Atmospheric black carbon (BC) particles are mostly in the submicron range because of their formation mechanisms (Bond et al., 2013). The average concentrations of main chemical components in PM1 (non-refractory species measured by AMS and BC by Aethalometer or MAAP) and gaseous pollutants, and meteorological conditions are listed in Table 1. The time series of meteorological parameters, including temperature, relative humidity (RH), atmospheric pressure, and wind speed and direction, can be found in Figs. S6–S9.

The average PM1 concentrations showed little difference in spring, summer, and autumn, which were 45.1 ± 45.8, 37.5 ± 31.0, and 41.3 ± 42.7 µg m-3, respectively. The average PM1 concentration of 81.7 ± 72.4 µg m-3 in winter was the highest among the four seasons. OA was the most important component (31–44 %) in PM1, similar to previous studies (Fig. 1). In autumn, the contribution of the carbonaceous components (OA+BC) to PM1 was over 50 %. In four seasons, SNA accounted for about 40–60 % in PM1. Secondary inorganics (sulfate and nitrate) correlated well with RH and/or LWC (Table S14), indicating that aqueous-phase reactions in aerosols played an important role in secondary inorganic formation in Beijing. The contribution of photochemical processes to secondary inorganics was likely less than that of aqueous-phase reactions according to the weaker correlations between secondary inorganics and odd oxygen (Ox= O3+ NO2). In particular, nitrate did not correlate with Ox in summer (Table S14). Lower temperatures in winter, spring, and autumn favored the partitioning to particulate nitrate and were partially related to higher concentrations of nitrate in PM1 (Table 1).

Compared with the results obtained over the world, PM1 and the concentrations of major chemical species in Beijing, especially in winter, were 2–10 times higher than those at US and European sites (Hu et al., 2016a). In China, PM1 in urban Beijing in winter was much higher than those results (27–48 µg m-3) obtained in the Yangtze River Delta (YRD) and Pearl River Delta (PRD) regions (Gong et al., 2012; He et al., 2011; Huang et al., 2011, 2012, 2013). However, PM1 in other seasons was approximate to those obtained at the sites at Changdao Island, and in the YRD and PRD regions in the same seasons (Hu et al., 2013, 2016a, and references therein). These results suggest that Beijing suffered from severe particulate pollution, especially in winter, which should be seriously taken into consideration.

Aerosol pollution in Beijing exhibited distinctive characteristics in four seasons, because of the significantly different meteorological conditions and emission sources of pollutants. Compared with the previous results in Beijing (Fig. 1), PM1 in summer was lower than before, which likely resulted from the more effective washout (Fig. S7) and lower concentrations of gaseous precursors (Table 1 and Table 1 in Sun et al., 2015). In summer, high temperature, strong solar radiation, and high oxidant concentrations generally enhance the secondary aerosol formation from gaseous precursors (such as SO2, NOx, and VOCs). As illustrated in Fig. 2, PM1 in summer was mainly in the range of less than 50 µg m-3 (∼ 80 %), and skew-normally distributed, with the highest frequency (∼ 45 %) in 15–35 µg m-3. The probability distribution of PM1 was similar to the previous results in Beijing in summer (Huang et al., 2010; Sun et al., 2010), but the range of PM1 was narrower. In contrast, PM1 in winter was higher than the previous results in Beijing before 2013, and equivalent to those in close periods from January to February 2013 (Fig. 1; Zhang et al., 2014; Sun et al., 2014). The heaviest particulate pollution in winter was the co-effect of the large amount of particles emitted from primary sources, prolonged control of weak weather system (i.e., low mixed layer and static air), and the rapid generation and accumulation of secondary particles from gaseous precursors (Hu et al., 2016a; Y. Sun et al., 2013).

In spring, autumn, and winter, the passage of strong cold air parcels transported from Siberia and Mongolia through Beijing (Sect. S11) was usually accompanied by strong winds, thus atmospheric relative humidity decreased, pressure increased, which was conductive to the dispersal of pollutants (Figs. S6, S8, and S9). The heavy pollution processes were usually ended by the passages, demonstrating periodic cycles of aerosol pollution (Guo et al., 2014; Wu et al., 2009). In spring and autumn, the biomass burning emissions in North China occurred often due to agricultural activities and traditional activities around Tomb-Sweeping Day (early April). Dense fire points were frequently observed by satellites (https://firms.modaps.eosdis.nasa.gov/firemap) in Beijing and surrounding areas (Figs. S10–S11). For instance, during 7–8 and 26–28 April, 18–20 and 23–27 October, and 6–11 November 2012, the concentrations of OA and other pollutants increased substantially (Figs. S6 and S8). Higher nitrate concentrations in spring and autumn, likely caused by the secondary conversion of gaseous nitrogen oxides emitted from biomass burning (Fabian et al., 2005). The probability distributions of PM1 in spring, autumn and winter were quite different from that in summer (Fig. 2), because the scavenging effect of strong wind was apparent due to the intrusion of long-range-transported air parcels. PM1 mainly concentrated in the range of low levels (< 20 µg m-3), with frequencies of 40–60 %. However, PM1 ranged much more broadly, with the highest concentrations of over 200 or 300 µg m-3, resulting from accumulated pollutants under strong primary emissions coupled with extremely unfavorable meteorological conditions, e.g., long-lasting stagnant weather, high humidity, and temperature inversion.

The contributions of chemical components in atmospheric aerosols at different concentration levels can help better understand the origins of chemical components. The proportions of chemical compositions as a function of PM1 are shown in Fig. 2.

In all seasons, at low PM1, the contribution of carbonaceous components (OA+BC) was dominant, accounting for 50–80 % of PM1. With the increase in PM1, the proportions of SNA in PM1 increased gradually, indicating that the enhancement of SNA primarily contributed to the increase in PM1, consistent with previous studies (Huang et al., 2010; Sun et al., 2010, 2012). The proportions of nitrate increased significantly and the nitrate concentration increased rapidly under higher RH (Figs. 2 and S6–S9). Nitrate showed good correlations with RH (Pearson correlation coefficients r=0.34–0.79, p<0.01) and the LWC in aerosols (r=0.82–0.87, p<0.01). These results indicate that the aqueous-phase reactions in wet aerosols and/or clouds could substantially contribute to nitrate formation. It is worth noting that when PM1 higher than 150 µg m-3, carbonaceous aerosols dominated PM1 again in autumn, mainly resulting from large amounts of primary emissions. The proportions of chemical components in PM1 varied less significantly in winter, reflecting the characteristics of regional pollution over the North China under the control of static weather system.

Affected by different meteorological conditions, e.g., solar radiation, temperature, RH, boundary layer, and mountain-valley breezes (Fig. S7), as well as different emission sources, the chemical compositions in PM1 showed distinct diurnal patterns in four seasons. The diurnal patterns of gaseous and particulate pollutants during the seasonal observations are shown in Fig. 3.

Diurnal patterns of OA were similar in spring, summer, and autumn. OA showed two obvious peaks around noon and in the evening. Similar peaks were also observed in winter, but these were not as strong as in the other seasons. These two peaks corresponding to meal time were mainly caused by cooking emissions, which will be discussed later in Sect. 3.2.1. The peak concentration of OA in the evening in autumn was about 2 times higher than in spring and summer, consistent with the results in October–November 2011 (Sun et al., 2015), possibly because of the more intensive cooking activities. In autumn, charcoal grilling, or barbecuing, has become one of the most popular outdoor recreational activities in urban Beijing and surrounding areas, due to more moderate and pleasant weather conditions than in other seasons (Table 1). In winter study, OA increased at night due to extra primary emissions, e.g., coal combustion and biomass burning.

Compared with OA diurnal patterns, flatter diurnal cycles of sulfate were observed in four seasons, identifying the regional characteristics of sulfate formation (Sun et al., 2015). In spring, sulfate showed small peaks around 13:00 and 20:00. In the daytime, both active photochemical production and more favorable dispersion conditions due to a higher planetary boundary layer (PBL), possibly caused such a diurnal pattern of sulfate. In summer, sulfate enhanced gradually from morning to evening and peaked in the evening only, indicating that the photochemical production of sulfate might be significant. Guo et al. (2010) reported that in addition to the gas-phase processes, the formation of sulfate was mainly attributed to aqueous-phase reactions in clouds and/or wet aerosols (70–80 %) in the summer of Beijing. The aqueous-phase formation of sulfate in summer should also be more intense than in spring and autumn due to high temperature and high humidity (Fig. 3k–l). In winter, sulfate showed two peaks in the morning (06:00–10:00) and evening (around 20:00), which was likely influenced by the secondary formation of primarily emitted SO2 (Fig. 3f). The lower sulfate concentration in the daytime was quite related to the dilution effect of the planetary boundary layer (PBL). The PBL height decreases at night, and the atmospheric stability increases, thus the pollutants are difficult to spread (Hu et al., 2016a).

Nitrate showed distinct diurnal patterns in four seasons. Heterogeneous/aqueous-phase reactions and gas-to-particle condensation processes are the main pathways to form fine-mode nitrate (Guo et al., 2010). In summer, nitrate exhibited an obvious diurnal pattern with the concentrations gradually decreasing in the daytime, indicating that NH4NO3 evaporated due to high temperatures (Fig. 3k) and overcame the photochemical production of nitrate. The elevated PBL also plays a role in reducing the concentration of nitrate in the daytime (Sun et al., 2015). In winter, nitrate increased gradually during 09:00–20:00, which did not appear in other seasons, suggesting that in addition to photochemical processes in the daytime, the partitioning of NH4NO3 into particulate nitrate was more significant due to low temperatures (< 5 ∘C; Fig. 3k). Similar diurnal patterns in summer and winter have been observed in previous studies (Sun et al., 2015). In autumn, nitrate increased in the morning and then later in the evening, primarily driven by the photochemical production. The diurnal variation in nitrate in spring was insignificant. Because ammonium mainly existed in the forms of (NH4)2SO4 and NH4NO3, its diurnal patterns generally exhibited a combined effect of sulfate and nitrate formation mechanisms. In summer, the diurnal variation in ammonium correlated better (r=0.55, p<0.01) with that of nitrate than sulfate (r=0.19, p>0.05). In other seasons, the diurnal variations in ammonium showed good correlations with those of both sulfate and nitrate (r=0.50–0.98, p<0.01 or 0.05).

Chloride showed higher concentrations at night, with a peak in the morning, and then deceased in the daytime. Such a diurnal pattern was possibly caused by high primary emissions of chloride at night. It was also driven by the diurnal patterns of both temperatures and the PBL height, and controlled by the temperature-dependent gas–particle partitioning (Hu et al., 2016a; Sun et al., 2015). Chloride usually presents at semivolatile state in the form of ammonium chloride in urban areas (Zhang et al., 2005); therefore the diurnal cycles of chloride and temperature trend oppositely. In winter, chloride was higher at night because of coal combustion and biomass burning emissions. Hu et al. (2016a) estimated that in winter over 50 % of chloride in PM1 existed as NH4Cl, and part of chloride existed as KCl and NaCl. BC was mainly affected by the diurnal variation in the PBL height. BC showed a marked rise at night (around 21:00–00:00) and a small morning peak (07:00–08:00). Nocturnal heavy-duty vehicular exhausts (Lin et al., 2009; Hu et al., 2012), coal combustion, and biomass burning emissions in the urban atmosphere also contributed to the high concentration of BC at night. The morning peak of BC was consistent with that of NOx, probably due to vehicular emissions during rush hour.

The mass size distributions of chemical compositions in PM1 during seasonal campaigns are shown in Fig. 4. The size distributions of SNA concentrated in accumulation mode, peaking at about 600–800 nm (dva), indicating the internally mixed states of submicron aerosols. The mode diameters of SNA in Beijing were higher than those (500–600 nm) in Hong Kong in four seasons (Li et al., 2015). The mode size (700 nm) of sulfate was smaller than that (790 nm) of nitrate in autumn. In contrast, the mode size of sulfate (760 nm) was larger than that (700 nm) of nitrate in winter. The size distribution of ammonium was more consistent with that of nitrate than sulfate in winter. More nitrate likely existed in the form of NH4NO3, formed by the reaction of gaseous HNO3 and NH3 condensed on atmospheric particles (Weimer et al., 2006).

During the winter, the peak size of OA was close to those of SNA, while in other seasons it was smaller than those of SNA, indicating a more aged state of OA in winter. The mass size distributions of chemical components in winter were also similar to those in the heavy-pollution episodes in other seasons, consistent with the severity of aerosol pollution in winter. In spring, summer, and autumn, the size distributions of OA were wider than those of SNA. The OA concentrations were much higher than those of SNA in the range of small size (100–500 nm), similar to the results in other urban areas (Weimer et al., 2006; Aiken et al., 2009), probably caused by the contribution of POA (Huang et al., 2010; Hu et al., 2013). In particular, influenced by primary emissions, the concentration of OA at peak size was much higher than those of other species in autumn. During the spring, summer and autumn, the proportions of OA decreased, while those of SNA increased rapidly with the increase in particle size (> 200 nm). SNA contributed predominantly to the enhancement of PM1, in accordance with that mentioned in Sect. 3.1.2. Different OA sources played different roles with the increase in particle size in winter, and the proportion of OA (about 40 %) varied slightly. However, the proportions of sulfate rose with the increase in size, implying the significant contribution of regionally transported sulfate.

COA refers to OA emitted from cooking activities (Allan et al., 2010). The knowledge on the transformation of COA in the atmosphere is limited (Dall'Osto et al., 2015). The reductive alkyl fragment ions are abundant in the MS of COA (Figs. S28–S31). The abundance of fragments m/z 55 and m/z 57 (f55 and f57, mass fractions of m/z 55 and m/z 57) in the MS of COA were about 7–10 and 3–4 %, respectively. Higher f55 and f57 are the most remarkable characteristic in the MS of identified COA factors, which are crucial to identify COA and HOA (Mohr et al., 2012). In this study, f55 in COA factors was significantly higher than in corresponding HOA factors, and f57 in COA factors was comparable to f57 in corresponding HOA factors (Figs. S28–S31). The oxidation state of COA was low, with low O / C (0.13–0.23) and OA / OC (1.33–1.46) ratios in this study.

As the edible oil is rich in oleic and linoleic acids (Dyer et al., 2008), the fragments m/z 55 and m/z 57 are not only contributed by the reductive alkyl fragments C4H7+ and C4H9+ but also by oxygenated fragments C3H3O+ and C3H5O+. The mass fractions of the later ones were about one-third of those of the reductive alkyl fragments with the same m/z. The ratios of oxygenated and reductive fragments m/z 55 and m/z 57 are used to identify the sources of OA. The ratio of C3H3O+ to C3H5O+ in COA is about 2, while in HOA it is about 1; the ratio of C4H7+ and C4H9+ in COA is about 2.5 and that in HOA is approximately 1 (Mohr et al., 2012). In this study, the ratios of C3H3O+ to C3H5O+ in COA were 2.7, 3.4, 3.8, and 2.4, and in HOA were 4.1, 0.6, 0.0, and 1.3; the ratios of C4H7+ and C4H9+ in COA and HOA were 2.3, 1.8, 2.3, and 2.4, and 1.4, 1.6, 1.1, and 1.1, respectively.

The reasonability of resolved COA factors was investigated through their concentrations, diurnal patterns, correlations of COA with external tracers, and the uncentered correlations (UC) of MS. The COA concentrations could be simply estimated with the formula provided by Mohr et al. (2012) based on the fragments m/z 55, m/z 57 and m/z 44. The estimated COA concentrations were 4.7, 4.4, 6.5, and 5.3 µg m-3 in four seasons, accounting for 33, 35, 35 and 18 % of OA, respectively. Compared with the concentrations (2.6, 2.5, 5.2 and 4.3 µg m-3) and proportions (19, 20, 28, and 15 %) based on the AMS-PMF analysis (Fig. 5), the results were overestimated, especially in summer, which is likely related to the applicability of experience parameters in different regions and seasons. The COA factors showed apparent diurnal patterns during seasonal observations, which is a key feature in identifying COA (see Sect. 3.2.3).

COA correlated well with some gaseous and other particulate pollutants, e.g., NOx and BC (Tables S9–S12). Dall'Osto et al. (2015) considered that the COA factors at a rural site contained more sources other than cooking emissions, based on the good correlations between COA with HOA, BC, NOx, nitrate, etc. The UC coefficients (Ulbrich et al., 2009) between the MS of resolved COA factors in this study and the average MS of COA in previous studies (Lanz et al., 2008; Mohr et al., 2009; Huang et al., 2010; He et al., 2010) were calculated. The UC coefficients were in the range of 0.910–0.990 (Table S13), confirming the good similarity and rationality of the resolved COA factors.

In previous studies, it was reported that COA accounted for 14–25 % (∼ 20 % on average) of the total OA in summer and winter in Beijing, with an average concentration (∼ 6 µg m-3), which is a relatively stable component of OA (Hu et al., 2016a; Wang et al., 2009). In this study, the concentrations and proportions of COA in OA during seasonal observations were in the range of 2.5–5.2 µg m-3 and 15–28 %, respectively, comparable to previous results. The highest concentration and proportion of COA in autumn were likely caused by strong emissions from charcoal-grilling or barbecue activities, which are popular in Beijing and surrounding areas in autumn as mentioned above. Fewer cooking activities during and around the Chinese New Year holiday (7–19 February) and the lower evaporation rates of cooking oils (Ndiaye et al., 2005) led to the lower concentration and proportion of COA in winter (Fig. S35). Overall, COA is an important non-fossil POA sources in all seasons, which should be taken into consideration, especially in autumn.

The BBOA factors that were identified in spring and autumn accounted for 13 and 11 % of OA, respectively. In summer and winter, the contributions of biomass burning to OA might be relatively low (probably < 5 %) and cannot be resolved in our dataset by PMF. Levoglucosan is a significant tracer of biomass burning emissions, the fragment of which, C2H4O2+ (m/z 60) is regarded as a tracer of BBOA (Alfarra et al., 2007; Cubison et al., 2011). The abundance of m/z 60 (∼ 0.8–1.3 %) in the MS of BBOA was much higher than the background abundance (0.3 %) in the urban without biomass burning emissions, which is a primary feature in identifying BBOA factors. The O / C ratios (0.24–0.31) of BBOA factors were higher than those (0.07–0.18) of HOA, and comparable to those (0.2–0.4) in previous studies (Mohr et al., 2009; He et al., 2010, 2011). In spring, farmers start preparing fields and burn straw leftover on the field. In autumn, burning crop straw randomly happens quite often during the harvest season for corn and other crops in North China. These phenomena result in serious atmospheric particulate pollution in Beijing and surrounding areas (Duan et al., 2004; Zheng et al., 2005).

BBOA tracked well (Tables S9 and S11) with chloride, BC, C2H4O2+ and acetonitrile, similar to the results of previous studies (DeCarlo et al., 2010; He et al., 2011; Gong et al., 2012). Particles emitted from biomass burning could contain a high proportion of chlorides (Silva et al., 1999). The diurnal variations in BBOA were similar with those of HOA, with higher concentrations at night and in the morning and lower ones in the daytime (see Sect. 3.2.3). They could be partially attributed to similar emission processes, such as residential burning activities.

Coal accounts for two-thirds of the total primary energy consumption and coal combustion is an important source of air pollution in China (You and Xu, 2010; Huang et al., 2014). CCOA was only resolved in winter, consistent with the domestic heating period, suggesting that this CCOA factor was dominated by residential burning with higher OA emission factors (Hu et al., 2016a). Alkyl fragments (CnH2n+1+ and CnH2n-1+) are the most abundant in the MS of CCOA. The MS of CCOA also showed pronounced signals of ion fragments from polycyclic aromatic hydrocarbons (PAHs), e.g., C10H8+ (m/z 128) from naphthalene and C14H10+ (m/z 178) from anthracene (Fig. S31). The CCOA spectrum in winter is similar to those in our previous studies at Changdao Island and in the winter of Beijing (Hu et al., 2013, 2016a). Compared to the average H / C ratios (1.76–1.96) in other POA factors (Canagaratna et al., 2015), H / C in CCOA (1.45) is lower. The O / C and OA / OC ratios in CCOA were 0.14 and 1.32, respectively. CCOA accounted for 17 % of total OA on average, within the range of 10–33 % reported in other studies in the winter of Beijing (Fig. 5; Hu et al., 2016a).

HOA is the sum of unresolved reductive and primary OA except specific OA (e.g., COA, BBOA, and CCOA), which is generally considered to be related to the emissions of fossil fuel combustion. The mass spectra (MS) features of HOA in this study are similar to those in previous studies (Huang et al., 2010; Hu et al., 2013; and references therein). Alkyl fragments (CnH2n+1+ and CnH2n-1+) are abundant in the spectra (Figs. S28–S31). O / C and OA / OC ratios of HOA factors were 0.18, 0.19, 0.07, and 0.36, and 1.40, 1.41, 1.27, and 1.63 during four seasons, respectively. Compared with the corresponding ratios (O / C: 0.04–0.26; OA / OC: 1.21–1.50) of HOA in previous studies (Canagaratna et al., 2015), the higher values in winter were influenced by the contribution of BBOA. Absolutely independent sources cannot be identified by factor analysis (e.g., PMF model). Therefore, the typical MS features of OA from other primary sources can be found in the MS of HOA. In this study, the abundant fragment ions in the MS of POA factors, e.g., CCOA (m/z 67, 69, 91, etc.) and COA (m/z 55, 57, etc.), were also presented in the MS of HOA (He et al., 2011; Hu et al., 2013, 2016a). This can also be confirmed through the correlations between HOA and external tracers (see Tables S9–S12). HOA tracked well (r=0.6–0.9, p<0.01) with traces of primary emissions (e.g., coal combustion and vehicular exhausts), such as chloride, NOx, CO, BC, and C2H4O2+, consistent with the primary sources of HOA.

OOA, considered as a good alternative to SOA in most cases, has been extensively investigated in a large number of previous studies (Jimenez et al., 2009; Zhang et al., 2005; Ulbrich et al., 2009; Ng et al., 2011). During the seasonal observations, OOA was an important component of OA. In summer and winter, OOA was separated into MO-OOA and LO-OOA. Oxygenated fragments (CxHyOz+) are prominent in the MS of OOA. The abundance of CxHyOz+ in MO-OOA was higher than that in LO-OOA. For instance, the fragment of carboxylic acid CO2+ (m/z 44) accounted for 17–21 % in MO-OOA, and 7–14 % in LO-OOA. The average O / C ratios for SV-OOA, LV-OOA, and OOA reported in some studies around the world were 0.53, 0.84, and 0.67, respectively (Canagaratna et al., 2015). In spring and autumn, the O / C ratios of OOA were 1.0 and 0.88, respectively. In summer and winter, the O / C ratios in resolved MO-OOA factors were about 0.91 and 0.84, higher than those of corresponding LO-OOA (0.67 and 0.77). In all four seasons, O / C ratios of OOA were higher than the average values mentioned above, indicating that OA was much more oxygenated and secondary formation contributed significantly to OA in urban Beijing. Overall, OOA (MO-OOA+LO-OOA) correlated well (r=0.89–0.96, p<0.01) with secondary inorganics (sulfate and nitrate). MO-OOA, as an extremely aged secondary component, exhibited good correlations with SNA (r=0.78–0.95, p<0.01). However, the correlation coefficients (r=0.63–0.89, p<0.01) between LO-OOA and SNA were slightly lower (Tables S10 and S12). In summer, LO-OOA and nitrate trended well during most days; during the whole winter campaign, LO-OOA tracked well with nitrate. OOA had strong correlations with RH and/or LWC (Table S14), indicating that aqueous-phase reactions play a dominant role in OOA formation. The slope of OOA against Ox steepened with the increase in RH and LWC (Figs. S41–S42), also implying that the aqueous-phase oxidation was an important pathway of the OOA formation. The good correlations (r=0.53–0.73, p<0.01) between Ox and LO-OOA in summer, and between Ox and OOA in autumn and winter, suggested that photochemical processes also contributed substantially to SOA, especially LO-OOA, in these seasons (Xu et al., 2017). Note that gaseous pollutants and main chemical compositions in PM1 displayed good correlations with each other in winter, which may be caused by regional pollution characteristics under a weak weather system on heavy-pollution days.

The diurnal patterns of OA components during seasonal observations are shown in Fig. 6. As mentioned above, the COA factors showed obvious diurnal patterns during all four campaigns, with two peaks around noon (about 13:00) and in the evening (about 20:00), in accordance with living habits of residents. The concentrations of COA factors at the noon peak (about 12:00–14:00) were 3.1, 2.5, 4.0, and 4.4 µg m-3 on average, respectively. COA reached the highest concentrations in the evening (18:00–21:00), as 5.6, 6.2, 14.0, and 7.5 µg m-3 on average, respectively. HOA was lower in the daytime and higher at night. Nocturnal activities, such as heavy-duty diesel vehicles only permitted at night (Lin et al., 2009), biomass burning and coal combustion, could make more significant contributions to POA. In autumn, the OA concentrations began to raise in the evening, peaked at about 19:00, and kept higher at night (Fig. 2), reaching twice that in the daytime. As Fig. 6c shows, the anthropogenic emissions, e.g., biomass burning, coal combustion, and especially cooking emissions, contributed significantly to OA. In winter, CCOA showed clear diurnal variations with high concentrations at night, consistent with the residential heating periods. The high concentration (10 µg m-3) and fraction (∼ 30 %) of CCOA in total OA points to strong coal combustion emissions at night in Beijing.

The peaks of OOA (or LO-OOA) coincided with the peaks of primary emitted COA (spring, summer, and autumn) and HOA (winter) in diurnal patterns, probably because strong primary emissions favored SOA production. It has been found that primary emissions evaporate substantially upon dilution to ambient conditions and those vapors undergo photooxidation, which produces SOA efficiently (Robinson et al., 2007; Murphy and Pandis, 2009). Recently, Liu et al. (2017) reported that SOA formed from gas-phase emissions of heated cooking oils. SOA may also be formed after photochemical aging of the large amounts of POA emitted from cooking oils.

In summer, OOA showed obvious diurnal variations: MO-OOA peaked in the morning and afternoon; LO-OOA showed two pronounced peaks at noon and at night. This was likely the co-effect of SOA formation via gas-phase photochemical reactions in the daytime and aqueous chemistry in humid air at night (Ervens et al., 2011; Lim et al., 2010). The concentration and proportion of LO-OOA increased significantly in the afternoon (12:00–16:00), up to 7 µg m-3 and 50 %, respectively (Fig. 6b), suggesting that LO-OOA was a strong local/regional photochemical product despite the much higher PBL in the daytime (Hu et al., 2016a). Regardless of the air mass history (local, regional, and long-distance transported), LO-OOA dominated OOA, accounting for 29–48 % of OA (Fig. S43). In winter, despite the expanded PBL, MO-OOA increased gradually from noon, and peaked in the evening, implying that MO-OOA could form through photochemical oxidations in the daytime.

The probability distributions of OA and the fractions of OA components as a function of OA concentrations are illustrated in Fig. 7. In spring, autumn and winter, similar to the probability distributions of PM1, the OA concentrations were frequently (about 50–80 %) lower than 10 µg m-3, due to the scavenging effect of strong wind accompanying long-range-transported air parcels. In autumn and winter, primary emissions (e.g., cooking emission and coal combustion) influenced strongly, resulting in the wide ranges of OA concentrations, up to 120 and 140 µg m-3, respectively. In contrast, OA concentrations presented a skew-normal distribution, mainly in the range of 5–20 µg m-3 in summer, similar to previous results (Huang et al., 2010; Hu et al., 2016a).

During the spring observation (Fig. 7a), as the OA concentration was less than 10 µg m-3, OOA dominated OA (up to 55 %), which was likely associated with the new particle formation in clean atmosphere (Wu et al., 2007). When the OA concentration was higher than 10 µg m-3, OOA primarily contributed to the increase in OA. The fraction of OOA remained at about 50–60 % as the OA concentration was over 25 µg m-3. In summer, as the OA concentration was lower than 30 µg m-3, OOA accounted for about 60–70 % of OA (Fig. 7b). As mentioned above, LO-OOA formation was efficient due to both the strong photochemical and aqueous-phase processes under high-RH and high-temperature conditions (Xu et al., 2017). With the increase in OA, the fraction of MO-OOA enhanced gradually, implying that the aged MO-OOA contributed importantly to the OA increase. In autumn, with the increase in OA, the proportion of OOA decreased gradually (Fig. 7c). In winter, when the OA concentration was lower than 70 µg m-3, OOA and POA contributed equally to OA, while the fraction of LO-OOA increased gradually (Fig. 7d). In both autumn and winter, the fractions of OOA slightly increased around 100 µg m-3. More work is needed to accurately clarify the cause of the OOA increase within this range.

In spring, summer, and autumn, as the OA concentration was higher than 30 µg m-3, the fraction of COA in OA increased to a different extent, reaching 30–60 %. In autumn, the contributions of BBOA and HOA to OA were relatively stable, while that of COA dramatically enhanced with OA increase, probably because that the static air during some intervals (e.g., 23–25 October) was not conducive to the dispersal of heavy emissions from cooking activities (Figs. S8 and S34). Furthermore, in autumn, considerable barbecue activities in urban Beijing and surrounding areas could be an important source of COA as mentioned above. These results indicate that strong cooking emissions were partially responsible for high OA concentrations in these seasons. In winter, when the OA concentration reached 100 µg m-3, the fraction of CCOA dramatically increased, from 10 to about 40 %, indicating that the strong emissions from coal combustion contributed predominately to high OA concentrations (Hu et al., 2016a). The O / C ratios of OA at different OA concentrations were dependent on the source contributions, showing lower values when the POA were dominant in all seasons (Fig. 7).

Some important metrics or tools, such as elemental ratios, van Krevelen (VK) diagrams, and carbon oxidation state (OSC) of OA, are used to investigate dynamic evolution and oxidation mechanisms for bulk organic aerosols.

The elemental compositions of OA are closely related to their properties, e.g., density, hygroscopicity, and vapor pressure. The element ratios and OA / OC ratios in OA obtained from seasonal observations in Beijing and at other urban and rural/suburban sites are listed in Table 2. During the campaigns, the average O / C and H / C ratios (in atomic number), and OA / OC ratios were in the range of 0.47–0.53, 1.52–1.63, and 1.77–1.88, respectively. In spring, the average OA / OC ratio in Beijing was lower than those determined in Mexico City and Bologna and at Changdao Island (Table 2). In the summers of 2011–2012, the OA / OC ratios in Beijing were higher than those measured at urban sites in Shenzhen and Riverside, and rural/suburban sites in Jiaxing and Melpitz, indicating a high oxidation state of OA due to strong SOA formation via photochemical reactions in Beijing. However, in summers of 2008 in Beijing and 2010 in Shanghai, the OA / OC ratios were quite lower, maybe because the reduction of pollutant emissions reduced SOA formation during the Beijing Olympic Games and Shanghai World Expo, respectively (Pye et al., 2013). In autumn, the OA / OC ratio in Beijing was slightly higher than that in Shenzhen but far lower than those ratios obtained at rural/suburban sites, e.g., Kaiping, Heshan, and Melpitz. The observed OA / OC ratio in Beijing in winter 2012 was higher than those in Fresno and in Beijing 2010, but comparable to or lower than those at rural/suburban sites such as Ziyang, Jiaxing, and Melpitz. Overall, the OA / OC ratios in Beijing were higher among urban sites except for in spring, and lower among rural/suburban sites except for in summer.

The van Krevelen (VK) diagram is an important tool to investigate the evolution and functional group alteration of OA (Heald et al., 2010). As shown in Table 3 and Figs. S36–S38, H / C and O / C ratios of OA exhibit good negative correlation (coefficient of determination r2=0.70–0.79), with the slopes of -0.57, -0.62, and -0.67 and intercepts of 1.91, 1.94, and 1.90 in the VK diagram in spring, summer, and autumn, respectively. The slopes are flatter than those (-0.7 to -1.0) obtained across the world (Chen et al., 2015) but are steeper than that determined in Ziyang (Table 3). The intercepts are slightly lower than those (2.1–2.2) at urban and downwind sites (Chen et al., 2015). However, the VK diagram in winter (Figs. 8 and S39) does not show any correlation (r2=0.02) between H / C and O / C. The slope is nearly zero (-0.08). This is possibly caused by the more complex sources of OA in winter; for example, CCOA was only resolved in winter.

Heald et al. (2010) concluded that OA trends to evolve along with a slope of -1 in the VK diagram. Ng et al. (2011) derived that the evolution from SV-OOA to LV-OOA is mainly along with a slope of approximately -0.5, and is associated with the replacement of carboxyl functional group (OH-(C=O)-). The identified reactions related to the alteration of functional groups are shown in Fig. 8. In spring, summer, and autumn, the slopes fell between -1 (the addition of carboxyl functional groups without fragmentation or carbonyl and hydroxyl in different carbons) and -0.5 (carboxyl functionalization with fragmentation). In winter, the nearly zero slope of the VK diagram suggests that hydroxylation and/or peroxidation processes (slope = 0) were likely to occur in Beijing during winter. On the other hand, carboxylation process might be unlikely to occur since the slope is higher than -0.5.

The average carbon oxidation state (OSC‾), approximated by 2× O / C–H / C, can also reflect the oxidation degree of OA (Kroll et al., 2011). The estimated OSC‾ values of OA during seasonal observations are listed in Table 2 and the integrated Tri-VK-OSC‾ diagram is shown as Fig. 8. In summer and winter, the average carbon oxidation states were higher, with an average of -0.54 ± 0.27 (in the range of -1.42–0.23) and -0.58 ± 0.25 (-1.32–0.11), respectively. In spring and autumn, the average carbon oxidation states were -0.64 ± 0.37 (-1.96–0.55) and -0.66 ± 0.39 (-1.66–0.64). Compared with previous studies, in spring, the OSC‾ was significantly lower than those at Changdao Island and Bologna sites, but slightly higher than that in Mexico City. In summer, the OSC‾ was comparable to those in Beijing in 2011 and Melpitz and much higher than those in Beijing in 2008, Shenzhen, Shanghai, Riverside, and Jiaxing. In autumn, it was higher than that in Shenzhen, lower than those in Kaiping and Melpitz, and close to that at Heshan site. The OSC‾ in winter was significantly higher than those in Beijing 2010, Fresno, and Jiaxing, but lower than those in Ziyang and Melpitz. Compared with the oxidation states of SV-OOA, OOA, and LV-OOA summarized by Canagaratna et al. (2015), the oxidation states of OOA, especially LO-OOA, in Beijing were generally higher than in other areas (Fig. 8). The oxidation states of MO-OOA in Beijing were only lower than those in very aged air masses in Ziyang in the basin (Hu et al., 2016b), over Mexico City (DeCarlo et al., 2010) and in Barcelona (Mohr et al., 2012). The oxidation states of LO-OOA were only slightly lower than those of MO-OOA in Beijing and were comparable to those of MO-OOA in other urban areas (Fig. 8).

In summary, in urban Beijing, bulk organic aerosols were in a higher oxidation state in summer and in a medium oxidation state in other seasons and were less oxygenated than at several rural/suburban sites. The possible reasons for relatively higher oxidation states of OA in Beijing could be the high hygroscopicity of aerosols and high OH radical concentration in air (Jimenez et al., 2009). The trend of increasing hygroscopicity with increasing O / C is robust based on field observations in Beijing and at several other sites (Jimenez et al., 2009; Wu et al., 2016). Xu et al. (2017) suggest a major role of aqueous-phase processes in increasing the oxidation degree of SOA in Beijing based on the consistency of higher O / C of SOA and higher RH. Lim et al. (2010) suggest that reactions with OH radicals tend to be faster and form more SOA than non-radical reactions. High daily maximum concentrations (0.4–1.7×107 cm-3) for OH were observed in the North China Plain (Lu et al., 2013; Tan et al., 2017).

The aging process of the air mass indicates that the physical and chemical changes occur and that secondary organics or inorganics form continuously, and the physicochemical properties of the air mass are modified during the transport of the air mass from source areas. In previous studies, the ratio or standardized ratio of NOx to NOy concentrations is used as a criterion to characterize the aging degree of air masses (Liang et al., 1998). Here the metric -log (NOx / NOy) is used to investigate the relationship between OA oxidation and the aging of air masses. The larger the metric is, the more aged the air mass (Kleinman et al., 2008; DeCarlo et al., 2008, 2010).

When -log (NOx / NOy)<0.1, the air plume is considered fresh (Liang et al., 1998). The maxima of the parameter -log(NOx / NOy) were about 0.4, 0.4, 0.2, and 0.2 during seasonal observations. This was consistent with the photochemical reactions being more active in spring and summer due to strong solar radiation, while in autumn and winter the plumes of primary emissions had greater impacts. The urban Beijing was under the control of the relatively aged air masses during some periods (e.g., 23–24 April, 17–18 August, and 26–27 October 2012). However, compared with the aerial results in Mexico City, where the highest -log (NOx / NOy) exceeded 1.4 (DeCarlo et al., 2010), the aging degree of air masses in urban Beijing was much lower.

The scatterplot of H / C vs. O / C ratios in Tri-VK-OSC‾ diagram (Fig. 8) is colored with -log(NOx / NOy). There was a trend that the oxidation state of OA was higher (i.e., lower H / C ratio, and higher O / C ratio and OSC‾) as the air mass aged more (from the upper left to the lower right). The OA factors resolved by the AMS-PMF analysis during seasonal campaigns are also marked in Fig. 8. In the order from POA (HOA, COA, CCOA, and BBOA) to SOA (LO-OOA, OOA, and MO-OOA), the factors evolved along with the direction to a higher oxidation state, which is consistent with the oxidation characteristics of the factors.

The OA / ΔCO ratio is used to evaluate the contribution of SOA formation, where ΔCO is CO subtracted the regional background concentration (0.1 ppmv) to exclude the influences of emitted and transported OA (de Gouw and Jimenez, 2009; DeCarlo et al., 2010). With the SOA formation, the OA / ΔCO ratio increases. The average of OA / ΔCO ratios (in the range of 1st and 99th percentile) were 37.0 ± 30.7, 25.0 ± 16.4, 20.3 ± 16.5, and 15.1 ± 8.1 µg m-3 ppmv-1, respectively. The ratios were comparable to or higher than the urban ΔPOA / ΔCO ratios (< 15 µg m-3 ppmv-1), but were much lower than the OA / ΔCO ratios (70 ± 20 µg m-3 ppmv-1) in aged urban air (de Gouw and Jimenez, 2009), implying that the contribution of POA was considerable in Beijing during seasonal campaigns, especially in autumn and winter.

Investigating the variations in OA / ΔCO ratios with the aging process of air masses can help reveal the formation rate of SOA and parameterize it in the model simulation (Dzepina et al., 2011). The scatterplots of OA / ΔCO ratios over O / C ratios during the seasonal observations in Beijing are shown in Fig. 9, colored with the metric -log(NOx/NOy). In laboratory and field studies on OA aging under strong oxidizing conditions, it was found that the OA / ΔCO ratios remain relatively stable at high O / C ratios with the increase in O / C ratios because organics obtain oxygen atoms but lose carbon atoms in the oxidation processes (DeCarlo et al., 2008, 2010). During seasonal campaigns in Beijing, the OA / ΔCO ratios showed different trends with the increase in O / C ratios as well as aging degrees of air masses (Fig. 9). Due to the intricate sources of OA in urban Beijing, the O / C ratios cannot reach higher values and the OA / ΔCO cannot remain an obviously stable level. In autumn, the OA / ΔCO ratios decreased with the increase in O / C ratios in more aged air masses, suggesting that reductive POA contributed substantially to high OA concentrations. An important reason for the decrease in OA / ΔCO ratios in aged air in autumn is the mixing of urban and biomass burning plumes with high CO content (DeCarlo et al., 2010). In summer and winter, along with the increase in O / C ratios and the aging degree of air masses, the OA / ΔCO ratios trended to increase, indicating the material contribution of SOA formation. Guo et al. (2014) concluded that photochemical oxidations of VOCs from urban traffic emissions are primarily responsible for the secondary formation during severe aerosol pollution events. Many studies have reported the rapid increase in SOA significantly exceeding the initial POA emission ratios in urban atmosphere in a day in the absence of biomass burning (DeCarlo et al., 2010).

In mass spectra measured by AMS, some relatively abundant fragment ions have a certain representativeness. As mentioned above, m/z 43 and m/z 44, mainly as C2H3O+ and CO2+, respectively, are predominant in the MS of OOA. While in the MS of HOA m/z 43 is mainly composed of alkyl fragment C3H7+ (Ng et al., 2010). The difference between the relative contents of m/z 43 and m/z 44 in OA can reflect the oxidation degree, e.g., the abundance of m/z 44 in MO-OOA is higher than that in LO-OOA. The fragment m/z 60 (mainly C2H4O2+) is often used as a tracer of biomass burning (Alfarra et al., 2007; Cubison et al., 2011).

The scatterplots of f44 against f43, and f44 against f60 during seasonal campaigns are shown in Fig. 10, colored with -log(NOx / NOy). The data points in the scatterplots of f44 vs. f43 substantially fell into the triangle derived by Ng et al. (2010). As the aging degree of air masses increased, OA showed the evolution trends moving from the lower right to the upper left generally in the triangle. The POA factors (HOA, COA, CCOA, and BBOA) are concentrated in the bottom of the triangle, LO-OOA with a higher oxidation state is in an intermediate location, and MO-OOA with the highest oxidation state is at the top of the triangle. Ng et al. (2010) found that with the enhancement of OA oxidation, data points gradually move upward from the lower half of the triangle through summarizing the results of chamber experiments and field observations.

In the scatterplots of f44 vs. f60, the majority of the data points concentrated on the left side and presented in a band-shaped region (Fig. 10e–h). Cubison et al. (2011) found that in the f44 against f60 space, data from biomass burning appear in the lower right part, while data with negligible biomass burning influence are concentrated on the left side as a band shape. There are a part of data points appearing in the conceptual space for BBOA (Cubison et al., 2011), more obviously in spring, autumn, and winter (Fig. 10e, g, h). In spring and autumn, a large number of data points are of higher f60 than in other seasons, and the aging degrees of air masses were low (-log(NOx / NOy)<0.05), which may be subject to biomass burning emissions as mentioned above. HOA, COA, CCOA, LO-OOA, MO-OOA, and OOA are located on the left side, while BBOA appears on the right. With the increase in the aging degree of air masses, f60 did not trend to change dramatically, suggesting that the contribution of biomass burning to OA has no dependence on the aging degree of air masses.