10:45 AM - 12:15 PM
[SSS07-P06] Approach of source inversion with synthetic local seismograms of the Nankai Trough earthquake
Keywords:Seismic source process analysis, Nankai Trough
The JMA is due to issue “Nankai Trough Earthquake Information (Extra)” early when a large earthquake of MW7.0 or greater occurred at the plate boundary along the Nankai Trough (Shimoyama et al., 2021). Using JMA’s seismograms obtained in real time, we early want to know the rupture area of the large earthquake along the Nankai Trough.
The source inversion with local seismograms is one of methods to estimate earthquake rupture area. Now, it is difficult to confirm validation of this inversion for Nankai Trough earthquake because no large earthquakes have occurred at the plate boundary along Nankai Trough under current seismic network.
Therefore we took the approach of making synthetic seismograms of current staitions assuming that an earthquake like the Showa Tonankai earthquake occured and carrying out source inversion with them as observed ones.
1. Source model
As a model, we chose a simulated earthquake similar to the Showa Tonankai earthquake in Hirose et al. (2022). In their study, they set about 13,000 triangle cells on the plate boundary in their simulations. We simplified them as about 1,300 grids to make synthetic seismograms. These grids on the boundary were set with about 11 km intervals in horizontal projection.
2. Making synthetic seismograms treated as observed
We obtained each Green function between each grid and JMA seismic station by reciprocity mode on OpenSWPC (Maeda et al., 2017) with JIVSM (Koketsu et al., 2012) as a 3D-structure. We made synthetic seismograms by convolving them with source processes at all grids.
3. Source inversion
(1) Setting fault plane
The fault plane was set as a flat plane including rupture starting point with the same strike, dip and rake of CMT of the earthquake. To simplify, we used the center of gravity and average strike, dip and rake by calculation from all moment at all grids instead of CMT. The source grids for inversion were evenly set on that plane and arranged as one of them matched rupture starting point.
(2) Inversion
The calculation conditions at source grids were constrained as follows.
Slip direction at each grid should be within ±45 degrees from the rake of the CMT solution. Source process was estimated to minimize the residual of observed seismograms and synthetic ones by convolving slips at source grids with Green functions calculated for inversion (e.g., Ide et al., 1996). Slip changes shoud be smooth. That hyperparameter was decided from ABIC (Fukahata et al., 2003).
When using 1D-structure in inversion, Green functions for inversion were calculated by the discrete wavenumber integration method (Bouchon, 1981). First wave arrival time of observed and synthetic seismic wave were adjusted at each station.
4. Results
In inversion, we used several 1D-structures and one 3D-structure which was JIVSM. In the cases of the 1D-structures, we couldn't obtain source process of the model earthquake. In the case of the 3D-structure, we obtained it, but it was not better.
Takemura et al.(2018) showed that it was needed to consider 3D-structure for MT inversion of earthquakes around the Nankai Trough. Nishimiya (2022) also reported that using 1D-structures in source process analysis of plate boundary earthquakes along the the Nankai Trough no good results may be obtained. It is difficult to determine Green functions properly for inversion with 1D-structures because the effect of structural inhomogeneity around the Nankai Trough is stronger than that e.g. off the Pacific coast of Tohoku district.
Our results are consistent with these.
The source inversion with local seismograms is one of methods to estimate earthquake rupture area. Now, it is difficult to confirm validation of this inversion for Nankai Trough earthquake because no large earthquakes have occurred at the plate boundary along Nankai Trough under current seismic network.
Therefore we took the approach of making synthetic seismograms of current staitions assuming that an earthquake like the Showa Tonankai earthquake occured and carrying out source inversion with them as observed ones.
1. Source model
As a model, we chose a simulated earthquake similar to the Showa Tonankai earthquake in Hirose et al. (2022). In their study, they set about 13,000 triangle cells on the plate boundary in their simulations. We simplified them as about 1,300 grids to make synthetic seismograms. These grids on the boundary were set with about 11 km intervals in horizontal projection.
2. Making synthetic seismograms treated as observed
We obtained each Green function between each grid and JMA seismic station by reciprocity mode on OpenSWPC (Maeda et al., 2017) with JIVSM (Koketsu et al., 2012) as a 3D-structure. We made synthetic seismograms by convolving them with source processes at all grids.
3. Source inversion
(1) Setting fault plane
The fault plane was set as a flat plane including rupture starting point with the same strike, dip and rake of CMT of the earthquake. To simplify, we used the center of gravity and average strike, dip and rake by calculation from all moment at all grids instead of CMT. The source grids for inversion were evenly set on that plane and arranged as one of them matched rupture starting point.
(2) Inversion
The calculation conditions at source grids were constrained as follows.
Slip direction at each grid should be within ±45 degrees from the rake of the CMT solution. Source process was estimated to minimize the residual of observed seismograms and synthetic ones by convolving slips at source grids with Green functions calculated for inversion (e.g., Ide et al., 1996). Slip changes shoud be smooth. That hyperparameter was decided from ABIC (Fukahata et al., 2003).
When using 1D-structure in inversion, Green functions for inversion were calculated by the discrete wavenumber integration method (Bouchon, 1981). First wave arrival time of observed and synthetic seismic wave were adjusted at each station.
4. Results
In inversion, we used several 1D-structures and one 3D-structure which was JIVSM. In the cases of the 1D-structures, we couldn't obtain source process of the model earthquake. In the case of the 3D-structure, we obtained it, but it was not better.
Takemura et al.(2018) showed that it was needed to consider 3D-structure for MT inversion of earthquakes around the Nankai Trough. Nishimiya (2022) also reported that using 1D-structures in source process analysis of plate boundary earthquakes along the the Nankai Trough no good results may be obtained. It is difficult to determine Green functions properly for inversion with 1D-structures because the effect of structural inhomogeneity around the Nankai Trough is stronger than that e.g. off the Pacific coast of Tohoku district.
Our results are consistent with these.