Appraising the accuracy of GIS-based bivariate statistical model for groundwater potential mapping in South Africa

<h2>Abstract</h2> Groundwater is a reliable water source for human needs worldwide, especially in semi-arid and arid regions. Therefore, utilizing innovative methods is vital to ensuring the sustainable management of groundwater resources. In this study, the accuracy of the relative frequency ratio (RF) model with predictor rate (PR) in mapping groundwater potential was assessed in Amahlathi Municipality, South Africa. PR was used to measure the prediction capability of the groundwater controlling parameters while generating a groundwater potential zone (GWPZ) map. The bivariate statistical model (RF model) was employed to analyze the correlation between the borehole locations in the region and nine groundwater controlling factors (curvature, geology, aspect, land use/land cover (LULC), topographic wetness index (TWI), slope, lineament density, rainfall, and drainage density) in ArcGIS software. In total, 315 borehole points were chosen randomly to provide a training dataset (70 %) and a model validation dataset (30 %). The area under the receiver operating characteristic curve (AUC) was utilized to check the accuracy of the RF model. The final GWPZ map indicates five classes of GWPZ: very low, low, moderate, high, and very high, occupying 19.92 %, 52.99 %, 24.65 %, 1.39 %, and 1.05 %, respectively, of the area. The validation result shows that the RF model performs satisfactorily in prediction (AUC = 75 %). The outcomes of this study show a successful mapping of groundwater potentiality by integrating the RF model and PR. These results are reliable and useful for the sustainable management of groundwater resources in Amahlathi Municipality.