the project and team members contributed to management/industry activities such
as bycatch risk assessments, assessments of trawl plan targets and monitoring strategies. The nature of
this project meant that the final results could be delivered only after the complete dataset had been
analysed and synthesized, towards the end of the project. With this report and the delivery of the final
outputs, and activities outlined in Section 6, it is largely from this time forward that the anticipated
outcomes may be achieved. The planned potential industry, management and stakeholder outcomes
include:
•
•
•
•
•
Raising the level of stakeholder knowledge of the status of the region’s ecosystems, facilitating
development and improvement of regional ecosystem management plans.
Progress has been made through milestone reports, numerous presentations to management
industry community and scientific audiences and delivery of preliminary results and images on
the Project's website. Further dissemination activities are planned.
Objective information on which stakeholders can base consultation with respect to:
- reasonable use of the region that maintains the ecosystem, bycatch and benthos species
- development and implementation of management plans leading to an ecologically sustainable
fishery acceptable to stakeholders, sustainability of the seabed environment and future planning.
Again, progress has been made through reports, presentations, delivery of preliminary results
and further activities are planned.
Assessment of the current Trawl Plan targets by estimating (with specifiable uncertainty)
performance against the 40% reduction in bycatch and 25% reduction in benthos, as required
under State legislation to meet environmental sustainability objectives.
Team members have contributed to management/industry assessments of the trawl plan targets
and bycatch risk assessments. The 40% and 25% reductions were considered largely with
respect to reductions in trawl effort; the outputs from this project have provided an assessment
of their likely sustainability.
Facilitation of stakeholder development of reliable and widely accepted operational ecological
risk/sustainability indicators for identification of marine species at risk, including both bycatch
and seabed benthos species, filtering of low risk species and identifying high risk species that
need further management or information, as required under Commonwealth legislation to meet
environmental sustainability objectives. This will address a DEW condition that the fishery
conduct a risk assessment and develop biologically meaningful target reference points for high
risk species within 3 years. Significant progress against these criteria is required when the WTO
is reviewed 3 yrs after its initial approval. The WTO is conditional on demonstrating adequate
performance.
Ecological risk/sustainability indicators, with biologically reference points, have been
developed with management and industry involvement. These outputs have contributed to an
DEW condition on the WTO for this fishery.
Assessment of the implications for sustainability of recent management changes (including the
new GBRMPA RAP zoning changes due to be implemented in 2004) —current environmental
targets, risk/sustainability indicators, and MSE modelling will be estimated both with/without,
before/after recent management changes.
These assessments have demonstrated that the suite of recent management changes have had
positive implications for sustainability of the fishery.
•
•
GBR Seabed Biodiversity
7-270
Successful review of the adequacy of the current Qld East Coast Trawl Management Plan
(1999) by provision of critical information. A key component of the review will be evaluation of
the adequacy of the current suite of input controls in relation to ensuring the negative impacts of
trawl on bycatch and benthos are maintained within acceptable limits. The TMP review will
begin by Nov 2004 and completed by Nov 2006. Outputs from this project will help develop
relevant and measurable environmental indicators to be incorporated into an improved Plan.
The project outputs provide indicators of the level of impact under the current management
arrangements and have provided biological reference points. Further activities are planned with
respect to revision of the TMP.
Ability to evaluate alternative management strategies that may in future be needed to meet State
and Commonwealth environmental sustainability legislation, in a MSE context that would
estimate the outcomes for the environment and for the fishery for each option – and thus
contribute to decision making by Managers and industry.
This capability has been demonstrated by assessments of the implications of recent management
changes and can be available for evaluating future alternative management strategies.
GBR Seabed Biodiversity
8-271
8. CONCLUSIONS
The outputs from the project have delivered the "Form of Results" as originally proposed and as
mapped against the objectives. These are outlined below:
The project proposed and has produced a library of videos of the seabed habitat types of the GBR
shelf, including the QECTF in the region. Real-time on-vessel characterisation along video transect
recognised 9 substratum types, 24 broad biological habitat types and 14 class-level animal events.
Post-processing of randomised frames of these videos in the laboratory recognised 11 physical bottom
types, 30 sediment types, and 114 biological habitat component types.
The project proposed and has produced an inventory of the benthos, bycatch and fish species of the
GBR shelf, including the QECTF in the region, with catalogued museum voucher specimens to
authenticate records. In total, more than 5,300 taxa were identified, many of which are new species,
and further taxonomic work will identify additional species.
The project proposed and has produced a database of the distribution and abundance of seabed benthos
species and bycatch species at each sampling station. In total there are more than 140,000 records of
species weight and count at sampled sites. This dataset was the basis for the biophysical modelling and
risk assessment.
The project proposed and developed predictive models of bio-physical relationships between benthos
and bycatch species, seabed assemblages and communities and their physical environment. About 850
species have been successfully modelled in this way, and in the process the key environmental
variables likely to be important in structuring biotic distributions have been identified and may be
useful as surrogates.
The project proposed and has produced maps of the distribution and abundance of benthos and
bycatch species, and their assemblages, based on the biophysical models, giving full coverage of the
GBR shelf, including the QECTF in the region. Many preliminary maps were provided to trawl
managers as the project progressed, and this report and subsequent dissemination activities will deliver
the final maps to managers and industry to facilitate any further development of strategies that
minimise effects of trawling on habitats.
The project proposed and has produced estimates of the large scale effects of trawling on benthos and
bycatch quantified by measuring and analysing their abundances across a range of trawl effort
intensities, within and outside trawl grounds in the GBR, while taking into account habitat differences.
The Trawl Effort covariate was significant in 55 of ~840 species analysed (6.4%); little more than
expected by chance and suggesting that trawling does not have a strong influence on overall seabed
distribution patterns. Of these significant responses, 38 were negative and most effects were small; 6
species had significant moderate negative change in biomass of >25%–36%. Seventeen significant
responses were positive and again most were small; one species had a significant moderate positive
change in biomass of ~50% and one other species had a significant large positive change in biomass
estimated at ~96%.
The project proposed and has produced quantitative sustainability risk indicators for bycatch species,
developed from this projects’ estimates of the proportion of bycatch populations exposed to trawling,
estimates of the proportion of these populations removed by trawling (based on relative catch-rates
from this and other projects, e.g. FRDC 93/096), and bycatch life-history characteristics (from SRA,
FRDC 96/257 and FRDC 2000/160). Exposure risk was estimated for about 850 species: about half
the species had very low exposure, about a third had low exposure, ~218 had moderate-low exposure,
~23 had moderate-high exposure and 10 species had high exposure. Of these species, 1 had very high
estimates of proportion caught annually, 4 had moderate-high estimates of proportion caught, 28 had
moderate-low proportions caught, and the remainder (>800) had low estimates. Inclusion of the SRA
qualitative recovery ranks indicated that about 15 species (20 listed) stood out as being at higher
relative risk. Additionally, the project included another quantitative absolute sustainability indicator
(analogous to the "SAFE" method, FRDC 2004/024) based on the estimates of annual catch herein and
estimates of natural mortality rates from other sources. This method indicated that three species
GBR Seabed Biodiversity
8-272
exceeded a limit reference point, and another three other species exceeded one or two conservative
reference points. Another 10 species below the sustainability reference points included seven species
ranked highly by the SRA method and were also listed due to uncertainty in parameters. Further
research is recommended to address key uncertainties in estimates of catchability and natural mortality
rates.
The project proposed and has produced quantitative status and sustainability risk indicators for
selected seabed structural benthos in the region and evaluations of the environmental performance of
different management options implemented over the duration by relevant authorities, developed from
this projects’ estimates of the proportion of benthos populations exposed to trawling, together with
measurements of trawl-removal rates and recovery rates from other projects (FRDC 93/096,
GBRMPA Trawl Recovery) in an MSE modelling framework (GBRMPA Trawl Scenario Modelling
Project). The consistent result was that the generalized depletion trends up until the late 1990s have all
been arrested and reversed by the series of management interventions implemented between 2000 and
2005. The 2001 buyback and the subsequent progressive penalties appeared to have made the biggest
positive contributions; the 2004 rezoning made a small positive contribution for some species.
The project proposed and has produced transferable methodology and tools for regional marine
planning nationally, including: cost-effective representative survey design and techniques (including
use and assessment of video and acoustics), spatial-statistical classification and prediction methods,
and biodiversity and bycatch species risk assessment methods. Reliable assessment methods required
robust distribution and abundance data delivered by accurate identification of specimens collected by
conventional sampling devices deployed at sites selected carefully by a biophysical stratified design
strategy. Acquisition of video data delivered information on dominant visible habitat components, and
contributed to understanding of the catchability of structural biota, though identifications were
problematic and statistical treatment of data from video was less successful. BRUVS were capable of
being deployed in more rugged terrain and could provide information on fishes from such areas where
the trawl could not be deployed, but also had some identification issues. Acoustics was capable of
reliably discriminating only a few substratum types, such a soft sediment, rough ground and
intermediate sediments.
GBR Seabed Biodiversity
9-273
9. RECOMMENDATIONS
A key output from the project was the series of environmental risk indicators, which aimed to filter out
species of little or no sustainability concern so that attention could be focussed on those species at risk
or potentially at risk. Several lists were produced, based on different indicators, and species were
ranked in order of risk or potential risk. The top 20 (an arbitrary threshold) ranked species for five of
the indicators are reproduced below (Table 9-1). For two of the indicators (c and e), the ranking is
relative and no reference points are possible. For two (a and b) of the other three, the indicator is clear
and quantitative and though related to risk is not necessarily indicative of sustainability, and the
reference points are arbitrary. The C/M indicator (d) is clear, quantitative and directly related to
sustainability through well established population modelling, and the reference points are biologically
based. While only three species appear to be at risk and another three species exceed conservative
reference points, based on the C/M indicator, there is uncertainty in the indicators that requires a more
precautionary response. Further, many species (27 of 48 listed below) have multiple occurrences
across these five indicators: one species occur in all five lists, six species occur in four lists, five
species occur in three lists, and 15 species occur in two lists — which also suggests a more inclusive
response. It is recommended that the entire list, as well as the arbitrary top-20 cut off, should be
considered for consultation regarding future action. Such management, industry and stakeholder
consultation processes can decide the most appropriate strategies from options that may include
clarification of the identified uncertainties, monitoring, management interventions that reduce impacts
on these bycatch species or combinations of these actions.
As already noted, there are uncertainties in the risk assessments due to uncertainties in estimates of
catchability and natural mortality rates. In one worst case scenario of catchability=1, some additional
species would exceed the limit reference point, several would exceed the conservative reference points
and many species with unknown mortality might be of concern. Similarly, in another worst case
scenario of natural mortalities having been over-estimated by for example a factor of two, some
additional species would exceed the limit reference point, several would exceed the conservative
reference points and many species with unknown mortality might be of concern. Equally, it is possible
that clarification of these uncertainties may show that species currently thought to be at risk or
potentially at risk may be demonstrated to be of no sustainability concern. Thus, it is recommended
that further analyses of relative catchability based on existing data from multiple sources and, if
necessary, field studies of actual catchabilities be conducted to address this key uncertainty. Similarly,
it is recommended that further analyses of natural mortality rates based on existing data and, if
necessary, biological studies leading to more precise estimates of natural mortality rates be conducted
to address this key uncertainty. Such results are likely to have wide application in risk assessments
being conducted in multiple jurisdictions.
During the course of the project, preliminary recommendations for monitoring seabed areas affected
by the rezoning have been provided and discussed with GBRMPA staff. In particular, this focussed on
identifying 3-4 seabed soft-sediment areas where high levels of trawl effort have been excluded by rezoning, with suitable reference areas open and subject to trawling, in a management evaluation
framework, with preliminary costings. Given the nature of the seabed habitats and fauna in potential
candidate areas, it was noted that any possible trawl impact may be difficult to detect initially and
observable recovery may occur quite rapidly. Therefore it would have been necessary to establish the
sampling program for monitoring at the time of the rezoning, with a review of future sampling
frequency based on the results of the initial monitoring. A monitoring program of this type was not
supported by the first year's ARP of the M&TSRF.
On a less-immediate timeframe, it is recommended that key seabed habitats and constituent species be
identified on the basis of the Seabed Biodiversity dataset, for long-term monitoring of trends in
ecological condition and their responses to regional pressures, in particular climate change. Candidate
habitats should include those that have been demonstrated to be particularly biodiverse such as
vegetated areas and epibenthic gardens. The vegetated areas such as deepwater seagrass/algal beds in
the mid-shelf off Townsville, the inner-shelf Capricorn region and the Turtle-Howick Is group
vicinity; the offshore/shelf-edge algal beds of the Central Section and the northern Swains/T-Reef; and
GBR Seabed Biodiversity
9-274
the outer shelf Halimeda banks of the Northern and Far-northern Sections, may well be vulnerable to
climate change as there is an expectation that the thermocline may deepen and upwellings may
become weaker and less frequent with potential consequences for productive habitat dependent on
nutrients from such sources. The potential consequences of changed runoff patterns for inner and
midshelf vegetation and epibenthic gardens are unknown.
Other further development activities outlined in Section 6 are also recommended with expected
benefits for greater understanding of the seabed ecosystem, sustainability of the trawl fishery, zoning
assessment and ongoing marine park planning arrangements and nationally for fisheries risk
assessments and regional marine planning more generally.
GBR Seabed Biodiversity
9-275
Table 9-1. Top 20 ranked species for four different indicators. (a) Percent of biomass directly exposed to effort (red = >75%, orange = >50%). (b) Estimated percent of biomass
caught per annum (red = >75%, orange = >50%, pale = >25%). (c) highest relative risk ranked species from plotting ‘Recovery’ rank from ‘SRA’ against estimated catch from b (no
reference points are possible). (d) Sustainability indicator: estimated catch b / natural mortality rate (red = exceeds limit reference point 1.0, orange = exceeds conservative reference
point 0.8, pale = exceeds conservative reference point 0.6). (e) Highest ranked species from assemblage exposure and species affinities for assemblages (no reference points).
(a) Effort exposed %
Genus
species
Penaeus
semisulcatus
Cryptolutea
arafurensis
Brachirus
muelleri
Pentaprion
longimanus
Pelates
quadrilineatus
Leiognathus
leuciscus
Upeneus
sundaicus
Portunus
gracilimanus
Terapon
puta
Enisiculus
cultellus
Brachaluteres
taylori
Trachypenaeus anchoralis
Metapenaeus
ensis
Erugosquilla
woodmasoni
Leiognathus
bindus
Melaxinaea
vitrea
Saurida
argentea/tumbil
Terapon
theraps
Myra
tumidospina
Calliurichthys
grossi
(b) Est. % caught
Genus
species
Brachirus
muelleri
Terapon
puta
Saurida
argentea/tumbil
Penaeus
semisulcatus
Psettodes
erumei
Scolopsis
taeniopterus
Sepia
pharaonis
Yongeichthys
nebulosus
Euristhmus
nudiceps
Tripodichthys
angustifrons
Apogon
poecilopterus
Saurida
grandi/undosquamis
Upeneus
sundaicus
Sepia
elliptica
Aplysia
sp1_QMS
Amusium
pleuronectes cf
Lamellaria
sp1
Leiognathus
leuciscus
Trixiphichthys
weberi
Sillago
burrus
(c) SRA rank
Genus
species
Brachirus
muelleri
Sepia
pharaonis
Terapon
puta
Saurida
argentea/tumbil
Penaeus
semisulcatus
Euristhmus
nudiceps
Apogon
poecilopterus
Sepia
elliptica
Scolopsis
taeniopterus
Psettodes
erumei
Amusium
pleuronectes cf
Tripodichthys
angustifrons
Saurida
grandi/undosquamis
Yongeichthys
nebulosus
Sepia
whitleyana
Upeneus
sundaicus
Leiognathus
leuciscus
Sepia
smithi
Portunus
gracilimanus
Chaetodermis
penicilligera
(d) C/M indicator
Genus
species
Fistularia
petimba
Brachirus
muelleri
Trixiphichthys
weberi
Pomadasys
maculatus
Psettodes
erumei
Sillago
burrus
Dasyatis
leylandi
Nemipterus
furcosus
Polydactylus
multiradiatus
Tripodichthys
angustifrons
Terapon
puta
Euristhmus
nudiceps
Saurida
argentea/tumbil
Nemipterus
peronii
Sepia
pharaonis
Saurida
grandi/undosquamis
Amusium
pleuronectes cf
Pristotis
obtusirostris
Nemipterus
sp juv/unident
Nemipterus
hexodon
(e) Assemblage exposure
Genus
species
Enisiculus
cultellus
Cryptolutea
arafurensis
Saurida
argentea/tumbil
Pentaprion
longimanus
Thenus
parindicus
Nassarius
cremmatus cf
Placamen
tiara
Tripodichthys
angustifrons
Upeneus
sundaicus
Erugosquilla
woodmasoni
Penaeus
semisulcatus
Psettodes
erumei
Charybdis
truncata
Aplysia
sp1_QMS
Metapenaeus
ensis
Calliurichthys
grossi
Melaxinaea
vitrea
Diogenidae
sp356-1
Portunus
tuberculosus
Sea pen
sp1
GBR Seabed Biodiversity
10-276
10. REFERENCES
Birtles, A., Arnold, P. (1983) Between the Reefs: some patterns of soft substrate epibenthos on the
GBR Shelf, pp. 159-163 in Baker, J.T., Carter, R.M., Sammarco, P.W., Stark, K.P. (Eds) Proc
GBR Conference, Townsville, 1983.
Birtles R.A., Arnold, P.W. (1988) Distribution of trophic groups of epifaunal echinoderms and
molluscs in the soft sediment areas of the central Great Barrier Reef shelf. In Proceedings of
the 6th International Coral Reef Symposium, Townsville, Australia, Volume 3: Contributed
Papers (Choat, J.H. & Barnes, D. eds) 6th International Coral Reef Symposium Executive
Committee, pp. 325-332
Blaber S.J.M., Brewer D.T., Harris A.N. (1994). Distribution, biomass and community structure of
demersal fishes of the Gulf of Carpentaria, Australia. Aust J Mar Freshw Res 45: 375-96
Breiman, L. (2001). Random Forests, Machine Learning 45(1), 5–32.
Brewer, D.T., Rawlinson, N., Eayrs, S., Burridge, C.Y., (1998). An assessment of bycatch reduction
devices in a tropical Australian prawn trawl fishery. Fish. Res. 36, 195–215.
Brewer, D.T., Griffiths, S., Heales, D.S., Zhou, S., Tonks, M., Dell, Q., Taylor, B.T., Miller, M.,
Kuhnert, P., Keys, S., Whitelaw, W., Burke, A., Raudzens, E. (2007). Design, trial and
implementation of an integrated, long-term bycatch monitoring program, road tested in the
Northern Prawn Fishery. Final Report on FRDC Project 2004/024. CSIRO Cleveland. pp. 416.
Brinkman, R., Wolanski, E., Deleersnijder, E., McAllister, F., Skirving, W. (2002) Oceanic inflow
from the Coral Sea into the Great Barrier Reef. Est Coast Shelf Sci 54: 655–668
Burridge C.Y., Pitcher C.R., Wassenberg T.J., Poiner I.R., Hill, B.J. (2003). Measurement of the rate
of depletion of benthic fauna by prawn (shrimp) otter trawls: an experiment in the Great
Barrier Reef, Australia. Fisheries Research 60: 237–253
Cappo, M., Harvey, E., Malcolm, H., Speare, P. (2003) Potential of video techniques to monitor
diversity abundance and size of fish in studies of Marine Protected Areas. In: “Aquatic
Protected Areas - what works best and how do we know?" (Eds Beumer JP, Grant A, Smith
DC). World Congress on Aquatic Protected Areas Proceedings, Cairns Australia, August
2002. pp 455-464
Cappo, M., Speare, P., D’eath, G. (2004) Comparison of Baited Remote Underwater Video Stations
(BRUVS) and prawn (shrimp) trawls for assessments of fish biodiversity in inter-reefal areas
of the Great Barrier Reef Marine Park. J Exp Mar Biol Ecol 302(2): 123-152
Cappo, M., De’ath, G., Speare, P. (subm.) Shelf-scale patterns in communities of aquatic vertebrates
in the inter-reefal waters of the Great Barrier Reef Marine Park determined by Baited Remote
Underwater Video Stations. Mar Ecol Progr Ser
Cohen, A., Daubechies, I., Vial, P., (1993). Wavelets on the Interval and Fast Wavelet Transforms,
Applied and Computational Harmonic Analysis, 1 p54-81.
Coles, R., McKenzie, L., De’ath, G., Roelofs, A., Lee Long, W. (2006) Modelling factors that
influence the spatial distribution of deepwater seagrass in the inter reef lagoon of the Great
Barrier Reef World Heritage Area, northern Australia. Submitted manuscript.
Condie, S.A., Dunn, J.R. (2006). Seasonal characteristics of the surface mixed layer in the
Australasian region: implications for primary production regimes and biogeography. Mar
Freshw Res 57: 569–590
Courtney, A.J., Haddy, J.A., Campbell, M.J., Roy, D.P., Tonks, M.L., Gaddes, S.W., Chilcott, K.E.,
O’Neill, M.F., Brown, I.W., McLennan, M., Jebreen, J.E., Van Der Geest, C., Rose, C., Kistle,
S., Turnbull, C.T., Kyne, P.M., Bennett, M.B. and Taylor, J. (2006). Bycatch weight,
composition and preliminary estimates of the impact of bycatch reduction devices in
Queensland's trawl fishery. FRDC Project #2000/170 Final Report 353p.
GBR Seabed Biodiversity
10-277
Courtney, A.J., Tonks, M.L., Campbell, M.J., Roy, D.P., Gaddes, S.W., Kyne, P.M., O’Neill, M.F.
(2006). Quantifying the effects of bycatch reduction devices in Queensland’s (Australia)
shallow water eastern king prawn (Penaeus plebejus) trawl fishery. Fisheries Research
80:136–147
De’ath, G. (2006) Boosted trees for ecological modeling and prediction. Ecology
De’ath, G. (2002) Multivariate regression trees: a new technique for modeling species-environment
relationships. Ecology 83:1105-1117.
DeVantier, L.M., De’ath, G., Turak, E., Done, T.J., Fabricius, K.E. (2006). Species richness and
community structure of reef-building corals on the nearshore Great Barrier Reef. Coral Reefs
25(3): 329-340
Dredge, M.C.L. (1985) Estimates of natural mortality and yield-per-recruit for Amusium japonicum
balloti Bernardi (Pectinidae) based on tag recoveries. Journal of Shellfish Research, 5:103–
109.
Drew, E.A. (2001). Ocean nutrients to sediment banks via tidal jets and Halimeda meadows. In:
Wolanski, E. (Ed.), Oceanographic processes of coral reefs: physical and biological links in
the Great Barrier Reef. CRC Press, Boca Raton, Florida, pp. 127-144
Duda, R., Hart, P., Stork, D., (2001). Pattern Classification, 2nd Ed, Wiley.
Dufrêne, M., Legendre, P. (1997) Species assemblages and indicator species: the need for a flexible
asymmetrical approach. Ecol Monogr 67(3): 345-66
Ellis, N., Pantus, F. (2001) Management strategy modelling: tools to evaluate trawl management
strategies with respect to impacts on benthic biota within the Great Barrier Reef Marine Park
area. CSIRO Marine Research Final Report to the Great Barrier Reef Marine Park Authority.
Engel, J. & Kvitek, R. (1998) Effects of otter trawling on benthic community in Monterey Bay
National Marine Sanctuary. Cons.Biol.12(6):1204-214.
Environment Australia, (2000). Biodiversity Research: Australia’s Priorities — a Discussion Paper.
Fabricius, K., De’ath, G. (2001) Biodiversity on the Great Barrier Reef: Large-scale patterns and
turbidity-related local loss of soft coral taxa. In: Wolanski, E. (Ed.), Oceanographic processes
of coral reefs: physical and biological links in the Great Barrier Reef. CRC Press, Boca Raton,
Florida, pp. 127-144
Fager, E.W., Longhurst, A.R. (1968) Recurrent group analysis of species assemblages of demersal fish
in the Gulf of Guinea. J Fish Res Board Can 25(7): 1405–21
Fogarty, M.P., and Murawski, S.A. 1998. Large-scale disturbance and the structure of marine systems:
fishery impacts on Georges Bank. Ecological Applications, 8(1), Supplement: S6eS22.
Friedman, J.H. (2001) Greedy function approximation: a gradient boosting machine. Ann Stat 29(5):
1189-1232
Fulton, E.A., Smith, A.D.M., Webb H., Slater, J. (2004). Ecological Indicators for the Impacts of
Fishing on Non-Target Species, Communities and Ecosystems: Review of Potential Indicators.
AFMA Final Research Report Number R99/1546
Gribble, N. A. (2003). GBR-prawn: modelling ecosystem impacts of changes in fisheries management
of the commercial prawn (shrimp) trawl fishery in the far northern Great Barrier Reef.
Fisheries Research 65(1-3): 493-506
Gribble, N. A. (2004). A spatially explicit multi-competitor coexistence model of penaeid (shrimp)
distribution on the Australian Great Barrier Reef. Ecological Modelling: Vol 177(1-2):61-74
Gribble, N.A., Good, N., Peel, D., Tanimoto, M., and Officer, R. (2007). Innovative stock assessment
and effort mapping using VMS and electronic logbooks. Final Report to FRDC 2002/056.
Griffiths, S.P., Brewer, D.T., Heales, D.S., Milton, D.A., Stobutzki, I.C., (2006). Validating ecological
risk assessments for fisheries: assessing the impacts of turtle excluder devices on
GBR Seabed Biodiversity
10-278
elasmobranch bycatch populations in an Australian trawl fishery. Marine and Freshwater
Research, 57, 395–401
Gulland, J.A. (1983). Fish Stock Assessment: a manual of basic methods. FAO/Wiley Inter-Science,
New York, 223 pp.
Gust, N., Choat, J.H., and McCormick, M.I. (2001) Spatial variability in reef fish distribution,
abundance, size and biomass: a multi-scale analysis. Mar Ecol Prog Ser 214, 237-251
Hancock, G.J., Webster, I.T., Stieglitz, T.C. (In Press) Horizontal mixing of Great Barrier Reef waters:
offshore diffusivity determined from radium isotope distribution. JGR Oceans
Haywood, M., Hill, B., Donovan, A., Rochester, W., Ellis, N., Welna, A., Gordon, S., Cheers, S.,
Forcey, K., Mcleod, I., Moeseneder, C., Smith, G., Manson, F., Wassenberg, T., Thomas,
Steve, Kuhnert, P., Laslett, G., Burridge, C., and Thomas, Sarah, (2005). Quantifying the
effects of trawling on seabed fauna in the Northern Prawn Fishery. Final Report on FRDC
Project 2002/102. CSIRO, Cleveland. 488 pp.
Hobday, A. J., Smith, A., Webb, H., Daley, R., Wayte, S., Bulman, C., Dowdney, J. Williams,,A.,
Sporcic, M., Dambacher, J., Fuller,M., Walker, T. (2006) Ecological Risk Assessment for the
Effects of Fishing: Methodology. Report R04/1072 for the Australian Fisheries Management
Authority, Canberra
Hilborn, R., Walters, C.J. (1992). Quantitative fisheries stock assessment: choice, dynamics and
uncertainty. Chapman & Hall, London. 570p.
Hill, B.J., Haywood, M., Venables, B., Gordon, S.R., Condie, S., Ellis, N.R., Tyre, A., Vance, D.,
Dunn, J., Mansbridge, J., Moeseneder, C., Bustamante, R. and Pantus, F., (2002). Surrogates I
- Predictors, impacts, management and conservation of the benthic biodiversity of the
Northern Prawn Fishery. Final Report on FRDC Project 2000/160. CSIRO, Cleveland. 437 pp.
Jensen, A., la Cour-Harbo, A., (2001). Ripples in Mathematics, the Discrete Wavelet Transform,
Springer.
Kaufman, L. and Rousseeuw, P. J. (1990) Finding Groups in Data, Wiley, New York.
Kaufman, L. and Rousseeuw, P. J. (1990). Finding Groups in Data: An Introduction to Cluster
Analysis. Wiley, New York.
Koch, V., Wolff, M. (2002) Energy budget and ecological role of mangrove epibenthos in the Caeté
estuary, North Brazil. Mar Ecol Prog Ser 228: 119–130
Kodama, K., Horiguchi, T., Kume1, G., Nagayama, S., Shimizu, T., Shiraishi, H., Morita, M.,
Shimizu, M., (2006) Effects of hypoxia on early life history of the stomatopod Oratosquilla
oratoria in a coastal sea. Mar Ecol Prog Ser Vol. 324: 197–206.
Larcombe, P., Carter, R.M., (2004) Cyclone pumping, sediment partitioning and the development of
the Great Barrier Reef shelf system: a review. Quatern Sci Rev 23:107–135
Lee, H-H., Hsu, C-C. (2003). Population biology of the swimming crab Portunus sanguinolentus in
the waters off northern Taiwan. Journal of Crustacean Biology 23:691–699
Letourneur, Y., Kulbicki, M., Labrosse, P. (1998) Spatial structure of commercial reef fish
communities along a terrestrial runoff gradient in the northern lagoon of New Caledonia.
Environ Biol Fish 51: 141–159
Long, B.G., Skewes, T.D., Taranto, T.J., Smith, G., Mcleod, I., Pitcher, C.R., Poiner, I.R. (1997) PNG
Gas Project: Torres Strait Marine Infill Survey: Report To NSR Re Papua New Guinea–
Queensland Gas Project. CSIRO Division of Marine Research.
Longhurst, A.R., Pauly, D. (1987). Ecology of tropical oceans. Academic Press, San Diego, California
(USA). 407p
Lowe-McConnell, R.H. (1987). Ecological studies in tropical fish communities. Cambridge University
Press. pp 213-223
GBR Seabed Biodiversity
10-279
Magurran, A.E., Henderson, P.A. (2003) Explaining the excess of rare species in natural species
abundance distributions. Nature 422(6933): 714-716
Mathews, E.J. and Heap, A.D., (2006). Sedimentology and Geomorphology of Inter-reefal Seabed
Environments of the Great Barrier Reef, northeast Australia. Geoscience Australia, Record
2006/11.
McLeod, I., Skewes, T.D., Gordon, S.R., Pitcher, C.R. (2007). A method for seabed habitat mapping:
Integrating acoustic information with biogeophysical observations; Case study – Scott Reef, in
Todd, B.J., and Greene, H.G., eds. Mapping the Seafloor for Habitat Characterization:
Geological Association of Canada, Special Paper 47, pp.
Newman, S.J., Williams, D.M., Russ, G.R. (1997). Patterns of zonation of assemblages of the
Lutjanidae, Lethrinidae and Serranidae (Epinephelinae) within and among mid shelf and outer
shelf reefs in the central Great Barrier Reef. Mar Freshw Res 48:119–128
Penn, J.W. (1976) Tagging Experiments with Western King Prawn, Penaeus latisulcatus Kishinouye
II. Estimation of Population Parameters. Aust. J, Mar. Freshwater Res. 27, 239-50
Pitcher, C.R., Venables, W., Ellis, N., McLeod, I., Pantus, F., Austin, M., Cappo, M., Doherty, P.,
Gribble, N. (2002). GBR Seabed Biodiversity Mapping Project: Phase 1. CSIRO/AIMS/QDPI
Report to CRC-Reef, pp. 192.
Pitcher C.R., Austin, M., Burridge, C.Y., Bustamante, R.H., Cheers, S.J., Ellis, N., Jones, P.N.,
Koutsoukos, A.G., Moeseneder, C.H., Smith, G.P., Venables, W., Wassenberg, T.J., (2004).
Recovery of Seabed Habitat from the Impact of Prawn Trawling in the Far Northern Section
of the Great Barrier Reef Marine Park. CSIRO Marine Research Final Report to GBRMPA,
pp. 370
Pitcher, C.R., Turnbull, C., Atfield, J., Griffin, D., Dennis, D.M., Skewes, T.D. (2005). Biology, larval
transport modelling and commercial logbook data analysis to support management of the NE
Queensland rock lobster Panulirus ornatus fishery. FRDC Project 2002/008 report to Fisheries
Research and Development Corporation. CSIRO Marine & Atmospheric Research pp. 148
Poiner, I.R., Glaister, J., Pitcher, C.R., Burridge, C., Wassenberg, T., Gribble, N., Hill, B., Blaber,
S.J.M., Milton, D.A., Brewer, D., Ellis, N., (1998). The environmental effects of prawn
trawling in the far northern section of the Great Barrier Reef Marine Park: 1991–1996. Final
Report to GBRMPA and FRDC. CSIRO Division of Marine Research – Queensland
Department of Primary Industries Report, 554 pp.
Porter-Smith, R., Harris, P.T., Andersen, O.B., Coleman, R., Greenslade, D., Jenkins, C.J. (2004).
Classification of the Australian continental shelf based on predicted sediment threshold
exceedance from tidal currents and swell waves. Mar Geol 211: 1– 20
R Development Core Team (2005) R: a language and environment for statistical computing. R
Foundation for statistical computing, Vienna.
Ramm, D.C., Pender, P.J., Willing, R.S., Buckworth, R.C. (1990). Large-scale spatial patterns of
abundance within the assemblage of fish caught by prawn trawlers in northern Australian
waters. Aust J Mar Freshw Res 41: 79-95
Ridgway, K. R., J. R. Dunn and J. L. Wilkin, 2002. Ocean interpolation by four-dimensional least
squares - Application to the waters around Australia. Journal of Atmospheric and Oceanic
Technology, 19, 1357-1375.
Robins, J.B. & McGilvray, J.G. (1999). The AusTED II, an improved trawl efficiency device 2.
Commercial performance. Fisheries Research 40: 29-41.
Robins-Troeger, J.B., Buckworth, R.C. & Dredge, M.C.L. (1995). Development of a trawl efficiency
device (TED) for Australian prawn fisheries II. Field evaluations of the AusTED. Fisheries
Research, 22: 107-117.
Sainsbury KJ, Campbell RA, Lindholm R, Whitelaw AW (1997). Experimental management of an
Australian multispecies fishery: Examining the possibility of trawl induced habitat
GBR Seabed Biodiversity
10-280
modification. In: Pikitch EK, Huppert DD, Sissenwine MP (Eds) Global Trends: Fisheries
management. pp 107-112. American Fisheries Society, Maryland, USA. John Wiley and Son
Saito, N., Coifman, R., (1995). Local Discriminant Bases and their Applications. Journal of
Mathematical Imaging and Vision, 5: 337-358.
Schwarz, G. (1978). "Estimating the dimension of a model". Annals of Statistics 6, 461-464.
Scott, D. W. (1992) Multivariate Density Estimation: Theory, Practice, and Visualization. Wiley.
Simrad EY500 Operating Manual ht—://www.simrad.com/html
Skewes, T.D., Pitcher, C.R., Long, B.G., Mcleod, I., Taranto, T.J. Gordon, S., Smith, G., Poiner, I.R.
(1996) Torres Strait Infill Survey: Report to IPC Re Pandora Gas Development Project.
CSIRO Division of Fisheries.
Sternlicht, D., de Moustier, C., (2003). Time-dependent seafloor acoustic backscatter (10-100kHz), J.
Acous. Soc. Am. 114 (5), p2709--2725 (2003).
Stobutzki, I.C., Miller, M.J., Brewer, D.T. (2001a). Sustainability of fishery bycatch: a process for
assessing highly diverse and numerous bycatch. Environmental Conservation 28, 167–181.
Stobutzki IC, Miller MJ, Jones P, Salini JP (2001b) Bycatch diversity and variation in a tropical
Australian penaeid fishery; the implications for monitoring. Fish Res 53(3): 283-301
Stobutzki, I.C., Miller, M.J., Heales, D.S., Brewer, D.T. (2002). Sustainability of elasmobranchs
caught as bycatch in a tropical shrimp trawl fishery. Fish. Bull. 100:800-821.
Trefethen, L. N. 1980. Numerical Computation of the Schwarz-Christoffel Transformation, Siam J.
Sci. Stat. Comp. 1, 82–102
Wahba, G. (1990) Spline Models for Observational Data. National Science Foundation -Conference
Board of Mathematical Sciences Regional Conference Series in Applied Mathematics,
Philadelphia: Society for Industrial and Applied Mathematics.
Ward, J.H. (1963). Hierarchical Grouping to optimize an objective function. Journal of American
Statistical Association, 58(301), 236-244
Wassenberg TJ, Blaber SJM, Burridge CY, Brewer DT, Salini JP, Gribble N (1997). The effectiveness
of fish and shrimp trawls for sampling fish communities in tropical Australia. Fish Res 30:
229-240
Watson RA, Dredge MLC, Mayer DG (1990). Spatial and seasonal variation in demersal trawl fauna
associated with a prawn fishery on the central Great Barrier Reef, Australia. Aust J Mar
Freshw Res 41: 65-77
Watson, R.A., Goeden, G.B. (1989) Temporal and spatial zonation of the demersal trawl fauna of the
central Great Barrier Reef. Mem. Qld. Mus. 27(2): 611-620
Wilkinson, C.R., Cheshire, A.C. (1988). Cross shelf variations in coral reef structure and function—
influences of land and ocean. In: Proceedings of the 6th International Coral Reef Symposium,
Townsville, Australia, Volume 3: Contributed Papers (Choat, J.H. & Barnes, D. eds) 6th
International Coral Reef Symposium Executive Committee, pp 227–233
Williams, D.M. (1991). Patterns and processes in the distribution of coral reef fishes. In: Sale PF (ed)
The ecology of fishes on coral reefs. Academic Press San Diego. pp 437–474
Williams, A., Last, P.R., Gomon, M.F., & Paxton, J.R. (1996) Species composition and checklist of
the demersal ichthyofauna of the continental slope off western Australia (20–35ºS) Rec. WA
Mus. 18, 135–155.
Wolanski, E. (1994). Physical oceanographic processes of the Great Barrier Reef. CRC Press Marine
Science Series.
Zhou, S., Griffiths, S.P. (2007) Sustainability Assessment for Fishing Effects (SAFE): a new
quantitative ecological risk assessment method and its application to elasmobranch bycatch in
an Australian trawl fishery. Fisheries Research (In Press)
GBR Seabed Biodiversity
11. ABBREVIATIONS
BRUVS – Baited Remote Underwater Video Stations
CARS2000 – CSIRO Atlas of Regional Seas
CERF National Marine Biodiversity Hub
CLARA – Clustering for large Datasets
CRC – Cooperative Research Centre
CTD – Conductivity Temperature Depth sensor array
DEW – Department of Environment & Water Resources
DEM – Digital Elevation Model
DLI – Dufrêne-Legendre Index
EPBC – Environment Protection & Biodiversity Conservation
ERA – Ecological Risk Assessments
FRDC – Fisheries Research & Development Corporation
GA – Geoscience Australia
GBRWHA – Great Barrier Reef World Heritage Area
GPS – Global Positioning System
GU zone – General Use zone
HO – Hydrographic Office
HPA – Highly protected area
IQR – Inter-Quartile Range
JCU – James Cook University
MaxN – Maximum number of fish seen at any one time in field of view of BRUVS
MSE – Management Strategies Evaluations
NOO – National Oceans Office
NPF – Northern Prawn Fishery
NSRMPA – National Representitive System of Marine Protected Areas
OSI – Ocean Sciences Institute, Sydney University
PAM – Partitioning Around Medoids
PAR – Photosynthetically Active Radiation
PC-space – Principal Components Space
SRA – Susceptibility-Recovery Analysis
QDPI&F – Queensland Department of Primary Industries & Fisheries
QECTF – Queensland East Coast Trawl Fishery
QSIA – Queensland Seafood Industry Association
RAP – Representative Areas Program
RMS – Residual Mean Square
SRA – Susceptibility Recovery Analysis
SVD – Singular Value Decomposition
VMS – Vessel Monitoring System
VTRs – Video Tape Recorders
WHA – World Heritage Area
WTO – Wildlife Trade Operation
11-281
GBR Seabed Biodiversity
12. APPENDIX 1: INTELLECTUAL PROPERTY
There is no intellectual property of a commercial nature arising from the outputs of this Project.
12-282
GBR Seabed Biodiversity
13-283
13. APPENDIX 2: STAFF
Roland Pitcher
Peter Doherty
Peter Arnold
John Hooper
Neil Gribble
Chris Bartlett
Matthew Browne
Norm Campbell
Toni Cannard
Mike Cappo
Giovannella Carini
Susan Chalmers
Sue Cheers
Doug Chetwynd
Andrew Colefax
Rob Coles
Stephen Cook
Peter Davie
Glenn De'ath
Drew Devereux
Barbara Done
Tim Donovan
Barry Ehrke
Nick Ellis
Gavin Ericson
Ida Fellegara
Karl Forcey
Melodyrose Furey
Dan Gledhill
Norm Good
Scott Gordon
Mick Haywood
Ian Jacobsen
Jeff Johnson
Michelle Jones
Stuart Kinninmoth
Sarah Kistle
Peter Last
Anita Leite
Shona Marks
Ian McLeod
Sybilla Oczkowicz
Cassanda Rose
Denise Seabright
Jacquie Sheils
Matt Sherlock
Posa Skelton
David H Smith
Greg Smith
Peter Speare
Marcus Stowar
Colleen Strickland
Patricia Hendriks
Claire Van der Geest
CSIRO Marine and Atmospheric Research
Australian Institute of Marine Science
Museum of Tropical Queensland
Queensland Museum South Bank
Queensland Department of Primary Industries & Fisheries
Museum of Tropical Queensland
CSIRO Maths and Information Sciences
CSIRO Maths and Information Sciences
CSIRO Marine and Atmospheric Research
Australian Institute of Marine Science
Queensland Museum South Bank
Queensland Department of Primary Industries & Fisheries
CSIRO Marine and Atmospheric Research
CSIRO Marine and Atmospheric Research
Queensland Museum South Bank
Queensland Department of Primary Industries & Fisheries
Queensland Museum South Bank
Queensland Museum South Bank
Australian Institute of Marine Science
CSIRO Maths and Information Sciences
Museum of Tropical Queensland
Australian Institute of Marine Science
Queensland Department of Primary Industries & Fisheries
CSIRO Marine and Atmospheric Research
Australian Institute of Marine Science
Queensland Museum South Bank
CSIRO Marine and Atmospheric Research
CSIRO Marine and Atmospheric Research
CSIRO Marine and Atmospheric Research
Queensland Department of Primary Industries & Fisheries
CSIRO Marine and Atmospheric Research
CSIRO Marine and Atmospheric Research
University of Queensland
Queensland Museum South Bank
Queensland Museum South Bank
Australian Institute of Marine Science
Queensland Department of Primary Industries & Fisheries
CSIRO Marine and Atmospheric Research
Museum of Tropical Queensland
CSIRO Marine and Atmospheric Research
CSIRO Marine and Atmospheric Research
Queensland Department of Primary Industries & Fisheries
Queensland Department of Primary Industries & Fisheries
Museum of Tropical Queensland
CSIRO Marine and Atmospheric Research
CSIRO Marine and Atmospheric Research
Queensland Department of Primary Industries & Fisheries
CSIRO Marine and Atmospheric Research
CSIRO Marine and Atmospheric Research
Australian Institute of Marine Science
Australian Institute of Marine Science
Queensland Museum South Bank
Queensland Museum South Bank
Queensland Department of Primary Industries & Fisheries
Cleveland
Townsville
Townsville
Brisbane
Cairns
Townsville
Cleveland
Perth
Cleveland
Townsville
Brisbane
Cairns
Cleveland
Cleveland
Brisbane
Cairns
Brisbane
Brisbane
Townsville
Perth
Townsville
Townsville
Cairns
Cleveland
Townsville
Brisbane
Cleveland
Cleveland
Hobart
Cairns
Cleveland
Cleveland
Brisbane
Brisbane
Brisbane
Townsville
Cairns
Hobart
Townsville
Cleveland
Cleveland
Cairns
Cairns
Townsville
Cleveland
Cleveland
Townsville
Cleveland
Cleveland
Townsville
Townsville
Brisbane
Brisbane
Cairns
Qld.
Qld
Qld
Qld
Qld
Qld
Qld
WA
Qld
Qld
Qld
Qld
Qld
Qld
Qld
Qld
Qld
Qld
Qld
WA
Qld
Qld
Qld
Qld
Qld
Qld
Qld
Qld
Tas
Qld
Qld
Qld
Qld
Qld
Qld
Qld
Qld
Tas
Qld
Qld
Qld
Qld
Qld
Qld
Qld
Qld
Qld
Qld
Qld
Qld
Qld
Qld
Qld
Qld
GBR Seabed Biodiversity
Bill Venables
Cath Walsh
Ted Wassenberg
Andrzej Welna
Gus Yearsley
CSIRO Maths and Information Sciences
Queensland Department of Primary Industries & Fisheries
CSIRO Marine and Atmospheric Research
CSIRO Marine and Atmospheric Research
CSIRO Marine and Atmospheric Research
13-284
Cleveland
Townsville
Cleveland
Cleveland
Hobart
Qld
Qld
Qld
Qld
Tas
GBR Seabed Biodiversity
14. APPENDIX 3: PROJECT STEERING COMMITTEE MEMBERS
Chairperson:
Peter Doherty — AIMS, CRC-Reef Program C Leader
Task Leader:
Roland Pitcher — CSIRO
Task Associate: David Williams — CRC-Reef
Task Associate: Dorothea Huber Æ Phil Cadwallader — GBRMPA
Task Associate: Brigid Kerrigan Æ Malcolm Dunning — QDPI&F
Task Associate: Duncan Souter Æ Martin Hicks — QSIA
Task Associate: Barry Ehrke — QSIA
Task Associate: Vicki Nelson — NOO
Task Associate: Vern Veitch — Sunfish
14-285
GBR Seabed Biodiversity
15-286
15. APPENDIX 4: SINGLE SPECIES TRAWL EXPOSURE
Table 15-1: Summary of species exposure estimates for all 840 modelled species, ranked in decending order by
percent of biomass exposed to trawl effort intensity; showing also species group membership, total estimated
biomass, percent of biomass available in General Use zone, percent of biomass potentially exposed in trawled
cells. Pale orange: >25% biomass exposed; dark orange: >50% biomass exposed; red: >50% biomass exposed.
Group
29
29
29
29
21
9
9
9
29
29
8
13
29
22
24
9
29
22
22
9
22
9
22
22
29
14
22
13
9
9
35
38
29
29
22
22
9
21
14
22
9
9
9
22
14
21
14
14
10
22
13
13
10
14
Class
Crustacea
Crustacea
Actinopterygii
Actinopterygii
Actinopterygii
Actinopterygii
Actinopterygii
Crustacea
Actinopterygii
Bivalvia
Actinopterygii
Crustacea
Crustacea
Crustacea
Actinopterygii
Bivalvia
Actinopterygii
Actinopterygii
Crustacea
Actinopterygii
Actinopterygii
Crustacea
Gastropoda
Actinopterygii
Bivalvia
Actinopterygii
Actinopterygii
Actinopterygii
Actinopterygii
Bivalvia
Actinopterygii
Actinopterygii
Actinopterygii
Actinopterygii
Anthozoa
Actinopterygii
Actinopterygii
Crustacea
Cephalopoda
Actinopterygii
Bivalvia
Crustacea
Actinopterygii
Actinopterygii
Crustacea
Actinopterygii
Crustacea
Crustacea
Bivalvia
Actinopterygii
Actinopterygii
Actinopterygii
Holothuroidea
Gastropoda
Genus
Penaeus
Cryptolutea
Brachirus
Pentaprion
Pelates
Leiognathus
Upeneus
Portunus
Terapon
Enisiculus
Brachaluteres
Trachypenaeus
Metapenaeus
Erugosquilla
Leiognathus
Melaxinaea
Saurida
Terapon
Myra
Calliurichthys
Upeneus
Thenus
Nassarius
Psettodes
Placamen
Scolopsis
Leiognathus
Repomucenus
Cynoglossus
Amusium
Yongeichthys
Apogon
Euristhmus
Tripodichthys
Sea pen
Gerres
Selaroides
Penaeus
Sepia
Nemipterus
Modiolus
Penaeus
Cynoglossus
Caranx
Metapenaeus
Saurida
Charybdis
Portunus
Amusium
Suggrundus
Leiognathus
Parapercis
Bohadschia
Lophiotoma
Species
semisulcatus
arafurensis
muelleri
longimanus
quadrilineatus
leuciscus
sundaicus
gracilimanus
puta
cultellus
taylori
anchoralis
ensis
woodmasoni
bindus
vitrea
argentea/tumbil
theraps
tumidospina
grossi
sulphureus
parindicus
cremmatus cf
erumei
tiara
taeniopterus
splendens
belcheri
maculipinnis
pleuronectes cf
nebulosus
poecilopterus
nudiceps
angustifrons
sp1
filamentosus
leptolepis
latisulcatus
pharaonis
peronii
elongatus
esculentus
sp 1 punctate
bucculentus
endeavouri
grandi/undo
truncata
tuberculosus
balloti
macracanthus
cf bindus
diplospilus
marmorata cf
acuta
Biomass Kg %Available
301314
74
480
57
80330
69
61963
62
129842
69
171753
59
370945
63
204641
59
60300
56
984
61
62129
71
45119
64
31126
67
19542
66
76017
42
171979
59
1109937
58
359964
63
14791
57
171819
54
723274
70
518607
55
35852
55
361247
61
3225
55
1016419
51
270168
54
98260
64
78915
60
824663
60
66438
42
121050
50
1374323
56
43969
45
507
57
84315
56
586810
56
235627
59
139386
51
1355758
64
39291
56
1031505
62
80719
56
1236784
64
534272
52
8331858
59
437520
48
394
47
2355308
55
559472
59
22870
59
3855
59
270670
69
4385
54
%Exposed
64
41
59
48
47
43
50
38
47
46
60
44
49
49
28
38
38
43
38
39
46
36
39
40
35
33
44
42
38
37
25
34
33
36
37
41
36
39
34
37
35
36
34
39
31
37
31
31
37
33
36
36
54
34
%EffortExp
174
128
119
117
103
95
93
86
78
75
72
67
67
65
63
63
63
62
60
59
58
57
57
56
55
54
54
53
52
52
51
51
51
50
50
50
49
49
48
48
47
47
47
47
46
46
46
46
45
45
45
45
44
44
GBR Seabed Biodiversity
Group
21
9
21
13
29
22
9
10
38
13
21
10
13
14
22
10
9
13
22
29
22
21
27
9
22
9
21
27
29
14
22
13
14
22
10
7
10
13
13
10
25
21
33
24
25
33
8
9
21
10
3
2
33
33
31
29
9
21
8
19
21
2
33
8
Class
Actinopterygii
Actinopterygii
Crustacea
Crustacea
Crustacea
Crustacea
Crustacea
Actinopterygii
Crustacea
Actinopterygii
Crustacea
Actinopterygii
Actinopterygii
Gastropoda
Gastropoda
Actinopterygii
Echinoidea
Crustacea
Actinopterygii
Cephalopoda
Gastropoda
Actinopterygii
Bivalvia
Gastropoda
Actinopterygii
Ophiuroidea
Crustacea
Actinopterygii
Bivalvia
Crustacea
Crustacea
Actinopterygii
Actinopterygii
Holothuroidea
Actinopterygii
Gastropoda
Actinopterygii
Gastropoda
Actinopterygii
Gymnolaemata
Gastropoda
Actinopterygii
Actinopterygii
Crustacea
Actinopterygii
Crustacea
Chlorophyceae
Bivalvia
Holothuroidea
Phaeophyceae
Crustacea
Anthozoa
Asteroidea
Actinopterygii
Demospongiae
Crustacea
Asteroidea
Actinopterygii
Gastropoda
Actinopterygii
Gymnolaemata
Demospongiae
Actinopterygii
Actinopterygii
Genus
Pseudorhombus
Inegocia
Ixa
Leucosia
Liagore
Portunus
Calappa
Ambiserrula
Oratosquillina
Leiognathus
Portunus
Aploactis
Trixiphichthys
Vexillum
Bufonaria
Inimicus
Brissopsis
Cryptopodia
Torquigener
Sepia
Aplysia
Upeneus
Leionucula
Lamellaria
Sillago
Dougaloplus
Dorippe
Grammatobothus
Dosinia
Diogenidae
Paguristes
Cynoglossus
Nemipterus
Holothuria
Suezichthys
Strombus
Apogon
Murex
Pomadasys
Iodictyum
Strombus
Paramonacanthus
Pseudorhombus
Calappa
Scorpaenopsis
Portunus
Chaetomorpha
Trisidos
Holothuria
Sporochnus
Paradorippe
Virgularia
Luidia
Apistus
Xenospongia
Pronotonyx
Astropecten
Paracentropogon
Strombus
Pseudorhombus
Selenaria
Ircinia
Minous
Asterorhombus
Species
arsius
japonica
inermis
ocellata
rubromaculata
hastatoides
sp44
jugosa
gravieri
moretoniensis
pelagicus
aspera
weberi
obeliscus cf
rana
caledonicus
luzonica
queenslandi
whitleyi
elliptica
sp1_QMS
asymmetricus
superba
sp1
burrus
echinata
quadridens
polyophthalmus
altenai
sp356-1
sp2358-2
sp juv/unident
nematopus
ocellata
gracilis
vittatus
nigripinnis
brevispina
maculatus
spp
campbelli
otisensis
elevatus
terraereginae
furneauxi
sanguinolentus
crassa
semitortata
sp2
comosus
australiensis
sp1
hardwicki
carinatus
patelliformis
leavis
sp4_AIM
longispinus
dilatatus
spinosus
maculata cf
1255
versicolor
intermedius
Biomass Kg %Available
329560
68
1096930
60
2544
62
13523
59
49419
48
5197
55
10969
57
501376
68
48611
48
47237
52
2172862
60
21363
66
59106
56
2302
44
19213
56
711097
65
1377669
46
5162
54
150537
52
158747
51
450338
51
367368
60
5499
56
5697
46
307944
46
2513
47
3584
63
358459
59
191530
50
442
45
30865
52
14818
54
693470
37
858968
52
14695
61
56120
57
59272
65
4747
61
1542585
65
17890
56
22441
64
402900
58
775731
67
10382
45
2174
60
1018755
65
360585
52
402307
47
110155
64
1515084
57
2758
49
2422
41
26743
58
1073477
56
599
59
329
46
11187
45
89968
49
92276
53
969118
58
288844
59
7482318
46
92283
58
154382
61
15-287
%Exposed
41
36
40
36
25
37
34
51
30
34
37
47
32
25
33
47
27
32
34
30
32
37
32
27
30
27
38
33
27
25
31
31
21
32
44
38
41
35
35
45
40
37
36
28
41
40
34
26
41
38
27
25
34
34
38
24
26
30
38
32
37
27
34
37
%EffortExp
44
44
44
44
43
43
43
43
42
41
40
40
40
39
39
39
38
38
38
38
38
38
37
37
37
37
37
37
37
36
36
36
36
36
36
36
35
35
35
35
35
35
35
34
34
34
34
34
34
34
33
33
33
33
33
33
33
33
33
33
33
33
32
32
GBR Seabed Biodiversity
Group
25
20
19
8
10
29
21
7
22
14
22
21
8
20
8
7
25
24
29
25
33
10
22
21
12
2
9
19
12
22
29
10
25
10
29
10
8
8
10
21
21
7
31
9
8
14
10
13
10
25
20
20
19
8
24
29
6
10
12
7
10
20
30
10
Class
Actinopterygii
Actinopterygii
Crustacea
Crustacea
Phaeophyceae
Actinopterygii
Actinopterygii
Echinoidea
Crustacea
Actinopterygii
Gastropoda
Liliopsida
Bivalvia
Echinoidea
Gymnolaemata
Actinopterygii
Actinopterygii
Actinopterygii
Actinopterygii
Holothuroidea
Foraminifera
Rhodophyceae
Actinopterygii
Actinopterygii
Crustacea
Holothuroidea
Asteroidea
Bivalvia
Crustacea
Echinoidea
Crustacea
Crustacea
Bivalvia
Liliopsida
Actinopterygii
Liliopsida
Rhodophyceae
Chlorophyceae
Phaeophyceae
Rhodophyceae
Actinopterygii
Actinopterygii
Actinopterygii
Bivalvia
Gastropoda
Liliopsida
Gastropoda
Gastropoda
Bivalvia
Rhodophyceae
Anthozoa
Asteroidea
Rhodophyceae
Crustacea
Bivalvia
Bivalvia
Crustacea
Bivalvia
Crustacea
Actinopterygii
Chlorophyceae
Actinopterygii
Gymnolaemata
Actinopterygii
Genus
Paraploactis
Nemipterus
Scyllarus
Actumnus
Padina
Nemipterus
Dactylopus
Salmacis
Portunus
Fistularia
Gemmula
Halophila
Ctenocardia
Ova
Hippothoa
Torquigener
Chaetodermis
Cynoglossus
Apogon
Holothuroidea
Discobotellina
Chondrophycus
Nemipterus
Centriscus
Ebalia
Stichopus
Astropecten
Lomopsis
Austrolibinia
Chaetodiadema
Ceratoplax
Penaeus
Barbatia
Halophila
Epinephelus
Halophila
Dasya
Udotea
Lobophora
Osmundaria
Nemipteridae
Calliurichthys
Trachinocephalus
Antigona
Xenophora
Halophila
Philine
Natica
Annachlamys
Polysiphonia
Alertigorgia
Astropecten
Laurencia
Scyllarus
Paphia
Corbula
Neopalicus
Annachlamys
Oratosquillina
Cynoglossus
Codium
Carangidae
Schizomavella
Repomucenus
Species
kagoshimensis
furcosus
demani
squamosus
sp
hexodon
dactylopus
sphaeroides
tuberculatus
petimba
sp2
decipiens
virgo cf
lacunosus
distans
sp1 (gloerfelt-tarp)
penicilligera
sp kopsi group
fasciatus
sp2
biperforata
sp1
sp juv/unident
scutatus
lambriformis
ocellatus
granulatus cf
sp1
gracilipes
granulatum
ciliata
plebejus
parvillosa cf
ovalis
sexfasciatus
spinulosa
sp1
argentea
variegata
fimbriata
sp juv/unident
ogilbyi
myops
lamellaris
indica
tricostata
sp1
vitellus
kuhnholtzi
sp1
orientalis
spp
sp2
martensii
undulata cf
sp2
jukesii
flabellata
quinquedentata
maccullochi
geppii
sp juv/unident
spp
limiceps
Biomass Kg %Available
18985
62
4012361
50
135376
55
915
54
658602
56
1421345
52
63493
54
342726
56
1226
46
135435
44
7259
46
3925942
50
6808
54
136339
44
404
55
734504
57
119061
59
58222
53
223485
43
44967
50
151281
58
29227
54
6496
47
19885
57
1021
52
2416172
49
16683
46
61039
51
1570
49
80329
48
562
42
129674
54
1283
49
4093618
51
285546
48
13547972
53
60829
55
785198
51
14640448
54
2542368
48
7775
52
123272
53
1028380
53
23273
45
25049
44
911642
45
8236
54
8642
53
365468
48
36529
54
33318
50
76778
45
259365
50
6856
50
7989
45
828447
48
12459
47
289876
51
62852
49
32047
59
224230
52
19532
38
2378
39
267345
52
15-288
%Exposed
38
25
33
36
40
21
33
38
28
26
28
30
34
23
36
38
38
28
24
29
35
38
28
35
27
28
25
29
26
27
22
35
28
32
22
33
34
34
36
28
30
35
36
25
27
25
33
30
35
33
27
26
31
31
25
27
29
32
26
35
36
20
22
34
%EffortExp
32
32
32
32
32
32
32
32
32
32
32
32
32
31
31
31
31
31
31
31
31
31
31
31
31
31
31
31
30
30
30
30
30
30
30
30
30
30
30
30
30
30
30
30
30
30
29
29
29
29
29
29
29
29
29
29
29
29
29
29
29
29
29
29
GBR Seabed Biodiversity
Group
7
10
9
7
7
7
10
7
20
11
33
7
7
7
21
21
16
10
25
2
2
34
2
10
19
10
12
10
25
28
7
8
11
11
33
12
2
25
9
29
1
11
3
25
8
11
19
31
20
30
19
20
27
20
10
2
10
36
16
27
21
1
10
25
Class
Phaeophyceae
Rhodophyceae
Bivalvia
Crustacea
Cephalopoda
Crustacea
Asteroidea
Crustacea
Chlorophyceae
Phaeophyceae
Polychaeta
Chlorophyceae
Cephalopoda
Actinopterygii
Echinoidea
Gymnolaemata
Gymnolaemata
Rhodophyceae
Cephalopoda
Crustacea
Demospongiae
Actinopterygii
Anthozoa
Chlorophyceae
Crustacea
Gastropoda
Holothuroidea
Gymnolaemata
Crustacea
Gastropoda
Actinopterygii
Crustacea
Rhodophyceae
Phaeophyceae
Gastropoda
Actinopterygii
Chlorophyceae
Actinopterygii
Crustacea
Bivalvia
Cephalopoda
Gastropoda
Actinopterygii
Asteroidea
Demospongiae
Bivalvia
Echinoidea
Gymnolaemata
Demospongiae
Gastropoda
Actinopterygii
Rhodophyceae
Crustacea
Echinoidea
Bivalvia
Crustacea
Rhodophyceae
Actinopterygii
Brachiopoda
Crustacea
Rhodophyceae
Phaeophyceae
Actinopterygii
Actinopterygii
Genus
Dictyotales
Griffithsia
Corbula
Leucosia
Sepiidae
Portunus
Oreasteridae
Sicyonia
Udotea
Sporochnus
Chloeia
Cladophora
Sepiadariidae
Torquigener
Laganum
Thalamoporella
Retelepralia
Lithophyllum
Cephalopoda
Dromidiopsis
Reniochalina
Cynoglossus
Trachyphyllia
Halimeda
Pagurid
Conus
Holothuroidea
Robertsonidra
Porcellanid
Atys
Sorsogona
Arcania
Heterosiphonia
Dictyopteris
Nassarius
Adventor
Halimeda
Upeneus
Dorippe
Corbula
Sepia
Biplex
Cynoglossus
Stellaster
Mycale (arenochalina)
Spondylus
Peronella
Orthoscuticella
Disyringa
Xenophora
Engyprosopon
Amansia
Nursilia
Laganidae
Chama
Portunus
Gracilaria
Minous
Brachiopoda
Trachypenaeus
Gracilaria
Lobophora
Erosa
Tragulichthys
Species
sp
sp
macgillvrayi
formosensis
spp
rubromarginatus
sp1
rectirostris
glaucescens
moorei
flava
sp
sp5
cf pallimaculatus
depressum
spp
mosaica
sp1
spp
edwardsi
stalagmitis
sp4
geoffroyi
sp2
sp17
ammiralis
sp22
spp
sp4154
cylindricus cf
tuberculata
elongata
muelleri
sp2
conoidalis cf
elongatus
borneenses
sp juv/unident
sp7142-12
fortisulcata
whitleyana
pulchellum
ogilbyi
equestris cf
mirabilis
wrightianus
orbicularis cf
spp
sp1
solarioides
grandisquama
glomerata
sp nov
sp3
spp
tenuipes
sp1
trachycephalus
sp1_MTQ
granulosus
sp2
sp
erosa
jaculiferus
Biomass Kg %Available
433762
53
10773
52
205900
47
1341
53
792725
49
5914213
55
2814889
56
259
56
1402555
42
2797072
53
15742
49
39457
52
17032
52
358774
52
2276625
50
56643
53
62
45
21086914
50
750874
47
27212
46
500459
41
466488
49
3171803
46
477419
46
2067
47
17677
51
1113121
51
12430
50
693
49
4963
49
1332608
53
11264
47
1083464
52
1307425
52
2625
52
11470
44
10124447
45
5365
48
601019
45
6462
43
493757
50
59197
47
28505
44
2055943
49
401414
45
103088
50
32469
50
26462
48
9021
46
37081
47
1306624
57
1902406
41
3789
50
190493
44
13590
50
1911035
42
1475393
48
649193
47
79799
41
424353
50
975404
48
6615675
53
175940
47
637207
47
15-289
%Exposed
35
33
25
34
30
34
37
34
23
32
28
33
33
31
30
32
33
33
27
26
23
31
27
30
27
34
26
33
29
28
32
27
31
31
29
22
26
26
24
23
32
26
25
29
31
30
28
30
23
27
30
21
26
24
32
22
29
26
31
25
29
30
32
25
%EffortExp
29
29
29
28
28
28
28
28
28
28
28
28
28
28
28
28
28
28
27
27
27
27
27
27
27
27
27
27
27
27
27
27
27
27
27
26
26
26
26
26
26
26
26
26
26
26
26
26
26
26
26
26
26
26
26
26
26
26
25
25
25
25
25
25
GBR Seabed Biodiversity
Group
21
4
8
7
10
11
1
25
1
31
36
27
1
15
7
6
25
1
2
34
11
36
24
16
20
34
20
3
3
7
30
2
8
23
1
3
11
11
36
25
6
11
4
3
1
7
2
19
8
11
20
16
20
10
14
Class
Actinopterygii
Actinopterygii
Crustacea
Crustacea
Rhodophyceae
Bivalvia
Phaeophyceae
Actinopterygii
Crustacea
Gymnolaemata
Crustacea
Actinopterygii
Crustacea
Crustacea
Actinopterygii
Demospongiae
Ophiuroidea
Demospongiae
Anthozoa
Bivalvia
Crustacea
Demospongiae
Actinopterygii
Crustacea
Crustacea
Gastropoda
Chondrichthyes
Crustacea
Mollusca
Crustacea
Anthozoa
Anthozoa
Crustacea
Actinopterygii
Crustacea
Demospongiae
Crustacea
Demospongiae
Chlorophyceae
Crustacea
Bivalvia
Gastropoda
Gymnolaemata
Gastropoda
Bivalvia
Asteroidea
Gymnolaemata
Demospongiae
Gastropoda
Chlorophyceae
Echinoidea
Gymnolaemata
Bivalvia
Rhodophyceae
Chlorophyceae
Gastropoda
Gastropoda
Chlorophyceae
Bivalvia
Crustacea
Gymnolaemata
Echinoidea
Actinopterygii
Bivalvia
Genus
Choerodon
Zebrias
Pilumnus
Dardanus
Dasya
Chama
Dictyota
Upeneus
Allogalathea
Scuticella
Carinosquilla
Pseudochromis
Actaea
Phalangipus
Synodus
Demospongiae
Ophiothrix
Mycale
Cycloseris
Mimachlamys
Myra
Demospongiae
Priacanthus
Parapenaeopsis
Penaeid unknown
Philine
Dasyatis
Thalamita
Mollusca
Alpheidae
Pteroides
Pteroides
Phalangipus
Choerodon
Parthenope
Spirastrella
Lisoporcellana
Demospongiae
Halimeda
Quollastria
Fragum
Strombus
Beania
Tudivasum
Modiolus
Anthenea
Hiantopora
Ircinia
Nassarius
Udotea
Temnotrema
Cranosina
Parahyotissa
Laurencia
Bornetella
Dolabella
Phos
Avrainvillea
Chlamys
Carinosquilla
Emballotheca
Breynia
Parapercis
Fulvia
Species
cephalotes
craticula
longicornis
callichela var
sp
pulchella
sp1
tragula
elegans
plagiostoma
redacta
quinquedentatus
jacquelinae
filiformis
tectus group
sp11
sp14
sp9
cyclolites
gloriosa
mammillaris
sp89
tayenus
venusta
penaeid unknown
angasi
leylandi
hanseni
eggs
sp2434
sp1
sp2
australiensis
sugillatum
longimanus
sp3
sp3194
sp61
cuneata
subtilis
retusum
variabilis
spp
armigera
sp1
sp1_AIM
intermedia cf
2710
glans cf
flabellum
sp3
coronata
imbricata
sp1
sphaerica
sp1
senticosus
sp1
sp2
thailandensis
spp
desorii
nebulosa
scalata
Biomass Kg %Available
421829
46
206568
49
2472
46
29106
52
365331
49
874718
46
82767
46
843838
47
2905
49
233332
48
44868
41
1907
54
958
46
30500
44
707843
49
48060
40
4806
46
1633320
47
112329
55
24876
44
25080
46
63728
50
1576906
45
10034
51
2674
44
5859
44
176702
44
54500
44
18397
44
164
44
22889
44
9105
44
10355
44
525263
46
8339
49
68071
41
354
43
303087
40
5618
43
5814
36
3345
47
6886
47
4544
43
80914
44
2064840
39
72818
51
936
44
10672550
43
39749
48
642280
44
72568
44
9
46
52508
42
3004213
44
243558
50
1882244
51
145869
46
18392
39
15871
46
87298
48
605916
43
170066
46
705874
47
3799
40
15-290
%Exposed
27
30
26
31
29
27
31
25
29
30
23
26
30
23
29
28
27
30
32
24
27
27
19
32
24
24
24
24
24
24
24
24
24
24
25
20
24
22
30
20
28
28
23
30
21
30
30
23
29
27
24
29
24
24
29
30
26
29
27
22
31
20
29
19
%EffortExp
25
25
25
25
25
25
25
25
25
25
25
25
25
25
25
25
25
25
24
24
24
24
24
24
24
24
24
24
24
24
24
24
24
24
24
24
24
24
24
24
24
24
24
24
24
24
24
24
24
24
24
24
24
24
24
24
24
24
24
24
24
24
24
23
GBR Seabed Biodiversity
Group
17
31
10
35
11
12
10
20
19
10
18
18
37
7
1
1
6
12
36
16
1
2
17
36
11
26
8
21
4
17
2
1
17
1
1
34
4
7
4
13
1
1
8
3
11
3
1
11
31
7
1
12
4
23
14
31
17
2
3
1
17
28
17
10
Class
Cephalopoda
Gymnolaemata
Actinopterygii
Crustacea
Crustacea
Bivalvia
Chlorophyceae
Anthozoa
Anthozoa
Echinoidea
Asteroidea
Actinopterygii
Demospongiae
Crustacea
Actinopterygii
Gymnolaemata
Actinopterygii
Actinopterygii
Holothuroidea
Gymnolaemata
Actinopterygii
Chlorophyceae
Gymnolaemata
Asteroidea
Gymnolaemata
Asteroidea
Crustacea
Actinopterygii
Bivalvia
Cephalopoda
Crustacea
Bivalvia
Demospongiae
Rhodophyceae
Chlorophyceae
Actinopterygii
Crustacea
Actinopterygii
Crustacea
Actinopterygii
Rhodophyceae
Actinopterygii
Crustacea
Crustacea
Gastropoda
Crustacea
Holothuroidea
Crustacea
Anthozoa
Gastropoda
Chlorophyceae
Crustacea
Anthozoa
Actinopterygii
Bivalvia
Actinopterygii
Echinoidea
Gastropoda
Demospongiae
Actinopterygii
Rhodophyceae
Echinoidea
Crustacea
Phaeophyceae
Genus
Sepia
Exochella
Batrachomoeus
Trachypenaeus
Bathypilumnus
Cucullaea
Caulerpa
Antipatharia
Heterocyathus
Gymnechinus
Metrodira
Onigocia
Hyrtios
Thenus
Paramonacanthus
Reteporella
apogon
Repomucenus
Holothuroidea
Lepralia
Grammatobothus
Caulerpa
Aetea
Poraster
Nelliella
Astropecten
Portunus
Pegasus
Dosinia
Sepiadariidae
Thalamita
Chama
Ianthella
Lenormandiopsis
Codium
Pseudomonacanthus
Jonas
Kanekonia
Isopoda
Polydactylus
Rhodophyceae
Pterois
Sicyonia
Parthenope
Gastropoda
Pilumnus
Stichopus
Myra
Junceella
Volva
Ventricaria
Clorida
Truncatoflabellum
Rhynchostracion
Anadara
Upeneus
Peronella
Atys
Ircinia
Paramonacanthus
Laurencia
Clypeaster
Hyastenus
Distromium
Species
smithi
conjuncta cf
dubius/trispinosus
fulvus
pugilator
labiata
sp2
spp
sulcatus cf
epistichus
subulata
sp juv/unident
sp6
australiensis
sp juv/unident
spp
kiensis
sublaevis
spp
elimata
pennatus
taxifolia
capillaris
superbus
spp
sp4_QMS
granulatus
volitans
histrio cf
sp2
sima
sp3
quadrangulata
lorentzii
sp2
peroni
luteanus
queenslandica
sp1
multiradiatus
sp3
russelii (e form)
lancifer
hoplonotus
eggs
semilanatus
horrens
australis
juncea
volva
ventricosa
obtusa
spp
nasus
ferruginea cf
filifer
macroproctes cf
naucum
spp
lowei
sp4
sp3
sebae
flabellatum
Biomass Kg %Available
664816
45
2952
40
124764
43
2663
42
11839
46
238633
43
610672
44
22957
36
407411
46
1300029
47
59546
45
125245
44
78222
38
3019359
46
79035
45
17023
41
21415
45
119448
44
1285178
43
1214
44
1989097
46
49177
47
1456
45
939166
41
15815
45
11617
42
21379
46
59795
44
273127
43
74692
44
49171
39
182124
49
23579428
45
2877749
45
106881
44
110656
44
49552
44
4114
42
10183
45
418667
30
10156570
45
132648
44
80109
43
5110
43
1359905
46
5379
40
2145324
46
138701
45
64983
45
8649
46
2060
36
5954
45
45894
42
578162
45
11718
46
480644
45
135607
43
23300
40
248870
37
565936
43
232107
43
983055
53
288
43
3445399
42
15-291
%Exposed
26
28
28
26
29
21
27
17
24
27
25
24
19
28
26
27
24
23
23
29
28
24
26
28
28
22
26
26
26
25
23
26
25
26
28
26
25
27
27
20
25
26
26
21
28
22
26
24
30
25
26
21
24
24
22
27
25
22
18
27
26
27
24
27
%EffortExp
23
23
23
23
23
23
23
23
23
23
23
23
23
23
23
23
23
23
23
23
23
23
23
23
23
23
23
23
23
22
22
22
22
22
22
22
22
22
22
22
22
22
22
22
22
22
22
22
22
22
22
22
22
22
22
22
22
22
22
22
22
22
22
21
GBR Seabed Biodiversity
Group
17
26
25
20
18
1
1
16
4
16
17
17
1
17
1
16
17
1
17
23
4
34
17
36
1
1
17
18
31
10
34
6
23
4
17
31
17
19
8
6
17
18
17
24
37
20
1
16
1
4
17
1
23
1
4
16
1
16
16
20
6
32
16
34
Class
Rhodophyceae
Demospongiae
Bivalvia
Echinoidea
Crustacea
Bivalvia
Actinopterygii
Actinopterygii
Gymnolaemata
Bivalvia
Crustacea
Anthozoa
Anthozoa
Crustacea
Actinopterygii
Gymnolaemata
Asteroidea
Crustacea
Crustacea
Gymnolaemata
Cephalopoda
Actinopterygii
Bivalvia
Demospongiae
Cephalopoda
Cephalopoda
Cephalopoda
Gastropoda
Gymnolaemata
Crustacea
Actinopterygii
Chlorophyceae
Demospongiae
Gastropoda
Crustacea
Gymnolaemata
Echinoidea
Holothuroidea
Demospongiae
Crustacea
Gymnolaemata
Ophiuroidea
Gymnolaemata
Ophiuroidea
Demospongiae
Asteroidea
Cephalopoda
Gymnolaemata
Gymnolaemata
Ophiuroidea
Anthozoa
Rhodophyceae
Demospongiae
Chlorophyceae
Actinopterygii
Crustacea
Rhodophyceae
Gymnolaemata
Gymnolaemata
Anthozoa
Demospongiae
Echinoidea
Gymnolaemata
Actinopterygii
Genus
Laurencia
Callyspongia
Placamen
Peronella
Camposcia
Laternulidae
Pristotis
Cottapistus
Didymozoum
Lima
Crustacea
Actiniaria
Caryophyllia
Metapenaeopsis
Eurypegasus
Chaperia
Goniasteridae
Thacanophrys
Parthenope
Synnotum
Sepiadarium
Bothidae
Arca
Callyspongia
Sepioloidea
Metasepia
Octopodidae
Gastropoda
Beania
Paguridae
Dactyloptena
Caulerpa
Clathria
Cymatium
Petalomera
Arachnopusia
Nudechinus
Holothuroidea
Crella
Palicoides
Antropora
Ophiochasma
Mimosella
Ophiopsila
Demospongiae
Astropecten
Sepia
Smittina
Escharina
Ophiacantha
Zoanthidae
Haloplegma
Dendrilla
Microdictyon
Tathicarpus
Thacanophrys
Hypoglossum
Crepidacantha
Calyptotheca
Pteroides
Hippospongia
Echinodiscus
Beania
Pseudorhombus
Species
sp3
sp2
sp2
lesueuri
retusa
sp1
obtusirostris
cottoides
spp
sp1
spp
spp
spp
rosea
draconis
spp
spp
sp245
harpax
spp
austrinum
sp juv/unident
sp1
sp6
lineolata
pfefferi
spp
spp
discodermiae cf
sp213
papilio
brachypus
sp9
caudatum
pulchra
spp
spp
sp46
1525
whitei
spp
stellatum
verticillata cf
pantherina
sp88
zebra
plangon
spp
pesanseris
indica cf
spp
duperreyi
sp4
sp1
butleri
longispinus
sp1
spp
spp
sp3
elastica
tenuissimus
plurispinosa cf
dupliciocellatus
Biomass Kg %Available
13655
43
27739
40
119881
44
370761
45
13044
41
613938
44
927471
43
45746
50
3120
49
31065
40
3542
43
397342
43
11317
42
444889
50
19336
46
242
39
13864
43
33062
43
1416
42
497
39
118069
43
133389
42
223887
43
70105
43
177633
42
125341
44
1232237
42
86018
41
564
40
53739
42
325477
42
27068
39
71609
45
6294
41
332
42
17312
34
9200
42
416848
43
81406
55
10059
39
645
42
949536
42
2138
42
4629
36
84681
38
12709
52
1303519
40
1702
43
5030
43
39636
44
5012
42
3474678
44
11362
37
690175
42
63577
42
900
39
511486
43
531
39
228473
41
90797
42
727107
42
116762
40
1481
38
1656810
44
15-292
%Exposed
24
21
23
21
22
26
27
29
27
27
24
24
27
25
26
27
24
26
24
22
25
23
27
23
26
24
24
23
25
26
25
24
24
25
25
25
25
24
31
23
25
22
25
19
20
26
26
24
23
25
25
24
20
23
25
24
25
25
26
21
22
25
25
24
%EffortExp
21
21
21
21
21
21
21
21
21
21
21
21
21
21
21
21
21
21
21
21
21
21
21
21
21
21
21
21
21
21
21
21
21
21
21
21
21
21
21
21
21
21
21
21
21
20
20
20
20
20
20
20
20
20
20
20
20
20
20
20
20
20
20
20
GBR Seabed Biodiversity
Group
16
17
32
1
1
30
15
37
4
14
3
11
17
16
1
36
2
17
10
1
17
34
4
18
17
16
20
1
7
1
18
17
37
37
16
28
16
16
23
4
18
16
2
36
2
18
1
20
23
18
6
16
16
16
16
16
4
2
15
18
4
17
17
16
Class
Gymnolaemata
Rhodophyceae
Actinopterygii
Crustacea
Gastropoda
Rhodophyceae
Gastropoda
Demospongiae
Crustacea
Crustacea
Bivalvia
Crustacea
Crustacea
Actinopterygii
Crustacea
Demospongiae
Chlorophyceae
Crustacea
Gymnolaemata
Actinopterygii
Bivalvia
Actinopterygii
Actinopterygii
Echinoidea
Rhodophyceae
Stenolaemata
Gastropoda
Chlorophyceae
Actinopterygii
Holothuroidea
Actinopterygii
Anthozoa
Gastropoda
Demospongiae
Calcarea
Crustacea
Anthozoa
Gymnolaemata
Holothuroidea
Liliopsida
Asteroidea
Rhodophyceae
Rhodophyceae
Demospongiae
Chlorophyceae
Ophiuroidea
Cephalopoda
Annelida
Animalia
Gastropoda
Crustacea
Crustacea
Gymnolaemata
Actinopterygii
Gymnolaemata
Holothuroidea
Bivalvia
Chlorophyceae
Ophiuroidea
Crustacea
Actinopterygii
Gymnolaemata
Chlorophyceae
Gymnolaemata
Genus
Micropora
Coelarthrum
Engyprosopon
Dardanus
Chicoreus
Peyssonnelia
Distorsio
Hyattella
Trachypenaeus
Scyllarus
Pteria
Parthenope
Achaeus
Choerodon
Metapenaeopsis
Sponge substrate
Halimeda
Hyastenus
Calyptotheca
Antennarius
Glycymeris
Lepidotrigla
Pseudorhombus
Salmaciella
Hypnea
Mecynoecia
Cymatium
Caulerpa
Orbonymus
Holothuria
Onigocia
Turbinaria
Cypraea
Dysidea
Calcarea
Metapenaeopsis
Solenocaulon
Rhynchozoon
Pseudocolochirus
Halophila
Luidia
Hydrolithon
Peyssonnelia
Demospongiae
Caulerpa
Ophiopsammus
Sepia
Annelida
Encrusting
Fusinus
Portunus
Hyastenus
Vesicularia
Onigocia
Pleurocodonellina
Holothuroidea
Circe
Penicillus
Euryalida
Pontocaris
Apogon
Plesiocleidochasma
Halimeda
Figularia
Species
variperforata cf
sp1
maldivensis
callichela
sp1
inamoena
reticulata
intestinalis
curvirostris
sp3418
coturnix cf
sp 67
sp5993
venustus
lamellata
substrate
bikensis
campbelli
wasinensis cf
striatus
hedleyi
japonica-like
diplospilus
oligopora
sp1
spp
pfeifferanium
cupressoides
rameus
dofleinii
sp b
spp
walkeri cf
sp10
calcareous sp4
metapenaeopsis sp
sp1
spp
violaceus
capricorni
maculata
reinboldii
sp1
sp17
sertularioides
yoldii
papuensis
spp
conglomerate
colus
argentatus
convexus
papuensis_AIM
cf macrolepis
spp
sp36
sp1
nodulosus
fragment
orientalis
timorensis
spp
gracilis
clithridiata cf
Biomass Kg %Available
14843
40
924225
42
444474
38
66220
37
533290
40
974895
38
57223
44
454226
40
565850
41
29264
38
1806
38
202017
43
2086
41
386846
36
263643
39
50696
42
7008553
31
43855
41
121815
52
31064
44
186408
42
1074762
43
308042
42
100584
38
136462
42
2764
40
4305
37
599194
39
131532
44
957738
37
45271
40
8638653
42
5844
37
23982
36
41835
37
983880
42
244307
40
99667
40
2281916
36
76915
41
365271
41
9313083
38
2379581
37
18469
45
265840
37
873462
41
363864
40
5213418
42
47582896
38
463836
42
470239
38
3668
40
87465
40
167134
33
1414
39
4379
37
93117
40
956607
33
47945
40
15667
41
123542
38
62133
42
5839218
36
1135
33
15-293
%Exposed
26
23
25
25
25
22
22
22
24
19
20
25
25
25
25
22
19
23
32
22
25
23
21
25
23
23
17
24
24
24
22
24
19
19
20
22
22
23
19
23
23
24
22
22
19
22
23
21
21
24
23
24
21
23
23
23
22
18
20
21
23
22
23
23
%EffortExp
20
20
20
20
20
20
20
20
20
20
20
20
20
20
20
20
20
20
20
20
20
20
20
20
20
20
20
20
19
19
19
19
19
19
19
19
19
19
19
19
19
19
19
19
19
19
19
19
19
19
19
19
19
19
19
19
19
19
19
19
19
19
19
19
GBR Seabed Biodiversity
Group
29
16
16
23
16
2
4
6
4
4
30
16
25
2
34
6
16
17
8
16
4
37
5
16
23
16
13
17
16
15
10
18
5
23
23
16
37
16
15
18
4
20
4
23
10
17
16
32
24
17
34
20
16
36
16
16
16
5
17
5
5
16
37
37
Class
Crustacea
Crustacea
Gymnolaemata
Asteroidea
Anthozoa
Ophiuroidea
Bivalvia
Chlorophyceae
Echinoidea
Crustacea
Demospongiae
Gymnolaemata
Crustacea
Actinopterygii
Gymnolaemata
Gymnolaemata
Gymnolaemata
Crustacea
Gastropoda
Echinoidea
Crustacea
Crustacea
Stenolaemata
Demospongiae
Demospongiae
Gymnolaemata
Crustacea
Echinoidea
Asteroidea
Crustacea
Bivalvia
Gastropoda
Ophiuroidea
Gymnolaemata
Octocorallia
Gymnolaemata
Demospongiae
Stenolaemata
Demospongiae
Actinopterygii
Actinopterygii
Actinopterygii
Actinopterygii
Asteroidea
Actinopterygii
Anthozoa
Gymnolaemata
Gymnolaemata
Actinopterygii
Phaeophyceae
Gymnolaemata
Crustacea
Gymnolaemata
Actinopterygii
Gymnolaemata
Gymnolaemata
Gymnolaemata
Anthozoa
Crustacea
Crustacea
Gymnolaemata
Gymnolaemata
Gastropoda
Calcarea
Genus
Iphiculus
Thacanophrys
Trypostega
Pentaceraster
Dichotella
Ophiomaza
Cardita
Phyllodictyon
Temnopleuridae
Izanami (matuta)
Hyattella
Beania
Gonodactylaceus
Richardsonichthys
Cribralaria
Teuchopora
Smittipora
Micippa
Xenophora
Temnotrema
Takedana
Gaillardiellus
Nevianipora
Hyattella
Pseudoceratina
Celleporina
Pagurid
Nudechinus
Euretaster
Arcania
Plicatula
Chicoreus
Ophiuroidea
Nolella
Octocorallia
Parasmittina
Fascaplysinopsis
Mesonea
Raspailia
Halicampus
Hippocampus
Pseudorhombus
Samaris
Goniodiscaster
Lethrinus
Sphenopus
Macropora
Savignyella
Elates
Dictyota
Exostesia
Lupocyclus
Pleurocodonellina
Lagocephalus
Didymosellidae
Conopeum
Mucropetraliella
Solenocaulon
Parthenope
Oreophorus
Conescharellina
Arthropoma
Ceratosoma
Calcarea
Species
spongiosus
sp879
spp
gracilis
sp1
cacaotica cf
sp1
sp1
sp2_QMS
inermis
intestinalis (form b)
regularis
graphurus
leucogaster
spp
verrucosa cf
abyssicola cf
philyra
cerea cf
bothryoides
eriphioides
rueppelli
spp
sp2
sp6
spp
sp2358-1
sp4_MTQ
insignis
gracilis
chinensis cf
banksii cf
spp
spp
spp
spp
sp1
radians
sp2
grayi
queenslandicus
argus
cristatus
rugosus cf
genivittatus
marsupialis
spp
spp
ransonnetii
sp2
didomatia
rotundatus
laciniosa cf
sceleratus
spp
spp
serrata cf
sp2
longispinus
reticulatus
spp
spp
tenue
calcareous sp5
Biomass Kg %Available
364
43
336
39
1350
38
2811190
40
72514
38
11435
37
160513
39
47831
38
31430
36
238633
39
579612
38
763
38
6851
49
46959
37
4312
42
1197
39
2036
37
155538
39
289919
30
41067
38
3374
33
5494
35
3235
40
408228
38
189002
40
5556
38
15945
44
133160
38
205115
40
14190
39
3083202
47
52142
37
12185
38
1536
39
71346
34
142379
38
367349
35
80
37
648941
38
8694
39
9042
37
353532
38
151891
43
90526
41
6198005
45
923474
48
13357
34
211
37
431203
30
150709
33
348205
43
73747
44
16498
35
210666
44
173000
35
550
37
167605
38
3573
37
12376
38
89789
37
667
41
215
35
15808
36
28284
37
15-294
%Exposed
18
22
23
20
22
20
23
21
23
23
21
22
26
22
22
21
23
21
21
22
22
18
23
22
22
23
20
22
21
19
29
22
19
19
18
21
18
21
19
21
21
15
23
20
24
29
22
21
12
18
22
21
22
22
22
21
22
18
21
18
22
21
18
21
%EffortExp
18
18
18
18
18
18
18
18
18
18
18
18
18
18
18
18
18
18
18
18
18
18
18
18
18
18
18
18
18
18
18
18
18
18
18
18
18
18
18
18
17
17
17
17
17
17
17
17
17
17
17
17
17
17
17
17
17
17
17
17
17
17
17
17
GBR Seabed Biodiversity
Group
18
16
4
23
3
5
3
16
36
6
23
16
23
16
31
4
24
16
37
1
37
16
17
16
13
6
5
23
17
5
17
37
2
23
18
32
30
37
18
4
15
31
21
3
16
4
17
16
23
16
16
16
6
6
18
24
5
5
32
23
16
5
23
4
Class
Chlorophyceae
Gymnolaemata
Crustacea
Echinoidea
Gymnolaemata
Actinopterygii
Rhodophyceae
Gymnolaemata
Actinopterygii
Gastropoda
Gymnolaemata
Phaeophyceae
Gymnolaemata
Gymnolaemata
Holothuroidea
Gymnolaemata
Bivalvia
Gymnolaemata
Demospongiae
Gymnolaemata
Demospongiae
Gymnolaemata
Ophiuroidea
Gymnolaemata
Bivalvia
Actinopterygii
Actinopterygii
Holothuroidea
Crustacea
Actinopterygii
Gymnolaemata
Demospongiae
Anthozoa
Actinopterygii
Echinoidea
Gymnolaemata
Demospongiae
Demospongiae
Ophiuroidea
Crustacea
Echinoidea
Actinaria
Actinopterygii
Gastropoda
Asteroidea
Gymnolaemata
Gymnolaemata
Asteroidea
Cephalopoda
Gymnolaemata
Gymnolaemata
Gymnolaemata
Crustacea
Demospongiae
Gymnolaemata
Actinopterygii
Actinopterygii
Anthozoa
Crustacea
Gymnolaemata
Polychaeta
Demospongiae
Ascidiacea
Echinoidea
Genus
Halimeda
Calloporina
Penaeus
Salmacis
Biflustra
Rogadius
Cryptonemia
Smittoidea
Apogon
Murex
Celleporaria
Sargassum
Fenestrulina
Cellaria
Holothuroidea
Nellia
Pitar
Lacernidae
Tethya
Cosciniopsis
Demospongiae
Elzerina
Ophiothrix
Schizomavella
Solen
Parapercis
Apogon
Cercodermas
Diogenidae
Choerodon
Adeonella
Demospongiae
Scolymia
Liocranium
Temnopleuridae
Amastigia
Dysidea
Acanthella
Ophiarachnella
Myrine
Lovenia
Anemone
Sillago
Scutus
Ophidiasteridae
Hippopetraliella
Triphyllozoon
Iconaster
Photololligo
Phonicosia
Tubulipora
Schizomavella
Leucosia
Demospongiae
Steginoporella
Arnoglossus
Diagramma
Nephthyigorgia
Actumnus
Microporella
Polychaete
Demospongiae
Ascidiacea
Pseudoboletia
Species
discoidea
sigillata
longistylus
belli
savartii
patriciae
sp
incucula cf
truncatus
tenuirostrum cf
sp1_QMS
sp
spp
spp
sp44
tenella cf
sp2
sp2
sp2
spp
conglomerate
blainvillii cf
sp6
australis cf
siphons only
snyderi
sp juv/unident
anceps
sp2
sp juv/unident
lichenoides cf
sp14
spp
praepositum
sp5
rudis
sp5
cavernosa
infernalis cf
kesslerii
elongata
sp9
ingenuua
unguis
sp1
magna cf
spp
longimanus
chinensis
circinata
spp
triquetra cf
magna
sp13
spp
waitei
pictum labiosum
sp1
setifer
spp
spp
fragment
spp
indiana
Biomass Kg %Available
70817
36
8060
32
1997680
38
1045645
37
2006
36
557252
40
38912
34
481
37
736274
42
648731
42
213092
36
1025790
35
37498
40
42280
39
38221
37
24071
40
7904
41
380
33
15694
32
621
32
442636
34
824
30
2696
36
9036
37
69535
49
4966
31
11812
36
95366
33
69702
35
2959
36
84065
40
15430
37
14597
34
81364
37
41136
38
3429
34
90011
32
37239
35
35459
37
1228
39
92701
41
6537
39
400298
48
12865
33
12687
34
10537
31
150801
34
17227
33
350906
37
91
34
707
36
1490
34
13372
36
302059
31
13187
33
2352
37
185953
35
7191
35
1499
32
1529
33
6371
34
2417
34
6019123
32
45457
43
15-295
%Exposed
21
22
20
20
18
21
16
20
20
20
20
21
21
21
19
20
18
21
17
20
18
21
20
20
23
20
17
19
20
17
23
18
19
20
18
17
17
19
20
20
19
19
24
19
19
20
19
19
19
20
19
19
17
20
19
17
16
17
19
19
17
18
18
24
%EffortExp
17
17
17
17
17
17
17
17
17
17
17
17
17
17
17
17
17
17
16
16
16
16
16
16
16
16
16
16
16
16
16
16
16
16
16
16
16
16
16
16
16
16
16
16
16
16
15
15
15
15
15
15
15
15
15
15
15
15
15
15
15
15
15
15
GBR Seabed Biodiversity
Group
16
12
16
28
4
31
6
16
20
16
32
16
2
13
23
30
20
23
30
26
16
16
16
32
23
23
5
15
5
16
23
16
5
16
35
37
34
16
5
34
3
5
32
23
5
5
14
37
2
30
26
16
23
5
5
16
23
23
37
16
6
23
32
18
Class
Actinopterygii
Crustacea
Holothuroidea
Gastropoda
Gymnolaemata
Chlorophyceae
Crustacea
Gymnolaemata
Gymnolaemata
Gymnolaemata
Actinopterygii
Ophiuroidea
Holothuroidea
Actinopterygii
Gymnolaemata
Demospongiae
Anthozoa
Crustacea
Demospongiae
Demospongiae
Gymnolaemata
Anthozoa
Gymnolaemata
Gymnolaemata
Bivalvia
Crustacea
Gastropoda
Bivalvia
Gastropoda
Anthozoa
Demospongiae
Echinoidea
Asteroidea
Bivalvia
Actinopterygii
Demospongiae
Actinopterygii
Gymnolaemata
Gastropoda
Crustacea
Demospongiae
Bivalvia
Actinopterygii
Holothuroidea
Gymnolaemata
Gymnolaemata
Crustacea
Demospongiae
Actinopterygii
Crustacea
Cyanophyceae
Gymnolaemata
Actinopterygii
Crustacea
Actinopterygii
Gymnolaemata
Gymnolaemata
Gymnolaemata
Demospongiae
Asteroidea
Actinopterygii
Anthozoa
Gymnolaemata
Crinoidea
Genus
Abalistes
Urnalana
Holothuroidea
Atys
Retiflustra
Microdictyon
Parthenope
Scrupocellaria
Tetraplaria
Porina
Apogon
Euryale
Actinopyga
Apogon
Cyclostomata
Dendroceratid
Studeriotes
Charybdis
Niphates
Dysidea
Adeonellopsis
Dichotella
Amathia
Filicrisia
Malleus
Actaea
Haustellum
Nemocardium
Latirus
Umbellulifera
Callyspongia
Temnopleurus
Anseropoda
Lima
Rogadius
Dendrilla
Nemipterus
Metroperiella
Chicoreus
Myra
Callyspongia
Liochonca
Fistularia
Holothuroidea
Bugula
Bugula
Portunus
Clathria (thalysias)
Paramonacanthus
Carinosquilla
Lyngbya
Tetraplaria
Apogon
Carinosquilla
Engyprosopon
Phidoloporidae
Puellina
Hippopodina
Demospongiae
Tamaria
Choerodon
Mopsella
Parmularia
Crinoidea
Species
stellatus
whitei
sp43
sp1
spp
umbilicatum
turriger
spp
ventricosa cf
vertebralis cf
9(dg)
asperum
spinea cf
cavitiensis
spp
sp1
sp2
jaubertensis
sp17
sp3
pentapora
gemmacea
spp
geniculata
albus
Tuberculosa
tweedianum
bechei
paetelianus cf
sp1
sp23
alexandri
rosacae cf
vulgaris
pristiger
sp5
theodorei
spp
spp
eudactyla
schultzi
polita
commersoni
sp38
robusta cf
spp
spinipes
vulpina
oblongus
australiensis
sp
immersa
cf fuscomaculatus
carita
sp juv/unident
sp1
spp
feegeensis cf
sp6
sp3
frenatus
sp1
spp
spp
Biomass Kg %Available
766691
35
41726
47
70289
32
32478
42
33182
36
1285617
31
2190
27
12955
36
9058
36
1154
34
62289
33
905776
36
5107122
29
7972
47
466
33
86107
32
11794
41
76639
31
306985
33
50439
28
7400
34
21485
35
668802
35
959
32
311540
31
6454
28
12376
35
14205
37
21199
35
775140
32
87841
39
57525
31
34571
33
21760
30
447881
34
72702
32
5326939
35
6097
33
49916
32
5119
42
451419
31
11024
37
8270
31
68431
30
2346
33
2368
34
5017
35
1011817
34
139092
41
66990
28
50986
21
30440
29
196097
32
10419
38
14226
34
10087
28
274
30
1957
32
873739
30
22911
30
96061
37
81629
32
2514
30
2425469
32
15-296
%Exposed
19
22
19
17
19
19
15
18
15
18
16
19
16
22
19
17
15
15
16
14
18
18
17
18
17
18
18
16
16
18
17
18
16
18
14
14
17
18
17
17
16
16
15
18
17
18
12
17
20
13
12
17
16
17
17
17
17
17
15
17
16
17
17
16
%EffortExp
15
15
15
15
15
15
15
15
15
15
15
15
14
14
14
14
14
14
14
14
14
14
14
14
14
14
14
14
14
14
14
14
14
14
14
14
14
14
14
14
14
14
14
14
14
14
13
13
13
13
13
13
13
13
13
13
13
13
13
13
13
13
13
13
GBR Seabed Biodiversity
Group
31
16
32
5
37
5
37
32
16
23
23
23
16
6
23
23
32
15
2
25
16
32
3
2
13
15
23
23
23
28
3
3
2
5
32
37
31
37
18
23
32
23
13
16
6
32
16
32
6
37
2
23
37
15
38
23
5
28
20
15
23
23
26
18
Class
Actinopterygii
Gymnolaemata
Chlorophyceae
Echinoidea
Gymnolaemata
Gymnolaemata
Gastropoda
Actinopterygii
Gymnolaemata
Gymnolaemata
Gymnolaemata
Actinopterygii
Crustacea
Actinopterygii
Gymnolaemata
Gymnolaemata
Actinopterygii
Crustacea
Chlorophyceae
Asteroidea
Gymnolaemata
Actinopterygii
Echinoidea
Chlorophyceae
Actinopterygii
Gymnolaemata
Gymnolaemata
Gymnolaemata
Ophiuroidea
Anthozoa
Bivalvia
Demospongiae
Chlorophyceae
Gymnolaemata
Gymnolaemata
Demospongiae
Anthozoa
Demospongiae
Actinopterygii
Holothuroidea
Anthozoa
Gymnolaemata
Crustacea
Actinopterygii
Crustacea
Actinopterygii
Gymnolaemata
Gymnolaemata
Crustacea
Ophiuroidea
Chlorophyceae
Gymnolaemata
Demospongiae
Crustacea
Gymnolaemata
Demospongiae
Gymnolaemata
Demospongiae
Actinopterygii
Crustacea
Hydrozoa
Gymnolaemata
Crustacea
Anthozoa
Genus
Lutjanus
Catenicella
Struvea
Asthenosoma
Canda
Celleporidae
Chicoreus
Engyprosopon
Hippaliosina
Bicrisia
Crisia
Apogon
Parthenope
Goby
Caberea
Celleporaria
Dactyloptena
Pandalidae
Halimeda
Tamaria cf
Bugula
Apogon
Prionocidaris
Halimeda
Siganus
Schizomavella
Margaretta
Thornleya
Placophiothrix
diaseris
Arca
Dendrilla
Caulerpa
Codonellina
Chaperiopsis
Demospongiae
Subergorgia
Chondrilla
Pentapodus
Holothuroidea
Acanthogorgia
Celleporaria
Eucrate
Pentapodus
Naxoides
Dendrochirus
Escharoides
Cigclisula
Lupocyclus
Ophiopeza
Udotea
Catenicella
Fascaplysinopsis
Carid
Amathia
Callyspongia
Hippopetraliella
Gelliodes
Upeneus
Quollastria
Hydroida
Turbicellepora
Thalamita
Dendronephthya
Species
adetii
spp
elegans
sp1
spp
spp
territus cf
latifrons
spp
spp
elongata cf
brevicaudatus
sp32091
sp juv/unident
spp
spp
orientalis
sp916
gigas
sp3
dentata cf
septemstriatus
bispinosa
opuntia
canaliculatus
inclusa cf
spp
spp
sp2
distorta cf
avellana_MTQ
sp6
serrulata
montferrandii
spp
sp53
suberosa
sp1
paradiseus
sp30
sp1
sp1_AIM
affinis
nagasakiensis
taurus
brachypterus
longirostris
spp
tugelae
spinosa cf
orientalis
sp1_CMR
sp3
sp4931
crispa
sp26
concinna
sp1
luzonius
gonypetes
spp
laevis
parvidens
spp
Biomass Kg %Available
1042450
33
44158
34
25785
21
746377
29
27116
33
1453
33
15996
31
13430
29
59823
25
39
30
175749
32
125557
28
1191
25
1042
34
15510
30
15622360
31
168574
29
11205
38
3224417
29
66687
43
212133
29
47319
31
210039
29
269000
25
377618
42
3672
33
43670
30
32344
30
3343
29
1117
37
325791
29
243595
29
346047
25
42
34
49
25
442819
31
14020
32
17714
29
2615371
34
1110532
27
10489
27
548232
26
2137
48
1132057
23
3086
26
49521
27
15371
27
2170
26
22309
30
347
26
842312
34
46744
34
380560
39
714
32
22713
27
128191
25
6873
34
74197
31
584420
36
21257
31
338440
26
648863
25
112316
20
1302022
29
15-297
%Exposed
21
16
18
14
16
15
15
15
16
14
16
14
16
15
16
16
15
17
14
20
16
14
14
15
19
15
16
15
15
13
15
14
13
14
15
12
16
13
16
15
13
14
18
15
13
13
14
14
12
12
16
14
12
13
14
14
14
11
15
12
13
13
6
13
%EffortExp
13
13
13
13
13
13
13
13
13
13
13
12
12
12
12
12
12
12
12
12
12
12
12
12
12
12
12
12
11
11
11
11
11
11
11
11
11
11
11
11
11
11
11
11
11
11
11
11
11
10
10
10
10
10
10
10
10
10
10
10
10
10
10
10
GBR Seabed Biodiversity
Group
5
32
6
13
37
5
23
6
15
12
16
3
24
31
3
23
26
3
24
23
35
32
26
23
31
31
23
18
23
20
24
23
3
3
17
23
15
3
32
32
12
16
15
16
37
2
23
26
23
15
23
23
28
23
15
28
15
15
23
37
23
37
15
15
Class
Crustacea
Crustacea
Crustacea
Bivalvia
Demospongiae
Actinopterygii
Gymnolaemata
Crustacea
Demospongiae
Bivalvia
Gymnolaemata
Anthozoa
Actinopterygii
Gastropoda
Crustacea
Gymnolaemata
Demospongiae
Gymnolaemata
Actinopterygii
Gymnolaemata
Cephalopoda
Chlorophyceae
Bivalvia
Demospongiae
Gymnolaemata
Gymnolaemata
Gymnolaemata
Actinopterygii
Asteroidea
Actinopterygii
Actinopterygii
Demospongiae
Crustacea
Demospongiae
Anthozoa
Anthozoa
Demospongiae
Bivalvia
Crustacea
Actinopterygii
Actinopterygii
Anthozoa
Actinopterygii
Crustacea
Cephalopoda
Actinopterygii
Gymnolaemata
Gastropoda
Gymnolaemata
Actinopterygii
Calcarea
Demospongiae
Anthozoa
Anthozoa
Demospongiae
Crustacea
Crustacea
Ophiuroidea
Crustacea
Demospongiae
Anthozoa
Demospongiae
Crustacea
Crustacea
Genus
Diogenidae
Paguridae
Solenocera
Solen
Dysidea
Siphamia
Celleporaria
Oncinopus
Spirastrella
Globivenus
Stylopoma
Euplexaura
Sirembo
Bolma
Hyastenus
Hippomenella
Tethya
Micropora
Lepidotrigla
Steginoporella
Photololligo
Caulerpa
Fulvia
Demospongiae
Telopora
Crassimarginatella
Euthyrisella
Parupeneus
Goniasteridae
Lutjanus
Paramonacanthus
Demospongiae
Austrolabidia
Oceanapia
Iciligorgia
Melithaea
Demospongiae
Arca
Naxoides
Apogon
Choerodon
Junceella
Upeneus
Barnacle
Loligo
Oxycheilinus
Adenifera
Terebellum
Sinupetraliella
Parapriacanthus
Clathrina
Demospongiae
Heteropsammia
Mopsella
Demospongiae
Calappa
Arcania
Ophionereis
Thalamita
Coelocarteria
Echinogorgia
Demospongiae
Pilumnus
Solenocera
Species
sp379
sp444
pectinata
sp3
arenaria
versicolor
sp2_QMS
aranea
sp2
embrithes cf
spp
sp6
imberbis
aureola
elatus
avicularis
sp3
angusta cf
calodactyla
magnilabris
sp1
racemosa
undatopicta
sp27
spp
spp
obtecta
heptacanthus
sp5
vitta
filicauda
sp26
gracilipes
tubes only
sp1
sp2
sp16
navicularis
sp53287
capricornis
monostigma
sp2
moluccensis
sp1
sp1
bimaculatus
armata
terebellum
spp
ransonneti
sp1
sp10
cochlea
sp2
sp146
sp 1984
heptacantha
semoni cf
intermedia
singaporensis
sp5
sp109
spinicarpus
choprai
Biomass Kg %Available
318
30
3476
22
14069
29
35340
45
267675
26
14154
27
87816
28
8497
23
528299
35
113931
42
13328
24
483859
36
91767
34
48007
24
18772
33
67
27
17612
28
1459
24
873992
25
43117
20
126860
32
485706
25
14593
24
12702
26
442
28
222
27
191941
24
331277
23
4161
19
316853
37
8764207
39
22270
21
12992
33
3664
33
129051
31
61500
22
43858
34
7242
22
2151
20
73308
21
285266
36
137591
28
337152
25
5009088
23
87486
30
48838
19
196265
19
635
31
3960
30
210816
23
24817
19
184966
18
22620
30
26395
24
6870
31
16292
33
962
23
7324
30
3456
22
110035
11
11065
26
119911
25
1620
26
124686
26
15-298
%Exposed
13
12
11
17
10
12
12
12
14
15
12
15
9
12
13
12
11
11
10
11
14
10
10
11
14
14
11
10
9
13
11
9
12
13
13
10
12
9
9
9
9
12
7
12
11
8
9
6
9
8
8
8
8
9
9
6
5
8
8
4
8
6
7
5
%EffortExp
9
9
9
9
9
9
9
9
9
9
9
9
8
8
8
8
8
8
8
8
8
8
8
8
8
8
7
7
7
7
7
7
7
7
7
7
7
7
7
7
7
7
6
6
6
6
6
6
6
5
5
5
5
5
5
5
5
4
4
4
4
4
4
4
GBR Seabed Biodiversity
Group
17
18
3
37
3
37
23
3
32
37
18
2
14
30
3
Class
Crustacea
Ophiuroidea
Actinopterygii
Demospongiae
Anthozoa
Demospongiae
Anthozoa
Actinopterygii
Actinopterygii
Demospongiae
Actinopterygii
Chlorophyceae
Crustacea
Echinoidea
Crustacea
Genus
Thacanophrys
Ophiarachnella
Centrogenys
Oceanapia
Carijoa
Cinachyrella
Echinogorgia
Inegocia
Crossorhombus
Cinachyrella
Canthigaster
Dictyosphaeria
Cloridina
Mespilia
Metapenaeopsis
Species
sp165
paucigranula cf
vaigiensis
sp21
sp1
sp1
sp3
harrisii
howensis
australiensis
rivulata
cavernosa
chlorida
globulus
novaeguineae
Biomass Kg %Available
885
25
15532
18
20426
23
5236461
25
78340
25
3327419
25
110904
18
217420
22
76179
22
125583
20
31856
12
1430663
21
374
29
22836
12
16882
9
15-299
%Exposed
6
5
5
5
5
5
5
5
4
5
4
2
3
3
2
%EffortExp
3
3
3
3
3
3
3
3
3
3
3
2
2
2
2
GBR Seabed Biodiversity
15-300
AUSTRALIAN INSTITUTE
OF MARINE SCIENCE
Australian Institute of
Marine Science
PMB 3, Townsville MC
TOWNSVILLE, Qld.
4810
Australia
Marine & Atmospheric
Research
Mathematics &
Information Sciences
233 Middle Street
CLEVELAND, Qld.
4163 Australia
Queensland
Department of Primary
Industries
Northern Fisheries
Centre
Tingara Street
CAIRNS, Qld. 4870
Australia
Queensland Museum
SOUTH BRISBANE,
Qld. 4101
Australia
Museum of Tropical
Queensland
TOWNSVILLE, Qld.
4810