Accelerating seismic interpretation with machine learning: A case study from the Gato do Mato Field, Santos Basin, Brazil

Gato do Mato, an ultra-deepwater oil and gas field in the Santos Basin, is a key exploration target within Brazil’s prolific pre-salt province. However, the thick salt layer overlying the reservoirs complicates seismic imaging, posing significant challenges for conventional interpretation methods. Traditional workflows are time-consuming, laborintensive, and susceptible to interpreter bias, highlighting the need for more efficient and automated solutions. This study explores the application of machine learning (ML) to accelerate seismic interpretation in this structurally complex setting. Both pre-trained and user-trained deep learning models were tested, with the user-trained model proving more effective for fault and horizon detection on a 3D seismic dataset covering 176 sqkm. The methodology includes seismic data conditioning, ML-based predictions, and quality control to refine results. CNN-based predictions reduced interpretation time and improved fault and horizon continuity. The implementation of the Multi-Horizon tool further accelerates horizon extraction, streamlining the workflow. The findings confirm that ML-assisted interpretation enhances both efficiency and geological accuracy, reducing manual workload and accelerating project timelines. These advancements have substantial economic implications, optimizing resource evaluation and reservoir management. Future efforts will refine ML models, extend their application to other pre-salt fields, and incorporate additional geological constraints to enhance predictive reliability. This research underscores the transformative potential of ML in offshore hydrocarbon exploration, advocating for broader adoption of AI-driven workflows in the oil and gas industry.