Pollution removal efficiency enhancement by agricultural biomass additions in constructed wetlands: A framework integrating meta-analysis with explainable machine learning

Abstract

The agricultural biomass addition on constructed wetlands (CWs) is a sustainable strategy that integrates pollution control with waste valorization. However, their widespread applications remain constrained by variations in biomass types, pretreatments, and system designs. This study combined meta-analysis and explainable machine learning to assess how agricultural biomass addition influences Chemical Oxygen Demand (COD) and Total Nitrogen (TN) removal in CWs, particularly under low C/N conditions, using 272 and 1,283 independent observations for meta-analysis and machine learning, respectively. Results showed that bamboo biochar significantly enhances COD removal efficiency in CWs by an average of 54.8% (SMD = 2.50), while lotus leaf biochar improves TN removal efficiency by 32.5% (SMD = 1.20) compared with controls. Additionally, multiple machine learning models were tested and the results showed that the XGBoost model demonstrated the robustperformance in simulating TN removal (R 2  = 0.83), whereas the Random Forest model is effective for COD removal (R 2  = 0.76). SHAP analysis further indicatedthat increasing wetland volume can simultaneously enhance both COD and TN removal efficiencies. This study underscores the power of combining meta-analysis and explainable machine learning to optimize CW design and management, offering a robust framework for improving pollution removal in CWs.