Unraveling the Context-Dependent Effects of Microplastics on Soil Enzymatic Activities via Interpretable Machine Learning
Year: 2026
Type: article
Abstract: Microplastics (MPs) are increasingly accumulating in agricultural soils; however, their variable impacts on soil enzymatic activity hinder consistent ecological risk assessments. To resolve these discrepancies, a comprehensive dataset collected from published studies was analyzed using an interpretable machine learning framework. Among evaluated algorithms, an eXtreme Gradient Boosting (XGBoost) regressor coupled with a Random Forest (RF) classifier successfully captured non-linear interactions between MP characteristics and enzymatic responses. Feature importance identified particle size as the primary determinant governing the directional response of soil enzymes. Partial dependence plots (PDP) further resolved specific functional ranges, revealing a critical “size-sensitive window” of 200-600 μm where inhibitory effects were most pronounced. Dosage exhibited a marginal saturation effect, with the inhibitory response leveling off beyond 100 g/kg. Exposure time exhibited a biphasic pattern, with an initial toxicological shock within 0-105 days followed by functional stabilization, while PE and PVC were identified as the most disruptive polymers. Furthermore, Shapley additive explanations (SHAP) quantified individual feature contributions, reinforcing the robustness of these identified regularities. The overall importance ranking (size > dosage > type > exposure time) provides new quantitative insights into soil enzyme responses, supporting a critical shift toward size-fractionated risk assessments in MP-contaminated agricultural environments.
Source: Biology
Institution: Hainan University
Language: English
Subfield: Pollution
Field: Environmental Science
Domain: Physical Sciences
SDG: Life on land
Open Access status: gold
Estimated APC paid: $3,258
Award: G4507208040
OpenAlex ID: W7222972908