09:15 〜 09:30
[SSS09-17] Non-ergodic Ground Motion Models for Japan Subduction Zone Earthquakes
キーワード：non-ergodic GMM, subduction zone, aleatory variability, path effect, seismic hazard
After the 2011 Tohoku earthquake, the local ground-motion models (GMMs) for the subduction zones of Japan were updated in the NIED (Morikawa and Fujiwara, 2013) and the NGA-Sub project (Si et al., 2022). In addition, three of the NGA-Sub global models also included terms for the regional differences in the ground motion scaling for the Japan region (e.g., Abrahamson and Gulerce, 2022). These GMMs are ergodic models in that they apply to all of Japan; however, many studies (e.g., Sung et al., 2022) have shown that there can be systematic source, path, and site effects for specific earthquake/site pairs which can be captured using the varying coefficient model. With the source/site-specific features, non-ergodic GMMs have shifted in the median and reduced aleatory variability as compared to ergodic GMMs. In this study, we modify a suite of alternative ergodic GMMs for Japan to capture the non-ergodic behavior from the systematic site and path effects for the interface and slab subduction zone earthquakes in Japan.
The non-ergodic path effects are modeled by the cell-specific linear-distance scaling which mimics the effects from a 2-D Q structure (Dawood and Rodriguez-Marek, 2010) and by source/site-specific constant terms that capture the path effects related to the 3-D velocity structure (Sung et al., 2023). The final model consists of the smooth spatially varying non-ergodic source, site, and path terms that can be applied to any source/site pair, not just the locations of earthquakes and sites in the data set. This allows the non-ergodic GMM to be applicable to probabilistic seismic hazard analyses. For the region with little or no data, the central estimate of the non-ergodic terms is close to zero, and the epistemic uncertainty for the non-ergodic terms is large. The standard deviation of the combined non-ergodic path effects (Q effects and 3-D velocity structure effects) is about 0.30-0.35 in natural logarithm units for all periods, which is similar to the result found for the other regions (e.g., Lin et al., 2011). Overall, the fully non-ergodic models lead to an aleatory variance of residual values for the GMMs that is reduced by 65% compared to ergodic GMMs. The combined effect of the shift in the median with the reduced aleatory variability can significantly affect seismic hazard calculations for the Japan region.
The non-ergodic path effects are modeled by the cell-specific linear-distance scaling which mimics the effects from a 2-D Q structure (Dawood and Rodriguez-Marek, 2010) and by source/site-specific constant terms that capture the path effects related to the 3-D velocity structure (Sung et al., 2023). The final model consists of the smooth spatially varying non-ergodic source, site, and path terms that can be applied to any source/site pair, not just the locations of earthquakes and sites in the data set. This allows the non-ergodic GMM to be applicable to probabilistic seismic hazard analyses. For the region with little or no data, the central estimate of the non-ergodic terms is close to zero, and the epistemic uncertainty for the non-ergodic terms is large. The standard deviation of the combined non-ergodic path effects (Q effects and 3-D velocity structure effects) is about 0.30-0.35 in natural logarithm units for all periods, which is similar to the result found for the other regions (e.g., Lin et al., 2011). Overall, the fully non-ergodic models lead to an aleatory variance of residual values for the GMMs that is reduced by 65% compared to ergodic GMMs. The combined effect of the shift in the median with the reduced aleatory variability can significantly affect seismic hazard calculations for the Japan region.