Massive High-Fidelity Focal Mechanisms Reveal Detailed Structure of Re-Activated Faults During Hydraulic Fracturing in Western Canada

Jiachen Hu, Yunfeng Chen, Hongyu Yu, Fangxue Zhang, Xing Li, Yangkang Chen

Microseismic focal mechanism solutions (FMSs) are essential for understanding reservoir stress changes and rock fracturing during hydraulic fracturing. While machine learning has shown strong performance in seismic data processing tasks, including phase picking and magnitude estimation, as well as identifying P-wave first-motion polarity for moderate to large earthquakes to invert FMSs, its application to microseismic events remains limited. This limitation arises from the distinct characteristics of microseismicity, such as lower signal-to-noise ratios (SNR) and different rupture mechanisms, which challenge the effectiveness of existing polarity pickers. At the same time, the increasing deployment of dense seismic arrays has generated vast amounts of data, creating both the need and opportunity to develop AI models specifically tailored to microseismic events. In response to the challenges of determining the P-wave first-motion polarity for microseismic events, we propose Micro-EQpolarity, a fine-tuned model based on the EQpolarity framework. The model combines convolutional blocks for feature extraction, transformer blocks for feature enhancement, and an MLP network for classification. Utilizing transfer learning, the model is pre-trained on the Southern California Seismic Network (SCSN) dataset and fine-tuned with 19,724 manually selected waveforms from the Tony Creek Dual Microseismic Experiment (ToC2ME) dataset, achieving an accuracy of 99.20%. Applied to seismic data from Western Canada, Micro-EQpolarity successfully inverted 2,519 high-quality focal mechanism solutions, creating a comprehensive catalog that extends analysis to events with magnitudes as low as -1.4. The model identified four distinct FMS types, revealing fine-scale fault structures and detailed patterns of fault reactivation. These findings provide new insights into fault reactivation mechanisms in the NS-Fault cluster and fluid diffusion processes in the NE-Fault cluster. In the NS-Fault cluster, our analysis reveals two possible reactivation mechanisms: higher friction coefficients and enhanced cohesion, with fault reactivation driven by the combined effects of Coulomb static stress and pore-fluid pressure. In the NE-Fault cluster, two-stage hydraulic fracturing facilitated fluid propagation, initially reaching the southwestern part of Fault 1 before spreading to Faults 2-6, with Fault 3 acting as a "transfer station" directing fluid diffusion both eastward and westward through low-dip fault conduits.