An Enhanced Focal Mechanism Catalog of Induced Earthquakes in Weiyuan, Sichuan, from Dense Array Data and a Multitask Deep Learning Model

Abstract Determining focal mechanisms of abundant small-magnitude (M < 3) earthquakes can better reveal subsurface fault structures and stress features, but it remains challenging due to insufficient records or inefficient methods. In the past decade, seismicity in the Weiyuan region of the southern Sichuan basin has increased dramatically following massive hydraulic fracturing activities. Here, we apply a multitask deep learning model, PhaseNet+, to local dense seismic records to enhance the focal mechanism catalog. 38,518 earthquakes are first detected and well located using predicted phase arrivals and the improved 1D velocity models. From predicted P polarities and measured S/P ratios, 20,740 focal mechanisms of varying qualities are then computed. Among 4399 high-quality focal mechanisms, about 92% are primarily associated with reverse faulting. It agrees well with moment tensor focal mechanisms from previous studies and the local tectonic settings, suggesting that the enhanced catalog can provide reliable faulting mechanisms. We observe clear spatial perturbations in their P-axis azimuths, mainly associated with variations in fault strikes. Nevertheless, changes in the stress field at short-length scales are also obvious, likely caused by intersected fault structures. Recent fluid injection operations and moderate earthquakes in Weiyuan may have further altered the subsurface stress state, demanding more detailed investigations.