We start from the standard version v9-02 of the GEOS-Chem Hg model (www.geos-chem.org; Amos et al., 2012; Song et al., 2015) and implement a new 2010 inventory for speciated anthropogenic emissions (Zhang et al., 2016) including contributions from commercial products (Horowitz et al., 2014). We then add original updates, described below, for atmospheric chemistry and atmosphere–ocean coupling. The standard v9-02 model includes cycling of Hg0 and HgII between the atmosphere, land, and the surface mixed layer of the ocean (Selin et al., 2008; Soerensen et al., 2010). The atmospheric model transports Hg0 and HgII as separate tracers. Gas–particle partitioning of HgII is parameterized following Amos et al. (2012) as a thermodynamic equilibrium function of local temperature and mass concentration of fine particulate matter (PM2.5). This parameterization is based on observed relationships between the gas / particle HgII concentration ratio and PM2.5, and it does not resolve effects from particle composition or surface area.

Holmes et al. (2010) presented the last detailed global atmospheric budget analysis of Hg in GEOS-Chem, and we use it as a point of comparison for this study. Their model version included Hg0 oxidation by Br atoms, with Br concentrations specified in the troposphere and stratosphere from the p-TOMCAT and NASA Global Modeling Initiative models, respectively (Yang et al., 2005; Strahan et al., 2007). The standard v9-02 model uses Br concentration fields from the GEOS-Chem tropospheric bromine simulation by Parrella et al. (2012) and features other model updates relative to Holmes et al. (2010), including the gas–particle HgII equilibrium partitioning described above, a corrected washout algorithm (Amos et al., 2012), and improvements to the 2-D surface ocean model (Soerensen et al., 2010). In this work we update the Br concentration fields to a more recent version of GEOS-Chem (Schmidt et al., 2016), as discussed further in Sect. 3.2.

We conduct a 3-year simulation for 2009–2011 driven by GEOS-5 assimilated meteorological data from the NASA Global Modeling and Assimilation Office (GMAO) with native horizontal resolution of 1/2∘ × 2/3∘ and 72 vertical levels extending up to the mesosphere. The GEOS-Chem simulation is conducted at 4∘ × 5∘ horizontal resolution by regridding the GEOS-5 meteorological data. We use anthropogenic Hg emissions for the year 2010 from Zhang et al. (2016). The spatial distribution of soil Hg is determined following the method of Selin et al. (2008), with updated soil respiration emissions (910 Mg Hg yr-1) for consistency with the mechanistic terrestrial model developed by Smith-Downey et al. (2010). Emissions from snow and re-emission of deposited Hg follow Selin et al. (2008). The model is initialized with a 15-year simulation to equilibrate the stratosphere. We present model results as averages for the three simulation years (2009–2011).

Hg0 / HgII redox chemistry in the standard GEOS-Chem v9-02 model includes Hg0 oxidation by Br atoms and HgII in-cloud photoreduction as described by Holmes et al. (2010), with Br concentration fields from Parrella et al. (2012). Shah et al. (2016) updated that chemistry following Dibble et al. (2012) to include second-stage oxidation of HgBr by HO2, NO2, and BrO, as well as new kinetics for HgBr dissociation. Here we further expand and update the chemistry using the mechanism described in Table 1. Br and Cl radical concentrations are taken from a new GEOS-Chem simulation of tropospheric oxidant–aerosol chemistry by Schmidt et al. (2016). Unlike Parrella et al. (2012), this simulation does not include a bromine radical source from sea salt debromination as recent evidence suggests that BrO concentrations in the marine boundary layer (MBL) are generally sub-ppt (Gómez Martín et al., 2013; Wang et al., 2015). Simulated concentrations of Br in the free troposphere are, on the other hand, a factor of 2 higher than Parrella et al. (2012), reflecting more extensive heterogeneous chemistry. Schmidt et al. (2016) show that their simulation provides a better simulation of aircraft and satellite observations of tropospheric BrO. Stratospheric concentrations of Br and Cl species are from the GEOS Chemistry Climate Model (Liang et al., 2010) and Global Modeling Initiative (Considine et al., 2008; Murray et al., 2012), respectively.

We use the global 3-D monthly archive of oxidant and radical concentrations from the GEOS-Chem simulation of Schmidt et al. (2016) that included detailed gas and aerosol chemistry. We apply diurnal scaling following Holmes et al. (2010) to the monthly mean concentrations of Br, BrO, Cl, ClO, and HOCl; a cosine function of the solar zenith angle for OH and HO2; and NO–NO2–O3 photochemical equilibrium for NO2. Schmidt et al. (2016) evaluated their simulated tropospheric BrO concentrations (global daytime mean of 0.96 ppt) with observations from satellite and aircraft and found no systematic bias. GEOS-Chem concentrations of NO2 and H2O2 (proxy for HO2) have been evaluated successfully in a number of aircraft campaigns (Martin et al., 2006; Hudman et al., 2007; Singh et al., 2007; Lin and McElroy, 2010; Mao et al., 2010; Travis et al., 2016). GEOS-Chem NO2 columns have also been evaluated against satellite observations over China (Lin et al., 2012), North America (e.g., Lamsal et al., 2014), and Africa (Marais et al., 2012). Fast Hg0 oxidation in the polar spring boundary layer is simulated by specifying high BrO concentrations when conditions for temperature, sea ice cover, sunlight, and atmospheric stability are met (Holmes et al., 2010). Monthly mean OA concentrations are archived from a separate v9-02 GEOS-Chem simulation including primary emissions from combustion and secondary production from biogenic and anthropogenic hydrocarbons (Pye et al., 2010). An evaluation of modeled OA against aircraft observations globally is presented in Heald et al. (2011), which showed no systematic bias in remote environments but an underestimate of median concentrations in polluted regions. Aqueous-phase concentrations of Hg0, O3, and HOCl are determined from the gas-phase partial pressures and Henry's law coefficients (Table 1). We estimate aqueous in-cloud OH concentrations following Jacob et al. (2005) as [OH(aq)] = β [OH(g)], with β=1×10-19 M cm3 molecule-1.

GEOS-Chem v9-02 uses a 2-D surface–slab ocean model with no lateral transport and with fixed subsurface ocean concentration boundary conditions (Soerensen et al., 2010). Oceanic circulation and mixing between the surface mixed layer and deeper ocean strongly impact Hg0 concentrations in the surface ocean through the supply of reducible HgII, driving changes in oceanic evasion (Soerensen et al., 2013, 2014). Here we present a new two-way coupling of the GEOS-Chem atmospheric Hg simulation to the MITgcm 3-D oceanic general circulation model with embedded ecology (Y. X. Zhang et al., 2015). First, the GEOS-Chem Hg model with the 2-D slab ocean is initialized over 15 years of repeated present-day emissions and meteorological data (15 years is required to equilibrate the stratosphere). Starting from these initial conditions, we conduct the GEOS-Chem simulation for the desired period (here 3 years: 2009–2011) and archive monthly surface air Hg0 concentrations and total HgII deposition fluxes. We then initialize the Y. X. Zhang et al. (2015) ocean model for 20 years with these archived surface boundary conditions to equilibrate the upper several hundred meters of the ocean. The monthly mean surface ocean Hg0 concentrations from the final year of the simulation are then input to GEOS-Chem, replacing the 2-D slab ocean model concentrations. The 15-year GEOS-Chem simulation is repeated with this new input. This process is iterated until convergence of results is achieved. We find that two iterations are sufficient.

Figure 1 (top panel) shows the simulated vertical and latitudinal distributions of annual zonal mean Hg0 and HgII mixing ratios. The fraction of tropospheric HgII in the aerosol phase ranges from less than 10 to more than 90 % (Amos et al., 2012). Modeled Hg0 decreases rapidly in the stratosphere, while HgII increases with altitude and dominates total Hg in the stratosphere. This vertical structure is driven by chemistry (Selin et al., 2007; Holmes et al., 2010), and these general vertical trends in the two species are consistent with limited available observations (Murphy et al., 2006; Talbot et al., 2007; Lyman and Jaffe, 2012). The exact partitioning of total Hg in the stratosphere between Hg0 and HgII is uncertain, and GEOS-Chem predicts a higher HgII fraction relative to other models (Bieser et al., 2016). Photodissociation of gas-phase HgII halides may be possible at ultraviolet wavelengths (Maya, 1977), but whether this is important in the mid-to-upper stratosphere requires further investigation. The stratosphere in the model accounts for 12 % of total Hg atmospheric mass and is in a very different redox regime from the troposphere. Here we focus on tropospheric budgets, which are most relevant to Hg deposition. The stratosphere is discussed further in Sect. 4.4.

Figure 2 shows the global tropospheric Hg budget from our GEOS-Chem simulation, including the rates of the major Hg0 / HgII cycling pathways. The global tropospheric Hg reservoir is 3900 Mg, including 3500 Mg as Hg0 and 400 Mg as HgII, smaller than the 4500 Mg reservoir in Holmes et al. (2010). In both cases, the reservoir was adjusted through the HgII photoreduction rate coefficient to match the observed annual mean TGM at long-term monitoring sites, mainly located at northern midlatitudes (Fig. 3). Holmes et al. (2010) used observations for 2000–2008 averaging 1.87 ± 1.00 ng m-3 STP (p=1 atm, T=273 K; n=39 sites), whereas we use observations for 2007–2013 averaging 1.46 ± 0.25 ng m-3 STP (n=37). Observed atmospheric concentrations in North America and Europe have declined on average by 1–2 % yr-1 since 1990, which has been attributed to the phase-out of Hg from commercial products and declines in coal Hg emissions as a co-benefit of SO2 and NOx emission controls (Zhang et al., 2016). Observations used in Holmes et al. (2010) also included very high concentrations at four East Asian sites (3–7 ng m-3 STP; Nguyen et al., 2007; Sakata and Asakura, 2007; Feng et al., 2008; Wan et al., 2009), whereas the more recent observations over East Asia used here (n=6) are all less than 3 ng m-3 STP (Sheu et al., 2010; Fu et al., 2012a, b; H. Zhang et al., 2015). Only one site (Mt. Changbai, China) has data available during both time periods, and observed TGM concentrations decreased there by a factor of 2 (Fu et al., 2012b). Thus, our atmospheric loadings may be more representative of recent (ca. 2010) conditions.

We calculate several atmospheric lifetimes to understand the processes driving Hg deposition: the chemical lifetime of Hg0 against oxidation, the chemical lifetime of HgII against reduction, the lifetime of HgII against deposition, and the lifetime of total gaseous mercury (TGM ≡ Hg0 + HgII(g)) against deposition. We find that the lifetime of Hg0 against oxidation to HgII in the troposphere is 2.7 months, with Br atom-initiated pathways contributing 97 % of total oxidation. This chemical lifetime is smaller than most prior model estimates, e.g., 6 months in Holmes et al. (2010), which used lower Br concentrations. The addition of NO2 and HO2 as second-stage oxidants is also important in increasing the rate of HgBr conversion to HgII, more than compensating for faster thermal decomposition of HgBr relative to Holmes et al. (2010). Our results are consistent with Shah et al. (2016), who estimated a global annual tropospheric Hg0 lifetime against oxidation to HgII by Br atoms of 1.2 to 2.8 months, based on their simulation of measurements of HgII over the southeastern US in summer during the Nitrogen, Oxidants, Mercury and Aerosol Distributions, Sources and Sinks (NOMADSS) campaign (Gratz et al., 2015). The dominance of NO2 and HO2 as HgBr oxidants reflects their high atmospheric abundance and is consistent with results from a box modeling study over the Pacific (Wang et al., 2014). We find Cl atoms contribute less than 1 % of Hg0 oxidation globally because the supply of Cl atoms (mean tropospheric mixing ratio of 2×10-4 ppt) is limited by fast conversion to the stable reservoir HCl. Aqueous-phase pathways in clouds contribute 2 %. Our work suggests that the dominant HgII species produced in the gas phase are BrHgONO and BrHgOOH, though the actual speciation of HgII in the atmosphere would likely be modified through cycling in aerosols and clouds (Hedgecock and Pirrone, 2001; Lin et al., 2006). We did not consider heterogeneous Hg0 oxidation on particle surfaces (e.g., Vidic et al., 1998; Flora et al., 1998; Lee et al., 2004) due to inadequate information to formulate atmospheric rates. Further study of heterogeneous Hg0 oxidation is needed (Ariya et al., 2015).

The vertical distribution of Hg0 → HgII oxidation rates is shown in the bottom panel of Fig. 1 in units of number density (relevant to the mass budget) and mixing ratio (relevant to transport). Only 1 % of Hg0 oxidation occurs in the stratosphere, consistent with previous work (Holmes et al., 2010). Most Hg oxidation by mass occurs in the extratropical free troposphere, consistent with the Br distribution (Schmidt et al., 2016), and is faster in the Northern Hemisphere because of higher NO2 concentrations. Hg0 oxidation in Holmes et al. (2010) was fastest in the Southern Ocean MBL due to high Br concentrations from sea salt aerosol debromination, but debromination in that region is unlikely since the sea salt aerosol remains alkaline (Murphy et al., 1998; Alexander et al., 2005; Schmidt et al., 2016). In general, the addition of second-stage oxidants HO2 and NO2 shifts Hg oxidation to lower latitudes relative to Holmes et al. (2010). This has implications for Hg deposition, which we discuss in Sect. 5. Hg0 oxidation in terms of mixing ratio features a secondary maximum in the tropical upper troposphere, where the Br / BrO ratio is high (Parrella et al., 2012; Fernandez et al., 2014).

We find that the shorter chemical lifetime of Hg0 in our simulation relative to Holmes et al. (2010) must be balanced by faster atmospheric reduction to reproduce observed TGM concentrations. This is implemented by adjusting the photoreduction coefficient α in Table 1. The resulting mean lifetime of HgII against reduction in the troposphere is 13 days. Shah et al. (2016) similarly found that faster Hg0 oxidation as needed to match NOMADSS observations required faster HgII reduction, with a tropospheric HgII lifetime against reduction of 19 days. The lifetime of tropospheric HgII against deposition is relatively long, 26 days (see Fig. 2), because most HgII production occurs in the free troposphere, where precipitation is infrequent (Fig. 1). HgII reduction in the model is faster in the Northern Hemisphere than in the Southern Hemisphere because of higher OA concentrations. Hence HgII concentrations are higher in the Southern Hemisphere than in the Northern Hemisphere (Fig. 1).

We find here that the HgII lifetime against reduction is shorter than against deposition, emphasizing the importance of reduction in controlling the atmospheric Hg budget. By contrast, Holmes et al. (2010) found an HgII tropospheric lifetime of 50 days against reduction and 36 days against deposition, which led them to conclude that no reduction would be needed if Hg0 oxidation kinetics were decreased within their uncertainty. This is no longer possible in our simulation because of the much faster Hg0 oxidation. We conclude that atmospheric HgII reduction must take place, although the mechanism is not established. With HgII reduction, the overall lifetime of tropospheric TGM against deposition in our simulation is 5.2 months, similar to the estimate of 6.1 months in Holmes et al. (2010). We discuss the consistency of this estimate with observations in the next section.

Figure 3 compares our simulation to observed 2007–2013 TGM surface concentrations. TGM includes Hg0 and the gaseous component of HgII. Observations include annual means at 37 land sites plus ship cruises. The spatial correlation coefficient between model and observations is r=0.57. The mean at land stations is 1.47 ± 0.27 ng m-3 STP in the observations and 1.44 ± 0.25 ng m-3 STP in the model, with the good agreement in the mean reflecting the adjustment of the HgII photoreduction rate coefficient as explained above.

Beyond this simulation of the mean, we successfully reproduce the observed standard deviation of TGM concentrations, which places an independent constraint on the atmospheric lifetime of Hg against deposition (Junge, 1974; Hamrud, 1983). This constraint had previously been expressed in terms of the interhemispheric gradient of TGM from ship cruises, leading to TGM lifetime estimates ranging from 4.4 months to 2 years (Slemr et al., 1981; Fitzgerald et al., 1983; Lindqvist and Rodhe, 1985; Lamborg et al., 2002). However, we find the interhemispheric gradient in surface concentrations is not a sensitive diagnostic of lifetime because surface atmospheric Hg in the Southern Hemisphere is controlled more by atmosphere–ocean exchange than by transport from the Northern Hemisphere. For example, changing the simulation of atmosphere–ocean exchange from the slab ocean to the MITgcm (Sect. 3.3) increases the TGM lifetime while also increasing the interhemispheric gradient due to changed ocean emissions. We conducted a sensitivity simulation doubling the tropospheric TGM lifetime in the model (to 10.4 months) by decreasing the rate of Hg0 oxidation. This produced less than 20 % of a decline in the interhemispheric gradient, but the relative standard deviation of TGM concentrations across land stations decreased by 40 %. Thus the overall variability of TGM concentrations provides a more sensitive constraint on the TGM atmospheric lifetime, while the interhemispheric gradient can be misleading. The model's ability to reproduce the observed standard deviation across land sites supports our simulated TGM atmospheric lifetime of 5.2 months.

Figure 4 compares simulated and observed seasonal cycles of TGM concentrations at land stations at northern midlatitudes (30 to 60∘ N) and in the Southern Hemisphere (see Fig. 3 for site locations). Also shown are results from a sensitivity simulation with the 2-D surface–slab ocean model (Soerensen et al., 2010). The observed seasonal cycle is successfully reproduced in the Northern Hemisphere. The February maximum and September minimum are driven in the model in part by Hg0 oxidation and in part by ocean evasion. Seasonality in radical oxidant concentrations outweighs temperature effects, leading to fastest oxidation in summer and slowest in winter. Modeled oceanic Hg0 evasion from the North Atlantic Ocean peaks in winter and early spring, because higher wind speeds increase mixing and entrainment of reducible HgII from subsurface waters and enhance the rate of air–sea exchange, and Hg removal via particle settling during this time is low (Sorensen et al., 2010, 2013). Similar to Song et al. (2015), we find that the 2-D slab ocean overestimates the observed seasonality at northern midlatitude sites (Fig. 4). Atlantic and Pacific Ocean seawater Hg0 measurements indicate this overestimate results from oversimplified ocean physics and the fixed boundary condition for subsurface ocean concentrations in the slab ocean model (Soerensen et al., 2014). Previous work shows one-way coupling to the MITgcm ocean simulation improves simulated surface ocean Hg0 concentrations compared to observations (Soerensen et al., 2014). Here, the two-way coupling to the MITgcm also improves the seasonality in atmospheric Hg.

In the Southern Hemisphere, Amsterdam Island in the Indian Ocean and Cape Point on the South African coast have similar monthly average concentrations and show no significant seasonal variation (Angot et al., 2014; Slemr et al., 2015). By contrast, the model at those sites has strong seasonality, as previously shown for the standard model version v9-02 (Song et al., 2015). We find a modest improvement in moving from the slab ocean simulation to the MITgcm, but the seasonal bias is still large. Long-range transport of atmospheric Hg from Antarctica could influence observed seasonality at these sites (Angot et al., 2016). Capturing the seasonality of atmospheric Hg in the Arctic using GEOS-Chem required parameterization of unique sea ice, oceanic, and riverine dynamics (Fisher et al., 2012, 2013; Y. Zhang et al., 2015); a similar analysis for the Southern Ocean region, with its distinct productivity dynamics impacting Hg cycling (e.g., Cossa et al., 2011; Gionfriddo et al., 2016), has not yet been performed.

Hg0 concentrations in the model drop rapidly above the tropopause (Fig. 1), consistent with observations (Talbot et al., 2007; Slemr et al., 2009). Modeled Hg0 oxidation in the lower stratosphere is almost exclusively from Br atoms. There is no significant HgII reduction there because OA concentrations and relative humidity are very low. Extensive TGM observations in the lowermost stratosphere at northern extratropical latitudes are available from the Civil Aircraft for the Regular Investigation of the atmosphere Based on an Instrument Container (CARIBIC) program, which collects measurements aboard commercial aircraft (Slemr et al., 2009, 2014, 2016). The early stratospheric data were biased low, and we focus on corrected data available for April 2014–January 2015 (Slemr et al., 2016). TGM in the stratosphere is expected to be mainly Hg0 because HgII is incorporated into aerosol (Murphy et al., 2006; Lyman and Jaffe, 2012).

Figure 5 compares the stratospheric TGM concentrations measured in CARIBIC to model values sampled along the flight tracks. The CARIBIC data also include ozone (O3) and CO concentrations (Brenninkmeijer et al., 2007). Here we use O3 concentration as a chemical coordinate for depth into the stratosphere and exclude tropospheric data as diagnosed by [O3] / [CO] < 1.25 mol mol-1 (Hudman et al., 2007). We correlate the logarithm of TGM concentrations, as a measure of first-order loss, to the O3 concentrations. The observations reach higher ozone than the model, indicating that they sample air with greater stratospheric influence. The model underestimates the log(TGM)–ozone slope by a factor of 2. To test whether this underestimate is due to slow Hg0 oxidation, we performed a sensitivity simulation doubling the stratospheric Hg0 oxidation rate (green line in Fig. 5). The model slope increased by only 32 %. This suggests that the log(TGM)–ozone relationship in the lowermost stratosphere is set in part by mixing rather than solely by chemistry (Xiao et al., 2007). The model underestimate of the log(TGM)–ozone slope could reflect excessive dynamical mixing in the lower stratosphere, a well-known problem in stratospheric transport models (Schoeberl et al., 2003; Tan et al., 2004), or errors in the timescales of air transit across the tropopause, which vary on the order of months to years (Orbe et al., 2014; Ploeger and Birner, 2016).