Towards Operational AI-based Flood Forecasting using Deep Data Assimilation of Multi-source Earth Observations

Flooding results in large economic and loss of life, which are further aggravated by the lack of precise forecasts of flood inundation depth and extent. Recent extreme flood events have highlighted the need for reliable operational flood forecasting systems. Conventional physics-based flood models are subject to multiple sources of uncertainty and are computationally demanding, which limits their applicability for real-time operational services. Artificial intelligence (AI)-based flood models, on the other hand, can significantly reduce computational cost and enable near real-time forecasting at high spatial resolutions. Despite recent advances, most AI-based flood models lack mechanisms to correct evolving prediction errors using real-time observations. Flood processes are highly nonlinear, with errors that evolve rapidly in space and time, while Earth Observation (EO) data provide only intermittent and spatially incomplete snapshots of the true system state. Deep data assimilation (DDA) addresses this gap by learning state-dependent error propagation and dynamically integrating multi-source EO information into AI-based flood forecasting models. In the recently funded Indo-German project FLAIR (Flood Forecasting using AI for Regional Sustainability, funded by BMBF), we develop observation operators linking simulated flood states to EO-derived flood extent and water surface elevation within a two-dimensional convolutional long short-term memory framework. DDA is then implemented through a state-parameter augmented approach to update model states in real time, accounting for dynamically evolving flood conditions. The proposed framework is evaluated for two human-altered test catchments with contrasting hydrological characteristics in India and Germany. Forecast performance is benchmarked against an open-loop configuration and a DDA-based CaMa-Flood model across multiple forecast lead times ranging from one to seven days. A specific innovation is the assimilation of reservoir Water Surface Elevations from EO altimeters which help determine their influence on the resulting flood propagation as well as enable reservoir optimization for dampening the flood wave. FLAIR demonstrates the potential of deep data assimilation and multi-source EO data to improve the accuracy and robustness of AI-based flood forecasts as well as builds trust in such forecasts through detailed benchmarking against physics-based models, supporting their application in operational flood risk management.