Polyparameter linear free energy relationship for wood char–water sorption coefficients of organic sorbates

Abstract

Black carbons (BCs), including soots, chars, activated carbons, and engineered nanocarbons, have different surface properties, but we do not know to what extent these affect their sorbent properties. To evaluate this for an environmentally ubiquitous form of BC, biomass char, we probed the surface of a well-studied wood char using 14 sorbates exhibiting diverse functional groups and then fit the data with a polyparameter linear free energy relationship (ppLFER) to assess the importance of the various possible sorbatechar surface interactions. Sorption from water to water-wet char evolved with the sorbate's degree of surface saturation and depended on only a few sorbate parameters: S reflects polarity, and B represents the electron-donation basicity. As generally observed for activated carbon, the sorbate's size encouraged sorption from water to the char, while its electron donation/proton acceptance discouraged sorption from water. However, the magnitude and saturation dependence differed significantly from what has been seen for activated carbons, presumably reflecting the unique surface chemistries of these two BC materials and suggesting BC-specific sorption coefficients will yield more accurate assessments of contaminant mobility and bioavailability and evaluation of a site's response to remediation.

Predicting Aqueous Adsorption of Organic Compounds onto Biochars, Carbon Nanotubes, Granular Activated Carbons, and Resins with Machine Learning

“…As discussed earlier, the biochar and GAC/CNT differ in that biochar generally contains a large amount of organic/inorganic impurities, which complicates the interactions between the biochars and the adsorbed chemicals. Among the descriptors, the E , S , B , V , and BET are the dominant ones for biochars, agreeing with the reported MLR models in which the S , B , and V are the most important (BET and V t are not included in the reported MLR models). ,, The coefficient for E is not recognized as significant; this may be because the adsorption data are only available for 1 biochar and 14 chemicals . The dependency on E may emerge if the data set is to be expanded in both chemicals and adsorbents.…”

“…However, the calculations showed that their contributions varied within a broad range (Figure S16), so the adsorbent properties cannot be neglected in the adsorption prediction. This can explain part of the higher deviations in commonly used MLRs as the absence of adsorbent properties has failed to capture the differences in the adsorption among the adsorbents, even if they belong to the same class. ,, …”

“…Among the descriptors, the E, S, B, V, and BET are the dominant ones for biochars, agreeing with the reported MLR models in which the S, B, and V are the most important (BET and V t are not included in the reported MLR models). 11,68,81 The coefficient for E is not recognized as significant; this may be because the adsorption data are only available for 1 biochar and 14 chemicals. 68 The dependency on E may emerge if the data set is to be expanded in both chemicals and adsorbents.…”

“…11,68,81 The coefficient for E is not recognized as significant; this may be because the adsorption data are only available for 1 biochar and 14 chemicals. 68 The dependency on E may emerge if the data set is to be expanded in both chemicals and adsorbents. The agreement between this and previous work indicates that the NN-LFER models are both robust and chemically meaningful for adsorption prediction.…”

“…The pp-LFERs approach, however, generally needs to build one MLR model per C e , and the predictions based on the established MLR models are only limited to the concentration levels involved in the modeling, which is not helpful when predictions under different equilibrium concentrations are needed. To expand the prediction ability to multiple concentration levels, a few studies have attempted to unify several MLR models into one equation by treating the fitting parameters ( e , s , a , b , v , and c ) in eq as concentration-dependent. , Unfortunately, the concentration levels are still limited by the available experimental data (typically covering several concentration levels such as C e = 0.001 × S w or 0.01 × S w ) and are not able to cover wide concentration ranges in real applications. This is because some chemicals are much less soluble than others such that we cannot obtain their experimental log K d under high equilibrium concentrations; as a result, developing pp-LFERs at multiple concentration levels will greatly reduce the number of chemicals (and hence the amount of data available) that can be included in the modeling process, which further limits the applicability and accuracy of the MLR models.…”