Coral size patterns across environmental gradients and morphologies

Updated: 2026-06-29

Coral size patterns across environmental gradients and morphologies Marine Lechene, Mariana Álvarez-Noriega, Madison Becker, Maren Toor, Sophie Gordon, Tiny Remmers, Will Figueira, Mia Hoogenboom, Renata Ferrari Abstract: Coral reefs are essential ecosystems that provide important ecological services and support the livelihoods of millions of people worldwide. Understanding the size structure of coral populations is critical for assessing recovery potential, yet our knowledge of how environmental conditions influence coral size structure, across a range of coral taxa, remains limited. This study tested three hypotheses: that larger corals are more prevalent (1) at higher latitude, (2) on inshore reefs, and (3) in sheltered habitats due to environmental forcing. We also expect that morphotaxa and depth will modulate coral size structure responses to the three environmental gradients examined. We thus examined spatial patterns in the size structure of six dominant morphotaxa across 84 sites on the Great Barrier Reef and the Torres Strait. We used large-area orthomosaics to measure the planar area of 5880 coral colonies, across gradients in latitude, cross-shelf position, wave exposure and depth. Our main findings indicate that larger corals are more prevalent on inshore reefs and less exposed habitats but not at higher latitude. These patterns were particularly pronounced in top-heavy morphologies, which are more vulnerable to breakage in high-exposure environments. As larger colonies produce more gametes these results can be used to identify reefs, or habitats within reefs, that disproportionately contribute large numbers of larvae to replenish coral populations. Understanding how environmental forces shape coral populations helps inform resilience-based management by identifying source reefs. METHODS Data collection and processing Imagery was collected as part of the Ecological intelligence for reef restoration and adaptation (EcoRRAP) sub-program (https://gbrrestoration.org/program/ecorrap/) following photogrammetry imaging procedures described in (Gordon et al., 2023). The total study sampling size included nine reefs along the Great Barrier Reef and three in the Torres Strait grouped into six clusters (Torres Strait, Offshore North, Offshore Central, Inshore Central, Inshore South and Offshore South; latitude range: -23.917571 to -9.73404). Within each reef, sites on the reef slopes and in lagoons were chosen across a wave exposure (to the prevailing wind direction) gradient (Front = high, Flank = mid, Back = low and Lagoon = lowest exposure) and depths (shallow ~5 m and deep ~12 m – where possible). The combination of site and depth is hereafter referred to as habitat (e.g., back shallow becomes low exposure shallow). Reefs were consideredoffshore or inshore based on water clarity assessment (i.e., inshore more turbid). The total sampling size was 85 sites with four replicates each, hereafter referred to as plots, or a total surface area of approximately 23900 m2 (340 plots of 72 m2 each) sampled in the period between January and May 2021 when no known major disturbance was recorded. A total of 2000-3000 images were taken at each plot and imported into Agisoft Metashape version 1.7.4 (Agisoft LLC, 2023) for 3D processing. For each plot, a two-dimensional orthomosaic was created from acquired images using custom python scripts with the parameters described in (Lechene et al., 2024). Each orthomosaic was standardised to a size of 12 x 6 m and imported into TagLab v2022.04.21 (Pavoni et al., 2022) for coral colony tracing. Despite variable visibility across sites, all orthomosaics were of sufficient quality to detect small colonies meeting the digitisation criteria detailed below. Corals were identified to the highest taxonomic resolution possible using six focal morphotaxa: tabular Acropora, Acropora millepora, massive Porites, Pocillopora damicornis, Pocillopora verrucosa, and Stylophora pistillata. These taxa were selected based on their high abundance across sites – quantified from 2021 orthomosaics (see details about species selection in Toor et al. (2025) section 5.2) – and their representation of contrasting life-history strategies. For instance, Acropora spp. exhibit rapid growth and higher susceptibility to mechanical and thermal disturbances, while massive Porites grow slowly but are generally more resistant to stress. A subset of colonies was identified by a highly trained benthic ecologist from the high-resolution images and the orthomosaics. This subset was then validated by a highly trained taxonomist in situ. When the in situ ID did not match the orthomosaic ID, the later was corrected. Validation analyses showed all morphotaxa were identified from images with 80-95 % accuracy. On each plot, colonies were digitised by delineating them with polygons using TagLab and assigned a class. In TagLab, the ‘4-clicks segmentation’ tool was used to trace colonies rapidly. This tool only requires the user to click on the four extremes of each colony before using an interactive segmentation model to trace its complex boundaries. These boundaries were then manually adjusted using the ‘Edit border’ tool for each colony. Colonies were digitised if they could be assigned to a class with at least 80% certainty, clearly reconstructed (i.e., not blurry or distorted), clearly visible (i.e., not obscured or overtopped by another organism or substrate) and entirely within the boundaries of the orthomosaic. Finally, colony sizes were extracted as 2D planar area in square meters using ArcGIS Pro version 3.0 (ESRI, 2023) and custom ArcPy and python scripts. Statistical analysis To examine spatial patterns and the relationship between environmental gradients and coral size structure, we ran three analyses that match the objectives and hypotheses’ structure presented in the introduction. With these three analyses we explored: 1) whether coral size varied between reef clusters distributed along a latitudinal gradient; 2) whether coral size varied between inshore and offshore locations; and 3) whether coral size varied between habitats distributed across gradients of wave exposure (back, deep sites = lowest exposure; front, shallow sites = highest exposure). Each of these analyses included different coral taxa (and morphologies). For all analyses, we fitted Bayesian Generalized Linear Mixed Effect Models (bGLMMs) using the packages brms (Bürkner 2017) and rstan version 2.26.22. (Stan Development Team 2023) within the R Statistical and Graphical Environment version 4.3.1 (R Core Team, 2023). Colonies with partial mortality on the periphery were included, but our method precluded us from accurately determining the size of colonies with central partial mortality (mainly massive Porites), which were subsequently excluded from the analyses (n = 162). We also checked that these excluded colonies were evenly spread among sites therefore it is unlikely that their exclusion would change any observed patterns. It should therefore be noted that we may be slightly overestimating average colony size for massive Porites. Because of higher uncertainty associated with the taxonomic identification of smaller colonies from orthomosaics and to ensure results are not impacted by our ability to detect smaller corals using this method, colonies smaller than 10 cm in diameter (~78 cm2) were excluded from the analyses (n = 3606). Instead of partitioning size data into arbitrary size classes and to keep size as a continuous variable, we performed quantile regressions for three quantiles corresponding to three size classes: smallest: quantile = 0.1 (i.e., the smallest 10%); median: quantile = 0.5 (i.e., the middle of the size distribution) and largest: quantile = 0.95 (i.e., the largest 5%). While other quantiles could have been selected (i.e., 0.25 or 0.75), we opted for more extreme quantiles to emphasise the tails of the distribution where shifts in size structure might be most informative for detecting demographic responses. All models used the same response variable, Coral Colony Planar Area, which was log-transformed and modelled using an asymmetric Laplace distribution adequate for quantile regression. Default priors were used (with the function get_prior of the brms package). Please note, not all taxa were included in all analyses, as for some combination of clusters, sites and cross-shelf position certain taxa were absent and would cause the models to not converge. Spatial variation in the size structure of coral morphotaxa across a latitudinal gradient was analysed by fitting three bGLMMs (one for each quantile or size class) with predictor variable Cluster interacting with Taxa and nested random effects of Plot within Site. P. verrucosa and S. pistillata were excluded as they were not present at all reef clusters. The six reef clusters were defined to capture both latitudinal position and cross-shelf variation (when applicable), while preserving ecological distinctiveness across reef systems. Clusters were explicitly ordered by latitude, with inshore and offshore reefs treated separately where relevant (e.g., in the southern region) to avoid conflating environmentally distinct habitats.A total of 20,000 iterations with a warmup of 5,000 and thinning of 12 were performed on each of three chains of each model. Posterior contrasts were calculated for each model to compare the estimated marginal means of colony size between clusters using the functions emmeans and pairs of the package emmeans v. 1.8.7 (Lenth, 2023). Spatial variation in the size structure of coral morphotaxa between inshore and offshore locations was analysed by fitting three bGLMMs with predictor variables: Cross-shelf position interacting with Taxa and Latitudinal cluster and nested random effects of Plot within Site. Cross-shelf position had two levels: inshore and offshore and Latitudinal cluster had two levels: central and south. Only tabular Acropora and P. damicornis were included as they were present at each combination of Cross-shelf position and Latitudinal cluster. A total of 20,000 iterations with a warmup of 5,000 and thinning of 12 were performed on each of three chains of each model except the model corresponding to the smallest size class which needed a total of 50,000 iterations with a warmup of 5,000 and thinning of 20. Posterior contrasts were calculated for each model to compare the estimated marginal means of colony size between inshore and offshore locations. Spatial variation in the size structure of coral morphotaxa between habitats was analysed by fitting three bGLMMs with predictor variables Habitat interacting with Taxa and nested random effects of Plot within Site within Reef. Habitat was a combination of wave exposure (lowest, low, mid and high) and depth (shallow and deep). All morphotaxa were included. A total of 20,000 iterations with a warmup of 5,000 and thinning of 10 were performed on each of three chains of each model. Posterior contrasts were calculated for each model to compare the estimated marginal means of colony size between habitats. For all models in all analyses, chains mixing, and convergence were checked using Markov Chain Monte Carlo diagnostics using the rstan package. All chains were well mixed and converged upon a stable posterior (all Rhat values < 1.05). For each model, goodness of fit was checked visually using posterior predictive checks and all models were validated using the DHARMa package version 0.4.6 (Hartig, 2022) to check simulated residuals. The random effect structure in each model was tested and chosen to capture non-independence due to spatial grouping at multiple scales and reflects the expectation that plots within the same reef or site are more similar. We conducted visual inspections of residuals for spatial patterning and overdispersion using randomised quantile residuals. While these checks did not reveal obvious spatial structure or model misspecification in the residuals, we acknowledge that residual spatial autocorrelation may persist even with appropriate random effects. Posterior predictions and marginal effects were extracted using the emmeans v. 1.8.7 and tidybayes v. 3.0.4 (Kay, 2023b) packages. To visualise the distributions of predicted colony sizes across clusters. cross-shelf position, habitats and morphotaxa, we used the stat_slab() function from the ggdist v. 3.3.0 package (Kay, 2023a). This function displays posterior distributions as smoothed density shapes (“slabs”), which allow visual comparison of distributional spread, skew, and central tendency. By default, slab height is normalised across groups, enabling consistent scaling across facets and facilitating visual comparison of distribution shape across categories. Slab width reflects the relative density of predicted values; narrower and more peaked curves indicate lower posterior variability. code and data are available at: https://doi.org/10.5281/zenodo.17668070 For further details please refer to the published manuscript: