Optimized intelligent learning for groundwater quality prediction in diverse aquifers of arid and semi-arid regions of India

Khan, Imran, Nizam, Sarwar, Bamal, Apoorva, Sajib, Abdul Majed, Mahammad Diganta, Mir Talas, Shaida, Mohd Azfar, Ashekuzzaman, S. M., Nash, Stephen, Olbert, Agnieszka I., Uddin, Md Galal. (2025). Optimized intelligent learning for groundwater quality prediction in diverse aquifers of arid and semi-arid regions of India. Cleaner Engineering and Technology, 26, 100984. https://doi.org/10.1016/j.clet.2025.100984

Abstract

Ensuring access to safe, affordable drinking water while implementing sustainable management practices is vital for achieving the United Nations Sustainable Development Goals-2030. Accurate groundwater (GW) quality assessment plays a crucial role in enhancing water management strategies. This study evaluates GW resources across the diverse aquifer systems of arid and semi-arid regions of northwest India using the recently developed Root Mean Squared-Water Quality Index (RMS-WQI) model, optimized with machine learning (ML) techniques. A total of 772 GW samples from 36 districts of state Rajasthan were analyzed for 16 water quality (WQ) indicators/parameters, including pH, Electrical Conductivity (EC), Total Dissolved Solids (TDS), major cations (Ca2+, Mg2+, Na+, K+), anions (Cl−, CO32−, HCO3−, SO42−, NO3−, F−, PO43−), Alkalinity (ALK), and Total Hardness (TH). The results indicate slightly alkaline GW (average pH 7.9), with elevated concentrations of Na+, Cl−, SO42− and NO3− exceeding Bureau of Indian Standards (BIS). This study employs the eXtreme Gradient Boosting (XGB) algorithm, demonstrating strong predictive capabilities within the RMS-WQI model across diverse aquifers of Rajasthan. This marks the first application of RMS-WQI at a state-wide scale in India. Model performance assessment indicated groundwater quality ranging from ‘fair’ to ‘marginal’, generally meeting BIS standards, with high sensitivity and low uncertainty. Statistical metrics (Root Mean Square Error-RMSE, Mean Squared Error-MSE, Mean Absolute Error-MAE, and Percentage of Absolute Bias Error-PABE) validated the model's efficiency, with minimal error and high sensitivity. Optimization using “Optuna” further enhanced model performance, confirmed by Tukey's Honest Significant Difference (HSD) test. Sensitivity analysis demonstrated robust goodness-of-fit, while uncertainty analysis indicated minimal discrepancies, with overall uncertainty below 2 %. Spatial analysis revealed varying WQ across districts, ranging from marginal to poor, while efficiency metrics demonstrated the model's effectiveness in providing accurate assessments. The configured WQI model could substantially contribute to informing aquatic managers and strategic planners for sustainable water resource management and policy development aimed at enhancing GW quality.

Publisher

Elsevier

Rights

CC BY

cb

Collections

Publications