Abstract:
The Mubarak Block in the Lower Indus Basin of Pakistan presents significant challenges for hydrocarbon exploration due to its complex structural setting, lithological variability, and incomplete well-log data. Traditional seismic inversion methods often fail to resolve thin beds and lateral heterogeneity, leading to uncertainties in reservoir characterization. To overcome these limitations, this study integrates seismic interpretation, petrophysical analysis, model-based inversion, and machine learning (ML) to improve the prediction of reservoir properties within the Lower Goru Formation. Seismic interpretation of 2D lines established the structural and stratigraphic framework of the reservoir, revealing NW–SE dipping geometries, normal fault systems, and channel-like seismic facies. Petrophysical evaluation of wells Rehmat-1, Rehmat-2, and Saqib-1A provided estimates of porosity, shale volume, and water saturation, confirming that clean Lower Goru sands typically show porosity values of 12–20% and water saturation below 50%. These well insights served as the rock-physics foundation for inversion and ML workflows. Model-based seismic inversion generated acoustic and shear impedance volumes that reproduced broad lithological trends but were limited by band-limited seismic data, resulting in smoothed property distributions and loss of thin-bed resolution. To address these shortcomings, ML techniques were applied. The Extra Trees Regressor successfully predicted missing shear sonic logs (R² ≈ 0.90–0.95), enabling complete elastic input for inversion. The ML workflow simultaneously predicted multiple reservoir properties, including Vp, Vs, RHOB, PHIE, SWE, VSH, and Vp/Vs. The ML inversion results captured thin interbeds, channel geometries, and lateral heterogeneity with higher fidelity than deterministic methods. A NE– SW trending sand fairway was delineated, characterized by porosity up to 20%, low SWE (<50%), low shale volume (<15%), and Vp/Vs anomalies consistent with hydrocarbon-bearing intervals. These outputs align with well data and seismic facies, validating the workflow. This research demonstrates that integrating ML with seismic inversion significantly enhances resolution, reduces uncertainty, and provides geologically consistent reservoir property predictions. The approach improves the identification of sweet spots and hydrocarbon-prone zones, offering a robust framework for future exploration and development in the Mubarak Block and similar geologically complex settings.
