Japan Geoscience Union Meeting 2022

Presentation information

[J] Oral

S (Solid Earth Sciences ) » S-CG Complex & General

[S-CG51] Driving Solid Earth Science through Machine Learning

Sun. May 22, 2022 9:00 AM - 10:30 AM 102 (International Conference Hall, Makuhari Messe)

convener:Hisahiko Kubo(National Research Institute for Earth Science and Disaster Resilience), convener:Yuki Kodera(Meteorological Research Institute, Japan Meteorological Agency), Makoto Naoi(Kyoto University), convener:Keisuke Yano(The Institute of Statistical Mathematics), Chairperson:Masaru Nakano(Japan Agency for Marine-Earth Science and Technology), Shinya Katoh(Disater Prevention Research Institute, Kyoto University), Hisahiko Kubo(National Research Institute for Earth Science and Disaster Resilience)

10:00 AM - 10:15 AM

[SCG51-05] Detection and hypocenter determination of volcanic earthquakes using Machine Learning: Application to Kirishima volcano

*Yohei Yukutake1, Ahyi KIM2 (1.Earthquake Research Institute, University of Tokyo, 2.Yokohama City University)

Keywords:Machine Learning, Volcanic earthquake

Volcanic earthquakes are triggered by the migration of magmatic fluid or related stress changes and are an essential indicator for evaluating volcanic activity. In a conventional procedure, the volcanic earthquakes were detected using a threshold such as STA/LTA. Their hypocenter and magnitudes were determined based on the information of phase-picking for the arrival times. The phase-picking data is often checked visually, which takes much time and makes it difficult to process in real-time. Several methods using machine learning frameworks have been developed for earthquake detection and phase-picking (e.g., Ross et al., 2018; Zhu & Beroza, 2018). However, these methods were mainly based on learning models obtained from the training data based on the crustal earthquakes, which may be difficult to apply to earthquakes in volcanic regions. Kim et al. (2021, SSJ) applied the architecture of PhaseNet developed by Zhu & Beroza (2018) to detect volcanic earthquakes. Then, they developed the learning model for the volcano earthquakes by using the training data obtained from the seismic catalog of 30,000 volcanic earthquakes in Hakone volcano from 1999 to 2020. In the present study, we applied this framework for the seismic data at Kirishima volcano. We used the continuous seismic waveform data obtained at 30 permanent stations in and around the Kirishima volcano. For the phase-picking data obtained by the framework, we conducted the phase association using the method by Zhang et al. (2019). We obtained the hypocenters of approximately 17.000 events from 2017 to 2020. The hypocenters were relocated by the Double-difference method (Waldhauser and Ellsworth, 2000), using the differential arrival times based on the phase-picking data by the machine learning frameworks and cross-correlation. Numerous small volcanic earthquakes were detected in the shallow areas of Shinmoe-dake and Iwo-yama, indicating an increase in seismic activity near the craters several months prior to the 2017-2018 eruption. This result also suggests that the present method may also be helpful for real-time monitoring of volcanic earthquakes.

Acknowledgments
We used the seismic waveform data obtained by the Japan Meteorological Agency, the National Research Institute for Earth Science and Disaster Prevention, Kyushu University, and the Earthquake Research Institute of the University of Tokyo. JMA provided us with the hypocenter catalog of volcanic earthquakes. This research was supported by the Next Generation Volcano Research and Human Resource Development Project of the Ministry of Education, Culture, Sports, Science and Technology of Japan.