Environ. Sci. Technol. 2009, 43, 1134–1140
Modeling the Global Fate and Transport of Perfluorooctanoic Acid (PFOA) and Perfluorooctanoate (PFO) Emitted from Direct Sources Using a Multispecies Mass Balance Model JAMES M. ARMITAGE,† MATTHEW MACLEOD,‡ AND I A N T . C O U S I N S * ,† Department of Applied Environmental Science (ITM), Stockholm University, SE-10691 Stockholm, Sweden, and Institute for Chemical and Bioengineering, Swiss Federal Institute of Technology, ETH Zurich, HCI G129, CH-8093 Zu ¨ rich, Switzerland
Received October 14, 2008. Revised manuscript received December 10, 2008. Accepted December 11, 2008.
The global-scale fate and transport processes of perfluorooctanoic acid (PFOA) and perfluorooctanoate (PFO) emitted from direct sources were simulated using a multispecies mass balance model over the period 1950 to 2010. The main goal of this study was to assess the atmospheric and oceanic longrange transport potential of direct source emissions and the implications for the contamination of terrestrial and marine systems worldwide. Consistent with previous modeling studies, ocean transport was found to be the dominant pathway for delivering PFO(A) associated with direct sources to the Arctic marine environment, regardless of model assumptions. The modeled concentrations for surface ocean waters were insensitive to assumptions regarding physical-chemical properties and emission mode of entry and were in reasonable agreement with available monitoring data from the Northern Hemisphere.Incontrast,modeloutputscharacterizingatmospheric transport potential were highly sensitive to model assumptions, especially the assumed value of the acid dissociation constant (pKa). However, the complete range of model results for scenarios with different assumptions about partitioning and emissions provide evidence that the atmospheric transport of directly emitted PFO(A) can deliver this substance to terrestrial environments distant from sources. Additional studies in remote or isolated terrestrial systems may provide further insight into the scale of contamination actually attributable to direct sources.
Introduction Perfluorocarboxylic acids (PFCAs) are highly persistent substances, ubiquitous in the global environment (1-4). Atmospheric transport is recognized as an important pathway for long-range transport (LRT) of many persistent organic pollutants (POPs) that are globally distributed (5). PFCAs, however, are weak acids that exist predominantly as nonvolatile anions (PFCs) in the environment and thus do not * Corresponding author e-mail:
[email protected]; phone: +46 (0)8 16 4012. † Stockholm University. ‡ Swiss Federal Institute of Technology. 1134
9
ENVIRONMENTAL SCIENCE & TECHNOLOGY / VOL. 43, NO. 4, 2009
fit the typical POP profile. It has been hypothesized that global dispersion is a result of atmospheric transport of volatile precursor substances that subsequently degrade to PFCAs in remote areas such as the Arctic (6, 7). Indeed, it is well established that fluorotelomer alcohols (FTOHs) and POSFbased chemicals (e.g., perfluorooctyl sulfonamidoethanols) can degrade in the environment to yield PFCAs with a range of chain lengths (6-9). An alternative pathway for global distribution of these substances is transport via ocean currents (3, 10, 11). As anions, PFCs have exceptionally high potential for transport in oceans compared to most conventional POPs because they do not extensively associate with sinking particulate matter, an important loss process for more hydrophobic substances (12). In the context of LRT, PFCs would be classified as “swimmers” (13), and ocean transport, although relatively slow compared to atmospheric transport, is still an efficient pathway for distributing these substances globally. Model-derived estimates of long-range transport potential that are independent of emission rate, such as Arctic contamination potential (ACP10) (5) and characteristic travel distance (CTD) (14), are frequently used to rank chemicals in terms of this property. Interestingly, PFCs were recently shown to have a relatively high ACP10, approximately 10 times higher than the ACP10 related to transport and degradation of FTOHs (15). However, since this measure does not distinguish between terrestrial and marine environments, it is possible that the potential of PFCs to contaminate remote areas is largely confined to ocean waters. Our research group first reported results of a global-scale simulation of the fate and transport of APFO (ammonium perfluorooctanoate) emitted from direct sources conducted using a latitudinally resolved, single species model (11). In that initial mass balance exercise, we assumed that the fate and transport of this substance could be represented by the anion (PFO) because the majority of the compound (>99%) was expected to be in ionized form in the environmental media included in the model (e.g., fresh and ocean water, soils, sediments). The main purpose of that study was to evaluate emission estimates and model performance through comparisons between modeled and observed concentrations in surface ocean waters, the compartment representing the most significant global reservoir in terms of contaminant mass (10). While the agreement between modeled and observed concentrations supported the plausibility of emission estimates, discrepancies such as substantial overestimation of observed concentrations in the central Pacific Ocean suggested the potential value of adopting a more spatially resolved model. Another limitation of our previous study was that only the anion (PFO) was explicitly modeled, whereas in reality, both the neutral and anionic forms will be present in the environment. Since PFOA has an appreciable vapor pressure, it is subject to surface-air exchange processes and will also be present in the gas phase of the atmosphere. Atmospheric transport may therefore play a more important role in the dispersion of this substance than was evident in ref 11, particularly to terrestrial environments distant from source regions. The goal of this modeling study is to investigate the globalscale fate and transport of PFO(A) (collective abbreviation for PFOA and PFO) by conducting a series of model simulations using a multispecies global-scale model with both latitudinal and longitudinal resolution. As in ref 11, only direct sources of PFO(A) are considered; however, model outputs characterizing both the oceanic and atmospheric LRT potential are presented. The results of the simulations 10.1021/es802900n CCC: $40.75
2009 American Chemical Society
Published on Web 01/21/2009
are then used to evaluate the alternative hypotheses regarding the global distribution of PFO(A) in the environment and identify monitoring data gaps that may provide important insights into these questions.
Methods Model Description. The fate and transport of PFO(A) emitted from direct sources was simulated using the BETR Global model (16). BETR Global is a multimedia environmental fate model that describes the global environment as 288 regions based on a 15° × 15° grid. Each region is subdivided into well-mixed compartments representing the atmosphere (two layers), soil, vegetation, freshwater, freshwater sediments, and surface ocean water. The model uses fugacity-based mass balance equations to simulate chemical fate (17) and is capable of both steady-state and dynamic calculations. The model also incorporates a representation of the intermittent precipitation approximation described in ref 18. We selected the BETR Global model because it is conceptually similar to the model used in our previous work (11) but includes increased spatial resolution. The greater spatial resolution was desirable particularly because available monitoring data from surface oceans indicate that substantial regional differences exist (4). For this work, we developed a multispecies version of BETR Global that can simulate ionizing species through specification of the acid dissociation constant (pKa), physical-chemical properties for both forms of the compound, and environmental pH. The multispecies calculations use the distribution ratio approach (19) to describe phase partitioning and exchange processes for PFO and PFOA simultaneously. Example calculations are presented in the Supporting Information (section S1). Using this approach, conversion between PFO and PFOA is assumed to be instantaneous, thus the ratio between anion and acid in each compartment is always 10(pH-pKa). We also adapted BETR Global to allow the user to directly enter environmental partition coefficients such as the organic carbon-water partition coefficient (KOC) and aerosol-air partition coefficient (KQA). This option was implemented to allow empirical information on partitioning behavior to be used directly and in its absence, to develop model scenarios covering a range of possibilities. Parameterization of Surface Ocean Exchange. We updated and improved the description of surface ocean flow patterns in BETR Global using data from the global drift buoy arrays now available on a 1° × 1° resolution from NOAA (20). Details of the approach used to derive the surface ocean exchange from these data are provided in the Supporting Information (section S2). Model parametrization for these simulations also incorporated deep-water formation. Lohmann et al. (21) discussed the importance of deep-water formation in the North Atlantic as a sink for low molecular weight PCBs, which are not strongly associated with particulate organic carbon in the water column. Since both the anionic and neutral form of PFO(A) are predominantly in the dissolved phase in aqueous systems, the deep-water formation pathway could be an important global sink. This hypothesis is supported by Yamashita et al. (4), who recently reported detectable concentrations of several perfluoroalkylated substances (PFAS) at depths up to 3500 m in the Labrador Sea, a region of deep-water formation. Our model parametrization of surface-deep ocean exchange is described in the Supporting Information (section S3). Model output from simulations (section S4) of iodine-129 transport into the Arctic Ocean, used to evaluate the model parametrization, is also presented (Figure S4, Supporting Information) along with a figure illustrating how PFO entrained in inflowing Pacific and Atlantic water is distributed in the surface waters
of the Arctic region in BETR Global (Figure S5, Supporting Information). Physical-Chemical Properties. Key model inputs include pKa, KQA, KOC, and first-order degradation half-lives for both species. The most recent direct experimental determinations of the pKa of PFOA reported a value of 3.8 ((0.1) at infinite dilution (22), whereas two indirect estimations have implied a value of approximately 1.3 (23, 24). Another recent publication (25) argued that the pKa of PFOA is most likely closer to zero on the basis of consideration of nonfluorinated analogues and property estimation software. The pKa value of 3.8 for PFOA was explained in terms of the helical conformation exhibited by this molecule, influencing its acidity (22). The authors suggested that only linear PFCA isomers with 8 or more carbon atoms are expected to exhibit helicity and hence an elevated pKa compared to shorterchain homologues. However, Moroi et al. (26) reported a positive relationship between number of carbons and pKa for longer-chain PFCAs, but the trend began from perfluoropentanoic acid (C5). The authors rationalized these results in terms of the presence of acid oligomers (AnHn) in solution (27, 28), which were suggested to have lower acid dissociation constants (Ka), and hence higher pKa, in comparison to monomeric acid (AH). According to this hypothesis, the positive relationship with chain length reflects enhanced hydrophobic interactions between alkyl tails of the acid molecules as the number of CF2 groups increases, leading to a higher proportion of AnHn in solution. Because of the lack of consensus, simulations with pKa values of 0, 1.5 and 3.5 were conducted. Our initial expectation was that similar results would be obtained regardless of pKa because the major bulk environmental compartments are not highly acidic; default pH values assumed in the model are soil 6.5, sediment 6.0, freshwater 6.0, and surface ocean water 8.1. With these pH values, only a minor fraction of total PFO(A) is protonated and the overall chemical fate is still determined largely by the properties of the anion. However, the speciation and behavior of PFO(A) could be quite different in aqueous phases associated with the atmosphere (e.g., rain, fog, aerosol humidity). For example, the pH of aerosols recently sampled in an urban area of the United States ranged from 0 to 4.5 (29), whereas estimated marine aerosol pH ranged from approximately 1.4-4.1 offshore the east coast of the United States and 2.6-5.4 in a more remote location (30, 31). Since higher values of pKa and lower values of aerosol/rain pH favor partitioning of PFOA into the gas phase, the modeled behavior of PFO(A) in the atmosphere (including deposition fluxes and LRT potential) could be highly dependent on the assumptions made regarding these parameters. Experimental determinations of the partitioning behavior of PFCAs are complicated by the strong tendency of these compounds to adsorb to surfaces (32). For example, while air measurements remote from point-sources indicate that the fraction of total PFO(A) mass associated with aerosols (φ) is 30-40% or higher (33, 34), these findings are uncertain because of the possibility of irreversible sorption to fiber filters used for air sampling (35). The influence of acidity on aerosol-air partitioning is also uncertain because of the range of possible pKa values and the variability in aerosol pH. As currently there are no aerosol-air partitioning models validated for PFCAs, we selected values of KQA corresponding to φ of 99% to span the range of partitioning behavior in the atmosphere. The pH of rain was assumed to be 4.5 in industrialized regions impacted by acid rain (e.g., U.S. east coast) and 5.6 elsewhere. Partitioning to other solids (soil, sediment, and suspended solids in water) was estimated from empirical KOC values reported in refs 36 and37. On the basis of these studies, values of 3400 (PFOA) and 100 (PFO) L kg-1 OC were assumed. Model outputs in VOL. 43, NO. 4, 2009 / ENVIRONMENTAL SCIENCE & TECHNOLOGY
9
1135
FIGURE 1. Modeled surface ocean concentrations (pg L-1) in 2005 with the maximum emission scenario assuming pKa ) 0 and air emissions diverted to land in region of origin. the major bulk compartments are expected to be relatively insensitive to these assumptions, based on our findings in ref 11. A value of -2.4 was selected for log KAW of PFOA (24, 32), while the log KAW of PFO was assumed to be negligible. Degradation rate constants in all environmental media were assumed to be 0.01% per year for both the neutral and anionic form. Global Emission Estimates. APFO emission rates from direct sources were estimated on the basis of ref 10. According to these estimates, the manufacturing of APFO and its subsequent use in fluoropolymer (FP) production accounts for nearly 90% of the total direct releases of APFO from 1951 to 2004. The majority of APFO was manufactured through the electrochemical fluorination (ECF) process, which yields linear (70-90%) and branched isomers (10-30%) across a range of chain lengths (10). Similar branched:linear isomer ratios would therefore be expected for geographical regions predominantly influenced by ECF-based sources (assuming isomers have similar pKa). For our model simulations, production was assumed to begin in 1951 and continue to 2010. Significant emission reductions from production were introduced after 2005 to reflect industry’s commitments to make improvements to its manufacturing plants. The total estimated emissions of APFO over the simulation period (1950-2010) range from ∼2600 to 5050 t, and therefore, two scenarios were considered, representing the minimum and maximum estimate. Emissions were assigned to regions with known APFO or FP manufacturing sites and production capacity was used as the primary basis to apportion the total estimated historic emissions. This simplifying assumption means that emissions per unit APFO or FP produced are identical across all sites. On the basis of this approach, the total historic emissions were distributed approximately 37% to North America (regions 78-80, see Figure 1), 37% to Asia (regions 92-94 and 114-116), and 26% to Europe/Russia (regions 60-62, 64, and 85). Further details of the derivation of the direct source inventories are presented in the Supporting Information (section S5). We assume that APFO will rapidly dissociate to PFO in the environment. Mode of Entry. Following ref 11, emissions from FP manufacturing facilities were assumed to be released to air (23%), freshwater (65%), and land (12%), while emissions from APFO manufacturing were assumed to be predominantly to freshwater (95%) with minor releases to air (5%). Stack emissions from FP manufacturing plants occur largely as vapors which cool and condense first to fumes (i.e., 1136
9
ENVIRONMENTAL SCIENCE & TECHNOLOGY / VOL. 43, NO. 4, 2009
ultrafine particulates) and then fine particulates (0.1-5 µm in diameter, ∼55% < 0.3 µm), likely consisting of solid PFOA or APFO as well as organic particulates with PFOA or APFO sorbed to them (38, 39). Because BETR Global lacks sufficient spatial resolution to model dispersion of stack emissions directly and no further information is available to characterize the composition of this source, two scenarios were considered in these simulations. In the first case, no modifications were made to the preliminary source profile. Emissions to air become distributed between the gas phase, aerosol phase, and rain as PFO(A) according to the aerosol-air and air-water partitioning coefficients and the pKa assumed for the simulation. In the second case, particulates emitted to air were assumed to deposit rapidly to surfaces in the source region (e.g., because of further particle aggregation/enhanced gravitational settling) and stack emissions were redirected to land in the emission input files. In these simulations, volatilization from surface compartments plays a critical role in supplying contaminant to the atmosphere. Differences in atmospheric model output between these two scenarios were therefore expected to depend strongly on pKa. We consider 60 different model scenarios representing two assumptions each about emission levels and mode of entry, three assumptions about pKa, and five assumptions regarding aerosol-air partitioning. Because modeled concentrations and fluxes are directly proportional to emission rates, only the maximum emission scenario was actually modeled since results can be scaled to represent the minimum emission rate. To explore the sensitivity of model results to bulk compartment pH values, we conducted additional simulations with the default pH in soil, freshwater and sediment (i) reduced by 0.5 log units, (ii) increased by 0.5 log units, and (iii) increased by 1.0 log unit.
Results and Discussion Modeled Distribution of PFO(A) in Surface Oceans. Modeled surface ocean concentrations (and hence transport fluxes) for a given emission rate were relatively insensitive to other model assumptions (e.g., within a median factor of ∼1.4 in the Northern Hemisphere and 1.6 for the Arctic). This variability is comparable to the factor of 2 related to the emission estimates. In terms of LRT, the highest concentrations in the Arctic marine environment were generated by the scenarios that assumed direct emissions to air and pKa of 3.5. For a given pKa, modeled PFO(A) concentrations in surface ocean waters for scenarios with emissions to air were
FIGURE 2. Measured (solid bars) versus modeled (white bars) surface ocean concentrations 2002-2006 (pg L-1). Measured values represent the range of reported concentrations in the geographical areas sampled, while modeled values represent the range of modeled concentrations (all scenarios) in the corresponding model regions (indicated in parentheses), assuming the maximum emission scenario. Modeled concentrations with the minimum emission scenario are approximately 50% lower. generally 10-30% higher in the Northern Hemisphere by 2005 and up to 2-fold higher in the Southern Hemisphere. These results reflect the rapid dispersion of contaminants in the atmosphere and the fact that PFO(A) transported northward in the atmosphere is not subject to deep-water formation, a major contaminant sink in these simulations. For example, approximately 25% of the emission total for North America and Europe (15% of global total) were transported to the deep ocean over the simulation period. The average modeled PFO(A) concentrations in surface ocean water in 2005 for the scenarios assuming pKa ) 0 and emissions to air diverted to land are shown in Figure 1. The corresponding values under the minimum emission scenario produce the same spatial pattern but are approximately ∼50% lower (i.e., a factor of 2600/5050). As in ref 11, the mass transfer of directly emitted PFO(A) into the Arctic marine environment via ocean transport is substantially higher than estimated fluxes for precursor substances (40, 41). The highest modeled concentrations in the surface oceans are found in the receiving waters adjacent to the major source zones (e.g., the Gulf of Mexico, U.S. east coast, North Sea, Mediterranean Sea, Sea of Japan/East Sea). Once in the marine environment, emissions from North America tend to be distributed northwards and eastwards with relatively little exchange across the equator. Emissions from northern continental Europe (regions 60-62) tend to be distributed into the North Atlantic, while emissions from southern continental Europe (region 85) enter the Mediterranean and are slowly exchanged into the open ocean. The Pacific Ocean is predominantly influenced by emissions from Asian source regions. The distribution in the Pacific shows that sufficient mass of PFO(A) from Asian sources is transported eastwards and northwards, such that modeled concentrations in the surface ocean at latitudes 30-60 °N are substantially higher than further south. Doubling times of modeled concentrations in Arctic regions from 1975-2005 ranged from 9-11 years, in agreement with the modeling work presented in ref 11. By 2005, modeled concentrations in the Arctic Ocean are similar at both “gateways” (i.e., Bering Strait/region 25 and north Atlantic/regions 12-14) despite the fact that nearly two-thirds of the estimated emissions were released from North American and European sources. This result reflects the influence of deep water formation in the North Atlantic, which acts to reduce the mass flow of PFO(A) into the surface waters of the Eastern Arctic Ocean. Modeled concentrations decline toward the interior regions of the Arctic but the relative change depends on the scenario. Hudson’s Bay (region 31 and 55) is particularly sensitive to assumptions regarding
mode of entry because of its proximity to source regions and potential for elevated atmospheric inputs. Comparison to Surface Ocean Monitoring Data. Global surveys of perfluorinated compounds in the marine environment were published in 2005 (3) and 2008 (4). Theobald et al. (42) sampled waters from the North and Greenland Sea and have also analyzed samples in the eastern Atlantic along a latitudinal transect from 53 °N to 30 °S (43). The ranges of observed and modeled surface concentrations (assuming maximum emissions) for the corresponding regions are presented in Figure 2. Analytical bias in measured water concentrations is of concern (44), but for our purposes, we assume the measurements are reliable. As shown in Figure 2, there is agreement between modeled concentrations and the range of reported values in the north Atlantic and Greenland/Norwegian Seas as well as offshore Japan. Modeled concentrations under the maximum emission scenario in these areas are within the range of observed concentrations and rarely exceed the peak reported values. Modeled and observed values in the mid-Atlantic (0-30 °N) showed a substantial discrepancy in terms of maximum concentration. The reported concentrations in refs 3 and 4 range from 67-439 pg L-1, whereas modeled values range from 17-64 pg L-1. However, concentrations reported in ref 43 for locations offshore of West Africa (region 107 and 131) are in better agreement with the modeled values. Interestingly, the value of 439 pg L-1 was from offshore Brazil (region 129), while the other samples from the mid-Atlantic (region 130 and 131) were lower. These observations may indicate that significant regional sources in South America exist, which were not included in the emission inventory. The comparison between modeled and observed concentrations in the Pacific Ocean (north of the equator) suggests that direct emissions from some parts of Asia (e.g., region 115 and 116) may also be underestimated. For example, modeled concentrations in the South China Sea (regions 116 and 117) under the maximum emission scenario (65-120 pg L-1) underestimate the reported values, which range from 160 to 420 pg L-1. Additional monitoring data could provide further insight but sampling in these areas should not be a priority. The priority for any future sampling campaigns should be surface waters of the north Pacific (>45 °N), particularly the Bering Sea, for which no published data are available. Monitoring data from this region would be valuable because they could corroborate the estimated PFO(A) mass flow into the Arctic via the Bering Strait. Measurements of surface water in the Mediterranean Sea, a water body with high modeled concentrations, would VOL. 43, NO. 4, 2009 / ENVIRONMENTAL SCIENCE & TECHNOLOGY
9
1137
FIGURE 3. Gross deposition flux (kg year-1) to High Arctic (regions 1-24) in 2005. Height of black (pKa ) 0), gray (pKa ) 1.5), and striped (pKa ) 3.5) bars represent model output with emissions to air and default bulk compartment pHs. The white bars underneath represent model output with no emissions to air (i.e., redirected to land in source regions). be useful to assess the plausibility of estimated emission rates in Europe. Atmospheric LRT Potential. While the simulated fate and transport potential of directly emitted PFO(A) is highly dependent on the assumptions made in each model scenario, the results indicate that emissions from point sources can contribute to the contamination of the terrestrial environment in both source and remote regions. Modeled annual gross deposition fluxes to North American regions below 60 °N (excluding regions 78-80) in 2005 range from approximately 45-60 ng m-2, 60-65 ng m-2 and 160-285 ng m-2 for pKa 0, 1.5, and 3.5, respectively, when emissions to air are included (assuming maximum emissions, see Supporting Information, Table S1-S3). Within source regions, the modeled gross deposition fluxes exceed 500 ng m-2 irrespective of model assumptions. These deposition fluxes are similar to or exceed the modeled PFO(A) deposition fluxes calculated by Yarwood et al. (45) related to the degradation of FTOHs released from North American sources (10-100 ng m-2 in U.S. East Coast, 99% because rain dissolution is minimized. For pKa ) 1.5, rain dissolution and particle deposition (wet + dry) tend to be comparable until φ > 50% after which particle deposition increasingly dominates. Comparison to Atmospheric Monitoring Data. Measured precipitation levels of PFO(A) from North American locations range from
