Modeling biphasic sorption and desorption of hydrophobic organic contaminants in sediments

A model was developed to predict biphasic sorption and desorption of hydrophobic organic compounds in contaminated sediments. The model was based on relatively rapid porous diffusion in amorphous organic carbon and slow solid-phase diffusion in condensed-phase organic carbon. The model was used to simulate measured solid-fluid phase desorption (rates and pore-water concentrations) for four polycyclic aromatic hydrocarbons exhibiting a range of hydrophobicities (phenanthrene, anthracene, pyrene, and benzo[a]pyrene) in two field-contaminated sediments from Utica Harbor (Utica, NY, USA) and Rouge River (Detroit, MI, USA). Pore-water concentrations have been related to bioavailability, indicating the potential usefulness of the model to predict bioavailability. Key model parameters included the fraction of condensed-phase carbon (estimated by combustion at 375 degrees C), partition coefficient to the condensed-phase carbon (estimated by desorption measurements on coal-like particles physically separated from Utica Harbor sediments), and diffusivity and ratio of volume to surface area of the condensed-phase organic matter (fitted to measured desorption data on both sediments and for the measured polycyclic aromatic hydrocarbons). Best fit for the diffusion coefficient in the condensed-phase organic matter was 8.5 x 10(-20) m2/s, and ratio of volume to surface area was 2 microm. These parameters estimated measured pore-water concentrations of all polycyclic aromatic hydrocarbons in both sediments with an average error of 46% and a correlation coefficient of 0.76 and the fast-desorbing fractions (as measured by the fraction removed with a nonpolar polymeric sorbent XAD-2) with an average error of approximately 30% and a correlation coefficient of 0.54 (14% and 0.76, respectively, for all but benzo[a]pyrene). Modeling results were relatively insensitive to the two fitted parameters, with changes of an order of magnitude or more being required to affect the correlation between the model and observations significantly.