Measurements of the cloud-free, cloud RES, and cloud INT particles were performed at Mt. Tianjing (24∘41′56′′ N, 112∘53′56′′ E, 1690 m a.s.l.) in Guangdong province, China between 18 November and 5 December 2020. This site is located at a national forest reserve and is less affected by anthropogenic sources. The cloud event determination threshold was set as visibility less than 3 km and relative humidity (RH) larger than 95 %. During the cloud events, the cloud RES and cloud INT particles were alternately introduced into the instruments through a ground-based counterflow virtual impactor (GCVI, model 1205, Brechtel Mfg., Inc., USA) and PM2.5 cutoff, respectively, at a frequency of 1 hour. The GCVI cut size was set to 7.5 µm, where the transmission efficiency of cloud droplets was 50 % (Shingler et al., 2012). It should be noted that the PM2.5 inlet may introduce possible uncertainty for the collection of cloud INT particles due to the interference of cloud droplets, although the size distribution of cloud droplets was mainly concentrated on 6–9 µm at mountain sites (Li et al., 2017). However, this would not be the case when cloud residual particles are mainly focused on, as in the present study. The collected cloud droplets passed through an evaporation chamber (40 ∘C), resulting in the cloud RES particles for downstream analysis. An aethalometer (model AE-33, Magee Scientific., USA) was used to measure the light-absorption coefficients of particles at wavelengths of 370, 470, 520, 590, 660, 880, and 950 nm. AE-33 uses two parallel spot measurement technologies to compensate for the light attenuation due to the filter loading effect (Drinovec et al., 2015). The BC concentration was calculated by the light-absorption coefficient at 880 nm. The detection limit of BC is less than 10 ng m-3 (equal to 0.077 Mm-1 at 880 nm) and the uncertainty is ∼ 2 ng m-3 (equal to 0.015 Mm-1 at 880 nm), with a time resolution of 1 min.

Cloud water samples were collected by a Caltech Active Strand Cloud Water Collector, Version 2 (CASCC2) (Demoz et al., 1996; Y. Yang et al., 2021) when the visibility was less than 200 m (from 14 November to 4 December 2020). The cut size was 3.5 µm, the flow rate was 5.8 m3 min-1, and the overall collection efficiency was 86 %. During the sampling period, 53 cloud water samples were collected. The 0.22 µm quartz fiber filter was used immediately to remove insoluble components after collection of cloud water and then frozen at -20 ∘C until analysis. Meanwhile, PM2.5 samples were collected by a mid-volume (300 L min-1) aerosol sampler (PM-PUF-300, Mingye, China). Daily samples (during 14 November to 8 December 2020) were collected on the quartz fiber filters, which were prebaked at 450 ∘C for 4 h in a muffle furnace to remove residual organics before use. After collection, all samples were frozen at -20 ∘C until analysis. In this study, PM2.5 samples collected at the same time as cloud water samples were regarded as INT-PM2.5 (n=13), and the others as FREE-PM2.5 (n=19). It should be noted that some FREE-PM2.5 samples also experienced short cloud events during collection. Blank samples of the cloud water and PM2.5 were collected and processed following the same procedure as the samples.

The light-absorption coefficient (AbsBrC(λ), Mm-1) of BrC in different wavelengths can be obtained by AE-33, assuming that the absorption Ångström exponent (AAE) of BC is 1 and the light absorption at 880 nm only due to BC (Drinovec et al., 2015). The cloud RES, cloud INT, and cloud-free particles were generally located in submicron size (Fig. S1), and thus were unlikely to have originated from noncombustion sources and are mostly biogenic and mainly exist in the coarse mode (Perrino and Marcovecchio, 2016). The AbsBrC(λ) contributed by the combustion sources can be estimated through a BC-tracer method (Wu et al., 2018) 1Abspri,combλ=AbsλBCpri×[BC]. Where Abs(λ) is the total light-absorption coefficient of carbonaceous aerosol that was measured by AE-33, (AbsλBC)pri can be determined by the minimum R squared (MRS) method to further evaluate the relative contribution of primary BrC and secondary formation BrC to the overall AbsBrC(λ). Firstly, Abspri,combλ is calculated based on (AbsλBC)pri, which is assumed to be step increasing from 0 to 120 with a rate of 0.1. The target (AbsλBC)pri value can be retrieved when the correlation coefficient (R2) between AbsBrC,sec(λ) with BC concentration reaching the minimum (see Fig. S2). Previous studies showed that the bias of MRS method is less than 23 % when the measurement uncertainty is less than 20 % (Wu and Yu, 2016). It should be noted that when the measured ratio of AbsλBC is lower than the retrieved (AbsλBC)pri, the AbsBrC,sec(λ) could be negative. In these cases, AbsBrC,sec(λ) is set to zero for subsequent analysis (Kaskaoutis et al., 2021; Wang et al., 2019a). These cases account for less than 5 % in the cloud RES and 28 %–70 % in the cloud INT and cloud-free particles

PM2.5 samples were ultrasonically extracted with ultrapure water (resistivity: 18.2 MΩ cm) for 30 min, then filtered by 0.22 µm polytetrafluoroethylene (PTFE) filters to obtain the PM2.5 aqueous extract. The concentrations of water-soluble ions, water-soluble heavy metals, WSOC in PM2.5 aqueous extract and cloud water samples were analyzed by ion chromatography (Metrohm 883 IC plus, Switzerland), inductively coupled plasma mass spectrometry (ICP-MS, Thermo Fisher, USA), and total organic carbon analyzer (TOC-V, Shimadzu, Japan), respectively. Parallel analyses showed that the relative standard deviation of each analysis was generally less than 15 %. The reported concentration data herein was after blank subtraction.

The light-absorption coefficient (AbsWSOC,λ) of WSOC can be obtained (Hecobian et al., 2010) by the measurement of cloud water and PM2.5 aqueous extract, with UV-Vis (UV1901, Kejie, China) 2AbsWSOC,λ=(Aλ-A700)×VlVa×L×ln⁡(10). Where Aλ is the absorbance of the sample, A700 is used to account for any drift; Vl is the volume of ultrapure water used to extract the sample (for cloud water it is the total sample volume), Va is the volume of sampled air through the PTFE filter (for cloud water it is the total volume of sampled air), and L is the cuvette path length (0.01 m).

The AAE values describing the spectral dependence of WSOC light absorption can be further deduced by exponential fitting AbsWSOC,λ between 300–500 nm. The MAEWSOC,λ can be calculated by dividing AbsWSOC,λ by the mass concentration of WSOC (µg m-3). The E250/E365 (the ratio of absorbance at 250 nm to that at 365 nm) is used to describe the humification of organic matter, which is inversely related to aromaticity and molecular weight of WSOC (Kristensen et al., 2015). Specific UV absorbance (SUVA, m2 g-1) at 254 and 280 nm has been proved to be qualitatively related to the structural characteristics (aromaticity and molecular weight) of WSOC to a certain extent (Weishaar et al., 2003), which can be calculated using the following equation: 3SUVA254or280=AL×CWSOC, where A is the absorbance of sample at 254 or 280 nm and CWSOC is the concentration of WSOC (mg L-1).

The excitation-emission matrix fluorescence spectroscopy (EEMs) of PM2.5 extract and cloud water was measured by a fluorescence spectrophotometer (F97pro, Lengguang, China). The sample blank was deducted before analysis, and the EEMs were normalized to the Raman units (R.U.) by using the Ramen peak (Ex = 350 nm, Em = 365–430 nm) of ultrapure water measured simultaneously with the sample (Lawaetz and Stedmon, 2009). Parallel factor (PARAFAC) analysis was performed on the acquired spectra with drEEM toolbox (version 0.3.0) based on MATLAB (Murphy et al., 2013). According to the outlier tests of PARAFAC, six samples with high leverage and high residual signals were removed in the modeling of PM2.5 aqueous extract. The details for obtaining maximum fluorescence intensity (Fmax), fluorescence index (FI), recent autochthonous contribution (BIX), and humification index (HIX) were described in​​​​​​​ Sect. S1 in the Supplement.

The presence of BrC could be indicated by the AAE values derived from AE-33 data, which are 1.30 ± 0.12 for cloud-free, 1.36 ± 0.22 for cloud INT, and 1.32 ± 0.15 for cloud RES particles. The light-absorption coefficient of BrC at 370 nm (Abs370) of cloud-free, cloud INT and cloud RES particles are 1.47 ± 1.23, 1.16 ± 1.14, and 0.25 ± 0.15 Mm-1, respectively (Fig. 1), with the AAE values of BrC at 2.71 ± 0.69, 3.13 ± 0.97, and 2.76 ± 0.89, respectively. The contribution of BrC light absorption to the total particle light absorption in the cloud-free, cloud INT, and cloud RES particles shows no significant difference, on average decreasing from ∼  23 % at 370 nm to ∼ 7 % at 660 nm, as shown in Fig. 2.

For the cloud water and PM2.5 aqueous extracts, light absorption properties of WSOC at 365 nm are taken as the representative to those of water-soluble BrC (WS-BrC) in the present study. As expected, there is a positive correlation between Abs365 and WSOC concentration in both cloud water and PM2.5 aqueous extracts (r>0.61, p<0.01). As shown in Fig. 1, there is great difference in Abs365 of WSOC in FREE-PM2.5, INT-PM2.5, cloud water-Day, and cloud water-Night, which are 0.49 ± 0.34, 0.27 ± 0.18, 0.09 ± 0.04, and 0.13 ± 0.10 Mm-1, respectively. The Abs365 of WSOC in PM2.5 in this study is at the same magnitude as that of PM10 in the Tibetan Plateau (Kirillova et al., 2016), and much lower than those in urban areas (3.4–33.9 Mm-1, as summarized in Table S1) (Chen et al., 2018, 2020; Huang et al., 2020; Kim et al., 2016). The AAE of WSOC has no significant difference among FREE-PM2.5, INT-PM2.5, cloud water-Day, and cloud water-Night, which are 6.01 ± 0.81, 5.37 ± 1.08, 5.81 ± 1.47, and 6.31 ± 1.51, respectively, within the reported range.

The MAE365 of WSOC in FREE-PM2.5, INT-PM2.5, cloud water-Day, and cloud water-Night are 0.31 ± 0.17, 0.31 ± 0.26, 0.17  ± 0.07, and 0.17 ± 0.07 m2 g-1, respectively. The MAE365 of WSOC in cloud water and PM2.5 are much lower than those in urban and alpine areas, and various source emission samples (Table S1) (Chen et al., 2018, 2020; Fan et al., 2016; Huang et al., 2020; Kim et al., 2016; Kirillova et al., 2016; Li et al., 2019; Park and Yu, 2016; Soleimanian et al., 2020; Wu et al., 2019). The MAE365 of WSOC shows no significant difference between the FREE-PM2.5 and INT-PM2.5, which is similar to the result observed in the Indo-Gangetic Plain (Choudhary et al., 2018), but their values are quite a bit higher, i.e., 1.6 and 1.8 m2 g-1 for the INT-PM1.0 and FREE-PM1.0, respectively. The MAE365 of WSOC in cloud water (0.06–0.32 m2 g-1) is slightly lower than the previously reported values in fog water (0.1–0.6 m2 g-1) in California (Kaur and Anastasio, 2017). Both the MAE365 of WSOC in cloud water and PM2.5 show a positive correlation (r>0.84, p<0.01) with SUVA254/280, and a medium negative correlation (r>0.43, p<0.05) with E250/E365, which may indicate that a higher MAE365 of WSOC has a higher aromatic and molecular weight; the aromaticity and molecular weight of WSOC may influence the light absorption capacity of cloud water and PM2.5 (Fig. S3).

Although there are tight correlations between the Abs370 for cloud water and the cloud RES particles, and for the INT-PM2.5 and the cloud INT particles (Fig. 3, r>0.60, p<0.01), the Abs370 of WSOC in cloud water (0.12 Mm-1) and INT-PM2.5 (0.27 Mm-1) are considerably lower than those in the cloud RES (0.24 Mm-1) and cloud INT particles (1.08 Mm-1) that collected simultaneously. Such differences may be attributed to the contribution of water-insoluble organic carbon (WIOC). The different optical properties for the whole BrC and WS-BrC may also be reflected by the AAE values. They are generally in the range of 4–8 at 300–500 nm in cloud water and PM2.5, much higher than those for BrC (2–4) calculated from AE-33 data at 370–660 nm. The contribution of water-insoluble BrC to the light absorption is estimated to be ∼ 75 % for the cloud INT particles and ∼ 48 % for the cloud RES particles on average, based on these differences (Fig. 3). It is also noted that the light absorption of WIOC might still be underestimated by ∼ 16 % when sampling size is considered for the GCVI and cloud sampler (as discussed in Sect. S1). High contributions of WIOC to BrC light absorption have also been observed in the Indo-Gangetic Plain (77 %) (Satish et al., 2020), Beijing (62 %), and Xi'an (51 %) (Huang et al., 2020).

Figure 2 shows the contribution of secondary BrC to the total BrC in cloud-free, cloud INT, and cloud RES particles estimated by the MRS method; 11 %–16 % and 9 %–23 % of the total absorption of BrC come from the secondary BrC for the cloud-free and cloud INT particles, respectively. Only a slight difference was observed for the cloud-free and cloud INT particles, indicating that cloud processing may have limited influence on the cloud INT particles. It is noted that even during the cloud-free periods, RH was generally higher than 70 % (Fig. S1). The contribution of secondary BrC in cloud INT and cloud-free particles are in the low range of reported values (as summarized in Table S2) (Gao et al., 2022; Kaskaoutis et al., 2021; Lin et al., 2021; Q. Wang et al., 2019a, b, 2020, 2021; Zhang et al., 2020, 2021; Zhu et al., 2021).

Differently, the contribution of secondarily formed BrC to the total BrC light absorption is 67 %–85 % in the cloud RES particles, which is surprisingly higher than those in the cloud-free and cloud INT particles. Such a high contribution may suggest how critical the role of cloud processing in the formation of BrC is. Compared with the relative contributions for the cloud-free and cloud INT particles, the importance of such a process in cloud droplets remarkably overrides that in cloud-free and cloud INT particles. The significance of secondary water-soluble BrC formation in cloud droplets may also be reflected by the significant correlation between the Abs365 of cloud water and PM2.5 aqueous extract with SNA (sulfate, nitrate, and ammonium) (r>0.77, p<0.01), and NOx (r>0.58, p<0.01), as shown in Fig. S4. The SNA and NOx concentrations are higher at night than in the daytime (Fig. S5), which is consistent with higher Abs365 of cloud water at night. NO2- resulting from dissolved NOx can react with benzene and finally forms nitrophenol in the presence of UV-A (Harrison et al., 2005; Vione et al., 2004). Various of reactive oxygen and nitrogen species generated from the photolysis of inorganic nitrate in aqueous-phase could also facilitate the photooxidation of organic compounds to form BrC (Seinfeld and Pandis, 2016; J. Yang et al., 2021) and potentially contribute to the light absorption of cloud water (Desyaterik et al., 2013). In-cloud aqueous processes leading to more CHON compounds in cloud water than below-cloud atmospheric particles have also been observed (Boone et al., 2015). In addition, a comparison between the WSOM (WSOM = WSOC × 1.8) normalized by K+ (as a primary source tracer) in cloud water and INT-PM2.5 (Fig. S6) also clearly indicates the enhanced formation of WSOM in cloud water. It is consistent with that the light absorption of WSOC contributed more to the cloud RES (∼ 52 %) than the cloud INT (25 %) particles, as estimated in Fig. 3.

The source apportionment of BrC in cloud water (i.e., Abs365) evaluated by the PMF model (see Supplement for data analysis and evaluation methods) also supports the critical role of aqueous processes on the formation of BrC, as shown in Fig. 4. Factor 1 is associated with relatively higher K+, NH4+, NO3-, SO42-, and C2O42-, contributing 64.3 % to WSOC and 86.9 % to Abs365. It may be appropriately recognized as a secondary product with contribution from biomass burning, as K+ represents a tracer for biomass burning, and NH4+, NO3-, C2O42-, and SO42- are regarded as secondary species (Cheng et al., 2015; Wang et al., 2012). Note that C2O42- is generally considered as a tracer of aqueous-phase processes (Zhang et al., 2017b). As previously observed, the aqueous SOA formed from biomass burning might contributed to the BrC budget in fog water (Gilardoni et al., 2016). Factor 2 is characterized by high levels of crustal trace elements such as Mg2+, Ca2+, Mn, and Zn, and thus identified as crustal materials, contributing 21.9 % to WSOC and 8.7 % to Abs365. Factor 3 shows extremely high loading with Na+ and relatively high Mg2+, Cl-, and Ni, which may originate from marine environments, contributing 13.8 % to WSOC and 4.4 % to Abs365.

The results from the EEMs measurements further indicate the different characteristics of WSOC and WS-BrC in PM2.5 and cloud water. Based on the PARAFAC model calculation (Fig. 5), two independent fluorescence components (P1–P2) assigned as humic-like substances are found in PM2.5, whereas four independent fluorescence components (C1–C4) assigned as humic-like substances (C1–C3), and tyrosine-like and/or protein-like substances (C4) are found in cloud water (Catalá et al., 2015; Coble, 2007). The fluorescence components of cloud water are similar to those at Mt. Tai (Zhao et al., 2019) and in France (Bianco et al., 2016b), where humic-like and protein-like substances are the main chromophores in cloud water. Compared with PM2.5, tyrosine-like and/or protein-like substances are unique to cloud water in the present study, which may partly be due to their relative enrichment in cloud water (Kristensson et al., 2010; Zhang and Anastasio, 2003).

In addition, the relative contribution of individual chromophores indicated by Fmax in PM2.5 and cloud water also exhibits different characteristics, although humic-like substances are the dominant fluorescent fraction in both PM2.5 and cloud water. The relative contribution shows no obvious difference between P1 and P2 components in PM2.5 (FREE-PM2.5 and INT-PM2.5), whereas the C3 component contributes the most (40.0 %) to the fluorescent intensity in cloud water. Further analysis of the relationship between the fluorescent components (Fmax) and the light absorption of WSOC (Abs365) in PM2.5 and cloud water shows significant positive correlations between Fmax of all fluorescent components with Abs365 (r>0.63, p<0.01, see Fig. 5). It suggests that these fluorescent components are tightly linked to the light absorption of WSOC. The FI, BIX, and HIX of cloud water are 1.58 ± 0.22, 0.57 ± 0.09, and 4.99 ± 3.83, respectively, which indicates limited humified WSOC in cloud water, and also that it is less affected by microorganisms and local sources (Huguet et al., 2009; McKnight et al., 2001; Zsolnay et al., 1999). Therefore, it is most likely that the organic components in cloud water may be significantly affected by in-cloud aqueous formation, which is consistent with the PMF results. With respect to the secondary processes, humic-like substances may be formed through Maillard reaction involving carbonyls with ammonium or amines (Bones et al., 2010; Hawkins et al., 2016), and also the photo-transformation of tyrosine (Berto et al., 2016).