4:30 PM - 4:45 PM
[SSS08-05] Three-dimensional tectonic stress fields estimated from seismic data for long-term periods with direct prior information
Keywords:stress, earthquake, inversion
Stress is one of the important physical quantities for understanding earthquake generation. The CMT data inversion is a statistical method that derives the tectonic stress pattern from centroid moment tensor data of seismicity, based on the Bayesian statistical inference and Akaike’s Bayesian Information Criterion (ABIC) (Terakawa and Matsu’ura, 2008). With this method Terakawa and Matsu’ura (2010) estimated 3-D tectonic stress pattern in and around Japan from 12500 moment tensor data of seismicity (period: January 1997 to January 2007) listed in the F-net moment tensor catalog. More than ten years have passed after the study period of the previous study. New data have been certainly stored in the catalog.
The original method of CMT data inversion incorporates indirect prior information that stress fields are smooth to some degrees into observed data to construct a Bayesian model with hyperparameters. In order to estimate tectonic stress pattern from data for long-term periods, we improved the method of CMT data inversion so that we can consider results of stress fields obtained by a previous analysis as prior information as well as information from new observed data. In the improved method, we can incorporate both indirect and direct prior information into new observed data in a proper way, based on the formulation for geodetic inversion by Matsu’ura et al. (2007). In general, this improvement enables us to estimate the stress pattern from more data with smaller uncertainties. If uncertainties become larger in an analysis with the direct prior information and new data, it will indicate that statistical feature for the new data are inconsistent with the direct prior information obtained from previous data. This suggests that the stress pattern would have changed between the two study periods.
I targeted the Kyushu district in Japan where the 2016 Kumamoto earthquake (Mw 7.0) and the largest pre-shock (Mw 6.1) occurred on 16 and 14 April 2016. I applied moment tensor data of seismicity for 22 years (January 1997 to April 2019) listed in the F-net catalog to the improved method, and estimated the tectonic stress pattern in the district. I divided the whole study period into three periods: the period 1 (1 January 1997 to 31 January 2007), the period 2 (1 February 2007 to 13 April 2016), and the period 3 (14 Aril 2016 to 30 April 2019). Firstly, considering the results of stress patterns estimated from the data during the period 1 (Case 1) as the direct prior information, I estimated the stress pattern from data during the period 2 (Case 2). The stress pattern in the Kyushu district is characterized by normal to strike-skip faulting with north-south tension. The results of the stress pattern in Case 2 are consistent with those obtained only from the original data of the period 1 (Case 1). Furthermore, the uncertainties become smaller on the whole, indicating that we obtained more reliable results of the stress pattern from more data for the longer period. Next, considering the results of Case 2 as the direct prior information, I estimated the stress pattern from data during the period 3. The stress pattern which reflects all the data during the whole period (Case 3) is roughly consistent with those of Cases 1 and 2. However, the stress patterns at a shallow depth around the Futagawa fault, which is the main part of the source region of the Kumamoto earthquake, is different from those in Case 1 and 2, indicating that the stress pattern would have locally changed after the large earthquake. Furthermore, the uncertainties around the Futagawa fault becomes larger than those in Case 2, although the number of data there in Case 3 is larger than that in Case 2. These results suggest that types of earthquake would have changed after the Kumamoto earthquake. Large coseismic slip on the Futagawa fault basically released shear stress at the source region. Therefore, temporal changes in types of earthquake may be caused by some factors other than stress changes. A plausible mechanism is a decrease in fault strength due to an increase in pore fluid pressure, which triggers many aftershocks.
The original method of CMT data inversion incorporates indirect prior information that stress fields are smooth to some degrees into observed data to construct a Bayesian model with hyperparameters. In order to estimate tectonic stress pattern from data for long-term periods, we improved the method of CMT data inversion so that we can consider results of stress fields obtained by a previous analysis as prior information as well as information from new observed data. In the improved method, we can incorporate both indirect and direct prior information into new observed data in a proper way, based on the formulation for geodetic inversion by Matsu’ura et al. (2007). In general, this improvement enables us to estimate the stress pattern from more data with smaller uncertainties. If uncertainties become larger in an analysis with the direct prior information and new data, it will indicate that statistical feature for the new data are inconsistent with the direct prior information obtained from previous data. This suggests that the stress pattern would have changed between the two study periods.
I targeted the Kyushu district in Japan where the 2016 Kumamoto earthquake (Mw 7.0) and the largest pre-shock (Mw 6.1) occurred on 16 and 14 April 2016. I applied moment tensor data of seismicity for 22 years (January 1997 to April 2019) listed in the F-net catalog to the improved method, and estimated the tectonic stress pattern in the district. I divided the whole study period into three periods: the period 1 (1 January 1997 to 31 January 2007), the period 2 (1 February 2007 to 13 April 2016), and the period 3 (14 Aril 2016 to 30 April 2019). Firstly, considering the results of stress patterns estimated from the data during the period 1 (Case 1) as the direct prior information, I estimated the stress pattern from data during the period 2 (Case 2). The stress pattern in the Kyushu district is characterized by normal to strike-skip faulting with north-south tension. The results of the stress pattern in Case 2 are consistent with those obtained only from the original data of the period 1 (Case 1). Furthermore, the uncertainties become smaller on the whole, indicating that we obtained more reliable results of the stress pattern from more data for the longer period. Next, considering the results of Case 2 as the direct prior information, I estimated the stress pattern from data during the period 3. The stress pattern which reflects all the data during the whole period (Case 3) is roughly consistent with those of Cases 1 and 2. However, the stress patterns at a shallow depth around the Futagawa fault, which is the main part of the source region of the Kumamoto earthquake, is different from those in Case 1 and 2, indicating that the stress pattern would have locally changed after the large earthquake. Furthermore, the uncertainties around the Futagawa fault becomes larger than those in Case 2, although the number of data there in Case 3 is larger than that in Case 2. These results suggest that types of earthquake would have changed after the Kumamoto earthquake. Large coseismic slip on the Futagawa fault basically released shear stress at the source region. Therefore, temporal changes in types of earthquake may be caused by some factors other than stress changes. A plausible mechanism is a decrease in fault strength due to an increase in pore fluid pressure, which triggers many aftershocks.