Passive air samplers (PASs), which provide time-averaged concentrations of
gaseous mercury over the timescale of weeks to months, are promising for
filling a gap in the monitoring of atmospheric mercury worldwide. Their
usefulness will depend on their ease of use and robustness under field
conditions, their availability and affordability, and most notably, their
ability to provide results of acceptable precision and accuracy. Here we
describe a comparative evaluation of three PASs with respect to their
ability to precisely and accurately record atmospheric background mercury
concentrations at sites in both southern Italy and southern Ontario, Canada. The study includes the CNR-PAS with gold nanoparticles as a sorbent, developed by the Italian National Research Council, the IVL-PAS using an activated carbon-coated disk, developed by the Swedish Environmental Research Institute, and the
Mercury (Hg) is a highly toxic pollutant, which due to its significant adverse impact on ecosystem and human health has been added to the environmental political agenda at national, regional, and global levels. In recent years, the adoption of the Minamata Convention has aimed to protect human health and the environment from Hg releases and emissions (UNEP, 2013). Article 22 of the convention requires parties to formally assess, through the provision of “comparable monitoring data on the presence and movement of mercury and mercury compounds in the environment”, how effective the structure and implementation of the convention is at meeting its primary goal (Article 1). Article 19 of the Convention highlights the importance of environmental monitoring. While such efforts should build on existing monitoring networks (UNEP, 2013), this will also require research and development of monitoring technologies.
The accurate assessment of air pollutants has increasingly come into focus as the need to understand their transport and mechanisms of deposition to ecosystems grows (Dinoi et al., 2017; Moretti et al., 2020; Naccarato et al., 2018, 2020; Tassone et al., 2020). Special attention is given to the atmosphere because it is a well-recognized pathway for Hg distribution throughout various environmental compartments (Driscoll et al., 2013). In this context, many regional atmospheric networks have been operating since the mid-1990s, including the US National Atmospheric Deposition Network-Mercury Deposition Network (NADP-MDN) (Vermette et al., 1995), the Environment and Climate Change Canada Atmospheric Mercury Measurement Network (ECCC-AMM), the Arctic Monitoring and Assessment Programme (AMAP) group of long-term measurements (Arctic Council, 1991), and the European Monitoring and Evaluation Programme (EMEP) (Tørseth et al., 2012). In 2010, the Global Mercury Observation System (GMOS) was created in an attempt to establish a global atmospheric Hg measurement network, integrating EMEP and AMAP, with more than 40 monitoring sites distributed worldwide. Since their beginning, there has been a growing interest in improving global monitoring of Hg by increasing the spatial resolution of gaseous Hg data, especially in remote locations and in developing countries (Pirrone et al., 2013), in order to meet Minamata Convention objectives.
Current methodologies, however, have a limited ability to monitor Hg on a truly global scale. Indeed, the use of active automated sampling systems based on sorbent traps with gold amalgamation, which are desorbed at relatively fine time resolution (3–5 min) for Hg quantification (Brown et al., 2010; Landis et al., 2002; Munthe et al., 2001; Steffen et al., 2012; Wängberg et al., 2001), may be limited by cost and the need for reliable electricity, consumables, and maintenance by well-trained operators (Huang et al., 2014; McLagan et al., 2016b; Pirrone et al., 2013). Given these constraints, passive sampling has been proposed as a viable alternative or supplemental system to fill the gaps in worldwide Hg monitoring. Compared to active sampling instruments, passive air samplers (PASs) for gaseous Hg are relatively inexpensive and thus can be deployed in large numbers, allowing for the identification and characterization of Hg sources through finely resolved spatial mapping (Huang et al., 2014; McLagan et al., 2016b; Pirrone et al., 2013). PASs are also suitable for deployment at remote sites because they require no power supply and are based on the unassisted molecular diffusion of gaseous Hg. Moreover, they are compact, portable, and easy to use. The adoption of PASs raises the very real possibility of a sustainable, long-term global network of atmospheric Hg measurements that includes regions not covered by existing efforts.
Over the past few years, a number of mercury PASs have been developed, each with different materials and geometries (Macagnano et al., 2018; McLagan et al., 2016a; Wängberg et al., 2016). While each sampler has its merit, the performance of different designs has yet to be compared systematically. This remains an impediment for understanding which PASs may be most appropriate for possible adoption in monitoring networks or whether a mix of designs can be reliably employed.
In this paper, for the first time, we report the results of a field-based intercomparison campaign and a controlled, blind performance comparison among different Hg PASs. The performance of three different PASs were evaluated at two monitoring sites, located in Italy and Canada, over a 3-month period. The PASs involved in this study were developed by the Italian Institute of Atmospheric Pollution Research (CNR-IIA) (Macagnano et al., 2018), the Swedish Environmental Research Institute (IVL) (Wängberg et al., 2016), and the University of Toronto (McLagan et al., 2016a). Data were submitted for compilation to a blind third party to control for reporting bias. The PASs were assessed for accuracy through comparison with active sampling data, for precision via replication, and for sensitivity through the assessment of blanks and detection limits, as well as in terms of the linearity of uptake over extended deployment periods.
Characteristics of the three PAS designs included in the comparison are
summarized in Table 1. The CNR-IIA PAS (CNR-PAS) consists of a fibrous
quartz filter coated with sorbent material, which is attached to the bottom
of a borosilicate glass vessel equipped with a double cap system to minimize
operator handling and to avoid contamination that could result from the cap
opening (Macagnano et al., 2018). The IVL-PAS consists of a disk coated with an activated carbon sorbent that is inserted in a badge-type device (Wängberg et al., 2016). The geometry of CNR-PAS and IVL-PAS makes them both axial diffusion samplers.
The
Characteristics of the three passive air samplers for gaseous mercury that were compared in this study.
The study design involved the side-by-side deployment of the three PAS types
in the vicinity of existing active air sampling sites in Rende (Italy) and
Toronto (Canada) during late winter and early spring of 2019. At both sites,
11 overlapping PAS deployment periods ranged in length from 2 to 12 weeks,
whereby each deployment involved triplicate PASs and one unexposed PAS as
a field blank, for a total of 88 PASs of each type. In some cases, additional
storage blanks were taken. Each participating research group supplied their
PASs along with deployment instructions, performed the chemical analysis, and
reported volumetric air concentrations and basic QA/QC results to an
independent third party. Gaseous elemental mercury concentrations recorded
by active air sampling instruments, averaged for the 22 deployment periods,
were reported at the same time. After data submission, only the following
changes were made to the data: a typographical error affecting the assumed
uncertainty of the sampling rate of the IVL-PAS was corrected. The blank
correction for the CNR-PAS was performed using the average of the field
blanks at one location instead of using field blanks specific for a
deployment, in order to be consistent with the blank correction applied for
the other two PASs. Additional submitted results for the
Two existing active air monitoring sites in Italy and Canada were used for
the performance evaluation of the three different PASs. The Italian site was
a monitoring station close to the CNR Institute of Atmospheric Pollution
Research (39
At both sites, gaseous mercury concentrations were obtained at 5 min
intervals using Tekran 2537x and 2537a automated mercury analyzers (Tekran
Instruments Corporation, Toronto, ON, Canada). In Toronto, two systems –
both a Tekran 2537x and a Tekran 2537a (5037 and 0075 units, respectively)
were operated in parallel to quantify the duplicate precision of the active
air sampling technique. These systems collect air onto gold traps, which are
thermally desorbed for quantification of mercury by atomic fluorescence
spectroscopy (Ebinghaus et al., 2011). The sampling was performed with airflow rates of 1.5 and 1.0 L min
Calibration results and data acquisition were quality controlled according to established quality assurance and quality control procedures (QA/QC). The GMOS-Data Quality Management (G-DQM) approach (D'Amore et al., 2015) was used to check the Tekran 2537x mercury concentration data collected at the Rende site and to monitor the performance of the instrument in terms of baseline shifts, sample volume cell bias, and difference between gold traps, thus verifying that it adhered to standard procedures in a way that minimizes losses and inaccuracies in data production. The Toronto QA/QC system used to check all data collected by Tekran analyzers was based on the research data management quality (RDMQ) standards defined in Steffen et al. (2012). These standards invalidate data based on cell bias and sample volume, while also monitoring for baseline shifts and deviation amongst other warning flags.
Samplers were sent by international courier from each participating laboratory to the two sampling locations shortly before the first deployment period. Following instructions provided by each participating research group, the samplers were deployed on a metal support rack at a height of about 4 m above ground to facilitate free air circulation (Fig. S1 in the Supplement). At both sites, all PASs were within 2 m of each other and from the inlet of the active air sampler. When not deployed, samplers were stored on-site at room temperature. Samplers were returned to the participating laboratories, again by international courier, shortly after the end of the last deployment.
While the three PASs were treated the same as much as possible, there were
some unavoidable differences. The IVL-PASs made return air trips by
international courier to both sampling sites and were deployed at both sites
by personnel with no experience with this sampler. The CNR-PAS did not need
to undergo extended travel to the Rende site and the
After removal of the top cap, CNR-PASs and IVL-PASs were positioned in the
seats of the shelter with the diffusive membrane or steel mesh net facing
downwards. After exposure, CNR-PASs were removed from the seat, closed with
the top cap, and placed into an aluminum bag containing a mercury scrubber
cartridge. IVL-PASs were similarly removed, placed in a plastic container,
and then in a plastic bag.
The PASs were deployed from 5 February to 30 April 2019, following a sampling plan that included four deployments of 2 weeks, three deployments of 4 weeks, two deployments of 6 weeks, and one deployment each
of 8 and 12 weeks (Table S1 in the Supplement). All PAS deployments were in triplicate, with
the addition of a field blank for each type of PAS to check the potential
for contamination during transport, storage and handling of the samplers.
The CNR-PAS and
Mercury in CNR-PASs was quantified using a CNR-IIA-designed thermal
desorption system, comprising a glass cylinder housed in a heater furnace
connected to a mercury vapor analyzer (Tekran 2537a) for Hg detection by
CVAFS. The sorbent membrane is placed into the cylinder, which is heated to
550
For mercury determination in IVL-PASs, the carbon filters were carefully
removed from each sample and individually boiled in an acid solution
(
Determination of mercury concentration in the activated carbon sorbent used
in the
For QA/QC of mercury analytical data during sorbent analyses, both
analytical and field blanks were used. Analytical blanks were analyzed
before deployment and sampling to ensure sorbent materials (HGR-AC,
AuNPs-
The samplers were deployed in triplicate during the campaign to assess the
precision of each PAS. Method detection limits (MDLs) and practical
quantification limits (PQLs) in ng were calculated as 3 and 10 times
the standard deviation of the amount of mercury in field blanks,
respectively. The limits of detection (LOD) and quantification (LOQ) in ng m
The average Hg concentration in the atmosphere measured by each sampler (
All statistical analyses were performed using R v. 3.3.3 software (R
Foundation for Statistical Computing, Vienna, Austria). We evaluated the
relative accuracy of different PASs, by first calculating the percentage
concentration differences between actively sampled concentrations
[Hg]
Based on these results, we then sought to calculate and compare mean
absolute concentration percentage differences across both PAS types and
sites, while accounting for (1) the non-independence of samples, (2) unbalanced sample sizes across sites and PASs, and (3) potentially
confounding effects of (a) sampling deployment times and (b) sites. Therefore,
we parameterized a second linear mixed effects model where absolute
concentration differences were predicted as a function of PAS type, site,
and a PAS-by-site interaction term as fixed factors; this mixed model
statistically accounted for the non-independence of samples by including
deployment period and Tekran identity as nested random effects. Based on
this model, we then used the “
The concentration of gaseous mercury in ambient air was determined by
averaging the values recorded by the Tekran Hg analyzers every 5 min during a specific PAS deployment period. The complete series of valid Hg
concentration data is displayed in Fig. S2 in the Supplement. Interruptions are due to
instrument calibration or maintenance. As mentioned in Sect. 2.4, the
values obtained at Rende were validated against the GMOS-Data Quality
Management (G-DQM), resulting in 98.9 % valid data. The average measured
Hg concentration at Rende over the 12 weeks was
At Toronto, the use of the RDMQ standards for data quality assessment and
measurement gaps during daily calibration periods, hourly standard additions,
and instrument maintenance resulted in 82.5 % valid data coverage
throughout the entire deployment period for the primary 2537x analyzer. The
secondary co-located 2537a analyzer experienced an 8 % shift in the
mass flow meter calibration during the study. Since it was not possible to
determine when the shift occurred, data from this analyzer were not used for
comparison with the PAS (but were included in the statistical analysis
described in 2.8). The active Hg concentration ranged between 1.17 and 34.6 ng m
The amount of Hg in the field blanks of the different passive air samplers
is summarized in Table S2 in the Supplement. The averages of those values are displayed in
the top row of panels in Fig. 1. The amounts in field blanks are similar
between the different passive samplers, ranging from generally less than 0.2 ng in the CNR-PAS to slightly above 0.4 ng in the IVL-PAS. The blank levels of the CNR-PAS are the lowest recorded during the campaign, especially for
exposure in Rende. The
The relative standard deviation (RSD) of levels in field blanks was also
similar between the three samplers, being slightly lower in the IVL-PAS
(
Mean and standard deviation of field blank levels, method detection limit (MDL), practical quantification limit (PQL), limit of detection (LOD), and limit of quantification (LOQ) for the three passive air samplers deployed in Rende, Italy, and Toronto, Canada.
The amount of mercury detected in field blanks was used for the calculation
of the method detection limit (MDL), the practical quantification limit
(PQL), the limit of detection (LOD), and limit of quantification (LOQ). Field
blank levels, MDLs, PQLs, LODs, and LOQs for the three samplers separated for
the two locations are displayed in Fig. 1. The numerical results can be
found in Table S3 in the Supplement. The MDL and PQL are derived from the variability in the field blank levels. Therefore, they are similar between the three samplers (middle row of panels in Fig. 1). Even though the RSD of the field blank levels is smaller for the IVL-PAS, the larger absolute SD means that it has slightly higher MDL and PQL (
In terms of volumetric air concentrations, the LODs and LOQs decrease with
the sampled air volume, which, in turn, increases with a sampler's SR and
deployment period. The bottom row of panels in Fig. 1 therefore displays the
LODs and LOQs for each of the five deployment times used in this study.
Larger differences between the
The very large number of triplicate deployments in this study allows for a thorough characterization of the precision of the different PASs. Specifically, we assess the replicate precision of three PASs deployed simultaneously, both before and after blank correction. Table S5 in the Supplement reports the amount of Hg quantified in the PAS during the 22 different deployments. The precision of the quantified amount in a PAS reported in this table is a combined measure of the consistency and reproducibility of PAS manufacturing, deployment, and handling as well as the laboratory analytical process. Table S6 in the Supplement reports the blank-adjusted amount of Hg in the PASs. The precision of the blank-corrected amounts reported in this table additionally accounts for the consistency and reproducibility of the blank contamination.
The relative standard deviation in percent (RSD%) of the mean amount of Hg quantified in three samplers is used as a measure of precision. Blank correction was performed using the average value of all field blanks deployed at one location, because the field blanks did not show a dependence on deployment length for any sampler but did display differences between Rende and Toronto deployments for some samplers. The precision of the blank-corrected amount was calculated by propagating the standard deviations of the amount in exposed samplers and of the amounts in field blanks. Fig. 2 displays the replicate precision for the three samplers, averaged for different deployment lengths, across the two locations, and across all replicated deployments. Numerical results are presented in Table S7 in the Supplement.
When judged based on the amount of Hg quantified in triplicated samplers,
Precision expressed as the relative standard deviation in percent of the amounts of mercury quantified in triplicate PASs, both before (blue) and after blank correction (orange), averaged over different deployment lengths, across different locations, and over all replicated deployments. Note that in some cases a sampler was lost and therefore some deployments were only duplicated.
When judged based on the blank-corrected amounts in replicate samplers,
precision was 4 %, 15 %, and 14 % for
The average air concentration during each of the 22 deployment periods was
derived by dividing the blank corrected amounts in a PAS by the product of
the deployment period and the SR. These concentration values are reported in
Table S8 in the Supplement. The SR and its estimated uncertainty for each PAS was provided by
each participating laboratory. Specifically, the uncertainty of the
The accuracy of the PAS-derived time-averaged air concentrations in Table S8 in the Supplement was judged by comparing them to the average value derived by the active Tekran instruments operating alongside the PAS. The sampling rate of all three PASs used in this study had originally been obtained in calibration studies involving active Tekran instruments, i.e., any uncertainties as to the precise speciation of gaseous mercury being sampled by either a PAS or a Tekran instrument would have affected calibration and evaluation equally. A direct comparison therefore should be valid. Tekran values were considered as a benchmark for pragmatic reasons, knowing full well that this measurement itself may provide biased results (Aspmo et al., 2005; Slemr et al., 2015; Temme et al., 2007). This is true even though flow and detector accuracy audits of the active instruments were performed before, during, and after PAS deployments at both locations. The possible size of this bias was estimated from the data collected by the two Tekran systems operating side-by-side in Toronto, although the comparison was somewhat hampered by an inconsistency between measured flow at the beginning and end of the sampling period for the 2537a (unit 0075) (see Sect. 3.1). If we disregard that uncertainty, the 2537a (unit 0075) instrument yielded values averaged over the deployment periods that were consistently lower than those measured by the 2537x (unit 5037) instrument that was chosen as the reference. This bias was on average 3.2 % for the 11 sampling periods and ranged from a low of 1.0 % for the third 2-week period to a high of 6.5 % for the last 4-week period.
Average bias and average absolute discrepancy between the time-averaged volumetric air concentrations of Hg derived by the passive air samplers and the Tekran instrument.
Table 2 summarizes the average bias and the average absolute difference
between the average concentrations measured by the Tekran instrument at each location and the various PASs. This compilation reveals a number of
features: the accuracy of all three PASs is much better during the
deployments in Rende than the deployments in Toronto. On average, the
IVL-PAS and CNR-PAS results for Rende show no bias, whereas the
Figure 3 displays the discrepancies of the PAS results from the average
concentrations measured by the Tekran analyzers for each of the 22 sampling
periods. This illustration reinforces the noticeable differences in sampler
accuracy between deployments in Rende and Toronto. It additionally shows
that there is no apparent relationship between the accuracy of the PASs and
the length of the deployment period in Rende. For the
Discrepancies of the time averaged air concentrations of Hg during 22 deployment periods as derived by the three PASs from the average concentration obtained by an active Tekran system deployed at the same time. Deployments in Rende/Toronto are displayed in the upper/lower panel, respectively. Positive/negative discrepancies indicate a PAS-derived concentration higher/lower than the Tekran value, respectively.
Our variance partitioning analysis, coupled with mixed effects models,
confirmed that all PAS-derived concentrations were significantly closer to
the Tekran values for the deployments in Rende than they were for the
deployments in Toronto (Fig. 4). The asterisks in Fig. 4 designate the
significance level by which the mean absolute concentration difference of a
“dataset” differs significantly from 0, i.e., whether PAS-derived
concentrations (based on Eq. 2) differ significantly from
Tekran concentrations. We can see that the mean concentration differences in
Rende, for all three PASs individually, and when the data are “pooled”
among all PASs, they are not significantly different from 0. On the other hand,
all of the mean concentration differences in the Toronto site are significantly
different from 0, again both for all PASs individually and when the data are
“pooled” among all PASs. When data from both sites and all PASs are
“pooled” together, the mean concentration values differ significantly from
0, which is mainly driven by the poorer agreement of values in Toronto. Note
that we use here the terms “pooled” and “datasets”, even though the
results in Fig. 4 are based on the single mixed effects model and are not
the results of multiple
Least square means and standard errors of the differences in concentrations measured by the PASs and by the Tekran units. Results are shown either for each PAS individually (colored markers) or for the three PASs together (black markers). They are also shown either separately for the two sampling sites or both of them together. The asterisks below each bar indicate whether or not the least square mean concentration differences from
that “dataset” differed significantly from 0 (where
Of course, the same concentrations measured by different techniques should not be significantly different. It implies that the uncertainty of the concentrations derived from all three PASs deployed in Toronto must have been underestimated, i.e., the assumed uncertainty of the SR applied in the calculation of the concentrations must have been too small. We may surmise that if meteorological conditions during a deployment deviate considerably from those prevailing during the calibration of a PAS (as they did for all three PASs during a Toronto winter), the SR incurs considerably higher uncertainty than if calibration and application take place under similar environmental conditions.
The variance decomposition analysis attributed roughly half of the variance
in percentage concentration differences to the PAS type (48.3 %) and most
of the other half of total variance by differences observed between Toronto
and Rende (site
A PAS's performance depends on having an uptake capacity that is
sufficiently high for mercury to remain in a linear uptake phase throughout
the entire deployment period. We can test this by assessing the linearity of
uptake. While this is sometimes done by plotting the blank-corrected amount
quantified in the samplers
Figure 5 shows these uptake plots for all three samplers at the two sampling
locations. Also shown are the linear regression lines fitted to the
displayed data. Table 3 reports the slopes with standard error of the
regression line, which has been forced through the origin, and the
coefficient of correlation
Plot of the blank-corrected amount of Hg quantified in three types a passive air sampler deployed in Rende or Toronto against the product of the deployment time of a sampler
Results of the linear regressions displayed in Fig. 5. The slope of the regression line corresponds to the sampling rate of a passive air sampler.
All uptake curves are linear with high
Table 3 also compares the site- and deployment specific SRs obtained from the
regressions with the generic a priori ones that were used in the calculation of the volumetric air concentrations from the PASs. Deviations between these SRs should be roughly similar to the bias of the PAS-derived air concentrations reported in Table 2. In the case of the
We can also compare the relative size of the fitted SRs at the two locations.
Interestingly, for both the
Generally, the three PASs performed better in Rende than in Toronto. This is
most apparent in the assessment of accuracy (Figs. 3 and 4). However, this
did not apply to all performance indicators. For example, the magnitude and
variability in field blanks was comparable between the two sites for the
A major difference between the two sites is the harshness of the weather conditions during the deployment period, which comprised the 3 months of February to April 2019. Winter and early spring in Toronto can be very cold, can have large temperature fluctuations over short time periods, and can have precipitation in different forms (snow, freezing rain, sleet, and rain). As was discussed in the preceding section, temperature and wind speed can influence the rate of diffusion to the passive sampling sorbent, causing variability in the SRs. It is also conceivable that during inclement weather, hoarfrost forms on the surfaces of the diffusive barriers or blowing snow could accumulate on the samplers, potentially impeding the path of Hg to the sorbent. However, it will often not be possible to attribute discrepancies to weather conditions, for example when deviations occur in opposite directions during overlapping deployments (e.g., the third 2-week and the second 4-week deployments overlap, yet the CNR-PAS shows positive bias in the former and negative bias in the latter).
Another possible source of the discrepancies between Tekran and PAS concentrations in Toronto (Fig. 3) is the higher fraction of missing or rejected data from the Tekran instrument operating in Toronto. While 98.9 % of Tekran data were valid in Rende, the data coverage for the PAS deployment was 82.5 % for the Tekran 2537x unit providing the reference value in Toronto. The reason for the lower percentage in Toronto is a sampling method that relies on daily calibrations (2.4 % daily loss of coverage) and hourly (8.3 % daily loss of coverage) spikes, which together account for a 10.7 % per day loss of coverage, yet improve confidence in the data. The distribution of the standard addition spikes throughout the day, however, means they are unlikely to result in any bias of the results. The remainder of data loss was due to regular maintenance and a power outage. A different, but equally valid, sampling method in Rende ran calibrations every 3 d with no spikes. For individual deployments, data coverage ranged as low as 66 % for the fourth 2-week deployment in Toronto. However, the discrepancy between the PASs and Tekran instrument are not unusually large during that deployment period (Fig. 3).
A final difference between the two study locations is the occurrence of several short spikes of elevated gaseous mercury concentrations in Toronto. If these had been caused by a local source in the immediate vicinity of the sampling site, it is conceivable that spatial gaseous mercury concentration gradients may have been present within the assembly of PASs and Tekran inlets. However, no relationship between the occurrence of such spikes and the discrepancies in PAS results is apparent. In any case, it is more likely the spikes were caused by sources sufficiently far from the sampling site to not result in concentration gradients on the scale of a few meters.
Table 4 compiles the key performance indicators for the three passive air samplers. In contrast to most of the sections above, this table provides the average of all values obtained from the Rende and Toronto deployments. This compilation reveals that the
Summary of the key metrics describing the performance of the three passive air samplers for Hg as determined in this study.
Table 4 also shows that IVL-PAS and CNR-PAS are remarkably similar in their performance characteristics with very similar LODs and replicate precision. While the average bias of the CNR-PAS overall is very small, this is largely because fairly large discrepancies occur in either direction and therefore cancel each other out. Overall, the IVL-PAS-derived air concentrations agree better with the Tekran-derived data than those of the CNR-PAS (12.5 % vs. 19.1 %).
The
The data presented in this study are available from the authors upon request.
The supplement related to this article is available online at:
AN conceived and coordinated the study, planned the PAS deployment strategy, assembled the CNR-PAS data submission, collected the research data, and wrote the original draft. AT, MM, and SM were responsible for the field deployments and TEKRAN measurements in Rende and contributed to data analysis and writing. AM, EZ, PP, JA prepared and analyzed the CNR-PAS. NP and FS acquired funding and managed some of the projects leading to this publication. MN, JM, and IW prepared and analyzed the IVL-PAS and assembled the IVL-PAS data submission. GWS was responsible for the field deployments and the TEKRAN measurements in Toronto. DB performed the analysis of the
The author declare the following competing financial interest(s): Tekran Instruments Corp. pays some licensing fees to the University of Toronto related to the
This article is part of the special issue “Research results from the 14th International Conference on Mercury as a Global Pollutant (ICMGP 2019), MercOx project, and iGOSP and iCUPE projects of ERA-PLANET in support of the Minamata Convention on Mercury (ACP/AMT inter-journal SI)”. It is not associated with a conference.
We are grateful to David Gay for compiling the data generated by the different participants and making them available after they had all been received, and to Emily Alvarez for help with sampler deployments in Toronto. The Ontario Ministry of Environment, Conservation and Parks is acknowledged for site access and meteorological data.
The work in Toronto was supported by a Discovery Grant of the Natural Sciences and Engineering Research Council of Canada to Frank Wania and a grant and contribution agreement from Environment and Climate Change Canada (fund no. GCXE19S042) with Carl P. J. Mitchell. The work in Rende was funded by the European Commission – H2020, the ERA-PLANET program (
This paper was edited by Pierre Herckes and reviewed by four anonymous referees.