Enabling real-time high-resolution flood forecasting for the entire state of Berlin through multi-GPU accelerated physics-based modeling

This study demonstrates how advances in multi-GPU computing can enable high-resolution, operational pluvial flood forecasting for large urban areas. Focusing on the hydrodynamic model RIM2D (Rapid Inundation Model 2D), the research evaluates its performance for large-scale urban flood simulations across the entire state of Berlin (891.8 km²) using spatial resolutions of 2, 5, and 10 m and GPU configurations ranging from 1 to 8 units. Two scenarios are analysed: the June 2017 real-world pluvial flood and a standardized 100-year return period (HQ100) event used for official hazard mapping. Results show that with multi-GPU processing, RIM2D can simulate the 48-hour 2017 event within operationally relevant timeframes—approximately 8 minutes at 10 m resolution, 34 minutes at 5 m, and 5.5 hours at 2 m resolution using 8 GPUs—making real-time integration into early warning systems feasible. 

The findings highlight that multi-GPU architectures are essential for enabling high-resolution simulations at urban scale, while also revealing diminishing performance gains beyond certain GPU thresholds. Overall, the study shows that large-scale, high-resolution flood simulations can now support operational early warning and impact-based forecasting, with significant implications for urban flood risk management under climate change.