Predictive Machine Learning Framework for Ion-Tuned Low-Salinity Enhanced Oil Recovery in Tight Niger Delta Formations

Abstract Tight reservoirs in the Niger Delta hold significant remaining oil but remain challenging to produce due to low permeability, complex mineralogy, and strong capillary trapping. Conventional waterflooding performs poorly in these intervals, leaving substantial oil stranded. Low-salinity waterflooding (LSWF) offers a promising enhanced oil recovery (EOR) solution by altering wettability through controlled brine chemistry, yet its application in tight formations remains underexplored, and predictive tools for rapid screening are lacking. This study integrates physics-based reservoir simulation with machine learning to develop a predictive framework for ion-tuned LSWF in tight Niger Delta sandstones. A compositional reservoir model representing a typical tight formation was built in CMG-GEM, capturing multiphase flow, aqueous geochemistry, and ion exchange reactions. Two scenarios were simulated: conventional high-salinity waterflooding and ion-tuned LSWF with engineered concentrations of Ca2+, Mg2+, and SO42- based on regional produced water data. Results demonstrate that ion-tuned LSWF achieves 28.9% incremental oil recovery over conventional waterflooding, delaying water breakthrough by 32.7% and improving the oil-to-water ratio by 35.2%. Geochemical analysis confirms multi-ion exchange and double-layer expansion as dominant mechanisms: divalent cation concentrations decreased by 17–20% during flooding, while sulfate exhibited active surface adsorption. Pearson correlation revealed strong negative relationships between total ionic strength, Ca2+, Mg2+, and cumulative oil production (r = −0.96 to −0.99), and a strong positive correlation between SO4/Cl ratio and recovery (r = +0.98).Among machine learning models, Random Forest achieved the best predictive performance (test R2 = 0.89, RMSE = 36.7 bbl/day, MAPE = 8.2%), effectively capturing nonlinear ion-rate relationships with minimal overfitting. Feature importance analysis identified lagged production (45.2%), temporal features (32.2%), total ionic strength (8.2%), and chloride concentration (6.5%) as dominant predictors. Overall, this work provides a framework for rapid screening capability that can guide pilot design and full-field implementation of low-salinity EOR in the Niger Delta's most challenging tight formations, by transforming low-salinity EOR evaluation from a time-intensive experimental process into a fast, data-driven decision tool.