Flood fOREcasting with Satellite Imagery and Generative High-resolution Techniques

Floods are among the most devastating natural disasters, causing widespread destruction and economic losses globally. Despite advancements in flood early warning systems, they remain limited in their ability to mitigate such losses, primarily due to a lack of accurate real-time flood inundation maps (FIMs) forecasting and reliable uncertainty quantification. The overall goal of the FORESIGHT project is to develop a framework to address this critical gap by leveraging cutting-edge Generative Artificial Intelligence (GenAI) techniques to forecast real-time FIMs with high resolution (30 meter/6hr) up to 5 days ahead. The proposed approach offers multiple advantages: it enables fast generation of FIMs by employing GenAI and omitting the time-consuming physical hydraulic model component of traditional approaches. It ensures accuracy by modifying dynamic river networks and improving precipitation forecast uncertainties. Physical consistency is maintained by training GenAI with data output from physical models. Crucially, it provides reliable uncertainty quantification for the generated forecasts. This innovative approach will be applied across four diverse large river basins - the Rhine, Indus, Niger, and Severn - to ensure broad applicability. Additionally, I will test the approach in ungauged basins to assess its potential applicability in data-scarce regions. By utilizing only satellite earth observation data and open-source global coverage data, the project aims to pave the way for the development of global operational flood forecasting systems.