Real-time flood forecasting in lowland rivers exposed to levee breaches using GPU hydrodynamics

With the increasing frequency of extreme hydrometeorological events due to climate change and the raise of floods in lowland regions triggered by sudden levee failures, early warning systems capable of alerting at-risk populations have become a cornerstone of contemporary flood risk management. Despite recent advances in flood forecasting, achieving accurate predictions in low-lying areas remains a major operational challenge due to the need of coupling high-resolution hydrological and hydrodynamic models while ensuring a remarkable computational efficiency. Moreover, the presence of extensive and aging levee systems, whose failure is often driven by multiple and uncertain processes, calls for the enhancement of these tools to account for flooding scenarios induced by potential breaches. This study develops an integrated flood forecasting system to predict real-time floods, including those triggered by levee failures. The modelling chain combines within a unique operational platform the gathering of meteorological weather forecasts and observed data, a rainfall-runoff model, and a two-dimensional hydrodynamic model taking advantage of Graphic Processing Units (GPUs). The procedure is designed to identify potentially at-risk areas in advance, supporting timely decision-making for flood risk management. The application to a lowland area in Northern Italy highlights the capability of the proposed framework to simulate complex flood dynamics at high-resolution by reproducing a real event occurred in March 2025 and by providing hypothetical levee-breach flood scenarios with lead times of up to two days for civil protection actions, even in fast reactive basins.