Efficiently quantifying thermodynamic and kinetic effects of biochar pyrolysis parameters on antibiotic adsorption...

Antibiotic pollution in aquatic environments threatens ecological and public health. Biochar is a promising biomass-derived adsorbent, yet its performance is profoundly influenced by pyrolysis parameters. This study aimed to develop an interpretable causal machine learning (ICML) framework to quantify the net causal effects of pyrolysis parameters, thereby providing targeted guidance for biochar pyrolysis processes. The framework integrated ML prediction, SHAP interpretability, and causal inference to analyze thermodynamic and kinetic adsorption behavior, identify key pyrolysis-related features, and estimate average and heterogeneous treatment effects of selected pyrolysis parameters. The results indicate that this framework achieved high-precision predictions for adsorption capacity (q_m) and rate constant (k₂) (test R² = 0.928-0.972) and revealed the dominant roles of pyrolysis temperature on q_m (8% contribution) and pyrolysis time on k₂ (23% contribution). By controlling for confounders, the positive causal effects of pyrolysis temperature on q_m (ATE = 0.116) and pyrolysis time on k₂ (ATE = 0.031) were confirmed. Heterogeneity analysis showed that the temperature effect on q_m was strongest under short duration and low heating rate (CATE = 0.208), while excessively long durations diminished this effect. The time effect on k₂ peaked under synergistic high-temperature and slow-heating conditions (CATE = 0.036). Based on these causal insights, this study proposes differentiated pyrolysis strategies targeting high capacity (> 600°C, 1-3 h) versus rapid removal (600-800°C, > 3h). This work extends biochar adsorption modeling from prediction and interpretation to causal effect estimation, providing a generalizable framework for the rational production and environmental application of high-performance biochar adsorbents.