Drivers of microplastic accumulation in urban agricultural soils: A data-driven source apportionment of plastic mulch...

Microplastic (MP) accumulation in urban agricultural soils arises from complex sources, yet the relative contributions of agricultural practices versus diffuse urban emissions remain inadequately quantified. A watershed‑scale interpretable machine‑learning framework was applied to quantify source contributions across 45 sampling sites in urban‑agricultural fringe. The framework encompassed 6 algorithms, which were integrated through a stacking ensemble approach. Predictors included mulching time, principal components (PC) of land‑use composition, and soil properties. Model performance was evaluated through repeated cross‑validation, with the ensemble and Random Forest algorithms showing the highest prediction accuracy. MP abundance ranged from 24.7 to 431.3 items/kg, with a mean concentration of 144.7 items/kg. Mulching time was identified as the dominant predictor, showing nonlinear growth with a potential plateau after extended use. Urban land cover, represented by PC1, exerted a predominantly linear influence. Soil unit weight exhibited a sigmoid‑like response, indicating a retention‑release mechanism. Direct measurements confirmed the progressive degradation of mulch films with aging, and a supplementary meta‑analysis supported a positive temporal trend in MP accumulation. It is concluded that MP accumulation is governed by the interplay of a saturating agricultural point source, linear urban diffuse sources, and soil‑mediated retention. A source‑apportionment framework for mixed‑use landscapes is thus provided.