Integrating explainable AI with machine learning for water quality assessment: Unveiling physical drivers and spatial dynamics

Sustainability of biodiverse ecosystem is highly dependent on fresh water availability and as rivers are main source of fresh water accessing their water quality and its proper management is an essential component. This study undertakes evaluation of water quality across 15 major Indian rivers from Himalayan and peninsular region having 403 water quality monitoring stations. Physicochemical and microbiological indicators are adopted by national assessment framework to analyse and capture the spatial disparities in river health across diverse hydro climate and human-influenced gradients like agriculture usage, domestic and livestock waste and industrial pollution. Out of 403 monitoring stations 78.7% of monitoring stations exhibit excellent water quality, 20.3% show good quality, and notably 1.0% are classified as poor, suggesting urban-industrial pollution hotspots rather than basin-wise degradation. The finding indicates that generalized management is insufficient and specific hotspots needs precise policies to avoid further deterioration. All 403 monitoring stations collect 7 parameters which are evaluated to form Water Quality Index which labels each monitoring stations in 5 different categories. Unsupervised clustering technique is then applied to segment river basins into three distinct regions, near-pristine river stretches, agricultural influenced, and severely impaired segments affected by microbial contamination and oxygen depletion due to urbanization. Interpretable AI methods are applied to strengthen the claims and identify the parameters which are responsible for the result. The assessment of water quality of 15 major rivers shows the effectiveness of combining WQI based evaluation with interpretable analytical methods can enhance evidence-based strategies for effective management of riverine ecosystem.