Our approach combines two parts: one model predicts water levels, and another identifies which areas are wet or dry. This is especially important in places like tidal marshes and intertidal zones, where land areas frequently transition between wet and dry states due to tides. We also used graphics processing units (GPUs) to speed up training, reducing the time it takes to build each model to just over one second. Once trained, the emulator can predict flooding patterns across the whole coastal region almost instantly.

This new tool is accurate, fast, and physically realistic. It makes it easier to map flood risks and respond quickly to storms, and it can be applied to many different coastal areas for both current and future flooding scenarios.