5:15 PM - 6:45 PM
[U15-P35] Estimation of the Strong Motion Generation Areas of the 2024 Noto Peninsula Earthquake based on Envelope Inversion
Keywords:2024 Noto Peninsula Earthquake, GIT, cGAN, Envelope Inversion, Strong Motion Generation Areas
1. Introduction
As rupture process estimation for short-period components with periods of 1 second or less, envelope inversion (Kakehi and Irikura, 1996; Nakahara, 1998) and estimation of Strong Motion Generation Areas (SMGA) using EGF (Miyake et al., 2003) have been conducted. The authors have developed a method (SS-GMG) that combines generalized inversion technique (GIT) and conditional generative adversarial networks (cGAN), a type of machine learning, to output site-specific acceleration time history waveforms with variations in Mw, distance, and observation points (Yamaguchi et al., 2024). In this study, acceleration waveforms generated by SS-GMG are used as elemental waveforms from sub-faults to estimate the short-period generation areas with a period of less than 1 second for the 2024 Noto Peninsula earthquake using envelope inversion, and the SMGA is estimated based on the results.
2. SS-GMG model
In GIT, records with Mj 4.0 to 6.5 and source distances of 150 km or less were used to evaluate Q values for propagation path characteristics, empirical site characteristics for each observation point, and source spectra for each earthquake. The mean stress drop for all earthquakes was 3.1 MPa. A model was constructed to predict the mean and standard deviation of the Fourier Amplitude Spectrum (FAS) based on the source spectrum of ω-2 and the GIT results. For the temporal characteristics, a time history model was constructed by learning the observed waveforms using cGAN. Given a random number and conditions (Mw, distance, depth, and observation point), the model outputs a set of time waveforms whose amplitudes are normalized from -1 to 1. The model outputs waveforms that simultaneously satisfy the mean and standard deviation of the FAS by multiplying the standardized time history waveforms by a factor in the amplitude direction.
3. Envelope inversion
The multi-time window method was used to estimate the rupture process. While waveform inversion uses velocity waveforms to estimate the rupture process, this study uses acceleration envelopes. Therefore, the period of 1 second or less, when the waveforms from each small fault are considered to be randomly summed up together, is the target of the study.
Element seismic waveforms from each sub-fault to each station were generated assuming Mw 5.0 and a stress drop of 3.1 MPa. Ten cases of elemental waveforms were created. For the acceleration envelopes, the average of the NS and EW components were used in the observations, and the output result of one horizontal component was used in the calculations. The distribution of observation points used for inversion is shown in Figure 1.
The spatio-temporal distribution of the moment release of each small fault with respect to the moment of the elemental earthquake of Mw 5.0 was estimated using a non-negatively constrained least squares method to minimize the error between the envelope of the observed waveform and the envelope of the calculated waveform. In the time direction, we assumed 10 time window impulses at 1 second intervals. A condition of minimum discrete Laplacian was added for spatio-temporal smoothing, and the degree of smoothing was determined by ABIC. The velocity at which the fracture front spread was assumed to be 3.0 km/s by grid search. The final slip distributions estimated from the 10 cases of elemental seismic waves are shown in Figure 2.
4. SMGA Estimation
The averaged results of the slip estimated by envelope inversion for the 10 cases are also shown in Figure 1. The location, size, initial rupture point, and rupture start time of the SMGAs were referred to the envelope inversion results, and stress drop of each SMGA were estimated by GA. The rupture propagation velocity in the SMGA was fixed at 2.4 km/s. The estimated stress drops for each SMGA were 13.2, 12.2, 15.7, and 14.6 MPa, respectively. The short-period level was 3.50 × 1019 Nm/s2, which was also consistent with the scaling of Tohdo et al. (2022, 2023). Examples of acceleration waveforms evaluated using the estimated SMGA are shown in Figure 3, and a comparison of the observed and calculated envelopes is shown in Figure 4. Although the maximum acceleration is generally reproduced at many observation points, there are some points where there are deviations in the envelopes. Future issues include reconsideration of the number of SMGAs and estimation including parameters that are assumed to be fixed values.
As rupture process estimation for short-period components with periods of 1 second or less, envelope inversion (Kakehi and Irikura, 1996; Nakahara, 1998) and estimation of Strong Motion Generation Areas (SMGA) using EGF (Miyake et al., 2003) have been conducted. The authors have developed a method (SS-GMG) that combines generalized inversion technique (GIT) and conditional generative adversarial networks (cGAN), a type of machine learning, to output site-specific acceleration time history waveforms with variations in Mw, distance, and observation points (Yamaguchi et al., 2024). In this study, acceleration waveforms generated by SS-GMG are used as elemental waveforms from sub-faults to estimate the short-period generation areas with a period of less than 1 second for the 2024 Noto Peninsula earthquake using envelope inversion, and the SMGA is estimated based on the results.
2. SS-GMG model
In GIT, records with Mj 4.0 to 6.5 and source distances of 150 km or less were used to evaluate Q values for propagation path characteristics, empirical site characteristics for each observation point, and source spectra for each earthquake. The mean stress drop for all earthquakes was 3.1 MPa. A model was constructed to predict the mean and standard deviation of the Fourier Amplitude Spectrum (FAS) based on the source spectrum of ω-2 and the GIT results. For the temporal characteristics, a time history model was constructed by learning the observed waveforms using cGAN. Given a random number and conditions (Mw, distance, depth, and observation point), the model outputs a set of time waveforms whose amplitudes are normalized from -1 to 1. The model outputs waveforms that simultaneously satisfy the mean and standard deviation of the FAS by multiplying the standardized time history waveforms by a factor in the amplitude direction.
3. Envelope inversion
The multi-time window method was used to estimate the rupture process. While waveform inversion uses velocity waveforms to estimate the rupture process, this study uses acceleration envelopes. Therefore, the period of 1 second or less, when the waveforms from each small fault are considered to be randomly summed up together, is the target of the study.
Element seismic waveforms from each sub-fault to each station were generated assuming Mw 5.0 and a stress drop of 3.1 MPa. Ten cases of elemental waveforms were created. For the acceleration envelopes, the average of the NS and EW components were used in the observations, and the output result of one horizontal component was used in the calculations. The distribution of observation points used for inversion is shown in Figure 1.
The spatio-temporal distribution of the moment release of each small fault with respect to the moment of the elemental earthquake of Mw 5.0 was estimated using a non-negatively constrained least squares method to minimize the error between the envelope of the observed waveform and the envelope of the calculated waveform. In the time direction, we assumed 10 time window impulses at 1 second intervals. A condition of minimum discrete Laplacian was added for spatio-temporal smoothing, and the degree of smoothing was determined by ABIC. The velocity at which the fracture front spread was assumed to be 3.0 km/s by grid search. The final slip distributions estimated from the 10 cases of elemental seismic waves are shown in Figure 2.
4. SMGA Estimation
The averaged results of the slip estimated by envelope inversion for the 10 cases are also shown in Figure 1. The location, size, initial rupture point, and rupture start time of the SMGAs were referred to the envelope inversion results, and stress drop of each SMGA were estimated by GA. The rupture propagation velocity in the SMGA was fixed at 2.4 km/s. The estimated stress drops for each SMGA were 13.2, 12.2, 15.7, and 14.6 MPa, respectively. The short-period level was 3.50 × 1019 Nm/s2, which was also consistent with the scaling of Tohdo et al. (2022, 2023). Examples of acceleration waveforms evaluated using the estimated SMGA are shown in Figure 3, and a comparison of the observed and calculated envelopes is shown in Figure 4. Although the maximum acceleration is generally reproduced at many observation points, there are some points where there are deviations in the envelopes. Future issues include reconsideration of the number of SMGAs and estimation including parameters that are assumed to be fixed values.