Sorption of organic compounds to humic and fulvic acids: development of a unifying model covering sorbate and sorbent variability
Sorption in natural organic matter (NOM) is a key process in determining the transport as well as the bioavailability of organic pollutants in the environment. A dataset containing more than 1000 NOM/air partition coefficients of nonionic organic chemicals measured in 10 different NOM from terrestrial and aquatic origins at different temperatures and different relative humidities is presented in this work. The conclusions that can be drawn from this extended sorption dataset provide a deeper understanding of the mechanisms involved in the sorption of organic compounds in NOM. In addition, this dataset allows the evaluation as well as the development of predictive models for NOM sorption. Relative humidity had a rather small influence (less than a factor of three) on the experimental partition coefficients. However, these results provided interesting mechanistical insights into the sorption process in dry NOM compared to completely hydrated NOM. Polar compounds generally sorbed more strongly than nonpolar compounds due to H-bonds (electron donor/acceptor interactions) with the NOM. No glass transitions in the range of 5 to 75 °C that would be relevant in respect to the sorption properties of Leonardite humic acid were observed. We found differences of more than one order of magnitude in the sorption coefficients of a given compound measured in NOM from different origins. The terrestrial HA exhibited substantially higher sorption coefficients compared to aquatic HA and FA. The difference between any two types of NOM is mainly reflected by a constant shift in the partition coefficients that applies to all compounds in the same way. This indicates that it is the number of available sorption sites per mass of sorbents rather than the types of intermolecular interactions between the sorbate and the sorbent that governs the major differences between the sorption properties of various types of NOM. An empirical correlation between the aromaticity and the differences in the sorption capacities of each NOM compared to Leonardite HA was found. While several models for the prediction of sorption coefficients of different compounds in a given NOM system have been published, no systematic approach has yet been developed for modeling of the natural variability in the sorption properties of NOM from different origins. The former models were evaluated using a subset of 200 experimental Leonardite HA/air partition coefficients. This evaluation revealed that none of the regression models based on partitioning into octanol yielded satisfactory fits for polar compounds although the octanol-based Karickhoff model showed good performance for nonpolar compounds. For PcKocWIN, a model based on molecular connectivity indices, some major shortcomings became apparent. SPARC, another increment method, predicted the experimental Leonardite HA partition coefficients with good accuracy. However, like other increment methods, SPARC suffers from the general disadvantage that its application domain is limited by its calibration dataset that is unknown to the user. A good description of the whole dataset is achieved with a polyparameter linear free energy relationship (pp-LFER) that explicitly accounts for cavity formation, nonpolar (van der Waals) and polar (electron donor/acceptor) interactions between the sorbate molecule and the sorbent phase. With this pp-LFER model, most of the Leonardite HA/air partition coefficients could be predicted within a factor of 2. The quantum chemical model COSMOtherm predicted the experimental Leonardite HA partition coefficients within a factor of 3 to 5 using a suggested 3-dimensional structure of Leonardite HA. COSMOtherm can be expected to be very robust with respect to new and complex structures because its calculations are based on a fundamental assessment of the underlying intermolecular forces; calibrations with experimental compound descriptors are not required. The pp-LFER model evaluated with the big experimental dataset for Leonardite HA was successfully applied to all other NOM. These pp-LFER equations provide for the first time a tool that allows including the variability in the sorption properties of NOM in environmental fate models. The pp-LFER model also successfully predicted organic-C/water partition coefficients collected from the literature when it was combined with experimental air/water partition coefficients. This expands the applicability of the results of this study because NOM/water sorption processes are of equal or even higher importance compared to NOM/air partitioning.