Environ. Sci. Technol. 2009 43, 8847–8853
Triple Domain in Situ Sorption Modeling of Organochlorine Pesticides, Polychlorobiphenyls, Polyaromatic Hydrocarbons, Polychlorinated Dibenzo-p-Dioxins, and Polychlorinated Dibenzofurans in Aquatic Sediments A . A . K O E L M A N S , * ,†,‡ K . K A A G , † A. SNEEKES,† AND E. T. H. M. PEETERS‡ Wageningen IMARES, P.O. Box 57, 1780 AB Den Helder, The Netherlands, and Aquatic Ecology and Water Quality Management Group, Wageningen University, P.O. Box 47, 6700 AA Wageningen, The Netherlands
Received July 15, 2009. Revised manuscript received October 15, 2009. Accepted October 15, 2009.
Here we analyze the potential of black carbon (BC) and oilinclusive models to explain in situ sorption of 1,1-dichloro-2,2bis(p-chlorophenyl)ethylene (DDE), 1,1-dichloro-2,2-bis(pchlorophenyl)ethane (DDD), organochlorine pesticides (OCP), polychlorobiphenyls (PCB), polyaromatic hydrocarbons (PAH), polychlorinated dibenzo-p-dioxins (PCDD), and polychlorinated dibenzofurans (PCDF) to harbor sediments. Such models are important to understand bioavailability and mobility limitations of these chemicals in the aquatic environment. Separate BC- or oilinclusive models have been described before. However, it is unclear whether oil could dominate in situ sorption in sediments that also contain BC, and whether the relative importance of phases would differ for different compounds. A BC- and oilinclusive model was evaluated against chemical data and measured sediment characteristics. Parameter uncertainty was assessed using Monte Carlo simulations. Fitted model parameters were consistent with literature data and were satisfactory from a statistical as well as a chemical perspective. Sorption to oil was strong, proportional to octanol-water partitioning (Log Kow) and of similar magnitude for OCP, PCB, PCDD, and PCDF. For PAH a single oil sorption coefficient was found. Oil dominated sorption only for PCBs, at oil levels above 50-250 mg oil/kg sediment. BC dominated sorption of most other compounds, especially high molecular PAHs, PCDD, and PCDFs.
Introduction Sediment quality assessment becomes increasingly important now that aquatic sediments have been recognized as important sources of hydrophobic organic chemicals (HOC). Consequently, sediments may limit achievement of chemical and ecological targets as set by the European Water Framework Directive (1). Bioavailability of contaminants is a crucial * Corresponding author e-mail:
[email protected]; phone: +31 317 483201; fax: +31 317 484411. † Wageningen IMARES. ‡ Wageningen University. 10.1021/es9021188 CCC: $40.75
Published on Web 10/26/2009
2009 American Chemical Society
variable to address in ecotoxicological risk assessment (RA) as well as in site characterization or monitoring in the framework of remediation of contaminated sediments. In retrospective RA or site characterizations, bioavailable concentrations may be directly measured using nondepletive passive samplers (solid-phase micro-extraction, SPME; polyoxymethylene solid-phase extraction, POM-SPE (2)) or mild exhaustive SPE techniques (XAD, Tenax (3)). In prospective assessments, typically models are used which attempt to accurately calculate mobile or bioavailable contaminant fractions from either emission data or from known concentrations per total volume of sediment. As for HOCs, models and legislations have traditionally adopted the equilibrium partitioning theory (EPT) concept (4). However, recent work shows that the traditional EPT model assuming linear partitioning to amorphous organic matter (AOM) (4, 5) is obsolete. Sediments appear to contain considerable fractions of condensed carbon phases (soots, coal, kerogen) that exhibit strong nonlinear sorption of hydrophobic, especially planar, organic compounds (6, 7). In this paper we refer to these materials as black carbon (BC). Furthermore, weathered oil residues (mineral oil, MO) have been shown to exhibit sorption strengths of similar or even higher magnitudes as compared to BC, for polycyclic aromatic hydrocarbons (PAH) and polychlorobiphenyls (PCB) (8-12). Obviously, presence of soot and/or oil fractions in sediments may significantly limit bioavailability and risks of HOCs. A limited number of earlier papers showed that in situ partitioning of PCBs and PAHs in sediments can be explained using a soot plus organic matter dual domain sorption model. A novel question, however, is whether association with oil should be accounted for when calculating or interpreting partitioning to oil- and BC-containing sediments, and whether the BC and oil sorption parameterizations for PCBs and PAHs also hold for other important compound classes such as 1,1-dichloro2,2-bis(p-chlorophenyl)ethylene (DDE) and 1,1-dichloro-2,2bis(p-chlorophenyl)ethane (DDD), organochlorine pesticides (OCP), polychlorinated dibenzo-p-dioxins (PCDD), and polychlorinated dibenzofurans (PCDF). Furthermore, the importance of oil was speculated from laboratory sorption data (10). It is not yet clear to what extent these laboratory derived sorption parameters agree with parameters derived from field data. Typically, field validation studies of sorption models are subject to large uncertainties due to environmental variation, model, and measurement error. Consequently, this calls for explicit consideration of propagation of uncertainty, preferably using probabilistic modeling. In this paper we interpret a data set of 12 contaminated sediments and 40 compounds (4 DDD/DDE, 4 OCP, 15 PAH, 7 PCB, 4 PCDDs, and 6 PCDFs) (13), in terms of dual and triple domain sorption models as specified above. No earlier study considered a single data set of this size and chemical variety for this purpose. Besides total pollutant concentrations, oil content, BC content and general sediment characteristics 24-h SPE (XAD) extraction data were used to calculate robust estimates of contaminant pore water aqueous concentrations. Model performances were compared for default (literature) versus optimized parameter settings. Uncertainty was quantified using Monte Carlo probabilistic modeling. Primary aims of this work were to (a) assess to what extent laboratory derived sorption parameters are consistent with in situ sorption data, (b) assess and evaluate differences in in situ sorption behavior among different compound classes, and (c) assess the importance of different VOL. 43, NO. 23, 2009 / ENVIRONMENTAL SCIENCE & TECHNOLOGY
9
8847
sorption domains, i.e., the added value of triple and dual sorption domain models compared to dual and single domain models.
Materials and Methods Sampling and Analysis. Twelve sediments were sampled in 2005 and 2006 in the North Sea Canal (The Netherlands) and two adjacent harbor areas (13) (Table S1 in the Supporting Information). In the harbor areas, samples were taken from the surface sediment (30 cm) using a box corer. In the main channel, samples were taken from the bed sediment below the accreted mud using a Jenkins core sampler. Wet sediment subsamples received internal standards (MBP-MXE 13C12labeled PCBs, Wellington Laboratories Inc.; Cil PAH US EPA 16 Cocktail, ES2528; EN-1948SS, EN-1948IS, and EN-1948ES 13 C12-labeled PCDD and PCDF sampling, syringe, and extraction standards, Wellington Laboratories, Campro Scientific) and were Soxhlet extracted 16 h with hexane/acetone (1:1 v:v; J.T. Baker, HPLC grade). For OCPs external standards were used, homemade with pestanal grade chemicals (Riedel de Hae¨n, Seelze, Germany). OCP injection standard was 1,2,3,4-tetrachloronaphtalene. Extracts were split in a part for PCDD/PCDF analysis and a part for PCB, OCP, DDD/ DDE, and PAH analysis. Representative sediment subsamples of 50 g were extracted 24 h with 25 g of XAD-2 (Amberlite, Sigma Aldrich) in 0.01 M CaCl2 solution. XAD-2 was salted out with 150 g/L NaCl (aq), washed with milli-Q water, airdried, and Soxhlet extracted as described above. Sediment and XAD-2 extracts for PCDD/PCDF analysis were cleaned by subsequent elution over multilayer silica and alumina columns. The cleaned extracts were concentrated, received an injection internal standard, and were analyzed using GCHRMS (Thermo Finnigan MAT 95) against internal standards (isotope dilution). Sediment and XAD-2 extracts for PCB, PAH, OCP, DDD/DDE, and mineral oil (hereafter denoted as oil) analysis were cleaned using activated silica, dried, and taken up in a recovery standard. Subsequently, PCB, PAH, OCP, and DDD/DDE were analyzed using GC-MS (HP5890, Agilent) against internal standards (PAH, PCB; isotope dilution), or external standards (OCPs, DDD/DDE). HOC cleanup recoveries were 80-100% (PCDD/PCDF), 30-100% (PAH), and 70-100% (PCB). Detection limits are specified in Table S2. Extracts for oil received an extra cleanup over florisil and were analyzed using GC-FID (HP5890, Agilent) using external standards. Sediment organic matter was measured as loss on ignition (550 °C, 3 h), after removal of carbonates. Black carbon was measured using the CTO375 chemothermal oxidation method (6, 14). Model Definition. In situ distribution of POPs was evaluated using the triple sorption domain model equation: Kd )
Cs nF,BC-1 ) focKoc + foilKoil + fBCKF,BCCw Cw
(1)
in which Kd is the sediment-pore water distribution coefficient (L/kg), Cs (µg/kg) and Cw (µg/L) are the HOC concentrations measured in sediment and water respectively, foc, foil, and fBC are the measured organic matter, oil, and black carbon weight fractions, Koc (L/kg), Koil (L/kg), and KF,BC ([µg/kg]/[ µg/L]nF,BC) are sorption constants for these respective phases, which are calculated from compound-class specific regressions with logKow, and nF,BC is the Freundlich exponent for nonlinear sorption to BC (set to 0.7 6, 15, 16). Combinations of the first and second, as well as the first and third, terms in eq 1 have been applied to field data in earlier studies (8, 15, 17, 18), but never the combination of all three of them, i.e., explicitly accounting for sorption contributions of organic matter, oil, and BC. Laboratory studies with pure BC have shown the superiority of Polanyi-Dubinin-Manes (PDM) or two-site Langmuir models in explaining HOC 8848
9
ENVIRONMENTAL SCIENCE & TECHNOLOGY / VOL. 43, NO. 23, 2009
sorption (19, 20). Field studies with “environmental” BC also attemped to prove the superiority of Langmuir or PDM approaches in modeling the BC adsorption term but never succeeded due to the uncertainty in the field data (ref 16 and discussion below). This is why we adopt a simple model approach where the Freundlich parameters are conditional. Modeling of Kd was performed with default parameters (see below) obtained from the literature as well as with optimized parameter sets (Table 1). Default Parameter Values. The variables Cs, foc, and foil in all samples were directly measured. BC content (fBC) was known for 18 samples from the area, partly overlapping with our 12 locations. BC levels were close (fBC ) 0.013 ( 0.0050, n ) 18) and showed no spatial trend or gradient. Consequently, fBC was set as a parameter at the average of measured values. Cw values were estimated from the “fast” XAD extractable concentrations (CXAD), by dividing them by Koc: Cw ) CXAD/Koc. The Koc values for amorphous carbon were calculated from compound-class specific quantitative structure property relationships (QSPRs) as recently provided by Van Noort (21). LogKow values were taken from refs 21 (PCBs, PCDD, PCDF), 2 (PAH), or 22 (OCP). Default KF,BC values for PCB and PAH were taken from logKF,BC-LogKow regressions calculated from measured in situ fBC, Cw, and Cs data from 23, as further specified in Figures S1 and S2. Default Koil values for PCB and PAH were estimated using new logKoil-LogKow regressions calculated from original laboratory data for sorption to weathered oil in the presence of sediment (10, 12, specified in Figures S3 and S4). Default BC and oil parameters for the other chemicals were set equal to those for PCBs. Optimization of Parameters. Parameters for organic carbon sorption were not optimized. Parameters for BC and oil sorption were optimized on measured Kd values using the Microsoft Excel Solver tool, using a relative least-squares criterium. Optimization was performed in three steps. First, with all parameters at their default values, parameters for black carbon sorption were optimized using data from four locations containing no oil. Subsequently, parameters for sorption to oil were optimized for samples with the highest oil levels (Table S1, locations 2, 22, 28, 36, 46). In a third iteration, fine-tuning of parameters was performed by optimizing all samples simultaneously. Typically, these final optimizations yielded only minor changes to parameter values and minor improvement (less than 20%) of the quality of fit. Model Evaluation. The added value of including BC and oil terms in eq 1 was statistically analyzed using ANOVA. Uncertainty Analysis. Accumulation of uncertainty in the optimized three domain model (eq 1), was assessed for one of the locations with relatively high oil content; location 36 (Table S1). For this location, Monte Carlo simulations with 60.000 iterations each were performed using the Microsoft Excel plug in MCSim version June 8, 2006 (24). For five parameters, values were randomly selected from uniform distributions (nF,BC) or normal distributions (LogKoil, LogKF,BC, LogKoc, and fBC). The fraction of BC was included in the uncertainty analysis because an average value was used instead of a separate value for each individual sample. An overview of ranges, averages, and errors used as input in the Monte Carlo simulations is presented in Table S3.
Results and Discussion Aqueous Concentrations Derived from Solid-Phase Extraction Data. Fast desorbing concentrations as estimated by 24-h XAD extractions decreased in the order PCBs > PAH > PCDD ≈ PCDF and showed a negative trend with increasing LogKow (Figure S5). Previously, these trends have been related to strong sorption tosand therefore slow desorption fromssedimentary black carbon fractions (15). So, the low fast desorbing fractions for high-LogKow PCB and PAH, and
TABLE 1. Black Carbon and Oil Sorption Parameter Values in Equation 1, with Resulting Root Mean Squares (RMSE) for Default and Optimized Parameter Settings compound classes DDD/DDE
OCP
PAH
PCB
PCDD
PCDF
0.955 -0.128 0.7 0.995 0.979
0.955 -0.128 0.7 0.995 0.979
0.919 2.11 0.7 0.129 6.29
0.955 -0.128 0.7 0.995 0.979
0.955 -0.128 0.7 0.995 0.979
0.955 -0.128 0.7 0.995 0.979
n) RMSE
18 0.221
7 0.145
71 0.145
28 0.0913
56 0.101
n) RMSE
3 0.0783
default parameters parametera slopeBC interceptBC nF,BC slopeoil interceptoil
n) RMSE optimized parameters parametera slopeBC interceptBC nF,BC slopeoil interceptoil
0.475 1.34 0.7 1.00 1.71
0.955 -0.128 0.7 0.995 0.979
n) RMSE
18 0.0970
7 0.145
n) RMSE
3 0.0501
n) RMSE
all locations 180 0.382 location 2 15 0.194 location 36 15 0.376
0.998 -0.22 0.6 0 6.27 all locations 180 0.168 location 2 15 0.163 location 36 15 0.0921
6 0.0431
1 0.0458
5 0.0752
3 0.0294
5 0.0650
1.03 -2.18 0.7 1.15 1.24
1.25 -2.79 0.7 0.995 0.979
1.29 -3.08 0.7 0.995 0.979
71 0.0797
28 0.0748
56 0.0752
6 0.0435 5 0.0454
1 0.0406 3 0.0222
5 0.0372
a SlopeBC, interceptBC, slopeoil, and interceptoil are slopes and intercepts of linear regressions of LogKF,BC and LogKoil versus LogKow respectively. nF,BC is the exponent in the Freundlich sorption isotherm equation.
PCDD and PCDF can be interpreted as a relatively low portion that is bound to organic matter, and thus a high portion to oil or BC. The fast desorbing concentrations are translated into Cw values to evaluate the in situ distribution of the HOC according to eq 1 as discussed in the next sections. The Cw data were obtained by dividing 24-h XAD extractable concentrations by Koc. One might argue that this is not the most logical procedure because more direct approaches are available, such as, for instance, employment of passive samplers (2, 25). Here, we present three main reasons for using short-term SPE extraction to estimate Cw values in this study. First, several recent reports show that SPE-derived rapidly desorbing concentrations are proportional to aqueous phase concentrations (through a partition coefficient), which in turn are proportional to concentrations in SPME fibers (26-30). Consequently, amorphous organic carbon fractions in sediments can be considered as “passive samplers” for HOC in a fashion essentially similar to SPME fiber material, and Cw values can be estimated either from rapidly desorbing concentrations or SPME concentrations using their respective solid-water partition coefficients: Koc and Kfiber. Second, in earlier work (15), we calibrated a dual domain model (eq 1, first and third terms) using Cw estimated from SPE-derived fast desorbing fractions for PCBs in flood plain lake sediments. Recently, we measured PCB Kd values for the same flood plain sediments as the ratio of measured PCB concentration in the solids and the pore water PCB concentration measured with POM passive samplers (2, 23). The measured PCB Kd values show excellent agreement to the Kd values modeled using parameters based on SPE fast desorbing concentrations
(17) (Figure S6). As the only difference between the procedures relates to the pore water concentration measurement, this directly shows the equivalence of the procedures in our laboratory. Third, passive samplers are difficult to apply in deep sediment layers, a drawback that does not apply to exhaustive SPE procedures which typically imply lengthy extractions in the laboratory. Modeling in situ Sorption: DDE and DDD. The p,p′DDE, p,p′-DDD, o,p′-DDE, and o,p′-DDD LogKd values (n ) 18) were modeled first with default PCB-based parameters. For LogKd > 5, modeled values fairly agreed to measured values (Figure S7). However, the lower values were overestimated up to 2 orders of magnitude. The overestimated LogKd values related especially to locations 74 and 77 which featured the lowest organic carbon content (1.2 and 1.7%, Table S1). Assuming that partitioning to amorphous carbon complies to the default QSPR (21), these compounds apparently bind less effectively to BC or oil than PCBs do. As oil contents were also very low in these samples, a plausible explanation is that not the oilsbut the default BC parameters (either the binding parameters or the BC abundance data)s were too high for these chemicals. The disagreement would be most obvious if BC sorption still dominated, i.e., in samples with low organic matter content, like 74 and 77. This explanation is most likely as the same locations also deviated for PAH and PCB (as will be discussed below). Optimizing the parameters using all data yields a relatively low LogKF,BC-LogKow regression slope of 0.47 (Table 1, Figure 1). This means that for DDE and DDD ∆LogKF,BC/∆LogKowis twice as low as for PCBs. This may be explained by DDE/ VOL. 43, NO. 23, 2009 / ENVIRONMENTAL SCIENCE & TECHNOLOGY
9
8849
FIGURE 1. Modeled Log Kd (L/kg) using optimized model parameters (Table 1) against measured Log Kd (L/kg) for six compound classes: DDD/DDE, OCP, PAH, PCB, PCDD, and PCDF. Numbers in the legend refer to location numbers and dashed line is the 1:1 line. DDD being more bulky than PCBs and being less able to adopt a planar configuration. Organochlorine Pesticides. For the OCPs only 7 measured Kd values were available (Figure 1, Figure S7), due to the low concentrations in either total concentrations or XAD extractable concentrations. These relate to R-, β-, and γ-HCH (Figure 1, measured Kd values 3 L/kg), in sediment from locations 62, 73, and 74. These locations had very low oil levels. The OCPs we studied do not belong to one structurally related compound class, like the PAH, PCB, DDD/ DDE, PCDD, or PCDFs do. Accordingly, we did not a prori assume the individual chemicals to behave similarly with respect to sorption to sediment constituents. Still, the modeled Kd estimates with PCB-based default parameters agree very well with the measured values (Figure 1). Note that chlorobenzenes are structurally related to PCBs. Because 8850
9
ENVIRONMENTAL SCIENCE & TECHNOLOGY / VOL. 43, NO. 23, 2009
of the small number of data points and the good agreement, the parameters for OCPs were not optimized. Polyaromatic Hydrocarbons. Using default parameters for PAH yielded reasonable predictions of Kd (n ) 180) for three locations (2, 23, 46) but generally, sorption was overpredicted and variability was large; up to 2 orders of magnitude (Figure S7). Interestingly, locations 2, 23, and 46 had the three highest ΣPAH levels (19-32 mg/kg, Table S1), whereas the two locations with the lowest ΣPAH levels (74 and 77, ΣPAH) 0.33, 0.32 mg/kg), deviate most from the 1:1 line. We provide two possible explanations for the lack of fit. First, we hypothesize that the current wide range of PAH concentrations of 2 orders of magnitude is not adequately captured by the default parameters. Second, the BC parameters may have been too high for locations 74 and 77. Obviously, optimizing the LogKF,BC-LogKow intercept from 2.11 to -0.22 improved the average fit (Figure 1 and RMSE
in Table 1). Oil contributed to total sorption with a fitted constant LogKoil ) 6.27 (Table 1), i.e., no variation of Koil with the hydrophobicity of the PAH was detected. This field-dataderived observation confirms our earlier laboratory results where a constant LogKoil of 7.0 ( 0.24 was observed for sorption of PAH to weathered oil (10). Tuning of parameters did not reduce the variation in modeled Kd values. Because BC was the dominant sorbing phase for most PAH, this may be caused by unaccounted variability in BC sorption parameters. Furthermore, use of an average BC content may have contributed to some unaccounted variability in the Kd values for PAH. Polychlorobiphenyls. As for the DDD/DDE and PAHs, modeled PCB Kd values (n ) 71) for locations 74 and 77 were overestimated 1 order of magnitude (Figure S7), which strengthens the idea that BC levels or affinities in these samples may have been overestimated. Optimizing the LogKF,BC-LogKow intercept to -2.18 and marginal modifications to other parameters yielded a good fit (Figure 1). The optimized LogKoil-LogKow regression parameters (Table 1) were still within error limits of the literature default value (95% CI slope ) 0.79-1.20, recalculated from data in ref 12). Polychlorinated Dibenzo-p-dioxins. Because PCDD specific sorption parameters for BC or oil in our study area are not available, first the default parameters for PCBs were evaluated. The measured Kd values (n ) 28) show reasonable agreement with Kd values modeled using PCB-based parameters (Figure S7). Apparently, PCB and PCDD partitioning in sediments is not fundamentally different. Optimization of the LogKF,BC-LogKow regression parameters resulted in a 25% higher slope, but with lower intercept, effectively yielding comparable values (Table 1). Regression parameters for oil required no change (Table 1). Polychlorinated Dibenzofurans. Similar to the PCDDs, PCDF Kd values (n ) 56) were modeled fairly well with default parameters values for PCBs, although an overprediction occurred at low Kd values (Figure S7). Optimization of the LogKF,BC-LogKow regression parameters resulted in a 30% higher slope, but with lower intercept, also yielding comparable values. In the LogKow ) 5-8 range, PCDF sorption to BC was not different from PCDD sorption to BC (∆LogKF,BC < 0.1). Regression parameters for oil needed no further optimization (Table 1). Differences in in Situ Sorption Behavior for Different Compound Classes. With respect to the optimized LogKF,BC-LogKow regression parameters (Table 1) two observations need to be considered. First, the regression parameters indicate less strong sorption than in previous calibrations. We provide several explanations for this, but our data do not allow quantification of the relative importance of the alternative explanations. For PAHs and PCBs, earlier calibrations were for flood plain lakes (15) with lower total concentrations and no oil. Lower concentrations may imply higher sorption in nonlinear isotherms or less sorption competition. For PCDD and PCDF earlier calibrations were for pure BC laboratory experiments (16). It is well-known that in the presence of dissolved organic matter, “fouling” of the carbon surfaces occurs, attenuating sorption for 1-2 orders of magnitude (19, 31). The current presence of oil may also contribute to attenuation of sorption to BC or organic matter (10, 11). Oil may cover or attenuate BC sorption similarly as organic matter fouling. Oil-inclusive modeling also implies that a smaller part of total sorption is attributed to BC, i.e., the apparent sorption to BC becomes less. Finally, steady-state aqueous HOC concentrations may have been enhanced due to ongoing degradation of the oil phase (9, 11). In summary, higher concentrations, fouling, presence of oil, and ongoing degradation of oil may explain lower apparent sorption to BC than in earlier studies.
FIGURE 2. Linear relationship between intercept and slope of LogKF,BC-LogKow regressions for six compound classes: DDE/ DDD, OCP, PAH, PCB, PCDD, and PCDF. The second intriguing observation is that the LogKF,BCLogKow slopes and intercepts for the six compound classes DDE/DDD, OCP, PAH, PCB, PCDD, and PCDF appear to be negatively correlated: low slopes coincide with high intercepts and vice versa (Figure 2). Slopes and intercepts of linear regressions are known to be correlated, but because these regression parameters relate to different compound classes, a mechanistic explanation is more likely. Low slopes of a regression against LogKow may indicate that process other than partitioning contributes to sorption, i.e., adsorption, which is acounted for by a higher intercept. The ultimate case would be a constant and high intercept representing a single distribution coefficient, like the single coefficient which is found for PAH sorption to weathered oil (10). Still, it is not clear why, for instance, DDD/DDE would have a lower slope and larger intercept than PAH. Importance of Multiple Domain Sorption Modeling: Relative Importance of Amorphous Organic Matter, Black Carbon, and Oil. The relative importance of the three sorption domains AOM, BC, and oil for the chemicals in sediments of the North Sea Canal was assessed in three manners. First, an ANOVA procedure was applied to test the superiority of a three-domain (AOM-BC-Oil) model over an AOM-only traditional model. The test used the residual sums of squares corrected for the decrease in degrees of freedom due to the larger number of parameters in the extended model. It appeared that the three domain model statistically is to be preferred over an AOM-only model for PAH, PCBs, PCDD, and PCDF (F-test, p < 10-5). For DDD/DDE and OCP the number of data was insufficient to yield a statistically significant result. We also tested the superiority of an AOM-BC-Oil triple domain model (eq 1) over an AOM-BC dual domain model, thus specifically testing the added value of oil. Including oil improved the fit compared to a BCinclusive model only for PCB (F-test, p ) 0.0113). For the other compound classes the improvement was not statistically significant. The fact that oil did not yield a better model according to statistical criteria does not prove that oil will never be relevant for these other classes. After all, the results are highly conditional and specific for the current data set. The significance of an improvement of the model fit depends on the number of measurements, measurement accuracy, the relative affinities of the chemicals for the three phases as well as the relative proportions of these phases. For instance, if oil levels would have been higher in these sediments, the significances of including oil in the lastmentioned test would have been better. VOL. 43, NO. 23, 2009 / ENVIRONMENTAL SCIENCE & TECHNOLOGY
9
8851
FIGURE 3. Modeled distribution of OCP, DDE/DDD (location 2), PCB (location 36), PAH (location 36), and PCDD/PCDF (location 36), over the three sorption domains OM, BC, and oil. PCB and PCDD/PCDF numbers relate to congener numbers. Apart from statistical testing, the relevance of the phases can be shown using distribution graphs. To illustrate the relevance of oil for in situ distribution of HOCs we selected the locations with highest oil levels, location 2 (DDD/DDE) or 36 (PCB, PAH, PCDD, PCDF). Results for these locations clearly illustrate the distinct differences among the compound classes (Figure 3). Typically, BC is most relevant for PCDDs, PCDFs, HCB, and >4-ringed PAH. Despite the presence of BC, oil is dominating the binding of PCBs and low molecular PAHs due to strong sorption to oil and less strong sorption to BC (Table 1). AOM takes care of up to 50-60% of sorption for DDD/DDE, and midmolecular weight PAHs. Without the extra BC and oil terms, eq 1 would underestimate Kd a factor 10-100 (PCDD/PCDF), 2-30 (PAH), 2-20 (OCP and DDD/ DDE), and 5-10 (PCB) (Figure S8). For PCBs, omitting the oil term but keeping the BC term yields an underestimation of a factor 5. Third, the relative importance of oil for capturing HOCs was explored for PCB153, using the in situ sorption parameters from Table 1 (Figure S9). Figure S9 shows the fraction of PCB153 present in 0-500 mg/kg (foil ) 0-0.0005) oil for a “default” sediment with foc ) 0.05 and fBC ) 0.005. At low pore water concentration (i.e., Cw ) 10-5 ng/L) oil dominates sorption (i.e., the fraction bound to oil is >50%) at 250 mg/kg oil or higher. At higher PCB concentrations (Cw ) 10 ng/L), oil dominates sorption at 50 mg/kg oil or higher. The reason for this dependence on Cw is the fact that sorption to black carbon is stronger at low PCB concentrations, and thus better competes with oil for binding of PCB153. Conversely, increasing the PCB concentration to a level higher than 10 ng/L did not further increase the fraction bound to oil due to saturation of BC. For sediments without phases that exhibit nonlinear sorption (i.e., no BC, kerogens, or coal), results for all pore water concentrations would be identical. Uncertainty in Multiple Domain Sorption Modeling. Our model accounted for three sorption domains and had five 8852
9
ENVIRONMENTAL SCIENCE & TECHNOLOGY / VOL. 43, NO. 23, 2009
parameters in its most complete form. A valid model parameterization was obtained but appeared to be rather conditional and thus site-specific. Considering the model complexity and the uncertainties in parameter values and some of the variables, quantification of error propagation is most relevant. Monte Carlo simulations for sediment location 36 showed uncertainties in modeled Log Kd that generally increased in the order PCB < PAH < PCDD/PCDF, with SDs of 0.15, 0.15-0.43, and 0.36-0.57 LogKd unit (Figures S10 and S11), and 90% confidence intervals of 0.5, 0.5-1.4, and 1.3-1.8 LogKd unit, respectively. These propagated uncertainties are similar to the variation in measured LogKd values, which also ranged from 0.5-1 (PCB) up to 1-2 Log Kd units (PCDD/PCDF, PAH). It should be noted that the SDs are average indicators of uncertainties that in fact are featured by continuous distributions (Figure S11). The fact that uncertainty was highest for PAH and PCDD/PCDF can be explained by the fact that for these chemicals BC was a relative important sorption domain (Figure 3). The BC term of eq 1 has a relatively high uncertainty because it is governed by four uncertain parameters: fBC, KF,BC, nF,BC, and Koc. Note that Koc also acts upon this term because in our model implementation Cw is defined by Cw ) CXAD/Koc, as mentioned before. Implications. The results from this paper show that two or three domain sorption models improved the estimation of in situ Kd values up to 2 orders of magnitude, which is very useful for fate modeling and prospective RA of HOCs. Including BC was shown to be necessary for statistical (i.e., significance) as well as quantitative (i.e., factor change of Kd) reasons, for all chemicals studied. This warrants the general use of dual domain sorption models for HOCs, instead of applying them only to PCB or PAH. A triple domain sorption model that also included oil was shown to be required for PCB, again for statistical as well as quantitative reasons.
Considering the fact that oil levels in this study were not that high, oil may be important for other HOCs too, even though this could not be statistically shown. This calls for further validation of this model with in situ HOC data at higher oil levels. Despite the advantages, probabilistic modeling clearly identified the limitations in multiple domain Kd modeling. Propagated errors are such that for retrospective RA or monitoring purposes direct chemical measurement may be more accurate and thus preferable.
Acknowledgments We thank the Department of Noord-Holland of the Dutch Ministry of Public Works, Transport and Water Management, for financial support. Comments by four anonymous reviewers greatly helped to improve the manuscript.
Supporting Information Available Sediment characteristics; limits of detection; Monte Carlo analysis specifications; sediment characteristics; LogKF,BC-LogKow regressions for sorption of PCB and PAH to BC and oil; fast desorbing fractions for PAH, PCB, PCDD, and PCDF; validation of estimation of Cw using fast desorbing concentrations; model performance using default parameters; model performance neglecting BC and oil; predicted fractions of PCB153 in oil as a function of sediment oil content; Kd values modeled with Monte Carlo simulations; probability distributions of modeled Kd values for BAP, PCB153, and PCDF131. This material is available free of charge via the Internet at http://pubs.acs.org.
Literature Cited (1) Fo¨rstner, U. Sediments and the European Water Framework Directive. J. Soils Sed. 2003, 2, 2–3. (2) Jonker, M. T. O.; Koelmans, A. A. Polyoxymethylene solid phase extraction as a partitioning method for hydrophobic organic chemicals in sediment and soot. Environ. Sci. Technol. 2001, 35, 3742–3748. (3) Pignatello, J. J. Slowly reversible sorption of aliphatic halocarbons in soils. I. Formation of residual fractions. Environ. Toxicol. Chem. 1990, 9, 1107–1115. (4) Di Toro, D. M.; Zarba, C. S.; Hansen, D. J.; Berry, W. J.; Swartz, R. C.; Cowan, C. E.; Pavlou, S. P.; Allen, H. E.; Thomas, N. A.; Paquin, P. R. Technical basis for establishing sediment quality criteria for nonionic organic chemicals using equilibrium partitioning. Environ. Sci. Technol. 1991, 10, 1541–1583. (5) Koelmans, A. A.; Lijklema, L. Sorption of 1,2,3,4-tetrachlorobenzene and cadmium to sediments and suspended solids in Lake Volkerak/Zoom. Water Res. 1992, 26, 327–337. (6) Gustafsson, o¨; Haghseta, K.; Chan, F.; McFarlane, A.; Gschwend, P. M. Quantification of the dilute sedimentary soot phase: Implications for PAH speciation and bioavailability. Environ. Sci. Technol. 1997, 31, 203. (7) Koelmans, A. A.; Jonker, M. T. O.; Cornelissen, G.; Bucheli, T. D.; Van Noort, P. C. M.; Gustafsson, o¨. Black Carbon: The Reverse of its Dark Side. Chemosphere 2006, 63, 365–377. (8) Sun, S.; Boyd, S. A. Sorption of Polychlorobiphenyl (PCB) Congeners by Residual PCB-Oil Phases in Soils. J. Environ. Qual. 1991, 20, 557–561. (9) Ghosh, U.; Weber, A. S.; Jensen, J. N.; Smith, J. R. Relationship between PCB desorption equilibrium, kinetics, and availability during land biotreatment. Environ. Sci. Technol. 2000, 34, 2542– 2548. (10) Jonker, M. T. O.; Sinke, A.; Brils, J. M.; Koelmans, A. A. Sorption of polycyclic aromatic hydrocarbons to oil contaminated sediments: Unresolved complex. Environ. Sci. Technol. 2003, 37, 5197–5203. (11) McNamara, S. W.; Ghosh, U.; Dzombak, D. A.; Weber, A. S.; Smith, J. R.; Luthy, R. G. Effect of Oil on Polychlorinated Biphenyl Phase Partitioning During Land Biotreatment of Impacted Sediment. J. Environ. Eng. 2005, 131, 278–286.
(12) Jonker, M. T. O.; Barendregt, A. Oil is a sedimentary supersorbent for polychlorinated biphenyls. Environ. Sci. Technol. 2006, 40, 3829–3835. (13) Kaag, N. H. B. M.; De Weerd H.; Peters R. J. B. Analysis of toxic components of sediment and biota samples from the North Sea Canal (report in Dutch); Report TNO Bouw en Ondergrond: The Netherlands, 2006. (14) Poot, A.; Quik, J. T. K.; Veld, H.; Koelmans, A. A. Quantification of Black Carbon in sediments and soils: Comparison of Rock Eval Analysis with traditional methods. J Chromatogr. A 2009, 1216, 613–622. (15) Moermond, C. T. A.; Zwolsman, J. J. G.; Koelmans, A. A. Black Carbon and Ecological Factors affect in situ biota sediment accumulation factors for hydrophobic organic compounds in flood plain lakes. Environ. Sci. Technol. 2005, 39, 3101–3109. (16) Armitage, J. M.; Cousins, I. T.; Persson, N. J.; Gustafsson, O.; Cornelissen, G.; Saloranta, T.; Broman, D.; Naes, K. Black carboninclusive modeling approaches for estimating the aquatic fate of dibenzo-p-dioxins and dibenzofurans. Environ. Sci. Technol. 2008, 42, 3697–3703. (17) Accardi-Dey, A.; Gschwendt, P. M. Assessing the combined roles of natural organic matter and black carbon as sorbents in sediments. Environ. Sci. Technol. 2002, 36, 21–29. (18) Bra¨ndli, R. H.; Hartnik, T.; Henriksen, T.; Cornelissen, G. Sorption of native polyaromatic hydrocarbons (PAH) to black carbon and amended activated carbon in soil. Chemosphere 2008, 73, 1805–1810. (19) Pikaar, I.; Koelmans, A. A.; Van Noort, P. C. M. Sorption of organic compounds to activated carbons. Evaluation of isotherm models. Chemosphere 2006, 65, 2343–2351. (20) Koelmans, A. A.; Meulman, B.; Meijer, T.; Jonker, M. T. O. Attenuation of PCB sorption to charcoal by humic acids. Environ. Sci. Technol. 2009, 43, 736–742. (21) Van Noort, P. C. M. Estimation of amorphous organic carbon/ water partition coefficients, subcooled liquid aqueous solubilities, and n-octanol/water partition coefficients of nonpolar chlorinated aromatic compounds from chlorine fragment constants. Chemosphere 2009, 74, 1024–1030. (22) Shen, L.; Wania, F. Compilation, evaluation, and selection of physical-chemical property data for organochlorine pesticides. J. Chem. Eng. Data 2005, 50, 742–768. (23) Koelmans, A. A., Poot, A.; Gillissen, F. Estimation of in situ fluxes of organic micropollutants from sediments (In Dutch); Report Wageningen University;Commissioned by RWS Waterdienst, 2008. (24) Barreto, H.; Howland, F. M. Introductory Econometrics: Using Monte Carlo Simulation with Microsoft Excel; Cambridge University Press: New York, 2006. (25) Stuer-Lauridsen, F. Review of passive accumulation devices for monitoring organic micropollutants in the aquatic environment. Environ. Pollut. 2005, 135, 503–524. (26) Kraaij, R.; Mayer, P.; Busser, F. J. M.; Van het Bolscher, M.; Seinen, W.; Tolls, J. Measured pore-water concentrations make equilibrium partitioning work - a data analysis. Environ. Sci. Technol. 2003, 37, 268–274. (27) Oen, A. M. P.; Schaanning, M.; Ruus, A.; Cornelissen, G.; Kållqvist, T.; Breedveld, G. D. Predicting low biota to sediment accumulation factors of PAHs by using infinite-sink and equilibrium extraction methods as well as BC-inclusive modeling. Chemosphere 2006, 64, 1412–1420. (28) You, J.; Pehkonen, S.; Landrum, P. F.; Lydy, M. J. Desorption of hydrophobic compounds from laboratory-spiked sediments measured by Tenax absorbent and matrix solid-phase microextraction. Environ. Sci. Technol. 2007, 41, 5672–5678. (29) Yang, Y.; Hunter, W.; Tao, S.; Gan, J. Relationships between desorption intervals and availability of sediment-associated hydrophobic contaminants. Environ. Sci. Technol. 2008, 42, 8446–8451. (30) Trimble, T. A.; You, J.; Lydy, M. J. Bioavailability of PCBs from field-collected sediments: Application of Tenax extraction and matrix-SPME techniques. Chemosphere 2008, 71, 337–344. (31) Jonker, M. T. O.; Hoenderboom, A. M.; Koelmans, A. A. Effects of sedimentary soot-like materials on bioaccumulation and sorption of polychlorinated biphenyls. Environ. Toxicol. Chem. 2004, 23, 2563–2570.
ES9021188
VOL. 43, NO. 23, 2009 / ENVIRONMENTAL SCIENCE & TECHNOLOGY
9
8853
