Particle swarm and grey wolf optimization: enhancing groundwater quality models through artificial neural networks

This study investigates the application of Artificial Neural Networks (ANN) supplemented with optimization algorithms for modeling and mapping groundwater quality in an extensive unconfined aquifer in Northern Iran, a task traditionally performed through labor-intensive and costly water sampling and lab analysis. A comprehensive collection of groundwater samples from monitoring wells scattered across the region facilitated the calculation of the Groundwater Quality Index (GWQI) for each well. These GWQI readings were subsequently categorized into four distinct quality classes very poor, poor, good, and excellent. Key variables impacting groundwater quality were identified, including proximity to industrial and residential areas, population density, aquifer transmissivity, precipitation, evaporation, geology, and elevation. These factors were compiled and processed within a GIS environment. To establish a relationship between the GWQI and these determinants, an ANN model was employed. This was enhanced by the application of two optimization algorithms, Particle Swarm Optimization (PSO) and Grey Wolf Optimizer (GWO), to determine the optimal weight and structure of the ANN model. The study's results indicated that the ANN-PSO model (overall accuracy = 0.88) surpassed both the standard ANN (overall accuracy = 0.71) and ANN-GWO (overall accuracy = 0.83) models in accuracy. The region with the best and poorest groundwater quality was identified in the west and the northern section of the study area respectively. The feature analysis identified precipitation and population as the critical factors influencing groundwater quality in the region.