Automatic seismic station assessment in real-time with Deep Learning
Abstract
The increase in the number of seismic stations allows us to improve the detection and to shorten the reporting time for earthquake early warning system (EEW). However, the more increasing the number of stations, the more difficult to manage the quality of seismic data. The goal of this study is to apply an automatic method for seismic station assessment to a real-time seismic monitoring system. For the assessment, we used the power spectral density (PSD) of seismogram, which has characteristics of background seismic noise over a wide range of frequencies (McNamara et al, 2004). The PSDs from seismic networks in South Korea between 2016 and 2017 were manually labeled into three categories based on the maintenance report and visual inspection of seismograms: quiet (e.g. background seismic noise), bad (e.g. anthropogenic noise or instrumental glitches), and event condition (e.g. local and regional earthquake). We trained a convolutional neural network on the labeled PSDs for the classification. It is shown that the trained model could assess the condition of the seismic station successfully. For real-time application, we used the Earthworm that supports earthquake monitoring, management of seismic network, and seismic processing. The trained model was modularized to assess seismic station within the Earthworm.
- Publication:
-
AGU Fall Meeting Abstracts
- Pub Date:
- December 2019
- Bibcode:
- 2019AGUFM.S53G0549K
- Keywords:
-
- 4341 Early warning systems;
- NATURAL HAZARDS;
- 7212 Earthquake ground motions and engineering seismology;
- SEISMOLOGY;
- 7215 Earthquake source observations;
- SEISMOLOGY