Tsunami height estimation via Gaussian process regression using the maximum absolute pressure change and time from seafloor sensors off the Kii Peninsula, Japan

Abstract The Dense Ocean-floor Network for Earthquakes and Tsunamis (DONET) was recently installed to monitor tsunamis in the Nankai Trough. In this study, an advanced tsunami prediction model using Gaussian process regression that is suitable for seafloor pressure observations is proposed. In traditional approaches, only the maximum absolute pressure change recorded by seafloor pressure sensors is used as an explanatory variable. The proposed method includes the time when the maximum absolute pressure change is recorded as an explanatory variable. Because tsunami data obtained at ocean observatories are insufficient for constructing Gaussian regression relationships, numerical tsunami simulations are used for learning and validation. After a tsunami is detected by DONET, the tsunami height prediction accuracy along the coast is increased by considering the time of the maximum absolute pressure change at seafloor pressure sensors. The proposed model enables rapid and effective estimation of coastal tsunami heights.