← Back to Sessions

Deep Learning-based Detection of Explosions and Earthquakes in South Korea

Description: 

The classification of earthquake and explosive signals in seismology is imperative for understanding the physical mechanisms of explosions and for accurately analyzing and interpreting earthquakes. Individual signals can be classified as explosions and earthquakes based on physical factors such as differences in P/S amplitudes, differences between regional and coda magnitudes, and focal depths of earthquakes. However, accurate classification is occasionally not achievable when the data lay on the decision boundary of classifiers. Furthermore, as the magnitude of detectable events decreases due to improvements in networks and sensors, technologies such as machine learning that can automatically process substantial amounts of signals are required, and thus massive volume of label data are naturally needed to improve the performance of supervised machine learning. Located inside the Asian tectonic plate, South Korea has low seismic activity, which makes it easier to discriminate earthquakes and explosions generated from mining operations distributed across the country. In this study, we detect body waves from explosions and earthquakes using machine learning-based phase detection techniques applied to 421 stations in South Korea, which were augmented for early warning after recent two Mw 5 earthquakes. Automatically detected phases are associated into events based on the backpropagation method. We detected and confirmed ~150,000 events in 7 years from 2016 through 2022. Over 180 clusters containing more than 100 events were identified based on differential travel time measurements and clustering algorithms, and each cluster was classified as an earthquake or explosion based on the proximity of the epicenters to mines and consistency in onset times of day. We verified over 100,000 explosions from clustering analysis. They can be used for the discrimination of explosions and earthquakes and analysis of physical mechanisms of explosions. In parallel, the microseismicity is useful for characterizing previously unidentified fine-scale active faults, characterizing intraplate swarms, and analyzing aftershock sequences.

Session: Network Seismology: Recent Developments, Challenges and Lessons Learned

Type: Oral

Date: 4/20/2023

Presentation Time: 10:15 AM (local time)

Presenting Author: Jeong-Ung Woo

Student Presenter: No

Invited Presentation: 

Authors