Machine learning exploration of the direct and indirect roles of Fe impregnation on Cr(VI) removal by engineered biochar

Data mining and knowledge discovery by machine learning (ML) have recently come into application in environmental remediation, especially the exploration for the multifactorial process such as hexavalent chromium [Cr(VI)] removal by iron-biochar composite (Fe-BC). The Cr(VI) removal capacity of Fe-BC was concurrently controlled by impregnated iron species (Fe⁰/Fe²⁺/Fe³⁺), carbon properties, and iron-carbon interactions, while the current lab-scale research could hardly untangle the overall relationships with the Cr(VI) removal experiments of one or several Fe-BCs under different research frameworks. Herein, we investigated the impacts of various microscopic material properties of Fe-BC on aqueous Cr(VI) removal by ML approach and highlighted the variations of biochar properties after iron impregnation. Our results suggested that the direct impacts of impregnated-iron contents on the Cr(VI) removal were limited, possibly related to undistinguished Fe species in the ML models, in which the roles of different iron species on Cr(VI) removal might be counteracted. Instead, the impacts of impregnated iron on the Cr(VI) removal were embodied indirectly by altering the biochar properties. Surface oxygen-containing functional groups (SOFGs) contents on biochar played a pivotal role in Cr(VI) removal according to the ML models. The condensed polyaromatic carbon matrices of BC and Fe-BC with a high content of non-polar carbon were also proved to be conducive to Cr(VI) removal. The ML models developed in this study consider surface functionalities information of BC and Fe-BC and offer a more accurate prediction for Cr(VI) removal, and the information mining behind models can act as a vital reference for the rational design of engineered biochar to remove aqueous Cr(VI).