Physics-informed and data-efficient machine learning for storm surge forecasting during typhoons

This study addresses the challenge of accurately forecasting storm surges during tropical cyclones, which is crucial in reducing coastal damages. We develop a physics-informed and data-efficient model, a storm surge neural network (SSNN), that embeds shallow water equations into a machine learning (ML) framework with a specifically designed piecewise loss function to effectively incorporate physical constraints. The SSNN is iteratively trained to make rolling forecast and can produce more accurate forecasts than a standard data-driven model even with a smaller size of training data. The improvement in the performance by the SSNN model is evident, particularly in coastal areas with varying strong winds. This is especially true for forecasts beyond 16 hours, where the data-driven model struggles to provide accurate predictions. Combining physics with ML could offer a promising way to improve storm surge forecasting and support coastal hazard preparedness.