Decoding structure-performance relationships in biochar-based phosphate adsorbents: A systematic review and machine learning approach

Phosphate pollution in water resources presents a critical environmental challenge that requires effective separation technologies. This review systematically analyzes biochar-based adsorbents for phosphate removal, integrating traditional literature analysis with machine learning to decode complex structure-performance relationships. A comprehensive evaluation of over 50 studies identified key synthesis parameters: pyrolysis temperatures below 600 °C and strategic metal modification significantly enhance phosphate separation efficiency. To bridge the gap between diverse experimental conditions and predict practical performance, an extreme gradient boosting (XGB) model was developed using 532 experimental data points encompassing various biochar types, modification methods, and operational conditions (pH 2–11, initial phosphate 5–500 mg L⁻¹, adsorbent dosage 0.1–10 g L⁻¹). The model achieved excellent prediction accuracy (R² test = 0.963) and revealed previously unrecognized patterns: while metal loading enhances capacity, the impact of metal loading plateaus is above certain thresholds, suggesting there are optimal ranges for modifying the material. Operational parameters that contribute 41.5 % to removal efficiency were identified using SHAP, addressing limitations of the conventional focus on material properties alone. Advanced characterization coupled with model interpretability demonstrated that metal-modified biochars exhibit hierarchical binding mechanisms, where surface complexation dominates at low concentrations, and precipitation becomes significant at higher loadings. This data-driven systematic evaluation provides quantitative guidelines for the rational design of biochar-based adsorbents and process optimization, marking a transition from empirical to predictive approaches in separation technology development.