A systematic review on machine learning-aided design of engineered biochar for soil and water contaminant removal

Le résumé fourni par la source

The design and application of engineered biochar is crucial for removing contaminants from soil and water,yet its development and commercialization still depend on time- and labor-intensive experimental methods. Machine learning (ML) offers a faster alternative, but despite its growing use in biochar research, no review systematically covers ML-driven design of engineered biochar for large-scale contaminant removal. This work fills that gap by analyzing ML’s role in optimizing biochar properties using pilot and industrial-scale datal. We examine key biochar characteristics, including physical (e.g., surface area, pore volume), chemical (e.g., ultimate/proximate analysis, aromatization), electrochemical (e.g., cation exchange capacity, electrical conductivity), and functional group properties, and their optimization for various contaminants. With special attention on three mechanistic dimensions, this review offers the first thorough study of ML applications for designing biochars based on pilot and industrial-scale data: ML forecasts micropore-mesopore synergies controlling diffusion-limited adsorption of heavy metals (Pb² + , Cd² + ); surface chemistry optimization - including oxygen functional group (-COOH, -OH); and electrochemical tuning - of redox-active sites for contaminant transformation. The paper emphasizes how ML models—such as Random Forest (RF) and Gradient Boosting Regression (GBR)—elucidate the nonlinear links between pyrolysis conditions (temperature, feedstock composition) and biochar performance. For adsorption, surface area and pore volume are distinctly important; in redox reactions for heavy metal removal, functional groups like C-O and C=O play vital roles. Unlike earlier studies mostly on the adsorption capacity of biochar, this work expands the scope to investigate how ML can customize biochar properties for optimal contaminant removal using interpretability tools like SHAP analysis. These instruments expose parameters including nitrogen-to-carbon (N/C) ratios and pyrolysis temperature in adsorption efficiency. The review also covers hybrid methods combining ML with molecular simulations (e.g., DFT) to link mechanistic knowledge with data-driven predictions. Emphasizing the need for multidisciplinary collaboration, the review finally shows future directions for ML-driven biochar design, guiding fieldwork by pointing out shortcomings of present techniques and opportunities for ML.

Ce résumé expose les affirmations des auteurs. BNTIC ne l’interprète pas comme une validation indépendante des résultats.