Machine-learning-based spatiotemporal reconstruction and risk assessment of 16 priority PAHs in mainland China's surface waters, 2000-2024.

Abstract

Polycyclic aromatic hydrocarbons (PAHs) are widespread aquatic contaminants, but their long-term national-scale dynamics remain poorly constrained by fragmented monitoring. We integrated measurements of 16 priority PAHs with multisource predictors to reconstruct annual surface-water concentrations across mainland China at 1 km resolution from 2000 to 2024. Four tree-based machine-learning models were evaluated for each compound, and model robustness was examined through random holdout testing, spatial block cross-validation, bootstrap resampling, and hotspot stability analysis. Reconstructed ΣPAHs concentrations were consistently higher in eastern and southern China, especially eastern coastal areas, the middle and lower Yangtze River, and southern coastal areas. Nationally, 58.07% of valid water pixels decreased significantly, whereas only 1.42% increased significantly. However, significant TEQ decreases occurred in far fewer pixels, indicating that declining total concentrations did not necessarily produce synchronous reductions in toxic risk. Low-molecular-weight PAHs dominated ΣPAHs. By linking compound-specific reconstruction with TEQ assessment, hotspot stability, and SHAP-based driver interpretation, this study provides a long-term data basis for understanding PAH evolution, toxicity-weighted risk, and priority control regions in mainland China's surface waters.