Seismic-scale fault characterization for geologic carbon storage using legacy data and machine learning

Fault characterization is crucial for evaluating containment integrity and supporting Area of Review (AoR) delineation under the U.S. Environmental Protection Agency Class VI framework for geologic carbon storage. We present an integrated workflow that combines prestack depth migration, seismic image enhancement, and pretrained deep-learning fault detection to provide seismic-scale structural constraints relevant to AoR delineation from legacy seismic data. We apply the approach to a 1998 three-dimensional seismic survey at the San Juan Basin CarbonSAFE Phase III site (New Mexico, USA), where CO 2 injection is planned in the Jurassic Entrada Formation at ∼2.5 km depth. Iterative velocity-model updating improves structural positioning, and anisotropic diffusion filtering enhances reflector continuity. A pretrained nested-residual U-Net model generates three-dimensional fault probability volumes without site-specific retraining. Estimated vertical seismic resolution at injection depth (λ/4 ≈ 20–40 m) constrains fault detectability. Detected seismic discontinuities are shallow, discontinuous, and spatially limited. No laterally extensive or vertically continuous faults are resolved within or adjacent to the injection interval, and no basement-rooted systems are identified within imaging limits. These results constrain seismic-scale fault continuity and vertical connectivity relevant to AoR delineation and containment-risk assessment. The study demonstrates how modern reprocessing and transferable machine learning can convert legacy seismic datasets into reproducible structural constraints that support Class VI AoR evaluation when integrated with geologic interpretation.