5:15 PM - 6:45 PM
[HDS11-P09] Fault Model Estimation of the 1946 Showa Nankai Earthquake Based on the Nonlinear Joint Inversion Method
Keywords:Tsunami, Nonlinear inversion, Joint inversion, Levenberg-Marquardt algorithm
In this study, we developed a new nonlinear inversion method that considers tsunami run-up calculations and estimated a fault model for the 1946 Showa Nankai earthquake. The tsunami calculations were performed using a nonlinear long-wave equation with three-layer nesting domains with 18-, 6-, and 2-arcsec grid spacings. An integral was set at 3 h with a time step width of 0.1 s. For tsunami excitation, the effects of horizontal displacement of the seafloor slope and the Kajiura filter were considered. Numerical tsunami calculations were performed repeatedly using the nonlinear long-wave equation on a large computer, and the residuals between the theoretical and observed values were used to modify the amount of slip by the Levenberg-Marquardt method (Levenberg 1944; Marquardt 1963). As an initial model of the fault, we used the fault model of the 1946 Showa Nankai Earthquake proposed by Annaka et al. (2003), but it was divided into eight subfaults.
We used 122 tsunami trace heights higher than 0.5 m and within 100 m from the shoreline with confidence level A in the Tsunami Trace Database of Tohoku University. For the tsunami waveforms, we used 3 h of tsunami waveforms at 1-minute intervals since the earthquake at seven tide gauges in Hosojima, Uwajima, Sakai, Morozaki, Fukue, Uchiura, and Ito. For crustal deformation data, we used data from 49 sites used by Murotani et al. (2015). In Model 1, we used only tsunami trace heights. In Model 2, we performed joint inversion using the tsunami trace heights, tsunami waveforms, and crustal deformation. The weights among the three datasets in Model 2 were adjusted for each iteration. The ratio of the sum of squared residuals of the crustal deformation, tsunami waveform, and tsunami trace height was assumed to be 1:2:20.
The K-κ and the residual sum of squares of the tsunami trace height were 0.99-1.26,43.8 m2 for Model 1 and 0.96-1.39,82.0 m2 for Model 2, respectively, both of which satisfied the criterion values (0.95<K<1.05, κ<1.45) for good reproducibility. However, there was a difference in the reproducibility of crustal deformation, with Model 2 explaining the observed data better. From the obtained slip distribution, we estimated the seismic moment and the moment magnitude to be 6.02×1021 Nm and 8.45 for model 1 and 5.38×1021 Nm and 8.42 for model 2 assuming a rigidity of 50 GPa. Model 1 showed a large slip from off Shikoku to off the Kii Channel, with a maximum slip of 7.16 m off Shikoku. Model 2, on the other hand, has a slip distribution similar to that of the Baba et al. (2002) fault model but differs from Baba's model in that a maximum slip of 7.18 m was observed near the epicenter.
We used 122 tsunami trace heights higher than 0.5 m and within 100 m from the shoreline with confidence level A in the Tsunami Trace Database of Tohoku University. For the tsunami waveforms, we used 3 h of tsunami waveforms at 1-minute intervals since the earthquake at seven tide gauges in Hosojima, Uwajima, Sakai, Morozaki, Fukue, Uchiura, and Ito. For crustal deformation data, we used data from 49 sites used by Murotani et al. (2015). In Model 1, we used only tsunami trace heights. In Model 2, we performed joint inversion using the tsunami trace heights, tsunami waveforms, and crustal deformation. The weights among the three datasets in Model 2 were adjusted for each iteration. The ratio of the sum of squared residuals of the crustal deformation, tsunami waveform, and tsunami trace height was assumed to be 1:2:20.
The K-κ and the residual sum of squares of the tsunami trace height were 0.99-1.26,43.8 m2 for Model 1 and 0.96-1.39,82.0 m2 for Model 2, respectively, both of which satisfied the criterion values (0.95<K<1.05, κ<1.45) for good reproducibility. However, there was a difference in the reproducibility of crustal deformation, with Model 2 explaining the observed data better. From the obtained slip distribution, we estimated the seismic moment and the moment magnitude to be 6.02×1021 Nm and 8.45 for model 1 and 5.38×1021 Nm and 8.42 for model 2 assuming a rigidity of 50 GPa. Model 1 showed a large slip from off Shikoku to off the Kii Channel, with a maximum slip of 7.16 m off Shikoku. Model 2, on the other hand, has a slip distribution similar to that of the Baba et al. (2002) fault model but differs from Baba's model in that a maximum slip of 7.18 m was observed near the epicenter.