Chemical Characteristics and Source Apportionment by Two Receptor Models of Size-segregated Aerosols in an Emerging Megacity in China

PM2.5, PM2.5-10, and PM10 samples were collected in Zhengzhou in 2014 to examine the chemical characteristics and sources of aerosols in this area. The PM concentrations, nine water soluble inorganic ions, organic carbon, elemental carbon, and twenty-two elements were determined, and positive matrix factorization (PMF) and chemical mass balance (CMB) were used for source apportionments. The meteorological impact was also evaluated by back-trajectory cluster analysis. Severe PM pollution was present in the study area, and the aerosol concentrations of PM2.5 samples (92%) and PM10 samples (85%) significantly exceeded the recommended levels of the Chinese National Ambient Air Quality Standard (NAAQS), with the average annual mass concentrations of PM2.5 and PM10 reaching 187 and 281 μg m, respectively. Secondary inorganic aerosols were the major ions in PM and accounted for 36%, 10%, and 27% of PM2.5, PM2.5-10, and PM10, respectively. The annual concentration of As (0.029 μg m) and Cd (0.010 μg m) in PM10 also exceeded the Chinese NAAQS levels, indicating a high health risk. Results from source apportionment by PMF modelling indicated that dust, vehicular traffic, coal combustion, secondary aerosols, and industry were the main pollution sources, accounting for 13.1%, 14.1%, 16.1%, 35.8%, and 14.6% of PM2.5; 25.1%, 20.8%, 21.8%, 10.5%, and 11.6% of PM2.5-10; and 19.8%, 15.8%, 18.5%, 22.5%, and 13.5% of PM10, respectively. Dust sources played an important role in PM pollution, especially coarse particles; however, secondary aerosol sources contributed the most to PM2.5. Both of these observations were consistent with the results of mass reconstruction of the size-segregated aerosols. The CMB results coincided with the PMF results for PM2.5. Cluster analysis showed that air quality in the study area across the four seasons was mainly affected by air masses from the northeast and the east.


INTRODUCTION
Through transport and dispersion, air pollution influences climate and weather patterns at relatively large spatial scales (100-1000 km) (Hadley, 2017).Especially, PM pollution has become a common environmental problem in megacities worldwide and is attracting considerable attention in different research fields (Chen et al., 2003;Kang et al., 2004;Sun et al., 2006;Zhang et al., 2013a).This pollution type is closely related to particle constituents and meteorological conditions, and holds important implications for human health, visibility, economy, weather, and the global climate (Charlson et al., 1992;Chameides et al., 1999;Ramanathan et al., 2001).The climatic and health effects of size-segregated aerosols vary considerably because of different particle sizes (Wang et al., 2012).Among the particles, PM 10 and PM 2.5 are widely studied.
Understanding of mass concentrations, chemical compositions, and sources is essential to reduce PM pollution.Atmospheric PM is a complex mixture of both primary and secondary particle species, including water-soluble inorganic ions (WSIIs), elements, organic carbon (OC), and elemental carbon (EC).Studying the chemical compositions of PM is important in assessing the effects of PM on air quality and human health and for conducting source apportionment (Tao et al., 2013).Understanding the sources and their contributions to atmospheric PM formation is also important to mitigate air pollution.Mathematical models, together with PM sampling and chemical characterization, have been shown to be extremely useful in order to interpret and understand the geographical distribution, temporal evolution, and origin of pollutants.Receptor models, particularly positive matrix factorization (PMF) and chemical mass balance (CMB), are advanced source apportionment methods for successfully assessing particle source contributions and have been applied in numerous locations worldwide (Xie et al., 1999;Kim et al., 2003;Gianini et al., 2013;Shi et al., 2014;Manousakas et al., 2015;Contini et al., 2016).
Fig. 1 shows the 2014-based annual average Moderate Resolution Imaging Spectroradiometer Terra Deep Blue aerosol optical depth (AOD) at 550 nm over the entirety of China (a) and Henan Province (b).The data indicate that the aerosol problem in China exhibits distinct regional characteristics, and in terms of AOD, the most deteriorated regions include the North China Plain (including Henan Province), the Yangtze River Delta region, the Pearl River Delta region, and the Sichuan Basin.A similar result was also reported by Luo et al. (2014).In brief, most of Henan Province experiences high AOD problems, particularly in Zhengzhou, the capital city of Henan Province.
Several studies have investigated the chemical composition and sources of PM 2.5 in Zhengzhou, which is predicted by the Economist Intelligence Unit as an emerging megacity in China at around 2020, similar to Beijing, Shanghai, and Guangzhou (Economist Intelligence Unit, 2012).Wang et al. (2016b) investigated the WSIIs and carbonaceous components of PM 2.5 in Zhengzhou from 2011 to 2013.Geng et al. (2013) used PMF and determined that soil dust, secondary aerosol, and coal combustion were the PM 2.5 main sources in 2010.However, given the rapid urbanization and economic development in Zhengzhou, sources of PM 2.5 possibly varied correspondingly.For example, in 2014, floor space of buildings under construction totaled 1.76 × 10 8 m 2 , the number of civil vehicles reached 1.93 million, and coal consumption by the above designated size industrial enterprises amounted to 35.21 million tons in Zhengzhou (Bureau of Statistics of Henan Province, 2015); these values were much higher than those in 2010 (8.88 × 10 7 m 2 , 0.96 million, and 26.66 million tons, respectively) (Bureau of Statistics of Henan Province, 2011).According to the data from the Ministry of Environmental Protection of the People's Republic of China (MEPPRC, http://www.mep.gov.cn/),Zhengzhou is among the 10 Chinese cities with the poorest air quality in 2014 (MEPPRC, 2015) and featured severe PM pollution.Data from national monitoring sites reveal that PM 2.5 and PM 10 are the primary pollutants in Zhengzhou in 2015 and suggest the importance of systematic studies.However, no study has been conducted on chemical compositions and source apportionment of PM 2.5 in Zhengzhou in 2014, nor compared pollution characteristics and sources of size-segregated aerosols (i.e., PM 2.5 , PM 2.5-10 , and PM 10 ).
Therefore, in this study, PM 2.5 , PM 2.5-10 , and PM 10 samples were collected in an urban area in Zhengzhou through a oneyear observation program, and concentrations of PM, WSIIs, OC, EC, and elements were determined.Given the chemical characterization, source apportionment of PM 2.5 , PM 2.5-10 , and PM 10 were performed via PMF and CMB.Meteorological impact was also analyzed by back-trajectory cluster analysis.This study is expected to provide fundamental information, including pollution characteristics and sources of PM, for regulatory agencies to mitigate PM pollution in this area.

Sampling
PM samples were collected between December 2, 2013, and October 24, 2014, at the same sampling site of our previous study in Zhengzhou (Wang et al., 2015).Fig. 1(c) shows the location of sampling site.Sampling was conducted at approximately 9:00 a.m. to 8:00 a.m. of the following day by using two high-volume PM samplers (TE-6070D, Tisch Environmental, USA) at a flow rate of 1.13 m 3 min -1 .Sampling was performed during the four seasons, and 12 to 14 samples for each size fraction were collected for each season.Quartz fiber filters (20.3 cm × 25.4 cm, Pall, USA) for PM 2.5 and PM 10 were used simultaneously for each sampling.Flow calibration was performed before sampling for each season.
The quartz filters were baked at 450°C for 5 h to remove adsorbed organics.The filters were then placed in a clean room for at least 48 h under constant temperature (25 ± 5°C) and relative humidity (45% ± 5%) before they were weighed using an analytical balance (Mettler Toledo XS205, Switzerland).

Chemical Analysis
Quartz filters were cut into pieces, with each piece featuring an area of 10.9 cm 2 .Two filter pieces were used to determine the mass concentrations of WSIIs.Samples were extracted in 25 mL ultrapure water for 30 min in an ultrasonic bath.Four anions (F -, Cl -, NO 3 -, and SO 4 2- ) and five cations (Na + , NH 4 + , K + , Mg 2+ , and Ca 2+ ) were analyzed by ion chromatography (ICS-90, Dionex, USA).Additional details were described by Wang et al. (2016b).
OC and EC concentrations were determined through thermal/optical transmittance method (Chow et al., 2001) by using a carbon analyzer (Model 5L, Sunset Laboratory, USA).Further information was described in our previous studies (Geng et al., 2013;Wang et al., 2016b).
Six pieces of filters (10.9 cm 2 each) were placed in a digestion cell (polytetrafluoroethylene) with a mixture of acids (11.1% HNO 3 /33.5% HCl).The digestion cell was then placed in a microwave digestion instrument (ETHOS ONE, MILE STONE, Italy).Digestion occurred under high-pressure with advantages of low dosage of reagents, which were digested almost completely.After cooling and filtering, 22 elements, including Be, Mg, Al, Ti, Mo, Mn, Fe, Co, Ni, Cu, Zn, V, Se, As, Sr, Ag, Cd, Sn, Sb, Ba, Tl, and Pb, were analyzed in the present study via inductively coupled plasma-mass spectrometry (Agilent7500cx, Santa Clara, CA, USA).

Quality Assurance and Quality Control
Field blank filters were measured as blank concentrations.According to US Environmental Protection Agency (EPA) (2016), standard deviation of replicate instrumental measurements of spiked blanks was used to calculate the method detection limit (MDL).For chemical analysis, each filter was measured twice, with errors considered acceptable within 5%.
For OC and EC, the analyzer was calibrated via a sucrose standard solution before measurement.Field blank concentrations measured 0.5 and 0.0 µg m -3 for OC and EC, respectively.MDL amounted 0.2 µg cm -2 for OC and EC.
For the elements, standard recovery efficiencies were determined, with values ranging from 80% to 120%.MDLs were calculated and ranged from 0.002 ng m -3 (Ag) to 3.362 ng m -3 (Sb).R 2 values of the standard curve for the 22 trace elements studied were all higher than 0.9989.

PMF and CMB Models
PMF 5.0 and CMB 8.2 models, as recommended by the US EPA, were both used in this study for source apportionment of atmospheric PM.Intercomparisons of PMF and CMB were conducted for mutual validation of the two model outputs given the unknown overall uncertainty of the applied models (Gianini et al., 2013).By contrast, CMB requires a specified priori information of emission sources and their profiles; however, PMF only necessitates qualitative or semi-quantitative a posteriori information of source emission profiles (Gianini et al., 2013).
PMF is a convenient factor analysis model (Paatero and Tapper, 1994) based on the weighted least square fit approach.Input data include concentrations and uncertainties of all species.The sample data matrix was decomposed into factor contribution matrix and factor profile matrix by PMF.For species concentrations, as data pretreatment, missing data and values below MDL were replaced by the species median and 0.5 × MDL, respectively (Brown et al., 2015;Cesari et al., 2016a;Jiang et al., 2018c).For uncertainties, the values considered were 0.1 × concentration + MDL/3 (species concentration > MDL), 0.2 × concentration + MDL/3 (species concentration < MDL), or four times the species-specific median (missing data) (Tauler et al., 2009;Jang et al., 2013).According to the US EPA (2014), bootstrap (BS) was run with 100 resamples, which is the recommended value of the model, and results were considered acceptable with all factors for PM in PMF analysis mapping above 80% of the BS runs; to determine a more optimal solution, Fpeak was conducted with strengths between -3 and 3; variations in G-space plots, profiles, and contributions were compared for solution optimization.We also adopted the constrained model to optimize the solution.When all parameters meet the performance requirements of PMF, i.e., intercept ~0, slope ~1, R 2 > 0.6, and signal-tonoise ratio and residuals for reliable model run are appropriate, results of source apportionment are considered acceptable (US EPA, 2014).Other details, including species categorization, are described by the US EPA (2014), Cesari et al. (2016b), andWang et al. (2017).
CMB requires ambient data and source profiles as inputs to estimate and quantify source contributions, on the basis of effective variance-weighted least squares fitting.Missing data (i.e., invalid) are automatically removed from calculation, and concentrations below MDL are also replaced by 0.5 × MDL (US EPA, 2004;Yatkin and Bayram, 2008).Performance of CMB result is controlled by four diagnostic parameters, and when all parameter values meet the requirements of CMB, i.e., T-statistics > 2.0, coefficient of determination > 0.8, Chi-square < 2, and 80% < percent mass < 120%, source apportionment results are considered acceptable (US EPA, 2004).

Back-trajectory Cluster Analysis
The Hybrid Single Particle Lagrangian Integrated Trajectory Model (HYSPLIT) is a powerful tool for studying long-range transport route of air masses.In this study, HYSPLIT_4 transport model was applied to assess the potential influences of regional transport and PM compositions, which vary widely under the prevailing wind, in four seasons.The 24-h backward wind trajectories at 500 m above ground level above the monitoring site (34°48′N, 113°31′E) were calculated with each campaign day including four backward trajectories.Then, on the basis of the similarity in spatial distribution, trajectories were clustered, and the suitable cluster number was selected before the dramatic increase in total spatial variation percentage (Wang, 2014).

General Characterization Major Precursors and Meteorological Variables
Table 1 lists the major precursors of PM (data from the national monitoring site) and meteorological variables in the study area.Wind speeds measured between 0.5 m s -1 to 4.5 m s -1 , and annual average value of 1.9 ± 0.7 m s -1 was half of that in Shanghai (2011-2013: 3.9-4.2m s -1 ; Wang et al., 2016a).The low wind speed was the disadvantage of contaminant dispersion.Mean ambient temperature was the highest in summer (28.1 ± 1.9°C) and lowest in winter (4.3 ± 3.4°C).Annual average value of relative humidity reached 57% ± 14%.The major precursors of PM, i.e., SO 2 , NO, and NO 2 (Krotkov et al., 2016;Jiang et al., 2017), showed notable seasonal variations, with the highest concentrations in winter (142 ± 55, 105 ± 46, and 87 ± 17 µg m -3 , respectively), and the lowest in summer (24 ± 9, 7 ± 3, and 46 ± 8 µg m -3 , respectively).During winter, enhanced emission of coal consumption for central heating, subsidence of the atmospheric mixing layer, and the prevailing stable atmospheric conditions (Zheng et al., 2015) resulted in high-level pollution.Daily NO 2 and SO 2 concentrations varied from 29 µg m -3 to 108 µg m -3 and from 17 µg m -3 to 244 µg m -3 , with approximately 29% and 8% of sampling days exceeding the Chinese National Ambient Air Quality Standard (NAAQS) (daily standard: 80 and 150 µg m -3 for NO 2 and SO 2 , respectively).Annual values of NO 2 and SO 2 were 68 ± 20 and 63 ± 52 µg m -3 , respectively, which exceed the NAAQS values (annual standard: 40 and 60 µg m -3 for NO 2 and SO 2 , respectively), especially for NO 2 .High precursor levels possibly play an important role in PM pollution.

Main WSIIs and Carbon
Fig. 3 and Table S1 in the Supplemental Materials provide the mean values of PM mass concentrations, WSIIs content, OC, and EC.Average concentrations of PM 2.5 , PM 2.5-10 , and PM 10 showed remarkable seasonal characteristics, i.e., winter (376, 149, and 524 µg m -3 ) > autumn (160, 89, and 249 µg m -3 ) > spring (136, 90, and 226 µg m -3 ) > summer (100, 54, and 155 µg m -3 ).These results were attributed to the comprehensive influence of various meteorological conditions and source emissions.In winter, frequent stagnant atmospheric condition and extra coal consumption for domestic heating lead to the highest PM concentrations, which can be demonstrated by the high levels of SO 4 2-, OC, Cl -, and EC in this season.Secondary inorganic aerosols (SIAs), including SO 4 2-, NO 3 -, and NH 4 + , were the major ions in PM, and accounted for 36%, 10%, and 27% of PM 2.5 , PM 2.5-10 , and PM 10 , respectively.This finding suggested that SIAs were mainly present in fine particles, as reported by previous studies (Kong et al., 2010;Long et al., 2014).The ratios of Ca 2+ , Mg 2+ , and F -in PM 2.5 and PM 2.5-10 were relatively comparable and featured small gaps.Correlation coefficients should be evaluated to identify similar sources of atmospheric particulates.As shown in Table S2 in the Supplemental Materials, Spearman correlation coefficients among WSIIs in PM 2.5 , PM 2.5-10 , and PM 10 were determined via IBM SPSS for Windows, Version 21.0.The ions possibly originated from the same source when correlation coefficients were close to 1. NO 3 -and SO 4 2exhibited good correlations with NH 4 + (R 2 = 0.87 and 0.88 in PM 2.5 , R 2 = 0.44 and 0.71 in PM 2.5-10 , R 2 = 0.87 and 0.86 in PM 10 , respectively), suggesting the presence of NH 4 NO 3 and (NH 4 ) 2 SO 4 .Cl -exhibited a high correlation with K + , thereby indicating that PM pollution could be affected by biomass burning (Silva et al., 1999).Ca 2+ and Mg 2+ are marker elements of crust matter (Winchester et al., 1979) and reflect the valuable contributions of dust sources.

Elemental Concentration
Fig. 4, Tables S3 and S4 in the Supplemental Materials show the mean concentrations and size distributions of trace element in PM 2.5 , PM 2.5-10 , and PM 10 .Total elemental concentrations accounted for approximately 3.5% of PM and exhibited the same seasonal characteristics as the PM level, i.e., they were highest in winter and lowest in summer.For PM 2.5 , Al and Fe, with concentrations exceeding 1000 ng m -3 , accounted for 58% of total elements, whereas Zn, Mg, and Pb in fine particles, with concentrations between 100 and 1000 ng m -3 , accounted for 34%.With concentrations less than 100 ng m -3 , the remaining 17 elements in PM 2.5 , i.e., Mn, Ba, Ti, Cu, Sr, As, Sn, Cd, Sb, Se, Ni, Mo, V, Tl, Ag, Co, and Be, accounted for 8% of the total elements.Similarly, for PM 10 , Al and Fe accounted for 65% of the total elements; Zn, Mg, Pb, Mn, Ti, and Ba accounted for 27%; the remaining 14 elements accounted for 8%.Annual concentration of As (0.029 µg m -3 ) and Cd (0.010 µg m -3 ) in PM 10 significantly exceeded the Chinese NAAQS (0.006 and 0.005 µg m -3 for As and Cd, respectively, in atmospheric environment including both gas and particle phase), with a relative high health risk.For different size distributions, Zn, Pb, Mn, Cu, As, Sn, Cd, Sb, Se, Ni, Mo, Tl, and Ag mainly fell under PM 2.5 , with ratios between 67% (Mo) and 96% (Se).However, Al, Fe, Mg, Ti, Ba, Sr, V, Co, and Be were relatively comparable under PM 2.5 and PM 2.5-10 , especially Ba and Sr, which exhibited a higher ratio in PM 2.5-10 than in PM 2.5 .In a previous study, Al, Fe, and Mg were reported to be crust elements mainly originating from dust emission (Winchester et al., 1979); this observation indicates that high PM level in this study area is influenced by dust.Zn, Pb, and Mn are closely related to vehicle brake, wear, and fuel combustion emissions (Garg et al., 2000;Cheng et al., 2010;Lin et al., 2014), whereas Ba, Cd, Cu, Sb, Sn, and Mo are emitted from tire tread and brake lining wear (Garg et al., 2000;  In our study, major PM components comprised SIAs, organic mass (OM), EC, and crustal minerals (CM, i.e., Al, Si, Ca, Fe, and Ti).Fig. 5 illustrates the mass reconstruction of the chemical compositions of PM.OM is estimated as OC multiplied by 1.8 (Hand et al., 2011;Simon et al., 2011), and dust mass is calculated as follows (Malm et al., 1994): (1) where Si was estimated on the basis of Al-to-Si ratio (0.46) (Chow et al., 2015) in PM 2.5 and Al-to-Si ratio (0.26) (Taylor and Mclennan, 1995) in PM 2.5-10 and PM 10 .CM was calculated using Ca 2+ concentrations given the lack of values for Ca.OM, SIAs, and CM were the most abundant species in PM, and accounted for 24%, 36%, and 12% of PM 2.5 ; 27%, 10%, and 26% of PM 2.5-10 ; and 25%, 27%, and 19% of PM 10 , respectively.By comparing size-segregated aerosols, CM was more abundant in coarse particles (PM 2.5-10 ) than in other particle types.This finding indicates that appropriate measures for dust control are more effective in decreasing PM 10 level than PM 2.5 level.However, SIAs mainly exist in PM 2.5 , suggesting that control measures of precursor gases (i.e., SO 2 , NO x , NH 3 , and VOCs) may be more successful in mitigating fine particle (PM 2.5 ) pollution than coarse particle pollution.Although EC and other elements only approximately accounted for 3% and 1% in PM, these components, especially heavy metals (As, Cu, Zn, Pb, Cd, Cr, and Ni), must still be paid further attention given their environmental and health effects (Martuzevicius et al., 2011;Liu et al., 2015).

Comparison of Source Appointment by PMF and CMB
Four scatter plots of PM 2.5 and PM 10 are shown in Fig. S1 in the Supplemental Materials.R 2 ranged from 0.79 to 0.90, indicating similar sources.In this study, PMF 5.0 was used to identify and quantify potential sources of PM 2.5 , PM 2.5-10 , and PM 10 , and 27 species concentrations and uncertainty of 159 PM samples (53 samples for each size fraction) were used as input data for the model.On the basis of the requirement of analysis input data of PMF 5.0 (US EPA, 2014), the species were classified as "strong," "weak," and "bad" variables."Weak" variables triple the uncertainty, whereas "bad" variables are excluded from analysis.Additional details are shown in Table 2. To obtain a further realistic solution of iterations and decreased rotational ambiguity (Amato et al., 2016), we applied different constraints to the PMF model during source appointment of size-segregated aerosols, with the dQ values satisfying the requirement of PMF 5.0 (US EPA, 2014); additional details and other parameters (Q robust /Q true , slope, and R 2 ) are provided in Table 2. Comprehensively considering the Other soluble ions: Na + , K + , Mg 2+ , F − , and Cl − .Other elements: measured elements except Al, Fe and Ti.Crustal minerals: Al, Si, Ca (Ca 2+ ), Fe, and Ti.Others: the rest of PM except the species shown.Q robust /Q true = 0.83 Slope = 0.86 R 2 = 0.88 Q robust /Q true : a measure of the impact of data points with high scaled residuals.Slope and R 2 : evaluate how well each species is fit for the model; calculated using the observed and predicted data.distributions of scaled residuals, solution stability, and physically reasonable solutions, the resulting six factors are the optimal choice in the three size fractions.The combination of displacement (DISP), BS, and BS-DISP were conducted to evaluate random errors and rotational ambiguity, and results showed that no swaps were diagnosed for dQ max of 4 and 8; bootstrapped factors were also mapped to base factors over 80%, conforming to the requirement of PMF 5.0 (US EPA, 2014).To obtain a further optimal solution, rotations with an Fpeak value from -3 to 3 were adopted, and the nonrotated solution (Fpeak = 0.0) was selected as the rotated results showed significantly increased Q values, and non-optimizing results, including those of G-space plots, profiles, and contributions, appeared as expected.Fig. 6 presents the PMF results.
Dust was the first factor, and it is characterized by Ca/Ca 2+ , Mg, Al, Fe, Ti, Mn, Pb, and Si (Lough et al., 2005;Wang et al., 2016c); dust was mixed with EC because of the effect of road dust (Jiang et al., 2018a).Dust source accounted for 13.1%, 25.1%, and 19.8% of PM 2.5 , PM 2.5-10 , and PM 10 masses in the study area, respectively (Table 3).Contribution values are consistent with the results of mass reconstructions, revealing that dust source plays an important role in PM pollution, especially in coarse particles, similar to a previous study (Lasun et al., 2016).
The second factor is vehicular traffic, which is mixed with road dust.Vehicular traffic source is characterized by high loads of EC, OC, NO 3 -, Fe, Zn, Cu, and Ni (Viana et al., 2006;Charlesworth et al., 2011).For example, Cu is linked to brake abrasion, whereas Ni is generated from oil combustion (Garg et al., 2000).EC is mainly emitted by heavy-duty vehicles (Manousakas et al., 2015).In this source, the relative high EC content was attributed to site location, which lies close to the west side of 4 th Beltway in Zhengzhou, where a considerable quantity of heavy-duty diesel trucks pass.This source contributed 14.1%, 20.8%, and 15.8% to PM 2.5 , PM 2.5-10 , and PM 10 in the study area, respectively.This source should be paid additional attention to alleviate PM levels caused by the rising number of vehicles in this area.
The third factor is coal combustion, with high loadings of Cl -, OC, Na + , SO 4 2-, As, and Pd, which are closely related to this source (Bhangare et al., 2011).Coal combustion is a major PM source and accounted for 16.1%, 21.8%, and 18.5% of PM 2.5 , PM 2.5-10 , and PM 10 mass, respectively, with slightly higher contributions in coarse particles, as similarly reported in a previous study (Tian et al., 2016).In previous works, coal combustion has been identified as a highly valuable PM source in China (e.g., coal combustions accounted for 12.4% and 10.5% of PM 2.5 at two sites in Nanjing in 2013 and 18% of PM 2.5 in Taian in 2014 [Li et al., 2016;Liu et al., 2016]).According to the China Statistical Yearbook (National Bureau of Statistics of China, 2015), total energy consumption in Henan in 2014 reached 228.9 million tons of standard coal equivalents, with a percentage of 77.7% for coal consumption; this value was much higher than the average proportion in China (64.0%).The energy structure of Henan heavily depends on coal, and this area may face serious PM pollution caused by long-term coal combustion.
The fourth factor is secondary aerosols, which feature high contents of SO 4 2-, NH 4 + , and NO 3 -.This source accounted for 35.8%, 10.5%, and 22.5% of PM 2.5 , PM 2.5-10 , and PM 10 mass, respectively.Secondary aerosols represent the most important source of PM 2.5 and PM 10 , and this (a) PM 2.5 (b) PM 2.5-10 Fig. 6.Source profiles of PM 2.5 , PM 2.5-10 , and PM 10 obtained by PMF model analysis., and NO 3 -mainly exist as fine particles, as also observed in a previous study (Cheng et al., 2015).This source is highly related to the photochemical reactions of SO 2 , NH 3 , and NOx (Wang et al., 2006;Perrone et al., 2010).Therefore, stricter measures of precursor gas control are necessary to decrease PM 2.5 pollution.
The fifth factor is identified as industry and mainly includes Fe, Mn, V, Cu, Zn, Cd, Se, As, Ag, V, Ba, Pb, OC, and EC.These elements are generally associated with iron/steel or other metals used in manufacturing (Chan et al., 1997;Turpin and Lim, 2001), whereas carbonaceous aerosols are also emitted from industrial sources (Zhang et al., 2013b).The output of 10 kinds of nonferrous metals (e.g., Cu, Pb, and Zn) in Zhengzhou totaled 0.53 million tons in 2014 (Bureau of Statistics of Henan Province, 2015), suggesting that related industries serve as important sources of PM emission.The industry source contributed 14.6%, 11.6%, and 13.5% of PM 2.5 , PM 2.5-10 , and PM 10 , respectively, implying that these species slightly but highly contribute to fine particles.This conclusion agrees with the results of chemical analysis, i.e., Zn, Pb, Cu, As, Cd, and Ag mainly exist in PM 2.5 .
The sixth factor includes the unidentified sources, particularly unorganized emission sources, including biomass burning and garbage incineration.This source contributed 6.3%, 11.6%, and 10.3% of PM 2.5 , PM 2.5-10 , and PM 10 , respectively.
Only CMB 8.2 (US EPA, 2004) was applied in this study to estimate source contributions to PM 2.5 owing to the lack of local chemical profiles of main sources for PM 2.5-10 and PM 10 .According to a previous study (Samara, 2005) (Jiang et al., 2018b), dust, secondary aerosols, coal combustion, biomass burning, vehicular traffic, and industrial emission were determined as the main sources of PM 2.5 in Zhengzhou.The profiles of PM 2.5 are summarized as follows: dust and biomass profiles were established with the support by the Public Welfare Project from MEPPRC, and additional details were described by Jiang et al. (2018a) and Wang et al. (2016d), respectively; the vehicle profile was acquired from the Jingguang North Road Tunnel in Zhengzhou via a tunnel experiment and was also supported by the project.Details of this study will be reported in another paper.Secondary aerosol profiles, including those of sulfate and nitrate, were purely stoichiometric, and profiles of coal combustion and industry were acquired from previous studies (Chen et al., 1994;Zheng et al., 2005).Contributions of seven sources, namely, dust, biomass burning, vehicular traffic, nitrate, sulfate, coal combustion, and the industry, were calculated by CMB.According to model requirements (US EPA, 2004), R 2 (0.86), X 2 (0.43), and percentage mass (93.5%) were all in the qualified range.
Table 3 lists the CMB results of source appointment for PM 2.5 .Secondary aerosols, including sulfate (18.2%) and nitrate (11.7%), are the most important sources and accounted for 29.8% of PM 2.5 .Coal combustion, dust, vehicular traffic, industry, and biomass burning contributed 18.8%, 12.7%, 11.8%, 11.5%, and 8.8%, respectively, of PM 2.5 .These findings were consistent with mass reconstruction and PMF results for PM 2.5 , excluding that for biomass burning source.In the results of PM source apportionment by PMF, biomass burning was not identified alone.For comparison, source contributions of dust and coal combustion under CMB (12.7% and 18.8%, respectively) were similar to those under PMF (13.1% and 16.1%, respectively), whereas contributions of secondary aerosols, vehicular traffic, and industry sources under the CMB model (29.8%, 11.8%, and 11.5%, respectively) were lower than those from PMF model (35.8%, 14.1%, and 14.6%, respectively).In this study, the source profiles obtained by PMF analysis were combined with those of redundant species, i.e., the species beyond specific sources, as a potential important reason.

Back-trajectory Cluster Analysis
The back trajectories for each of the identified cluster in four seasons during the sampling period are shown in Fig. 7, and the total spatial variation percent in all the clusters was below 10%.Cluster analysis results revealed that air mass transport significantly influenced PM pollution in Zhengzhou.Air quality across the four seasons was mainly affected by air masses from different directions.Air masses from the northeast were relative constant across the four seasons and accounted for 17.2% (cluster 3 in spring) to 29.6% Fig. 7. Back trajectories for each of the identified cluster in four seasons during the sampling period.
(cluster 2 in winter) of all air masses.In terms of AOD, the Beijing-Tianjin-Hebei region (i.e., northeast direction) was one of the most highly deteriorated regions of aerosol pollution in China in 2014 (Fig. 1(a)).Therefore, air masses from the northeast carried a considerable amount of pollutants and probably aggravated PM pollution levels in the study area.The sustained air masses from the east accounted for 15.9% (cluster 4 in winter) to 28.9% (cluster 2 in spring).Air masses from the south, with a relatively low PM pollution (Fig. 1(a)), exhibited a more prominent effect in spring and summer than in the other two seasons and accounted for 38.5% and 42.9%, respectively, of total air masses.Notably, PM level in the study area decreased because of dilution function.Obstructed by the Taihang Mountains, air masses from the northwest direction presented much better air quality (Fig. 1(a)) and only appeared in winter (25.0%) and autumn (14.3%) under strong air circulation.However, owing to their high altitude, airflow on mountains caused no significant effects on ground wind speed in the study region (average wind speeds of 1.4 and 1.7 m s -1 in winter and autumn, respectively [Table 1]).Thus, air quality in winter and autumn remained poor.The highest PM 2.5 concentrations appeared with short air-mass trajectory lines, with air masses accounting for 15.9% (cluster 4 in winter; average PM 2.5 : 160 µg m -3 ), 38.5% (cluster 1 in spring; average PM 2.5 : 92 µg m -3 ), 19.6% (cluster 4 in summer; average PM 2.5 : 78 µg m -3 ), and 16.1% (cluster 5 in autumn; average PM 2.5 : 122 µg m -3 ) of total air masses in different seasons (PM 2.5 data from the national monitoring site).These results indicate the notable transport influence from the adjacent regions.

CONCLUSIONS
This study investigated the major precursors, mass levels, chemical compositions, and source apportionments of sizesegregated aerosols, i.e., PM 2.5 , PM 2.5-10 , and PM 10 , in Zhengzhou, an emerging megacity of east-central China.Results showed that high levels of major PM precursors, i.e., NO 2 and SO 2 , play an important role in serious PM pollution.The average annual concentrations of PM 2.5 and PM 10 were 187 and 281 µg m -3 , respectively, which are much higher than those of the Chinese NAAQS.Concentrations of the size-segregated aerosols exhibited remarkable seasonal characteristics, with the highest average concentrations in winter and the lowest values in summer.
The results of the chemical analysis indicated that SIAs were the major ions in PM and mainly existed in the fine particles: SIAs accounted for 36%, 10%, and 27% of PM 2.5 , PM 2.5-10 , and PM 10 , respectively.Spearman correlation coefficient analysis revealed biomass burning and dust as important sources of PM.The annual concentrations of As (0.029 µg m -3 ) and Cd (0.010 µg m -3 ) in the PM 10 remarkably exceeded those of the Chinese NAAQS and implied a relatively high health risk.After comparing the elemental compositions of the size-segregated aerosols, we found that Zn, Pb, Cu, As, Sn, Cd, Sb, Tl, and Ag mainly existed in the PM 2.5 .By contrast, Al, Fe, Mg, Ti, Ba, and Sr were relatively comparable in the PM 2.5 and the PM 2.5-10 .
Results of mass reconstruction of the size-segregated aerosols revealed that CM was more abundant in coarse particles than in fine particles but SIAs were primarily present in PM 2.5 .
The PMF results indicated that source contribution characteristics differed among the size-segregated aerosols.Overall, dust, vehicular traffic, coal combustion, secondary aerosols, and industry served as the main pollution sources, accounting for 13.1%,14.1%,16.1%,35.8%,and 14.6% in PM 2.5 ;25.1%,20.8%,21.8%,10.5%,and 19.8%,15.8%,18.5%,22.5%,and 13.5% in PM 10 , respectively.Dust sources played an important role in coarse particles, indicating that appropriate dust control measures more effectively decrease PM 10 levels than PM 2.5 levels.However, secondary aerosol sources provided the highest contribution to PM 2.5 , suggesting that controlling precursor gases (i.e., SO 2 , NO x , NH 3 , and VOCs) can be more effective in reducing fine particles than coarse ones.The CMB findings agreed with the PMF results for PM 2.5 .For comparison, the contributions of secondary aerosol, vehicular traffic, and industrial sources were lower in the CMB model than the PMF model.
Cluster analysis showed that air quality across the four seasons was mainly affected by air masses from the northeast (17.2%-29.6%)and the east (15.9%-28.9%),indicating the significant influence of transport from the Beijing-Tianjin-Hebei region and the adjacent regions, respectively.Air masses from the south exhibited an important dilution function.

Fig. 1 .
Fig. 1.Location of the sample site (c) and the annual average 10 × 10 km 2 spatial resolution Terra MODIS Deep Blue AOD at 550 nm over the entirety of China and in Henan Province (a, b) (data from Tao et al. (2016)).

Fig. 3 .Fig. 4 .
Fig. 3. Mean values of PM 2.5 , PM 2.5-10 and PM 10 mass concentration and the main species.rest: the rest of PM except the species shown.

Table 1 .
Major precursors and meteorological variables during the sampling period.Hourly concentration data of SO 2 , NO and NO 2 are from the national monitoring site.
Bozlaker et al., 2013;Lin et al., 2014), demonstrating that high PM concentration in this area is affected by vehicles.

Table 2 .
Summary of input species, constraints and parameters for PM 2.5 , PM 2.5-10 and PM 10 by PMF.

Table 3 .
Comparison of source contributions (%) to PM by PMF and CMB models.
, using source profiles entirely from literature as input data for CMB presents substantial bias in source contribution.In