Project title: Global operational storm surge prediction using scientific machine learning and satellite altimetry
PI: Thomas Adcock
Storm surge events will increase in frequency and intensity due to sea-level rise. Critical to mitigating their devastating socioeconomic impacts is the ability to accurately forecast them. This project leverages SciML to learn the oceanic response to compound forcing and develops a global operation storm surge forecast system.
Conventionally, storm surges are predicted using numerical simulation. These approaches are computationally expensive and sensitive to inaccuracies in input data. Both aspects are problematic in developing countries. While standard ML avoids computational limitations, prediction of extremes is difficult as these events are infrequent in data. SciML combines the computational efficiency of ML with physics-informed extrapolation allowing for the exploitation of new data sources, namely satellite altimetry.
This project leverages SciML to produce global operational models for tidal and storm surge prediction. These non-parametric physics-informed approaches enable the first empirical studies of dynamics of sea levels under compound forcing and the usage of satellite altimetry. Causal ML will help answer fundamental questions about how oceanic processes respond to climate change: specifically, why, and how tides are changing? This work illustrates how ML can enable scientific discovery and prediction of extreme events. The global models will be invaluable to developing countries that lack forecasts.
