[S-TT43] Seismic Big Data Analysis Based on the State-of-the-Art of Bayesian Statistics

[STT43-P02] Seismic detection based on combination of station-wise phase picks by deep learning and application to dense seismic observation network data

*Tokuda Tomoki¹, Shutaro Sekine², Shintaro Abe², Naoshi Hirata¹, Hiromichi Nagao¹ (1.Earthquake Research Institute, The University of Tokyo, 2.Association for the Development of Earthquake Prediction)

Keywords:Seismic detection, Multiple stations, Deep learning

Recently, deep learning has gained much attention for seismic detection, replacing the conventional STA/LTA method based on abrupt change of amplitude. Due to the powerful representation capability of neural networks, a deep learning model is remarkably flexible for fitting to waveform data, which allows for high-performed seismic detection. On top of that, a further extension of the deep learning framework is to apply to multi-station waveforms. Using full information of waveforms observed at various stations, the multi-station framework can largely enhance EQ detection capability. Typically, seismic detection is performed in a station-wise manner, which is in turn combined to yield a network-based detection. However, a conventional approach involves complicated tuning of relevant parameters, which requires labeled waveform data as well as expertise in the field. In the present study, we use a simple but effective method for a network-based detection, which requires neither labeled seismic data nor complicated parameter tuning. We considered to combine station-wise phase picks (P-phase and S-phase) yielded by a pre-trained deep learning model (Generalized Phase Detection, GPD). Concretely, for a particular time-window, we defined a propensity score for earthquake based on maximum values of station-wise phases. For the propensity score, both the number of stations and cutoff value were determined in an unsupervised manner. We applied the method to continuous waveforms from Metropolitan Seismic Observation network (MeSO-net) and Nagaoka network (AN-net). We set the parameters using one-week continuous waveform data and performed earthquake detection. It was implied that the method effectively detected not only cataloged events but also non-cataloged ones.