The smoke particles were collected by the instrument coupled with a dilution channel which was designed to simulate fire emissions representative of “real-world” open BB and household CC activities (Fig. S1 in the Supplement). In the present study, a total of 27 BB samples (IDs1–27) were collected at Xishuangbanna, Yunnan, from 20 May to 3 June 2016, and the detailed sampling process was described in our previous article (Cui et al., 2018). Briefly, raw fuels (rough 20×3×2 cm3) were air-dried for several days and ignited in a stainless-steel bowl, and then the rising smoke was collected through a dilution system. The sampling system mainly consists of a dilution tunnel, a residence time chamber, three PM samplers, and so on. Every biomass was burned three times at about 1–2 kg of fuel per burn. Every combustion process lasted for 20 min. The collection of smoke particles started when the fuel ignited, and it ended when the concentration of CO2 was down to atmospheric CO2 level. Dilution ratios of each experimental process were calculated using the CO2 concentrations before and after dilution. The collection flow rate and average dilution ratio were 180 L min-1 and 2.1, respectively. And the other six BB samples (IDs 28–33) were collected in Guangzhou, Guangdong.

The smoke particles of CC (IDs 34–50) were collected in the same way as that of the BB experiment, but they used a stove in Guangzhou, Guangdong, from 18 November 2017 to 23 January 2018. The tested stove is a technically improved stove (named Jin-Yin stove). Due to the difficulty of ignition of coal, we used smokeless charcoal to ignite one third (about 300 g) of the raw-coal chunk (2–5 cm in size) in the stove, removed the charcoal after ignition, and then added the remaining raw-coal chunk (about 700 g) to start to collect the smoke particles. Every coal was also burned three times at about 1 kg of fuel per burn. Every combustion process lasted for about 40–150 min. The collection flow rate and average dilution ratio were 150 L min-1 and 1.5, respectively. Additionally, the modified combustion efficiency (MCE) was calculated to characterize the relative amount of smoldering and flaming combustion phase (Lin et al., 2016; Cui et al., 2018). The average MCE value was 0.73±0.08 for CC experiments, but it was unavailable for the BB experiments because the CO sensor did not work in the field work, which was mentioned in our previous paper (Cui et al., 2018).

Tunnel aerosols (total eight samples; IDs 51–58) were collected at the Siping tunnel from 1 to 2 November 2017 and the Xiaoyang Shan tunnel from 1 to 2 December 2017, in Shanghai and two vehicle exhaust particles (IDs 59–60) were collected from the direct emission of two different trucks (more fresh aerosols). With no other instructions, we used “vehicle emissions” to represent all tunnel aerosols and vehicle exhaust particle samples. These filters were wrapped in aluminum foil and pre-baked at 450 ∘C for 5 h before sampling and stored at -20 ∘C after sampling. Overall, there was a total of 60 total suspended particulate (TSP) matter samples on source emissions in this study, and blank samples that were collected at different times and locations were used for correcting filter samples.

WSOC for UV–vis absorption and EEM analysis were extracted with purified water (resistivity of >18.2Ω) via the ultra sonication of quartz filter punches for 30 min. Because water cannot effectively extract the BrC (Liu et al., 2013; Shetty et al., 2019), the remaining filter was further freeze-dried and extracted with methanol (high-performance liquid chromatography grade; HLPC) to obtain the MSOC constituent for a better understanding of the optical properties and molecular compositions of BrC. It is worth noting that the MSOC in this study is not necessarily like that of the same name in other studies. All the extracts were filtered through a 0.22 µm polytetrafluoroethylene membrane into amber colored-glass vials to remove the insoluble material.

We measured both OC and elemental carbon (EC) using an aerosol carbon analyzer (Sunset Laboratory, Inc., USA), following the National Institute for Occupational Safety and Health (NIOSH) thermal–optical transmittance (TOT) standard method (Mo et al., 2017), and the emission factors (EFs) of PM, OC, and EC were calculated, and detailed information was presented in the Supplement. We also analyzed the elements of biomass (C, H, O, and N) and coal (C, H, O, N, and S) using an elemental analyzer (Vario EL cube, Elementar, Germany), and the results were listed in Tables S1 and S2. The carbon content of WSOC was measured using total organic carbon analysis (Vario TOC cube, Elementar, Germany) before acidification with phosphoric acid to remove inorganic carbon, while the concentration of MSOC was assessed using the method developed by a previous study (Chen et al., 2017b). Briefly, the extracted MSOC was dried gently under nitrogen and then re-dissolved in 500 µL of methanol. Subsequently, 50 µL of the solution was added to the clear quartz filter (area of 1.5 cm2) until dry and analyzed using the TOT standard method.

The UV–vis absorption and EEM spectra of WSOC and MSOC were analyzed using a UV–vis spectrophotometer (UV-4802, Unico, China) and an Aqualog fluorometer (HORIBA Scientific, USA), respectively. The wavelengths used to characterize the UV–vis spectra were between 200 to 800 nm at a step size of 2 nm. Purified water and methanol were used as a baseline correction for WSOC and MSOC before measurement, respectively. Mass absorption efficiency (MAE; m2 g-1C) was obtained from the following equation (Li et al., 2018): 1MAEλ=Aλ×ln⁡(10)/(C×L), where Aλ is the value of light absorption at the given wavelength of the spectrophotometer, C (µg C mL-1) is the concentration of WSOC and MSOC, and L is the optical path length (in this study, 0.01 m). Moreover, the pH of WSOC was measured for all samples within the ranges of 5.5–6.5, which generally did not affect the absorbance according to the prior study (Chen et al., 2016a).

The emission and excitation wavelengths of the fluorescence spectra were from 245 to 580 nm and 240 to 500 nm, respectively. The wavelength increments of the emission and excitation scans were 4.66 and 3 nm, respectively. Further, the contributions of solvents to the fluorescence spectra were subtracted.

The WSOC and MSOC of six selected samples including two BB aerosols (Musa and Hevea), two CC aerosols (a anthracite and a bituminous coal), 1 d of tunnel aerosol (combining the aerosols in the inlet and outlet of the tunnel in the same day), and one vehicle exhaust particle sample were analyzed using FT-ICR MS. To remove inorganic ions before instrumental analysis, WSOC was further adjusted to a pH of 2 by the addition of hydrochloric acid (HCl) and then passed through a solid-phase extraction (SPE) cartridge (Oasis HLB – hydrophilic-lipophilic balance, 30 µm, 60 mg/cartridge, Waters Corporation, USA). The constituent retained on the SPE cartridge was eluted with methanol containing 2 % ammonia (v/v). Eluants were evaporated until dry under a gentle nitrogen gas stream. The solutions extracted by methanol were evaporated under a gentle nitrogen gas stream for preparation.

We used the analysis method of FT-ICR MS described in detail in our previous study (Mo et al., 2018). Briefly, ultrahigh-resolution mass spectra were obtained using a solariX XR FT-ICR MS (Bruker Daltonics GmbH, Bremen, Germany) equipped with a 9.4 T superconducting magnet and an ESI ion source. The system was operated in negative-ionization mode. The ion accumulation time was set to 0.6 s. The lower and upper mass limit was set to a mass-to-charge ratio (m/z) of 150 and 800 Da, respectively. The mass spectra were externally calibrated with arginine clusters using a linear calibration and then internally recalibrated with typical O6S1 class species peaks using quadratic calibration in DataAnalysis version 4.4 (Bruker Daltonics). A typical mass-resolving power of >450 000 at m/z 319 with <0.2 ppm absolute mass error was achieved. The mass spectra of field blank filters was analyzed to detect possible contamination following the same procedures. More data processing was presented in Sect. S1 of the Supplement.

PARAFAC analysis with non-negativity constraints was used to explore the fluorescent components in dissolved BrC based on the method established by the previous studies (Murphy et al., 2013; Andersson and Bro, 2000), which was performed using the decomposition routines for Excitation Emission Matrices (drEEM) toolbox version 2.0 using MATLAB (http://models.life.ku.dk/drEEM, last access: June 2014). This method had been widely used in the analysis of fluorescence spectra in aerosols (Chen et al., 2016a, b; Matos et al., 2015; Wu et al., 2019). Absorbance measurements were used to correct the EEM for inner-filter effects (IFEs) according to the previous studies (Luciani et al., 2009; Gu and Kenny, 2009; Fu et al., 2015). The highest light absorbance in the calibrated wavelength range of WSOC and MSOC was not greater than 2 (mostly below 1 at 254 nm), which was appropriate for the inner-filter corrections of the EEMs (Gu and Kenny, 2009; Murphy et al., 2013). Each EEM was normalized to the Raman peak area of purified water collected on the same day to correct fluorescence in Raman units (RUs) at an excitation of 350 nm and corrected for the dilution factor (Murphy et al., 2013, 2010). Additionally, the signals of the first-order and second-order Rayleigh and Raman scattering in the EEM were removed by an interpolation method (Bahram et al., 2006). Repeated convergence of the model was examined based on the iteration of the minimum square principle. The exploration phases of two- to seven-component PARAFAC models contained an evaluation of the shape of spectral loading, leverage analysis, an examination of the core consistency, residual analysis, and split-half analysis (Figs. S2–S7). A six-component PARAFAC model was identified and successfully passed the split-half validation with the split style of S4C6T3 for the WSOC and MSOC in 60 samples, respectively.

The PM, OC, and EC EFs of 27 biomass and 17 coal combustion experiments were summarized in Table S3. The relevant EFs of some biomass species have been previously reported (Cui et al., 2018). In this experiment, the EFs of PM, OC, and EC from 27 types of biomass burning were 15±11 g kg-1 fuel, 8.0±6.4 g kg-1 fuel, and 7.7×10-1±3.4×10-1 g kg-1 fuel, respectively. The EFs emitted from bituminous CC (PM = 9.1×10-1±6.5×10-1 g kg-1 fuel, OC = 4.2×10-1±3.3×10-1 g kg-1 fuel, and EC = 9.4×10-2±1.9×10-1 g kg-1 fuel) were much higher than those of anthracite combustion (PM = 1.5×10-1±8.9×10-2 g kg-1 fuel, OC = 1.2×10-2±4.5×10-3 g kg-1 fuel, and EC = 1.6×10-4±1.4×10-4 g kg-1 fuel) in the same stove. These differences could be attributed to the high volatile matter content of bituminous coal (Tian et al., 2017; Chen et al., 2005). Note that the CC smoke collection began when the flame had been ignited with one third of the material, and the rest was added. Therefore, the results of our study would be lower than the real values.

MAE can be used to characterize the efficiency of solar energy absorption, which is represented by the degree of conjugation and the amount of electron delocalization in molecules (Chen et al., 2016a). As shown in Fig. 1 and Table S4, MAE at 365 nm (MAE365) was significantly higher in the case of BB and CC aerosols than in vehicle emissions in this study, consistent with the previous findings (Xie et al., 2017; Fan et al., 2016). Bituminous CC aerosols had higher MAE365 values than anthracite combustion aerosols. Here, we introduced the EC / OC ratios, which could be used as an indicator of fire conditions (Xie et al., 2017). Figure S8 showed the MAE365 of WSOC vs. EC / OC relationships for all BB and CC aerosols. The data clearly showed that the WSOC light absorption of BB aerosols was dependent on combustion conditions. However, a weak relationship (p>0.05) in CC aerosols suggested another factor might influence the light absorption, such as maturity (Li et al., 2018). Compared to WSOC, higher MAE365 values were observed in the MSOC collected from BB (2.3±1.1 m2 g-1C) and bituminous CC (3.2±1.1 m2 g-1C) aerosols. This could be due to the fact that these strongly light-absorbing fat-soluble components are likely to be PAHs of a large molecular weight and quinones from BB and fossil fuel combustion (Sun et al., 2007; Chen and Bond, 2010), which were more soluble in a low-polarity solution, but we obtained the opposite results in the case of anthracite combustion and vehicle emissions.

The MAE365 of WSOC in this study was compared with the other studies (Fig. 1). The BB aerosols in this study had a higher MAE365 value than those in other controlled BB experiments, while it was comparable to corn straw burning emissions (Park and Yu, 2016; Fan et al., 2016). Further, the simulated BB aerosols exhibited a higher MAE365 value than those in highly BB-impacted areas (Hecobian et al., 2010), indicating that the aging in the transport process could reduce the light absorption (Dasari et al., 2019). The CC aerosols showed a higher MAE365 value than the other coal experiments (Li et al., 2018; Fan et al., 2016), while comparable values to water-soluble BrC were observed in winter of Beijing (Cheng et al., 2011; Yan et al., 2015). The result indicated the strong influence of BrC in this season in this region. In addition, the simulated combustion aerosols in this study exhibited higher MAE365 values than the other areas (such as Guangzhou, Nanjing, Los Angeles, South Korea, Nepal, and so on) (see Fig. 1).

Methanol has a lower polarity than water and can extract the water-insoluble compounds that are generally stronger chromophores. Chen et al. (2017b) extracted organic matters in aerosols using different polar solutions, and they found water-insoluble organic matter (WIOM) had a higher MAE value than the water-soluble organic matter (WSOM), consistent with our result for the BB and bituminous CC aerosols. Vehicle emission aerosols generally had a lower MAE value such as methanol-soluble BrC (0.62±0.76 m2 g-1C) in the controlled emission experiment (Xie et al., 2017), which was comparable to WSOC (0.71±0.30 m2 g-1C) but higher than MSOC (0.26±0.09 m2 g-1C) in this study.

Fluorescence spectra was used to characterize the organic chromophores of different sources. We applied the PARAFAC model (Murphy et al., 2013) to determine the underlying chromophore components of the 60 source samples. Six typically independent components (abbreviated P1–6) in WSOC were resolved, as shown in the top of Fig. 2 and Table 1. Compared to the previous studies, the fluorescence of P1 and P6 were similar to those of 7CM–C1 (the C1 component of a seven-component model) and 7CM–C3, named humic-like substances (HULIS-1) (Chen et al., 2016b). Further, there were peaks in the emission wavelengths (>400 nm) of P1 and P6, which were probably derived from conjugated systems (Chen et al., 2016b). The peak of the P3 component was mostly located in the region categorized as protein-like (cytidine) or tryptophan-like fluorophore (Qin et al., 2018; Fan et al., 2016). Generally, peaks at shorter excitation wavelengths (<250 nm) and shorter emission wavelengths (<350 nm) were associated with simple aromatic proteins such as tyrosine (Cory and Mcknight, 2005), which was similar to the fluorescence of the P2 component observed in this study. The P5 component was similar to tryptophan- and tyrosine-like components (Chen et al., 2017a). Therefore, the P2, P3, and P5 components were named protein-like substances (PLOM). The P4 component was reported relatively rarely but was similar to previously observed peaks that were considered to arise mainly in surface water and algal secretions (Yu et al., 2015). It is worth noting that the origins and chemical structures of the chromophores studied are not necessarily like those of chromophores with the same name in other types of organic matter.

The results from the six-component model (abbreviated C1–6) of MSOC identified by PARAFAC, as described in the bottom of Table 1 and Fig. 2, were different from those observed in WSOC, indicating MSOC contained different compound types from WSOC after water extraction. The peak of the C1 component was similar to that of the P3 component of WSOC, but the excitation wavelength was higher than that of the P3 component. The higher excitation wavelength indicated the presence of conjugated unsaturated bond systems shifting towards the high wavelengths of the C1 component (Matos et al., 2015). Moreover, as reported, the C3 component was similar to component 2 of urban alkaline-soluble organic matter (ASOM) collected from the city of Aveiro, Portugal (Matos et al., 2015).

The maximum fluorescence intensity (Fmax) was calculated by multiplying the maximum excitation loading and maximum emission loading for each component by its score (Murphy et al., 2013). Generally, changes in the relative abundance of a component (Fmax/∑Fmax) could indicate changes in its overall importance, which had been successfully applied to study the origins of chromophores (Yan and Kim, 2017; Chen et al., 2017a, b; Wu et al., 2019). In this study, the relative abundances of fluorescent components in different types of samples were highly variable, depending on the sources (Fig. 3a). The P1 component accounted for an average of 34±4.7 % of the total fluorescence intensities in the case of tunnel aerosols, which was higher than BB aerosols (mean ± SD: 19±4.8 %), CC aerosols (14±3.8 %), and vehicle exhaust particles (17±1.0 %). The difference of the P1 component between tunnel aerosols and vehicle exhaust particles indicated that the P1 component had an aged vehicle exhaust origin. In contrast, the fluorescence of the P6 component was weak in all the samples, but higher in vehicle emissions (9.4±2.3 %) than in BB and CC aerosols (both 2.5 %). The P5 component was more intense in vehicle exhaust particles (30±1.6 %) than in other sources. The P2 component was abundant only in BB aerosols (33±11 %), but not in vehicle emissions, which suggested that some structures responsible for this chromophore could not exist in vehicle emissions. The P4 component was the more abundant chromophore in CC aerosols (34 ± 7.7 %) and vehicle emissions (29±5.9 %), especially in vehicle exhaust particles (38±1.1 %). In contrast, the P4 component in BB aerosols was weak (11±7.9 %), indicating a fossil origin. The P3 component was almost equal across all samples. The possible reason was that the P3 component was similar to the peak of tryptophan-like compounds which were common in practically all published models and were likely to be found in almost all sources (Yu et al., 2015).

The relative intensities of fluorescent components in MSOC exhibited similar characteristics to WSOC (Fig. 3b). The C1 component was the most intense substance in the case of BB aerosols (38±14 %) compared to the other sources. The C2 component was enriched in BB aerosols (21±6.9 %) and tunnel aerosols (17±6.9 %) compared to those in CC aerosols and vehicle exhaust particles. In addition, C2 exhibited a difference between bituminous CC and anthracite combustion aerosols, as well as tunnel aerosols and vehicle exhaust particles, indicating that the C2 component could be used to identify these sources. The C4 component was intense in CC aerosols (41±6.0 %) and vehicle emissions (26±4.4 %). The C3 component was not abundant among the three sources and not observed in the vehicle exhaust particles, not suggesting a fresh vehicle exhaust emission origin. The C5 and C6 components were more intense in vehicle exhaust particles (25±6.8 % and 50±6.8 %, respectively), suggesting they were the primary vehicle emission chromophores. The last study observed that the relative abundances of various chromophores in aerosols with different particle sizes were different (Chen et al., 2019). Therefore, the fluorescence technique is sensitive for chromophores with different sources, sizes, and chemical structures, and so on. In summary, the variation of the fluorescent components from different sources obtained by EEM–PARAFAC method could be helpful to the source apportionment of BrC in environmental applications.

The molecular compositions of WSOC and MSOC extracted from BB and CC aerosols and vehicle emissions were determined by negative ESI-FT-ICR MS. ESI is a soft ionization method, and it can only ionize polar organic compounds and hydrophilic molecules (Wozniak et al., 2008), but nonpolar or less polar compounds such as polycyclic aromatic hydrocarbons and saturated hydrocarbons are not easily ionized by ESI (Lin et al., 2018). In addition, ESI- cannot detect the N-heterocyclic alkaloid compounds (Laskin et al., 2009). Thus, this study mainly discussed these readily ionizable polar organic compounds by ESI-.

Figure 4 showed the reconstructed negative-ion ESI FT-ICR mass spectra of WSOC for the six selected samples. Lots of peaks with intensive mass ranges between m/z 150 and 600 were showed in the mass spectra, with the most massive numbers of ions within the ranges of m/z 200–400. Additionally, more formulas were detected in BB aerosols (total of 7708) than CC aerosols (5305) and vehicle emissions (4047) (Table 2), suggesting a higher observed chemical complexity (i.e., the observed peaks). According to the intensity of each ion, the average molecular formulas of WSOC in the six aerosol samples were calculated and listed as follows: C18.7H23.5O6.99N0.73S0.09, C19.9H21.5O7.65N0.34S0.03, C16.1H13.3O5.37N0.68S0.23, C15.2H13.7O4.24N0.45S0.41, C13.4H18.0O7.52N0.45S0.40, and C17.3H21.1O5.65N0.53S0.08 for Musa, Hevea, anthracite, bituminous coal, tunnel, and vehicle exhaust, respectively. The BB aerosols had higher contents of C and H, while the CC aerosols and tunnel aerosol had higher contents of S.

In this study, these identified molecular formulas were classified into four main compound groups based on their compositions: CHO, CHON, CHOS, and CHONS. CHO compounds refer to the compounds that contain carbon, hydrogen, oxygen, and the other compound groups that are defined analogously. The relative abundances of the four compound groups were calculated by the magnitude of each peak divided by the sum of magnitudes of all identified peaks and showed in Fig. 4. CHO was the most abundant component in WSOC, accounting for 43 %–69 % of the total intensities of BB aerosols, 36 %–37 % of CC aerosols, and 36 %–47 % of vehicle emissions, respectively. CHO values in BB and CC aerosols were lower than those of mass spectra from simulated combustion experiments (BB of 53 %–72 % and CC of 43 %; Song et al., 2018). Generally, CHO formulas were consistent with species reported previously as lignin-pyrolysis products (Fleming et al., 2017), and they detected this fraction with 43.1±14.6 % in brushwood chulha cook firers. CHON was abundant in the three sources. This result was different from the findings that CHON species had a higher percentage in BB smoke and were not abundant in CC smoke (Song et al., 2018). The high fraction of CHON in CC aerosols could be due to that the N-containing compounds in the BB smoke PM2.5 coming from the nitrogen content in the fuels (Coggon et al., 2016), and the contents in coal fuels were comparable to biomass fuels (See Tables S1 and S2). However, S-containing compounds were more abundant in CC aerosols (9.2 %–21 % for CHOS and 13 %–20 % for CHONS, respectively) and tunnel aerosol (24 % for CHOS and 16 % for CHONS, respectively) than those in BB aerosols (2.0 %–5.6 % for CHOS and 0.62 %–3.7 % for CHONS, respectively) and vehicle exhaust particle (7.5 % for CHOS and 0.25 % for CHONS, respectively), consistent with the previous studies (Song et al., 2018; Wang et al., 2017). ESI was more efficient in ionizing S-containing compounds and most of them were selectively ionized by ESI-, suggesting that they were polar species such as organosulfates (Lin et al., 2018). Our study reported that S-containing compounds in WSOC were associated with CC emissions by combining with carbon isotope data (14C) (Mo et al., 2018). Furthermore, the relative abundances of group species in CC aerosols and tunnel aerosol were similar to those of water extracts in the hazy day (Jiang et al., 2016), indicating both sources could be important contributors of haze. However, differences between tunnel aerosol and vehicle exhaust particles were observed, indicating S-containing compounds in the tunnel aerosol were more secondary formation.

Van Krevelen (VK) diagrams are a useful tool that provides a visual graphic display of compound distribution, and to some extent, are used to qualitatively identify different composition domains in organic mixtures (Song et al., 2018; Lv et al., 2016; Smith et al., 2009). In this study, each source showed similar VK patterns. As shown in Fig. S9, Musa and Hevea burning had a VK diagram similar to that of WSOC in straw burning and fog water (Schmitt-Kopplin et al., 2010; Mazzoleni et al., 2010). S-containing compounds in tunnel aerosol with high O/C and H/C ratios were similar to the aerosol-derived WSOC in New York and Virginia (Wozniak et al., 2008). Six dominate domains were identified in WSOC, including lignins, carbohydrates, tannins, proteins, condensed aromatic, and unsaturated hydrocarbons. In addition, results showed that compounds observed in CC aerosols had lower H/C and O/C ratios than those in BB aerosols and vehicle emissions, indicating a higher unsaturated degree and lower oxidation level. There were compounds outside the specified regions, which had a high H/C ratio (≥2.2) and DBE of 0, corresponding to saturated oxygenated species, and there could be some long-chain polyalcohols (Lin et al., 2012a).

The mass spectra of MSOC exhibited differences from WSOC (Fig. S10), especially in BB aerosols and vehicle emissions that exhibited larger m/z values in the range of 350–600. The detected formulas in MSOC were much lower than those in WSOC, with the total number of 4502, 3628, and 1069 for BB, CC, and vehicle emission aerosols, respectively (Table S5). The reason could be that ESI- can efficiently ionize the polar compounds, and the methanol extracts after water-extracted may contain more moderate- and low-polarity compounds that were not easily ionized. The average molecular formulas were C26.9H46.2O4.27N0.24S0.02, C23.3H34.9O5.18N0.20S0.02, C18.2H19.2O4.24N0.92S0.03, C22.4H20.7O3.01N0.38S0.05, C22.6H44.1O5.70N0.74S0.11, and C25.2H48.5O4.86N0.58S0.08 of MSOC in the six aerosol samples, respectively, showing higher C and H content than their corresponding formulas of WSOC but a decreasing trend in O content.

CHO and CHON were the main components in MSOC, accounting for about 90 % of the total intensities (CHO plus CHON). CHO was the most abundant category observed in BB aerosols (78 %–80 %). The elemental compositions observed in CC aerosols were different between bituminous coal and anthracite combustion. The abundance of CHON in anthracite combustion was higher (73 %), while the CHO in bituminous combustion was higher (60 %), which was consistent with the corresponding WSOC. It might be due to a higher N content and lower O content of anthracite than that of bituminous coal (see Table S2). However, CHON in BB aerosols (18 %–20 %) exhibited lower abundances than those in CC aerosols and vehicle emissions. S-containing compounds were not abundant in MSOC. It may be due to the combination of an S element and an O atom showing higher polarity.

Figure S11 showed the VK diagram of MSOC in the six aerosol samples. More formulas in BB aerosols exhibited two distinct groups with H/C of 1.4–2.2 and 0.6–1.4 vs. O/C of 0.1–0.5 in three domains (lignins, proteins, and lipids). Compounds in CC aerosols with lower H/C and O/C ratios were dominant in the domains of lignins and condensed aromatic, especially in bituminous CC aerosol with more unsaturated hydrocarbon. Tunnel aerosol showed a wide range of O/C in S-containing compounds and a wide range of H/C in non-S-containing compounds. In contrast, compounds in vehicle exhaust particles had a wide range of H/C but a narrow O/C ratio. The VK diagram with fewer S-containing compounds in vehicle exhaust particle showed a similar characteristic to the distribution of non-S-containing compounds in tunnel aerosol, indicating the difference was mainly due to the S-containing compounds.

Tables 2 and S5 presented the relative abundance of weighted molecular weight (MWw), double-bond equivalence (DBEw), and modified aromaticity index (AImod,w) of WSOC and MSOC, respectively. DBE was used as a measure of the unsaturated level in a molecule, and AImod could be used to estimate the fraction of aromatic and condensed aromatic structures (Song et al., 2018; Lv et al., 2016; Koch and Dittmar, 2006). BB aerosols had higher MWw values than CC and vehicle emissions in WSOC. Further, higher DBEw and AImod,w values were observed in CC aerosols than the others. MSOC had higher MWw but lower AImod values than the corresponding WSOC. Furthermore, CHO and CHON compounds had higher DBEw and AImod,w values than S-containing substances, consistent with the earlier results (Lin et al., 2012a, b).

Figure S12 showed the fraction of AImod values of WSOC in the six aerosol samples, where the formulas were classified according to their AImod (aliphatic – AImod=0, olefinic – 0<AImod ≤0.5, and aromatic – AImod>0.5). The results illustrated that the fraction of the aromatic structure in non-S-containing compounds was higher than those in S-containing compounds. CC aerosols had a higher aromatic fraction than BB aerosols and vehicle emissions, especially in CHO and CHON (up to 89 % of total ion intensities). In BB aerosols, the non-S-containing compounds had a high fraction of the olefinic structure, followed by the aromatic structure, but the S-containing compounds had a higher aliphatic and olefinic structure than the aromatic structure. A higher fraction of the aliphatic structure in vehicle emissions was observed in the S-containing compounds (especially in tunnel aerosol; exceeding 81 %). These aliphatic S-containing compounds might form by the precursors (long-chain alkanes) from vehicle emissions (Tao et al., 2014), which had higher H/C and lower DBE values (see Table 2). However, the previous study showed that AI must be regarded as the most conservative approach and may result in an underestimate of the aromatic structures (Koch and Dittmar, 2006), which was observed in Beijing aerosols (Mo et al., 2018). Although AImod identified more compounds as aromatic and condensed aromatic components than AI, the AImod may introduce uncertainties for individual molecules, which was demonstrated by Koch and Dittmar (2006).

For MSOC, the aromatic structure fractions in non-S-containing compounds were higher than those in S-containing compounds, and the aromatic structure fractions in CC aerosols were higher than those in BB aerosols and vehicle emissions (Fig. S13), which was consistent with WSOC. Furthermore, we found that the fraction of the aliphatic structure in MSOC was higher than that in WSOC, indicating more fat-like compounds.

In the above statements, we discussed the light absorption and fluorescence properties from aerosols in the three different sources. The light absorption capacity of WSOC and MSOC was essential to assess the evolution of BrC, and fluorescence spectra were sensitive to different sources and could help for the source apportionment of BrC. In addition, we evaluated the molecular compositions of the three sources. Therefore, understanding the factors affecting the optical properties of BrC is important. It was reported that the MAE value in the BB experiments depended largely on burning conditions (Chen and Bond., 2010), and in the CC experiments it depended on coal maturity (Li et al., 2018). Chen et al. (2017b) illustrated that the higher light absorption capacity was associated with the low- and medium-polarity fractions that contained aromatic and polar functional groups (O or both O and N atoms). Sources play an important role in light absorption capacity, consistent with our current study. The MAE365 values of WSOC in highly BB-impacted areas were 2 times higher than in lowly BB-impacted areas in the southeastern United States (Hecobian et al., 2010). Atmospheric aging has a significant effect on the light absorption capacity of BrC (Li et al., 2019), but the mechanism involved is very complex. The response of the light absorption capacity of different types of BrC to aging is highly variable, and enhancement or reduction in the light absorption capacity of BrC is possible (Li et al., 2019). These results indicated that light absorption capacity might be affected by various factors. In this study, the higher MAE365 values were observed in BB and CC aerosols than vehicle emissions, and the chemical structures and unsaturation degree of different sources were discussed. Next, we further discussed the relationship between optical properties and chemical structures below.

Before discussing their relationship, we firstly determined the compounds that were potentially able to absorb light radiation based on the above statement to reduce the influence of non-absorbing substances (Lin et al., 2018). Mo et al. (2018) reported that MAE365 of HULIS in aerosols was affected by oxidation level and unsaturation degree. In this study, the MAE365 value had no significant correlation with O/C, indicating that light absorption capacity does not appear to be affected by their oxidized properties in the source emission aerosols. Instead of O/C, the MAE365 value had a significant positive correlation with the average DBE and MW, respectively (Fig. 7), suggesting the unsaturation level and MW played a vital role in the light absorption capacity of source samples. Field experiments indicated that the majority of absorption was the larger molecules (>500 Da) (Di Lorenzo et al., 2017). It is crucial to understand the relationship between light absorption of the source samples and their molecular compositions due to the compounds in fresh emissions that may undergo a secondary process and introduce more uncertainty for their optical properties.

Fluorescence spectra can provide more information than UV–vis spectra. A red shift in the excitation–emission maximum could indicate increased aromaticity and higher molecular weight (Ghidotti et al., 2017). Field observations have demonstrated that chromophore components were associated with chemical structures (Chen et al., 2016a, b; Stubbins et al., 2014). Chen et al. (2016b) illustrated that the fluorescent components of HULIS-1 and HULIS-2 were correlated positively with CO+ and CO2+ and CxHy+ and CxHyO1+ groups ions, respectively, using the correlation analysis of the relative intensities of ion groups in high-resolution aerosol mass spectrometers (HR-AMSs) and the relative content of fluorescence components. In another study, Chen et al. (2016a) demonstrated that fluorescent components had strong links with chemical groups in the Fourier transform infrared (FT-IR) spectra, including the oxygenated functional groups (nonacidic carbonyl C = O and carboxylic COOH groups), aliphatic C-H group, amine C-NH2, and alcohol C-OH groups. The chromophores are sensitive to sources, and it is very important to understand the molecular composition of chromophores for the classification and source apportionment of atmospheric BrC. However, the ESI- cannot ionize the most typical BrC chromophores such as O-heterocyclic PAHs (O-PAHs) or N-heterocyclic PAHs (N-PAHs) (Lin et al., 2018), which was not enough to discuss the relationship between the fluorescence spectra and molecular composition. The combination of atmospheric pressure photoionization (APPI+ and APPI-) and ESI (+ and -) may provide more ionizable compounds, but these techniques were not within the scope of our study.