Organic aerosols (OAs), which consist of thousands of complex compounds emitted from various sources, constitute one of the major components of fine particulate matter. The traditional positive matrix factorization (PMF) method often apportions aerosol mass spectrometer (AMS) organic datasets into less meaningful or mixed factors, especially in complex urban cases. In this study, an improved source apportionment method using a bilinear model of the multilinear engine (ME-2) was applied to OAs collected during the heavily polluted season from two Chinese megacities located in the north and south with an Aerodyne high-resolution aerosol mass spectrometer (HR-ToF-AMS). We applied a rather novel procedure for utilization of prior information and selecting optimal solutions, which does not necessarily depend on other studies. Ultimately, six reasonable factors were clearly resolved and quantified for both sites by constraining one or more factors: hydrocarbon-like OA (HOA), cooking-related OA (COA), biomass burning OA (BBOA), coal combustion (CCOA), less-oxidized oxygenated OA (LO-OOA) and more-oxidized oxygenated OA (MO-OOA). In comparison, the traditional PMF method could not effectively resolve the appropriate factors, e.g., BBOA and CCOA, in the solutions. Moreover, coal combustion and traffic emissions were determined to be primarily responsible for the concentrations of PAHs and BC, respectively, through the regression analyses of the ME-2 results.
Atmospheric aerosols are generating increasing interest due to their adverse effects on human health, visibility and the climate (IPCC, 2013; Pope and Dockery, 2006). Among different particulate compositions, many studies focus on organic aerosols (OAs) because they contribute 20–90 % to the total submicron mass (Jimenez et al., 2009; Zhang et al., 2007). OAs can be either directly emitted by various sources, including anthropogenic (i.e., traffic and combustion activities) and biogenic sources, or produced via secondary formation after the oxidation of volatile organic compounds (VOCs) (Hallquist et al., 2009). Therefore, reliable source identification and quantification of OAs are essential before developing effective political abatement strategies.
Aerodyne aerosol mass spectrometer (AMS) systems are the most widely adopted online aerosol measurement systems for acquiring aerosol chemical compositions (Canagaratna et al., 2007; Pratt and Prather, 2012). An AMS provides online quantitative mass spectra of non-refractory components from the submicron aerosol fraction with a high temporal resolution (i.e., seconds to minutes) (Canagaratna et al., 2007). The total mass spectra can be assigned to both several inorganic compounds and the organic fraction through mass spectral fragmentation tables (Allan et al., 2004). To further investigate the different types of organic fractions, numerous studies have exploited the positive matrix factorization (PMF) algorithm and apportioned the AMS organic mass spectra in terms of their source emissions or formation processes (Zhang et al., 2011). PMF is a standard multivariate factor analysis tool (Paatero, 1999; Paatero and Tapper, 1994) that models the time series of measured organic mass spectra as a linear combination of positive factor profiles and their respective time series. Most of the earlier PMF studies were conducted on unit-mass resolution (UMR) mass spectrometers (Lanz et al., 2007, 2010; Ulbrich et al., 2009), although more have recently focused on high-resolution (HR) mass spectra PMF (Aiken et al., 2009; Docherty et al., 2008; Huang et al., 2010). The use of HR mass spectra data to constrain PMF solutions can reduce their rotational ambiguity and result in more interpretable OA factors. For example, Aiken et al. (2009) found that hydrocarbon-like OA (HOA) and biomass burning OA (BBOA) were better separated using HR-AMS data than with UMR data. However, even HR-AMS-PMF can also yield mixed factors (especially in heavily polluted areas) due to their complex emission patterns.
The abundant characteristic fragments for cooking-related OA (COA) (e.g.,
In this study, a novel source apportionment technique using the multilinear engine tool (ME-2) was successfully applied to organic mass spectra obtained with an HR-ToF-AMS at two urban sites during pollution-heavy periods during the same year. The improved OA source apportionment results are discussed and compared with an unconstrained PMF analysis.
Measurements at Qingdao (36.10
An HR-ToF-AMS was deployed for the online measurement of non-refractory
PM
The locations and the average PM
A routine analysis of the HR-ToF-AMS data was performed using the software
SQUIRREL (version 1.57) and PIKA (version 1.16) written in Igor Pro 6.37
(Wave Metrics Inc.) (
PMF is a mathematical technique used to solve bilinear unmixing problems
(Paatero and Tapper, 1994) that enables a description of the
variability of a multivariate database as the linear combination of static
factor profiles and their corresponding time series. The bilinear factor
analytic model in matrix notation is defined in Eq. (1), where the measured
matrix
In this study, PAH mass concentrations were quantitatively determined from
the HR-AMS data. The steps outlined are as follows: first, the PAH
molecular ions [M]
In this section, a conventional PMF without any prior information is performed to analyze the OA source apportionment. Then, we use the ME-2 method to optimize the OA source apportionment based on the information obtained from the PMF method.
We performed unconstrained runs with a range from 2 to 10 factors.
Generally, PMF solutions with large numbers of factors are not considered
due to possible mathematical splits of the factor profiles. However, some
factors that have small contributions or that have similar mass profiles as
other factors (but different time series) may only be found in solutions
with large numbers of factors. We observe that most of the solutions
provided via PMF include either multiply split factors or mixed factors that
are not properly separated from one another. In other words, PMF does not
produce an appropriate solution. The six-factor solutions for Qingdao and
Dongguan are shown in Fig. S1 and S2 in the Supplement, and three types of primary OAs
(POAs) were identified for each sampling site, including HOA), coal
combustion OA (CCOA) and cooking OA (COA) for Qingdao and HOA, biomass
burning OA (BBOA) and COA for Dongguan. Oxygenated OA (OOA) seems to be
excessively split in the six-factor solutions for both of the sites. HOA is
distinguished by alkyl fragment signatures with prominent contributions of
The anchor mass spectra for
Before operating ME-2, feasible and reasonable prior input profiles must be
determined. To the best of our knowledge, this is the first HR-OA dataset
that employs anchor profiles extracted from an unconstrained PMF solution
with a higher number of factors, and the same approach has been successfully
applied to source apportionment efforts using UMR ME-2 (Fröhlich et
al., 2015). In our case for Qingdao, the BBOA anchor profile should be
investigated, and we attempted to look for it from the unconstrained PMF
results based on the same dataset and found that the BBOA factors in the seven-
and eight-factor solutions might be used as the anchor profiles. They both had
good correlation with the BBOA MS in Chinese biomass burning emission
simulation (He et al., 2010), confirming their
basic BBOA characteristics. Although these two BBOA factors are quite
similar, the BBOA from the eight-factor solution is better suited to be a
constraining profile due to its smaller
Mass spectra of the OA factors, average fractions of the OA factors,
diurnal variations of the OA factors and time series of the OA factors
identified by the ME-2 method for
According to the unconstrained PMF results, the best interpretable results
for both two sites are the six-factor solutions with factors that include
HOA, COA, BBOA, CCOA, less-oxidized oxygenated OA (LO-OOA) and more-oxidized
oxygenated OA (MO-OOA) (Fig. 3a and b). As inputs, we constrain the
In this study, we used two simple and reasonable criteria to obtain a better
environmental OA source apportionment: the reasonability of the O
In order to prove the improvement of using the anchor profiles generated by the unconstrained PMF run with the same local datasets, we also run the ME-2 analysis using the anchor profiles available in the literature, with the results shown in Table S5 and S6. For Qingdao, the correlations between POAs and their tracers and the Q/Qexp values using the three BBOA profiles in the literature are poorer than using the BBOA obtained in this study (Table S5). For Dongguan, the results from ME-2 using the HOA profiles in the literature are also poorer than using the HOA profiles obtained in this study (Table S6). Therefore, it can be clearly seen that the method to obtain an anchor profile in this study is easier (it does not depend on the results in the literature) and more valid.
Figure 1 shows the chemical compounds of PM
For Qingdao, the final result is the average of all of the ME-2 runs, with
constraints including
For Dongguan, similar to the OA source apportionment using ME-2 in Qingdao,
the final result is the average of two accepted
Meteorological conditions (especially wind) play a crucial role in the dilution and transport of air pollution. We used the relationships between the component concentrations and wind to profoundly understand the origins of the OA factors and their nature. The distributions of the OA factor concentrations versus the wind direction and speed are plotted in Fig. S7. For both of the urban sites, higher mass concentrations of the POA factors were mostly accompanied by low wind speeds, denoting their local emission characteristics. Additionally, for the OOA factors, a large proportion of their higher concentrations were maintained at higher wind speeds, indicating that the OOAs were formed by transport processes. However, the small fraction of high-level OOAs that was concentrated within the low wind-speed region represents the fast formation of OOAs from some local POA.
BC and PAHs are mainly derived from incomplete combustion processes (Schmidt and Noack, 2000; White, 1985), and thus they were used as tracers for the POAs. In this study, the BC was directly measured by the AE-31, and the PAHs were quantified using the method developed by Bruns et al. (2015) based on AMS data. Both the BC and PAHs showed pronounced diurnal cycles similar to those of the POAs (see Fig. S8). In addition, POAs are properly split into different subtypes via the ME-2 method, thereby providing the possibility to better understand the contributions of different POAs to BC and PAHs and to verify the POA source identification. In this section, we use a multilinear regression method to analyze the POA factors for BC and PAHs.
Figure 4 shows the average contributions of OA sources to BC and PAHs in Qingdao and Dongguan. At both sites, HOAs were the dominant attribute of BC (51 % for Qingdao and 40 % for Dongguan) and CCOAs contributed the most to the PAHs (59 % for Qingdao and 43 % for Dongguan), indicating that BC mainly originates from traffic emissions and that PAHs in the Chinese urban polluted atmosphere are dominated by coal combustion during the wintertime. These findings are consistent with results reported in similar studies (Elser et al., 2016; Huang et al., 2015, 2010; Sun et al., 2016; Xu et al., 2014; Zhang et al., 2008). Moreover, the ratio of PAHs to OAs (1.8 %) in Qingdao was similar to that in the northern Chinese urban site of Xi'an (1.9 %) (Elser et al., 2016) but was higher than that in Dongguan (0.9 %). This is likely because a larger fraction of coal combustion to the total OA concentration would enhance the ratio of PAHs to OAs (Elser et al., 2016). Biomass burning was the second-most important source for both BC and PAHs; it was responsible for 33 and 29 % of the BC at Qingdao and Dongguan, respectively, and for 29 and 34 % of the PAHs at Qingdao and Dongguan, respectively. Cooking emissions were a minor source of BC and PAHs, accounting for less than 10 %. These results are also consistent with the published findings. For example, biomass burning is an important source for BC (Kondo et al., 2011; Reddy et al., 2002) and, in some regions with fewer traffic emissions, BC has the best correlation with BBOAs (Schwarz et al., 2008). In addition, in Beijing and California, PAHs are correlated well with BBOAs but are much more weakly correlated with COAs (Ge et al., 2012; Hu et al., 2016; Sun et al., 2016).
In this study, we used PMF to interpret the organic aerosol sources at two
Chinese urban sites in winter, and found that PMF did not work properly
(i.e., it did not allow for the separation of several primary sources of
OAs). Therefore, we adopted the ME-2 approach, which yields more reliable
solutions. Technically, there are three important steps when using the ME-2
method to interpret the sources of OAs. The first step is to investigate the
mixed and unidentified factors that are constrained according to issues in
the unconstrained PMF results. Generally, we constrained one or more POA
factors (i.e., HOA, COA, BBOA and CCOA) for the polluted urban sites. The
second step is to search for a reasonable anchor profile for each
constrained factor. Two approaches were used: searching for anchor profiles
via an increase in the number of unconstrained PMF factors from the same
dataset and using mass profiles derived from other similar studies. The
third step is to choose the criteria for obtaining the optimal results. The
choice of a reasonable range of O
Data are available by contacting the corresponding author at huangxf@pku.edu.cn.
The supplement related to this article is available online at:
The authors declare that they have no conflict of interest.
This work was supported by the National Natural Science Foundation of China (91744202, 41622304) and the Science and Technology Plan of Shenzhen Municipality (JCYJ20170412150626172, JCYJ20170306164713148). Edited by: Mingjin Tang Reviewed by: two anonymous referees