Tail-dependent groundwater quality regimes in the Neogene arid multi-aquifer system of Saudi Arabia using copula-informed machine learning

Study region: The Neogene aquifer in the Al-Hassa Oasis, Eastern Saudi Arabia, is an arid multi-aquifer system characterized by intensive groundwater abstraction and growing water quality stress. Study focus: Conventional groundwater quality studies rely on correlation or direct machine learning (ML) approaches. The current methods do not separate marginal effects from dependence structure and rarely capture joint behaviour under extreme conditions. The present study develops a dependence-based framework that combines empirical copula transformation, tail dependence analysis, and ML for Water Pollution Index (WPI) prediction using 300 groundwater samples. New hydrological insights for the region: The copula transformation standardized all variables into a uniform probabilistic space. This was verified using the Kolmogorov–Smirnov (KS) test (KS = 0.003, p = 1.0), which supports reliable dependence analysis. The RBF-SVR model showed the best generalization performance (DC = 0.9616, RMSE = 0.0073, MAE = 0.0062, MAPE = 0.4111). Strong lower-tail dependence (λL ≥ 0.8 at q = 0.05) between WPI and chromium (Cr), manganese (Mn), nickel (Ni), and barium (Ba) suggests that baseline groundwater quality is mainly controlled by natural geochemical processes rather than salinity. On the other hand, upper-tail dependence appears only beyond threshold levels, indicating selective coupling under elevated pollution conditions. The findings point to distinct groundwater quality regimes and provide a basis for early detection of transitions from stable to degraded conditions.