Comparisons of mercury sources and atmospheric mercury processes between a coastal and inland site

Authors


Abstract

[1] Comparisons of mercury sources and atmospheric mercury processes were conducted between a coastal and inland site in northeastern North America. Identifying sources of atmospheric Hg is essential for understanding what is potentially contributing to Hg bioaccumulation at these two sites. A data set consisting of gaseous elemental mercury (GEM), gaseous oxidized mercury (GOM), particle-bound mercury, ozone, trace gases, particulate ions, and meteorological data were analyzed using principal components analysis (PCA), absolute principal component scores (APCS), and back trajectories. The PCA factors representing gaseous Hg condensation on particles during winter and combustion and industrial sources were found at both sites. However, the PCA factor for combustion/industrial sources was not found in 2010 at either site, likely because of SO2 emissions reductions from coal utilities from 2008 to 2010. Using APCS and back trajectories, the combustion/industrial factor at the coastal site was narrowed down to shipping ports along the Atlantic coast. Hg sources affecting coastal sites are different from those affecting inland sites because of the influence of marine airflows. GEM evasion from the ocean was evident from a PCA factor containing GEM, relative humidity, wind speed, and precipitation along with significantly higher contributions of this source (APCS) from oceanic trajectories compared to land/coastal trajectories. Analysis of the effects of ozone and water vapor mixing ratio on %GOM/total gaseous mercury suggest that Hg-Br photochemistry occurred at lower ozone concentrations (<40 ppb) at the coastal site and the absence of free troposphere transport of GOM.

1 Introduction

[2] Source-receptor studies on atmospheric mercury (Hg) reveal the major sources and atmospheric Hg processes that are potentially affecting ambient air Hg concentrations. In the current literature, receptor-based methods [Viana et al., 2008; Watson et al., 2008; Hopke and Cohen, 2011] have identified sources contributing to atmospheric Hg in major cities [Lynam and Keeler, 2006; Liu et al., 2007; Huang et al., 2010; Xu and Akhtar, 2010], rural areas [Han et al., 2007; Choi et al., 2008], coastal sites [Engle et al., 2008], and remote sites [Cheng et al., 2012]. Often, these studies are investigating a particular site or a few sites with a limited focus on identifying potential Hg sources affecting each site. In this study, receptor-based and statistical methods are being used to compare similarities and differences in sources and atmospheric Hg processes between a coastal site (Kejimkujik National Park, Nova Scotia, Canada) and an inland site (Huntington Wildlife Forest, New York, USA). The two sites were selected for this study because they are recognized biological Hg hotspots in North America [Evers et al., 2007]. Previous studies suggest that atmospheric Hg deposition could be a potential factor contributing to mercury bioaccumulation at the two sites [VanArsdale et al., 2005; Evers et al., 2007; Wyn et al., 2010].

[3] Atmospheric Hg is emitted from anthropogenic and natural sources. Pirrone et al. [2010] estimated that the global Hg emissions from anthropogenic sources was 2320 tons/year and reported that the largest Hg emissions come from fossil fuel combustion and artisanal gold mining. Natural emissions and re-emissions of previously deposited Hg originating from anthropogenic sources accounted for 5207 tons/year. The top two sources of these Hg emissions are from oceans and biomass burning [Pirrone et al., 2010].

[4] The coastal site is potentially influenced by Hg sources on land and from the marine boundary layer (MBL) due to its proximity to the ocean. Speciated atmospheric Hg studies conducted in the MBL revealed atmospheric Hg sources and processes that may be specific to marine environments. One of the sources or processes is the evasion of gaseous elemental mercury (GEM) from the ocean [Laurier et al., 2003; Sigler et al., 2009a; Soerensen et al., 2010], which was estimated to range from 800 to 2800 tons/year [Soerensen et al., 2010]. Scavenging of gaseous oxidized mercury (GOM) from the atmosphere by sea salt aerosols is believed to be a major sink for GOM in the marine environment [Engle et al., 2008; Holmes et al., 2009; Malcolm et al., 2009]. It is estimated that more than 80% of GOM in the MBL is absorbed by sea salt aerosols, which is then deposited into the ocean [Holmes et al., 2009] and potentially in coastal environments [Malcolm et al., 2009]. The in situ photochemical production of GOM by mercury-bromine reactions is particularly active in the polar regions and the MBL because the ocean is a major source of reactive bromine [Steffen et al., 2008; Obrist et al., 2011]. However, recent modeling research suggests these reactions can occur globally [Holmes et al., 2010]. The study examines whether these MBL sources and processes affect the coastal and inland sites.

[5] Principal components analysis (PCA) was used to determine Hg sources and atmospheric processes from a recent (2009–2010) data set containing speciated atmospheric Hg, trace gases and particulate inorganic ion concentrations, and meteorological data. Absolute principal components scores (APCS) were analyzed with back trajectories generated from the Hybrid Single-Particle Lagrangian Integrated Trajectory (HYSPLIT4) model to quantitatively interpret sources and processes identified in PCA.

2 Methods

2.1 Site description

[6] The Kejimkujik (KEJ) coastal site is a national park located in south central Nova Scotia, Canada (latitude, 44.32°; longitude, −65.2°; elevation, 170 m). The Huntington Wildlife Forest (HF) inland site is part of the Adirondack Park located in central New York (latitude, 43.97°; longitude, −74.22°; elevation, 500 m). The two sites are similar in terms of the forested landscape, but the terrain is quite different with the KEJ coastal site being mostly a flat plain and the HF inland site having a more complex topography with mountains and valleys.

[7] Both sites are distant from major cities and Hg point sources. The nearest major city from the coastal and inland site is ~120 km (Halifax) and ~180 km (Syracuse) away, respectively. There are no significant Hg point sources within a ~175 km radius of either sampling site. There are, however, a high density of Hg point sources in eastern parts of Canada (especially in southern Ontario and Quebec) and in the United States (especially in Ohio, West Virginia, Pennsylvania, and New Jersey; Figure 1).

Figure 1.

Site map showing the locations of the speciated atmospheric Hg sampling sites and regional Hg point sources emitting more than 5 kg of Hg/y (circles). Hg point source locations in 2009 and 2010 are from Environment Canada National Pollutant Release Inventory and U.S. EPA Toxics Release Inventory.

2.2 Hg measurements and data

[8] The two sites are part of the National Atmospheric Hg Network (AMNet). Measurements of speciated atmospheric Hg concentrations began in January 2009 at the KEJ coastal site and in November 2007 at the HF inland site. The data analyzed in this study spans a 2 year period from January 2009 to December 2010. The Tekran Speciation System (Models 1130/1135/2537) was used to measure GEM and the operationally defined species, GOM and particle-bound mercury (PBM). The speciation system operates on 3 h cycles at both sites: 2 h for measurement of GEM and collection of GOM and PBM and 1 h for analysis and quantification of GOM and PBM. The two sites are also monitoring hourly ground level O3 concentrations [Environment Canada, 2011a; USEPA, 2011a] and hourly meteorological data [Environment Canada, 2011b; USEPA, 2011a]. Hourly PM2.5 from 2009 was also available for the KEJ coastal site [Environment Canada, 2011c]. SO2, HNO3, and particulate inorganic ion concentrations (Ca, K, Mg, Na, Cl, NH4, NO3, and SO4) were measured daily at the KEJ coastal site [Environment Canada, 2011a] and weekly at the HF inland site [USEPA, 2011a].

2.3 Data analysis methods

[9] PCA is a data reduction method in which the variables in a data set are reduced to a smaller number of factors that explains the data set. It has been applied to speciated atmospheric Hg data [Lynam and Keeler, 2006; Liu et al., 2007; Huang et al., 2010] and used as an exploratory data model to explain potential sources affecting Hg measurements. Prior to running PCA, the daily averaged data set from the KEJ coastal site and weekly averaged data set from the HF inland site were normalized. The criteria to determine the suitability of a data set for PCA and the number of factors to retain are discussed in statistics textbooks [e.g., Pallant, 2005]. The suitability of the data sets for PCA was determined by Kaiser-Meyer-Olkin measure of sampling adequacy (>0.5) and Bartlett's test of sphericity (p < 0.05). The number of factors to retain depended on Kaiser's criterion (retain factors with eigenvalues > 1), scree plot (retain factors above a “bend” in eigenvalues plot), and Monte Carlo parallel analysis (retain factors when the eigenvalue obtained from PCA > eigenvalue obtained from parallel analysis).

[10] Although the algorithm for generating the PCA factors is very quantitative, there is a qualitative component to PCA. The interpretation of the PCA factors is based on examining pollutant markers or tracers with large factor loadings. The factor loading is a correlation coefficient between a variable (e.g., pollutant or meteorological variable) and a factor. Positive loadings mean that there is a positive correlation between a variable and a factor, whereas negative loadings mean that there is a negative correlation. Therefore, variables with high factor loadings can be used to represent sources because more of the variance in those variables is being explained by a factor. The data in this study consisted of air pollutants that are indicators of combustion and industrial sources (SO2, HNO3, SO42-, and PM2.5), urban and combustion/industrial/wildfires sources (O3), soil/crustal (Ca2+), biomass (K+), marine sources or road salt/deicing components (Na+, Cl, and Mg2+), and agricultural emissions (NH4+) [Lynam and Keeler, 2006; Viana et al., 2008; Watson et al., 2008; Zhang et al., 2008]. Meteorological variables, such as temperature, can be used to approximate the time of year, e.g., positive loading represents warmer seasons, whereas negative loading represents colder seasons. Relative humidity and precipitation are indicators of wet (positive loadings) and dry (negative loadings) conditions. The meteorological variables are useful because some Hg sources or processes are more likely to occur under certain meteorological conditions, e.g., increased coal/wood combustion for wintertime heating, wildfires occurring in hot and dry conditions, condensation of volatile Hg species on particles at low temperatures. Although this data set could identify a wide range of anthropogenic and natural Hg sources, the pollutant markers are not unique enough to identify specific types of industrial sources. Trace elements and organic carbon measurements would be better suited for this purpose.

[11] Another method being used in this study is the analysis of back trajectories generated from the HYSPLIT4 model [Draxler and Rolph, 2012; Rolph, 2012]. The trajectory duration was 48 h, and three trajectories were generated at 0:00, 12:00, and 24:00 for each sampling day at the KEJ coastal site, using the National Weather Service's National Centers for Environmental Prediction EDAS (Eta Data Assimilation System) 40 km archive meteorological data. The starting height of the back trajectories was 10 m above the model ground level, which was determined from vertical sounding data (mean sea level pressure and surface pressure) and the actual elevation of the KEJ coastal site. Trajectory model uncertainties could be due to errors in modeled wind fields, turbulence in the PBL, and starting height of trajectories [Stohl, 1998; Stohl et al., 2002], which affect the three-dimensional path traveled by the trajectories. Each sampling day at the KEJ coastal site was categorized as land, coastal, oceanic trajectories/airflow, and mixed trajectories/airflow, as previously done for another coastal site by Engle et al. [2008]. A comparable trajectory analysis was not carried out for the HF inland site because the weekly pollutant samples could not be easily characterized by any one airflow regime. The categorization of the trajectories was done by visual analysis. The three trajectories generated per sampling day were plotted on a map. When all three trajectories could be traced back to land areas (over the United States and/or Canada), the sampling day would be characterized by land airflows. Sampling days in which trajectories trace back along the Canadian or U.S. Atlantic coasts were considered coastal airflows. These trajectories lie close to the coastline over land and/or water. Sampling days associated with the oceanic trajectory origin had trajectories that trace back to the open Atlantic Ocean. These trajectories lie far away from the coastline in the remote Atlantic Ocean. When the three trajectories for each sampling day could not be categorized by any one airflow regime, the sampling day would be categorized as mixed trajectory origins (e.g., land/coast, land/ocean, coast/ocean). Pollutants (PBM, SO2) and meteorological (temperature, relative humidity, wind speed) data statistics were analyzed to determine whether the trajectories were representative of land, coastal, and oceanic conditions. For the purpose of categorizing the trajectories to land, coastal, and oceanic airflows, a high level of trajectory accuracy would not be required as compared to using trajectories for the purpose of identifying Hg point sources. Statistical comparisons of the pollutant and meteorological variables between the back trajectory origins suggest that the categorized back trajectories were representative of land and oceanic conditions. The frequencies of land, coastal, and oceanic back trajectories were also comparable to those estimated from measured wind data.

[12] To link the sources identified from PCA with trajectory origins, APCS for each factor/component were calculated for each sampling day. APCS represents the rotated absolute component score for each source/factor to each sampling day [Thurston and Spengler, 1985; Song et al., 2006].

[13] The study of sources, chemistry, and transport of atmospheric mercury has its complexities because air pollutant emissions from sources and all types of atmospheric processes are occurring simultaneously. It would not be feasible to experimentally control for Hg emissions one source at a time to study its effect on receptor sites, for example. Therefore, models are used to represent the complex environmental system. The receptor modeling approach used in this study is a fundamentally sound and well-researched method of exploring source-receptor relationships [Viana et al., 2008; Watson et al., 2008; Hopke and Cohen, 2011]. There are numerous atmospheric Hg source-receptor studies that have applied PCA [Lynam and Keeler, 2006; Liu et al., 2007; Huang et al., 2010] and a combined pollutants data and back trajectory modeling approach (e.g., potential source contribution functions [Han et al., 2007; Choi et al., 2008; Xu and Akhtar, 2010], and concentration fields analysis [Rutter et al., 2009]).

3 Results and discussion

3.1 Overall Concentrations

[14] GEM, GOM, and PBM concentrations at the KEJ coastal site (0.43–4.80 ng/m3, 0–52 pg/m3, and 0–20 pg/m3 in 2009, respectively, and 0.5–1.76 ng/m3, 0–12 pg/m3, and 0–34 pg/m3 in 2010, respectively) were within the range of the HF inland site (0.2–1.96 ng/m3, 0–32 pg/m3, and 0–39 pg/m3 in 2009, respectively, and 0.4–3.38 ng/m3, 0–44 pg/m3, and 0–39 pg/m3 in 2010, respectively) and other coastal-rural/remote sites listed in the supporting information. A comparison with measurements over oceans is necessary due to the proximity of the KEJ site to the Atlantic Ocean. Even though the open ocean is far away from mercury point sources, it could be a source of GEM and GOM [Laurier et al., 2003; Sigler et al., 2009a; Soerensen et al., 2010; Sprovieri et al., 2010a]. Total gaseous mercury (TGM = GEM + GOM) and GOM concentrations over the Atlantic Ocean in the Northern Hemisphere range from 1.4 to 2.2 ng/m−3 and 2.5 to 10 pg/m−3, respectively [Soerensen et al., 2010; Slemr et al., 2011].

3.2 PCA Factor Loadings

[15] PCA was applied to the daily averaged Hg, pollutants, and meteorological data for the KEJ coastal site and weekly averaged data set for the HF inland site. Weekly averaged data was used for HF because trace gases and particulate ions were sampled weekly. To ensure that the same averaging period was used for data from both sites, the higher sampling interval data (e.g., Hg, O3, and met variables) were averaged to correspond with the trace gases and particulate ions data. The analysis was conducted separately for each year at the KEJ coastal and HF inland sites. In the initial factor loadings table, all the variables were included in the analysis. Upon inspection of the factor loadings (i.e., correlation coefficients between the pollutant/meteorological variables and each factor), it was found that some variables were not present in any of the factors that contained GEM, GOM, and/or PBM. Because a factor loading is a correlation coefficient between a variable and a factor, variables in a factor must have some correlation in order to belong in the same factor. When mercury is not present in a factor, it means Hg is not correlated with the other variables; therefore, these variables would not be useful for characterizing sources of Hg and were not utilized in the second PCA run. Even though PCA was carried out with many pollutant variables initially, not all of them will be related to mercury. The subsequent analysis was refined by removing the variables that were not related to mercury. The following discussion is on the potential sources and processes represented by the factors in the second PCA run. These factor loadings are shown in Table 1. To aid the interpretation of the PCA factors, APCS for a factor were compared between seasons to determine when a particular source was most likely to occur (Table 2).

Table 1. PCA Factor Loadings and Sources or Processes Identified for the KEJ Coastal and HF Inland Site
Sources or Processes InterpretedText IDGEMGOMPBMO3PM2.5SO2HNO3CaKMgNaClNH4NO3SO4TRHWSPrec% var
  1. a

    In the second column, the Text ID is a label for each factor referred to in the text; KEJ = coastal site; HF = inland site; 09 = year 2009; 10 = year 2010. On the right side, T = air temperature; RH = relative humidity; WS = wind speed; Prec = precipitation; % var = % variance explained by each factor. n/a indicates PM2.5 were not available at the coastal site in 2010 and at the inland site from 2009 to 2010. Blanks indicate the absolute value of the factor loadings lower than 0.3, which is considered a weak correlation between a variable and a factor.

Combustion/industrial/wildfiresKEJ09-1  0.37 0.850.430.820.560.69   0.880.540.840.45   28.6
Coal combustion/industrial (winter)HF09-20.440.480.62 n/a0.900.84   0.38  0.59 −0.49   21.2
WildfiresHF10-1  0.350.78n/a  0.860.300.82      −0.87 −0.5236.4
HF10-20.710.880.68 n/a  0.360.41          23.3
Condensation on particles in winterKEJ09-2  0.770.66 0.69       0.51 −0.83   17.1
KEJ10-30.60 0.730.46n/a          −0.89   24.0
HF09-3  0.36 n/a    0.360.830.92 0.64 −0.64   18.9
Photochemical production of GOM;KEJ09-30.660.77 0.56            −0.55  12.4
transport of free troposphere air;KEJ10-10.340.89 0.69n/a           −0.82  26.9
spring seasonal trendHF09-1 0.68 0.65n/a  0.95 0.88      −0.87  24.4
GEM evasion from oceanKEJ09-40.39               0.580.780.7611.7
                      
GEM evasion from ocean/regional backgroundKEJ10-20.55  0.41n/a           0.380.860.7725.4
Urban emissionsHF09-40.690.37 0.60n/a          −0.38 0.94 14.1
Hg wet depositionHF09-5  −0.36 n/a           0.33 0.898.0
Table 2. APCS (Average ± Standard Deviation) Associated with Each PCA Factor and Season for the 2009 and 2010 Coastal and Inland Site Data
  1. a

    Top: KEJ = coastal site. Bottom: HF = inland site; 09 = year 2009; 10 = year 2010. APCS for each PCA factor were calculated from the PCA output as described in the supporting information and the means and standard deviations of the APCS were reported for each season. Source (PCA factor) contributions to the measurements can be compared between the seasons. Larger APCS indicates a larger source contribution to the sampling days in a particular season.

 20092010
SeasonKEJ09-1KEJ09-2KEJ09-3KEJ09-4KEJ10-1KEJ10-2KEJ10-3
Winter1.61 ± 1.041.63 ± 0.990.26 ± 0.555.02 ± 0.74−0.62 ± 0.514.61 ± 1.274.26 ± 0.95
Spring1.93 ± 1.240.78 ± 0.771.95 ± 1.154.89 ± 1.190.77 ± 1.164.42 ± 0.683.28 ± 0.65
Summer2.00 ± 0.58−0.31 ± 0.380.71 ± 0.715.27 ± 1.22−0.22 ± 0.694.00 ± 0.722.16 ± 0.28
Fall1.70 ± 0.640.44 ± 0.42−0.08 ± 0.434.76 ± 0.79−0.85 ± 0.624.16 ± 1.132.69 ± 0.61
 
 20092010
SeasonHF09-1HF09-2HF09-3HF09-4HF09-5HF10-1HF10-2
Winter−1.73 ± 0.771.60 ± 1.603.10 ± 1.375.48 ± 0.904.20 ± 0.94−1.84 ± 0.673.38 ± 0.96
Spring−0.26 ± 1.260.61 ± 0.521.81 ± 0.795.88 ± 0.774.61 ± 1.02−0.57 ± 0.723.96 ± 0.58
Summer−1.72 ± 0.420.41 ± 0.251.56 ± 0.174.33 ± 0.694.71 ± 1.04−1.94 ± 0.634.11 ± 1.02
Fall−1.86 ± 0.330.44 ± 0.422.01 ± 0.344.6 ± 0.804.45 ± 1.08−2.72 ± 0.583.83 ± 1.30

3.2.1 KEJ Coastal Site 2009

[16] Four factors were extracted from the 2009 coastal site data after removing Na+, Cl, and Mg2+. The first factor (KEJ09-1 in Table 1) could be characterized as the transport of combustion and industrial emissions due to the presence of PBM, PM2.5, SO42−, HNO3, NO3, and SO2. Other than PBM and PM2.5, all of these pollutants were also found in a factor representing the transport of industrial emissions at a remote site from another study [Cheng et al., 2012]. Wildfires are also a potential source of atmospheric Hg to the coastal site because the factor also included soil/dust and biomass pollutant markers, such as Ca2+ and K+, and had positive loadings on temperature. The positive loading on temperature in KEJ09-1 is consistent with warmer seasons (i.e., spring and summer). The drier conditions that often occur in warmer seasons are conducive to wildfires. In addition to GEM, PBM is also emitted from biomass combustion [Wiedinmyer and Friedli, 2007; Obrist et al., 2008; Wang et al., 2010; Huang et al., 2011]. Wang et al. [2010] suggested that PBM may be a better indicator of forest fires than GEM. Due to much higher background concentrations of GEM, more significant increases in PBM than GEM would likely be observed [Wang et al., 2010; Huang et al., 2011]. Therefore, the loading on PBM is likely associated with wildfires. It is possible for a factor to represent both combustion and industrial emissions and wildfires because these sources may be affecting the KEJ coastal site simultaneously. The factor loadings on NH4+ and NO3 may be indicative of long-range transport from agricultural areas, such as the U.S. Midwest, which supplies excess ammonia needed (the amount above what is required to react with particulate sulfate) to form particulate ammonium nitrate [Pitchford et al., 2009].

[17] The second factor might be related to increased coal combustion and condensation of gaseous Hg on particulate matter in winter because of the positive loadings on PBM and SO2, and the negative loading on temperature (KEJ09-2 in Table 1). Increases in atmospheric Hg concentrations in the winter at Northern Hemisphere sites were attributed to increases in coal or wood combustion for residential heating [Poissant et al., 2005; Choi et al., 2008; Sigler et al., 2009b]. APCS for KEJ09-2 were significantly higher in winter compared to summer (p < 0.05), which suggests that the coal/wood combustion source contributed more to winter measurements than in the summer (Table 2). The condensation of Hg on particulate nitrate is a strong possibility because lower temperature is one of the factors favoring particulate nitrate formation [Pitchford et al., 2009]. The presence of O3 is consistent with coal combustion because it emits ozone precursors, e.g., NOx.

[18] The third factor (KEJ09-3), which contained strong positive loadings on GEM, GOM, and O3 and negative loadings on relative humidity, may be attributed to several atmospheric sources and processes. The presence of GOM and O3 suggest that GOM was produced by photochemical reactions [Lynam and Keeler, 2006; Liu et al., 2007; Huang et al., 2010]. Although there are uncertainties on the atmospheric oxidants involved in the oxidation of GEM, studies suggest that mercury-bromine chemistry is particularly active in marine environments [Laurier et al., 2003; Holmes et al., 2009; Obrist et al., 2011], which may be applicable to the KEJ coastal site. The transport of free troposphere air due to high-pressure zones and katabatic winds could result in higher concentrations of GOM and O3 and drier conditions, but these atmospheric circulation patterns mainly affect high-altitude sites, e.g., western United States [Fiore et al., 2003; Swartzendruber et al., 2006; Weiss-Penzias et al., 2009]. This factor might also represent a common seasonal cycle for these variables because GEM, GOM, and O3 concentrations were highest in spring and relative humidity was also at its lowest. A further analysis will be conducted in a later section to examine relationships between GEM, GOM, O3, and relative humidity.

[19] The presence of GEM, precipitation, wind speed, and relative humidity in the fourth factor (KEJ09-4) are representative of evasion of GEM from the ocean. Back trajectories at the KEJ coastal site were categorized to land, coastal, oceanic, and mixed trajectory origins. GEM concentrations in 2009 and air temperature, wind speed, and relative humidity in both years were significantly higher for oceanic trajectories compared to land and coastal trajectories (p < 0.05). Precipitation on the sampling days associated with oceanic trajectories in 2009 and 2010 were also higher than land/coastal trajectories (p < 0.05), but significant differences were found only in 2010. The higher relative humidity is representative of oceanic airflows and also increases the likelihood for rain. Sigler et al. [2009a] suggested that a nor'easter (major Atlantic storm) event in New Hampshire could enhance the evasion of GEM from the ocean because of strong turbulence. Hence, there was a strong factor loading on precipitation. Higher solar radiation intensity (which is strongly related to air temperature) and wind speeds also increase the evasion of GEM from oceans [Laurier et al., 2003]. These findings are consistent with the higher air temperatures and wind speeds for oceanic trajectories compared with land and coastal trajectories (p < 0.05) and the positive factor loading on wind speed. Na+, Cl, and Mg2+ are tracers of sea salt aerosols and represent marine airflow, which suggest that these ions should be present in KEJ09-4. But unlike GEM, sea salt aerosols are efficiently removed by precipitation [Tsyro et al., 2011]. Hence, the factor loading on precipitation suggests these ions were scavenged by wet deposition. See APCS–back trajectory results for further quantitative interpretation of the KEJ09-4 factor.

3.2.2 HF inland site 2009

[20] NH4+, SO42−, and K+ were removed from the 2009 inland site data prior to the second PCA run. Several factors extracted from the 2009 inland site data in Table 1 were similar to those from the coastal site, such as the factor with GOM, O3, and relative humidity (HF09-1), transport of combustion and industrial emissions in winter (HF09-2), and condensation of GEM or GOM on particulate matter in winter (HF09-3).

[21] The factor with GOM, O3, and relative humidity (HF09-1) may be related to several Hg processes similar with the coastal site. These processes likely occurred in the spring because the highest GOM, O3, Ca2+, and lowest relative humidity were also observed in this season. Unlike the coastal site, the factor representing combustion and industrial emissions in winter at the inland site also contained GEM and GOM (HF09-2). The factor, however, did not contain NH4+, SO42−, K+, and Ca2+. The lack of secondary pollutants (e.g., NH4+, SO42−) and the presence of GOM might be due to coal combustion sources being relatively closer to the inland site than the coastal site. The negative factor loading on temperature could represent colder seasons, such as winter, which is consistent with increased coal or wood combustion for residential heating. APCS for HF09-2 were also statistically higher during winter than any other season (p < 0.05), which confirms that the coal combustion source contributed more to winter measurements (Table 2). The third factor (HF09-3) represents the condensation of gaseous Hg on particulate nitrate, road salt, and deicing substances (e.g., MgCl2) that are often applied in winter at the Adirondack Park [Kelting and Laxson, 2010]. Na+, Cl, and Mg2+ are also markers for sea salt aerosols, but their influence on the inland site might not be significant because it is located approximately 400 km from the Atlantic coast.

[22] There were two factors that were extracted from the inland site data but not from the coastal site data. One of these factors contained positive loadings for GOM, O3, wind speed, and GEM and negative loadings on temperature (HF09-4). The variables in this factor could characterize the transport of oxidizing air mass from urban and industrial areas. This factor was observed at the inland site rather than the coastal site because it is relatively closer to urban and industrial areas. The negative loading on temperature is a potential indicator of colder seasons, but because of the factor loading on wind speed, it can be interpreted as colder airflows (i.e., from north of the HF site). There are several Hg point sources north of the HF site near Montreal, Quebec (see Figure 1), which were also identified as a potential Hg source region to the HF site by Choi et al. [2008]. The factor HF09-5 could represent Hg wet deposition due to PBM because of the negative loading on PBM and positive loadings on relative humidity and precipitation.

3.2.3 2009 to 2010 comparisons

[23] Trace gases and particulate ions were excluded in the second PCA. The three factors retained are similar to those from the 2009 data, such as the factor with GOM, O3, GEM, and relative humidity (KEJ10-1 in Table 1), evasion of GEM from oceans (KEJ10-2), and condensation of gaseous Hg on aerosols in winter (KEJ10-3). The presence of O3 in KEJ10-2 was unexpected because O3 concentrations over the ocean are typically lower due to photolysis in the presence of water vapor, reactions with reactive halogens, and a lack of NOx [Read et al., 2008].

[24] A major difference between the PCA results for 2009 and 2010 data was a lack of a combustion and industrial factor in the 2010 data for both the KEJ and HF sites. Na+, Cl, NO3, NH4+, SO42−, HNO3, SO2, wind speed, and temperature from the inland site data were removed prior to the second PCA due to the lack of relationships with speciated atmospheric Hg. The two factors (HF10-1 and HF10-2) that were kept were likely related to wildfires because of the factor loadings on Ca2+ and K+ (i.e., soil and biomass markers). The negative loading on relative humidity and precipitation in HF10-1 could represent the dry conditions needed to sustain wildfires. Wang et al. [2010] previously reported that forest fires in Quebec, Canada affected speciated atmospheric mercury concentrations at several rural sites in New York.

[25] The agreement in the interannual results between the coastal and inland sites suggests that both sites were affected by the same combustion and industrial sources. The absence of the combustion and industrial source profiles may be due to large reductions in SO2 (>887,000 tons or 84% reduction) and NOx (>60,000 tons or 37% reduction) emissions from coal-fired power plants across the United States between 2008 and 2010. A previous study conducted in Rochester, New York observed decreases in RGM and SO2 correlations as the operation of a local coal-fired power plant was being phased out [Huang et al., 2010]. However, despite the shutdown of the local coal-fired power plant, the coal combustion signature was found in the PCA results for the Rochester site because regional coal combustion sources were still in operation. In this study, the large-scale emissions reductions in coal combustion sources were reflected in the PCA results.

[26] Source emission reductions have important implications on the mercury bioaccumulation issues reported at the two sites because studies suggest that Hg accumulated in forest canopies are primarily derived from the atmosphere [Risch et al., 2012]. However, understanding the causes of the biological Hg hotspots would require further studies on wet and dry Hg deposition, mercury cycling in forest canopies, and post deposition processes.

3.3 APCS–back trajectory relationships

[27] APCS were used to determine when the sources identified from the PCA factors contributed to the sampling days. The PCA factor loadings in the previous section are the correlation coefficients between a variable and a factor; therefore, they do not explain how the sources affected the individual sampling days. APCS was calculated for each sampling day at the KEJ coastal site, which has been previously assigned a back trajectory origin (e.g., land, coastal, oceanic). It was not applied to the HF inland site because the data was averaged weekly, and there could be large variability in the back trajectory regimes over a 1 week period. APCS were calculated from the PCA output, which was derived from the input of daily averaged (for KEJ coastal site) speciated atmospheric Hg, trace gases, particulate ions, and meteorological data. The purpose was to quantitatively link sources to trajectory origins because some of the sources identified from the factor loadings could be associated with a specific trajectory regime. For example, combustion and industrial emissions are likely related to land and coastal airflows; evasion of Hg from the ocean should be closely associated with oceanic trajectories, etc.

[28] The average APCS for the factor representing transport of combustion and industrial emissions and wildfires (KEJ09-1) were significantly higher for coastal trajectories than land-based trajectories (p < 0.05; Table 3). APCS for oceanic trajectories were not significantly different from coastal or land trajectories. A potential source of combustion/industrial emissions along the Atlantic coastline in Canada and in the United States may be the numerous shipping ports and vessels [USEPA, 2011b]. Ships that use large marine diesel engines emit large quantities of NOx, SOx, and PM2.5 [USEPA, 2011b]. Particulate matter from diesel exhaust also contains trace amounts of air toxics including Hg, Cr, Mn, and Ni [USEPA, 2011c]. This source is consistent with the presence of the factor loading for PBM instead of GEM or GOM. Measurements of speciated atmospheric Hg over the Adriatic Sea were also attributed to shipping and port activities [Sprovieri et al., 2010b]. The soil pollutant markers present in the KEJ09-1 factor are consistent with the soil dust factor identified in aerosol compositions from the U.S. East Coast [Castanho et al., 2005].

Table 3. APCS (Average ± Standard Deviation) Associated with Each PCA Factor and Back Trajectory Origin for the 2009 and 2010 Coastal Site Data
Back Trajectory Origin20092010
KEJ09-1KEJ09-2KEJ09-3KEJ09-4KEJ10-1KEJ10-2KEJ10-3
  1. a

    KEJ = coastal site; 09 = year 2009; 10 = year 2010. APCS for each PCA factor were calculated from the PCA output as described in the supporting information and the means and standard deviations of the APCS were reported for each back trajectory origin. Source (PCA factor) contributions to the measurements can be compared between the back trajectory origins. Larger APCS indicates a larger source contribution to the sampling days associated with a particular trajectory origin.

  2. b

    Indicates few sampling days with mixed land/ocean airflows.

Land1.43 ± 0.701.31 ± 0.980.46 ± 0.714.58 ± 0.67−0.31 ± 0.903.89 ± 0.573.58 ± 1.16
Land, coast1.79 ± 0.960.87 ± 0.830.56 ± 1.074.94 ± 0.90−0.07 ± 1.174.20 ± 0.762.96 ± 0.82
Land, ocean1.80 ± 0.660.24 ± 0.740.31 ± 0.935.44 ± 1.100.01 ± 1.14b4.10 ± 0.99b3.17 ± 0.83b
Coast2.59 ± 1.480.74 ± 0.701.56 ± 1.314.41 ± 0.900.06 ± 1.144.11 ± 0.472.65 ± 0.60
Coast, ocean2.40 ± 1.090.51 ± 0.941.22 ± 1.425.46 ± 0.88−0.53 ± 0.705.03 ± 1.022.59 ± 0.50
Ocean1.74 ± 0.78−0.29 ± 0.401.21 ± 1.515.91 ± 1.11−0.46 ± 0.685.42 ± 1.722.29 ± 0.56

[29] Significantly higher APCS were found for land and coastal trajectories than oceanic trajectories for the factor associated with winter coal combustion and condensation of Hg on particulate nitrate (KEJ09-2 and KEJ10-3). This finding is expected because both the coal combustion and particulate nitrate sources are from land and were observed in both the 2009 and 2010 data (Table 3).

[30] In the 2009 data, the average APCS was higher for coastal trajectories than land-based trajectories for the factor containing GEM, GOM, O3, and negative loadings on relative humidity (KEJ09-3 in Table 3). The NOx emissions from marine mobile sources and transportation emissions from large metropolitan areas along the East Coast likely contributed to O3 formation [USEPA, 2011b]. However, the APCS were not statistically different between the trajectory origins in 2010 (KEJ10-1 in Table 3).

[31] For the factor related to evasion of GEM from oceans, the average APCS for ocean and coast/ocean trajectories were statistically higher than both land and coastal trajectories. This result was observed in both the 2009 (KEJ09-4) and 2010 (KEJ10-2) data.

3.4 Relationships between GEM, GOM, O3, and water vapor content

[32] The springtime averages and correlations and the PCA results illustrated that there is strong link between GEM, GOM, O3, and relative humidity, but not with any other pollutant variables. However, it was not clear what these variables represent and could be attributed to several sources and processes, such as in situ photochemical production of GOM.

[33] All measurements of GEM, GOM, O3, and relative humidity in 2009 and 2010 were used to determine the effect of O3 concentration ranges (O3 bins) on %GOM/TGM for each site and each year (Figures 2a and 2b). The first O3 bin was 0 to 19 ppb, which increased by intervals of 10 ppb for subsequent O3 bins. %GOM/TGM represents the degree of oxidation of GEM. Even though neither Br nor OH concentrations were measured at the coastal or inland sites, the lower O3 bins (<40 ppb) could represent the presence of Br and OH. This is because reactive Br leads to O3 depletion [Read et al., 2008; Finlayson-Pitts, 2010; Obrist et al., 2011] and OH is primarily produced by the photolysis of O3 in the presence of water vapor [Atlas et al., 2003; Read et al., 2008]. However, the focus will be on the potential role of Br because modeling studies suggest that OH is likely a minor oxidant of GEM [Calvert and Lindberg, 2005]. One-way ANOVA was used to determine whether %GOM/TGM was significantly different (p < 0.05) between the O3 bins. Independent samples t test was used to determine whether %GOM/TGM was significantly different (p < 0.05) between sites at each O3 bin.

Figure 2.

Average %GOM/TGM and WVMR at different O3 concentration ranges (bins) in (a) 2009 and (b) 2010 at the coastal site (KEJ) and the inland site (HF). Bars represent average %GOM/TGM and the error bar is the standard deviation. Triangles represent average WVMR.

[34] At the KEJ coastal site in 2009 and 2010, there were no significant differences in %GOM/TGM between the lower O3 bins (i.e., 0–19 vs. 20–29 vs. 30–39 ppb O3). At higher O3 bins (≥40 ppb), %GOM/TGM increased significantly with O3 bins. At the HF inland site in 2009, %GOM/TGM increased significantly with all O3 bins. In 2010, no significant differences in the %GOM/TGM were observed between the 20 to 29 and 30 to 39 ppb O3 range. Once again at higher O3 bins (≥40 ppb), %GOM/TGM increased significantly with O3 bins. At lower O3 concentrations, it was observed that %GOM/TGM did not increase significantly with O3 in both years at the coastal site and only in 2010 at the inland site. This suggests that at lower O3 concentrations, atmospheric oxidants such as Br and OH may be involved in the oxidation of GEM. A further important observation is the higher %GOM/TGM in the 70 to 79 ppb O3 bin in 2010 at the HF site compared to the KEJ site. This might not be related to Hg photochemistry and could be explained by the HF site experiencing stronger pollution episodes than the KEJ site in 2010. Based on PCA results, HF10-1 was related to wildfires and contained a positive loading on O3. At the KEJ site in 2010, there were three PCA factors with O3 (see KEJ10-3, KEJ10-1, and KEJ10-2 in Table 1), but the factors were not related to strong pollution (e.g., Hg condensation on particles in winter, Hg photochemistry, evasion of GEM from oceans).

[35] Significant differences in %GOM/TGM between the KEJ coastal site and HF inland site were found at each O3 bin < 40 ppb. In 2009, %GOM/TGM at the coastal site was higher than the inland site for the < 40 ppb O3 bins (Figure 2a). This is aligned with previous field studies, which suggested the enhanced GEM oxidation by reactive halogens in the MBL [Laurier et al., 2003; Obrist et al., 2011]. But the trend was reversed in 2010, with higher %GOM/TGM observed at the inland site (Figure 2b).

[36] Although theoretical studies suggest that ozone cannot oxidize atmospheric mercury efficiently [Calvert and Lindberg, 2005], the strong relationship between GOM and O3 concentrations observed at the coastal and inland sites from this study and other sites [Poissant et al., 2004; Abbott et al., 2008; Sigler et al., 2009b; Cheng et al., 2012] have not been completely explained. If the GOM-O3 relationship is believed to be related to mercury photochemistry, then it might be competing with emerging evidence of mercury-bromine chemistry. As reported by Obrist et al. [2011], the time series trends for GOM and BrO concentrations were opposite from those of O3 concentrations. If it is due to strong pollution episodes, a strong GOM-O3 correlation could be expected because GOM may be directly emitted from sources. These sources should also be emitting precursors of ozone, such as NOx and VOCs. However, the PCA factors (KEJ09-3, KEJ10-1, and HF09-1) from this study did not have other pollutants present.

[37] An alternative explanation for the GOM-O3 relationship is the transport of free troposphere air, which is elevated in GOM and O3, and associated with low relative humidity conditions [Weiss-Penzias et al., 2009]. Unfortunately, the relationship between GOM, O3, and relative humidity is also not very clear when the water vapor mixing ratios (WVMR) are analyzed. WVMR is the mass of water vapor to the mass of dry air (g/kg) and is calculated from relative humidity and saturated WVMR (WVMRs) [Poissant, 1997]. WVMRs is a function of air temperature and atmospheric pressure. Figures 2a and 2b show that low WVMR (i.e., drier air) was associated with moderate O3 concentrations (30–40 ppb), whereas high WVMR was associated with both low and high O3 concentrations at both the coastal and inland sites. Therefore, it is likely that the elevated GOM or O3 concentrations are not from the transport of free troposphere air. Modeled trajectories have also been applied at two other coastal sites in the Gulf of Mexico to examine this mechanism [Engle et al., 2008; Weiss-Penzias et al., 2011]. The results from these studies were mixed despite the proximity of the sites (~50 km) and use of HYSPLIT model at both sites.

4 Conclusions

[38] Hg sources and atmospheric Hg processes at coastal and inland sites were compared by analyzing a recent data set consisting of speciated atmospheric Hg, trace gases and particulate ion concentrations, and meteorological data. The analysis demonstrated that the coastal and inland sites were affected by combustion and industrial emissions and wildfires. However, emission reductions from coal combustion sources in the following year may have led to the absence of pollutant markers in the PCA factors. APCS-trajectory analysis suggested the source of combustion and industrial emissions at the coastal site were the shipping ports along the Atlantic coast. The condensation of gaseous Hg on particulate matter in winter was identified at the two sites, but the particle compositions were different, which reflects the different sources of particulate matter, e.g., particulate NO3 at the coastal site and road salt at the inland site. The analysis also revealed major differences in sources and atmospheric Hg processes between a coastal and inland site, which are attributed to the influence of marine airflows to the coastal site. PCA and back trajectory data suggested that the coastal site was affected by the evasion of GEM from the ocean. Analysis of the effects of O3 concentrations on the degree of GEM oxidation (%GOM/TGM) showed that the oxidation of GEM by Br could occur at the coastal site when O3 concentrations are < 40 ppb.

Acknowledgments

[39] We thank Amy Hou, Mike Shaw, and Robert Vet of Environment Canada and Greg Skelton for data verification and extraction for the Kejimkujik site and site operators at KEJ and HF sites who contributed to AMNet, NatChem Particulate Matter Database, NAPS data, National Climate Data Archive, and U.S. EPA CASTNET. NOAA Air Resources Laboratory is acknowledged for the provision of the HYSPLIT transport and dispersion model and READY web site (http://www.arl.noaa.gov/ready.php) used in this publication.

Ancillary