The PM2.5 samples were collected on the campus of Guangzhou Institute of Geochemistry, Chinese Academy of Sciences, Guangzhou, China (23.14∘ N, 113.35∘ E), which is an academic and residential region. Traffic emissions and residential activities are the potential pollution sources in the sampling area. The 24 h PM2.5 sampling was conducted using a high-volume sampler (Tianhong Intelligent Instrument Plant, Wuhan, China, with a flow rate of 1.0 m3min-1) during 7 to 30 January 2018, and a total of 24 samples were collected on the prebaked quartz filters (20.3cm×25.4 cm, Whatman, Maidstone, UK). Field blank samples were collected by keeping a blank filter in the sampler without pumping air. Before sampling, the filters were wrapped in aluminum foil and prebaked at 450 ∘C for 6 h to remove carbonaceous impurities. Before and after sampling, the filters were weighed at 25 ∘C and 50 % RH on a microbalance (Sartorius Model BP210D). The PM2.5 concentrations were determined by weighing the filters before and after collection. Finally, all filter samples were stored in a refrigerator at -20 ∘C until analysis. The mass accuracy achieved was <2 % based on triplicate analyses of filter samples. Meteorological data (http://www.wunderground.com/history/airport/ZGGG, last access: 10 March 2020), including wind speed; temperature; relative humidity; and concentrations of SO2, O3, and NO2, for the sampling days are presented in Fig. 1 and Table S1 in the Supplement.

HULIS were isolated using a water extraction and solid-phase extraction (SPE) procedure as described previously (Zou et al., 2020). This method has been used in most previous studies because of its easy operation and high reliability and reproducibility and low limit of detection (Fan et al., 2012); therefore, it was also used in this study. Briefly, portions of the PM2.5 samples (100 cm2) were ultrasonically extracted with 50 mL of ultrapure water for 30 min. The extracts were filtered through a 0.22 µm PTFE syringe filter to remove the suspended insoluble particles. About 50 mL of water extracts was obtained from each sample, of which 20 mL was used for the isolation and analysis of HULIS, 20 mL for analysis of water-soluble organic carbon (WSOC), and the remaining extracts for the analysis of inorganic ions. Then, the 20 mL water extracts were adjusted to a pH of 2 with HCl and loaded on a preconditioned SPE cartridge (Oasis HLB, 200 mg, 6 mL, Waters, USA). The hydrophilic fraction (i.e., inorganic ions, high-polarity organic acids, etc.) was removed with ultrapure water, whereas the relatively hydrophobic HULIS fraction was retained and eluted with 2 % (v/v) ammonia/methanol. Finally, HULIS solution was evaporated to dryness with a gentle N2 stream and redissolved with ultrapure water for the analysis.

It is noted that the HULIS here is the hydrophobic portion of water-soluble organic matter, which can be isolated with different types of SPE columns (e.g., HLB, C-18, DEAE, XAD-8, and PPL) (Fan et al., 2012, 2013; Lin et al., 2012; Zou et al., 2020; Jiang et al., 2020). Although each resin type has its special chemical properties, the hydrophobic HULIS isolated with different sorbents were similar in chemical, molecular properties based on previous studies (Fan et al., 2012, 2013; Zou et al., 2020). Therefore, for better comparison with other studies, the hydrophobic fractions isolated by SPE methods were all termed as HULIS in the present paper.

The absorption spectra of the WSOC and HULIS fractions were measured by a UV–Vis spectrophotometer (UV-2600, Shimadzu) between 200 and 700 nm. Each spectrum was corrected for the filter blanks. The light absorption coefficients, absorption Ångström exponent (AAE), and mass absorption efficiency (MAEλ) were calculated, and the detailed methods for this are presented in the Supplement.

For FT-ICR MS analysis, the HULIS samples were isolated from PM2.5 collected during four periods: before haze days (clean-I days, 7–12 January), haze bloom days (haze-I days, 13–18 January), haze decay days (haze-II days, 19–24 January), and after haze days (clean-II days, 25–30 January). A filter punch (18 cm in diameter) was taken from every sample, and all six samples in each period were combined for the isolation of HULIS fractions. The obtained HULIS samples were measured with an ESI FT-ICR MS instrument (Bruker Daltonics GmbH, Bremen, Germany) equipped with a 9.4 T refrigerated actively shielded superconducting magnet. The system was operated in both ESI- and ESI+ modes. The scan range was set to m/z from 100 to 1000, with a typical mass-resolving power >450000 at m/z 319 with <0.2 ppm absolute mass error. The mass spectra were calibrated externally with arginine clusters and internally recalibrated with typical O5-class species peaks in DataAnalysis 4.4 (Bruker Daltonics). Due to the inherent differences in the ionization mechanisms between ESI- and ESI+ modes, the data detected by the two ionization modes can provide complementary information on the molecular composition of atmospheric HULIS (Lin et al., 2012, 2018). The details of data analysis are provided in the Supplement.

The amounts of organic carbon (OC) and elemental carbon (EC) were determined by an OC–EC analyzer (Sunset Laboratory Inc., USA) (Mo et al., 2018). The concentrations of WSOC and HULIS were determined by a total organic carbon (TOC) analyzer (Shimadzu TOC-VCPH, Kyoto, Japan). The water-soluble inorganic species (NO3-, SO42-, Cl-, NH4+, K+, Na+, Ca2+, Mg2+) were measured with a Dionex ICS-900 ion chromatography system (Thermo Fisher Scientific, USA) as described previously (Huang et al., 2018). The concentrations of Lev were analyzed with gas chromatography–MS after derivatization with N,O-bis(trimethylsilyl)trifluoroacetamide and pyridine at 70 ∘C for 3 h (Huang et al., 2018). Detailed information regarding these measurements is provided in the Supplement.

Figure 1 shows the meteorological conditions, PM2.5 concentration, and concentrations of major chemical constituents, including carbon fractions and water-soluble inorganic ions in PM2.5 samples obtained during a haze bloom-decay process. Based on the variation in PM2.5 concentration, these samples were categorized into four groups: clean-I days (before haze, 14–24 µgm-3), haze-I days (haze bloom, 45–114 µgm-3), haze-II days (haze decay, 58–115 µgm-3), and clean-II days (after haze, 9–35 µgm-3). As indicated in Table S1 and Fig. 1, the PM2.5 concentrations increased from 18 ± 3.3 µgm-3 on clean-I days to 82 ± 26 and 84 ± 22 µgm-3 on haze-I and haze-II days, respectively, and then decreased to 21 ± 10 µgm-3 on clean-II days. This finding obviously indicated that the average PM2.5 concentrations during the examined haze episode are higher than the second-grade national ambient air quality standard in China (75 µgm-3, 24 h), whereas those during clean days are lower than the first-grade national ambient air quality standard in China (35 µgm-3, 24 h). However, the average PM2.5 concentrations during the haze event are lower than those in the cities in winter haze, including Shenyang (108 µgm-3) (Zhang et al., 2020) and Nanjing (123 ± 28.5 µgm-3) (Li et al., 2020), Beijing (158 µgm-3), and Xi'an (345 µgm-3) (Zhang et al., 2018).

As shown in Table S1, the average concentrations of OC and EC were 2.2–15 and 0.36–2.7 µgCm-3 in the four stages, respectively, implying that the distinct changes in OC and EC were higher during haze episodes than on clear days. During the entire study period, WSOC concentration ranged from 0.5 to 12.5 µgCm-3 (4.3 ± 1.2 µgCm-3), which contributed to 53 %–57 % of OC in PM2.5. The HULIS concentration noted in the present study ranged from 0.15 to 6.1 µgCm-3 (2.2 ± 1.9 µgCm-3), which was comparable to those concentrations observed in the PRD region, such as Hong Kong (2.38 ± 1.62 µgCm-3) (Ma et al., 2019), Guangzhou (2.4 ± 1.6 µgCm-3) (Fan et al., 2016), and Heshan (2.08 ± 1.16 µgCm-3) (Jiang et al., 2020), but lower than those in northern cities of China, such as Xi'an (12.4 ± 6.5 µgCm-3) (Huang et al., 2020), Beijing (3.79 ± 3.03 µgCm-3) (Mo et al., 2018), and Lanzhou (4.7 µgCm-3) (Tan et al., 2016). As shown in Fig. 1, HULIS also exhibited obvious variations during the entire sampling period. The average HULIS concentration was 0.46 ± 0.22 µgCm-3 on clean-I days, which sharply increased to 4.5 ± 1.2 µgCm-3 on haze-I days, then decreased to 3.1 ± 1.2 µgCm-3 on haze-II days, and rapidly declined to 0.75 ± 0.52 µgCm-3 on clean-II days. This result was consistent with the changing trend of WSOC, OC, and EC. In addition, the HULIS / WSOC ratios were about 0.50 ± 0.13 in the PM2.5 samples, which are in broad agreement with other studies showing that HULIS comprise the major fraction of WSOC (Fan et al., 2016; Ma et al., 2019; Jiang et al., 2020).

As illustrated in Fig. 1, obvious variations in chemical compositions were also observed in these PM2.5 samples. Secondary inorganic aerosols (SIA) (i.e., SO42-, NO3-, and NH4+), OC, and EC exhibited a similar variation during the entire study period, and their contents sharply increased from 10 January on clean-I days to 13–18 January on haze-I days, then slowly decreased on haze-II days, and finally reached lower levels on clean-II days. It must be noted that the increasing rate of EC was similar to that of SIA on haze-I days, indicating that direct emissions and atmospheric reactions may play similar roles in PM2.5 increase during this haze bloom period. As indicated in Fig. 1f, the highest values of NO3-/SO42- were observed on haze-I days, implying the important influence of traffic exhaust emissions in the haze bloom period (Mo et al., 2018). In addition, the high NO2 and O3 concentrations and the stable meteorological conditions with high temperature also led to the outburst of fine particulate pollution in this period. During haze-II days, the SIA and organic matter (OM) contents in PM2.5 slowly decreased, whereas the concentrations of Na+, Cl-, and unidentified materials in PM2.5 increased (Fig. 1e and h), suggesting that local contribution weakened and regional contribution via sea salt became more important (Jiang et al., 2021). This phenomenon was also observed to be consistent with the changes in the pollutant sources transported by air masses. As indicated in Fig. S1, the PM2.5 samples on haze-II days included some contributors transported from the coastal area of eastern Guangdong Province and Fujian Province, and the PM2.5 is likely to be enriched with sea salt materials and mineral dusts.

The light absorption properties of WSOC and HULIS (Fig. 1d, i, and j and Table S2) exhibited obvious temporal variations during the sampling period. The AAE values of WSOC and HULIS ranged from 4.1 to 6.4 and 5.6 to 6.6, respectively. The AAE values for HULIS were obviously higher than those for WSOC in the same sample (Fig. 1i), indicating that light absorption of HULIS is more wavelength-dependent than that of WSOC. Similar results have also been observed in previous studies (Park et al., 2018; Jiang et al., 2020; Cao et al., 2021). These differences could be attributed to the differences in chemical composition of chromophores in WSOC and HULIS. As shown in Table S2, the E250/E365 values of HULIS (5.3–5.6) are all higher than those of WSOC (4.4–5.1), suggesting that the light-absorbing species in HULIS may have lower aromaticity and/or lower molecular weight than those in WSOC (Chen et al., 2016; Li and Hur, 2017). Therefore, the HULIS fractions exhibit relatively high absorption at UV and short visible wavelengths and relatively low absorption at long visible wavelengths, which results in relatively high AAE values. Moreover, the AAE values of HULIS did not present significant variation during the entire haze process, which could be related to the evolution of HULIS chromophores in different stages (Jiang et al., 2020; Huang et al., 2018; Deng et al., 2022). At first, the enhanced oxidation of aromatic species on haze days could lead to the bleaching or degradation of chromophores (a detailed explanation is provided in Sect. 3.3) and thus a lower wavelength dependence (Forrister et al., 2015; Zhan et al., 2022). In contrast, the outburst of secondary organic aerosols and the photolysis of organic aerosols on haze days both tended toward higher AAE values (Saleh et al., 2013; Dasari et al., 2019). Consequently, the different trends in AAE were counterbalanced during the haze days, which resulted in no significant AAE variations being observed for HULIS in the entire sampling process. This is also consistent with the trends of the E250/E365 ratios of HULIS in the four stages (Table S2).

Light absorption at 365 nm (Abs365) for WSOC and HULIS was 2.5 ± 2.0 and 1.8 ± 1.6 Mm-1, respectively (Table S2). HULIS contributed to about 72 % of light absorption coefficients by WSOC, implying that they enriched the major light-absorbing components in WSOC. As shown in Fig. 1d, the Abs365 values for HULIS presented obvious temporal variations. The Abs365,HULIS value was 0.55 ± 0.06 Mm-1 on clean-I days, which first increased to 3.4 ± 1.5 Mm-1 on haze-I days, then slowly decreased to 2.6 ± 0.85 Mm-1 on haze-II days, and finally rapidly declined to 0.64 ± 0.32 Mm-1 on clean-II days. This result was similar to the variations in the mass concentration of HULIS. Furthermore, the Abs365 values for HULIS in Guangzhou were found to be higher than those observed in the southeastern Tibetan Plateau (0.38–1.0 Mm-1) (Zhu et al., 2018) but obviously lower than those in Xi'an (7.6–36 Mm-1) (Shen et al., 2017) and Beijing (3.7–10.1 Mm-1) (Du et al., 2014).

In general, MAE365 values can be used to assess the light absorption capacity of target organic compounds (M. Li et al., 2019). As shown in Fig. 1j and Table S2, the average MAE365 value for WSOC was 1.0 ± 0.21 m2g-1C (0.68–1.3 m2g-1C), nearly the same as 1.1 ± 0.27 m2g-1C (0.77–1.8 m2g-1C) for HULIS, during the entire sampling period. Moreover, the MAE365 values for HULIS measured in the present study were noted to have dropped to the ranges of those determined in Beijing (1.43 ± 0.33 m2g-1C) (Mo et al., 2018), Xi'an (0.91–1.85 m2g-1C) (Yuan et al., 2021), and Hong Kong (1.84 ± 0.77 m2g-1C) (Ma et al., 2019). The average MAE365 values for HULIS exhibited some temporal variations. The MAE365 values for HULIS were 0.91 ± 0.03 and 0.95 ± 0.11 m2g-1C on haze-I and haze-II days, respectively, which were lower than those (1.3 ± 0.22 and 1.3 ± 0.27 m2g-1C, respectively) observed on clean-I and clean-II days, suggesting that HULIS have a relatively weaker light absorption capability on haze days. This finding is consistent with the results reported by Zhang et al. (2017), who found that the MAE365 values in the heating or non-heating seasons during hazy days were lower than those on clean days. These differences in MAE365 values may potentially contribute to the enhanced oxidation reaction that was derived by the increased O3 levels and high temperature and relative humidity (RH) during haze days (Fig. 1). This oxidation process would lead the chromophores containing C=C unsaturated bonds to be severely degraded (J. Wang et al., 2017; Zhang et al., 2017). Besides, an increase in additional sources for HULIS in the study area, such as weaker or non-light-absorbing compounds formed by atmospheric oxidation, could also result in weaker light absorption of HULIS during the haze episode (Liu et al., 2018).

For an in-depth understanding of the variation in HULIS at the molecular level during the haze process, the four HULIS samples collected in different stages of the haze process were analyzed by ESI FT-ICR MS in both negative and positive modes. As shown in Fig. 2, thousands of peaks were detected in the mass range between m/z 100 and m/z 700, with the high-intensity ions noted within m/z 150–400. It is obvious that some organic compounds with stronger arbitrary abundance were labeled, and their formulas, double-bond equivalent (DBE), modified aromaticity index (AImod), and potential sources are listed in Table S3. Compounds a (C7H7NO3) and b (C8H6O4) both have high DBE values, which might be assigned to aromatics such as methylnitrophenol and phthalic acid, whereas compound d (C8H18O4S) with a low DBE value and high O/S ratio was probably aliphatic organosulfate. According to previous studies, these organic molecules might be derived from BB and diesel fuel, and thereby these results suggest that both BB and vehicular emissions are important sources of BrC in ambient aerosols (Mohr et al., 2013; Riva et al., 2015; Blair et al., 2017). Furthermore, compound e (C10H17NO7S) and compound f (C10H18N2O11S) in Table S3 were found to be identical to the oxidation products of monoterpenes, suggesting that biogenic sources could contribute to the formation of HULIS (Surratt et al., 2008; Wang et al., 2019). Thus, HULIS could be affected by multiple sources during the haze process, possibly including BB, biogenic sources, and anthropogenic emissions.

The identified formulas could be divided into seven compound categories, namely, CHO-, CHON-, CHOS-, and CHONS- detected in ESI- mode and CHO+, CHN+, and CHON+ detected in ESI+ mode. As illustrated in Fig. 2, the CHO compounds were the most abundant group in all the HULIS, accounting for 43 %–50 % and 51 %–57 % of the overall compounds detected in the ESI- and ESI+ modes, respectively. It must be noted that relatively low contents of CHO- were detected during the haze episode (haze-I and haze-II days) and of CHO+ molecules in haze-I HULIS. The CHON compounds were the second most abundant group in all the HULIS. As shown in Fig. 2, the relative content of CHON- was 23 % on clean-I days, which slightly increased to 24 %–25 % in the haze episode and then decreased to 23 % on clean-II days. In contrast, the relative content of CHON+ compounds was 41 % on clean-I days, which increased to 45 % on haze-I days and then fell to 42 % on haze-II days and 41 % on clean-II days. Both CHOS- and CHONS- compounds were identified in all the four HULIS, accounting for 19 %–22 % and 8 %–11 % of the total identified compounds, respectively. The CHN+ compounds were the least abundant (1.3 %–3.6 %) in the four HULIS samples and were relatively high during the haze episode, especially on haze-I days.

Tables S4 and S5 show the relative-abundance-weighted elemental ratios, molecular weight (MW), DBE, AImod, and carbon oxidation state (OSC) for the identified compounds in HULIS. The MWw values for HULIS determined in the ESI- mode on haze-I and haze-II days were 302 and 283, respectively, which were higher than those on clean-I and clean-II days (266 and 264, respectively). Similar variation was also observed for MWw for HULIS detected in ESI+ mode (Table S5). These results clearly indicated that more higher-MW compounds constituted HULIS obtained during the haze episode. Furthermore, the molecular properties of HULIS in different stages of the haze process also exhibited some observable differences. As shown in Table S4, the HULIS samples in the haze episode detected by the ESI- mode presented relatively low AImod,w values and relatively high O/Cw, O/Nw, and O/Sw ratios compared to those on clean days, indicating that haze HULIS exhibited relatively low aromaticity and a higher oxidation degree than clean HULIS. These differences can be attributed to the enhanced oxidation degradation of aromatic compounds (e.g., phenols, nitroaromatic compounds, and polycyclic aromatic hydrocarbons (PAHs)) during the haze process. In addition, an increased contribution from traffic emissions and secondary reactions of biogenic VOCs (bio-VOCs) also decreased the aromaticity and increased the oxidation degree of HULIS (Liu et al., 2016; Tang et al., 2020). These changes in HULIS compounds led to the decrease in their MAE365 values during the haze episode, as described above (Zhong and Jang, 2014; Song et al., 2019).

As described earlier, the light absorption properties of HULIS exhibited obvious variation during the haze bloom-decay process. The average Abs365 value for HULIS was 0.55 ± 0.06 Mm-1 on clean-I days, which first increased to 3.4 ± 1.5 Mm-1 on haze-I days, then slowly decreased to 2.6 ± 0.85 Mm-1 on haze-II days, and finally rapidly declined to 0.64 ± 0.32 Mm-1 on clean-II days. In general, the light absorption of HULIS can be related to their chemical and molecular properties that are influenced by factors such as sources, secondary formation, and the aging process. The results of principal component analysis (PCA) obviously showed a positive loading for principal component 1 (PC1), and the Abs365 values for HULIS were clustered with EC, Kbb+, Lev, NH4+, and NO3- (Fig. 5). These results suggested that BB and other sources such as new particle formation could contribute to light absorption of HULIS (An et al., 2019; Song et al., 2019). Similarly, the findings of Pearson correlation coefficient analysis revealed that the Abs365 values for HULIS exhibited significant positive correlations with Kbb+ (r=0.728, p<0.01) and Lev (r=0.800, p<0.01) (Table S8). As Lev and Kbb+ are generally considered tracers derived from BB, these results suggested the significant contribution of BB to light absorption of HULIS. This observation was also supported by the abundance of BBOA compounds detected in all the four HULIS samples (Table S6). The significant positive relationships between the Abs365 values for HULIS and secondary ions (i.e., NO3- (r=0.702, p<0.01), SO42- (r=0.554, p<0.05), and NH4+ (r=0.899, p<0.01)) indicated the important impact of secondary formation on the light absorption of HULIS. Besides, the Abs365 values for HULIS were also strongly correlated with NO2, O3, and NO2, which confirmed the important impact of atmospheric oxidation reactions on the light absorption of HULIS.

It must be noted that MAE365 is a key parameter signifying the light absorption ability of HULIS. As listed in Table S2, the MAE365 values for HULIS varied in different stages and were lower on haze days owing to the variation in the chemical and molecular composition of HULIS during the haze bloom-decay process. Furthermore, the AImod values for HULIS varied in different stages (Tables S4) and were relatively low on haze days, indicating that haze HULIS have a comparatively low degree of conjugation or aromaticity. This finding suggested that the HULIS compounds may undergo higher oxidation during the haze episode, causing a decline in chromophores and reduction in the light absorption capacity of HULIS (Lin et al., 2017). Besides, the accumulated contribution of organic compounds from vehicle emission and secondary chemical reactions of bio-VOCs may also dilute light-absorbing compounds in haze HULIS (Tang et al., 2020; Liu et al., 2016).

Lin et al. (2018) reported that potential light-absorbing chromophores can be determined in the region between DBE=0.5×C (linear conjugated polyenes CxHyC2) and DBE=0.9×C (fullerene-like hydrocarbons). In the present study, most of the high-intensity CHON, CHO, and CHN compounds with high AI values were clustered in a potential BrC chromophore region (Figs. S3 and S4), which mainly comprised CHON (46 %–50 % in ESI- mode and 56 %–62 % in ESI+ mode) and CHO (44 %–48 % in ESI- mode and 29 %–38 % in ESI+ mode) compounds (Table 1). Although the contribution of CHN+ compounds to BrC was relatively low, the content of potential chromophores among the total CHN+ compounds was higher than those in CHON+ and CHO+ compounds. Therefore, these three groups of light-absorbing compounds (i.e., CHON+, CHN+, and CHO+ compounds) were further examined. As shown in Table 1, the IntC/IntBrC values of CHO- (i.e., content of CHO- chromophores in the total chromophores) decreased from 48 % to 44 %, whereas the IntC/IntBrC values of CHON- increased from 46 % to 50 % during the haze bloom process. These findings indicated that more NOC chromophores were formed during this stage in which higher NO2 concentration may be preferred for the formation of N-containing chromophores such as nitrophenols. However, it must be noted that the proportions of both CHO- and CHON- chromophores among the total identified compounds decreased from clean-I to haze-I days, suggesting the occurrence of a stronger photo-bleaching process during the haze bloom stage (Zeng et al., 2020). Likewise, both CHO+ and CHON+ compounds presented similar variation during the entire study period. In addition, the CHN+ compounds also exhibited higher IntC/IntBrC values during the haze bloom process, suggesting the accumulated contribution from the local combustion process. Furthermore, the proportion of CHON+ chromophores in the total CHON+ compounds increased with the decreasing content of CHN+ chromophores, possibly implying that some aromatic CHN compounds were transformed to CHON+ compounds during the aging process.