Development of predictive equations for the early identification of uranium contamination using basic hydrochemical...

Groundwater uranium (U) prevalence in Punjab potentially threatens human health, with increasing incidence of kidney ailments, and other issues. Different aspects of U contamination have been thoroughly investigated in the region; however, a simple and cost-effective approach for the early detection of U at the local scale is lacking. Consequently, we aim to develop reliable and efficient predictive equations using commonly measured water quality parameters. The Generalized Additive Model (GAM) and Logistic Regression (LR) were adopted to model continuous U levels and categorical U class (for U30: Class 1 (U > 30 µg/L) and Class 0 (U ≤ 30 µg/L), and similarly for U60: Class 1 (U > 60 µg/L) and Class 0 (U ≤ 60 µg/L)), respectively. The adjusted R² and deviance explained for multivariate GAM are 0.73 and 75.7%, respectively, with TDS-HCO₃¯, TDS-Na⁺, and HCO₃¯-Na⁺ as significant interacting pairs. Groundwater U increased with the increase of HCO₃¯ and Na⁺ above 400-600 mg/L and TDS at 1000-1500 mg/L. This shows that greater ionic strength, excess desorption from mineral surfaces, and uranyl-carbonate complexation facilitated U enrichment in the region. The stepwise LR performed exceptionally for both U30 and U60, achieving accuracies of 94% and 96%, respectively. The resulting predictive equations are $P_{U30}=1/1+{\text{e}}^{-[-32.10+6.72(log10\_TDS)+1.20(log10\_Na)+1.57(log10\_K)+1.19(log10\_Mg)+1.05(pH\left)\right]}$  and $P_{U60}=1/1+{\text{e}}^{-[-39.42+7.32(log10\_TDS)+1.40(log10\_Na)+1.27(log10\_NO3)+1.99(pH\left)\right]}$ . The equations were cross-validated using an independent dataset, achieving more than 80% accuracy, indicating their reliability. Thus, it can serve as an efficient tool for determining groundwater U contamination at unmonitored locations in Punjab, thus aiding in groundwater management and attaining SDG 6.