Estimating Geomechanical Characteristics for Optimal Reservoir Characterization in Niger Delta Deepwater Fields Using Machine Learning Integration: A Case Study of the Agbami Field, Nigeria

This study presents an integrated geomechanical and machine‑learning workflow to characterize the deepwater Agbami Field reservoir (OML 127/128) in the central Niger Delta, Nigeria. Wireline logs (gamma ray, density, neutron, sonic) and a 3D seismic volume were used to derive key elastic moduli—Poisson’s ratio (ν), Young’s modulus (E), bulk modulus (K), shear modulus (G)—and unconfined compressive strength (UCS) via calibrated empirical correlations. Core measurements from two wells calibrated log‑derived estimates, reducing bias to below 4 %. Supervised ML models (Random Forest and Artificial Neural Network) were trained on combined log and seismic‑attribute datasets, achieving high predictive accuracy (RF R² ≥ 0.91; RMSE ≤ 0.85 GPa for E; R² ≥ 0.92; RMSE ≤ 8 MPa for UCS). Trained models were applied across the 3D seismic grid and upscaled via Sequential Gaussian Simulation to honor well control and spatial continuity. Results reveal that poorly consolidated turbidite sands exhibit lower ν (0.26), E (8.3 GPa), K (13.0 GPa), G (3.25 GPa), and UCS (33 MPa) with higher porosity (0.22), whereas shaley seals display higher ν (0.34), E (17.0 GPa), K (19.5 GPa), G (7.5 GPa), and UCS (60 MPa) with reduced porosity (0.08). Voxel‑wise uncertainty mapping (50 realizations) identifies elevated variance (σ(ν) up to 0.06) near major faults and inter‑well gaps, guiding targeted data acquisition and conservative well designs. The integrated approach delivers a robust 3D mechanical earth model, informing safer drilling operations, optimized completion strategies, and improved reservoir management under deepwater conditions.