Comparative immobilization of labile Pb and Zn fractions in contaminated soil using jackfruit seed-, sugarcane bagasse-, and taro stem-derived biochars: a machine learning-assisted mechanistic elucidation

Lead (Pb) and zinc (Zn) persist in mining-affected soils due to their association with labile fractions that control mobility and potential bioavailability, necessitating fraction-resolved approaches to evaluate stabilization processes. Biochars derived from sugarcane bagasse, jackfruit seed, and taro stem were produced at 400 °C and applied to contaminated soil at rates of 3%, 5%, and 10% (w/w), followed by a 30 days incubation. Metal fractionation was assessed using the Tessier sequential extraction scheme, coupled with interpretable machine learning. Biochar amendment reduced the exchangeable fraction of both metals and promoted redistribution to less labile pools, with Pb exhibiting a more pronounced shift (up to 61% reduction) than Zn. The extent and direction of redistribution were strongly feedstock-dependent: taro stem biochar preferentially stabilized Pb, whereas jackfruit seed biochar exerted a greater influence on Zn partitioning, demonstrating distinct metal-specific stabilization pathways. Model interpretation using SHAP and partial dependence analysis revealed consistent, metal-specific controls on fraction redistribution, with soil pH, organic carbon, electrical conductivity, and amendment rate emerging as dominant predictors, thereby linking soil chemical conditions to stabilization behavior. Together, these findings indicate that metal stabilization is governed by metal-specific redistribution mechanisms rather than uniform immobilization pathways, providing a quantitative and mechanistically informed framework for optimizing biochar selection in contaminated soils.