Global atmospheric model for mercury including oxidation by bromine atoms

. Global models of atmospheric mercury generally assume that gas-phase OH and ozone are the main oxidants converting Hg 0 to Hg II and thus driving mercury deposition to ecosystems. However, thermodynamic considerations ar-gue against the importance of these reactions. We demon-strate here the viability of atomic bromine (Br) as an alterna-tive Hg 0 oxidant. We conduct a global 3-D simulation with the GEOS-Chem model assuming gas-phase Br to be the sole Hg 0 oxidant (Hg + Br model) and compare to the previous version of the model with OH and ozone as the sole oxidants (Hg + OH/O 3 model). We specify global 3-D Br concentration ﬁelds based on our best understanding of tropospheric and stratospheric Br chemistry. In both the Hg + Br and Hg + OH/O 3 models, we add an aqueous photochemical reduction of Hg II in cloud to impose a tropospheric lifetime for mercury of 6.5 months against deposition, as needed to reconcile observed total gaseous mercury (TGM) concentrations with current estimates of anthropogenic This added reduction would not be necessary in the Hg + Br model if we adjusted the Br oxidation kinetics downward within their range of uncertainty. We ﬁnd that the Hg + Br and Hg + OH/O 3 models are equally capable of reproduc-ing the spatial distribution of TGM and its seasonal cycle at northern mid-latitudes. The Hg + Br model shows a steeper decline of TGM concentrations from the tropics to southern mid-latitudes. Only the Hg + Br model can reproduce the springtime depletion and summer rebound of TGM observed at polar sites; the snowpack component of GEOS-Chem suggests that 40% of Hg II deposited to snow in the Arctic is transferred to the ocean and land reservoirs,


Introduction
Mercury is a neurotoxic pollutant that is dispersed globally by atmospheric transport. Emissions are mostly elemental mercury (Hg 0 ) and atmospheric observations of Hg 0 imply an atmospheric lifetime on the order of one year (Lindberg et al., 2007). The oxidized product Hg II is highly water soluble and deposits rapidly through precipitation and surface uptake. Understanding the global budget of atmospheric mercury and the source-receptor relationships for mercury deposition therefore requires global atmospheric transport models with accurate redox chemistry.
Published by Copernicus Publications on behalf of the European Geosciences Union.
A fundamental limitation of current models is the uncertainty in the atmospheric chemistry of mercury (Lin et al., 2006;Ariya et al., 2008Ariya et al., , 2009. Atmospheric observations imply that oxidation of Hg 0 to Hg II must be photochemical (Shia et al., 1999;Selin et al., 2007). Models generally assume that gas-phase OH and ozone are the main oxidants, and also include aqueous-phase reduction of Hg II to Hg 0 that competes with deposition as a sink for Hg II (Bergan and Rodhe, 2001;Petersen et al., 2001;Cohen et al., 2004;Lin et al., 2006;Seigneur et al., 2006;Selin et al., 2007;Pongprueksa et al., 2008). However, recent work suggests that gas-phase oxidation of Hg 0 by OH and O 3 is too slow to be of atmospheric relevance (Calvert and Lindberg, 2005;Hynes et al., 2009). Heterogeneous oxidation in clouds and aerosols is conceivable but hypothetical (Calvert and Lindberg, 2005;Snider et al., 2008;Ariya et al., 2009). There is also no accepted kinetics or mechanism for Hg II atmospheric reduction (Ariya et al., 2009;Hynes et al., 2009). Present-day measurement techniques cannot determine the molecular identity of atmospheric Hg II oxidation products, but instead quantify all gas-phase Hg II as reactive gaseous mercury (RGM). Holmes et al. (2006) proposed that gas-phase Br atoms might be the dominant global oxidant of Hg 0 , with most of the oxidation taking place in the free troposphere. Several pieces of evidence support this idea. Oxidation of Hg 0 by Br is thought to explain the mercury depletion events (MDEs) in polar spring (Goodsite et al., 2004;Steffen et al., 2008;Xie et al., 2008). Diurnal patterns of Hg II in the marine boundary layer (MBL) are consistent with oxidation by Br (Hedgecock et al., 2005;Holmes et al., 2009). Column measurements suggest a background concentration of 0.5-2 ppt BrO in the free troposphere Van Roozendael et al., 2002;Sinnhuber et al., 2005) that could be accounted for by photolysis and oxidation of bromocarbons (Yang et al., 2005). Br atom concentrations deduced from photochemical equilibrium with this background BrO could yield an Hg 0 atmospheric lifetime of less than a year (Holmes et al., 2006). The lower stratosphere also contains elevated BrO (Salawitch et al., 2005), which might explain the rapid depletion of Hg 0 observed above the tropopause .
Constructing a plausible global model of Hg + Br chemistry is challenging because of the large range of reported Hg + Br kinetics (Holmes et al., 2006;Ariya et al., 2008;Hynes et al., 2009) and because of uncertainties in the concentrations of atmospheric Br. Gaseous inorganic bromine (Br y ) originates from atmospheric degradation of bromocarbons and debromination of sea-salt aerosol (von Glasow et al., 2002;Pszenny et al., 2004;Yang et al., 2005). Shortlived bromocarbons are thought to dominate the supply of Br y in the free troposphere and lower stratosphere; these include CHBr 3 and CH 2 Br 2 emitted by the ocean and CH 3 Br of both biogenic and anthropogenic origin (Yang et al., 2005). Sea salt dominates Br y supply in the MBL. Br y cycles between radical forms (Br and BrO) and non-radical reservoir species (HOBr, HBr, BrNO 3 , BrNO 2 , and Br 2 ) (Pszenny et al., 2004). Br and BrO are in fast photochemical equilibrium during daytime and disappear into the reservoir species at night. Heterogeneous reactions of HOBr, HBr, and BrNO 3 on aerosols could also be important for maintaining radical concentrations (von Glasow et al., 2004;Yang et al., 2005). Br y is eventually removed from the atmosphere by wet and dry deposition.
Three previous global mercury model studies have included Br as an Hg 0 oxidant in addition to OH and O 3 Seigneur and Lohman, 2008;Dastoor et al., 2008). The studies of Ariya et al. (2004) and Dastoor et al. (2008) focused on simulation of Arctic MDEs, where the importance of Br is well established. Seigneur and Lohman (2008) evaluated the sensitivity of the simulated interhemispheric and vertical gradients of Hg 0 to the Hg + Br reaction kinetics. Their simulated mean surface Hg 0 concentrations changed by 20-40% across the range of the kinetic data (Ariya et al., 2002;Donohoue et al., 2006), with the best results obtained with the slow kinetics. In contrast, Dastoor et al. (2008) reported that the fast kinetics gave a better simulation of Hg 0 in the Arctic at Alert, Canada. Seigneur and Lohman (2008) also presented a sensitivity test in which Br was the sole oxidant of Hg 0 . This showed an unrealistic peak of Hg 0 in the tropics and minima at the poles.
Here we use the GEOS-Chem global chemical transport model  to evaluate whether a model with gas-phase Br as the sole Hg 0 oxidant can in fact be consistent with atmospheric observations. The above model studies derived their tropospheric bromine concentrations from satellite observations of BrO columns (Chance, 1998;Burrows et al., 1999), which feature polar maxima of BrO. Recent aircraft observations in the Arctic, however, show that the troposphere contributes less to these polar maxima than previously expected (Salawitch et al., 2010), so earlier models likely overestimated tropospheric Hg 0 oxidation at high latitudes. Here we use a combination of field measurements and process-based models to estimate the distribution of Br from the surface to the stratosphere. We also describe several other recent improvements to the GEOS-Chem mercury model including updated anthropogenic emissions, mechanistic uptake by sea-salt aerosol, scavenging by snow and ice, and a coupled snowpack reservoir. We evaluate the ability of this new model version to reproduce atmospheric observations through comparisons with multiple data sets.

Model description
The previous version of the GEOS-Chem atmosphere-oceanland mercury model (v7.04) was described by . The model includes a global 3-D atmosphere (here 4 • × 5 • horizontal resolution, 55 vertical levels, hourly time steps) coupled to 2-D surface ocean and soil reservoirs. The atmospheric component is driven by assimilated meteorological data from the Goddard Earth Observing System Atmos. Chem. Phys., 10, 12037-12057, 2010 www.atmos-chem-phys.net/10/12037/2010/ (GEOS) of the NASA Global Modeling and Assimilation Office (GMAO). It includes three transported species: Hg 0 , Hg II , and inert, nonvolatile particulate mercury (Hg P ). The surface ocean component  includes three species: Hg 0 , reactive dissolved Hg II , and inert particlebound Hg P . These ocean species undergo chemical interconversion and vertical exchange with the atmosphere and with a deep ocean reservoir of fixed mercury concentrations. Horizontal transport in the ocean is neglected. Natural soil mercury concentrations are specified on the 4 • × 5 • grid by steady state of emissions and deposition in the preindustrial atmosphere . They are augmented for present-day on the basis of the modeled deposition patterns of anthropogenic mercury.
In the present model we have updated the emissions, atmospheric chemistry, and deposition modules used by . We elaborate on these improvements below. We have also updated the transport component by using meteorological input from the GEOS-5 assimilation data, which have 0.5 • ×0.67 • horizontal resolution and 72 vertical layers. As before, we degrade the resolution to 4 • ×5 • and 47 layers for computational expediency. Tracer transport algorithms are from the current GEOS-Chem version (8.02.03), which includes improved cross-tropopause transport (MacKenzie, 2009) and a non-local parameterization of boundary layer mixing (Lin and McElroy, 2010). Figure 1 presents our updated global mercury budget, which we will discuss in Sect. 4.  previously used the GEIA anthropogenic emissions for 2000 (Pacyna et al., 2006) but increased Hg 0 emissions globally by 30% (by 50% in China) to 3400 Mg a −1 total Hg in order to accommodate atmospheric observations. Those emissions exceed the 1900-2600 Mg a −1 range of recent estimates and are likely too high (Mason and Sheu, 2002;Streets et al., 2009;Pacyna et al., 2010;Pirrone et al., 2010). Here we use the Streets et al. (2009) global inventory for 2006 partitioned into 17 regions; emissions within each region follow the GEIA distribution. In addition, Hg 0 emissions from artisanal gold mining total 450 Mg a −1 (Hylander and Meili, 2005;, which is very close to the independent estimate of 400 Mg a −1 by Telmer and Veiga (2009). Our anthropogenic emissions thus total 1300 Mg a −1 Hg 0 , 650 Mg a −1 Hg II , and 100 Mg a −1 Hg P . While these are 30% lower than in , our simulation remains consistent with the observed Hg 0 concentrations (as we will show below) because changes in the redox chemistry prolong the Hg 0 lifetime.

Emissions
The soil emissions specified by  were an exponential function of both soil temperature and solar radiation, producing a strong summer peak. With the smaller anthropogenic emissions and slower oxidation in the present model, these emissions would result in a summer Hg 0 max- imum at northern mid-latitudes that is at odds with observations. Here we specify soil emission E as a function of solar radiation only following Zhang et al. (2001), where C s is the soil mercury concentration (g g −1 ), R g is the solar radiation flux at the ground, and α = 1.1×10 3 m 2 W −1 . The scaling factor β = 0.02 g m −2 h −1 is derived here from global mass balance in the preindustrial period, as described by . With this change, simulated Hg 0 concentrations follow the observed seasonal cycle, but total present-day soil emissions, 1200 Mg a −1 , are unchanged from . As in , soil and vegetation emit an additional 260 Mg a −1 through rapid photoreduction of deposited Hg II . Biomass burning emits 300 Mg a −1 following the distribution of biomass burning CO, using a new Hg/CO emission ratio of 100 nmol mol −1 derived in Sect. 3.5. Friedli et al. (2009) estimate larger biomass burning emissions of 675 ± 240 Mg a −1 based on satellite-derived fire area and biome-specific emission factors, but our results here are not sensitive to this difference because these emissions are relatively small in any case. The model no longer includes emissions through plant transpiration because of field evidence that this process is unimportant (Gustin et al., 2004).
Arctic field studies find large Hg 0 emissions from sunlit snowpacks in spring and summer, following surface enrichment caused by MDEs (Cobbett et al., 2007;Steffen et al., 2008, and references therein). Some of the mercury deposited during MDEs may be retained in ecosystems during snowmelt (Dommergue et al., 2003;Brooks et al., 2006;Johnson et al., 2008), but low Hg concentrations in lateseason snow and meltwater suggest that most of the MDEdeposited mercury returns to the atmosphere (Kirk et al., 2006). We add a snowpack reservoir on the 4 • × 5 • grid that accumulates mercury deposition and releases it as Hg 0 under sunlit conditions. The reservoir lifetime is 180 d, decreasing to 21 d when T > 270 K to fit observations by Fain et al. (2007Fain et al. ( , 2008) that re-emission accelerates sharply when melting begins. This simple parameterization reproduces the seasonal cycle of atmospheric Hg 0 at Arctic sites as will be shown in Sect. 3.2. We find that 60% of mercury deposited to snow is eventually reemitted and 40% enters the underlying ocean or soil. Global snow emissions are 210 Mg a −1 . Figure 1 summarizes model emissions. Net ocean Hg 0 emissions respond dynamically to changes in emissions and chemistry and are now 2000 Mg a −1 , which is 40% smaller than the earlier model and closer to central estimates from other studies (Lamborg et al., 2002;Mason and Sheu, 2002;Sunderland and Mason, 2007;. Global mercury emissions are 8300 Mg a −1 if we include gross ocean Hg 0 emissions or 6300 Mg a −1 if we include only net ocean emission.

Chemistry
A major update in this work is to oxidize Hg 0 by Br atoms instead of by ozone and OH. Table 1 lists the reactions involved. Atomic bromine initiates Hg 0 oxidation in the gas phase following a mechanism described by Goodsite et al. (2004). The unstable product, HgBr, may either dissociate or react with Br or OH to form Hg II . We use kinetic coefficients from Donohoue et al. (2006), Goodsite et al. (2004) and Balabanov et al. (2005). These coefficients are at the low end of the published range (Holmes et al., 2006) and are similar to the ones chosen by Seigneur and Lohman (2008) to fit observed vertical Hg 0 gradients and by Xie et al. (2008) to model MDEs. OH concentrations for the HgBr + OH → HgBrOH reaction are archived from a GEOS-Chem full-chemistry simulation (Park et al., 2004).
Global bromine concentrations are specified on the model grid by combining estimates of the contributions from major precursors: bromocarbons, halons, and sea-salt aerosol bromide. For the troposphere, except the MBL, and the lower stratosphere we use monthly archived Br from the p-TOMCAT model with biogenic bromocarbon and methyl bromide as the only source gases (Yang et al., 2005). In the middle stratosphere and above, where halons decompose, we use archived Br from the NASA Global Modeling Initiative (GMI) Aura4 model with halon and methyl bromide source gases (Strahan et al., 2007). These model estimates are constrained by observations of the bromocarbon source gases (e.g. Douglass et al., 2004;WMO, 2007;Warwick et al., 2007) and standard gas-phase chemistry of Br y (Sander et al., 2006). They may be lower limits because we do not account for ventilation of MBL air containing Br y from sea-salt aerosol (Yang et al., 2005) or heterogeneous reactivation of Br y on aerosols (von Glasow et al., 2004;Yang et al., 2010).
Sea-salt aerosol bromide is an additional source of Br y for the MBL. Here we assume a uniform daytime concentration of 1 ppt BrO, consistent with the few observations available (Leser et al., 2003;Saiz-Lopez et al., 2004;Martin et al., 2009;O'Brien et al., 2009) and with the observed diurnal cycle of RGM (Holmes et al., 2009). We calculate the associated Br concentrations from photochemical steady state (Platt and Janssen, 1995), where J BrO is the BrO photolysis frequency, and k 1 = 2.1 × 10 −11 cm 3 molecule −1 s −1 and k 2 = 1.2 × 10 −12 cm 3 molecule −1 s −1 are the rate coefficients for the BrO + NO → Br + NO 2 and Br + O 3 → BrO + O 2 reactions, respectively (Platt and Janssen, 1995).
[NO] = 10 ppt is assumed, and [O 3 ] and mean daytime J BrO are archived from GEOS-Chem full-chemistry simulations (Park et al., 2004;Parrella et al., 2010). We impose a diurnal cycle for Br throughout the atmosphere as done by Holmes et al. (2009) and find that the global model reproduces the observed diurnal cycles of Hg II in the MBL as reported in that earlier study.
Springtime photochemistry of sea salt on sea ice can produce unusually high Br concentrations in the polar boundary layer in spring, resulting in fast oxidation of mercury and ozone (Simpson et al., 2007). BrO concentrations are typically 5-15 ppt . Here we specify 10 ppt BrO in the Arctic (Antarctic) boundary layer during March-May (August-October) over areas with sea ice, sunlight, stable conditions, and temperatures below 268 K. We calculate the Br concentration in steady state as above, assuming that O 3 is depleted to 2 ppb. Figure 2 shows the resulting GEOS-Chem Br mixing ratios for the months of January and July. We also show BrO for reference although it does not oxidize Hg 0 in the model. Br and BrO have a strong photochemically driven seasonal cycle in the extratropics. Concentrations increase with altitude due to photochemical production. Minima in the tropical lower troposphere reflect wet deposition of soluble inorganic bromine species. Br concentrations peak at the tropical tropopause due to strong radiation and relatively low ozone, but otherwise show little latitudinal variation in the summer hemisphere. Monthly mean BrO columns range from 1 × 10 13 cm −2 in the tropics to 4 × 10 13 cm −2 at the summer pole, which agrees well with values and latitudinal trends observed from satellites (Chance, 1998;Richter et al., 2002;Sioris et al., 2006), after we account for the two-fold difference between 24-h averages shown here and the daytime concentrations detected from space.
From these Br concentration fields and the mechanism in Table 1 we obtain a global Hg 0 chemical lifetime of 6 months, with most of the oxidation taking place in the free troposphere. We find that HgBrOH is the major product, but it and other Hg II species are expected to undergo ion exchange in cloud and aerosol water to produce HgCl 2 primarily (Hedgecock and Pirrone, 2001;Lin et al., 2006).  Subsequent deposition of Hg II depends on its gas/aerosol partitioning, for which observations show considerable variability (Jaffe et al., 2005;Caldwell et al., 2006;Liu et al., 2007;Valente et al., 2007;Cobbett et al., 2007;Weiss-Penzias et al., 2009). This partitioning is expected to depend on temperature, aerosol load, and aerosol composition (Lin et al., 2006;Rutter and Schauer, 2007a,b). Future work will link Hg II partitioning to aerosol concentration and composi-tion in the model, while here we assume 50/50 partitioning of Hg II between the gas and aerosol phase for the purpose of calculating Hg II deposition as described in the following sub-sections.
Our initial simulation without reduction of Hg II produced mean Hg 0 surface concentrations that were smaller than observed. Early global models for mercury included aqueous reduction of Hg II by HO 2 and SO 2− 3 , but these reactions are now thought to be negligibly slow (Van Loon et al., 2000;Gårdfeldt and Jonsson, 2003). More recent models have hypothesized gaseous or aqueous reactions and tuned the kinetics to match the Hg 0 observations Pongprueksa et al., 2008). Laboratory studies have reported fast UV photoreduction of aqueous HgCl 2 in the presence of organic acids (Pehkonen and Lin, 1998;Ababneh et al., 2006;Si and Ariya, 2008). We assume Hg II reduction in liquid water clouds to be proportional to the NO 2 photolysis frequency, archived from a GEOS-Chem full-chemistry simulation (J. , and adjust the reduction rate to best match the global mean surface Hg 0 measurements. The best fit yields a Hg II global tropospheric lifetime of 1.7 months against reduction. After including reduction, the mean atmospheric lifetime of mercury is 7.3 months (6.5 months in the troposphere). We will also discuss results from a sensitivity simulation without Hg II reduction and instead decreasing the overall rate of Hg 0 + Br + X →→ Hg II (X ≡ Br, OH) reaction by 60% to yield the same atmospheric lifetime of mercury as in the standard simulation. This decrease in oxidation lies within the range of theory-derived kinetic coefficients for HgBr + X → HgBrX (Goodsite et al., 2004;Balabanov et al., 2005) or could be accommodated by lower atomic Br concentrations.
An important objective of this study is to evaluate the ability of GEOS-Chem to fit observations using Br as the sole Hg 0 oxidant instead of OH and ozone. We will compare results from a simulation with Br chemistry (the "Hg + Br model") versus one with OH and ozone chemistry ("Hg + OH/O 3 model"). While both oxidation mechanisms, and possibly others, may operate together in the real atmosphere, these idealized simulations enable us to explore the constraints that observations place on the atmospheric chemistry of mercury. For the Hg + OH/O 3 model we follow the kinetics of Sommar et al. (2001) and Hall (1995), as used by Selin et al. (2007), with OH and ozone concentrations specified from a full-chemistry GEOS-Chem simulation. The resulting oxidation of Hg 0 is faster than by Br and takes place at lower altitudes where Hg II deposits faster, so we com-pensate by increasing the reduction rate coefficient 4 fold. The Hg 0 lifetime in that simulation is 3.7 months with OH contributing 80% of the sink, but with the faster reduction the atmospheric lifetime of total mercury is the same as in the Hg + Br model. Figure 3 shows the zonal distribution of Hg 0 oxidation in the Hg + Br and Hg + OH/O 3 models. Oxidation by bromine is fast in the MBL where Br number density is largest, but most of the global oxidation occurs in the free troposphere due to low temperatures and increasing Br mixing ratios with altitude (Holmes et al., 2006). Oxidation is also fast in the stratosphere but limited by the small concentrations of Hg 0 . The Southern Hemisphere has faster oxidation than the Northern Hemisphere because of the oceanic source of bromocarbons and the low temperatures over Antarctica. Springtime bromine explosions drive secondary oxidation maxima in the polar boundary layers. Oxidation by OH and O 3 follows the general distribution of OH concentrations, with a maximum in the lower tropical troposphere and symmetry about the equator. Reduction of Hg II (not shown) peaks at 1-2 km altitude, where cloud liquid water is high, and no reduction occurs above 10 km where clouds are entirely ice.

Sea-salt aerosol as a sink for mercury
Building on earlier work by Hedgecock and Pirrone (2001) and Selin et al. (2007), we previously suggested that uptake of Hg II by sea-salt aerosol as HgCl 2− 4 is the dominant sink for Hg II in the MBL and the major source of mercury to the surface ocean (Holmes et al., 2009). We calculated the Hg II uptake rate and subsequent deposition flux (F dep ) in a box model of the MBL on the basis of the local 10-m wind speed (u 10 ), relative humidity (saturation ratio S) and mixing depth (H ). Fast winds enhance uptake through increased sea spray, while low relative humidity increases [Cl − ] within the seasalt particles and hence promotes formation of HgCl 2− 4 . We accounted for mass-transport limitations at the gas-particle interface over the sea-salt aerosol size distribution. Here we parameterize the results from this MBL box model for implementation in GEOS-Chem as a first-order rate coefficient (k) for Hg II net uptake and subsequent deposition, where [Hg II ] is the MBL concentration. Figure 4 shows k(u 10 ,S) simulated in the box model with full physics over the range of conditions expected in the marine atmosphere (Holmes et al., 2009). We fit k to the following form: exp a 2 u 10 + a 3 u 1/2 10 + a 4 u 3/2 10 with coefficients a 0 = 1 × 10 −10 s −1 , a 1 = −59.91, a 2 = −1.935 s m −1 , a 3 = 9.009 s 1/2 m −1/2 , and a 4 = 0.1477 s 3/2 m −3/2 . This simplified model closely fits the 24-h mean loss rate in the full-physics model (r 2 = 0.97) over the parameter range 0.7 ≤ S ≤ 0.99 and 0.1 ≤ u 10 ≤ 20 m s −1 .

Other deposition processes
GEOS-Chem includes wet scavenging of Hg II and Hg P following the scheme of Liu et al. (2001), and dry deposition of Hg 0 , Hg II , and Hg P following the resistance-in-series scheme of Wesely (1989).  describe how these schemes apply to mercury in the previous version of the model. They assumed Hg II to be gaseous HgCl 2 for the purpose of computing deposition; the Henry's law solubility constant of HgCl 2 is 1.4 × 10 6 M atm −1 (Lindqvist and Rodhe, 1985), sufficiently high for near-100% scavenging in clouds and fast dry deposition limited by aerodynamic resistance. Here we assume 50/50 partitioning of Hg II between the gas and aerosol phase, which increases the lifetime of Hg II against dry deposition as compared to the previous model version.  assumed no scavenging of Hg II in cold (frozen) clouds and snow, and zero retention efficiency of Hg II upon cloud freezing, in order to reproduce the observations of low wet deposition fluxes of mercury at northern US sites in winter. However, observations by Douglas et al. (2008) indicate high mercury concentrations in rime ice, implying high retention efficiency. Therefore we now assume that supercooled water in mixed-phase clouds retains all Hg II and Hg P during freezing. Douglas et al. (2008) and Johnson et al. (2008) found by contrast very low Hg concentrations in ice grown from the vapor phase, so we still assume no mercury scavenging by cloud ice. Below-cloud scavenging by snow is included only for aerosol Hg II and Hg P , with the same efficiency as by rain (Murakami et al., 1983;Feng, 2009). Sigler et al. (2009) found that snowfall has little effect on ambient RGM, so we do not include below-cloud scavenging of gaseous Hg II by snow. Adding low-temperature scavenging as described above increases deposition at high latitudes, but also allows low-latitude convective rainfall to scavenge from higher altitudes.

Model evaluation
We test here whether the Hg + Br model (simulation with Hg 0 oxidation initiated by Br only) can reproduce the general patterns seen in atmospheric observations, and compare these results to the Hg + OH/O 3 model (simulation with Hg 0 oxidation by OH and O 3 ). All simulations are initialized over 15 years of repeated present-day meteorological data to reach annual steady state in the stratosphere. We then analyze model results averaged over 2006-2008 and compare to observed air concentrations and wet deposition fluxes. Figure 5 shows annual mean observed surface concentrations of total gaseous mercury (TGM ≡ Hg 0 + RGM) compared to the Hg + Br model. TGM in the model is calculated as Hg 0 + 0.5 Hg II . The measurements include annual means at 39 land sites during 2000-2008, plus data from ship cruises (Lamborg et al., 1999;Laurier et al., 2003;Temme et al., 2003;Laurier and Mason, 2007;Soerensen et al., 2010a). Trends in mean TGM during the last decade are small (of order 1% a −1 ) or negligible at most background sites in the Northern Hemisphere Wangberg et al., 2007). Southern Hemisphere data contain larger trends (Slemr et al., 2010) which we discuss below. The model reproduces the spatial variability observed at at the 39 land sites (r 2 = 0.81). The mean and  2007), plus additional sites in Europe , EMEP 2009), North America Yatavelli et al., 2006;Stamenkovic et al., 2007;Temme et al., 2007;Choi et al., 2008;Sigler et al., 2009, E. Edgerton, personal communication, 2008, East Asia (Nguyen et al., 2007;Sakata and Asakura, 2007;Feng et al., 2008;Wan et al., 2009), South Africa (Slemr et al., 2010)  A prominent deficiency in the model, previously identified by Selin et al. (2007), is that it does not reproduce the high concentrations observed over the North Atlantic and Pacific Oceans during ship cruises. This is likely due to upwelling mercury from the sub-surface ocean, possibly reflecting the legacy of past anthropogenic emissions. Although this is not captured in our simulation, where uniform sub-surface ocean mercury concentrations are assumed globally , Soerensen et al. (2010b) find that forcing GEOS-Chem with observed sub-surface North Atlantic concentrations can reproduce the high atmospheric concentrations observed over the North Atlantic. This will be implemented in a future version of the model. Figure 6 shows that the Hg + Br and Hg + OH/O 3 models diverge in their surface TGM predictions for the Southern Hemisphere because of the different oxidant distributions (Figs. 2 and 3). The Hg + Br model predicts 110-120 ppq TGM at southern mid-latitudes vs. 140-150 ppq in the Hg + OH/O 3 model. The Hg + Br model better simulates land stations in Antarctica (Temme et al., 2003) and  (Slemr et al., 2010) spans the range between the Hg + Br and Hg + OH/O 3 models. Ship data at southern mid-latitudes likewise encompass a wide range (110-160 ppq), likely caused by variability in marine emissions, that does not discriminate between the two simulations. Additional long-term measurements at a southern mid-latitude site, complementing the record at Cape Point, together with greater constraints on Southern Hemisphere ocean emissions would further test the Hg 0 oxidation mechanism.

Global distribution of mercury
The meridional gradient in Fig. 6 differs markedly from the model of Seigneur and Lohman (2008), which predicted peak Hg 0 in the tropics and unrealistically low concentrations in the extra-tropics when Br was the sole oxidant. Seigneur and Lohman inferred Br concentrations from the GOME BrO columns, imposing vertical distributions and Br/BrO ratios from the p-TOMCAT CTM (Yang et al., 2005). That CTM does not include halons and would therefore greatly underestimate the contribution of the stratosphere to the BrO column. Considering that the stratospheric contribution is what causes the BrO column increase with latitude ( Fig. 2), this method would particularly overestimate tropospheric Br and, hence, Hg 0 oxidation at high latitudes. attributed this seasonal cycle to the photochemical sink from OH, and we obtain the same result with oxidation by Br which also peaks in summer. At Cape Point, the only site with long-term data in the Southern Hemisphere outside Antarctica, TGM displays little seasonal variation during recent years, as shown. Before 2004, however, TGM had larger variation with TGM maxima in summer (December-February), in contrast to the Northern Hemisphere . The Hg + OH/O 3 model reproduces the earlier pattern, which reflects the maximum of ocean emissions in austral summer caused by biological and photochemical reduction of aqueous Hg II Slemr et al., 2008). The Hg + Br model has smaller seasonal variation due to the offsetting effect of strong oxidation by Br at southern mid-latitudes in summer and this coincides with the recent seasonal data. This reinforces the value of additional measurements of mercury concentration and interannual variability of ocean fluxes in Southern Hemisphere.

Seasonal cycle at surface sites
Observations at Arctic sites and at the Neumayer Antarctic site show a springtime minimum driven by MDEs and a summertime maximum driven by re-emission from the snowpack Cobbett et al., 2007). The Hg + Br model can reproduce this seasonal variation but not the Hg + OH/O 3 model, which does not include MDEs. We find that atmospheric concentrations are consistent with reemission of 60% of Hg deposited to the snowpack during springtime and 40% net incorporation into the ocean and soil. The area within the Arctic Circle receives 60 Mg a −1 net deposition in the Hg + Br model vs. 40 Mg a −1 in the Hg + OH/O 3 model without MDEs. The Antarctic Circle similarly receives 70 Mg a −1 in the Hg + Br model, but only 20 Mg a −1 in the Hg + OH/O 3 model. Dastoor et al. (2008) estimate a similar re-emission fraction from snow, but 3 times larger net deposition to the Arctic surface.

Testing oxidation chemistry through Antarctic subsidence events
Observations at Antarctic sites show frequent summertime events of depleted Hg 0 and enhanced RGM together with elevated ozone (Sprovieri et al., 2002;Temme et al., 2003;Aspmo and Berg, 2009). These differ from springtime depletion events in that O 3 is anti-correlated with Hg 0 . From four events in the published Neumayer and Terra Nova Bay data (Sprovieri et al., 2002;Temme et al., 2003), we estimate ranges of −6.0 to −11.5 for Hg 0 / O 3 and 1.5 to 4.0 for RGM/ O 3 . Aspmo and Berg (2009) used backtrajectories to identify the mid-troposphere as the source region for such events. Brooks et al. (2008) also found that subsiding air at the South Pole contains elevated Hg II . These observations provide a sensitive test for Hg 0 oxidation chemistry in the model because the cold, dry Antarctic atmosphere minimizes the confounding effect of aqueous reduction. In addition, Br is an effective Hg 0 oxidant over Antarctica in summer (Fig. 3) while OH is ineffective. Figure 8 shows simulated Hg 0 and RGM at Neumayer for January 2008, and O 3 from a GEOS-Chem full-chemistry simulation at the same time and location (J. . The model time series shows several subsidence events with enhanced O 3 and RGM, and depleted Hg 0 . These events last 1-3 days, as found by Temme et al. (2003). We derive the model Hg 0 / O 3 and RGM/ O 3 ratios shown in Fig. 8B from a reduced major-axis fit to the January time series. The ratios in the Hg + Br model are consistent with observations while those in the Hg + OH/O 3 model are much too weak. Observed ratios were calculated from data reported by Temme et al. (2003). Modeled ranges are 95% confidence intervals for the reduced major-axis slope from the bootstrap method. Evaluation Programme (EMEP) over Europe. These networks collect weekly (MDN) or monthly (EMEP) integrated samples. We use sites with at least 75% of annual data available for the simulated years, 2006-2008. We also require fewer than 5 consecutive missing samples for MDN. Both MDN and EMEP have been used extensively to test atmospheric mercury models (e.g. Bullock et al., 2009;Gusev et al., 2009) and to evaluate the impact of mercury emission reductions (e.g. Butler et al., 2008;Wangberg et al., 2007;Pacyna et al., 2009;Prestbo and Gay, 2009).

Wet deposition
Wet deposition is very similar in the Hg + OH/O 3 model except where discussed below.
The model predicts the highest wet deposition in the coalburning regions of Europe and North America, reflecting near-field deposition of Hg II and Hg P emissions. Observations over Europe are elevated in the industrialized central region and show a poleward decrease in deposition with similar magnitude to the model. Over the Eastern US, the observations likewise show high deposition stretching from Texas to the Mid-Atlantic states, where anthropogenic mercury emissions are largest. At the northern end of this band the model exceeds observations, regardless of oxidant. Figure 10 shows the seasonal cycle of wet deposition in the Eastern US and reveals that the positive model bias in the Northeast occurs mainly in winter. Suppressing cold scavenging can eliminate the bias, as found by , but observations indicate that such scavenging occurs , and suppressing cold scavenging in the model would cause 40% underestimates of deposition in Alaska, Alberta and Finland. Fast reduction may compete with near-field deposition for the fate of Hg II emissions. We assume in the model that reduction is photochemical and therefore ineffective in winter, but it is possible that reduction occurs in all seasons or, equivalently, that the fraction of mercury emitted as Hg II is too high in current inventories Pongprueksa et al., 2008). This would decrease wet deposition over the Mid-Atlantic and Midwest emission regions Vijayaraghavan et al., 2008).
Sites around the Gulf of Mexico report the highest mercury wet deposition in North America, even though regional mercury emissions are lower than in the Northeast US. Convective scavenging of mercury from the free troposphere likely causes this regional feature (Guentzel et al., 2001;. The Hg + Br model underpredicts wet deposition here by 50%. While the Hg + OH/O 3 model is closer to observations in the northern Gulf region, it is still 40% lower than MDN sites in southern Florida. On a monthly basis, both models overlap the observed wet deposition range in the Gulf region during November-May, as seen in Fig. 10, but only the Hg + OH/O 3 model has a strong deposition peak during the wet summer months, which accounts for its better comparison with the annual mean. During these months OH provides a vigorous subtropical Hg II source available for convective scavenging in the Hg + OH/O 3 model, while there is little Br present in the Hg + Br model. Br concentrations could be larger than are specified here if ventilation of sea-salt-derived Br y from the MBL or heterogeneous reactivation of Br y are important (see Sect. 2.2). Figure 11 shows mean vertical profiles measured from aircraft during the INTEX-B and ARCTAS campaigns over North America and the Pacific and Arctic Oceans (Talbot et al., 2007, plus CARIBIC flights over the Atlantic Ocean, Eurasia and North America Slemr et al., 2009). Due to uncertain inlet loss of RGM, the measurements include Hg 0 plus some fraction of gaseous Hg II . This provides an upper limit for Hg 0 and a lower limit for TGM, and we refer to it here as Hg 0 * . We increase the INTEX-B measurements by 40% based on an in-flight intercomparison (Swartzendruber et al., 2008). The aircraft data are still ∼10% lower on average than the  Singh et al., 2009). We correct a low bias of 40% in the observations based on an in-flight intercomparison (Swartzendruber et al., 2008). Hg 0 . Of most interest here is the observed decline of concentrations above the tropopause. This decline is generally reproduced in the model if we assume that measurements quantify Hg 0 only (or equivalently that Hg II is present mainly in the aerosol). Vertical gradients across and above the tropopause are similar in the Hg + Br and Hg + OH/O 3 simulations, and so do not provide an effective test of the chemical mechanism.

Aircraft measurements
One prominent discrepancy is the inability of the model to simulate the steep decline above the tropopause in the Arctic springtime. Complete Hg 0 * depletions were common during ARCTAS in stratospheric air with O 3 > 100 ppb, and Hg 0 * was rarely detectable when O 3 exceeded 200 ppb, suggesting that oxidation increases abruptly above the tropopause (Talbot et al., 2007;Kim et al., 2009). We tested whether additional Br could be responsible by doubling it in the model throughout the stratosphere, corresponding to 4 ppt BrO in the lowermost stratosphere. Simulated Hg 0 decreased by only 10 ppq at 10 km. Much higher bromine concentrations are unlikely based on satellite observations (Chance, 1998) and constraints on the stratospheric bromine budget Salawitch et al., 2010). In another sensitivity test, we added Hg 0 oxidation by BrO to the model , but this reaction enhanced Hg 0 oxidation throughout the column rather than specifically in the stratosphere.
Additional oxidants in the springtime polar stratosphere might include Cl, Cl 2 , and BrCl generated by heterogeneous chemistry. Hg 0 oxidation reactions with these species are fast (Ariya et al., 2002;Donohoue et al., 2005) but limited by the low oxidant concentrations. The GMI Aura model predicts mean values of ∼1 ppt ClO in the lowermost stratosphere during spring ARCTAS, corresponding to 0.5 ppq Cl and up to 100 ppt Cl 2 and 30 ppt BrCl (Strahan et al., 2007). Based on the available kinetic data (mainly 298 K), the resulting lifetime of Hg 0 exceeds 1 year, too long to account for Hg 0 depletion. However, Thornton et al. (2003) observed much greater chlorine activation (∼10 ppt ClO) in the Arctic winter stratosphere than predicted by the GMI model. At these levels, Cl, Cl 2 and BrCl could become important Hg 0 oxidants.
The ARCTAS flights over California and Nevada  provided a first opportunity for detailed boundary layer mapping of a continental source region (Fig. 12). Polluted conditions during these flights caused intermittent low bias in one of the two instrument channels, which we correct by removing the lower value of each consecutive measurement pair. The highest concentrations were in biomass burning plumes sampled in both northern and southern California. The three most concentrated plumes had Hg/CO enhancement ratios of 90-130 nmol mol −1 and the mean enhancement ratio for all fire plumes (identified by CH 3 CN > 0.25 ppt) was 80 nmol mol −1 . Weiss-Penzias et al. (2007) and Finley et al. (2009) found similar Hg/CO enhancements (136 ± 60 nmol mol −1 ) in the Pacific Northwest during summers  found 60 nmol mol −1 during summer ARC-TAS flights over Canada, which are similar to ratios of 70-240 nmol mol −1 observed worldwide Friedli et al., 2009). Based on these measurements, we reduced the Hg/CO emission ratio for biomass burning in GEOS-Chem to 100 nmol mol −1 in this work (previously 210 nmol mol −1 ), as discussed in Sect. 2.1.
Apart from the fire plumes, the California observations show highest Hg 0 * near industry and ports in Los Angeles and Long Beach. Typical concentrations exceeded 200 ppq throughout the Los Angeles basin, following a pattern that closely resembles the emission distribution in the EPA source inventory (EPA, 2008). A fresh anthropogenic plume with high SO 2 encountered near the Mexican border does not correspond with any nearby sources in the inventory, suggesting that some industrial emissions in the border region are underestimated. Offshore marine airmasses contained up to 200 ppq Hg 0 * as well as elevated dimethyl sulfide indicative of ocean emissions. Mercury levels also persisted above 220 ppq for 150 km on a flight over active and inactive mines in western Nevada. These elevated concentrations are typical for summertime at surface sites in Nevada and may result from a mix of mining operations and naturally Hg-enriched soils (Lyman and Gustin, 2008).

Is atmospheric reduction necessary to explain observations?
Global models of atmospheric mercury require atmospheric reduction of 4000-10000 Mg a −1 Hg II to achieve an unbiased simulation of mean TGM observations with current emission inventories (Bergan and Rodhe, 2001;Lin et al., 2006;Seigneur et al., 2006;Selin et al., 2007, this work). While several reductants for Hg II have possible atmospheric relevance (see review by Ariya et al., 2008), atmospheric importance has not been established for any; so the role of Hg II reduction in the global atmosphere remains conjectural (Hynes et al., 2009). Fast reduction may occur in fresh power plant plumes Landis et al., 2009), but the global effect would be small because the anthropogenic Hg II source is only 700 Mg a −1 . We find in our Hg + Br model that all simulation results can be replicated without atmospheric reduction if we decrease the overall rate of Hg 0 conversion to Hg II by 60%. This could be accommodated within the range of theory-derived kinetic coefficients for the reaction HgBr + X → HgBrX (X ≡ Br, OH) (Goodsite et al., 2004;Balabanov et al., 2005) or by smaller Br concentrations. Until better constraints on Hg 0 oxidation rates are available, it appears that atmospheric reduction is not required to explain any of the major features of the global mercury cycle. Figure 1 shows the global atmospheric mercury budget derived from our Hg + Br simulation in GEOS-Chem. Emissions and deposition in our Hg + OH/O 3 simulation differ from the figure by less than 10%. The troposphere accounts for 99% of total atmospheric Hg 0 but only 50% of Hg II , reflecting the lack of Hg II chemical or depositional loss in the stratosphere. The Hg II burden in Fig. 1 includes inert particulate mercury Hg P but it contributes only 2 Mg. Nearly all redox fluxes occur in the troposphere, as seen in Fig. 3. Anthropogenic emissions here are 2050 Mg a −1 and total emissions are 8300 Mg a −1 , both within the literature range as described in Sect. 2.1. Although the original GEOS-Chem model of Selin et al. (2007) used similar anthropogenic emissions (2200 Mg a −1 ),  increased these to 3400 Mg a −1 to match observed TGM after adding Hg 0 dry deposition to the model. Our Hg + Br model matches observed TGM with smaller emissions because oxidation is slower, resulting in a longer atmospheric lifetime for mercury. Our Hg + OH/O 3 model achieves the same result by assuming faster reduction. Atmospheric reduction can be eliminated entirely in the Hg + Br model if the oxidation kinetic coefficients are reduced within their uncertainty, as described in Sect. 3.6. Land in the model emits 1200 Mg a −1 from soils plus 260 Mg a −1 from rapid photoreduction of Hg II deposited to vegetation and 260 Mg a −1 from snow. Even though we eliminated mercury evapotranspiration, the total land emissions are unchanged from  because of the constraint from preindustrial steady state.  extrapolated field flux measurements to estimate that terrestrial ecosystems emit 1650 Mg a −1 (range 860-3800 Mg a −1 ) including primary geogenic sources but excluding biomass burning. In our model the corresponding emission is 2200 Mg a −1 , well within that range.

Global mercury budget
Most atmospheric mercury is removed as Hg II (5100 Mg a −1 ) with the spatial pattern shown in Fig. 13. Wet deposition accounts for 3100 Mg a −1 and dry deposition for 800 Mg a −1 . Sea-salt particles take up an additional 1200 Mg a −1 and this accounts for 35% of Hg II deposition to the ocean. Global Hg 0 dry deposition is 3200 Mg a −1 , but emissions offset this so that oceans and soils everywhere are net sources of atmospheric Hg 0 . After accounting for all mercury species, the deep ocean in the model sequesters 1600 Mg a −1 from the atmosphere, similar to the previous GEOS-Chem model version (2100 Mg a −1 ) .
The Hg + OH/O 3 model generates a very different Hg II deposition pattern from the Hg + Br model, except for close to Hg II sources, as seen in Fig. 13. With the Hg + OH/O 3 oxidation mechanism, deposition is largest in the tropics where [OH] is greatest and deep convective rain occurs frequently. The Hg + Br model has greater Hg II deposition in the Southern Hemisphere due to the oxidation differences seen in Fig. 3. Despite these large-scale differences, both oxidation mechanisms predict similar wet deposition at monitoring sites in North America and Europe because of the anthropogenic influence (see Sect. 3.4). Figure 13 implies that wet deposition measurements in the tropics and Southern Hemisphere could distinguish between oxidation mechanisms. Furthermore, the Hg + Br model suggests that mercury inputs to Southern Ocean ecosystems may be much greater than previously thought.

Conclusions
We have added Hg 0 oxidation by gas-phase bromine atoms to a global 3-D atmospheric model (GEOS-Chem) to test whether this reaction is consistent with the observed patterns of atmospheric mercury concentration and deposition. We compare the model performance with Br as the sole oxidant ("Hg + Br model") against a model in which OH and ozone are the only oxidants ("Hg + OH/O 3 model"). While both oxidation mechanisms, and possibly others, may operate together in the real atmosphere, these idealized simulations enable us to explore the constraints that observations place on the atmospheric chemistry of mercury.
Total mercury emissions in the model are 8300 Mg a −1 , including 2050 Mg a −1 from anthropogenic sources. A new snowpack reservoir stores deposited mercury and reemits it under sunlight at a temperature-dependent rate. The seasonal cycle of Arctic Hg 0 implies that 60% of mercury deposited to snow is eventually reemitted while the remainder (60 Mg a −1 ) transfers to the ocean and soils.
Hg + Br kinetics here follow Goodsite et al. (2004) and Donohoue et al. (2006), while Hg + OH/O 3 kinetics are identical to . Global bromine distributions derive from tropospheric and stratospheric chemistry models, which are constrained by precursor gas measurements. We also specify BrO concentrations in the marine and polar springtime boundary layer based on observations. Hg 0 has a 6 month chemical lifetime in the Hg + Br model and a 3.7 month lifetime in the Hg + OH/O 3 model. Matching atmospheric observations with the imposed anthropogenic emission inventory requires an atmospheric lifetime of 7.3 months for TGM, which we achieve by invoking photochemical reduction of Hg II in clouds (at a faster rate in the Hg + OH/O 3 model). This reduction would be unnecessary in the Hg + Br model if we decreased the overall Hg 0 oxidation rate by 60%, which is within the uncertainty of the HgBr + X + M reaction step.
The Hg + Br and Hg + OH/O 3 models both provide unbiased simulations of TGM surface concentrations and their spatial variance. In particular, the Hg + Br model reproduces the interhemispheric gradient of TGM, which contradicts an earlier study (Seigneur and Lohman, 2008). Observed seasonal cycles of TGM at mid-latitudes are also consistent with both models. However, only the Hg + Br model reproduces the spring depletion and summer rebound observed at polar sites. The Hg + Br model also provides a better simulation of Hg 0 oxidation during subsidence events over Antarctica.
Wet deposition flux patterns of mercury observed over Europe and North America are generally reproduced in the model. Simulated deposition in the Northeast US in winter is too high regardless of oxidant, which could reflect excessive scavenging by snow, reduction of Hg II in power plant plumes, or speciation error in the emission inventory. The Southeast US summer maximum in mercury wet deposition is better simulated by the Hg + OH/O 3 model, where it reflects scavenging of Hg II from the free troposphere by deep convection.
Vertical profiles from CARIBIC, INTEX-B, and ARCTAS aircraft show uniform concentrations in the troposphere and declines above the tropopause. We reproduce these features in the Hg + Br model except in Arctic spring where the observed stratospheric depletion is strongest. Neither Br nor BrO can explain the extreme stratospheric Hg 0 depletion in Arctic spring and the Hg + OH/O 3 model does no better. We suggest that Cl, Cl 2 , or BrCl might be important in the springtime stratosphere, and estimate that the required concentrations are within the range of ClO observations. Two major effects of using Br as the Hg 0 oxidant instead of OH and ozone are to lower Hg 0 concentrations in the Southern Hemisphere and to increase mercury deposition to the Southern Ocean. The Hg + OH/O 3 model, in contrast, has peak deposition in the tropics. Mercury concentration and deposition data in the tropics and southern mid-latitudes would be necessary to test these patterns.
Our results show that gas-phase bromine is viable as the main global oxidant for Hg 0 , producing a TGM lifetime and distribution consistent with nearly all available observations. Most of the oxidation occurs in the free troposphere where Br concentrations are constrained by bromocarbon measurements. We also find that atmospheric reduction of Hg II may not be necessary to match observed Hg 0 concentrations if we decrease Hg 0 oxidation kinetics within its uncertainties.