Abstract
This study aims to evaluate the performance of the previous sequential Bayesian update for synthesizing tsunami scenarios in a diverse database consisting of complex fault rupture patterns with heterogeneous slip distributions. We utilize an existing database comprising 1771 tsunami scenarios targeting the city of Westport (WA, U.S.), which includes synthetic wave height records and inundation distributions resulting from a fault rupture in the Cascadia subduction zone. After preprocessing the training dataset according to the developed framework, Bayesian updates are performed sequentially to evaluate the probability that each training scenario is a test case. In addition to detecting the scenario with the highest probability, i.e., the most likely scenario, we synthesize the scenario by the weighted mean of all the learning scenarios by their probabilities. The accuracies of tsunami risk evaluation based on both resultant scenarios are evaluated from the maximum offshore wave, inundation depth, and its distribution. The results of the cross-validation with five different testing/training datasets showed that the weighted mean scenario has almost comparable performance to that of the most likely scenario. Additionally, the sequential Bayesian update improves the accuracy of both methods if a 3–4 minute observation time window is given, and has an advantage over the benchmark results provided by dynamic time warping with full-time series data.
