Linking water quality index structure to seasonal contaminant dynamics and predictive stability in a semi-arid aquifer in the F'kirina Plain of North-eastern Algeria

Groundwater quality in semi-arid aquifers is strongly controlled by inter-seasonal/interannual hydroclimatic dynamics, which regulate hydrochemical evolution and contaminant mobilization. However, the extent to which the mathematical architecture of water quality indices (WQIs) influences hydroclimatic responsiveness, predictive stability, and model robustness remains insufficiently understood. This study refined the previously developed PCA-based Water Quality Index (WQI_P) using a contribution-weighted component formulation and compared it with eight other WQIs, namely WA_WQI, CCME, EWQI, GWQI, WQI_D1, WQI_D2, WQI_D3, and WQI_D4, in the F'kirina Plain, northeastern Algeria. Thirteen physicochemical parameters were measured from 21 groundwater samples collected during the dry season (September 2024) and 21 samples during the wet season (March 2023) (n = 42). Principal Component Analysis (PCA), hybrid GLM–Pearson variable selection, and penalized regression (Lasso, Ridge, and Elastic Net) were applied to evaluate predictive performance. Predictor-exclusion sensitivity analysis and Monte Carlo cross-validation (100 repeated train-test splits) were performed to assess predictor dependence and predictive stability. The dry campaign exhibited greater mineralization than the wet campaign, reflecting contrasting inter-seasonal/interannual hydroclimatic dynamics. The refined WQI_P achieved the highest predictive performance (test RMSE = 3.42, MAE = 3.10, R² = 0.97), whereas the sensitivity analysis demonstrated marked differences in predictor dependence among WQI architectures, with WQI_D1 showing the highest performance after predictor exclusion (test RMSE = 0.75, MAE = 0.64, R² = 0.999). Overall, WQI architecture strongly influenced predictive performance, robustness, and responsiveness to inter-seasonal/interannual hydroclimatic dynamics, providing a framework for machine-learning-based groundwater quality assessment.