Bridging the implementation gap: artificial intelligence and machine learning applications in heat risk management

Extreme heat is the largest weather-related cause of death in the United States and a rapidly increasing risk globally. However, its severe human health impacts are among the most avoidable consequences, underscoring the important role of heat risk management. Artificial intelligence (AI) and machine learning (ML) offer new capabilities for forecasting and managing heat-related risks. This paper conducts a systematic review to understand the growing role of AI/ML in extreme heat risk management. To structure the analysis, we draw on methodological and thematic applications and propose an adapted Predict-Prevent-Prepare-Respond-Recover (PPPRR) risk management framework that reflects the critical role of prediction in the field. We found a clear, overlapping concentration in the literature on the “Predict” (98.4%) and “Prevent” (59.2%) functions, utilizing large, structured datasets for modeling and vulnerability mapping. Conversely, “Prepare,” “Respond,” and “Recover” are significantly underrepresented, making up (1.6%, 10.4%, and 3.2%, respectively). This imbalance highlights a “research-to-implementation gap” and fractured learning cycle that reflects a focus on modeling rather than operational needs. Bridging this gap requires an emphasis on the latter PPPRR phases and the integration of participatory and less structured datasets, such as real-time community input, to train the next generation of models, reduce bias, and achieve a complete learning cycle.