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

A physics-informed storm surge neural network (SSNN) forecasts typhoon storm surge more accurately than a standard data-driven model, especially beyond 16 hours with smaller training datasets.

• Why it matters: Coastal damage depends on reliable storm surge predictions during tropical cyclones, yet purely data-driven approaches can fail when conditions change and data are limited.
• What they did: The SSNN embeds shallow water equations into an ML framework using a piecewise loss to enforce physical constraints, then is iteratively trained for rolling forecasts.
• The result: The SSNN’s accuracy improves most in coastal areas with varying strong winds and is particularly evident for forecasts beyond 16 hours, where the data-driven model struggles.

AI summary of the abstract · may contain errors · not medical advice · read the paper.