Rapid classification of local seismic events using machine learning
Regional seismic networks often observe artificially induced seismic events such as blasting and collapses. Misclassified seismic events in the earthquake catalog can therefore interfere with assessments of natural seismic activity. Traditional methods rely on the period, and phase features of seism...
Saved in:
Published in | Journal of seismology Vol. 26; no. 5; pp. 897 - 912 |
---|---|
Main Authors | , , |
Format | Journal Article |
Language | English |
Published |
Dordrecht
Springer Netherlands
01.10.2022
Springer Nature B.V |
Subjects | |
Online Access | Get full text |
Cover
Loading…
Summary: | Regional seismic networks often observe artificially induced seismic events such as blasting and collapses. Misclassified seismic events in the earthquake catalog can therefore interfere with assessments of natural seismic activity. Traditional methods rely on the period, and phase features of seismic waves to determine the nature of seismic events. We designed three seismic event classifiers with reference to convolutional neural network structures such as VGGnet, ResNet, and Inception. The designed classifiers were tested and compared using three-channel seismic full-waveform time-series data and spectral data. Our classifiers are shown to only require 60 s of full-waveform seismic event data and first-arrival times for alignment; additional phase labeling or numerical filtering is unnecessary. Rapid classification of earthquakes, blasting, and mine collapses can be achieved within approximately 1 min of an event. As a test case, this study uses 6.4 k observations of actual local seismic events with magnitudes ranging from M
L
0.6 to M
L
4.5 obtained from 47 broadband seismic stations in the Henan Regional Network of the China Seismological Network Center; these observations include natural earthquakes, blasting, and collapse events. The results indicate that our classifiers can reach a lower classification magnitude limit of M
L
0.6 and that their recall and accuracy exceed 90%, outperforming manually performed routine classifications and similar approaches. These findings provide an important reference for the rapid classification of small and medium earthquakes. |
---|---|
ISSN: | 1383-4649 1573-157X |
DOI: | 10.1007/s10950-022-10109-5 |