Predicting Concurrence of Heatwaves, Droughts, and Wildfires with Spatiotemporal Deep Learning (Proposals Track)
Ana Trisovic (MIT)
Abstract
Extreme climate events—heatwaves, droughts, and wildfires—pose critical challenges to sustainability. Their concurrence suggests shared risk factors, yet has not been explored in a deep learning predictive model. We aim to identify a state-of-the-art (SotA) multi-task spatiotemporal model to predict heatwaves, drought severity, and wildfire occurrences simultaneously. By unifying these hazards in a single framework, we aim to capture both their individual behaviors and common causes. Our goal is to advance climate extreme event modeling by (1) constructing a comprehensive dataset of these climate exposure indicators, (2) benchmarking models to identify the SotA approach, and (3) demonstrating that shared representation learning improves concurrent hazard modeling. Our findings will help inform climate preparedness, mitigate public health risks, and support risk-aware decision-making in a changing climate.