Floods are among the world’s most destructive natural hazards, and their impacts are growing as climate change and urban development increase vulnerability. Yet accurately mapping floods in real time remains a challenge. Limited observations, diverse terrain, and the difficulty of ensuring physical consistency make predictions uncertain. This research combines multimodal satellite data with terrain information to build a physics-guided deep learning framework. We employed an AI approach to capture detailed spatial patterns and to model large-scale hydraulic dynamics, and by enforcing physical constraints from the shallow water equations, our model produces flood predictions that are both accurate and physically coherent. Tested across multiple floodplains, this approach delivers reliable estimates of flood extent and depth, while keeping mass and momentum consistent. These results show that embedding physics into AI can directly support operational monitoring, early warning, and climate resilience efforts worldwide.
Speaker
Dr. Leila Hashemi-Beni
Research Fellow, Geospatial Analytics for Environmental Management
