Abstract
Accurate estimation of water saturation (Sw) is essential for reservoir characterization and reliable calculation of hydrocarbons in place. This study presents the first systematic benchmarking a total of 83 water-saturation models for the Mishrif Formation in southern Iraq. This includes 64 Archie-based formulations employing Indonesian, Dual Water, Waxman-Smits, Simandoux and Juhasz models along with 19 data-driven models including combinations of Random Forest, Light Gradient-Boosting Machine (LGBM) and Self-Organizing Maps (SOM) clustering approaches. The analysis integrates conditioned well-log data with 131 depth-matched core samples from three wells (A-C). Dean-Stark water-saturation measurements were used as ground-truth calibration for quantitative benchmarking. Model performance was evaluated using four complementary metrics: mean squared error (MSE), concordance correlation coefficient (CCC), Pearson correlation coefficient (R) and coefficient of determination (R2). To resolve inconsistencies among individual metrics, two dual-ranking algorithms combining these measures were implemented, enabling robust model selection. Results indicate that among the Archie-based models, the Borai Dual-Water Linear model outperforms all others, while Random Forest model ranks highest among data-driven approaches. Overall, advanced machine learning models-based models consistently outperform Archie-based counterparts in the studied shaly carbonate reservoir. Probabilistic uncertainty analysis confirms acceptable reliability for volumetric applications (|error|≤ 0.20). The proposed benchmarking workflow reduces model-selection ambiguity and is methodologically transferable to similar shaly carbonate reservoirs worldwide.