Groundwater quality index prediction and aquifer failure risk analysis using metaheuristic-tuned artificial neural networks.

Quantifying the uncertainty associated with groundwater quality is essential for the sustainable management of aquifer systems, particularly in arid and semi-arid regions. This study aims to assess the variability of the groundwater quality index (GQI) in the Jiroft Plain aquifer, located in southeastern Iran, by incorporating the uncertainty of hydrochemical parameters through geographic information system (GIS) techniques and machine learning (ML) modeling. To this end, a hybrid framework combining feedforward neural networks (FFNN) with metaheuristic optimization algorithms was developed to construct an accurate GQI prediction model. The model was trained and validated using 14 years of groundwater quality data and evaluated through multiple statistical performance indicators. Subsequently, the probabilistic distribution of key hydrochemical variables was characterized, and 1.9 × 10⁶ synthetic samples were generated for each parameter using a Monte Carlo simulation. The optimized ML model was then applied to estimate GQI values for the synthetic dataset, enabling the computation of aquifer reliability and the development of spatial vulnerability maps. The results indicate that accounting for uncertainty substantially increases the estimated vulnerability of the aquifer and reduces its reliability to levels below the permissible threshold, highlighting a critical groundwater condition. Overall, the robustness of the proposed probabilistic framework demonstrates its potential as a reliable tool for assessing the true status of aquifers under uncertainty.