Size-segregated particle samples were collected by a six-stage micro-orifice uniform deposit impactor (MOUDI) with respective aerodynamic cut-point diameters of 0.26, 0.44, 0.77, 1.4, 2.5, and 10 µm. Sample collection started at 08:00 LT (UTC-8) and concluded at 07:00 LT the next day, leaving 1 h for operation. All samples were collected on quartz filters (Whatman), which were prebaked for 5 h (500 ∘C) before sample collection, and were wrapped in aluminum foil and stored at -20 ∘C.

A total of 20 sets of six-stage size-segregated aerosol samples were collected at a rural site in Huairou District, Beijing, from 14 November to 30 December 2016 and from 30 June to 8 September 2017. The sample collection days were randomly selected, and samples were later categorized according to the degree of air pollution. Winter sampling days covered six levels of air quality from excellent to severe pollution, whereas summer sampling days only covered good and moderate air quality. The air-quality-index-weighted 72 h backward trajectories during the sampling period are exhibited in Fig. 1.

Organic and elemental carbon (OC and EC, respectively) were determined using a thermal–optical carbon analyzer (DRI), and the IMPROVE (Interagency Monitoring of Protected Visual Environments) thermal evolution protocol was adapted. Detailed information can be found in earlier studies (Cheng et al., 2009; Tan et al., 2016). The detection limit of OC and EC was 1.0 µgm-3, as quantified by a filter and filter blank. Quality assurance (QA) and quality control (QC) were performed by replicate analyses every 10 samples, and the repeatability was better than 5 %.

A quarter of the filter sample was ultrasonically extracted twice with 5 mL ultrapure water each time and mixed up after extraction. The extracts were then filtered through a 0.22 µm membrane filter to remove impurities (Xiang et al., 2017). The measurement of WSOCs was performed by a TOC (total organic carbon) analyzer (multi N/C 3100, Analytic Jena AG, Germany).

The extraction procedures for water-soluble ions (WSINs) were similar to those for WSOCs, but a 0.22 µm Teflon filter was used to remove impurities. Ion chromatography (IC, Dionex ICS 900 and 1100) was used in the detection, with eight WSIN species analyzed (Cl-, NO3-, SO42-, NH4+, Na+, K+, Ca2+, and Mg2+). The recovery (90 %–110 %) and reproducibility (relative standard deviation of each ion lower than 5 %) of the ions were implemented as well.

The extraction procedures for samples subject to fluorescence and ultraviolet–visible (UV–Vis) sampling were the same as for WSOC detection. The excitation–emission spectra were obtained using a fluorescence spectrophotometer (F-7000, Hitachi, Japan), and UV–Vis spectra were obtained using an ultraviolet spectrophotometer (UV-2401PC, Shimadzu, Japan). Briefly, the wavelength ranges were 200–400 nm for excitation and 250–500 nm for emission with a 5 nm interval for fluorescence spectroscopy (Qin et al., 2018). UV–Vis was measured between 200 and 500 nm with a 5 nm interval. All EEM data in the present study are in Raman units (RU). The background signals, interfering signals (first- and second-order Rayleigh and Raman scatterings), and the inner-filter effects were removed by subtracting a blank EEM spectrum, replacing Rayleigh scatters with a band of missing values, and inserting a triangular set of zeros in the “emission far below excitation” area (Bahram et al., 2006). Data correction and standardization followed the procedures described in Xiao et al. (2016). As shown in Fig. 2, the EEM spectra were partitioned into five regions (Birdwell and Engel, 2010), and the fluorescence regional integration (FRI) method was applied to examine the fluorescence intensities of the accordant regions with respect to the total fluorescence intensity. The specific fluorescence intensity (SFI) was the fluorescence intensity divided by the WSOC concentrations.

Table 1 shows the size-segregated mass concentrations of WSINs, WSOCs, and OC as well as their ratios generated from the data collected at a rural site in Beijing during winter and summer. WSOCs showed a monomodal distribution in winter and a bimodal distribution in summer, with a dominant mode between 0.26 and 0.44 µm in both seasons and a small secondary mode in particles larger than 1 µm in summer, indicating that carbonaceous species were mainly rich in fine particles (Huang et al., 2020). Contemporary reports by other researchers have observed a bimodal distribution in WSOCs, with peaks located at 0.8 and 7 µm, respectively, in Shenzhen, China, and at 0.4–0.5 and 2–3 µm, respectively, in Gwangju, Korea (Yu et al., 2016; Huang et al., 2020).

The WSOC/OC ratios were 0.24–0.56 in winter and 0.16–0.31 in summer. These values were smaller than those previously reported for a polluted period in Beijing and in other cities in China (Tian et al., 2014; Wu et al., 2020). Earlier studies have suggested higher WSOC/OC ratios in summer than in winter (Xiang et al., 2017; Qin et al., 2018), which is in contrast with the results of the present study. Contrasting seasonal patterns in WSOC/OC ratios have also been reported between urban and rural sites in Georgia, USA (Zhang et al., 2012), which seems to support our results presented above. In this work, the WSOC/OC ratios were higher in particles with an aerodynamic diameter smaller than 1.4 µm compared with the coarse mode (PM2.5–10), which is accordant with findings previously reported for clear days in Beijing (Tian et al., 2016).

The size-segregated EEM spectra of winter and summer WSOCs are depicted in Fig. 2a and b, respectively, and their specific fluorescence intensities (SFIs) per unit WSOC are shown in Fig. 2c and d, respectively. The overall fluorescence peaks were mainly produced among regions II–V, and the peaks were peak A, peak T, and peak M, which could be categorized as humic-like, tyrosine-like, and oxygenated organic substances, respectively (Qin et al., 2018). The fluorophores first increased with increasing particle size, reached their highest intensities at particle sizes of 0.26–0.44 µm, and then decreased with increasing particle size in both seasons. Although the fluorescence peaks of WSOCs were mainly produced in similar regions between the two seasons, the relative abundance was different (more quantitative analysis below). The aggregated fluorescence spectra of all size-segregated samples resembled the spectra of total suspended particles (TSP) and PM2.5, shown in Fig. S1 in the Supplement, with some subtle nuance in border shape (Chen et al., 2016a; Qin et al., 2018).

The detailed characteristics of the fluorescence spectra could be found in the SFI spectra. The SFI showed evident differences between fine- and coarse-mode particles in both seasons. The spectra of coarse-mode WSOCs covered a wide range of natural sources (according to our unpublished research), whereas the spectra of fine particles widely overlapped with that of PM2.5 in Fig. S2 in the Supplement (matched with anthropogenic sources and secondary sources from our study), indicating that the sources of WSOCs affected their fluorescence properties. Moreover, the SFI spectra showed a clear blue shift within regions I to III with increasing particle size in winter as well as humble variations in summer.

Figure 3 shows the size distribution of WSOCs and their average fluorescence intensity (AFI) in the two seasons. The AFI showed a monomodal distribution with a peak at a particle size of between 0.26 and 0.44 µm in winter as well as a bimodal distribution in summer, which was accordant with the size distribution of WSOCs. AFI/WSOC ratios could represent the overall average fluorescence density of WSOCs (Xiao et al., 2016). The AFI/WSOC ratios ranged from 0.22 to 0.57 in winter and from 0.18 to 0.34 in summer. These values were higher than that in the industrial city of Lanzhou (Qin et al., 2018). Our unpublished research found that the AFI/WSOC ratios were lower than 0.2 for anthropogenic source samples.

The fluorescence regional integration (FRI) was calculated to quantify the relative strength of the fluorescence intensity in regions I to V, represented by FRI1 to FRI5, respectively (Fig. 4). FRI I and FRI II (protein-like species) increased with increasing particle size and peaked at the coarse mode in winter. FRI III and FRI V (HULIS) were mainly abundant in fine particles. FRI IV (microbial-related species) showed little variation in the particle size range from 0.26 to 2.5 µm but decreased with particle size from 2.5 to 10 µm. In summer, the sum of FRI I–FRI III increased with increasing particle size, peaked at 1.4 µm, and decreased with particle size from 1.4 to 10 µm. FRI IV showed the inverse: decreased with particle size in the range from 0.26 to 1.4 µm and increased with particle size in the size range from 1.4 to 10 µm. FRI V did not show a clear tendency, but these species displayed high abundance from 0.26 to 0.44 µm and 0.77 to 1.4 µm.

Inclusive information was stored in fluorescence spectra, with some regularities being extracted by dividing fluorescence intensities between wavelengths. The humification index (HIX) represents the humification degree or aromaticity of fluorescent organics. PeakT/PeakC, the ratio between tryptophan and humics, can reflect the biodegradability of organics. Some other fluorescence indices are listed in Table S1 in the Supplement. Figure 5 shows the size distribution of the HIX and PeakT/PeakC ratio. The HIX showed a monomodal distribution, peaking between 0.26 and 0.44 µm in summer and between 0.44 and 0.77 µm in winter, indicating that the aromaticity of size-segregated WSOCs initially increased and then decreased with increasing particle size. The PeakT/PeakC ratio increased gradually with increasing particle size in winter, whereas it first decreased in fine particles and then increased with particle size in summer. PeakT/PeakC peaked in the coarse mode in both seasons, indicating that fluorescent microbial-related species likely existed on large atmospheric particles. It was reported that biogenic oxygenated organics are more inclined to adhere to coarse-mode particles (Huang et al., 2020).

The Stokes shift (SS) is the energy loss due to fluorophore relaxation, which might be associated with the π-conjugated system and electron cloud density (Lakowicz, 2006). High SS values indicate greater energy loss due to relaxation in the excited states. Organic compounds with larger π-conjugation scales possibly exhibit a high fluorescence intensity in the high SS region (Xiao et al., 2020). Xiao et al. (2019) found that a SS near 1.2 µm-1 is an important border for hydrophobic and hydrophilic components. Hydrophobic fractions tend to have higher intensity in SS > 1.2, possibly as a result of the large scale of the π-conjugated system. In contrast, hydrophilic fractions usually have ionogenic groups bond with fluorescent aromatics, reducing π-conjugated systems and leading to high fluorescence intensities existing on both sides of a SS of 1.2. Note that the same research also reported that hydrophobic fractions tended to present fluorescence peaks at SS > 1 (Xiao et al., 2016). Thus, the ratios of fluorescence intensity for high SS (> 1.1) values are calculated as follows: 5ηSS>1.1=∑Ex∑EmI|SS>1.1∑Ex∑EmI.

The harmonic mean of the excitation and emission wavelengths (WH) reflects the average energy level of the excited states. In a large π-conjugated system, the electron in the ground state needs relatively low excitation energy to jump to the excited state (Berberan-Santos and Valeur, 2012). The ratios of the fluorescence intensity in the low-energy state (WH > 320) are calculated as follows: 6ηWH>320=∑Ex∑EmI|WH>320∑Ex∑EmI.

The size-segregated SS, average SS, and ηSS>1.1 are shown in Figs S4 and S5 in the Supplement. The SS values of all particle sizes showed similar distributions, in terms of the consistency of the fluorescence energy for WSOCs. The intensities for SS < 1.1 were of the same level as those for SS > 1.1, indicating the predominance of hydrophilic fluorescent contents in WSOCs. The average SS showed unobtrusive variations with increasing particle size in both seasons, and ηSS>1.1 was slightly higher for particle sizes < 1.4 µm compared with other particle sizes in winter. ηWH>320 tended to increase from a particle size of 0.26 to 0.44 µm and then decrease afterwards in both winter and summer (Fig. 5c), indicating the existence of a large-scale π-conjugated system or high π-electron density around 0.44 µm, and then decreased with particle size from 0.44 to 10 µm.

PARAFAC is a mathematical method capable of separating chemically independent but spectrally overlapping fluorescence components, based on the assumption that EEM spectra are independent, linearly related, and additive (Murphy et al., 2011). Several prior studies have been carried out using the PARAFAC method to investigate fluorescent WSOCs in atmospheric aerosols (Pohlker et al., 2012; Chen et al., 2019; Yue et al., 2019). The results have shown that bioaerosols exhibited high bimodal signals at an excitation wavelength of 275 nm and at an emission wavelength of 320 nm, which is classified as protein-like organic matter. In a typical coal-burning city in China, fluorophores emerging at excitation wavelength between 230 and 250 nm and an emission wavelength between 380 and 410 nm were associated with humic-like substances with a high molecular weight.

The present study conducted PARAFAC analysis for winter and summer samples separately to reveal seasonally dependent fluorescence spectra. Three components (C) were extracted from winter spectra: C1, defined as HULIS-1; C2, representing a protein-like component; and C3, defined as HULIS-2 (Chen et al., 2016b). However, only two recognizable components were identified in summer: C1, characterized as HULIS-1, and C2, characterized as protein-like components. Component C3 in the summer EEM spectra was of no physical significance (multiple emission peak points at one single excitation wavelength) and was characterized as a noise signal.

The proportions of the extracted components are shown in Fig. 6 along with the PARAFAC results. Protein-like compounds were more abundant in particles larger than 2.5 µm in both seasons (37 %–40 % in winter and 20 %–21 % in summer), and HULIS showed higher fractions in fine-mode than coarse-mode particles in both seasons. This quantitatively demonstrated that microbial-related WSOCs more likely existed in large particles and that HULIS was rich in fine particles. The ratios of HULIS-1 / HULIS-2 in winter were higher in fine particles with an aerodynamic diameter of 0.44–2.5 µm than in ultrafine particles (< 0.26 µm) or coarse-mode particles. HULIS-2 was likely freshly emitted fluorescent WSOCs, and HULIS-1 exhibited fluorescence characteristics of oxidized HULIS (Vione et al., 2019). The low HULIS-1 / HULIS-2 ratios in ultrafine and coarse-mode particles might be due to abundant sources of freshly emitted WSOCs.

A high GRD represents a strong connection between the reference and comparison factors. By setting the WSOCs (or AFI and UV) of particles < 0.26 µm as a reference and using those of larger particles for comparisons, the relations among particle sizes can be depicted by the GRD of size-segregated WSOCs (or AFI and UV), as shown in Fig. 7a for winter samples and in Fig. 7b for summer samples. In winter, GRD0.44–GRD2.5 showed a downward tendency, varying from 0.88 to 0.76 for WSOCs and from 0.88 to 0.78 for the AFI, indicating that the WSOC concentration and the AFI gradually deviate from their original situation with increasing particle size. In summer, GRD0.44–GRD2.5 showed little variation, with average values of 0.64 for WSOCs and 0.73 for the AFI, but decreased in GRD10, indicating that WSOCs in particles larger than 0.26 µm had little inheritance from the primal fine particle.

The relations of WSOCs with the AFI and average UV (referred to as UV below) of different particles are shown in Fig. 7c and d for winter and summer, respectively. The AFI and UV showed a high GRD in both seasons for all particle sizes (with an average GRD > 0.9), indicating that the fluorescence intensity and light absorption were closely connected with the WSOC concentration. However, clear variations in the GRD were observed with increasing particle size, with contrasting patterns to those of fluorescence indices. Thus, it was speculated that these variations resulted from the secondary transformation of WSOCs, as indicated by the fluorescence indices. Moreover, the GRD was strongly negatively correlated with the estimated secondary organic carbon (SOC) concentration, with a correlation efficient r of -0.64 (p < 0.000) in winter and -0.63 in summer. The lowest GRD was found for particle sizes of 0.26–0.44 µm. The high AFI, large π-conjugation scale, rich HULIS, and low GRD in this particle size range indicated that the fluorescent WSOCs of these particles were highly affected by secondary processes. Thus, the GRD between WSOCs and AFI could serve as an indicator of secondary formation.