Location Invariant Flood Prediction using Fourier Neural Operator

As coastal cities are increasingly vulnerable to climate- related flooding, the need for accurate and timely flood predictions is pressing, particularly for emergency response planning. This study presents an approach to predict flood events at high spatial resolution, leveraging inundation maps and 15-minute rainfall data. We employ Fourier Neural Operator (FNO), a method that learns efficient approximations of solutions to parametric partial differential equations. By incorporating multi-modal data types, our FNO model captures the complex dynamics of flood events including water levels in affected areas and long-range spatial and temporal dependencies. This interdisciplinary research contributes to the development of generalizable methods for predicting floods in coastal cities.