Coral micro-fragmentation assays for optimizing active reef restoration efforts

In-situ nursery

The field-based nursery was located on Moku o Lo‘e (Coconut Island) at the Hawai'i Institute of Marine Biology (HIMB) in Kānéohe Bay (Figs. 1A, 1B) and was constructed to conduct research on improving the time and cost efficiency of reef restoration. The coral nursery consists of floating walkways surrounding and supporting suspended midwater platforms for coral cultivation. It was constructed in 2017 from recycled materials salvaged from decommissioned aquaculture and marine mammal pens. The primary source of the over 1,000 corals housed in the nursery is from retired pens and other ‘corals of opportunity’ relocated to the nursery from marine debris or other salvage which would otherwise be discarded. Collection and monitoring work was approved by the Division of Aquatic Resources (DAR) under Special Activities Permits (SAP) 2018-03, 2019-16, and 2020-25.

Ex-situ nursery

The Division of Aquatic Resources’ Hawai'i Coral Restoration Nursery’s (HCRN) ex-situ nursery is located at the Ānuenue Fisheries Research Center (AFRC) on Sand Island (Figs. 1C, 1D). It was built for the purpose of improving methods of coral culture and outplanting, to restore and conserve Hawai'i’s coral reefs, and consists of small to large indoor and outdoor tanks with varying levels of filtration and control of temperature, lighting, water chemistry, and biotic communities. A full-time staff of professional aquarists provide daily husbandry for the maintained corals. The facility primarily outplants a range of Hawaiian species ranging in size from 15 cm to over 1 m in length. The source material for this land-based nursery is predominantly coral that would otherwise have been destroyed from various state and federal projects such as harbor improvements or dredging.

Outplanting sites

Each nursery had an adjacent natural reef outplanting site. The in-situ nursery outplanting site (Fig. 1B) was located in an enclosed bay with low water flow and composed of a sandy substrate situated next to an existing mature M. capitata and P. compressa coral reef structure with roughly 70% coral cover (Caruso et al., 2021; McGilly, 2019). This site was selected so assays could be placed on the adjacent sand flat rather than affixed to the reef. The site for the ex-situ nursery outplanting (Fig. 1D) was located in Māmala Bay on the south shore of O‘ahu with hard bottom pavement. Due to high water flow conditions here (Grigg, 1998) assays were fixed to the reef with Z-SPAR A-788 splash zone epoxy (Pettit Paint, Rockaway, NJ, USA) in open patches among live coral colonies.

Assay design and deployment

The pyramid assay design was a modified smaller version of the ones regularly used at the HCRN with three sides and a flat top, leaving space for a label (Fig. 2). The pyramid shape was selected because it reduces the horizontal surface upon which sediment is retained, while minimizing the 3D surface area over which corals must grow to fuse and rapidly cover the artificial substrate. Once completely covered, this design also blends well into the reef substrate which is why such coral modules were selected for outplanting by the HCRN. A polyvinyl chloride sheet (Celtec®) was cut into a form consisting of three sides and a top (Fig. 2) and joined together by drilling holes on the edge and joining them together with zip ties. The forms were nestled upside-down into tubs of sand and a fast-setting concrete (Portland Type II cement) was poured into them. Once the concrete set the zip ties joining the forms sides were cut and the concrete pyramids were removed. All pyramids were soaked in seawater for a month to cure, and then allowed to air dry in the sun for an additional month. The labels were made on a Dymo® label maker and affixed with All-Fix two-part epoxy putty prior to coral fragment attachment.

Three unique M. capitata and three P. compressa parent colonies, roughly 30 cm in diameter, were collected at each site from within or adjacent to the nursery (Figs. 1A, 1C, and Appendix 1A, 1B) resulting in six parent colonies from a combination of natural reef and coral nursery origin at each location. The standard quarantine period was also observed for the ex-situ nursery samples, whereby any parent colonies not already in the nursery had all epifauna removed before being placed in a quarantine tank where they were required to remain clear of aquatic invasive species (AIS), parasites or visible disease for at least 1 month before experimentation (Appendix 1A, 1B). Coral predatory nudibranchs (Phestilla sp.) emerging during quarantine on the P. compressa parent colonies at the ex-situ nursery resulted in additional cleaning and delayed fragmentation and deployment by 2 months at this site.

The parent corals were micro-fragmented to yield seven small (≈1 × 1 cm or 1 cm² each, 7 cm² total), three medium (≈1.75 × 1.75 cm or 3 cm² each, 9 cm² total), and one large (≈3 × 3 cm, 9 cm² total) fragment(s) per pyramid (Fig. 2). Each of the six unique colonies per site were fragmented into three identical replicates (A, B and C), which were then outplanted at 0, 4 and 8 months resulting in 27 assays per species (54 total) per reef outplanting site (see Appendix 2 for detailed timeline of coral assay deployment). The assays outplanted at 0 months (T0) had no nursery time and were outplanted immediately following confirmation of solid attachment of the fragments (Figs. 1B, 1D). The remaining two sets were kept in nursery conditions (Figs. 1A, 1C) until outplanting of the second set of 18 pyramids at the same reef sites in month 4 (T1), leaving the final replicate in the nursery. After a further 4 months (T2, 8 months since T0 was deployed) the third set was then also outplanted at the same locations.

The top (the location of the label) and all sides of the pyramid were photographed with a ruler, for scale, before placing them in the nursery or outplanting site. In order to avoid effects of position on the assay, the location of each fragment size was alternated around the pyramid with a clearly visible label on the top i.e., all ‘A’ replicates had the large fragment on the upper face, the medium on the left and the small on the right, relative to the label (Fig. 2). The pyramids were then outplanted with labels all facing the same direction, ensuring that each fragment size was exposed to all the potential variable light and water flow conditions.

In addition to recording survivorship (alive or dead), fragment size (cm²) of all 1,188 corals was manually measured from planar scaled digital photos (Fig. 3) using the program ImageJ (Schneider, Rasband & Eliceiri, 2012) at the point of fragmentation, outplanting, and at the end of the experiment. The final survivorship and growth measurements were collected at the in-situ nursery 4 months after the final deployment, but were delayed due to COVID-19 lockdowns until 9 months after final deployment at the ex-situ nursery (Appendix 2).

3D photogrammetric measurements

The coral fragments at the in-situ nursery (Figs. 1A, 1B) grew with a much higher degree of three-dimensional structure than those of the ex-situ nursery (Figs. 3A–3D). Therefore, we added 3D photogrammetry to estimate surface area of living coral tissue at the in-situ nursery in addition to estimating growth from planar scaled digital images that was performed at both locations.

A Three-Dimensional Structure from Motion (3D SfM) photogrammetric model of coral assays at the in-situ nursery was constructed using Agisoft Metashape Pro v 1.5.5 (AgiSoft Metashape Professional, 2019), from approximately 500 photos taken with a Canon Rebel EOS in an Ikelite underwater housing (Fig. 3E). Camera settings and assembly of the SfM model followed recommendations in Suka et al. (2019). Briefly, manual camera settings were selected (auto ISO exposure, f-stop = F10, shutter speed = 1/320, -⅓ exposure, broad point autofocus, repeat shutter, and large format photos). A batch script in Metashape Pro was run with the following settings (alignment = high accuracy and generic precision, 40k key point limit, 4k tie point limit, adaptive camera model fitting, Optimization = fit f and cx, cy, build dense cloud = medium quality, mild depth filtering, build mesh = arbitrary surface type, medium depth map quality, build texture = generic mapping mode, texture from all cameras, and hole filling enabled, build tiled model = source data dense cloud with medium depth map quality). The resulting SfM model was scaled with a series of six printed targets, fixed in pairs 10 cm apart. The accuracy of the three scale bars was 0.1 cm, with an overall estimated error of 0.07 cm. The scaled 3D SfM model was exported into Cloudcompare v2.11 (GPL Software) and areas with living coral tissue were segmented for inclusion with the segmentation tool. Corals on each side of the pyramid were grouped, labeled and colorized using an elevation model to highlight upward growth along the Z axis (Fig. 3E). Surface area was estimated for each size category (e.g., fused or unfused corals were grouped together for estimation of total surface area for each size category). Finally, we compared two-dimensional area (from top-down measurements) with three-dimensional surface area estimates from the SfM model by linear regression in R (RStudio, Inc. RsT, 2019) (Fig. 3F).

Statistical Analyses

Generalized linear mixed models (glmm) were used to assess the likelihood of individual fragments surviving to the end of the experiment based on their total time in nursery and their size class (see Appendices 3–5 for raw data and analyses). To avoid bias from fragments that did not survive their time in nursery, only fragments surviving to their designated outplanting time were included in these analyses. A glmm was fit for each species within each nursery using the glmer() function from the lme4 R package, with a binomial response defined in the model specification. Linear mixed models allowed the use of random effects to account for variation due to parent colony (genotype) and deployment pyramid within colony. Fixed effects included an interaction between experimental day outplanted (total time in nursery) and size group on the pyramid face being estimated. To ensure the reliability of glmm results we inspected model convergence as well as variance inflation using the vif() function from the R package car. Overall model performance was evaluated as adjusted marginal versus conditional R² using the r.squaredGLMM() function from the MuMIn R package. An example glmm model specification would be:

$$Survival\sim Day\mathrm{\_}Outplanted{}^{\ast }Size+(1|Genotype/Pyramid)$$

Percent net growth was calculated as the percent difference between total living coral tissue area at the beginning and end of the experiment and was assessed across all fragments on each face of a pyramid. Linear Mixed Models (lmm) were fit using the R package lme4 for both M. capitata and P. compressa datasets within each nursery, resulting in a total of four separate fitted models to estimate the effect of nursery time and size class on pyramid face net growth. Similar to models used previously to assess survivorship, these models included nested random effects for parent colony genotype and pyramid within genotype, in addition to a fixed effect interaction between experimental day outplanted and fragment size class. After model fitting a type three ANOVA was performed using the Anova() function in the R package car to assess the significance of marginal effects in each model. An example lmm model specification would be:

$$PercentNetGrowth\sim Day\mathrm{\_}Outplanted\ast Size+(1|Genotype/Pyramid)$$

Statistical Analyses

We define survivorship as those fragments which survived after deployment to the reef site until they were retrieved, therefore corals that died in the nursery prior to deployment were excluded. In order to incorporate both the fixed effect of ‘site’ and the random effect of ‘genotype’ (parent colony) we analyzed the binary response of survivorship using glmm models with a binomial (logit link) error distribution from the lme4 package (Bates et al., 2015) in RStudio (See Appendices 7 and 8 for raw data and analyses). Similar to the pyramid assays, variance inflation was inspected and overall model performance was assessed as adjusted marginal vs. conditional R².

$$Survival\sim Site+(1|Genotype)$$

Percent net growth included only those fragments which survived to the end and was calculated as 100*(cm² growth at the end/cm² growth at the beginning), which factors in the differing initial sizes of the coral fragments that had been grown for variable periods in the nursery prior to field deployment. We used lmm models in the lme4 package to analyze percent net growth with site as a fixed and genotype as a random effect.

$$PercentNetGrowth\sim Site+(1|Genotype)$$

Net growth for P. compressa was square-root transformed as required to meet assumptions of normality and homoscedasticity, but M. capitata did not require transformation as determined by Q-Q plots, histograms, and residuals over fitted plots. Similar to the pyramid assays after model fitting a type three ANOVA was performed using the Anova() function.

Fragment survival likelihood among sizes and residence times

Overall, small fragments were significantly less likely to survive to the end of the experiment (42%) compared to medium or large fragments (67% and 70%, respectively) which were not significantly different from one another (Appendices 9–11). GLMM models of individual fragment survivorship indicated that survivorship of M. capitata fragments at the ex-situ nursery outplanting site was significantly improved through increased time in the nursery and this effect was greatest among the smallest fragments (Appendix 10A). By comparison, there was no significant effect of nursery time on P. compressa survivorship for the ex-situ nursery outplanting site (Fig. 6 and Appendix 10B). A large proportion of survivorship variance was attributed to the nested effect of pyramid within genotype with approximately 80.8% (40.7% marginal) for M. capitata and 68.2% (13.9% marginal) of variation for P. compressa explained (Appendix 10). In contrast to the ex-situ outplanting site, there were no significant effects of nursery time, size class, or the interaction of these effects for fragment survivorship at the in-situ outplanting site for either coral species (Appendix 11). A positive, but marginally insignificant, effect was observed for the interaction between medium size and nursery time for fragments of Porites compressa at the in-situ nursery outplanting site. For the in-situ outplanting site, more of the variation in survivorship was attributed to fragment genotype than to pyramid within genotype for M. capitata, whereas survivorship of P. compressa varied more with pyramid within genotype. (Appendix 11).

Percent net growth among sizes and residence times

Pyramid assays revealed that the in-situ and ex-situ outplanting experiments differed greatly in their observed effects on net growth (Fig. 6). Because it was not possible to obtain permits for a fully reciprocal transplantation design and we have only the local outplanting site for each nursery, we cannot determine whether it is the site, the nursery design, or some interaction of both that contributed to this difference. However, neither the in-situ or ex-situ nursery duration showed consistently dramatic improvement for net percent growth after outplanting, and only M. capitata showed a positive effect of both total time in the ex-situ nursery, and its interaction with size class on overall net growth (Fig. 6 and Appendix 12A). Despite some trends in the data, no such effects were significant for P. compressa (Appendix 12B). For M. capitata at the ex-situ nursery, small fragments experienced the largest net-growth benefit from increased nursery time, but this effect was reduced for the larger size classes. In contrast, neither M. capitata nor P. compressa showed a significant benefit to net growth from prolonged duration in the in-situ nursery (Appendix 13).

Assay performance

The pyramid assay performed well in both low- and high-flow environments with 94% assay recovery at the in-situ nursery site (low flow) and 74% recovery at the ex-situ outplanting site (high flow). Pyramids were easy to handle and small enough to be mass produced and housed in either nursery. The design also made outplanting extremely easy because their weight allowed for placement in sandy rocky areas at the low-flow in-situ outplanting site and attachment to the reef with epoxy at the high-flow ex-situ outplanting site. Although we did not quantify sediment accumulation directly, visual inspection confirmed that the sloped sides of the pyramids reduced issues with sedimentation while still allowing for a clearly visible label on top during outplanting (Fig. 2). The design provided a weighty solid substrate for fusion of coral tissue (Figs. 3A–3D) without appearing artificial, rapidly blending into the reef substrate and making some difficult to detect by the end of the experiment (Appendix 1C, 1D). However, the small size of each face limited how long the assay could be used to assess individual survivorship, because fragments began to fuse making it hard to distinguish individuals. The compact size also limited the number of colony fragments which could be affixed, which is why the block assay was used to assess spatial variation among individuals.

Fragment survival likelihood among sites

Overall survivorship was 55% for M. capitata and 56% for P. compressa across all sites (Appendix 15). There was significant variation among the species across sites for both survivorship and growth. M. capitata had the lowest survivorship (28%) at site six and the highest survivorship (72%) at sites two and 10. Porites compressa had only 42% survivorship at sites five and eight to a maximum of 72% at site seven (Fig. 4B). There was approximately twice as much variation in survivorship across individual colonies of M. capitata compared to P. compressa (36.4% vs. 12.4%, respectively). For both species a larger proportion of survivorship variance was explained by the fixed effect of ‘site’ while a smaller proportion of variance was attributed to the random effect of ‘genotype’ and there appears to be an interaction because colonies that did among the best and worst at one site would reverse that trend at another. The models of M. capitata and P. compressa explained approximately 32.65% (25.2% marginal), and 11.89% (8.59% marginal) of variation in fragment survivorship, respectively (Appendix 16).

Percent net growth among sites

The growth of both species was significantly different among sites, but M. capitata had more than an order of magnitude higher variation in growth (641% vs. 49.5%) among individuals compared to P. compressa (Appendix 17). Percent net growth increased for both species by over 100% during the 8 months, with a mean 104% increase for M. capitata and 129% for P. compressa (Appendix 15). Even with individual variation, there were consistent differences among sites. The lowest growth rate for M. capitata was at site six (27%), and the highest at site two (155%), whereas for P. compressa the lowest growth was seen at site two (78%) and the highest at site 10 (272%). Looking across the environmental gradient of the bay, with the exception of site five, M. capitata tended to show the lowest growth rates (27–81%) at the central sites 3–8, and highest growth rates (123–155%) at the northern and southern ends of the bay (1–2 and 9–10). P. compressa, by comparison, had no obviously reduced growth regions across the bay, and only the northern sites (9–10) stand out (211–272%) with site 10 P. compressa showing the highest net growth than any other location in the bay (Fig. 4C and Appendix 15).

Assay performance

The block assays were easily deployed without epoxy due to the low wave energy environment in Kānéohe Bay, and all were successfully retrieved without loss in the field. Three top labels were lost, but in each case the backup labels remained attached. Although the flat horizontal surface appeared to retain more sediment than the sloped faces of the pyramid assay, sedimentation was minimal at most sites, and survivorship and growth were of the same magnitude between each design. However, the block design accommodated a larger number of unique colony fragments (n = 18) than the pyramid assays (Fig. 2), while still maintaining sufficient spacing for several months of growth without direct fragment-to-fragment interaction. This design allows for rapid performance testing of individual genotypes in potential restoration locations and could help to identify and focus efforts on which species and individuals are most likely to thrive at a given location.