An unknown fraction of mercury that is deposited onto the cryosphere is emitted back to the atmosphere. Since mercury that enters the meltwater may be converted to highly toxic bioaccumulating methylmercury, it is important to understand the physical and chemical processes that control the ultimate fate of mercury in the cryosphere. In this study, we review deposition mechanisms as well as processes whereby mercury is lost from surface snow. We then discuss redox reactions involving cryospheric mercury. We address the conditions under which reduction and oxidation occur, the stabilizing effect of halides, and the reducibility of reactive gaseous mercury versus mercury associated with particles. We discuss physical processes including the aging of the snowpack, the penetration of insolation through the cryosphere, the vertical motion of gaseous elemental mercury molecules through the cryosphere, the melting of snowpacks, and the loss of mercury from snowpacks during snowmelt both to the atmosphere and with the meltwater's ionic pulse. These physicochemical processes are universally applicable. Variations in the behavior of cryospheric mercury between open high-latitude, open high-altitude, and forested regions, which are caused by differing environmental conditions, are also discussed. Finally, we review observed concentrations of mercury in surface snow, seasonal snowpacks, meltwater, and long-term cryospheric records. The information presented here can be used to develop a parameterization of the behavior of cryospheric mercury that is dynamically linked to environmental variables.
 Gaseous elemental mercury (GEM) is emitted by anthropogenic processes and from land surfaces through natural processes. Previously deposited mercury may also be revolatilized from both land and oceanic surfaces. Between 2000 and 2005, anthropogenic mercury emissions declined in Europe but increased significantly in Asia [Pacyna et al., 2006, 2010]. Once emitted or revolatilized, GEM is transported through the atmosphere, where it undergoes chemical reactions, and is subsequently deposited. Once deposited, it can be converted to a potent bioaccumulating neurotoxin, methylmercury, in the presence of bacteria. Methylation can occur in aqueous environments including oceans [Sunderland et al., 2009], freshwater wetlands [Loseto et al., 2004; Goulet et al., 2007], and peatlands [Mitchell et al., 2008a]. This is an issue of great concern in Arctic countries, where country foods that include large marine mammals and fish constitute a high proportion of the aboriginal peoples' diet [Van Oostdam et al., 2005]; the mean mercury intake of Inuit sometimes approaches the provisional tolerable daily intake set by the World Health Organization [Van Oostdam et al., 2005]. Elevated maternal methylmercury contents have been linked to neurobehavioral damage in children [Van Oostdam et al., 2005].
 Atmospheric mercury is deposited onto the cryosphere–snow- and ice-covered surfaces–as onto other surfaces. It is known that some of the mercury deposited onto the cryosphere is rapidly emitted back to the atmosphere [Lalonde et al., 2002]. However, the precise fraction of the deposited mercury that is revolatilized has been hotly debated by scientists working at high latitudes. Some studies have suggested that net deposition associated with springtime Atmospheric Mercury Depletion Events (AMDEs) is low or insignificant [St. Louis et al., 2005; Kirk et al., 2006; St. Louis et al., 2007; Hedgecock et al., 2008] and that there is no firm evidence that AMDEs exert a significant influence on mercury concentrations in the Arctic Ocean [Outridge et al., 2008]. However, other studies have concluded otherwise. Steffen et al.  determined that there was a net annual loss of surface-level atmospheric GEM in 5 of 7 years at Alert. Hirdman et al.  concluded that “a substantial fraction” of mercury deposited in association with AMDEs accumulates in the snow. Johnson et al.  found that snowpack loads of mercury remain fairly constant from the springtime AMDE season to snowmelt. Loseto et al.  concluded that snowmelt may be the primary source of mercury to lakes in the Canadian High Arctic and a significant source to water bodies downstream, including the Arctic Ocean; concentrations of total mercury in lakes were barely affected by the presences of catchment wetlands. Dommergue et al.  estimated that cryospheric meltwater contributes 8–21% of the mercury content of the fjord near Ny-Ålesund in the Norwegian Arctic, despite an average total mercury concentration in local runoff streams of only 3.5 ng L−1; a concentration of 10 ng L−1 might be considered significant. Brooks et al. [2008a] estimated that over the entire Antarctic Polar Plateau, ∼60 tons of mercury is sequestered yearly below the photoreduction zone of the snowpack. Bargagli et al.  observed mercury concentrations in some Antarctic mosses and lichens that were higher than in similar lichens from polluted Northern Hemispheric environments, indicating a significant net deposition of mercury in the Antarctic. These results suggest strongly that mercury deposited onto the cryosphere at high latitudes does, indeed, have an important impact on the environment. Concentrations of cryospheric mercury can help in evaluating the long-term impact of mercury deposited onto this medium.
 Long-term cryospheric records are not alone in their ability to assess net mercury deposition over long periods: peat and lake sediment cores also provide long-term records of net mercury deposition. A net deposition rate of only ∼1 μg m−2 yr−1 was deduced by Givelet et al.  from peat hummocks in the Canadian High Arctic for 4000 B.C. to A.D. 1200. Pre-industrial accumulation rates calculated by Shotyk et al.  from peat cores in Greenland and Denmark are also low (0.3 to 3 μg m−2 yr−1). However, in 1953 and 1994–1995, accumulation rates of ∼174 and 14 μg m−2 yr−1, respectively, were found by Shotyk et al. . Unfortunately, the interpretation of mercury concentrations in long-term peat and lake sediments is more complicated than in long-term cryospheric records. In peat, mercury is able to move both vertically and laterally and binds preferentially to more decomposed material [Biester et al., 2007]. For lake sediments, the rate of transfer of mercury to these sediments may be affected by the level of algal productivity, which may, in turn, be affected by climate change [Outridge et al., 2005, 2007].
 The issue of the fate of mercury deposited onto the cryosphere will likely become even more important over the next few decades than it is currently. During the next few decades, Asian anthropogenic mercury emissions and, consequently, Northern Hemispheric mercury deposition are expected to increase [Faïn et al., 2009]. Furthermore, climate change may also impact the deposition of mercury onto the cryosphere. The nature of this impact is as yet unknown. Also unknown is how a changing climate will affect the behavior of cryospheric mercury.
 Geographically, the magnitude of the net deposition of mercury onto the cryosphere is of particular significance in polar regions, given that high-latitude AMDEs are accompanied by important mercury deposition [Steffen et al., 2008]. Nonetheless, it is also an important issue at midlatitudes, where seasonal snowpacks may be collocated with sources of mercury emissions and/or with high population densities [Faïn et al., 2007]; Susong et al.  found a 2.6-fold increase in the mean mercury concentration of seasonal snowpacks in industrial versus remote regions of Idaho. Moreover, it is estimated that seasonal snowpacks annually cover 27 million km2 from 40°N to 60°N, while they only cover ∼18 million km2 from 60°N to 90°N [Pielke et al., 2004]. In the Southern Hemisphere, the Antarctic Polar Plateau covers 5 million km2 [Brooks et al., 2008a].
 During field campaigns, some studies have measured concentrations of mercury in snow, meltwater, and/or ice, while others have measured fluxes of mercury between the atmosphere and cryosphere. Unfortunately, the observations from such studies are only valid at specific locations and times. Since the observations we described above suggest that the flux of mercury at the cryosphere/atmosphere boundary is heterogeneous, varying both spatially and temporally, data from individual field studies may not be representative of processes occurring elsewhere or at other times. It is, therefore, difficult to extrapolate results from these studies either temporally or spatially [Steffen et al., 2008]. Thus, it is necessary to examine all available data together to derive a consistent theory of the physical and chemical processes governing the behavior of mercury in the cryosphere.
 A benefit of large-scale numerical models is their ability to estimate mercury concentrations and fluxes over extensive geographical regions and time spans. However, for model estimates to be reliable, the models must represent physical and chemical processes accurately. Dastoor et al.  described a representation of AMDEs and their associated deposition and revolatilization in a global three-dimensional model. However, the representation of the processes involved was simplistic. Furthermore, few small-scale process models that simulate the behavior of cryospheric mercury have been described in the literature to date. Faïn et al.  used a diffusion model to deduce atmospheric GEM concentrations from 1940 to 2006 from concentrations of GEM in firn air. Similarly, both Ferrari et al.  and Faïn et al.  modeled the diffusion of GEM in the interstitial air of snow. Poulain et al. [2007a] presented a mass balance for mercury in snowpacks. In this balance, wet deposition, dry deposition, and throughfall constitute mercury sources, while revolatilization and snowmelt constitute sinks. Observations of wet deposition and the concentration of mercury in the springtime snowpack, along with calculated rates of reduction under different canopies, were used to estimate the values of wet deposition, revolatilization, and snowmelt. Given these estimates and the observed snowpack mercury concentration, the sum of dry deposition and throughfall was derived. Thus, no one, to our knowledge, has yet simulated the behavior of mercury in the cryosphere in anywhere near its full complexity.
 Models will be able to simulate the present-day behavior of cryospheric mercury accurately only if they represent the physical and chemical processes that determine this behavior, as opposed to proscribing a constant rate of mercury revolatilization at all locations and times. Such dynamical physically based models will also be able to simulate how the revolatilization of cryospheric mercury is likely to respond to a future climate. We hope that this review will provide a theoretical foundation for dynamical physically based models of varying levels of complexity.
 This review is organized as follows. Background information that is required to understand fully the remainder of the manuscript is presented in section 2, which includes a discussion of processes through which mercury is deposited to and lost from the cryosphere. High-latitude AMDEs, which are accompanied by important mercury deposition, are discussed in detail. Section 3 describes the physical and chemical processes that govern the behavior of cryospheric mercury. The chemical processes consist of redox reactions that involve mercury and that occur within snowpacks. Physical processes include snowpack aging, the penetration of insolation through snowpacks, the motion of mercury within snowpacks, and the melting of snowpacks. The composition of meltwater is also discussed in section 3. Section 4 provides a summary of the universally applicable physicochemical processes that determine the behavior of cryospheric mercury, along with a discussion of regional variations in this behavior. Such variations are produced by differing environmental conditions. Sections 1–5 discuss observations of mercury in surface snow, seasonal snowpacks, meltwater, and long-term cryospheric records, with section 5 focusing on the geographical distributions of these observations. In section 6 we have provided a list of measurements that would be very useful for model development. In section 7 we estimate the impact of climate change on the behavior of cryospheric mercury. Finally, section 8 presents our conclusions.
 Conventions used in this review are listed here as an aid to the reader. Table 1 provides the geographical coordinates of all locations referred to in this article. Studies using multiple sites in the same general area have been assigned an average location. The entries of Tables 2–9 are ordered first by location, following the order of Table 1, then by date of data, and finally by alphabetical author order. The dates listed for a given study represent the study's main time period. Where observation values are not provided explicitly in the text, they have been estimated from article figures. To facilitate the comparison of values measured by different studies, units have been standardized. Where necessary, conversions use a density of water of 1 g cm−3. Estimates of errors provided in articles and reproduced in Tables 2–9 almost always represent the standard deviation of the observations. We define the sample size (n) as the number of snow samples collected; even if a single snow sample is analyzed multiple times, we consider the sample size to be unity. Unless an article states otherwise, we assume that a single sample was collected at each site and time.
Table 1. Locations of Sites Discussed
South Pole Station
Terra Nova Bay
Tibetan Plateau, Asia
Mountainous regions, western USA
Upper Fremont Glacier, Wyoming, USA
Flin Flon, Manitoba, Canada
Northwestern Ontario, Canada
North-central Minnesota, USA
Laurentians, Quebec, Canada
Hanover, New Hampshire, USA
Sainte-Foy, Quebec, Canada
Cairngorm mountains, Scotland
Col du Dôme glacier, France
Churchill, Manitoba, Canada
Kuujjuarapik/Whapmagoostui, Quebec, Canada
Svartberget catchment, Sweden
Barrow, Alaska, USA
Ship, Arctic Ocean
Resolute (Bay), Nunavut, Canada
Cornwallis Island, Nunavut, Canada
Hudson Bay, Canada
Ellesmere Island, Nunavut, Canada
Baffin Bay, Canada
Alert, Nunavut, Canada
Camp Century, Greenland
Site 2, Greenland
Station Milcent, Greenland
Station Nord, Greenland
Ship: North Atlantic, Arctic oceans
2. Vertical Fluxes
 Mercury is deposited onto and may subsequently be lost from snowpacks. Deposition mechanisms include dry and wet processes. Both elemental and oxidized species of mercury are deposited. Deposition processes are discussed further in section 2.1.
 Loss mechanisms for cryospheric mercury include (1) percolation, (2) the settling of particulate mercury (PHg), and (3) the reduction of deposited oxidized mercury, which may then be emitted back to the atmosphere as GEM [Lalonde et al., 2002]. Although particulate settling may have contributed to anomalously high mercury concentrations in the snowpack overlying the Commonwealth Glacier [Witherow and Lyons, 2008] and particulate settling or percolation may have transported mercury from surface to basal snow at Ny-Ålesund [Larose et al., 2010], it was not important in a snowpack at Sainte-Foy [Lalonde et al., 2002]. Furthermore, Poulain et al. [2007a] estimated that, of the mercury lost from a Laurentian snowpack, 67% was lost through revolatilization and 31% through presnowmelt percolation. Thus, it seems likely that outside of melting periods, loss of mercury from snow is due primarily to the emission of GEM back to the atmosphere [Lalonde et al., 2002, 2003].
 Vertical gradients of atmospheric GEM concentration above snowpacks, and the fluxes they imply, can be significant; differences in concentration of up to ∼50% between two sampling heights above the snowpack, where the lower and upper sampling heights are at approximately 10 to 20 cm and 150 to 180 cm above the snowpack surface, respectively, have been observed [Steffen et al., 2002; Lahoutifard et al., 2005; Sommar et al., 2007]. Given the occurrence of cryospheric mercury reduction, the GEM emitted from snowpacks may have been deposited previously as any mercuric species by any process at any time. Unfortunately, the deposition of oxidized mercury to the cryosphere may coincide with the emission of GEM; deposition may mask the emission signal in observations of total mercury in snow [Kirk et al., 2006]. In the Laurentians, Poulain et al. [2007a] observed a balance between deposition and emission, with constant mercury concentrations in surface snow. On a seasonal scale, Steen et al.  found that deposition and emission dominated during polar winter and spring, respectively, at Ny-Ålesund. Emission is discussed further in section 2.2.
 In the literature, the emission of previously deposited GEM is sometimes referred to as “reemission.” However, according to the usage recently adopted in the assessment report for the Task Force on Hemispheric Transport of Air Pollution, the revolatilization of mercury deposited previously is considered a secondary emission. Primary emissions consist of anthropogenic mercury emissions and emissions of mercury through natural processes such as volcanoes and geological sources. Cryospheric mercury emissions are likely a combination of primary and secondary emissions. In this article we discuss the behavior of the secondary emissions. For the sake of simplicity, we refer to these secondary emissions as “emissions.”
 Both elemental (GEM) and oxidized species of mercury are deposited. The oxidized species may be gaseous (reactive gaseous mercury, RGM) or associated with particles (PHg). RGM and PHg are both operationally defined as their exact chemical structures are unknown [Simpson et al., 2007b; Steffen et al., 2008]. It is known, however, that the depositions of RGM and PHg are much more efficient than the deposition of GEM [Lin et al., 2006]. Nonetheless, GEM deposition is substantial as a result of the predominance of GEM in the atmosphere [Lin et al., 2006]. Wet and dry processes produce, in general, comparable amounts of mercury deposition [Lin et al., 2006]. Wet deposition requires precipitation. The rate of dry deposition varies with mercuric species, boundary layer stability, and land surface cover [Lin et al., 2006]. Unfortunately, predicting the behavior of RGM and PHg, including their depositions, will remain imperfect until their physical and chemical properties are fully understood. Defining their properties is currently an area of active research. Under investigation are basic questions such as whether RGM is typically formed first and then converted to PHg or vice versa.
 When AMDEs are observed, they can be of local or nonlocal origin [Constant et al., 2007]. Local AMDEs are those where the chemical processes occur in situ. The chemical processes of nonlocal AMDEs have occurred elsewhere; the resulting air mass has been transported to a new location. Local AMDEs occur in the presence of weak winds and are characterized by both high ambient concentrations and significant deposition of RGM and PHg [Constant et al., 2007; Johnson et al., 2008]. Kirk et al.  reported that concentrations of both RGM and PHg at Churchill on Hudson Bay, which were frequently above 500 pg m−3, were often sufficiently high to replace the depleted atmospheric burden of GEM. These observations suggest that the AMDEs recorded had occurred so recently that the pool of atmospheric oxidized mercury created had not yet been significantly depleted through deposition; bromine explosions and the subsequent oxidation of atmospheric mercury are apparently faster than the deposition of RGM and PHg. Nonlocal AMDEs occur in the presence of stronger winds [Simpson et al., 2007b]. During these events, atmospheric ozone and GEM concentrations drop at a precipitous rate that cannot be explained by chemistry [Bottenheim and Chan, 2006]. Ambient concentrations and deposition of RGM and PHg may not be elevated [Constant et al., 2007; Ferrari et al., 2008]. Bottenheim and Chan  calculated that air parcels took on the order of 1 day to travel from bromine explosion sites to Barrow but up to 6 days to travel to Alert and Ny-Ålesund; upwind of Alert and Ny-Ålesund, the environment is not normally conducive to depletion events.
 As a result of both the local/nonlocal nature of AMDEs and the high degree of spatiotemporal heterogeneity exhibited by bromine explosions, AMDE-related mercury deposition is episodic and highly heterogeneous [Ferrari et al., 2005; Kirk et al., 2006; Steffen et al., 2008]. Near Churchill, mercury concentrations in surface snow varied 2.5-fold within 2 km [Kirk et al., 2006]. On and near Ellesmere Island, total snowpack mercury concentrations differed by ∼2.5 orders of magnitude over ∼300 km [St. Louis et al., 2007]. In northern Alaska, both mercury and sea salt deposition decreased with increasing distance from the coast, reflecting the maritime nature of AMDEs [Snyder-Conn et al., 1997]. Similarly, mercury concentrations in surface snow [Constant et al., 2007; Brooks et al., 2008b] and vegetation [Landers et al., 1995] declined with distance from a high-latitude coast. Mosses and lichens collected near a frequently refreezing lead in Antarctica were characterized by mercury concentrations up to fourfold to fivefold higher than in specimens collected elsewhere in the region [Bargagli et al., 2005]. Finally, modeling results show elevated deposition north of Alaska and eastern Russia [Dastoor et al., 2008]. These results indicate the danger of extrapolating an individual study's deposition values either spatially or temporally at high latitudes.
 At lower latitudes where AMDEs do not occur, canopies affect mercury deposition [Lin et al., 2006]. Vegetation scavenges gaseous-phased mercury from the atmosphere [Nelson et al., 2010]. The mercury is subsequently flushed to the forest floor by precipitation and/or contributes to litterfall [Poulain et al., 2007a; Nelson et al., 2010]. Nelson et al.  found that the average mercury deposition from individual significant snowfall events in Maine was fourfold higher in the presence of a canopy than in the open. Although canopies do not reproduce the important mercury deposition associated with AMDEs, the increased deposition of mercury onto canopy-covered snowpacks is important nonetheless; AMDE-related deposition is sporadic both temporally and spatially while canopy-covered snowpacks have an extensive global coverage.
 Regardless of latitude, high-altitude sites are influenced almost exclusively by the free troposphere during winter when the atmosphere is stable. However, they may be exposed to regional pollution through convective lifting during spring and summer [Maupetit et al., 1995; Cozic et al., 2008]. This springtime polluting is extremely unfortunate: the mercury is being deposited near snowmelt, so that a substantial portion of the deposited mercury likely enters the high-elevation meltwater (section 3.2.4). Since snowmelt coincides with a biologically sensitive point in time [Lindberg et al., 2002], elevated concentrations of mercury in meltwater likely enhance the incorporation of mercury into biota, whether through methylation or other processes. This effect is likely more pronounced in regions where the growing season is short, such as at high latitudes or at high altitudes.
 Although much of the deposited mercury is in oxidized form, it is GEM that is emitted from snow-covered surfaces; in-snow reduction (section 3.1.1) is indicated. The produced GEM molecules move vertically to the surface via diffusion and ventilation (section 3.2.3). The molecules that reach the surface of the snowpack may escape to the atmosphere. Emitted GEM molecules are sometimes quickly reoxidized either at or just above the surface of the snowpack [Skov et al., 2006].
 Although insolation drives GEM emission in laboratory studies [Dommergue et al., 2007; Faïn et al., 2007], mercury emission and insolation are not truly correlated: daily maximum GEM net emission values remain fairly constant at Ny-Ålesund from March through June despite the fact that daily maximum solar insolation values are assuredly increasing throughout the period [Steen et al., 2009]. The lack of continuous increase in mercury emission throughout spring may reflect the fact that the supply of cryospheric mercury available to be emitted does not increase throughout the period. However, global radiation and GEM emission at Ny-Ålesund were correlated during March but not during any of the remaining spring months [Steen et al., 2009]. Thus, the fact that mercury emission maximizes at midday may be only indirectly related to insolation: insolation initiates a chain reaction that eventually influences mercury emission. This issue will be discussed further in section 3.2.3.
 Concerning the effectiveness of GEM emission, Brooks et al. [2008a] estimated the lifetime of mercury within the top 15 cm of snow at ∼16 days. Sherman et al.  estimated that ∼35–50% and ∼65–75% of mercury in surface snow and drifted snow samples, respectively, was lost during 10.5 h of photoreduction in chamber experiments prior to snowmelt. It appears as though canopies reduce mercury emission as the average net mercury deposition to the snowpack over the entire cold season was tenfold higher in the presence of a canopy [Nelson et al., 2008], despite an only fourfold increase in deposition under canopies. Indeed, Poulain et al. [2007a] suggested that only 2% to 3% and 18% to 24% of mercury deposited onto the ground by wet and throughfall processes under coniferous and deciduous canopies, respectively, was emitted. Mechanisms by which canopies inhibit GEM emission are discussed in sections 3.1.1, 3.1.2, and 3.2.3.
 It is important to realize that a decreasing mercury concentration in surface snow between successive samplings does not necessarily indicate emission, even in the absence of percolation and particle setting: the occurrence of either fresh snowfall or blowing snow between samplings can result in different batches of snow being sampled [Domine et al., 2004; Poulain et al., 2004; St. Louis et al., 2005; Kirk et al., 2006; Cobbett et al., 2007]; in such cases, comparing mercury concentrations becomes meaningless.
 Fresh snowfalls are also important in that they can bury mercury contained in the former surface layer, rendering it unavailable for emission. Of the three glacial snowpacks analyzed by Dommergue et al. , the layer containing the highest mercury concentration was at the surface in only one instance. In that snowpack, which is on the Austre Lovénbreen glacier, the layer containing the second highest concentration was at a depth of 225 cm; this layer likely represented mercury deposited during an AMDE and buried soon after by a significant snowfall [Dommergue et al., 2010]. The Austre Lovénbreen glacier near Ny-Ålesund is 1–2 km from the coast [Kaštovská et al., 2005]. Witherow and Lyons  also described the creation of a snowpack layer characterized by a high mercury concentration via the burial of this layer by fresh snow. Thus, regions experiencing frequent snowfalls, which render mercury unavailable for emission, are more likely to accumulate mercury in the snowpack. Similarly, wind action can bury a surface layer [Brooks et al., 2008a]. However, it can also expose a formerly buried layer, making its mercury content once again available for emission.
3.1. Chemical Reactions Within the Snowpack
 It is possible that atmospheric GEM may be trapped within the snowpack by in-snow oxidation or by being buried by falling snow. Otherwise, GEM, being highly labile [Steen et al., 2009], moves easily between the atmosphere and cryosphere. On the other hand, deposited oxidized mercury must be reduced before GEM can be emitted. The reduced mercury can, in turn, be reoxidized within the snowpack. Individual chemical pathways involved in the redox reactions are not investigated by this study. Interested readers are referred to any of the available review articles on chemical reactions involving mercury, including that by Lin and Pehkonen .
Brooks et al. [2008a] estimated that ∼490 tons of deposited oxidized mercury is reduced annually to elemental mercury in the Antarctic Polar Plateau snowpack. Observations of GEM in the interstitial air of snow and surface snow suggest that reduction of oxidized mercury is possible in the dark [Faïn et al., 2007; Ferrari et al., 2008], either as a continuation of photolytically initiated reactions [Lalonde et al., 2003; Dommergue et al., 2007] or through a reaction requiring no insolation at all [Ferrari et al., 2004]. However, laboratory and flux chamber experiments [Lalonde et al., 2003; Dommergue et al., 2007; Johnson et al., 2008] and observations of GEM within the snowpack [Poulain et al., 2004; St. Louis et al., 2005; Faïn et al., 2007] indicate that the overwhelming majority of the GEM emission occurs in the presence of solar radiation. Furthermore, Sherman et al.  reported the mass-independent fractionation of mercury in surface snow collected at Barrow. This fractionation, where odd-mass-number isotopes of mercury are lost preferentially, suggests that photochemical reduction has been active. The authors also reported the occurrence of mass-dependent fractionation, a behavior which has been associated with photochemical reduction of mercury in aqueous solutions. In mass-dependent fractionation, light isotopes of mercury are lost preferentially.
 The photoreduction of mercury has been reported to be forced by both visible (400–750 nm) [Poulain et al., 2004; Johnson et al., 2008] and ultraviolet-A (UV-A) radiation (320–400 nm) [Poulain et al., 2004; Faïn et al., 2007]. However, the photoreduction is forced primarily by ultraviolet-B (UV-B) radiation (280–320 nm) [Poulain et al., 2004; Dommergue et al., 2007; Faïn et al., 2007], with 305 to 320 nm the most important bandwidth [Dommergue et al., 2007]. Lalonde et al.  ascertained that the inclusion of UV-B light increased reduction fivefold. Thus, an important limitation to the rate of reduction is a sufficient supply of UV-B radiation. Indeed, insufficient UV-B radiation is almost certainly the most important factor in limiting the reduction of mercury within snowpacks. Perovich  observed nonzero irradiances at wavelengths of 320 and 305 nm at the surface of the Beaufort Sea starting from ∼1 March and ∼15 April, respectively, with ∼40 times as much irradiance measured at 320 nm than at 305 nm during May.
 Even in the presence of sufficient insolation and an ample supply of reductants, not all oxidized mercury is reduced and emitted [Dommergue et al., 2010]; not all forms of oxidized mercury are reducible by photodissociation or photoreduction [Dommergue et al., 2007]. Faïn et al.  reported that less than 10% of total mercury in the Summit, Greenland, snowpack is easily reducible by tin(II) (stannous) chloride (SnCl2) or sodium borohydride (NaBH4). Similarly, Ferrari et al.  determined that the concentration of oxidized mercury easily reducible by SnCl2 was less than the method detection limit of 0.8 ng L−1 in Alpine snow. Presumably, most of the easily reducible mercury had already been emitted. Lindqvist and Rodhe  reported that some fraction of both RGM and PHg is not easily reducible. Unfortunately, the precise fraction of each that is not easily reducible is unknown. This is an important issue as their reducibility will have a significant impact on the geographical distribution of GEM emission; if PHg is far less reducible than RGM, as is likely, the presence of PHg would affect the photoreactivity of the snowpack's mercury content [Larose et al., 2010]. Consequently, regions with higher PHg deposition would retain a higher percentage of deposited mercury.
 There is solid evidence that PHg in the snowpack is at most minimally reduced and emitted: Witherow and Lyons  associated anomalously high mercury concentrations in layers in a snowpack overlying Commonwealth Glacier with dust and particles; Jitaru et al.  found that total mercury was significantly correlated with their proxy for insoluble dust during cold periods in ice cores from Dome C; Loewen et al.  discovered that the highest mercury content in a snowpack on the Tibetan Plateau was located in a layer that also contained dust, which may have been transported from South and central Asia; Schuster et al.  determined that the three most important elevations in mercury concentration above the background level in the Upper Fremont Glacier coincided with major volcanoes, which emit both mercury and dust. In fact, 6% of the net mercury deposited in this glacier was produced by volcanoes; Balogh et al.  concluded that more than 90% of the total mercury in a Minnesotan snowpack was in the form of PHg; Poulain et al. [2007a] found that while PHg constituted ∼44% of mercury in wet deposition in the Laurentians, PHg constituted 69%, 72%, and 82% of the snowpack mercury under no, coniferous, and deciduous canopies, respectively. The authors concluded that PHg is likely created within the canopy and deposited as throughfall. However, the increase of PHg in the open also suggests the preferential emission of RGM; Poulain et al. [2007b] determined that both the concentration of total mercury and the fraction of PHg were significantly higher in both surface snow and throughout the snowpack over sea ice than over land in the Canadian Archipelago; St. Louis et al.  determined that PHg constituted up to 65% of total mercury in snow from the Canadian Archipelago; and Cobbett et al.  suggested that a sudden almost sixfold difference in snow core mercury concentrations at two sites that had previously been characterized by similar concentrations was caused by mercury associated with dust and debris having been blown to only one of the sites. Similarly, Witherow and Lyons  suggested that the lower mercury concentrations in snowpacks in central Greenland compared to snow overlying the Commonwealth Glacier, Antarctica, may be a result of lower PHg values. If this conjecture is accurate, the wide range of mercury concentrations observed in snow and firn by Faïn et al.  at Summit may partially reflect the fact that the transport of dust to Summit exhibits significant seasonable [Dibb et al., 2007, 2010] and interannual [Dibb et al., 2007] variability.
 Interestingly, Ferrari et al.  discovered that the fraction of total mercury in surface snow that consisted of species reduced easily by SnCl2 and/or NaBH4 declined suddenly at snow temperatures above ∼−5°C. The authors proposed that stable compounds form at higher temperatures through adsorption onto particles or via microbial activity. Alternatively, the high fraction of stable mercury found at warmer temperatures may simply indicate the onset of melting, which promotes simultaneously increased photoreduction of reducible mercury followed by increased GEM emission, and the swift removal of RGM in the meltwater's ionic pulse (sections 3.2.4 and 3.3). PHg exits the snowpack with the meltwater at a later date (section 3.3). Thus, if RGM is considerably more reducible than PHg, the Ferrari et al.  data are explained by the behavior of RGM and PHg at the onset of snowmelt. However, the possibility of the conversion between RGM and PHg within the snowpack is an important issue. If we assume that PHg is considerably less reducible than RGM, the formation of PHg within the snowpack will decrease GEM emission. Thus, to represent accurately the geographical distribution of GEM emission, we need to know onto which particles RGM is likely to adsorb within the snowpack, and the geographical distribution of those particles. We also need to know how likely the conversion from PHg to RGM is, which particles are involved, and the geographical distribution of those particles.
 Indications of dark oxidation were found by Poulain et al. [2007a] and Faïn et al. , with the process more active under a canopy than in the open [Poulain et al., 2007a]. Poulain et al.  found that oxidation was found to be favored over reduction in both the dark and under UV-A or visible radiation. However, this behavior is not ubiquitous: Ferrari et al.  found that oxidation dominated in general during the day, with reduction dominating at night, probably because oxidizing activity diminished at night while, as mentioned in section 3.1.1, reduction can occur in the dark. Seasonally, Mann et al.  discovered that total mercury concentrations in firn at Summit tended to increase during spring, likely as a result of the in-snow photochemical production of oxidants. Similarly, Faïn et al.  found that in-snow concentrations of GEM at Summit were lower during spring and higher during summer, as a result of oxidation dominating during spring and reduction dominating during summer; dark oxidation processes were likely active. Interestingly, concentrations in snow of both bromide and a dust proxy are higher in spring than summer at Summit [Dibb et al., 2007, 2010]. As mentioned in section 2.1, Summit is too far inland to experience AMDEs. Since halides stabilize oxidized mercury (this section) and increased dust concentrations are likely accompanied by increased concentrations of less-reducible PHg (section 3.1.1), the seasonality of both the bromides and dust proxy would promote higher concentrations of cryospheric oxidized mercury during spring.
 In relation to spatial distributions of cryospheric mercury and halides, St. Louis et al.  found that mercury concentrations in surface snow generally decreased from 345 ng L−1 at ∼1 km from a lead edge with distance from the lead. Similarly, Brooks et al. [2008b] observed concentrations of mercury in surface snow that decreased from ∼430 ng L−1 at the edge of the contiguous sea ice with distance from the edge. Interestingly, St. Louis et al.  found that total mercury concentrations tended to be either consistently high or low over all snowpack layers, suggesting that the spatial distribution of chloride ions likely remains fairly consistent over time. Similarly, Poulain et al. [2007b] found that the mercury content of snowpacks was considerably higher in snowpacks over sea ice than over land. The greatest difference, up to 2 orders of magnitude, was just above the sea ice. Over sea ice, chloride concentrations increased with increasing snowpack depth. This increase may reflect the fact that ions are able to migrate up into the bottom 17 cm of snowpacks covering sea ice [Domine et al., 2004].
 Concerning the source of the cryospheric stabilizing halide ions observed at continental sites, chloride ions found in Alpine snowpacks were likely emitted by anthropogenic activities [Ferrari et al., 2002]. Loewen et al.  determined that the chloride ions found in a Himalayan snowpack were likely transported to the Tibetan Plateau from South and central Asia by the jet stream. Dibb et al.  reported that bromide concentrations in surface snow at Summit often increased with vertical mixing; concentrations in the free troposphere had likely been enhanced by the springtime bromine explosions (section 2.1).
 In maritime regions, concentrations of stabilizing chloride ions in snow can be enhanced by sea spray and deposited sea salt aerosols [Domine et al., 2004; Simpson et al., 2005]. The aerosols are generated through the formation of new sea ice [Rankin and Wolff, 2003] and by snow blowing over sea ice [Yang et al., 2008]; even sites out of the range of sea spray can contain a significant concentration of sea salt [Beine et al., 2006]. Contamination by salts of coastal snow and snow covering first-year sea ice is ubiquitous during spring in polar regions [Simpson et al., 2007a, 2007b]. Multiyear ice is less contaminated [Yang et al., 2010]. Of great importance is the fact that AMDEs are also active during spring and also lead to increased mercury deposition near coastal regions and first-year sea ice (section 2.1); there is a strong likelihood that at least at some of the locations where mercury deposition is greatly enhanced, elevated chloride concentrations retain a greater fraction than usual of the mercury content of the snowpack.
 Calculated rates of net in-snow GEM production range from 0.0001 to 0.33 ng L−1 h−1 (Table 2). Assuming that the rate of net production is limited by the supply of oxidized mercury, we calculated reaction rate constants for the net reduction of oxidized mercury by dividing each article's given rate of net production by its mean mercury concentration. Mean total mercury and oxidized mercury concentrations are basically interchangeable in this calculation, given that the concentration of GEM in the interstitial air of snow typically represents less than 1% of the snowpack's total mercury content [Dommergue et al., 2003a; Ferrari et al., 2005; Faïn et al., 2007]. The calculated reaction rate constants vary from 7 × 10−6 to 0.6 h−1 (Table 2). Assuming that sufficient UV-B radiation and sufficient reductants were available at all sites, lower reaction rate constants reflect higher rates of reoxidation, higher concentrations of stabilizing halides, and/or higher fractions of mercury that are not easily reducible. Dommergue et al. [2003a], who calculated the lowest reaction rate constants, mentioned the presence of particulate matter on the snow. This suggests the presence of PHg in the snowpack, which is probably less reducible than RGM (section 3.1.1). We can expect that their reaction rate constants were depressed further by elevated snowpack concentrations of oxidizing and stabilizing halogens, given the coastal location and the report of high chloride concentrations in the snow at that location by Dommergue et al. [2003b]. It is very likely that the presence of PHg was the dominant factor in depressing the reaction rate constant. This supposition is supported by the fact that the reaction rate constant from Cornwallis Island, where halogen concentrations were also elevated, was third largest. Quite remarkably, at the onset of snowmelt, the estimated reaction rate constant for Kuujjuarapik/Whapmagoostui increased by an order of magnitude. The two highest reaction rate constants, as one might expect, are associated with inland continental locations that are not under canopies.
Table 2. Rates of Net Production of GEM and Reaction Rate Constants for the Net Reduction of Mercury Within the Snowpack
Rate of Net GEM Production (ng L−1 h−1)
Mean Mercury Concentration in Snow (ng L−1)
Reaction Rate Constant for the Net Reduction of Oxidized Mercury (h−1)
 Regarding the effect of canopies on reduction, Poulain et al. [2007a] calculated the rate of net GEM production to be almost 7 times greater in the open than under a canopy. These authors suggested that the reduction in incident radiation caused by the canopy's shadowing effect was the primary reason behind this decrease in production; snowpacks under deciduous and coniferous canopies receive only 0.52 and 0.07, respectively, of the incoming irradiance measured in the open [Poulain et al., 2007a]. However, the canopy's shadowing effect is reinforced by the elevated concentrations of both PHg and oxidants found in snowpacks under canopies (sections 3.1.1 and 3.1.2). Combining the lower rate of net production with the higher mercury concentrations observed in snow under canopies yields a reaction rate constant that is over 25-fold lower under canopies than in the open; the presence of canopies has an extremely important impact on GEM emission.
3.2. Physical Processes Within the Snowpack
 Since the photochemical production of GEM molecules within the snowpack and their subsequent motion toward the surface of the pack are mediated by the physical characteristics of the snowpack, we start this section with a description of snowpacks. Radiative transfer within the cryosphere is then discussed, followed by a description of vertical motion within the cryosphere and the impact of snowmelt on mercury emission.
 In general, the densification rate of a snowpack is correlated with density in early and midwinter, while it tends to be constant in spring. At the onset of melting, snowmelt fills the snowpack pores, snow grains are rounded further, thereby increasing compaction, and snowpack depth decreases; density increases [Mizukami and Perica, 2008]. Density typically ranges from 0.2 to 0.4 g cm−3 [Domine et al., 2008], although it can sink to 0.01 g cm−3 or lower for fresh dendritic snow [Domine et al., 2008] and rise to 0.61 g cm−3 for hard windpack [Beine et al., 2006]. Across the central Arctic Basin, snowpack density is typically 0.34 g cm−3 [Sturm et al., 2002]. Warm temperature snowpacks tend to have higher densities than cold temperature snowpacks. While density tends to decrease with increasing distance from a large body of water [Mizukami and Perica, 2008], snow on lakes is 21% denser than on nearby land [Sturm and Liston, 2003].
 All ice surfaces, including snow grains, are always covered by a quasi-liquid layer above ∼200 K [Abbatt, 2003]. The quasi-liquid layer is a disordered layer that strongly promotes molecular mobility [Kuhn, 2001]. It has some liquid-like properties but is not simply super-cooled water; such a layer would freeze rapidly, being in contact with ice [Domine et al., 2008]. The thickness of the quasi-liquid layer increases progressively as melting approaches [Abbatt, 2003].
 In general, the rate of cryospheric photoreduction of a chemical species is proportional to the actinic flux, or the amount of spherically integrated insolation reaching a given point, and the concentration of the reactants [Domine et al., 2008]. There are two regions of actinic flux. In the top layer, the sum of the upwelling radiation and the reflected incident radiation yields an actinic flux that is greater than the amount of radiation incident on the snowpack [Warren, 1982; King and Simpson, 2001]. The ratio of the top-layer actinic flux to the incident radiation is approximately fourfold, approximately twofold, and less than unity for UV radiation with a solar zenith angle of 0°, 60°, and 84°, respectively, while it is always approximately twofold for diffuse incident UV radiation [Simpson et al., 2002]. The depth of the top layer decreases from 2 to 0.1 cm for UV radiation as the solar zenith angle increases [Lee-Taylor and Madronich, 2002; Peterson et al., 2002].
 Below the top layer, the actinic flux decreases exponentially with depth as a result of scattering by individual snow grains, following the Bouguer-Lambert Law [King and Simpson, 2001; Domine et al., 2008]. Increasing solar zenith angles decrease the radiation's penetration: for UV radiation, the actinic flux at a depth of 60 cm at a angle of 25° is comparable to the actinic flux at a mere 1 cm depth at an angle of 65° [Lee-Taylor and Madronich, 2002]. The morphology of the snow grains drives the scattering [Grannas et al., 2007; Domine et al., 2008]; snow grains that are larger, and water-covered snow grains that are effectively larger, reduce scattering, enabling the radiation to penetrate the snowpack farther [Warren, 1982; Fisher et al., 2005]. Absorption of the radiation reduces the actinic flux. Since ice absorbs UV radiation weakly, snowpack absorption is determined primarily by its impurities, such as organic matter, soils, and/or aerosols [King and Simpson, 2001; Warren et al., 2006; Grannas et al., 2007; Domine et al., 2008]. Thus, impurities in the snowpack can be of considerable importance [Lee-Taylor and Madronich, 2002]. This ability of particulate matter to absorb radiation may contribute to the tendency for PHg to be retained by the snowpack (section 3.1.1); in the presence of PHg, less transmitted insolation is available for photoreduction.
 The impact of a snowpack's scattering and absorbing characteristics on the penetration of incident radiation is described by the e-folding depth, or the depth at which the actinic flux's magnitude is 1/e, or 37%, of its incident value [Perovich, 2007]. For a given snowpack, the e-folding depth decreases with decreasing radiation wavelength from visible to UV-B [King and Simpson, 2001; Galbavy et al., 2007; Perovich, 2007]. The increased absorption of melting snowpacks tends to reduce e-folding depths [Perovich, 2007]. Table 3 provides e-folding depths for various surface snow types. It is important to remember that the stated snow type may not have extended right to the e-folding depth [Beine et al., 2006]. As Table 3 demonstrates, insolation penetrates into sea ice farther than into any of the snow types listed. Sea ice is a combination of ice platelets, brine pockets, precipitated salts, and air bubbles, with temperature determining the relative proportions [King et al., 2005].
Table 3. The e-Folding Depths and Densities for Various Wavelengths and Types of Snow
 It is often estimated that ∼85% of the photoreduction occurs within the depth equal to twice the e-folding depth [e.g., King and Simpson, 2001]. At this depth, the magnitude of the actinic flux is ∼15% that of the incident flux according to the Bouguer-Lambert Law. Unfortunately, this calculation ignores the thin top layer of the snowpack where the actinic flux may be substantially greater than the magnitude of incident radiation, as discussed above. Furthermore, since the e-folding depth describes only the impact of the snowpack's physical characteristics on the cryospheric penetration of radiation, the depth equal to twice that of the e-folding depth provides no absolute measure of the magnitude of the actinic flux at that depth. As mentioned above, the magnitude of the actinic flux itself depends strongly on solar zenith angle; this issue is important at higher latitudes. Mercury-related photochemistry can be active to depths of ∼60 cm [Dommergue et al., 2003a; Faïn et al., 2008] or limited to the top ∼3 cm [Poulain et al., 2004; Brooks et al., 2008a]; 60 and 3 cm are far greater and smaller, respectively, than twice the value of almost all e-folding depths listed in Table 3. This range of chemically active depths may be explained by physical differences in the snowpacks and perhaps also by the fact that e-folding depths provide no absolute measure of the magnitude of the actinic flux.
3.2.3. Vertical Motion
 GEM molecules produced by reduction within the snowpack must travel to the snowpack's surface before being emitted to the atmosphere; the depth of the chemically active layer does not by itself describe the depth of the layer contributing mercury to emission. Gaseous molecules are transported within the snowpack by diffusion and ventilation [Albert and Shultz, 2002].
 Ventilation, which is caused by turbulence-induced pressure variations, is a very effective transporter of molecules within the cryosphere and over the atmosphere/cryosphere interface. Turbulent diffusivity is estimated at 10−3 to 1 m2 s−1 by Hutterli et al.  in a study of snowpack formaldehyde and at 10−5 m2 s−1 under moderate winds (7–9 m s−1) by Albert et al.  in a study of snowpack ozone. These estimates are 1 to 5 orders of magnitude greater than the effective molecular diffusivity values of Table 4. The influence of ventilation may extend to depths of 50–100 cm [Domine et al., 2008]. However, the impact of ventilation depends strongly on the snow's permeability [Kuhn, 2001; Albert and Shultz, 2002; Domine et al., 2008]. Wind blowing over a rough surface is an important source of turbulence [Kuhn, 2001; Albert and Shultz, 2002; Anderson and Neff, 2008]. The ventilation generated by this turbulence is known as wind pumping [Kuhn, 2001]. On the upstream and lee sides of a rough surface feature the surface pressure increases and decreases, respectively [Colbeck, 1989]. Ventilation forced by smaller surface roughness features is stronger near the top of the snowpack but decays more rapidly with depth [Colbeck, 1989]. Wind speed, which affects the intensity of the ventilation [Albert and Shultz, 2002], significantly affects mercury emission [Steffen et al., 2002; Lahoutifard et al., 2005; Steen et al., 2009]. Albert and Shultz  found that a threefold increase of the overlying wind speed from 3 to 9 m s−1 resulted in a 6.5-fold increase in the speed at which the inert gas sulphur hexafluoride (SF6) was transported through the interstitial air of snow.
 Wind-generated turbulence is likely not the only important driver of ventilation. Emission has been repeatedly observed to maximize near midday (section 2.2). However, oxidation tends to dominate over reduction in daylight (section 3.1.2); the midday emission does not necessarily reflect maximum concentrations of cryospheric GEM. These two facts can be reconciled if the turbulence resulting from atmospheric convective thermal bubbles forced by radiational heating is the primary driver of the midday emission. The stable boundary layer is almost ubiquitous in high-latitude snow-covered environments as a result of radiational cooling [Anderson and Neff, 2008]. Since stable boundary layers act to reduce turbulence [Anderson and Neff, 2008], insolation-driven atmospheric turbulence would tend to maximize near midday in conjunction with the insolation maximum and subsequently subside. Supporting these arguments, the ventilation-driving midday maximum in boundary layer instability is radiationally forced in the one-dimensional snow model of Thomas et al. . The fact that emission and insolation are uncorrelated in the long term (section 2.2) suggests that under sufficiently windy conditions, wind-driven turbulence is more effective at ventilating the snowpack than turbulence generated by atmospheric thermal convection.
 Ventilation-driving turbulence may also be generated by cryospheric thermal convection [Powers et al., 1985]. Such convection is driven by the temperature gradient between the base of the snowpack and the ambient air. This gradient can be significant over land during winter at high latitudes especially at night under the influence of radiational cooling; Helmig et al.  recorded nighttime differences of up to ∼25°C between the surface of the snowpack and a depth of 30 cm. Although Sturm and Johnson  determined that convection was prevalent within a snowpack in Alaska, Helmig et al.  found no concrete evidence that ventilation driven by crysopheric convection impacted cryospheric ozone concentrations at Summit; the effect of wind-driven ventilation was far more obvious.
 The emission of cryospheric GEM is likely affected by the ability of a canopy to inhibit ventilation. The shadowing effect of a canopy (section 3.1.2) likely inhibits ventilation by depressing the generation of radiationally forced convective thermal bubbles. Furthermore, a canopy absorbs momentum [Fatnassi et al., 2006] and dissipates energy throughout its entire depth through the drag it exerts [Yue et al., 2008]. This momentum sink is proportional to the leaf area index and to the square of the velocity [Fatnassi et al., 2006]. Yue et al.  found that turbulence generated just below the top of the canopy diminishes rapidly below the canopy top. Similarly, both momentum and its flux nearly vanish by the bottom of the canopy. Thus, turbulence present at the top of the canopy is likely unable to drive ventilation in canopy-covered snowpacks. Moreover, even if the large-scale horizontal surface-level pressure-driven “sloshing” motions described by Yue et al.  were able to induce ventilation and promote emission, it is probable that little of the emitted mercury would be able to escape the canopy. Even though it is likely that the decreased photoreduction of deposited oxidized mercury caused by a canopy's shadowing effect (section 3.1.2) is the primary reason behind the reduced mercury emission associated with canopies (section 2.2), the ability of a canopy to reduce ventilation by both inhibiting the generation of thermal bubbles and absorbing momentum, and its ability to inhibit the transport of emitted mercury to the canopy top, are likely strong contributing factors to the reduced mercury emission associated with canopies.
Dommergue et al. [2003a] suggested that the abundant GEM produced in a melting snowpack is emitted, while the high concentrations of oxidized mercury suddenly released from the snow grains exit the snowpack with the meltwater; emission accounted for less than 7% of the mercury lost. A combination of the mercury's location within the snowpack, the depth of the snowpack, and the rate of melting, which determines the strength of the percolation within the snowpack, likely determines whether the snowpack's mercury is emitted or enters the meltwater; the likelihood that the entire mercury burden of a snowpack will be emitted probably increases with decreasing snowpack depth and decreasing melt rate.
 Meltwater mercury concentrations are diminished by evasion [Lindberg et al., 2002; Lalonde et al., 2003; Simpson et al., 2007b]. Evasion is likely promoted by flatter terrain, which reduces drainage and increases residence time, and frozen surfaces such as permafrost and sea ice, which impede infiltration [Loseto et al., 2004]. In contrast, the rate of evasion is decreased by oxidants in the meltwater [Lalonde et al., 2003]. Evasion may explain why the average mercury concentration in isolated meltwater ponds on Arctic sea ice from Aspmo et al.  was only half that of surface snow. Furthermore, evasion from such meltwater ponds may cause the summertime increase in concentration of surface-level atmospheric GEM observed in the Arctic [Steffen et al., 2005; Aspmo et al., 2006]. Indeed, Hirdman et al.  found that the highest 10% of atmospheric GEM concentrations at Ny-Ålesund during summer are strongly associated with low-level transport across the Arctic Ocean. Unfortunately, an in-depth analysis of physical and chemical processes determining the fate of mercury in meltwater is beyond the scope of this review. However, it is important to remember that from an ecological point of view, the concentration of mercury in meltwater as it exits the snowpack is less important than the meltwater's mercury concentration as it enters potential methylation sites such as wetlands, peatlands, and oceans [Loseto et al., 2004; Goulet et al., 2007; Mitchell et al., 2008a; Sunderland et al., 2009]; it is methylmercury that bioaccumulates and poses a health risk to human populations [Van Oostdam et al., 2005].
 Given that evasion reduces the concentration of mercury in the meltwater and that mercury is included in the ionic pulse, meltwater sampling must occur at or very near the onset of melting in order to capture the meltwater's maximum mercury concentrations. It seems highly likely that Loseto et al. , Poulain et al. , St. Louis et al. , and Aspmo et al.  all sampled meltwater after the passing of the ionic pulse. Larose et al.  remarked that their first meltwater sample likely corresponded to the very end of the ionic pulse. Unfortunately, the interannual variability of the onset of snowmelt at high northern latitudes is high; 1–2 month differences are common in the central Arctic, while differences often exceeding 2 months are found in more southerly areas, such as Hudson Bay [Drobot and Anderson, 2001].
4. Summary of Physical and Chemical Processes
 In section 3 we discussed in detail the physical and chemical processes that govern the behavior of cryospheric mercury. These processes apply equally to mercury in snow, firn, and ice, although the different media will affect the outcome of these processes. For instance, the vertical motion of GEM molecules will be slower in firn and ice than in snow as a result of their greater densities. We also discussed regional variations of this behavior that are caused by differing environmental conditions. Since section 3 presents a considerable amount of information, section 4.1 provides a short summary of the fundamental processes that are common to all locations. Section 4.2 summarizes regional differences.
4.1. Fundamental Processes Common to All Regions
Figure 1 presents a schematic of the fundamental physical and chemical processes that govern the behavior of cryospheric mercury regardless of location. These processes were discussed in depth in section 3. Progressing from left to right, the schematic indicates the strong likelihood that all GEM deposited onto snow-covered surfaces is emitted immediately. In contrast, most if not all of the deposited PHg is likely retained by the cryosphere.
 The fate of RGM that is deposited onto snow-covered surfaces is far more complicated than the fates of GEM and PHg; deposited RGM may undergo a series of chemical reactions. Deposited RGM is reduced to GEM primarily under UV-B radiation in the 305–320 nm wavelength range. Possible reductants are listed in the schematic. The GEM produced may be oxidized if any of the listed potential oxidants are present in the snowpack. Oxidation tends to dominate over reduction in daylight. Hydrogen peroxide is listed as both a reductant and an oxidant as it behaves as the former in pH-neutral snow and as the latter in acidic snow. Oxidized mercury may be reduced again, restarting the chemical redox cycle. However, subsequent reduction is prevented by halides in the snowpack; these ions tend to stabilize oxidized mercury.
 GEM produced by the cryospheric reduction of deposited oxidized mercury is available for emission to the atmosphere once it has traveled to the snowpack surface. The diffusive transport of GEM molecules is ubiquitous but slow. This transport is enhanced by ventilation, which is driven by turbulence. The turbulence is induced by radiationally induced atmospheric convection and by wind pumping. Apparently, emitted GEM is sourced from only the top ∼2 cm of the snowpack. At the onset of snowmelt, GEM emission to the atmosphere increases significantly. Simultaneously, a considerable fraction of the snowpack's burden of oxidized mercury exits the snowpack in the meltwater's ionic pulse.
4.2. Regional Differences in the Behavior of Cryospheric Mercury
 In sections 2 and 3, both the deposition of mercury to the cryosphere and the emission of cryospheric mercury were reported to vary among three broad categories of location: open high-latitude, open high-altitude, and forested regions. In this section, we summarize the main features of the behavior of cryospheric mercury in these three regions. Differences in this behavior are caused by varying environmental conditions; the chemical and physical processes summarized in section 4.1 apply equally to all three regions.
 While wet and dry mercury deposition processes are active at high latitudes as well as elsewhere, high-latitude coastal and marine regions are notable for the springtime AMDEs, during which important amounts of mercury are deposited locally and episodically. Such sites are also notable for high concentrations of halides in the cryosphere. Halides stabilize the oxidized mercury within the snowpack, reducing emission. Thus, at high-latitude coastal and marine sites the deposition of atmospheric mercury and the retention of cryospheric mercury may simultaneously be enhanced. If these two processes are enhanced simultaneously, significant concentrations of cryospheric mercury are observed.
 During the short high-latitude snowmelt period a significant amount of mercury may rapidly exit the cryosphere with the meltwater's ionic pulse. However, given that the underlying surface at high latitudes may well be impermeable permafrost or sea ice, a considerable fraction of the meltwater's mercury content may be evaded to the atmosphere, especially in areas with flat terrain, before reaching possible methylation sites.
 At high-altitude sites, concentrations of surface-level atmospheric mercury reflect background upper tropospheric concentrations during the convectively quiet fall through midspring period. However, during the convectively active season from late spring through summer, regional pollution may be lifted to these high-elevation sites. This lifting leads to increased concentrations of surface-level atmospheric pollution. Mercury deposition by both wet and dry processes increases. If halides are also emitted regionally, deposition of halides will also increase during this period. The stabilizing effect of halides on cryospheric oxidized mercury leads to a greater retention of the elevated cryospheric mercury concentrations produced by the increased mercury deposition; the springtime lifting of regional pollution may have a double impact.
 If the high-latitude sudden springtime increase in cryospheric mercury concentrations precedes snowmelt, the concentration of mercury in meltwater may be significant. Since the drainage of water in steep terrain is efficient, a substantial fraction of the mercury in high-altitude meltwater is likely to reach possible methylation sites such as lakes before being evaded to the atmosphere.
 Canopies affect the behavior of cryospheric mercury in multiple ways. Mercury throughfall onto snowpacks under canopies is greater than the total deposition from dry and wet processes in adjacent areas. Moreover, the fraction of PHg that is deposited to snowpacks under canopies is enhanced over the fraction deposited in adjacent open areas. This is important, as PHg is strongly retained by the cryosphere. Reduction is depressed under canopies as a result of their shadowing effect. This is considered the primary effect of canopies on the behavior of cryospheric mercury. Snowpacks under canopies also contain more oxidants than in adjacent open areas. Depressing the reduction and enhancing the oxidation reduces the amount of GEM available for emission. Finally, the ability of canopies both to inhibit the generation of thermal bubbles and to absorb momentum reduces ventilation, which also depresses emission. Thus, more mercury is deposited onto snowpacks under canopies than in adjacent open areas, and a smaller fraction of the deposited mercury is emitted to the atmosphere.
 Concerning the fate of mercury in meltwater from canopy-covered snowpacks, the shadowing effect of canopies as well as their ability to absorb momentum likely also dampens the evasion of mercury from meltwater. Thus, if the meltwater's route to potential methylation sites remains covered by canopies, the likelihood that this mercury will reach these sites is increased.
5. Analysis of Mercury Data
5.1. Emission Estimation
Table 5 provides values of daily net loss of mercury from surface snow calculated here from concentrations provided in the literature. Since, outside of melting periods, the daily net loss is likely primarily due to emission (section 2), we refer to the daily net loss as percent emission. Although we calculated percent emission values using consecutive days at a given site during which the concentration of mercury in surface snow declined, values from all sites discussed in an article contributed to the statistics in Table 5 pertaining to that article. Statistics pertaining to “no snow on second day” (hereafter NoS) excluded pairs of days where, to the best of our knowledge, new snow fell on the second day. This avoids the masking of the emission signal by wet deposition (section 2.2). Statistics pertaining to “with snow on second day” (hereafter WithS) used only pairs of days where new snow fell on the second day. Unless stated otherwise, the following discussion of emission estimates relates to the NoS set of statistics.
Table 5. Mercury Lost From Surface Snow Within a 24 h Period
 Investigating the effect of fresh snow falling on the second day of the 24 h period, Table 5 demonstrates that the mean WithS percent emission value is greater than the mean NoS percent emission value at Sainte-Foy, Barrow, Ellesmere Island and Ny-Ålesund. This suggests that at these sites, the amount of mercury deposited through wet processes tends to be small. However, the fact that fresh snow increased the mercury concentration of surface snow more often than it decreased the concentration at Sainte-Foy and Ny-Ålesund indicates the variability at these sites of the amount of mercury deposited through wet processes. At Ny-Ålesund, 75% of the observations from Sommar et al.  of concentrations of mercury in surface snow greater than 30 ng L−1 were recorded during a period characterized by frequent snowfalls.
 Considering only the NoS statistics of Table 5, the four minimum percent emission values of under 10% in Table 5 (Sainte-Foy, Barrow, Ellesmere Island, and Ny-Ålesund from Dommergue et al. ) strongly suggest significant coincident dry deposition. Concentrations of rapidly deposited atmospheric oxidized mercury may frequently be elevated at Barrow as a result of the recurrent offshore leads (section 2.1). In contrast, the low maximum percent emission recorded at Ny-Ålesund from Sommar et al.  is likely not meaningful, given that only a single percent emission value is available from this study. The fact that the next lowest maximum (43%) occurred at Barrow is not unexpected, given this site's coastal location with its presumably elevated snowpack oxidizing and stabilizing halogen concentrations (section 3.1.2). Daily percent emission values greater than 50% were exhibited by eight of the 11 studies, showing that such large bursts of emission are not uncommon. Interestingly, only three studies were characterized by daily percent emission values greater than 80%. However, in the absence of further deposition or fresh snowfalls, a daily loss of 50% of the mercury content of the surface snow can reduce the concentration to a negligible value.
 Mean NoS percent emission values, which provide a summary of Table 5's statistics, range from a low of 15% at Ny-Ålesund using concentrations from Sommar et al.  and 20% at Barrow, up to 51% at Churchill and 52% at Kuujjuarapik/Whapmagoostui using concentrations from Constant et al. . Unfortunately, the high Constant et al.  mean is based on only two 24 h periods. With an increase in sample size to four for Dommergue et al. [2003a], the mean value calculated for the same location decreases to 41%; this lower mean value is expected, given that this is a coastal location. Complicating the interpretation of the high Churchill mean percent emission value is the fact that this mean was derived from loss values that were calculated by Kirk et al. . Their calculation was based on the number of days elapsed since an AMDE. There is no indication that these authors omitted 24 h periods where the second day received fresh snow. Moreover, the windy conditions noted by Kirk et al. , where wind speeds up to 90 km h−1 were recorded, may have generated drifting snow, rendering emission estimates meaningless (section 2.2). Nonetheless, it is quite possible that strong ventilation generated by the windy conditions did, indeed, force significant emission (section 3.2.3).
 Average mean percent emission values for each location are plotted in Figure 2. The average values were calculated from individual means weighted by sample size. If a sample size was not available, a value of 5 was assigned arbitrarily. The geographical distribution of the mean percent emission values is surprising. Given that snowpacks in coastal locations are expected to have higher contents of oxidizing and stabilizing halogen species (section 3.1.2), these locations are expected to have lower mean percent emission values. However, only three of the seven coastal sites are characterized by an average mean below 35%. Furthermore, the two continental sites, where one might expect greater average mean percent emission values, are characterized by an average mean below 40%. It is possible that ventilation is stronger at the coastal locations; greater sample sizes and more information on mercury speciation and reducibility in snowpacks might solve this puzzle. Nonetheless, we can state that the average mean daily percent emission is consistently lower than 55%.
5.2. Mercury Concentrations
Figures 3, 4, 5, and 6 present mean concentrations of mercury in surface snow, seasonal snowpacks, meltwater, and long-term cryospheric records. Since mercury concentrations in snow collected on tables or trays represent wet mercury deposition rather than the concentration of mercury in the surface layer of the snowpack, such concentrations are not included in our analysis. For the purposes of this study, seasonal snowpacks are defined as having accumulated snow over no more than 2 years. The long-term cryospheric records thus consist of ice cores and snowpacks accumulated over more than 2 years. The plotted concentrations are based on data contained in Tables 6, 7, 8, and 9. A logarithmic scale was used to accommodate the considerable variability in the mercury concentrations.
 Caution must be exercised when interpreting Figures 3, 4, 5, and 6. Sampling and analysis techniques vary between studies. Also, many high-latitude studies focus on the spring period, when AMDE-associated deposition is active. Consequently, their mean concentration of mercury in surface snow is most likely elevated over the mean that would have characterized the entire snow season. However, since snowmelt is a springtime phenomenon, it is the springtime mercury concentrations that are relevant. A further complicating factor when interpreting Figures 3, 4, 5, and 6 is the fact that a plotted concentration may represent a single data value or the average of several mean values, where the individual means may themselves be based on multiple data values. For instance, only one sample contributed to the plotted concentration of mercury in meltwater at Kuujjuarapik/Whapmagoostui. On the other hand, 17 mean values from 8 studies contributed to the plotted concentration of mercury in surface snow at Ny-Ålesund, with Berg et al.  providing 2 concentrations, while Dommergue et al.  provided 71 concentrations.
 At a given location the mean values from the contributing studies can be very similar or vary hugely. In Wisconsin the three mean concentrations of mercury in surface snow provided by Bloom and Watras , Fitzgerald et al. , and Lamborg et al.  differ by less than 3 ng L−1. Similarly, minimum and maximum concentrations of mercury in surface snow in northwestern Ontario from St. Louis et al.  and Lalonde et al.  differ by less than 0.5 ng L−1. On the other hand, at McMurdo, the mean concentration of mercury in surface snow over sea ice from Brooks et al. [2008b] is 200 ng L−1 greater than the mean concentration in surface snow from Sheppard et al. . However, the mean concentrations of mercury in seasonal snowpacks in the McMurdo area from Sheppard et al.  and Capelli et al.  differ by less than 3 ng L−1, while mean mercury concentrations in offshore seasonal snowpacks at Barrow from Garbarino et al.  and Douglas et al.  differ by more than 150 ng L−1. In view of this sometimes large degree of variability between the mean values of contributing studies at a given location, and in view of the disparate sample sizes, the average mean mercury concentration at a location was calculated from individual means weighted by sample size. If a sample size was not available, a value of 5 was assigned arbitrarily. In contrast, the maximum/minimum mercury concentration at a location referred to in discussions below is the maximum/minimum concentration of all individual studies at that location.
 Concerning the plotting of the tabulated concentrations, a study's mean concentration is represented in Figures 3, 4, 5, and 6 only if a mean concentration is provided or can be derived. In consequence, since no mean concentration of mercury in long-term cryospheric records is available from Planchon et al.  for Coats Land, and since their expedition is the only one to have worked at this location, Coats Land is not represented in Figure 6. Since the reference locations for Resolute Bay and Cornwallis Island are separated by a mere ∼22 km (Table 1), their average mean mercury concentrations were combined for Figure 3.
5.2.1. Surface Snow
 A minimum mercury concentration in surface snow of less than 2.5 ng L−1 has been recorded at least once at almost all of the sites listed in Table 6. This suggests that emission likely occurs universally. Demonstrating a consistency in the behavior of mercury in surface snow, the highest maximum concentrations of mercury in surface snow are often collocated with the highest average mean concentrations.
Table 6. Concentrations of Total Mercury in Surface Snow
 Average mean concentrations of mercury in surface snow are presented in Figure 3. Over half (12 of 20) of the average mean concentrations are at least 10 ng L−1. In general, these concentrations tend to be high in coastal regions. This is expected, given that AMDEs deposit mercury on or near sea ice and that the concentrations of oxidizing and stabilizing halogen species (section 3.1.2) are also expected to be elevated in coastal regions. The high mercury concentrations measured in the Alps were accompanied by high levels of anthropogenically produced chloride ions (section 3.1.2). The elevated mercury concentrations at the South Pole are likely produced by the deposition of the high concentrations of atmospheric oxidized mercury [Brooks et al., 2008a].
 Low average mean mercury concentrations in surface snow are found in midlatitude continental North America. The snow samples involved were mainly collected in the open. Since canopies did not promote mercury retention in these snow samples (sections 3.1.1, 3.1.2, and 3.2.3), and since the concentration of oxidizing and stabilizing halogens is likely low in this region, these low mercury concentrations are expected. The higher mean mercury concentration in Maine can be explained by the fact that most of the sampling sites involved were affected by canopies. The low mercury concentrations in surface snow at Summit are also expected, given that Summit is located too far inland to be affected by AMDEs (section 2.1), and concentrations of halides in surface snow are extremely low at Summit [Dibb et al., 2010]. Less expected are the low concentrations over the Arctic Ocean northwest of Ny-Ålesund, at Cornwallis Island and at Alert. However, the snow samples providing the Arctic Ocean concentrations were collected from the middle of June through the end of August by Aspmo et al.  (Table 6). The AMDE season with its significant mercury deposition ends with the onset of snowmelt (section 2.1), which occurs in late May or early to mid June in the Arctic (section 3.2.4). Since most of the mercury burden leaves the snowpack soon after the onset of snowmelt (section 3.3), we can assume that many of the concentrations contributing to this mean value had already declined significantly. Since the Cornwallis Island Mel site was characterized by an average concentration of stabilizing chloride ions that is less than half the average concentration in Alpine snow [Ferrari et al., 2002; Poulain et al., 2004], the low concentrations of mercury in surface snow are expected. Furthermore, the variability of the concentrations of both chloride ions and oxidized mercury in the surface snow samples of the Cornwallis Island study by Poulain et al.  indicates that deposition of mercury in the Canadian Archipelago is highly variable. Finally, since Alert is known to experience nonlocal AMDEs, which tend to be characterized by low deposition values (section 2.1), low concentrations of mercury in surface snow at Alert are appropriate.
5.2.2. Seasonal Snowpacks
 Concentrations in Table 7 of mercury in seasonal snowpacks indicate that the highest maximum and minimum concentrations tend to be collocated with the highest mean seasonal snowpack concentrations. Noticeable exceptions are the concentrations from the Tibetan Plateau, Barrow, Cornwallis and Ellesmere Islands, and Ny-Ålesund. These exceptions suggest the presence of spatial and/or temporal variability in mercury concentrations within the seasonal snowpack. Indeed, seasonal differences in mercury concentrations are strong on the Tibetan Plateau [Loewen et al., 2007] and spatial differences are strong near Barrow [Snyder-Conn et al., 1997; Garbarino et al., 2002] and on/near Cornwallis and Ellesmere Islands [St. Louis et al., 2007], while both temporal and spatial differences are important in the vicinity of Ny-Ålesund [Dommergue et al., 2010].
Table 7. Concentrations of Total Mercury in Seasonal Snowpacks
 Average mean concentrations of mercury averaged over the depth of a seasonal snowpack are plotted in Figure 4. Almost half (10 of 23) of the seasonal snowpack mercury concentrations plotted are at least 10 ng L−1. Average mean snowpack mercury concentrations in the Alps are high, as were the Alpine average mean surface snow concentrations. Similarly, mercury concentrations are consistently low in both surface snow and seasonal snowpacks at Alert and at midlatitudes in the middle of North America. In contrast, the average mean seasonal snowpack concentration at McMurdo is considerably lower than its average mean surface concentration. This reduction suggests the occurrence of strong emission. However, this suggestion is not supported by the estimated mean daily loss of mercury from surface snow at McMurdo (Table 5 and Figure 2). In fact, the surface snow mercury concentrations contributing to Figure 3 were measured both onshore and offshore, whereas the seasonal snowpack mercury concentrations were measured onshore only. Table 6 indicates that the mean concentration of mercury in surface snow at McMurdo is threefold higher offshore than onshore when using concentrations from Brooks et al. [2008b] alone and fourfold higher when including concentrations from Sheppard et al. , which were probably derived from samples collected onshore. Thus the dissimilarity of the average mean mercury concentrations in surface snow and seasonal snowpacks at McMurdo likely reflects the variability of mercury deposition (section 2.1) and the variability of concentrations of oxidizing and stabilizing halogen compounds (section 3.1.2) far more than the strength of the emission. On the other hand, the much lower seasonal snowpack mercury concentrations at the South Pole likely do suggest the occurrence of strong emission. The high seasonal snowpack mercury concentrations in southern Greenland versus the low surface snow mercury concentrations in the middle of Greenland are also surprising at first glance. Again, this likely demonstrates the spatial heterogeneity of both mercury deposition and the oxidant/stabilizing content of seasonal snowpacks; chloride ions were observed at both Dye-3 in southern Greenland [Weiss et al., 1975] and station Milcent slightly to its north [Herron et al., 1977]. Sample contamination was unlikely at station Milcent, which is characterized by the highest seasonal snowpack mercury concentrations, as Herron et al.  reported finding low mercury concentrations at Barrow (reported by Weiss et al. ) using the same collection and analysis technique. Weiss et al.  also discussed the issue of sample contamination and deduced, based on the accumulation rates of the five chemical species studied, that serious contamination was highly unlikely. The highest average mean concentration of mercury in seasonal snowpacks was measured at Flin Flon. This extremely elevated concentration reflects a local anthropogenic mercury source; a copper smelter was operational in Flin Flon during the study by Hicks et al. .
 Mercury concentrations in meltwater (Table 8) are consistently the highest in the Alps, whether the minimum, mean, or maximum concentration is considered. This continues the trend of consistently high values at that location in both surface and seasonal snowpack mercury concentrations (Tables 6 and 7). On the other hand, the high maximum mercury concentration in meltwater observed in northwestern Ontario by Allan et al.  is somewhat surprising; it may have been augmented by mercury flushed from the soil (section 3.3).
Table 8. Concentrations of Total Mercury in Meltwater
 Of the average mean meltwater mercury concentrations plotted in Figure 5, almost half (4 of 9) are at least 10 ng L−1. The high average mean mercury meltwater concentrations at Barrow and in the Alps are consistent with the high average mean surface snow and seasonal snowpack concentrations at those locations. Similarly, the lower average mean mercury meltwater concentrations in the middle of North America are expected. On the other hand, the average mean concentration of mercury in meltwater at Ny-Ålesund is surprisingly low, even though only the average of the first eight daily surface meltwater concentrations (8.5 ng L−1), which is sevenfold greater than the average of the last eight daily concentrations (Table 8), and the average snowpack meltwater concentration contributed to the average mean concentration plotted. However, the maximum mercury concentration in meltwater observed at Ny-Ålesund (24.0 ng L−1) is significant.
5.2.4. Long-Term Cryospheric Records
 The minimum and maximum concentrations of mercury in long-term cryospheric records (Table 9) show the same basic spatial distribution as the mean concentrations. However, some of the maximum concentrations are somewhat surprising. The record from Commonwealth Glacier, Antarctic, contains a layer characterized by a mercury concentration of 40 ng L−1, which is far above the mean value of ∼3.8 ng L−1. A similarly valued maximum concentration (34 ng L−1) from the Upper Fremont Glacier in the United States coincided with the eruption of Mount St. Helens [Schuster et al., 2002]. The four maximum concentrations in western Greenland (Camp Century, Site 2, Station Milcent and Dye-3) are extremely high. However, they are entirely possible. Douglas et al.  recorded a mercury concentration of 820 ± 160 ng L−1 on surface hoar 25 m from a lead near Barrow. The 90th percentile of mercury concentrations in both surface hoar and diamond dust collected by Douglas et al.  onshore near Barrow was over 600 ng L−1. The maximum concentrations in diamond dust and surface hoar were 1370 and 975 ng L−1, respectively. Douglas et al.  recorded even higher mercury concentrations from rime on trays (1580 and 5200 ng L−1) and on an airplane wing (15,500 ng L−1). The two highest rime concentrations were observed on the same non-AMDE day.
Table 9. Concentrations of Total Mercury in Long-Term Cryospheric Records
 Average mean concentrations of mercury observed in long-term cryospheric records are shown in Figure 6. Just under half (4 of 10) of the average mean concentrations are at least 10 ng L−1, while two further average mean concentrations are at least 7.5 ng L−1. Given that the Alps had consistently high mercury concentrations in surface snow, seasonal snowpacks, and meltwater, the far lower average mean mercury concentration in the long-term record appears surprising. However, the concentrations of mercury in surface snow, seasonal snowpacks, and meltwater were all based on samples collected at altitudes no higher than 2448 m above sea level (asl) (mean altitude is 1645 m asl), while the snow and ice cores were taken from the Col du Dôme glacier at 4250 m asl. Thus, the Col du Dôme glacier is apparently sufficiently elevated that it remains uncontaminated by the regional pollution that affects the lower-altitude sites, even during the spring/summer convective season (section 2.1). However, given that Saharan dust events, which may be accompanied by elevated concentrations of PHg, are known to affect this glacier [Maupetit et al., 1995], and that PHg is strongly retained by the cryosphere (section 3.1.1), one might expect to find at least a few layers characterized by high mercury concentrations in the long-term cryospheric record. Unfortunately, if such layers did exist, their presence may well have been hidden by the prevailing low free-tropospheric mercury concentrations and the use by Jitaru et al.  of an ∼5 year averaging time period.
 The six long-term cryospheric records in Greenland are also interesting. Summit, in central Greenland, and Crete, ∼200 km to its southeast, are characterized by the lowest average mean mercury concentrations. However, the average mean concentrations from the four sites in western Greenland are considerably higher. It can be argued that these four western concentrations, along with the average mean concentration of 49 ng L−1 from Weiss and Bertine  (Table 9) that was not plotted due to its imprecise Greenlandic location, are unreliable, given that the studies responsible for these concentrations are from the 1970s and that, therefore, the samples were contaminated and/or analyses were insufficiently accurate due to a lack of scientific sophistication [Jackson, 1997]. However, the four high concentrations plotted are all closer to the west coast than Summit and Crete. Their coastal location likely leads to enhanced deposition of both mercury associated with AMDEs (section 2.1) and mercury-retaining halogen species (section 3.1.2) associated with seawater. Indeed, chloride ions were found in the records by Herron et al.  at station Milcent and by Weiss et al.  at Dye-3, both of which stations are in southern Greenland. Herron et al.  deduced that their chloride ions were of marine origin. Furthermore, Carr and Wilkniss  remarked that the variation in the mercury concentrations obtained from sites 80 km apart at and near Camp Century (the westernmost site in Greenland) by Weiss et al.  may easily represent the spatial heterogeneity of mercury deposition rather than a significant increase in mercury deposition over time, as claimed by Weiss et al. . Furthermore, Herron and his coauthors, using identical sampling and analysis techniques [Herron et al., 1977], observed low mercury concentrations at Barrow (mean concentration in surface snow of ∼12 ng L−1 reported by Weiss et al. ; Table 6). Weiss et al.  discussed the issue of sampling technique and deduced, based on the accumulation rates of the five chemical species examined, that it was highly unlikely to have been a serious problem in their study. The fact that at Dye-3 the mean from Weiss et al.  is fivefold greater than the mean from Appelquist et al.  may not be important, as the highest mean from Summit, from an article published in 2008, is over 20-fold greater than the lowest mean at that site. Nonetheless, it would be interesting to remeasure the mercury concentration in ice cores from these four and other sites in western Greenland.
 A noteworthy result from the Antarctic ice cores is the dependence of mercury concentration on climactic period. Both Vandal et al.  and Jitaru et al.  observed a clear distinction in the behavior of mercury between the coldest and warmest periods. Jitaru et al.  estimated that total net deposition increased almost fivefold from the warmer to the coldest periods. This temperature dependence may be related to the availability of refreezing sea ice leads, which are associated with the release of bromine radicals (section 2.1), and/or to the fact that the formation of HgBr2, which deposits fairly rapidly, requires temperatures lower than ∼260 K [Jitaru et al., 2009].
6. Measurements Wish List
 A comprehensive field campaign undertaking the measurements listed in Figure 7, as contemporaneously and interoperably as possible, would provide a complete instantaneous picture of mercury in the cryosphere and atmosphere, and its flux between the two media. Optimally, these sets of measurements should be performed at the same time of day for as long a period as possible. Through this comprehensive study of speciated data, the reducibility of RGM and PHg and the rate of conversion between RGM and PHg could be deduced. The relative importance of dry and wet deposition could also be estimated for each species. If possible, campaigns would be conducted in a wide variety of locations. For instance, data from the middle of regions of both first-year and multiyear sea, and from a variety of high-latitude and midlatitude urban and remote areas both near coastlines and inland, at varying altitudes and under a variety of canopies would be extremely informative. To promote the comparison of results from individual field campaigns, it would be useful if a standard depth were defined for the surface layer, e.g., 2 cm; definitions of the surface layer ranging from 1 to 15 cm have been encountered in the literature to date.
7. Impact of Climate Change
 Given that the behavior of mercury in the cryosphere is complex, it is difficult to estimate what effect a warming climate will have on this behavior. At high latitudes in a warmer climate, fewer AMDEs may occur, as the season where sea ice refreezes in the presence of sunlight will grow shorter; AMDE-associated mercury deposition may decline. On the other hand, during this shorter AMDE season, the sea ice will probably be thinner and more dynamic, with leads forming and refreezing more frequently. The frequency of AMDEs would likely then increase, augmenting deposition of mercury onto high-latitude snowpacks. With more areas of open water, deposition of halogen species onto high-latitude snowpacks would likely increase. Thus, even though the change in high-latitude AMDE-associated mercury deposition in a warming climate is uncertain, the retention of the deposited mercury would likely be enhanced.
 At all latitudes, possible changes in many environmental parameters will affect mercury emission. For instance, the snowpack season can be expected to grow shorter as the climate warms. To assess the impact of such a change on mercury emission, the rates of emission from the underlying surface and from the cryosphere need to be compared; their difference will vary according to the nature of the subsurface. However, if coniferous forests extend their range in a warmer climate so that they cover a greater fraction of the global snowpack, cryospheric mercury emission will decline globally. Similarly, if the changing climate is accompanied by more frequent snowfalls, mercury will be trapped more often through burial. If wind patterns change in the future, ventilation and emission will be enhanced at some locations and depressed at others. Emission of mercury from snow will also decrease if concentrations of atmospheric PHg increase as a result of greater numbers of organic and inorganic aerosols. If convection becomes more vigorous earlier in the spring as the climate warms, the mercury content of high-altitude snowpacks will increase earlier in the season; the resulting meltwater may contain even higher concentrations of mercury.
 In this review article we have discussed in detail the universally applicable physical and chemical processes that govern the behavior of mercury in the cryosphere, whether snow, firn, or ice. The issue of the emission of mercury deposited onto cryospheric surfaces is of particular importance at the moment, given that our climate is changing and that Asian anthropogenic emissions and, consequently, Northern Hemispheric mercury deposition are expected to increase in the future.
 Although many questions remain unanswered, we do know that mercury deposited to the cryosphere is reduced, primarily in the presence of UV-B radiation. Oxidation may follow. Cryospheric oxidized mercury is stabilized by halides. Emitted GEM molecules, which are likely sourced from the top ∼2 cm of the snowpack, are transported to the surface by diffusion and ventilation. The fraction of mercury that is not emitted enters either long-term storage or the meltwater. At the onset of snowmelt, GEM emission increases significantly, while oxidized mercury leaves the cryosphere rapidly with the meltwater's ionic pulse. Some of the mercury in the meltwater is evaded; significant mercury methylation is possible only if the mercury in the meltwater reaches possible methylation sites before being evaded.
 Since the physical and chemical processes discussed in this review are completely general, a dynamic parameterization based on these processes would be valid at all locations and times. Given that the variations in the behavior of cryospheric mercury between open high-latitude, open high-altitude and forested regions are caused by differing environmental conditions, such a dynamic physically based parameterization should be able to reproduce these variations. The dynamic aspect of such a parameterization is also important when simulating the behavior of cryospheric mercury under different climatic conditions.
 This study was funded by the projects INCATPA and CARA. The former project used funding from Government of Canada Program for International Polar Year. The authors would like to thank Lars R. Hole of the Norwegian Meteorological Institute for providing meteorological data measured by the Alfred Wegener Institute in Ny-Ålesund. Extensive discussions with Florent Domine, Aurélien Dommergue, Tom Douglas, Daniel Figueras-Nieto, Kaj Hansen, Alexandre Poulain, Andrei Ryjkov, and Kenjiro Toyota were very helpful. We are indebted to the three anonymous reviewers who made such perceptive comments and greatly increased the quality of the manuscript.