日本地球惑星科学連合2022年大会

講演情報

[E] 口頭発表

セッション記号 S (固体地球科学) » S-CG 固体地球科学複合領域・一般

[S-CG45] 地球深部とダイナミクス理解の新展開

2022年5月22日(日) 10:45 〜 12:15 301A (幕張メッセ国際会議場)

コンビーナ:渡辺 寛子(東北大学ニュートリノ科学研究センター)、コンビーナ:阿部 なつ江(国立研究開発法人海洋研究開発機構研究プラットフォーム運用開発部門マントル掘削プロモーション室)、小俣 珠乃(国立研究開発法人海洋研究開発機構)、コンビーナ:McDonough William F(Department of Earth Science and Research Center for Neutrino Science, Tohoku University, Sendai, Miyagi 980-8578, Japan)、座長:阿部 なつ江(国立研究開発法人海洋研究開発機構研究プラットフォーム運用開発部門マントル掘削プロモーション室)、渡辺 寛子(東北大学ニュートリノ科学研究センター)

11:45 〜 12:00

[SCG45-05] Big data of small earthquakes to reveal the crustal stress in Japan

★Invited Papers

*内出 崇彦1椎名 高裕1今西 和俊1 (1.産業技術総合研究所 地質調査総合センター)

キーワード:地震、応力場、応力インバージョン解析、深層学習、震源メカニズム解

For solid earth studies, seismic observations, in-situ ones (such as ones by drilling) for the current status, and geological ones for the past have their own merits and complement each other. Seismic waves are radiated from and travel through the earth's interior, and therefore, they tell us the current structure and the dynamics of the earth.


Here, we focus on the properties of small earthquakes that indicate the stress field around the source regions. The stress field at depth is essential to understand the tectonics and assess the earthquake occurrence. Focal mechanisms of earthquakes, corresponding to the fault orientation and the slip direction, are influenced by the stress orientation and, therefore, they indicate the stress orientation. Since small earthquakes occur much more frequently than large earthquakes, we can obtain a large amount of data for small earthquakes.


Such massive data requires an automated and efficient process. The focal mechanism determination needs P-wave first-motion polarities, whether the first motion in the vertical direction is upward or downward. The automated picking of P-wave first-motion polarities is not easy since the very first motions are close to the noise level. Therefore, we introduced deep learning using existing polarity data by manual picking, including the Japan Meteorological Agency's catalog for earthquakes with magnitude (M) greater than 3. Since relatively simple neural network models had worked (e.g., Ross et al., 2018; Hara et al., 2019), we constructed a neural network model with four convolution layers and two fully-connected layers (Uchide, 2020). We applied the trained model to more than 660 thousand earthquakes with M < 3 at depths < 20 km which occurred in Japan in 2003 - 2020. Using the obtained first-motion polarities and HASH code (Hardebeck and Shearer, 2002), we finally derived reasonable focal mechanism solutions for more than 210 thousand earthquakes. The automated process enabled us to deal with much more data than prior studies (e.g., Terakawa and Matsu'ura, 2010; Yukutake et al., 2015).


Using the derived focal mechanism solutions, we estimated the spatial distribution of the crustal stress orientations and stress ratios over the Japanese Islands. The method is based on Hardebeck and Michael (2006). The spatial smoothing was optimized to minimize Akaike's Bayesian Information Criterion (ABIC) (Akaike, 1980; Yabuki and Matsu'ura, 1992). We computed the estimation error of horizontal principal stress axes by a bootstrap method, and we eliminated the stress estimations with an error of more than 15 degrees.


The obtained stress map indicates the overall east-west compressional stress and the regional variations, in addition to local stress anomalies. The stress orientation distribution has contrasts at some geological borders, such as Hinagu and Futagawa faults in Kyushu, Median Tectonic Line in Shikoku, and Itoigawa-Shizuoka Tectonic Line in Shizuoka. The stress map helps understand the overall vision of the crustal stress in Japan and further study the local stress distribution.


Acknowledgment
We used seismic data from NIED Hi-net, JMA, and GSJ at AIST, and the JMA Unified Earthquake Catalog. This work was partially supported by the EDGE Runners Project of AIST. Deep learning was done using AI Bridging Cloud Infrastructure (ABCI) of AIST.


References:
Akaike, H. (1980) in Bayesian Statistics. doi: 10.1007/BF02888350
Hara, S., Fukahata, Y., & Iio, Y. (2019) EPS. doi:10.1186/s40623-019-1111-x
Hardebeck, J.L., & Shearer, P. M. (2002) BSSA. doi:10.1785/0120010200
Hardebeck, J. L., & Michael, A. J. (2006) JGR. doi:10.1029/2005JB004144
Ross, Z. E., Meier, M.-A., & Hauksson, E. (2018) JGR. doi:10.1029/2017JB015251
Terakawa, T., & Matsu'ura, M. (2010) Tectonics. doi:10.1029/2009TC002626
Uchide, T. (2020) GJI. doi:10.1093/gji/ggaa401.
Yabuki, T., & Matsu'ura, M. (1992) GJI. doi:10.1111/j.1365-246X.1992.tb00102.x
Yukutake, Y., Takeda, T., & Yoshida, A. (2015) JGR. doi:10.1016/j.epsl.2014.12.005