11:00 AM - 1:00 PM
[HDS10-P05] Sensitivity Analysis of Tsunami Hazard Curve: A case study in the eastern Shikoku
Keywords:Probabilistic Tsunami Hazard Analysis, Hazard curves, Sensitivity analysis, Computational load reduction, G-R law
This study constructed hazard curves for tsunamis caused by the interplate earthquakes in the Nankai Trough. We relied on the G-R law defining the earthquake occurrence probability using the JMA earthquake catalog. Earthquake data used in the estimation of the G-R law were shallower than a depth of 50 km, greater than M5.0 from 1919 to 2020 under the sea in the Nankai region. We estimated the 95% confidence interval of the G-R law. Drawing tsunami hazard curves required many tsunami calculation results (tsunami DB). We used tsunami DB constructed in our previous study, where it solved the two-dimensional nonlinear long-wave equations for 3480 earthquake scenarios proposed by Fujiwara et al. (2020). The calculations applied a five-layer nesting domain to reduce the computational load. Based on the earthquake occurrence probability and tsunami DB, the tsunami hazard curves indicated the annual exceedance frequency of a tsunami height for the eastern Shikoku.
A comparison between the tsunami hazard station (J-THIS) and the tsunami hazard curve obtained in this study shows a higher probability at J-THIS. However, a simple comparison is not reasonable due to the different treatment of earthquake occurrence probability. We interpreted this as reflecting the difference between the Brownian Passage Time process adopted by J-THIS and the homogeneous Poisson process in this study.
We also investigated the sensitivity of the G-R law to the tsunami hazard curves by modifying the parameters construing in the G-R law. We, thus, changed the data period, magnitude, and epicentral depth. The tsunami hazard curves were out of the 95% confidence interval of the original tsunami hazard curves for the following three cases of (1) data period from October 1997 to March 2020, (2) only earthquakes with M>6.0, and (3) only earthquakes with M>7.0. On the other hand, epicentral depths were not sensitive to the tsunami hazard curves.
In addition, we proposed a method to reduce the number of earthquake scenarios for the tsunami computation load saving. The number of scenarios was reduced by simply thinning out the scenarios. As a result, even with the number of scenarios of a factor of 10, the tsunami hazard curve was still within the 95% confidence limit of the original tsunami hazard curve.
A comparison between the tsunami hazard station (J-THIS) and the tsunami hazard curve obtained in this study shows a higher probability at J-THIS. However, a simple comparison is not reasonable due to the different treatment of earthquake occurrence probability. We interpreted this as reflecting the difference between the Brownian Passage Time process adopted by J-THIS and the homogeneous Poisson process in this study.
We also investigated the sensitivity of the G-R law to the tsunami hazard curves by modifying the parameters construing in the G-R law. We, thus, changed the data period, magnitude, and epicentral depth. The tsunami hazard curves were out of the 95% confidence interval of the original tsunami hazard curves for the following three cases of (1) data period from October 1997 to March 2020, (2) only earthquakes with M>6.0, and (3) only earthquakes with M>7.0. On the other hand, epicentral depths were not sensitive to the tsunami hazard curves.
In addition, we proposed a method to reduce the number of earthquake scenarios for the tsunami computation load saving. The number of scenarios was reduced by simply thinning out the scenarios. As a result, even with the number of scenarios of a factor of 10, the tsunami hazard curve was still within the 95% confidence limit of the original tsunami hazard curve.