Characterization and Mobility Modeling as a Means of Stimulation Treatment Design for the Complex Makhul Carbonate Reservoir

Abstract Exploiting the significant oil reserves in the Makhul reservoir has been consistently challenged by its complex geology, low matrix permeability, unconnected fracture networks, and the presence of kerogen-rich intervals. Historical stimulation attempts have repeatedly failed to deliver sustainable post-treatment production rates. The reservoir is a tight dual-porosity carbonate system interpreted as an anticline comprising three major compartments at depths approaching 14,000 ft, all exhibiting the persistent issues of poor permeability and limited fracture connectivity. This study aims to develop permeability models for the Makhul reservoir using conventional well-log data. The workflow began with (i) detailed rock typing and (ii) modeling permeability as a function of porosity, deep resistivity, compressional sonic travel time, and shale index. The dataset, 278 pairs of laboratory-measured porosity and permeability combined with triple-combo logs, exhibited significant scatter and repeated permeability values, necessitating log transformation, normalization, and thorough data diagnostics. Statistical analysis confirmed the absence of meaningful linear or nonlinear correlations between permeability and any individual log-derived variable. Classical nonlinear regression yielded a modest coefficient of determination (~0.2). Rock typing identified seven distinct lithofacies ranging from calcareous mudstones to coarse-grained limestones, including fractured crystalline and grain-supported dolopackstones. Reservoir quality shows a consistent upward trend, improving progressively from the calcareous mudstone facies to the grain-dominated dolopackstone lithofacies. Incorporating discrete rock types into nonlinear regression elevated the regression coefficient of determination to ~0.95, with obtained permeability predictions within less than one-third of an order of magnitude, underscoring the value of rock-type delineation. To further enhance prediction reliability, the study integrates advanced machine-learning tools with comprehensive petrophysical analysis to construct an AI-based permeability modeling framework capable of capturing the reservoir's complex nonlinear behavior. Decision-tree models developed in MATLAB, tested under different train–test splits and pruning levels, demonstrated that unpruned trees achieve perfect training accuracy but suffer from overfitting, whereas moderate pruning yields models that are more interpretable and generalize more effectively. Overall, the results demonstrate that AI-based decision-tree modeling, when integrated with FZI-guided rock typing, provides a practical, and an improved approach for predicting permeability in the Makhul complex carbonate reservoir.