Prediction of maximum tsunami height and arrival time using machine learning and inverse variance weighting

Abstract

Tsunamis can cause catastrophic damage to coastal communities when they strike the coast. This study aims to establish a rapid forecasting system for maximum tsunami height and arrival time using machine learning techniques. Imwon Port, located on the eastern coast of the Republic of Korea, was selected as the target site. To address uncertainties in seismic fault parameters and earthquake magnitude, a weighted logic tree approach was employed to generate a comprehensive tsunami dataset. Additionally, nine offshore observation points in the East Sea were considered to predict the maximum tsunami height and arrival time at Imwon Port. Before developing the machine learning models, gradient boosting was applied to identify important explanatory variables. Subsequently, machine learning models were constructed to predict maximum tsunami height and arrival time, and an inverse variance weighting method was introduced to combine the predictions from high-performing models. The estimates produced by the proposed machine learning framework showed good agreement with results from numerical simulations. The rapid predictions of tsunami characteristics enabled by this approach have the potential to reduce tsunami-related damages and save lives in vulnerable coastal areas.