9:00 AM - 9:15 AM
[SCG52-01] Crustal stress inferred from focal mechanism solutions based on P wave first-motion polarities picked using deep learning
Keywords:Seismology, Crustal Stress, Focal Mechanisms, Stress Inversion, Deep Learning
Understanding of crustal deformation and forecast of inland earthquakes need knowledge on the crustal stress orientation. Estimated stress orientations have been compiled as the World Stress Map (Heidbach et al., 2018). In the case of the Japan Islands, Terakawa et al. (2010) estimated the stress orientations using moment tensor solutions from the routine catalog of NIED F-net for typically M 3.5 or greater. Yukutake et al. (2015) used focal mechanism solutions that were determined using manual picks of P wave first-motion polarities. The resolution of the stress field estimation is limited by data. Moreover, there are many areas with no focal mechanism solutions and then the stress field cannot be estimated in this way. There is a case that stress orientations inferred from focal mechanism solutions of microearthquakes were clearly different from their surroundings (e.g., Imanishi et al. (2012)), which suggests that the interpolation of the stress estimation does not always work. Therefore, we desire the focal mechanisms of even more earthquakes. Thanks to the recent development of deep learning, Uchide (2020) studies focal mechanisms of ~ 110 thousand events with M 1.5 or larger beneath Japan Islands. Using this technique, we estimate focal mechanisms of even more earthquakes and the crustal stress orientation of the Japan Islands.
First, we determined focal mechanisms of small to microearthquakes around Japan Islands by the HASH method (Hardebeck and Shearer, 2002) using P-wave first-motion polarities. Seismic data were from NIED Hi-net, JMA, and GSJ/AIST. P-wave arrival times were from the JMA catalog and first-motion polarities were picked by the neural network model of Uchide (2020). We selected earthquakes with JMA magnitudes greater than 0.5, inside of or within 50 km from coastlines of Japan Islands, and at depths less than 20 km. As a result, we analyzed ~ 573 thousand events, which is approximately five times more than those studies by Uchide (2020). We obtained “A” to “C” rank solutions for more than 195 thousand events, which was used for the following stress inversion analysis.
The stress orientation was estimated using the SATSI code (Hardebeck and Michael, 2006). The grid interval was 0.2° for both latitude and longitude. Several small-scale stress anomalies were seen: normal faulting areas in northeast Japan and stress anomaly around Cape Shionomisaki of Kii peninsula, beneath which a high-density block was detected (Honda and Kono, 2005). The spatial changes in the stress orientation were often seen on tectonic lines and faults, such as the southern end of Itoigawa-Shizuoka Tectonic Line, the Median Tectonic Line in Shikoku island, Nagato-Hida marginal Tectonic Line in Chugoku region, and Hinagu and Futagawa faults in Kyushu island. This spatial correspondence can be a clue to investigate the cause of the stress orientation changes.
Acknowledgment
We used phase data from JMA Unified Earthquake Catalog, and seismic data from NIED Hi-net, JMA, and GSJ/AIST. We also used the HASH code (Hardebeck and Shearer, 2002) and the SATSI code (Hardebeck and Michael, 2006). This work was supported by the AIST EDGE Runners project.
First, we determined focal mechanisms of small to microearthquakes around Japan Islands by the HASH method (Hardebeck and Shearer, 2002) using P-wave first-motion polarities. Seismic data were from NIED Hi-net, JMA, and GSJ/AIST. P-wave arrival times were from the JMA catalog and first-motion polarities were picked by the neural network model of Uchide (2020). We selected earthquakes with JMA magnitudes greater than 0.5, inside of or within 50 km from coastlines of Japan Islands, and at depths less than 20 km. As a result, we analyzed ~ 573 thousand events, which is approximately five times more than those studies by Uchide (2020). We obtained “A” to “C” rank solutions for more than 195 thousand events, which was used for the following stress inversion analysis.
The stress orientation was estimated using the SATSI code (Hardebeck and Michael, 2006). The grid interval was 0.2° for both latitude and longitude. Several small-scale stress anomalies were seen: normal faulting areas in northeast Japan and stress anomaly around Cape Shionomisaki of Kii peninsula, beneath which a high-density block was detected (Honda and Kono, 2005). The spatial changes in the stress orientation were often seen on tectonic lines and faults, such as the southern end of Itoigawa-Shizuoka Tectonic Line, the Median Tectonic Line in Shikoku island, Nagato-Hida marginal Tectonic Line in Chugoku region, and Hinagu and Futagawa faults in Kyushu island. This spatial correspondence can be a clue to investigate the cause of the stress orientation changes.
Acknowledgment
We used phase data from JMA Unified Earthquake Catalog, and seismic data from NIED Hi-net, JMA, and GSJ/AIST. We also used the HASH code (Hardebeck and Shearer, 2002) and the SATSI code (Hardebeck and Michael, 2006). This work was supported by the AIST EDGE Runners project.