Benchmarking water saturation models for the Mishrif formation using dean-stark data.

Area of Science:

• Petroleum Geoscience
• Reservoir Engineering
• Machine Learning Applications in Earth Science

Background:

• Accurate water saturation (Sw) estimation is critical for hydrocarbon reserve calculations and reservoir characterization.
• Shaly carbonate formations present unique challenges for traditional Sw estimation methods.
• A comprehensive comparison of diverse Sw models is needed for optimizing reservoir analysis.

Purpose of the Study:

• To systematically benchmark 83 water saturation (Sw) models for the Mishrif Formation in southern Iraq.
• To compare the performance of traditional Archie-based models against advanced data-driven machine learning approaches.
• To establish a robust model selection methodology for shaly carbonate reservoirs.

Main Methods:

• Benchmarking 64 Archie-based formulations (Indonesian, Dual Water, Waxman-Smits, Simandoux, Juhasz) and 19 data-driven models (Random Forest, LGBM, SOM).
• Integration of well-log data with 131 depth-matched core samples from three wells.
• Quantitative performance evaluation using MSE, CCC, R, and R2 metrics, with dual-ranking algorithms for robust selection.

Main Results:

• The Borai Dual-Water Linear model demonstrated superior performance among Archie-based models.
• The Random Forest model achieved the highest rank among data-driven approaches.
• Machine learning models consistently outperformed Archie-based models in this shaly carbonate reservoir, with acceptable uncertainty for volumetric applications (|error|≤0.20).

Conclusions:

• The proposed benchmarking workflow effectively reduces model selection ambiguity.
• Advanced machine learning models offer superior accuracy for water saturation estimation in shaly carbonate reservoirs.
• The methodology is transferable to similar geological settings globally, enhancing reservoir characterization practices.