Decoding microplastic pollution in China freshwater: interpretable machine learning insights into...

Microplastics (MPs) pollution in freshwater ecosystems, combined with the leaching of toxic additives such as phthalate esters (PAEs), poses a pervasive and escalating threat to aquatic environmental health. As the top plastic producer/consumer, China exhibits widespread MP pollution, yet national-scale, composition-specific analyses of MP distribution and the associated PAE leaching risks remain limited. This study established a comprehensive dataset from 846 sampling sites (523 rivers, 323 lakes) across China (2014-2024), with MP sizes unified to 20-5000 μm. An interpretable multi-output multilayer perceptron (MLP) model was developed to simultaneously map spatial distribution of five dominant MPs (polyethylene [PE], polypropylene [PP], polyethylene terephthalate [PET], polyvinyl chloride [PVC], and polystyrene [PS]). Although the MLP exhibited a slightly lower average R² (0.61) relative to random forest (0.69) and XGBoost (0.68), its shared‑parameter architecture retains inherent co‑occurrence patterns across MP compositions. Independent validation confirmed model robustness and generalizability. Shapley additive explanation analysis highlighted key associated factors: precipitation was closely linked to PVC, and PET in rivers, while, temperature was correlated with PE and PP in lakes. National risk assessment of composition-specific PAE leaching identified di(2-ethylhexyl) phthalate (DEHP) as the primary concern. Under average leaching scenarios, low ecological risks were observed in rivers of Beijing, Tianjin, Taiwan and southeastern coastal regions as well as southern lakes; under maximum leaching scenarios, risks escalated to medium levels in focal river regions. This study provides the national-scale quantification of composition-specific MP distribution and PAE leaching risks, offering a transferable framework for targeted pollution management and policy formulation.