Introduction

Many regions of the world far exceed the World Health Organization (WHO) air quality guidelines for ambient fine particulate matter (PM2.5)1,2 air pollution levels, with substantial impacts on human health3. The Global Burden of Disease estimated 3 million deaths (9% of all deaths) and 80 million years of lost healthy life were attributable to exposure to outdoor PM2.5 globally in 20174. Nonetheless, ground-based monitoring of PM2.5 mass concentration is inadequate for exposure assessment5, and the select studies of PM2.5 chemical composition measurements6,7 are even sparser. In some locations, the majority of metal concentration measurements made to date have been of the PM10 fraction8,9,10. Global observations of PM2.5 mass concentration and composition can inform aerosol model development and exposure assessment, improve understanding of emission sources and help prioritize mitigation policies11 to reduce health impacts.

The relationship between PM2.5 and human health12,13 including the association of PM2.5 with cardiovascular disease14, respiratory disease15, cancer16, and type 2 diabetes17 has become better understood over the last few decades. However, more analysis of these relationships is needed before effects of particular components (specifically trace metals) are well understood at a global scale. The oxidative potential of PM2.5 is related to its metal content6 (as well as its carbon content18) and increased abundance of redox-active elements can induce oxidative stress19,20,21,22. Increased cardiovascular disease rates have been associated with exposure to enhanced relative concentrations of K, Al, Ni, Zn and V14,23 and mortality risks have been associated with preferential bioaccumulation of heavy metals such as As, Pb, and Al24. Further, many metals have known health effects, such as As, Cd, and Cr, which are classified by the WHO’s International Agency for Research on Cancer (IARC) as known human carcinogens (IARC Group 1)25, and Pb, which is associated with impaired cognitive function26. Measurements of PM2.5 composition are needed to assess the global distribution of these deleterious metals in fine particulate matter, as they could pose health risks to populations living in areas of high PM2.5 pollution.

Ground-based elemental composition can also provide information about airborne PM2.5 sources. For example, K has associations with wood burning27,28,29, Zn can be linked with traffic through tire wear30, and V derives mainly from heavy fuel oil combustion such as from shipping31,32. Coal is a source of multiple potentially harmful elements such as Pb, Cr, Mn, As, and Se31,33 whereas non-ferrous metal production is a large source of As, Cd, and Zn31. Vehicle traffic contributes a mix of elements including the heavy metals Ba, Zn, and Pb, as well as the crustal components Fe, Al, Mg, and Ti27,34. To our knowledge, no other global network has measured the trace metal concentrations in PM2.5. These observations are needed to better understand particulate matter sources and loadings, to assess emerging emission inventories and evaluate spatially-resolved concentration predictions made by chemical transport models on a global scale35,36 and to understand local and regional impacts of emission sources.

The ground-based Surface PARTiculate mAtter Network (SPARTAN; http://www.spartan-network.org) is building up a long-term data set measuring PM concentration at globally distributed sites37 and provides new data to evaluate PM composition. In this study, we investigate the trace metal composition of PM2.5, supplemented by coarse PM (PMc = PM10-2.5) and PM10 data, in ambient air samples from SPARTAN sites around the world. Our focus is on PM2.5 due to its importance for health.

Results and discussion

The mean PM2.5 mass concentrations at each SPARTAN site are shown in Fig. 1, with standard error bars (SE = \(\sigma/\sqrt{n}\), σ = sample standard deviation, n = number of samples) for each site mean. Of the 19 SPARTAN sites, Kanpur had the highest mean PM2.5 levels at 102.8 ± 20.2 (SE) µg/m3, followed by Beijing at 58.1 ± 2.2 µg/m3, then Dhaka (49.0 ± 3.3 µg/m3) and Hanoi (47.1 ± 7.6 µg/m3).

Figure 1
figure 1

Mean PM2.5 mass concentrations at SPARTAN sites with standard error bars shown. Overlaid green bars show total measured trace metal mean mass concentrations for each site.

Full size image

Sites located in urban areas situated outside of North America generally had the highest levels of PM2.5. The one notable exception was the Mammoth Cave site, which had the 11th highest PM2.5 levels. This can be explained through two aspects of the sampling; firstly, the site is located in a heavily wooded national park; sampling occurred in the summer when carbonaceous emissions are highest from biogenic sources, and secondary organic aerosol is prevalent38. Secondly, the site has the second-lowest percentage of its PM2.5 mass present in the measured trace metal mass (4%), indicating other sources (i.e. biogenic) were dominant. Sites located in Canada generally had the lowest levels of PM2.5, which is consistent with their intended categorization as low-PM environments.

Table 1 details the mean mass concentrations of 15 trace metals analyzed in PM2.5 samples from SPARTAN sites. Elements above ICP-MS detection limits (detailed in supplemental Table S1) for > 10% of samples were considered; Li, Co, Ag, and Ce did not satisfy this requirement, and as such are not discussed here. Individual elements were examined to understand their distributions across sites (a detailed breakdown across sites can be found in the supplemental section S1, with a corresponding table of standard deviations in supplemental Table S4). Elemental breakdowns for the PM10 fraction are found in supplemental Table S7, while relative concentrations of each trace element in PM2.5 can be found in supplemental Table S8.

Table 1 Full elemental breakdown of mean mass concentrations of trace metals in PM2.5 at SPARTAN sites.
Full size table

Whole-system uncertainties for the SPARTAN network were calculated in the same manner as previously reported by Weagle and coauthors39. The results for the uncertainties can be found in the supplemental Table S3, and individual plots in supplemental Figure S1.

Of interest is the anthropogenic contribution to such trace metals in general. Crustal enrichment factors (EFs) were used to distinguish naturally occurring crustal elements from those released by anthropogenic sources. We compared filter-extracted elemental concentrations X to background continental concentrations from Taylor and McLennan40, normalized by measured crustal Fe concentrations, which is predominantly from natural sources (e.g.41) as seen in Eq. (1).

$$E{F_{X,2.5}} = \frac{{{{\left[ {{\raise0.7ex\hbox{$X$} \!\mathord{\left/ {\vphantom {X {Fe}}}\right.\kern-\nulldelimiterspace} \!\lower0.7ex\hbox{${Fe}$}}} \right]}_{{\text{P}}{{\text{M}}_{2.5}}}}}}{{{{\left[ {{\raise0.7ex\hbox{$X$} \!\mathord{\left/ {\vphantom {X {Fe}}}\right.\kern-\nulldelimiterspace} \!\lower0.7ex\hbox{${Fe}$}}} \right]}_{{\text{Taylor}}}}}}$$
(1)

Although natural variability in the Fe fraction in soil will affect these results, the Fe fraction in soil tends to be quite consistent42. Anthropogenic sources of Fe imply that the resultant enhancement ratios will be a conservative indicator of anthropogenic contribution. Nonetheless, to ensure that conclusions are not affected by using Fe as the reference element due to extraction efficiency, a complementary analysis was performed by replacing site Fe concentrations with corresponding coarse PM concentrations, seen in Eq. (2). Coarse particulate matter over land has been established as predominantly stemming from suspended dust43,44 (with other potential sources including urban dust and organic material), making it a suitable measure of crustal source abundance to verify the validity of results using Fe. An estimate of the concentrations of the metal oxides present in coarse PM (estimated coarse PM = PMc*) was obtained by use of the soil reconstruction equation described by Malm et al.43 and shown in Eq. (3), in combination with continental concentrations from Taylor and McLennan.

$$E{F_{X,2.5}} = \frac{{{{\left[ {{\raise0.7ex\hbox{$X$} \!\mathord{\left/ {\vphantom {X {P{M_c}}}}\right.\kern-\nulldelimiterspace} \!\lower0.7ex\hbox{${P{M_c}}$}}} \right]}_{{\rm{P}}{{\rm{M}}_{2.5}}}}}}{{{{\left[ {{\raise0.7ex\hbox{$X$} \!\mathord{\left/ {\vphantom {X {PM_c{^*}}}}\right.\kern-\nulldelimiterspace} \!\lower0.7ex\hbox{${PM_c{^*}}$}}} \right]}_{{\rm{Taylor}}}}}}$$
(2)
$${\text{PM}}_{{\text{c}}}{^{*}} = { 2}.{2}0\left[ {{\text{Al}}} \right] + { 2}.{49}\left[ {{\text{Si}}} \right] + { 1}.{63}\left[ {{\text{Ca}}} \right] + { 2}.{42}\left[ {{\text{Fe}}} \right] + { 1}.{94}\left[ {{\text{Ti}}} \right]$$
(3)

Enrichment factors can be considered in three subsets: EF < 10, elements which we attribute to sources that are primarily crustal45,46,47; EF between 10 and 100, elements with mixed anthropogenic and natural sources; and EF > 100, elements with largely anthropogenic sources. Figure 2 displays the EF elemental breakdown for all SPARTAN sites as well as the site-specific PMc-scaled EF results.

Figure 2
figure 2

(a) Crustal enrichment factors (EF) for PM2.5 relative to crustal ratios of the given element and iron (Eq. 1), and (b) replacing measured iron with measured PMc (Eq. 2). Elements are sorted by mean enrichment factor across all sites; sites are listed alphabetically. Singapore is not shown in the bottom plot due to unavailable PMc data.

Full size image

Similar results when normalizing by either Fe or PMc provides confidence in the analysis. Enrichment factors vary by orders of magnitude depending on element and location. The highest levels of enrichment are found in anthropogenically-dominant elements such as Pb, As, and Zn, and generally found at sites with the highest levels of PM2.5. Of the four sites with the greatest PM2.5 concentrations, elemental EF values > 350 are found in Dhaka, Kanpur, and Hanoi for Pb, Zn, and As, and in Beijing for As. These enrichments highlight the effects of anthropogenic activities not only on total PM2.5 but on trace metal PM2.5 specifically, with potentially harmful levels of these heavy metals found in these high-PM cities.

Of the individual elements measured, two elements in particular demand further investigation: the carcinogenic metalloid As, and the heavy metal Pb. These elements can originate from multiple industrial sources, including smelting, waste incineration, and coal burning12,14. There are guidelines for the concentrations in air of both elements. For example, the US National Ambient Air Quality (NAAQS) exposure limit for Pb is a 3-month mean concentration of 150 ng/m348, and multiple samples taken from the Kanpur and Dhaka sites exceeded these Pb concentrations—leading to 3-month mean concentrations over the US guideline for both sites. Figure 3 shows the lead PM2.5 concentrations measured in samples taken from the two sites, with the NAAQS guideline as a reference. This guideline refers to Total Suspended Particulate matter (TSP), a wider size range of particulates, so measured concentrations of lead PM2.5 are underestimating the TSP concentrations.

Figure 3
figure 3

Concentrations of Pb in SPARTAN samples at Kanpur (red circles) and Dhaka (black triangles). Dotted blue line represents the NAAQS 3-month exposure guideline for lead concentrations (150 ng/m3). Samples are plotted by the final date of sampling, as sampling occurs over a time period of 9 days. Solid red (Kanpur) and black (Dhaka) lines represent 3-month rolling mean Pb concentrations.

Full size image

A comparison of the PM2.5 (Table 1) and PM10 (Table S7) values reveals that site-mean PM10 values are 41% higher at Kanpur and 74% higher at Dhaka than the PM2.5 values. Comparing PM2.5 at the two sites, Dhaka has both a higher mean Pb concentration (280 ng/m3) than Kanpur (209 ng/m3), and more samples measuring above the 150 ng/m3 guideline (38% vs. 30%). This is in contrast to their respective levels of total PM2.5, as mean PM2.5 concentrations from Kanpur samples are roughly double those in Dhaka. The Dhaka site exhibits some seasonality, with peak Pb values occurring around January 2015 and generally high values in winter months, when mixed layer depths are shallow. Similarly high Pb levels in PM2.5 have previously been reported in Dhaka49,50, and have been attributed to combustion of fossil fuels, battery industries, paints and varnishes, and lead-containing waste water.

The second element of interest in this study is arsenic, as the World Health Organization recommends no safe level of arsenic exposure due to its carcinogenic risk. An estimated lifetime excess risk is 1:1,000,000 at 0.66 ng/m3, or 1:100,000 at 6.6 ng/m351. As is associated with industrial activities such as smelting, burning of coal, and waste incineration52,53,54. Mean As concentrations for SPARTAN sites are shown in Fig. 4. The majority of SPARTAN sites fall under this level, but there are four sites with mean As concentrations near or above 6.6 ng/m3—Kanpur, Hanoi, Beijing, and Dhaka. The Beijing and Dhaka sites were sampled for an entire year facilitating direct comparison with the guideline. Although sampling at Kanpur and Hanoi was less than a year, concentrations at these sites were sufficiently high that the standard would be approached or exceeded even in the unlikely event that ambient concentrations were zero for the remainder of the year (yielding roughly 11.5 ng/m3 for Kanpur and 6.1 ng/m3 for Hanoi).

Figure 4
figure 4

Concentrations of As in PM2.5 samples taken from SPARTAN sites, with standard deviation bars shown. Dotted red line represents 1:100,000 excess lifetime risk of cancer due to As exposure (6.6 ng/m3). Dotted black line represents 1:1,000,000 excess lifetime risk of cancer due to arsenic exposure (0.66 ng/m3).

Full size image

Kanpur has significantly higher As concentrations than any other SPARTAN site with 15.3 ng/m3. The mean As mass concentrations for Hanoi, Beijing, and Dhaka were determined to be 8.1 ng/m3, 7.1 ng/m3, and 6.3 ng/m3 respectively. These sites are also the four sites with the highest levels of PM2.5, so it is not unexpected that they have greater abundances of trace metal PM2.5 such as arsenic.

A complementary approach to understanding the levels of trace metals in fine PM at SPARTAN sites is to compare them not only to the crustal abundance, but to another site with relatively low levels of these trace metals. Figure 5 shows the relative abundance (RA) of PM2.5 trace metals at the eight SPARTAN sites with at least one elemental RA of 10 or greater, compared to a natural reference site, in this case the Mammoth Cave (M.C.) National Park site. The Mammoth Cave site has the second-lowest trace metal mass percentage of total PM2.5 mass. Low trace metals and the natural environment of the site make it an insightful reference point against which to compare the various types of SPARTAN sites. Relative abundances for trace metal PM2.5 are calculated using Eq. (4) and are unitless.

Figure 5
figure 5

Relative abundances of trace metals in PM2.5 in representative SPARTAN sites with at least one element with RA of 10 or greater. Abundances are relative to the natural, low-trace metal PM2.5 Mammoth Cave site. Sites are sorted by total PM2.5 mass concentrations L-R, then top to bottom. Relative abundance values are unitless.

Full size image
$$R{A}_{X}={\left[{\text{X}}\right]}_{\text{Site}}/{\left[{\text{X}}\right]}_{\text{M.C.}}$$
(4)

Sites with an elemental RA of 10 or greater are shown in the figure below to highlight particular areas and species of note. The full set of relative abundances can be found in the supplemental information (Table S5), as well as a similar calculation using the PM2.5-relative elemental concentrations instead of the absolute concentrations (supplemental Table S6). Generally, combustion-related elemental concentrations increase together, as do crustal components.

Particularly high RA values (12–311) of anthropogenic elements Zn, As, Pb, and Cd are found for Kanpur, Beijing, Dhaka, and Hanoi; these abundances are in alignment with the crustal enrichments seen for these heavy metals at these sites, and with the higher PM2.5 levels at these sites.

Examining the relative abundances on a site-specific basis reveals more local information—one notable instance being the Beijing site, which shows markedly elevated levels of selenium compared to any other site. Chinese coal has been found to be particularly rich in selenium55, which in concert with the large quantity of coal burned in China implies coal-burning as a major anthropogenic source of selenium PM2.5 in the region. Coal emissions in China have also been shown to contain high levels of As, Cd, and Pb56, all of which are significantly elevated by RA in the PM2.5 samples from Beijing.

There are a few isolated, notable RA values at the Bandung and Ilorin sites that can be linked to specific regional industries. Bandung has highly elevated Pb (RA of 38, EF > 700), even after the phasing out of leaded gasoline in 200657. One likely contributor to the extremely high Pb levels is lead smelting, as Indonesia is one of the largest lead acid battery recyclers in Asia58. Rapidly growing numbers of lead smelters produce large amounts of Pb-enhanced emissions, the transport of which could explain the elevated lead levels seen at the Bandung site. The Ilorin site shows abundant levels of Cr (RA of 40, EF > 260); this may be associated with the prevalent tanning industry in the region, in which chromium compounds are used prominently. Analysis of effluents from Nigerian tanneries found high levels of Cr present59, and high levels of the metal in PM2.5 imply that some of this Cr is being converted to or released in particulate form.

Singapore has significant enrichment of V (49), which likely stems from the nearby burning of shipping fuel, with the Port of Singapore being one of the busiest ports in the world. Proximity to petroleum refineries is another likely contributor to these elevated V levels60, as the Singapore site is located approximately 10 km east-northeast of Jurong Island, an industrialized artificial island home to several refineries61. Significantly enhanced levels of V in Singapore can also be seen in the crustal EFs (EF > 240, the highest for V of any SPARTAN site), which further validates the key impact of nearby anthropogenic emissions. The influence of these and other nearby anthropogenic activities (e.g. vehicular traffic and metal production facilities) is also seen in enhanced levels of Zn, with an RA value of 13 and EF > 600. Unlike some of the other regions with high PM2.5 and significant anthropogenic activity, Singapore does not exhibit large relative abundances for other anthropogenically dominant elements such as Pb, As, and Cr. This is likely due to the near-total absence of coal-burning in the region, as Singapore relies heavily on natural gas for energy purposes61.

Two non-North American sites did not exhibit any RA values of 10 or higher: Manila and Rehovot. This merits attention as Manila in particular is a densely populated city yet shows no significantly higher levels of trace metals than our natural site in Mammoth Cave. One partial explanation is that Manila has relatively low PM2.5 mass concentrations among the non-North American sites (8th out of 11 sites). For Rehovot, the sampling site is the least densely populated non-North American site, which would align with the generally low amounts of trace metals observed at that location. Of the North American sites, Toronto, Bondville, Sherbrooke, Atlanta, Halifax, Kelowna, and Lethbridge all had RA values below 5 for all elements measured. This aligns well with expectations that these sites should generally have less PM pollution, especially for the Canadian MAPLE (Mortality Air Pollution Associations in Low Exposure Environments) sites designated as low-PM environments.

Although SPARTAN does not yet have sites in Europe, we surveyed prior measurements from the region to place our findings elsewhere in context. We find that at European background reference sites, heavy metal concentrations tended to be lower than at SPARTAN sites in densely populated regions62,63. These findings reinforce our conclusions about the enrichment of heavy metals compared with background reference sites.

Conclusions and implications

Globally consistent measurements of airborne metal concentrations in PM2.5 are needed to assess and understand the distributions of ambient concentrations. Over 800 samples of fine particulate matter (PM2.5) from 19 unique sampling sites located across four continents were collected and analyzed using consistent protocols, with concentrations of 15 different trace metals reported. As the SPARTAN network continues to develop, these sampling protocols will continue to be improved to better report the chemical composition of trace metals in PM2.5. In general, it was found that several elements were enriched compared to their concentrations in the crust—in particular, elements such as Pb, As, and Zn that have anthropogenic sources were greatly enriched in sampled PM2.5. This enrichment was most notable in large, densely populated urban areas such as Beijing, Dhaka, Kanpur, and Hanoi, but was seen generally across many SPARTAN sites.

The enrichment of potentially harmful elements in fine particulate matter from anthropogenic sources is of global relevance to public health and warrants further attention. For example, we found in Dhaka and Kanpur that Pb concentrations exceeded the US National Ambient Air Quality 3-month guideline of 150 ng/m3, and that concentrations of the carcinogen As approached or exceeded the World Health Organization’s 1:100,000 excess lifetime risk level (6.6 ng/m3) in Kanpur, Hanoi, Beijing, and Dhaka. Emerging evidence indicates that anthropogenic fugitive, combustion, and industrial dust can comprise a significant fraction of PM2.5 in densely populated regions64. More generally, the high concentrations of several potentially harmful elements (e.g. Zn, Pb, Cd, Se, and As) at densely populated cities such as Beijing, Dhaka, Kanpur, and Hanoi motivate expanded measurements in other cities worldwide, especially in rapidly developing economies.

Materials and methods

SPARTAN site-selection favors densely populated, globally dispersed regions that are underrepresented in terms of availability of representative and long-term air quality data. Local site-selection favors representative environments that avoid anomalous sources; low rooftops in urban environments are desirable to increase fetch, diminish local traffic influence, and offer instrument security. Locations of SPARTAN sites are shown in Fig. 6 and further site information (e.g. elevation, latitude, longitude, sampling period) is detailed in Table 2.

Figure 6
figure 6

Open blue circles indicate the location of SPARTAN sampling sites used here, overlaid on a background color map of population density (NASA SEDAC GPW65, made using MATLAB R2019b—https://www.mathworks.com/products/matlab.html).

Full size image
Table 2 SPARTAN site-by-site information.
Full size table

As SPARTAN site selection prioritizes under-sampled locations, some regions that are well-represented in terms of air quality data have not established a SPARTAN site to date. Further expansion of the network is planned for the future, in order to cover regions not sampled by SPARTAN, as well as add more sites in underrepresented continents.

Sampling through SPARTAN has also occurred at several North American sites through various projects. Three pilot sites in the United States sampled in 2014–2016, and five sites located across Canada sampled over 2017–2019 as part of the MAPLE project67. Overall filter sampling occurred over an average of 20 months across sites (range of 2 to 50 months). Over 800 PM2.5 filters have been analyzed as of October 2019 and were included in the data set used for this study. While some sites may have lower numbers of samples, the SPARTAN sampling protocol (detailed below) was designed to better capture long-term averages in sampled PM, so even relatively few samples could be representative for the season in which they were sampled (detailed in Table 2). Ongoing measurements are expected to increase seasonal representativeness over time.

An overview of SPARTAN is provided by Snider et al.37,68, in which the basis of the sampling procedure, filter analysis, and measurement ranges are presented. As the SPARTAN project developed, some aspects of the process were updated. To briefly summarize the original procedure, PMc (coarse PM) and PM2.5 filter masses were collected on a two-stage stacked filter unit inside rooftop-mounted AirPhoton SS4i automated air samplers over 9-day periods. A removable filter cartridge protected seven sequentially active pairs of coarse Nuclepore and fine Teflon filters, plus a pair of travelling blanks. Beginning in late 2017, sampling stations at SPARTAN were upgraded to the AirPhoton SS5 models, which use a cyclone inlet to separate particles by varying flow rates through the station (5 and 1.5 L/min, respectively, for PM2.5 and PM10 size-cuts). This allowed for elimination of the Nuclepore filters and the introduction of filter cartridges with eight total stretched Teflon filters that included six sampling PM2.5, one sampling PM10, and one travelling blank. These cartridges were pre-assembled in the SPARTAN central laboratories at Dalhousie University, and shipped to sites for installation by site operators.

Once the filter cartridges are installed, the sampling stations run autonomously for the cartridge duration. For international sites, each of the PM2.5 filters is sampled for rotating 3-h spans over 9 days, totaling 24 h of sampling. Measuring across the entire diurnal cycle over a 9-day period, rather than sampling for a consecutive 24-h period every 9 days, helps to better capture long-term averages37. The PM10 filter was sampled for a 30-min period after each 3-h PM2.5 sample for the entire 54-day sampling period, providing a 54-day average PM10 concentration over a total of 24 h. For sites that are part of the MAPLE project, each filter was sampled for a total of 48 h (6-h periods for the PM2.5 filters and 1-h periods for the PM10 filters) to ensure a quantifiable amount was deposited in these low-PM environments. Once sampling was complete, filter cartridges were removed from the sampling station, sealed, and returned to the central laboratory at Dalhousie University for chemical and physical analysis of the filters. Filters were analyzed to determine PM2.5 or PM10 mass (gravimetric), water-soluble ions (ion chromatography with conductivity detection), black carbon (absorbance, determined through smoke-stain refractometry), and trace metals (inductively-coupled plasma mass spectrometry).

The focus of this study is trace metal content, quantified here via inductively coupled plasma mass spectrometry (ICP-MS, Thermo Scientific X-Series 2). For each filter a small quantity of isopropyl alcohol (previously 30 µL, currently 10 µL) was added, then the filter was extracted (97 °C for 2 h) with 5% trace metal grade nitric acid solution in a process similar to Fang et al.20 and Herner et al.69. Filters were boiled in the acidic solution and the liquid extract submitted for quantitative analysis via ICP-MS, using 25–500 ppb (μg/L) trace metals standards and three reference elements for atomic mass (45Sc, 115In, and 159Tb) for each analysis. Measured concentrations from the field blank filters for each filter cartridge were subtracted from the seven corresponding samples of each cartridge to account for variable trace metal baselines. Consistent ICP-MS analysis in the central Dalhousie laboratory facilitated consistency of results across sites. It is established that nitric acid extraction efficiencies for some crustal elements such as Fe can be as low as 50%70 as confirmed by our internal tests, and as such, measurement methods are being re-evaluated for future SPARTAN use. More recent analyses include the use of HCl to improve extraction efficiencies. In addition, filters will also be analyzed with XRF at Washington University. For now, this initial analysis offers perspective on this emerging dataset, with observed variability that far exceeds the factor of 2 uncertainty associated with some extraction efficiencies.

In order to compare SPARTAN trace metal measurements with independent concurrent measurements, a joint sampling campaign was conducted in the US in concert with the IMPROVE (Interagency Monitoring of Protected Visual Environments) network that has been previously described68 and is summarized in the supplemental information (section S2).

Whole system uncertainties for the SPARTAN network are estimated through use of collocated filter sampling stations. The process is described in previous work39 but briefly, three sites in typically low (Halifax, Canada), moderate (Toronto, Canada), and high (Beijing, China) PM environments performed collocated sampling over 3 week periods. Over this period, each station recorded 24-h samples (48-h in Halifax to ensure adequate loading) which were then analyzed to evaluate uncertainties across the network as further described in the supplemental material (section S3).