A review of current knowledge concerning PM 2 . 5 chemical composition , aerosol optical properties , and their relationships across China

South China Institute of Environmental Sciences, Ministry of Environmental Protection, Guangzhou, China Air Quality Research Division, Science and Technology Branch, Environment and Climate Change Canada, Toronto, Canada Key Laboratory of Aerosol Chemistry and Physics, Institute of Earth Environment, Chinese Academy of Sciences, Xi'an, China Key Laboratory of Regional Climate-Environment for Temperate East Asia, Institute of Atmospheric Physics, Chinese Academy of Sciences, Beijing, China

emission in coastal cities are other major source factors, each of which contributed 7-27 % to PM 2.5 mass in applicable cities.
The geographical pattern of scattering coefficient (b sp ) was similar to that of PM 2.5 , and that of aerosol absorption coefficient (b ap ) was determined by elemental carbon (EC) mass concentration and its coating.b sp in ambient condition of relative humidity (RH) = 80 % can be amplified by about 1.8 times that under dry conditions.Secondary inorganic aerosols accounted for about 60 % of aerosol extinction coefficient (b ext ) at RH greater than 70 %.The mass scattering efficiency (MSE) of PM 2.5 ranged from 3.0 to 5.0 m 2 g −1 for aerosols produced from anthropogenic emissions and from 0.7 to 1.0 m 2 g −1 for natural dust aerosols.The mass absorption efficiency (MAE) of EC ranged from 6.5 to 12.4 m 2 g −1 in urban environments, but the MAE of water-soluble organic carbon was only 0.05 to 0.11 m 2 g −1 .Historical emission control policies in China and their effectiveness were discussed based on available chemically resolved PM 2.5 data, which provides the much needed knowledge for guiding future studies and emissions policies.

Introduction
Knowledge of spatiotemporal variations in chemical and optical properties of atmospheric aerosols is needed in addressing regional and global air quality and climate issues (Fuzzi et al., 2015;Ginoux et al., 2012;Li et al., 2016c;Liao et al., 2015;Monks et al., 2009;Qian et al., 2015).Aerosol concentrations across China have been at extremely high levels in the recent two decades, largely caused by rapidly increased energy consumption (Chan and Yao, 2008;Fang et al., 2009;Guan et al., 2014;Wang and Hao, 2012;Zhang et al., 2013b).The frequency of haze has also been increased significantly due to light extinction of atmospheric aerosols, especially PM 2.5 (Li and Zhang, 2014;Pui et al., 2014;Watson, 2002).The Ministry of Environmental Protection of China thus promulgated NAAQS to include daily and annual PM 2.5 standards starting in early 2012.As a result, real-time PM 2.5 data in 74 Chinese cities have been recorded since 2013.
Light extinction, the sum of light scattering and absorption, is controlled by not only PM 2.5 levels, but also its chemical composition, size distribution and hygroscopic potential of its major components, and meteorological conditions (Hand and Malm, 2007a;Malm et al., 2003;Pitchford et al., 2007;Zhang et al., 2014a).High humidity combined with large fractions of hygroscopic chemical components (e.g., sulfate, nitrate, ammonium and some organic matter) can enhance light extinction and haze intensity (Liu et al., 2011(Liu et al., , 2013b;;Zhang et al., 2015b;Zieger et al., 2013).A large number of studies has been conducted in China in recent years investigating PM 2.5 composition, aerosol optical properties, aerosol hygroscopic properties and haze formation mechanisms (Guo et al., 2014;Jing et al., 2015;Liang et al., 2016;Liu et al., 2011Liu et al., , 2012;;Pan et al., 2009;Tao et al., 2014b;Wang et al., 2015b;Yan et al., 2008Yan et al., , 2009;;Yang et al., 2011b;Zheng et al., 2016).However, knowledge of long-term trends of PM 2.5 concentration, especially its major chemical components, is still limited (Fontes et al., 2017), and few studies have focused on the geographical pattern of PM 2.5 composition across China and its impact on aerosol optical properties (Li et al., 2017b).The present study aims to gain such knowledge through a thorough review of available studies.
Considering the large number of publications, only ground measurement data of chemical composition of PM 2.5 , aerosol scattering and absorption coefficients, and aerosol hygroscopic properties published after the year 2000 in scientific papers of Science Citation Index (SCI) journals are reviewed and summarized in this study.A total of about 150 articles met the above criteria including 100 articles on PM 2.5 chemical composition and source apportionment, 40 articles on aerosol optical properties and 10 articles on aerosol hygroscopic properties.Many of these articles focused on several of the biggest cities, such as Beijing, Shanghai, Guangzhou and Hong Kong, while other studies focused on cities such as Tianjin, Shijiazhuang, Jinan, Nanjing, Hangzhou, Fuzhou, Xiamen, Shenzhen, Chengdu, Chongqing, Xi'an, Lanzhou, Zhengzhou, Wuhan, Changsha, Haikou and several background sites (Fig. 2).Geographical and temporal patterns of PM 2.5 and its major chemical components including SO 2− 4 , NO − 3 , NH + 4 , organic carbon (OC) and EC, and aerosol optical properties are generated, source-apportionment analysis results are summarized and relationships between aerosol optical properties and PM 2.5 chemical composition are explored.Recommendations are also provided for alleviating PM 2.5 levels and reducing haze occurrence.
2 Spatiotemporal patterns of PM 2.5 and its major chemical components In this section, available measurements of chemically resolved PM 2.5 are reviewed and summarized in terms of geographical distributions, interannual variations and seasonal patterns.Measurements are grouped based on geographical regions, such as the Beijing-Tianjin-Hebei (BTH) in the North China Plain, the Yangtze River Delta (YRD), the Pearl River Delta (PRD), the Sichuan Basin and other regions (Fig. 1).Five dominant chemical components of PM 2.5 (SO 2− 4 , NO − 3 , NH + 4 , OC and EC) are discussed in detail.Data reviewed in this section are all listed in Table S1 in the Supplement.
Filter-based measurements of PM 2.5 at rural sites in China were limited and mainly conducted at Shangdianzi of Beijing, Conghua and Tianhu of Guangzhou, and Hok Tsui of Hong Kong (Hagler et al., 2006;Lai et al., 2016;Louie et al., 2005a;Zhao et al., 2013c).Rural PM 2.5 was around half of that in the cities of the same region.A similar geographical pattern was seen in rural PM 2.5 as in the urban, e.g., annual PM 2.5 at the rural site of BTH (Shangdianzi) was 72 µg m −3 , which was 2 times that (35 µg m −3 ) at the rural sites of PRD.

Seasonal patterns
In BTH, the highest seasonal average PM 2.5 concentrations were observed in winter and the lowest in summer in all the cities with seasonal variations up to factors of 1.7, 1.5, 1.6 and 1.8 in Beijing (Cao et al., 2012b;Chan et al., 2005;J. Chen et al., 2014a;Dan et al., 2004;Duan et al., 2006;Han et al., 2014;He et al., 2001;Huang et al., 2014b;Lin et al., 2016;Okuda et al., 2011;Pathak et al., 2011;Song et al., 2006aSong et al., , 2007;;Sun et al., 2004Sun et al., , 2006;;Tan et al., 2016a;Tao et al., 2016aTao et al., , 2015a;;Tian et al., 2015;Wang et al., 2005;Yang et al., 2005aYang et al., , 2016;;Zhang et al., 2013a;Zhao et al., 2013c), Tianjin (Cao et al., 2012b;Gu et al., 2010Gu et al., , 2011;;Li et al., 2009;Tian et al., 2016;Zhao et al., 2013c), Shijiazhuang (Zhao et al., 2013c) and Chengde (Zhao et al., 2013c), respectively.It is noted that major pollutant sources in BTH were located south of Hebei Province and the prevailing winds in BTH were from the north in winter and from the south in summer (Li et al., 2016b;Lu et al., 2010Lu et al., , 2011;;Wang et al., 2013;Xu et al., 2011).The location and distribution of major industrial sources, intensity of local minor sources such as winter heating, and prevailing wind directions together caused the slightly different magnitudes of seasonal variations among the four cities discussed above.Moreover, extreme weather events such as weakening monsoon circulation, depression of strong cold air activities, strong temperature inversion and descending air masses in the planetary boundary layer also played important roles in the strong PM 2.5 pollution during winter (Niu et al., 2010;Wang et al., 2014b;Zhao et al., 2013d).Several extreme wintertime air pollution events in recent years covered vast areas of northern China and were all correlated to some extent with extreme weather conditions (Zou et al., 2017).
Similar seasonal patterns as above were also observed in cities of other regions in China, such as Chengdu (Tao et al., 2013a(Tao et al., , 2014a)), Zhengzhou (Geng et al., 2013), Jinan (Yang et al., 2012) and Fuzhou (Xu et al., 2012b), with seasonal variations between a factor of 1.8 and 2.5.In conclusion, the highest seasonal average PM 2.5 observed in winter at all urban sites in China was likely due to high emissions from winter heating and/or poor pollutant dispersion.
To ensure the comparability of the data collected using different instruments, measurement uncertainties were first briefly discussed here.Most studies in China analyzed OC and EC using a DRI carbon analyzer or Sunset carbon analyzer.IMPROVE is the most widely used thermal-optical protocol for OC and EC analysis for the DRI analyzer, while NIOSH is the one used for the Sunset analyzer.OC and EC measured by the two analyzers are comparable if the same analysis protocol is used.For example, Wu et al. (2011) showed that OC from the Sunset analyzer was only 8 % lower than that from the DRI analyzer, while EC was only 5 % higher.However, when different protocols were used by the two analyzers, the differences were much larger, e.g., EC from NIOSH was almost 50 % lower than that from IM-PROVE (Chow et al., 2010;Yang et al., 2011a).Note that OC and EC were also measured using a CHN elemental analyzer in 2001-2002 in Beijing, which uses a similar protocol to NIOSH (Duan et al., 2006).In any case, the measurement uncertainties of total carbon (TC, the sum of OC and EC) were less than 10 % (Chow et al., 2010;Wu et al., 2011).
The ions including SO 2− 4 , NO − 3 and NH + 4 were measured by ion chromatography.Measurement uncertainties should be less than 15 % in most cases under strict QA/QC procedures (Orsini et al., 2003;Trebs et al., 2004;Weber et al., 2003), but could be larger for ammonium nitrate (NH 4 NO 3 ) since it can evaporate from the filters before chemical analysis under high temperature and low RH conditions, and this applies to both quartz fiber filter and Teflon filter (Keck and Wittmaack, 2005;Weber et al., 2003).The loss of NO − 3 due to evaporation was found to range from 4 to 84 % depending on the ambient temperature (Chow et al., 2005).Although the exact magnitudes of measurement uncertainties cannot be determined for NO − 3 and NH + 4 , they are not expected to significantly affect the interannual variations discussed below for the three cities (Beijing, Shanghai and Guangzhou) considering the small year-to-year temperature changes.

Interannual variations in Beijing
Chemically resolved PM 2.5 data in BTH covering multiple years are only available in Beijing and the interannual variations are discussed for this city below (Duan et al., 2006;He et al., 2001;Song et al., 2006a;Yang et al., 2011b;Zhang et al., 2013a;Zhao et al., 2013c).Interannual variations in OC and EC were generally small, e.g., a factor of 1.5 for OC and 1.8 for EC (Fig. 3a).OC decreased from 23.6-25.8µg m −3 in earlier years (1999)(2000)(2001)(2002)(2003)(2004)(2005)(2006) to below 17.6 µg m −3 after 2008.EC increased from 6.3 µg m −3 in 1999-2000 to 9.9 µg m −3 in 2001-2002, and then gradually decreased to 5.7 µg m −3 in 2009-2010.TC increased from 29.8 µg m −3 in 1999-2000 to 32.7-35.7 µg m −3 in 2001-2006, and then decreased to 23.3 µg m −3 in 2009-2010.The nearly 30 % reduction in TC in recent years in Beijing can be taken as a real trend since measurement uncertainties were believed to be around 10 % as mentioned above.OC is produced from both primary emissions and secondary formation, while EC (also known as black carbon or BC) is mainly from primary emissions.The anthropogenic emission for OC and BC over the entire China showed an increasing trend in 1996-2010 (Lu et al., 2011), while BC emissions showed a slightly decreasing trend in Beijing and Tianjin in 2005-2009(Qin and Xie, 2012).Meanwhile, BC emissions sharply increased in Hebei Province in [2005][2006][2007][2008][2009].The amount of BC emissions in Hebei Province was much higher than the sum of those in Beijing and Tianjin (Qin and Xie, 2012).Thus, the decrease in EC concentration in Beijing was likely dominated by local emission reduction instead of regional transport from Hebei province.
Annual SO 2− 4 concentration increased slightly during 1999-2010 and ranged from 10.2 to 16.4 µg m −3 in Beijing.SO 2 emission in China increased by about 60 % during 2000-2006 and then decreased about 9 % during 2006-2010 due to the compulsory flue-gas desulfurization equipment applied in power plants (Lu et al., 2011).However, the sum of the SO 2 emissions in BTH (including Beijing, Tianjin and Hebei Province) increased sharply from 2097 Gg yr −1 in 2000 to 2916 Gg yr  (Zhang et al., 2012b;Zhao et al., 2013b).Different from those of SO 2 emissions, NO x emissions in all the cities and provinces in BTH showed increasing trends in 2005-2010.NO x emissions in Beijing slightly increased from 410 Gg yr −1 in 2005 to 480 Gg yr −1 in 2010 (Zhao et al., 2013b).However, annual average surface NO 2 concentrations in Beijing showed a decreasing trend and fluctuated in the range of 49-66 µg m −3 during 2006-2009 (http://www.zhb.gov.cn/).There were some inconsistences between the trends of surface NO 2 concentrations and column NO 2 or NO x emissions, likely due to the impact of photochemical reactions on surface NO 2 concentrations in urban areas.To some extent, the increasing trend of NO − 3 in Beijing was likely related to the increases in NO x emissions in both Beijing and the surrounding cities or provinces.
Considering the potential large uncertainties in NH + 4 measurements, its trends should only be discussed qualitatively.NH + 4 concentrations were relatively steady in Beijing during 1999-2006, ranging from 5.7 to 7.3 µg m −3 .NH 3 emissions changed little (13 400-13 600 Gg yr −1 ) before 2005 in China, and increased slightly in BTH region during 2003-2010 (Zhou et al., 2015b).The small increase in NH + 4 in 2009-2010 in Beijing was consistent with the NH 3 emission trend in this region (Zhang et al., 2013a;Zhao et al., 2013c).Moreover, the increase in NO − 3 in Beijing was also an important factor contributing to the increase in NH + 4 .In summary, a decreasing trend was identified in TC and increasing trends were found for SO 2− 4 , NO − 3 and NH + 4 in Beijing.The interannual variations in EC agreed with the local emission trends in Beijing, but those in SO 2− 4 , NO − 3 and NH + 4 agreed more with the regional scale emission trends of their respective gaseous precursors in BTH rather than the local emission trends in Beijing.Nonlinear responses of concentration changes of these aerosol components to their respective emission trends were found, demonstrating other important factors potentially affecting aerosol formation.It is worth to note that several recent studies have highlighted the important role NO 2 might play in sulfate formation in the polluted environment in China (Cheng et al., 2016;Wang et al., 2016a;Xie et al., 2015a).Nevertheless, the aqueous SO 2 + H 2 O 2 /O 3 oxidation should still be the dominant mechanism in most cases, especially at a background site (Lin et al., 2017).The aqueous SO 2 + oxygen (catalyzed by Fe (III)) reaction can also be important under heavy haze conditions in north China (Li et al., 2017a).Extensive measurements of stable oxygen are needed to confirm the relative contributions of different sulfate formation mechanisms.

Relative contributions to PM 2.5
To investigate the relative contributions of dominant chemical components to PM 2.5 mass, the measured PM 2.5 mass was reconstructed based on SO 2− 4 , NO − 3 , NH + 4 , OM (organic matter) and EC.The conversion factor between OC and OM was 1.8 considering the prevailing biomass burning in BTH (Cheng et al., 2013a;Du et al., 2014a).
Data collected in 2009-2010 were first discussed since multiple cities in BTH have data during this period (Zhang et al., 2013a;Zhao et al., 2013c).Secondary inorganic aerosols (the sum of sulfate, nitrate and ammonium) contributed 36-39 % of PM 2.5 annually in the majority of the cities, but only 25 % in Chengde, a tourist city located in the northeast part of BTH and 200 km away from Beijing.Generally, the percentage contribution of secondary inorganic aerosols to PM 2.5 decreased with decreasing PM 2.5 level, e.g., from Shijiazhuang to Tianjin, Beijing and then Chengde, a phenomenon that is consistent with what was found within the same city but for different pollution levels in a winter season (Tao et al., 2015a).Carbonaceous aerosols (OM + EC) contributed 29-32 % to PM 2.5 in most cities, but as high as 45 % in Chengde, and had an opposite trend to that of secondary inorganic aerosols in terms of city-to-city variations.At the rural site Shangdianzi near Beijing, secondary inorganic aerosols and carbonaceous aerosols accounted for 42 and 32 %, respectively, of the PM 2.5 mass, which were not significantly different from those in cities located south of Yan Mountains.The sum of secondary inorganic aerosols and carbonaceous aerosols accounted for 65-70 % of the PM 2.5 mass in cities of BTH.

Interannual variations in Shanghai
Chemically resolved PM 2.5 data in YRD covering multiple years are only available in Shanghai (Ming et al., 2017;Wang et al., 2016b;Ye et al., 2003;Zhao et al., 2015b).Interannual variations in OC in this city were within a factor of 1.6 for OC and a factor of 4.1 for EC (Fig. 3b).OC concentrations were relatively steady (14.0-14.9µg m −3 ) during 1999-2009, but sharply decreased to 9.9-10.(Lu et al., 2011).The annual variations in SO 2 emission in YRD (including Shanghai, Jiangsu and Zhejiang) were relative small, ranging 3171 Gg yr −1 in 2000, 3506 Gg yr −1 in 2004, 3376 Gg yr −1 in 2007, and 3397 Gg yr −1 in 2010 (Lu et al., 2010;Zhao et al., 2013a).Annual average SO 2 concentrations in Shanghai were in the range of 45-61 µg m −3 during 2000-2005 and decreased by around 50 % to 24-29 µg m −3 during 2010-2013 (http: //www.zhb.gov.cn/).Note that SO 2 emissions in Shanghai only accounted for less than 20 % of the total SO 2 emissions in YRD and with small annual variations.The high concentrations of SO 2− 4 observed in Shanghai were also closely related to regional transport from north China (e.g., BTH and Shandong Province) (Li et al., 2011;Wang et al., 2016b).
Annual NO − 3 concentrations in Shanghai were relatively steady (6.0-7.7 µg m −3 ) during 1999-2009, but sharply increased to 13.3 µg m −3 in 2011-2014.NO x emissions in YRD also showed an increasing trend during these years, consistent with satellite retrieved vertical column NO 2 densities during 2000-2010 (Zhang et al., 2012b;Zhao et al., 2013b).In contrast, surface-level annual NO 2 concentrations in Shanghai sharply decreased from 90 µg m −3 in 2000 to a range of 48-61 µg m −3 during 2003-2013 (http://www.zhb.gov.cn/).The inconsistency in the trends between emissions and gaseous and particulate surface air concentrations was also found in Beijing.Photochemistry and regional transport of related pollutants should be the major causes of this phenomenon.
Annual NH + 4 concentrations decreased from 5.9 µg m −3 in 1999-2000 to the levels of 4.1 µg m −3 in 2009 and then increased to 6.6 µg m −3 in 2011-2014.NH 3 emissions increased in 2000-2005 in east China (including BTH, YRD and PRD) and possibly also increased in 2006-2010 due to the lack of control measures for NH 3 in China (Wang et al., 2011).The recently increased NH + 4 concentrations in Shanghai were likely due to the concurrent increases of NH 3 emissions and NO − 3 concentrations.In summary, a decreasing trend was identified in TC, increasing ones for NO − 3 and NH + 4 , and a stable one for SO 2− 4 in Shanghai.The interannual variations in EC agreed with its local emission trends in Shanghai rather than regional emission trends.In contrast, interannual variations in SO 2− 4 , NO − 3 and NH + 4 agreed more with the regional scale emission trends of their respective gaseous precursors in YRD.Similar to Beijing, nonlinear responses of concentration changes of these aerosol components to their respective emission trends were also found in Shanghai.

Relative contributions to PM 2.5
The chemical compositions in PM 2.5 between the cities in YRD were compared between Shanghai and Nanjing due to the lack of continuous annual data in Hangzhou.A conversion factor of 1.6 between OC and OM was chosen for YRD, slightly smaller than that (1.8) chosen for BTH considering the lower impact of biomass burning on PM 2.5 in this region (Feng et al., 2006;Li et al., 2016a).Secondary inorganic aerosols contributed 25-54 % of PM 2.5 annually in Shanghai and Nanjing, while carbonaceous aerosols contributed 28-47 % (Li et al., 2016a;Wang et al., 2016b;Ye et al., 2003;Zhao et al., 2015b).The sum of secondary inorganic aerosols and carbonaceous aerosols accounted for 76 and 66 % of PM 2.5 mass in Shanghai and Nanjing, respectively, which was comparable with that (65-70 %) in BTH.
In Hangzhou, seasonal contributions can only be estimated for summer and winter 2003 (Cao et al., 2012b).Seasonal contribution of secondary inorganic aerosols in winter was 44 %, which was evidently higher than that in summer (34 %), while carbonaceous aerosols contributed 33-35 %.At the rural sites (Ningbo and Lin'an) in Zhejiang Province, seasonal contributions of carbonaceous aerosols varied within a small range (28-34 %) in four seasons in 2008-2009, which were comparable with those in Hangzhou (Feng et al., 2015;Liu et al., 2013a).
In summary, the different seasonal average contributions of secondary inorganic aerosols and carbonaceous aerosols in Shanghai and Nanjing were likely due to different local sources in YRD.The seasonal patterns of these chemical components in Shanghai were a result of both local emissions and regional transport, but in Nanjing, the seasonal pattern was mainly determined by local emissions because it is an inland city surrounded by many industrial enterprises, including power plants, petrochemical plants and steel plants.
In contrast to the TC trend, annual SO  2000, 2004, 2007and 2010, respectively (Lu et al., 2010;;Zhao et al., 2013a).However, SO 2 emissions in PRD decreased by more than 40 % between 2005 and 2009, due to flue gas desulfurization facilities in power plants and large industrial boilers installed in this region (Lu et al., 2013).Annual average SO 2 concentrations in Guangzhou gradually increased from 45 µg m −3 in 2000 to 77 µg m −3 in 2004, and then decreased to 17 µg m −3 in 2014 (http://www.gzepb.gov.cn/).Thus, the increased SO 2− 4 concentration before 2010 in Guangzhou was largely due to regional transport of pollutants from outside of PRD.The decreased SO 2− 4 concentration in 2014 in Guangzhou was likely due to flue gas desulfurization facilities in power plants and large industrial boilers implemented across the entire Guangdong Province (http://www.gdep.gov.cn/).
Meanwhile, NO x emissions increased in Guangdong Province as well as across the entire PRD, similar to the trends in BTH and YRD (Lu et al., 2013;Zhang et al., 2012b;Zhao et al., 2013b).However, annual average surface NO 2 concentrations in Guangzhou fluctuated from 61 to 73 µg m −3 during 2000-2007 and from 48 to 56 µg m −3 during 2008-2014 (http://www.gzepb.gov.cn/).An opposite trend was also found between NO 2 and NO x emissions with the former persistently decreased, while the latter increased in Guangzhou, although NO − 3 concentrations also increased.Thus, emissions as well as chemical processes both affected these ion concentrations in air.Annual NH + 4 concentrations slightly increased about 10 % before 2010 in Guangzhou and Shenzhen although NH 3 emissions changed little during 2002-2006 in PRD (Zheng et al., 2012).Thus, the slightly increased NH + 4 concentrations, if not caused by measurement uncertainties, in Guangzhou and Shenzhen during 2002-2010 were largely due to the increased SO 2− 4 and NO − 3 , which enhanced the conversion of NH 3 to NH + 4 .In summary, a decreasing trend was identified in TC and increasing trends were found for SO 2− 4 , NO − 3 and NH + 4 in Guangzhou and Shenzhen before 2010, while all chemical components decreased after 2010 in Guangzhou.The interannual variations in EC were inconsistent with BC emission trends in Guangdong Province.In contrast, interannual variations in SO 2− 4 , NO − 3 and NH + 4 agreed with regional scale emission trends of their respective gaseous precursors in Guangdong Province rather than PRD.Similar to Beijing and Shanghai, nonlinear responses of concentration changes of these aerosol components to their respective emission trends were also found in Guangzhou and Shenzhen.
Relative contributions to PM 2.5 Data collected in 2002-2003 were discussed since multiple cities (e.g., Guangzhou, Conghua, Zhongshan, Shenzhen and Hong Kong) in PRD have data during this period (Hagler et al., 2006).The conversion factor between OC and OM was chosen to be the same as in YRD (1.6).Secondary inorganic aerosols contributed 33-38 %, depending on location, of PM 2.5 annually, while carbonaceous aerosols contributed 37-46 %.It is noted that PM 2.5 in Guangzhou was much higher than those in other coastal cities (including Zhongshan, Shenzhen and Hong Kong), but the contributions of secondary inorganic aerosols and carbonaceous aerosols were not significantly different between these cities.At rural sites (Tianhu and Conghua near Guangzhou and Hok Tsui near Hong Kong), secondary inorganic aerosols and carbonaceous aerosols accounted for 35-48 and 24-43 %, respectively, of the PM 2.5 mass, which were similar to those obtained in the cities in PRD (Hagler et al., 2006;Lai et al., 2016;Louie et al., 2005b).Thus, the sum of secondary inorganic aerosols and carbonaceous aerosols accounted for 68-83 % of the PM 2.5 mass in the PRD region, similar to Shanghai (YRD).Although many studies have been conducted in PRD, many of them were short term.Studies covering all four seasons were mainly carried out in Guangzhou and Hong Kong (Fig. 6) (Andreae et al., 2008;Cao et al., 2004Cao et al., , 2012b;;Cui et al., 2015;Ho et al., 2006a;Huang et al., 2014b;Jung et al., 2009;Lai et al., 2007;Liu et al., 2014a;Louie et al., 2005a;Tan et al., 2009;Tao et al., 2009Tao et al., , 2014cTao et al., , 2015bTao et al., , 2017;;Yang et al., 2011b).Seasonal average contributions of secondary inorganic aerosols were generally higher in spring and autumn than in summer and winter in both Guangzhou and Hong Kong.If all the years of data were averaged together, secondary inorganic aerosols contributed 43, 31, 38 and 33 % in spring, summer, autumn and winter, respectively, to PM 2.5 in Guangzhou and 45, 25, 46 and 37 %, respectively, in Hong Kong.However, different seasonal patterns were found between Guangzhou and Hong Kong for carbonaceous aerosols.Carbonaceous aerosols contributed 34, 37, 35 and 34 % in spring, summer, autumn and winter, respectively, to PM 2.5 in Guangzhou and 54, 47, 49 and 38 %, respectively, in Hong Kong.Seasonal variations in OC / EC ratios ranged from 1.6 to 3.4 in Guangzhou and from 1.2 to 2.1 in Hong Kong, suggesting coal combustion and vehicle exhaust were the dominant sources in Guangzhou, while vehicle exhaust was the dominant source in Hong Kong (He et al., 2008;Watson et al., 2001).

Other cities
Besides the cities in BTH, YRD and PRD, synchronous measurements of PM 2.5 and the dominant chemical components have also been conducted in several cities in other regions of China, mainly in the capital city of a province (e.g., Zhengzhou of Henan Province: Geng et al., 2013;Xi'  was chosen for Fuzhou, Xiamen and Haikou and 1.8 for other cities based on their geographical locations. Annual average contributions of secondary inorganic aerosols and carbonaceous aerosols to PM 2.5 were 30 and 36 %, respectively, in the island city Haikou, similar to what was found in Beijing, and were 43-46 and 29-36 %, respectively, in the coastal cities Fuzhou and Xiamen, similar to what was found in Shanghai.Annual contributions of secondary inorganic aerosols ranged from 29 to 39 % in inland cities (Zhengzhou, Xi'an, Jinan, Chengdu, Chongqing and Changsha) except Lanzhou (15 %), which were comparable with those observed in PRD (33-41 %).In contrast, large differences were found in the annual contributions of carbonaceous aerosols, ranging from 23 % in Zhengzhou to 47 % in Chongqing.The sum of secondary inorganic aerosols and carbonaceous aerosols accounted for 56-79 % of the PM 2.5 mass in these cities.
At an Asian continental outflow site (Penglai in Shandong Province), annual average contribution of secondary inorganic aerosols to PM 2.5 reached 54 % (Feng et al., 2012b), evidently higher than those in urban and inland rural sites in China, while that of carbonaceous aerosols was 31 %, close to those in BTH.This finding suggested that intensive emissions of SO 2 and NO x in China enhanced the downward transport of secondary inorganic aerosols to the Pacific Ocean.
Seasonal average contributions are only shown here for Jinan (Yang et al., 2012), Zhengzhou (Geng et al., 2013), Fuzhou (Xu et al., 2012b), Chengdu (Tao et al., 2013a(Tao et al., , 2014b)), Lanzhou (Tan et al., 2016b;Wang et al., 2016c), Xiamen (Zhang et al., 2012a), Changsha (Tang et al., 2017) and Haikou (Liu et al., 2017a) due to the incomplete data in Xi'an and Chongqing (Fig. 7).Seasonal contributions of secondary inorganic aerosols were evidently higher in summer than in other seasons in Zhengzhou, Jinan and Lanzhou (typical northern cities), similar to what was seen in BTH.In the southwest city Chengdu and the central city Changsha, seasonal contribution of secondary inorganic aerosols in spring was only 30 and 27 %, respectively, lower than other seasons (40-42 and 30-31 %, respectively).In the two southern coastal cities Fuzhou and Xiamen, the highest seasonal average contribution of secondary inorganic aerosols was observed in winter (53 and 33 %, respectively), much higher than in other seasons (34-42 and 21-24 %, respectively).In the southern island city Haikou, seasonal contributions of secondary inorganic aerosols were also slightly higher in winter (30 %) than in other seasons (21-27 %), similar to the coastal cities Fuzhou and Xiamen.
Seasonal average contributions of carbonaceous aerosols were evidently higher in cold seasons than in warm seasons in the three northern cities (Zhengzhou and Jinan and Lanzhou) due to heating activities and biomass burning, similar to BTH. Surprisingly, a similar seasonal pattern was also found in one coastal city Xiamen, e.g., 38 % in winter versus 27-30 % in other seasons.In contrast, higher seasonal contributions were found in warm season than in cold seasons in the southern coastal city (Fuzhou) and the southern island city (Haikou).No seasonal variations were found in the southwest inland city Chengdu (29-32 %) and the central inland city Changsha (28-33 %).The summed contributions of secondary inorganic aerosols and carbonaceous aerosols were evidently lower in spring than in other seasons in most of the northern cities (e.g., Jinan, Lanzhou, Zhengzhou and BTH), likely due to the frequent spring dust storm events in northern China.

Summary of PM 2.5 chemical properties
Carbonaceous aerosols showed decreasing trends over the last 10 years (2000)(2001)(2002)(2003)(2004)(2005)(2006)(2007)(2008)(2009)(2010) in Beijing, Shanghai and Guangzhou, consistent with BC emission trends in these cities and surrounding areas.SO 2− 4 and NH + 4 remained at high levels with no significant trends in Beijing and Shanghai, but with an increasing trend in Guangzhou.NO − 3 showed increasing trends in all of the above-mentioned megacities.Annual mass concentrations of PM 2.5 , secondary inorganic aerosols and carbonaceous aerosols showed similar spatial gradients decreasing from high-to low-latitude regions.Annual average contributions of secondary inorganic aerosols to PM 2.5 ranged from 25 to 48 % with higher values in southern regions, and those of carbonaceous aerosols ranged from 23 to 47 %, also with higher values in southern regions (Fig. 8).The percentage contributions of the sum of secondary inorganic aerosols and carbonaceous aerosols were higher in southern cities than in northern cities due to the frequent dust events in the north.
The highest seasonal average contributions of secondary inorganic aerosols to PM 2.5 were observed in summer in most of the northern cities, but can occur in different seasons in southern cities.In contrast, the highest seasonal contributions of carbonaceous aerosols were observed in cold seasons in most of the northern cities, and in warm seasons in most of the southern cities.The different seasonal patterns were largely caused by heating and biomass burning in cold seasons in north China.

Source apportionment of PM 2.5
Advantages of receptor-based methods used for source apportionment analysis for various pollutants were discussed in I. Cheng et al. (2015) and Hopke (2016).Source apportionment studies of PM 2.5 in China using receptor models have also been reviewed recently covering a wide range of topics (Liang et al., 2016;Lv et al., 2016;Pui et al., 2014;Zhang et al., 2017b).However, a general summary of spatial-temporal patterns of PM 2.5 source factors and their relative contributions is still lacking, which is the focus of the discussion below.Data collected in this section are listed in Table S2 in the Supplement.
Commonly used receptor models in source apportionment of PM 2.5 in China include Principal component analysis/absolute principal component scores (PCA/APCS), chemical mass balance receptor (CMB), positive matrix factorization (PMF) and UNMIX and the Multilinear Engine-2 model (ME-2).Among these, PMF and CMB models were the most widely used in China.Quantitative assessments of the uncertainties in using these methods are rare; studies us-ing the same dataset collected in 2000 in Beijing and applying the above-mentioned models suggested that, while the models still identified the same dominant source factors, the relative contributions from these source factors differed by as much as 30 % between the different models (Song et al., 2006a, b).Similar magnitudes of uncertainties could also be caused by using different biomass burning tracers despite using the same receptor model (Tao et al., 2016b).
Major source factors identified for PM 2.5 in most Chinese cities include secondary inorganic aerosols (SIA), coal combustion (COAL), biomass burning (BIOM), traffic emission (TRAF), dust emission (DUST) and industrial emission (INDU).Other source factors were also identified (and sometimes due to using more specific source names), such as metal manufacturing (including iron and steel industry, copper smelting) in industrial cities (e.g., Dongying and Tai'an of Shandong Province, Nanjing, Hangzhou, Lanzhou, Chengdu, Chongqing and Changsha), and sea salt and ship emissions in coastal cities (e.g., Longkou of Shandong Province, Nanjing, Guangzhou, Zhuhai and Hong Kong).
Contributions of dominant source factors to PM 2.5 are discussed below in detail on a regional basis.
A study in Tianjin in 2013-2014 only identified four dominant sources (SIA, COAL, TRAF and DUST) using the Multilinear Engine-2 model (ME-2), which accounted for 41, 25, 14 and 20 %, respectively, of the annual average PM 2.5 mass (Tian et al., 2016).Compared with results in 2012-2013 in Beijing (Zíková et al., 2016), the contributions of SIA were comparable in the two cities, but those of COAL and DUST were much higher in Tianjin than Beijing.However, the results from an earlier study in Tianjin in 2009-2010 were much more comparable to those in Beijing during the same years in terms of PM 2.5 levels and source attributions (Zhao et al., 2013c), implying faster decrease in COAL contribution in Beijing than Tianjin.
Seasonal results of source apportionment analysis are also available for Beijing (Huang et al., 2014b;Song et al., 2007;Wu et al., 2014;Zheng et al., 2005;Zhang et al., 2013a;Zíková et al., 2016).In most cases, SIA was the largest contributor in spring, summer and autumn, accounting for 26-61 % of the PM 2.5 mass, while COAL was the largest contributor in winter, accounting for 13-57 % of the PM 2.5 mass.The contributions of the other sources were lower than those of SIA and COAL, but subject to seasonal variations.For example, the largest seasonal contribution of BIOM was in autumn and of DUST in spring.

The Yangtze River Delta region
Studies for 1 year or longer were only made in Nanjing (Li et al., 2016a) and Hangzhou (Liu et al., 2015).Metal manufacturing was identified as a source factor in both Nanjing and Hangzhou, while ship emissions were also identified in Nanjing.Annual contributions of SIA to PM 2.5 mass reached 68 % in Nanjing, while all the other sources (COAL, DUST, sea salt and ship emissions, and metal manufacturing) each contributed 10 % or less.In contrast, metal manufacturing, SIA, TRAF and COAL accounted for 32, 28, 17 and 13 %, respectively, of the PM 2.5 mass in Hangzhou.Evidently, the contributions of SIA in Nanjing were much higher than those in Hangzhou and cities in BTH.The contributions of COAL in Nanjing and Hangzhou were similar, but were evidently lower than those in cities in BTH.
Similar to the cities in BTH, the largest seasonal average contribution of SIA in Nanjing was in summer and of COAL in winter.Only winter data were available in Shanghai (Huang et al., 2014b), and the contributions of SIA and DUST to PM 2.5 were similar between Shanghai and Nanjing.
Slightly different sources factors were identified in suburban studies.Annual contributions from mixed source (from regional transport), secondary nitrate and chloride, ship emissions, COAL, and electronic industries accounted for 36, 20, 18, 13, and 13 %, respectively, of the PM 2.5 mass in suburban Zhuhai, while SIA, BIOM, sea salt, residual oil combustion (related to Ni and V, or ship emission), DUST,and TRAF accounted for 39,20,17,12,7, and 5 %, respectively, of the PM 2.5 mass in suburban Hong Kong.
Despite the slightly different source factors identified between urban and suburban sites in PRD, SIA was the largest contributor to PM 2.5 mass in this region.Ship emissions were identified in this region, but not in northern China, and this source factor contributed more than 10 % of the PM 2.5 mass in all the studies except the one for Shenzhen.Similar to the cities in northern China, a high contribution from coal combustion was also found in Guangzhou and suburban Zhuhai.
Seasonal results of source apportionment analysis were available for four seasons in suburban Hong Kong (Huang et al., 2014c), winter in Guangzhou (Huang et al., 2014b), and summer and winter in Foshan (Tan et al., 2016c) and Hong Kong (Ho et al., 2006b).SIA was the largest contributor to PM 2.5 among all the identified source factors in every season in suburban Hong Kong (30-45 %) and in winter in Guangzhou (59 %).In contrast, INDU was the largest contributor in winter in Foshan (39 %), a typical industrial city in PRD (Tan et al., 2016c).In suburban Hong Kong, seasonal average contribution of SIA was the lowest in summer, different from what was found for cities in BTH and YRD, while that of sea salt and ship emissions were the highest in sum-mer due to the prevailing air masses from the South China Sea (Huang et al., 2014c).

Other cities
Studies covering 1 year or longer were mostly conducted for provincial capital cities, including Jinan (Yang et al., 2013), Zhengzhou (Geng et al., 2013), Xi'an (Wang et al., 2015c), Lanzhou (Wang et al., 2016c), Chengdu (Tao et al., 2014a), Chongqing (Chen et al., 2017), Changsha (Tang et al., 2017), Wuhan (Xiong et al., 2017), Xiamen (Zhang et al., 2016) and Haikou (Liu et al., 2017a), and for an inland city Heze (Liu et al., 2017b) and a regional background site (located in Yellow River Delta National Nature Reserve in the city of Dongying) (Yao et al., 2016), both in Shandong Province.Annual results were available from most studies, but were aggregated from seasonal results for Wuhan and Haikou.All the sites were grouped into four regions for easy discussion, i.e., northwest China (Lanzhou and Xi'an), southwest China or Sichuan Basin (Chengdu and Chongqing), eastern and central China (Jinan, Zhengzhou, Heze, Dongying, Wuhan and Changsha), and south coastal cities (Xiamen and Haikou).
The two northwest cities showed the same top four dominant source factors, although with slightly different percentage contributions to PM 2.5 mass, e.g., 29 % from SIA, 19 % from COAL, 17 % from DUST and 15 % from TRAF in Xi'an, and 17 % from SIA and 22 % from the other three sources in Lanzhou.The lower SIA contribution in Lanzhou was likely due to the dry climate inhibiting formation of SIA.Similar results to those in Xi'an were also obtained in rural Xi'an, with SIA, COAL, DUST and TRAF contributing 31, 16, 20 and 13 %, respectively, to the PM 2.5 mass (Wang et al., 2015c).The two southwest cities (Chengdu and Chongqing) showed nearly the same source-apportionment analysis results with SIA contributing just below 40 % and COAL and INDU each contributing around 20 % to PM 2.5 mass.
The same top four dominant source factors (SIA, COAL, DUST and TRAF) found in the two northwest cities were also found in other capital cities (Jinan, Zhengzhou and Wuhan, except Changsha) and a medium-sized city (Heze) in eastern and central China, which accounted for 24-55, 14-23, 5-26 and 5-27 % of the PM 2.5 mass, depending on location.In Changsha, SIA, a mixed source of INDU and BIOM, and DUST accounted for 60, 27 and 13 %, respectively, of PM 2.5 mass.SIA, BIOM and INDU were the most important sources, accounting for 54, 16 and 16 %, respectively, of PM 2.5 mass in a regional background site in Dongying.Similar source-apportionment results were found between the two south coastal cities (Xiamen and Haikou) with SIA, TRAF, DUST, COAL and sea salt accounting for 20-27, 16-21, 12-22, 8-9 and 6-10 %, respectively, of the PM 2.5 mass.
2.3.5 Summary of PM 2.5 source apportionment studies SIA, COAL and TRAF were the dominant source factors in most cities in China (Fig. 10).On an annual average, the sum of these three factors accounted for 63-80 % of PM 2.5 mass in the cities of BTH region, 58-78 % in the cities of YRD region, 51-67 % in the cities of PRD region, 51-61 % in the northwest cities, 57-60 % in the southwest cities, 57-82 % in the eastern and central cities, and 44-57 % in the south coastal cities.The contributions of DUST were significant (7-26 %) in northern cities and a central city (Zhengzhou), of INDU significant (19-27 %) in typical industrial cities (e.g., Chengdu, Chongqing, Changsha), and of ship emission significant (7-19 %) in coastal and river cities (e.g., Longkou, Nanjing, Guangzhou, Zhuhai, Hong Kong).High seasonal contributions were found for SIA in summer, COAL in winter, DUST in spring and ship emission in summer in applicable cities.
It should be noted that SIA chemical compounds are formed by reactions in the atmosphere involving primary emissions of gaseous precursors that can be produced from any of the identified sources factors discussed above as well as from sources seldom mentioned in source apportionment studies such as agricultural emissions and many natural sources.However, sources of the gaseous precursors are often undetermined in source apportionment studies resulting in a large proportion of PM 2.5 that cannot be explained (Karagulian et al., 2015).If the SIA contributions can be allocated to specific types of primary emissions, the overall percentage contributions from each of the identified source factor should be much higher, especially for COAL, TRAF, INDU and BIMO due to their high emission rates of primary pollutants of gaseous species.To identify the various types of primary emission sources, datasets containing trace element and other chemical markers need to be included in source apportionment models (Lee and Hopke, 2006).Combining receptor-based analysis results with source-based studies us- ing chemical transport models can provide a more complete picture by quantifying contributions of dominant emission sources to PM 2.5 pollution.
Aerosol optical depth (AOD), representing the integrated light extinction coefficient in a vertical column, can be obtained from MODerate-resolution Imaging Spectroradiometer (MODIS) data.Satellite retrievals of AOD have been widely applied to estimate surface PM 2.5 concentrations using statistical models (Liu et al., 2005;Hu et al., 2013;Ma et al., 2014;Wang and Christopher, 2003).Although the correlation between AOD and PM 2.5 mass concentration depends on many factors, such as aerosol size distribution, refractive index, single-scattering albedo and meteorological factors (Che et al., 2009;Guo et al., 2009bGuo et al., , 2017)), the predicted PM 2.5 mass from satellite AOD data compared well with ground-level measurements (Ma et al., 2014;Xie et al., 2015b).Moreover, the spatial distributions of AOD measured using sun photometers mostly agreed with those retrieved from satellite data (Che et al., 2014(Che et al., , 2015;;Liu et al., 2016b;Pan et al., 2010).Spatial distributions of annual average AOD in 2014 are shown in Fig. 11.The spatial distributions of PM 2.5 shown in Fig. 2 are similar to the patterns of AOD shown in Fig. 11.Differences in fine structures of their patterns were due to surface PM 2.5 versus column AOD comparison and spatial variations in PM 2.5 chemical composition.
In this section, geographical patterns of the aerosol optical properties including b sp and b ap measured on ground base in major Chinese cities are first discussed (Sect.3.1).Temporal patterns of b sp and b ap on annual and seasonal scales are then discussed for major regions (Sect.3.2).Fewer studies were available for b ap than b sp ; however, the measured BC concentrations (at 880 nm wavelength) can be converted to b ap (at 532 nm wavelength) by a factor of 8.28 m 2 g −1 (Wu et al., 2009).
Annual average b ap ranged from 37 to 96 Mm −1 with higher values observed in Chengdu and Xi'an (likely due to popular biomass burning besides large amount of coal burning) (Cao et al., 2012a;Tao et al., 2014a, b;Wang et al., 2017a;Zhang et al., 2014b), and lower values in Shouxian and rural Beijing (Fan et al., 2010;Yan et al., 2008;Zhao et al., 2011).b ap in Guangzhou was higher than that in Beijing and Shanghai despite their similar PM 2.5 EC levels, likely due to the different coating of EC in Guangzhou than in other cities.For example, the mass absorption of EC in Guangzhou was 8.5 m 2 g −1 (at 532 nm) in autumn 2004 (Andreae et al., 2008), which was higher than that (4.2 m 2 g −1 at 870 nm, equivalent to 7.2 m 2 g −1 at 532 nm) in winter 2013 in Beijing (Wu et al., 2016).

Temporal patterns
3.2.1The Beijing-Tianjin-Hebei region b sp measurements in BTH longer than 1 year were only available in Beijing, including the years of 2005, 2006, 2008-2009and 2009-2010(He et al., 2009;;Jing et al., 2015;Zhao et al., 2011).Annual b sp in Beijing increased by 36 % from 264 Mm −1 in 2005 to 360 Mm −1 in 2009-2010, while PM 2.5 increased by 20 % from 107 to 129 µg m −3 during the same period.However, annual b ap in 2009-2010 was 64 Mm −1 , which was slightly higher than 56 Mm −1 in 2005-2006, although the annual EC in 2009-2010 was evidently lower than that in 2005-2006.Meanwhile, annual secondary inorganic aerosols in 2009-2010 were evidently lower than that in [2005][2006].The coating by secondary inorganic aerosols likely enhanced the absorption of EC (Bond et al., 2006;Cheng et al., 2009;Yu et al., 2010).et al., 2008;Zhao et al., 2011).Considering all of the abovementioned data together, we can conclude that b sp and b ap showed slightly increasing tendencies in urban and rural Beijing in recent years.
Seasonal variations in b sp and b ap at urban and rural sites in Beijing are plotted in Fig. 13.The highest seasonal average b sp in Beijing was observed in winter and the lowest in spring with seasonal variations up to a factor of 1.7 (Bergin et al., 2001;Garland et al., 2009;Han et al., 2014;He et al., 2009;Jing et al., 2015;Li et al., 2013b;Liu et al., 2009;Tao et al., 2015a;Tian et al., 2015;Zhao et al., 2011).A different seasonal pattern was seen at the rural site located north of Beijing, which showed 10-26 % higher values in summer than in the other seasons (Yan et al., 2008;Zhao et al., 2011).The highest seasonal b sp in winter in Beijing was consistent with the highest seasonal PM 2.5 mass.However, in rural Beijing the highest PM 2.5 mass was observed in spring due to the frequent dust storm events, and the second highest seasonal average PM 2.5 mass in summer, which corresponded to the highest seasonal b sp .This is because scattering efficiency of dust aerosols was lower than that of anthropogenic aerosols (Zhao et al., 2011).
The highest seasonal b ap in Beijing appeared in autumn and the lowest in spring with seasonal variations up to a factor of 2.0 (Bergin et al., 2001;Garland et al., 2009;He et al., 2009;Jing et al., 2015;Li et al., 2013c;Liu et al., 2009;Tian et al., 2015;Wu et al., 2016).Seasonal variations in b ap were different from those of b sp due to their dependence on different chemical compounds, i.e., b sp mainly on PM mass while b ap mainly on EC mass in PM and its coating.In rural Beijing b ap was lower by 19-57 % in summer than in other seasons, and with similar seasonal variations to b sp , suggesting aerosols in rural Beijing mainly came from regional transport (Yan et al., 2008).
At the rural site in Tianjin (Wuqing) located between Beijing and Tianjin, only spring and summer 2009 and winter 2010 data were available, which gave a seasonal average of 280 Mm −1 in spring, 379 Mm −1 in summer and 485 Mm −1 in winter for b sp , and 47 Mm −1 in spring and 43 Mm −1 in summer for b ap (Fig. 13) (J.Chen et al., 2014b;Ma et al., 2011).These seasonal values in Wuqing were higher than those observed at the rural sites near Beijing, likely because Wuqing is close to and downwind of Tianjin and Hebei Province, where major pollutant sources are located.

Other cities
In Chengdu of southwest China, the highest b sp appeared in winter and the lowest in summer with seasonal variations up to a factor of 1.9, which was consistent with the seasonal pattern of PM 2.5 (Tao et al., 2014a, b).However, the highest b ap appeared in spring despite the highest EC in winter (Tao et al., 2014b).One explanation could be the large amount of OC emitted from biomass burning in spring, which enhanced the absorption of EC (Schnaiter et al., 2005;Tao et al., 2013b).b sp and b ap in winter were evidently higher than those in summer in Xi'an in northwest China, consistent with the seasonal patterns of PM 2.5 and EC, respectively (Cao et al., 2009(Cao et al., , 2012a;;Wang et al., 2015c).Seasonal measurements of b sp and b ap were also made at remote sites (Dunhuang, Yulin and Zhangye of Gansu Province, Dongsheng of Inner Mongolia) focusing on dust aerosols and only covered spring and winter (Li et al., 2010;Xu et al., 2004;Yan, 2007).b sp in winter ranged from 303 to 304 Mm −1 , which is double that in spring (126 to 183 Mm −1 ).
4 Relationships between aerosol optical properties and PM 2.5 mass concentrations 4.1 Mass scattering efficiency of PM 2.5 b sp and PM 2.5 mass concentration have been found to correlate well in numerous field studies (Andreae et al., 2008;Han et al., 2015;Hand and Malm, 2007b;Jung et al., 2009;Pu et al., 2015;Tao et al., 2014bTao et al., , c, 2015aTao et al., , 2016a;;Tian et al., 2015;Wang et al., 2012b;Zhao et al., 2011).A parameter describing their relationship is defined as mass scattering efficiency (MSE), which is the slope of the linear regression of b sp against PM 2.5 mass.MSE was found to vary with location and season due to the variations in PM 2.5 chemical composition.Some of the variations may due to different sampling conditions, e.g., ambient (controlled RH < 60 %) versus dry condition (controlled RH < 40 %), online versus filter-based PM 2.5 sampling.Available MSE data are discussed here, although uncertainties from measurements will not be addressed in this study.
In BTH, annual average PM 2.5 MSE was higher in Beijing (5.9 m 2 g −1 ) than in rural Beijing (4.8 m 2 g −1 ) based on online PM 2.5 mass (Zhao et al., 2011).In urban Beijing in winter of 2013, PM 2.5 MSE increased to 4.9 m 2 g −1 during a heavily pollution episode and decreased to 3.6 m 2 g −1 during clean days, due to a large fraction of soluble inorganic components (e.g., (NH 4 ) 2 SO 4 and NH 4 NO 3 ) in PM 2.5 under heavy polluted conditions (Tao et al., 2015a).In rural Beijing in 2005-2010, dust episodes had lower PM 2.5 MSE (0.7 m 2 g −1 ) and anthropogenic pollution episodes had higher PM 2.5 MSE (4.3 m 2 g −1 ) (Pu et al., 2015).
In YRD, annual average PM 2.5 MSE ranged from 3.8 m 2 g −1 in Ningbo to 5.3 m 2 g −1 in Hangzhou with a regional urban average (including the cities of Nanjing, Shanghai, Suzhou, Hangzhou and Ningbo) of 4.1 m 2 g −1 in 2011-2012 (Cheng et al., 2013b).PM 2.5 MSE in Lin'an (4.0 m 2 g −1 ), a rural site of YRD, was close to the regional urban average value in YRD (Xu et al., 2002).PM 2.5 MSE in Shanghai reached 5.6 m 2 g −1 in winter of 2012 (Han et al., 2015), which was higher than that in Beijing in the same season (Tao et al., 2015a).
In PRD, annual average PM 2.5 MSE in Guangzhou was 3.5 m 2 g −1 with seasonal average ranging from 2.3 m 2 g −1 in summer to 4.5 m 2 g −1 in autumn in 2009-2010 (Tao et al., 2014c).These values were close to 4.2 m 2 g −1 (Andreae et al., 2008) and 2.7 m 2 g −1 (Jung et al., 2009) measured in the same city in autumn of 2004.However, PM 2.5 MSE in rural Guangzhou (Wanqingsha, south of Guangzhou) was 5.3 m 2 g −1 (Wang et al., 2012), which was evidently higher than that in Guangzhou in the same season (Tao et al., 2014c).
In southwest China, seasonal average PM 2.5 MSE ranged from 3.5 to 4.4 m 2 g −1 in Chengdu in 2011 (Tao et al., 2014b).In northwest China, PM 2.5 MSE was 3.0 m 2 g −1 for anthropogenic pollution and 1.0 m 2 g −1 for dust pollution at a remote site (Yulin, located at the interface of the desert and loess regions, Shanxi Province), which was similar to rural Beijing (Xu et al., 2004).
In summary, annual PM 2.5 MSE typically ranged from 3.5 to 5.9 m 2 g −1 in urban areas in China with higher values in north China and lower values in south China.Seasonal average PM 2.5 MSE typically ranged from 2.3 to 5.6 m 2 g −1 with higher values in winter and autumn and lower values in spring and summer.Generally, PM 2.5 MSE typically ranged from 3.0 to 5.0 m 2 g −1 for anthropogenic pollution and from 0.7 to 1.0 m 2 g −1 for natural dust aerosols.

Mass absorption efficiency of EC and organic matter
EC is the dominant absorption species in PM 2.5 .Similar to PM 2.5 MSE, the slope between b ap and EC mass was defined as mass absorption efficiency (MAE) of EC.
Various instruments have been used to measure b ap including an Aethalometer, multi-angle absorption photometer (MAAP), Radiance Research particle soot absorption photometer (PSAP) and photoacoustic spectrometer (PAS), with the first two instruments measuring attenuation of the sample on the filter for estimating BC mass concentration, and the last two measuring b ap directly.Most studies in China used an Aethalometer and MAAP.BC mass concentrations (880 nm) were converted to b ap (532 nm) using an empirical constant of 8.28 m 2 g −1 , which was obtained by the regression between BC mass and b ap synchronously measured in autumn in Guangzhou, keeping in mind that application of an empirical constant obtained from one specific study to other cases may cause large uncertainties (Wu et al., 2009).EC MAE was 7.5-8.5 m 2 g −1 in winter and 9.4 m 2 g −1 in summer in Beijing (632 nm) (Cheng et al., 2011;Wu et al., 2016).The higher EC MAE in summer was likely due to more coating of EC in higher ambient humidity (Wu et al., 2016).BC MAE was 6.5 m 2 g −1 at 532 nm in autumn in Shenzhen of PRD (Lan et al., 2013).However, BC MAE was 12.4 m 2 g −1 at 532 nm in winter in Xi'an (Wang et al., 2014a).Moreover, EC MAE of diesel was 8.4 m 2 g −1 (632 nm), which was higher than those (3.0-6.8 m 2 g −1 ) of biomass burning sources (e.g., crop residual and wood) (Cheng et al., 2011).
Organic matter or brown carbon is also a strong light absorbing material at short wavelengths.Available MAE values of OC include 0.76 m 2 g −1 (532 nm) in autumn in 2008 in Guangzhou (Andreae et al., 2008).Moreover, available MAE values of water-soluble organic carbon (WSOC) include 1.79 and 0.71 m 2 g −1 (365 nm) in winter and summer, respectively, in Beijing (Cheng et al., 2011).The WSOC MAEs of wood, grass, corn and diesel tractors were 0.97, 0.90, 1.05 and 1.33 m 2 g −1 (365 nm), respectively, which were much higher than that of gasoline motorcycles (0.20 m 2 g −1 , 365 nm) (Du et al., 2014b).Evidently, the MAEs of OC or WSOC should not be neglected for short wavelength absorption.

Aerosol hygroscopic properties
b sp under ambient condition can differ significantly from dry conditions due to hygroscopic properties of soluble aerosol chemical components.A relationship between ambient and dry b sp is thus developed for estimating ambient b sp from measured dry b sp , which is often described by the hygroscopic growth curve (f (RH)) as a function of RH: Here, a and b are empirical fitting parameters.Only a few studies conducted in Beijing, Wuqing, Lin'an and Guangzhou provided the aerosol hygroscopic curves (Table S4 in the Supplement).Three different methods have been used to obtain f (RH).The first one measures simultaneously dry and wet b sp using a nephelometer and visibility meter, respectively.The second one measures wet b sp by integrating a nephelometer equipped with a humidifier, and the third one estimates dry and wet b sp based on Mie theory with size-resolved chemical components.
Available f (RH) curves in China are summarized in Fig. 14.The three f (RH) curves in autumn of 2007, 2011 and 2014 in urban Beijing were all measured using the first method (Fig. 14a) (Liu et al., 2013b, c;Yang et al., 2015).
The two f (RH) curves measured in 2011 and 2014 were quite close, but the one in 2007 was lower under RH < 80 % and higher under RH > 80 %, likely due to aerosol chemical composition and size distribution changes in these years.
The f (RH) curves at four rural sites were measured using the second method, including Baodi of Tianjin in spring, Wuqing of Tianjin in winter (Fig. 14b) (J.Chen et al., 2014b;Pan et al., 2009), Raoyang of Hebei Province in summer (Wu et al., 2017) and Lin'an of Zhejiang Province in spring (Fig. 14c) (Zhang et al., 2015b).It is known that the hygroscopic chemical components are mostly water-soluble inorganic salts (e.g., (NH 4 ) 2 SO 4 , NH 4 NO 3 ), while mineral dust and organic matter are mostly hydrophobic.In Baodi in the spring season, the concentrations of (NH 4 ) 2 SO 4 and NH 4 NO 3 and their mass fractions in fine-mode particles were higher during polluted episodes than during clean periods or dust storm episodes, resulting in higher f (RH) values during the polluted episode.f (RH) values measured in winter in Wuqing were evidently higher than those measured in spring in Baodi under RH < 80 %, likely due to more hygroscopic chemical components in winter in Wuqing.In Raoyang, a different fitting curve of f (RH) was obtained with a much higher f (RH = 80 %) value (2.3) than in other rural sites in BTH mentioned above, likely due to higher frac- tions of hydrophilic components in PM 2.5 (> 56 %).In all the BTH sites, f (RH) value increased continuously with increasing RH.However, in a different study an abrupt increase in f (RH) at RH values of 73-81 % was observed in summer in Wangdu of Hebei Province due to the deliquescence of ammonium sulfate (Kuang et al., 2016).Similar to what was found in Baodi, f (RH) values during polluted episodes were also higher than those during dust episodes in Lin'an, but the differences between polluted and dust periods were smaller in Lin'an than in Baodi.Noticeably, the f (RH) values during polluted episodes were similar in Lin'an and Baodi, e.g., f (RH = 80 %) was 1.5 and 1.6, respectively.
The f (RH) curves (solid lines) in summer in urban Guangzhou were measured by the first method, while those (dotted lines) in autumn in rural Guangzhou and in summer and autumn seasons in urban Guangzhou were measured by the third method (Fig. 14d) (Cheng et al., 2008b;Lin et al., 2014;Liu et al., 2008b).f (RH = 80 %) values were 2.04 and 2.68, respectively, for urban aerosols originating from the north and marine aerosols originating from the South China Sea.f (RH < 80 %) curves were similar in urban and rural Guangzhou; however, f (80 % < RH < 90 %) values in rural Guangzhou were evidently higher than those in urban Guangzhou, likely due to the much higher fraction of secondary inorganic aerosols in fine-mode particles in rural Guangzhou than urban Guangzhou in the dry season (Lin et al., 2014;Liu et al., 2008a).
If averaging all available f (RH) curves shown in Fig. 15, the empirical fitting parameters a and b were found to be 2.87 ± 0.03, 5.50 ± 0.06, respectively (Fig. 15a).But if excluding dust episodes in Baodi and Lin'an (Fig. 15b), the empirical fitting parameters a and b were 3.17 ± 0.03, 5.54 ± 0.06, respectively (Fig. 15b).Based on the average f (RH) curve, b sp under ambient condition (RH = 80 %) can be amplified by about 1.8 times that under dry conditions in China.This suggests that reducing inorganic water-soluble salts is critical in alleviating hazy weather in China.

Source apportionment of haze in China
To investigate the contributions of PM 2.5 chemical components to b sp , a revised formula developed by the original IM-PROVE method is applied in this section (Pitchford et al., 2007).The revised IMPROVE formula can be simplified as follows: (1) where X represents (NH 4 ) 2 SO 4 , NH 4 NO 3 and OM.RH growth curves of f S and f L for (NH 4 ) 2 SO 4 and NH 4 NO 3 can be found in Pitchford et al. (2007).
Using the chemical composition data shown in Fig. 8 and annual average RH values in major cities in China as input (http://data.cma.cn/), the estimated annual b ext and its load percentages under dry and ambient conditions are plotted in Fig. 16.For b ext under dry conditions, carbonaceous aerosols had similar percentage contributions to secondary inorganic aerosols in Shijiazhuang, Tianjin, Shangdianzi, Shanghai, Hok Tsui, Zhengzhou, Xi'an, Jinan, Chengdu, Fuzhou and Xiamen, but the percentage contributions were 11-65 % higher in other urban and rural sites.However, under ambient conditions the contributions of secondary inorganic aerosols were evidently higher (by 2-54 %) than those of carbonaceous aerosols in most cities except in Beijing, Chengde, Lanzhou and Chongqing.Noticeably, the contributions of secondary inorganic aerosols for b ext sharply increased by about 18-25 % under ambient conditions than dry conditions in humid (RH > 70 %) cities (e.g., Haikou, Changsha, Xiamen, Nanjing, cities in PRD, and Chengdu).

Implications for aerosol pollution controls
There is no doubt that reduction of PM 2.5 will be the ultimate approach for improving visibility and alleviating hazy weather.Industrial emission contributions to secondary inorganic aerosols were the dominant sources of PM 2.5 in urban areas in China (Liang et al., 2016).Aerosols produced from traffic emissions, biomass burning and soil dust were also important sources in north China.Secondary inorganic aerosols were formed from atmospheric reactions involving SO 2 and NO x , which were mainly emitted from coal combustion, the major energy source in China for decades.
A series of regulations controlling coal combustion emissions have been made since the first version of the NAAQS was promulgated in 1982.The Air Pollution Prevention law of the People's Republic of China was promulgated in 1987, which was the milestone in air pollution prevention history in China.It also marked the beginning of a new era for preventing air pollution based on the national law, followed by a series of regulations for controlling coal combustion.During 1990-2000, most of the control measures or technologies (e.g., desulfurization and dedusting for coal combustion) were focused on reducing SO 2 emissions.The measure for gross control of SO 2 emissions had been enforced since 1996.Despite these efforts, the amount of SO 2 emissions increased by about 28 % from 2000 to 2005 (http: //www.zhb.gov.cn/).The amount of SO 2 emission began to decrease in 2006 and gradually reduced to the emission level of 2000 in 2010 (http://www.zhb.gov.cn/).Meanwhile, am-bient annual SO 2 concentration in urban cities in China also decreased from 57 µg m −3 in 2005 to 40 µg m −3 in 2010 (http://www.zhb.gov.cn/).Apparently, the emission control efforts for reducing SO 2 emissions since 2006 have been effective.
The control measures for NO 2 only began with the control of vehicular emissions in 1995, but the inclusion of NO 2 in the gross control indexes did not happen until 2010.New coal power plants were also required to denitrate after 2010.The emissions of NO x actually increased from 1996 to 2010, as seen in the vertical column NO 2 derived from satellite data (Zhang et al., 2012b).Although annual average ambient NO 2 at the surface fluctuated from 30 to 40 µg m −3 during 2000-2010 in China (http://www.zhb.gov.cn/),annual average ambient NO 2 in megacities (e.g., Beijing, Shanghai and Guangzhou) slowly increased.Evidently, the control of emissions of nitrate gaseous precursors was not very effective during 2000-2010.
Despite the above-mentioned control measures, sulfate concentrations remained high and nitrate concentrations even gradually increased in megacities in China.More recently, the Clean Air Action Plan (CAAP) for improving air quality was promulgated and implemented by the State Council of the People's Republic of China in 2013 (http://www.gov.cn).This plan aims to reduce the PM 2.5 annual mass concentrations by 25, 20 and 15 % of the 2012 levels in BTH, YRD and PRD, respectively.Key industries, including power plants, iron and steel smelting, petroleum chemical, cement, nonferrous metals smelting and chemical production, were required to implement stricter emission standards in key regions including most megacities in China (http://www.zhb.gov.cn).Accordingly, annual average PM 2.5 in China from online monitored data at 74 cities gradually decreased from 72 µg m −3 in 2013 to 50 µg m −3 in 2015, showing some promising results from the series of control measures.
One factor that needs to be considered in future pollution reduction is the nonlinearity of chemistry (Cheng et al., 2016).For example, a model sensitivity study suggested a potential increase in NO − 3 mass concentrations due to the increased atmospheric oxidizing capacity, even with decreasing NO x emissions (Zhao et al., 2013a).Furthermore, increased atmospheric oxidizing capacity may also enhance the conversion of volatile organic compounds to OM.In fact, the contribution of secondary organic aerosols to PM 2.5 was also high and could increase further in typical megacities in China (He et al., 2011;Huang et al., 2014b;Sun et al., 2013).Another factor that requires more attention is ammonia emissions from agricultural activities in rural areas and human activities in cities. Ammonia emissions can enhance PM 2.5 pollution substantially, especially in ammonia-limited (acid aerosols) areas (Wang et al., 2011).This topic needs further investigation through both modeling simulation and field observations.
To improve the air quality across China, the following recommendations are provided based on the major chemi-cal components contributing to PM 2.5 and their impact on aerosol optical properties.Emissions produced from coal combustion, in both the industrial sectors and in residential areas, need to be further reduced.While advanced pollution control technologies should be adopted in the medium term in major industrial sectors consuming coal, cleaner energy sources should be considered a long-term goal (Cao et al., 2016).Providing cleaner energy to the vast rural and urban areas in north China for heating and cooking can reduce not only coal combustion emissions but also biomass burning emissions.Efficient use of fertilizers in agriculture is needed to reduce nitrogen emissions especially ammonia (Behera et al., 2013).Educating the public to reduce meat consumption in their daily lives, especially in the more affluent developed regions, can reduce the nitrogen footprint substantially and thus nitrogen emissions (Galloway et al., 2014), in addition to the potential benefits to human health.Traffic emissions in megacities may also need to be constrained, such as developing more efficient public transportation systems and limiting the use of personal automobiles.Having more vegetation coverage is especially important in arid or semi-arid areas as well as urban areas in reducing dust emissions (Baldauf, 2017), aside from the biological benefits.The continued expansion of the three northern region shelter forests in north China can potentially reduce dust emissions by increasing the dry deposition removal of aerosols (Zhang et al., 2017a).
Data availability.Data used in this study are from literature surveys and are available in the Supplement of this paper.

Figure 1 .
Figure 1.Geographical regions and location of cities with measurements.

Figure 2 .
Figure 2. Annual PM 2.5 mass concentration in various Chinese cities with filter-based measurements: (a) categorized into regions and (b) lined with latitude.

Figure 3 .
Figure 3. Interannual variations in PM 2.5 and dominant chemical components in Beijing (a), Shanghai (b) and Guangzhou (c).

Figure 4 .
Figure 4. Seasonal PM 2.5 and dominant chemical components in BTH.
1 µg m −3 in 2011-2014.EC varied in the range of 4.1 to 6.5 µg m −3 during 1999-2009, and also sharply decreased to 1.6-2.1 µg m −3 in 2011-2014.TC decreased from 19.5 µg m −3 during 1999-2009 to 11.9 µg m −3 in 2011-2014, or nearly 40 % reduction, which was much higher than the known measurement uncertainties.Noticeable reduction of OC and EC occurred after the 2010 Shanghai World Expo, which resulted in a decrease in TC after 2010.BC emission slightly decreased in Shanghai in 2005-2009, but increased in the adjacent Zhejiang and Jiangsu provinces (Qin and Xie, 2012).BC emissions in Jiangsu Province were much higher than the sum of those in Shanghai and Zhejiang.Thus, the decreased EC concentration in Shanghai was largely a result of local emission reductions.Annual SO 2− 4 concentration decreased from 14.0 µg m −3 in 1999-2000 to the range of 11.7 µg m −3 to 12.5 µg m −3 during 2009-2014.The trend of SO 2 emission in YRD generally agreed with that across the entire China, which showed an increasing trend during 2000-2006 and a decreasing one during 2006-2010

Figure 5 .
Figure 5. Seasonal PM 2.5 and dominant chemical components in YRD.

Figure 6 .
Figure 6.Seasonal PM 2.5 and dominant chemical components in PRD.

Figure 7 .
Figure 7. Seasonal PM 2.5 and dominant chemical components in other cities.

Figure 8 .
Figure 8. Annual PM 2.5 and dominant chemical components in China.

Figure 9 .
Figure 9. Interannual variations in PM 2.5 and the dominant six sources in Beijing.

Figure 10 .
Figure 10.Annual contributions of PM 2.5 dominant sources across China.

Figure 11 .
Figure 11.Spatial distribution of annual average aerosol optical depth (AOD) across China in 2014.

Figure 12 .
Figure 12.Annual b sp and b ap in China.

Figure 14 .
Figure 14.The hygroscopic growth curves in different sites in China.

Figure 15 .
Figure 15.Distribution of the hygroscopic growth curves in China.

Figure 16 .
Figure 16.Annual b ext percentage loading under dry and ambient conditions at urban sites in China.