AI fault prediction for ultra-deepwater OBN seismic data: advancing structural interpretation in carbonate reservoirs

Accurate interpretation of complex fault networks is crucial for ultra-deepwater hydrocarbon exploration, yet traditional towed-streamer (TS) seismic systems frequently fail to image subtle faults due to environmental noise and complex topographies. This study develops an AI-driven fault prediction framework specifically tailored for high-fidelity Ocean Bottom Node (OBN) seismic data. The proposed methodology uses structure-oriented filtering for data preprocessing and an enhanced High-Resolution Network (HR-Net) that adaptively fuses multi-scale features to maintain spatial resolution and suppress noise. A progressive post-processing workflow involving fault enhancement, skeletonization, and ant-tracking is then applied to automate 3D fault model generation. Application in the Santos Basin’s pre-salt carbonate reservoirs demonstrated that this approach performs better traditional TS methods, effectively resolving intricate Y-shaped fault geometries and providing reliable structural support for established carbonate depositional models in the region. This workflow reduces structural interpretation cycles from months to hours, improving geological accuracy and mitigating drilling risks in offshore exploration.