Detection of seismic activity in South Korea using Machine learning

Jeongung Woo, & William L. Ellsworth

Submitted September 11, 2022, SCEC Contribution #12559, 2022 SCEC Annual Meeting Poster #065

South Korea, located hundreds of kilometers away from the nearest plate boundaries, is a tectonically inactive intraplate region with low levels of seismicity. Characterizing the source properties of intraplate earthquakes can be challenging due to the scarcity of earthquakes. Seismic monitoring networks in Korea have recently been upgraded and now contain over 300 seismic stations distributed across the peninsula. Two recent moderate > Mw 5 earthquakes in 2016 and 2017 provide a good opportunity to study such rare events. In this study, we create a comprehensive earthquake catalog of South Korea from the Korean seismic networks. Phase arrivals were picked using the deep-learning phase picker PhaseNet, and associated using back-projection techniques with a 1-D local velocity model. The hypocenters were initially determined with Geiger’s method and clustered before applying the double-difference relocation method to determine locations within each cluster. Our preliminary result successfully recalled 92% of the earthquakes reported by Korean Meteorological Administration (KMA) and all the undetected events in our catalog are located tens of kilometers away from seismic networks. We also found approximately three hundred additional seismic clusters with > 20,000 events in 2019, which is more than 20 times the number in the KMA catalog for the same period. Most of these events are located near mines visible from satellite, and probably do not represent earthquakes. The averaged S/P amplitude ratio was generally less than 1 for clusters close to mining sites and above 1 for confirmed seismic sequences. This suggests that the S/P ratio is useful for classification of mining blasts and earthquakes, and potentially provides plenty of training dataset for grouping events as blast signals. The continuing aftershocks of the two moderate earthquakes reveal structures similar to those revealed by immediate aftershocks. Automatic microseismic monitoring applied in this study confirms the utility of machine-learning based phase pickers in tectonically inactive regions with low levels of seismicity and shed light on seismic source characteristics in South Korea.

Key Words
Machine learning, Intraplate earthquake, Earthquake location

Woo, J., & Ellsworth, W. L. (2022, 09). Detection of seismic activity in South Korea using Machine learning. Poster Presentation at 2022 SCEC Annual Meeting.

Related Projects & Working Groups