Gaseous dry deposition of atmospheric mercury: A comparison of two surface resistance models for deposition to semiarid vegetation



[1] In the United States, atmospheric mercury (Hg) deposition, from regional and international sources, is the largest contributor to increased Hg concentrations in bodies of water leading to bioaccumulation of methyl mercury in fish. In this work, modeled dry deposition velocities (vd) for gaseous Hg are calculated using two surface resistance parameterizations found in the literature. The flux is then estimated as the product of the species concentration and modeled vd. The calculations utilize speciated atmospheric mercury concentrations measured during an annual monitoring campaign in southern Idaho. Gaseous elemental mercury (GEM) and reactive gaseous mercury (RGM) were monitored with Tekran models 2537A and 1130, respectively. Two anemometers collected meteorological data, including one fast-response three-dimensional sonic anemometer to measure turbulence parameters. For the flux calculation, three resistances are required to model the mechanisms that transport gaseous Hg from the atmosphere to the surface, with the surface resistance being the largest source of error. Results from two surface resistance models are presented. In particular, the downward flux is sensitive to the choice of model and input parameters such as seasonal category and mesophyll resistance. A comparison of annual GEM and RGM fluxes calculated using the two models shows good agreement for RGM (3.2% difference for annual deposition); however, for the low-solubility species of GEM, the models show a 64% difference in annual fluxes, with a range of 32% to 200% in seasonal fluxes. Results indicate the importance of understanding the diurnal variation of the physical processes modeled in the surface resistance parameterization for vd.

1. Introduction

[2] Stringent regulations have been implemented to prevent the direct release of toxic chemicals into watersheds; however, samples taken from remote lakes throughout the world with no direct mercury (Hg) input indicate high Hg concentrations in the fish [Hanisch, 1998]. This suggests that deposition (dry and wet) of atmospheric Hg is the largest contributor to the bioaccumulation of methylmercury in fish. In the western U.S., the state of Idaho currently has a fish consumption advisory due to high levels of mercury in the fish caught from the Salmon Falls Creek Reservoir (SFCR) in southern Idaho [Idaho Fish Consumption Advisory Program, 2009]. While there are no local sources in the area, regional sources of Hg emissions include ore smelting associated with mining activities in northern Nevada, ∼220 km southwest of SFCR [Abbott et al., 2008]. The mercury emissions can be deposited locally, but research shows that a fraction of the emitted mercury can be transported to other locations by the prevailing winds in the area [Bullock and Brehme, 2002].

[3] Mercury is emitted from natural (i.e., volcanic eruptions, crustal degassing) and anthropogenic processes (e.g., coal combustion, cement manufacturing, mining operations, waste incineration, chlorine production), and deposited Hg can be reemitted to the atmosphere and enhance the regional and global transport of mercury [Schroeder and Munthe, 1998]. Modeling the flux of atmospheric Hg species is difficult due to the physical, biological and chemical processes influencing the net Hg flux at the Earth's surface. Specifically, Hg fluxes are difficult to model and measure due to the complex chemical reactions in the atmosphere and the presence of multiple Hg species. For example the oxidation of gaseous elemental mercury (GEM) to reactive gaseous mercury (RGM) [Lin et al., 2006]. In addition to these complicated physical and chemical processes, measurement limitations exist. The most direct method for measuring gaseous fluxes is eddy correlation; however, due to slow instrument response it is not possible for Hg [Fritsche et al., 2008]. Therefore, the gradient method with either the modified Bowen ratio [Meyers et al., 1996] or flux gradient [Edwards et al., 2005] approach is common for measuring Hg fluxes [Lindberg et al., 2002; Lindberg and Meyers, 2001; Poissant et al., 2004]. While both emission and deposition of GEM is likely of importance at this location, the focus of this paper is to evaluate the dry deposition (downward flux component) modeling of gaseous Hg species from the atmosphere to semiarid vegetation and snow-covered ground. Measurement limitations during the experiment prevented calculation of the net flux, therefore the bidirectional atmospheric process that includes both emission and deposition is not investigated.

[4] Recent advancements for the calculation of Hg fluxes in air quality models have included a transfer velocity to model the bidirectional exchange, for example see Bash [2010] regarding this implementation in the Community Multiscale Air Quality Model (CMAQ). However, current air quality models use a unidirectional resistance parameterization scheme to obtain a deposition velocity that is used to calculate the Hg flux [Bash et al., 2007; Lin et al., 2007]. The primary difference between the bidirectional and unidirectional model formulations is that the species concentration can not be assumed to be zero in the leaf, soil or water at the surface for bidirectional exchange. Thus, bidirectional parameterizations are required that are based on the compensation rate for mass transfer of the species. Often times emission is the largest contributor to the net flux, yet parameterizations are necessary to estimate the deposition component of the physical and biological processes governing the transport of Hg from the atmosphere to the surface. Several studies from the literature cite a large sensitivity of the calculated deposition velocity on the modeled net flux of Hg [Lin et al., 2006, 2007; Bash et al., 2007]. Therefore, the comparison presented in this work is important to assess the influence of the parameterization schemes on the calculated deposition velocity. Until air quality models fully implement the bidirectional exchange process, the sensitivity of the modeled results to the parameterization for deposition velocity is critical.

[5] The flux density for a specific gaseous species can be estimated by multiplying a modeled deposition velocity (vd) by the species concentration at a known height [Seinfeld and Pandis, 2006]. For this work, resistance models from the literature were used to calculate vd for GEM and RGM during seasonal Intensive Observation Periods (IOPs) at SFCR, located in Southern Idaho. These models incorporate turbulence parameters, meteorological quantities, surface characteristics and chemical species information in three resistance components: aerodynamic, quasi-laminar and canopy resistance [Wesely, 1989; Walmsley and Wesely, 1996; Zhang et al., 2003a; Seinfeld and Pandis, 2006]. Typically, the largest source of error in these models comes from the surface resistance model, especially for a low-solubility species like GEM where surface interactions (canopy resistance) dominate the modeled vd [Lin et al., 2006]. Several studies have reported vd values for GEM and RGM based on resistance models or flux measurements (see Table 1). Typically, mean values are reported and these can be misleading due to the large differences in daytime and nighttime vd. This diurnal pattern is particularly apparent in the vd for GEM, where there are large differences between the maximum vd values that occur during the day and the nighttime minimums. Not only is there a large seasonal and diurnal dependency on vd, the method of obtaining vd greatly influences the result, this is evident in the large range of literature values reported in Table 1. A more complete review of Hg vd values from the literature is provided by Zhang et al. [2009], including a discussion on the measurement techniques and modeling methodology. One conclusion from their review is that measurement and calculation uncertainties are quite large, and a comparison of different modeling schemes for vd would be useful.

Table 1. Reported Deposition Velocity Values From Literature (cm s−1)
Land UseMethodReported ValuesGEMRGMReference
Regionalmodeled (United States)apeak summer0.093.44Lin et al. [2006]
Short grass/forbsmodeledbdaytime∼0.11∼1.8Zhang et al. [2009]
Semiarid naturalsurrogate surface6 day collection1.72, 1.52Lyman et al. [2007]
Tall grassgradient methods0.06–0.1literature [Lindberg et al., 2002]
Alfalfa plantschamber studies0.03–0.1literature [Lindberg et al., 2002]
Forestcalculatedpeak day, mean night∼0.2, 0.036Lindberg et al. [2002]
Wetlandflux measurementsmedian0.197.6Poissant et al. [2004]
Hardwood forestmodeledcmean0.0541.9Rea et al. [2001]
Evergladesmodeledbmean day, night0.42, 0.061.27, 0.47Marsik et al. [2007] (1999)
Evergladesmodeledbmean day, night0.46, 0.082.10, 0.67Marsik et al. [2007] (2000)
SFCRmodeledamean summer, winter0.043, 0.000360.40, 0.76Abbott et al. [2008]

[6] The current study is an evaluation of the unidirectional resistance parameterization used to model the gaseous dry deposition velocity for atmospheric Hg. Net flux estimates presented in this paper are given to quantitatively compare results from the two canopy resistance parameterizations being evaluated. These fluxes are not representative of the total deposition or the net flux of Hg, they are only indicative of gaseous dry deposition. Typically, the metric used to compare fluxes across models is the deposition velocity, given in Table 1. The work presented here demonstrates the importance of canopy resistance formulation on vd, and is essential because it is the methodology commonly used in air quality modeling (i.e., standard CMAQ formulation) [Bash, 2010; Lin et al., 2006, 2007].

[7] The objective of the 2008 annual monitoring campaign at SFCR was to determine Hg sources and to calculate the contribution of Hg to the watershed via dry deposition. The findings from the first objective are discussed in other publications [Abbott et al., 2008; Richards, 2010]. In this work, two surface resistance models are compared to calculate the annual and seasonal dry deposition of GEM and RGM. Both models use a similar resistance scheme, and the same aerodynamic and quasi-laminar resistance models to calculate vd, but the models for canopy resistance differ. Results indicate a large discrepancy in the calculated fluxes depending on which canopy resistance scheme is used and on the selection of input parameters for the model. To begin, a brief site description, experimental methods and descriptions of the gaseous dry deposition models are given. This is followed by the presentation of results in section 3 and a discussion in section 4 where the major differences between the two models are compared.

2. Methods

2.1. Experimental Setup

[8] SFCR is located in southern Idaho, 28 km west of the SFCR dam (42.1456°N, 115.014°W), on a private ranch at House Creek (elevation of 1734 m). The site is situated in a semiarid climate and is characterized by grass and mixed shrub vegetation less than 50 cm high. Average daytime temperatures are ∼30°C in summer and 0°C in winter, with an annual average relative humidity of 34–67%. A more detailed description of the SFCR study site, regional watershed motivation for mercury research and results from a previous atmospheric mercury monitoring campaign are given in Abbott et al. [2008]. Using the log wind profile during neutral boundary layer conditions and the sonic anemometer data, the aerodynamic roughness length (z0) was calculated for the site [Stull, 1988]. The purpose of this work is to evaluate differences in the modeled dry deposition of gaseous Hg, and compare our findings with previous results from the same location to show the discrepancies that exist with current modeling methodologies. Therefore, although z0 was calculated, a value of z0 = 50 mm was used for all non-snow-covered IOPs, to be consistent with the calculations presented by Abbott et al. [2008]. This value is within the range of mean neutral z0 values calculated for each IOP, 46–82 mm. During snow-covered IOPs z0 = 10 mm [Zhang et al., 2003a] was used. The leaf area index (LAI) was not measured at the site, therefore an estimate was made based on land use category (LUC) selection [Zhang et al., 2003a]. National Weather Service (NWS) observations at the Twin Falls airport 57.5 km northeast of SFCR were used for rain and cloud cover information needed to model canopy wetness.

[9] Time series concentrations of speciated atmospheric mercury were measured using Tekran Instruments Corporation (Toronto, ON, Canada) monitors from the Idaho National Laboratory (INL) [Poissant et al., 2005; Abbott et al., 2008]. Continuous GEM was measured using the Tekran 2537A Mercury Vapor Analyzer with a 5 min sampling time. The Tekran 1130 Mercury Speciation Unit (denuder module) made it possible to collect RGM data. The 1130 unit operated on a sampling schedule where ambient air was collected for a set period of time, then a measurement cycle was performed to obtain the RGM concentration for the time period. For this experiment ambient air was collected for 1 h, and a measurement cycle occurred every 2 h. Therefore, one 60 min average RGM was recorded per 2 h period. Particulate mercury data will not be presented in this publication, but for the majority of the IOPs, it was monitored using the Tekran 1135 unit, measuring Hg(p) less than 2.5 micrometers in diameter. All Tekran devices were operated in the INL temperature controlled trailer with the inlets mounted outside the trailer at a height of 4 m above the ground, the sampling lines were heated and 15 m long. A meteorological station, equipped with devices from Met One Instruments (Grants Pass, OR), was also mounted on the trailer at the same height to monitor wind direction, wind speed, relative humidity, air temperature, solar radiation and barometric pressure.

[10] Although deposition can be modeled with mean meteorological variables, several equations in the models require inputs that are calculated from turbulence data. As a result, additional models are needed to correlate the mean atmospheric data to turbulence parameters [Golder, 1972]. In this study, a micrometeorological tower was installed near the INL trailer. It was equipped with one Campbell Scientific Inc. (Logan, UT) CSAT-3 three-dimensional sonic anemometer (Sonic). The Sonic measured three components of wind speed and sonic temperature at a frequency of 10 Hz, 3.7 m above the ground. Three-dimensional wind speed measured with a fast-response anemometer is necessary to calculate the turbulence parameters that influence the transport of gases in the atmosphere. The Sonic data were filtered in postprocessing to exclude data when the wind passed over a stand of trees located directly behind the tower. Hence, winds from 320° to 100° were omitted. This filter typically removed less than 6% of the data, with the exception of the November IOP where 10% were excluded; see Figure 1.

Figure 1.

July IOP GEM hourly mean flux roses for (a) GEM and (b) RGM using W96 methodology and (c) GEM and (b) RGM using the Z03 methodology with Rm = 100 s m−1.

2.2. Gaseous Dry Deposition Model

[11] Assuming measurements are taken in the constant flux layer, the flux density (F, ng m−2 s−1) of Hg can be interpreted as the vertical turbulent flux of the species. This is shown in equation (1), where the overbar indicates a time average and w′ is the turbulent fluctuations of vertical wind speed [Meyers et al., 1996].

equation image

Using conservation of mass, the full scalar transport equation for GEM can be derived [see, e.g., Baldocchi, 2003]. The resulting equation can be simplified, as shown in equation (2), assuming the flow is steady (does not change with time) and horizontally homogeneous (no horizontal advection or flux divergence), the influence of vertical advection is small and the measurements are made in the constant flux layer of the atmospheric surface layer [Stull, 1988; Baldocchi et al., 1988].

equation image

The Sc term here represents the net flux from vegetation (sources and sinks) into the control volume at height z. Integrating equation (2) from the surface to the measurement height (zm) results in an expression for the mean flux density, shown in equation (3) [Baldocchi, 2003].

equation image

Here, the measured turbulent flux of C at height zm represents the difference between the mass of C added (per unit area per unit time) at the canopy floor (equation imagez=0) and that which is emitted or absorbed within the canopy (Sc) (per unit volume per unit time). The measurement of equation imagez=zm can be made using the eddy correlation technique, or other methods briefly described in section 1. However, limitations often exist with measurement techniques, thus requiring a model to approximate the flux as the product of a velocity (vd, m s−1) and species concentration (C, ng m−3) at a known height [Seinfeld and Pandis, 2006], as shown in equation (4).

equation image

Hence, vd is modeled with a resistance analogy scheme that includes three resistances, Aerodynamic Resistance (Ra), Quasi-laminar Resistance (Rb) and Canopy Resistance (Rc). The transport of the species to the surface is the sum of these resistances (equation (5)).

equation image

The treatment of this modeled velocity as a unidirectional process depends on the resistance analogy model, and the assumption that the surface concentration of the species is zero. If the species concentration at the surface is accounted for, the velocity can be modeled with a modified resistance scheme for a bidirectional velocity. This is commonly called an exchange or transfer velocity instead of deposition velocity to indicate that the process accounts for deposition and emission [Bash et al., 2007; Bash, 2010]. The models described in this paper use the same equations for Ra(s m−1) and Rb(s m−1); however, two formulations for Rc(s m−1) will be compared.

2.2.1. Aerodynamic Resistance

[12] Aerodynamic resistance (Ra) is calculated to account for the transport of gaseous species across the constant flux atmospheric surface layer (ASL). The model used for Ra, equation (6), is general and can be used for all species. Turbulent transport is the primary process governing Ra, therefore atmospheric stability and turbulence parameters are inputs to the model [Seinfeld and Pandis, 2006]. When turbulence data are not available, approximations using meteorological data can be made [Golder, 1972]. Aerodynamic resistance depends on friction velocity (u* = [(equation image2 + equation image2)1/4] (m s−1), where u′, v′ and w′ are the turbulent fluctuations of the streamwise, transverse and wall normal velocities [Stull, 1988]), surface roughness and atmospheric stability (ζ = z L−1, where z = measurement height (m) and L = Monin-Obukov Length [−Tou*3gequation image] (m), where To, Ts, g and κ are the mean surface temperature, turbulent fluctuations of virtual (Sonic) temperature, gravitational constant and the von Karman constant (0.4) [Seinfeld and Pandis, 2006; Stull, 1988]). The ζ parameter utilizes L to relate the turbulence generated from mechanical and buoyant mechanisms to determine the stability of the atmosphere. Note that the turbulent fluctuations of wind speed are utilized in the u* and L calculations, and L also includes the turbulent fluctuations of temperature.

equation image

In equation (6), η0 = (1 − 15ζ0)1/4, η = (1 − 15ζ)1/4, ζ0 = z0L−1 and z0 is the aerodynamic roughness height (m). Equation (6) is derived from gradient transport theory, momentum transfer similarity and Monin-Obukov similarity theory. First, from gradient transport theory the flux across the ASL can be expressed as the product between the concentration gradient and an eddy diffusivity. Applying Monin-Obukov similarity theory the eddy diffusivity can be estimated using a dimensionless temperature profile function. The resulting equation is integrated across the constant flux layer (from z0 to the measurement height z) to obtain Ra. The method for utilizing gradient transport theory and eddy diffusivity in the calculation of Ra, and the formulation of the dimensionless temperature profile function is provided by Seinfeld and Pandis [2006] and Stull [1988], respectively. According to Seinfeld and Pandis [2006] the stability criteria listed in equation (6) is valid from −1 to 1; however, the criteria was relaxed in this work to include data from −5 < ζ < 5. The validity of Ra within a range of atmospheric stability comes from the use of experimental data in the model development for flux profile relationships in the ASL, with the applicable stability range differing for each study [Businger et al., 1971; Webb, 1970].

2.2.2. Quasi-laminar Resistance

[13] The quasi-laminar resistance (Rb) model is species specific, depends on the molecular properties of the substance and the site surface characteristics, accounted for with friction velocity (equation (7)) [Seinfeld and Pandis, 2006]. Physically, Rb models the molecular transport of the gaseous species across the quasi-laminar sublayer, the thin layer of stagnant air (<1 cm) between surfaces (i.e., ground and vegetation) and the ASL.

equation image

The Schmidt number (Sc = ν D−1) is the ratio of the kinematic viscosity of air (cm2 s−1) to the molecular diffusivity of the species (cm2 s−1). As with Ra, the model for Rb requires turbulence data (Sonic) for the u* calculation.

2.2.3. Canopy Resistance

[14] The largest uncertainty in the present model for vd comes from canopy resistance (Rc), the modeled uptake of the gaseous species to the surface, and for GEM Rc is the dominant term in the vd model [Lin et al., 2006]. Rc is a complex term and includes parameterization for physical, biological and chemical processes that are effected by vegetation type, solubility of the chemical being deposited, structure of the canopy, ground characteristics and meteorology. To illustrate the difference in dry deposition models two parameterizations for canopy resistance are compared (equation (8)): the model first presented by Wesely [1989] with the modifications discussed by Walmsley and Wesely [1996] (referred to as W89 and W96 respectively) and the resistance model utilized by Zhang et al. [2003a] and discussed by Brook et al. [1999]; Zhang et al. [2002a] (identified as Z03).

equation image

[15] For simplification in model comparisons the terms in equation (8) are numbered so a physical description can be assigned to each term in the discussion. T1 in equation (8) includes the stomatal (Rst) and mesophyll (Rm) resistances for the leaf and is corrected for canopy wetness using the model presented by Janssen and Römer [1991] (Wst for Z03 and incorporated in calculations for W89/W96). T2 includes the in-canopy aerodynamic (Rac) and ground surface (Rgs) resistances, T3 is the cuticle (Rcut) resistance and is modeled on a per leaf basis for W89/W96 and for the canopy for Z03. The last term in the W89/W96 equation, T4, includes the buoyant convection (Rdc) resistance term and the resistance due to exposed surfaces in the lower canopy (Rcl). Wesely [1989] and Zhang et al. [2003a] both give schematics and in-depth descriptions for each of the resistance terms implemented in equation (8). In both of the models chemical species information is scaled to SO2 and O3 by using the dimensionless Henry's Law (H*) and reactivity (f0) constants [Wesely, 1989] or by using the scaling parameters α and β [Zhang et al., 2003a]. The RGM in each model is assumed to be HgCl2 [Lin et al., 2006; Marsik et al., 2007]. The following values for the molecular diffusivity of the species (Dx) and the ratio of water vapor diffusivity image to the species are used, for GEM: DGEM = 0.12 cm2 s−1, image = 1.82 and for RGM: DRGM = 0.09 cm2 s−1, image = 2.53 [Marsik et al., 2007; Lin et al., 2006].

2.2.4. Wesely Surface Resistance Model

[16] Wesely [1989] developed a surface resistance parameterization to model the dry deposition of a gaseous species due to surface interactions. While the implementation of the model establishes a framework for deposition estimates, there are some drawbacks and limitations. The benefits include being able to input, otherwise unknown, chemical properties of the modeled species with dimensionless Henry's Law (H*) and reactivity (f0) constants as scaling parameters, for GEM: H* = 0.139, f0 = 10−5 and for RGM: H* = 2.8 · 106, f0 = 1 [Lin et al., 2006]. The main drawbacks are the limited land use categories (11 LUC), requirement of a seasonal category (SC) selection and chemical species limitations (i.e., chemical properties of the species, varying chemical form in atmosphere). The model depends heavily on the LUC and SC selected for the reference values listed in Table 1 of Wesely [1989]. Our Table 2 lists the IOP, SC and description for the SFCR field study, all IOPs were modeled with the LUC for range land. While the impact on RGM modeling is minimal, the updates from Walmsley and Wesely [1996] greatly impact the GEM results, primarily due to the treatment of infinite resistances being changed from 9999 s m−1 to 1025 s m−1, increasing the Rc value thus lowering vd. The W89 model should be used with caution because when the W96 changes were added, Walmsley and Wesely [1996] showed that the calculated vd changed by as much as 33% depending on the LUC and species.

Table 2. Wesely [1989, Table 1] Seasonal Category (SC) Selected for Each IOP During the Salmon Falls Creek Reservoir Field Study Used for Our Implementation of the W89 and W96 Models
IOP NameSCDescription
March Data 14winter: snow on ground, subfreezing
March Data 25transitional spring with partially green short annuals
June1midsummer with lush vegetation
July SC11midsummer with lush vegetation
July SC22autumn with unharvested cropland
November2autumn with unharvested cropland

2.2.5. Zhang Surface Resistance Model

[17] An improved surface resistance scheme was presented by Zhang et al. [2003a], which includes a more complex parameterization of the inner canopy resistance with turbulence and meteorological variables, a larger LUC selection (26 versus 11 for W89/W96) and no SC requirement. Again, if the specific chemical properties of the species are not known, scaling values are used to relate species to SO2 and O3. A detailed description for the Z03 model is given by Zhang et al. [2003a]; however, Zhang et al. [2002a] give the most complete formulation of Rst and the development of the nonstomatal resistances is provided by Zhang et al. [2003b] for SO2 and Zhang et al. [2002b] for O3. Rst calculation requires photosynthetically active radiation (PAR) measurements. Since PAR was not measured in this experiment, the model presented by Weiss and Norman [1985] was used, and requires only incoming solar radiation to be measured. The PAR model, derived from experimental data separates direct and diffuse, as well as visible and near-infrared radiation components. For the leaf water potential coefficient, the general flow equation for the passive absorption of water through short vegetation from Brook et al. [1999] was used, and is a function of temperature. For the Table 1 values listed in the work by Zhang et al. [2003a], LUC = 2 (ice) were used for the winter (March Data 1) and LUC = 13 (short grass and forbs) were used at all other times. The LUC for ice was chosen for the winter IOP because no vegetation was above the snow, therefore the canopy resistances are not applicable. An LAI of 1.0 was used as suggested by Zhang et al. [2003a] based on the LUC selection. The snow fraction on leaves was not accounted for because the vegetation in the area does not hold snow. Also, a spherical leaf angle distribution was assumed. Previous research has shown that the modeled deposition for RGM (20–26% uncertainty) and GEM (minimal) are relatively insensitive to this parameter [Marsik et al., 2007]. For RGM the scaling parameters α = 10 and β = 10 were used [Marsik et al., 2007], and Rm = 0 s m−1 based on work by Zhang et al. [2002a] for a highly soluble species. The scaling parameters were not needed for GEM because values from literature [Marsik et al., 2007] were used for Rgs = 21400 s m−1 and Rcut = 33400 s m−1. For the GEM base case, Rm = 100 s m−1 for a weakly soluble species [Zhang et al., 2002a] was used. The model was also evaluated with Rm = 1000 s m−1 as this value is close to the Rm value obtained from the H* and f0 scaling by W89/W96 [Wesely, 1989; Lin et al., 2006]. Initially, the temperature dependent Rm from Du and Fang [1982] was utilized; however, further investigation of this equation for Rm indicated that it was not applicable for this work because it was obtained as a residual term from experimental analysis.

3. Results

[18] Annual deposition fluxes calculated with the two canopy resistance parameterization schemes are shown in Table 3 and compared to modeled fluxes from a previous field study at the same location [Abbott et al., 2008]. Measurements were only taken during five IOPs, therefore the annual calculations are based on summing the seasonal deposition estimates. The seasonal values were determined by multiplying the deposition flux by the number of days in each season, and assuming four equal seasons in one year. The five IOPs are listed in Table 2; March Data 1 was used for the winter deposition estimates and the July and November IOPs were used for autumn.

Table 3. Annual Dry Deposition (μg m−2) of Mercury for All Surface Resistance Models and Configurations Compared for the Salmon Falls Creek Reservoir Field Study, and for a Previous Field Study at the Same Locationa
Model NameGEMRGMTotal
W89 SC1−11.20−1.92−13.12
W89 SC2−8.63−1.65−10.29
W96 SC1−9.13−1.91−11.04
W96 SC2−6.56−1.64−8.20
Z03 Rm = 100−17.66−1.85−19.51
Z03 Rm = 1000−9.84−1.85−11.69
Abbott et al. [2008]−10.8−1.1−11.9

[19] For all cases in the current study the RGM fluxes are larger than the previous study, while the GEM fluxes are only larger for the Z03Rm = 100 s m−1 and W89 July SC1 cases. The W89 values are shown here for completeness and to show the effect of the W96 modifications, where W96 values are less than the W89 values due to the treatment of infinite resistance components in the model. This primarily effects GEM because Rc is the dominant term. For RGM, all three resistances are a similarorder of magnitude and the infinite resistance in Rc is not an issue. Due to the similar order of magnitude of the RGM resistance terms, it is assumed that the RGM annual flux differences between this study and Abbott et al. [2008] are due to differences in the Ra and Rb calculations. In the current study, turbulence data were used to calculate Ra and Rb, with a stability filter to eliminate unrealistically large Ra values that decrease the calculated deposition. Notice that the SC selection for the W89/W96 model greatly influences the deposition result, decreasing the total flux by as much as 26%.

[20] The seasonal fluxes are presented in Table 4 by IOP, to compare the W96 and Z03 models, with two Rm values for GEM in the Z03 model. The RGM values for Z03 are only listed in one column because the Rm value does not change and is always zero, for W96 and Z03 the RGM flux is similar. Whereas for GEM, the Z03Rm = 100 s m−1 always has the largest deposition, while W96 and Z03Rm = 1000 s m−1 are more similar. This result indicates the importance of Rm selection on the modeled flux for GEM, using the Z03 model. For reference, previous seasonal flux values at the same location for GEM were −0.04, −3.7, −6.9 and −0.10 μg m−2 for winter, spring, summer and fall; and for RGM were −0.22, −0.44, −0.33 and −0.12 μg m−2 for winter, spring, summer and fall [Abbott et al., 2008]. The seasonal and annual calculations of Abbott et al. [2008] use the same methodology as this study to obtain a seasonal deposition by multiplying the flux by the number of days in each season (assuming four equal seasons). The Wesely [1989] model was also used to calculate their fluxes; however, no turbulence data were used for the calculations and no stability restriction was placed on Ra.

Table 4. Seasonal Deposition (μg m−2) of GEM and RGM for Each Intensive Operating Period (IOP)a
Species NameIOP NameW96Z03 Rm = 100Z03 Rm = 1000
  • a

    W96: land use category (LUC) = 3 (range land), seasonal category (SC) listed in Table 2; Z03: LUC = 13 (short grass and forbs), except for winter LUC = 2 (ice).

GEMMarch 1 (winter)−0.00−0.51−0.51
GEMMarch 2 (spring)−1.99−2.74−1.86
GEMJuly (SC = 2)−0.00
GEMJuly (SC = 1)−2.58−6.26−2.66
RGMMarch 1 (winter)−0.27−0.26
RGMMarch 2 (spring)−0.36−0.39
RGMJuly (SC = 2)−0.65NA
RGMJuly (SC = 1)−0.92−0.82

[21] Two variables contribute to the calculated flux, the modeled vd and C. The value for C is measured, and while measurement error exists in C and is approximately 6% for GEM and 16% for RGM [Abbott et al., 2008], the more relevant error in this study is assumed to come from vd. The ensemble averaged hourly vd and C are shown for GEM in Figure 2 and Figure 3 for RGM, with error bars representing the standard deviation. A diurnal pattern can be seen in the vd plots, especially for GEM where the nighttime vd is nearly zero and summer daytime values (July IOP) almost reach 0.2 m s−1, 0.1 m s−1 and nearly zero for Z03, W96 SC1 and W96 SC2, respectively. The attenuated daytime peak values from W96 SC2 are due to the absence of vegetation uptake for this SC, while the other two models show a clear diurnal trend. This suggests that SC2 is the incorrect choice for the late summer July IOP if vegetation is present and enhances deposition. During the March Data 1 IOP (winter) both models show a very small vd and less of a diurnal pattern due to the vegetation and canopy parameterizations becoming infinite to model the snow-covered ground, therefore limiting the transfer of GEM to the surface. While the diurnal pattern for RGM vd is not as distinct, larger values are present in the afternoon, and the concentration of RGM shows more of a diurnal pattern than GEM, especially in the July IOP where RGM concentrations are at a maximum.

Figure 2.

Ensemble averaged GEM concentration and vd for each IOP in the SFCR field study: (left) GEM concentration and (right) modeled deposition velocity. Error bars indicate the standard deviation and seasonal category information for W96 in Table 2.

Figure 3.

Ensemble averaged RGM concentration and vd for each IOP in the SFCR field study: (left) RGM concentration and (right) modeled deposition velocity. Error bars indicate the standard deviation and seasonal category information for W96 listed in Table 2.

[22] These diurnal variations in vd are important because reporting mean vd values for seasonal or annual IOPs can result in an incomplete picture due to the large changes between daytime and nighttime values. As was done in this study and by Abbott et al. [2008], the hourly averaged vd values are typically used to calculate the annual and seasonal deposition; however, mean values are usually reported in the literature. Seasonal and annual fluxes are important for comparison with other studies, but for model improvement and to determine the physical relevance of the parameters modeled in Rc the diurnal variations in vd need to be investigated. Taking note of this precaution the mean seasonal vd values from Abbott et al. [2008] calculated with W89 are given here for comparison GEM: 0.034 ± 0.032, 0.043 ± 0.040, 0.00084 ± 0.0017 and 0.00036 ± 0.0011 cm s−1 for spring, summer, fall and winter and RGM: 0.5 ± 0.39, 0.40 ± 0.31, 0.51 ± 0.43 and 0.76 ± 0.57 cm s−1 for spring, summer, fall and winter. Even if one considers the full range of vd values, they are much lower than the values shown in Figures 2 and 3 for the same field study location. The values for RGM are more similar to the current study, but the GEM values are considerably lower. In addition to Rc differences, the lower values could be due to the turbulence parameters being used to calculate Ra and Rb or the stability criteria filter used in this study that removed unrealistically large values of Ra.

[23] The temporal variation and influence of the modeled vd and measured C on the calculated flux can be seen in Figure 4, a time series plot of GEM concentration, vd and Flux for the July IOP. Notice that the deposition flux is greatest when GEM concentrations are elevated and there is a large vd, with maximum hourly deposition fluxes of −0.008, −0.0034 and −0.0032 ng m−2 s−1 for Z03Rm = 100 s m−2, Z03Rm = 1000 s m−2 and W96 SC1. Since the W96 SC2 vd is minimal during the day, a nearly zero flux of GEM is calculated. This reinforces the importance of SC selection in the W96 model. While some studies have neglected GEM deposition, this shows that it should not be neglected due to large concentrations of GEM in the atmosphere, which at times is well above the global background of 1–3 ng m−3 for total Hg [Lin et al., 2006]. There is a large difference in the calculated daytime flux using Z03 when the Rm value is changed, and while the Z03Rm = 1000 s m−1 seems to correlate well with W96 (typically less than 10% difference in calculated daytime vd; see Figure 5, bottom right) it should not be assumed that this is the correct value. In addition to the correlation between the vd modeled with the Z03 models and W96 SC2, Figure 5 shows the correlation between the modeled vd and atmospheric parameters for the Z03Rm = 100 s m−1 and W96 SC2 models.

Figure 4.

SFCR July IOP hourly mean temporal variation of (top) GEM concentration, (middle) modeled deposition velocity, and (bottom) calculated GEM flux for W96 with two SC selections and Z03 with two Rm inputs; W96 (SC = 2) and Z03 (Rm = 1000 s m−1) are nearly indistinguishable. (Note that tic marks are every 24 h and denote 00:00.)

Figure 5.

The first three pairs of plots show July IOP GEM hourly mean linear correlation plots for temperature (T), solar radiation (SR), and friction velocity (u*) versus (left) W96 modeled vd and (right) Z03 modeled vd. The fourth pair of plots shows correlation between modeled vd with Rm = 100 s m−1 and Rm = 1000 s m−1.

4. Discussion

[24] The results presented in this work show a large discrepancy in the calculated fluxes depending on which canopy resistance scheme is used and the selection of input parameters for the model. Specifically, for W96 the SC selection is of great importance and for Z03 the Rm value for GEM is critical. Additionally, when comparing Hg fluxes with measurements or simulations from the literature, caution needs to be used to ensure similar quantities are being compared. When Hg flux calculations are made, the vd for GEM is ignored in many cases [Seigneur et al., 2001], therefore underestimating the value for total Hg deposition. Different modeling and measurement techniques lead to different vd results, which is apparent in the large range of literature values presented in section 1. Also, the resistance parameterization for dry deposition does not account for the emission of Hg from the surface, which recent studies have found to be of great importance to the global Hg cycle [Xu et al., 1999; Lindberg and Meyers, 2001; Lindberg et al., 2007; Poissant et al., 2004].

[25] To better investigate the differences between the models, the deposition components (labeled terms in equation (8)) were plotted to compare the two base case models from Z03 (with Rm = 100) and W96 (SC = 1). Values for the resistance components from equation (8) are also shown in Table 5 for a comparison between the two models. To assist in the interpretation of the term plots, shown in Figure 6, note that as a term (T1, T2, T3 or T4) goes to zero Rc becomes very large, decreasing the calculated vd. Additionally, the scales for each of the term plots shown in Figure 6 differ by orders of magnitude. Briefly, these terms represent deposition pathways for the inner leaf (T1), ground and canopy (T2), outer leaf (T3) and additional Wesely terms (T4); more detail can be found in section 2.2. T3 is not plotted because there is no temporal variation in Rcut, for Z03T3 = 3 · 10−5 m s−1 and for W96T3 = 6 · 10−9 m s−1 and T3 = 2 · 10−8 m s−1 for dry and dew wetted canopies, respectively. The variation in T4 was minimal; therefore it is also not plotted in Figure 6 and T4 = 1.1· 10−8 m s−1. Based on this information and the plot of deposition pathways (T1, T2) in Figure 6, it can be seen that for both models the largest contributor to an increased vd is the increase in T1 primarily due to Rst or the plant stomata opening during the day to increase uptake of GEM to the vegetation (see difference in minimum T1−1 values in Table 5). For Z03Rst is based on asunlit and shade model that is a function of temperature, PAR, leaf water potential, water vapor pressure deficit and table values for each of these that determine the minimum and maximum for stomatal opening. Here, W96 depend only on temperature, solar radiation and one table value corresponding to the minimum stomatal resistance. Included in T1 is the resistance due to the leaf mesophyll, the section of leaf between the outer coating (cuticle) and the leaf pore (stoma), making T1 based only on interior leaf properties.

Figure 6.

July IOP GEM hourly mean temporal variation of terms (T1, T2) in equation (8) for the surface resistance model and all resistance terms (R) in equation (5) for vd model. Two models are shown: those of W96 (SC = 1) and Z03 (Rm = 100 s m−1). (Note that tic marks are every 24 h and denote 00:00; plots on right are from 23 July 2008 11:00 to 25 July 2008 11:00 to show diurnal cycle.)

Table 5. July IOP GEM Resistance Components (s m−1) From Equation (8) for the Surface Resistance Modela
ResistanceW96 (SC = 1)Z03 Rm = 1000
  • a

    Two models are shown: W96 (SC = 1) and Z03 (Rm = 100 s m−1). Note that T1 is used instead of Rst and Rm to account for the wetness parameter (Wst) from Z03, where W96 includes the canopy wetness directly in the calculation of Rst and Rcut.

T1−1calculated: 1185–2.59 · 106calculated: 446–1.0 · 1025
Ractable: 100calculated: 16–61,337
Rgscalculated: 1.85 · 107literature: 2.14 · 104
Rcutdew: 4.81 · 107, dry: 1.76· 108literature: 3.34 · 104
Rdccalculated: 197–6811
Rclcalculated: 9.35 · 107

[26] For the case of GEM, the behavior between the gaseous species and the leaf mesophyll is not well understood and greatly impacts the modeled vd results [Lindberg et al., 1992; Marsik et al., 2007]. In the case of Z03, when Rm is increased it causes term T1 to decrease during the day, resulting in a smaller vd. Marsik et al. [2007] show the impact of Rm selection on the Z03 calculated vd. A change of Rm = 100 s m−2 to Rm = 10000 s m−1 resulted in a vd change from 0.46 cm s−1 to 0.02 cm s−1. They note that more research is needed to determine the physically appropriate value for Rm. The W96 results in a Rm of 955.7 s m−1; however, this is obtained using a scaling equation and involves several assumptions for O3 and SO2 behavior in the water within the plant leaf stomata.

[27] The other terms shown in Figure 6 for W96 are very small, causing a large Rc and lowering vd. At night, the largest term in equation (8) for W96 is T1; this is physically improbable because the leaf stomata are assumed to be closed at night, and therefore deposition cannot be dominated by vegetation uptake. T2 for Z03 includes an equation with u* and LAI to calculate a physically meaningful value for the in-canopy aerodynamic resistance, where the value for W96 is looked up in a table. While this does not impact the daytime vd values, the nighttime values for Z03 are larger than those of W96 because this term and T3 are larger. In Figures 4 and 5, the Z03Rm = 1000 s m−1 correlates well with W96 SC1, with less than 10% difference between the calculated daytime vd. This result is coincidental as the two models are not physically modeling the same processes, this is evident when comparing the nighttime vd for the two models, with differences typically exceeding 180%.

[28] The application of these canopy resistance schemes requires several atmospheric parameters as inputs. In the present case, many of these parameters were modeled; however, to improve the accuracy of flux calculations in the future it is recommended that these models be replaced with measurements. Recommended measurements include: fast-response 3-D velocity measurements (e.g., Sonic, as in this study), accurate LAI measurements, PAR, global radiation, precipitation, snow cover and dew formation. The correlation between atmospheric parameters (temperature, solar radiation, u*) and modeled vd for W96 and Z03 is shown in Figure 5 for the July IOP.

[29] This work compares calculated deposition velocities from two surface resistance parameterizations, using field data to compare realistic values for vd and to enable a comparison with previous vd calculations at the same location [Abbott et al., 2008]. Previous work has shown that the net flux of mercury depends on both the deposition and emission processes. Advancements are being made to incorporate this bidirectional exchange into air quality models for Hg (for a detailed description see Bash et al. [2007] and Bash [2010]). When modeling a bidirectional exchange, the resistance parameterizations for the vegetation and soil can be similar to the unidirectional deposition because both model similar physical processes Zhang et al. [2009]. Therefore, it is important to evaluate these resistance schemes with the simplest approach to determine which parameterizations contribute to the largest source of error in vd. The work presented here shows that for GEM, the resistance model formulation, seasonal category selection and scaling for the mesophyll resistance all contribute to a large amount of error in the calculated deposition. The following is a list of recommendations to improve the state of Hg modeling in air quality applications.

[30] 1. The model proposed by Wesely [1989] should be implemented with the modifications presented by Walmsley and Wesely [1996]. In the present work, these changes resulted in a 20–27% difference in annual GEM deposition.

[31] 2. For the Wesely [1989] model, it is recommended that quality control guidelines be established to ensure the proper selection of the seasonal category, including documentation justifying the selection.

[32] 3. Experiments are needed to quantify Rm for GEM; in this work there is a 57% difference in the annual deposition flux using the Zhang et al. [2003a] model due to changes in Rm. This is not the first time this recommendation has been made: Marsik et al. [2007] also recommend research to determine an accurate value for Rm.

[33] 4. For experimental studies using field data, Ra and Rb should be calculated with data and not a meteorological model.

[34] 5. Bidirectional models still rely on resistance parameterizations, similar to the present work [Zhang et al., 2009]. Therefore, the resistance components should be investigated in a similar manner to evaluate the physical relevance of each term.

[35] 6. Researchers should carefully specify how all parameters are obtained when results are presented, models are often used when data is not available and this must be explicitly stated.

5. Summary

[36] The results indicate large differences in the calculated downward flux for GEM depending on the parameterization chosen for the surface resistance, and the selection of tabular input parameters. Annual deposition fluxes calculated for GEM indicate a 64% difference depending on model selection, with up to a 200% difference for the calculated seasonal deposition. While these differences are large, the total gaseous mercury deposition calculated using the two canopy models presented provides a plausible range of deposition values for the region. There is a large dependency on seasonal and diurnal variations, and the method of obtaining the deposition velocity can influence the flux results. Flux measurements, gradient methods, chamber studies and models are all tools utilized to estimate the deposition velocity for a species. Therefore, a large range of deposition velocities for mercury exists in the literature and caution is needed when interpreting these values to ensure similar quantities are being compared. Particularly, this is the case for GEM, where a bidirectional flux may exist and can be dominated by the emission of Hg at the surface.


[37] The authors would like to acknowledge the support of the U.S. Environmental Protection Agency (EPA) Regional Applied Research Efforts (RARE) program; funding was provided through the RARE program for EPA Region 10. In addition, the authors are grateful to Che-Jen Lin at Lamar University for providing guidance in mercury deposition calculations and to Sage Aslet for allowing us to utilize his ranch at House Creek for our sampling location. The SFCR field experiment was completed with assistance from University of Utah graduate students Lance M. Richards and Joel R. Lisonbee.