Item

Abstract

Continuous seismic records include various earthquake signals such as ordinary fast earthquakes and slow earthquakes like tectonic tremors. A machine-learning-based automatic classification approach would allow us to process a large amount of waveform data and to understand geophysical phenomena around a target seismic network. In this study, we propose an automatic unsupervised classification algorithm for continuous records based on frequency characteristics.Our proposed algorithm first extracts frequency features by calculating running spectra, and then the vector quantization is performed in the feature space. After that, the data points are converted by the kernel principal component analysis and are clustered by the Ward hierarchical clustering algorithm in the mapped space. Finally, classification results are obtained by cutting the dendrogram at 1/4 of the maximum height.We tested the proposed algorithm by applying to one-week-long continuous waveforms recorded at five temporary ocean-bottom seismometers to observe aftershocks of the 2004 M7.4 off the Kii Peninsula earthquake (Yamazaki et al., 2008). For every station, tectonic tremors with large amplitudes were assigned for unique class(es) different from those for background noises and fast earthquakes, indicating that the proposed algorithm successfully detected tectonic tremors with a good S/N ratio. We also compared the classification results to a tremor catalog compiled by manual inspection and found that the detection rate was 87%, which suggested that the proposed algorithm could detect tremors with a high detection rate although the algorithm did not use specific knowledge of tectonic tremor such as template.