Global review of recent source apportionments for airborne particulate matter

Abstract

Source apportionments have become increasingly performed to determine the origins of ambient particulate pollution. The results can be helpful in designing mitigation strategies to improve air quality. Source specific particulate matter (PM) concentrations are also being used in health effects studies to be able to focus attention on those sources most likely to be responsible for the observed adverse health effects. In 2015, the World Health Organization (WHO) released its initial compilation of source apportionment studies published through August 2014. This initial database was described by Karagulian et al. (Atmospheric Environment120 (2015) 475–483). In the present report, a new compilation has been prepared of those apportionments published since 2014 through December 2019. In addition, the database has been expanded to include apportionments of heavy metals, water-soluble components, and carbonaceous components in ambient PM. As a result of this work, we have developed and presented some perspectives on source apportionment going forward. We also have made a series of recommendations for source apportionment studies and reporting them. It is essential for papers to provide a minimum set of information so that the study can be adequately assessed, and the results utilized by others in making policy decisions or as part of other scientific studies.

Graphical abstract

Highlights

• •Comprehensive review of the PM_2.5/PM₁₀ source apportionment literature since 2014
• •Includes apportionments for heavy metals, water-soluble species, and carbonaceous PM
• •Major increase in source apportionment studies in China
• •Most commonly used source apportionment method is positive matrix factoriztion

1. Introduction

Identification and quantitative apportionment of atmospheric pollutants to their sources is an important step in providing information that guides the development of air quality management strategies. These strategies are implemented through regulations and other measures such as subsidies for controls. In addition, changing economic conditions can also affect source emissions of particulate matter precursor species such as SO₂ and NO_x that will form particulate sulfate and nitrate through atmospheric oxidation. For example, the availability of low-cost natural gas in the United States has displaced a significant amount of coal used for electricity generation (Squizzato et al., 2018a). Trends in the airborne particulate matter (PM) concentrations suggest that the policy, the economics or both have resulted in improved air quality (e.g., Squizzato et al., 2018a). Source apportionments can help to determine if policy implementation and/or economic drivers have actually changed the mixture of sources contributing to the measured pollution (Squizzato et al., 2018b). The resulting source-specific particulate matter (PM) concentrations can then be used in health effects studies to identify those sources most strongly associated with adverse health outcomes (e.g., Rich et al., 2019; Croft et al., 2020; Hopke et al., 2020). Most source apportionment studies have focused on PM_2.5 (particles with aerodynamic diameters ≤2.5 μm) and PM₁₀ (particles with aerodynamic diameters ≤10 μm).

Compositions measurements of PM emissions have not been extensively made in Europe. Thus, their compendium focused on source profiles estimated from ambient data using methods like positive matrix factorization (PMF). It can be found at SPECIEUROPE (https://source-apportionment.jrc.ec.europa.eu/Specieurope/index.aspx). Through the leadership of the European Union's Joint Research Centre in Ispra, guidance on how to perform source apportionment studies (Belis et al., 2014, Belis et al., 2019; Thunis et al., 2020) and several intercomparison studies (Hopke et al., 2006; Belis et al., 2015, Belis et al., 2020) have been conducted to improve the skills in using these techniques across Europe.

In China, source profiles measured in China from 1987 to 2017 are available at https://doi.org/10.5194/acp-19-3223-2019. In addition, the source profiles were deposited in the Mendeley data repository and can be downloaded from doi:10.17632/x8dfshjt9j.2. In 2013, the Air Pollution Prevention and Control Action Plan was promulgated by the State Council of China as part of the implementation of the 13th 5-year plan. Since 2013, there have been a series of research initiatives to improve air quality through the more effective use of source apportionments including requiring apportionment studies to be conducted in 35 megacities. Through the national key research programs and the associated guidance documents, apportionments have been completed in >40 cities. However, not all this work has been published in the scientific literature and may be forthcoming as more data become available given ongoing efforts to reduce particulate pollution in China and to document the resulting changes in concentrations and source contributions.

Tools such as positive matrix factorization (PMF), chemical mass balance (CMB), and Unmix have become readily available. These methods have been extensively employed in a variety of studies and were reviewed by Hopke (2016). They can be downloaded from the U.S. EPA website and used without cost (https://www3.epa.gov/scram001/receptorindex.htm). Compilations of source profiles for the United States, Europe, and China are available. Measured U.S. profiles are compiled in the Speciate database (https://www.epa.gov/air-emissions-modeling/speciate). PMF and related non-negative least-squares methods have become the most widely used method because of the limited availability of up-to-date, locally relevant source profiles in most of the world.

Given the need for such studies and the resources to conduct them, there has been increasing numbers of source apportionment studies around the world. Karagulian et al. (2015) compiled and reviewed all the source apportionment studies published through August 2014. Major efforts have also been made to perform source apportionments in all major cities of China beginning in 2013 with an initial training workshop held in 2014 (https://cleanairasia.org/node12387/) followed by several other for local authorities. Thus, there have been many studies made across Europe and China in addition to continuing work in North America that have occurred in the past 5 years. There are also more efforts being started in other countries to make use of source apportionments to support their air quality management efforts. To support these efforts, a new database of the papers reporting PM source apportionments based on compositional data published since the final quarter of 2014 has been developed. We have retrieved, reviewed, and compiled the results in the spreadsheets provided in the supplemental material files for this report. This paper presents the approach and summary results describing the current known apportionments of PM_2.5 and PM₁₀ from around the world using receptor models and measured ambient PM composition data.

2. Methods

To initiate the work, searches were performed using the Web of Science, SCOPUS, and Google Scholar using the keywords: source apportionment, chemical mass balance, positive matrix factorization, particulate matter, PM_2.5, PM₁₀, and ambient PM. Each paper was then examined to ascertain what was apportioned and if it provided quantitative results in which PM concentrations were assigned to specific source types. In this assessment, it was found that many of the published papers did not always provide critical details of their study. Quantities that were commonly omitted included mean measured values for the parameter being apportioned, specific locations of the sampling site(s), actual period of the ambient sampling campaign, total number of samples, and the details of the application of the apportionment methods. For example, Brown et al. (2015) provide useful guidance as to what should be reported for any PMF analysis. In many cases, the apportionments were only provided as pie or bar graphs without a table of the mass or % contributions of the identified sources. In the case of missing information, the corresponding author was contacted, and these additional details are provided in the tables. However, in the absence of a response, values were estimated by digitizing the graphs using Un-Scan-It (V7).

3. Results and discussion

3.1. Scope of the study

Table 1 presents summary numbers for the identified source apportionment records and the number of specific sources identified for PM_2.5 and PM₁₀. The sites at which samples have been collected are displayed in Fig. 1. In Table 1 and the results tables in the supplemental material, the list of source types was expanded from that of Karagulian et al. (2015) since more recent works have reported more detailed source apportionments. We have also added comments to clarify what specific apportionments were for individual publications. We have provided references in the form of DOIs when available since searching for the DOI will allow a quicker retrieval of the original publication.

Table 1.

Summary of the typology of the collected apportionment reports.

	PM	Metals	Water soluble	Carbonaceous
Typology
No. of publications	414	20	11	25
No. of countries	58	2	2	7
No. of PM10 values	243	4	1	8
No. of PM2.5 values	564	25	14	40
No. of OC values	–	–	–	37
No. of EC values	–	–	–	17
Source type – PM2.5
Sulfate	300	0	0	1
Nitrate	282	1	1	1
Mixed SIA	186	21	6	–
Secondary organic	–	–	–	9
Vegetative detritus	–	–	–	25
Sea Salt	240	23	3	–
Dust	510	22	14	23
Traffic	507	21	2	50
Diesel	0	0	0	27
Industry	272	0	7	0
Biomass burning	384	0	9	38
Coal combustion	363	1	5	28
Food cooking	–	–	–	7
Other (unspecified-human origin)	332	0	2	25
Source type – PM10
Sulfate	116	0	0	0
Nitrate	102	0	0	0
Mixed SIA	43	2	1	–
Sea salt	157	2	0	0
Dust	182	2	1	2
Traffic	183	0	0	2
Diesel	106	1	1	4
Industry	116	1	1
Biomass burning	112	0	0	2
Coal combustion	101	0	0	3
Other (unspecified-human origin)	119	–	–	0

Fig. 1.

Map showing the locations of the sampling sites for PM_2.5 (blue points), PM₁₀ (red points), and combined PM_2.5/PM₁₀ (purple points) reported in the identified apportionment publications.

In addition to the apportionment of PM mass, additional results have been obtained for heavy (toxic) metals, water-soluble species, and carbonaceous species. In the case of heavy metal and water-soluble species, some of the reports attempt to apportion the PM mass even though the missing species such as carbonaceous component are missing and represent a substantial fraction of the PM mass. In general, the carbonaceous species (organic carbon (OC) and/or elemental carbon (EC)) analyses only apportioned the OC and/or EC and did not report apportionments of the PM mass concentrations. A higher fraction of these reports compared to the PM apportionments did not report the mean values of the measured OC or EC concentrations.

In all, we included 741 PM apportionments from 414 published papers covering 401 different locations, 28 apportionments of metals in 19 papers from 13 locations, 18 water soluble species apportionments in 13 papers from 14 locations, and 32 papers reporting 70 apportionments of carbonaceous species in 42 locations. The publications report apportionments that provide information on 1 to 9 source categories. Fig. 2 shows the distributions of the number of resolved sources for PM_2.5 and PM₁₀. During this period of 2015 to 2019, more reports of PM_2.5 were available than that for PM₁₀. There were two peaks in the PM_2.5 distribution with 5 source categories having the largest number of reports but significant numbers of papers reported 6 to 9 source types. Many of these studies were of shorter duration and with apparently more limited numbers of samples resulting in fewer resolved source types. For PM₁₀, there was a rising number of reported source types up to 7 and a sharp drop off at 8 and 9 sources.

Fig. 2.

Histogram of the number of papers reporting a given number of resolved source categories.

We have not made quality assessments as was done in the prior study (Karagulian et al., 2015). All of the studies published in peer-reviewed journals were included. It is very difficult to truly assess these studies given the frequent lack of key details such as the number of samples or details of their application of the source apportionment method(s). There are many studies that only covered short periods with relatively few samples and yet used factor analysis methods such as PMF. Since such methods depend on the existence of “edge” points (Henry, 2003) to be able to resolve the sources, these studies generally only report a few source types and often with mixed compositional profiles. There are still papers appearing using eigenvector-based methods such as principal components analysis (PCA) without a subsequent multilinear regression analysis (MLR). PCA apportions variance, typically after mean values are subtracted from the measured data value, and thus, do not provide a quantitative apportionment of the mass. Many of the PCA papers used the inappropriate eigenvalue >1 criterion for the number of retained factors when they should use the number of factors contributing total variance >1 after rotation (Hopke, 1982). Even with a multilinear regression or an unnormalized solution such as in absolute principal components analysis (APCA), eigenvector analyses are inappropriate tools for heteroskedastic data such as species concentrations in airborne particulate matter since an eigenvector analysis is actually an unweighted least squares fit (Lawson and Hanson, 1974; Malinowski, 2002). Several journals have now indicated that they will no longer consider manuscripts that employ PCA as their primary approach to source apportionment (Hopke, 2015; Hopke and Jaffe, 2020).

There were 539 reported PMF results. However, in many of the publications, important details of the PMF analyses were missing including the method for estimating the data point uncertainties, the criteria for choosing the number of factors to retain, and for the more recent papers, failure to report the results of the displacement (DISP), boot-strap (BS), and combined (BS-DISP) analyses as suggested by Brown et al. (2015).

For most of the 89 CMB studies, local, current profiles were not measured, and they relied on existing source profile libraries. In these approaches, it is possible to obtain a good fit to the data with incorrect profiles and thereby estimate incorrect source contributions. Subramanian et al. (2006) provides a very useful analysis of this problem in estimating the contributions of motor vehicles to OC and PM_2.5 in Pittsburgh. Thus, trying to assign quality indices to the publications would involve significant amounts of subjective judgments. If prior source apportionments are of interest for comparisons with current work, it is strongly suggested that investigators use the tables to identify relevant papers. They can then be carefully reviewed, and the investigators can make their own quality assessments.

3.2. Source types

3.2.1. Secondary inorganic species

There were 3 inorganic source types reported: sulfate, nitrate, and mixed secondary inorganic aerosol (SIA). Sulfate and nitrate in PM typically arise from the oxidation of SO₂ and NO₂, respectively. SO₂ emissions come from the burning of sulfur containing fossil fuels like coal and residual oil typically burned in larger scale combustion systems such as coal-fired power plants and marine diesel engines. However, there can be emission of SO₃ to form primary sulfate in marine diesel engines (Agrawal et al., 2008a, Agrawal et al., 2008b, Agrawal et al., 2010) and residential coal combustion (Dai et al., 2019). NO₂ is formed from NO emitted by high temperature combustion of gas, liquid fuels, and coal. Both NO₂ and SO₂ can be oxidized homogeneously by hydroxyl radicals and heterogeneously in droplets and on particle surfaces (Seinfeld and Pandis, 2016). In general, with sufficient data, it should be possible to separate nitrate from sulfate since nitrate tends to dominate in winter when temperatures are lower favoring the formation of particulate ammonium nitrate. Sulfate typically peaks in the summer when the increased photochemical activity permits more homogeneous formation of sulfate from the emitted SO₂. Thus, short sampling campaigns may not provide sufficiently different mixtures to permit resolution of the sulfate and nitrate and thus, mixed SIA is reported. In a number of cases, it was reported that there was secondary organic carbon (SOC) associated with these source types. Since oxidative chemistry drives the formation of SIA and SOC and the SIA particles represent a significant amount of the ambient aerosol surface area, condensation of the SOC on the SIA particles would account for their covariance and the assignment of SOC mass to the SIA particle types. It also needs to be noted that there can be emissions of primary sulfate from any combustion sources such as diesel engines burning high sulfur oil (Cordtz et al., 2013) and low temperature combustion of coal (Dai et al., 2019). The sulfate may not appear clearly in the “diesel” profile since the variations in sulfate can be fully subsumed by the “sulfate” factor. However, the relationship with the primary emissions can be observed in the g-space plots (Kim and Hopke, 2008: Pandolfi et al., 2020) and adds to the total impact of the primary source.

3.2.2. Dust

This generic name for crustal material includes both natural soil and desert dust suspended in the atmospheric by winds and resuspended dust arising largely from traffic. The resuspended road dust is generally characterized by additional metals arising from brake wear (e.g., Cu, Sb, Si, Fe), tire wear (Zn), and exhaust emissions from burnt lubricating oil (Ca, Ba, Zn). In some studies, these were distinguishable and in others, they were combined into dust or crustal designations.

3.2.3. Sea salt

Sea salt is noted by the presence of sodium and can be reported as both fresh sea salt with a significant chloride concentration in the profile and/or aged sea salt with the chloride displaced by sulfate and/or nitrate ion. Thus, in some cases, sea salt represents the sum of fresh and aged particles with the aged particles being a combination of natural and anthropogenic source emissions.

3.2.4. Traffic

Motor vehicle emissions can include both exhaust and non-exhaust emissions from spark-ignition (gasoline) and compression-ignition (diesel) vehicles. In most cases, a single combined traffic or vehicular source was identified. In some cases, exhaust and non-exhaust were reported separately while in others, they apportioned gasoline and diesel vehicles, and road dust although all of the separate apportionments were not always reported. We have noted in the comments any additional information that was provided about how the exhaust and non-exhaust traffic emissions were apportioned.

3.2.5. Industry

Industry is a broad category including both specific industries such as a neighboring steel mill whose emissions are being characterized and multiple source types are identified to a generic “industry” consisting typically of metallic elements whose origins are not well known. In some reports, multiple specific types of industry were reported. Their contributions were summed and noted in the table comments.

3.2.6. Biomass burning

This category also includes multiple sources including agricultural burning and residential heating/cooking using wood or other biomass (crop residues, dung, etc.). The source was typically characterized by the presence of high potassium concentrations and sometimes supplemented with levoglucosan or other molecular markers. In some areas, both biomass and coal are burned for heating/cooking, and they covary sufficiently so as to be inseparable. If the predominant or distinguishing species in the profile was potassium, the contribution as assigned to the biomass burning category.

3.2.7. Coal/oil combustion

Some reports only provide a “fossil fuel” source while others specify “coal” or “heavy oil.” Coal combustion can be multiple processes. There can be high temperature coal combustion in coal-fired power plants or other industrial activities where pulverized coal is injected into the boiler. Many of these sources have particulate controls so that the major emissions are SO₂ and NO_x although there have been significant efforts to add or improve the gaseous pollutant controls on coal-fired power plants such has occurred in China (Zheng et al., 2018). In many cities, local district heating plants burn pulverized coal at high temperatures, but without controls leading to emissions of both particulate and gaseous pollutants, but little or no primary sulfate. The profile for this type of coal combustion depends strongly on the nature of the coal. For example, in Beijing, the coal has a very high chlorine content and chloride in the particles is the marker species (Yu et al., 2013). Finally, there is residential coal combustion for heating/cooking in which lump coal or briquettes are burning. Large pieces of solid fuel burn less efficiently that pulverized coal and during the startup and burnout phases of the combustion cycle, large quantities of primary sulfate (Dai et al., 2019) and oxidized primary organic carbon (Li et al., 2019a) are emitted. Thus, understanding the nature of local sources is important in the correct identification of source profiles derived by factor analysis methods like PMF.

“Heavy oil” is more generally referred to as “residual oil,” “No. 6 oil,” or “Bunker-C oil”. This material is the residue left after the fractional distillation used to provide liquid fuels (liquid petroleum gas, gasoline, kerosene, No. 2/diesel oil). It contains the solid materials that were suspended in the crude oil including the porphyrins that are enriched in nickel and vanadium as well as significant amounts of sulfur. Thus, No. 6 oil has a distinctive Ni/V signature although this ratio varies depending on the origin of the crude oil. It is primarily used as fuel in marine diesel engines and represents a significant sulfur source in major ports. However, the sulfate contribution is often subsumed by the sulfate factor as noted previously and its contribution can be assessed using the g-space edges (Kim and Hopke, 2008). Marine diesel emissions have been declining through regulations requiring cleaner fuels within the territorial waters of the port's location. However, after January 1, 2020, rule IMO 2020 (http://www.imo.org/en/MediaCentre/HotTopics/Pages/Sulphur-2020.aspx) requires all ships to use fuel containing <0.5% sulfur, down from the prior global limit of 3.5%. Ships may comply by being equipped with SO₂ scrubbers or using low sulfur fuels as ships come into port. However, there may continue to be significant emissions while the ships are in international waters.

3.2.8. Other

It is not always possible to assign all the measured mass to the resolved sources either in PMF or CMB analyses. Thus, the unaccounted-for mass is the difference between the mean measured mass and the sum of the mean contributions of the resolved sources. In many reports, the “Other” category is used to report this unapportioned mass. However, in other cases, source types that do not fall into the defined categories such as “secondary organic carbon” or “construction dust” were reported. Notes were placed in the comment column to provide additional information on the meaning of “Other” for each paper as appropriate.

4. Metals, water-soluble, and carbonaceous

These 3 additional tables in the supplemental material file provide information on major portions of the ambient PM although not the total mass concentration. Many of the “metals” papers included health risk estimates based on standard exposure-risk relationships (EPA, 1989) for carcinogenic and non-carcinogenic endpoints. These reports have not been combined with the PM apportionments given their limited species coverage, and in many cases, the lack of a quantitative PM measurement. However, they provide additional information for researchers who are also studying these more limited apportionment problems. In most cases, there is a report of the fraction of the secondary organic aerosol or carbon, but the apportionment of the SOA/SOC to biogenic or anthropogenic sources is not provided. There are marker species for SOA (e.g., Kleindienst et al., 2007; Fu et al., 2016), but they have not been widely used. As the secondary inorganic fraction of PM_2.5 continues to diminish with increased controls on their precursor sources, more emphasis will be needed on full apportionments of the carbonaceous fraction of the PM_2.5 to provide an adequate understanding of the PM_2.5 mass sources to permit appropriate control strategies to be formulated.

4.1. Regional average contributions for the specified source categories

Regional or country-specific average values were calculated for each source type. These population weighted averages were calculated following a similar approach as in Karagulian et al. (2015). The equation for these calculations is as follows:

$$S_i=\frac{{\sum }_{j=1}^n\mathit{Cpo}{p}_j\times s_{\mathit{ji}}}{{\sum }_{j=1}^n\mathit{Cpo}{p}_j}$$ (1)

where S_i is the population weighted average fractional contribution of source i, Cpop_j is the city population for study j, and s_ji is the fractional apportioned contribution of source i to the PM_2.5 or PM₁₀ for study j. If the source category apportionments did not sum to 100%, the difference was assigned to the “Other” category.

The results for PM_2.5 are presented in Fig. 3 and those for PM₁₀ are shown in Fig. 4. The numerical values are provided in Table 2. In the present analyses, there are many more reports of PM_2.5 apportionments compared to PM₁₀ results and very few where both size fractions were collected and analyzed at the same location. Thus, the results for the two size fractions are not directly comparable since they do not directly represent the same specific locations, analysis methods, completeness of data, and data analytic methods. The secondary inorganic species have been combined into a single category by taking the sum of sulfate and nitrate and performing the population weighted average with the mixed SIA category. It is difficult to compare these figures directly with those in Karagulian et al. (2015) since they reduced the categories to Traffic, Industry, Domestic fuel burning, Natural sources (dust and sea salt), and Unspecified sources. Thus, important source types like mixed SIA that are generally regional in origin were admixed into more local source types.

Fig. 3.

Map showing regional or country average source apportionments for PM_2.5.

Fig. 4.

Map showing regional or country average source apportionments for PM₁₀.

Table 2.

Tabulation of the fractional apportionments by global region or country.

Source	Australia and New Zealand	Central America & Caribbean	Eastern Asia -Not China	Eastern Europe	Northern Africa	Northern America -Canada	Northern America -USA	Northern Europe	South America - Brazil	South America - other	South-eastern Asia	Southern Asia	Southern Europe	Western Africa	Western Asia (M49)	Western Europe (M49)	Northern China	Southern China
No. of reports	4	4	27	20	4	23	148	8	7	4	14	46	86	10	12	60	205	41
PM2.5 (μg/m3)	15.8		27.5	24.7	25.0	8.5	11.0	14.9	22.4	35.1	24.8	102.5	20.4	50.7	46.7	15.6	100.9	50.3
Mixed SIA (%)	11.5		42.3	31.5	0.0	26.8	46.2	25.0	17.1	20.1	23.9	34.3	19.1	21.3	51.8	36.0	31.0	31.0
Sea salt (%)	7.0		6.8	27.1	0.2	7.5	7.2	10.2		16.0	7.8	4.8	19.5	1.4	5.6	4.5	4.1	4.1
Dust (%)	5.1		7.2	18.5	20.0	5.4	2.4	18.8	3.8	13.5	19.1	9.8	19.8	18.4	6.0	12.3	8.9	8.9
Traffic (%)	10.4		17.6	23.8		18.6	8.0	50.4	23.5	35.5	23.0	25.2	26.6	17.9	14.9	19.2	19.3	19.3
Industry (%)	21.5		8.9	15.9		6.7	17.8	8.5	32.5	38.6	15.4	6.4	5.9	8.7	1.0	17.7	16.8	16.8
Biomass. burning (%)	75.1		10.1	17.8		13.8	4.4	15.4	22.4	12.2	16.3	14.9	10.7	2.8	13.5	12.0	10.3	10.3
Coal or no. 6 oil combustion (%)			14.5	32.4		7.9	1.4	11.1	4.4	29.2	13.8	7.1	56.4	13.0	15.9	16.1	10.9	10.9
Other (%)	12.0		11.9	18.5	24.0	10.1	26.0	4.1	17.9	15.7	23.9	10.8	14.5	15.7	1.5	15.2	10.5	10.5
PM10 (μg/m3)	20.5	35.4	40.5	13.8	56.0	16.2	18.4	28.9	42.5	51.9	92.8	190.0	32.3	220.2	85.3	30.6	164.6	110.0
Mixed SIA (%)	6.0		0.0	48.5			18.6	32.7	17.9	21.0		17.5	26.2	22.2	16.7	35.3	22.3	28.9
Sea salt (%)	13.0	1.0		8.0	4.5		3.3	15.6	14.0			16.5	8.9	2.2	6.2	10.1	6.9	6.0
Dust (%)	5.0	1.0	44.0	16.7	44.1	58.0	24.3	24.6	25.3	28.0		33.5	22.2	26.9	37.3	13.0	30.5	15.2
Traffic (%)		1.0	8.0	17.2	6.0	3.0	15.8	4.9	38.4	23.0		21.4	20.9	20.0	16.6	19.8	14.6	25.6
Industry (%)	48.0			5.9			11.2	6.6	11.1	4.9		19.8	7.4	17.9	9.3	4.5	12.1	19.0
Biomass. burning (%)				14.0		17.0	12.3	3.3	7.6			17.0	11.8		9.4	14.4	14.3	5.9
Coal or no. 6 oil combustion (%)		1.0	25.0	41.7		14.0	4.9	5.6				12.9	5.7	22.1	18.0	7.8	20.2	17.3
Other (%)			23.0	8.1	8.7	7.0	31.4	22.8		23.0		23.2	8.7	35.0	13.8	9.5	11.0	9.0

Sulfate continues to be a dominant part of the PM_2.5 mixed SIA category of the countries with major economies such as North America, Europe, mainland China, and other East Asian countries including the Republic of Korea and Japan. Coal combustion remains an important category in eastern and southern Europe and China. In these areas, there have been efforts to reduce the extent of coal-burning to reduce both air pollutants and greenhouse gas emissions. However, many of the reported apportionments are based on samples collected as far back as 2000. Thus, these results do not fully reflect the changes in sulfate and nitrate concentrations that have been observed in the eastern US (Squizzato et al., 2018a), China (Zhang et al., 2018), and in Europe (Li et al., 2017). There have been increases in sulfate in South Asia, Central Asia, North Africa and the Middle East, Eastern and Southern Sub-Saharan Africa, Andean South America, and Oceania and relatively constant sulfate concentrations elsewhere globally (Li et al., 2017). Thus, areas where there have been active efforts to reduce SO₂ emissions have seen reductions although they are not always reflected in the source apportionment results being reported here.

Nitrate also decreased in the United States although not as rapidly as sulfate (Squizzato et al., 2018a). However, in the North China Plain, the 31.8% reduction in NO_x emissions from 2010 to 2017 only lowered surface nitrate by 0.2% and even increased nitrate in some polluted areas (Fu et al., 2020). In both cases, the increased availability of oxidants such as hydroxyl radical resulting from the decrease in other pollutants like SO₂ has led to the slower or no reductions in nitrate. Similar limited declines in nitrate are observed in Europe (Li et al., 2017). It can be anticipated that future apportionments will see further declines in the SIA concentrations, but with the potential for increased concentration of secondary organic aerosol whose origins can only be ascertained with more extensive chemical characterization and related expense.

Other apportionments are what might be expected in terms of dust dominating the PM sources in North Africa although there are only 3 results from Tunisia and one from Algeria. The results for Australia and New Zealand represent only 4 reports. The 2 of the 3 from Australia are from rural areas (Cape Grim and the Snowy Mountains) so the apportionments are dominated by the two urban sites, Nelson, NZ and Muswellbrook, Australia. Central America and the Caribbean also only represent 4 reports, 3 of which are from Monterey, Mexico and the other from Cuba. Ten of the 14 Southeastern Asia PM_2.5 reports are from Malaysia and the remaining 4 are from generally rural sites in the Philippines. Some areas have only values for one or the other PM size range given the published reports that could be found. For example, there are no PM₁₀ reports from southeastern Asia and no PM_2.5 studies in Central America or the Caribbean. Thus, there are differences in the patterns of pie charts between the two maps (Fig. 2, Fig. 3).

The focus of apportionment studies remains on higher income countries in North America, Europe, and East Asia. There were no reports from Central Asia, Southern Africa, much of Central and South America, and Southeastern Asia although many of these areas have significant particulate air quality issues that could benefit from source apportionment information. The lack of consistent, long-term sampling and analyses programs preclude detailed trend analyses except in the United States where speciated data are available beginning in 2001 (Solomon et al., 2014) and in south and southeastern Asia where data have been collected under a series of IAEA projects (Hopke et al., 2008; Atanacio et al., 2016).

5. Perspectives

Currently, the least-squares approaches for source apportionment are relatively mature. In addition to EPA-PMF and EPA-CMB, non-negative constrained alternating least square (Tauler et al., 1993), and non-negative least squares (e.g., Lee and Seung, 1999; Camp, 2019) can be downloaded and used freely. There are good reviews of these methods and guides to their use that were mentioned above. It is important for people using these methods to take time to carefully read the literature so that they understand the underlying basis for these methods. Papers such Henry (2003) and Paatero et al. (2014) are essential literature to review prior to applying factor analysis methods. Thus, some of the major improvements that can be made in future studies are going to be improving the input data for the analyses.

An issue in any study of the ambient aerosol is the quality of the sampling and measurements. Very few studies report using reference materials to ascertain the quality of their analytical procedures. At the present time, there are very limited standard reference materials that are available for validating the analytical methods for any of the reported species. The U.S National Institute of Standards and Technology (NIST) have a few relevant materials including SRM 2783 which provides PM_2.5 deposits on polycarbonate filter. SRM 2783 contains certified elements (e.g. Na, Mg, Al, K, Ca, Ti, V, Cr, Mn, Fe, Co, Ni, Cu, Zn, As, Sb, and Pb) and reference elements (e.g., Si, and S). However, most sampling for elemental analyses using XRF collect the samples on Teflon filters. Yatkin et al. (2018) have described an approach to produce multi-element reference samples, but there is currently no production and certification of these reference materials. Schantz et al. (2016) report the development by NIST of 2 new fine PM reference materials, SRM 2786 Fine Particulate Matter (<4 μm) and SRM 2787 Fine

Particulate Matter (<10 μm), that are provided as bottles of powder. However, these standards would not provide a test of extracting the species of interest from the particles on a filter material. Thus, additional certified reference materials are required to provide realistic and effective tests of the analytical methods being applied to characterize collected PM samples and provide the input data to source apportionment studies.

One improvement in source apportionments that can be easily done using commonly available data is the addition of pollutant gases to the input data. Species like CO and NO_x will covary with the particulate species particularly those associated with local source emissions. CO is typically higher in spark-ignition emissions while NO_x is more dominant in diesel engine emissions. Gases had been initially incorporated into CMB analyses by Marmur et al. (2005). Emami and Hopke (2017) have shown that adding gases to a typical PMF data set provides additional source resolution and reduces rotational ambiguity. Li et al. (2019b) provided a better resolution of PM_2.5 in Beijing by adding gases, OC/EC IMPROVE thermal fractions, and water-soluble OC and HULIS. Thus, including additional species in the data set can add to the quality of the source apportionment.

Another way to incorporate additional information into the analysis is to use multiple site data. Depending on the locations of the sampling sites and the point sources are relative to the wind direction, there can be samples from one site that are affected by a source that provides no impact at another site that is directionally separated from the source plume. Kara et al. (2015) were able to resolve a number of specific sources using a network of sampling sites. Thus, combining multiple site data can potentially improve source resolution and reduce the rotational ambiguity.

For all methods, increasing the time resolution of the methods would be very useful. As noted above, having edge points is essential to factor analysis. Edge points exist because the contribution from a source is zero. However, when does traffic emissions cease in major urban centers? Hourly data can be obtained from online instruments such as in situ XRF, OC/EC, and ion chromatographic analyzers, or an Aerosol Chemical Speciation Monitor (ASCM) and a multiple wavelength aethalometer. Then, it is possible to measure minimal traffic emissions in the early hours of the day. Photochemically driven PM peaks in the early afternoon. This added time resolution provides additional resolution and accuracy in the apportionments. Data from different time resolution instruments can be combined (e.g., Srivastava et al., 2019) although not with currently freely available software. In addition, these methods provide 168 hourly samples per week so even short-term sampling campaigns provide a reasonable number of samples although they do not provide seasonal or long-term trend results. However, these methods are now being deployed in long-term monitoring stations and publications making use of such data should be increasing in future publications.

As many countries reduce SO₂ and NO_x emissions and the subsequently formed secondary inorganic PM decreases in concentration, the origins of the carbonaceous constituents in PM becomes more important. There have also been improvements in in situ systems for organic constituents in PM (e.g., Williams et al., 2006). Such data provide increased source resolution including atmospheric processes that lead to the formation of secondary organic aerosol.

There are possible methodological approaches that may improve source apportionments. There are now several studies using Bayesian approaches (e.g., Balachandran et al., 2013; Park et al., 2019). Bayesian models assume that source contributions and profiles vary probabilistically. Bayesian models incorporate existing knowledge about each of these parameters by the assignment of a “prior” probability distribution that reflects the current state of knowledge of the parameter. There may be some information about the profiles similar to what is used in CMB analyses, but very little is known about the distributions underlying the measured profile values since there have not been sufficient measurements made on any source at any time. Clearly very little is known a priori regarding the contributions and thus, defining the prior distributions are problematic.

6. Recommendations

To improve the quality of the data used for source apportionment, improvements in sampling and analyses are needed. There are a wide variety of samplers employed, most of which have been well characterized in the literature and some that have not. The sampling systems need to be fully described either in the reports or by reference. Analyses should be evaluated routinely using standard reference materials. Currently there are not many standards available and what are available may be too expensive for some programs to buy. Yatkin et al. (2018) provided a method to make XRF reference filters that have been evaluated by Hyslop et al. (2019). However, these filters are produced using aerosol technologies that typical analytical laboratories do not have. Thus, production of such reference filters by a third party would provide a useful addition to those reference materials issues by national standards organizations. There should also be use of solid reference materials such as NIST coal fly ash or some of the IAEA reference materials in the QA/QC of dissolution methods used to prepare samples for inductively coupled plasma/mass spectrometry (ICP/MS) analyses. At least summaries of the QA/QC data should be provided in any report as part of the supplementary materials.

The increased use of semicontinuous, in-situ instrumentation is highly encouraged. Although such instrumentation is expensive, the availability of hourly data provides a much more robust data base from which to do source apportionment. Prior work had shown how data collected on different bases can be combined to produce more detailed and likely more accurate apportionments (Zhou et al., 2004; Ogulei et al., 2005; Srivastava et al., 2019). Such measurements will also provide continuous data that will be more useful for chemical transport model evaluations and epidemiological studies.

To make reported source apportionments more useful to other researchers and policy makers, it is recommended that authors, reviewers, and editors carefully ensure that all potential source apportionment papers contain the following information:

• •Coordinates of the sampling site (latitude and latitude)
• •Start and stop dates of the sampling campaign and total number of collected samples
• •Mean value of the apportioned variable(s); If there are multiple sites and/or multiple apportionments, provide site-by-site and/or variable-by-variable results
• •
  Full details of the apportionment calculations

  • o
    For CMB, provide the profiles, where they come from, the basis for choosing them, and the regression diagnostics related to the profile matrix
  • o
    For PMF, provide the process for the error estimations, choice of numbers of factors and the rationale for that choice, and results of the error analyses as described by Brown et al. (2015).
  • o
    For Unmix, provide the selection of profile numbers, profiles, and related details.
• •
  Table reporting the mean apportionment values

  • o
    A pie or bar chart can go into the paper but put a table of numerical values into the supplemental material file.
  • o
    Report mean values for each site; mean values across multiple sites can go into the main text but have a table of the site-by-site values in the supplemental material file so the spatial variability of the source contributions can be assessed.
  • o
    It is preferable to report both the mass and fraction (%) contributions even if the mass contributions can be calculated from the mean mass and % contributions.

More uniform reporting of the details of source apportionment studies and their results will make them much more useful to review, assess, and incorporate in future work that looks to examine changes in concentrations and source contributions over time and space or to provide a context for other studies such as human or ecological health assessments based on broader parameters such as PM_2.5, PM₁₀, or water-soluble species. It will also make them more credible to air quality managers who will use the apportionment results in developing air quality management plans.

Declaration of competing interest

The Authors certify that there were no conflicts of interest in searching, collecting, assimilating and reporting the work presented in this manuscript.

Editor: Pavlos Kassomenos

Appendix A. Supplementary data

Supplementary material