The experimental setup is illustrated in the Supplement Fig. S1. Detailed procedures for sample preparation and collection may be found in previous studies (Li et al., 2020, 2021; Niu et al., 2020). Briefly, fresh smoke was generated by burning dry biomass fuels (i.e., rice, maize and wheat straw) in a combustion chamber, and the smoke was then passed through a potential aerosol mass oxidation flow reactor (PAM-OFR) (Aerodyne Research, LLC, Billerica, MA, USA) to simulate aging processes on the timescale of hours to days. The biomass combustion chamber had a volume of ∼ 8 m3 (1.8 m (W) × 1.8 m (L) × 2.2 m (H)) and was made of 3 mm thick aluminum to withstand high-temperature heating. The combustion chamber was equipped with a thermo anemometer, an air purification system, a heated sampling line, a dilution sampler and so on. More detailed information about the design and evaluation of the combustion chamber was described in Tian et al. (2015).

In order to get sufficient aerosol samples for measurements of chemical composition, around 1 kg biomass fuels were burned inside the chamber in 10 burning cycles. The entire burning cycle, including ignition, flaming, smoldering and extinction, intends to simulate real-world source characterization. Each burning cycle, containing ∼ 100 g biomass fuels, lasts around 12–18 min. The fresh smoke was diluted by 4.6 times using clean air controlled by the flow balance. A portion of the diluted smoke by a dilution sampler (Model 18, Baldwin Environmental Inc., Reno, NV, USA) was drawn through a quartz fiber filter (47 mm diameter, Whatman QM/A, Maidstone, UK) at 5 L min-1 using a miniVol PM2.5 sampler (Airmetrics, OR, USA) to capture fresh emission.

The PAM-OFR can be used to simulate an environment with extremely high oxidant concentrations with short residence times (Kang et al., 2007). Another portion of the exhaust (∼ 9 L min-1) was directed through a 19 L cylinder PAM-OFR (with a diameter of 20 cm and length of 60 cm) to simulate atmospheric aging. The residence time of PAM-OFR is estimated to be 90 ± 1 s at flow rate of 9 L min-1 (Li et al., 2021). Three oxidants (O3, ⋅OH and ⋅HO2) were generated in the PAM chamber using irradiation from ultraviolet (UV) lamps. The OH exposure values (OHexp) can be calculated by the concentration of SO2 and CO at the OFR inlet and outlet in a laboratory setting. Relative humidity (RH) inside the OFR was varied by passing different amounts of carrier gas through the OFR humidifier (MH-110). Additional details on smoke generation condition, test study and evaluation of the PAM-OFR were described by Cao et al. (2020).

In this study, the UV lamps operated at a voltage of 2 and 3.5 V, and OHexp in the chamber was estimated at 2.6 × 1011 and 8.8 × 1011 molecules s m-3, respectively. These levels corresponded to ∼ 2 and 7 d of aging (Chow et al., 2019; Watson et al., 2019), assuming a representative atmospheric ⋅OH level of 1.5 × 106 molecules m-3 (Mao et al., 2009). The aged aerosols were sampled by another miniVol PM2.5 sampler (5 L min-1) following the reactions in the PAM-OFR chamber. Each test was conducted in triplicate to account for experimental errors and to provide a measure of variability, which was calculated as standard deviations. A total of 36 samples were collected and analyzed for chemical composition.

For diacids, ketocarboxylic acids and α-dicarbonyls analysis, one-quarter of each filter sample was extracted three times (15 min each) with purified (18.2 MΩ) water (Milli-Q, Merch, France) and ultrasonication. The pH of the aerosol extracts was adjusted to 8.5 to 9.0 using a 0.1 M potassium hydroxide solution prior to drying that converts carboxylic acids into their salts (Bikkina et al., 2021). This drying step improves the recovery of smaller diacids, such as C2 (Hegde and Kawamura, 2012). Water extracts were concentrated to near-dryness with a rotary evaporator under vacuum and then reacted with 14 % BF3/n-butanol at 100 ∘C for 1 h to derivatize carboxyl groups to dibutyl esters and oxo groups to dibutoxyacetals.

After derivatization, n-hexane was added and washed with pure water three times to remove the water-soluble inorganics such as hydrogen fluoride and boric acid. The hexane layer was concentrated to near-dryness using a rotary evaporator under vacuum and a N2 blow-down technique, and then the esters and acetals of target analytes were dissolved in known amounts of n-hexane. Finally, the hexane layers were concentrated to 100 µL and analyzed using a capillary gas chromatograph (GC; HP 6890, Agilent Technology, Santa Clara, CA, USA) equipped with a split/splitless injector and a flame ionization detector (FID). Peak identification was performed by comparing the GC retention times with those of authentic standards and confirmed by a thermal desorption (TD) unit coupled with a gas chromatograph/mass spectrometric detector (TD-GC/MS, Models 7890A/5975C, Agilent Technology, Santa Clara, CA, USA). The detection limits for those organic compounds were 0.1 ng m-3, and the analytical errors, based on the replicate analyses, were less than 15 %. Recoveries of the target compounds were 83 % for C2 and 87 % to 110 % for the other species.

Concentrations of the various species in the aged samples were affected by their initial emission and also undergo degradation and production through secondary chemical processes. Fresh and aged fuel-based EFs for each measured chemical compound were calculated by dividing its filter mass by the mass of combusted dry biomass fuel (Andreae and Merlet, 2001; Li et al., 2020; Tian et al., 2015); that is 1EFi=mi×vStk×D×tsampleQp×mfuel×DR, where EFi (mg kg-1) is the EF of chemical compound i for the specific crop, mi (mg) is the mass of chemical compound i collected on the filter, vStk is the average stack flow velocity (m s-1) at standard conditions, D is the stack cross section (m2), tsample is the sampling duration (s), Qp is the sampling volume through the filter (m3) at standard temperature and pressure, and mfuel is the mass of burned biomass fuel (kg, dry weight).

The dilution ratio (DR) was determined from the CO2 concentrations measured at the stack, diluted stack and background, where 2DR=CO2,Stk-CO2,BkgCO2,Dil-CO2,Bkg, where CO2,Stk is the CO2 concentration in the stack, CO2,Bkg the background CO2 concentration in the atmosphere and CO2,Dil the CO2 concentration in the diluted smoke.

Stable carbon isotopic determinations (δ13C) of diacids, ketocarboxylic acids and α-dicarbonyls followed the techniques of Kawamura and Watanabe (2004). The isotope values of the derivatized samples were determined using a gas chromatography–isotope ratio mass spectrometer (GCIR-MS; Thermo Fisher, Delta V Advantage, Franklin, MA, USA). The δ13C values were then calculated for free organic acids using an isotope mass balance equation based on the measured δ13C values of derivatives and BF3/n-butanol (Kawamura and Watanabe, 2004). To ensure the analytical error of the δ13C values less than 0.2 ‰, each aerosol sample was analyzed in triplicate to obtain the average values.

Fresh and aged PM2.5 EFs for a homologous series of diacids, ketocarboxylic acids (glyoxylic acid, ωC2, and pyruvic acid, Pyr), α-dicarbonyls (glyoxal, Gly, and methylglyoxal, mGly) and benzoic acid are presented in Table 1. The EFs for most fresh and aged diacids varied by several orders of magnitude, with higher EFs after atmospheric aging. The highest fresh EF (i.e. EFfresh) was found for wheat straw ranging 44–122 mg kg-1 for succinic acid (C4) and 67–102 mg kg-1 for Gly, higher than EFs found in maize and rice. The arithmetic means and standard deviations for the EFfresh of total diacids from burning of rice, maize and wheat straws were 63 ± 24, 117 ± 39 and 285 ± 135 mg kg-1, respectively.

As is shown Fig. 1, distributions of diacids in fresh emissions varied by crop type and species. Of the saturated n-dicarboxylic acids, C4 acid was the most abundant species in the maize and wheat straw, with average EFfresh of 22 ± 12 and 83 ± 46 mg kg-1, respectively. Azelaic acid (C9) and C4 were the most abundant species from rice burning, with EFfresh of 11 ± 2.9 and 10 ± 9.0 mg kg-1, respectively. These findings are consistent with the fresh smoke aerosols in Siberian BB plumes (Kalogridis et al., 2018), in which C4 and C9 were more abundant than C2. Previous studies also showed C9 to be an oxidation product of unsaturated fatty acids in biomass smoke (Kawamura and Gagosian, 1987; Kawamura et al., 2013; Agarwal et al., 2010; Cao et al., 2017). C2 is the most abundant species of diacids and is one of the final products of the SOA reaction chain. In the fresh BB sample, C2 emissions were lower due to the short aging time.

Similar to the diacids, the highest EFfresh for ketocarboxylic acids and α-dicarbonyls was also found in wheat straw samples, with 44 ± 31 and 138 ± 91 mg kg-1, respectively. Gly was the highest α-dicarbonyl, with an average EFfresh of 27 ± 3.9, 42 ± 10 and 84 ± 41 mg kg-1 for rice, maize and wheat straw, respectively. This is consistent with previous studies which showed that Gly is often more abundant than mGly in polluted aerosols collected from China (Pavuluri et al., 2010; Ho et al., 2007). Benzoic acid also was determined, and its EFfresh for rice, maize and wheat aerosols was 1.9 ± 0.2, 2.5 ± 0.4 and 3.1 ± 0.3 mg kg-1 (Table 1).

Patterns in the relative abundances of diacids have been used to evaluate biogenic versus anthropogenic source strengths and the photochemical processing of organic aerosols (Kawamura et al., 2012). Previous studies have shown that C4 can be directly oxidized into C2 or via C3 into C2 (Jung et al., 2010; Sorooshian et al., 2007), with C2 being an end product of the photochemical oxidation (Wang et al., 2012). The ratios of C3 / C4, C2 / C4 and C2 / total diacids can be regarded as indicators of aerosol aging (Cheng et al., 2013; Kunwar et al., 2019; Meng et al., 2018; Pavuluri et al., 2010), with higher ratios indicative of more aged aerosols (Kawamura and Sakaguchi, 1999). As shown in Table 2, the ratios in this study showed a clear atmospheric aging trend from fresh to 7 d aging, with ratios of 0.7 to 6.4 for C2 / C4, 0.1 to 0.6 in C2 / total diacids and 0.2 to 0.5 in C3 / C4, indicating obvious photochemical oxidation.

Ratios of ωC2 / C2 and Gly / mGly can also be used to evaluate the oxidation of organic aerosols (Cheng et al., 2013, 2015; Kawamura et al., 2013). In the study, apparent reduction of the ωC2 / C2 ratios from 1.3 (fresh) to 0.2 (7 d) supports the potential oxidation pathways from precursor glyoxylic to oxalic acids. Aqueous-phase oxidation by OH is faster for Gly than for mGly, and the abundance of Gly relative to mGly is an indicator of aerosol aging (Cheng et al., 2013). The ratio of Gly / mGly in Xi'an samples was lower on haze days than on clean days and lower in summer than in winter. Similarly, the Gly / mGly ratios in the aged BB samples were higher in the fresh PM2.5 samples (3.8) compared to the 2 d (2.3) and 7 d (2.0) aging.

Ratios of C3 / C4, C2 / diacids, ωC2 / C2 and Gly / mGly are similar among studies, except for the higher C3 / C4 ratio of 3.9 found in marine aerosols of over the Pacific region (Kawamura and Sakaguchi, 1999) and lower C3 / C4 ratios in Siberian BB emissions in a large aerosol chamber (< 0.03) (Kalogridis et al., 2018). The largest difference was found for C2 / C4, which varied from < 1 for fresh aerosol in Siberian BB (Kalogridis et al., 2018) to 25.2 from forest fire in Thailand (Boreddy et al., 2021). Elevated C2 / C4 ratios exceeding 10 were found in the aged ambient atmosphere of Xi'an, China (10.4) (Cheng et al., 2013), Mt. Hua, China (10.7) (Meng et al., 2014), marine aerosol, Pacific Ocean (14.3) (Kawamura and Sakaguchi, 1999), and the ambient atmosphere of Okinawa Island, Japan (15.5) (Kunwar and Kawamura, 2014). These C2 / C4 ratios are ∼ 63 % to 142 % higher than these reported in this study. Overall, these comparisons show the importance of photochemical aging; however, the atmospheric oxidation evidently was more extensive in aerosols from some remote mountain and marine environments.

Stable carbon isotope ratios (δ13C) can provide insights into the sources of aerosols. Pavuluri and Kawamura (2016) reported that average δ13C values for C2 from biogenic aerosols (-15.8 ‰) were less negative; i.e., they contained more 13C and were isotopically more enriched than those from anthropogenic aerosols (-19.5 ‰). Data for δ13C can also provide information on the processing or aging of organic aerosols because isotopic fractionation results from chemical reactions or phase transfer (Pavuluri and Kawamura, 2016; Zhang et al., 2016). Mass loading of δ13C for diacids in the fresh BB samples was too low to be detected by the GCIR-MS, but the δ13C values for C2 ranged from -23.3 % to -21.0 ‰ (with an average of -21.9 ± 1.2 ‰) in 2 d and -19.1 ‰ to -15.5 ‰ (-17.3 ± 1.7 ‰) for 7 d aged samples (Table 3).

Table 3 shows that the average δ13C values of C2 from aged maize samples were higher than those of rice and wheat. The reason for the isotope difference may be that maize is a C4 plant, while wheat and rice are both C3 plants. Song et al. (2018) showed that δ13CTC in C4 plants is isotopically heavier than in C3 plants. Moreover, the δ13C of C2 is more abundant in 7 than 2 d samples (Table 3), with -13.1 ± 1.6 ‰ (2 d) and -7.1 ± 1.4 ‰ (7 d) in maize; -26.2 ± 1.8 ‰ (2 d) and -20.8 ± 3.3 ‰ (7 d) in rice and -26.5 ± 0.2 ‰ (2 d) and -24.0 ± 0.5 ‰ (7 d) in wheat combustion. The δ13C data for C3, C4 and ωC2 (Table S2) showed similar trends, consistent with previous studies. For example, Zhao et al. (2018) found that the δ13C values of C2 were related to aging. Pavuluri and Kawamura (2016) analyzed diacids, ωC2 and Gly for δ13C in anthropogenic and biogenic aerosol samples by UV irradiation and reported more δ13C less negative with longer irradiation times. During atmospheric oxidation reactions, organic compounds react with OH radicals, causing the release of CO2 and CO which contain relatively the lighter 12C isotope and thus leaving the remaining substrate enriched in 13C (Hoefs, 1997; Sakugawa and Kaplan, 1995).

A comparison of δ13C values for C2 in the aerosols from selected environments is shown in Fig. 4. The average δ13C value (-21.9 ± 1.2 ‰) of 2 d aged biomass burning of C2 was comparable to the values reported for urban regions, such as Beijing (-21.8 ± 2.8 ‰) (Zhao et al., 2018) and Liaocheng (-19.8 ± 3.1 ‰) (Meng et al., 2020) (Table 3). With continued aging, the C2δ13C of the 7 d samples (-17.3 ± 1.7 ‰) was more similar in samples from Mt. Tai (-16.5 ± 1.8 ‰) (Meng et al., 2018) and western Pacific and Southern Ocean aerosol (-16.8 ± 0.8 ‰) (Wang and Kawamura, 2006), but it was significantly lighter than that of samples from the Gosan Climate Observatory at Gosan, South Korea (-13.7 ± 2.5 ‰), which is a mountain background site in East Asia (Zhang et al., 2016).

The chamber experiment (Niu et al., 2020) measured VOC compounds. Table S3 presents the correlations between decreases in VOCs and increases in diacids from fresh to 2 d aged BB samples. Significant (0.01 < p < 0.05) correlations (R) were observed for toluene with Gly (R= 0.75), mGly (R= 0.81), Pyr (R= 0.78), ωC2 (R= 0.78) and C2 (R= 0.67) (Fig. 5), suggesting that toluene was converted to diacids during the aging process. Indeed, it has been reported that the photooxidation of toluene is a potential source of secondary organic aerosol (SOA) in urban air (Sato et al., 2007), and the major chemical components of SOA include hemiacetal, peroxy hemiacetal oligomers and diacids. It also can be seen that benzene had significant correlations with mGly and C2 (R>0.59 in Fig. 5), implying that the oxidation of benzene led to diacid formation. Photooxidation of Gly and mGly is a major global and regional source of C2 diacid, and the two formation pathways are Gly–ωC2–C2 and mGly–Pyr–ωC2–C2, respectively (Yasmeen et al., 2010; Wang et al., 2012). As shown in Fig. 5, the slope (0.20–0.59) between the decrease of toluene and the increase of intermediates (Gly, mGly, Pyr and ωC2) is significantly higher than C2 (0.04). This was the same for benzene; the slope between the decrease of benzene and the increase of mGly is 0.55, while for C2 it is only 0.05.

On the global scale, isoprene is the most important precursor for C2, contributing 70 % to global C2, while anthropogenic VOCs contribute 21 % to C2 production (Myriokefalitakis et al., 2011). Thus, it is not surprising that isoprene correlated with C2 (R= 0.58) (Fig. 5). In addition, several alkenes and alkanes also had a significant correlation with C2 (Table S3), indicating that these species may react in secondary oxidation processes to generate C2. Previous studies have confirmed that diacids can be oxidation products of aromatic hydrocarbons (Borrás and Tortajada-Genaro, 2012) and cycloolefins (Hamilton et al., 2006) and may originate from diesel vehicle exhaust (Samy and Zielinska, 2010). However, no significant correlation was found between decreases in VOCs and increases in 7 d aged diacids. For the longer aging times, the particulate-phase compounds may be further oxidized to generate other compounds besides diacids. Such a correlation between decreases in VOCs and increases in diacids again suggests that 2 d aging may be sufficient to oxidize VOCs to diacids.