This repository also contains the supplementary tables for a global analysis of climate-resilient coral reefs. These materials provide full documentation of the datasets, model inputs, performance metrics, and optimization methods used to generate global predictions of coral cover and life-history composition, as well as the spatial prioritization results. Table S1. Coral reef observation datasets used for model training and validation. Each dataset is summarised by the number of observations, availability of life history data, temporal coverage (first and last observation year), and full citation. These datasets provide the empirical basis for modelling coral cover and life-history composition across global reef systems. Table S2. Life-history trait assignments for coral taxa. Coral genera and species were classified into competitive, stress-tolerant, and weedy life-history strategies, with some taxa allocated proportionally across groups to reflect variation in morphology and ecological behaviour (e.g. branching vs. massive forms). These proportional assignments were used to derive life-history composition for all modelling analyses. Table S3. Predictor variables (n = 42) used in coral-cover and life-history models. For each environmental and anthropogenic variable, the table reports a brief description, units, spatial and temporal resolution, and source dataset. Table S4. Training and testing performance of machine-learning models predicting total coral cover and life-history group cover. Metrics include mean R² and mean absolute error (MAE) across 100 model runs (± SD). High training R² values indicate strong within-sample fits, while lower testing R² and higher MAE values reflect the challenge of predicting fine-scale coral cover patterns under real-world variability. Life-history models generally show lower predictive skill than total coral cover, consistent with greater noise and fewer observations in taxonomically resolved datasets. Table S5. Regional model performance and selection criteria for coral-cover predictions. For each coral province, the table reports skill metrics (mean absolute error skill and median absolute deviation skill), their variability (SD), and one-sided permutation test p-values assessing improvement over a null model. Columns also show the proportion of models from the global pool meeting predefined MAE and MAD thresholds, and the percentage meeting both criteria. Models passing both thresholds were retained for regional ensemble predictions of 2020 and 2050 coral cover. Table S6. Spatial optimization outcomes by countries, territories and jurisdictions under SSP3-7.0. For each coral reef area, we estimate total reef extent and the area selected as climate-resilient refugia under the 50 Reefs plus optimisation, including the proportion of newly identified priority area and the overlap with the original Beyer et al. (2018) prioritisation. Values for areas selected only by Beyer et al., as well as reef habitat not selected by either approach, are provided in km² and as percentages of total reef extent. Table S7. Spatial optimization outcomes by countries, territories and jurisdictions under the higher-emissions SSP5-8.5 scenario, presented as a supplementary sensitivity analysis to Table S6. Columns and units are as in Table S6. Minor differences in total extent values between SSP3-7.0 and SSP5-8.5 (<0.01% of prioritised area) reflect gridding artefacts from the prioritisation step and do not affect interpretation. Appendix S1. Gurobi solver equations and configurations. This appendix outlines the mathematical formulation of the spatial optimization problem and the Gurobi-specific settings used to obtain stable, spatially coherent solutions.

DescriptionThis dataset contains the global 250 m resolution binary prioritization surface generated from a coral reef portfolio optimization analysis described in the associated manuscript. The raster identifies selected planning units resulting from the optimization framework designed to maximize climate resilience, life history representation, spatial cohesion, and risk minimization. Each pixel represents a 250 m × 250 m grid cell (~0.003° × 0.003° at the equator). Pixel values are binary:0 = not selected1 = selected The raster corresponds to two optimization solutions, one for a moderate emissions scenario (SSP3-7.0) used in the main anlaysis and one for a sensitivity analysis of a high emissions scenario (SSP5-8.5). Format: GeoTIFFNumber of layers: 1 Spatial PropertiesCoordinate Reference System (CRS): WGS 84EPSG: 4326Units: degrees (longitude/latitude)Resolution: 0.003° × 0.003° (~250 m at the equator)Extent:xmin: -179.9907xmax: 179.9763ymin: -34.29605ymax: 32.51395 DimensionsRows: 22,270Columns: 119,989Total cells: 2,672,494 (automatically defined by raster dimensions) Data TypeSingle-band rasterValue range: 0–1Mean value: 0.3042389 InterpretationValues of 1 indicate 250 m pixels included in the optimized portfolio. Values of 0 indicate pixels not selected. The mean value reflects the proportion of reef pixels selected globally under the specified optimization constraints. GenerationThe raster was produced using an integer programming framework and solved using Gurobi Optimizer. Input layers to the prioritization included 250 m global predictions of coral cover, life history groups, spatial cohesion, and model uncertainty. Full methodological details are provided in the associated manuscript.