Source Apportionment of Total Suspended Particles (TSP) by Positive Matrix Factorization (PMF) and Chemical Mass Balance (CMB) Modeling in Ahvaz, Iran

The World Health Organization (WHO) has reported that among the 3.7 million premature deaths annually linked to outdoor air pollution, 90% are in developing countries. Most of these deaths are related to cardiovascular, respiratory, and lung diseases (WHO 2014). Particulate matter (PM) is among the most important air pollutants, with adverse effects on human health and the environment. Several studies have been conducted regarding the association between PM and different diseases, disability, and mortality (Brunekreef and Forsberg 2005; Hadei et al. 2017a). Epidemiological studies have proven the association between pulmonary and respiratory diseases and high concentrations of PM (Brunekreef and Forsberg 2005; Mohseni Bandpi et al. 2017).

Iran is faced with serious air pollution problem in recent years, which had led to significant loss in public health (Hadei et al. 2017b). The concentrations of PM in several cities of Iran exceed the guideline values from the World Health Organization (WHO) (Asl et al. 2015; Conti et al. 2017; Miri et al. 2016; Mohammadi et al. 2016). Ahvaz, with a population of 1.3 million people, is surrounded by large deserts, which raises a high possibility of dust storms (Shahsavani et al. 2012a). This city is known for its relatively low vegetation, strong surface wind, and high temperature and humidity. The vast deserts around Ahvaz (especially in western areas and in countries, such as Iraq and Saudi Arabia) are reported as the main sources of dust particles (Shahsavani et al. 2012b). High concentrations of dust originating from the southern and western neighboring countries of Iran pass through the southern, western, and central areas of Iran. Dust events have caused severe unhealthy conditions in many cities, especially megacities, such as Tehran and Ahvaz, within recent years. Ahvaz has been recognized as one of the most polluted cities in the world because of high concentrations of PM and deaths attributed to PM (Khaniabadi et al. 2017; WHO 2016). Hence, there is a compelling need for quantification, identification and apportionment of the pollutants’ sources to facilitate their reduction through proper control strategies and plans (Taiwo et al. 2014).

Receptor modeling (RM) uses the physical and chemical characteristics of air pollutants to identify and apportion their contributing sources. RM approaches use information from the measurements of PM chemical composition and estimates of its uncertainty and have been extensively applied to evaluate the contributions of emission sources to atmospheric pollution in specific sites (Bove et al. 2014; Taiwo et al. 2014). Common receptor models can be categorized into univariate models, such as chemical mass balance (CMB) models, and multivariate models, such as principal component analysis (PCA), positive matrix factorization (PMF), and the EPA’s Unmix model (Choi et al. 2015).

The CMB model uses both the concentrations of chemical species in PM (and their uncertainties) and the chemical fingerprints of the sources (i.e., source profiles) as input. Alternatively, the PMF model uses only the concentrations and uncertainties of chemical species as inputs, because the factor/source profiles are outputs of the model together with their contributions to measured PM concentrations. Both models are sensitive to the selection of sources and to their collinearity (i.e., similarity of the tracers characterizing the source in the profile). Some methodologies have been developed to deal with the collinearity problem, including the one based on diagnostic Si/Al ratio for crustal elements and coal-fired power plants (Contini et al. 2016), and the one to distinguish between two forms of nitrate (Cesari et al. 2018). The PMF model identifies factors that must be attributed to specific emission sources, and in some cases, the obtained factors could include contributions from multiple sources (Hopke 2010; Watson et al. 2008). CMB is the most conventional method among those requiring very detailed information of sources and emission profiles. However, PMF requires relatively little quantitative knowledge of sources and emission profiles, although it does require initial qualitative information of the sources present in the study area. It is suggested that the integrated application of RM methods can solve the limitations of each of the models by constructing a more robust solution based on their strengths (Viana et al. 2008).

Using individual or combined receptor models, several studies have been conducted on the source apportionment of particulate matter in various countries (Cesari et al. 2016; Lee et al. 2008; Shi et al. 2014). Manousakas et al. (2017) used PMF 5.0 for source apportionment of PM_2.5 in a coastal area of Greece and found the possible sources: biomass burning, sea salt, shipping emissions, vehicle emissions, mineral dust, and secondary sulfates (Manousakas et al. 2017). Gianini et al. (2013) conducted a study, using PMF and CMB models, on source apportionment of PM₁₀, organic carbon (OC), and elemental carbon (EC). The authors found some differences in the results obtained from the two models (Gianini et al. 2013). Cesari et al. (2016) compared the application of PMF 3.0, PMF 5.0, and CMB models and observed a significant difference in contributions of some sources (Cesari et al. 2016). Viana et al. (2008) proposed the integrated use of factor analysis techniques (PCA and PMF) to identify and interpret emission sources and obtain a first quantification of their contributions to the PM mass and the subsequent application of CMB (Viana et al. 2008).

There are few studies on the source apportionment and emission estimation of PM in Iran and the Middle East. Sowlat et al. (2012) applied the PMF model for source apportionment of total suspended solids (TSP) in Ahvaz. Among seven estimated sources, crustal dust was identified as the major source of TSP. Other sources included road dust, motor vehicles, marine aerosols, secondary aerosols, metallurgical plants, and petrochemical plants and fossil fuel combustion (Sowlat et al. 2012). Likewise, in another study, the results of the PMF model showed that the crustal dust has the highest contribution to the mass of PM₁₀ in the ambient air of Ahvaz (Sowlat et al. 2013). Using the PMF 1.1 model, Engelbrecht and Jayanty (2013) assessed sources of airborne mineral dust and other aerosols in 6 cities of Iraq.

Five individual factors were modeled, four of which were assigned to components of geological dust and the fifth to gasoline vehicle emissions together with battery smelting operations (Engelbrecht and Jayanty 2013). In a study using PMF (version 3.0.1.2), sand dust showed the highest contribution among five estimated sources of PM_2.5 in Kuwait (Alolayan et al. 2013).

Alternatively, the CMB model requires source profiles to determine the contribution of PM’s components. The source profiles could be obtained from site-specific measurements and the use of receptor models, such as PMF, or from previously reported profiles, such as those in the SPECIATE database. SPECIATE is the U.S. Environmental Protection Agency’s (EPA) repository of volatile organic gas and PM speciation profiles of air pollution sources. SPECIATE could be applied as the input to chemical mass balance (CMB) receptor models or as a reference to verify profiles derived from ambient measurements using multivariate receptor models (e.g., factor analysis and positive matrix factorization) (Simon et al. 2010).

In a study, the performance and the uncertainty of ten different RMs were compared with a reference by evaluating the apportioned mass, the number of sources, the chemical profiles, the contribution-to-species, and the time trends of the sources. Approximately 87% of the 344 source contribution estimates (SCEs) reported by participants in 47 different source apportionment model results met the 50% standard uncertainty quality goal. Furthermore, 68% of the SCE uncertainties reported in the results were coherent with the analytical uncertainties in the input data. The most used models, EPA-PMF v.3, PMF2, and EPA-CMB 8.2, presented quite satisfactory performances in the estimation of SCEs. In general, RMs were capable of estimating the contribution of the major pollution sources with a level of accuracy that is in line with the needs of air quality management (Belis et al. 2015).

The present study was designed to determine the contribution of each possible source of total suspended particles (TSP) in Ahvaz’s ambient air using the measured results of 33 elemental and 10 ionic species and three modeling approaches: positive matrix factorization (PMF), the integrated use of PMF and chemical mass balance (CMB), and the integrated use of the SPECIATE database and CMB.

The average concentration of TSP in the study period was 1494.80 μg/m³. The results of elemental and ionic analyses are presented in Tables 1 and 2, respectively. In addition, the composition of TSP is presented in Fig. 1 and is classified into six groups of crustal dust, organic carbon, total ions, elemental carbon, heavy metals, and residuals. A mass closure analysis was performed on the results of each sampling day according to a common method (Ma et al. 2001), and the differences between the measured and calculated TSP concentrations were approximately 4–12%.

Table 1

Summary results of the elemental concentrations of the total suspended particles (TSP) in Ahvaz

Element	Mean	SD	Min.	Max.	Unit
Si	237.5	272.9	9.8	1065.0	µg/m3
Ba	6.7	8.1	0.2	42.0	µg/m3
Be	2.7	63.4	0.1	28.1	µg/m3
Ca	109.5	125.4	5.8	483.0	µg/m3
Fe	24.5	31.3	1.3	137.4	µg/m3
K	30.0	22.1	5.5	97.7	µg/m3
Mg	22.5	26.4	1.7	129.3	µg/m3
Na	81.3	71.2	8.3	294.3	µg/m3
P	4.3	5.7	0.2	25.0	µg/m3
Al	108	122.9	6.7	514.9	µg/m3
Li	0.2	0.2	0.1	0.9	µg/m3
Mn	18.4	22.3	1.7	116.3	µg/m3
As	6.0	5.5	0.1	26.0	ng/m3
Cd	3.93	4.9	0.2	24.9	ng/m3
Co	2.65	3.1	0.4	14.5	ng/m3
Cr	6.0	4.1	1.5	22.9	ng/m3
Cu	8.8	6.7	1.0	34.3	ng/m3
Hg	5.9	4.7	1.0	22.7	ng/m3
Ni	5.1	5.8	0.5	30.7	ng/m3
Pb	8.0	7.0	1.1	30.3	ng/m3
Se	5.8	4.7	0.7	24.0	ng/m3
Sn	6.9	6.9	0.3	28.9	ng/m3
Sr	6.6	5.7	0.4	28.8	ng/m3
Ti	6.4	4.9	0.7	20.4	ng/m3
V	5.6	5.3	0.8	28.0	ng/m3
Zn	33.0	30.0	3.9	129.7	ng/m3
Mo	1.1	1.3	0.1	6.0	ng/m3

Table 2

Summary results of the ionic concentrations of the total suspended particles (TSP) in Ahvaz

Ion	Mean	SD	Min.	Max.	Unit
Na+	9.6	6.56	0.2	28.0	µg/m3
Ca2+	22.3	17.3	3.0	90.0	µg/m3
Mg2+	2.8	2.8	0.2	12.9	µg/m3
K+	2.4	2.2	0.1	9.6	µg/m3
NH4−	5.5	5.3	0.3	27.1	µg/m3
Cl−	17.6	21.7	1.3	118.8	µg/m3
NO3−	30.9	32.3	0	187.5	µg/m3
SO42−	48.7	47.7	4.3	254.3	µg/m3
F−	0.5	0.56	0	3.1	µg/m3
NO2−	0.02	0.02	0	0.1	µg/m3

Open image in new window

The contribution of emission sources of TSP in Ahvaz resulting from all the three approaches are presented in Fig. 2. In addition, TSP concentrations from different sources according to the three approaches are presented in Table 3.

Open image in new window

Table 3

TSP concentrations (μg/m³) from different sources according to the three approaches

Sources	PMF (μg/m3)	PMF and CMB (μg/m3)	SPECIATE and CMB (μg/m3)
Crustal dust	438.8 (± 17.5)	492.7 (± 22.9)	347.2 (± 16.5)
Industrial and mining activities	356.7 (± 21.4)	311.4 (± 21.4)	300.5 (± 21)
Motor vehicles	314.6 (± 31.4)	284.9 (± 32)	261.4 (± 29.8)
Marine aerosols	146 (± 21.9)	165.8 (± 27.7)	184.6 (± 31.2)
Petrochemical activities	57.9 (± 18.5)	136.9 (± 48)	72 (± 25.4)
Road dust	47.7 (± 2.8)	139.8 (± 9.6)	78.5 (± 5.4)
Nondetermined sources	–	–	249.4 (± 30.7)

Input data were validated and entered into the CMB model. For the combination of CMB and PMF (PMF–CMB approach), six profiles derived from PMF, including crustal dust, industrial and mining activities, motor vehicles, marine aerosols, secondary inorganic aerosols, and road dust, were used in the CMB model. For the CMB and SPECIATE approach (SPECIATE–CMB), another profile was added as “nondetermined sources” to the six above, because CMB was not able to reconstruct 100% of measured concentrations.

As shown in Fig. 2 and Table 3, the results of PMF–CMB indicate that crustal dust has the largest contribution to TSP in Ahvaz, followed by industrial and mining activities, motor vehicles, and marine aerosols. Road dust and secondary inorganic aerosols have the smallest contribution among the six sources.

The results of the SPECIATE–CMB approach also show that crustal dust has the largest contribution to TSP mass. However, this value is less than the results of the other two approaches. Industrial and mining activities, motor vehicles, and marine aerosols are in the next ranks, and road dust and secondary inorganic aerosols have the smallest contribution among the seven sources. As mentioned, a factor named “nondetermined sources” was added to the SPECIATE–CMB approach, which was allocated 16.7% of the TSP mass.

The sources of TSP in Ahvaz were estimated and apportioned using three approaches, namely, PMF, PMF, and CMB, and SPECIATE and CMB. The results of elemental and ionic analyses showed that the major portion of particulate matter consists of crustal elements. Shahsavani et al. (2012a, b) conducted a study on the characterization of the ionic composition of TSP and PM₁₀ during the Middle Eastern Dust (MED) storms in Ahvaz and found that crustal ions were most abundant during dust days, whereas secondary ions were dominant during nondust days (Shahsavani et al. 2012a).

Table 7 (Appendix) presents the results of the PMF model. The first factor was influenced by species such as Ca, Al, K, Pb, OC, and EC as well as small amounts of other metals. According to these values and taking into account the suitable designations of other factors, this factor can be named as industrial (metal smelting and rolling, profile pipes, and parts manufacturing) and mining centers (brick, cement, asphalt, etc.). The elements, such as K, Ti, Pb, Cd, and Zn, are introduced as markers of the metal smelting industry (Alleman et al. 2010). High concentrations of organic and elemental carbon are other indicators of industrial activities (Zhang et al. 2013).

The second factor has high percentages of ions and elements, such as NO³⁻, SO₄²⁻, Na, Mn, Ni, and Fe, and can be associated with secondary inorganic aerosols. These results are consistent with those in the study by Alleman et al. (2010). The third factor’s dominant species are Al, Br, Cu, Fe, K, Pb, Ti, Si, and Zn. In fact, the species found in this factor are a combination of elements in the dust of the earth and those caused by human activities (Zhang et al. 2013). According to Lim et al. (2010), the factor with high amounts of Si, Ce, Al, Ba, and Zn can be assumed as road dust originating from mobile transportation on unpaved or sandy roads by the resuspension of precipitated soil particles (Lim et al. 2010). Sowlat et al. (2013) reported that the same species are tracers of road dust and are responsible for 5.5% of the total mass of PM₁₀ (Sowlat et al. 2013). Because of the arid climate of Ahvaz and the existence of deserts and unpaved roads, the aforementioned reasons can be supported.

The fourth factor is mainly influenced by Na and Cl. Other elements, such as K, Mg, and Ca, also play a significant role in this factor. According to previous studies, such as the one by Santoso et al. (2008), the presence of higher levels of Na and Cl represents “marine aerosols,” which can be acceptable in the case of Ahvaz, which is located in the proximity of the Persian Gulf and has a high humidity (Santoso et al. 2008).

Species, such as EC, OC, K⁺, Zn, Mn, and SO₄, are dominant in the fifth factor and are classified as pollutants emitted from human activities. Santoso et al. (2008) have considered Zn, SO₄, and C as appropriate indicators for this factor, which can be denoted the “motor vehicle” category (Santoso et al. 2008). Vehicles have been considered as one of the major sources of OC (Zhang et al. 2013). Species, such as Pb, Zn, Ni, Mn, and OC, are introduced as indicators of motor vehicle emissions (Lee et al. 2010). Ahvaz is one of the most populated cities of Iran, having 1.4 million residents. The increasing number of vehicles and traffic is one of the problems of Ahvaz city (Tabatabaiee and Rahman 2011).

The greatest part of TSP is allocated to the sixth factor, named “crustal dust”. Different studies have introduced various species as its indicators. For example, Alleman et al. (2010) introduced Si, Al, Fe, and Ca, whereas Lim et al. (2010) considered elements, such as Mn, Mg, Ti, and Br for the identification of the crustal dust category (Alleman et al. 2010; Lim et al. 2010). The sixth factor contains considerable amounts of elements, such as Al, Si, Fe, and Br, as well as species, such as K, Mg, Ti, and Mn; therefore, the best designation for this factor can be “crustal dust”.

These results are consistent with a study on source apportionment of TSP in Ahvaz using PMF. Sowlat et al. (2012) showed that crustal dust is responsible for approximately 56% of TSP. Other sources included road dust (7%), motor vehicles (8%), marine aerosols (9%), secondary aerosols (7%), metallurgical plants (4.5%), and petrochemical plants and fossil fuel combustion (8.5%). They also found that crustal dust has higher relative contributions during spring and summer, whereas motor vehicles had higher contributions during fall, winter and weekdays (Sowlat et al. 2012). In another study, research conducted by Sowlat et al. (2013) found that crustal dust is the most dominant source of PM₁₀ mass in Ahvaz. The PMF model identified eight factors: crustal dust (41.5%), road dust (5.5%), motor vehicles (11.5%), marine aerosols (8.0%), secondary aerosols (9.5%), metallurgical plants (6.0%), petrochemical industries and fossil fuel combustion (13.0%), and vegetative burning (5.0%) (Sowlat et al. 2013). The high contribution of crustal dust may be due to the improper management of water resources in Middle East (Moghaddam et al. 2017).

The results of source apportionment of TSP obtained from the PMF, PMF–CMB, and SPECIATE–CMB approaches are presented in Fig. 2 and Table 3 for comparison. Crustal source is determined as the major source of TSP in all three methods. Industrial and mining centers, motor vehicles, and marine aerosols are in the next ranks. Nevertheless, the ranks of the two sources of “secondary inorganic aerosols” and “road dust” are different in these three methods, possibly because of the closeness of their contribution values and occurrence of errors related to the similarity in the source profiles (common pollutants associated with source profiles). The variability in RM results has often been associated with the different logical approaches of the models used and/or the local specificity of the sites, suggesting that the simultaneous application of several RM methods to the same dataset could provide statistically more robust results (Cesari et al. 2016).

However, in this case, because the contributions of source profiles were identified according to classified pollutants’ data, it does not include any category designated “nondetermined sources”. In contrast, the third method (CMB and SPECIATE) contains a “nondetermined sources” category, because the total percentage of all profiles’ data is less than the total data associated with the measured concentration of chemical species and TSP mass. Despite the presence of the “nondetermined sources” category, there is a relatively acceptable consistency among the results of this method and the other two.

Cesari et al. (2016) studied the source apportionment of PM₁₀ using PMF (both 3.0 and 5.0 versions) and CMB in three sites near an industrial area in Italy. The approach determined nine sources: marine, traffic, resuspended dust, biomass burning, secondary sulfate, secondary nitrate, crustal, coal combustion power plant, and harbor-industrial. The intercomparison of PMF and CMB showed significant differences for secondary nitrate, biomass burning, and harbor-industrial sources because of the noncompatibility of these source profiles, which have local specificities (Cesari et al. 2016). Lee et al. (2008) conducted a survey on source apportionment of PM_2.5 using PMF and CMB models in the southeastern United States. The dominant species in the CMB and PMF results were the same. They observed a site-to-site variation in PMF profiles caused by atmospheric processes and local source variability. In comparison, the CMB profiles obtained from a limited number of emission measurements may not be locally representative, even if they are regionally so. Identified or unidentified sources also can cause the differences in the results due to a lack of proper source profiles or proper ‘‘marker’’ species (Lee et al. 2008).

Gianini et al. (2013) conducted a study on source apportionment of PM₁₀, OC, and EC using PMF and CMB models. The authors found some differences in the results obtained from the two models but concluded that method intercomparison can be useful to detect the strengths and the weaknesses of the different methods (Gianini et al. 2013). In another study on the intercomparison of receptor models, Viana et al. (2008) used PCA, PMF, and CMB for a PM₁₀ dataset. The results showed that there is good overall performance by the three models, with good agreement in source identification as well as high correlations between input (CMB) and output (PCA and PMF) source profiles. However, larger differences were found regarding the quantification of source contributions. The authors came to the conclusion that the integrated application of RM methods can solve the limitations of each of the models by constructing a more robust solution based on their strengths. The authors suggested the integrated use of PCA and PMF models to identify and interpret emission sources and obtain a first quantification of their contributions to the PM mass and the subsequent application of CMB (Viana et al. 2008).

Godoy et al. (2005) performed an absolute principal component analysis (APCA) to quantify the contribution of possible sources of aerosols in a Brazilian site. They compared the APCA obtained profiles with SPECIATE 3.1 profiles and observed good similarity between them (Godoy et al. 2005). Taiwo et al. (2014) compared PMF obtained profiles with the USEPA SPECIATE database for different processes of iron- and steel-making and showed that there are several differences between these two types of profiles, possibly because of differing materials’ inflow at the steelworks processing units and/or the application of air pollution control systems in the plants. These discussions indicate that even with PMF modeling, it is preferable to obtain the profiles from on-site measurements to support the choice of factors and tracer elements. In addition, the differences may be due to the gap in time between the study period and the time of SPECIATE profile measurements, as observed in the case of Pb in SPECIATE. The application of SPECIATE-derived profiles is found to be reasonable in areas without sufficient on-site measurements (Taiwo et al. 2014).

On occasions without adequate knowledge of emission characteristics (i.e., source profiles) of local sources, PMF is a more appropriate method for source apportionment than CMB because CMB source apportionment results are sensitive to the source profiles used. In contrast, in situations with sufficient data, the CMB model can yield more satisfactory and precise results. However, it is difficult to assert that one method is superior to the other, given the uncertainties discussed previously and the lack of elaborate studies and a priori knowledge of the emission characteristics of all the major local sources at a site (Lee et al. 2008). Limitations of PMF can appear when source emissions have a strong temporal correlation or when meteorological factors have a significant impact on PM variability. The use of PMF in these situations can mix the source profiles and consequently under- or overestimate the real-world sources. The application of CMB models can be interrupted by the lack of availability of source profiles and the nonrepresentativeness of the available profiles for local source emissions (Gianini et al. 2013). In another study, authors reported that PCA and PMF models were able to identify the origin of certain elements and were more efficient in computation (input data and software requirements, especially PMF). In contrast, CMB could determine a larger number of sources and had the highest performance when experimental source profiles were used (Viana et al. 2008).

In conclusion, despite the different contributions of sources, there is a noticeable consistency between the results of these methods. Furthermore, because of the proven performance of the combined RM methods in previous studies and the presence of sufficient data on the number of species and samples, the results of the PMF–CMB approach are possibly the most realistic among those of the three applied approaches.

The present study was designed to determine the contribution of each possible source of total suspended particles (TSP) in Ahvaz’s ambient air using three approaches: PMF, PMF–CMB, and SPECIATE–CMB. Crustal source was determined to be the major source of TSP in all three methods. Industrial and mining centers, motor vehicles, and marine aerosols were in the next ranks. Nevertheless, the ranks of the two sources of “secondary inorganic aerosols” and “road dust” were different in these three methods, possibly because of the closeness of their contribution values and the occurrence of errors related to the similarity in the source profiles. These results were consistent with another study on source apportionment of TSP in Ahvaz. Despite the different contributions of the sources, a noticeable consistency was found between the results of these methods. Furthermore, because of the proven performance of combined RM methods in previous studies and the presence of sufficient data on the number of species and samples, the results of the PMF–CMB approach are possibly the most realistic among the three applied approaches.

The datasets used and/or analyzed during the current study are available within the article body and supplementary files; the materials also are available from the corresponding author upon reasonable request.