Polychlorinated biphenyls (PCBs) are ubiquitous contaminants in the global biogeosphere. Based on extrapolation of reaction rates determined at temperatures in laboratory experiments that were 60–100 degrees above global average tropospheric temperatures, it has been suggested that the gas-phase reaction with the hydroxy radical (HO·) in the troposphere constitutes the strongly dominating sink for PCBs. An assessment of a broad set of field observations relevant to the actual environmental behavior of individual PCB congeners suggests that such an assertion should be treated with some caution. First, if the proposed reaction rates were applied to published tropospheric contents of individual PCB congeners, a continuous order-of-magnitude depletion of the lighter congeners relative to more chlorinated ones would be predicted. However, a large shift in PCB fingerprint is inconsistent with congener-specific sediment archives, where a disproportionate historical depletion of lighter PCBs cannot be found. Alternatively, such HO·-reaction rates would require a continuous source flux of PCBs to today's environment grossly enriched (again by several orders-of-magnitude) in lighter PCB congeners to counterbalance the implied selective congener losses; a scenario that is not supported by available information on emission rates. Further, the existence of a latitudinal trend in tropospheric concentrations, where the highest concentrations are found at lower latitudes, suggests that the fate of PCBs is controlled by temperature-driven partitioning processes rather than a HO·-reaction sink, which would be expected to be more efficient at lower latitudes. Finally, the observed tropospheric variability of individual PCB concentrations conform poorly with both the absolute values and congeneric trends in predicted reaction lifetimes. A reconciliation of these apparent discrepancies is attempted.
1.1. The Need for a Global Perspective to Assess Environmental Residence Time of Polychlorinated Biphenyls
 A quantitative understanding of the integrated interactions of persistent organic pollutants (POPs) with large-scale biogeochemical processes is required for predicting their global removal rates and thus environmental longevity. Due to their persistence, high ecotoxicity, and well-known span of physico-chemical properties, the notorious polychlorinated biphenyls (PCBs) serve well as POP model compounds. Persistent contaminants of semivolatile and hydrophobic nature such as the PCBs are found throughout the global biogeosphere. This environmental pool of PCBs is dynamically recycling and partitioning between the major environmental media in response to the physico-chemical properties of the individual compounds (Figure 1). Because of their large organic-matter inventories, sediments and soils would be expected to constitute the largest environmental reservoirs of POPs due to the generally low vapor pressure and water solubility of the substances [Harrad et al., 1994; Axelman and Broman, 2001]. Hence, because PCBs are continuously seeking to equilibrate between the different reservoirs, a permanent loss process of PCBs occurring in any one reservoir will be reflected also in the PCB inventory of the others. While PCBs are dynamically partitioning between the atmosphere-ocean-sediment/soil reservoirs, the net or permanent environmental sinks of PCBs are (1) burial to deeper layers of soils and sediments (2) reaction with hydroxy radical in the troposphere, and (3) export to deep seawater masses (removal on 100-year timescale) (Figure 1). Progress in elucidating the rate of these global environmental sink processes holds the key to estimate the “global cleaning time” and thus developing the capacity to predict at what rates the concentration of these compounds will decrease in various ecosystems and regions.
 Historically, there have been many reports on the occurrence of organic pollutants at various remote locations, leading to qualitative conclusions regarding their propensity for global dispersal. Such monitoring information is of limited use for describing the global fate of anthropogenic chemicals. Therefore, a plea has been made to emphasize process-oriented studies on molecular-level interactions and macroscopic transport phenomena [Blumer, 1975; Stumm et al., 1983; Farrington and Westall, 1986; Gschwend and Schwarzenbach, 1992]. A requirement for developing accurate global-scale mass balance models for POPs is to gain comprehension of the most important processes and the ability to correctly extrapolate molecular-scale processes to a macroscopic level. Once sufficient understanding of anticipated key processes enables prediction of a chemical's large-scale distribution and fate, it is of utmost importance that field measurements of its actual distribution and transport/transformation rates be performed in order to verify, and refine, the model.
 Recent large-scale mass balance models, incorporating many processes, have indicated our present inability to account for the majority of the emitted POPs [e.g., Simonich and Hites, 1994; Axelman and Broman, 2001]. Alternative multimedia approaches are attempting to model the global distribution of POPs by assuming a suite of rates and properties of the system [e.g., Wania and Mackay, 1995, 1996]. However, while the output from these fugacity models are frequently presented with a high resolution and apparent predictive capacity, they are seldom based on, or verified by, actual field data. Nevertheless, a common insight gained in all these investigations is the dire lack of information on the magnitude of permanent environmental sinks such as deep-ocean export, burial in shelf sediments, and tropospheric reaction. Some limited progress is being made on estimating the ocean export sink of POPs [e.g., Schulz et al., 1988; Schulz-Bull et al., 1998; Gustafsson et al., 1997a, 1997b]. Atkinson and Hites with coworkers have put forth a major contribution by demonstrating that the tropospheric vapor-phase reaction between several POP compound classes and the hydroxy radical (HO·) in the troposphere may represent a globally significant sink [Atkinson, 1987; Anderson and Hites, 1996a, 1996b; Brubaker and Hites, 1997, 1998a]. Based on extrapolations of laboratory results to the global environment, it has been proposed that the reaction with HO· in the troposphere is, by far, the most dominant sink of PCBs [Anderson and Hites, 1996a]. If so, this process could be anticipated to leave decipherable imprints on, for instance, their tropospheric longevities, the PCB distributions, and congener profiles in any environmental matrix. Given the wide acceptance of this process as a key global sink for PCBs, and, by inference, for other important POPs, it behooves us to test this assertion by investigating the signals of the actual behavior of PCBs in the environment. To this end, several approaches are employed here to test the HO· sink hypothesis against environmental data. The analysis, based on congener-specific mass balance considerations in combination with temporal and spatial distribution patterns, suggests that the HO· reaction may be significantly overestimated for several PCBs and, in effect, not the dominant environmental sink for others. Important gaps in our understanding of the global fate of PCBs still do exist, potentially on the orders-of-magnitude scale.
 The standard PCB nomenclature that is used throughout this paper is (1) number of chlorine substitutions on the biphenyl ring (e.g., trichlorobiphenyl is denoted 3CB) and (2) the IUPAC numbering scheme of the 209 individual PCB congeners (e.g., PCB 28).
1.2. HO· Reaction as an Environmental Sink
 The hydroxy radical is an important reagent in ambient air, governing the atmospheric fate of many trace chemicals. It is highly reactive and is formed through several different pathways, all of which ultimately originate in actinic reactions and depend on the presence of water vapor. Its concentration is, therefore, highly variable in the atmosphere and strongly correlated with the intensity of sunlight, with nighttime concentrations orders-of-magnitude lower than those observed during the day. The HO· is highly reactive toward aromatic compounds due to their high electron density. However, the relatively low vapor pressure of compounds such as PCBs complicates investigations of this reaction under realistic conditions. More than 15 years ago, Atkinson and Aschmann  estimated reaction tropospheric lifetimes for biphenyl (2.7 days) and monochlorobiphenyl (1CB) (4–8 days) based on laboratory studies of the HO· reaction rates at 22°C. Extrapolation methods then extended their lifetime estimates up to 5CBs [Atkinson, 1987], for which the estimated ambient tropospheric lifetime range was 60–120 days. More recently, Anderson and Hites [1996a] made measurements of the reaction rates at elevated temperatures (+50–+90°C) and extrapolated their results to ambient tropospheric temperatures (−48–+13°C). A reaction pathway for the HO·-mediated degradation of PCBs has been suggested by Brubaker and Hites [1998b]. The HO· is suggested to initially attack the ipso position of the biphenyl followed by a series of reactions, involving O2 and O3, of which benzoic acid has been identified as the major stable product. None of the suggested HO· reaction products have so far been analyzed in environmental air samples.
Anderson and Hites [1996a] relied on a set of simplifying assumptions to arrive at a global tropospheric loss rate of a bulked ΣPCB parameter from their laboratory-extrapolated reaction rates. A bulk reaction rate for ΣPCB was based on an average 3.5 chlorine biphenyl. To estimate the global tropospheric load of ΣPCB, they used an assumed average surface concentration of 0.2 ng m−3, which is approximately 50% of the average concentration of ΣPCB, measured at Bermuda in 1993 [Panshin and Hites, 1994]. They also assumed that ΣPCB exhibits a vertical gradient. This is to be expected if the reaction lifetimes are comparable to, or shorter than, vertical tropospheric mixing times. However, in such a case, one would expect a steeper gradient for the more reactive low-chlorinated congeners than for the less reactive highly chlorinated congeners. Based on their estimated global loss of ΣPCB of 8271 t yr−1, Anderson and Hites [1996a] concluded that tropospheric HO· reaction is, by far, the dominant sink of PCBs in the global environment, swamping a single ocean sink estimate taken from Duce et al.  of 240 t yr−1 quoted as being to “deep ocean sediments” [Anderson and Hites, 1996a]. In reality, the estimate by Duce et al.  concerned atmospheric input to the surface ocean and was relying on the product of a limited data set of tropospheric concentrations from the 1970s and 1980s and a number of assumptions concerning washout ratios, gas/particle distribution, Henry's law constants and seawater concentrations. For instance, a potentially significant underestimate was the gas-exchange term, which was assumed to correspond to 10% of the maximum flux (i.e., that oceanic fugacity of PCBs corresponded to 90% of that of the atmospheric boundary layer). Furthermore, Duce et al.  had stressed the limitations of their single number, considering there to be lots of “caveats, assumptions, and uncertainties in our estimate”. More recent direct estimates of net export from the surface ocean to deeper layers of individual PCB congeners suggest a ΣPCB (assuming that PCB 52 is 5%) net ocean flux of 100 t yr−1 in 1993 to just a limited sector of the NW Atlantic Ocean, corresponding to 4% of the earth's total ocean area [Gustafsson et al., 1997b]. It is a major objective of the following section to use available diagnostic field signals and mass balance constraints to evaluate the environmental consistency and reasonableness of the postulation that rapid vapor-phase reactions are dominating the global PCB fate.
2. Assessment of Environmental Consistency of a Large HO· Sink
2.1. Postulated Magnitude of HO· Sink in Relation to Magnitude of Emissions and Recyclable Environmental Pool
 Individual budgets for four PCB congeners in the Northern Hemisphere was recently used to contrast the magnitude of a set of loss processes for congeners spanning a range of physico-chemical properties [Axelman and Broman, 2001]. Background concentrations of PCBs in soils, sediments, water and air were assessed in conjunction with the following removal fluxes: burial in shelf sediments, sedimentation to the deep ocean, tropospheric degradation and burial on land. In the budget calculations of Axelman and Broman , the tropospheric sink was based on the reaction rates presented by Anderson and Hites [1996a]. Axelman and Broman  selected a one-box approach and assumed that a pool of PCBs exists, which recycles between different environmental media on a decadal timescale or faster. A dynamic intermedia exchange is supported by field studies of the air-water and air-soil interactions [e. g., Hornbuckle et al., 1994; Cousins et al., 1999; Hippelein and McLachlan, 1998]. Most PCBs appear to be found in remote reservoirs to which they must have arrived largely through atmospheric transport [Axelman and Broman, 2001]. The analysis in sections 2.1 and 2.2 (below) uses the concept of the one-box recyclable pool (Figure 1). Hence, even with existing fugacity gradients between recyclable compartments, the annual relative loss rate (kglobal [yr−1]) from the total environmental pool of individual congeners can be estimated by summing the sinks in the individual reservoirs:
where FHO· is the hydroxy radical sink flux (t yr−1), Focean is the settling flux to the deep-ocean (t yr−1), Fsediment is the burial flux in continental shelf sediments (t yr−1), Fsoil is the burial flux in soils (t yr−1), and mtotal is the sum of the inventories in the atmosphere, the surface ocean, the continental shelf surface sediments, and the surface soils. The sum of all sink fluxes in equation (1), here denoted Fsink, contrasted to estimated global emission source fluxes (Femission) gives a predicted environmental rate of decline (kpred [yr−1]):
This predicted rate of decline is possible to test against observed concentration time trends in field matrices such as dated sediment cores and biota used in long-term monitoring programs.
 The Northern Hemisphere budget estimated for PCB 28 (3CB) indicated that FHO· constituted as much as 99% of Fsink, or 2 500 t yr−1 [Axelman and Broman, 2001]. For PCB 28 kglobal was 0.58 yr−1, in strong contrast to PCB 153 (6CB) for which kglobal was only 0.02 yr−1. For the HO· reaction rates to be correct and applicable, this magnitude, and the implied stark intercongener gradient in loss rate, has to mean that either (1) the congeneric composition of the PCBs found in the environment has changed in a correspondingly dramatic fashion, or (2) that Femission has been balancing the sink, both in terms of absolute numbers and in terms of congeneric differences. Axelman and Broman  used available emission estimates by Harrad et al.  and Berdowski et al.  to obtain emission fluxes to the Northern Hemisphere. These figures suggested that Femission was 21 t yr−1 for PCB 28, which is a factor 100 lower than what is needed to balance the postulated large HO· sink term for this congener. Even for PCB 153 Femission is a factor 100 lower than FHO·. A recent, more comprehensive, analysis of the historical production and emissions of individual PCBs by Breivik et al. [2002a, 2002b] gave an estimate of the current global emission rate of 2.0 t yr−1 for PCB 28, of which more than 90% would occur in the Northern Hemisphere. Considering the maximum limit of the large uncertainty range (0.04–55 t yr−1) provided by Breivik and coworkers, also this analysis indicates that current emissions are far from balancing a FHO· of 2 500 t yr−1 for PCB 28. Based on available estimates of current emission fluxes it therefore appears unlikely that continuous emissions can balance the suggested large reaction term.
 The environmental concentrations after the major bans in the 1970s should thereby also be largely controlled by permanent environmental loss processes. Taking conservatively, the maximum limit of the cumulative historic emissions of 11 700 t for PCB 28 [Breivik et al., 2002b], the suggested FHO· for PCB 28 would still in only 5 years consume all of the PCB 28 historically released to the global environment. The maximum cumulative emission estimate by Breivik and coworkers turns out to be in accordance with the 14 000 t for PCB 28 suggested by Axelman and Broman , which was recalculated from a figure originally suggested by the National Academy of Sciences . Such a rapid decline of PCB 28 in the environment is inconsistent with observations of concentration time-trends for PCBs in different environmental media, which instead indicate that rates of decline are in the range 0–0.2 yr−1 depending on the position of the sampling site [e.g., Panshin and Hites, 1994a; Gregor et al., 1995; Bignert et al., 1998; Simcik et al., 1999]. Furthermore, a HO· sink of this magnitude would have to mean that the environmental concentration of PCB 28 relative to PCB 153 must have decreased dramatically since the ban of PCBs in the 1970s.
2.2. Historical Records of Congener Composition
 A dominant sink of lighter congeners occurring in any one medium would affect the congener patterns in all media, irrespective of the differences in fingerprints found in different media resulting from the thermodynamically driven partitioning. The large difference between the estimated [Axelman and Broman, 2001] environmental removal rates of PCBs 28 and 153 (due to FHO·) would result in a 200 times larger depletion of PCB 28 than PCB 153 over a time period of 10 years. Such a large difference should be detectable in any environmental media allowing time-trend studies. Peat bogs and accumulation areas of marine and lake sediments constitute historical records of congener composition. For instance, data on the relative abundance of individual PCB congeners in a lake sediment [Sanders et al., 1992] and in a peat core [Sanders et al., 1995] showed no trend of decreasing relative amounts of less chlorinated congeners (Figure 2). Similarly, the congeneric composition in Baltic Sea sediments was found to be relatively constant in sediment layers from the period 1920–1990 [Axelman et al., 1995]. Oliver et al.  found in a number of locations that the more recent sediment layers of Lake Erie actually contained a higher proportion of lower-chlorinated congeners. An indication of a systematic relative depletion of less chlorinated congeners over time could neither be found in a study of 11 Canadian lakes [Muir et al., 1996]. Historical records thus give no support for a rapid decline in lighter congeners relative to heavier congeners. In fact, the available data on the evolution of the congeneric composition suggest that it has not changed significantly over the last decades.
2.3. Spatial Trends
 So far, we have considered residence time, sources and sinks of the entire recyclable pool of PCBs in the environment (Figure 1). However, it is also informative to consider residence times, sources and sinks in subcompartments of this recyclable pool. Specifically, important information regarding the magnitude of the HO· sink can be deduced from the spatial distribution of PCBs in the troposphere and how compliant this is with the HO· concentration field and other tropospheric sinks such as deposition (note the distinction from permanent total environmental sinks in this case). Model calculations suggest that the ratio between the average tropospheric concentrations of HO· at 30°N and at 60°N is approximately 20 in January and 2 in July [Spivakovsky et al., 2000]. This means that if reaction with HO· were to be the dominant atmospheric sink of PCBs, this sink would be concentrated to lower latitudes leading to relatively higher concentrations in higher latitudes. Hence, if the reaction with the HO· was the dominating tropospheric sink a trend of increasing concentration with latitude should be observable. Axelman and Broman  demonstrated that the opposite was the case; a significant trend of decreasing tropospheric concentrations of PCB 28, PCB 52 and PCB 153 with increasing latitude, which is exemplified by PCB 28 in Figure 3. The regression, which was based on average concentrations at 16 stations (in total more than 300 individual measurements), predicted that the concentration of PCB 28 was about 10 times higher at 40°N than at 80°N. Analogical support for this observation can be found in recent reports of direct measurements of CH3CCl3, for which the dominating tropospheric sink is reaction with HO·. Since the anthropogenic sources of this chemical have been greatly reduced it is assumed that its tropospheric distribution presently is controlled by the tropospheric removal processes rather than the source distribution [Montzka et al., 2000]. A global network of sampling stations shows that the concentrations of CH3CCl3, in stark contrast to the trend for PCBs, are consistently lower in lower latitudes [Montzka et al., 2000]. Hence, the inverse relationship between latitude and tropospheric concentration for PCBs is incompatible with the HO· reaction being its dominating tropospheric sink. The latitudinal trend in tropospheric concentrations suggests the presence of a sink, the magnitude of which is inversely related to temperature, controlling the removal of PCBs from the troposphere. A net temperature-driven partitioning to aerosols and land/ocean surfaces would lead to higher deposition rates at higher latitudes.
 The observed negative latitudinal trend could theoretically coexist with a large HO· sink under either of two different conditions: (1) if an emission flux, positively correlated with temperature, would counterbalance the higher HO· loss of PCBs in lower latitudes, or (2) if deposition rates (mass per area and time) would be significantly higher than the HO· reaction rates (mass per area and time). The first condition can be discounted based on the finding above that ample literature data suggest that current emissions rates are orders of magnitude too low to counterbalance the suggested large HO·-sink term. In the second case, the temperature-dependent deposition process would control the atmospheric fate and give rise to a negative dependence of atmospheric PCB concentration with respect to latitude. This condition was therefore tested for PCB 28 with a number of available literature data on deposition fluxes. The area-specific average HO· loss of PCB 28 in the Northern Hemisphere, derived from the total hemispheric HO·-reaction loss given by Axelman and Broman  is 16 ng m−2 d−1. Reported wet and dry particle deposition fluxes for PCB 28 are in the range 0.18–0.7 ng m−2 d−1 [Franz and Eisenreich, 1993; Brorström-Lundén et al., 1994; Brorström-Lundén, 1996]. Similarly, typical short-term gas-deposition fluxes are on the order of 3 ng m−2 d−1 for the entire homologue group 3CBs [Hornbuckle et al., 1994]. Hence, for PCB 28 the postulated area-average HO· loss are much larger than the observed deposition fluxes. The implication is that the postulated large atmospheric HO·-reaction sink for PCB 28 indeed is inconsistent with the observation of a negative correlation between atmospheric concentration and latitude.
2.4. Tropospheric Residence Times
Junge  postulated on the basis of theoretical considerations that there should be an inverse relationship between tropospheric residence time and variability in concentration of tropospheric trace gases. Bringing together field data for a large set of different trace gases, he found an empirical relationship between total tropospheric lifetime (τtotal; year) and standard spatial variation in concentration:
where RSDtotal is the total variability observed in the concentration data expressed as one relative standard deviation. Although the theoretical framework developed by Junge has been debated [Hamrud, 1983; Slinn, 1988] the empirical relationship is remarkably consistent over 7 orders of magnitude. It should be recognized that the inverse total tropospheric lifetime is the sum of the inverse lifetimes caused by a number of processes occurring in the troposphere. In a simplified manner, the overall tropospheric residence time (τtotal) may therefore be described as:
where τreact is the residence time due to chemical reactions, τwet is the residence time due to wet deposition and τdry is the residence time due to dry deposition. Each of these three processes consists, in turn, of several different processes. For instance, the dry deposition consists of both gas adsorption and deposition of the particle-associated species. If τHO·, here believed to equal τreact, is significantly longer than τtotal, some other, and faster, loss process must control τtotal. For instance, an increasing number of chlorine substitutions (nCl) yields lower vapor pressure and hence deposition rate can be expected to increase. However, if τHO·, or the estimated residence time of any of the other loss processes, is shorter than τtotal it means that at least one of the two estimates is wrong.
 We have calculated τtotal (equation (3)) for homologue groups from reported tropospheric variabilities at Bermuda (U.K., western North Atlantic, remote marine [Panshin and Hites, 1994a]), Bloomington (Indiana, USA, suburban [Panshin and Hites, 1994b]), and Svalbard (Norway, remote marine, Eurasian Arctic [Oehme et al., 1996]). Remarkably good consistency was found between the estimates of τtotal from these widely distributed and diverse locations (Figure 4). While no large weight is placed on the absolute numbers of Junge τtotal estimates, the congeneric trends in τtotal is diagnostic. The Bermuda and Bloomington sites exhibit a significant negatively sloped linear regression between τtotal and nCl (p < 0.01, based on all congeners). Intriguingly, the independently predicted τHO· for 3–5CBs, and in particular for the 3CBs, is significantly shorter than, and thus inconsistent with, the total residence times, observed at all three sites. However, for the more highly chlorinated congeners (6CB to 9CB) the data is feasible in this respect. For those PCBs, the higher values for τHO· than for τtotal suggests that some process, or several processes, other than the HO· reaction plays an equally or more important role in the control of τtotal. In fact, the overall negative dependence of τtotal on nCl for the entire chlorination range, at these widely different locations, indicates that the process(es) exerting the dominant control on τtotal is more dependent on hydrophobicity (log Kow), or the inverse vapor pressure (po), than on molecular properties favoring the HO· sink. This evaluation of tropospheric residence times as a function of nCl suggests that net scavenging/deposition processes may be the dominant tropospheric fate for PCBs due to an increased partitioning tendency toward organic matter in aerosols [e.g., Goss and Schwarzenbach, 1998], toward aerosol surfaces [e.g., Pankow, 1987] and land surfaces [Hippelein and McLachlan, 1998] as well as toward raindrops and surface waters [e.g., Mackay et al., 1979].
Anderson and Hites [1996a] did contrast their calculated tropospheric τHO· for PCBs with the τtotal calculated with Junge's method by Panshin and Hites [1994a] from tropospheric concentrations measured at Bermuda. They held that the range 5–30 days for τHO· did not differ considerably from the 40–70 days for τtotal observed at Bermuda. As a basis for this claim they invoked uncertainty in the Jungeian τtotal, presumably stemming from uncertainties in the source and sink patterns of PCBs. However, Anderson and Hites [1996a] did not present a closer analysis of the rationale behind this claim.
Hamrud  investigated the effect of a few different source and sink patterns on the atmospheric variability of hypothetical gases by using a model coupled to a general tropospheric circulation model. From tropospheric residence times, Hamrud's model predicted the expected tropospheric variability of the hypothetical gases, throughout the entire troposphere and within the lowest 1 km of the troposphere. To test the reasonableness of some of the assumptions made by Anderson and Hites [1996a] we compared Hamrud's modeled relationship between residence time and observed tropospheric variability with the proposed reaction lifetimes and observed atmospheric variability of PCBs. We chose a model scenario with an anthropogenic source pattern (other alternatives were rain forest source or ocean source) and a uniform sink (which would be the most applicable case for chemical degradation within the troposphere) [Hamrud, 1983].
 We sought to address the question: Are the estimated τHO· for PCBs consistent with the observed tropospheric variability? To this end, the τHO· of 3–7CBs, along with their field-observed variability in tropospheric concentrations, were added to a Junge plot and compared with model calculations and observations for other compounds with similar residence times (Figure 5). The PCBs fall within the range of relationships observed for other hydrocarbons (acetylene, benzene and a series of C2–C4 alkanes), [Jobson et al., 1998]). The absolute deviation from the Junge relationship appears to be similar to findings for these volatile hydrocarbons. However, in contrast to the trend for the hydrocarbons, there appears for the PCBs to be a lack of dependence between tropospheric variability and lifetime as would be anticipated from the span of HO· reaction rates. Jobson et al.  found this to be the case for volatile hydrocarbons only in extreme situations where samples were taken in the direct vicinity of known sources. A conversion to express the variability in PCB concentration as standard deviation on a log-transformed data set (σlnx) instead of as RSDtotal was not possible for the Bermuda and Bloomington data sets since the published data was given as average values only [Panshin and Hites, 1994a, 1994b]. However, given the consistency between all three sites with regard to RSDtotal (Figure 4) we believe that the Bermuda data [Panshin and Hites, 1994a], used in Figure 5 indicate that there is not a trend of increasing tropospheric variability despite the presumed increase in reaction rates.
 As pointed out by Junge , the basis for the calculation of the tropospheric residence time according to equation 3 is strictly the variation in tropospheric concentration. In the case of the PCBs, a greater geographical coverage than currently available would obviously be useful. Another tangible consideration is the contribution to the observed variation that is caused by sampling and chemical analysis. This means that the true variation in tropospheric concentration is some value less than the measured/obtained value. To elucidate the variance components properly, an experimental design allowing analysis of variance is needed. A crude analysis of the impact of analytical variability can be done by comparing the standard deviation of the apparent tropospheric variation, discussed to this point, with the standard deviation of sampling and analysis. A simple illustrative approach to estimate the contribution of these two variance components to the overall “true total” variability could be:
where accounting for the fraction of total variance measured that is contributed by analytical variability (“τsampling + analysis”; within quotations because obviously not a true time term) could result in a refined τtrue − total−1 parameter. This deduced entity would thus represent the variability caused exclusively by processes in the atmosphere. Unfortunately, we could not find a good estimate of the variation caused by the high-volume sampling method normally used to measure tropospheric concentrations of PCBs. However, Oehme et al.  reported that the relative standard deviation for chemical analysis (repeated analysis of one sample) of air samples was typically less than 0.1 for PCB 153. It can be expected that the variation caused by the sampling procedure adds significantly to this figure. Hence, the τtrue − total may actually be much lower than the τtotal to the effect that the relationship between RSDtotal (i.e., RSDtrue − total would be lower) and τHO· illustrated above (Figure 5) would fall even more outside the range anticipated from Junge's relationship.
3. Toward a Reconciliation of Existing Discrepancies
 Significant inconsistencies appear to exist in our collected view of the biogeospheric fate of PCBs, and by inference this may hold true also for many other similar compound classes. We have attempted in the above section to dissect and analyze the consistency of the HO· sink with available field data. The objective of this section is to consider the quality of related current concepts so as to find a direction toward reconciliation of the existing discrepancies.
3.1. HO·-Reaction Rates
 We have scrutinized the experimental approach taken by Anderson and Hites [1996a] to obtain the intrinsic reaction rates and find no flaws or aspects that easily could have been done better. The main shortcoming of the experimental design was the relatively narrow and elevated temperature interval over which the experiments had to be conducted, necessitating an extrapolation with the Arrhenius equation over a large temperature range down to ambient temperatures. Additionally, in the paper by Anderson and Hites [1996a] uncertainty ranges were given for the reaction rates at +25°C. It would be more relevant to evaluate the uncertainty range at the global tropospheric mean temperature, which may be somewhere near −10°C.
 We have therefore estimated the uncertainty caused by the extrapolation, i.e., the temperature correction outside the measured temperature interval, by evaluating the raw data on measured reaction rates from the original paper [Anderson and Hites, 1996a] in Arrhenius plots (Figure 6). In general, the uncertainty (95% C.I. divided by mean) became 100–200% larger when extrapolating the data down to −10°C compared to the uncertainty at +25°C. However, this expanded uncertainty range is still too small to alone explain the discrepancy between the sink estimate based on the reaction rates and environmental observations.
3.2. Spatial and Temporal Variations in [PCB] and [HO·]
 The parameterization of the PCB and HO· concentration fields is of obvious importance to the estimation of the large-scale sink of PCBs resulting from their bimolecular reaction. Anderson and Hites [1996a] and Axelman and Broman  put forth two separate estimates of the total environmental sink of individual PCBs due to their reaction with the HO·. A comparison of these two estimates, based on the budget for PCB 28 (its contribution to the sum-PCB budget of the first work), shows that they were of similar magnitude. The same intrinsic reaction rates were used, but there were some significant differences in the assumptions concerning the parameterization of the PCB and HO· concentration fields. Anderson and Hites [1996a] assumed a homogeneous PCB concentration horizontally but assumed a decreasing concentration gradient with altitude, assumed to follow the better known vertical distribution of propane, which has a similar HO·-reaction rate as their average PCB with “3.5 chlorines”. In contrast, Axelman and Broman  assumed a homogenous vertical distribution in the troposphere, but used extensive environmental data to develop a relationship describing the latitudinal dependence of individual congener concentration (Figure 3). It is quite plausible that both studies are overestimating the total amount of tropospheric PCBs but for different reasons. As illustrated in Figure 3, the PCB concentrations decrease significantly toward higher latitudes. Hence, although Anderson and Hites [1996a] used half the annual averaged concentration at the low-latitude Bermuda station this likely leads to an overestimation of the tropospheric inventory of PCBs. The shortage of data for many regions of the Earth, in particular the Southern Hemisphere, complicates estimating this inventory. However, atmospheric concentrations of non-ortho substituted PCB 77, PCB 126 and PCB 169 from an Atlantic transect from 52°N to 75°S [Lohmann et al., 2001] showed that concentrations in the Southern Hemisphere were generally only a factor 1.5–3 lower than in the Northern Hemisphere, and that the concentrations of PCB 126 (5CB) and PCB 169 (6CB) were decreasing with increasing latitude in both hemispheres. It is known that the non-ortho substituted PCBs have additional emission sources than the technical mixtures constituting the overwhelming source of ortho-substituted PCBs. For instance, non-ortho substituted PCBs are formed during incineration of polyvinyl chloride plastic (PVC) [Katami et al., 2002] and municipal waste [Sakai et al., 2001]. Nevertheless, the data of Lohmann et al.  indicate that the PCBs behave similarly in the troposphere of both hemispheres and that significant amounts can be expected to exist also in the Southern Hemisphere. Another potential source of error is that concentrations measured over the North Atlantic, and adjacent continental areas, may overestimate the total global tropospheric inventory of PCBs, considering that most of the continuous “new” emissions and secondary emissions may occur in this part of the world.
 The HO· concentration field may also exhibit some significant dynamics that neither of these two budget calculations has taken into consideration. Anderson and Hites [1996a] used a global 24-hr average HO· concentration of 9.7·105 [Prinn et al., 1995], while Axelman and Broman  considered modeled latitudinal background distribution of the HO· concentration by Spivakovsky et al. . Still, recent findings suggest that NOx-emissions from the heavy ship traffic across the North Atlantic give rise to HO· concentrations elevated by up to one order of magnitude [Lawrence and Crutzen, 1999]. This is another indication that a longitudinal resolution of the tropospheric concentration of both PCBs and HO· is warranted.
 The magnitude of the vapor-phase HO· reaction sink discussed here is directly dependent on an accurate determination of the gas-phase concentration (Cg) of the PCBs. The vast majority of data on the aerosol-gas distribution of PCBs and other semi-volatile organic compounds in the atmosphere is based on a method of sampling particles on glass- or quartz-fiber filters followed by passing the air through an adsorbent, such as XAD or polyurethane foam (PUF). Different mechanisms potentially causing sampling artifacts with this approach have been suggested. Cg may be overestimated due to blow-off of PCBs associated with the water-air interface and/or the organic-matter film covering the particles collected on the filter. The pressure drop across the filter may also change the partitioning, causing the compounds to volatilize from the particles trapped on the filter and be collected on the adsorbent [Bidleman, 1988]. Conversely, Cg may also be underestimated if the gas-phase species is sorbed to the large surface area of the glass/quartz-fibers [Hart and Pankow, 1994]. This latter effect is, however, considered most efficient for low-volatile compounds such as larger PAHs. In urban areas some 10–30% of 6–7CBs may be collected on the filters, whereas the same figure for 3–4CBs is a few percent. At present, the fraction collected on the glass-fiber or quartz filter using the above techniques is so low in remote locations that they are generally either discarded or lumped together with the PUF adsorbent used to collect the gas phase [Panshin and Hites, 1994; Oehme et al., 1996]. An exception is the finding by Lohmann et al.  that the particulate fraction of PCB 126 and PCB 169 was more than 50% in marine atmospheric samples south of 50°S. This phenomenon may be restricted to non-ortho substituted PCBs since these planar compounds have been shown to sorb considerably stronger to soot carbon than ortho-substituted PCBs such as PCB 28 and 153 (T. Bucheli and O. Gustafsson, unpublished data, 2002).
 Partitioning of semivolatile organic compounds onto the molecular air-water interface (AWI) is increasingly being recognized as a potentially important process affecting the speciation-driven fate of hydrophobic compounds in many different situations. For instance, this AWI species has been suggested as one explanation to much elevated fog-droplet association of pesticides [e.g. Glotfelty et al., 1987] and to elevated concentrations found in precipitation [Capel et al., 1991], both well above expectations based on the compounds' Henry's law constants. It is conceivable that PCBs associated with the AWI of small hydrometeors and/or with the water skin common to all hydrophilic aerosols are blown off on impact when these fragile nongaseous forms strike the filter surface and are hence collected on the downstream adsorbent and assigned to the gaseous species. Pankow  found linear correlations between the subcooled liquid vapor pressure and the air-water interface to air partition coefficient (Ka−awi; mol m−2/mol m−3) for PAHs up to and including benz[a]anthracene:
 Waterside partitioning to the AWI of methylperylene [Gustafsson and Gschwend, 1999] provided an independent confirmation of the obtained relationship as - when converted from waterside to airside partitioning with Henry's constant - the Ka − awi of methylperylene fell on the extrapolated line of the PAH regression in Pankow . This extended the range of vapor pressures investigated to cover those for most of the PCBs. While equation 6 is strictly applicable only to PAHs, we applied it to PCBs to test whether the AWI species could be important for PCBs in the atmosphere. However, even under conditions with high aerosol concentrations of 105 cm−3 and a typical aerosol size of 0.1 μm, the fraction of total 3–4CBs associated with the aerosols as a result of this process would be on the order of 10−5. In summary, it appears unlikely that a missed AWI-associated species could significantly attenuate the large HO·-reaction sink term whereas sampling artifacts cannot be fully excluded.
3.4. Net Environmental Sink: Settling to Deep Sea Versus HO· Reaction
 It is instructive to quantitatively compare the postulated HO· sink with other potentially large sinks for a range of PCB congeners. Consider a vertical column, including the troposphere (6000 m; when corrected for decreasing pressure with height) and the mixed surface ocean (200 m; year-round average for midlatitude open ocean). The major sinks of PCBs considered are reaction with HO· in the atmosphere and deep-sea export on sinking particles. The relative sink magnitudes may be estimated by formulating their apparent first-order sink rates (time−1):
where fair and fpart are the fraction of a congener in the vapor-phase air and surface water particle forms, respectively. This phase speciation is calculated for the air and water columns, assuming equilibrium particle-water and air-water partitioning, using typical values for Kow − Koc and KH (Table 1), an ambient temperature of 20°C and a typical open ocean particulate organic carbon concentration of 30 μg/L. Relative to the intrinsic rates on which the HO-mediated reaction lifetimes in Figure 4 is based, these effectively system-wide HO-mediated sink is attenuating effect of the phase distribution. In equations (7a) and (7b) it is seen that the HO reacts only with the fraction of a given PCB in this troposphere-ocean system that exists as a tropospheric gas. This fair is becoming decreasingly smaller with increasing degree of chlorination, thus resulting in very sluggish overall HO-mediated removal in an environment consisting of other compartments in addition to the atmosphere. The apparent first-order ocean settling rate coefficient, ksettle, (day−1) is defined for particles collected on GF/F filters and is derived from the surface water disequilibrium between 238U − 234Th [Gustafsson et al., 1997b]. A value of 0.044 day−1 obtained at Bermuda Atlantic Time series Station [Gustafsson et al., 1997b] was employed and this is representative for the year-round average of the oligotrophic North Atlantic [Buesseler, 1998]. Using the best data currently available for parameterization (Table 1), this simple calculation would predict that the HO· route is the dominating sink for less-chlorinated congeners whereas deep-sea export should be anticipated to be prevalent for heavier PCBs even in a low-export oligotrophic regime (Figure 7). As indicated above and by Axelman and Broman , there are several additional removal routes, including shelf sediment and soil burial, also deserving of detailed attention.
Table 1. Parameters Used to Calculate the Characteristic Removal Times in Figure 7
 One of the weakest aspects in the current state of assessing global budgets of PCBs is the treatment of “new” emissions. Scrutinizing the basis of the existing emission estimates reveals that they are based on very few and frequently out-of-date information sources. For example, estimates given by Harrad et al.  and the frequently referred to “European Emission Inventory” [Berdowski et al., 1997] rely on the same seminar proceedings concerning the production and use of PCBs until the early 1980s [Bletchly, 1986]. In these emission assessments [Berdowski et al., 1997, Bletchly, 1986] the largest specified source of PCB emission is leakage from electrical equipment. Little discussion was devoted to possible additional emission routes from diffuse sources. This despite the fact that Bletchly  concluded that as much as 500 000 t PCB found its way to open applications, small capacitors and export to non-OECD countries, meaning that this amount was unaccounted for in the original emission estimates. The open applications include carbonless copying paper, plasticizers, hydraulic fluids, adhesives, paints, window sealants, light fixtures and other building materials. It is possible that these materials still are leaking either from their original application or from contaminated land, sediments and waste deposits. Bletchly  also commented on this large unaccounted-for amount and suggested that a major portion could be free in the environment. In the analysis of the global use and emission of PCBs by Breivik et al. [2002a, 2002b], an attempt was made to quantitatively account also for emissions as a result of the “open use” of PCBs. Other diffusive emission routes considered were fires and accidental spills to soils. Detailed information on production and use was fed into an emission mass balance model based on available information on product lifetime, emission factors, PCB destruction and accidental release factors. A number of assumptions had to be made to complete the model and the uncertainty range of the output was thus large. For instance, the mid value of the annual current emission flux (Femissionmid) for PCB 52 was 0.9 t yr−1 (by 1999) with an uncertainty range (Femissionmin − Femissionmax) of 0.01 − 27 t yr−1. Still, not even the high limit of the range is near the level needed to account for the large suggested HO· loss rate of, for instance, PCB 28 (3CB) or PCB 52 (4CB). For PCB 28 Femissionmax is only 2% of FHO· and for PCB 52 this figure is 5%.
 However, the approach of Breivik et al. [2002b] was in essence an allocation of the production into different types of use with different emission factors. Therefore, the diffusive emission of PCBs from known point sources with a significant level of contamination was not accounted for by Breivik et al. [2002b]. Any medium with a fugacity significantly higher than the background fugacity is a potential source for continuous emissions. One example is New Bedford Harbor, where high sediment concentrations of PCBs up to 103 times higher than background shelf concentrations have been found [Ikalainen and Hall, 1986]. Similar concentrations were found at a contaminated site near a paper mill in Västervik, Sweden. The PCBs at this location are believed to originate from self-copying paper processed at a paper recycling facility [Axelman and Broman, 1999]. In the pulp fiber basin and in the adjacent sediments approximately 100 kg PCB 28 was found, of which about 0.16 kg yr−1 was estimated to be leaking to the surrounding environment ([Axelman and Broman, 1999], congener specific data previously unpublished). Estimated volatilization leakage from New Bedford Harbor is in the same range, 2.2 kg yr−1 for ΣPCB [Garton et al., 1996]. The relatively few numbers of sites investigated makes it difficult to assess the global importance of this type of source. However, the reported emission rates can be contrasted to the postulated large global HO·-reaction sink of PCB 28. In the order of 106 such contaminated sites would be required to balance the reaction sink term.
4. Future Research Needs
 Mass balance constraints and process-oriented evaluation of available congener-specific field data dispute the proposition that the HO· reaction constitutes the dominant global sink for PCBs. The number of observations and the consistency between them makes the case so strong that we argue that there clearly is a gap in our understanding of the environmental fate of PCBs. It is distressing that, after 30 years of research, there remains a fundamental lack of comprehension regarding the environmental fate of one of the most well studied classes of organic contaminants. Their global dispersal has long been known in a qualitative manner but we still cannot answer the question: which sink process dominates the large-scale fate of PCBs, and by inference other semivolatile organic compounds, and approximately how large is this sink? These features should be among the first to be considered in any risk assessment of the emission of toxic chemicals as they hold the clue to estimating environmental longevity and “cleaning times.” We suggest that there is a need to much better understand the global behavior of PCBs, both in terms of permanent loss fluxes and significant environmental inventories. The deep ocean has been suggested to be a significant sink for PCBs, but this process is extremely poorly investigated for PCBs, as well as for many other toxic organic substances. Given that burial in marine sediments is the largest organic carbon sink from the active reservoirs in the global carbon cycle and that 90% of this burial takes place in shelf sediments [Hedges and Keil, 1995], this obvious inventory and sink for hydrophobic organic pollutants ought to be quantified on a global scale. While latitudinal coverage of PCBs in the remote atmosphere is becoming available (e.g., Figure 3), a better coverage in the vertical and at longitudes away from North America and Europe, and ideally for 1–2CBs, would allow improved field testing of the magnitude of the tropospheric HO· reaction as a globally significant sink of PCBs.
 Fruitful discussions on various aspects of the large-scale fate of POPs with Karl-Heinz Ballschmitter, Dag Broman, Ron Hites and Kevin Jones are gratefully acknowledged. We also thank Anders Bignert for advice on statistical matters. This work was supported by funding from EU DG XII, contract ENV4-CT97-0638 (GLOBAL-SOC).