induced seismicity, machine learning, weiyuan shale gas block, Science, Geophysics. Cosmic physics, QC801-809, Environmental sciences, GE1-350

Seismic hazard assessment and risk mitigation depend critically on rapid analysis and characterization of earthquake sequences. Increasing seismicity in shale gas blocks of the Sichuan Basin, China, has presented a serious challenge to monitoring and managing the seismicity itself. In this study, to detect events we apply a machine-learning-based phase picker (PhaseNet) to continuous seismic data collected between November 2015 and November 2016 from a temporary network covering the Weiyuan Shale Gas Blocks (SGB). Both P- and S-phases are picked and associated for location. We refine the velocity model by using detected explosions and earthquakes and then relocate the detected events using our new velocity model. Our detections and absolute relocations provide the basis for building a high-precision earthquake catalog. Our primary catalog contains about 60 times as many earthquakes as those in the catalog of the Chinese Earthquake Network Center (CENC), which used only the sparsely distributed permanent stations. We also measure the local magnitude and achieve magnitude completeness of ML0. We relocate clusters of events, showing sequential migration patterns overlapping with horizontal well branches around several well pads in the Wei202 and Wei204 blocks. Our results demonstrate the applicability of a machine-learning phase picker to a dense seismic network. The algorithms can facilitate rapid characterization of earthquake sequences.