Spatiotemporal dynamics of PFAS ecological risk in major Chinese river networks: a data-driven assessment (2011-2024)

Per- and polyfluoroalkyl substances (PFAS) pose persistent and widespread risks to aquatic ecosystems, yet their long-term risk dynamics remain poorly quantified. Here, we reconstruct the spatiotemporal evolution of PFAS ecological risk in China's river networks from 2011 to 2024 by integrating a national historical monitoring database (n = 1,110 sites) with interpretable machine learning and statistical modeling. Ecological risk was quantified using a mixture-based risk index and predicted at 2 km resolution using an XGBoost classifier trained on 19 dynamic environmental and socioeconomic covariates. Model transferability was evaluated using an independent out-of-time field survey conducted in 2024. The model achieved stable predictive performance (AUC = 0.84 for internal testing; AUC = 0.86 for independent external validation), indicating its out-of-time generalization potential in a typical mixed-use watershed. High-risk regions were persistently concentrated in eastern China. Temporal analysis revealed a transient reduction in national high-risk area temporally coincident with the 2019 PFOS ban (-18.5% relative to the 2019 peak), followed by an observed rebound trajectory in 2023-2024, temporally associated with the increasing prevalence of short-chain alternatives. SHAP-SEM analysis suggests that natural hydrogeological conditions are associated with lower baseline vulnerability, whereas anthropogenic pressures are associated with elevated risk, potentially through the attenuation of soil retention capacity. These results provide a decadal-scale, policy-resolved assessment of PFAS ecological risk and provide data-driven insights into the limitations of substance-by-substance regulation.