Black Carbon-Inclusive Modeling Approaches for Estimating the

Environ. Sci. Technol. 2008, 42, 3697–3703

Black Carbon-Inclusive Modeling Approaches for Estimating the Aquatic Fate of Dibenzo-p-dioxins and Dibenzofurans JAMES M. ARMITAGE,† I A N T . C O U S I N S , * ,† N . J O H A N P E R S S O N , † ÖRJAN GUSTAFSSON,† G E R A R D C O R N E L I S S E N , †,‡ TUOMO SALORANTA,§ DAG BROMAN,† AND KRISTOFFER NÆS| Department of Applied Environmental Science (ITM), Stockholm University, SE-10691 Stockholm, Sweden, Department of Environmental Engineering, Norwegian Geotechnical Institute (NGI), P.O. Box 3930, Ulleval Stadion, N-0806 Oslo, Norway, Norwegian Institute for Water Research (NIVA), Gaustadalléen 21, N-0349, Oslo, Norway, and NIVA, Televeien 3, N-4879, Grimstad, Norway

Received October 18, 2007. Revised manuscript received February 12, 2008. Accepted February 15, 2008.

A novel black carbon (BC) inclusive modeling tool is applied to estimate the distribution and long-term fate of dibenzo-pdioxins and dibenzofurans (PCDD/Fs) in the Norwegian Grenland Fjords. Three versions of the model were developed in which sediment-water partitioning was described using (i) an amorphous organic carbon (AOC) partitioning sorption model without BC sorption, (ii) a combined AOC and BC sorption model based on the Freundlich isotherm, and (iii) a combined BC-AOC model based on the Langmuir isotherm. The predictive ability of the three different models was evaluated for 17 PCDD/Fs by comparison of model predictions with observed organic carbon normalized sediment-water partition coefficients (KTOC) and with measured concentrations. All three versions of the model were able to predict concentrations that were in reasonableagreementwithmeasuredparticulateconcentrations (i.e., within a factor of 4 of median values). Estimated particulate concentrations were less sensitive to the model choice because the majority of the mass of these hydrophobic chemicals is associated with particulates regardless. However, for estimation of KTOC or dissolved water concentrations, both versions of the combined AOC and BC sorption models provided greatly improved estimates compared to the AOC-only model.

solid and the organic carbon normalized solid-water partition coefficient (KAOC) (1), Kd ) fAOCKAOC

(1)

where fAOC is the mass fraction of AOC in dried soil or sediment and KAOC (LW/kgAOC) is the organic carbon normalized solid–water partition coefficient. It has further been shown that KAOC is closely related to the octanol–water partition coefficient (KOW) (2, 3). The approach suggested by Karickhoff (3) for estimating KAOC (i.e., KAOC ) 0.41KOW) is the most common approach used in fate modeling (1) and was recently shown to predict the extent of sorption of nonpolar compounds to humic acid in good agreement with experimental values (4). There is, however, increasing evidence to suggest that the existing AOC partitioning model paradigm is not sufficient to explain the sorption behavior of all hydrophobic organic compounds. For example, Niederer et al. (4) demonstrated that sorption to humic acids can be substantially underestimated by the Karickhoff approach (3) if the compound can engage in specific molecular interactions such as hydrogen bonding (e.g., polar compounds). Furthermore, the presence of strongly sorbing pyrogenically derived materials, such as unburned coal, kerogen, cenosphere, soot, and charcoal (the remnants of incomplete combustion, commonly termed “black carbon” or BC), in sediment particles has been shown to lead to enhanced sorption of organic compounds, especially planar and aromatic ones including PCDD/Fs (see review by Cornelissen et al., ref 5). In this paper, a previously developed aquatic fate model (DIG [Dioxins in Grenland]-modeling tool) (6) is used to simulate the fate of PCDD/Fs in the Grenland Fjords in Norway to test the importance of BC sorption. In the previous version of the DIG-modeling tool described in ref 6, the problem with estimating sorption to BC was avoided by directly inputting observed values of the KTOC determined from field studies (7). Here, this problem is revisited by investigating the ability of the DIG-modeling tool to simulate the particulate-phase concentration and sediment-water distribution of PCDD/Fs in various parts of the fjord using both an organic matter partitioning (AOC) sorption model and two versions of a combined AOC and BC sorption model. The simulated results are evaluated through comparison to observations from three field studies (7). It is known from previous work that the distribution of PCDD/Fs in the fjord is strongly related to the fraction of BC (8) in the solids. In previous studies, a framework was developed for the inclusion of BC in multimedia chemical fate models (8) and for including BC in modeling bioaccumulation (9), but this is the first attempt to evaluate BC-inclusive multimedia models using monitoring data.

Methods Introduction Multimedia fate models typically assume that amorphous organic carbon (AOC) is entirely responsible for the sorbing capacity of solids for hydrophobic compounds and that the solid-water distribution coefficient Kd (LW/kgdw) can be readily estimated from the mass fraction of organic carbon in the * Corresponding author e-mail: [email protected]. † Stockholm University. ‡ Norwegian Geotechnical Institute. § Norwegian Institute for Water Research, Oslo. | Norwegian Institute for Water Research, Grimstad. 10.1021/es702638g CCC: $40.75

Published on Web 04/12/2008

 2008 American Chemical Society

Selected Compounds. Analytical data for model comparison are available for 7 PCDD congeners and 10 PCDF congeners from three separate campaigns that sampled the water column and bottom sediments in December 1998, June–July 1999, and May 2000, respectively (7). Additional sediment samples were also available from 1997. The compounds and their physical-chemical properties are listed in Table 1. Model and Study Site Description. The DIG modeling tool (version 0.29) is a multimedia box model based on the fugacity principle, which was developed to describe the longterm fate of PCDD/Fs in the marine environment of the Grenland Fjords in Norway (6) (Figure 1). The Grenland Fjords are five jointed fjords in the south of Norway (59° 5′ N, 9° 38′ VOL. 42, NO. 10, 2008 / ENVIRONMENTAL SCIENCE & TECHNOLOGY

9

3697

TABLE 1. Physical-Chemical and Phase Partitioning Properties of PCDD/Fs PCDD/F

log KOWa

log Kf,BCb (pg/kgBC)/(pg/L)n

log KBCc

2378-TCDD 12378–5D 123478–6D 123678–6D 123789–6D 1234678–7D OCDD 2378-TCDF 12378–5F 23478–5F 123478–6F 123678–6F 123789–6F 234678–6F 1234678–7F 1234789–7F OCDF

6.96 7.50 7.94 7.98 8.02 8.40 8.75 6.46 6.95 7.11 7.52 7.57 7.76 7.65 8.01 8.23 8.60

8.7 9.1 9.3 9.3 9.4 9.6 9.8 9.4 9.8 9.9 10.2 10.3 10.4 10.3 10.6 10.8 11.1

10.3 10.6 11.2 11.4 11.3 11.8 12.6 9.3 9.7 9.9 10.5 10.6 11.1 11.1 10.4 11.5 12.2

n CS ) fAOCKAOCCW + fBCKf,BC CW

a Values are from ref 17. b Values are from ref 15 and are corrected from 1 ug/L to 1 pg/L. Values at 1 ug/L estimated using LFERs modified from ref 16 (see ref 15 for justification). PCDDs: log Kf,BC ) 0.6 log KOW + 2.76 ; PCDFs: log Kf,BC ) 0.83 log KOW + 2.19, c Values were estimated from thermodynamic calculations as described in ref 16 using melting temperatures from ref 18 and subcooled liquid solubilities from ref 17.

E), which were modeled as a series of linked boxes representing water, sediments, and an overlying air compartment (see Supporting Information, Figure S1). The change in contaminant mass with time (dm/dt) for each of the boxes in the model was described by a differential mass balance equation: j

Ei +

∑ f D )f (∑ D tr

j

i

i

tr rx ij + Di

)

(2)

where the left-hand side of the equation comprises the input rate to the box, as Ei is emission (mol/h), and the summed term fjDji is the fugacity (f in Pa) in the adjacent boxes (subscript j) times the total intercompartmental transfer Dtr values (mol Pa1- h-1). The right-hand side comprises the loss rates from the box, that is, the fugacity in the box (subscript i), times the summed intercompartmental transfer Dtr values and reaction Drx value (mol Pa1- h-1). The system of equations was numerically solved for the fugacity by finite difference approximation over the simulation period. In the innermost fjord, the Frierfjord, a magnesium production plant began operating in 1951 and closed in 2002. Large amounts of PCDD/Fs and other chlorinated organic pollutants were formed as byproduct during the chlorination of magnesium oxide to yield water-free magnesium chloride. The long-term release of PCDD/Fs to the fjord (see Supporting Information, Figure S2) has left a legacy of contaminated sediments. As a result, concentrations in biota in the fjord exceed the EU recommended dietary limit of 4 ng/kg wet weight (sum of PCDD/Fs in 2,3,7,8-TCDD toxic equivalents). The model used here is parametrized exactly as in Persson et al. (6), except with regard to how the distribution between water and suspended particulates and water and bottom sediment and the flow of particulate material are described. In this modified DIG-model, adsorption to BC and transport of chemical in association with particulate BC is additionally considered. The modification to mechanistically model sorption to BC is described below and in the Supporting Information. 3698

9

Solid-Water Distribution of the PCDD/Fs. Both Freundlich and Langmuir isotherms have been used to describe the combined sorption into AOC (absorption) and onto BC (adsorption) (10). It is debatable as to which of these two models provides the best approximation, and indeed, neither may be ideal. The Freundlich nonlinear model represents a situation where the BC is heterogeneous and has multiple types of sorption sites acting in parallel, with each site exhibiting a different sorption energy and total site abundance (11). The Freundlich model for combined sorption to AOC and BC is represented by eq 3 (12)

ENVIRONMENTAL SCIENCE & TECHNOLOGY / VOL. 42, NO. 10, 2008

(3)

where CS is the concentration on the solid (µg/kg solid), fBC is the mass fraction of black carbon in the sediment (fTOC, fraction total organic carbon, ) fAOC + fBC), CW is the freely dissolved concentration of the compound modeled in water (µg/L), Kf,BC is the BC-normalized adsorption coefficient ((µg/ kgBC)/(µg/L)n) defined at 1 µg/L, and n is the Freundlich coefficient for sorption to BC. The value of n is typically in the range of 0.3-0.7 (5), which represents a nonlinear sorption isotherm at all solute concentrations in water; the average n value for 21 literature values for phenanthrene is 0.61 (5). In the Langmuir isotherm, at low sorbent concentrations, where the strongest sorption sites are far from being saturated, the relationship between the concentration on the solid and the water concentration is linear. This is contrary to the Freundlich isotherm where sorption is nonlinear at all water concentrations. In the Langmuir isotherm, a nonlinear isotherm only exists when there is competition for adsorption sites. The Langmuir model is represented by eq 4 (13),

[

CS ) fAOCKOCCW + fBC

bCWQmax 1 + bCW

]

(4)

where Qmax and b are the maximum site adsorption capacity of the sorbent (mol/kg BC) and the Langmuir site sorption affinity (L/mol), respectively. At low solute water concentrations eq 4 reduces to eq 5 (13). CS ) fOCKOCCW + fBCbQmaxCW

(5)

In the Langmuir isotherm, bQmax is thought to be equivalent to the BC-normalized distribution coefficient KBC ((µg/kgBC)/ (µg/L)). Unlike Kf,BC, KBC does not need to be corrected for the water concentration at which it was measured. According to the Langmuir isotherm, KBC must be measured at very low water concentrations, that is, when competitive sorption does not occur. KBC can also be derived from theoretical calculations as discussed later. Equation 5 simplifies to eq 6, which was first presented by Gustafsson et al. (14). CS ) fOCKOCCW + fBCKBCCW

(6)

Van Noort et al. (13) have suggested that if the solute concentrations of the sorbing chemical (CW) are less than 0.01 of the solid solubility in water (SS) then a linear sorption model can be applied (eq 6). Because the dissolved concentrations of PCDD/Fs in the Grenland Fjords are typically around the pg/L level in sediment porewater (15), the condition (CW , 0.01SS) is fulfilled, as the solid solubilities for the selected compounds are not lower than 100 nmol/m3 (1). Such a simple approach, however, does not account for the wide range of other organic contaminants present in the Grenland Fjords (e.g., PAHs) that could compete for sorption sites. Three versions of the DIG-model were generated in which sorption was estimated by (i) an organic carbon partitioning model AOC (eq 1), (ii) a combined AOC and black carbon (BC) sorption model based on the Freundlich isotherm (eq

FIGURE 1. The Grenland Fjords area with sampling sites indicated. Solid lines show locations of sills at Brevik and towards Skagerrak.

FIGURE 2. Predicted versus observed organic carbon–water partition coefficients (log KTOC, L/kg) for the AOC-only model (open square), AOC + Langmuir sorption isotherm (closed circle) and AOC + Freundlich sorption isotherm (closed triangle) in all boxes with observations for all congeners. Observed data represent discrete samples taken from 1998 to 2000 across the model domain. 3), and (iii) a combined AOC and BC sorption model based on the Langmuir isotherm (eq 6). Application of the Freundlich model based on eq 3 and the Langmuir model based on eq 6 provided some challenges for model parametrization. Fractions of AOC and BC have been previously measured in the suspended and bottom

sediments of the Grenland Fjords (7). Kf,BC values for PCDD/ Fs were recently measured by Bärring et al. (16) for monoto tetrachlorinated PCDD/Fs in experiments with high performance liquid chromatography (HPLC) where CW values were held close to 1 µg/L. However, measured values were not available for all the PCDD/Fs selected in this study. It VOL. 42, NO. 10, 2008 / ENVIRONMENTAL SCIENCE & TECHNOLOGY

9

3699

was therefore necessary to estimate Kf,BC (see Table 1) using the linear free relationships (LFERs) in refs 15 and 16 that related KOW and Kf,BC. No values were available for the Freundlich n term in eq 3 for PCDD/Fs, so a value of 0.7 [a common default value in other modeling studies (8, 9)] was used, and the sensitivity of this value on model results was tested. In the version of the Freundlich model used, CW is taken to represent only the concentration of the individual PCDD/F being modeling and thus does not account for the fact that other compounds could compete for the same adsorption sites. For the Langmuir model, KBC values for the selected PCDD/Fs were derived (Table 1) using knowledge of the thermodynamics behind the sorption process as described in ref 16, log KBC ) log (γW) - log (γBC) + log

( VA) + ln(10)R ( T - 1) + ∆Sφ

Tm

log (ABC) (7)

where γW is the activity coefficient, calculated according to yw ) (CwsatV) - 1; where Cwsat (l) (mol/L) is the subcooled liquid aqueous solubility of the pure compounds taken from Govers and Krop (17), and V is the specific molar volume of the dilute aqueous solution, 0.018 L/mol. The activity coefficient for a compound on the BC surface (γBC) is assumed to represent an ideal solid phase and is thus set to unity. In a dilute aqueous solution, the specific surface area of the BC (A) can be approximated to 3000 m2/mol, representing the product of Avogadro’s number and the two-dimensional area of water (0.5 Å2). The entropy loss for an aromatic molecule (∆Sφ) is assumed to be 33 J/mol/K (16), R is the gas constant (8.314 J/mol/K), Tm is the melting temperature (K) taken from Iorish et al. (18), and T is the ambient temperature (K). An experimentally determined area of BC (ABC) of 48 000 m2/kg (15) is used to convert KBC from surface-area based units into the more frequently used mass units. Intermedia Particulate Fluxes. Another important difference between the modified BC-inclusive DIG model presented here and the previously published DIG model (6) is that flows of particles between the different model compartments in the BC-inclusive DIG model are estimated with respect to both AOC and BC, respectively (see Supporting Information). Model Simulations and Comparisons with Monitoring Data. The model was originally designed to provide both historical and future contamination trends in the fjord (from 1950 to 2050) with the aim to determine how long it will take the fjord to recover from contamination (see previous modeling studies, refs 6 and 19). For the model evaluation exercise presented here, model simulations were carried out for 17 PCDD/F congeners for the period from 1950 to 2005. The measurements used for model evaluation were compiled from studies between 1997 and 2000, but it was necessary to undertake a long-term dynamic simulation for PCDD/Fs because of the long residence times of these compounds in sediments. Further details about the monitoring data are presented in the Supporting Information (see Figures S3 and S4). The predictive ability of the three different models was evaluated for all PCDD/Fs by comparison of model predictions with observed organic carbon normalized sedimentwater partition coefficients (KTOC) and with monitoring data for particulate-phase concentrations from field studies undertaken between 1997 and 2000 (7). KTOC (L/kgTOC) was computed from monitoring data according to eq 8, KTOC ) (CS ⁄ CW) ⁄ CTOC

(8)

where CS is the observed particle sorbed concentration (g/ L), CW is the freely dissolved water concentration (g/L), and CTOC is the total particulate organic carbon concentration (kgTOC/L). The freely dissolved water concentration was 3700

9

ENVIRONMENTAL SCIENCE & TECHNOLOGY / VOL. 42, NO. 10, 2008

calculated by correcting for association with dissolved organic carbon (DOC) using eq 9, app

corr

CW )

CW (1 + DOC × KD)

(9)

where appCW is the observed “apparently” dissolved concentration measured in the PUF sorbent (g/L) (7), DOC is the concentration of DOC in the water (kgDOC/L), and KDOC is the DOC-water partition coefficient (L/kgDOC). The DOC was measured in ref 7, and KDOC for the hydrophobic compounds such as PCDD/Fs may be estimated from a LFER relating KDOC and KOW (20, 21). According to the review of Krop et al. (21), KDOC may be anywhere between 0.1 and 1.0 KOC or, from application of KOC ) 0.41KOW, KDOC is between 0.04KOW and 0.4KOW. To be consistent with earlier studies (6), the LFER from Burkhard (20) (KDOC ) 0.08KOW) was applied using KOW values proposed in ref 17. The DIG model also accounts for association of the PCDD/Fs with DOC (see Table S1 in the Supporting Information). Scatter plots of predicted versus observed values were created to assess the performance of the different versions of the model. Model bias (MB), as described in the Supporting Information, was used to quantify the overall predictive power of the different version of model for the 17 congeners. An MB > 1.0 implies model overestimation, whereas an MB < 1.0 implies model underestimation. The average MB together with the 95-percentiles of the MB determine if there is systematic under- or overestimation. The 95-percentiles of the MB additionally inform us on the precision of the different versions of the model.

Results and Discusson Comparison of Predicted and Observed KTOC Values. Scatter plots for predicted and observed log KTOC (1998–2000) for the three versions of the model are presented in Figure 2. It is clear from this figure that the AOC model substantially underpredicts KTOC. The overall average MB for the AOC model is 0.003, and the 5th and 95th percentiles of MB, 0.0001 and 0.11, respectively, indicating a systematic underestimation of KTOC. Both the Langmuir and the Freundlich BCinclusive models perform much better. The overall average MB for the Langmuir model is 0.66 (5th and 95th percentiles; 0.02 and 25, respectively), whereas for the Freundlich model MB is 0.52 (5th and 95th percentiles; 0.003 and 85, respectively). Although there is considerable error in the model predictions, the average model bias for the BC-inclusive models is acceptable for models of this type, especially considering the complexity of the system being modeled. Additional insight into model performance can be gained by examining model bias on a congener-specific basis. A summary of this information is presented in the Supporting Information (Figure S5.) Whereas the AOC and Langmuir models do not show any large difference in average MB between dioxins and furans, the Freundlich model tends to underestimate the KTOC for dioxins by an order of magnitude (average MB ) 0.06) but provides good estimates for the furans (average MB ) 1.4). These results are partly related to the fact that the estimated Kf,BC values of the dioxins (corrected to 1 pg/L) are lower than for the furans (see Table 1). The relative magnitude of emissions for each congener must be considered as well because this factor influences the dissolved concentration in the water column and, hence, the degree to which the Kf,BC values are corrected compared to the values presented in Table 1. Regardless of the uncertainty in emission estimates, the results of this model evaluation suggest that the estimated Kf,BC and/or Freundlich n values used for the dioxins in the BC-inclusive Freundlich version of the model may need to be re-evaluated. In a field situation, competitive sorption would occur from the many different contaminants present in the fjord,

FIGURE 3. Predicted vs median observed water column particulate concentrations (pg/m) from 1998 to 2000 for the AOC-only model (open square), AOC + Langmuir sorption isotherm (closed circle), and AOC + Freundlich sorption isotherm (closed triangle) in all boxes with observations for all congeners.

FIGURE 4. Predicted vs median observed sediment solid concentrations (ng/g dw) from 1997 and 2000 for the AOC-only model (open square), AOC + Langmuir sorption isotherm (closed circle), and AOC + Freundlich sorption isotherm (closed triangle) in all boxes with observations for all congeners. especially considering the relatively low concentrations of PCDD/Fs compared to other contaminants present [e.g., polycyclic aromatic hydrocarbons (PAH)]. The version of the Langmuir model used assumes no competitive sorption on BC from other contaminants in the fjord, and the version of the Freundlich model used assumes only competition from other molecules of the single compound being modeled. The success of these versions of the Langmuir and Freundlich BC-inclusive models in predicting KTOC in the Grenlandsfjords would suggest that competitive sorption from other contaminants is not influencing the sorption of PCDD/Fs to BC. Conceptually, this behavior is well described with the version of the Langmuir model used here (eq 6), which assumes that competitive sorption is not occurring when concentrations of contaminants competing for sorption sites are sufficiently low. Given that the Langmuir model is simpler, requiring fewer parameters, as well as more accurate and precise, it could be argued that this approach is preferable for modeling this system, assuming that the monitoring data are representative. Key parameters affecting the prediction of KTOC (see eqs 3 and 6) are the Freundlich n, KBC or Kf,BC, KAOC, and fOC and fBC. KDOC is also important because it affects the value of the freely dissolved water concentration, CW, in eq 3. Because BC-sorption is much stronger than sorption to AOC, fOC and KAOC prove not to be sensitive input parameters when estimating KTOC for PCDD/Fs in this system. The remaining input parameters of interest are either based on measurements (i.e., Kf,BC or fBC) or estimated (i.e., KBC or KDOC). Although the review of Krop et al. (21) suggests that there is an order of magnitude uncertainty in the estimation of KDOC, the

uncertainties in the parameters defining the LFERs used to estimate Kf,BC (16) were not reported. The Freundlich coefficient, n, is another source of uncertainty. It has been shown to typically vary between 0.3 and 0.7 for PAHs in the laboratory (5). There are no values of n available for PCDD/ Fs, but it could conceivably be somewhere between 0.3 and 1.0 (with lower values of n yielding higher values of Kf,BC for a given dissolved water concentration). Experimentation with different values of n during model development demonstrated the large sensitivity of this parameter on model results (predicted KTOC varied by 3 orders of magnitude), and it is thus clear that it needs to be well-constrained in BC-inclusive models based on the Freundlich isotherm. The results of this model evaluation also suggest that it may be necessary to estimate congener-specific values of n for the water column particulates and bottom sediments. Ideally, Kf,BC values should be measured at the pg/L range to avoid having to adjust them so greatly for concentration, which would in turn reduce the high sensitivity of the Freundlich n. An obvious followup to this work would be to undertake a rigorous sensitivity and uncertainty analyses of the BC-inclusive models. A simpler model of the Grenlands Fjords system together with a novel technique for uncertainty analysis has recently been published, which would make undertaking and interpreting uncertainty analyses of this complex system feasible (19). Comparison of Predicted and Observed Particulate Concentrations. The evaluation of the model-predicted concentrations is focused on the particulate fraction since we have data for both the water column and sediments. Scatter plots of predicted and median observed particulate concentrations for the three different models are presented VOL. 42, NO. 10, 2008 / ENVIRONMENTAL SCIENCE & TECHNOLOGY

9

3701

in Figures 3 (water column) and 4 (sediments). The overall average model bias of the AOC-only model for the water particulates is 0.3 (5th and 95th percentiles; 0.03 and 7, respectively), and the averages are 3.9 (5th and 95th percentiles; 0.3 and 50, respectively) and 3.1 (5th and 95th percentiles; 0.3 and 30, respectively) for the Langmuir and Freundlich models, respectively. The overall average model bias of the AOC-only model for the sediment solids is 0.3 (5th and 95th percentiles; 0.005 and 20, respectively), and the averages are 4.0 (5th and 95th percentiles; 0.3 and 60, respectively) and 3.8 (5th and 95th percentiles; 0.1 and 105, respectively) for the Langmuir and Freundlich models, respectively. Model performance on a congener-specific basis is summarized in the Supporting Information (Figures S6 and S7). Although it can be concluded that the AOC-only model generally under-predicts and that both BC-inclusive models generally overpredict the measurements, the overall average model biases are essentially equivalent (i.e., all models are within a factor of 4 of the median values) and, thus, no model can be declared superior based on this assessment alone. This is especially true considering the sample sizes and variability in the measured values (e.g., range of measurements span over 1-order of magnitude in some cases). As shown in Supporting Information Figures S6 and S7, the AOC-only model tends to underestimate the median concentrations of the lower molecular weight PCDD/ Fs but exhibits improved model performance for the more hydrophobic congeners (e.g., OCDD). Interestingly, the opposite pattern was found for both BC-inclusive models, which demonstrates that the inclusion of BC-sorption leads to an overall increase in the total mass of contaminant retained within the model domain for all congeners. It is also interesting to note that the differences between the predicted particulate concentrations for the three models are much less than the differences between the predicted KTOC values. This has previously been observed by Prevedouros et al. (8) and relates to the fact that particulate concentrations of hydrophobic contaminants are less sensitive to the choice of model than are the dissolved concentrations because the majority of the mass of the chemical is already associated with particulates. For example, BC-sorption that leads to an increase from 97 to 99.999% of the mass associated with particles will have little influence on predicted concentrations. On the other hand, a decrease of 3% of the mass in the dissolved phase to 0.001% will have a huge impact on the predicted dissolved water concentration and, hence, the KTOC value. Models are often evaluated against bulk/total water or sediment concentrations (e.g., ref 22). In this study all models predicted bulk/total concentrations reasonably well overall, but only the BC-inclusive models were able to estimate the phase partitioning in agreement with measured values. It is recommended, therefore, that evaluations of models include comparisons of phase partitioning within media whenever possible, especially because dissolved phase concentrations are generally assumed to represent the bioavailable fraction for most aquatic organisms. From the results of this study, it was not possible to conclude whether the Langmuir or Freundlich model provides a better description of phase partitioning. Although the greater discrepancy between measured KTOC values and predictions generated by the Freundlich model for the dioxins suggests the superiority of the Langmuir model, the uncertainty in the estimated Kf,BC and n values used in the model simulations must be acknowledged. It is recommended that further model evaluations are conducted for aquatic systems with differing levels of contamination. It would be particularly interesting to undertake a model evaluation exercise in a system in which competitive sorption was occurring. 3702

9

ENVIRONMENTAL SCIENCE & TECHNOLOGY / VOL. 42, NO. 10, 2008

It is unfortunate that there is often a paucity of quality input parameters for parametrizing BC-inclusive sorption models, which will inevitably cause large uncertainties in model predictions. It is recommended that further effort be devoted to measuring fBC in soils and sediments and, in particular, to better constraining Kf,BC and the Freundlich coefficient n for a wider range of contaminants.

Supporting Information Available The Supporting Information contains a description of the modifications made to the original DIG model, a brief description of the model bias calculation, five tables of input parameters, one figure of the model structure, one figure showing the temporal trend in emissions for TCDD, two figures with details on measurement data, and three figures showing congener-specific model bias for KTOC, predicted water column particulate concentrations, and predicted sediment solid concentrations. This information is available free of charge via the Internet at http://pubs.acs.org.

Acknowledgments This study was financed by a grant from the Norwegian Research Council Programme Profo (139032/720) and Norsk Hydro through the Norwegian Institute for Water Research (NIVA, Project Dioksiner i Grenlandsfjordene). The EU integrated project NoMIRACLE (European Commission FP6 Contracts No.003956) is acknowledged for partially funding the Ph.D. studies of J. M. A.

Literature Cited (1) Mackay, D. Multimedia Environmental Models: The Fugacity Approach. Lewis, Boca Raton, Florida, 2001. (2) Chiou, C. T.; Peters, L. J.; Freed, V. H. Physical concept of soilwater equilibria for non-ionic organic-compounds. Science 1979, 206, 831–832. (3) Karickhoff, S. W. Semi-empirical estimation of sorption of hydrophobic pollutants on natural sediments and soils. Chemosphere 1981, 10, 833–846. (4) Niederer, C.; Goss, K-U.; Schwarzenbach, R. P. Sorption equilibrium of a wide spectrum of organic vapours in leonardite humic acid: Modeling of experimental data. Environ. Sci. Technol. 2006, 40, 5374–5379. (5) Cornelissen, G.; Gustafsson, Ö.; Bucheli, T. D.; Jonker, M. T. O.; Koelmans, A. A.; Van Noort, P. C. M. Extensive sorption of organic compounds to black carbon, coal, and kerogen in sediments and soils: Mechanisms and consequences for distribution, bioaccumulation, and biodegradation. Environ. Sci. Technol. 2005, 18, 6881–6895. (6) Persson, N. J.; Cousins, I. T.; Molvær, J.; Broman, D.; Naes, K. Modelling the long-term fate of polychlorinated dibenzo-pdioxins and furans (PCDD/Fs) in the Grenland Fjords, Norway. Sci. Total Environ. 2006, 369, 188–202. (7) Persson, N. J.; Gustafsson, Ö.; Bucheli, T. D.; Ishaq, R.; Naes, K.; Broman, D. Soot-carbon influenced distribution of PCDD/F in the marine environment of the Grenlandsfjords, Norway. Environ. Sci. Technol. 2002, 36, 4968–4974. (8) Prevedouros, K.; Palm-Cousins A.; Gustafsson, Ö.; Cousins, I. T. Development of a black carbon-inclusive model: Application for PAHs in Stockholm. Chemosphere 2008, 70, 607–615. (9) Hauck, M.; Huijbregts, M. A. J.; Koelmans, A. A.; Moermond, C. T. A.; van den Heuvel-Greve, M. J.; Veltman, K.; Hendriks, A. J.; Vethaak, A. D. Including sorption to black carbon in modeling bioaccumulation of polycyclic aromatic hydrocarbons: Uncertainty analysis and comparison to field data. Environ. Sci. Technol. 2007, 41, 2738–2744. (10) Cornelissen, G.; Gustafsson, Ö. Sorption of phenanthrene to environmental black carbon in sediment with and without organic matter and native sorbates. Environ. Sci. Technol. 2004, 38, 148–155. (11) Schwarzenbach, R. P.; Gschwend, P. M.; Imboden, D. M. Environmental Organic Chemistry. 2nd Edition. John Wiley & Sons. Inc., New York, 2003. (12) Accardi-Dey, A.; Gschwend, P. M. Assessing the combined roles of natural organic matter and black carbon as sorbents in sediments. Environ. Sci. Technol. 2001, 36, 21–29.

(13) Van Noort, P. C. M.; Jonker, M. T. O.; Koelmans, A. A. Modeling maximum adsorption capacities of soot and soot-like materials for PAHs and PCBs. Environ. Sci. Technol. 2004, 38, 3305–3309. (14) Gustafsson, Ö.; Haghseta, F.; Chan, C.; McFarlane, J.; Gschwend, P. M. Quantification of the dilute sedimentary ”soot-phase”: Implications for PAH speciation and bioavailability. Environ. Sci. Technol. 1997, 31, 203–209. (15) Persson, N. J.; Bucheli, T. D.; Gustafsson, Ö.; Broman, D.; Næs, K.; Ishaqa, R.; Zebühr, Y. Testing common sediment-porewater distribution models for their ability to predict dissolved concentrations of POPs in the Grenlandsfjords, Norway Chemosphere 2005, 59, 1475–1485. (16) Bärring, H.; Bucheli, T. D.; Broman, D.; Gustafsson, Ö. Sootwater distribution coefficients for polychlorinated-p-dioxins, polychlorinated dibenzofurans and polybrominated diphenylethers determined with the soot cosolvency column method. Chemosphere 2002, 49, 515–523. (17) Govers, H. A. J.; Krop, H. B. Partition constants of chlorinated dibenzofurans and dibenzo-p-dioxins. Chemosphere 1998, 37, 21392152. (18) Iorish, V. S.; Dorofeeva, O. V.; Moiseeva, N. F. Thermodynamic properties of dibenzo-p-dioxin, dibenzofuran, and their poly-

chlorinated derivatives in the gaseous and condensed phases. 2. Thermodynamic properties of condensed compounds. J. Chem. Eng. Data 2001, 46, 286–298. (19) Saloranta, T. M.; Armitage, J. M.; Haario, H.; Næs, K.; Cousins, I. T.; Barton, D. Modelling the effects and uncertainties of contaminated sediment remediation scenarios in a Norwegian fjord by Markov chain monte carlo simulation. Environ. Sci. Technol. 2008, 42, 200–206. (20) Burkhard, L. P. Estimating dissolved organic carbon partition coefficients for nonionic organic chemicals. Environ. Sci. Technol. 2000, 34, 4663–4668. (21) Krop, H. B.; Van Noort, P. C. M.; Govers, H. A. J. Determination and theoretical aspects of the equilibrium between dissolved organic matter and hydrophobic organic micropollutants in water (KDOC). Rev. Environ. Contam. Toxicol. 2001, 169, 1–122. (22) Armitage, J.; Cousins, I. T.; Hauck, M.; Harbers, J. V.; Huijbregts, M. A. J. Empirical evaluation of spatial and non-spatial European-scale multimedia fate models: Results and implications for chemical risk assessment. J. Environ. Monit. 2007, 9, 572–581.

ES702638G

VOL. 42, NO. 10, 2008 / ENVIRONMENTAL SCIENCE & TECHNOLOGY

9

3703