Aqueous Adsorption of Pharmaceutical Pollutants on Biochar: a Review on Physicochemical Characteristics, Classical Sorption Models, and Advancements in Machine Learning Techniques

Biochar, a carbon-rich substance obtained through the pyrolysis of various biomass sources like wood, agricultural residues, and municipal solid waste, exhibits exceptional adsorption properties, making it highly effective in removing a wide range of pollutants, including emerging pharmaceutical contaminants. This is due to its porous structure and large surface area. This review presents a comprehensive analysis of the physicochemical characteristics of biochar as a pharmaceutical adsorbent. The study includes evaluating traditional models used to describe the adsorption process and exploring the potential of advanced machine learning algorithms in predicting pesticide behavior on the biochar matrix. The literature assessment reveals the use of spectroscopic techniques to assess the interaction between pharmaceutical drugs and biochar. The conventional Langmuir, pseudo-second-order, and thermodynamic equations were identified as the most suitable models for explaining the nature of adsorption on biochar. However, these models have limitations as they fail to consider the simultaneous effects of adsorption conditions, the type of pharmaceutical drug, and the type of aqueous source. To address these limitations, advanced learning models such as artificial neural networks (ANN), extreme gradient boosting (XGBoost), and random forest (RF) have been employed. These models have been trained on extensive datasets and have demonstrated reliable predictions, even under conditions that extend beyond the experimental range.

Automated summary

This review analyzes the physicochemical characteristics of biochar as a pharmaceutical adsorbent and explores the use of traditional models and advanced machine learning algorithms to explain its adsorption behavior. The study finds that biochar exhibits excellent adsorption properties, but traditional models have limitations. Therefore, advanced learning models are employed to improve prediction capabilities.