The site discussed herein is maintained by the New York State Department of Environmental Conservation (NYSDEC) as a part of AMNet under the National Atmospheric Deposition Program (NADP) and the National Toxics Network (NTN). The monitoring site is located on the rooftop of the Pfizer Plant Research Laboratory on the northern edge of the New York Botanical Garden in the north Bronx (40∘52′05′′ N, 73∘52′42′′ W; US EPA site ID 36-005-0133). The height of the measurement point is about 9 m from ground surface, and winds arriving at the location are not significantly obstructed by immediate surroundings. The 100 ha New York Botanical Garden is surrounded by highways and mixed residential–commercial areas. New York City is a metropolitan area with a population > 19 million, and the region has a long manufacturing, petrochemical, and industrial legacy that includes contamination from Hg and other toxic compounds. The Bronx site is also downwind of many regional sources (Fig. 1). Continuous measurements of meteorological variables and trace gas and toxic air pollutants are conducted at this site. Additional details on the site can be found on the NYSDEC website (http://www.dec.ny.gov/docs/air_pdf/2017 plan.pdf).

GEM was measured every 5 min using a Tekran (Toronto, ON) model 2537B (27 August 2008 through 24 October 2013) or 2537X (25 October 2013 onward) cold-vapor atomic fluorescence (CVAF) analyzer with a nominal detection limit of < 0.1 ng m-3 (∼ 11.2 ppqv). The instrument was calibrated daily with an internal permeation source. The Tekran system was operated according to standard operating procedures from the NADP's AMNet. The AMNet site liaison performs annual site visits, which include manual injections to verify the internal permeation source, and is responsible for quality assurance of the data (Civerolo et al., 2014). Additional details can be found in Landis et al. (2002) and Gay et al. (2013).

Measurement data of sulfur dioxide (SO2), nitrogen dioxide (NO2), carbon monoxide (CO), temperature, wind direction, and wind speed were averaged hourly. The SO2 measurements were taken using a TEI 43C and a 43i TLE instrument using pulsed fluorescence. The NO2 measurements were taken using a TEI 42C instrument using conversion on a heated molybdenum catalyst followed by chemiluminescence. It is acknowledged that this method is not specific to NO2 and suffers from interferences due to other oxides of nitrogen. However, in this dense urban area with ample fresh anthropogenic emissions, this artifact is relatively small in an absolute sense since nitric oxide (NO) and NO2 account for a substantial fraction of total reactive nitrogen. CO was measured by a TEI 48C and an API 300EU instrument using reference method 054 with nondispersive infrared absorption. The technical details of the deployment of these instruments are given by the NYSDEC at www.dec.ny.gov/chemical/8541.html and in the 2016 Annual Monitoring Network Plan (www.dec.ny.gov/docs/air_pdf/2016plan.pdf).

The impact of regional and local anthropogenic sources was simulated using the NOAA HYSPLIT Atmospheric Transport and Dispersion Modeling System (Draxler and Hess, 1997, 1998; Draxler, 1999; Stein et al., 2015) for the winters and summers of 2009–2015. HYSPLIT was driven by the EDAS 40 km model output over a domain extending westward to OH, southward to northern VA, and northward to include New England (Fig. 10a). The model was run in the forward mode for 120 h starting from each day of a season.

The dispersion of a pollutant is calculated by assuming a fixed number of particles being advected about the model domain by the mean wind field and spread by a turbulent component. By assigning certain mass to a particle, emissions are also incorporated in the model (Stein et al., 2015). Within the domain there was a total of 522 counties reporting Hg emissions that were extracted from the National Emissions Inventory (NEI) 2011 of the US Environmental Protection Agency (EPA) (https://www.epa.gov/air-emissions-inventories/2011-national-emissions-inventory-nei-data). Note that the total emissions of Hg were treated as 100 % GEM emissions. The US EPA's NEI documents emissions on an annual county basis. For model simulations the EPA emission amount for each county was broken down to an hourly rate by the annual emission amount divided by (365 × 24). Two emission scenarios were designed. One scenario included emissions from all the 522 counties, and the other scenario excluded emissions from the five boroughs in NYC. Output of scenario no. 2 quantified the effect of anthropogenic emissions outside of NYC (denoted as regional sources) on NYC ambient GEM concentrations due to long-range transport only. The difference in NYC GEM concentrations between the two scenarios was used to approximate the effect of only local sources on NYC GEM concentrations.

In the analysis of the US East Coast trough, the trough axis index (TAI) and trough intensity index (TII) defined by Bradbury et al. (2002) were used to quantify the position and intensity of the US East Coast trough. The seasonal TAI quantifies the mean longitudinal position of the quasi-stationary midtropospheric East Coast trough. The TAI domain extends from 120 to 30∘ W and the southern to northern boundaries ranged from 40 to 50∘ N. The TAI index was calculated by averaging the longitudinal positions (Long) of the minimum 500 hPa heights (Hmin) observed at each of the four latitudinal steps (j=40, 42.5, 45, and 47.5∘ N) within the index range to produce a practical index in longitudinal units (relative to the prime meridian): TAI=average[Long(Hmin)j].

The TII is an estimate of wave amplitude at 42.5∘ N and is the mean height change at the 500 hPa surface from equal distances east and west of the East Coast trough axis. It was calculated using TII=Hmini-Hi+30∘+Hmin⁡i-Hi-30∘2. The more negative TII is, the stronger the influence of the US East Coast trough would be. The 2.5∘×2.5∘ reanalysis data from the National Center of Environmental Protection/National Center of Atmospheric Research were used to calculate average seasonal TAI and TII. Additional details about TAI and TII can be found in Bradbury et al. (2002).

Annual cycles of 2009, 2010, and 2014 displayed larger GEM mixing ratios (> the 75th percentile value of the entire study) during the warm seasons (summer–spring) than the cool seasons (fall–winter) (Fig. 2, Table 1), in agreement with previous urban site studies (Denis et al., 2006; Liu et al., 2007; Zhu et al., 2012; Zhang et al., 2013; Civerolo et al., 2014). The pattern of such annual cycles was evidenced in > 20 % (< 10 %) of the warm (cold) season in 2009 and 2010 and 67 % (31 %) of the warm (cold) season in 2014 experiencing larger GEM mixing ratios (Table 1). However, this pattern was not reproduced in 2011 and 2012, when the frequency of occurrence of larger GEM values was either comparable between the two seasons or slightly higher in the cold season.

Three salient features were evident in the interannual variation in a range of percentile mixing ratios of GEM (Fig. 3, Table 1). First, the 2009–2010 cool season percentile values of GEM were the lowest of all cool seasons. Second, the 2011 warm season percentile values were the lowest of all warm seasons, even lower than in the cool season of the same year, not reproducing the 2009 and 2010 annual cycles. Third, the 2014 and 2015 seasonal percentile values were mostly the highest of the study period and for the first time since 2011, warm season values exceeded the cool season ones, reproducing the 2009 and 2010 annual cycles.

The most pronounced diurnal cycles occurred in summer, as shown in seasonal average diurnal cycles in Fig. 4, with a peak between 02:00 and 06:00 UTC and a minimum between 10:00 and 16:00 UTC, which is consistent with previous studies for urban locations (e.g., Denis et al., 2006; Liu et al., 2007; Zhu et al., 2012; Lan et al., 2012). In the summers of 2009–2012, the daily maximum was ∼ 170–190 ppqv and the daily minimum ∼ 140–160 ppqv. The diurnal amplitude, defined as the difference between the daily maximum and minimum, was up to ∼ 50 ppqv in summer, ∼ 20 ppqv in fall and spring, and < 10 ppqv in winter.

During 2008–2013, the cool seasons experienced much larger interannual variability in GEM than the warm seasons did, whereas in 2014 and 2015 GEM concentrations were elevated significantly above other years in all seasons (Fig. 4). Over 2008–2013, the largest interannual variability of up to ∼ 40 ppqv difference was observed between the lowest GEM mixing ratios in fall 2009 and the largest in fall 2012, whereas spring and summer experienced much less interannual variability, except spring and summer 2011, as aforementioned, which saw the lowest GEM mixing ratios, ∼ 20 ppqv lower than all other warm seasons.

The 2009–2010 cool season exhibited the lowest percentile values, whereas most of winter 2014 and cool season 2014–2015 percentile values were the highest of the study period (Table 1, Fig. 3). The difference in percentile values between the two cool seasons ranged from 30–40 ppqv in the 25th percentile and median values to 37–67 ppqv in the 90th percentile. The possible effect of anthropogenic emission changes on those interannual variations in GEM concentrations was the very first to be examined. EPA national emission inventories showed a 13 % decrease from 2008 (31 810 kg) to 2011 (27 695 kg) and then an increase of 2 % to 2014 (28 270 kg) in total emissions from the eastern US, including states east of the Mississippi River, of which NYC emissions increased from 125 kg in 2008 to 145 kg in 2011 and to 199 kg in 2014 (https://www.epa.gov/air-emissions-inventories/2014-national-emissions-inventory-nei-data). Using an average planetary boundary layer height of 1000 m over the eastern US (the surface area for the eastern US is 2.483×1012 m2), a decrease of 4115 kg from 2008 to 2011 emissions was converted to a total decrease of 200 ppqv over all days of the three years and averaged at a decreasing rate of 0.2 ppqv d-1, and an increase of 575 kg from 2011 to 2014 emissions was converted to a rate of ∼ 0.03 ppqv d-1. The potential change in NYC atmospheric concentrations was estimated to be ∼ 3 ppqv d-1 from the 2008–2011 NYC emission increases alone and ∼ 6 ppqv d-1 from the 2011–2014 increase. The potential changes in ambient concentrations caused by the regional emission decrease–increase were negligible compared to the observed interannual difference. Those possibly caused by NYC emission increases could be significant but appeared to be inconsistent with the changes in ambient concentrations in two ways. First, from 2010 to 2011 NYC emissions increased and yet summertime ambient concentrations decreased by 10 ppqv throughout the averaged seasonal diurnal cycle (Fig. 4). Second, if the residence time of emitted GEM was 1 day, the total increase in ambient mixing ratio would be 6 ppqv d-1 due to anthropogenic emission increases and would be even smaller, spreading throughout the day, which was negligible compared to the ∼ 60 ppqv increase observed in the spring 2011 average seasonal diurnal cycle compared to the spring 2014 one (Fig. 4). The contribution from the NYC anthropogenic emissions to ambient GEM was further demonstrated using the HYSPLIT simulations in Sect. 5.

Legacy and natural emissions could be another driver for the observed interannual variations in GEM. However, seasonal mean temperature and GEM from the Bronx location were not found to be correlated, which suggested that the effect of changes in legacy and natural emissions on ambient GEM might not be dominant. In addition, using the estimated annual natural and re-emissions of 9.4 to 13.0 µg m-2 from Zhang et al. (2016) for the Bronx site during 2009–2014, the maximum year-to-year change was calculated to be ∼ 1 ppqv d-1, assuming an average planetary boundary layer height of 1000 m. This change alone could not explain the observed interannual variations. It alludes to the potential effect of regional legacy and natural emissions as well as chemistry, which needs to employ modeling tools and is beyond the scope of this study. Here, it was hypothesized that atmospheric circulation was one predominant factor contributing to the observed interannual variation in ambient Bronx GEM concentrations.

To validate this hypothesis, circulation patterns were examined first using the Bronx site wind data. In the falls of 2008, 2009, 2011, and 2013, wind came from all four quadrants, with comparable frequency ranging from 15 to 30 % of the season, whereas in fall 2010 the northwesterly (270–360∘) was more frequent (37 %), and the northeasterly (0–90∘) became predominant (∼ 50–74 %) in the falls of 2012 and 2014 (Fig. 5a). The winters experienced northwesterly winds (270–360∘) more often ranging from ∼ 40 to 65 % of the season, with the exception of winter 2015 when a little below 40 % of the season experienced northwesterly winds, on par with southwesterly wind (180–270∘) (Fig. 5a). Wind speed was averaged seasonally for the four wind quadrants (Fig. 5b). Northwesterly wind (270–360∘) was the strongest (> 3 m s-1) in the winters and springs of 2009–2013 and was reduced to ∼ 2 m s-1 in 2014 and 2015. Southwesterly (180–270∘) wind hovered around 2 m s-1 in the cool seasons, except 2009–2010 when it decreased to 1 m s-1. Wind speed in the two easterly quadrants (0–180∘) varied comparably from 1 to 2 m s-1, except in spring 2013 when it reached 2.5 m s-1 and was particularly low (0.5 m s-1) in the 2014 cool season.

Since anthropogenic Hg sources are mostly concentrated to the west, southwest, south, and northeast of the Bronx site with much fewer sources to the northwest (Fig. 1), GEM mixing ratios would vary expectedly corresponding to air masses arriving from different directions. This was clearly suggested by mixing ratios of GEM averaged seasonally for the four wind quadrants (Fig. 5c). Generally, seasonally averaged GEM mixing ratios were larger by ∼ 20–50 ppqv in the two southerly quadrants than those in the northerly quadrants. Overall, in addition to local emissions, interannual variability in the origin of the air masses reaching the Bronx appeared to cast significant influence on the ambient concentrations of GEM in the city. This argument was strongly supported by SO2 values in the four wind quadrants (Fig. 5d). Consistent with GEM (Fig. 5c), southwesterly (180–270∘) wind brought in air masses with the highest SO2 levels in 2008–2011, especially in winter, reaching 13–14 ppbv, followed by half the values in the winters of 2012–2015. In contrast, the SO2 mixing ratios were close in the other three wind quadrants. One difference between the variation patterns of GEM and SO2 in the four wind quadrants was that air masses from the southeast appeared to also be rich in GEM, whereas SO2 in air from the southeast was low, close to that from the northwest and northeast. One confounding factor for this difference could be due to the ocean being a major source of GEM, and also the only landmass southeast of the Bronx is Long Island, with limited major polluters.

Two cases, the lowest percentile values in the 2009–2010 cool season and the highest in 2014–2015 were used to elaborate on this point. What was most striking about the 2009–2010 cool season was the very low frequency (14 %) of wind from the southwesterly quadrant (180–270∘) in fall 2009 and the largest frequency of wind from the northwesterly quadrant (67 %, 270–360∘) in winter 2010 combined with nearly the lowest wind speed (≤ 1 m s-1) in the three quadrants (0–270∘) (Fig. 5a–c). This indicates that the particularly low mixing ratios in the cool season of 2009–2010 were likely caused by an influx of relatively cleaner Canadian air masses that were over 4 times more frequent and the slowest southerly flow of more polluted air.

The second case is winter 2014 when GEM averaged in the four wind quadrants reached the maximums of all time (Fig. 5c). Coincidently the frequency of wind from the northwesterly quadrant (270–360∘) was nearly the lowest of all cool seasons, barely reaching 40 % of the season compared to up to 67 % in winter 2010 (Fig. 5a). Meanwhile, the frequency of wind from the southwesterly quadrant (180–270∘) reached a high of 34 % of all cool seasons, and the wind speed of ∼ 2 m s-1 was comparable to the northwesterly quadrant. This is a strong indication of the arrival of air masses rich in GEM originating from the heavy emitters in the northeastern US urban corridor via flow nearly as frequent and as fast as the relatively clean northwesterly quadrant. Winter 2015 showed similar wind patterns, also coinciding with high GEM concentrations.

Such variations in wind direction and speed at the Bronx site can be better understood in the context of large-scale circulation. The climatological 500 hPa geopotential height (GPH) (1980–2010) for cool seasons during 1980–2010 exhibited the US East Coast trough centered over coastal southeastern Canada extending southwestward over the eastern US (Fig. 6a). All cool seasons experienced variations in this pattern, except the cool seasons of 2009–2010 and 2013–2014, which appeared to be anomalous (Fig. 6b and c). Specifically, the trough in winter 2010 shifted eastward farthest out over the ocean and was the weakest, evidenced in the maximum TAI (62∘ W) and nearly the least negative TII value (-80 m) (Fig. 6d). In contrast, the trough in winter 2014 was situated the farthest over land and had the strongest of all winters, backed by the most negative TAI (85∘ W) and nearly the most negative TII value (-201 m) (Fig. 6d) . This suggested that in winter 2010 the northeast US was most frequently under the influence of air masses from higher latitudes via flow on the backside of the US East Coast trough, whereas this was much less prevalent in winter 2014 due to the eastern US in winter 2014 being positioned near the axis in the front of the trough. This was further clearly reflected in the maps of sea level pressure (SLP) for the two winters. The unusual winter 2010 circulation was signified by northerly gradient flow (Fig. 6f) from the backside of the Icelandic Low, which shifted toward the south and west near Newfoundland compared to its 1980–2010 climatological position right between and below Greenland and Iceland (Fig. 6e). This indicated predominant transport of relatively clean air from Canada combined with strong ventilation of continental pollution, likely leading to the least polluted air in winter 2010 of all seven winters. In contrast, in winter 2014 NYC appeared to be on the periphery of high-pressure systems in predominantly slow northwesterly and southwesterly flow regimes (Fig. 6g). This explains the least frequent, lowest wind speed in the easterly wind quadrants during winter 2014 (Fig. 5b), which is conducive to regional buildup of air pollution, resulting in the highest mixing ratios of GEM of all winters. More evidence was shown in Sect. 6 using modeled contributions to NYC ambient concentrations from local versus regional anthropogenic sources.

Correlations between Hg and several tracers (e.g., CO, SO2, and NO2) have been commonly used to identify Hg anthropogenic sources, source–receptor relationships, and/or emission ratios. The linear correlation between CO and GEM, especially in winter, in rural locations despite their different sources, reflects their emission ratios in regionally well-mixed air masses (e.g., Mao et al., 2008). At the Bronx site, seasonal GEM and CO values were found to be correlated with r up to 0.69 (p∼ 0) in all seasons during 2008–2013, indicating significant, year-round regional influence, and the two were notably not or minimally correlated in all the seasons from winter 2014 through spring 2015. During 2008–2010 and 2012 r2 values of GEM–CO were larger in warm than in cold seasons, with the maximums exceeding 0.40, and r2 values remained high from winter 2011 through winter 2012 (Fig. 9). The slope value varied from the smallest (∼ 0.02–0.03 ppqv ppbv-1) in winters of 2009–2010 to the largest (0.21 ppqv ppbv-1) in summer 2012 (Fig. 9), with the largest higher than the upper end of the range, 0.06–0.14 ppqv ppbv-1, from rural southern New Hampshire during winters 2004–2007 (Mao et al., 2008). This was greatly different from the GEM–CO correlation in rural southern New Hampshire in winter only due to confounding factors such as legacy emissions and wet deposition in summer (Mao et al., 2008; Lombard et al., 2011). The Bronx experiencing more significant GEM–CO correlation in warm seasons indicated better regionally mixed air masses, influenced predominantly by anthropogenic emissions, than in cool seasons. This is consistent with the cool and warm seasonal circulation patterns as discussed in Sect. 3.2 and 3.3, which is that in warm seasons NYC was predominantly under the influence of the subtropical high conducive to regional mixing and buildup of pollutants.

No correlation between GEM and CO during 2014–2015 could be due in part to the more dramatic emission reductions in CO than changes in GEM in the eastern US. The high percentile values of CO at the Bronx site had been affected by anthropogenic emission reductions over the years, while the 10th and 25th percentile values (referred to as baseline CO in the literature) remained fairly constant in all seasons (Fig. S5). Zhou et al. (2017) suggested that baseline CO in rural northeastern US areas was controlled by a multitude of factors, including global biomass emissions, large-scale circulation, and cyclone activity. At the Bronx site, the low percentile value, close to regional baseline levels, was possibly determined by a range of factors, whose importance could have varied from year to year.

Unlike previous studies (e.g., Jen et al., 2013; Choi et al., 2013), GEM at the Bronx site was found to be poorly correlated with SO2 while somewhat to moderately correlated with NO2 (r= 0.13–0.71, p< 0.0001) (Table 2), despite abundant sources co-emitting GEM, SO2, and NO2 locally and upwind. In addition to different lifetimes, different magnitude and timing of emission reduction implementations and source types of the three compounds could have affected their relation. Total Hg anthropogenic emissions in NYC were increased by 16 % from 2008 to 2011, mainly in miscellaneous nonindustrial not elsewhere classified (NEC) and waste disposal emissions, and further increased by 37 % from 2011 to 2014 primarily in fuel combustion. As aforementioned, emissions of Hg in the eastern US decreased by 13 % from 2008 to 2011 and increased by 2 % from 2011 to 2014. In contrast, total SO2 emissions in NYC decreased steadily by 30 % from 2008 to 2011 followed by a further decrease of 43 % to 2014, while over the eastern US total SO2 emissions decreased by 48 % from 2008 to 2011, furthered by another 29 % decrease in 2014. Specifically, NYC launched the Clean Heat program in winter 2008–2009 resulting in a 69 % decrease in SO2 concentrations averaged over the city-wide street-level monitoring sites in winter 2012–2013 (NYC Health, 2013; https://www1.nyc.gov/assets/doh/downloads/pdf/environmental/air-quality-report-2013.pdf). The Bronx data also reflected the effect of such emission reductions, with a 58 % decrease in the seasonal median mixing ratio of SO2 from 9.2 ppbv in winter 2009 to 2.8 ppbv in winter 2015 (Fig. S6). As for NO2, fuel and mobile combustion emissions comprised > 99.5 % of the total NOx emissions in NYC and ∼ 90 % over the eastern US. NYC NOx emissions changed insignificantly (1 %) from 2008 to 2011 and by 15 % from 2011 to 2014, while eastern US mobile and fuel combustion emissions were decreased by 16 and 33 %, respectively, from 2008 to 2011, and further decreased by 13 and 9 %, respectively, to 2014. These varying changes possibly contributed to confounding the emission signature of GEM vs. NOx and altered that of GEM vs. SO2.

The effect of local emissions can be accentuated by the correlation between GEM and SO2 and between GEM and NO2 for the SO2 and NO2 mixing ratios, exceeding their respective seasonal 95th percentile concentrations. However, nearly no correlation between GEM and SO2 or between GEM and NO2 was found in this subset of data (Table 2). It should therefore be cautioned that tracer correlation could not be used to identify source types of GEM or estimate emission ratios of GEM to SO2 or NO2 in NYC.

HYSPLIT dispersion simulations were used to obtain a quantitative comparison of the effects of sources outside and inside NYC on NYC ambient concentrations of GEM. As stated in Sect. 2.3, the modeling domain extended westward to OH and southward to northern VA and northward to include New England (Fig. 10a), with a total of 522 counties reporting Hg emissions. As described earlier, two scenarios were designed for model simulations:

a scenario with the emission sources in all 522 counties within the domain

Simulations of scenario no. 2 quantify the contribution of sources outside of NYC to Hg concentrations in NYC, and the difference in the concentrations in NYC between the two scenario s quantifies the contribution of NYC local sources to Hg concentrations in NYC.

Shown in Fig. 10b is the contribution, in percentage of the total contribution from all anthropogenic emissions in the domain, to NYC ambient concentrations of GEM from anthropogenic emissions alone from local sources, and in Fig. 10c is the contribution of emissions from regional anthropogenic sources. There was clearly interannual variability in the contribution of local versus regional anthropogenic sources. Local emissions averaged a contribution of 25 % in all winters of 2009–2015, with the period minimum of 17 % in winter 2011 and the maximum of 33 % in winter 2013 (Fig. 10b). Conversely, the contribution of regional sources averaged a contribution of 75 % in all winters, with the largest 83 % in winter 2011 and the lowest 67 % in winter 2013 (Fig. 10c). Compared to contributions in the winter of the same year, those from local sources were larger (by up to 12 % in 2009) in the summers of 2009, 2011, 2012, and 2014, close in summer 2010, and 10 % smaller in summer 2013 (Fig. 10b).

A close examination revealed largely consistent relation between NYC GEM mixing ratios and source contributions. As suggested in Sect. 3.2, the Bronx experienced the lowest concentrations of GEM in all percentile values in winter 2010, and yet, interestingly, the simulated local contribution in winter 2010 was in the midrange of the seven winters. This indicates that the particularly low background concentration in the sweeping northerly flow led to less regional contribution to NYC Hg concentrations than regional sources would in other years. In contrast, winter 2014 saw the highest 25th, 50th, 75th, and 90th percentile concentrations of GEM, and yet the contribution of local sources (∼ 22 %) was not even higher than average (25 %). As aforementioned, in winter 2014 the eastern US was most likely under the least dynamic conditions conducive to regional buildup of air pollution, which resulted in a higher-than-average contribution from regional sources and conversely lower-than-average contribution from local sources (Fig. 10c). Consistent with GEM, the lower percentile mixing ratios of CO, SO2, and NO2 appeared to be elevated or stopped decreasing compared to those in the previous year (Figs. S5, S6).

The HYSPLIT dispersion model simulations suggested that close to three-quarters of the anthropogenically induced concentration of GEM in NYC was from regional sources. It should be pointed out that other factors and/or processes, such as legacy and natural emissions, deposition, meteorology, and/or large-scale circulation, might have competed with the effect of anthropogenic emission reductions. Nearly 90 % of the model simulation domain is covered by vegetation. SMOKE model output in Ye et al. (2017) suggested that the ratio of anthropogenic to legacy and natural emissions was 0.3 over the domain. Legacy and natural emissions could become dominant under warmer and wetter conditions in summer. Moreover, Hg deposition could be impacted by changes in physical parameters such as light, temperature, and plant species (Rutter et al., 2011). Indeed, changes of -30 to 50 % in Hg deposition were simulated for the eastern US from the 2000s to the 2050s due to changes in precipitation (Megaritis et al., 2014). Net GEM surface emissions were estimated to be dominant in summer and net dry deposition was estimated to be dominant in other seasons at the majority of AMNet monitoring sites in eastern North America (Zhang et al., 2016). Since Hg deposition and legacy emissions are closely linked, these studies indicate potential changes in legacy emissions in response to variable meteorological conditions and changing climate with subsequent effects on atmospheric Hg concentrations. Therefore, with legacy and natural emissions accounted for, regional contributions to NYC ambient Hg concentrations would be even more dominant.

Caution needs to be taken in interpreting the model results due to the limitation of the modeling exercise. First, to save computational time, the simulation domain used in this study was smaller than is ideal, and thus with a larger regional domain, the significance of regional anthropogenic sources could be enhanced. Second, the HYSPLIT dispersion model accounts only for long-range transport of a pollutant from sources within the domain, without considering chemical transformation, gas-to-particle partitioning, atmosphere–surface exchange of mercury, loss through deposition, and background concentrations. This being said, for a compound such as GEM with a lifetime of 6–12 months, dispersion model simulations would be adequate for providing relative contributions of regional and local sources to ambient concentrations at a location of interest in continental midlatitudes.