A convolutional neural network-based classification of local earthquakes and tectonic tremors in Sanriku-oki, Japan, using S-net data

Abstract Low-frequency tremors have been widely detected in many tectonic zones, and are often located adjacent to megathrust indicating that their spatiotemporal evolution provides important insights into events. The envelope correlation method (ECM) is commonly used detect tremors. However, the ECM also detects regular earthquakes, which requires separation of these two signals after initial detection. In addition, weak, so classifying from noises an essential problem. We develop a convolutional neural network (CNN)-based using single S-net station off Sanriku region, Northeast Japan, classify local tremors, noise. Along Japan Trench, especially region focused this study, earthquakes occurred coexistence within small detection, location, discrimination events key understand relationship between slow earthquakes. spectrograms three-component velocity waveforms were recorded during 16 August 2016 14 2018 as training test datasets for CNN. CNN successfully classified 100%, 96%, 98% noise, respectively. showed successful application our continuous waveform data including tremor explore feasibility proposed noise streaming data. output probabilities true classifications decrease with increasing epicentral distance and/or decreasing event magnitude. This highlights need train proximal seismic stations detecting multiple stations.