COMPUTATIONAL SIMULATION SET SELECTION FOR STORM SURGE SURROGATE MODELING

Flooding by tropical cyclones poses a significant threat to life and livelihood in coastal communities worldwide. Recent advances in machine learning have enabled efficient estimation of tropical cyclone surge using surrogate models (e.g., Kajbaf and Bensi 2020 and refs therein). In turn, the application of these surrogate models supports the probabilistic hazard characterization and flood warning essential for prosperity in the coastal zone. Typically, storm surge surrogate models are developed from discrete sets of high-fidelity storm surge simulations with physics-based models such as ADCIRC (e.g., Westerink et al., 2008). Herein, we explore the significance of storms represented in these physics-based simulation sets—hereafter training sets—as they relate to the performance of storm surge surrogate models.