12:00 PM - 12:15 PM
[SSS09-11] Estimating finite fault slip distributions of large earthquakes based on time-reversal wave propagation simulation with data assimilation
Larmat et al. (2006) proposed Time-Reversal Imaging (TRI), which has shown great success in earthquake source estimation. This method is different from general source inversion; the source imaging is performed by back-propagating array observation waveform data to the hypocenter by time inversion calculation of the equations of motion. This feature enables us to estimate the source without the prior setting of a fault plane. In addition to the significant large earthquakes, the application of the TRI to long-lasting phenomena such as tremors is being considered. On the other hand, TRI generally requires dense observation to obtain high-resolution source images (e.g., Kawakatsu and Montagner, 2008; Fukahata et al., 2014; Nakahara and Haney, 2015). Recently Furumura and Maeda (2020) introduced a calculation and observation data assimilation mechanism into TRI (DA-TRI), and showed that the K-NET/KiK-net data whose interstation distance is 20 -25 km can be used for source imaging of small and medium earthquakes occurring around Japan. In this study, we apply the TRI to estimate slip distribution on the finite fault of large earthquakes.
We set the DA-TRI calculation area as 460.8 x 460.8 km in horizontal and 57.6 km depth directions, respectively, in central Japan. The subsurface structure model of the JIVSM (Koketsu et al., 2012) was used for sedimentary layer, crust, mantle structure, and Pacific/Philippine Sea plate. Waveform data recorded at 234 KiK-net borehole stations for a duration of 180 seconds of two earthquakes that occurred off Ibaraki prefecture in 2008 (Mw6.9) and 2011 (Mw7.9) were backpropagated to the hypocenters by the 3D finite-difference calculation to estimate slip distribution (or asperity) on the fault plane, respectively. The optimal interpolation method was used for the data assimilation. The assimilation parameter was set based on the assumption that the error distribution of the seismic wavefield follows the Gaussian distribution of the correlation distance 5 km, and that the error of calculation and observation data is equivalent.
First, we analyzed the 2008 off-Ibaraki earthquake. DA-TRI was used to find the time variation of elastic energy of all grids in the finite difference method calculation (Top of Figure 2a). The time before and after the peak was regarded as the fault rupture time. The maximum value of elastic energy at the upper surface of the Pacific plate at that time was displayed as a source image (Bottom of Figure 2a). The obtained source image showed a strong concentration of elastic energy in a range of 25 km x 35 km, on about 10 km northeast of the hypocenter after five seconds from the earthquake origin time based on the catalog of the Japan Meteorological Agency. This position coincides well with the position of the large source fault slip estimated by Yamanaka (2008) from the far-field body wave analysis.
For the larger 2011 earthquake, the temporal variation of elastic energy (Top of Figure 2b) obtained from DA-TRI had two peaks at 2 and 23 seconds from the catalog origin time, suggesting the existence of two large fault slips around each time. Thus, we obtained the slip distribution of this earthquake by superimposing wave amplitudes on the Pacific plate at the two time-steps. A concentration of strong elastic energy was obtained in an area of 30x20 km at about 20 km southeast of the hypocenter, while a concentration of weak elastic energy in 30 x 20 km at 40 km northwest. The locations of energy concentration coincide with the asperity locations obtained from near-field strong motion records (Kubo et al., 2013), however, our results were unstable. This is because the calculation of the backpropagation of seismic waves continues to the past beyond the focal time, which we are tackling to improve.
We set the DA-TRI calculation area as 460.8 x 460.8 km in horizontal and 57.6 km depth directions, respectively, in central Japan. The subsurface structure model of the JIVSM (Koketsu et al., 2012) was used for sedimentary layer, crust, mantle structure, and Pacific/Philippine Sea plate. Waveform data recorded at 234 KiK-net borehole stations for a duration of 180 seconds of two earthquakes that occurred off Ibaraki prefecture in 2008 (Mw6.9) and 2011 (Mw7.9) were backpropagated to the hypocenters by the 3D finite-difference calculation to estimate slip distribution (or asperity) on the fault plane, respectively. The optimal interpolation method was used for the data assimilation. The assimilation parameter was set based on the assumption that the error distribution of the seismic wavefield follows the Gaussian distribution of the correlation distance 5 km, and that the error of calculation and observation data is equivalent.
First, we analyzed the 2008 off-Ibaraki earthquake. DA-TRI was used to find the time variation of elastic energy of all grids in the finite difference method calculation (Top of Figure 2a). The time before and after the peak was regarded as the fault rupture time. The maximum value of elastic energy at the upper surface of the Pacific plate at that time was displayed as a source image (Bottom of Figure 2a). The obtained source image showed a strong concentration of elastic energy in a range of 25 km x 35 km, on about 10 km northeast of the hypocenter after five seconds from the earthquake origin time based on the catalog of the Japan Meteorological Agency. This position coincides well with the position of the large source fault slip estimated by Yamanaka (2008) from the far-field body wave analysis.
For the larger 2011 earthquake, the temporal variation of elastic energy (Top of Figure 2b) obtained from DA-TRI had two peaks at 2 and 23 seconds from the catalog origin time, suggesting the existence of two large fault slips around each time. Thus, we obtained the slip distribution of this earthquake by superimposing wave amplitudes on the Pacific plate at the two time-steps. A concentration of strong elastic energy was obtained in an area of 30x20 km at about 20 km southeast of the hypocenter, while a concentration of weak elastic energy in 30 x 20 km at 40 km northwest. The locations of energy concentration coincide with the asperity locations obtained from near-field strong motion records (Kubo et al., 2013), however, our results were unstable. This is because the calculation of the backpropagation of seismic waves continues to the past beyond the focal time, which we are tackling to improve.