Detection of low-frequency earthquakes by the matched filter technique using the product of mutual information and correlation coefficient

Abstract

The matched filter technique is often used to detect microearthquakes such as deep low-frequency (DLF) earthquakes. It compares correlation coefficients (CC) between waveforms of template earthquakes and the observed data. Conventionally, the sum of CC at multiple seismic stations is used as an index to detect the DLF earthquakes. A major disadvantage of the conventional method is drastically reduced detection accuracy when there are too few seismic stations. The new matched filter method proposed in this study can accurately detect microearthquakes using only a single station. It adopts mutual information (MI) in addition to CC to measure the similarity between the template and target waveforms. The method uses the product of MI and CC (MICC) as an index to detect DLF earthquakes. This index shows a distinct peak corresponding to an earthquake signal in a synthetic data set consisting of artificial noise and the waveform of a DLF earthquake. Application of this single-station method to field observations of Kirishima volcano, one of the most active volcanoes in Japan, detected a total of 354 events from the data in December 2010, whereas the catalog of the Japan Meteorological Agency shows only two. Of the detected events, 314 (89%) are likely DLF earthquakes and other events may be false detections. Most of the false detections correspond to surface-wave arrivals from teleseismic events. The catalog of DLF earthquakes constructed here shows similar temporal behavior to that found by the conventional matched filter method using the sum of the CC of the six stations near the volcano. These results suggest that the proposed method can greatly contribute to the accurate cataloging of DLF earthquakes using only a single seismic station. Graphical Abstract

Deep low-frequency earthquake activity associated with the 2018 eruptions in the Kirishima volcano group, Japan

“…We applied a new matched-lter technique that was developed by Kurihara et al (2021), which uses the product of the mutual information and correlation coe cient for earthquake detection, to the continuous data recorded at a single seismic station (N.SUKH in Fig. 1).…”

“…1). We applied a 1-8-Hz bandpass lter, 25-Hz sampling frequency, and 8-s window length to the waveforms prior to the analysis, as outlined in Kurihara et al (2021). We employed 293 DLF earthquakes from the April 2004-December 2018 period in the JMA catalog as template events.…”

“…The lack of template events from the target analysis period may be the reason for the apparent low rate of DLF earthquake activity associated with the 2018 eruptions. Here, we apply a new matched-lter technique that was developed by Kurihara et al (2021) to the continuous waveform data surrounding the 2018 eruptions, with a su cient number of template events employed for DLF earthquake detection, to more accurately capture the temporal variations in DLF earthquake activity. Furthermore, we conduct the same analysis on the waveforms associated with the 2011 eruptions.…”

Deep low-frequency earthquake activity associated with the 2018 eruptions in the Kirishima volcanic complex, Japan

“…In addition, because DLF earthquakes generally have small waveform amplitudes, they can be identified from waveforms recorded only at stations with high signal-to-noise (SN) ratios. Previous studies demonstrate that the SN-ratio of station N.SUKH is the highest around the Kirishima volcano (Kurihara et al 2019(Kurihara et al , 2021. Thus, template matching using waveform at the single station N.SUKH is expected to lead to a more comprehensive detection of DLF earthquakes compared to the previous studies.…”

“…Here, we apply a new matched-filter technique that was developed by Kurihara et al (2021) to the continuous waveform data surrounding the 2018 eruptions, with a sufficient number of template events employed for DLF earthquake detection, to more accurately capture the temporal variations in DLF earthquake activity. Furthermore, we conduct the same analysis on the waveforms associated with the 2011 eruptions.…”