University of Birmingham Characterization and Source Apportionment of Carbonaceous PM2.5 Particles in China - A Review

Organic aerosols (OAs) account typically for 20 – 90% of ﬁ ne particles (PM 2.5 ) in the lower troposphere. They contribute to a wide range of environmental problems, from local issues (e.g., urban haze) to global problems (e.g., climate change). Huge e ﬀ orts have been dedicated to studying the composition, abundance, spatial and temporal distribution and sources of organic aerosols in China. This review aims to summarize recent studies on characteristics and sources of OAs and assesses the current state of understanding of the organic aerosol pol- lution in China. The OA constitutes ∼ 20 – 45% of the PM 2.5 with an annual mean value of 5.2 – 44.5 μ g/m 3 at sites across China, which is about ﬁ ve times higher than that reported in North America and Europe. There are thousands of di ﬀ erent organic compounds in atmospheric aerosols, but only about 5 – 10% of them have been identi ﬁ ed and quanti ﬁ ed. OAs show pronounced spatial variations with much higher concentrations in Northern than Southern China. Seasonally, the highest OA concentrations are observed in the winter, whereas the lowest are in the summer. This is due to higher emission rates from anthropogenic sources (e.g., biomass and coal combustion) and poorer dispersion conditions in the winter. Approximately 60 – 80% of total OA is apportioned by receptor modeling (such as Chemical Mass Balance, CMB) and tracer-yield method, based on the source pro ﬁ les of primary organic aerosols (POA) and secondary OA (SOA) derived from local emission sources. A number of OA sources are identi ﬁ ed, including motor vehicles, industrial emissions, biomass combustion, food cooking, and coal combustion for POA and anthropogenic and biogenic emissions for SOA. Industrial emissions and motor vehicular exhaust are the dominant sources of organic aerosols in the industrialized areas of Northern China, as well as the Pearl River Delta and Eastern China, whereas in other urban areas, residential coal combustion and motor vehicular exhaust are the dominant sources in winter and summer respectively.


Introduction
Atmospheric aerosols have been well-known as a major air pollutant in China's megacities during the last decades (Chan and Yao, 2008;Yang et al., 2016). Organic aerosol (OA) can make up a significant fraction of atmospheric particles, accounting for 20-90% of fine particulate matter (PM 2.5 ) mass in the lower troposphere Kanakidou et al., 2005;Murphy et al., 2006;Zhang et al., 2007a). Over the last decade, a number of field observation and lab studies of organic aerosols have been carried out in China, which have reported that organic matter constitutes a substantial fraction (30-50%) of PM 2.5 in Beijing, Shanghai, Guangzhou and Xi'an Huang et al., 2014;Zheng et al., 2005). Cao et al. (2003a) has reported that in the Pearl River Delta Region (PRDR), carbonaceous aerosols (those that include organic carbon (OC) and elemental carbon (EC)), accounted for 40.2% of PM 2.5 and 35.9% of coarse particles (PM 2.5-10 ), while organic aerosols were the most abundant components contained in PM 1 (39.7%) . OC can be directly emitted from combustion sources or formed through gas to particle conversion processes. While organic carbon is always bound to other elements (H, O, N etc) in organic compounds, elemental carbon is that not bound to other elements. The latter can be measured by thermal-optical techniques which differentiate it from organic carbon, but it is often measured optically as black carbon which is an ideally light-absorbing substance composed of carbon. Recommended definitions are given by Petzold et al. (2013). Because organic carbon is associated with oxygen, hydrogen and nitrogen, the organic carbon (OC) can be converted to OA by multiplying its concentration by factors of 1.6 ± 0.2 in urban areas (Duan et al., 2007;Turpin and Lim, 2001), although a wide range of values has been used from 1.2 close to traffic emissions to > 2.0 for a large secondary component.
Evidence from a large number of studies shows that OAs adversely affect not only human health, but also influence climate. In addition, OAs cause visibility reduction, material and ecosystem damage, and affect cloud condensation nuclei (CCN) (Cao et al., 2006;Dentener and Crutzen, 1993;Mauderly and Chow, 2008;Seinfeld and Pandis, 2016). For example, epidemiological studies have found strong relationships between the level of OA in the ambient particulate matter and cardiovascular health outcomes and adverse respiratory health (Bi et al., 2002Janssen et al., 2011). Considering effects on climate, more attention has been paid to polar organic aerosols owing to their hygroscopic character and their ability to reduce visibility by causing uptake of water into the aerosol (Cao et al., 2012;Huang et al., 2014;Watson, 2002). In addition, OA can modify the earth's climate through scattering and absorption of solar radiation, altering cloud properties and lifetimes Kanakidou et al., 2005;Penner et al., 2003). OAs in different size ranges may have different physicochemical properties, such as water solubility, due to different sources and composition.
In the ambient air, OA occurs as primary organic aerosol (POA) which is directly emitted from numerous anthropogenic and biogenic sources, or secondary organic aerosols (SOA) which are formed through chemical oxidation of high vapour pressure compounds (Fuzzi et al., 2006;Han et al., 2016b;Kanakidou et al., 2005). Han et al. (2016b) investigated the composition and distribution of OA over eastern China, finding that POA and SOA contributed 52% and 48% of total OA, respectively. Huang et al. (2014) reported that the SOA contributes 44-71% of OA (averages for all four cities of Beijing, Shanghai, Guangzhou, and Xi'an). Numerous source apportionment studies of OA have been conducted in China. For example, Lu et al. (2011) have  (Song et al., 2007); Hongkong: OA=OC*1.9 ; other sites: OA=OC*1.6 (Cao et al., studied the inventory of primary organic aerosol from 1996 to 2010 and reported that the dominant contributor to anthropogenic emissions is residential emissions (69%), in which residential biofuel (56%) is combusted in the home for cooking and heating. Biomass burning, industrial and traffic emissions, and coal combustion are also considered as major sources of OA in Beijing (Dan, 2004;Duan et al., 2004;Song et al., 2006aSong et al., , 2006bSong et al., , 2007. Similar results were observed in the south of China (Cao et al., 2006;Feng et al., 2016;He et al., 2011;Zhang et al., 2008c), and the organic species were more enriched in the accumulation mode (particularly, in the size range of 0.43-0.65 μm) . In order to devise effective policies for the control of the particulate pollution, it is very important to quantify the fractions of POA and SOA. Several literature reviews on aspects of OA have been published, Hahn (1980) summarized knowledge of the organic constituents of natural aerosols and described a cycle of tropospheric organic matter. Given the availability of quantitative molecular data, Jacobson et al. (2000) presented an extensive and comprehensive description of characteristics and analysis method of OA. Based on earlier reviews, Kanakidou et al. (2005) reviewed knowledge of OA of importance for global climate modelling and identified significant gaps to reduce the uncertainties. De Gouw and Jimenez (2009) reviewed the sources of OA in the global atmosphere. Iavorivska et al. (2016) summarized the deposition of organic carbon by precipitation. Wang et al. (2016b) provided insights into the characteristics of OC and EC pollution in China on a national scale on the basis of published data.
Despite numerous published literature studies on organic aerosols in China, few systematic attempts have been made to analyze spatial and temporal variations. Different protocols were used for OA analysis, leading to uncertainties in OA characterization. In this review, we will (1) present the general sampling and analysis methods used for OA; (2) summarize the sources of POA and SOA; (3) examine spatial and seasonal variations of OA concentrations and (4) review source apportionment of OA using receptor models. In this paper, unless otherwise stated, the term OA refers to the OA contained in fine particles (particulate matter with an aerodynamic diameter smaller than 2.5 μm). For this reason, the rapidly increasing volume of data collected by Aerosol Mass Spectrometry which is approximately PM 1 is not reviewed in detail. This is included in a recent review by Li et al. (2017).

Sampling and analysis of OA
Traditionally, aerosols are collected on filters and analyzed by chemical and optical methods. More recently, online instruments such as the AMS (Aerosol Mass Spectrometer) or online OC/EC analyzer have been developed to measure the concentration of OAs with high time resolution. This section will review the analytical methods used for OA between 2000 and 2017 (Table S1 and Table 1).

Aerosol sampling
Aerosol sampling for OA analysis needs to consider two key aspects. One is the choice of sampler considering cost, commercial availability, ease of usage and quality of operation. Different instruments have been used for distinct studies (Table S2). For example, high volume samplers are often used to collect organic compounds in the particulate phase at a typical flow of 1.13 m 3 /min Feng et al., 2009;Hart et al., 1992). In some studies, medium volume samplers (TH-150C, X. Wu et al. Atmospheric Environment 189 (2018) 187-212 URG-3000K) giving a flow of 30-100 L/min (Li et al., 2015b), or mini volume samplers (Airmetrics, USA) at a flow of 5 L/min have been used for OA collection (Cao et al., 2003b;Zhang et al., 2013a). The second issue to consider during sampling is the choice of suitable filters, which depends on the type of chemical analysis, the capture efficiency and resistance of filters. The filters, usually of glass fibre or quartz, are weighted before and after sampling to determine the mass of particulate matter and are usually baked at 500-850°C for several hours before use to remove all residual organic compounds and to drive off water (Zhang et al., 2008a). After sampling, the exposed filters are stored in a freezer at about −20°C until analysis. However, a number of studies have suggested that it would be better to store filters at −80°C to prevent organic compounds from evaporating and to avoid oxidation of semi-volatile organic compounds and to prohibit microbial growth, thus ensuring minimization of organic compound losses (Harrison et al., 1996;Zheng et al., 2000). This paper synthesizes data relating to organic aerosols from many sites in China, whose locations appear in Fig. 1.

OA analysis
Quantification of OA employs a range of instruments depending upon the information sought. Thermal-optical analysis (TOA) quantifies carbon and seeks to differentiate organically bound carbon from elemental carbon. The AMS is able to differentiate source-related components of organic carbon in real time, while carbon isotope analysis can differentiate modern and fossil carbon. Chromatographic separation by gas chromatography (GC) or liquid chromatography (LC) allows analysis of individual compounds by GC/MS and LC/MS when coupled with mass spectrometric detection. Techniques and appropriate applications of these methods are listed in Table 2.
Different protocols for heating temperature and gas flows are available for TOA. The most popular TOA protocols include IMPROVE (Interagency Monitoring of Protected Visual Environments) (Chow et al., 1993, IMPROVE_A (a modification of the IMPROVE protocol), NIOSH (National Institute of Occupational Safety and Health) (Birch and Cary, 1996;Birch, 1998) and EUSAAR_2 (European Supersites for Atmospheric Aerosol Research). Different protocols yield a varying OC/EC split, resulting in uncertainty in concentrations of OC and EC that are operationally defined (as Table S3). NIOSH exhibits negative EC bias relative to IMPROVE due to premature EC evolution in pure helium (OC4 step), whereas, IMPROVE shows an opposite EC preference owing to the lower temperature of OC4 (580°C). Lower temperature leads to a fraction of OC being carried into the oxidative condition (Cavalli et al., 2010;Cheng et al., 2011). EUSAAR_2 extends time steps at lower temperature (OC1, OC2) to promote volatilization of OC over pyrolysis, and the highest temperature of 650°C is reported to be the best for OC yield without premature creation of pyrolysis OC (POC) (which is OC evolved in the EC phase of the analysis). Laser reflectance and laser transmittance are used for correction of POC. The former is mainly affected by near surface char and the latter is more influenced by within filter char leading to longer times to return to initial value of the laser signal and thus more delay in the split point (Bautista et al., 2015;Cao et al., 2013;Chow et al., 2004;Han et al., 2016a). This is consistent with previous observations where the EC estimated from reflectance was 1.57 times as high as that from transmittance in samples from Lhasa (Li et al., 2016a).
GC/MS can be used to isolate a wide range of compounds on a single GC column according to each compound's retention time, and to make compound identification relatively straightforward through mass spectrometry (MS). Major analytical procedures are listed in Table S4 ( Chauhan et al., 2014;Fu et al., 2008;Yin et al., 2015). Both thermal desorption (TD) GC/MS and solvent extraction (SE)-based GC/MS have been used in the literature. Although TD-GC/MS does not need sample extraction and requires less filter for detecting molecular organics than that based on SE-GC/MS, polar compounds are lost to the GC column (Ho et al., 2008). In order to increase the fraction of OAs to be analyzed, a number of previous studies used high-resolution GC and high-resolution MS to measure individual compounds (Hildemann et al., 1994). Recently, two dimensional (2D) GC-time of flight MS (GC × GC-TOFMS) was developed to enhance the GC resolution and sensitivity by separating unresolved complex matter (UCM) with two comprehensively coupled columns . Welthagen et al. (2003) reported that the UCM fraction account for 70% of the semivolatile organic materials, and cannot be resolved by 1-D GC. Meanwhile, the TOFMS provides mass spectral data of all separated compounds .
There are limitations to the analysis of polar organic compounds by GC/MS due to low volatility. Thus different methods of chemical analysis, such as LC/MS, are used to quantify more polar fractions in OA and to fill in the gaps for quantitation of carbonaceous mass. The LC/ MS method employs various types of columns to separate the compounds within a narrow compound class (Sato et al., 2012;Fairbaugh, 2015).
Traditional methods of OA analysis such as GC/MS and LC/MS typically identify a limited number of compounds accounting for 10-30% of OA mass due to inadequate separation capability and the requirement for authentic standards to provide the retention time in chromatograms. In recent years, two dimensional gas chromatography coupled with time-of-flight mass spectrometry (GCxGC-ToFMS) has greatly enhanced separation capability , and when combined with variable energy ionisation offers great power to distinguish between isomeric compounds . High-resolution mass spectrometry techniques (HRMS), such as Orbitrap MS and Fourier transform ion cyclotron resonance mass spectrometry (FT-ICR MS) , coupled with a soft ionization technique such as electrospray ionization (ESI) Roach et al., 2010;Gautier et al., 2014) can identify the molecular formulae and elemental composition of organic compounds without authentic standards, especially for determination of secondary organic aerosols (Liu et al., 2017b;Tong et al., 2016). For example, Reinhardt et al. (2007) quantified approximately 60% of organic compounds using FT-ICR MS. However, this gives information on the molecular formulae without defining the structure of a molecule.
Recently, online AMS has been widely used for chemical analysis of OA in real time with high time resolution (seconds to minutes). It is a quantitative instrument, with some models offering unit mass resolution (UMR) (Q-AMS, C-TOF-AMS and ACSM). The Aerodyne high resolution time-of-flight mass spectrometer (HR-TOF-AMS) is commonly used to characterize the size-resolved chemical composition of submicron particles with high mass resolution (Canagaratna et al., 2007;Drewnick et al., 2005;Jayne et al., 2000;Zhang et al., 2015a). A brief description of AMS is given in Table 2 (Aiken et al., 2008;DeCarlo et al., 2006;Zhang et al., 2011b). Normally, the Positive Matrix Factorization (PMF) algorithm is applied to AMS datasets to study sources of OA (Zhang et al., 2011c), and reports the temporal variations of the composition and concentration of the OA in the form of factors with mass spectra and temporal variation characteristics of specific source types and secondary aerosol types.
The ratios of 13 C/ 12 C (δ 13 C) and 14 C/ 12 C (δ 14 C) can be measured by accelerator mass spectrometry, and can be used to quantify the source of carbonaceous aerosol . The variation of δ 13 C is affected by chemical fractionation processes, like photosynthesis. δ 14 C is a function of the age of the carbon, and is an effective isotopic tracer to distinguish contemporary from fossil origin (Heal et al., 2011;Xu et al., 2016;Zhang et al., 2012), because 14 C is completely absent in fossil fuel OA owing to its short half-life (5726 years) and can only be found in modern non-fossil carbon sources (biogenic and biomass burning). Comparing δ 13 C ratios in samples with values in known sources can be useful for source apportionment. For instance, Chesselet et al. (1981) inferred that fine particles collected over the Tropical North Atlantic and Pacific mainly come from continental sources with Table 2 Measurement methods for studying organic aerosols. and "W" every 7.5 or 10 min. The V mode is more sensitive, which is to receive the mass concentrations of the non-refractory species for processing UMR data. The W mode could obtain higher mass resolution to make the elemental composition in the mass spectrum to be quantitatively estimated and evaluate the elemental δ 13 C of −26‰ ± 2‰, based on the measurements of aerosols collected in urban areas of France. Many 14 C studies have measured 14 C in both OC and EC rather than in total carbon (TC), which enhances OA source apportionment even for non-fossil sources Szidat et al., 2009;Zhang et al., 2012). The analytical procedures for 14 C in Water Soluble Organic Carbon (WSOC), Water Insoluble Organic Carbon (WINSOC) and EC have been reported previously (Liu et al., , 2017a.

Sources and emissions of OA
In this section, the four main sources of primary organic aerosols (traffic emissions, industrial emissions, residential coal combustion and biomass burning) and their emission rates in China are discussed. This is illustrated by Fig. 2a and b which presents OC and EC emissions from generic sources over China in 2012. The main sources of secondary organic aerosols (biogenic and anthropogenic SOAs) are also considered. Furthermore, the source profiles for source apportionment of POA by receptor modelling and SOA tracers are reviewed in this section.

Traffic emissions
Road traffic is a major source of carbonaceous aerosols in Chinese cites (Lu et al., 2011). For example, Zhao et al. (2012) estimated that OC and EC from traffic emissions accounted for about 10.3% and 15.0% of total OC and EC in Northern China, respectively. Organic aerosols from road traffic can be released directly in the exhaust due to incomplete combustion of fuels and lubricating oil or can be formed in the atmosphere by the oxidation of traffic generated-VOCs such as aromatics (Kanakidou et al., 2005). Lang et al. (2017) found a very high correlation (r 2 ) between annual average OC concentration with vehicular OC emissions (r 2 = 0.95) and VOC emissions (r 2 = 0.9) in Beijing over a 10 year period (2004)(2005)(2006)(2007)(2008)(2009)(2010)(2011)(2012)(2013)(2014), suggesting a high contribution of traffic emission to the atmospheric OC level. This section will summarize the emission factors (EFs) of POA released from vehicles in China and factors affecting the measurement of EFs. The contribution of traffic emission to SOA will be discussed later in Section 5.    Cao et al. (2006) reported that the emissions were 60 Gg/year of organic aerosols from diesel vehicles and 5.1 Gg/year from gasoline vehicles.
3.1.1.2. Effects of fuel types and properties. The type of fuel (e.g. gasoline, diesel, or biofuel) has a significant effect on the EF of organic aerosol. Cheng et al. (2010) measured the EFs of aerosols in a traffic tunnel in Hong Kong and found that the EF of OC emitted from a diesel vehicle (67.9 mg km −1 ) is 8 times higher than that from gasoline vehicles. This is consistent with the later observation by Cui et al. (2016). In another study, Zhang and Balasubramanian (2016) investigated the impacts of alternative fuels such as biodiesels on EFs. This study found that the blending n-butanol and n-pentanol with biodiesel could reduce significantly the emissions of PM and EC, but increased the proportion of organic aerosol. Similarly, Yang et al. (2017) reported that the blended biodiesel show a higher fraction of OC at low and medium engine loads. In addition, Xu et al. (2013) characterized the exhaust diesel particles from different oxygenated fuels (soybean-based biodiesel blend, dimethyl carbonate blend, and dimethoxy methane blend) and found that the oxygenated fuels could increase the fuel density and viscosity, leading to degradation of combustion efficiency, therefore enhancing the emission rate of OC.
3.1.1.3. Effects of vehicle operating modes. The operating modes (speed, acceleration and load) have a significant influence on the emission factor of traffic-generated organic aerosols. Cui et al. (2017) reported that poor driving conditions (low average and highly variable speeds) could result in a higher emission of OC. This concurs with Wu et al. (2016a) who conducted the measurement of OC under highway and non-highway driving conditions. The lower speeds under non-highway driving conditions lead to a lower combustion temperature, causing higher OC emissions due to less pyrolysis of fuel and lubricant oil. Furthermore, the effect of fuel injection pressure, and the application of exhaust gas recirculation (EGR) was investigated by Li et al. (2014a,b). Those studies found that a high level of EGR could enhance the soot formation from a diesel engine, leading to increased emission of carbonaceous aerosols.

Industrial emissions
Industry is one of major contributors of organic aerosols in China. Cui et al. (2015) estimated that the emissions from the industrial sector account for 17-21% and 30-34% of the total OC and EC emissions, respectively. This section will summarize the emission factors and emissions of OA from coal fired combustion, oil combustion and power plants based on different fuels and industrial processes. Fig. 2c and d present the OC and EC emissions from different industrial sectors.
3.1.2.1. Industrial coal combustion. The emission factor (EF) of industrial boilers is highly affected by the coal maturity. For examples, Zhang et al. (2008c) found that emission factors for industrial bituminous coal combustion are 0.3 and 0.062 mg kg −1 coal for OC and EC, respectively, which is much lower than that emitted from industrial brown coal (17.1 mg kg −1 for OC and 2.8 mg kg −1 for EC). Mixed coal combustion produced moderate levels of organic aerosols (1.9 and 0.7 mg kg −1 coal for OC and EC, respectively). Bond et al. (2002) indicated that the types of coal with low maturity have a relative high volatile content that is the precursor component for particle formation, leading to increase of emission factors for OA.
Coal combustion in industry generates carbonaceous aerosols with a range from 394 to 993 Gg/yr for OC and from 430 to 591 Gg/yr for EC from investigations in the field and calculation using models (Cao et al., 2006;Cui et al., 2015;Lu et al., 2011). The estimate of OC and EC emissions in previous studies shows different trends. Cui et al. (2015) estimated that the annual trends of OC and EC emissions from industrial coal combustion showed steady growth during 2000-2005 (387-510 Gg OC per year) and then levelled off for 2005-2012 (455-494 Gg OC per year). However, their estimates are higher than those from Lu et al. (2011) and Zhao et al. (2011). The differences can be attributed to two aspects. First, some studies have omitted some emissions from certain non-combustion industrial processes. Second, Cui et al. (2015) added wood combustion in industry and used larger emission factors for small coal and biofuel stoves. Cao et al. (2006) estimated the highest annual emissions (1116.1 Gg OC and 543.9 Gg EC per year) among all studies in 2000, due to inclusion of the emissions from uncontrolled coal-fired boilers, kilns and furnaces in rural industry.
3.1.2.2. Oil combustion. The estimated emission from industrial oil combustion shows a clear growth from 2005, with emissions of 2-3 Gg/yr OC and 3-5 Gg yr −1 EC before 2005 which increased to 7-8 Gg/yr OC and 14-15 Gg yr −1 EC for the following years (Cui et al., 2015). Cao et al. (2006) estimated much higher emissions from oil combustion in industry for 2000 (25.5 Gg OC and 17 Gg EC), mainly due to inclusion of emissions from rural industry. As shown in Table 4, the emission factor for industrial oil combustion that is approximately 8-11 g/GJ for OC and 25-35 g/GJ for EC, is higher than that from an oil or coal-fired power plant (Lu et al., 2011).

Power generation.
According to the State Power Corporation of China (2017), electricity was mainly derived from coal combustion (66%) in China in 2016. According to the National Bureau of Statistics of China (NBSC) data, power plants consumed 1845 Tg of coal, which is 44.8% of total coal combustion in 2014. However, the OC and EC emissions are relatively low, only 11 and 21 Gg/yr in 2010, constituting about 0.3 and 1.9% of total national emissions of OC and EC, respectively within China (Cao et al., 2006;Cui et al., 2015;Lu et al., 2011). This is because pulverized coal is the dominant coal used in power plants, and emits lower levels of carbonaceous materials at high temperatures. In addition, the emissions of carbonaceous aerosols from coal-fired power plants are reduced by electrostatic precipitators (ESP)  Zhang et al. (2009) reported that net PM 2.5 emission factors declined from 2.0 g/kg coal to 1.2 g/kg coal (a reduction of 40%) using FGD. For coal-fired power plants, OC and EC emission rates reduced from 1.2 g/GJ and 0.9 g/GJ in 1998 to 0.3 g/GJ and 0.6 g/GJ in 2010, respectively. These OC and EC emission factors were higher than those for oil combustion power plants, but by 2010 these were equal (Table 4) (Lu et al., 2011). Cui et al. (2015) estimated that the annual EC emission decreased from 11 Gg/yr in 2000 to 6 Gg/ yr in 2010. Besides ESP, some other dust collectors have been also used, including fabric filter systems, wet scrubbers and cyclones Zhao et al., 2010). However, the coal-fired power plants have resulted in a tremendous rise in greenhouse gas emissions. For instance, Chen et al. (2010) found that the direct CO 2 emissions from electric power and hot water production accounted for 42.8% of the total.

Residential combustion emissions
Residential emissions include those from heating and cooking by coal combustion, natural gas combustion and biofuel burning. In this section, we focus on the emissions from domestic coal combustion.
The emission factors for residential coal combustion varied from 0.014 to 13.71 g/kg for OC and from 0.005 to 10.96 g/kg for EC (shown in Table 5). Higher EF OC and EF EC values are found in the field from combustion of anthracite briquettes rather than from burning under laboratory test conditions. For instance, the EF OC and EF EC of anthracite briquettes burned in a brick stove with a chimney were 0.47 g/kg and 0.028 g/kg, respectively (Zhang et al., 2008c). It is also reported that the EF OC of anthracite burned in an improved stove with a chimney in Shanxi was 0.14 g/kg . However, the results of an experimental test with anthracite briquettes were 0.014 g/kg and 0.005 g/kg for EF OC and EF EC , respectively . It is noted that great differences exist in EFs for coals of different maturity. As shown in Table 5, medium-volatile bituminous coal has the highest EF OC (13.71 g/kg) and EF EC (10.96 g/kg), which are 334 and 2192 times higher than those from semi-anthracite respectively (Chen et al., 2006), suggesting that estimation of carbonaceous emissions from residential coal would be very difficult in China, because coals have a wide range of maturity. Other studies also gave similar results that the EF OC and EF EC are 0.014-0.47 g/kg and 0.005-0.028 g/kg respectively for anthracite briquettes and 0.59-13.713 g/kg and 0.2-10.96 g/kg for bituminous coal. More mineral matter and less OC are emitted from anthracite coals due to the higher burning temperature.
The largest OA source in China is coal combustion, which accounts for about 70% of total energy consumption . According to Fig. 3, coal consumption increased from 1.2 × 10 9 tonnes in 2001 to 2.75 × 10 9 tonnes in 2014, contributing 68-72.5% of total energy consumption during this period (NBSC, 2015). Power plants are the largest coal consumer (1.95 × 10 9 tonnes), followed by industry (1.19 × 10 9 tonnes) and residential use (3.6 × 10 8 tonnes). Coal burning is a very common fuel in Northern China for heating in winter. Zhang et al. (2008c) estimated that residential coal consumption emits more than five times more OC than from industry. The large contribution from residential sources is mainly because of the lack of aftertreatment in the residential stove. Meanwhile, improved stoves with galvanized flue pipes have been used to reduce the EF of carbonaceous aerosols. For example, Zhi et al. (2009) found that approximately 61% and 98% of OC and EC emissions were reduced in the improved stove.

Biomass burning
Biomass burning emissions in this review include open biomass burning and household biofuel use. The main sources of household biofuel for cooking and heating in the countryside of China include crop wastes (rice straw, maize residue, wheat residue, bean straw, cotton stalk, and kaoliang stalk), and woody fuels (wood and branches). The emission factors (EFs) of organic aerosols are strongly related to the type of fuel burning. For example, a stove-fuel combination test was conducted to compare emissions from crop residue combustion and woody fuel combustion (as in Table 6), indicating that crop waste combustion emitted much higher OC (10.86-14.66 g/kg) but lower EC (2.71-3.53 g/kg) than those emitted from woody fuel combustion (1.97-3.16 g/kg for OC and 2.62-4.91 g/kg for EC, respectively) . However, the efficiency of biomass burning depends on the temperature of burning, heat intensity, aeration and process of smoldering, which determine the organic compounds emitted from burning (Simoneit, 2002). For instance, the EFs of OC and EC were reported to be 10.53 and 0.49 g/kg dry fuel for rice straw burning in flaming combustion in the laboratory, respectively, which is a little higher than those emitted from smoldering combustion with 8.77 g/kg of OC and 0.37 g/kg of EC (Zhang et al., 2013b).
Biomass combustion is estimated to have emitted about 41-79% of OC (1264-2750 Gg/yr) and 27-50% of EC (398.3-716 Gg/yr) in China during 2000-2012 by different studies (Cui et al., 2015;Lu et al., 2011). Agricultural waste burning emitted about 13% of OC, forest fires contributed 2-11% of OC, and residential biofuel burning accounted for 56% of anthropogenic OC emissions in 2010 in China (Lu et al., 2011). Based on statistical data from NBSC, straw residue combustion has reached about 600 million tonnes currently, accounting for 70% of rural energy. Alternatively  burning in Beijing (Wang et al., 2015c). The distribution of fire locations indicates that the fires occurred mostly in the plains, such as east of Henan, south of Shandong, north of Anhui and Jiangsu, which is where winter wheat is cultivated.

Source profiles of primary organic aerosols
In the CMB (Chemical Mass Balance) receptor model, it is very important to obtain representative source profiles for emissions. The selected molecular markers should be stable during transportation from the source to receptor. CMB inputs often include OC, EC, n-alkanes, nalkanoic acids, PAHs, hopanes, levoglucosan and sterols (Fraser et al., 2003;Schauer et al., 1996Schauer et al., , 2002. The major source profiles and molecular tracers are summarized in Table 7 and Table 8. Gasoline and diesel engine source profiles were measured by Cai et al. (2017) in China. This study included several types of gasoline vehicles (light duty and heavy duty with electronic fuel injection and catalyst, heavy duty with carburetor and catalyst, heavy duty carburetor, and motorcycles) and two types of diesel vehicles (light duty and heavy duty). OC contributed respectively 31.7% and 54.9% to PM 2.5 from light and heavy duty gasoline vehicles. For diesel vehicles, OC accounted for 56.9% of PM 2.5 , while EC was 17.6% and 17.7% of PM 2.5 for heavy and light duty diesel, and 8% on average for gasoline. These results are consistent with previous work by Zhang (2006), who found that the average content of OC and EC in fine particles is 38% and 4% from gasoline cars and 58% and 16% from diesel cars, respectively. In other studies, Pio et al. (2011) measured the OC and EC at both roadside and urban background sites in Portugal and the UK, and obtained the lowest OC/EC ratio ranging from 0.3 to 0.4 for the roadgenerated aerosols. The results of Pio et al. (2011) are in agreement with the findings by Yu et al. (2011), but they are lower than those measured by Hildemann et al. (1991) for particles emitted from gasoline (OC/EC = 2.2) and diesel vehicles (0.8).
Heptadecane (C17) is one of the major tracer compounds for vehicle emissions, and the dominant carbon number of the n-alkanes (C max ) from PM in vehicle exhaust can range from C17 to C25. The carbon preference index (CPI) of gasoline and diesel exhaust is reported as 0.93 and 1.07, respectively (Cai et al., 2017). The hopane concentrations in gasoline emitted particles is higher than that from diesel vehicles and the predominant hopane is 17α(H),21β(H)-hopane (HP30), which has been used as a tracer for vehicle emissions (Shrivastava et al., 2007). Vehicle emissions are found as a major source of PAHs in Chinese megacities such as Hong Kong, Guangzhou and Beijing Guo, 2003;Wu et al., 2014). This is consistent with observations from the western countries. For example, Harrison et al. (1996) identified benzo(ghi)perylene (BGP) and coronene along with phenanthrene (PHE) and benzo[b]naphtha[2,1-d]-thiophene (BNT) as a source fingerprint for vehicle emissions; Miguel et al. (1998) and Masclet et al. (1986) found that the gasoline engine emission was enriched in BGP and coronene and diesel exhaust emitted mainly chrysene (CHR), fluoranthene and pyrene (as in Table 10).
Source profiles of coal combustion were measured from industrial boilers and residential stoves, including anthracite, bituminite and subbituminite coal (Zhang et al., 2008c). The contribution of OC and EC to PM 2.5 is 45% and 12%, respectively, and the POM from coal combustion was predominantly comprised of n-alkanes (20%), PAHs (38%) and aliphatic acids (46-68%). The CPI of n-alkanes is 1.2-1.4 and dominated by C17 to C25. PAHs are by-products of incomplete combustion of carbonaceous fuels with emissions of many individual compounds higher than those from other sources, especially picene, benzo [b] fluoranthene, pyrene and retene . For example, Wei et al. (2016) observed that naphthalene, fluoranthene, pyrene, phenanthrene, fluorene, chrysene and benzo(a)pyrene were dominant PAHs emitted from a coal-fired power plant. For a heavy oil and natural gas fuelled-boiler, naphthalene, phenanthrene, fluoranthene, pyrene, fluorene and benzo(b)chrysene were found to be the major PAHs. OC/ EC ratios for the industrial emissions are 3.5-5, which are higher than those from traffic emissions. A study by Yin et al. (2010) in the UK found that picene is predominantly from coal combustion.
The composition of organic aerosol from biomass burning varies significantly depending on the type of biomass fuel and combustion conditions. In China, there is a large quantity of crop straw produced, Wang et al.
Zhao et al. Wang et al. (2009a) 0 in table refers to the value below detection and the blank means that was not analysed.
X. Wu et al. Atmospheric Environment 189 (2018) 187-212 up to about 6 × 10 8 tonnes per year, and half is combusted as an energy source in rural regions or burned for disposal . Many previous papers have focussed on the composition characteristics of cereal straw and wood burning aerosol as in Table 9. The most abundant species in biomass combustion is levoglucosan, contributing about 4.5% and 1.7% to the mass of fine particulate matter from straw and wood burning, respectively, followed by n-hexadecanoic acid (0.4% for straw burning and 0.2% for wood burning). Moreover, the methoxyphenols from lignin and dehydroabietic acid from conifer smoke are significant tracers for biomass burning (Simoneit, 2002) as shown in Table 10. Potassium is also well-known as a tracer for biomass burning. The high K + /OC ratio values (0.19-0.21) in summer in Beijing suggest that biomass burning is active in this period (Duan et al., 2004;Sun et al., 2014). High OC/EC ratios of 3.8-13.2 characterize biomass  GloBEIS: Global Biosphere Emissions and International System (Guenther et al., 1999); MEGAN: Model of Emissions of Gases and Aerosols from Nature (Guenther, 2006); G95: Global BVOC emission model (Guenther et al., 1995): PRD-Pearl River Delta.    Wang et al. (2008) PRD refers to Pearl River Delta Region; HK refers to HongKong; BJ refers to Beijing; HN refers to Hainan; Su means summer; F means fall; W is winter. a The concentration of C5-alkenetriols.
X. Wu et al. Atmospheric Environment 189 (2018) 187-212 burning . A Chinese cooking source profile was measured in the studies of Zhao et al. (2007a) and Zhao et al. (2015b). During these studies, fatty acids (tetradecanoic acid, hexadecanoic acid, octadecanoic acid, oleic acid and linoleic acid) were the major contributor to OC, comprising 75.7% of the total mass of quantified organic compounds and with nhexadecanoic acid as the predominant compound. They are released by hydrolysis and thermal oxidation of glycerides (Rogge et al., 1991). Sterols, which include cholesterol, campesterol and β-sitosterol, are also found in Chinese cooking emissions (Zhao et al., 2007a). This observation is similar to that reported by He et al. (2004) that fatty acids comprise over 90% of the quantified organic mass emitted from cooking in Shenzhen, and fatty acids comprise 78% of measured compounds from western style fast cooking (Zhao et al., 2007b). However, in some previous studies in other countries, fatty acids emitted from meat cooking have been low (Rogge et al., 1991;Schauer et al., 1999). The composition of organic compounds is different among various cooking styles. For example, Cantonese style cooking emitted 39 μg of fatty acids per mg of POA, and released high amounts of sterols with 2 μg/mg of POA. The typical OC/EC ratios of 4.3-7.7 are used as an alternative indicator of kitchen emissions (See and Balasubramanian, 2008).

Secondary organic aerosols
Secondary organic aerosols (SOA) are formed from gas to particle conversion of volatile organic compounds (VOC) from both biogenic and anthropogenic sources by oxidation primarily with ozone (O 3 ) and hydroxyl radicals (OH), forming products of lower volatility that subsequently partition into the particle phase. Biogenic VOCs (BVOC) are released from terrestrial vegetation, grassland, shrublands, peatlands and forest, and include isoprene, monoterpenes, sesquiterpenes, and oxygenated hydrocarbons (Guenther, 2006). Anthropogenic VOC (AVOC) mainly come from biomass burning, fossil fuel combustion, transportation, solvent utilization and industry .

BVOC
Emission factors of BVOC have been estimated either from direct field measurements, or assigned based on the vegetation class of the plants which have not been measured experimentally. Most studies focus on isoprene, monoterpene and other volatile organic compounds (other VOC) emission factors (Tsui et al., 2009;Wang et al., 2003). In Table 9, the average emission rate of isoprene of 33.1 μg C g −1 h −1 is calculated from 39 vegetation types in Beijing (Wang et al., 2003), which is comparable to the result by Tsui et al. (2009), who measured the annual average EF of BVOC in Hong Kong from 13 tree species with 35 μg C g −1 h −1 for isoprene. Overall, these regional EFs of isoprene are higher than the average EF from China (10.3-11.8 μg C g −1 h −1 ) due to the different measurement methods and variable vegetation types present. For monoterpenes, the highest reported value of emission factor is from tree emissions (17.9 μg C g −1 h −1 in Hong Kong), and the lowest EF is from forest types (0.24 μg C g −1 h −1 ) (Klinger et al., 2002;Tsui et al., 2009). The other VOC emission factors for different plants and forests have been based on a value of 1.5 μg C g −1 h −1 (Geron et al., 1994;Wang et al., 2003) except for the value of 5.28 μg C g −1 h −1 from a study by Klinger et al. (2002), who measured the EF from 12 forest types. Estimates of isoprene and monoterpene emission rates from forest types have been calculated from weighted averages based on the species distribution (Li et al., 2013a). Some emission rates from shrub and crop species have been estimated from measurements made in China (Tsui et al., 2009;Wang et al., 2003).
The estimated annual BVOC emissions in China, Beijing, Hong Kong and Pearl River Delta (PRD) from 1994 to 2013 are shown in Table 9. It shows that the BVOC emissions in 2013 in China were 5.6 × 10 13 g C, including 3.2 × 10 13 g C isoprene, 0.6 × 10 13 g C monoterpenes and 1.5 × 10 13 g C other VOC (Li et al., 2016c). Li et al. (2013a) and Li and Xie (2014) estimated that the annual emissions during 1999-2003 ranged from 4.3 to 4.9 × 10 13 g C. This result is two and three times as high as the annual emissions from measurements by Klinger et al. (2002) and Chi and Xie (2012), which are 2.1 and 1.4 × 10 13 g C, respectively. The isoprene contributes the highest percentage of 55% (2.3 × 10 13 g C) to the total emission in 2003, followed by other VOC with 31.8% (1.4 × 10 13 g C) and monoterpenes with a lower emission at 13.1% (0.6 × 10 13 g C). The discrepancy in the estimates is likely due to the isoprene emissions, which used different emission factors, information on land cover distribution, meteorological parameters and models (Wang et al., 2016c). Moreover, the BVOC emission estimates exhibit varying proportions of isoprene, monoterpene and other VOC between different regions. The highest contribution of total emission in the PRD was other VOC of about 41% and isoprene contributed least with 25% (Zheng et al., 2010). However, the isoprene emission accounted for 48% of total BVOC emission in Beijing, which is higher than that from monoterpene (22%) and other VOC (30%) (Wang et al., 2003). The area average BVOC emission in Hong Kong is higher than that from PRD, Beijing and China, because the vegetation coverage in Hong Kong is higher than that in PRD and Beijing, and there is less forest cover in China as a whole (Klinger et al., 2002;Tsui et al., 2009).

AVOC
Emission factors of AVOC (anthropogenic VOC) have been estimated from different emission sources. Estimates of EF from anthropogenic categories were provided in previous studies (Li et al., 2009a;Wang et al., 2009bWang et al., , 2013Wang et al., , 2014Wu et al., 2016c). In general, the highest AVOC emission rates are from solvent utilization in China (pesticide utilization and dry cleaning) with 276-1000 g/kg, followed by the EF for VOC from biomass burning with 2.12-15.7 g/kg. The industrial processes emitted about 0.1-81.4 g VOC per kg original materials, except for the raw chemical medicine (430 g/kg) and polyfoam (770 g/kg). For transportation, the VOC EF of on-road vehicles, dependent on the technology level and type of fuels, ranged from 7.43-9.98 g/km for pre-Euro 1 gasoline to 0.14-0.2 g/km for Euro 5 gasoline in China, but there is less variation around 0.22-0.31 g/kg for the diesel EF. On the other hand, the EFs of off-road vehicles ranged from 6.1 to 91.5 g/kg, including railway engines, marine vessels, construction machinery, agricultural machinery, agricultural tractors and trucks .
The total annual AVOC emissions increased from 2.2 × 10 13 g in 2008 to 3.1 × 10 13 g C in 2013, dominating in many urban areas and regions Wu et al., 2016c). However, the contribution of different AVOC sources to the total emission varied in different studies. For example, Wu et al. (2016c) reported that the industry sector is the largest contributor to the AVOC emissions, accounting for 29-39% of total AVOC during 2008-2012, followed by transportation with 26% and then solvent utilization with 17%. On the contrary, the emissions in 2013 estimated by Li et al. (2016c) are dominated by solvent utilization, transportation and industry, which emitted 1.0 × 10 13 g (32.8%), 0.8 × 10 13 g (25.7%) and 7.6 × 10 12 g C (24.7%), respectively. The discrepancy is probably due to the different classifications between industry and solvent utilization. Generally, it indicates that industry and transportation are the main sector to target to reduce VOC emission, and solvent utilization plays an increasingly important role in the total AVOC emissions in China. Emissions from biomass burning are relatively stable (3 × 10 12 g-4.2 × 10 12 g C), but the relative contribution to total AVOC emissions decreased slightly from 17% in 2008 to 10% in 2013, which was due to the rapid growth of industrial emissions.
As discussed above, the annual BVOC emissions are higher than the annual AVOC emissions in China, but the AVOC dominate in urban and some economic regions. For example, Li et al. (2016c) estimated that the emissions of BVOC and AVOC are 5.3 × 10 13 g and 3.1 × 10 13 g in China, respectively. However, the AVOC emissions account for more than 80% of total VOC emissions in Beijing, Tianjin, Shanghai, and Jiangsu Province . A similar result was reported for Hong Kong by (Tsui et al., 2009), who estimated that AVOC and BVOC accounted for 79% and 21% of the total annual VOC emissions, respectively, owing to the high population and vehicle density in the megacity.

SOA yield
In previous studies, most parameters used in traditional modeling SOA mass loadings are based on observations from laboratory chambers. The SOA mass yield (Y) is defined as the ratio of aerosol mass concentration formed per mass of hydrocarbon reacted (ΔM, in μg m −3 ) to the amount of total precursor VOCs reacted (ΔHC, in μg m −3 ) Kroll and Seinfeld, 2008). Odum et al. (1996) have developed a semi-empirical model based on the gas-particle partitioning of semivolatile compounds as follow: Where α i is the mass of gas phase fraction for semi-volatile species i and K p,i is the partitioning coefficient of compound i. For example, Hao et al. (2011) determined that the SOA yield from α-pinene oxidation in O 3initiated chemistry was 28.9%. For aromatics, the SOA yield of 30% was presented by Gordon et al. (2014). Then, the SOA production from precursors could be estimated by the SOA yield Y and mass of the reacted precursor ΔHC . However, the calculated SOA abundance is normally less than ambient levels using the current chamber/flow tube yields. Table 10 lists commonly used molecular tracers of biogenic and anthropogenic SOA.

Anthropogenic SOA tracers
Several anthropogenic SOA tracers were measured in previous studies in China, including phthalic acids (o-, m-, and p-isomers), hydroxybenzoic acids (2-hydroxybenzoic acid, 3-hydroxybenzoic acid and 4hydroxybenzoic) and 3,4-dihydroxybenzoic acid for SOA derived from aromatic compounds, and 2, 3-dihydroxy-4-oxopentanoic acid for SOA from toluene (Baltensperger et al., 2005;Emanuelsson et al., 2013;Zhao et al., 2004). Table 1 lists annual mean concentrations of OC, OA, EC and PM 2.5 of major cities in China since 2008 (Lang et al., 2017;Wang et al., 2016b;Wu et al., 2016b). Annual concentrations of OC and EC in Chinese megacities (e.g., Beijing, Shanghai, Guangzhou, Xiamen, Nanjing and Tianjin) range from 3.3 to 27.8 μg/m 3 and 1.3 to 9.3 μg/m 3 respectively. The mean value of OC (14.5 μg/m 3 ) is five times as high as those typically observed in North America (3 μg/m 3 ) and three times higher than those in Europe; the mean EC concentration (4.4 μg/m 3 ) is approximately seven and four times higher than those in North America and Europe respectively (Table 11).

Atmospheric concentration of carbonaceous aerosols
The OC/EC ratio is commonly used to diagnose sources of organic aerosols and for estimation of SOA (Chen et al., 2006;Pio et al., 2011). Reported OC and OC/EC ratios from various functional areas in China are plotted in Fig. 4. Samples collected from near road sites are characterized by low OC/EC ratios ranging from 2.5 to 5.0, similar to that of gasoline vehicle exhaust reported by Schauer et al. (2002). Industrial Table 11 Mean carbonaceous aerosol mass concentrations in PM 2.5 observed in the primary megacities of the world cited from the literature (μg/m 3 ). TOT refers to NIOSH-TOT. a The values here refers to median concentration.
X. Wu et al. Atmospheric Environment 189 (2018) 187-212 emissions have a slightly higher OC/EC ratio ranging from 3.5 to 5.5, which is comparable with those from coal smoke reported in other studies with a OC/EC ratio of 2.5-10.5 (Chen et al., 2006). Residential and commercial sites exhibit a significantly lower OC/EC ratio (2.5-4.0) relative to the result from See and Balasubramanian (2008), who estimated the OC/EC ratio from typical kitchen emissions was 4.3-7.7. The difference in ambient data from these cooking sources is likely due to the contribution from vehicle exhaust and coal combustion to those sites. It is worth noting that OC/EC ratio at residential sites is relatively high in data collected at residential sites in Xiamen, which is attributed to the fuel used for cooking which changed to natural gas . Very high OC/EC ratios were observed in particles from biomass burning (7-11.5) (Wang et al., 2016a;Sun et al., 2014), consistent with that (4.1-14.5) measured by Watson et al. (2001). In Fig. 4, OC shows a wide range of concentrations from 3.2 to 26.4 μg/m 3 in residential areas.

Temporal variations of OA in China
In China in both urban and suburban areas (with the exception of some rural sites), organic aerosols were generally higher in winter and lower in summer. As shown in Fig. 5, in most of the Chinese cities the average concentrations of OA revealed a winter maximum (urban: 46.4 μg/m 3 ; suburban: 29.7 μg/m 3 ), followed by fall (urban: 29.0 μg/ m 3 ; suburban: 20.6 μg/m 3 ) and spring (urban: 20.0 μg/m 3 ; suburban: 16.9 μg/m 3 ), and a summer minimum (urban: 14.7 μg/m 3 ; suburban: 12.8 μg/m 3 ) (Feng et al., 2015;Ji et al., 2016;Li et al., 2015a;Tao et al., 2017b;Zhang et al., 2008bZhang et al., , 2014aZhao et al., 2013). Generally, the wintertime OA in urban ambient air is 2-5 times higher than that in summertime, while the values of OA during spring and fall are similar. However, in some suburban sites (e.g., Shanghai), the highest concentrations were found in fall, likely due to biomass burning activities (Feng et al., 2009).
This seasonality is mainly determined by the meteorological conditions and the variability in emission intensities. The low temperature and low boundary layer heights have been shown to affect significantly the accumulation of OA . For example, the low boundary layer height can promote SOA condensation and enhance the accumulation of semi-volatile organic compounds (SVOC) that can be subsequently adsorbed onto the existing aerosol surface (Cao et al., 2007;Zhang et al., 2013a). The low temperature facilitates SVOC condensation, thus increasing the OA concentrations (Pankow and Bidleman, 1992). Biomass burning is an important source of OA in China. Biomass burning activities are most prevalent in fall, leading to very high OA concentrations especially at the suburban sites Feng et al., 2009). In northern China, domestic heating, using both coal and biofuels, contributes to a high OA loading in winter . Moreover, the cold start of vehicles was suggested to have a stronger impact on the OA winter concentrations (Zheng et al., 2005). OA commonly exhibits lower concentrations in summer as a result of the greater boundary layer height that enhances the vertical dispersion of pollutants , although the strong solar radiation facilitates photochemical reactions leading to formation of oxidised VOCs (Tuet et al., 2017), and the high temperature, shifts the SVOC to the gas phase. In southern China, the organic aerosols can be washed out by the summer rain and are affected by air masses origins (Tao et al., 2017b). For instance, since the Pearl River Delta is in a transitional climate region experiencing both tropical and subtropical air masses, significant variations of air masses are found between different seasons. In summer, the PRD is mainly affected by the air masses from the China South Sea with low pollutant loadings, leading to the lowest concentrations of OA. On the contrary, the air masses in winter predominantly come from the north and northeast of the PRD where numerous pollutant sources exist, causing elevated OA levels (Zheng et al., 2009).
Hong Kong is one of the exceptions. Gao et al. (2016) indicated that the highest concentration of PM 10 OA is in fall and summer when taking an average from clean and haze days, not in winter, since the trafficrelated OA makes a relatively high contribution to OA in summer, when air masses come from the ocean and bring local emissions from downtown areas to this sampling site. In addition, the OA is likely dominated by SOA in fall.
Distinct seasonal variations in organic aerosols on the haze and nonhaze days have been observed in more recent studies (Huang et al., . These were approximately two times higher than those observed on non-haze days. The higher OC/EC ratio on haze days indicates that SOA is a key contributor to OAs during haze events .

Spatial variations of OA in China
In this section, concentrations are annual means, unless stated otherwise. Concentrations reported as OC in the original papers have been converted to OA using factors of 1.4 for Beijing (Song et al., 2007), 1.9 for Hong Kong  and 1.6 for all other cities (Cao et al., 2007). Table 1 summarizes the OA concentrations measured in different regions in China with at least one year of sampling from 2010 to 2014. Geographical locations of the sampling sites for the compiled data are presented in Fig. 1, which are grouped as the North China Plain (NCP: Beijing, Tianjin, Tanshan, Jinan, Zhengzhou), the Yangtze River Delta (YZRD: Shanghai, Nanjing, Hangzhou, Lin'an), Pearl River Delta (PRD: Guangzhou, Hong Kong, Zhuhai, Shenzhen), Southeastern coast (Fuzhou, Xiamen, Putian, Quanzhou), Southwest (Chengdu, Chongqing, Neijiang, Wanzhou), and other regions (Northeast and Northwest).
In the NCP, the annual mean concentrations of OA in 2010-2014 range from 17.3 to 42.2 μg/m 3 with an average of 28.9 μg/m 3 Gu et al., 2014;Ji et al., 2016;Zhao et al., 2013). OA concentrations show a decreasing trend in the NCP. For instance, the annual mean OA level of 28.0-67.5 μg/m 3 measured in 2006 in some cities of the NCP by Zhang et al. (2008b) exceeds current concentrations. Such a decrease is presumably due to improved implementation of emission control policies, such as a widespread switch from coal and biofuel to natural gas in residential areas (Ge et al., 2004). The concentration of EC in Beijing ranges from 2.1 to 8.6 μg/m 3 (mean: 4.8 μg/ m 3 ) and an average OC/EC ratio of 4.58 suggests that a high proportion of SOC is present in OC (Castro et al., 1999). This high level of OA and EC in Beijing is explained not only by the local anthropogenic sources X. Wu et al. Atmospheric Environment 189 (2018) 187-212 due to the large population living there, but also by transported pollution from surrounding industries. In the YZRD, the annual OA and EC concentration range from 12.8-28.8 μg/m 3 and 2.1-9.3 μg/m 3 , with mean values of 20.9 μg/m 3 and 4.5 μg/m 3 , respectively (Chang et al., 2017;Chen et al., 2010Chen et al., , 2016Feng et al., 2015;Li et al., 2015a). The OA concentration varied significantly depending on the level of industrialization and urbanization, population density and lifestyles (Feng et al., 2013). For example, the highest OA concentration was measured in downtown Nanjing surrounded by industry, and was relatively steady (22-28.8 μg/m 3 ) during 2007-2015 Li et al., 2016b). Moreover, there was a higher OA concentration (18.3-28 μg/m 3 ) at a suburban site than in urban Shanghai, suggesting that the organic pollutants were mainly affected by local sources. Highly effective strategies were used to reduce the particles, leading to a sharp decrease in the OA level from 23.5 μg/m 3 to 12.8 μg/m 3 from 2006 to 2014 due to the 2010 Shanghai World Expo (Tao et al., 2017a).
As one of the most economically developed regions with many industries and electric power plants in southern China, the PRD has not suffered from haze episodes as much as northern China. The annual mean OA is 8.6-15 μg/m 3 Tao et al., 2017b), which is 30-50% of that in the NCP. Moreover, the average OA in urban Guangzhou decreased by 45.9% and 9% in 2014 when compared to concentrations (24.2 μg/m 3 and 14.4 μg/m 3 ) measured in 2003, respectively (Cao et al., 2004Tao et al., 2014). However, the EC concentration was reduced by 33% from 2009 to 2014. Tao et al. (2017b) concluded that the dominant source of organic aerosols in Guangzhou is vehicle emissions owing to a good correlation (R 2 = 0.81) of OC and EC with a regression slope of 1.96 found in 2014. Therefore, the decline of OA concentration in recent years may be due to the efficient control measures applied to vehicle emissions (Lai et al., 2016;Tao et al., 2014). For the other coastal cities in the Southeast adjacent to the YRD and PRD (Fig. 1), the annual mean OA concentration was 12.9-19.1 μg/m 3 , which is lower than that in YRD and slightly higher than the value in PRD. Among these cities, the pollutant concentrations in Xiamen and Putian are influenced by a sea breeze (Zhang et al., 2011a).
Organic aerosol concentrations were high in Southwestern and Northwestern China with an average of 24.3-30.4 μg/m 3 and 29.8-34.9 μg/m 3 , respectively Wang et al., 2015d;Zhang et al., 2015b). The Sichuan Basin is in the agricultural heartlands of the Southwest, including Chengdu and Chongqing, which are influenced by anthropogenic organic matter sources, like biomass burning, coal combustion, high relative humidity and located in basins surrounded by mountains which inhibit the dispersion of pollutants . Compared with previous studies, the levels of OA in Chengdu in 2010-2015 were 38% lower than the 53.4 μg/m 3 OA measured in 2006 (Zhang et al., 2008b). The OA level in Chongqing was also 54% lower than 25.1-76.8 μg/m 3 measured in 2003 (Cao et al., 2007). It is clear that the organic aerosols in the Southwest have declined in the last decade.
The downward trends are mainly attributable to the pollution controls implemented to improve air quality including changes in fuels. A similar trend was observed in Xi'an in Northwestern China. The initial higher OA concentrations in Xi'an were mainly attributable to coal combustion emissions and unfavorable meteorological conditions (Cao et al., 2005). The annual concentration of OA there has decreased to 33.8 μg/m 3 in 2013 from 68.5 μg/m 3 in 2005 and 34.9 μg/m 3 in 2010 (Cao et al., 2007;Han et al., 2016a;Wang et al., 2015d), which is probably the consequence of pollution control measures, including the replacement of coal by natural gas. Furthermore, the higher OC/EC ratio (4-5.1) in winter corresponded to elevated residential coal combustion and biofuel for heating, and the mass of OC and EC resided predominantly in the PM 2.5 fraction Wang et al., 2015d). The difference between the OA concentration at the rural (24.5 μg/m 3 ) and the urban site (30.1 μg/m 3 ) is not very large, indicating that the organic component loadings at the rural site are affected by a combination of outflow from urban areas and local emissions (Wang et al., 2015d).

SOA tracer concentrations
As summarized in Table 10, the SOA tracer concentrations are highly dependent on locations and seasons. For example, the concentrations of 2-methylglyceric acid, C5-alkenetriols and 2-methyltetrols were 45, 18, 98 ng/m 3 at Mt. Tai in eastern central China, respectively (Fu et al., 2010). Their concentrations were higher at nighttime than in daytime owing to reduction of the planetary boundary layer (PBL) depth (Fu et al., 2010). Comparing the concentrations of total isoprene-SOA tracers between urban and rural locations in Beijing, the former site concentration of 73.8 ng/m 3 is lower than that at the latter of 93.8 ng/m 3 , presumably due to higher isoprene emissions . Isoprene-derived SOA is the predominant contributor to BSOA in early June (accounting for 65% of BSOA), which is higher than that in fall-winter owing to the higher emission and reaction rates in the summer (Ding et al., 2011). This result is similar to that in the Canadian High Arctic . The concentration of monoterpene tracers was 20-306 ng/m 3 in urban Beijing and 28-230 ng/m 3 in the rural Beijing area Guo et al., 2012;Yang et al., 2016). Ding et al. (2012) observed that the level of these compounds is slightly higher in the fall-winter (16.4 ng/m 3 ) than in the summer (11.6 ng/m 3 ) in PRD. This was attributed to the lower temperature, which results in an increase of tracer levels in the particle phase. On the contrary, Hu et al. (2008) estimated that monoterpenes and sesquiterpenes are the largest contributors to OA in the summer in Hong Kong due to high emissions of these biogenic hydrocarbons, consistent with those found in Hong Kong by Li et al. (2013b). Generally, there is no difference between the contribution of β-caryophyllinic acid in summer and fall-winter, with an average concentration of 3 ng/m 3 in both season (Ding et al., 2012). Moreover, Fu et al. (2010) observed that the sesquiterpene SOA tracers have the same level in daytime (12 ng/m 3 ) and nighttime (12 ng/m 3 ).
The contribution of SOA from anthropogenic precursors to total SOA is highly dependent on the location and season. They are the dominant source of SOA in the highly urbanized and industrialized PRD region. For instance, Ding et al. (2012) found that a tracer of precursor toluene (taken to represent all aromatics) ranged from 2.84 to 52 ng/m 3 and 1.7 to 48.9 ng/m 3 with the average of 17.8 and 13.1 ng/m 3 summer and winter-fall (Table 10), accounting for 76% and 79% of estimated SOC in summer and winter-fall, respectively. Recently, Hu et al. (2016) estimated in a modelling study that the country-average level of ASOA is 2 μg/m 3 , representing 21% of annual average SOA, 31% in the winter and 14% in summer. This was attributed to lower biogenic emissions of SOA precursors and higher emissions of ASOA in the winter, as well as the effect of cold temperatures and high humidity. These conditions favor the condensation of ASOA compounds. The toluene-derived SOA in the urban atmosphere (1.33 ng/m 3 ) contributes as much as that in the rural atmosphere (1.33 ng/m 3 ) in Shanghai, consistent with that in Hong Kong (1.77 ng/m 3 ) and in Beijing (13.3 ng/m 3 at urban and 11.7 ng/m 3 at rural sites). This suggests that ASOA is formed on a regional scale (Feng et al., 2013;Guo et al., 2012). A possible explanation for the higher ASOA in Beijing than other cities is the high regional SOA background concentration. Wang et al. (2009c) reported that aromatic precursors contributed about 55-65% of the SOA in the PRD.

Source apportionment results for OA
The OC/EC ratio, and individual molecular markers such as hopanes, PAHs and levoglucosan are commonly used to estimate the source contribution to POA and SOA (Shrivastava et al., 2007;Simoneit, 2002). Also, the application of chemical mass balance (CMB) and positive matrix factorization (PMF) receptor models has been widely reported in previous studies. In this section, we begin with a survey of recent laboratory and field observations, and discuss the estimation of POA and SOA concentrations.

Receptor modeling of OA
The most commonly used receptor models include the Chemical Mass Balance (CMB), isotopic mass balance using 14 C and factor analytical methods that include Positive Matrix Factorization (PMF) and Principal Component Analysis-Absolute Principal Component Scores (PCA-APCS), UNMIX, and multi-linear engine (ME) (Cao et al., 2005;Hu et al., 2010;Li et al., 2013c;Sun et al., 2012aSun et al., , 2013Zhang et al., 2014a). The application of CMB requires locally determined and quantitative knowledge of the emission source profiles. Given that the source profiles are used as inputs to CMB, it has the potential to separate sources clearly but is unable to identify unknown sources and secondary aerosols. On the contrary, PMF and PCA do not require known source profiles but are based on tracer species, e.g., those listed in Table 8 (Paatero and Tapper, 1994;Viana et al., 2008;Xu et al., 2016). These methods have the disadvantage that the interpretation of sources is often subjective as one tracer probably comes from more than one emission source, and that they are unable to separate the contribution from sources that covary in time. Hopke (2015) suggested that PCA is not really a receptor model, because it cannot provide an apportionment of the pollutants to the source. Absolute principal components analysis (APCA) was used by Thurston and Spengler (1985), which can provide a mass apportionment but the limitation of this model is that it does not consider the uncertainties of values of inputted variables. Mass spectrometric, online techniques, such as AMS, have merit for detecting the chemical processing of OA due to the output data carrying much chemical information and mass sensitivity . As mentioned above, AMS instruments have been widely used for PM 1 analysis and PMF analysis frequently conducted on the mass spectra (m/z 12-150) measured with AMS instruments to identify major organic components. For example, Sun et al. (2013) and Zhang et al. (2014a) divided the organic aerosol into five components by PMF, including low-volatility and semi-volatile oxygenated organic aerosol (LV-OOA and SV-OOA) that represents SOA, a hydrocarbon-like OA (HOA) that refers to POA related with urban emissions, a coal combustion OA (CCOA), and a cooking-emitted OA (COA). In addition, radiocarbon 14 C analysis combined with organic tracers is used to improve the direct differentiation of fossil fuel (FF) sources from modern carbon sources (non-fossil fuel, NF) Sun et al., 2012a).

Source contribution of POA
Results of receptor modelling of OAs in China are summarized in Table 12 and Fig. 6. About 60-80% of total OA was apportioned. Shown in Fig. 6, Feng et al. (2006) and Wang et al. (2009a) used a CMB model to estimate the contributions of seven sources (diesel/gasoline vehicles, coal burning, kitchen emission, vegetative detritus, cigarette smoke, biomass burning and other OC, usually taken to be SOC) to particulate organic matter in PM 2.5 . In Hong Kong, OC was dominated by the primary sources with 74% in locally influenced samples but the secondary sources were dominant in samples deriving from regional air masses with approximately 50-60% of OC (Li et al., 2013c). In Fig. 7, Liu et al. (2017a) have combined 14 C analysis with unique molecular organic tracers to apportion winter carbonaceous aerosols in 10 cities of China, resulting in the proportions of different carbon fractions as shown in Fig. 7. It includes EC bb (biomass burning-derived EC), EC f (fossil fuel combustion derived EC), POC bb (biomass burning-derived primary OC), POC f (fossil fuel combustion derived POC), SOC nf (nonfossil fuel derived SOC), and SOC f (fossil fuel-derived SOC). On average, the largest contributor of carbonaceous aerosols was the SOC nf , which represented 46%, followed by SOC f (16%), POC bb (13%), POC f (12%), EC f (7%), and EC bb (6%). Therefore, the SOC (62%) in carbonaceous aerosols sampled in winter in China was higher than the proportion of POC.
In Beijing, the seasonal trend in traffic-generated organic aerosols obtained by CMB models is unclear, but AMS-PMF showed an obvious diurnal trend with two distinct peaks corresponding to rush hours around 7:00-9:00 a.m. and 7:00-9:00 p.m. (Cheng et al., 2013a;Song et al., 2007;Zheng et al., 2005). For instance, Zheng et al. (2005) estimated that the contribution of vehicle emissions in Beijing is higher in autumn (6.73 μg/m 3 , 26% of OA) but lower in winter (3.11 μg/m 3 , 11.9% of OA) in 2000, which was attributed to reduced outdoor activities under cold synoptic conditions. By contrast, Cheng et al. (2013a) reported a higher level of OA derived from vehicle emissions in spring (18.5%) and lower in winter (9%) in 2012, which was attributed to relatively lower emissions from other sources (e.g., power plants and biomass burning) in spring. On the basis of isotopic and other analyses, Sun et al. (2012a) and Zhang et al. (2010) estimated that vehicular emissions are the most important contributor to the fossil source in the summer, but coal combustion dominates in the winter in Beijing. Traffic-related AMS hydrocarbon-like organic aerosol (HOA) and EC displayed two peaks corresponding to morning and evening traffic rush hours in Beijing, and the highest concentration maintained at midnight due to more heavy-duty vehicles and a lower mixed layer height (Ji et al., 2016;Xu et al., 2016). Feng et al. (2006) reported that the spatial distribution of hopanes is similar to that of engine exhaust. That study also found that the concentration of hopanes at an urban site (16.1 ng/ m 3 ) is higher than that at a rural site (12 ng/m 3 ) in Shanghai. This is consistent with the results in Beijing, Pearl River Delta Region, Hebei and Tianjin (Cai and Xie, 2007;Cao et al., 2003a;Wang et al., 2009a).
Coal combustion is shown by AMS to be a significant source of PM 2.5 organic aerosols that contributes 52% of OA during haze periods in Beijing (Elser et al., 2016). As shown in Table 12, coal combustion OA (CCOA) comprised of 33% of OA and 17% of non-refractory submicrometer aerosol (NR-PM 1 ) in the winter . In contrast, it contributes little to OA in the summer (10%) in 2011 in Beijing . This is due to the intensified emissions from coal combustion during the heating season. However, 15% of the OC came from coal combustion in Shanghai outside the heating season, suggesting that coal usage there mainly comes from industry (Fig. 6).
Regarding the diurnal trends in summer, the CCOA showed the highest values at midnight (18 μg/m 3 ) and lowest in the morning (approximately 7 μg/m 3 ). This trend was explained by the higher humidity at night that promotes transfer of the semi-volatile water-soluble organic aerosols from the gas phase to the particle phase (Sun et al., 2012b). A similar conclusion is drawn by Xu et al. (2016). Liu et al. (2017a) reported that industrial coal combustion and biomass burning are the main contributors to OAs in northern China and vehicle exhaust has a greater impact in south and central China from 14 C analysis (Fig. 7).
Biomass burning is a significant primary source of organic particles, accounting for 18-38% of the fine OA (Streets et al., 2003;Wang et al., 2007;Zhang et al., 2008a). In Beijing, biomass burning was an important contributor to OA (26.1%) in the winter, but only accounted for about 10% of OAs in the summer (Fig. 6). Levoglucosan exhibits seasondependent characteristics with highest concentrations (around 806 ng/ m 3 ) in fall and winter, but lowest levels in summer (Zhang et al., 2008b). This conclusion is consistent with the analysis of organic aerosols in Beijing by the 14 C technique (Yang et al., 2005). Duan et al. (2004) compared the contribution of biomass burning to OC between background and urban sites, and reported that a nearly double contribution of biomass burning was found in the background area (53%) compared to the urban area (28.3%).
The contribution of cooking emissions to the OA did not vary greatly between seasons. For instance, Wang et al. (2009a) estimated the contributions of cooking emissions to ambient OA in summer and winter in Beijing, were 7.7 μg/m 3 (accounting for 23.8% of OA) and 8.0 μg/m 3 (17.3% of OA) in 2007, respectively. There is approximately 14.5% of OA that was accounted for by the organic species in the cooking emissions . Based on the PMF-AMS technique, Sun et al. (2013) have found that its contribution to OA in PM 1 is 19% with the highest values of 32% during lunch and dinner time and the lowest of 11% in late afternoon during winter in Beijing. This number is approximately 20% in OA in Lanzhou province during summer with an average concentration of 5.86 μg/m 3 .
In addition, COA on average increased to 36% in clean periods, suggesting that COA plays a significant role as a local emission source of OA (Sun et al., 2012b. Natural POA derives from disintegration of bulk plant material, humic acid in soil dust and the presence of viable biological microbes like viruses, bacteria and fungal spores, which have sizes ranging from 1 nm to 300 μm. In rural areas, the concentration of organic aerosols 1. Source apportionment for PAH; 2. Including biomass burning, coal combustion and gasoline; 3. Results from PM 1 . Fig. 6. Summary of CMB apportioned source contributions to OC in PM 2.5 in Shanghai , Beijing Zheng et al., 2005), and Hongkong (Hu et al., 2010;Li et al., 2013c).
was influenced by larger biological particles, whereas in the urban area it shows a higher proportion of smaller particles. The sugar and sugar alcohols are excellent tracers for quantification of microbial activity, including pollen, fungi and bacteria. The total concentration of these is 203 ± 63 ng/m 3 at a suburban site and 142 ± 54 ng/m 3 at an urban site, indicating an enhanced biological contribution at the suburban site Yang et al., 2016). Wang et al. (2009a) found that vegetative detritus is also present in air, contributing 0.3% in summer and 0.5% in winter. Some studies have hypothesized the possibility of marine phytoplankton or sea-salt aerosol as potential source for OA, because marine phytoplankton can emit unsaturated fatty acids into the air while sea-salt aerosols may have an organic component from scavenging of surface active material from seawater (Jacobson et al., 2000;Yang et al., 2016). Fu et al. (2013) studied the organic aerosols over the Arctic Ocean based on a tracer-method and found that primary saccharides were dominant compounds in the marine OA, reaching up to 24.5 ng/m 3 .

Estimation of SOA
It is impossible to measure the concentration of SOC directly due to its complexity. Therefore, both EC-tracer and SOA-tracer methods are used to estimate the contribution of SOC to the organic aerosols. The EC-tracer method is based on the hypothesis that the EC comes from POA and the POC/EC ratio in POA is relatively constant. Using it, the SOC is calculated as follows: Where OC sec is the ambient secondary organic carbon and (POC/EC) is the ratio of primary aerosols. The determination of the POC/EC ratio is crucial, because it varies between different sources and can be affected by meteorological conditions. Previous studies have estimated this ratio during periods when SOA is very low, such as during winter, morning traffic periods with low solar radiation, under low ozone concentrations, and without rain and cloud events (Lim and Turpin, 2002). Generally, the OC/EC in the lowest 5-10% of values, or below the threshold of 2.9 was used by Li et al. (2015b). The SOA tracer based approach to estimation of SOA was first used by Kleindienst et al. (2007), who reported the mass fraction of the SOA (fsoa) on the basis of chamber simulations. It is defined as [SOA] i soa Where the Σ [tracers] is the sum of field measured SOA tracers derived from a certain individual or group of precursors, i, and [SOA] is the mass concentration of SOA derived from those precursors. A comparison by Ding et al. (2012) of the EC-tracer and SOA tracer methods showed very poor agreement in fall-winter, but relatively good in summer.
Using a tracer-based method, Fu et al. (2010) found that SOA tracers were formed during long range atmospheric transport due to unclear diurnal variations and estimated that the concentration of biogenic SOA is 2.6 μg C/m 3 , accounting for 10% of OA in East China in the summer. It differs between Beijing (4.4%) and Hong Kong (30%) in the same season due to the different geographical locations with different meteorological conditions, local vegetation cover, and oxidant levels Yang et al., 2016). For example, the BSOA measured in Hong Kong is affected by regional sources due to the land-sea breeze circulation which advects the pollutants from the PRD. Using an air quality modeling system (WRF/Chem) coupled with an SOA model, Jiang et al. (2012) found that anthropogenic sources accounted for 35% of total SOA over China in 2006, with 41%, 26%, 39% and 59% in spring, summer, autumn and winter, respectively. Subsequently, Hu et al. (2016) studied the distribution of SOA in the whole country in 2013 by a revised Community Multiscale Air Quality model with updated SOA yields, resulting in estimates of the contribution of biogenic SOA of 75% in summer, 50-60% in autumn and spring, and 24% in winter. Some other simulations of SOA over China have found that 65%-85% of SOA in southern China comes from biogenic precursors (Han et al., 2008;Jiang et al., 2012;Li et al., 2013b).

Conclusion and recommendations
The main primary sources of organic and elemental carbon aerosols account for 30-50% of PM 2.5 in China. Motor vehicle emissions contribute 13-50% of OC, especially high in megacities, which is reflected in OC/EC ratios or AMS measurements. Industrial emissions are major contributors in developed regions, like the PRD region, Shanghai and Zhejiang. 70% of rural energy derives from combustion of crop residues, and biomass burning has been considered as a dominant source of OA (accounting for 18-38%). Also, food cooking and coal combustion are reported as significant sources, whose contribution to OA is approximately 20% and 25% respectively. Secondary organic aerosol, X. Wu et al. Atmospheric Environment 189 (2018) 187-212 averaged across China, derived from biogenic precursors accounts for 24-75% of total SOA, while estimates of the anthropogenic fraction are from 25-76%. From the review above, there are some remaining issues that need to be addressed: 1. Further indicative tracers for SOA precursors need to be developed towards a comprehensive characterization of SOA sources in the field due to the complexity of SOA composition; 2. More effort should be given to improving comprehension of SOA formation mechanisms, especially during haze episodes in China; 3. The source profiles available for use in CMB models are limited, and represent some typical emission sources of OA, but do not contain comprehensive information on all relevant sources; 4. The inventory of OA emissions needs refinement of estimates and extension to include all sources. For example, the emissions from cooking are highly variable and influenced by the style and conditions of cooking; 5. Although receptor modeling for source apportionment is an effective method to calculate the contribution of different sources, differences in the results from different models need to be reconciled.