Application of Water Quality Indices, Machine Learning Approaches, and GIS to Identify Groundwater Quality for Irrigation Purposes: A Case Study of Sahara Aquifer, Doucen Plain, Algeria

In order to evaluate and project the quality of groundwater utilized for irrigation in the Sahara aquifer in Algeria, this research employed irrigation water quality indices (IWQIs), artificial neural network (ANN) models, and Gradient Boosting Regression (GBR), alongside multivariate statistical analysis and a geographic information system (GIS), to assess and forecast the quality of groundwater used for irrigation in the Sahara aquifer in Algeria. Twenty-seven groundwater samples were examined using conventional analytical methods. The obtained physicochemical parameters for the collected groundwater samples showed that Ca2+ &gt: Mg2+ &gt: Na+ &gt: K+, and Cl− &gt: SO42− &gt: HCO3− &gt: NO3−, owing to the predominance of limestone, sandstone, and clay minerals under the effects of human activity, ion dissolution, rock weathering, and exchange processes, which indicate a Ca-Cl water type. For evaluating the quality of irrigation water, the IWQIs values such as irrigation water quality index (IWQI), sodium adsorption ratio (SAR), Kelly index (KI), sodium percentage (Na%), permeability index (PI), and magnesium hazard (MH) showed mean values of 47.17, 1.88, 0.25, 19.96, 41.18, and 27.87, respectively. For instance, the IWQI values revealed that 33% of samples were severely restricted for irrigation, while 67% of samples varied from moderate to high restriction for irrigation, indicating that crops that are moderately to highly hypersensitive to salt should be watered in soft soils without any compressed layers. Two-machine learning models were applied, i.e., the ANN and GBR for IWQI, and the ANN model, which surpassed the GBR model. The findings showed that ANN-2F had the highest correlation between IWQI and exceptional features, making it the most accurate prediction model. For example, this model has two qualities that are critical for the IWQI prediction. The outputs’ R2 values for the training and validation sets are 0.973 (RMSE = 2.492) and 0.958 (RMSE = 2.175), respectively. Finally, the application of physicochemical parameters and water quality indices supported by GIS methods, machine learning, and multivariate modeling is a useful and practical strategy for evaluating the quality and development of groundwater.