The concentrations of gas-phase peroxides were measured in both BJ-2018Summer and BJ-2018Winter using high-performance liquid chromatography (HPLC, Agilent 1200, USA) with a time resolution of 21 min. The HPLC coupled with the post-column enzyme derivatization method could distinguish H2O2 from organic peroxides. This method is well established (Hua et al., 2008; He et al., 2010; Wang et al., 2016) and is only briefly described here. Ambient air was drawn into a glassy scrubbing coil at a flow rate of 2.7 slpm (standard litre per minute). H3PO4 solution (5×10-3 M) was added to the scrubbing coil at 0.2 mL min-1 to dissolve H2O2 from ambient air. The collection efficiency of H2O2 was validated to be close to 100 %. Then, the mixture was injected into the HPLC instrument with the mobile phase (H3PO4, 5×10-3 M). Peroxides separated by the column reacted stoichiometrically with para-hydroxyphenylacetic acid (PHPAA) under the haemin catalyst, generating stable PHPAA dimers that were measured by a fluorescence detector. The peroxides were identified and quantified using standard samples, and the detection limit (DL) of the gas-phase H2O2 was about 10 parts per trillion by volume (pptv). The values below DL were replaced by DL divided by the square root of two (the same hereafter). The gas-phase samples during BJ-2018Summer were used for the partitioning analysis in the gas-liquid phase, while the data of BJ-2018Winter were used to study the partitioning in the gas-aerosol phase.

Rain samples were collected by a custom-built glass funnel and were used for the analysis of liquid-phase peroxides in BJ-2018Summer. During the observation period, the collection of rain samples was well organized depending on the intensity, amount, and duration of the rain. Because the peroxides were easy to break down, the collected rain samples were preserved in brown vials at 277 K until being analysed with HPLC within 6 h. The subsequent detection method for the liquid-phase peroxides was the same as for the gas-phase peroxides. In all, we collected 60 rain samples during seven rain episodes, and the DL of the liquid-phase H2O2 was about 8 nM. The specific dates of the rain events in chronological order were 24 July, 25 July, 5 August, 6 August, 8 August, 30 August, and 2 September.

Aerosol-phase samples were gathered on Teflon filters (Whatman™, 47 mm diameter and 2 µm pore size) using a four-channel filter sampler (Wuhan Tianhong TH-16A, China) at 16.7 slpm during BJ-2018Winter. Teflon filters were supported by stainless-steel filter holders during a 11.5 h sampling interval. We immediately disposed of two Teflon filters for the analysis of peroxides and total peroxides (TPOs), and the remaining filters were kept under refrigeration at 255 K for subsequent component analysis. For analysing the aerosol-phase peroxides, the Teflon filters were immediately extracted with 10 mL H3PO4 in conical flasks and placed on a shaker to be blended thoroughly at 277 K and 180 rpm for 15 min. Then, the extracted solution was measured with HPLC within 40 min. The remained extracted solution was stored at 255 K away from light for subsequent measurement of H2O2 concentration variation with time, and details of the experimental conditions of the extracted solution are shown in the Supplement. Photochemical reactions of aerosols may produce aerosol-phase H2O2 (Zhou et al., 2008), and the effect of the photochemical reactions on the level of H2O2 in the extracted solution is discussed in the Supplement. Furthermore, the extracted solution was also used for the measurement of TPOs using the iodometric spectrophotometric method, which could measure H2O2 as well as organic peroxides (Nozaki, 1946; Banerjee and Budke, 1964). The extraction efficiency has been discussed in our previous work (Li et al., 2016), and we did not correct the total peroxides level. After the oxygen in the extracted solution was blown off by bubbling with nitrogen for 5 min, 250 µL of potassium iodide solution (KI, 0.75 M) was added to the solution to react with TPOs in the dark for 12–24 h (Reactions R1 and R2). The reaction product I3- ion could be detected using UV–Vis spectrophotometry (Beijing PERSEE TU-1810, China) at the wavelength of 420 nm. A total of 31 aerosol-phase samples were analysed, and the DL of aerosol-phase H2O2 was close to 0.24 ng m-3 (0.006 ng µg-1).

To avoid the matrix influence on samples (i.e. Teflon filters and the H3PO4 solution), we measured the concentration of blank samples in every extraction. The level of H2O2 in three-quarters of the blank samples was equal to 0 µM, and the concentration of H2O2 in the remaining blank samples was below 10 % of that in the ambient aerosol samples. To prevent the matrix influence, we deducted the background values of the samples. In addition, to ensure that the measured H2O2 was attributed to aerosols collected on Teflon filters, we performed experiments to demonstrate that the physical adsorption on clean Teflon filters without aerosols was responsible for 15 % of the measured H2O2 in aerosol samples. The details are available in Fig. S1 in the Supplement. In this study, we did not correct for the effect of the physical adsorption on the aerosol-phase H2O2. R13I-+H2O2+2H+→I3-+2H2OR2R1OOR2+3I-+2H+→I3-+R1OH+R2OH

Water-soluble cations (Na+, NH4+, K+, Mg2+, and Ca2+) as well as anions (Cl-, NO3-, and SO42-) were measured with ion chromatography (IC, Dionex ICS2000 and ICS2500, USA). Transition metal elements deposited on Teflon filters were measured with inductively coupled plasma mass spectrometry (ICP-MS, Bruker aurora M90, Germany). The mass concentration of PM2.5 was measured with a TEOM 1400a analyser. Meteorological parameters (ambient temperature, relative humidity, and wind speed) and major trace gases (O3, SO2, NO, NO2, NOx, and CO) were monitored simultaneously using a series of commercial instruments (Met One Instruments Inc., Thermo 49i, 43i, 42i, and 48i).

The concentration of gas-phase H2O2 was measured to be 0.30±0.26 parts per billion by volume (ppbv) for the seven rainfall events (Fig. S2a in the Supplement) and 0.53±0.77 ppbv for the entire BJ-2018Summer period. The theoretical liquid-phase H2O2 value in pure water was 25.20 µM, and the level of measured H2O2 in the liquid phase was 44.12±26.49 µM (3.19–139.95 µM), as shown in Fig. S2b. The detailed values of the peroxides in the gas and liquid phases are shown in Table S1. Based on Eq. (2), the effective field-derived Henry's law constant, HAm, averaged 2.1×105 M atm-1 in rain samples, which was 2.5 times the theoretical pure-water Henry's law constant, HAt, at 8.4×104 M atm-1 and 298±2 K. The average of the ratio of predicted to measured levels of H2O2 in each rain sample was 88 %. The result suggested that 88 % of the measured liquid-phase H2O2 came from gas-phase partitioning, while 12 % of H2O2 was from other sources. We divided 52 rain samples into three types based on the comparison of the measured and predicted levels of H2O2. When the difference between levels of the measured and predicted liquid-phase H2O2 fell within ±20 %, we suggested that these samples (type B) followed Henry's law, and the remaining samples (types A and C) did not agree with Henry's law. The percentages of samples in types A, B and C were 69 %, 19 %, and 12 %, respectively. The details of each rain sample are presented in Fig. 1. In this paper, we focused on the type A samples with a high measured liquid-phase H2O2 level, and the difference between levels of the measured and predicted liquid-phase H2O2 averaged 30 µM with a maximum of 71 µM. Further, based on the ratio of the predicted to measured levels of H2O2 in rainwater, 59 % of the liquid-phase H2O2 in type A samples was produced from sources other than gas-phase partitioning.

To explain the difference between HAm and HAt, we should rule out the effects of pressure, pH, and TS on HAt. First, to our knowledge, the influence of pressure on HAt can usually be neglected under conditions of less than 1 atm (Lind and Kok, 1986). Also, HAt of H2O2 is independent of pH in the range of 4–7 (Xu et al., 2012); therefore, the present study does not consider the influences of pressure and pH on HAt. The temperature during BJ-2018Summer could be divided into three ranges: 294–296, 297–299, and 300–306 K. The percentages of samples in these three temperature ranges were 25 %, 63 %, and 12 %, respectively, and the ratios of HAm to HAt in the same temperature ranges were 1.4, 2.6, and 4.5, respectively. The maximum value of HAt in the range 294–306 K was 1.2×105 M atm-1, while HAm reached 4.2×105 M atm-1 at the 90th percentile. This suggested that the influence of TS on HAt was negligible. The non-linear relationship between HAm and TS, shown in Fig. S3, also indicated that TS played an unimportant role in determining HAm. Thus, other explanations were needed for understanding the difference between HAm and HAt.

The solubility of H2O2 in clouds is higher than that in the ground rainwater. There is a negative dependence of the solubility on temperature (Huang and Chen, 2010), which allows for the possibility of mass transfer of H2O2 from rainwater to the surrounding air when falling. Let us assume that the gas-phase H2O2 concentration is homogeneous and the rain droplet size remains constant during the falling process. The diameter of the raindrops (DP) is mainly distributed in the range of 0.05–2.50 mm. Calculations were performed for typical droplet diameters at 0.1, 0.5, 1.0, and 2.0 mm. The height of the precipitation cloud base during summer time in north China was always less than 2000 m (Shang et al., 2012), so we assumed the fall distance to be 500, 1000, 1500, and 2000 m, respectively, which remained the same as previous studies (Adamowicz, 1979; Levine and Schwartz, 1982). In the process of falling, it was necessary to consider the mass transfer resistance in the gas and liquid phases. However, it could be that the shear force generated on the raindrop surface when it fell improved the mixing rate in the droplet significantly; therefore, the liquid-phase mass transfer resistance was negligible (Pruppacher and Klett, 1997; Elperin and Fominykh, 2005). Thus, the overall mass transfer resistance reduced to the mass transfer resistance in the gas phase.

Here, we first discussed residual H2O2 in raindrops after a fall from a height of 1000 m. The temperature in clouds (TSc) was estimated to be 292 K, which was 6 K lower than the ground. HAt in pure water at 292 K was 1.4×105 M atm-1 (Sander et al., 2011). Provided that the droplet started at equilibrium with the cloud atmosphere and the level of H2O2 in the cloud atmosphere was equal to the level near the ground (0.30±0.26 ppbv), the initial level of liquid-phase H2O2 before falling (Caq0) was 42.87 µM. However, the equilibrium was broken when the raindrops fell as the ambient temperature increased. The mass transfer coefficient in the gas phase (kg) can be calculated by Eqs. (S1) to (S4) in the Supplement (Levine and Schwartz, 1982; Kumar, 1985). The concentration of H2O2 in the droplet at the ground (Caqd) can be estimated by Eq. (S5). The results are presented in Table 1, demonstrating that the large droplet has a small mass transfer coefficient. As a result, the liquid-phase H2O2 in the large raindrops was more slowly released into the air. Caqd of the droplet diameter at 2.0 mm was close to Caq0, while Caqd at 0.1 mm approximated the theoretical level of liquid-phase H2O2 in pure water at 298 K, as indicated by Fig. 2. The results demonstrated that the effect of residual H2O2 in large raindrops on the level of H2O2 in ground rainwater could be of great importance.

Next, we investigated the influence of different fall distances on Caqd. The decreasing temperature at increasing fall distances caused larger HAt and Caq0 in clouds, and Caqd also increased. The wide gap of Caqd between different fall distances was more visible for the large droplet, as seen in Fig. 2. Based on the above analysis, the residual H2O2 in large raindrops could increase the H2O2 level in rainwater to a maximum of 48.81 µM at a fall distance of 2000 m, which could explain to a large extent the difference between the measured and predicted levels of H2O2 in rainwater.

Based on the rain intensity, seven rain events during BJ-2018Summer could be divided into three types, as shown in Table S2 in the Supplement. Rain events in types I, II, and III have rain intensities <1, 1–10, and >10 mm h-1, respectively. The larger the diameter of raindrops, the greater the rain intensity (Kumar, 1985). According to the above relationship between the diameter of raindrops and the level of liquid-phase H2O2 in the ground rainwater, the difference between the measured and predicted liquid-phase H2O2 levels should be greater as the hourly rain intensity increases. We found that the differences between Caqm and Caqt increased during the rain periods on 25 July and 5 August, during which the maximum hourly rain intensities were more than 10 mm h-1. Because it is difficult for the liquid-phase H2O2 in heavy rains to diffuse into the gas phase, much H2O2 may be retained in the ground rainwater, which could well represent the level of H2O2 in cloud water.

During the rain episode on 1–2 September 2018, the concentration of gas-phase H2O2 decreased over time. However, there was a sudden rise from 0.47 ppbv at 01:03 local time (LT) to 0.66 ppbv at 01:46 LT, which subsequently dropped to 0.38 ppbv over time (Fig. 3a). Surprisingly, the difference between the measured and predicted levels of liquid-phase H2O2 reached a low value in the meantime, indicating that the increase in gas-phase H2O2 was due to the release of H2O2 from raindrops that contained high level of H2O2, as presented in Fig. 3b. Compared with Fig. S4 in the Supplement, which described the relationship between rain intensity and time, the rain intensity simultaneously dropped to 3.51 mm h-1 from 6.35 mm h-1, consequently decreasing the diameter of the raindrops and increasing the mass transfer of H2O2 from rainwater to the gas phase. Provided that 20 µM H2O2 in rainwater was released into ambient air, the increase in the gas-phase H2O2 level was 0.24 ppbv, which was in accordance with the sudden rise during 01:03–01:46 LT on 1–2 September 2018.

The above analysis assumes that the gas-phase H2O2 concentration is uniform. However, the distribution of gas-phase H2O2 at different heights may be complicated. We could use the average level of H2O2 in rainwater at the ground to estimate the concentrations of H2O2 in cloud water (Caqc) and the nearby atmosphere (Cgc), as presented in Table 2. Assuming the simplest case, DP is 1.0 mm, the fall distance is 1000 m, and the levels of H2O2 in the gas phase and rainwater at the ground are 0.30 ppbv and 44.12 µM at 298 K, respectively. Considering the release of H2O2 from raindrops into ambient air during the falling process, the level of H2O2 in cloud water should be 47 µM. Based on Henry's law, the surrounding gas-phase H2O2 should be 0.33 ppbv, a little higher than that at the ground. When the fall distance is 500, 1500, and 2000 m, H2O2 in cloud water should be 46, 49, and 51 µM, respectively, and H2O2 in nearby ambient air could be 0.41, 0.26, and 0.21 ppbv, respectively.

From 21 December 2018 to 5 January 2019, the gas-phase H2O2 level was 24.08±28.83 pptv, as shown in Fig. S5a in the Supplement. We eluted Teflon filters with H3PO4 solution and measured the level of H2O2 in the extracted solution to calculate the aerosol-phase H2O2 concentration. The mass concentration of aerosol-phase H2O2 and the normalized concentration to aerosol mass were 2.22±1.49 ng m-3 (DL – 6.75 ng m-3) and 0.093±0.085 ng µg-1 (DL – 0.409 ng µg-1), respectively. The mean concentration of the aerosol-phase H2O2 at night-time (0.107±0.102 ng µg-1) was higher than that at daytime (0.079±0.066 ng µg-1), as presented in Fig. S5b. The level of aerosol-phase H2O2 in the present study was lower than those reported in previous studies (Table S3), which may be due to the extraction method, the extraction time, reducing substance levels, or aerosol pH value, as shown in the Supplement. Assuming a molecular weight of 300 g mol-1 (Docherty et al., 2005; Epstein et al., 2014), the level of TPOs averaged 10.26±6.38 ng µg-1 (2.08–28.75 ng µg-1). It was calculated that H2O2 took up a small fraction of TPOs, equal to 8±6 % in molar concentration ratio, which indicated that organic peroxides accounted for a large proportion of peroxides and could play important roles in the formation of PM2.5 and secondary organic aerosols.

The measured level of H2O2 in aerosols was much higher than the predicted value using Pankow's absorptive partitioning theory, which suggested that the aerosols collected on the filter existed under non-equilibrium conditions, and the aerosol-phase H2O2 may arise from sources other than gas-phase partitioning. Based on Eqs. (3) and (4), KPm was equal to 3.8×10-3±4.8×10-3 m3 µg-1 at 270±4 K, which was 4 orders of magnitude higher than KPt, 2.8×10-7 m3 µg-1. The effect of parameter variation on calculating KPt could not account for the large discrepancy between KPm and KPt (Shen et al., 2018), and other factors are needed to explain the difference. In terms of the proportion of theoretical to measured concentrations, the partitioning of gas-phase H2O2 into aerosols could be neglected, and nearly all of the aerosol-phase H2O2 was generated from other reactions.

Because aerosol water content (AWC) cannot be correctly evaluated at low RH, the effective field-derived Henry's law constant (HPm) of H2O2 was estimated for the high-RH condition, e.g. a heavy haze episode from 2 to 3 January 2019 (RH, 30 %). Details regarding the estimation of AWC are shown in the Supplement. It was calculated that AWC, Cpm, and Cgm levels averaged 3.20 µg m-3, 6.63×103 µM, and 1.90×10-11 atm. Based on Eq. (5), the average HPm on 2–3 January 2019 was calculated to be 2.7×108±1.8×108 M atm-1. However, the theoretical Henry's law constant (HPt) at 270 K was 1.1×106 M atm-1 (Sander et al., 2011), which was lower than HPm by 2 orders of magnitude. In the study of Chung et al. (2005), the salting-in effect can improve the level of H2O2 by a factor of 2 when the concentrations in salt solutions were up to 10 M, and the most obvious salting-in effect of salt solutions was ammonium sulfate. In this paper, the levels of aerosol-phase NH4+ and SO42- on 2–3 January 2019 were 94 and 21 M, respectively, and the level of (NH4)2SO4 was assumed to be 21 M. The increase of HPm by the salting-in effect of (NH4)2SO4 was about 3.2×106 M atm-1 at 286 K based on equations in Chung et al. (2005). Even though aerosol particles were collected at 270±4 K and the increase may be greater, the salting-in effect could not fully explain the difference between HPm and HPt. Other sources need to be found later.

Figure 4 demonstrates that the concentration of aerosol-phase H2O2 is dependent on RH, with a trend that is first increasing (stage A) and then decreasing (stage B) as RH increases. The variation of H2O2 with RH was the result of competition between production and consumption processes. Here, the production process refers to either process that favours increasing the level of aerosol-phase H2O2, while the consumption process refers to the process consuming aerosol-phase H2O2. In the first stage, the higher RH could accelerate the heterogeneous uptake of H2O2 onto aerosols and enhance the level of aerosol-phase H2O2 (Pradhan et al., 2010; Shiraiwa et al., 2011; Zhao et al., 2013; Slade and Knopf, 2015). The level of H2O2 was negatively associated with RH in the subsequent stage, ascribed to much more rapid consumption of H2O2 due to its oxidizing of the reducing substances on polluted days, such as SO2 into SO42-.

We considered a heavy haze episode, from 2 to 3 January 2019, as an example to explain the important contribution of aerosol-phase H2O2 to SO42- growth on polluted days in detail. The PM2.5 mass concentration of the severe haze event was up to 201.20 µg m-3. Based on the measured H2O2, the reaction rate (RR) and sulfate formation rate (SFR) averaged about 3.03×10-3 µmol m-3 h-1 and 0.29 µg m-3 h-1 (Table S4), respectively. The detailed calculation process is provided in the Supplement. In addition, the growth rate of SO42- calculated by the measured data was 0.51 µg m-3 h-1, and H2O2 oxidation pathway contributed about 57 % of the measured growth of SO42- in PM2.5. This result strongly suggested that the aerosol-phase H2O2 indeed acted as an important oxidant in the formation of sulfate, and played significant roles in the rapid growth of PM2.5 during pollution events.

Next, we considered that the consumption rate of aerosol-phase H2O2 increased with an increase of RH. The extent of the concentration variations of H2O2, SO42-, and PM2.5 at 10th and 90th percentiles were 22, 6, and 5, respectively, suggesting that the inverse relationship still existed when we eliminated the interference of the dilution effect due to a high aerosol loading. The dilution effect was unimportant and could be neglected. It is suggested that larger levels of SO42- and PM2.5 are often accompanied by higher RH. In Fig. 5, the reverse curve between aerosol-phase SO42- and H2O2 became steeper with the formation of SO42-, indicating that the rate of consumption of H2O2 on polluted days was much higher than that on clean days, which could offer proof of rapid H2O2 consumption with increasing RH. In addition, the level of H2O2 in the aerosol phase exhibited a negative correlation with PM2.5 mass concentration, as shown in Fig. 5. In other words, the aerosol-phase H2O2 concentration was lower on polluted days than on clean days, further demonstrating that the removal rate of H2O2 by oxidizing SO2 into SO42- exceeded the production rate during pollution events with high RH.

In addition to the factors that influence the aerosol-phase H2O2 concentration, there are certain physical and chemical reactions other than gas-phase partitioning that could increase the level of aerosol-phase H2O2, e.g. heterogeneous uptake of H2O2 on aerosols. Previous studies have shown that heterogeneous uptake of H2O2 is positively related to RH. High RH is beneficial to the mass transfer of H2O2 from the gas phase to the aerosol phase, which could accelerate the reaction between H2O2 and reducing compositions of aerosols, thus contributing to more heterogeneous uptake of H2O2 (Huang et al., 2015; Wu et al., 2015). To quantitatively evaluate the importance of the heterogeneous uptake of H2O2 on aerosols to the aerosol-phase H2O2, we calculated the average rate of heterogeneous uptake based on Eqs. (S6) to (S11) in the Supplement. Details about each parameter were introduced in Table 3, and the heterogeneous uptake rate of H2O2 (d[X]pt,h/dt) averaged 0.02 ng µg-1 h-1.

As HO2 radical is a precursor of H2O2, the heterogeneous uptake of HO2 onto aerosols may also contribute to the formation of the aerosol-phase H2O2. We assumed that the reactive uptake coefficient of HO2 to aerosol particles was 0.2, and the product of HO2 was H2O2 (Li et al., 2019). At the same observation site in winter of 2017, HO2 concentration for noontime averaged (0.4±0.2)×108 and (0.3±0.2)×108 cm-3 on clean and polluted days, respectively (Ma et al., 2019). Since HO2 level data in 2018 was not available, we used the level of HO2 on clean days in winter of 2017 for calculations, and the average was about 0.2×108 cm-3 at daytime. The heterogeneous uptake rate of HO2 on aerosols was calculated in the same way as H2O2, and the formation rate of the aerosol-phase H2O2 by reactive uptake of HO2 averaged 0.22 ng µg-1 h-1 all day.

It was demonstrated that the concentration of H2O2 in the extracted solution first increased rapidly, then reached peaks at distinct hours that depended on the specific sample, and finally declined gradually over time. However, interestingly, there was large sample-to-sample variation, with samples classifiable into three types in terms of the change in trend and evolution duration (Fig. 6 and Table 4). The third type (Fig. 6c) occurred when the H2O2 level exhibited a steady decline without a growth stage within 13 h, and this was the case with samples 5 and 6 on a slightly polluted day on 2 January 2019. Samples 1 and 2 on 29 December 2018 during clean days belonged to the first type (Fig. 6a), during which H2O2 rapidly grew within 5 h and subsequently decreased at a slow rate over 25 h. The evolution trends of H2O2 in the second type (samples 3 and 4, Fig. 6b) during clean days from 31 December 2018 to 1 January 2019 were similar to those of the first type, except H2O2 approached its peak at about 40 h over the whole analysis process lasting for about 300 h. Given that the H2O2 concentration increased in the extracted solution as time went, the effects of the extraction and transportation processes on the effective gas-aerosol partitioning coefficient of H2O2 were discussed, as shown in the Supplement.

To seek the reasons for the elevated level of H2O2 in the extracted solution, we compared the ratio of the maximum (Cmax⁡) to initial (C0) H2O2 concentrations in the extracted solution with the molar concentration ratio of the aerosol-phase TPOs to H2O2 and found that the ratios of Cmax⁡/C0 and TPOs/H2O2 were in the same order of magnitude for the first and second types, as exhibited in Table S5 in the Supplement. This result provided evidence that part of the aerosol-phase H2O2 originated from the decomposition/hydrolysis of organic peroxides, as described in earlier studies (Wang et al., 2011; Li et al., 2016). In the second type, the concentration of TPOs normalized to aerosol mass reached a maximum, indicating that the second type had more TPO sources, which consequently caused higher TPOs/H2O2 and Cmax⁡/C0 ratios compared with the first type. Furthermore, the aerosol surface is semi-liquid or liquid under high RH (Liu et al., 2017), which provides reaction sites for the decomposition/hydrolysis of aerosol-phase organic peroxides. Aerosol-phase organic peroxides could decompose into H2O2 before the particle aerosols were collected (Zhao et al., 2018). Thus, the decomposition/hydrolysis of organic peroxides in the extracted solution could be applied to the ambient particles. The average rates of the decomposition/hydrolysis of organic peroxides to H2O2 in the rising stage for the first and second types were 0.01 and 0.10 ng µg-1 h-1, respectively.

The three types of samples were in accordance with the growth process of PM2.5. According to meteorological parameters and trace-gas data (Table S6 in the Supplement), static weather conditions gradually formed and were accompanied by lower wind speed, lower ozone level, higher RH, and higher gaseous pollutants levels. The mass concentration of PM2.5 increased from 13.45 to 63.11 µg m-3. In addition, the mass concentration of TPOs also showed a rising trend, whereas the level of TPOs normalized to aerosol mass increased at first and decreased afterwards due to the rapid growth of PM2.5. Because of the consumption of reactive TPOs, which formed SO42- during polluted days, the rest of the TPOs were stable organic peroxides that could not easily decompose into H2O2, e.g. peroxide esters (ROOR). The ratio of TPOs/H2O2 in the third type, collected on a slightly polluted day, was close to that in the second type on clean days, but a rising trend of H2O2 in the extracted solution could not be observed. It was calculated that the ratios of decomposable TPOs to total TPOs for the three types were 29 %, 98 %, and 0 %, respectively.

Recently, it was reported that organic peroxides accounted for a large proportion of secondary organic aerosol (SOA) mass, varying widely from less than 20 % to 60 % (Docherty et al., 2005; Li et al., 2016; Gong et al., 2018). Peroxy radicals also played important parts in the formation of highly oxygenated molecules (HOMs) via an autoxidation mechanism, which can form aerosols without sulfuric acid nucleation (Kirkby et al., 2016). The thermal decomposition of peroxide-containing SOAs and HOMs contributed to the formation of aerosol-phase H2O2 (Krapf et al., 2016). A similar phenomenon was also found by Li et al. (2016), in which the decomposition/hydrolysis of organic peroxides sustainably generated H2O2 accompanied by the attenuation of TPOs in the extracted solution, and about 18 % of gaseous organic peroxides underwent the heterogeneous decomposition on aerosols into H2O2. The decomposable organic peroxides are often peroxycarboxylic acids (PCAs, e.g. peroxyacetic acid, PAA; peroxyformic acid, PFA) and α-hydroxyalkyl-hydroperoxides (α-HAHPs, e.g. hydroxymethyl hydroperoxide, HMHP).