Surrogate modeling using Gaussian process regression for analyzing uncertainty associated with various tsunami scenarios of an electrical power infrastructure

The final objective of this study is to develop tsunami scenarios for coastal power plants that include site-specific uncertainties related to the maximum tsunami height, periodic variations, and the direction of tsunami.Initially, a framework was presented to quantify the uncertainties of the tsunami scenarios.Surrogate models for tsunami numerical calculations were constructed targeting the maximum tsunami height at the coastal line locations along the Pacific coast for the wave sources along the Japan Trench.In constructing the surrogate model, features representing the source characteristics in tsunami numerical calculations and the relative location to tsunami source and evaluation point were quantified, and the relationship between these features and the maximum tsunami heights at the coastal line was modeled using Gaussian process regression.Compared to discrete tsunami numerical calculation results, the surrogate model can provide a more detailed distribution of maximum tsunami height under continuous variation of input parameters.Through a detailed sensitivity analyses of parameters using the surrogate model, dominant parameters for the maximum tsunami height were identified.Furthermore, sensitivity analyses of the tsunami hazard curves generated by the surrogate model were conducted.