Scientists have created an AI model that predicts moderate thermal stress, a major precursor to coral bleaching at sites along the Florida coral reef up to six weeks in advance, with predictions usually accurate to within a week.
The study presents an explainable and site-specific machine learning framework to support coral scientists and restoration practitioners in local reef management and emergency response planning.
“This model gives coral scientists and resource managers a warning of the likelihood of thermal stress during a season — and, more importantly, the week it’s most likely to start,” said lead author Marybeth Arcodia of the Rosenstiel School of Marine, Atmospheric, and Earth Science at the University of Miami. Arcodia holds a dual position in the Department of Atmospheric Sciences and the Frost Institute for Data Science and Computing. « Thanks to explainable AI, we can also identify the environmental factors that underlie these predictions on each reef site. «
“Our model identifies potential factors that influence thermal stress at a given reef site,” explained Richard Karp, co-author and postdoctoral researcher at the Rosenstiel School’s Cooperative Institute for Marine and Atmospheric Studies, “this information gives managers the opportunity to identify trigger points in emergency action plans, which can support planning and response decisions. «
The research team combined atmospheric science, coral ecology and data science to build a prediction tool adapted to the Florida coral reef.
Localized AI predictions on actionable time scales
Using an XGBoost machine learning model, the team predicted the appearance of moderate thermal stress on corals at three reef sites using environmental data from 1985 to 2024. The entries included measurements of accumulated and instantaneous thermal stress, sea surface temperature anomalies, air temperature, winds, solar radiation, and indicators of Loop current and El Niño conditions, based on NOAA Coral Reef Watch and other public data sets.
« Our prediction system produced skillful forecasts up to six weeks in advance, and in most cases, it was accurate about a week before the thermal stress actually started, » said Karp. « He also surpassed two benchmark approaches – a multiple logistic regression model and a frequency-based method both to predict whether thermal stress would occur and to identify when it would begin. «
The researchers also applied explainable AI techniques using SHAP, a method that shows which environmental factors most influence each prediction – to understand how thermal stress factors differ by reef site and forecasting time.
Surface air temperature consistently ranked among the most important predictors, while other key environmental factors varied by site and forecast time, highlighting the value of the localized prediction.
« This information is provided on time scales where management actions are still possible, » Karp added. « They help prioritize surveillance, inform on the moment to initiate emergency actions and guide on where resources are targeted most effectively. «
Actionable forecasts to support proactive reef conservation
Reefs in Florida and the Caribbean are experiencing increasingly frequent and severe thermal stress and bleaching events – including the 2023 record sea heat wave – which increases the need for early warning tools at the site level.
The authors point out that the new AI framework aims to complement and not replace existing operational systems such as NOAA Coral Reef Watch, by adding a localized synchronization signal, season by season, for the beginning of thermal stress.
source : enerzine
