From mapping to modelling: the evolving multidimensional microplastic risks in China's farmlands.

Agricultural soils are major sinks for microplastic (MP) pollution, yet the cascading drivers that shape multidimensional MP risks remain poorly understood, limiting our ability to project future trajectories under evolving socioeconomic conditions. Herein, a national-scale survey combined with Laser Direct Infrared Spectroscopy (LDIR) revealed widespread MP contamination in China's farmland, ranging from 2.50 × 104 to 1.39 × 106 items/kg. A novel composite MultiMP risk index, integrating abundance, morphology, size, and polymer type, indicated 70.1% of the sites face moderate risk. Five interpretable machine learning (ML) models were applied in predicting the MultiMP, with LightGBM achieving superior performance (R2 = 0.845, MAE = 0.192, and RMSE = 0.155). Beyond the geochemical factors of longitude (16.5%) and PM2.5 (10.5%), key anthropogenic drivers-including agricultural film usage, GDP per capita and population density-collectively shaped MP risks, contributing 8.10-9.30% to feature importance. Pairwise interaction analysis further identified high-risk hotspots where elevated longitude (115-125°E) intersects with agricultural film usage exceeding 5000 tons. Finally, ensemble projections under five Shared Socioeconomic Pathways (SSPs) from 2021 to 2050 indicated strong socioeconomic dependence of future MP risks. The fossil-fueled development scenario (SSP5, 2.210) yields a risk approximately 4.14% higher than the sustainable development pathway (SSP1, 2.122). The East China region, which held the top MultiMP score in 2021, is projected to be surpassed by the North China region as the highest-risk region by 2050. This shift challenges static emissions-based risk paradigms, demonstrating that future MP threats will be dynamically governed by the interplay of socioeconomic pathways and geographic constraints, leading to a counter-intuitive re-ranking of regional risks by mid-century.