A 3D seismotectonic model constrained by a deep learning-based earthquake catalog: Implications for fault activity and buried structure detection in the northern section of Xiannvshan Fault Zone

This study integrates two-phase dense array observations and the LOC-FLOW machine-learning-based workflow to construct a high-precision microseismic catalogue for the northern section of the Xiannvshan Fault Zone in the Three Gorges Reservoir area. The relocated catalogue reveals microseismicity organised into three seismically active zones—Zone I (NW of the Xiannvshan Fault), Zone II (western extension of the Tianyangping Fault), and Zone III (EW strip above the Tianyangping Fault)—and two depth layers: a shallow layer (0-6 km) and a deep layer (6-15 km). A steeply dipping linear seismic band in Zone II is interpreted as an inferred hidden fault of the western Tianyangping Fault system, and the shallow and deep seismogenic layers may be connected by fluid-permeable fracture networks along the fault zones. Comparison with the regional network catalogue during the same period demonstrates a detection gain of over an order of magnitude, with the dense array capturing abundant shallow microseismicity previously invisible to the permanent network. These observations support the hypothesis of a dual-mechanism model of reservoir-modulated seismicity. The results highlight that dense-array and machine-learning approaches can reveal previously undetected shallow seismogenic structures and provide a reference framework for seismic hazard assessment in the Three Gorges Reservoir area.