Automated interpretation of deep learning-based water quality assessment system for enhanced environmental management decisions

Water quality assessment is a critical issue in the Aseer region of Saudi Arabia, where environmental and anthropogenic factors pose a major challenge to both drinking water and irrigation systems. The aim of this study was to carry out a detailed assessment of the water resources in the region, focussing on the most important aspects affecting water quality. The main objectives were to calculate various water quality indices for drinking and irrigation purposes, to develop an automated system using convolutional neural networks (CNN) to predict these indices and to increase the transparency of these models using explainable artificial intelligence (XAI) methods. Methodologically, the study used CNN algorithms optimised by Bayesian techniques for the prediction of eight water quality indices, coupled with SHAPley Additive exPlanations (SHAP) analysis under XAI to interpret the complex decision-making processes of these models. This dual approach enabled a comprehensive and insightful assessment of water quality. Using a robust dataset from the Aseer region, eight water quality indices were calculated, revealing significant variations and highlighting areas of concern. In this study, the entropy weight-based DWQI averaged 77.90 with a high standard deviation (std) of 39.08, reflecting considerable variability. The automated CNN models demonstrated robust performance in predicting water quality indices, with high accuracy (R2 = 0.959 in training and 0.945 in testing) for sodium percentage (Na%). However, the Magnesium Hazard (MH) index showed lower accuracy, suggesting possible overfitting and the need for further optimisation. SHAP analysis highlighted chloride, sulphate, and total dissolved solids as key contributors to the WQI, while sodium and calcium were significant for the sodium adsorption ratio. These insights enhance understanding of parameter influence on water quality assessments. This study introduces an advanced computational approach integrating CNN and XAI techniques, improving water quality evaluation and supporting informed environmental management in the Aseer region.