Analysis of seismicity in the Haicheng-Xiuyan region based on dense array data and deep learning methods

The aftershocks of the 1975 M_S7.3 Haicheng and 1999 M_S5.4 Xiuyan earthquakes have persisted for a long time. The ChinArray-III dense stations, deployed in eastern North China from 2018 to 2020, increased seismic monitoring capability in the Haicheng-Xiuyan region, which can facilitate the construction of high-precision earthquake catalogs to better clarify the fault structures and seismogenic mechanisms of the two earthquakes. In this study, we selected 15 permanent stations and 37 ChinArray-III stations within 150 km of the epicenter of the Haicheng Earthquake. Next, we used deep learning methods to pick P- and S-wave phases from continuous waveforms recorded at these stations from January 2018 to July 2020. Based on these picks, we constructed an automatic earthquake catalog of the Haicheng-Xiuyan region. Compared with the routine manual catalog by China Earthquake Networks Center (CENC), our catalog contains 9.7 times more seismic events, including 98.3% of the seismic events in the CENC catalog, and has a lower magnitude of completeness (M_c = 1.1 vs M_c = 1.8 for the CENC catalog). The relocated events indicate that the strike of the Haichenghe-Dayanghe fault varies considerably from northwest to southeast, indicating that the fault bends slightly around the hypocenter of the 1975 M_S7.3 Haicheng earthquake which may act as a channel for fluid migration. The weak seismicity in the area between Haicheng and Xiuyan indicates that the fault section may be locked. Furthermore, the 1999 M_S5.4 Xiuyan earthquake and its aftershock sequence occurred on the Kangjialing fault and its ENE-trending conjugate fault, and the intersection of the two faults coincides with the source areas of the 1999 M_S5.4 and 2000 M_S5.1 Xiuyan earthquakes. Therefore, the Xiuyan earthquake sequence may be controlled by the Kangjialing fault and its conjugate fault. This study shows that the automatic earthquake catalog, obtained by deep learning methods and dense seismic array, can provide valuable information for fault structures and the seismogenic mechanisms of moderate-to-strong earthquakes.