Harnessing AI-driven sedimentary interpretation for high-resolutionsubsurface modeling in reservoirs

The depositional complexity of sedimentary reservoirs presents a significant challenge in hydrocarbon exploration due to their intricate pore systems, diagenetic alterations and band-limited seismic resolution in stratigraphic delineation. Conventional interpretation methods, while foundational, often fall short in resolving subtle subsurface variations that influence hydrocarbon accumulation and recovery. To resolve this, the current study integrates artificial intelligence (AI) with seismic inversion and petrophysical analysis to enhance lithofacies prediction and porosity modeling in a mature carbonate basin with proven yet underdeveloped hydrocarbon potential.

The research utilizes a comprehensive dataset comprising 3D seismic volumes, wireline well logs and core analyses from a development carbonate field. Seismic inversion methods, including sparse-spike and model-based inversion are employed to create high-fidelity acoustic impedance volumes. These inverted attributes are then cross-correlated with well-derived porosity and lithofacies data to serve as input for supervised machine learning algorithms, specifically Random Forest (RF) and Gradient Boosting Machines (GB). Feature engineering and recursive attribute selection are introduced to optimize model performance and reduce overfitting.

The resulting 3D property models carves out the stratigraphic complexities and reservoir sweet spots that were previously undetected using conventional methods. These include thin, high-porosity layers trapped within low-contrast seismic facies and fracture corridors indicative of secondary porosity growth. The integration of ML algorithms allows for extraction of subtle impedance contrasts, which corresponded well with petrophysical variations across the reservoir. Furthermore, the application of machine learning significantly reduced interpretation time, enhanced objectivity, and provided consistent results across large volumes of data.

By automating attribute extraction, feature optimization, and classification, the proposed method reduces reliance on subjective interpreter bias, particularly in regions with sparse well control. Additionally, the research explores the potential of combining ML-based analysis with physics-based modeling to improve geological clarity in reservoir models.

The study demonstrates that integrating machine learning with geophysical and petrophysical workflows can significantly elevate the resolution and reliability of subsurface interpretations in carbonate terrains. This approach represents a step-change in exploration geoscience, offering a pragmatic and forward-looking approach for maximizing hydrocarbon discovery and recovery in challenging reservoirs.