Machine-learning and prioritization models reveal climate refugia for coral reefs into 2050

Kyle J A Zawada, Emily Darling, Stacy D Jupiter , Timothy R McClanahan, Joseph M Maina

Climate change is accelerating the decline of coral reefs, yet some locations may retain conditions that support persistence under future warming. We compiled 45,064 coral field observations from 1960–2025 and 42 climate, biophysical, and human-pressure predictors to train machine-learning ensembles estimating coral cover and community composition in 2020 and 2050. Predictions span a global 250 m-resolution map of potential coral reef extent: 8.8 million pixels covering 552,969 km2. Model performance varied by province and reflected field-data availability; predictions were restricted to regions where models passed pre-specified accuracy thresholds. We then used a multi-objective optimization framework to identify reef areas that jointly maximize current and future coral cover, community composition, spatial cohesion, and model certainty. Our results reveal 165,922 km² of potentially climate-resilient reef locations spanning 71 countries and 100 territories and jurisdictions, adding 30 countries and 54 jurisdictions beyond the original 50 Reefs assessment (Beyer et al 2018). More than half of this resilient habitat (61%; 100,755 km2) occurs within five countries with extensive reef systems: the Bahamas, Cuba, Australia, Indonesia, and the Philippines. These findings highlight the uneven but widespread global distribution of large-scale refugia and emphasize the opportunity for global efforts and countries to strategically focus policies, conservation finance, and management actions most likely to sustain coral reef futures.