Machine Learning–Assisted Fault Characterization and Seal Assessment for Reservoir Compartmentalization in the BEE Field, Niger Delta

Abstract Fault displacement analysis in fault-controlled reservoir compartmentalization remains a major challenge affecting well placement, reservoir connectivity assessment, drilling optimization, and production performance. Quantitative seismic fault interpretation is often time-consuming, subjective, and limited in resolving subtle fault segments that influence compartment boundaries. This study presents an integrated, data-driven workflow that combines traditional fault-displacement analysis with Machine Learning (ML) to improve the prediction of fault-related reservoir compartmentalization and associated sealing behavior in the BEE Field, Niger Delta Basin. Fault throw displacement data were extracted from 3D seismic and well information across key stratigraphic horizons and analyzed using displacement–distance (T–x) and displacement–depth (T–z) profiles. These analyses constrained fault growth styles, linkage evolution, relay zone development, and displacement gradients relevant to reservoir connectivity and seal potential. ML techniques were then applied to enhance interpretation consistency and prediction capability. This approach was adopted because quantitative fault-displacement metrics provide physically meaningful constraints on fault growth and sealing processes, while ML leverages these metrics to reduce interpretation subjectivity and extend fault characterization into areas where conventional seismic interpretation alone is limited. The integrated results show a clear relationship between fault displacement and compartmentalization behavior. A field-specific displacement threshold of approximately 40 m is associated with increased shale smearing potential and reduced cross-fault communication, while lower-displacement faults indicate higher connectivity potential. Unsupervised clustering grouped faults based on displacement magnitude, displacement gradients, and linkage attributes, enabling objective classification of fault maturity and seal-related behavior. Supervised Random Forest models predicted fault displacement in poorly imaged areas, providing continuous displacement fields that support indirect evaluation of fault sealing capacity through displacement-controlled shale smearing and connectivity indicators. These results provide quantitative constraints for fault seal risk assessment and reservoir modeling. Overall, the workflow reduces interpretation bias, improves confidence in compartment boundary and fault seal prediction, and supports more informed development planning. The approach is applicable to other deltaic basins where fault complexity and data limitations impact subsurface decision-making.