Integrated multi-scale heterogeneity analysis and explainable AI for improved reservoir rock typing: A case study of Reservoir-X, Southern Niger Delta Basin, Nigeria

The high cost of extensive coring in the Niger Delta Basin necessitates innovative, data-driven methods for accurate reservoir characterization. We hypothesized that a multi-scale analysis, the integration of global statistical averages, stratigraphic zonal measures, and pore-scale physics, would reveal significant flow complexities masked by standard metrics. Using wireline logs and core porosity-permeability data from Reservoir-X, Southern Niger Delta Basin, we calculated heterogeneity indices and hydraulic flow units. While global Lorenz (7.5%) and Dykstra-Parsons (13.2%) coefficients suggested a homogeneous reservoir, the Stratigraphic Modified Lorenz Plot (SMLP) provided critical depth-dependent resolution, uniquely exposing localized baffles and rhythmic cyclicity (2.2 m) that standard global applications failed to detect. To extrapolate these localized, core-derived hydraulic units to the continuous reservoir, we employed a supervised Random Forest model to predict rock types in uncored intervals, validating its performance against core-derived classes with 91% accuracy. To bridge the gap between black box algorithms and geological physics, we applied SHAP (SHapley Additive exPlanations) analysis, which identified bulk density as the dominant predictor, confirming that compaction is the primary physical control on reservoir quality. Furthermore, we integrated unsupervised K-means clustering, evaluated via the Elbow Method and sum of squared errors (SSE) reduction, to challenge the rigid supervised classification. We demonstrate that this combined workflow is necessary because the unsupervised analysis identified distinct transition zones (4–5 clusters) that the standard three-rock-type empirical models missed. We conclude that combining explainable AI with heterogeneity analysis provides a superior framework for identifying cryptic facies and optimizing reservoir models in complex deltaic systems.