City-Scale High-Resolution Flood Nowcasting Based on High-Performance Hydrodynamic Modelling

Abstract

Effective flood early warning systems serve as a cornerstone in mitigating the impacts of urban flood disasters. Nevertheless, contemporary urban flood warning systems encounter significant technical challenges, particularly the high uncertainty and low spatiotemporal resolution associated with rainfall forecasting and the inefficiency in flood modelling, especially for large-scale and high-resolution predictions. This study introduces an urban flood nowcasting system designed to tackle key technical challenges in flood prediction. The proposed framework employs a multi Graphics Processing Unit (GPU) accelerated shallow water equation (SWE) model, enabling high-resolution predictions of inundation distributions across urban surfaces within a limited time frame. To validate its effectiveness, the framework was applied to a vast urban area spanning 779 km2 in Guangdong Province, China. Through the simulation of a representative extreme flood event, the accuracy and computational efficiency of the flood nowcasting system were comprehensively demonstrated, showcasing its potential for real-world applications in urban flood early warning and disaster management. Furthermore, a comprehensive evaluation of the impact of rainfall spatial and temporal resolutions on flood modelling was conducted. The results reveal that the proposed model can predict a 4 h flood event with a spatial resolution of 4 m in just 10 min, harnessing the parallel computing capabilities of 16 GPUs. This established flood nowcasting framework offers strong technical support for the accurate prediction and early warning of urban flood disasters, enhancing disaster preparedness and response.

Keywords: Urban flooding, flood forecast, hydrodynamic modelling, multi-GPU acceleration