Microplastics alter soil phosphatase activity and phosphorus mobilization: a global meta-analysis integrated with machine learning

Agricultural soils are a major reservoir and conduit within the anthropogenic water cycle, receiving microplastics (MPs) through sewage-sludge/biosolid land application, reclaimed-water irrigation, plastic-mulch degradation and atmospheric deposition, and exporting dissolved nutrients to surface and ground waters via runoff and leaching. Phosphatase enzymes govern organic-phosphorus mineralization and therefore the size of the dissolved- phosphorus (P) pool that drives aquatic eutrophication, yet how MPs alter soil acid (ACP) and alkaline (ALP) phosphatase activity, and which conditions create the greatest soil-to-water transfer risk, remains unresolved. We synthesized 171 global studies (866 ACP and 688 ALP observations) in a meta-analysis integrated with an interpretable machine-learning ensemble that stacked XGBoost, LightGBM and Random Forest, with SHAP and accumulated local effects (ALE) interpretation. MPs significantly increased ACP activity by 9.89%, whereas the overall ALP effect was non-significant (+1.40%) but strongly context-dependent, ranging from significant stimulation (e.g., biodegradable MPs: +15.15%; aged MPs: +25.60%) to significant decrease under certain conditions (e.g., with nitrogen addition: −9.45%). Soil pH was the dominant control on ACP responses (mean |SHAP| = 0.088) and clay content on ALP responses (0.062), each ranking first across all five best-performing hyperparameter configurations, indicating divergent driver hierarchies for the two enzymes. ALP suppression was concentrated in clay-rich soils, whereas low-clay soils showed neutral-to-positive responses. By identifying where MP contamination most strongly perturbs phosphatase-mediated P mobilization, our findings connect a terrestrial contaminant to diffuse nutrient pollution of receiving waters and provide a transferable framework for contaminant–water-quality synthesis.