A stochastic rain-on-grid framework for handling spatio-temporal rainfall uncertainty in impact-based flood nowcasting

Predicting flash flood impacts remains a major challenge due to intrinsic uncertainty in rainfall spatial-temporal structure and limited understanding of how rainfall organization propagates through hydrological and hydrodynamic processes to generate urban-scale impacts. These limitations hinder the development of reliable impact-based early warning systems for small, fast-responding catchments. To address these challenges, we introduce a Stochastic Rain-on-Grid framework that explicitly accounts for rainfall uncertainty by coupling a high-resolution stochastic rainfall generator with a 2D hydrodynamic model operating at the watershed scale. The framework is applied to a representative high-impact flash flood event affecting a piedmont urbanized area characterized by complex interactions between mountain and urban flooding processes. Using 100 equiprobable synthetic storms reproducing the statistical properties of the observed radar rainfall (200 m, 2 min), we assess how rainfall spatio-temporal variability alone influences catchment response and street-level flood impacts. Results show substantial variability in simulated hydrographs despite statistically similar rainfall inputs, while this variability systematically attenuates at the street scale, leading to more stable hazard classifications. This indicates that impact-based hydrodynamic indicators are more robust targets for early warning systems than traditional hydrograph-based metrics. Analysis of rainfall structure metrics reveals that spatial and temporal coefficients of variation consistently correlate with impact severity. Building on these relationships, we propose the Storm Variability Diagram, which classifies equiprobable events by expected impact and significantly reduces uncertainty in hazard mapping through ensemble partitioning. Overall, this study provides a proof-of-concept for impact-oriented uncertainty assessment through a modular and transferable framework, supporting uncertainty-aware flash flood forecasting.