Influence of atmospheric degradation and deposition on LRT model results
The LRT results obtained with environmental fate models strongly depend on a chemical's effective residence time in air, ta,eff, which reflects the combined effect of degradation, physical removal, and net deposition. Chemical property data needed to estimate degradation in air and deposition from air to surface media are the rate constant for reaction with OH radicals in air, kOH; the aerosol–air partition coefficient, Kp (to quantify the effect of wet and dry particle deposition); and the Henry's law constant (to quantify the effect of rain washout and gaseous air–water exchange) . In addition to degradation and deposition, potential revolatilization from surface media influences ta,eff. The extent of revolatilization depends, among others, on the air-surface partition coefficients of a chemical (air–water, air–soil, air–vegetation, and air–snow partition coefficients), on the chemical's lifetime in the surface media , and on the air-surface equilibrium status (i.e., whether the surface is already close to equilibrium with the air above or whether net deposition of airborne chemical is still occurring). Particularly relevant for the global cycling of persistent SOCs is air-seawater exchange [32–35]; because of its large surface area and volume, seawater is an important reservoir of SOCs that is closely coupled to the atmosphere.
The available experimental data for kOH have been compiled by Klopffer and Wagner ; this compilation shows that for most semivolatile organic compounds, kOH has not been measured. Therefore, estimation methods, such as the Atmospheric Oxidation Program Software (AOPWIN)  (http://www.epa.gov/oppt/exposure/pubs/episuite.htm), have to be used for SOCs. The AOPWIN software estimates the rate constant of a molecule based on contributions from all relevant functional groups of the molecule [38–40]. The group contributions used in AOPWIN, however, have been derived from a set of small molecules, and whether the resulting values can be applied to complex molecules, such as many SOCs, is not clear [40,41]. Therefore, the AOPWIN results obtained for SOCs are fraught with considerable uncertainty and may systematically overestimate compound reactivity. In other words, a first problem for the calculation of ta,eff of SOCs is that kOH of the fraction of SOCs in the gas phase is uncertain.
A second problem for the calculation of ta,eff of SOCs is the following: To calculate the chemical's fraction sorbed to aerosols and, on this basis, the removal of the chemical by wet and dry particle deposition, the aerosol–air partition coefficient, Kp, is required. Values of Kp often are derived from the octanol–air partition coefficient, KOA, in a relationship such as Kp = cKOA . This approach is limited to aerosols whose sorptive capacity is determined by organic matter. For an average aerosol, as it is assumed in generic LRTP models, this is the case . In more highly resolved models, however, it is desirable to take aerosols consisting of mineral components and elemental carbon into account, because these aerosols may dominate the aerosol–air partitioning of SOCs in certain regions. For such aerosols, however, Kp cannot be derived from KOA, as is further discussed below.
A third problem is that the extent to which the particle-bound fraction, Φ, of an SOC is subject to degradation is unknown. It has been proposed that the particle-bound fraction is shielded against OH radicals and light and, therefore, is not significantly degraded [44–48]. Reactivity of organic components contained in atmospheric aerosols, however, has been observed in general [49,50]. Accordingly, it must be concluded that the half-life of SOCs bound to aerosol particles is largely unknown and that experimental investigations of the reactivity of SOCs on aerosol particles are highly desirable.
Finally, the effect of deposition to vegetation and to ice and snow on the LRT of SOCs also has been investigated [10, 51-56]. Vegetation (particularly forest canopies) has a large surface area that can filter SOCs out of the air, and measured air–vegetation deposition velocities are higher than those for other surfaces [57,58]. The capacity of a forest canopy to store and degrade SOCs, however, is limited . Deposition of SOCs to forest canopies may reduce the LRT of SOCs with KOA values between 107 and 1011 [51,59], but the overall importance of deposition to forest canopies to the LRT and global fate of chemicals is limited [60,61].
The effect of snow on the environmental fate of chemicals is complex [54,56] and can only be touched on here. Efficient scavenging of chemicals from air by snowfall is a factor that potentially reduces the LRT of airborne chemicals . After deposition to a snowpack, different processes take place, depending on the properties of the chemical and of the snowpack [56,62]. Relatively volatile chemicals, such as lighter PCBs, revolatilize to a significant extent. Involatile and water insoluble chemicals, such as heavy PCBs and polybrominated diphenyl ethers (PBDEs), are associated to particles in the snowpack and are released to soils during or after snowmelt; chemicals with relatively high water solubility and affinity to the ice surface, such as α-HCH, are released with meltwater. An interesting effect of large, snow-covered areas on the LRT of relatively volatile compounds (lighter PCBs and HCB) may be that a repulsive effect of the snow cover in high latitudes prevents the chemicals from deposition and reduces their northward transport .
Generic multimedia box models for LRTP screening
Several generic LRTP models were developed in the 1990s, including ChemRange [9,63], TaPL3 , and ELPOS . In addition, LRT calculations also were performed with already-existing box models, such as CalTOX  (http://eetd.lbl.gov/ied/era/) and SimpleBox . These generic LRTP models assume homogeneous environmental conditions and are based on average values of temperature, wind speed, precipitation, and so on. They yield LRT metrics as output, derived from air–water partition coefficient (KAW) and KOW values, and degradation half-lives (soil, water, and air) as chemical-specific input.
Because LRTP models such as ChemRange, TaPL3, and so on are generic, the question of how they can be evaluated and tested for validity is of particular interest. This question can be addressed by comparisons of model results to field data and by model intercomparisons. Shen et al.  measured 17 organochlorine compounds along a south–north transect from Central America to the Arctic and derived profiles of chemical concentrations in passive samplers as a function of latitude. From these profiles, they derived empirical travel distances (ETDs), which range from less than 500 km for heptachlor up to 13,000 km for pentachlorobenzene. In addition, they calculated CTDs and spatial ranges with three generic LRTP models (TaPL3, ELPOS, and ChemRange) and then compared results from these models to the empirical data. The three models were completely consistent in terms of ranking the chemicals, and the model results were in very good agreement with rankings according to the ETDs and also with the numerical values of the ETDs. This demonstrates that LRTP metrics also can be derived from field data and that the generic LRTP models yield results that describe the empirical LRTP results for organochlorine chemicals well.
Muir et al.  performed a similar analysis for five current-use pesticides (CUPs; atrazine/desethylatrazine, alachlor, disulfoton, dacthal, and diazinon). In their case, the empirical data are concentrations of the pesticides measured in the water column of Canadian lakes from 40°N to more than 70°N. However, whereas Shen et al.  had concentrations in air for the full range from 30°N to 80°N, the CUP data have a gap between 55°N and 70°N and only a few data points north of 70°N, which means that their ETDs are more uncertain. In the comparison with results from the three generic LRTP models, Muir et al.  found that all three models yield LRT results that are considerably below the values derived from the field data. To understand this discrepancy, it is helpful to consider the processes determining the effective atmospheric residence time of SOCs, ta,eff—namely, degradation in air and net deposition to surface media. For the CUPs investigated by Muir et al., both degradation and deposition likely are overestimated by the three generic LRTP models: degradation because the kOH values used are too high (applies to disulfoton, dacthal, and diazinon), and deposition because rain washout of chemicals with low Henry's law constant is overestimated (applies to atrazine, desethylatrazine, and alachlor).
To determine degradation of CUPs by OH radicals, second-order OH reaction rate constants have to be estimated with AOPWIN (see above), and then these rate constants are multiplied by an average concentration of OH radicals of approximately 7 × 105 molecules/cm3. Both of these components may contribute to the low LRTP results obtained with the LRTP models: OH radical concentrations in northern Canada likely are lower than the average of approximately 7 × 105 molecules/cm3 assumed in the models. In addition, the second-order rate constants obtained with the AOPWIN method may be too high (see above). As pointed out by Klöpffer and Wagner , the advent of quantitative structure–activity relationship methods in the 1990s has caused a sharp reduction in experimental studies. This is unfortunate, because further experimental studies are crucial to broaden the empirical basis of methods such as AOPWIN, and measurements of OH reaction rate constants of complex substances are highly desirable.
Rain washout of chemicals with Henry's law constants below 0.01 Pa·m3/mol is overestimated in the generic LRTP models, because continuous rainfall is assumed in the models. In reality, rainfall events are separated by dry periods in most regions of the world. In these dry periods, water-soluble chemicals also can undergo significant LRT . The effect of these transport events is not represented in the current versions of the generic LRTP models, which leads to the underestimation of the LRTP of atrazine and alachlor observed by Muir et al. . The effect of intermittent rain, however, can be approximated in generic LRTP models by introducing an upper bound of the wet deposition velocity that represents the effect of periods without rain .
A different approach to validating a generic LRTP model was used by Stroebe et al. . They investigated whether the so-called Junge hypothesis is reproduced by the ChemRange model, a generic global LRTP model. The Junge hypothesis  relates a chemical's atmospheric lifetime to the variability in the chemical's concentration that is measured in the air. Chemicals with long lifetimes are relatively homogeneously distributed in the environment, especially in regions distant from the source; therefore, their concentrations exhibit low scatter. This relationship can be expressed as ρ = a · t−ba, where ρ is the coefficient of variation (standard deviation divided by mean) of a chemical's concentrations measured in the field, ta is the chemical's residence time in air, and a and b are coefficients. Junge reported a = 0.14 years and b = 1, but in general, a and b depend on the type of chemical investigated, the location of the measurements in relationship to the source region, and other factors . A Junge relationship with b = 1 represents the long-term effect of the large-scale distribution dynamics of airborne chemicals. Using the ChemRange model, Stroebe et al.  calculated coefficients of variation of chemical concentrations in a region far from the source for a range of volatile chemicals with atmospheric lifetimes from 1 to 104 days. They found that the model reproduces the empirical relationship that Junge  derived from field data on atmospheric trace gases (b = 1), which indicates that the treatment of atmospheric LRT in the model correctly reflects the large-scale and long-term distribution dynamics of airborne chemicals.
If the Junge relationship is applied to SOCs, the distinction between the effective atmospheric lifetime and the degradation lifetime in air is essential. As introduced above, the effective residence time in air, ta,eff, reflects the overall removal of a chemical from the air by degradation and net deposition and, therefore, is shorter than or close to the degradation half-life. For SOCs with significant net deposition from air to the surface media, a Junge relationship is only observed if ta,eff is related to p .
Finally, it is of interest how different the various generic LRTP models actually are. Fenner et al.  compared nine multimedia box models that have been used for Pov and LRTP calculations. They defined a set of 3,175 hypothetical chemicals with half-lives in air from 4 h to one year, half-lives in water from 1 d to 10 years, log KAW from -11 to 2, and log KOW from 1 to 8. With all nine models, Fenner et al. determined the overall persistence, Pov, and LRTP of the 3,175 hypothetical chemicals, ranked the chemicals according to Pov and LRTP, and between any pair of models, calculated rank correlation coefficients (RCCs). For Pov, eight of the nine models show RCCs above 0.9. The rankings in Globo-POP, the ninth model, differ from those in the other eight models, because Globo-POP is the only model with different temperatures in different regions. For LRTP, the models are less consistent, both because some use transfer-oriented and some target-oriented LRTP metrics (but five models using transfer-oriented LRTP metrics still have RCCs above 0.9) and because of different amounts of ocean water and, thereby, different importance of LRT in water. On the basis of this model comparison, the OECD Expert Group for Multimedia Models decided that a consensus model for Pov and LRTP screening should be developed; this model is available from the OECD website as the OECD Pov and LRTP Screening Tool (http://tinyurl.com/66q47j). Comparison of the OECD Tool with the nine models investigated by Fenner et al. yields RCCs as high as those found for these nine models themselves . The OECD Tool has been used for the evaluation of the current POP candidate chemicals under the Stockholm Convention [72,73] (http://www.sust-chem.ethz.ch/downloads).
In conclusion, generic LRTP models are reliable and useful if they are employed within their application domain: for chemicals with a low Henry's law constant, the intermittent rain approach should be implemented. Depending on the region represented, appropriate OH radical concentrations and kOH values should be used, and the chemical properties may have to be adjusted to temperatures lower than 298 K.
Spatially resolved multimedia box models
Two spatially resolved multimedia box models developed in the 1990s are Globo-POP  and CliMoChem . These models are global models consisting of well-mixed latitudinal zones. The main purpose of the models is to investigate the south–north migration of chemicals, whereas transport in the east–west direction is assumed to be fast and cannot be resolved in these models. Transport of chemicals in the south–north direction is driven by large-scale eddy diffusion, which leads to exchange of air and water masses between the different latitudinal zones of the models. The eddy diffusion coefficients used in the models are based on empirical data for hemispheric and interhemispheric mixing of trace gases [75,76].
Wania and Mackay [77,78] discussed the pathways of different types of chemicals to the Arctic and the possibility that certain chemicals accumulate in polar regions, mainly because lower temperature leads to slower degradation and lower vapor pressure and Henry's law constant or, correspondingly, higher KOA values. This effect is called cold condensation, although it is not a condensation in the strict sense but, rather, a shift in partitioning equilibria from air toward surface media, which is more accurately reflected by the expression “cold-trap effect.” The cold-trap effect is most visible for volatile compounds, such as carbon tetrachloride or HCB, because these compounds have reached a nearly uniform global distribution in the atmosphere. Spatially constant concentrations in air in combination with stronger partitioning into condensed media in colder regions lead to an increase in chemical concentration in the condensed media along a transect from warmer to colder regions (Fig. 2, top). This has been observed in the field, for example, for CCl4 and chlorofluorocarbons  and for HCB [80,81]. For chemicals with low vapor pressure (e.g., PCB 180), the temperature dependence of the vapor pressure or KOA is even stronger than that of CCl4 and HCB (the enthalpy of vaporization, which determines the temperature dependence of the vapor pressure, is approximately 30, 60, and 95 kJ/mol for CCl4, HCB, and PCB 180, respectively). For chemicals with low vapor pressure, however, the main amount resides in the surface media, not in the air. For such a chemical, if it is very persistent, a nearly uniform distribution will, theoretically, be reached after very long times (Fig. 2, middle). In this case, the cold-trap effects lead to lower concentrations in air in colder regions (instead of higher concentrations in surface media), making the chemical less mobile. In reality, however, low-volatility chemicals are still far from having a uniform distribution, because their LRT is much slower than that of volatile compounds. The cold-trap effect is then masked by the gradient that is caused by incomplete transport away from the source region (Fig. 2, bottom).
Figure Fig. 2.. Conceptual illustration of the cold-trap effect: (A) persistent volatile compounds, such as CCl4; (B) persistent semivolatile compounds, such as polychlorinated biphenyl (PCB) 180 (scenario with uniform distribution in surface media as the long-term result of long-range transport [LRT]); and (C) persistent semivolatile compounds (scenario with inhomogeneous distribution; point in time before scenario B). Emission of chemicals is assumed to occur in the warm region.
Download figure to PowerPoint
In addition to the cold trap effect, which refers to the environmental distribution of a single chemical, a second question is how different the transport efficiencies of several chemicals in comparison are (e.g., in a suite of PCB homologues). To address this question, the term global fractionation has been introduced. Conceptually, global fractionation refers to different transport efficiencies of chemicals caused by different chemical properties, such as partition coefficients, degradation half-lives, and temperature dependencies of these properties. Ideally, the effect of different chemical properties would be analyzed in a homogeneous environment and with equal emissions of all chemicals at the same location and time. In reality, the environment is highly inhomogeneous—for example, in terms of precipitation, organic matter content of soils, and emissions, which are spatially and temporally diverse and differ from chemical to chemical. Therefore, concentration trends observed in the field are a complex superposition of the effects of release patterns, chemical properties, and varying environmental conditions. To reduce the effect of confounding factors, model investigations are useful, because environmental conditions and release patterns can be controlled in environmental fate models and the relationships between chemical properties and the efficiencies of LRT can be evaluated systematically [82,83].
One aspect of global fractionation that can be directly investigated in the field is how the relative composition of a mixture of chemicals changes as a function of latitude. For PCB concentrations measured in soils along a latitudinal transect from England to Norway, Meijer et al.  found an increase in the fraction of trichloro and tetrachloro PCBs, a constant fraction of pentachloro to hexachloro PCBs, and a decrease in the fraction of heptachloro and octachloro PCBs. Both the Globo-POP and CliMoChem models clearly reproduce this fractionation observed in PCB concentrations in soil [83,85]. Lighter PCBs, because of their higher vapor pressure and lower KOA, are more mobile than heavier PCBs and, therefore, are relatively enriched in remote regions. In the future, however, this picture may change for two reasons [82,83]: First, lighter PCBs are less persistent than heavier ones, so in the long-term, the fraction of lighter PCBs will decrease (whereas the heavier ones will form long-lasting reservoirs). Second, the low vapor pressure of heavier PCBs makes their volatilization and mobilization slow, but in the course of decades, substantial amounts will still be released and transported via air (whereas lighter PCBs are quick and reach the Arctic soon after their release, heavier PCBs can catch up in the long-term).
Figure Fig. 3.. Independent pieces of information needed for an environmental fate assessment of a chemical. All four components are uncertain, and the influence of the different uncertainties on the overall assessment needs to be evaluated.
Download figure to PowerPoint
Using the Globo-POP model and the ACP as the model endpoint, Wania  has investigated which combinations of chemical properties lead to high ACP values. Volatile chemicals (i.e., “flyers”) are transported quickly and in high amounts via the air (Fig. 2, top). “Swimmers” are chemicals with a low Henry's law constant and a persistence sufficiently high for transport by ocean currents (e.g., HCHs). “Multiple hoppers” are chemicals for which, after transport over a certain distance and deposition to the ground, the distribution between air and surface media is close to equilibrium. Higher temperatures in the summer then cause revolatilization of the chemical that was deposited earlier so that another “hop” of LRT can take place. Even in the case of these chemicals, however, substantial amounts reach remote regions in a single “hop” taking place within days or weeks after release . To what extent revolatilization after deposition (i.e., release from secondary sources) actually contributes to the transport to remote regions needs further investigation. “Single hop” chemicals, finally, have such a high KOA that the soil to which they are deposited after LRT remains undersaturated, and almost no revolatilization takes place. These chemicals have the slowest LRT, but in the long-term, substantial amounts of these chemicals also can undergo LRT.
Spatially revolved multimedia box models also have been used to investigate the large-scale distribution of several individual chemicals. These investigations comprise four main components (Fig. 3): An emission inventory defining the amounts of the chemical released to the environment as well as the spatial and temporal distribution of these releases, the chemical property data (i.e., partition coefficients and degradation half-lives), the model itself as well as the selection and parameterization of environmental processes used in the model; and the field data to which the model results can be compared.
Wania et al.  evaluated the Globo-POP model comparing model results and field data for α-HCH. On this basis, Wania and Mackay  investigated the global distribution and mass balance of α-HCH using the Globo-POP model, with a particular focus on the period after 1980, when the usage of technical HCH, which is the source of α-HCH to the environment, was strongly reduced in northern temperate countries. They found that in the northern polar region of the model, a long-lived reservoir of α-HCH is formed that accounts for more than 30% of the chemical's global environmental inventory after 1992. The burden of α-HCH in the northern temperate region, in contrast, strongly decreased during the same period. Similarly, a shift has occurred from a large reservoir of α-HCH in cultivated soils before 1984 (>40% of the global inventory) to a more persistent reservoir in surface ocean water (>50% of the global inventory after 1992). This shift toward more long-lived reservoirs in colder regions also is reflected by the evolution of the overall persistence of α-HCH over time. Wania and Mackay calculated the overall persistence of α-HCH as the global amount present in the model, Mtot, divided by global degradation mass flux, Fdeg:Pov = Mtot/Fdeg. Both components, Mtot and Fdeg, are a function of time in this calculation, and Mtot decreases from approximately 100 kt in 1985 to less than 10 kt in 1997. In parallel, Pov increases from approximately eight months to approximately 1.5 years.
Using the CliMoChem model, Schenker et al.  compared model results for DDT with measured DDT concentrations and estimated future DDT levels caused by DDT usage against malaria. Similar to the results of Wania and Mackay  for α-HCH, Schenker et al.  found long-lived residues of DDT in Arctic soils in their model.
Wania  and Schenker et al.  also used the two zonally averaged models to investigate the hypothesis that levels of perfluorooctanoic acid (PFOA) observed in Arctic snow are caused by atmospheric deposition of PFOA formed by degradation of volatile precursors . Schenker et al.  investigated airborne distribution, deposition, and transformation to PFOA of fluorotelomer alcohols and perfluorooctane sulfonamidoethanols. The model results are within a factor of two in agreement with PFOA flux estimates derived from PFOA measurements in ice from Greenland . This indicates that transport of both types of volatile precursors to the Arctic and subsequent transformation into PFOA is a possible explanation for PFOA levels observed in Arctic surface media, such as inland ice. For the overall mass flux of PFOA to the Arctic, transport of PFOA itself in ocean water also is relevant. The importance of these two pathways, PFOA in ocean water and PFOA from volatile precursors in air, is the subject of a controversial scientific debate. In absolute terms, the amount of PFOA entering the Arctic via volatile precursors is only a few percent of the amount transported via ocean currents . For inland ecosystems and, perhaps, also oceanic surface water, however, deposition of PFOA formed out of volatile precursors probably is still a relevant source of exposure to PFOA.
The problem that Kp values derived from the KOA are not appropriate for all types of aerosols has been addressed by Götz et al. . They used poly-parameter linear free-energy relationships (pp-LFERs) derived by Goss and Schwarzenbach  and by Roth et al.  to describe sorption of SOCs to the different components of a fine and a coarse aerosol fraction (organic matter, elemental carbon, sea salt, and silica). On this basis, regionally different concentration and composition of atmospheric aerosols [96,97] can be taken into account in spatially resolved environmental fate models. Götz et al.  implemented this approach in the CliMoChem model, and using the LRT of three pesticides (α-HCH, terbuthylazine, and trifluraline) to the Arctic as an example, they compared it to the approach using Kp values derived from the KOA. They found better agreement between model results and pesticide concentrations measured in Arctic air  if the pp-LFER description of aerosol–air partitioning is used: Discrepancies are a factor of two to eight (pp-LFER approach) and a factor of 10 to 50 (KOA approach). Because aerosol–air partitioning is a key factor determining the LRT of SOCs, regional differences in aerosol concentration and composition should be taken into account in spatially resolved environmental fate models.
In addition, PBDEs have been investigated with several environmental fate models. The LRT of PBDEs is not yet well understood. Wania and Dugani  used four generic LRTP models to estimate the transport distances of nine PBDE congeners (from monobromo to decabromo). In these calculations, constant environmental conditions were assumed, as is typical of generic LRTP models. Breivik et al. , in contrast, used the CoZMo-POP model with variable environmental conditions (continuous vs intermittent rain and presence vs absence of vegetation) to estimate CTDs of BDEs 47, 99, and 209. In addition, they investigated measured PBDE concentrations in sediments from lakes between 43°N and 80°N. Particularly difficult to understand is the LRT of deca-BDE (i.e., BDE 209), which sorbs strongly to aerosol particles because of its very high KOA and undergoes only slow reaction with OH radicals but fast direct photolysis. Whereas Wania and Dugani  reported relatively low LRTP of BDE 209 because of its fast deposition with aerosol particles, Breivik et al.  found in some model scenarios considerably higher CTDs of BDE 209 as well as an ETD of approximately 500 km, which is only by a factor of 1.5 smaller than ETDs derived from the lake sediment data for PCBs. Neither Wania and Dugani  nor Breivik et al. , however, included direct photolysis of PBDEs, which is an important degradation process in air. Raff and Hites  determined photolytic lifetimes of PBDEs from mono- to deca-BDE and investigated the influence of photolytic degradation, degradation by reaction with OH, and deposition with particles on the atmospheric lifetime of PBDEs. For BDE 209, they assumed a fraction of 99.999% in the particle-bound phase in combination with the assumption that photolysis of BDE 209 occurs only in the gas phase. Under these conditions, deposition with aerosol particles is still the most important removal mechanism of BDE 209 from the atmosphere. If, however, only a slow but nonzero photolytic reaction of particle-bound BDE 209 is assumed, then photolysis is the main removal mechanism of BDE 209 in the atmosphere in temperate and tropical regions . Under this assumption, the LRT and deposition mass fluxes of BDE 209 in remote regions are considerably smaller than those in the scenario without photolysis of the particle-bound fraction of BDE 209. The case of BDE 209 clearly demonstrates how important an improved understanding of the reactivity of particle-bound SOCs is.
A global multimedia box model with higher spatial resolution is BETR-Global . In contrast to Globo-POP and CliMoChem, BETR-Global has spatial resolution in both the latitudinal and the longitudinal direction. The model has been designed to bridge the gap between multicompartment models with low spatial resolution, on the one hand, and highly resolved general circulation models, on the other. It consists of 288 grid cells with a size of 15° × 15°. Air flow between the grid cells is based on global reanalysis data from the U.S. National Centers of Environmental Prediction .
The BETR-Global model has been used mainly to investigate the global distribution of PCBs. Using the PCB emission inventory by Breivik et al. , MacLeod et al.  demonstrated that PCB concentrations in air obtained with the model are in good agreement with PCB concentrations measured at 11 long-term monitoring sites of the Integrated Atmospheric Deposition Network around the Laurentian Great Lakes (USA and Canada) and of the European Monitoring and Evaluation Programme (agreement within a factor of 3.2 for 60% of the model-vs-field concentration pairs). On this basis, MacLeod et al. used the model to investigate the influence of changes in atmospheric conditions caused by climate change on the distribution and levels of PCBs. They related changes in PCB concentrations in several regions of the model to changes in the North Atlantic Oscillation (NAO) index. The NAO index is a climate indicator that describes conditions in the atmospheric pressure field between Iceland and the Azores. MacLeod et al. found that the maximum variability in PCB concentrations in air related to NAO variability is a factor of two. More pronounced effects of changing atmospheric conditions on the distribution and levels of SOCs may occur in the future as a consequence of ongoing climate change. The relationship between climate change and pollutant dynamics is receiving increasing attention and is being investigated on various scales from regional to global . In general, the BETR-Global model can be used to investigate source–receptor relationships between regions represented by different grid cells (see next section).
Atmospheric transport models
General circulation models of the atmosphere have increasingly been used to investigate the large-scale distribution of SOCs. These models provide much higher spatial and temporal resolution than multimedia box models, and they can be used to determine source–receptor relationships between different regions and to identify trajectories of chemical transport during certain transport events [106–109].
Semeena and Lammel  as well as Leip and Lammel  used the atmospheric transport model ECHAM coupled to reservoirs representing the surface media, soil, vegetation, and ocean water. They defined LRT metrics  and investigated the environmental distribution dynamics of γ-HCH and DDT [86,107]. Using model versions with and without revolatilization from surface media, they compared multiple- and single-hop transport of the two insecticides to remote regions, including the Arctic. According to their model results, both types of transport contribute to similar extents to the amounts of γ-HCH and DDT in the Arctic. γ-Hexachlorocyclohexane exhibits a slightly higher contribution of multiple-hop transport, because it has a lower Henry's law constant than DDT (favoring wet deposition) and is more volatile than DDT (favoring revolatilization after deposition).
The global distribution of γ-HCH also was investigated by Koziol and Pudykiewicz  with an atmospheric transport model to which modules for soil, ocean water, and also ice and snow are coupled . The model was evaluated by comparison of model results to measured concentrations in the Arctic and in southern Quebec (Canada), with deviations of a factor of five or less. Annual concentration averages show better agreement than short-term concentration patterns. For India as a source region, transport of γ-HCH to North America along westerlies and to Africa and Latin America along easterlies was demonstrated.
The POPs model of the Meteorological Synthesizing Centre–East (MSEC-POP) has been used to calculate the distribution of four selected PCB congeners in the northern hemisphere for the year 1996 . The discrepancies between calculated PCB concentrations in air and measurements from several locations from midlatitudes to the Arctic were around a factor of four. The model results show, for example, the contribution of different continental regions (Africa, America, different parts of Asia, different parts of Europe, and Russia) to PCB deposition mass fluxes in the Arctic. For PCBs 28 and 118, transport from Russia to the Arctic is higher than Russia's share in the emissions of these two congeners; for PCBs 153 and 180, transport from northwestern Europe to the Arctic is higher than the share of this region in the emission of PCBs 153 and 180. Holoubek et al.  compared results from the MSCE-POP model to concentrations of PCB 153, benzo[a]pyrene, and γ-HCH measured at Kosetice (Czech Republic) and found deviations between model results and measured concentrations of approximately a factor of five and below .
Hansen et al.  developed a version of the Danish Eulerian Hemispheric Model capable of describing the distribution of SOCs (DEHM-POP) . As a case study, they investigated α-HCH; similar to Koziol and Pudykiewicz , they found good agreement between model results and measured concentrations of α-HCH in air for annual concentration averages but less agreement for short-term concentration trends. Hansen et al.  further discussed the effect of dynamic snow–air exchange of chemicals on the concentration profiles of α-HCH obtained with the DEHM-POP model [112–114]. The snowpack acts as a highly dynamic reservoir receiving chemical with snowfall and releasing it by evaporization. Inclusion of the snowpack tends to improve the agreement between model results and field data, but some measurement sites still show pronounced discrepancies between model results and α-HCH concentrations measured in air.
Hansen et al.  compared the DEHM-POP model with the less highly resolved EVn-BETR model, which is a multimedia box model based on BETR North America [116,117], and found good agreement in many respects. A characteristic difference is that advective transport of airborne chemical out of an open model domain is more pronounced in the atmospheric transport model, because dynamic air flow data are used in the atmospheric transport model, whereas in the multimedia box model, long-term averages of air flows out of and into the model domain are used, which leads to a dampened atmospheric dynamics in the box model.
Atmospheric transport models adjusted for the treatment of SOCs are intended to provide the most accurate description of the environmental distribution dynamics of SOCs. Certainly, with these models valuable insights can be gained that go beyond the results from multimedia box models. The high resolution of these models, however, requires a correspondingly high resolution of emission data, descriptions of environmental media and processes other than those in air, and of chemical property data dependent on environmental conditions. In most cases, these components are not available with the same resolution as that used in the description of the atmospheric dynamics. As in multimedia box models, the overall uncertainty of the model results cannot be smaller than the uncertainty of the most uncertain component shown in Figure 3. In many cases, these components are emission data and/or chemical property data.