Mercury in the snow and ﬁrn at Summit Station, Central Greenland, and implications for the study of past atmospheric mercury levels

. Gaseous Elemental Mercury (Hg ◦ or GEM) was investigated at Summit Station, Greenland, in the interstitial extracted from the perennial snowpack (ﬁrn) at depths from the surface to 30 during 2006. Photolytic production and destruction of Hg ◦ were observed close to the snow surface during summer dark


Introduction
Mercury is a persistent and toxic heavy metal present in the environment in various chemical forms. In the atmosphere, gaseous elemental mercury (Hg • , GEM) is the predominant form of mercury (>95%) and has a global atmospheric residence time of about 6-24 months (Lamborg et al., 2002). The relatively inert GEM can undergo long-range transport whereas oxidized species of mercury such as particulate mercury (PM) and reactive gaseous mercury (RGM) are subject to rapid deposition near their emission sources. Atmospheric mercury depletion events (AMDEs) were first observed at Alert, Canada (Schroeder et al., 1998). They commonly occur during the spring in polar environments and are characterized by rapidly decreasing concentrations of GEM in the atmosphere. The decrease is caused by photochemically initiated oxidation reactions involving marine halogens (Lu et al., 2001;Lindberg et al., 2002;Skov et al., 2004) that transform GEM to RGM and PM. These oxidized species can be deposited onto snow surfaces. As a result, the arctic seasonal snowpack is suspected to contribute to the contamination of the aquatic reservoir during snowmelt. Conversely, the polar snowpack can also act as a source of GEM to the atmosphere. Direct photoreduction of divalent mercury (Hg(II)) complexes deposited onto snow surfaces lead to an emission of GEM to the atmosphere. AMDEs have only been observed in coastal areas. Brooks et al. (2007) reported preliminary observations about mercury chemistry at the South Pole, but X. Faïn et al.: Mercury in the snow and firn at Summit Station, Greenland there is still a clear lack of data on inland Greenland and Antarctica, even though these ice sheets represent a surface of ∼14 millions km 2 of snow which could strongly interact with the lower atmosphere and influence the global cycle of mercury.
Atmospheric mercury pollution over the last 150 years has greatly increased mercury deposition to terrestrial and aquatic ecosystems in remote areas. Modern measurements combined with historical records from lake sediments and peat suggest a threefold increase in mercury deposition since pre-industrial times (Engstrom and Swain, 1997;Lamborg et al., 2002). A single ice core profile from the Freemont Glacier (Wyoming, USA) showed a 20-fold increase since 1840 (Schuster et al., 2002) for Hg(II) deposition. Mason et al. (1994) and Mason and Sheu (2002) used archives of deposited mercury and modeling studies to estimate an increase in the global atmospheric reservoir's GEM concentrations by a factor of 2-5 since the beginning of the industrialized period. Only total and reactive mercury can be investigated in sediment, peat and ice archives, however, postdeposition processes may strongly modify the chemical information trapped into these records (see e.g. Biester et al., 2007). Consequently, these archives do not provide direct information on the evolution of the global background concentration of atmospheric GEM over time.
The potential of the polar ice sheets to serve as an archive for the reconstruction of past atmospheric GEM composition has been well recognized. From the top surface to ∼60-120 m depth is the firn, an openly porous and permeable media through which air can diffuse. The solid ice is located below the firn, and the transition between firn and ice is called the close-off. Ice core records of trapped gases and interstitial firn air have been used extensively to study the past history of atmospheric greenhouse gases (Schwander et al., 1988;Siegenthaler et al., 2005). In principle, both firn air and ice cores should also be useful for quantifying the histories of atmospheric GEM concentrations.
However, to correctly interpret GEM measurements from the deep firn air or in the ice air, we first need to quantify any perturbations in the GEM records arising from surface phenomena. For example, chemical processes leading to production and/or destruction of GEM close to the snow-air interface might distort atmospheric properties preserved in the glacial records. More generally, physical and chemical modifications of GEM signals in the lower atmosphere, the surface snowpack and the deep firn have to be understood to determine the transfer function. The GEM transfer function is defined as the relation between GEM atmospheric concentrations and GEM concentrations in deep ice bubbles. This function has to be determined in order to reconstruct the past history of GEM in the atmosphere.
We measured, for the first time, GEM concentrations from the top surface to a depth of 30 m at Summit Station, Central Greenland, during two field campaigns during summer 2005 and spring 2006. The aims of this study were (i) to docu-ment the chemical processes involving GEM in the shallow firn air and (ii) to understand the influence of these surface processes on the GEM levels in the deep firn air. We also discuss the feasibility of using the GEM transfer function to reconstruct past atmospheric GEM levels from deep firn air and air trapped in ice.

Study site
GEM was investigated on the Greenland ice sheet at the Greenland Environmental Observatory (72.6 • N, 38.5 • W, 3200 m elevation) from 23 to 29 July 200529 July (summer 2005 and 29 May to 6 June 20066 June (spring 2006. In 2005, atmospheric and snow air samplings were carried out close to the Science Trench: a clean area where access was restricted. In 2006, atmospheric air, snow air and firn air were sampled in a remote location 10 km away from the station. These measurements were part of the collaborative firn air sampling campaign "Firn structure, interstitial processes, and the composition of firn air at Summit, Greenland" led by CREEL, Hanover, N.H. (USA).

In situ air analysis
We measured GEM levels with two Tekran gas phase GEM analyzers (Model 2537A; Tekran Inc., Toronto, Canada). One was dedicated to ambient and shallow firn measurements, while the other was used exclusively for deeper firn measurements. A precision mass flow meter supplies the 2537A with a sample volume referenced to NTP (Normal Temperature and Pressure: 20 • C, 1 atm). The pre-filtered air stream (soda lime trap and 0.2 µm Teflon particle filter) is collected on two gold cartridges. GEM is thermally desorbed from the cartridges and detected by cold vapor atomic fluorescence spectrometry (AFS) at 253.7 nm (Tekran, 1999). Dual gold cartridges allow alternate sampling and desorption, resulting in continuous measurement of GEM on a predefined time base. The set-up, accuracy and precision of this instrument have been assessed during field intercomparisons at an urban/industrial site (Schroeder et al., 1995) and at a remote marine background location (Ebinghaus et al., 1999). The Tekran analyzer was operated with a 5-min sampling frequency and the air was sampled at a flow rate of 1 l min −1 . The analyzer was calibrated every 25 h with an internal automatic permeation source injection. The detection limit for GEM in this operational mode is roughly 0.15 ng m −3 (Aspmo et al., 2005). All GEM data discussed in this paper are presented with one StDv.
We measured GEM concentrations between depths of 40 and 330 cm below the snow surface using GAMAS probes (Gaseous Mercury in Interstitial Air in Snow), devices dedicated to the sampling of GEM and temperature measurements in the air of snowpacks (Dommergue et al., 2003a). X. Faïn et al.: Mercury in the snow and firn at Summit Station, Greenland

3443
This system has been used successfully in different Arctic sites, for example, at Station Nord, Greenland (Ferrari et al., 2004a, b), Kuujjuarapik, Canada (Dommergue et al., 2003b) and Col de Porte, French Alps . Ten GAMAS probes and an atmospheric sampling line were connected to the Tekran analyzer using an 11-port Teflon solenoid switch. This set-up allowed a cyclic sampling of each probe every 10 min.

Firn air sampling
Firn air was sampled using established methods (Schwander et al., 1988;Battle et al., 1996;Butler et al., 1999) from one hole at the depths of 15, 25, 30, 40, 50, 58, 63, 66, 70, 72, 74, 76, 78 and 79.5 m, but only the top three depths will be discussed in this manuscript. Briefly, a 5-m long bladder was lowered into the borehole after drilling to the sampling depth, and was pressurized, effectively sealing the borehole. Two Dekabon ® (polyethylene/aluminum composite) lines were used to pump firn air from a space left immediately below the bladder. These lines drew air from two separate openings separated by a horizontal stainless steel baffle nearly as wide as the hole. Air was pumped from the upper opening at ∼20 l min −1 and was directed to waste after measuring the CO 2 concentration (in situ measurements using a LICOR instrument). When CO 2 levels stabilized, indicating effective removal of contamination by younger and/or ambient air, sampling started from the lower opening. Pumping from the upper opening continued during sampling to remove any air leaking from within or around the bladder. It also served to keep sampled air out of contact with the bladder itself. Firn air was collected in pressurized flasks for analysis of other gas species at a flow of 4 l min −1 during approximately one hour. Afterwards, the sampling line was connected to a Vacuubrand PTFE pump (MZ-2C) and GEM analyses commenced. The pump delivered firn air to the inlet of a 2537A Tekran analyzer at a flow rate of 1 l min −1 . Blanks of the Dekabon ® sampling line were checked both at the beginning and at the end of the field work. They were 0.08±0.13 ng m −3 (n=8, before sampling, on 25 May) and 0.01±0.06 ng m −3 (n=18, after sampling, on 1 June). We measured blanks of the PTFE pump before and after sampling at all depths to quantify any contamination introduced by this additional pump.

Pressure influence on GEM analysis
Summit Station elevation is 3200 m, and an atmospheric pressure of ∼675 mbar was observed both during July 2005 and June 2006. While our 2537A analyzers are optimized for GEM determination in the air from the sea-level atmosphere up to 2000 m (Tekran Inc., Toronto, Canada), Ebinghaus and Slemr (2000) have shown that low pressures can influence the operation of these analyzers. Using ground calibrations, these authors reported a pressure dependency of the AFS detector of about 0.1% mbar −1 over the tested range of pressures (980 to 620 mbar). Unfortunately, we could not use such results because we calibrated both Tekran analyzers automatically using an internal calibration source every 25hrs at Summit Station. Instead, we carried out laboratory studies to obtain information about the effect of low pressure on internal calibrations. While both analyzers were at ambient pressure (∼990 mbar), all the inlet and outlets (including internal calibration inlets) were introduced into a lowpressure chamber. This equipment, located at the Centre de Recherche du Service de Santé des Armées (CRSSA, La Tronche, France) and initially designed for biological studies, had a volume of 2000 l and could reach a stable pressure from ambient to 500 mbar in a few minutes. The accuracy of the pressure measurement in the chamber was ±6 mbar. We first did internal calibrations at 675 mbar to reproduce the conditions of Summit Station. Then, we performed 14 manual injections (∼120 pg of Hg • ) using a primary mercury vapor source (Tekran 2505), using the manual injection port. Because the instruments were at a higher pressure than their internal plumbing, we also sampled mercury free air to check that no invasion of ambient air occurred. The response of both instruments was found to be pressure dependent: at 675 mbar we observed an increase of the GEM signal by 6.2 and 9.4% for the two analyzers (serial numbers 209 and 238, respectively). All data discussed in this study were pressure corrected.

Snow sampling
A 2-m snowpit was dug in the clean air sector of Summit Station, 500 m south of the Science Tower. The pit was sampled for total mercury (Hg T ) on 23 July 2005, in a location where the snow had not been previously disturbed. All samples were immediately stored in the dark at −20 • C until analysis. For Hg T samples, we used ultra clean Teflon bottles and clean snow sampling procedures (Ferrari et al., 2000). Hg T includes species such as HgCl 2 , Hg(OH) 2 , HgC 2 O 4 that are easily reducible by SnCl 2 or NaBH 4 , as well as stable complexes such as HgS and Hg 2+ bound to sulfur in humic compounds, and some organomercuric species (Lindqvist and Rodhe, 1985). Hg T was determined at the Department of Environmental Science of the University Ca'Foscari of Venice (Italy), using an Element ICP-SFMS (Thermo Finnigan MAT Instrument, Bremen, Germany). Planchon et al. (2004) gave a detailed description of the analytical technique. Instrumental calibrations were carried out with Hg standards prepared from serial dilutions of a monoelemental Hg 2+ solution at 1000 mg l −1 (CPI International Santa Rosa, CA, USA). Using these techniques, the laboratory achieved a lower detection limit of ∼0.18 ng l −1 . The precision of the measurements was estimated to be 15% (the relative standard deviation of five replicates). Snow samples were melted just prior to analysis to minimize mercury transformation during storage.

Meteorological conditions
During both field measurement campaigns, there was 24-h daylight. During summer 2005, atmospheric temperatures and wind speeds exhibited two different periods. From 18 to 25 July wind speeds were always below 3 m s −1 while from 26 and 27 July wind speed was typically between 3 and 5 m s −1 . From 18 to 22 July atmospheric temperatures ranged from −20 • C at night to −5 • C during the day. From 24 to 27 July 2005, air temperatures during the days were close to 0 • C. During spring 2006, the daily temperature ranged between −30 • C at night and −15 • C during the day. Weather conditions were generally calm and clear with a maximum wind speed of 4 m s −1 (measured two meters above the snow surface). The only exception was on 6 June when a maximum wind speed of 6 m s −1 was recorded. The hemispherical integrated incoming radiation measured at a height of two meters during both field campaigns are shown, along with the shallow firn air data, in Figs. 3 and 4.

Snow and firn characteristics
The physical properties of the Summit snowpack have been intensively studied before. We measured density in 2-m pits both in summer 2005 and spring 2006, finding a pattern similar to that of Albert and Shultz (2002). Our 2-m pit represented about three years of snow accumulation. In May and early June, most of the top 50 cm of snow is fine-grained wind-packed snow. Low-density hoar layers form occasionally during summer. In accumulated firn from previous years, the layers of hoar from the summer alternate visibly with high density wind-pack layers characteristic of winters. The In summer 2005, similar hoar layers were observed centered at ∼80 and ∼150 cm depths. At Summit, the permeability generally increases with depth by more than an order of magnitude in the first 3 m of the snowpack (Albert and Shultz, 2002), and generally decreases below 3 m.
To assess the role of total mercury in chemical processes, we sampled the full 2 m of our pit for Ca 2+ and Na + abundance, as well as for Hg T in summer 2005. We used Ca 2+ and Na + measurements (not shown here) for pit dating. Ca 2+ concentration peaks at ∼10, ∼90 and ∼150 cm depths indicating spring layers, while Na + concentration peaks at ∼40, ∼110 and ∼170 cm depths showing winter layers were observed. Hg T concentrations shown in Fig. 2 ranged from 2.6 to 20.2 ng l −1 , with a mean value of 10.3 ng l −1 and a mean analytical uncertainty of 0.7 ng l −1 . We collected duplicate samples at the snow surface and at 10, 20, 60, 100, 140, and 180 cm depths, the concentrations reported at these depths on Fig. 2 are the means of duplicate analysis and error bars represent the variability between these duplicates. While there are substantial uncertainties in our measurements, we observed neither a seasonal cycle, nor an annual change in Hg T deposition over this period.
Although these are the first measurements of Hg T made at Summit, reactive mercury (Hg R ) has been studied at this site in the past, both in the first seven meters of the snow and firn (Mann et al., 2004) and in the deep firn (Boutron et al., 1998). Hg R corresponds to the fraction of mercury bound in complexes that are easily reducible by SnCl 2 or NaBH 4 . Examples include HgCl 2 , Hg(OH) 2 and HgC 2 O 4 (Lindqvist and Rodhe, 1985). Reactive mercury represents less than 10% of Hg T in the Summit snowpack. Mann et al. (2004) reported a high variability of Hg R concentrations in the first 2 m of the snowpack, with values ranging from 0.25 to 0.68 ng l −1 (mean value of ∼0.46 ng −1 ).

GEM in the atmosphere
We measured GEM levels in atmospheric air at Summit Station during summer 2005 and spring 2006. Concentrations were stable, and no atmospheric GEM depletion events were observed. Mean concentrations were 1.61±0.14 ng m −3 (n=1102) from 27 May to 6 June 2006, and 1.80±0.14 ng m −3 (n=112) from 23 to 29 July 2005. GEM atmospheric levels measured during both spring and summer at Summit were a bit higher than concentrations reported for Alert at the same season . This slight difference (∼10%) was not confirmed by intercalibrations and comparison measurements of the instruments used at Alert and Summit. However, the air masses at Alert and Summit have different origins: the chemical loading to the atmosphere above Summit is influenced by air masses arriving at high altitude (500 hPa) (Khal et al., 1997). Moreover, springtime displays a strong variability of GEM concentrations in the atmosphere at Alert due to the atmospheric mercury depletion events (AMDEs). These phenomena, leading to a complete depletion of GEM in the atmosphere, occur simultaneously with the post-solar sunrise destruction of ozone. Helmig et al. (2007b) investigated the surface ozone mixing ratio at Summit Station from 2000 to 2004 and reported concentrations ranging from ∼40 to ∼70 ppbv without any ozone depletion event. No atmospheric ozone and mercury depletion events were observed at Summit, indicating no strong variations in their atmospheric concentrations throughout the year. In summer 2005, diel variations of GEM concentrations were observed from the surface to a depth of 200 cm. Of the many recordings, only very few measurements fell below the atmospheric level at a depth of 40 cm; most measurements in the near-surface in July were higher than the ambient atmospheric concentration. Due to equipment maintenance, we were not able to collect data on 22 and 23 July. In the following section, we investigate the different roles of physical processes and chemical mechanisms in the daily variations of GEM concentrations in the SIA.

Chemical or physical processes in the SIA?
Diel variations of GEM concentrations were observed both during spring and summer (see Figs. 4 and 5). Five possible phenomena could potentially explain the variations observed: i) adsorption of GEM on snow grains, ii) natural ventilation, iii) diffusion within the deeper snowpack, iv) chemical processes leading to destruction and/or production of GEM and v) advective flow induced by the sampling procedure. Due to the high adsorption energy of GEM on snow evaluated at ∼61 kJ mol −1 (Ferrari et al., 2004a), the adsorption of GEM is extremely reduced and cannot explain these observations. Similarly, natural ventilation could not induce the variation of GEM concentrations observed. Albert and Shultz (2002) showed at Summit Station that gas transport in high permeability layers 15 cm beneath the snow surface under low winds (∼3 m s −1 ) was not influenced by natural ventilation, but controlled by diffusion.
To identify the potential role of diffusion in the interstitial air in the snow and firn, we simulated the diffusion of the diel variation of GEM recorded in spring 2006 at a depth of 40 cm. We first determined diffusion constants characterizing GEM and the firn structure at Summit Station. The GEM diffusivity in air at Summit Station D Surface (T ,P ) was determined using the value reported by Massman (1999) corrected for Summit temperature and pressure (Schwander et al., 1988).  (1) gives the GEM diffusivity in the atmosphere above the firn surface. To obtain the effective diffusivity of GEM in the SIA (D eff ), we used the following relation: where φ is the snow porosity determined using the densities of snow and ice: τ is the firn tortuosity. The tortuosity of a porous medium represents the complexity of the pathway, and is commonly calculated as the ratio of the mean path length to the minimum possible (straight line) path length. Albert and Shultz (2002) reported that the tortuosity of the surface wind pack at Summit in June was ∼2.0. A value of 2.0 is the maximum tortuosity value that can be considered in the first several meters of the snowpack, because the top 50 cm of snow is wind-pack, and is less permeable than the underlying layers at Summit, as can be seen in Fig. 2. Considering a tortuosity of 1.4-2.0 and a snow porosity of ∼0.7, we obtained D eff for GEM of 0.06-0.08 cm 2 s −1 . Diel cycles of GEM concentrations in the SIA showed a time shift with increasing depth both during summer (Fig. 4) and spring (Fig. 5), suggesting that diffusion was occurring. To better assess the transport of GEM by diffusion in the SIA, we modeled this physical process from 40 to 270 cm depths during spring 2006. In this model, we used GEM spring measurements at a depth of 40 cm (GEM 40 ) and a constant concentration of 0.1 ng m −3 at a depth of 270 cm as boundary conditions. The modeled diffusion of GEM 40 deeper in the snowpack is shown on Fig. 6. A comparison of modeled data and measurements is also represented. Peak concentrations from field data exhibit a slight shift in time when depth increases but modeling results show that this shift should be larger if only diffusion took place in the SIA. This effect could be due of induced air advection in the firn, as we discuss in the following paragraph.  showed that interstitial air sampling in snow with sampler flow rates of 1 l min −1 (and higher) draws in a substantial amount of ambient air into the sampler, even when one assumes that the sampler has perfect contact with the surrounding snow. Thus, we investigated the role of induced interstitial advection in the shallow firn. Using the multidimensional model (Albert, 1996;, along with the measured permeability profile depicted in Fig. 1 for the permeability of the top two meters for spring 2006, and the published permeability profile deeper in the firn at Summit (Albert and Shultz, 2002), we investigated the impacts of sampler-induced advection on the concentration profiles measured in the firn air during spring 2006. Using a sampler intake rate of 1 l min −1 for each simulation, Fig. 7 depicts the air flow patterns when the intake is at depths of 40, 120, and 160 cm in the firn. The closer the intake is to the atmosphere/snow interface, the more air is drawn vertically down through the surrounding snow into the sampler intake, thus diluting the sampled air at the sampled depth with air from above. This is clearly evident when the sample intake is at a depth of 40 cm (Fig. 7a); most of the air into the inlet is coming directly down from the near-surface snow. As the inlet gets deeper, the contribution from the snow-air interface decreases, and the contribution from nearby permeable layers (whether above or below the inlet) increases. Considering the inlet at a depth of 120 cm (Fig. 7b), there is still surface air coming down, but a greater part of the sample comes from the high-permeability layer locate between 50-95 cm depths in the snow. At a depth of 160 cm (Fig. 7c), much of the air sampled comes from the high-permeability layer above the intake: this high-permeability layer serves as a channel for lateral flow in the firn. Sampler-induced advection transports down photochemically-induced changes in GEM concentrations from the near-surface snow. On one hand, we assume that diel cycles observed from 80 to 160 cm depths during spring 2006 are due to sampler-induced advection. The decrease with depth of the amplitude of the diel GEM variations is consistent with our modeling, showing less influence of the forced advection as depth increases. We note that dilution/mixing of interstitial air measurements by flow samplers is not unique to this study; in fact it is an unavoidable artifact of all firn air measurements that have been carried out by flow samplers to date Dominé et al., 2008). On the other hand, sampler-induced advection could not explain the decrease in minimum GEM concentrations with depth measured close to midday. Diffusion modeling, reported on Fig. 5, could not reproduce either this destruction of GEM, which may reflect chemical mechanisms involving GEM from 40 to 270 cm that will be discussed in the following section.
Snow permeability data were not available for summer 2005, therefore we could not model the impact of sampler induced advection on the GEM data. However, the diffusion model used previously to investigate GEM data from spring 2006 (Fig. 6) was applied to the data collected during summer 2005, and demonstrated that diffusion of GEM was too slow to explain the diel cycles deeper in the SIA. Similarly to spring data, we assume that diel GEM variations below a depth of 80 cm could be an artifact due to sampler-induced advection. Forced advection could also lead to an underestimation of GEM concentrations in the first meter due to the mixing of low-concentrated air from the atmosphere and high concentrated air from the SIA. Thus, the increase of the mean GEM concentrations with depth during summer 2005 could not be explained by diffusion, but may be an artifact of sampler-induced advection. Dark chemical production of GEM should also be considered as a possible mechanism for the observed increase in GEM concentrations.  In summary, measurements in the field when compared to gas transport modeling lead to the evidence that natural diffusion was not the main process driving the evolution of GEM concentrations in the SIA. On the other hand, photochemical changes in GEM that occur in the top centimeters to tens of centimeters of snow are likely advected down into the sampler, giving the appearance that photochemical reactions are occurring deeper in the snow than is actually the case. Despite this sampler-induced artifact, there is evidence of chemical processes involving GEM in the SIA, with both photolytic mechanisms close to the surface and dark processes in the first three meters of the snowpack.

Chemical mechanisms within the snowpack
The photochemical processes in snow and the exchanges of reactive trace gases between the SIA and the Atmospheric Boundary Layer (ABL) have been largely investigated at Summit Station . Galbavy et al. (2007) measured irradiance in the snow and firn at Summit, and found actinometry e-folding depths near 12 cm. Less than 2% of monochromatic light penetrates deeper than the top half-meter at Summit. Thus, photochemistry can directly play a role from the top surface to ∼60 cm depth in the snow at Summit. In this area, defined as the sunlit snow, diel variations of GEM observed during spring and summer in the SIA were probably the result of the co-existence of GEM oxidation and Hg(II) reduction. These two competitive phenomena have already been observed in coastal snowpacks in the Arctic (Dommergue et al., 2003b;Ferrari et al., 2004b). In our case, oxidation was predominant during spring and Hg(II) reduction was predominant during summer. According to e-folding depth values reported by Gavaldy et al. (2007), there is no evidence that significant, naturally-occurring photochemical reactions involving Fig. 7. Modeled air flow patterns into the sampler intake, given the measured permeability profile at Summit, when the intake is at depths of 40, 120, and 160 cm in the firn. Samples drawn from the top two meters of firn include firn air that has undergone photochemical reactions at shallower depths in the snow. GEM could occur below ∼60 cm depth. However, at Dome C (Antarctica), irradiance attenuation in the shallow firn depends on the snow layers considered: at higher depths, Warren et al. (2006) observed a lower irradiance attenuation related to an increase in snow grain size. A similar study is necessary at Summit Station to better characterised the photochemical processes in the snowpack. Assuming that light does not penetrate below ∼60 cm depth, the diel variation of GEM concentrations from 80 to 200 cm depths in the SIA is probably due to the transport of the photochemically-induced gas composition in the top half-meter to deeper depths by the sampler flow, as shown by our modeling (see Fig. 7). This effect occurred both during summer 2005 and spring 2006. However, the decrease of mean GEM levels during spring 2006, and their stabilization to zero (see Fig. 3) from 200 to 270 cm could not be explained by the transport of the depleted air but only by dark oxidation processes in the SIA. The active photochemistry involving both GEM and Hg(II) complexes implies that one must exhibit great caution when using Hg R and Hg T profiles to infer depositional histories. Hg(II) species could be perturbed by post-deposition processes occurring close to the surface, leading to Hg R or Hg T profiles in the firn that would not be simply linked to the history of mercury deposition on the surface. In the next sections we discuss the photolytic and dark mechanisms which could be involved in the destruction and production of GEM.

GEM depletion within the snowpack
In the sunlit snow, we observed GEM depletion with concentrations lower than atmospheric ones and a diel evolution which was anti-correlated to solar irradiation during both summer 2005 and spring 2006. During spring 2006, we also observed a depletion of GEM in the SIA below the sunlit zone, with GEM concentrations close to zero from 200 to 270 cm (see Fig. 3). Thus, both photolytic and dark oxidation of GEM were occurring in the SIA at Summit. Using backward transport model simulations,  showed that the boundary layer at Summit could be periodically impacted by halogens, and the bromine mixing ratio in the SIA at Summit was estimated at 1-4 ppbv by Peterson et al. (2001). Br • is probably the most efficient oxidant for GEM in the Arctic troposphere Skov et al., 2004), although BrO • could also be an active oxidant for GEM . Thus, GEM depletion within the snowpack could be the result of homogenous and/or heterogeneous chemistry with Br radicals (Ferrari et al., 2004b;Goodsite et al., 2004). The following mechanism has been proposed by Goodsite et al. (2004): These authors demonstrated that a recombination of Hg • with Br • (R1), followed by the addition of a second radical Br • (R2) was able to explain the observed rate of Hg • removal during AMDEs. Radicals I • and OH • could also interact with HgBr • , but their concentrations within the snowpack are to low for these compounds to be considered. Br 2 is also an oxidant for GEM according to the following reaction: However, the rate constant for reaction (R3) is much lower (by a factor ∼10 5 ) than the rate constants for (R1) and (R2) (Ariya et al., 2002;Goodsite et al., 2004). Since Reactions (R1), (R2) and (R3) are exothermic in gas and aqueous phases (Tossel, 2003), they could occur in a dark environment if Br • radicals or Br 2 are available.
Modeling and measurements of photolyzable bromine (Br 2 , BrCl, HOBr) and bromine radicals at Alert, Canada, showed that the snowpack is a bromine source (Tang andMcConnell, 1996). Foster et al. (2001) reported Br 2 concentrations in the SIA at Alert, Canada, that were the double of those measured in the overlying air during the sunlight period, which provides evidence for the production of Br 2 in natural snowpacks. There are now several observations suggesting that halogen chemistry may also be important at Summit: Swanson et al. (2007) provided evidence for production of alkyl halides in the SIA, and  suggested that HOx chemistry could be linked to halogen oxides at Summit Station. In the lower atmosphere, Br • radicals are formed by Br 2 photodissociation at wavelengths below 620 nm. Such wavelengths are available in the top centimeters of the snowpack at Summit Station (Galbavy et al., 2007). In July at Summit, Peterson and Honrath (2001) measured diel variations of ozone concentrations at a depth of 30 cm with minimum concentrations at midday, and levels close to atmospheric ones during the night. They demonstrated that only destruction by bromine radicals is sufficiently efficient to explain such rapid processes. Such ozone depletions are similar to GEM measurements we report for the sunlit zone both during summer and spring. We finally assume that the mechanisms involved in the destruction of GEM in the sunlit snow (from the surface to ∼60 cm depth) are similar to the ones happening during AMDEs and ODEs in the polar atmosphere during spring, as previously suggested (Ferrari et al., 2004b): GEM is actively converted to its oxidized form Hg(II) when solar irradiation is strong enough to produce Br • by the photolysis of Br 2 .
Deeper, the snowpack may also act as a source of Br 2 , even if there is no light. Oum et al. (1998) reported laboratory measurements suggesting that the oxidation of Br − ions by ozone at the surface of the snow crystals could produce Br 2 in the dark. These authors proposed the following reactions: This mechanism is consistent with field observations reported before the polar sunrise at Alert by Foster et al. (2001). These authors reported the simultaneous depletion of ozone and production of Br 2 in the SIA. Recently, Helmig et al. (2007a) reported O 3 measurements in the SIA at Summit from April to June 2004. These authors did not observed any ozone destruction in the SIA during the polar night (beginning of April) but they did not investigated ozone below a depth of 60 cm. Deeper measurements are necessary to conclude whether or not dark ozone oxidation occurred during this period of the year. On the other hand, they reported permanent minimum ozone levels at a depth of one meter for June. Such a decrease in concentration with depth cannot be explained by diffusion or sampler-induced advection. Considering that the actinic flux is null at this depth in the Summit snowpack (Galbavy et al., 2007), dark destruction of ozone in the SIA must be occurring. We assume that such dark oxidation reported for O 3 at Alert by Foster et al. (2001) also occurred in the SIA at Summit Station according to the mechanisms described previously (R4-R5-R6). This destruction of ozone is a source of Br 2 , which becomes available for GEM oxidation in the SIA via Reaction (R3).
More measurements during different seasons are now necessary to better understand the dark chemical processes involving both ozone and GEM. During summer 2005, we did not observe dark destruction of GEM directly, but rather diffusion of GEM produced in the sunlit snow through deeper layers. Sampling-induced artifacts could have also hidden this process.
3.4.6 GEM production in the snowpack GEM photo-production in the shallow firn was observed both during spring and summer, from the top surface to a depth of ∼60 cm where light could penetrate. GEM production processes were much more active relative to oxidation during the summer rather than during spring. GEM production was previously reported in interstitial air of both polar (Dommergue et al., 2003b;Ferrari et al., 2004b) and mid-latitude snowpacks (Poulain et al., 2004;Faïn et al., 2007). Most authors assumed that such production is the result of direct photodissociation of Hg(II) complexes. This reduction mechanism was also observed in water solutions (Xiao et al., 1994) and was suggested to occur in snow (Lalonde et al., 2002;Lalonde et al., 2003). GEM could also be produced by the reduction of Hg(II) complexes by photochemically produced compounds. Hydroperoxyl radical (HO • 2 ) was proposed as a potential reductant of Hg(II) in liquid water (Lin and Pehkonen, 1999). This hypothesis has to be considered carefully, since a recent study based on thermodynamic considerations showed that reduction of Hg(II) by HO • 2 radicals should be of minor importance (Gardfeldt and Jonsson, 2003). Sjostedt et al. (2007) reported the first measurements of peroxy radicals above the Summit snowpack. The majority of HO • 2 radicals in the SIA are generated via the photolysis of O 3 , H 2 O 2 , HONO and CH 2 O. Thus HO • 2 concentrations exhibit a diel pattern with maximum values at midday, but levels above zero at night. Humic acids may also play a central role in the degradation of Hg(II) molecules by producing reactive species. These mixtures of organic macromolecules can transfer electrons to species adsorbed on their surfaces when irradiated over broad spectral regions.
Laboratory flux chamber measurements indicated that GEM is first driven by solar radiation (especially UVA and UVB radiation), and then could be enhanced when liquid water is observed in the snowpack Faïn et al., 2007). Global irradiation and air temperatures were higher during summer 2005 as compared to spring 2006: the evolution of these two environmental parameters from spring to summer favored an increase in the photoproduction processes of GEM.
Our 2005 data showed an increase in the mean GEM concentration with depth, and a maximum was measured at a depth of 120 cm. This could be explained by samplerinduced advection from concentrated layers located close to the surface. However, dark production of GEM could also occur, as previously reported from snow samples collected in a mid-latitude snowpack (Lalonde et al., 2003), inside an alpine snowpack  and inside an arctic snowpack (Ferrari et al., 2004b).

Seasonal variations of GEM concentration
Environmental parameters such as irradiation and temperature, as well as the snow and the snow air compositions, evolved from spring to summer and led to a change in the relative importance of both GEM destruction and GEM production mechanisms in the SIA at Summit. However, we observed mean GEM concentrations in the SIA that were lower during spring and higher during summer than the atmospheric level. This suggests that GEM concentrations, up to a depth of three meters in the SIA, could exhibit a yearly variation with low concentrations during spring and higher concentrations during summer. However, no measurements were available for winter time. Using GEM concentrations collected from 15 to 30 m depths and diffusion modeling, we first investigated if any seasonal variations in GEM concentrations close to the surface could affect GEM levels in the deep firn air, thereby affecting the long-term record that is represented by the deep firn air. We then characterized the mean value and the amplitude of the seasonal variation expected in the shallow firn. These measurements were taken at depths too high to be significantly impacted by sampler-induced air flow from the snow surface down to these levels; rather, these concentrations represent interstitial air from the nearest permeable layer. These values are all close to the atmospheric level of ∼1.65 ng m −3 . As discussed in Sect. 3.4.2., dark destruction of GEM was occurring in the SIA close to the surface. Such processes obviously do not exit in the firn at depths of 15 to 30 m, because the GEM concentrations observed are not null. However, the temperature at both 15 and 2 m depths are similar during spring, and close to the mean annual temperature of the site (i.e. ∼−32 • C). We assume that the source of Br 2 required for dark GEM destruction no longer exists in the firn air at a depth of 15 m and below. To our knowledge, measurements of ozone concentrations have never been reported for deep polar firn air, but O 3 is a rather reactive gas which is not preserved in the air bubbles trapped in ice cores. The sinks of ozone are multiple, in the sunlit snow as well as in dark conditions. The atmospheric sources of ozone are much more limited and always require solar radiation. Thus, we suggest that the entire firn air column from a depth of 15 m to the bottom could exhibit null concentrations of ozone, and consequently null Br 2 concentrations, an oxidant for GEM in dark conditions.

Firn structure and modeling
Figure 8 presents a conceptual model of the firn at Summit Station. We can think of the firn as having three primary zones: the chemical zone (CZ), the diffusion zone (DZ) and the lock-in zone. The chemical zone extends from the surface to 2.7 m (z C ) and GEM levels within this zone are determined by both chemical and physical processes. These are discussed in detail in the previous section. Daily variations of GEM concentrations in the near-surface snow air lead to bi-directional fluxes between the shallow firn and the lower atmosphere. Below 2.7 m depth, we observed an increase in GEM concentrations during spring 2006. This evolution could be explained by a slowdown or a stop of the chemical processes involving GEM, combined with diffusion from deeper snow layers. The diffusion zone extends from the bottom of the chemical zone to the top of the lock-in zone, and is comprised of an openly porous and permeable matrix in which the air's composition is determined solely by physical processes (as shown in Sect. 3.5.1., there is no chemical alteration of GEM below z C ). The diffusion zone ends at the top of the lock-in zone (z D ). Within the lock-in zone, impermeable winter layers prevent the vertical diffusion of air, but persistent open porosity in the summer layers allows the extraction of samples (Martinerie et al., 1992). At still-greater depths, when the overburden is sufficient, all pores become closed and air can no longer be extracted. This defines the firn-ice transition.
We used a 1-D forward model developed by Rommelaere et al. (1997)  centrations occurring in the chemical zone could influence the GEM levels recorded in the air of the diffusion zone. Diffusion zone processes taken into account to simulate GEM transport were molecular diffusion, gravitational fractionation, and a downward air flux due to bubble closure sequestering air from open pores (see Fig. 8). This sequestered air must be replaced by air coming from the upper part of the firn, thus creating a downward flux. Surface temperature and accumulation rate were set to their present day values and assumed to be constant throughout the model run. We used firn structure parameters (density and closed porosity) from EUROCORE drilling (Schwander et al., 1993). Free-air diffusivity of GEM is given by Eq.
(2) gives the diffusivity of GEM within the firn (i.e. D eff , see Sect. 3.4.3). Deff depends on the firn tortuosity which increases progressively with depth. A null value of the ratio 1/τ at the top of the lock-in zone means that diffusion ceases at this depth. In short, the GEM effective diffusivity decreases from the snow surface value to zero at the close-off. We used tortuosity data reconstructed from CO 2 record to determine the GEM D eff profile at Summit. See Fabre et al. (2000) for a more complete description of the method.

Age distribution
Due to transport processes, the firn air at a given depth is not characterized by a single age, but by a distribution of ages (Schwander et al., 1993;Trudinger et al., 2002). We used our diffusion model to generate age distributions G(z, t) of GEM. This approach has been used in various studies for a variety of gases with seasonal atmospheric cycles (Trudinger et al., 2002;Montzka et al., 2004;Assonov et al., 2007). The distribution for depth z represents the relative contribution of different ages, where G(z, t)≡1. The age distributions at the depths of 30 m, 40 m, 50 m, 60 m, 69 m and 70 m are shown in Fig. 9. For example, point A indicates that 20% of GEM in the firn air at a depth of 30 m was in the atmosphere 2 years ago. Point B shows that 4% of GEM at a depth of 70 m is 15 years old. Above 70 m, the distributions show nonzero values for the first year (for example, ∼4% at a depth of 60 m). At 70 m and below, air is at least two years old. Consequently, the annual variations of GEM concentrations at the bottom of the chemical zone (depth z C ) cannot influence the firn air record below 70 m. In short, GEM values below a depth of 70 m record the annual mean concentrations of the bottom of the CZ.
For the firn air above 70 m, we have to examine to which extent GEM variations in the CZ influence the GEM concentrations that are measured deeper. Using an intuitive approach, we perceive that the amplitude of the variations of GEM concentrations in the CZ decreases as the depth increases. Indeed, GEM diffusion takes longer as the air is deeper. For a better estimate of the diffusion of GEM varia-tions in the CZ, we can adapt our transport model using different variation scenarios. We modeled their diffusion from the bottom of the CZ into the diffusion zone and compare this data to our field measurements.

Downward diffusion of seasonal GEM variations
As discussed in Sect. 3.4.5, GEM concentrations were stable in the atmosphere, but showed seasonal variations in the firn down to z C (2.7 m) due to chemical processes and transport. These variations have an influence even below the chemical zone as they diffuse downward. However, due to decreasing diffusivity of the firn with depth, the amplitude of the seasonal GEM cycle also steadily decreases with depth, vanishing completely at a depth of 70 m (see Sect. 3.5.2). In order to understand the downward propagation of this seasonal signal through the diffusion zone, we simulated seven different seasonal cycles and compared them to firn-air observations between 15 and 30 m depths. The cycles we considered are shown in Fig. 10 and are composed of three distinct modes: constant concentrations close to the atmospheric background, periods with elevated GEM concentrations relative to atmospheric background and periods with depleted GEM concentrations relative to background. A constant concentration of ∼1.65 ng m −3 at the bottom of the chemical zone will arise if there is no chemistry in the shallow firn, or if there is equilibrium between production and destruction of GEM. Concentrations higher than atmospheric values occur when production processes dominate, as we observed during summer 2005. GEM concentrations below atmospheric levels will occur when oxidation exceeds production in the shallow firn, as we observed during spring 2006. All cycles were run for 16 years with a time step of 7.3 days. The age distribution (see Fig. 9) shows that 98% of the firn air is between 0 to 16 year old at a depth of 30 m ( 16 t=0 G(30, t)=98%). For each of the scenarios described in Fig. 10, we calculated the depth z SEAS at which the amplitude of any GEM variation is less than 0.15 ng m −3 , the uncertainty of our GEM analyzer. These values are presented in Table 1. The location of z SEAS depends on both the amplitude of the GEM variations and the length of the constant concentration period (with GEM level of ∼1.65 ng m −3 ). Table 1 also includes the concentration of GEM at z SEAS . Below z SEAS , the GEM variations originating in the top layers of the snowpack are no longer observable. As shown by scenario 1, 2 and 3, an increase of the amplitude between maximum and minimum concentrations leads to an increase in the depth z SEAS . An increase in the period with stable concentrations (of about ∼1.65 ng m −3 ) reduces z SEAS (compare cycles 2, 6 and 7). As expected, for all cycles, concentrations at z SEAS are close to the mean concentration at the bottom of the chemical zone. While the different trial cycles do lead to a range of values for z SEAS , none has any influence on the firn air below a depth of 35 m.
Using our observations from 15 to 30 m depths, we are able to further constrain the shape of the seasonal GEM cycle Table 1. Characteristics of GEM diffusion in the deeper firn of the scenarios presented in Fig. 6. z D is the depth where fluctuations induced by surface scenario are lower than 0.15 ng m −3 . We also report GEM concentration at this depth z D . expected at the depth z c . We begin by assuming z SEAS is at most 15 m, since GEM concentrations were constant from 15 to 30 m depths. To be consistent with measurements carried out in the chemical zone in both summer 2005 and spring 2006, we considered GEM concentrations at depth z C ranging from a minimum at the beginning June (0.1 ng m −3 ) to a maximum at the end of July (2.9 ng m −3 ), and we adjust winter concentrations to the constant atmospheric background (i.e. 1.65 ng m −3 ). We fitted the mean annual concentration to ∼1.65 ng m −3 to match the observed value at a depth of 15 m, and we adjusted the length of the winter (i.e. constant concentration period) so as to not observe GEM variations deeper than 15 m. These constraints lead to cycle 7 in Fig. 10. The expression of this cycle in the firn between z C and 30 m is shown in Fig. 11. The blue line indicates the expected GEM concentration profile in June and is consistent with data collected at 3.3, 15, 25 and 30 m. As discussed in Sect. 3.4.4, production or destruction of GEM in the shallow firn can dominate, depending on season (spring vs. summer), leading to elevated or depleted GEM concentrations. The results from the model presented in Fig. 11 lead us to conclude that the annual mean concentration at a depth of 2.7 m in the firn air at Summit Station is close to the atmospheric level. This result is of prime importance: we can interpret the signal from a depth of 15 m and the close-off solely as a record of the long-term atmospheric trend. Moreover, the shape of cycle 7 (Fig. 10) indicates that GEM chemistry in the snowpack probably occurs only during the brief period between May and September, which suggests that chemical processes involving GEM in the shallow firn could be linked to solar radiation. We assumed in this study that no photochemical processes were occurring in the SIA below 60 cm depth, according to Galbavy et al. (2007). However we expect that future studies will investigate the transmission of solar irradiance deeper within the snowpack.

Implications for the transfer function
Our measurements of GEM close to the surface revealed a high variability of GEM concentrations on a daily time scale, as well as a seasonal shift in the chemical mechanisms affecting GEM levels. Daily fluctuations in GEM concentrations are too fast relative to diffusion times to influence the firn GEM record beyond a depth of ∼3 m. Moreover, our modeling work showed that seasonal variations are not expressed in the deep firn. In principle, the first 70 m of the firn could show signs of seasonality, but our best estimates of the seasonal variations of GEM lead us to expect that only the first 15 m of the firn air contain detectable influences of shallowfirn chemistry. This implies that firn air from a depth of 15 m to the close-off has the potential to be used to reconstruct the atmospheric history of GEM over the last decades. Further-more, the conservative nature of GEM below ∼3 m indicates that it is possible to calculate the GEM transfer function at Summit Station using more detailed information on the diffusivity of the firn between ∼3 m and the close-off.

Conclusions
Atmospheric Mercury Depletion Events were not observed at Summit Station during both summer 2005 and spring 2006, and are unlikely to occur, since Ozone Depletion Events have never been reported for this site. In air extracted from the snowpack between 40 and 200 cm depths, GEM concentrations showed variability at daily and seasonal time scales. Both production and oxidation of GEM were observed during spring (beginning of June) and summer (end of July), but the respective contributions of each process change with time during the season. Oxidation is the dominant mechanism during spring, and production dominates during summer. We postulate that dark oxidation chemistry can influence GEM levels in the SIA. While the shallowest firn at Summit Station acts as a source and a sink of GEM for the lower atmosphere, the deeper firn air contains a record of the evolution of atmospheric GEM concentrations since it is not perturbed by surface processes. In particular, the daily variations of GEM concentrations are too rapid to influence the firn record and the influence of seasonal variations does not extend below the first 15 m of the firn. From a depth of 15 m to the close-off (about ∼79.5 m at Summit Station), the firn air should provide a record of the past atmospheric GEM composition.