Cross-media dynamics and prioritized risks of PFAS in textile-impacted environments: using geospatial machine learning

Per- and polyfluoroalkyl substances (PFAS) from textile receiving environment pose multifaceted environmental risks, yet their cross-media dynamics remain poorly understood due to limitations in traditional correlation-based models. This study performed geospatially-informed machine learning (ML) framework to elucidate PFAS interactions between soils and tree barks across 48 sites in five major textile hubs in China. Target analysis of 31 PFAS revealed higher concentrations in barks (49.9-2004.3 ng/g) than in soils (1.14-73.3 ng/g). Ultra-short-chain and emerging compounds (e.g., TFA, PFPrA, 8:2 FTS) dominated the profiles, indicating the PFAS transition amidst evolving regulations. Among 5 ML algorithms, LightGBM achieved the superior predictive performance of ∑PFAS in both matrices (R² = 0.962-0.974). GeoSHAP analysis identified barks PFAS as strong predictors of soil contamination (18.0 % contribution), underscoring cross-compartment influence. Emission from textile and clothing industry yield contributions of 8.95 % and 9.92 % to soils and barks PFAS burdens. To prioritize environmental risks, a novel multi-criteria risk framework was developed, integrating ToxPi scoring, Risk Index and Environmental Hazard Priority Index (EHPi). 67.7 % PFAS were fell within low-to-medium risk category, with TFA and 8:2 FTS ranked highest. This study enhanced PFAS assessment by integrating ML model and chemical-specific risk ranking, enabling strategy for pollution mitigation in textile-impacted environments.