Constraints from observations and modeling on atmosphere–surface exchange of mercury in eastern North America

Mercury (Hg) is a trace metal that has adverse effects on human health and the environment (Driscoll et al., 2013). Unlike most other metals, mercury in the lower atmosphere exists primarily in its elemental form (Hg⁰; also known as gaseous elemental mercury, GEM), which is volatile and poorly soluble (Lin and Pehkonen, 1999). The major loss pathway of atmospheric Hg⁰ is via oxidation to its oxidized state (Hg^II), a much more soluble form that partitions between the gas and particle phases, and subsequently undergoes dry and wet deposition (Lindberg et al., 2007; Selin, 2009). After being deposited on land and water surfaces, Hg^II can be reduced photochemically and/or biochemically to Hg⁰, which can be reemitted back to the atmosphere (Selin et al., 2008). The atmosphere–surface exchange of Hg⁰ is thus bidirectional: both evasion (upward) and dry deposition (downward) occur at the interface (Gustin and Lindberg, 2005; Xu et al., 1999). Wet deposition of Hg⁰ is insignificant due to its low solubility (Maestas, 2011). A positive (negative) net flux indicates that Hg⁰ evasion exceeds (is lower than) its dry deposition (Gustin et al., 2006). Hg⁰ exchange fluxes are important components of global Hg emissions, which also include anthropogenic sources, geogenic activities, and biomass burning (Pirrone et al., 2010). Despite their global importance, Hg⁰ fluxes from terrestrial and oceanic surfaces remain poorly constrained at global and regional scales, leading to considerable uncertainty in our understanding of the biogeochemical Hg cycle.

Based on field flux measurements and modeling studies, it remains unclear whether the global terrestrial ecosystem acts as a net source or a net sink of atmospheric Hg⁰. More than 100 field studies have been conducted since the 1970s to measure terrestrial Hg⁰ fluxes from various surface types (e.g., bare soil, forest, grassland, and snow/ice) and to determine environmental factors influencing Hg⁰ exchange (e.g., Lindberg and Turner, 1977; Xiao et al., 1991; Poissant et al., 2004; Ferrari et al., 2005; Obrist et al., 2005; Fritsche et al., 2008; Fu et al., 2010). A recent global database summarizing these studies shows a high variability in measured Hg⁰ fluxes over terrestrial surfaces (Agnan et al., 2016). Small-scale (<0.1 to 10³ m²) measurements, although unevenly distributed in time and among regions and landscapes, have been extrapolated to estimate terrestrial Hg⁰ fluxes at a global scale (for example, (1000–3410 Mg yr^-1, Mason, 2009); -2810 to 4650 Mg yr^-1 (50% uncertainty range; Agnan et al., 2016)). Agnan et al. (2016) has suggested that the largest uncertainty in estimating global terrestrial Hg⁰ fluxes arises from the forest ecosystem, which has a median flux of -59 Mg yr^-1 (a small net sink) with a very large 50% uncertainty range (-2580–3276 Mg yr^-1).

Most atmospheric mercury models treat Hg⁰ emission from, and dry deposition to, land separately. Dry deposition is typically simulated using a resistance-in-series scheme (Wesely, 1989; Zhang et al., 2009). Some models calculate soil emissions with empirical equations involving important controlling factors such as solar radiation, temperature, and substrate Hg concentrations (Selin et al., 2008; Holmes et al., 2010; Lei et al., 2013; Lin et al., 2010; Shetty et al., 2008). Terrestrial emissions in some other models are mapped according to biogenic CO emission and/or historical Hg deposition and scaled by global Hg budgets (Dastoor et al., 2015; Chen et al., 2015; Jung et al., 2009). Simulated global Hg⁰ fluxes from terrestrial surfaces range from -1300 to 3500 Mg yr^-1 (Holmes et al., 2010; Chen et al., 2015; Lei et al., 2013; Corbitt et al., 2011; Smith-Downey et al., 2010; Kikuchi et al., 2013; Amos et al., 2013, 2014; De Simone et al., 2014). Net Hg⁰ fluxes from the contiguous United States may also be positive or negative based on flux scaling methods (-183 to 269 Mg yr^-1, 50% uncertainty range; Agnan et al., 2016) and a newly developed bidirectional exchange model (118–141 Mg yr^-1, Wang et al., 2014).

The global ocean is believed to be a net Hg⁰ source, but its magnitude is uncertain. Air–ocean Hg⁰ exchange can be estimated using flux chamber techniques or by measuring the gradient between air Hg⁰ and oceanic Hg⁰ (also known as dissolved gaseous mercury, DGM) levels (Gårdfeldt et al., 2003). Most surface ocean waters have been observed to be supersaturated in DGM, generating a positive net flux of Hg⁰ to the atmosphere (Sprovieri et al., 2010; Soerensen et al., 2014; Kuss et al., 2011; Ci et al., 2011a). Global Hg⁰ fluxes from oceanic surfaces have been estimated in the range of 800–5500 Mg yr^-1 by different numerical models (Strode et al., 2007; Soerensen et al., 2010; Zhang et al., 2014; De Simone et al., 2014; Lei et al., 2013; Chen et al., 2015; Amos et al., 2013, 2014; Sunderland and Mason, 2007) and using flux scaling methods (Mason and Sheu, 2002).

Atmospheric Hg⁰ data can provide constraints on terrestrial and oceanic sources and sinks. For example, Hg⁰ observations at a background site in Nova Scotia, Canada, show that air originating from the Northwest Atlantic has about 0.06 ng m^-3 higher (p < 0.05, independent samples t-test) concentrations than air originating over terrestrial surfaces, implying Hg⁰ evasion from the ocean (Cheng et al., 2014). A recent application of global-scale inverse modeling combined a global mercury model and atmospheric Hg⁰ concentrations observed at multiple ground-based sites to constrain Hg⁰ fluxes from terrestrial and oceanic surfaces (Song et al., 2015). However, the relatively large systematic uncertainty (or bias) of Hg⁰ concentration measurements at different sites (Slemr et al., 2015), about 10% estimated from several side-by-side intercomparison experiments (e.g., Ebinghaus et al., 1999; Temme et al., 2006), limits the assessment of terrestrial and oceanic Hg⁰ fluxes across multiple sites (Song et al., 2015). This systematic uncertainty is minimized, however, when a consistent instrument is used to measure atmospheric Hg⁰ at different locations. Here, we use aircraft measurements to accomplish this task. In this study, we conduct atmospheric mercury chemical transport modeling and quantitatively compare these model results with Hg⁰ data from the Nitrogen, Oxidants, Mercury and Aerosol Distributions, Sources and Sinks (NOMADSS) aircraft campaign as well as ground- and ship-based observations. Our goal is to provide constraints on the large scale terrestrial and oceanic fluxes of Hg⁰ from eastern North America.

In this section, we first present the aircraft-, ground-, and ship-based observations used for this study. A brief description of the GEOS-Chem Hg chemical transport model (CTM) is then given and different model simulations we perform are introduced.

The NOMADSS campaign was conducted onboard the National Science Foundation (NSF)/National Center for Atmospheric Research (NCAR) C-130 aircraft from June 1 to July 15, 2013, during which 19 research flights were made out of Smyrna, Tennessee (Figure 1). NOMADSS was part of the Southeast Atmosphere Study (SAS; http://www.eol.ucar.edu/field_projects/sas), a field campaign that focused on southeastern U.S. regional air quality and climate. Mercury concentrations were measured using the onboard University of Washington’s Detector for Oxidized Hg Species (DOHGS). Its design, configuration, and calibration have been described in detail by Lyman and Jaffe (2012) and Ambrose et al. (2013, 2015), and are briefly summarized here. The DOHGS was used to simultaneously and continuously measure total Hg (THg) and Hg⁰ concentrations using two parallel channels, with Hg^II determined by their difference. The two channels sample ambient air at 1 standard liter per minute from a heated (110 °C) Teflon sample line connected to a rear-facing aircraft inlet. In the THg channel, air passes through a heated (650 °C) quartz pyrolyzer to reduce all Hg^II species to Hg⁰. In the Hg⁰ channel, a trapping sorbent, either quartz wool or a cation exchange membrane, is used to selectively remove Hg^II from the air. Each channel employs a Tekran^® 2537B analyzer to detect Hg⁰ at a time resolution of 2.5 minutes, using cold vapor atomic fluorescence spectrometry after pre-concentration on two alternating gold traps and subsequent thermal desorption. A customized Hg⁰ permeation source was applied to perform the pre-, in-, and post-flight calibrations during NOMADSS. Mercury concentrations were reported in nanograms per cubic meter (ng m^-3) at standard temperature and pressure. The overall measurement uncertainty was calculated as the quadratic sum of the 1σ precision and calibration uncertainties, varying from 6–10% for Hg⁰ among different research flights. The NOMADSS speciated Hg observations have been used to assess oxidation of Hg⁰ by bromine in the subtropical Pacific free troposphere (Shah et al., 2016; Gratz et al., 2015), to quantify THg enhancement ratios for coal-fired power plants (Ambrose et al., 2015), and to evaluate the Hg outflow from the Chicago/Gary urban/industrial area (Gratz et al., in preparation).

Other NOMADSS measurements used here to help interpret mercury observations include isoprene, dimethyl sulfide (DMS), propane, SO₂, NO_x, and meteorological and state variables. Isoprene is known as a tracer for terrestrial forest emissions and DMS is a marker for marine emissions (Seinfeld and Pandis, 2006). Isoprene was measured using the Proton Transfer Reaction Mass Spectrometer (Yuan et al., 2015). Propane and DMS were measured by the Trace Organic Gas Analyzer, an online Gas Chromatograph Mass Spectrometer (Apel et al., 2010). A Thermo Scientific model 43i-TLE monitor was used to collect SO₂ data. NO_x (= NO + NO₂) data were collected from an in situ chemiluminescence instrument (Ridley and Grahek, 1990). We average all these measurements to a 2.5-minute resolution to be consistent with the DOHGS Hg data.

As shown in Figure 1, Hg⁰ concentration observations at 11 ground-based rural/remote sites in the eastern United States are drawn from the National Atmospheric Deposition Program’s Atmospheric Mercury Network (NADP AMNet). Gay et al. (2013) summarized site characteristics, as well as instrumentation, standard operating procedures, and quality assurance in the NADP AMNet network. Briefly, Tekran^® analyzers are used at these sites with sampling intervals of 5–30 minutes. The original high-frequency observational data from 2009 to 2013 are converted into hourly averages and then into monthly averages. We require at least 30 minutes of data to derive an hourly average and at least 10 days of data to derive a monthly average. We use mercury observations from three summertime ship cruises (denoted AUG08, JUN09, and AUG10 for the August 2008, June 2009, and August 2010 cruises, respectively) in the Northwest Atlantic Ocean, to facilitate the analysis of net oceanic Hg⁰ flux (Figure 1). Concentrations of air Hg⁰, aqueous Hg⁰ (Hg_aq⁰), and total aqueous Hg (Hg_aq^T) along the ship cruise routes are drawn from Soerensen et al. (2013), which also described in detail the methods of atmospheric measurement and seawater collection and analysis. In addition, we compare the model outputs with the wet deposition flux of Hg^II in the eastern United States measured by the NADP Mercury Deposition Network (MDN) (Prestbo and Gay, 2009).

The GEOS-Chem global CTM (version 9-02; http://www.geos-chem.org) of atmospheric composition is driven by assimilated meteorological fields from the NASA Global Modeling and Assimilation Office Goddard Earth Observing System. The GEOS-5 Forward Processing (FP) and GEOS-5.2.0 data are used for the simulation year of 2013 and the spin-up period of 2010–2012, respectively, due to their different time periods covered (http://gmao.gsfc.nasa.gov/products/). We run the CTM in a one-way nested grid formulation with the native GEOS-5 FP horizontal resolution of 0.25° × 0.3125° over North America (130–60° W, 10–60° N). The global simulation with a coarser resolution of 2° × 2.5° (degraded from the native grid) generates initial and boundary conditions for the nested grid simulation. Both simulations have 47 vertical layers in the atmosphere. The GEOS-Chem Hg nested grid simulation was developed by Zhang et al. (2012) based on GEOS-5.2.0 (with a native 0.5° × 0.667° resolution in North America), and is updated to the most recent GEOS-5 FP fields. The GEOS-Chem Hg model dynamically couples a 3D atmosphere (Selin et al., 2007), a 2D terrestrial reservoir (Selin et al., 2008), and a 2D mixed layer slab ocean (Soerensen et al., 2010). Speciated inorganic Hg tracers are tracked in both the atmosphere and the ocean mixed layer. Following Holmes et al. (2010), we include atomic bromine as the predominant oxidant of Hg⁰ in the atmosphere and use a two-step oxidation mechanism (Goodsite et al., 2004, 2012). A full chemistry GEOS-Chem simulation generates tropospheric bromine fields used here (Parrella et al., 2012). Some previous Hg model simulations (e.g., Pongprueksa et al., 2011) have hypothesized photo-reduction of Hg^II in gas and/or aqueous phases and adjusted the reduction rates to match global mean surface Hg⁰ measurements, but in our simulations, we do not include this process. Model results are sampled along flight and ship cruise routes. We use a non-local planetary boundary layer (PBL) mixing scheme developed by Holtslag and Boville (1993) and implemented in GEOS-Chem by Lin and McElroy (2010).

The surface fluxes of Hg⁰ at the Earth’s surface include anthropogenic sources, biomass burning, geogenic activities, as well as the bidirectional fluxes involved in the atmosphere–terrestrial and atmosphere–ocean exchange. Note that anthropogenic sources also emit a small fraction of Hg^II. In our model, the sink of atmospheric Hg⁰ is the oxidation by bromine radicals. For all the simulations, biomass burning emissions (382 Mg yr^-1) are estimated using a global CO emission database (van der Werf et al., 2010) and a volume mixing ratio of Hg/CO of 2 × 10^-7 (Slemr et al., 2014). Geogenic activities (90 Mg yr^-1) are spatially distributed based on the locations of mercury mines.

GEOS-Chem estimates the bidirectional exchange of Hg⁰ between the atmosphere and the terrestrial and oceanic surfaces in different ways. For atmosphere–terrestrial exchange, GEOS-Chem treats the evasion and dry deposition of Hg⁰ separately (Selin et al., 2008). Dry deposition is parameterized with a resistance-in-series scheme (Wesely, 1989). Hg⁰ evasion includes volatilization from soil and rapid recycling of newly deposited Hg. The former is estimated as a function of soil Hg content and solar radiation. The latter is modeled by recycling a fraction of wet/dry deposited Hg^II to the atmosphere as Hg⁰ immediately after deposition (60% for snow covered land and 20% for all other land uses) (Selin et al., 2008). The net terrestrial flux is calculated as the sum of the three individual processes. GEOS-Chem estimates the atmosphere–ocean exchange of Hg⁰ using a standard two-layer diffusion model (Liss and Slater, 1974):

where F is the net emission flux (ng m^-2 h^-1), K is the mass transfer coefficient estimated by wind speed and temperature-corrected Schmidt numbers for CO₂ and Hg⁰ (Nightingale et al., 2000), and H is a temperature dependent Henry’s law constant (Andersson et al., 2008). Hg_aq⁰ and Hg_air⁰ are the modeled concentrations of elemental Hg in the mixed layer waters and in the lowest layer of the atmosphere, respectively. In our model, the ocean mercury in the mixed layer interacts not only with the atmospheric boundary layer but also with the subsurface waters through entrainment/detrainment of the mixed layer and wind-driven Ekman pumping (Soerensen et al., 2010).

In this study, multiple global GEOS-Chem model simulations with nested grid over North America are performed to test different representations and hypotheses of atmosphere–surface exchange of Hg⁰, including a reference simulation following Song et al. (2015) referred to here as “REF”, a simulation “INV” that is based on global observational constraints from inverse modeling (Song et al., 2015), and four sensitivity simulations (HSO, HOX, HOXSO, and HOCEAN). The global mercury budgets of several model simulations are shown in Table 1.

In REF, global anthropogenic mercury emissions are taken from the United Nations Environment Programme (UNEP) for 2010 (AMAP/UNEP, 2013). Over North America, UNEP 2010 emissions are replaced by U.S. EPA’s National Emissions Inventory (NEI) and Environment Canada’s National Pollutant Release Inventory (NPRI), both based on activity data for 2011 (U.S. EPA, 2013; Environment Canada, 2013). These inventories result in global emissions of 1598 Mg yr^-1 Hg⁰ and 340 Mg yr^-1 Hg^II. Net terrestrial emission fluxes of Hg⁰ can be calculated as the sum of soil volatilization, rapid recycling, and dry deposition of Hg⁰ over land. In REF, the net terrestrial and oceanic fluxes of Hg⁰ are 960 and 2960 Mg yr^-1, respectively, at a global scale (see Table 1).

Song et al. (2015) optimized four model parameters and emission fluxes, including soil volatilization, Asian anthropogenic emission, and two ocean physicochemical parameters (the rate constant of dark oxidation of Hg_aq⁰ and the partition coefficient of Hg_aq^II on particles), using a Bayesian inversion approach and global ground-based atmospheric mercury observations. These changes are applied in the sensitivity simulation of INV. As described in detail in Song et al. (2015), the anthropogenic Asian emission of Hg⁰ in the UNEP 2010 inventory is increased by 90%, and the soil volatilization in the REF is reduced by 76% (see Figure 2). Changing the two ocean parameters affects the modeled global ocean mercury budget, particularly the mass exchange between the mixed layer and subsurface waters, and also the magnitude and seasonality of Hg⁰ evasion from the ocean surfaces. On a global scale, the net terrestrial and oceanic Hg⁰ surface fluxes modeled by INV are -280 and 3260 Mg yr^-1, respectively (see Table 1).

Several sensitivity GEOS-Chem model simulations (HOX, HSO, and HOXSO; HOX standards for “high oxidation” and HSO standards for “high soil volatilization”) are conducted by making additional changes to INV. The enhanced concentration levels of Hg^II and BrO were measured in the free troposphere during the NOMADSS aircraft campaign, supporting the role of bromine as the dominant Hg⁰ oxidant (Gratz et al., 2015). Using the default tropospheric bromine fields and the oxidation mechanism (as described above), modeled Hg^II levels are about a factor of 3 too low (Shah et al., 2016). Therefore, in HOX, we increase the rates of Hg⁰ oxidation to Hg^II by tripling the summertime bromine radical (BrO_x = Br + BrO) concentrations in the tropical and subtropical free troposphere (45° S–45° N; from 750 hPa to the tropopause), in order to match the observed high levels of Hg^II during NOMADSS. Shah et al. (2016) showed that using faster oxidation kinetics (e.g., Ariya et al., 2002) could also simulate the observed high Hg^II levels during NOMADSS. Recent studies have indicated that the bromine fields and kinetics used in current atmospheric models are both very uncertain (e.g., Wang et al., 2015; Ariya et al., 2015; Schmidt et al., 2015, submitted), but a detailed discussion on atmospheric oxidation and reduction is beyond the scope of this study. We hypothesize that the mid-latitude terrestrial flux of Hg⁰ in spring may be enhanced based on the evidence from both atmospheric mercury modeling and surface flux measurements (see details in the Discussion section). Therefore, in HSO, the springtime soil volatilization in the mid-latitudes (20°–60°) is increased by a factor of 4 in order to evaluate this hypothesis against atmospheric observations. We will show below that this increase improves the model’s ability to reproduce the seasonal pattern of observed Hg⁰ concentration in the NADP AMNet network. In “HOXSO”, we combine HOX and HSO by increasing both summertime oxidation and springtime soil volatilization. As shown in Table 1, these changes influence the modeled net terrestrial Hg⁰ emission, Hg⁰ oxidation, and Hg^II wet deposition. The sensitivity simulation “HOCEAN” is the same as HOXSO, except that net oceanic Hg⁰ flux from the Northwest Atlantic during summer is increased by 80% (derived from the comparison between model results of HOXSO and ship cruise observations, see details in the Results section).

In the rest of the paper, these model simulation results are compared to different types of observations in eastern North America.

During 19 research flights, 1589 2.5-minute observations of Hg⁰ were sampled by the DOHGS. We select those observations within the planetary boundary layer (PBL), where rapid atmosphere–surface exchanges of chemical species take place. PBL heights exhibit significant spatiotemporal variations, and are typically diagnosed from vertical profiles of meteorological parameters measured by radiosonde soundings (Liu and Liang, 2010). However, the PBL height for each NOMADSS data point cannot be determined in such a way since sounding measurements are limited in time and space (Seidel et al., 2010). Here, we select observations below the lower of 1.2 km (a fixed PBL height estimated from the vertical profile of potential temperature observed in all research flights, see Figure S1) and the modeled PBL height in the GEOS-5 FP meteorological fields. It should be noted that Kim et al. (2015) have recently shown, over the summertime southeastern United States, that the daytime GEOS-5 FP PBL heights may have a 30–50% positive bias against LIDAR and ceilometer measurements. From the sensitivity model simulations that decrease the daytime GEOS-5 FP PBL heights by 40%, we find that such potential bias does not significantly affect our simulation results and the conclusion of the paper remains unchanged. About 30% (466 data points out of 1589) of the NOMADSS Hg⁰ observations are identified within the PBL. The within-PBL Hg⁰ observations have a significantly larger (p < 0.001) median concentration of 1.47 ± 0.10 ng m^-3 than free tropospheric Hg⁰ observations (1.37 ± 0.12 ng m^-3), with significance determined by the non-parametric Mann-Whitney U test (Rosner and Grove, 1999). This vertical gradient indicates Hg⁰ oxidation in the free troposphere (Shah et al., 2016). The air within the PBL is characterized by a high water vapor mixing ratio (WVMR) of 12.7 ± 1.4 g kg^-1 whereas the free troposphere is much drier (WVMR ∼ 2.2 ± 2.7 g kg^-1). We further remove about 3% (13 data points out of 466) of the within-PBL observations with the highest 1% of concentrations of SO₂, NO_x, or propane to avoid the influence of nearby anthropogenic sources (Ambrose et al., 2015; Miller et al., 2013). Measurements made over Lake Michigan in RF-15 are also screened out.

The selected background within-PBL NOMADSS observations are then divided into two groups: those over terrestrial and over oceanic surfaces. The air over oceanic surfaces was sampled in two flights out to the Northwest Atlantic (RF-14 and RF-16) and one flight out to the Gulf of Mexico (RF-12) (Figure 1). Measurements over terrestrial surfaces were made in 14 out of the 19 flights. The extremely low mixing ratios of isoprene (0.00 ± 0.01 ppbv) and high levels of DMS (8.26 ± 4.37 pptv) observed over oceanic surfaces, compared to those observed over terrestrial surfaces (isoprene of 1.38 ± 0.49 ppbv and DMS of 1.95 ± 1.63 pptv), suggest that the two groups represent different types of air masses (Table 2). The median Hg⁰ concentration observed over oceanic surfaces (1.55 ± 0.11 ng m^-3, N = 73) is significantly higher than that over terrestrial surfaces (1.45 ± 0.09 ng m^-3, N = 360) (p < 0.001, Mann-Whitney U test).

Below, we compare model simulation results with Hg⁰ concentration observations from 11 ground-based sites, three ship cruises, and the NOMADSS campaign, as well as the wet deposition data of Hg^II.

Figure 3 shows the average Hg⁰ concentration measured at the 11 NADP AMNet ground-based sites. The average observation in 2013 is similar to that in 2009–2012, both of which indicate pronounced seasonal variations with a maximum of Hg⁰ in late winter and early spring (about 1.45 ng m^-3 in February–April) and a minimum in late summer and early fall (about 1.20 ng m^-3 in August–October). We use the Normalized Root Mean Square Error (NRMSE) to evaluate the performance of model simulations:

where Hg⁰_obs,i and Hg⁰_mod,i are the observed and modeled Hg⁰ concentrations in the ith month, respectively. A smaller NRMSE value indicates a better model performance. Compared to the average NADP AMNet observation in 2013, the NRMSE values for REF, INV, HOX, HSO, and HOXSO are 0.14, 0.09, 0.09, 0.10, and 0.05, respectively, and thus HOXSO has the best performance among them. As shown in Figure 3, HOXSO is also the only simulation that reproduces the observed seasonal pattern of Hg⁰ concentration. The two changes applied in HOXSO, the faster summertime Hg⁰ oxidation (also made in HOX) and the elevated springtime mid-latitude soil volatilization (also made in HSO), are both essential for modeling this seasonal pattern. The tripled summertime BrO_x accelerates the conversion of Hg⁰ to Hg^II through oxidation, and thus the modeled Hg⁰ concentration in HOXSO shows a similar decreasing gradient (about -0.06 ng m^-3 month^-1) during the summer months with the observation in 2013 (about -0.08 ng m^-3 month^-1) and in 2009–2012 (about -0.04 ng m^-3 month^-1). However, the model simulations without enhanced oxidation (REF, INV, and HSO) cannot reproduce this feature in the observed Hg⁰ concentration. In contrast, the simulations without elevated soil volatilization (REF, INV, and HOX) show a large decline of about 0.23 ng m^-3 in modeled Hg⁰ during the spring months, which is not seen in the ground-based observations. By increasing springtime soil volatilization, HOXSO better reproduces the observed Hg⁰ trend in this season. Since the two sensitivity simulations, HOX and HSO, cannot reproduce the observed seasonal pattern of ground-based Hg⁰, hereafter we do not compare their results with ship cruise and aircraft measurements. Figure S2 shows the comparison of Hg^II wet deposition fluxes between the MDN measurements and different model simulations (REF, INV, and HOXSO) during 2013 summer over eastern North America. REF and INV have similar insignificant positive biases of +2% and +1%, respectively. This is because only Hg⁰ surface fluxes differ between these two simulations. HOXSO shows a large positive bias of +59% due to the enhanced bromine radical fields used in this sensitivity simulation. It is noted that this bias should become smaller if an in-plume Hg^II reduction is applied in GEOS-Chem (e.g., Zhang et al., 2012).

Table 3 compares the observed and modeled concentrations of Hg in both the atmospheric boundary layer and the ocean mixed layer for three summertime ship cruises in the Northwest Atlantic Ocean (Soerensen et al., 2013). These simulations (REF, INV, and HOXSO) produce air Hg⁰ concentrations (1.2–1.4 ng m^-3) within the uncertainty ranges of the observations. Among these simulations, REF leads to the lowest mixed layer concentrations of both Hg_aq⁰ and Hg_aq^T, whereas HOXSO predicts the highest. Compared with measurements, REF and INV underestimate mixed layer Hg_aq⁰ concentrations by 17–54% and 12–51%, respectively. HOXSO models comparable Hg_aq⁰ concentrations for one ship cruise (AUG08), but still underestimates its levels for the other two (up to 40%). On average, HOXSO underestimates the observed mixed layer Hg_aq⁰ by about 20%. As shown in Figure 4 (b–d), the modeled net oceanic Hg⁰ emission fluxes also follow the order of REF < INV < HOXSO. Considering that HOXSO may underestimate mixed layer Hg_aq⁰ concentrations by 40% and that different gas exchange parameterizations may lead to a 30% variability in estimated oceanic Hg⁰ fluxes (Sunderland et al., 2010; Andersson et al., 2007), we conduct an additional sensitivity simulation referred to as HOCEAN, in which the net Hg⁰ fluxes from the Northwest Atlantic (100–60° W, 15–45° N) during the summer months (June–July–August) are increased by 80% above the oceanic Hg⁰ fluxes in HOXSO. HOCEAN thus represents a potential upper limit of oceanic emissions from the Northwest Atlantic determined from the above model–observation comparison of ship cruises.

Table 2 shows the background within-PBL Hg⁰ observations during the NOMADSS aircraft campaign, divided into two groups (over terrestrial and over oceanic surfaces), and the corresponding model results from REF, INV, HOXSO, and HOCEAN. All these model simulations predict lower Hg⁰ concentrations than the DOHGS observed. The modeled median concentrations of Hg⁰ over terrestrial surfaces range from 1.29–1.39 ng m^-3, about 4–11% lower than its observed median of 1.45 ng m^-3. This difference generally falls within the overall measurement uncertainty range of 6–10% of the DOHGS. In contrast, by comparing the observed and modeled Hg⁰ concentrations (1.33 and 1.30–1.42 ng m^-3, respectively) during the NOMADSS period (2013 June–July) for the 11 AMNet ground-based sites in the eastern United States, we do not find a consistent negative bias in our model simulations (Figure 3). As described earlier, the DOHGS observed a significantly higher Hg⁰ concentration of 0.10 ng m^-3 over oceanic surfaces in the Northwest Atlantic than over terrestrial surfaces in the eastern United States. However, our model simulations cannot reproduce this land–ocean difference. As shown in Table 2, REF, INV, and HOXSO indicate that the median Hg⁰ concentrations over oceanic surfaces are 0.07 (p < 0.001), 0.02 (p = 0.45), and 0.03 (p < 0.001) ng m^-3 lower than those over terrestrial surfaces. HOCEAN increases the summertime net Hg⁰ flux from the Northwest Atlantic by 80% than HOXSO, and predicts comparable (p = 0.67) Hg⁰ concentrations over terrestrial surfaces (1.29 ± 0.03 ng m^-3) and over oceanic surfaces (1.29 ± 0.05 ng m^-3). Among our sensitivity GEOS-Chem model simulations, HOCEAN can be considered as the best simulation with regard to the comparison with Hg⁰ observations.

We have described that HOXSO can reproduce the seasonal variation of Hg⁰ concentration observed at the NADP AMNet ground-based sites with an enhanced springtime soil volatilization in the mid-latitude region (20°–60°). As shown in Figure 2, the net terrestrial fluxes of Hg⁰ (the sum of soil volatilization, rapid recycling, and dry deposition) modeled by HOXSO are positive in the spring months. We hypothesize that the mid-latitude terrestrial flux of Hg⁰ in spring may be enhanced based on the following evidence from both atmospheric mercury modeling and surface flux measurements. Song et al. (2015) quantitatively constrained monthly soil volatilization using worldwide ground-based observations and a Bayesian inversion approach. As shown in Figure 2, the 1σ uncertainty range of soil volatilization from the emission inversion revealed an enhancement during the spring months (Song et al., 2015), and increasing the soil volatilization by a factor of 4 agrees well with this modeling result. The measured net Hg⁰ fluxes in the mid-latitudes for different land use types (i.e., forest, grassland, agriculture, and bare soil) and different seasons are summarized in Table S1. We find that most of them (7 out of 9) show higher terrestrial Hg⁰ fluxes in spring (by 40% to a factor of 6) than the averages in other seasons. Bash and Miller (2007) suggested that springtime agricultural tilling operations can significantly mobilize Hg into the atmosphere from its soil pool. The small vegetation coverage in spring, which allows the penetration of solar radiation to the soil surface, may also be another important factor for the measured high Hg⁰ emission fluxes (Choi and Holsen, 2009). It is important to note, however, that only limited terrestrial flux measurements are available and very large uncertainties exist in them. Net deposition fluxes of Hg⁰ during springtime have also been suggested (Mao et al., 2008; Converse et al., 2010).

The NOMADSS aircraft campaign found higher Hg⁰ concentrations over the Northwest Atlantic than over the eastern United States (Table 2). Gay et al. (2013) reported a similar finding by combining the NADP AMNet sites into several loosely defined groups, including a coastal/remote group (NS01 and NH06) and a continental/remote group (VT99, NY20, and GA40) (Figure 1). Three years of data show that the former group has about 0.07 ng m^-3 higher Hg⁰ concentrations than the latter. However, the Hg⁰ measurement systematic uncertainty for these NADP AMNet sites, which is estimated to be about 0.14 ng m^-3 (10% of the observed Hg⁰ concentration of 1.3–1.4 ng m^-3; Figure 3), is larger than this 0.07 ng m^-3 difference between the two site groups, limiting our ability to apply these results in a modeling context. NOMADSS minimizes potential systematic differences since the DOHGS was used to measure Hg⁰ over both terrestrial and oceanic surfaces.

This land–ocean difference of atmospheric Hg⁰ concentrations over eastern North America has important implications for regional terrestrial and oceanic Hg⁰ fluxes. The modeled net surface fluxes in 100–60° W and 20–50° N during summer are shown in Figure 4 (b–d). Note that all model simulations also include the same three unidirectional sources that occur over land (i.e., anthropogenic, geogenic, and biomass burning), which together emit 3.1 Mg month^-1 Hg⁰ from terrestrial surfaces (corresponding to an average Hg⁰ flux of 0.8 ng m^-2 h^-1), as shown in Figure 4a. REF emits an average net oceanic flux of 1.3 ng m^-2 h^-1 from the Northwest Atlantic and an average net terrestrial flux of 0.8 ng m^-2 h^-1 from the eastern United States. (Figure 4b). Thus, in REF, the total Hg⁰ flux emitted over land (1.6 ng m^-2 h^-1) is larger than that from oceanic surfaces (1.3 ng m^-2 h^-1). As described earlier, in REF, the median Hg⁰ over land is 0.07 ng m^-3 higher (p < 0.001) than that over ocean, which is opposite to the observed land–ocean difference in NOMADSS. Among our model simulations, HOCEAN has the lowest terrestrial Hg⁰ flux of -0.6 ng m^-2 h^-1 (total Hg⁰ flux emitted over land is thus 0.2 ng m^-2 h^-1) and the highest oceanic Hg⁰ flux of 3.3 ng m^-2 h^-1, and predicts the same Hg⁰ concentrations (p = 0.67) over both surfaces in NOMADSS. Consequently, the above comparison between our model results and NOMADSS observations implies either even higher oceanic Hg⁰ fluxes from the Northwest Atlantic or lower Hg⁰ fluxes emitted from land in the eastern United States, or both, than the fluxes simulated by HOCEAN.

We have demonstrated through the model–observation comparison of ship cruises that HOCEAN represents a potential upper limit of oceanic emissions from the Northwest Atlantic. However, riverine discharges of Hg, an oceanic source that is not considered in the 2D mixed layer slab ocean of GEOS-Chem, may lead to additional Hg⁰ emissions from coastal/shelf areas and help to reconcile the difference between model results and NOMADSS observations. Soerensen et al. (2013) found, during the ship cruise AUG08 in the Northwest Atlantic (Figure 1), that enhanced freshwater inputs due to anomalously high precipitation in July 2008 strongly increased the mixed layer Hg_aq⁰ concentrations and Hg⁰ fluxes in waters of the Gulf of Maine approximately 60 km offshore (270 ± 50 fM and 7.2 ± 4.2 ng m^-2 h^-1), when compared to those in more open waters (130 ± 14 fM and 3.8 ± 2.9 ng m^-2 h^-1). A spatial trend of higher Hg_aq⁰ levels in coastal waters (∼ 100 km) than in open waters has also been observed in other cruises (e.g., Ci et al., 2011b). Two NOMADSS research flights, RF-14 and RF-16, sampled the oceanic air 50–150 km off the coast of South Carolina on July 5 and July 8 of 2013, respectively (Figure 1). Unusually high water discharges were measured in July 2013 for rivers in South Carolina (see Figure S3a for an example). Additionally, measurements within the NADP Mercury Deposition Network (MDN; Prestbo and Gay, 2009) showed heavy rainfall and high wet deposition fluxes of Hg in late June and early July 2013, right before the sampling time of these two flights (see Figure S3b for an example). Overall, these results suggest that riverine discharges of mercury may contribute to the observed high Hg⁰ concentration over the Northwest Atlantic.

HOCEAN simulates an average net terrestrial flux of -0.6 ng m^-2 h^-1, meaning that the simulated terrestrial ecosystem in the eastern United States is a net sink of Hg⁰ during summer. The simulated dry deposition of Hg⁰ of 2.0 ng m^-2 h^-1 is only partially offset by its evasion that includes soil volatilization (0.6 ng m^-2 h^-1) and rapid recycling of newly deposited Hg (0.8 ng m^-2 h^-1). The global terrestrial flux measurement database (Agnan et al., 2016) suggests a small positive median Hg⁰ flux of 0.1 ng m^-2 h^-1 (50% uncertainty range: -0.1 to 0.5 ng m^-2 h^-1) over background bare soil, which is in agreement with measured vertical profiles of Hg⁰ in soil pores (Obrist et al., 2014). However, forest is the most important land use type in the eastern United States, especially during summer when the leaf area index is maximum (thus leaf surface areas are several times larger than underlying soil surface areas) (Drummond and Loveland, 2010; Buermann et al., 2001). Forest canopies can reduce Hg⁰ evasion from underlying soils by absorbing most of the incoming solar radiation and suppressing the rising of soil temperature (Wang et al., 2006; Choi and Holsen, 2009). Foliage is well-known to constitute a net sink of atmospheric Hg⁰ through stomatal and non-stomatal uptake (Wang et al., 2014; Stamenkovic and Gustin, 2009), and the median Hg⁰ flux measured over leaves at background forest sites is -0.12 ng m^-2 h^-1 with a 50% uncertainty range from -1.48 to 1.65 ng m^-2 h^-1 (Agnan et al., 2016). Hg translocation from soil to leaves is unlikely to be significant (Fay and Gustin, 2007; Cui et al., 2014). Therefore, it is not unreasonable to consider the summertime terrestrial ecosystem in the eastern United States as a net sink of Hg⁰, given relatively low soil evasion and high foliage uptake. However, it is not possible to provide an accurate estimate for the magnitude of this sink, since reliable flux measurements over forests are currently lacking, and the mechanism of Hg transport in plant/vegetation is not well understood (Agnan et al., 2016).

The unidirectional sources over land (i.e., anthropogenic, geogenic, and biomass burning) are also uncertain and their fluxes may be higher or lower than the values applied in our model. However, their uncertainties are small compared to those associated with terrestrial and oceanic fluxes in the studied region. In summer, simulated geogenic activities and biomass burning emit little atmospheric Hg⁰ from the eastern United States (< 0.1 ng m^-2 h^-1 on average), and thus the contributions of their uncertainties on Hg⁰ fluxes are insignificant in this context. The NEI 2011 inventory used in our model has an average Hg⁰ flux of 0.7 ng m^-2 h^-1 in the eastern United States. In general, anthropogenic emissions in North America are considered to be relatively well constrained (1σ error around ± 30%) (e.g., Pacyna et al., 2010; AMAP/UNEP, 2013). Although U.S. EPA will not release the NEI inventory in 2013 (the year in which NOMADSS took place), another estimate, the Toxics Release Inventory (TRI; U.S. EPA, 2015), shows that the magnitudes of estimated mercury emissions for 2011 and 2013 differ only slightly (< 2%).

Based on the comparison between model results and aircraft and ship cruise observations, our analysis suggests a high Hg⁰ flux from the Northwest Atlantic. In our simulations, this high Hg⁰ flux helps to explain the land–ocean differences of observed atmospheric Hg⁰ concentrations in the eastern North America, and also agrees with the observed high aqueous Hg⁰ levels in the mixed layer. In contrast, a recent study by Weigelt et al. (2015) classified Hg⁰ data observed at the Mace Head station on the Northeastern Atlantic coast of Ireland into different air mass groups according to their geographical origins. Air masses originating mostly from the Northeast Atlantic were found to have generally lower Hg⁰ concentrations (0.07 ± 0.04 ng m^-3 calculated from monthly means and 0.04 ± 0.05 ng m^-3 from monthly medians, both using data in 2010–2013) than those from continental Europe. As shown in Figure S4, our model simulations of the Northwest Atlantic, particularly the waters near the continental United States, have overall higher Hg⁰ fluxes when compared to the Northeast Atlantic. This is qualitatively consistent with the different land–ocean patterns of observed atmospheric Hg⁰ over the Northeast and Northwest Atlantic.

The model also enables us to identify physicochemical processes that lead to simulated high Hg⁰ fluxes in the Northwest Atlantic (and the relatively low fluxes in the Northeast Atlantic). The modeled Hg⁰ fluxes are positively correlated with the mixed layer Hg_aq⁰ concentrations (Equation 1), which are in turn determined by multiple processes in the mixed layer, including photochemical- and biological- redox reactions and adsorption/desorption on particles, and the vertical interactions of the mixed layer with the above atmosphere and subsurface waters. The wet/dry deposition of Hg^II from the atmosphere is a source of mercury in the mixed layer, whereas Hg⁰ evasion into the atmosphere is a sink. Vertical exchanges between the mixed layer and subsurface waters include entrainment/detrainment, wind-driven Ekman pumping, and particle sinking (biological carbon pump) (Soerensen et al., 2010; Batrakova et al., 2014; Song et al., 2015). Figure 5a shows the spatial distribution of net Hg⁰ fluxes (modeled by HOXSO) from the North Atlantic Ocean (100° W–20° E, 20–60° N) in summer. The spatial distributions of modeled Hg fluxes associated with several above-mentioned processes are shown in Figure 5(b–e). A comparison of the magnitude of these Hg fluxes indicates that the high wet deposition of Hg^II into the Northwest Atlantic is the dominant process determining the simulated high net oceanic flux of Hg⁰ from this region. The wet deposition of Hg^II is closely related to the precipitation rate, which also has a spatial pattern with generally higher values in the Northwest Atlantic, in particular the waters near the continental United States, and lower values in the Northeast Atlantic (see Figure 5f). The summertime precipitation in the North Atlantic is influenced by the North Atlantic Subtropical High (NASH, also known as the Bermuda High), a semi-permanent high pressure system in the lower troposphere (Li et al., 2012). As shown in Figure 5f, strong precipitation is located along the western boundary of the NASH, which can be represented by the 1560 m geopotential contour line at 850 hPa (Li et al., 2011), while the precipitation rate in eastern side of the NASH is very small. Overall, the high simulated Hg⁰ flux from the Northwest Atlantic is mainly a result of high wet deposition of Hg^II, which is in turn linked to high precipitation rates in this region during summer. Given the uncertainties in modeling oceanic mercury and the limited representation of these processes in our slab ocean model, however, our ability to draw process-based conclusions from this study is limited. Similarly, a more detailed 3D oceanic mercury model (Zhang et al., 2014) suggests that high wet deposition of Hg^II leads to high Hg⁰ flux from the Northwest Atlantic (Zhang Y, personal communication).

Atmosphere–surface exchange of Hg⁰ in eastern North America is constrained by combining aircraft-based observations made during the 2013 summer NOMADSS campaign (as well as ground- and ship-based measurements) and GEOS-Chem CTM simulations. As a consistent instrumentation (the DOHGS) measured Hg throughout NOMADSS, the systematic uncertainty of Hg⁰ concentration measurements at different locations is minimized. Within the PBL, significantly higher median Hg⁰ concentrations were observed over oceanic surfaces of the Northwest Atlantic than over terrestrial surfaces of the eastern United States during NOMADSS (p < 0.001). The model simulation (HOCEAN) with a low (negative) terrestrial Hg⁰ flux and a high (positive) oceanic flux in this region obtains the same Hg⁰ concentrations (p = 0.67) over both surfaces. Riverine discharges of mercury, an oceanic source that is not included in GEOS-Chem but may be significant in the NOMADSS period, may help to reconcile the model–observation discrepancy. By analyzing processes in the 2D mixed layer slab ocean of GEOS-Chem, we show that inferred high Hg⁰ emission fluxes from the Northwest Atlantic may be a result of high wet deposition fluxes of oxidized mercury, which are in turn linked to high precipitation rates in this region. Given relatively low soil evasion and high foliage uptake, it is likely that terrestrial ecosystem in the summer eastern United States acts as a net sink of Hg⁰. Increasing simulated terrestrial fluxes of Hg⁰ in spring compared to other seasons can better reproduce seasonal variability of observed Hg⁰ concentration at ground-based sites in eastern North America.

The NOMADSS observational data are publicly available from the SAS project on the NCAR/UCAR Earth Observing Laboratory’s website (http://data.eol.ucar.edu/master_list/?project=SAS). The NADP AMNet observational data are available on the NADP’s website (http://nadp.sws.uiuc.edu/data/AMNet/) but access to these data is password protected (please contact the program coordinator). Ship cruise data are published in Soerensen et al. (2013). GEOS-Chem is an open source chemical transport model for atmospheric composition (http://www.geos-chem.org). Model simulation results are available upon request.

© 2016 Song et al. This is an open-access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited.

We thank the NCAR’s Earth Observing Laboratory and its Research Aviation Facility for their support throughout NOMADSS. We acknowledge Wisconsin Department of Natural Resources, Oklahoma Cherokee Nation Air Quality Program, and all the other site operators and principal investigators within the NADP AMNet for providing observational data used in this study. We also thank two reviewers for constructive comments.