This study was conducted in order to identify possible sources and to estimate their contribution to particulate matter in a rural area. For this purpose, a commonly used receptor model, positive matrix factorization (PMF), was applied to a PM2.5 data set collected in a rural area of Madrid (Spain) between May 2004 and April 2005. A total of eighty nine samples were gathered. Chemical composition of particulate matter including major components, trace elements, total carbon, alkanes, PAHs, alcohols and acids were analyzed to study sources of atmospheric aerosols using the positive matrix factorization model. Present work is characterized by including some organic tracers within PMF analysis, through which we can get a more accurate source apportionment. To our knowledge, present work is the first one, which employing organic tracers, reported a source apportionment by PMF model in a rural area of Spain. PMF analyses of 31 different combinations of input species were performed to obtain a suitable solution. To link PMF factors with possible sources, authors directly compared factor profiles with source profiles. PMF apportioned the PM2.5 mass into nine factors (three were secondary and six primary emissions) identified as wax plants emissions, combustion processes, secondary nitrate, secondary sulphate, secondary organic aerosol, cooking process, uncompleted fossil fuel combustion, dust soil and biogenic emissions.
Keywords: Atmospheric aerosol; organic aerosol; PM2.5; PMF; source apportionment.
1. Introduction
Organic fraction is a major component of atmospheric particulate matter, being more significant in rural areas. As a result, during last years the number of works related to organic faction has increased significantly. This increment was due to the growing concern about atmospheric aerosol because their hazardous effects on human health (Arden Pope III et al., 2002; Schwartz, 2002; WHO, 2003) and as far as its role in climate change (IPCC, 2001).
Organic fraction of atmospheric aerosol includes thousand of compounds, being some of them more studied than others. In fact, alkanes and PAHs have been widely studied and there are numerous worldwide works concerning them since many years ago (Lao et al., 1973; Lee et al., 1976; Simoneit and Mazurek, 1982; Méndez et al., 1993). Over the years, studies become widen, characterizing a large number of compounds (Gogou et al., 1998; Pío et al., 2001a; Guo et al., 2004). Currently, some authors have characterized more than 300 organic compounds associated to particulate matter (Oliveira et al., 2007).
Since last years, the organic fraction of aerosols has been analyzed extensively, which identified a greater number of individually species. However, knowing composition is not enough to overcome hazardous effects related to atmospheric aerosols, it would be necessary to control emissions in origin. Nowadays there have been identified many sources of organic aerosols. Among them, anthropogenic sources include fossil fuel combustion, coal and wood burning, meat cooking, cigarette smoke, agriculture debris and resuspension of soil particles, while emission of plants wax, fungi, bacteria, pollen, algae and natural combustion processes as wildfires and volcano eruptions are the main natural sources (Bi et al., 2002; Brown et al., 2002; Cincinelli et al., 2003; Bi et al., 2005).
Many studies have been conducted using various statistical methods, such as factor analysis (FA), principal component analysis (PCA), chemical mass balance (CMB) and lately, positive matrix factorization (PMF) with the only aim of identifying sources of particulate matter. Each method requires different degrees of knowledge about the number of sources and the source profiles at the chosen site. One of the most important advantages of PMF is that it can identify particulate matter sources and provide the contribution of each source in absence of prior information on sources. Other advantage of PMF is that identification and quantification of sources are relatively faster than other models such as CMB and PCA, while the main disadvantage of PMF is the slower selection of the input files due to uncertainties calculation.
In light of the above, the aim of this study was double: firstly to characterize the particulate fraction of atmospheric aerosol and secondly to identify sources of PM in a rural area according to chemical characterization. This paper presents an analysis of chemical composition of 89 samples of PM2.5 collected from April 2004 to April 2005. This is the first study accomplished in this area and as results there is no prior information about sources. For this reason, PMF model was chosen. Thus, measurement results of 5 ion components, 11 metals, total carbon (TC), particulate matter (PM) and several organic compounds such as 24 alkanes, 11 PAHs, 18 alcohols and 27 organic acids have been used in order to estimate the fine aerosol sources and their contributions. Also, meteorological parameters and gaseous species were included to assist interpretation of the source factors.
2. Experimental
2.2. Samples collection and analysis
Sampling took place in Chapinería (altitude: 675 meters over sea level; latitude: 40° 22' 45'' North; longitude: 4° 12' 15'' West). Chapinería is a little town situated 50 kilometres from Madrid, has less than 2000 in habitants and there are not significant industrial activities around it. This area is surrounded by Quercus ilex forest and is influenced by long-range transport processes of desert dust from North Africa. For these reason, this site is considered a rural area
Particulate matter was collected using a high volume sampler with quartz filters, previously baked out. For each sample, PM2.5 mass concentrations and chemical composition were obtained.
The analysis was performed separately for inorganic and organic fraction. The analytical procedures followed for the analysis of these species has been already published (Pindado et al., 2009). Briefly, ion components, such as SO42-, NO3-, NO2-, Cl-, Na+, K+, NH4+, Ca2+ and Mg2+ were leached with water and them analyzed by ion chromatography (IC). On the other hand, an acid digestion with HNO3 and HCl were performed and 24 trace elements were analyzed by Inductively Coupled Plasma (ICP/MS) and Atomic Absorption. The TC was determined by carbon elemental analysis by combustion. In order to analyze organic composition, filters were Soxhlet extracted with a mixture of CH2Cl2/Acetone 3:1, extracts were purified via sequential elution though a glass column packed with 1.5 g of silica gel using solvent of increasing polarity. Four fractions were eluted to provide a separate chromatographic analyses; alkanes, alcohols and acids were subject to gas chromatographic-mass spectrometric (GC-MS) analysis, performing a derivatization with N,O-bis-(thrimethylsilyl)-trifluoracetamide (BSTFA) for polar compounds, meanwhile PAHs were submitted to high performance liquid chromatographic (HPLC) with fluorescence detector. Also, two gaseous species, NOx and Ozone, were measured and included into PMF analysis.
Uncertainties associated to organic compounds were individually calculated, being uncertainty combined ranging from 9 to 18 %.(Pindado et al., 2010). Uncertainties of inorganic compounds were estimated as measurement errors.
2.3. PMF description
The PMF model was developed by Paatero at the University of Helsinki in Finland in the mid 1990' (Paatero, 1997; Paatero, 1999). PMF assumes that concentrations at receptor sites are impacted by linear combinations of source emissions, which are derived as factors in the model. Thus, model assumes there are p sources impacting a receptor, and linear combinations of the impacts from the p sources give rise to the observed concentrations of the various species. Mathematically can be written as,
where Xij is the (i x j) matrix of ambient concentrations of j species on the i days, Gik is the (i x k) matrix of sources contributions of k factors on i days, Fkj is the (k x j) matrix of source profiles of k factors that is species j, and Eij is the (i x j) matrix of residuals not fitted by the model. The task of PMF model is to minimize the function Q using constrained, weighted least-squares. This function is defined as:
where Sij is an uncertainty estimate in the j species measured on the i days.
To perform the PMF model, a qualitative knowledge of the sources is only required, however PMF model also has limitations: source interpretation may be relatively subjective and the inability to clearly separate covariant sources.
PMF requires two input files, one file with the concentrations and one with the uncertainties associated with those concentrations. For this reason, the preparation of input files is time consuming due to formatting of the data to obtain the required input matrices. Choosing of modelling parameters and number of factors is not straightforward and is still largely by the experience of authors (Paatero and Hopke, 2003).
EPA PMF1.1 was used for the current analysis. This first version of the model determines signal-to-noise ratio (s/n) statistics for every input species. Moreover, the model generates regression diagnostics, including the intercept, slope, root mean squared error and r-square. It is hopefully s/n statistics are higher for all input species, intercept near 0, slope near 1 and r-squared larger than 0.6. All this information will help to optimize the solution.
PMF factors reveal which species temporally co-vary. Since the temporal variability of pollutant concentrations is not solely determined by changes in emissions, we should not link PMF factors to source profiles directly, although many studies refer to the PMF factors as sources.
3. Results and discussions
Measurements were taken between April 2004 and July 2005 in a rural area of Madrid. A total of 89 diary samples were collected. More than 90 organic compounds, including alkanes, PAHs, alcohols and acids, were separately determined using chromatography techniques. In addition the inorganic fraction (main ionic components and metals) were quantified. In order to simplify our analysis, some compounds were not included in the PMF analysis. Species whose values are often below detection limit were excluded; meanwhile some compounds were grouped according to the categorization previously made by other authors (Shrivastava et al., 2007). Thus, finally we chose 66 species to PMF analysis. Table 1 summarizes average concentrations of chemical components of fine particles from a year-long study included in the model.
Among the 66 species chosen, 14 were removed from the computation because they were frequently present at concentrations below the detection limit and 6 species were considered "weak". Hence, fifty-two variables have been chosen to develop PMF, including alkanes, PAHs, alcohols, saturated acids, unsaturated acids, α-pinene degradation products, metals, main inorganic ions, gases species, total carbon and PM2.5.
Table 1
3.1. PMF results
Different combinations of input data were tested. PMF analysis examined several solutions, by increasing the number of allowed factors; from 4 to 10. The missing data, which were 6 % in this study, were established as the geometric mean of all the concentrations measured for each species and uncertainty was set 4 times the geometric mean (Polissar et al., 1998). The final choice was based on the evaluation of the resulting sources profiles as well as the quality of the chemical species fits. Current work presents the optimum solution for 52 input dataset according statistics. Thus, a 9-factor solution was proved to be the best choice, in terms of both quality of the fit and physical sence for the studies system. The statistical of each of the 52 input datasets is listed in Table 2.
Table 2
The distribution factors and contributions for all studied compounds by the PMF model are presented in Figure 1 and 2, respectively. In order to identify source types, the resolved sources profiles from PMF analysis were compared with known profiles obtained from previous works (Rogge et al., 1998; Schauer et al., 2001; Hays et al., 2002; Zheng et al., 2002; Landis et al., 2007; Shrivastava et al., 2007).
Figure 1 and 2
It must be remarked that not all the factors specifically represent individual sources. According to our results, among the obtained factors, there are three identified as secondary factors (factor # 3, 4 and 5) and six identified as primary emissions (factor # 1, 2, 6, 7, 8 and 9). The resolved sources include not only natural sources such as biogenic emissions, wax plants, and dust soil, but also anthropogenic sources such as combustion process, cooking, and traffic. Moreover, secondary aerosol sources were separated in the form of secondary nitrate, secondary sulphate, and secondary organic aerosol.
3.1.1. Interpretation of factor profiles
The first factor characterized by alcohols C24, C26, C28, C30 and acids C24, C26 and C28 has been identified as "Wax plants". These key species have been associated with primary biogenic emissions such as plant waxes by several authors (Simoneit, 1989). This factor explains 60 % of the alcohols higher than 20 atoms of carbon and 90 % of the fatty acids with more than 20 atoms of carbon. It is well known that odd alkanes are associated with biogenic emissions; mainly they have been related to plant waxes. Thus the slight contribution of odd alkanes suggests wax plant emissions. Time series of contributions show a clear seasonal behaviour with higher autumn and winter values, due to leaf fall caused by wind abrasion. Furthermore, this factor showed similar pattern for weekdays and weekends, being consistent with biogenic emissions.
Second factor grouped 50 % - 90 % of aliphatic hydrocarbons between n-C24 and n-C33 atoms of carbon. Alkanes have been associated to any combustion processes. The rest of variables comprise less than 10 %, uniquely is remarkable 20 % of NOx involved in this factor. NOx is related with all combustion process. These results lead us to associate this factor with a combustion process. Also this factor shown a slightly weekly pattern, higher values are in weekdays implied an anthropogenic origin. Also, the time series show higher contributions during winter. This may reveal a residential burning contribution (Karanasiou et al., 2009).
Factor #3 grouped the highest contributions of NO3- and NH4+ so it was identified as secondary nitrate. The secondary nitrate particles contain high concentrations of NO3- and NH4+. Also NOx, which is a precursor of nitrate, has a large contribution to this factor. The main source of NOx in the atmosphere is traffic and stationary sources. The secondary nitrate do not shows a weekly variation, which may indicate a relationship with traffic emissions. Moreover, secondary nitrate showed an important content of TC, which could be attributed to the semi-volatile organic compounds condensing onto surface of ammonium nitrate particles (Amato et al., 2009). In addition, the temporal evolution shows higher concentrations in winter, when low temperature and high relative humidity help the formation of secondary nitrate particles (Kim and Hopke, 2008; Nicolas et al., 2008). In our study, secondary nitrate showed the highest contribution during 22/Nov and 12-13/Jan. Within these days, in addition to humidity was greater than 70 % and temperature was low, there is an accumulated period of atmospheric pollutants.
The species contributing to fourth factor are sulphate, nitrate, Na, Cl and ozone. Ion sulphate is formed through different oxidation reactions of SO2, which is released to the atmosphere by several combustion processes. This factor has also shown a high contribution for ozone, suggesting sulphur dioxide can react with ozone to create sulphate. Typically, secondary sulphate tends to be abundant in warmer days due to the fact it is formed by photochemical oxidation of SO2. Thus, the secondary sulphate shows seasonal variations with higher concentrations in summer when photochemical activity is highest. Is well known secondary sulphate is a tracer of long-range transport, so presence of NaCl, which is related to marine aerosol, confirms there is a regional transport. For these reasons, formation of secondary sulphate particles may have formed elsewhere and were transported to the sampling area. The age of the aerosol has been evaluated through the parameter C18/C18:1 by several authors (Guo et al., 2003; He et al., 2006; Oliveira et al., 2007). Elevated values of this parameter indicate that the aerosol has been issued in other areas and has suffered transport processes. Values calculated at Chapinería ranged between 0.5 and 33.3 which confirms a transport phenomenon.
The fifth factor explains 80 % of the α-pinene products degradation and unsaturated acids. The three measured species categorized as α-pinene products degradation were cis-pinonic acid, trans-norpinonic acid and pinic acid. These compounds are components of SOA and were measured in smog chambers from the oxidation of monoterpenes (Fick et al., 2004). Unsaturated acids involve oleic, linoleic and azelaic acid. The species associated with this factor are related to organic secondary components; therefore we can conclude that this factor represent the secondary aerosol. This factor showed a slightly seasonal pattern, exhibiting higher contribution during summer months. This is in agreement with the known tending of SOA to be mainly formed during warm days, when there are higher temperatures that encourage atmospheric reactions and thus secondary compounds formation.
The sixth factor was dominated by palmitic and stearic acids. Both compounds are elevated in source profiles for seed oil cooking and for meat cooking (Dutton et al., 2010). The occurrence of NaCl in this factor may also confirm that it was coming from cooking emissions. Also, this factor does not show a clearly seasonality, so is agree with relating it with cooking emissions.
Factor 7 describes 60 % - 80 % of most PAHs included in the model and a contribution of 15 % of NOx and K is also explained in this factor. PAHs are associated to uncompleted fossil fuel combustion meanwhile NOx and K are also related with combustion processes. In according with these results, factor 7 is related to combustion processes, same as factor 2. However, the model has separated both factors, because the factor 7 had a strong weekly variability, with higher values during the days of the week, a trend not showed in factor 2 clearly. This week variability suggests anthropogenic origin and can be attributed to emissions produced by traffic. The ratio of BaA/(BaA+Chry) was 0.31, which is very close to these ratios reported in previous studies: 0.27 in diesel vehicles (Moon et al., 2008) and 0.40 in gasoline vehicle (Kavouras et al., 1999).This factor can not differentiate between exhaust from diesel or gasoline motors, given that the first contained higher amounts of lower molecular weight PAHs whereas the gasoline motors higher molecular weight compounds were more abundant (Aynul Bari et al., 2009).
The eighth factor grouped Ca, Mg, Al, Fe, Sr and Ba. The percentage apportioned for these species to this factor ranged from 50 % to 80 %. All species associated with this factor are related to crustal material. This factor showed a contribution six times higher than mean value in three periods; September - October 2004 and March 2005. During these days, three long-range transport processes of desert dust from Sahara were registered (Gómez-Comino and Artiñano, 2006; Salvador et al., 2008). These dust transported from Sahara Desert have been extensively analysed (Nicolás et al., 2008; Amato et al., 2009; Viana et al., 2010) In light of the above, factor #8 has been allocated as dust soil.
Factor #9 explains 50 - 70 % of the alcohols and acids with less than 20 atoms of carbon. According to previous works, these alcohols are associated with microbiological emissions (Zheng et al., 2000; Alves et al., 2001; Pio et al., 2001b); meanwhile acids with less than 20 atoms of carbon are linked with microbiological emissions as well as anthropogenic emissions such as cooking (Rogge et al., 1993; Pio et al., 2001a; Schauer et al., 2001; Almeida Azevedo et al., 2002; Hays et al., 2002; Radzi Bin Abas et al., 2004; Hays et al., 2005; Oliveira et al., 2007). However, factor#6 grouped cooking emissions, and therefore the hypothesis that factor #9 represents mainly a biogenic source is very likely.
3.1.2. Model performance
The ability of the PMF model to reproduce the measured PM2.5 concentration was performed by comparison between concentrations measured and modelled. The 89 samples of concentrations measured vs. modelled were plotted in Figure 3. The correlation coefficient was 0,99, indicating that the 9 factor solution account for the variation in mass concentration of species. The slope of the correlation curve was 1.02, which implies the PMF model overestimates slightly concentrations.
Figure 3
4. Conclusions
Chemical composition data of fine aerosol collected from May 2004 to April 2005 in Chapinería (Spain) were analyzed by the PMF method to estimate and evaluate the contributions of possible emissions sources. Chemical characterization included five major ions, 11 species of trace elements, TC, more than 90 organic compounds, including alkanes, PAHs, alcohols and acids. Inclusion of inorganic components as well as carbonaceous fraction gives better model fit. Comparisons with source profiles were used to associate PMF factors with specific source classes. As result, nine factors were identified including wax plants, combustion processes, secondary nitrate, secondary sulphate, secondary organic aerosol, cooking process, uncompleted fossil fuel combustion, dust soil and biogenic emissions. Six of these factors are related to primary emissions and three of them are categorized as secondary aerosol. In light of these results, although the sampling area was considered a rural area, a clear influence of anthropogenic emissions was found, which points clearly Chapinería as a semi rural area.
This work exhibited the usefulness of PMF to identify and interpret emission sources, although some evidence of uncertainties remains and further research is needed to ensure that sources identified are robust enough.
