Learning Storm Surge with Gradient Boosting

Storm surge is a major natural hazard for coastal regions, responsible both for significant property damage and loss of life. Accurate, efficient models of storm surge are needed both to assess long-term risk and to guide emergency management decisions. While high-fidelity ocean circulation models such as the ADvanced CIRCulation (ADCIRC) model can accurately predict storm surge, they are very computationally expensive. Consequently, there have been a number of efforts in recent years to develop data-driven surrogate models for storm surge. While these models can attain good accuracy and are highly efficient, they are often limited to a small geographical region and a fixed set of output locations. We develop a novel surrogate model for peak storm surge prediction based on gradient boosting. Unlike most surrogate approaches, our model is not explicitly constrained to a fixed set of output locations or specific geographical region. The model is trained with a database of 446 synthetic storms that make landfall on the Texas coast and obtains a mean absolute error of 0.25 meters. We additionally present a test of the model on Hurricanes Ike (2008) and Harvey (2017).