CONSTRUCTION OF A TSUNAMI WATER LEVEL ESTIMATION MODEL BASED ON FAULT PARAMETERS USING DEEP LEARNING

Abstract

 In the probabilistic assessment of the maximum tsunami water level, large-scale parameter studies are required to comprehensively account for uncertainties associated with earthquake occurrence. Therefore, reducing the computational cost of tsunami simulations for individual parameter settings is desirable. In this study, a deep learning (DL) model was developed to impute coastal water levels, serving as a surrogate for part of the tsunami simulation process, based on fault parameters and event locations. Tsunami propagation over virtual submarine topography was simulated for 2,592 cases with varying fault parameters to generate a foundational dataset for DL model training. By appropriately optimizing the hyperparameters, the DL model demonstrated sufficient accuracy and stability, confirming its potential as a reliable imputation-based surrogate method for tsunami simulation results.