ERROR EVALUATION IN TSUNAMI ARRIVAL TIME PREDICTION USING MACHINE LEARNING

Abstract

 In this study, we used machine learning to instantaneously predict the tsunami arrival time to town A in Mie Prefecture from water level data immediately after the occurrence of an earthquake, targeting the Nankai Trough earthquake. Prediction errors were compared between a model for each earthquake scale and a model using data for all earthquake scales. Although there was a difference between the models for each earthquake size, the error tended to decrease as the number of teacher data increased. In the case of the initial water level and tsunami arrival time used in this study, the overestimation of the initial water level and tsunami arrival time at Mw 8.6 was improved when the number of teacher data increased compared to Mw 9.0, where the overestimation was more than 5 minutes at many points. We also evaluated the accuracy of the predictions when using different earthquake scenarios of different magnitudes as the teacher data.