Scientists at the University of Miami have developed a new artificial intelligence model designed to help protect coral reefs by predicting when they will experience heat stress — a precursor to coral bleaching — up to six weeks in advance. The model was trained using 40 years of environmental data, enabling it not only to forecast if heat stress will occur in a given season but also to estimate the specific week it’s likely to begin. This early warning ability could give conservation teams valuable time to prepare and act before deadly bleaching takes hold.
Bleaching happens when elevated ocean temperatures disrupt the symbiotic relationship between coral and the algae that live inside their tissues, causing corals to turn white and often die if stressful conditions persist. Coral reefs are critical ecosystems, supporting roughly a quarter of marine life and protecting coastal communities where about one billion people live. The new AI model also highlights the environmental factors most important to heat stress at each reef site, which helps scientists tailor localized strategies to safeguard vulnerable areas.
The research team used a machine-learning method called XGBoost and combined multiple types of environmental data — including sea surface temperatures, solar radiation, wind, and atmospheric indicators like El Niño and Loop Current effects — to achieve accurate site-specific forecasts. The model was tested at three reef locations with predictions that were generally accurate to within about one week, outperforming more traditional forecast approaches.
Beyond just timing, the model’s explainable AI techniques allow researchers and reef managers to see which variables are driving heat stress risks at different sites. That insight can inform decisions such as when to increase monitoring, prioritize restoration efforts, or implement emergency responses. While not replacing existing tools, this AI system aims to complement current coral reef monitoring frameworks by adding a more localized and actionable early-warning capability.
