Geophysical and machine learning investigation of saltwater intrusion into freshwater aquifer in some estuary environment in Niger delta

Geophysical and physiochemical approach, combined with machine learning (ML) prediction and feature classifiers, were employed in a research study to investigate saltwater intrusion in freshwater aquifers in some Estuary environment in Niger Delta, Nigeria. Five ML classifiers and regressor models were employed to classify, predict, and optimize 17 resistivity features based on vertical electrical soundings (VES) data ranging from 0.4 to 769.9Ωm to predict saline intrusion into aquifer layers. The ML established that the Gradient Boost classifier and Random Forest (RF) regress or yielded cross-validation accuracy of 96% and 98% at R2 ≤0.9642 and highlighted five significant predictors of saltwater intrusion in specified Estuary environment with F1 score ≤0.9300. The predictive performance of the selected ML confirmed the potential electrode was the most significant predictor of saltwater intrusion into freshwater aquifers, corresponding to an 85.3% normalized degree of importance. RF predicted saltwater intrusion in the freshwater aquifer based on optimum potential electrodes in the range of 0.5-9.24, layer depth of 0.5-4m, elevation (5-13), curve type (A), and resistivity (0.5-43.8Ωm)with a corresponding R2≥0.8457. The findings from the groundwater occurrence and depth (GOD) index classified the study area into low and moderate vulnerability classes, with values ranging from 0.168 to 0.420. The hydraulic resistance values at 2.877m-1 to 27.28m-1, determine the aquifer vulnerability index (AVI).Groundwater analysis indicated elevated levels of electrical conductivity, salinity, and total dissolved solids, exceeding WHO standards. The GOD index, AVI, and water quality index (WQI) from the coastal location were consistent with the ML prediction.