A Study on Tsunami Arrival Time Prediction by Machine Learning

Abstract

 In this study, as an initial step for instantaneous tsunami arrival time prediction after an earthquake, we used images of the initial water level obtained from numerical tsunami simulations for the Nankai Trough to predict the tsunami arrival time using machine learning. The arrival times were pre-processed and colored every 3 minutes and every 2 minutes for training. The accuracy of the tsunami arrival time prediction was improved by increasing the number of training data, and the error of the tsunami arrival time was within 3 minutes in some places. In addition, we compared the prediction accuracy of A town and B city in Mie prefecture to confirm the difference. Although the inundation area is overestimated and the arrival time is currently overestimated or underestimated, the arrival time of the tsunami is generally predicted well by machine learning based on the initial water level information.