Abstract Fault interpretation in pre-salt carbonate reservoirs remains challenging due to complex structural geometries and illumination issues. Traditional workflows based on geometric attributes such as coherence and dip are often sensitive to parameter selection and data quality, limiting their ability to capture subtle or sub-seismic discontinuities. In this study, we present a deep learning-based workflow for 3D seismic fault segmentation tailored to broadband ocean-bottom node (OBN) data from the Búzios Field in Brazilian pre salt. A synthetic training dataset was generated using point-spread function (PSF) convolution to better reproduce the frequency content and imaging characteristics of pre-salt seismic data. A 3D U-Net architecture was trained for binary fault segmentation using balanced cross-entropy loss to address class imbalance. The model successfully identified major fault trends and small- to medium-displacement discontinuities. To further enhance prediction continuity and reduce artifacts, we implemented a preconditioning workflow in which the predicted fault probability volume guides an edge-preserving smoothing filter. This approach improved fault delineation while maintaining structural detail. Comparisons with conventional coherence-based methods indicate that the proposed workflow reduces artifacts and enhances the detection of subtle discontinuities. The results demonstrate that combining realistic synthetic data generation with deep learning and targeted seismic preconditioning provides a robust alternative to traditional interpretation workflows in structurally complex pre-salt environments.
Nilo et al. (Mon,) studied this question.
Synapse has enriched 5 closely related papers on similar clinical questions. Consider them for comparative context: