Machine learning and computational chemistry to improve biochar fertilizers: a review

Traditional fertilizers are highly inefficient, with a major loss of nutrients and associated pollution. Alternatively, biochar loaded with phosphorous is a sustainable fertilizer that improves soil structure, stores carbon in soils, and provides plant nutrients in the long run, yet most biochars are not optimal because mechanisms ruling biochar properties are poorly known. This issue can be solved by recent developments in machine learning and computational chemistry. Here we review phosphorus-loaded biochar with emphasis on computational chemistry, machine learning, organic acids, drawbacks of classical fertilizers, biochar production, phosphorus loading, and mechanisms of phosphorous release. Modeling techniques allow for deciphering the influence of individual variables on biochar, employing various supervised learning models tailored to different biochar types. Computational chemistry provides knowledge on factors that control phosphorus binding, e.g., the type of phosphorus compound, soil constituents, mineral surfaces, binding motifs, water, solution pH, and redox potential. Phosphorus release from biochar is controlled by coexisting anions, pH, adsorbent dosage, initial phosphorus concentration, and temperature. Pyrolysis temperatures below 600 °C enhance functional group retention, while temperatures below 450 °C increase plant-available phosphorus. Lower pH values promote phosphorus release, while higher pH values hinder it. Physical modifications, such as increasing surface area and pore volume, can maximize the adsorption capacity of phosphorus-loaded biochar. Furthermore, the type of organic acid affects phosphorus release, with low molecular weight organic acids being advantageous for soil utilization. Lastly, biochar-based fertilizers release nutrients 2–4 times slower than conventional fertilizers.