Japan Geoscience Union Meeting 2022

Presentation information

[J] Poster

H (Human Geosciences ) » H-DS Disaster geosciences

[H-DS10] Tsunami and tsunami forecast

Thu. Jun 2, 2022 11:00 AM - 1:00 PM Online Poster Zoom Room (15) (Ch.15)

convener:Satoko Murotani(National Museum of Nature and Science), convener:Hiroaki Tsushima(Meteorological Research Institute, Japan Meteorological Agency), Chairperson:Yuichi Namegaya(National Institute of Advanced Industrial Science and Technology), Satoko Murotani(National Museum of Nature and Science)

11:00 AM - 1:00 PM

[HDS10-P03] Tsunami Height Estimation by Gaussian Processing Regression Using Tsunami Height and Arrival Time at Seafloor Pressure Measurement Points in the Kii Peninsula, Japan

*Yutaro Iwabuchi1, Toshitaka Baba2, Takane Hori3, Masato Okada4, Yasuhiko Igarashi1 (1.Graduate School of Science and Technology, The University of Tsukuba, 2.Graduate School of Science and Technology, Tokushima University, 3.R&D Center for Earthquake and Tsunami, Japan Agency for Marine-Earth Science and Technology, 4.Graduate School of Frontier Science, The University of Tokyo)


Keywords:Early Tsunami prediction, Gaussian process, Oceanfloor Pressure gauge

To mitigate tsunami damage, early warning systems are operated around the world. In Japan, the Dense Oceanfloor Network System for Earthquakes and Tsunamis (DONET) was recently developed in the Nankai trough (Kaneda et al., 2015). DONET are equipped with seismometers and ocean-bottom pressure gauges at 51 points on the sea floor and submarine data can be acquired in real time. These data are useful for early prediction of tsunamis caused by earthquakes and submarine landslides.
The offshore tsunami heights and coastal tsunami heights are considered to be correlated, but the offshore tsunami heights become higher because of the overlapping when the tsunami approaches coastal area. Therefore, we studied the relationship between offshore and coastal tsunami heights to use simulation data at points of DONET oceanfloor pressure gauges for early tsunami prediction. For the early tsunami prediction, previous works used only the maximum absolute values of the hydrostatic pressure changes at the point of oceanfloor pressure gauges during a tsunami (Baba et al., 2014, Igarashi et al., 2016). Although compressing time series of pressure gauges data, we have presented improved tsunami height prediction algorithms from the hydrostatic pressure gauge data by linear regression (Baba et al., 2014) and Gaussian process regression (Igarashi et al., 2016).
However, the time series information is compressed by extracting the maximum value of the seafloor hydrostatic pressure gauge data, but in order to further improve the prediction accuracy, it is necessary to utilize the spatial characteristics of each sensor for prediction.
The time information from each sensor should not only be useful for predicting the arrival time of the tsunami, but for having information about the direction in which the tsunami arrives. Since the tsunami speed depends only on the water depth, the arrival time of the maximum tsunami height at the sensor point is determined by the location of the sensor point and the epicenter.
In this study, we propose a method to estimate the coastal tsunami height from training data similar to the input sensor data by Gaussian process in which the maximum hydrostatic pressure at the sensor point and the arrival time when the maximum hydrostatic pressure recorded are used as explanatory variables.
In order to verify the effectiveness of the arrival time as a feature, the accuracy of the results estimated from the maximum hydrostatic pressure data only and the results estimated from the maximum hydrostatic pressure and arrival time data were compared by RMSE. In addition, to compare the robustness of the model to measurement errors, the estimation accuracy was also compared when the validation data were subjected to normal distribution noise. Moreover, the effectiveness of the arrival time would be verified by the arrival time was selected or not as a highly influential feature by sparse modeling.
The verage of RMSE of the maximum tsunami heights estimated at 19 coastal cities with the noise variance from 0 to 1 is shown in the representative figure.
The proposed method reduces RMSE by using arrival time in the entire range of variance 0 to 1 of normally distributed noise. In the sparse modeling by ARD, the arrival time remained for all cities. In the estimation for some coastal cities, the arrival time was selected at the sensor points in the direction of tsunami intrusion into the port, indicating that the arrival time functions as a feature that indicates geographical characteristics as expected.