Bridging source and risk: a combined PMF and unsupervised machine learning framework for heavy metal management in a transboundary himalayan river

Managing water quality in data-scarce transboundary basins is often challenging. This difficulty is due to an over-reliance on single-method source apportionment and a lack of structured, data-driven integration. This study introduces an integrated framework that combines receptor modeling with unsupervised machine learning. The framework supports source interpretation, uncovers complex pollution patterns, and identifies cost-effective monitoring proxies. The framework was applied using a comprehensive dataset of eight heavy metals (Cr, Ni, Cu, Zn, Pb, Hg, As, and Cd) and four physicochemical parameters (pH, EC, TDS, TH). Data were collected from 15 locations along the climate-sensitive Jhelum River across two seasons. Positive Matrix Factorization (PMF) revealed urbanization and industrial emissions (41.15%) as the primary year-round source. Unsupervised machine learning supports this with pattern recognition. Principal Component Analysis (PCA) performed separately for dry and wet seasons explained 86.7% and 84.3% of total variance, respectively, with strong loadings of Zn, Pb, Cd, Cu, Cr, and Hg (0.75–0.85) on PC1, identifying them as urban-industrial indicators. K-means clustering systematically divided the sites into three groups: urban-industrial hotspots, agricultural, and background. Correlation analysis showed that easily measured parameters, total hardness (TH) and electrical conductivity (EC), serve as useful screening indicators of specific metal contamination ( r > 0.35 for Zn and Ni). A health risk assessment, contextualized within these ML-derived patterns, indicated substantial carcinogenic hazards from As and Ni (CR > 10 −4 ) for children. Risks were highest in the urban-industrial cluster. The framework demonstrates a structured approach that could be adapted for moving from descriptive pollution accounting to predictive, cost-effective, and evidence-based management. It can be adapted for other data-sparse river basins, pending local validation. Integrated Framework for Heavy Metal Management in the Jhelum River. • A combined PMF and unsupervised ML framework for transboundary river systems. • Urban-industrial activities contributed the most to heavy metal pollution. • TH and EC correlated with key metal concentrations. • Clustering analysis helped prioritize higher-risk zones for management attention. • Framework enables cost-effective monitoring in data-limited Himalayan basins.