FNO Tsunami Flooding Simulation for Arbitrary Bathymetries, Shorelines, and Land Topographies

Global warming is rapidly melting glaciers, causing sea levels to rise and shorelines to encroach on communities and animal habitats. This results in coastal erosion, increased storm surges, and habitat loss, such as the breeding grounds of elephant seals. More critically, higher sea levels heighten community vulnerability to tsunamis, which can generate massive waves that overwhelm sea walls, as evidenced in Indonesia and Japan. Thus, simulating tsunami flooding is crucial. While SWASH, a tool based on shallow water equations, effectively models complex environments like ports and residential areas, it is slow, taking 4 hours on 16 CPU cores for one simulation. In contrast, we have developed neural operators that significantly speed up this process. Our neural operators, trained on hundreds of input-output image pairs, convert a 2.5D input image of the simulation environment into an output image showing the predicted flooding state, including wave height and wet-and-dry states. Benchmark tests show our neural operators complete simulations in just 1.2 seconds, compared to 7 hours for traditional finite volume solvers. This speed advantage remains regardless of input grid dimensions. While our neural operators excel at predicting the wet or dry state, further training is needed to improve wave height predictions. Figures demonstrate our neural operators' ability to predict flooding at various times given a simulation environment input. Our methodology offers a faster, efficient alternative for tsunami flooding simulations, which is crucial for enhanced disaster preparedness.