[3L1-OS-3a] OS-3

[3L1-OS-3a-04] Machine-learning-based detection of slow slip signal in the GNSS displacement time series of southwest Japan

〇Yusuke Tanaka¹, Masayuki Kano¹ (1. Graduate School of Science, Tohoku University)

Keywords:signal detection, earthquake forecast, slow slip events, crustal deformation

Automatic detection of seismic wave signals based on machine learning enhances the analysis of seismic big data. On the other hand, for the crustal deformation, Global Navigation Satellite System (GNSS) have been recently revealed transient fault slip called slow slip event (SSE). However, the detection of SSEs based on machine learning has rarely been attempted. Therefore, we conducted a single-site detection using CNN, from the real GNSS time series in southwest Japan where the detection of SSEs is the most abundant. We created approximately 16000 training data of noise and SSE signals with time window of 121 days. Then, we followed the model architecture which was originally developed for seismic wave detection. As a result of training, we obtained 80–90% of accuracy within 50 epochs. The smaller signals showed higher false negative ratio. The response to the noise data was quite complicated, reflecting the spatiotemporal variation of noise characteristics.