Assessment of groundwater potential in terms of the availability and quality of the resource: a case study from Iraq

A semi-confined aquifer from Kirkuk Governorate, northern Iraq was taken as a case study to map groundwater potential in terms of both the availability and quality of the resource. In terms of quantity, five machine learning (ML) algorithms were used to model the relationship between locations of 1031 wells with specific-capacity data and nine influential groundwater occurrence factors. The algorithms used were linear discriminant analysis, classification and regression trees, linear vector quantization, random forest, and K-nearest neighbor. The groundwater occurrence factors used were elevation, slope, curvature, aspect, aquifer transmissivity, specific storage, soil, geology, and groundwater depth. Analysis of the worthiness of the factors used in the analysis by the information gain ratio indicated that five out of nine factors were worthy (average merit > 0): groundwater depth, elevation, transmissivity, specific storage, and soil. The remaining factors were non-worthy (average merit = 0) and thus they were removed from the analysis. The performance of the five ML algorithms was investigated using accuracy and kappa as evaluation metrics. Applying the models in the carte package of R software indicated that random forest was the best model. The probability values of this model were used for mapping quantitative groundwater potential after classification into three zones: poor, moderate, and excellent. Groundwater quality for drinking was modeled using the water quality index and the weights of the chemical constituents used (pH, TDS, Ca²⁺, Mg²⁺, Na⁺, \({\mathrm{SO}}_{4}^{2-}\), \({\mathrm{Cl}}^{-}\), and \({\mathrm{NO}}_{3}^{-}\)) were assigned using entropy information theory. A map of the groundwater quality index revealed five classes: < 50 (excellent), 50–100 (good), 100–150 (moderate), 150–200 (poor), and > 200 (extremely poor). Combining the groundwater quality index map with the groundwater potential map using summation operators revealed three zones of groundwater potential: poor, moderate, and excellent. Comparing this combined map with the quantitative groundwater potential map showed different patterns for the distribution of potential classes, which confirms that analysis of the groundwater potential should include groundwater quality as an important factor.