11:30 〜 11:45
[SSS05-19] Development of a Modified Matched Filter Technique for Foreshock Detection in Large-Scale Laboratory Rock Friction Experiments
Understanding the process of foreshock occurrence is crucial for solving fundamental, long-standing and debated topics in Seismology, related to the preparation phase of a large earthquake and the possibility of earthquake prediction. Laboratory experiments can help understand the mechanisms of foreshock occurrence. For example, Yamashita et al. (2021) detected more than 1,000 foreshocks in rock friction experiments and showed that the mainshock preparation process and related foreshock activity differ depending on the heterogeneity of the laboratory fault surfaces. As a continuation of previous work (Yamashita et al.,2021), the current study aims to detect a larger number of foreshocks and analyze their space-time evolution in more detail.
We used a catalog of foreshock events observed by piezoelectric acoustic sensors in large-scale rock friction experiments (Yamashita et al., 2021). To detect small events that were possibly missed due to the difficulty in the configuration of the detection threshold for various sizes of foreshocks, with the standard STA/LTA technique, we developed a modified approach based on the Matched Filter Technique (MFT) (e.g., Peng and Zhao, 2009). We computed the coherency, which is the frequency-domain cross-correlation between the reference (i.e., template) and target waveforms, normalized by its spectral amplitude (e.g., Prieto et al., 2009). The coherency is a metric of detection more sensitive to high frequency components compared to the standard cross-correlation function because of the spectral normalization. Thus, it offers an improved way of detecting small events with high-frequency components masked by the relatively low-frequency background noises. After computing the coherency, we found that sometimes the detection signals were not clear when stacked, due to the distance between the epicenter of the target and template events, or were missed because of large amplitude noise. To detect such events that were evaluated as non-significant detections by the coherency metrics, we also computed the kurtosis of the coherency as a new procedure. The detected signal was considered as an event if either coherency or kurtosis exceeded their respective thresholds. The thresholds for the detection by coherency and kurtosis were set to be 8 times the standard deviations of each of these measures. We used as templates 745 foreshock events with the moment magnitude, Mw, smaller than -5, listed in the catalog of Yamashita et al. (2021).
Using the method described above, we detected 60 events for a 0.5 seconds data window preceding a laboratory mainshock, which was about 3 times more than the number of events detected by Yamashita et al. (2021) for the same period. Among these events, 45 were detected by using coherency and 15 by using kurtosis. This result shows that some of the detected events were located relatively far from the hypocenters of the template events or had very small amplitudes on seismograms, and thus highlights the potential of the coherency- and kurtosis-based method. Our method may also lead to an improved detection of natural earthquakes.
We used a catalog of foreshock events observed by piezoelectric acoustic sensors in large-scale rock friction experiments (Yamashita et al., 2021). To detect small events that were possibly missed due to the difficulty in the configuration of the detection threshold for various sizes of foreshocks, with the standard STA/LTA technique, we developed a modified approach based on the Matched Filter Technique (MFT) (e.g., Peng and Zhao, 2009). We computed the coherency, which is the frequency-domain cross-correlation between the reference (i.e., template) and target waveforms, normalized by its spectral amplitude (e.g., Prieto et al., 2009). The coherency is a metric of detection more sensitive to high frequency components compared to the standard cross-correlation function because of the spectral normalization. Thus, it offers an improved way of detecting small events with high-frequency components masked by the relatively low-frequency background noises. After computing the coherency, we found that sometimes the detection signals were not clear when stacked, due to the distance between the epicenter of the target and template events, or were missed because of large amplitude noise. To detect such events that were evaluated as non-significant detections by the coherency metrics, we also computed the kurtosis of the coherency as a new procedure. The detected signal was considered as an event if either coherency or kurtosis exceeded their respective thresholds. The thresholds for the detection by coherency and kurtosis were set to be 8 times the standard deviations of each of these measures. We used as templates 745 foreshock events with the moment magnitude, Mw, smaller than -5, listed in the catalog of Yamashita et al. (2021).
Using the method described above, we detected 60 events for a 0.5 seconds data window preceding a laboratory mainshock, which was about 3 times more than the number of events detected by Yamashita et al. (2021) for the same period. Among these events, 45 were detected by using coherency and 15 by using kurtosis. This result shows that some of the detected events were located relatively far from the hypocenters of the template events or had very small amplitudes on seismograms, and thus highlights the potential of the coherency- and kurtosis-based method. Our method may also lead to an improved detection of natural earthquakes.