Integrated Ensemble Weight of Evidence and Logistic Regression for Potential Groundwater Mapping: An Application to the Northern Piedmont of High Atlas Mountains (Morocco)

The objective of this study was to seek faster and cost-effective ways to assess groundwater potential in both bedrock and mixed bedrock and structurally deformed alluvial environments in an area located in the Piedmont of High Atlas Mountains (Morocco). Two data-mining techniques, namely, the Weights of Evidence (WofE) and Logistic Regression (LR) were implemented based on the spatial association between productive well/spring locations and effective factors governing the regional groundwater recharge. Prior to building these binary classification models, a variable screening and exploratory data analysis through Information Value (IV) and WofE were performed to quickly rate all the variables according to their predictive power. This step allows us to weed out 2 out of 10 variables that simply contain no additional information that will help predict GWP areas. Making use of the link that exists between WofE and LR, we then refitted the LR model using the WofE scale for recoding predictors. The predictive capability of each model was determined by the Receiver Operating Characteristic (ROC) curves and Area Under the Curve (AUC). The obtained AUC values were 0.80, 0.83, and 0.88 for LR, WofE, and (WofE- LR), respectively. This result indicates that the integrated LR-WofE model is a relatively good estimator of groundwater potential mapping in comparison with the single application of these models. The produced groundwater potential maps can serve for better planning and management of groundwater resources.