Passive air samplers (PASs) for gaseous mercury
(Hg) were deployed for time periods between 1 month and 1 year at 20 sites
across the globe with continuous atmospheric Hg monitoring using active
Tekran instruments. The purpose was to evaluate the accuracy of the PAS
vis-à-vis the industry standard active instruments and to determine a
sampling rate (SR; the volume of air stripped of gaseous Hg per unit of time)
that is applicable across a wide range of conditions. The sites spanned a
wide range of latitudes, altitudes, meteorological conditions, and gaseous
Hg concentrations. Precision, based on 378 replicated deployments performed
by numerous personnel at multiple sites, is 3. Subscripted numbers are not significant,
but are reported to reduce rounding errors in subsequent studies (see
Sect. 2.3 for details).
Article 19 of the Minamata Convention requests that “parties endeavour to cooperate to develop and improve … geographically representative monitoring of mercury (Hg) levels in environmental media … [and gain] … information on the environmental cycle, transport (including long-range transport and deposition), transformation and fate of mercury and mercury compounds” (UNEP, 2013). Given the atmosphere represents the primary pathway for the global distribution of mercury (Schroeder and Munthe, 1998; Selin, 2009; Driscoll et al., 2013), highly accurate and precise atmospheric monitoring of Hg is paramount in attaining these goals. Existing atmospheric Hg monitoring networks, such as the Global Mercury Observation System (GMOS), the Atmospheric Monitoring Network (AMNet), and the Environment and Climate Change Canada Atmospheric Mercury Monitoring (ECCC-AMM) network, have greatly improved the understanding of atmospheric Hg (Li and Lee, 2014) and the ability to develop and evaluate global atmospheric distribution models (Lin et al., 2006) such as GEOS-Chem, GLEMOS, GEM-MACH-Hg, and ECHMERITRADM (Travnikov et al., 2017). However, the spatial scope of these networks is limited, especially in the southern hemisphere (Li and Lee, 2014), leading to considerable gaps in the understanding of atmospheric Hg cycling. It is widely acknowledged that further network expansion will be required (Pirrone et al., 2013; Sprovieri et al., 2017).
The principal constraints limiting the spatial expansion of atmospheric Hg measurements are high costs and dependence on electricity, compressed gases, and technical training (McLagan et al., 2016a; Pirrone et al., 2013; Huang et al., 2014). Passive air samplers (PASs) operate without these constraints and have the potential to complement existing approaches and greatly improve the spatial resolution of measurements (McLagan et al., 2016a; Pirrone et al., 2013; Huang et al., 2014). One successful example of such a combined approach is the Global Atmospheric Passive Sampling (GAPS) network for persistent organic pollutant (POP) monitoring. Shortly after the implementation of the Stockholm Convention on POPs, the GAPS network was established to “complement high-volume, active air sampling activities in assessing the presence of POPs in the atmosphere and in evaluating their global distribution and long-range transport” (Pozo et al., 2006). The network now includes over 40 sites across all seven continents (Pozo et al., 2006; Shunthirasingham et al., 2010; Herkert et al., 2018).
McLagan et al. (2016b) calibrated a PAS for measuring gaseous Hg in the atmosphere that utilizes a Radiello® diffusive barrier and a sulfur-impregnated activated carbon sorbent. The sampler's replicate precision (2 %) is excellent (McLagan et al., 2016b). Also the variability of its sampling rate (SR; volume of air effectively stripped of gaseous Hg per unit time) caused by meteorological parameters is small, increasing slightly with both temperature and wind speed across ranges relevant to outdoor deployments (McLagan et al., 2017b). However, testing thus far has been restricted to deployments in laboratory settings and at only one outdoor location. To better understand the PAS's overall uncertainty, its performance under variable geographical, meteorological, and gaseous Hg concentrations must be evaluated.
In this study, we seek to assess the accuracy of the McLagan et al. (2016b) PAS by comparing ambient Hg concentrations derived from the PAS to those measured with established active sampling techniques. PASs were deployed at sites with ongoing active monitoring instruments in North America, Asia, Australia, and Europe. These sites cover a wide range of meteorological conditions and some variation in gaseous Hg concentrations. In addition to quantifying the overall uncertainty of the PAS, this study also allowed for the refinement of the previously calibrated SR using a much larger pool of data. Furthermore, some of the selected sites recorded the speciation of atmospheric Hg, making it possible to investigate whether gaseous oxidized Hg (GOM) is being taken up by the PAS.
The PAS used in this study has been described in detail by McLagan et al. (2016b). Briefly, a stainless-steel mesh cylinder filled with sulfur-impregnated activated carbon (HGR-AC; Calgon Carbon Corporation) is placed inside a white Radiello® diffusive barrier (Sigma Aldrich). This barrier is protected from wind and precipitation by attachment to the inside of a protective shield, which also serves as a storage and shipping container. The sampler works by diffusive uptake and accumulation of gaseous Hg onto the sorbent. Following deployment, the sampler is retrieved, the sorbent contents are analyzed on an automated thermal combustion atomic absorbance instrument (McLagan et al., 2016b), and the time-averaged gaseous Hg concentration is calculated using a previously calibrated sampling rate (see Sect. 2.5).
The term gaseous Hg is used to describe the sorbed analyte because it has not been confirmed whether this PAS takes up gaseous elemental Hg (GEM) or TGM (total gaseous Hg; both GEM and GOM). It is, however, unlikely that the highly reactive nature of GOM allows it to pass through the pores of the diffusive barrier. The most recent modelling estimations suggest that the “effective lifetime” of GEM is around 6 months (Corbitt et al., 2011; Horowitz et al., 2017). The atmospheric lifetime of GOM due to reduction and deposition is on the order of days to weeks (Ariya et al., 2015; Horowitz et al., 2017; Shah et al., 2016). Although uncertainties remain, the shorter atmospheric lifetime and higher deposition fluxes of GOM translate to GEM making up the majority of TGM in most places (typically > 95 %; Cole et al., 2014; Driscoll et al., 2013; Rutter et al., 2009; Slemr et al., 2015). As such any uncertainty related to the uptake of GOM by the PASs is likely small.
Sampling sites for passive air sampler accuracy testing. Sampling sites are coloured by intensity of PAS deployments: high (monthly, seasonal, half yearly, and yearly deployments), medium (seasonal, half yearly, and yearly deployments), low (half yearly and yearly deployments), and very low (yearly deployments).
For this study, PASs and standard operating procedures (see Sect. S1 in the Supplement) were sent from Toronto to 20 sampling sites on four continents (Fig. 1) using Canada Post or international couriers. The PAS were deployed for a period of 1 year at each site during the time between 2015 and 2017. Accessibility of AMNet and ECCC-AMM networks resulted in a greater number of sites in North America than in other global regions. Temporal resolution of sampling ranged from monthly, quarterly, biannual to annual deployments. The number of deployments varied between sites. Four sites with a deployment intensity categorized as “high” were sampled with monthly, quarterly, biannual, and annual resolution. Six “moderate” deployment intensity sites had quarterly, biannual, and annual resolution. Seven “low” deployment intensity sites had biannual and annual resolution and three “very low” deployment intensity sites had year-long deployments. Exact numbers, lengths, and dates of deployments at each site are shown in Tables S2.1, S2.2, and S2.3 in the Supplement. After deployments, PASs were stored at each site until the last PAS had been retrieved, at which point they were returned to Toronto by courier for analysis of the sorbent content. In total, there were 142 triplicated deployments (426 samplers) across all sites. Field blanks were obtained by transporting PASs to each site, removing the Teflon tape and solid cap, attaching an open cap with mesh screen, holding it up to a deployment position for 10 s, and then immediately taking it down, closing, and sealing it as described above. Temperature and wind speed for each deployment period, as recorded by weather stations at or near the sampling sites, are listed in Table S2.3.
TGM and GEM concentrations were measured with Tekran 2537 series cold-vapour
atomic fluorescence spectrometer (CVAFS) systems. Details of instrument
setup are given in Landis et al. (2002), Steffen et al. (2008), Cole et al. (2013), and NADP (2015).
Concisely, the instrument consists of a 2
Tekran 1130 and 1135 speciation units paired with a Tekran 2537 CVAFS were
deployed at Alert, Mauna Loa, Salt Lake City, Beltsville, and Grand Bay.
Again, full details are provided elsewhere (Landis et al., 2002; Steffen et
al., 2008; Cole et al., 2013). In brief, ambient air enters these systems
through an impactor inlet (to remove particles > 2.5
Information on the percentage data coverage across the deployment periods of the active measurement systems and PASs (successful samples/deployments out of actual samples/deployments) is shown in Table S2.2. The 5 min TGM and GEM data from the Tekran 2537 series instruments or the Tekran 2537 component of the speciation units were compared with the concentrations derived from the PASs. For Alert station, which operates both a Tekran 2537X sampler and a Tekran 1130 or 1135 speciation unit, data from the former were used for comparison.
Given that TGM is generally dominated by GEM at most sites (typically > 95 %; Cole et al., 2014; Driscoll et al., 2013; Rutter et al., 2009; Slemr et al., 2015), differences between the two concentrations are likely to be small. To ensure consistency with the nomenclature used for the PASs, the analyte sampled by non-speciating Tekran 2537 instruments is referred to as gaseous Hg. The PAS-derived concentrations were compared with the speciated data available for Alert, Mauna Loa, Salt Lake City, Beltsville, and Grand Bay to ascertain whether the PASs accumulate GOM.
Mauna Loa (3397 m a.s.l.) and Mt. Lulin (2862 m a.s.l.) are high-altitude
sites and required active concentrations to be adjusted for pressure. At Mt.
Lulin, to obtain a volumetric flow rate of 10 L min
The uncertainty of all reported values is given by one standard deviation.
The standard errors of regression coefficients were also converted to
standard deviations. Uncertainties are reported to one significant digit
unless the first non-zero digit is a 1, in which case there are two
significant digits (e.g. 5.43
All PASs were analyzed at the University of Toronto Scarborough for the
total mass of Hg using an AMA254 (Leco Instruments Ltd.) by means of thermal
combustion, amalgamation, and atomic absorption spectroscopy in pure oxygen
carrier gas (USEPA method 7473; USEPA, 2007). To prevent sulfur poisoning
of the instrument's catalyst by the high sulfur sorbent (S
The instrument was calibrated via the addition of diluted Hg liquid standard
(1000
Analytical blanks, i.e. clean (unexposed) HGR-AC sorbent, were analyzed
regularly and had a mean concentration of 0.30
Gaseous Hg concentrations in the atmosphere,
The second set of air concentrations was calculated by using a recalibrated SR
obtained from the data generated in this study plus the data from the
original calibration experiment (hence termed recalibrated SR). This recalibrated SR was calculated
using the slope method as described by Restrepo et al. (2015) and McLagan et al. (2016b). Rearranging Eq. (1), we can derive a SR for individual deployments
from the mass of sorbed Hg,
The third set of air concentrations was calculated using the recalibrated SR with
adjustments for the mean temperature,
The three sets of air concentrations derived from the PAS,
Mean passive and active air sampler data at each sampling
location.
Few PAS samples were lost during deployment, transport, or analysis. Reasons
for losses were poorly sealed samplers, errors in recording of deployment
time and dates, loss during analysis (e.g. catalyst failure), and a hail
storm (Hunter Valley site). Of 142 triplicated PAS deployments, 93 % (132 deployments, 378 samplers) and 89 % (129 deployments, 375 samplers) had
at least one and two successfully analyzed PAS, respectively. The
precision-based uncertainty of the PAS calculated from the successful
replications was 3.
Uptake curves and individual deployments of passive air samplers across all 20 sampling locations. 0 ng points mark the beginning of deployments. All samples on the same line are for deployments that began at the same time. All axes are scaled the same. Deployments of the same colour cover equivalent deployments periods (i.e. orange is 7–12-month deployment at all sites).
The amount of mercury quantified in each individual sampler is plotted against the deployment time in Fig. 2. These uptake curves are highly linear over 12 months at all sites, confirming that the PASs do not approach a limit to their uptake capacity throughout all deployments. At Xiamen and Ningbo, sites with the highest uptake rates, i.e. the mass of Hg sorbed per unit time, and hence the highest ambient gaseous Hg concentrations (Table 1), 144 and 133 ng of Hg, respectively, were taken up over a 12-month period. This is almost double the mass of Hg taken up in the original uptake experiment (McLagan et al., 2016b) and indicates the maximum deployment time of the sampler is at least 1 year even under the elevated concentrations observed in East Asia. The high uptake capacity of the HGR-AC sorbent for gaseous Hg is due to a large surface area to volume ratio and the affinity of gaseous Hg to the impregnated sulfur (McLagan et al., 2016a; Suresh Kumar Reddy et al., 2013; Zhang et al., 2012).
Uptake rates can be derived from the slopes of the curves in Fig. 2. Uptake
rates of PASs deployed at the same site (comparing slopes within each panel
in Fig. 2) are very uniform. The greatest relative variability in uptake
rate between samples at any one site occurred at Alert (Table 1) and was low
(< 25 % RSD; Table 1). This attests to the stability of both (i) the SR and (ii) the gaseous Hg concentrations at each site over the length of
the PAS deployments (1 month or longer). The time-averaged nature of the
concentrations measured by the PASs conceals much of the variability that
generally occurs at shorter time resolution. The higher variability in
uptake rates and gaseous Hg concentrations at Alert and Ningbo (Table 1) can
be attributed to springtime atmospheric Hg depletion events (Steffen et
al., 2008) and seasonal variability in the elevated East Asian background
concentrations, respectively. The differences in uptake rates between sites
(comparing slopes among different panels in Fig. 2) were caused by different
gaseous Hg concentrations at each location as evidenced by the significant
correlation between uptake rate and active gaseous Hg concentration data (
The Tekran 2537 active monitoring instruments successfully recorded Hg concentrations during 59 % of all deployment periods. Data gaps were caused by instrument failures, power outages, or removal of poor-quality data during RDMQ™ processing. The PASs covered a greater percentage of the deployment periods across all sites, indicating its high reliability and ease of use. A comparison of active and passive sampling was considered meaningful if an active instrument had recorded data during at least a quarter of a PAS's deployment time. This was the case in 113 of 142 deployments (80 %). 107 of these 113 deployments (306 samplers) had at least one successful PAS data point and therefore were used in the comparison of actively and passively derived gaseous Hg concentrations (Table 1). Because the sites in Sydney, Hunter Valley, and Xiamen did not have adequate data coverage for any of the deployments, the comparison included 17 of the 20 total sites.
Comparison of active (
The gaseous Hg concentrations calculated from the PAS using the original SR of
0.121
The recalibrated SR based on 343 passive samplers deployed at 17 sites with collocated active samplers (37 samples from McLagan et al., 2016a, and 306 samples
from this study) was 0.135
McLagan et al. (2016b) theoretically estimated the SR of the PAS based on the
molecular diffusivity of elemental Hg and an estimated effective diffusion
distance. Using an air-side boundary layer thickness of 15 mm, which has
been recommended for outdoor deployments with the protective shield (McLagan
et al., 2016b) and the mean temperature across all deployments in this study
(9.8
When the recalibrated SR is used for the derivation of air concentrations from the PAS, the
MND compared to active instrument-derived concentrations is significantly
(
Additionally, deployment length had no significant effect on the MND of
samples for either the recalibrated SR (
It is important to acknowledge that this is not a fully independent
evaluation of the performance of the sampler, as the same data were used in
the derivation of the recalibrated SR and in the determination of the air concentrations
from the PAS. However, we stress that data from all sites were used to
derive a single recalibrated SR that was used for all sites; i.e. the fitting involved was
not site-specific. For example, the Little Fox Lake site contributed the
most data points to the recalibration (
Our precision estimate (3.
In summary, we judge the overall uncertainty of the PAS to be higher than
the precision based uncertainty (3.
The performance of the PAS using either the recalibrated or adjusted SR represents a substantial
improvement over all existing gaseous Hg PAS designs to date, especially
those with sufficiently low detection limits to monitor background
concentrations (as summarized in a review on gaseous Hg PASs by McLagan et
al., 2016a). While the accuracy-based uncertainty of the 3M PAS by McCammon
and Woodfin (McCammon and Woodfin, 1977) was similar (8
Plots comparing active instrument-derived gaseous Hg concentrations with passive concentrations determined using the original, recalibrated, and adjusted SRs for each sampling site are presented in Sect. S4 (Figs. S4.1–S4.17). The data in Fig. 3 are colour coded by site categorization (urban, rural, altitude, northern/Arctic).
Of the 20 sampling locations from the current study, five were classified as
urban (Xiamen, Ningbo, Salt Lake City, New York City, and Sydney).
Additionally, the previous calibration study site in Toronto was included in
recalibrations and uncertainty assessments. Overall, there was good
agreement between active and passive concentrations using the recalibrated (MND:
8.
Salt Lake City may experience elevated GOM concentrations and atmospheric Hg
reactivity in general due to the increased presence of atmospheric
halogenated species in the atmosphere around the Great Salt Lake (Gay et
al., 2013; Peterson and Gustin, 2008; Stutz et al., 2002). Only one of the
seven deployments in Salt Lake City had sufficient actively measured
speciation data for comparison. The mean GOM concentration for that
deployment was 0.014 ng m
Concentrations derived from PASs deployed in New York City using
recalibrated (8.
Good agreement between active and passive concentrations both at typical
hemispheric and elevated East Asian background concentrations using either
the recalibrated or adjusted SR (uncertainties not significantly different;
Eleven sites (Beltsville, Put-in-Bay, Grand Bay, Kejimkujik, Ucluelet, St.
Anicet, Egbert, Waldhof, Hunter Valley, Cape Grim, and Gunn Point) were
located in rural settings. Put-In-Bay is situated on South Bass Island on the Ohio (USA) side of Lake Erie. Temperatures at the site were similar to the mean of all
sites, but wind speeds were high. Nonetheless, differences from active
concentrations were small using either recalibrated or adjusted SR, i.e. were not significantly
reduced by adjusting the SR for wind speed and temperature (
At both Beltsville and Grand Bay, the proportion of GOM in TGM measurements was < 1 % for all deployments. Hence, no information on the sampled analyte could be derived from these data.
The two high-altitude sampling sites at Mt. Lulin and Mauna Loa provided a
unique opportunity to not only examine the PAS's functionality under
relatively low atmospheric pressure conditions, but also to test its
performance in a likely more dynamic zone of atmospheric chemistry at or
above the planetary boundary layer (Bieser et al., 2017; Carbone et al.,
2016). Lower atmospheric pressure has the potential to affect the PAS in two
ways: (i) increasing the SR because diffusivity coefficients are inversely
proportional to pressure (Armitage et al., 2013; Klánová et al.,
2008; Seethapathy et al., 2008) and (ii) decreasing the SR as there is less
mass per volume of air. While the use of mixing ratios would prevent the
latter effects, they are not the preferred method of reporting atmospheric
Hg measurements (Weigelt et al., 2016). These two effects should
theoretically cancel each other out, and hence we would expect passive
concentrations to align with pressure-adjusted active concentrations. At Mt.
Lulin the MNDs were low for both the recalibrated (5.3
Active GOM concentrations measured by the Tekran speciation unit were
elevated at Mauna Loa and made up 9.
Two sites from Canada's north were included in the study: Alert (Fig. S4.16), in the high Arctic, and Little Fox Lake (Fig. S4.11), which is north
of Whitehorse in the Yukon territory. Both sites had high temporal
resolution data, except for the first 4 months of sampling (October to
February) at Alert, when PAS data were lost due to poorly sealed samplers
and contaminated field blanks. While wind speeds were moderate and not
excessively variable across deployments at either site, the mean
temperatures of each deployment ranged over 27.4 K at Little Fox Lake and
20.5 K at Alert (Table S2.3). Despite the larger range of temperatures at
these sites, mean temperatures across all deployments were not excessively
low (Alert: 5.9
The Alert site also employed a Tekran speciation system. The mean GOM
concentrations at Alert across the different PAS deployments represented
4.
From this much larger data set of collocated active and passive measurements
of gaseous Hg we were able to revise our original SR, which we determined to be
overestimating concentrations, to a recalibrated SR of 0.135
This study also attempted to address the exact nature of the analyte being taken up by the sampler. At Mauna Loa, where overall GOM made up the greatest proportion of TGM, PAS results were significantly improved when using active GEM data over TGM data, which agrees with the hypothesis that GOM is removed by the diffusive barrier. Data at Alert were inconclusive as there were no significant differences with the PAS results using either active data for GEM or TGM for either the atmospheric Hg depletion event period or the whole dataset. While the Mauna Loa data do suggest the PAS is taking up solely GEM, the same results were not apparent at Alert, and hence we cannot yet conclude with certainty that GEM is indeed the sole analyte sorbed by the PASs. Furthermore, in all cases, the proportion of GOM (TGM minus GEM) in TGM measurements was close to the level of PAS uncertainty, which further reduces the strength of the conclusions that can be drawn. The deployment of samplers in controlled chambers with a point source of GOM or isotopic analysis of the sorbed Hg may yield more definitive findings.
McLagan et al. (2016a) outlined three key rationales behind the development and use of a gaseous Hg PAS: (i) background concentration monitoring, especially at remote sites, (ii) measuring gaseous Hg gradients with high spatial resolution deployments, and (iii) personal exposure sampling. Results of this study indicate the PAS is a highly precise and accurate tool that can complement and even replace existing monitoring techniques in certain circumstances across the three aforementioned rationales. Additionally, their small size, low cost, non-electrical operation, and applicability across a range of conditions ascribe to their versatility and with consideration may unlock a number of additional deployment scenarios that were not previously viable or even considered with only active monitoring instruments.
All data used in calculations are available in the Supplement. Any additional data can be provided upon request to the corresponding author.
Supplement information includes graphical, written, (linked) video standard operating procedures, all concentration data, and active–passive concentration comparisons for each site.
The supplement related to this article is available online at:
Several of the authors have recently signed a licensing agreement between the University of Toronto and a private corporation concerning the sampler used in the present study. It is possible that this could eventually lead to the commercialization of the sampler.
We would like to sincerely thank all the site technicians and members of AMNET, ECCC-AMM Hg monitoring network, and those at sites outside of these networks that assisted in the deployment and collection of the PASs and the retrieval and upkeep of active samplers. These individuals are Dylan Nordin, Rob Tordon, Martin Pilote, Corrine Schiller, Helena Dryfhout-Clark, Kevin Rawlings, Melody Fraser, Matthew Hirsch, Ronald Cole, Justin Chaffin, Andy Hale, Larry Scrapper, Nash Kobayashi, Da-Wei Lin, and JinSheng Chen. We also acknowledge funding from Strategic Project Grant no. 463265-14 by the Natural Sciences and Engineering Research Council of Canada (NSERC) and an NSERC Alexander Graham Bell Canada Graduate Scholarship. Alexandra Steffen acknowledges funding from the Northern Contaminants Program of Indigenous and Northern Affairs Canada for atmospheric Hg monitoring at Alert and Little Fox Lake. Edited by: Aurélien Dommergue Reviewed by: three anonymous referees