A GPU-accelerated local time step-based shallow water model for integrated sea–land flood inundation in densely built urban areas during storm surges

Coastal flood models for storm surge often face significant computational challenges due to the large amounts of computational cells in integrated sea-land scenarios and the need for extremely small time steps in certain regions, which are typically constrained by the globally minimum time step. To address these limitations, this study presents an efficient integrated sea-land flood model for storm surge disaster prediction in coastal urban areas with dense buildings. The model leverages a Graphics Processing Unit (GPU)-accelerated framework to handle large computational grids and incorporates a local time step (LTS) approach to mitigate restrictions imposed by locally refined grids, extremely small time steps, and flow conditions disparities between sea and land. A GPU-optimized parallel algorithm enhances computational performance by refining the numerical framework, optimizing kernel functions, and improving memory utilization, all seamlessly integrated with the LTS approach. The model’s efficiency is demonstrated through storm surge and flood simulations covering the sea, straits, and densely built coastal land areas in Macau, China. The results show that LTS scheme can reduce computation time by approximately 40 times, significantly improving computational efficiency across different mess configurations. With GPU and LTS acceleration, the present model provides a powerful tool for real-time coastal flood forecasting.