Seismic Anisotropy of the Alpine Fault: Application of Machine Learning to Distributed Acoustic Sensing Borehole Seismics

Seismic anisotropy in fault zones provides valuable insights into subsurface deformation and fracture networks. This study focuses on the Alpine Fault in New Zealand, aiming to invert for the seismic anisotropy of the damaged rock mass from transit times recorded from borehole distributed acoustic sensing (DAS). A key challenge is the labor-intensive process of picking numerous transit times required for inversion. To address this, a novel machine learning-based approach that uses directly common shot gathers as images is introduced to automate the picking of qP- and qS-wave arrival times from borehole shot gathers. The deep learning neural network is trained with synthetic seismic data, generated using a range of elastic properties, and manually picked transit times. Preliminary results reveal significant reductions in stiffness compared to intact rock, suggesting the influence of fracturing within the fault damage zone. This machine learning-based method has proven effective for borehole seismic analysis, as it processes spatially continuous seismic traces while reliably distinguishing seismic phases, indicating a substantial promise of deep learning method for advanced seismic processing.