Extending Heat Wave Forecasts with Interpretable Machine Learning

Sub-seasonal forecasting has the potential to extend lead times for extreme weather events and improve targeted adaptation strategies, yet it remains underutilized. Heat waves, which have surged in both frequency and intensity over the past century, stand to benefit significantly from enhanced sub-seasonal predictions. This study presents a hyper-local sub-seasonal heat-wave forecasting model that integrates local meteorological conditions, climatology, global climate indices, land-surface variables, and ocean temperatures. Testing has demonstrated that this model accurately predicts over 85% of observed heat waves. The model’s predictions are correlated with emergency department data to develop an early-warning system for heat-related health impacts, identifying heat thresholds that best correlate with public health outcomes. This early-warning system aims to enhance health system preparedness and response by providing timely alerts based on advanced sub-seasonal forecasting techniques. It will guide the allocation of additional adaptation resources, such as cooling shelters. Although currently based on data from Austin, Texas, the model provides a replicable framework that can be extended to other regions.