Machine learning-based prediction of water quality index and pollution source apportionment in the Haraz river: A probabilistic risk assessment approach.

Abstract

The Haraz River faces serious environmental and ecological threats due to pollution, especially leachate generated from a nearby landfill. This study assesses water quality in the Haraz River by comparing samples collected upstream and downstream of landfill and mining sites along the river. The finding reveals increased pollution downstream of the landfill, with higher electrical conductivity (EC), total dissolved solids (TDS), total suspended solids (TSS), and nutrients, especially in the winter, indicating downstream pollutant transport. Leachate showed very low dissolved oxygen (DO) while upstream and downstream sites indicated relatively higher DO concentrations. The 88.89% of phosphate (PO43), 100% of turbidity and TSS, 58.33% of color, 77.78% of ammonium (NH4+), and 100% of bacterial indicators (total and fecal coliforms) exceeded the permissible limits set by WHO, EPA, BIS, and Iranian standards. Root Mean Square-Water Quality Index (RMS-WQI) was classified as good at 2.78% of the sampling sites, fair at 94.44%, and marginal at 2.78%. linear support vector regression (SVR-lin) demonstrated the best predictive performance, while sulfate (SO42-) showed the highest relative importance and the greatest influence on the model-derived RMS-WQI. Further, geochemical analysis confirmed that silicate weathering is the dominant factor controlling the Haraz River's water chemistry, with little impact from evaporation. Source apportionment using the Absolute Principal Component Score-Multiple Linear Regression (APCS-MLR) model identified geogenic hydrochemical processes as the dominant source (64.66%), followed by localized anthropogenic nutrient pollution from agricultural and domestic activities (35.34%). Children showed the highest vulnerability, with a 95th percentile hazard index (HI95th) of 1.81.