Predictive modeling of biochar pH using pyrolysis conditions and biomass elemental composition through machine learning approaches

Accurate prediction of biochar pH prior to pyrolysis is critical for designing customized environmental materials, yet traditional black-box machine learning models fail to capture the complex, non-linear thermochemical mechanisms dictating alkalinity. This study aims to develop a highly accurate, mechanistically interpretable predictive framework for biochar pH by coupling ensemble machine learning algorithms with SHapley Additive exPlanations (SHAP). A comprehensive dataset of 248 diverse biomass samples was curated, utilizing initial elemental compositions and pyrolysis conditions (temperature and residence time) to train, optimize, and cross-validate seven predictive algorithms. The optimized Random Forest model significantly outperformed other architectures, demonstrating robust generalization and precision on unseen test data (R 2 of 0.703, MSE = 0.737, AARE% = 5.8%). Crucially, SHAP analysis moved beyond simple feature ranking to reveal specific directional impacts: elevated pyrolysis temperatures positively drive biochar pH by accelerating acidic functional group volatilization and concentrating basic ash minerals, whereas higher initial biomass carbon contents and prolonged residence times inversely suppress pH generation by promoting structural graphitization and reducing reactive basic sites. By translating statistical feature importance into actionable physical chemistry, this interpretable framework equips process engineers with a transparent, highly accurate tool to pre-determine and optimize biochar alkalinity without relying on exhaustive, energy-intensive experimental trials.