Environ. Sci. Technol. 2006, 40, 4211-4218
Modeling Polychlorinated Biphenyl Mass Transfer after Amendment of Contaminated Sediment with Activated Carbon DAVID WERNER,† UPAL GHOSH,‡ AND R I C H A R D G . L U T H Y * ,§ School of Civil Engineering and Geosciences, University of Newcastle upon Tyne, NE1 7RU Newcastle, United Kingdom, Department of Civil and Environmental Engineering, University of Maryland Baltimore County, Baltimore, MD 21250, and Department of Civil and Environmental Engineering, Stanford University, Stanford, CA 94305-5080
The sorption kinetics and concentration of polychlorinated biphenyls (PCBs) in historically polluted sediment is modeled to assess a remediation strategy based on in situ PCB sequestration by mixing with activated carbon (AC). We extend our evaluation of a model based on intraparticle diffusion by including a biomimetic semipermeable membrane device (SPMD) and a first-order degradation rate for the aqueous phase. The model predictions are compared with the previously reported experimental PCB concentrations in the bulk water phase and in SPMDs. The simulated scenarios comprise a marine and a freshwater sediment, four PCB congeners, two AC grain sizes, four doses of AC, and comparison with laboratory experiments for up to 540 days of AC amendment slowly mixed with sediment. The model qualitatively reproduces the observed shifts in the PCB distribution during repartitioning after AC amendment but systematically overestimates the overall effect of the treatment in reducing aqueous and SPMD concentrations of PCBs by a factor of 2-6. For our AC application in sediment, competitive sorption of the various solutes apparently requires a reduction by a factor of 16 of the literature values for the AC-water partitioning coefficient measured in pure aqueous systems. With this correction, model results and measurements agree within a factor of 3. We also discuss the impact of the nonlinearity of the AC sorption isotherm and firstorder degradation in the aqueous phase. Regular mixing of the sediment accelerates the benefit of the proposed amendment substantially. But according to our scenario, after AC amendment is homogeneously mixed into the sediment and then left undisturbed, aqueous PCB concentrations tend toward the same reduction after approximately 5 or more years.
Introduction Numerous water bodies in the United States and worldwide contain polychlorinated biphenyl- (PCB-) contaminated * Corresponding author phone: 001 650-723-3921; fax: 001 650725-8662; e-mail:
[email protected]. † University of Newcastle upon Tyne. ‡ University of Maryland Baltimore County. § Stanford University. 10.1021/es052215k CCC: $33.50 Published on Web 05/24/2006
2006 American Chemical Society
sediment that poses long-term risks to public health and wildlife (1). Owing to the magnitude of the problem, the remediation of these sediments requires innovative and costeffective solutions, such as in-place treatment with activated carbon (AC) amendment to achieve PCB sequestration. Modeling the complex interactions between PCB desorption kinetics, PCB sorption by AC, and PCB availability assists understanding of the practical implementation of remediation strategies based on the concept of PCB repartitioning and sequestration with AC incorporated into the biologically active sediment layer. The in situ sequestration of PCBs and other hydrophobic organic chemicals in sediment by AC has the potential to reduce the contaminant’s aqueous-phase concentration (2-4) and the contaminant exposure to biota (5). Thus, in situ treatment with AC is expected to achieve a marked reduction in the uptake of PCBs at the base of the food chain in aquatic ecosystems. The sorption of PCBs to AC is exceptionally strong (6). Even a low dose of AC may change the thermodynamic partitioning equilibrium in sediment significantly, for which a portion of the PCBs will transfer from sediment particles to the added AC. Slow mass transfer kinetics can hinder this PCB mass transfer and the establishment of the new thermodynamic equilibrium. A common observation in desorption experiments with hydrophobic organic contaminants from sediment is an initial fast release of a readily available fraction followed by the incremental slow release of a more strongly bound fraction (7). The transfer of desorbed contaminants to an activated carbon particle will be affected by such release phenomena as well as the mixing regime (8). In addition, the sorption of contaminants within AC is also kinetically hindered, and the time required to reach the adsorption equilibrium can be months to years depending on the AC grain size (9, 10). In this paper we discuss a numerical model that simulates the release of PCBs from sediment, sorption by added AC, and resulting concentration in the aqueous phase. Mathematical models describing desorption or adsorption kinetics of contaminants in porous particles are based on either firstorder kinetics (11) or diffusion (12, 13). Only the later accounts for the kinetic effect of the sorbent’s particle size, and therefore a diffusion-based model was chosen for this investigation, since the AC particle size is an important design parameter for the proposed sediment amendment. We couple the simulation of geophysical processes with a first-order rate uptake model for biomimetic semipermeable membrane devices (SPMDs) (14) to approximate the impact of in situ sequestration on PCB uptake in SPMDs. The objective of our work is to further the understanding of the mechanistic processes and parameters affecting the mass transfer of PCBs in sediment after amendment with AC.
Materials and Methods Concept. As briefly outlined in ref 3, our modeling approach distinguishes between two different sediment particle types: a light-density fraction representing carbonaceous particles such as charcoal, coal, coke, cenospheres, or wood, and a heavy-density fraction representing the mineral phase with coatings of organic matter. A third particle type in our numerical model is AC. We based our model on the assumption that the sediment is mixed as in the laboratory treatments (2, 4). The current model also incorporates firstorder rate mass transfer to SPMDs and first-order rate degradation in the aqueous phase. VOL. 40, NO. 13, 2006 / ENVIRONMENTAL SCIENCE & TECHNOLOGY
9
4211
FIGURE 1. Five domains of the model: Light, low-density sediment particles; heavy, high-density sediment particles; AC particles; semipermeable membrane device (SPMD); and external aqueous phase. Each particle type is represented by a mean particle size (geometric mean); the surface of each particle type is in linear partitioning equilibrium with the external aqueous phase; and the movement of compounds in to and out of particles is described by apparent diffusivities. The contaminant uptake by SPMDs is described by first-order mass transfer kinetics. A first-order rate accounts for degradation in the external aqueous phase. The conceptual model is depicted in Figure 1. Each domain (light and heavy sediment particles and AC) is represented by a geometric mean grain radius, solid density, and porosity. We assume that the surface of each particle type is in linear partitioning equilibrium with the external aqueous PCB concentration. The movement of PCBs in or out of the different particles is described by intraparticle diffusion with a concentration-independent apparent diffusion coefficient (12). For each time step we calculate the amount of PCBs diffusing out of the sediment particles and into the AC from the PCB concentration gradients within the different particle types, and we calculate the amount exchanged with an SPMD if present and the amount degraded. The concentration change of a compound within the SPMD is given as
(
)
Cspmd dCspmd ) kspmd Caq dt Kspmd
(1)
where Cspmd (grams per cubic centimeter) is the PCB concentration in the SPMD, kspmd (per second) the first-order mass transfer rate, and Kspmd (dimensionless) is the SPMDwater partitioning coefficient. The various fluxes determine the overall change in the bulk aqueous PCB concentration Caq (grams per cubic centimeter):
[
] [
dCaq Vh d 3 Rh 2 r Sh(r) dr )dt Vaq dt R 3 0 h Vl d 3 Vac d 3 Rl 2 Rac 2 r Sl(r) dr r Sac(r) dr Vaq dt R 3 0 Vaq dt R 3 0 l ac Vspmd Cspmd k C - kdegCaq (2) Vaq spmd aq Kspmd
[
∫
∫
]
(
∫
)
]
where Vx (cubic centimeters), with x ) h for heavy, l for light, or ac for AC particles, denotes the total volume of each phase component; Sx (grams per cubic centimeter) denotes the volumetric PCB concentration in the particles; Rx is the particle geometric mean radius; and kdeg (per second) is the first-order degradation rate in the bulk aqueous phase. Implementation. The implementation of the intraparticle diffusion part of this model is based on the explicit numerical 4212
9
ENVIRONMENTAL SCIENCE & TECHNOLOGY / VOL. 40, NO. 13, 2006
scheme described by Wu and Gschwend (12), who wrote their numerical model to investigate the effect of particle size distributions on sorption dynamics. Our model also distinguishes among different particle types. Following ref 12, we use different spatial resolutions or number of nodes N for each particle type to achieve faster computation times. Typical choices for N in our model runs were Nheavy ) 11, Nlight ) 61, and NAC ) 1201. The system of differential equations is solved with an explicit Euler scheme. A time step constraint avoids instability in the most sensitive component of the system. The code was implemented in Matlab and is provided as Supporting Information. Quantification of the Input Parameters. We evaluate the model for PCB contaminated sediment from the Hunters Point Naval Shipyard, San Francisco Bay, California, and for sediment from Lake Hartwell, South Carolina. The sorbent properties and PCB contamination histories of these sediments are quite distinct and have been extensively characterized (2, 4, 15). The methodology of the experimental studies is summarized in Supporting Information. The mass transfer model requires a total of 19 input parameters as listed in Table 1. The mass percentage of the light particles was determined by density separation with a saturated cesium chloride solution of density 1.8 g/mL (15). The solid-water partitioning coefficients for the heavy, Kh, and light, Kl, sediment fractions were calculated from the measured solid-phase PCB concentrations of each sediment fraction and the aqueous PCB concentrations measured for the bulk sediment after 2 weeks of sediment-water contact in rotating batches (2, 4). The desorption rates for the light, Da,l/rl2, and heavy, Da,h/rh2, sediment fractions were quantified from PCB desorption experiments with the bulk sediment (2, 4) by fitting the equation
m(t) mtot
) 1 - fPCB,l
[
6
π
2
∞
1
∑n
n)1
(1 - fPCB,l)
2
[
(
Da,l
exp 6
∞
rl 1
∑n
π2 n)1
2
2
)]
n 2 π2 t
(
exp -
-
)]
Da,h n 2 π2 t 2 rh
(3)
to the data (13). In eq 3, m(t)/mtot is the contaminant mass fraction desorbed, and fPCB,l is the sediment PCB mass fraction associated with the light-density particles. The later was calculated from the sediment mass fraction of the light particles, fl, and the partitioning coefficients, Kh and Kl, of the two density fractions. The details of the calculation can be found in the “initial condition” section of the Matlab code provided as Supporting Information. The desorption rates of the light and heavy particles, Da,l/rl2 (per second) and Da,h/ rh2 (per second), were thus the two fitting parameters, and the nonlinear multiparameter fit was performed with a Marquard-Levenberg procedure. Figure 2 shows the parameter fit for Hunters Point and Lake Hartwell sediment desorption data. The Hunters Point data fit well with eq 3 (Figure 2a). The Lake Hartwell data were more problematic because the readily released PCB mass associated with the heavy particles is 82%, as calculated from the density fractions and corresponding partitioning coefficients, whereas desorption data (Figure 2b) suggests that approximately 75% of the total PCB mass is readily released. Because eq 3 has only one release rate for the PCBs bound to the heavy particles, it is unable to match the experimental data neatly. This could be due to the presence of some strong or occluded sorption sites within the sand, clay, and silt of the mineral sediment matter of Lake Hartwell or the experimentally determined mass in the light-density fraction could have been underestimated. To obtain a reasonably good fit of the experimental desorption data as shown in Figure 2b, we used a slightly higher mass percentage of light particles of 1.8%
TABLE 1. Input Parameters for Model Calculations and Exemplary Values and Corresponding Sensitivities for 22′455′-PCB (PCB-101) in Hunters Point Sediment parameter
parameter values
sourcea (ref)
Sb
AC particle radius AC solid-phase density AC porosity AC dose AC-water partitioning coefficient octanol-water partitioning coefficient water phase diffusion coefficient mass fraction light particles light particle porosity light particle density light particles partitioning coefficient apparent release rate, light particles heavy particle porosity heavy particle density heavy particles partitioning coefficient apparent release rate, heavy particles water/sediment ratio SPMD/sediment ratio aqueous phase first-order degradation rate
rac ) 0.0075 cm dac ) 1.96 g/cm3 pac ) 0.55 dose ) 0.034 g/g Kac ) 1.5 × 109 cm3/g Kow ) 2.4 × 106 Daq ) 5.0 × 10-6 cm2/s fl ) 0.06 g/g pl ) 0.11 dl ) 1.2 g/cm3 Kl ) 7.6 × 105 cm3/g Da,l/rl2 ) 9.7 × 10-10 s-1 ph ) 0.20 dh ) 2.5 g/cm3 Kh ) 2.5 × 104 cm3/g Da,h/rh2 ) 2.2 × 10-7 s-1 ratiows ) 1 cm3/g ratiospmds ) 0.075 cm3/g kdeg ) 0 s-1
measured (3) measured (3) measured (3) measured (3) literature (6) literature (6) estimated (19) measured (15) estimated (6) estimated (6) measured (2, 15) measured (2, 15) estimated (13) estimated (13) measured (2, 15) measured (2,15) measured (3) measured (3) estimated (18)
1.0 0.5 -1.6 -1.0 -0.5 0.5 0.5 0.0 0.0 0.6 0.1 -0.1 0.1 0.4 -0.1 0.0
a For measured data, some original publications report total PCB or PCB homologue data, not PCB congener data. b Sensitivities were determined by calculating the effect of a variation in the parameter value P on the calculated aqueous PCB concentration Caq(28d) after 28 days. The sensitivity coefficient S is defined as follows: (∆P/P)S ) ∆Caq(28d)/Caq(28d).
charcoal (light particles) from the literature (6, 13). As demonstrated later, a variation in these parameter values has little impact on the calculated results and an estimate for these parameter values was considered sufficient. For the AC, the values of the input parameters were measured (3) or taken from Jonker and Koelmans (6), who report AC-water partitioning coefficients Kac for a number of PCB congeners at the very low aqueous PCB concentrations (picograms to nanograms per liter) that fit the range of aqueous PCB concentrations in our experiments. Because kinetic sorption experiments with AC and PCB congeners were not available for the relevant concentration range, the apparent diffusion coefficient in AC was estimated from (12)
Da )
FIGURE 2. Quantification of the PCB release rates from the desorption data for (a) Hunters Point sediment and (b) Lake Hartwell sediment for 244′-PCB (9, PCB-28), 22′55′-PCB (O, PCB-52), 22′455′-PCB (2, PCB-101), and 2344′5-PCB (], PCB-118). The PCB mass fraction on the light and heavy sediment particles was calculated from the partitioning coefficients, Kl and Kh, and the sediment mass fractions of the light particles, fl. The release rates from the light sediment particles, Da,l/rl2, and the heavy sediment particles, Da,h/rh2, were then determined by fitting eq 3 to the measured desorption data of the bulk sediment. (in lieu of 1.2% experimentally) in our model calculations for Lake Hartwell, thus allocating approximately 22% (in lieu of 18%) of the total PCB congener mass to the light sediment fraction with the slower release rate. The density and porosity of the sediment particles were estimated by use of typical values for clay/silt (heavy particles, Lake Hartwell), sand (heavy particles, Hunters Point) and
pac2 (1 - pac)dacKac + pac
Daq
(4)
where Daq (square centimeters per second) is the aqueous molecular diffusion coefficient, pac (dimensionless) is the porosity of the particle, dac (grams per cubic centimeter) is the solid density of the activated carbon, and Kac (cubic centimeters per gram) is the linear partitioning coefficient at the relevant very low aqueous PCB concentration range. To describe the contaminant mass transfer into SPMDs according to eq 1, we use empirical relations for the mass transfer rate kspmd and the SPMD-water partitioning coefficient Kspmd published by Booij et al. (14); these are based on the octanol-water partitioning coefficient Kow, for which values are reported by Jonker and Koelmans (6). On the basis of the estimation formula for membrane diffusion coefficients published by Booij et al. (14), one expects tri-, tetra-, and pentachlorobiphenyls to diffuse within 1 day across the SPMD membrane, and a first-order rate model of PCB uptake was considered appropriate for our simulations. Data for the quantification of the required input parameters was drawn from several sources, and then only for a small number of PCB congeners was a complete set available. Among the 11 PCB-Kac values reported by Jonker and Koelmans (6), complete sets of data could be assembled for 22′455′-PCB (PCB-101) and 23′44′5-PCB (PCB-118) in Hunters Point sediment and for 244′-PCB (PCB-28), 22′55′-PCB (PCB52), PCB-101, and PCB-118 in Lake Hartwell sediment. Table 1 lists the parameters and their values for PCB-101 in VOL. 40, NO. 13, 2006 / ENVIRONMENTAL SCIENCE & TECHNOLOGY
9
4213
FIGURE 3. Predicted (curves) and measured (symbols) reductions in aqueous PCB concentrations for (a) 2344′5-PCB (PCB-118) in Hunters Point sediment, treated with a variable dose of AC with a mean radius, rac, of 0.075 mm; (b) 2344′5-PCB (PCB-118) in Hunters Point sediment, treated with a 3.4% dose of AC with a mean radius, rac, of 0.075 or 0.022 mm; and (c) 244′-PCB (PCB-28), 22′55′-PCB (PCB-52), and 2344′5-PCB (PCB-118) in Lake Hartwell sediment treated with a 2.0% dose of AC with a mean radius, rac, of 0.075 mm. The Kac values used in the model calculations were from Jonker and Koelmans (6). After addition of an SPMD to batches with a limited volume of sediment, an additional decrease in the aqueous PCB concentration is predicted. This effect is more pronounced for sediment treated with a low dose of AC because it has a lower overall sorption capacity. experiments with Hunters Point sediment, and all other parameter values are given as Supporting Information.
Results and Discussion Modeling Results. Some exemplary model results are shown in Figure 3 and Tables 2 and 3, which compare the predicted and measured aqueous or SPMD-concentrations of PCBs relative to those in untreated sediment from Hunters Point and Lake Hartwell after amendment with AC. As shown in Figure 3, a rapid initial reduction in the aqueous PCB concentration is predicted over several weeks, followed by a slow incremental reduction over months. These results show that understanding the sorption kinetics is clearly relevant for evaluating the efficacy of the proposed AC amendment sediment treatment strategy. The impact of a variable dose of AC on the aqueous concentration of PCB-118 congener in Hunters Point sediment is shown in Figure 3a as predicted by the mass transfer model (curves) and as measured in variable dose experiments 4214
9
ENVIRONMENTAL SCIENCE & TECHNOLOGY / VOL. 40, NO. 13, 2006
(symbols). The model predicts that increasing the AC dose by a certain factor will increase the observed reduction in the aqueous PCB concentrations proportionally, as long as the AC dose is sufficient to reduce the aqueous PCB concentration in the sediment significantly (i.e., on the order of 90%). The predicted effect of the particle size on reduction in aqueous concentrations of PCB-118 in Hunters Point sediment contacted with a 3.4% (w/w) dose of either coarser (rac ) 0.075 mm) or finer (rac ) 0.022 mm) AC particle sizes is shown in Figure 3b, together with the corresponding experimental measurements. According to the model, reducing the AC radius by a certain factor should initially reduce the aqueous PCB concentration by the same proportion. However, note that the thermodynamic equilibrium is not affected by the size of the AC. In theory, the effect of AC particle size on the time-dependent aqueous PCB concentrations becomes less apparent at longer times. This is shown in Figure S1 in the Supporting Information. The model results indicate that one would intermittently expect a trend of lower treatment efficiency of activated carbon for the higher molecular weight PCBs as compared to lower molecular weight PCBs based on the combination of both slower desorption rate from sediment and slower uptake in AC. This is shown in Figure 3c for tri-, tetra-, and pentachlorobiphenyls in Lake Hartwell sediment amended with a 2% (w/w) dose of AC. This congener effect is in qualitative agreement with the experimental observation. Comparison of Modeling Results with Experimental Data. Previously, we observed for aqueous PCB concentrations that the model systematically overestimates the reduction in the aqueous PCB concentrations 28 days after AC addition to Hunters Point sediment (3). Now, with a much larger data set shown in Tables 2 and 3, we find that the ratios of measured to predicted reductions in aqueous or SPMD concentrations of PCBs are in the range of 2-6 with only a few exceptions. Thus, the investigation of two sediments confirms the earlier finding of a consistent deviation. Of special interest are the measured SPMD data for AC-amended Hunters Point or Lake Hartwell sediment after 180 and 540 days. These data provide evidence for a continuous increase in PCB sequestration as predicted by the model at times when aqueous PCB concentrations were near or below the analytical detection limit. We conclude that the numerical mass transfer model qualitatively reproduces the long-term behavior and describes the impact on aqueous and SPMD concentrations for several key variables in the experiments such as PCB congener properties, AC dose, AC grain size, and different sediment properties. However, the model systematically overestimates the actual benefit of the amendment by severalfold. The following section addresses the issue of the systematic deviation. Systematic Underestimation of Aqueous and SPMD Concentrations of PCBs after AC Amendment. The systematic overprediction of the efficiency of the AC amendment is evident in Tables 2 and 3 and Figure 3. This could be due to an overestimation of the apparent diffusion coefficient in eq 4, in which the tortuosity factor was assumed to be inversely proportional to the AC porosity. However, if we compare our estimated apparent AC-water partitioning coefficients Kac,app after 2 months of sorbent/solute contact in pure water/sorbent batches with our experimental measurements, we observe no systematic overestimation of the Kac,app of various PCB congeners, as shown in Figure S2 in Supporting Information. The blocking of AC pores by sediment colloids could explain the observed systematic discrepancy. Pore blocking would decrease the intraparticular pore space available for diffusion and increase the tortuosity, and thus part of the sorption capacity of the AC would become inaccessible. To
TABLE 2. Measured and Modeled Ratios of Aqueous Concentrations in AC-Amended versus Untreated Sedimenta Caq_treated/Caq_untreated AC
measured
PCB
dose (%)
RAC (µm)
28 days
180 days
101 118 101 118 101 118 101 118
3.4 3.4 1.7 1.7 0.34 0.34 3.4 3.4
75 75 75 75 75 75 22 22
0.093 0.13 0.15 0.20 0.54 0.68 0.013 0.020
0.04 0.20
28 52 101 118
2.0 2.0 2.0 2.0
75 75 75 75
0.018 0.040 0.16 0.12
