Climate change is transforming coral reefs by increasing the frequency and intensity of marine heatwaves, often leading to coral bleaching and mortality. Coral communities have demonstrated modest increases in thermal tolerance following repeated exposure to moderate heat stress, but it is unclear whether these shifts represent acclimatization of individual colonies or mortality of thermally susceptible individuals. For corals that survive repeated bleaching events, it is important to understand how past bleaching responses impact future growth potential. Here, we track the bleaching responses of 1,832 corals in leeward Maui through multiple marine heatwaves and document patterns of coral growth and survivorship over a seven-year period. While we find limited evidence of acclimatization at population scales, we document reduced bleaching over time in specific individuals, primarily in the stress-tolerant taxa Porites lobata, indicative of acclimatization. For corals that survived both bleaching events, we find no relationship between bleaching response and coral growth in three of four taxa studied. This decoupling between bleaching and growth suggests that coral survivorship is a better indicator of future growth than is a coral’s bleaching history. Based on these results, we recommend restoration practitioners in Hawaiʻi obtain outplants from Porites and Montipora colonies with a proven track-record of growth and survivorship, rather than devote resources toward identifying and cultivating bleaching-resistant phenotypes. Survivorship followed a latitudinal thermal stress gradient, but because this gradient was small, it is likely that local environmental factors also drove differences in coral performance between sites. Efforts to reduce human impacts at low performing sites would likely improve coral survivorship in the future.

We used large-area imaging (photogrammetry) to capture a 3D snapshot of coral reef condition at our six sites in leeward Maui in 2014, 2015, 2017, 2019, and 2021. Large-area imaging is the process of generating a composite visual reconstruction (i.e., 3D model, orthoprojection, DEM, etc.) via the overlap of many component images. Large-area imaging has been used with increased frequency to archive coral reef structure and condition, and provides a basis for in silico field work to answer a variety of ecological questions. Details of our large-area imaging workflow are available elsewhere and will be described in minimal detail here.

At each fixed site, divers entered the water with two D7000 SLR Nikon cameras (focal lengths of 18mm and 55mm) in Ikelite underwater housings. Divers marked the boundaries of each 10 x 10 m long term monitoring plot with six calibration tiles, and placed four 0.5 m long scale bars within the plot. One diver swam approximately 1.5 m over the reef in a gridded pattern with the cameras, which were programmed to take a picture every second. This produced approximately 5,000 pictures per site. The camera diver imaged a core area of 10x10 m, and swam several meters beyond the calibration tiles to ensure that a buffer region (> 1 m wide) around the core area was also imaged.

Once imagery was collected, we used the software Agisoft Metashape (St. Petersburg, Russia) to build a dense point cloud, which we refer to here as a 3D model. After building the 3D models in Metashape, we loaded them into the custom software Viscore for postprocessing. These postprocessing steps included 1) scaling the 3D model using the 0.5 m scale bars, 2) entering depth measurements collected at each calibration tile so that the 3D model could be oriented with respect to the sea surface, 3) manually aligning 3D models of the same site collected in different years, and 4) exporting a high-resolution top-down view of the model known as an orthoprojection. Each orthoprojection was 12 x 12 m and had a resolution of 1 mm per pixel. We used these orthoprojections rather than the 3D models for all subsequent data collection steps.  

Researchers used TagLab and ArcGIS Pro to trace patches of live coral tissue following the approach of Rodrguez et al. 2021. We found no effect of software on traced planar area, and a single annotator QCed all tracings in TagLab to control for any potential effect of annotator or software. We used the high-resolution imagery that underlies the 3D model (raw images) as a reference to assist with tracing and species ID. For this study, we chose to identify Pocillopora to the genus level because morphology is not a good indicator of species ID for Pocillopora in Hawaiʻi, whereas the three other taxa could be confidently identified to species. For P. lobata, we focused on massive and submassive growth forms to avoid confusion with the related branching coral Porites compressa.

After corals were traced, we used TagLab to “match” patches of live tissue that represented the same individual through time. This created a network of temporally linked coral patches. This enabled us to more accurately identify individual genets of corals such as Porites and Montipora, which readily exhibit fusion and fission over time due to partial morality. Hereafter we use “patch” to denote a single contiguous region of live coral tissue in one timepoint, and “genet” to denote a network of patches interconnected through time. While we cannot definitively say that linked corals represent a single genetic individual (since we did no genetic testing), the precise tracking capabilities afforded by overlapping orthoprojections and associated raw images enable us to identify individual genets with a reasonable degree of confidence. Additionally, we visually inspected orthoprojections to identify potential instances of pseudoreplication arising from colony fission prior to our timeseries, and found minimal evidence to support this concern. We performed all subsequent analyses at the genet level, and considered a genet to have survived the full timeseries if at least one patch of that genet existed in both 2014 (our first sampling year) and 2021 (our final sampling year). For each genet, we calculated the total planar area (cm²) in each timestep by summing the planar area of individual associated patches.

To achieve an appropriate sample size of M. capitata, M. patula, and P. lobata, we traced corals within 10 randomly placed non-overlapping 0.5 m² quadrats within the orthoprojection. We identified and traced all coral patches with a diameter > 5cm whose centroid fell within the boundaries of a quadrat. To account for fission and fusion dynamics, we also traced patches outside of a quadrat or < 5cm in diameter if they were temporally linked to a genet inside a quadrat. At sites where < 40 genets were traced in our first timestep, we placed additional quadrats (up to 25 total) and continued tracing taxa until we reached 40 genets or until 25 quadrats had been placed. For Pocillopora, which was less abundant than other taxa, we identified and traced all patches within each 12 x 12 m orthoprojection. To ensure a balanced design for statistical analyses, we randomly selected 100 P. lobata genets for n = 2 sites where > 100 genets had been traced. 

We assessed the extent and severity of bleaching for every patch of coral tissue > 1cm² in each year of our timeseries. First, we visually estimated bleaching extent as the percent of a patch’s area (0-100%) with some degree of paling unrelated to coral growth or disease. Focusing on this bleached tissue only, we then scored bleaching severity (from 0 to 3) based on the overall degree of paling observed (0 for practically no pigmentation loss; 1 for slight paling, 2 for significant loss of pigmentation, and 3 for almost or completely stark white). When estimating both bleaching extent and severity, we used Viscore’s Virtual Point Intercept interface to reference the original imagery. This allowed us to view multiple angles of each coral patch to assess bleaching, enabling us to account for changes in tissue color due to lighting or image quality.

Once we assessed bleaching extent and severity for all patches in all years, we calculated genet-level bleaching extent and severity in each timepoint by summing bleaching extent and severity across patches, weighted by patch area. Then, we incorporated both bleaching extent and severity into a single bleaching metric for each genet in each timepoint. According to this metric, genets fell into one of five categories: no bleaching or very minor paling, minor bleaching, moderate bleaching, severe bleaching, or extreme bleaching. If a genet changed by more than one step between 2015 and 2019, the genet was considered to have an increased or decreased bleaching response over time. We considered genets that changed one bleaching step or less between 2015 and 2019 to have a stable bleaching response over time. Among these stable genets, those that exhibited moderate bleaching or higher were considered to have “high bleaching susceptibility”, and all other genets were considered to be “thermally tolerant”. These bleaching responses (thermally tolerant, decreased bleaching response, increased bleaching response, and high bleaching susceptibility) were used as fixed factors to compare genets over time.

To test for signs of acclimatization among surviving genets in each population (corals of the same taxa within a given site), we employed a bootstrapping approach using only genets that were present in both bleaching events. We calculated the difference between each genet’s bleaching score in 2015 and 2019, which produced an integer response variable ranging from -4 to 4, where 0 indicated no change in bleaching, -4 indicated extreme bleaching in 2015 and no bleaching in 2019, and 4 indicated no bleaching in 2015 and extreme bleaching in 2019. Then, we generated a null distribution of the median change in bleaching scores for each coral population via 100,000 resamples. P values were calculated as the proportion of all resamples where the population’s median change in bleaching score was ≥ 0, and we used α = 0.05 as our threshold of significance. We did not test for acclimatization in two populations of Pocillopora spp. which had a low sample size of surviving genets (< 10 genets).

To test if coral populations at each site exhibited different bleaching responses, we created a contingency table of surviving genets based on their bleaching responses over time. We performed Fisher’s exact test to test the null hypothesis that there was no relationship between site and bleaching response. We performed this test between all permutations of sites for each taxa and calculated adjusted p values using the Bonferroni correction for multiple comparisons. We assigned post hoc letters at a significance level of α = 0.05 to indicate significant differences in bleaching response between coral populations at different sites. We also pooled data by site and performed the same test to identify significant differences in bleaching response among coral taxa. Separately, we pooled observations across sites and ran a logistic regression of bleaching probability vs. genet size in 2015 and 2019 to quantify the relationship between bleaching and genet size. For this analysis, we coded bleaching as a binary variable (0 for genets with no signs of bleaching or minor bleaching, 1 for genets with moderate, severe, or extreme bleaching) in order to perform the logistic regression.

To test if acclimatized and thermally tolerant genets exhibited higher growth over the timeseries than other corals, we performed an ANCOVA for each coral taxa, focusing on surviving genets only. We used bleaching response (thermally tolerant, decreasing bleaching susceptibility, increasing bleaching susceptibility, and high bleaching susceptibility) as a fixed effect, only retaining those with a sample size ≥ 10 for analysis. Our continuous predictor variable was genet planar area at the start of the timeseries, and our response variable was genet planar area at the end of the timeseries. Genet planar area was natural log transformed for normality, and model assumptions were visually assessed using residual plots and quantitatively using Shapiro tests for normality and Cochran tests for homogeneity of variance. We performed a separate ANCOVA using site as a fixed effect to test for significant differences in coral growth between sites for each taxon. All models used a significance level of α = 0.05.

Finally, to test for differences in coral survivorship between sites, we performed a logistic regression using all traced genets (not just survivors) for each taxon. We used site as a fixed effect, log transformed planar area of genets in 2014 as our predictor variable, and coded genet survivorship as a binary response variable (0 = genet did not survive until 2021, 1 = genet survived from 2014 to 2021). There were significant interactions between multiple sites, so to facilitate interpretation we conducted pairwise comparisons between sites that didn’t interact. We calculated adjusted p values using the Bonferroni correction for multiple comparisons at a significance level of α = 0.05. To test for differences in coral survivorship between taxa, we also pooled genets by site and ran a logistic regression of survivorship vs. initial genet size. We completed statistical analyses using the ‘dplyr’, ‘emmeans’, ‘multcomp’, ‘coin’, and ‘rstatix’ packages in R v. 4.0.5.