The sampling site is situated on the rooftop of a building (8 m above ground level) in the Institute of Atmospheric Physics (39∘58′28′′ N, 116∘22′16′′ E), which is considered a representative urban site in Beijing (Sun et al., 2012). PM2.5 samples were collected onto preheated (450 ∘C, 6 h) quartz-fibre filters (Pallflex) by using a high-volume air sampler (TISCH, USA) at an airflow rate of 1.0 m3 min-1 for 23 h from September 2013 to July 2014 (n= 65). Field blanks were prepared before, during and after the campaign by putting pre-combusted filters onto the sampler for a few minutes without pumping. After sampling, filters were properly stored at -20 ∘C to avoid microbial degradation of organics and evaporation of semi-volatile components. Beijing is surrounded by Hebei Province and Tianjin Municipality with intensely developed industries (Xia et al., 2007), so the atmospheric visibility and quality in Beijing sometimes seriously deteriorate owing to the substantial primary aerosols from these areas.

Aerosol samples were analysed for diacids and related compounds using a method reported previously (Kawamura, 1993; Kawamura and Ikushima, 1993). In brief, water-soluble organic acids were obtained from ultrasonic extraction for small discs of PM2.5 samples submerged in Milli-Q water three times. Then, the sample extractions were concentrated to dryness and further reacted with 14 % BF3/n-butanol. Finally, the derivatized extracts were dissolved in n-hexane and analysed by a split/splitless Agilent 6980GC/FID equipped with an HP-5 column (0.2 mm × 25 m, 0.5 µm film thickness). The same procedure was used to analyse the field blank filters. Concentrations of the target organic acids in this study were corrected for the field blanks. Furthermore, recoveries of major organic acids of this method were better than 85 %.

The determination of stable carbon isotope ratio (δ13C) values for LMW organic acids relative to Pee Dee belemnite (PDB) was analysed using the technique developed previously (Kawamura and Watanabe, 2004). In short, an internal standard (n-alkane C13) was added to derivatized fraction of each sample at the correct proportion. δ13C values of the derivatized dibutyl esters or dibutoxy acetals, measured using GC (HP6890)/isotope ratio mass spectrometer (irMS), were then calculated for diacids, keto acids and α-dicarbonyls based on the isotopic mass balance equation. Each aerosol sample was analysed several times to make sure that the differences for major diacids in δ13C were below 1 ‰ in general. But for a few compounds, the analysis differences were less than 2 ‰.

A part of each filter was extracted with 20 mL of Milli-Q water under ultrasonication for 30 minutes and passed through a filter head of 0.22 µm nominal pore size (PVDF, Merck Millipore Ltd). Ion chromatography (ICS-2100) was used to determine the concentrations of cations (Na+, NH4+, K+, Mg2+ and Ca2+). The separation of cations was accomplished by using an IonPac CS 12A (4 × 250 mm) analytical column, with an eluent flow rate of 1.0 mL min-1. Another ICS-2100 system was used to measure the concentrations of anions (F-, Cl-, NO3- and SO42-). The separation of anions was accomplished using an IonPac AS11-HC analytical column. The eluent was 25.0 mM KOH at a flow rate of 1.0 mL min-1. The anions and cations were analysed separately after the extraction solution was divided into two paths.

For the WSOC measurement, 3.14 cm2 of each filter was extracted by Milli-Q water (20 mL). After 15 min sonication, the extraction was measured by a Shimadzu TOC-V CPH total carbon analyzer (Aggarwal and Kawamura, 2008). OC and EC were determined by using thermal optical reflectance (TOR) following the Interagency Monitoring of Protected Visual Environments (IMPROVE) protocol on a DRI Model 2001 thermal/optical carbon analyzer (Chow et al., 2005). The limit of detection (LOD) for the carbon analysis was 0.8 for OC and 0.4 µgC cm-2 for EC, with a precision of greater than 10 % for total carbon (TC). The concentrations of inorganic ions, WSOC and OC / EC reported here are all corrected for the field blanks.

To better evaluate the influences of air masses from different origins on organic aerosols in Beijing, 5-day backward trajectory analyses with fire spots were performed for each sample from the sampling site at a height of 500 m a.s.l. by using the Hybrid Single-Particle Lagrangian Integrated Trajectory (HYSPLIT4) model (Rolph et al., 2017). Burning activities in East Asia were illustrated by fire spots, and the datasets were downloaded from the MODIS website (https://earthdata.nasa.gov/earth-observation-data/near-real-time/firms). The backward trajectories were assigned to several major classes on the basis of the prevalent wind direction. And the numbers in each panel implied the percentages of hourly trajectories in the sampling season to better illustrate the air mass origins, as given below in this study (Fig. 1).

Table 1 summarizes the seasonal concentrations of LMW diacids (C2–C12), oxocarboxylic acids (ωC2–ωC9, pyruvic acid), and α-dicarbonyls (C2–C3) in PM2.5 particles in Beijing. Oxalic acid (30.8–1760 ng m-3, average 288 ng m-3) was the predominant individual diacid, showing a peak in autumn and a minimum in winter, whereas its relative abundances (0.39–0.58, average 0.52) to total measured diacids exhibited a maximal and a minimal ratio in summer and wintertime, respectively (Table 3). The predominance of C2 found in this study was coincident with the results of terrestrial and marine particles in previous studies (Kawamura and Yasui, 2005; Ho et al., 2007; Pavuluri et al., 2010; Fu et al., 2013). Among ω-oxocarboxylic acids (ωC2–ωC9), glyoxylic acid (ωC2) was detected as the dominant oxoacid.

Either succinic acid (C4) or occasionally tPh was the second most abundant compound, followed by ωC2 in cold seasons (autumn and winter) or malonic acid (C3) in warm seasons (spring and summer). Total diacids showed the largest abundance in autumn, followed by spring, whereas total oxoacids and α-dicarbonyls both displayed higher levels in cold seasons, especially in autumn (Table 1). The concentrations of individual dicarboxylic acid reduced along with the increase in carbon numbers, but in the range of longer-chain diacids (C6–C12), adipic (C6) and azelaic (C9) acids showed more abundances than other species in the atmosphere throughout the year (Fig. 2).

Seasonal trends in homologue series of diacids and other main species presented three different patterns. The first type, such as oxalic, malonic, succinic, adipic acids and methylglyoxal (MeGly), showed maximum concentrations in autumn with a relatively high abundance in late spring to early summer. The second type showed maximum concentrations in cold seasons (autumn, winter): phthalic, terephthalic and glyoxylic acid peaked in winter, while azelaic acid (C9), glyoxal (Gly), methylmaleic (mM), maleic (M) and fumaric (F) acids peaked in autumn. For the third type, concentrations of methylmalonic (iC4) and 2-methylglutaric (iC6) acids were almost constant throughout the year. These three seasonal patterns indicate different emission sources of the compounds and their precursors and evolution processes of organic aerosols in the atmosphere.

Total concentrations of diacids showed a wide range (110–2580 ng m-3) with an average maximum (763 ng m-3) and minimum (366 ng m-3) in autumn and winter, respectively. These values were comparable to those in Tanzania, East Africa (wet season: 329 ng m-3; dry season: 548 ng m-3 in PM2.5; Mkoma and Kawamura, 2013), slightly lower than those in Tokyo, Japan (726 in June, 682 in July and 438 ng m-3 in November; Kawamura and Yasui, 2005) and a rural site at Gosan, Jeju Island (735 in spring, 784 in summer, 525 in autumn and 500 ng m-3 in winter; Kundu et al., 2010b). The comparisons of the diacids in Beijing with those in other urban cities are presented in Table 2.

Daily variations in diacids and other major organic acids are given in Fig. 3. Oxalic acid has been recognized as the end product that is associated with atmospheric chain reactions of organic species with oxidants (Kawamura and Sakaguchi, 1999). C2 can be generated in abundant quantities by vehicular emission (Kawamura and Kaplan, 1987; Donnelly et al., 1988), biomass burning activities (Turnhouse, 1987; Destevou et al., 1998; Schauer et al., 2001; Kundu et al., 2010a), fossil fuel combustion (Kawamura and Kaplan, 1987; Rogge et al., 1993), and the photo-oxidation of volatile organic compounds and other precursors transported over a long distance (Kawamura and Yasui, 2005; Kundu et al., 2010b). Malonic acid was detected at relatively low concentrations in four sampling periods, with the highest abundance in autumn. The daily variation in malonic acid was almost the same as succinic acid. Concentrations of C4 diacid being in excess of C3 diacid implies that primary emissions contributed more to dicarboxylic acids, a typical pattern that is frequently obtained in aerosols emitted from biomass burning (Kawamura et al., 2013), vehicular exhaust (Ho et al., 2010) and fossil fuel combustion (Ho et al., 2007). The daily variation tendency of C2 resembled that of C3 and C4, indicating that these compounds may have similar photochemical oxidation pathways or emission sources in the atmosphere.

In addition to shorter-chain diacids (C2–C4), azelaic acid (C9) had the highest concentration among the saturated diacids in all seasons (Table 1). C9 is a photochemical oxidation product of unsaturated fatty acids derived from natural biogenic sources such as terrestrial higher plants and sea-to-air emission of marine organics as well as anthropogenic emissions including biomass burning (Kawamura and Gagosian, 1987). Under favourable atmospheric conditions, the photo-oxidation of biogenic unsaturated fatty acids to C9 with oxidants, such as O3, OH and HO2, is inclined to occur in air (Stephanou and Stratigakis, 1993). Additionally, tyre wear debris and traffic exhaust also make contributions to the abundance of LMW fatty acids like C18:0, a category of C9 precursors (Rogge et al., 1993).

Azelaic acid was observed in abundance throughout the whole sampling period, while the monthly mean ratios of C9 to total diacids (C9 / Tot) ranged from 0.05 to 0.09, with the highest values in winter (Table 3). Kawamura and Kaplan (1987) reported that C9 can be detected in motor exhaust and may originate from the oxidation of corresponding hydrocarbons, suggesting that dicarboxylic acids are combustion products of normal alkanes in fuels. A great deal of chloride in wintertime is linked to the increased emission of coal incineration in Beijing, particularly under stagnant meteorological conditions that facilitate the formation of particle-phase ammonium chloride (Y. L. Sun et al., 2013). Azelaic acid correlated well with K+ (0.3 ≤ r2≤ 0.4), a tracer for biomass burning (Andreae, 1983), and Cl- (0.4 ≤ r2≤ 0.5) in cold seasons (Fig. S1 in the Supplement), indicating that substantial amounts of C9 may stem from the local and surrounding combustion activities in Beijing.

Ph and tPh are both more abundant in cold seasons than in warm seasons. We found Ph to be the fourth most abundant species in winter (37.9 ± 27.2 ng m-3) and summer (24.9 ± 8.2 ng m-3). Concentration ranges of Ph (7.6–98.5 ng m-3; mean: 31.7 ng m-3) in cold seasons were larger than those (0.08–7.47 ng m-3; mean: 1.76 ng m-3) from Gosan, Jeju Island (Kundu et al., 2010b), but were obviously lower than those (53–278 ng m-3; mean: 150 ng m-3) in urban Xi'an (Y. L. Sun et al., 2013). Phthalic acid is either formed via photochemical pathways of naphthalene or directly released into air by fossil fuel burning and the incomplete combustion of aromatic hydrocarbons in motor vehicles. Moreover, the abundance of Ph may also be caused by increased phthalates emissions from plastic waste burnings in heavily polluted areas in China (Deshmukh et al., 2016). Phthalic acid esters are used as plasticizers in resins and polymers (Simoneit et al., 2005). Therefore, anthropogenic emissions contributed to relatively high concentrations of Ph in PM2.5 in Beijing.

Terephthalic acid (tPh), the second highest abundant diacid in winter (48.7 ± 41.1 ng m-3), showed a pattern in contrast to a previous study that reported Ph as the second most abundant compound (Ho et al., 2010). Terephthalic acid is directly emitted from plastic wastes incinerations in ambient air (Simoneit et al., 2005; Fu et al., 2010; Kawamura and Pavuluri, 2010). High concentrations of tPh observed in winter indicate substantial plastic waste incineration. Another phthalic isomer, isophthalic acid (iPh), was also detected in the samples; concentrations of this isomer had a seasonal pattern similar to those of Ph and tPh throughout the year. However, the concentrations of iPh were the lowest among these isomers.

Oxocarboxylic acids, which are understood as the intermediate products of the oxidation of mono-carboxylic acids, can further be photochemically oxidized to form diacids (Warneck, 2003; Carlton et al., 2007). Concentrations of all keto acids varied from 9.50 to 353 ng m-3 during sampling periods with a maximum (73.3 ± 76.3 ng m-3) in autumn and a minimum (46.5 ± 26.8 ng m-3) in summer. Except for ωC7 and ωC8, the oxoacids showed larger concentrations in cold seasons (autumn and winter), which might be attributed to accumulation under stagnant meteorological conditions. Their concentrations were higher than those in aerosols from Tanzania (60.0 ± 19.0 and 31.0 ± 18.0 ng m-3 in PM2.5 during dry and wet seasons, respectively; Mkoma and Kawamura, 2013) but much lower than these detected in Mangshan, a rural site in Beijing (159 ng m-3 in the daytime, 97.9 ng m-3 in the nighttime; He et al., 2014).

Glyoxylic acid (ωC2) is measured as the most abundant oxoacid, followed by pyruvic (Pyr) and 4-oxobutanoic (ωC4) acids. All of them are important intermediates in photo-oxidation processes for the production of low carbon-number diacids such as C2, C3 and C4 diacids (Hatakeyama et al., 1987). ωC2 and Pyr are more abundant in cold seasons (Table 1) with similar seasonal patterns (Fig. 3g–h). Both correlated well with K+ and Cl- in sampling seasons (Figs. S2–S3), which demonstrate that ωC2 and Pyr originated from common combustion emissions or similar secondary formation pathways. 9-Oxononanoic acid (ωC9), photochemically generated from unsaturated fatty acids (Kawamura and Gagosian, 1987), showed larger concentrations in autumn and winter. This concentration trend was consistent with that of azelaic acid. Additionally, a lower thermal inversion layer, less precipitation and a slower wind speed can enhance the accumulation of organic compounds.

Total concentrations of α-dicarbonyls varied across a wide range (1.50–85.9 ng m-3) and were relatively more abundant in cold seasons (25.1 ± 28.1 in autumn, 15.5 ± 15.9 ng m-3 in winter). The average seasonal concentrations are larger than those at Gosan, Jeju Island (Kundu et al., 2010b). Glyoxal (Gly) and methylglyoxal (MeGly) are semi-volatile gaseous organic precursors that can be produced by the oxidation of isoprene (Zimmermann and Poppe, 1996), monoterpenes (Fick et al., 2004) and other biogenic volatile organic compounds (VOCs; Ervens et al., 2004) and anthropogenic aromatic hydrocarbons (e.g. benzene and toluene; Volkamer et al., 2001). Both carbonyls can form less volatile polar organic acids including Pyr and ωC2 in subsequent oxidation processes, which are key intermediates to produce oxalic acid. Gly and MeGly correlated well with nss-K+ (Gly: 0.3 ≤ r2≤ 0.9, MeGly: 0.3 ≤ r2 ≤ 0.9) throughout the whole year (Fig. S2), whereas they showed good relations with Cl- (Gly: 0.3 ≤ r2≤ 0.8, 0.4 ≤ r2≤ 0.8) in autumn, winter and summer (Fig. S3). Concentrations of these two carbonyls are largely affected by biogenic precursors (e.g. isoprene and monoterpenes) emitted from vegetation and biomass burning activities during entire sampling periods in addition to coal burning and motor exhaust (aromatic hydrocarbons). Low temperature is favourable for the adsorption and condensation of gaseous organic species on existing particles in cold seasons.

The ratio of oxalic acid to total diacids (C2 / Tot) has been applied to estimate the relative contribution of secondary fraction to atmospheric aerosols during long-range transport. Typically, higher mass concentration ratios are associated with more aged aerosols (Kawamura and Sakaguchi, 1999). The ratios of C2 / Tot were the lowest in winter (0.39 ± 0.05; Table 3), indicating that wintertime organic aerosols may be less aged (Fig. 4a). After the primary emissions of PM2.5 particles from motor vehicles, fossil fuel and biomass burning activities in local regions in winter, aging processes occur during atmospheric transport. In contrast, C2 / Tot ratios are similar in the other three seasons (Table 3). In total, the seasonal mean values of C2 / Tot in this study were lower than the winter ratios (0.8 ± 0.04) in Central Himalayan aerosols owing to the aging of organic compounds occurring in the northerly wind but were close to those in summer (0.5 ± 0.01) due to increased temperature and high wind in the Central Himalayas (Hegde and Kawamura, 2012). Thus, the ratios of C2 / Tot and this seasonal trend indicate that the photochemical formation of dicarboxylic acids is insignificant in urban Beijing.

Two pathways for the generation of C2, C3 and C4 diacids in air were reported by Kawamura et al. (1996a). On the one hand, C4 diacid can be generated via the photo-oxidation of unsaturated fatty acids from terrestrial higher plants and domestic cooking over continental lands, as well as from phytoplankton emissions over the remote marine regions (Kawamura and Sakaguchi, 1999), and it can subsequently be oxidized to form C3 and C2 diacids (Kawamura and Ikushima, 1993). Typically, concentrations of C2, C4 and C9 diacids showed similar seasonal variation trends (Fig. 3), implying that they were derived from common primary emissions or photochemical processing. Furthermore, C2 / Tot (C2 %) showed strong correlations with C2 / C4 in all four seasons (Fig. S4), indicating the importance of biogenic unsaturated fatty acids, followed by secondary formation ways. On the other hand, aromatic hydrocarbons may be oxidized to produce Gly and ωC2, which are intermediates in the formation of C2 (Kawamura and Ikushima, 1993; Kawamura and Bikkina, 2016). Biogenic and anthropogenic VOCs (e.g. isoprene) can react with oxidants to generate Gly and MeGly in the gas phase. Hydrated α-dicarbonyls can ultimately produce C2 via the photochemical oxidation of Pyr and ωC2 as intermediates (Lim et al., 2005), whereas C3 and C4 diacids cannot be produced in this way.

Therefore, we investigated the importance of these formation pathways by examining the interrelationships between concentration ratios of C2 / Pyr, C2 / ωC2, C2 / Gly and C2 / Tot in this study. Relatively high ratios of C2 / Pyr, C2 / ωC2 and C2 / Gly were observed in autumn, spring and summer (Fig. 4c–e), but their values were much lower than those detected in particulate matters at Gosan, Jeju Island (Kawamura and Yasui, 2005; Kundu et al., 2010b), where the aerosols were relatively aged during long-range transport. Figure S4 shows that strong relationships between C2 / ωC2, C2 / Pyr and C2 / Tot (%) only existed in summer. No correlation was observed between C2 / Gly and C2 / Tot (%). A negative correlation between C2 / ωC2 and C2 / Tot (%) in summer has never been reported previously. These phenomena demonstrate that abundant C2, ωC2, Pyr and Gly were emitted directly from extensive agricultural residue burning, motor vehicles and fossil fuel burning in the studied regions. The supplement of ωC2, Pyr and Gly is faster than their photodegradation to form C2 in air. Only slightly stronger photochemical production of C2 from Pyr was observed in summer.

C3 diacid can be produced as a result of hydrogen abstracted by OH radicals, followed by decarboxylation processing of C4 diacid (Kawamura and Ikushima, 1993). The mass concentration ratio of C3 / C4 is a good indicator for evaluating the contributions of dicarboxylic acids from primary emissions or secondary oxidation production in the atmosphere (Kawamura and Sakaguchi, 1999). Lower C3 / C4 ratios were detected in vehicular exhaust by Kawamura and Kaplan (1987), ranging from 0.25 to 0.44 with an average of 0.35. Less thermally stable C3 diacid can degrade more preferentially to other species rather than remaining stable during incomplete combustion processes.

Figure 4b shows that the C3 / C4 ratios were relatively larger in the warm seasons (spring and summer). However, the temporal trend in the C3 / C4 ratios is relatively flat throughout the sampling year, and most values are less than or equal to unity in Beijing. A low C3 / C4 ratio is associated with the substantial emissions from motor vehicles (Kawamura and Kaplan, 1987). By contrast, the prolonged secondary oxidation of organic matter leads to C3 / C4 values much greater than unity (Kawamura and Ikushima, 1993; Kawamura and Sakaguchi, 1999). The ratios of C3 / C4 reported in this study are lower than that (1-year average of 1.49) in urban Tokyo (Kawamura and Ikushima, 1993) and in the remote Pacific Ocean (average 3; Kawamura and Sakaguchi, 1999), where dicarboxylic acids are largely produced by photochemical reactions. These results demonstrated that in addition to slightly enhanced atmospheric photo-oxidation in summer, incomplete combustions, like motor vehicles and biomass burnings, overwhelmingly contributed to dicarboxylic acids in Beijing.

Phthalic acid (Ph) was one of the most abundant compounds during the sampling period. The seasonal trend in the ratio of phthalic acid to total dicarboxylic acids (Ph / Tot) is shown in Fig. 4h. The Ph / Tot ratios in winter were nearly 2–3 times greater than those in spring and autumn, which implies that phthalic acid is largely emitted by anthropogenic sources in winter, mainly as a result of intensive coal burning for house heating. It is worth noting that the Ph / Tot ratio was also relatively high in summer. A previous study reported that a great amount of naphthalene obtained in Beijing is an important raw material for the substantial formation of phthalic acid (Liu et al., 2007). Therefore, increased ambient temperatures and stronger solar radiation in summertime facilitate the transformation of gaseous polycyclic aromatic hydrocarbons (PAHs; e.g. naphthalene) to produce relative high levels of Ph in Beijing.

C6 and Ph can be formed via secondary oxidation of anthropogenic cyclic olefins (e.g. cyclohexene) and aromatic hydrocarbons, respectively, whereas C9 is mainly produced by photochemical oxidation of biogenic unsaturated fatty acids (Kawamura and Gagosian, 1987; Kawamura and Ikushima, 1993). Thus, the mass concentration ratios of C6 / C9 and Ph / C9 may effectively indicate the source strength of anthropogenic and biogenic emissions to these organic acids.

The seasonally averaged ratios of C6 / Tot, C9 / Tot, C6 / C9 and Ph / C9 are displayed in Table 3. Mean values of C6 / Tot are constantly low in all four seasons, whereas the seasonal ratios of C9 / Tot are the highest (0.09) in winter and the lowest (0.05) in summer, which results in the lowest value of C6 / C9 ratios in winter (0.34 ± 0.13), and are almost constant in the other three seasons. This trend is different from the one detected in the Central Himalayan region (1.07 in winter and 0.56 in summer; Hegde and Kawamura, 2012) and Chennai, India (0.42 for winter and 0.29 for summer; Pavuluri et al., 2010). In contrast, the values of Ph / C9 are relatively high in winter (1.40 ± 0.69) and summer (1.33 ± 0.39), followed by spring (0.92 ± 0.33) and autumn (0.82 ± 0.39); its ratios are obviously lower than the values found in 14 other megacities in China (2.71 for winter and 3.37 for summer; Ho et al., 2007) but are a bit higher than those in Tokyo (0.65 1-year mean value; Kawamura and Ikushima, 1993). From the outcomes discussed above, we concluded that the contribution from anthropogenic emissions was the main source in megacities. Ph / C6 reached the highest values in winter (4.06 ± 0.78) and the lowest ratio in autumn (1.66 ± 0.78). Kawamura and Kaplan (1987) demonstrated that the Ph / C6 ratio from gasoline fuel vehicle (2.05) is lower than that from diesel fuel vehicles (6.58). This phenomenon shows abundances of diacids attributable to more emissions from a gasoline fuel vehicles than diesel burning.

Maleic acid (M), originated predominantly from photochemical oxidation of aromatic hydrocarbons (e.g. benzene and toluene), can be subsequently isomerized to its trans-isomer, fumaric acid (F), under favourable conditions. Lower M / F values have been detected in atmospheric aerosols over the North Pacific Ocean (0.3; Kawamura and Sakaguchi, 1999) as well as at Alert in the high Arctic (ratio range: 0.5–1.0; Kawamura et al., 1996a). The M / F ratios are almost constant in winter (2.0 ± 0.66) and spring (2.0 ± 0.67) and are higher than those in autumn (1.67 ± 0.81) and summer (1.35 ± 0.49). This trend may be associated with substantial amounts of precursors emitted from biomass burning in autumn, fresh aerosols initiated by high-speed winds in spring and enhanced isomerization reaction from M to F under intense solar radiation in summer. The conversion of maleic to fumaric acids can be restrained in polluted environments with minimum weak sunlight (Kundu et al., 2010a). Therefore, M may not be effectively isomerized to F during wintertime in Beijing. Though M / F had the lowest ratio values in summer, dicarboxylic acids and related compounds in Beijing are not seriously subjected to secondary oxidation processes over the whole year when compared to the strength of primary emissions.

Based on field observation, Kawamura and Ikushima (1993) hypothesized that C4 diacid can transform into malic acid (hC4) by means of hydroxylation. The hC4 / C4 ratios were the highest in warm seasons (0.04 ± 0.01 in summer and 0.03 ± 0.02 in spring), which supported this hypothesis. hC4 / C4 ratios in summer are 2–4 times larger than those in cold seasons, similar to the trends observed in Jeju Island, Korea (Kundu et al., 2010b), and urban Tokyo (Kawamura and Ikushima, 1993).

To assess the emission strength of anthropogenic activities in Beijing, the mean values of C3 / C4 (Fig. 5a), M / F (Fig. 5b), Ph / C9 (Fig. 5c), Ph / Tot (Fig. 5d) and tPh / Tot (Fig. 5e) mass ratios were compared with those at other sampling sites, including inland Xi'an city (Wang et al., 2012); Gosan, Jeju Island (Kundu et al., 2010a); and the western Pacific Ocean (H. Wang et al., 2006). Xi'an, a megacity in the Guanzhong Plain, is located in one of the regions heavily polluted by fossil fuel and biofuel combustion. Atmospheric aerosols at the Gosan site are mixtures of westerly winds from high-latitude regions of Eurasia (Fu et al., 2012b). Marine aerosols over the western Pacific Ocean are a combination of long-range transported continental aerosols and locally emitted marine aerosols (Chen et al., 2013).

Figure 5 presents the global distribution of diagnostic mass ratios of diacids and related compounds. Rather low C3 / C4 ratios were observed in urban aerosols, including Beijing and Xi'an, compared to those aged organic matter collected from Gosan, Jeju Island, and the western Pacific Ocean. Similarly, larger C3 / C4 ratios were obtained in summer than in the other seasons. The same observation in Beijing may be attributable to the enhancement of secondary oxidation that favours the conversion of C4 to C3 in the warm season; however, in that case, the photochemical activity is insignificant compared to the primary emissions. Similar to the C3 / C4 ratios, low M / F ratios indicate the importance of photochemical reaction routes (Kawamura and Sakaguchi, 1999). The mean values of M / F in the Beijing aerosols are larger than or comparable to those reported in Gosan, Jeju Island (spring: 1.4; summer: 0.76; autumn: 1.6; and winter: 2.2), but lower than those obtained in Xi'an aerosols (summer: 2.22; winter: 2.38), indicating that the PM2.5 aerosols in Beijing are mainly linked with regional primary emissions, whereas the photo-isomerization from cis- to trans-isomer is insignificant.

High Ph / C9 ratios were detected in continental samples owing to a relatively strong contribution from anthropogenic sources to dicarboxylic acids. Slightly larger values of Ph / C9 (on average) were obtained in Xi'an than in Beijing because the air masses in Xi'an were more heavily influenced by intense industrial emissions. Although the values of Ph / C9 in both megacities were higher than those in the western Pacific Ocean, the wintertime Ph / C9 ratios in Gosan were much greater than those in Beijing, which may be caused by the secondary generation of abundant precursors, such as naphthalene, which were transported across a long distance from East Asia.

In this study, we calculated the ratios (%) of Ph and tPh to total diacids to estimate the primary emission strength at different sampling sites. The largest mean mass ratios of Ph / Tot were observed during winter in Beijing, while the values in the other seasons were lower than those observed in Xi'an due to its basin-like topography. For tPh / Tot ratios, the mean values in Beijing were much higher than those in marine areas. However, the average value of tPh / Tot in winter was lower than that in Xi'an. Thus, these comparisons illustrate significant contributions from waste plastic burning and fossil fuel combustion in Beijing during wintertime.

Previous studies have utilized principal component analysis (PCA) to discriminate between the source apportionments of atmospheric aerosols (Hopke, 1985). In this paper, typical dicarboxylic acids with other major components were chosen for factor analysis. Compounds with common sources or photo-oxidation reactions would be likely to display similarities in mass variations and be assigned to one “factor”. High loadings of variables on the selected species reveal closer links of sources and formation pathways between these compounds (Wu et al., 2015). Here, “total varimax” maximizes the variance of the squared elements in the columns of a factor matrix. The PCA results for dicarboxylic acids and other main components in PM2.5 in Beijing from September 2013 to July 2014 are given in Table 4.

During the whole sampling period, the first factor accounted for 75.2 % of the total variance with high loadings of selected diacids, WSOC and EC (a tracer for incomplete combustion-generated carbon emissions). Typically, the prolonged photochemical oxidation of organics in the atmosphere leads to enhanced concentrations of polar organic matter. WSOC can account for 45–75 % of aerosol carbon mass in biomass burning emissions (Falkovich et al., 2005) and 20–60 % of that in fossil fuel combustion-derived particles (Pathak et al., 2011b). Agricultural waste burning is a serious pollution factor in Beijing (Fig. 1; Viana et al., 2008; Cheng et al., 2014), especially in late June and early October, resulting in substantial organic aerosols (Fu et al., 2012a). In many studies, open-waste burning aerosols are mixed with biomass and fuel burning aerosols (Akagi et al., 2011; Lei et al., 2013). C4, C9, tPh, ωC2, Pyr, Gly and MeGly showed strong correlations in the first factor, implying that burning activities contribute to a large fraction of their concentrations, including biomass burning, biofuel combustion and burning of municipal wastes. For example, the photo-oxidation of p-xylene, a main precursor of the terephthalic acid dimethyl ester, can produce glyoxal (Volkamer et al., 2001; Simoneit et al., 2005; Kawamura and Pavuluri, 2010).

EC and maleic and phthalic acids are clearly associated with other species, indicating that they originate from common mixed sources that are mainly produced by anthropogenic emissions, such as vehicular exhaust, fossil fuel combustion and biomass burning. Aromatic hydrocarbons from incomplete combustions are key materials for maleic and phthalic acids (Kawamura and Sakaguchi, 1999). Both M and Ph showed high abundances under hazy conditions (Mochida et al., 2003).

As for the second factor, Ph, tPh and EC have weak loadings, which seem to originate from motor emissions, fossil fuel combustion and waste plastic burning. WSOC also showed a slight loading in the second factor, which indicates that anthropogenic emissions also contribute to a certain amount of WSOC during the sampling periods.

The systematic differences in stable carbon isotope ratios of diacids and other polar acids were attributable to kinetic isotope fractionation processes in the atmosphere (Hoefs and Hoefs, 1997), while secondary oxidation of these water-soluble organic acids is more influential for diacid carbons to enrich in 13C (Wang and Kawamura, 2006). For example, the relatively short carbon-chain diacids enriched in 13C were ascribed to the kinetic isotopic effect (KIE) for the photochemical breakdown of longer-chain diacids (Anderson et al., 2004; Irei et al., 2006). Lower dicarboxylic acids with enrichment of 13C may be less active to oxidants (e.g. OH radicals). Therefore, the determinations of δ13C values of dicarboxylic acids and related compounds show vital information about the atmospheric aging processes of aerosols derived from local emissions or long-range transport in air.

Table 5 presents the stable carbon isotope ratios of major compounds. The mean δ13C values of C2, C3 and C9 were constant among seasons, but those of C4, ωC2, Pyr were smaller in summer than in winter. Because coal is more enriched in 13C than in petroleum fuel (Court et al., 1981; Kawashima and Haneishi, 2012), the 13C enrichment of these organic acids during wintertime may be attributable to the enhanced coal incineration for house heating. Mean δ13C values of malonic acid in autumn and spring were similar to those of succinic acid, suggesting that they may have similar sources or the same secondary formation pathways.

The mean seasonal δ13C values of C9 varying from -25.6 to -26.9 ‰ were smaller than those of C2–C4 diacids. It is noteworthy to state that the depletion of 13C in continental higher plants (C3 plants: -27 ‰) is greater than the particulate organic matter from marine plankton activities (around -20 ‰) (Turekian et al., 2003; Miyazaki et al., 2011). The δ13C values of azelaic acid suggested that unsaturated fatty acids derived from biomass burning in the surrounding areas are a key source of C9 in Beijing.

As mentioned earlier, Ph is mainly formed via the photochemical processes of polycyclic aromatic hydrocarbons, but it can be emitted directly from fossil fuel combustion as well (Kawamura and Kaplan, 1987; Fraser et al., 2003). The largest δ13C value of Ph in winter was linked with its peak concentrations. This finding may be ascribed to the intensity of coal and gasoline combustion in Beijing, especially the stagnant atmospheric conditions that favour the accumulation of organic matter during wintertime (Cao et al., 2011). In general, organic aerosols that are derived from coal and gasoline burning are more enriched in 13C than other emissions, including diesel combustion, aerosols released from C3 plants and secondary organic matter.

For terephthalic acid, the lowest δ13C value of tPh (ave: -33.5 ‰) in winter supports the finding that it is directly emitted from the burning of plastic wastes. Waste burning usually contains many plastics and occurs frequently in open areas without emission control (Kawamura and Pavuluri, 2010), in addition to other local anthropogenic emissions. Lighter δ13C values of major compounds in Beijing than those in the marine and Arctic areas may be explained by more contributions of primary emissions from anthropogenic sources.

Box plots of stable carbon isotope ratios (δ13C values) are displayed in Fig. 6 for seasonal distributions of diacids, glyoxylic and pyruvic acids in PM2.5. The δ13C values of oxalic acid ranged from -27.2 to -14.8 ‰ in sampling time, with similar seasonal mean δ13C values. Malonic acid was more enriched in 13C than others in autumn (-17.6 ‰) and summer (-18.7 ‰), while succinic acid showed the greatest δ13C value (-17.1 ‰) among all species in winter and spring. A previous study noted that increasing concentrations of oxalic and malonic acids inhibit the growth of total fungi number due to the lower pH, which in turn changes the efficiency of fungi to degrade malonic acid (Côté et al., 2008). Hence, an enrichment of 13C in the remaining malonic acid may be interpreted by the isotopic fractionations occurring in the breakdown of dicarboxylic acids or the photochemical degradation of C3 diacid (Pavuluri and Kawamura, 2012). In this study, the median δ13C values of ωC2 were much lower than those of C2 in all seasons, whereas the δ13C of Pyr showed median values similar to C2 in autumn and spring, which are surprisingly higher than that of C2 in autumn.

In order to further estimate the impacts of air mass source regions on δ13C values of specific compounds, 5-day backward trajectories for each aerosol sample are illustrated in Fig. 1. Data from urban Sapporo (Aggarwal and Kawamura, 2008), Gosan on Jeju Island (Zhang et al., 2016) and remote marine regions (Kawamura and Watanabe, 2004; Wang and Kawamura, 2006) are plotted together with the seasonal mean δ13C values of major species detected in this study (Fig. 7). The largest average δ13C value of oxalic acid was observed in the Gosan samples. The seasonal mean δ13C values of malonic acid in Beijing were higher than those in Sapporo and remote marine areas, but C3 was less enriched in 13C compared to Gosan owing to the degradation of C3 diacid or C2 diacid depleted in 13C. The mean δ13C values of succinic acid are comparable to those in the other three places, except for summer. The mean δ13C values of Pyr in autumn (-19.6 ‰) and spring (-22.3 ‰) were similar to the data in Sapporo (-20.3 ‰) and Gosan (autumn: -19; winter: -22.2; spring: -19.1; summer: -17.6 ‰) aerosols. The δ13C values of ωC2 and Ph in remote marine samples are the highest, followed by those for the Sapporo and Gosan sites and then Beijing. In contrast, the seasonal mean δ13C values of C6 and C9 in Beijing are similar to those in Sapporo and Gosan aerosols but lower than those in marine aerosols.

The air masses in Gosan, Jeju Island, and Sapporo are mixtures of the flows from the mainland of East Asia. The δ13C values illustrate that organic aerosols in Sapporo are formed via the photo-oxidation of precursors originated from anthropogenic and biogenic emissions (e.g. biomass burning) to a large extent, especially C6 and C9; however, the study in Gosan found that aerosol samples are more aged in the western North Pacific rim. Most importantly, particulate organic matter in remote marine areas is intensively aged during long-range transport and is affected by both the sea-to-air emissions and the terrestrial outflows. Moreover, the enrichment in 13C can be regarded as a result of the isotopic fractionation for aged aerosols. Urban aerosols from Beijing, where the air masses are mixed with those originating from Siberia and surrounding areas, are seriously affected by biomass/biofuel burning in the whole year. Compared with the δ13C values in Gosan, Sapporo and remote marine areas, the smaller δ13C values of organic compounds in Beijing may be caused by the different emission strengths of various primary sources.

C2 / Tot ratio is suggested to be a useful tracer to evaluate the aging of atmospheric aerosols (Kawamura and Sakaguchi, 1999). The mean δ13C values of oxalic acid showed the smallest value in winter (-22.9 ‰) and the highest value in autumn (-20.1 ‰), followed by spring (-21.9 ‰; Fig. 8a). Here, we compared the δ13C values of C2 and its concentration changes with the relative abundance of C2 to total diacids (Fig. 8b). The isotopic values of C2 positively correlated with C2 / Tot ratios in autumn (r2= 0.45) and winter (r2= 0.29), suggesting that the production of C2 from the oxidation of precursors can contribute to the increase in δ13C values (Pavuluri et al., 2011). Due to enhanced primary emissions from coal combustion and biomass burning, stagnant atmospheric inversion can favour the accumulation of pollutants. Furthermore, the δ13C values of tPh decreased from autumn to winter, followed by an increase toward summer (Fig. 9). Seasonal δ13C values of tPh decreased with the enhanced ratios of tPh to total diacids (tPh / Tot) in autumn (r2= 0.35) and winter (r2= 0.19), indicating large emissions from municipal waste burning activities in cold seasons, especially in winter.

Aged organic aerosols are characterized by a high abundance of polar and water-soluble organic species, leading to high WSOC / OC ratios (Fu et al., 2015). However, in the Beijing samples, the δ13C values of major species (C2, C3, C4, C9, ωC2, Pyr, Ph and tPh) did not show strong relationships with the WSOC / OC ratios in this paper (Fig. 10). The δ13C values of C3 only correlated well with the WSOC / OC ratios in summer (r2= 0.57), in conformity with the variation in C3 / C4 ratios, which illustrates an enhanced degree of photochemical processing of diacids during summertime. The δ13C values of C4 were negatively correlated with WSOC / OC ratios in the cold seasons (autumn: 0.31; winter: 0.45), demonstrating an enrichment of δ13C in C4 with decreasing WSOC / OC ratios. Ph displayed negatively weak correlations in summer, while ωC2 presented weakly positive and negative relations in autumn (0.2) and spring (0.29), respectively. The positive relationship between the δ13C values of Pyr and WSOC / OC in autumn (0.62) suggests that an isotopic enrichment of Pyr increases with high WSOC / OC ratios, which may have resulted in the largest δ13C values of Pyr in autumn. There are no correlations between C2, tPh and C9 with the WSOC / OC ratios in the Beijing samples. Thus, the results discussed above suggest that primary emissions in local regions significantly impact diacids and related compounds in Beijing.