Multi-Dimensional Assessment of Index-Based Groundwater Vulnerability Models Under Different Weighting Schemes: A Case Study of the Upper Cheliff Plain, Algeria

Groundwater vulnerability maps are widely used tools for aquifer protection. This study aims to assess the robustness and uncertainty surrounding parameter weighting in index-based overlay models, as well as the reliability of metrics used for their validation. To this end, DRASTIC and DRASTIC-LU models were applied to the Upper Cheliff Plain (northwestern Algeria) under original expert-based, Shannon entropy, and Frequency Ratio weighting schemes, resulting in six vulnerability maps (DRASTIC, DRASTIC-LU, DRASTIC-EN, DRASTIC-LU-EN, DRASTIC-FR, and DRASTIC-LU-FR). Sensitivity analysis identified net recharge as a dominant parameter among the six models, whereas aquifer media showed no spatial discriminatory contribution, despite its high original weight, as indicated by Shannon entropy. Validation against nitrate concentrations (Pearson and Spearman correlations, and AUC) showed that entropy-weighted models achieved the strongest performance under direct validation (DRASTIC-EN: r = 0.393, ρ = 0.328, AUC = 0.734). Frequency Ratio models performed well under in-sample validation (r = 0.546 and 0.533 for DRASTIC-FR and DRASTIC-LU-FR, respectively) but decreased by 48 and 63% under leave-one-out cross-validation. Monte Carlo analysis resulted in low coefficients of variability under weight perturbation (<7%) for all six models. These findings highlight that combining validation, sensitivity, and uncertainty analyses is essential to reliable groundwater vulnerability assessment and sustainable aquifer management.