Groundwater quality prediction for drinking and irrigation uses in the Murcia region (Spain) by artificial neural networks

Abstract

This research proposes the use of machine learning tools to assess groundwater quality in the semiarid Mediterranean region of Murcia, Spain, with a focus on the risk of aquifer salinization. Two groundwater quality indices were defined: one for drinking water (DWQI) and another for irrigation purposes (IWQI), calculated using ten and fifteen parameters, respectively. The weights of the parameters such as pH, electrical conductivity (EC), major ion concentrations, as well as the Kelly ratio, KR; magnesium hardness, MH; potential salinity, PS; sodium absorption rate, SAR; and the percentage of soluble sodium, %Na in the calculation of these indices were determined through principal component analysis (PCA). The developed artificial neural network (ANN) models included a resilient backpropagation multilayer perceptron (RProp-MLP) and a probabilistic neural network with dynamic decay adjustment (PNN DDA), both implemented within a KNIME framework. Input variables were selected based on Spearman correlation analysis, PCA, and scientific criteria related to the risk of saline intrusion and irrigation water infiltration. The dataset consisted of 1962 groundwater samples collected from 159 sampling points between 2000 and 2023, covering 38 groundwater bodies with diverse hydrochemical characteristics. Both models demonstrated strong predictive performance, with the RProp-MLP model outperforming the PNN DDA across all evaluated metrics. The best results were obtained using RProp-MLP with seven-input variables (Ca²⁺, Cl^‒, Mg²⁺, Na⁺, NO₃^‒, SO₄^2‒ and EC), although satisfactory accuracy was also achieved using only five-input variables. This study highlights the effectiveness of ANN-based models for groundwater quality assessment and management, contributing to the sustainable use of water resources in semiarid regions.