Deep Learning-Based Gravity Interpretation for Petroleum Reservoir Prediction and Prospectivity Zonation in Gongola Basin, Northeastern Nigeria
Authors: *Umar, B.A., Aleem, K.F., Abdulqadir, I.F. And Shuaibu, M.A.
DOI Info: http://doi.org/10.5281/zenodo.21047971
ABSTRACT
Residual gravity anomalies provide valuable information for delineating subsurface structures and evaluating petroleum prospectivity in sedimentary basins. This study presents a deep learning framework for gravity anomaly interpretation and petroleum reservoir prediction in the Gongola Basin, northeastern Nigeria, using satellite-derived gravity data from the GOCO06s global gravity field model. Regional gravity trends were removed through spectral and polynomial filtering, while gravity derivatives were computed to enhance structural discontinuities associated with faults, depocenters, structural highs, and stratigraphic traps. An attention-enhanced U-Net convolutional neural network was developed to simultaneously reconstruct residual gravity anomalies and classify petroleum-related structural features. The predicted petroleum prospectivity map identifies the central and southwestern sectors of the Gongola Basin as the most favorable exploration targets, consistent with known structural depocenters and gravity lows. Monte Carlo dropout uncertainty analysis further indicates high prediction confidence across structurally coherent regions while highlighting areas requiring additional geophysical constraints. The proposed framework demonstrates that integrating satellite gravity data with deep learning provides a rapid, scalable, and cost-effective approach for structural interpretation and petroleum prospectivity assessment in frontier sedimentary basins.
Affiliations: Department of Surveying and Geoinformatics, Modibbo Adama University Yola, P.M.B. 2076, Yola, Adamawa State, Nigeria.
Keywords: Residual Gravity Anomaly, Convolutional Neural Network, U-Net, Petroleum Exploration, Gongola Basin, Satellite Gravity Data
Published date: 2026/06/30
