How microplastics affect nitrogen removal in nature-based stormwater infrastructures: A machine learning and meta-analysis study.

Microplastics (MPs) pose a significant threat to ecosystem functions, yet their systematic impact on nitrogen removal in nature-based stormwater infrastructures (NBSIs, e.g., bioretention systems, constructed wetlands) remains poorly understood. This study integrates meta-analysis and machine learning to systematically elucidate how MP properties, system characteristics, and environmental conditions influence nitrogen removal performance in NBSIs. Data from 19 published studies, were used to train and evaluate five machine learning models: extreme gradient boosting (XGBoost), random forest (RF), light gradient boosting (LightGBM), multilayer perceptron (MLP), and Kolmogorov-Arnold network (KAN) models. Results show that MPs most significantly interfere with NH4+ -N removal, primarily influenced by particle size, particle concentration, and polymer type. In contrast, NO3--N removal is co-regulated by environmental conditions (pH, C/N ratio) and biotic components (e.g., Typha, Ophiopogon japonicus). Total nitrogen (TN) removal is predominantly controlled by C/N ratio and pH, with SHapley Additive exPlanations (SHAP) analysis showing their cumulative contribution exceeds 60%, indicating that environmental regulation exerts a stronger influence than MP-related variables. Mechanistically, MPs mainly impede the nitrification stage, with comparatively minor effects on denitrification. Among the models tested, XGBoost achieved the highest predictive accuracy (R2 > 0.86). These findings reveal stage‑specific mechanisms of MP interference and offer a theoretical basis for optimizing the design and operation of NBSIs.