Kinetics and impacting factors of HO 2 uptake onto submicron atmospheric aerosols during the 2019 Air QUAlity Study (AQUAS) in Yokohama, Japan

. HO 2 uptake kinetics onto ambient aerosols play We to precisely investigate the values of the HO reactivities of ambient gases and aerosols and (ii) HO 2 uptake coefﬁcients onto ambient aerosols ( γ ) during the 2019 Air QUAlity Study (AQUAS) in We identiﬁed the effects of individual chemical components of ambient aerosols on γ . The results were veriﬁed in laboratory studies on individual chemical components: transition metals play a key role in 2 and

Abstract. HO 2 uptake kinetics onto ambient aerosols play pivotal roles in tropospheric chemistry but are not fully understood. Field measurements of aerosol chemical and physical properties should be linked to molecular-level kinetics; however, given that the HO 2 reactivity of ambient aerosols is low, traditional analytical techniques are unable to achieve this goal. We developed an online approach to precisely investigate the lower-limit values of (i) the HO 2 reactivities of ambient gases and aerosols and (ii) HO 2 uptake coefficients onto ambient aerosols (γ ) during the 2019 Air QUAlity Study (AQUAS) in Yokohama, Japan. We identified the effects of individual chemical components of ambient aerosols on γ . The results were verified in laboratory studies on individual chemical components: transition metals play a key role in HO 2 uptake processes, and chemical components indirectly influence such processes (i.e., by altering aerosol surface properties or providing active sites), with smaller particles tending to yield higher γ values than larger particles owing to the limitation of gas-phase diffusion being smaller with micrometer particles and the distribution of depleting species such as transition metal ions being mostly distributed in accumulation mode of aerosol. The modeling of γ utilized transition metal chemistry derived by previous studies, further confirming our conclusion. However, owing to the high NO concentrations in Yokohama, peroxy radical loss onto submicron aerosols has a negligible impact on O 3 production rate and sensitivity regime. Jacob, 2000). However, the observed HO 2 concentration in field measurements has not yet been fully explained by sophisticated models (known as the "HO x dilemma") (Stone et al., 2012;Creasey et al., 1997;Kanaya et al., 2007b;Whalley et al., 2010;Millán et al., 2015), although it can be mostly solved in the conditions of clean marine air where NO concentration is low or aerosol loading is low enough to make the heterogeneous reaction of HO 2 unimportant (Sommariva et al., 2004;Kanaya et al., 2007a). Owing to the short atmospheric lifetime of HO x (= OH + HO 2 + RO 2 ), the HO x reactivity measurement can provide a robust test of its complex chemistry (Heard and Pilling, 2003). The HO 2 uptake kinetics onto ambient aerosols, including HO 2 reactivity (k a ) and uptake coefficient (γ ), influence many atmospheric processes, including ozone formation rate, ozone formation sensitivity to NO x , and H 2 O 2 formation (Sakamoto et al., 2019;Thornton et al., 2008). With γ > 0.1, HO 2 concentration can also be influenced under conditions such as low [NO] or high aerosol loading (Lakey et al., 2015;Mao et al., 2013b;Martínez et al., 2003;Tie et al., 2001;Jacob, 2000;Matthews et al., 2014). These effects make the HO 2 uptake onto ambient aerosols indirectly influence human health and climate change.
To our knowledge, aside from us, only one study has measured γ , using an offline method that integrated ambient aerosols over size and time (Taketani et al., 2012). Considering that the offline method may distort γ , we developed an online approach to evaluate HO 2 uptake kinetics onto ambient aerosols. This method was successfully applied in Kyoto, Japan, in summer 2018, using a versatile aerosol concentration enrichment system (VACES) and a technique combining laser-flash photolysis with laser-induced fluorescence (LFP-LIF) (Zhou et al., 2019b). The obtained average γ value (∼ 0.24) was comparable with the previous values used for modeling studies (∼ 0.2) (Stadtler et al., 2018;Jacob, 2000). However, the large standard deviation (± 0.20, 1σ ) of γ along with the measurement time suggest that many other parameters might play a role, e.g., the measurement setup, aerosol characteristics, T , and RH.
In this study, we chose Yokohama (Japan), a coastal city with higher pollutant levels than Kyoto and different properties of the air masses from mainland Japan and the coast, as the measurement site. This is part of the Air QUAlity Study (AQUAS) campaign. The chemical and physical properties of ambient aerosols were quantified in real time. To test their influence on k a and γ , we conducted correlation matrix analysis coupled with the bootstrap method and classified the arriving air masses from different directions. Further, the main mechanism of γ was investigated by comparing the real-time quantified γ values with the modeled values. The impact of the peroxy radical's loss onto ambient aerosols on air quality is evaluated through their impact on ozone formation rates and their sensitivity to NO x . The results obtained here will better estimate the heterogeneous reaction between HO 2 and ambient aerosols in sophisticated air quality models.

Sampling sites
The measurement campaign was conducted at the Yokohama Environmental Science Research Institute in Yokohama, Japan (location: 35 • 28 52.8 N, 139 • 39 30.3 E), from 24 July to 3 August 2019. The sampling ports of the instruments were placed approximately 25 m above the ground. Figure S1 shows the air mass directions during the campaign, which can be classified into two categories: (i) from the sea to the north, toward Yokohama (∼ 19 % of the experimental period: from 12:00 25 July to 12:00 27 July 2019), and (ii) from the mainland toward Yokohama (∼ 81 % of the experimental period). This classification was intended to distinguish the chemical properties of aerosols arriving from the mainland and the ocean and to consequently quantify their impacts on k a and γ .
2.2 Measurement strategies, instrumentation, and related data analysis

LFP-LIF
In situ ambient air HO 2 reactivity was measured using LFP-LIF, which was adapted from a laser-induced pump and probe OH reactivity measurement technique. This approach has been successfully employed for gas-and aerosol-phase HO x (= OH + HO 2 ) reactivity measurements (Sadanaga et al., 2004;Miyazaki et al., 2013;Sakamoto et al., 2018). Further details concerning the HO 2 reactivity measurements are given in the Supplement.

VACES
To compensate for the relatively low ambient aerosol concentrations and thus the low k a and the low limit of detection (LOD) for the HO 2 reactivity measurement (∼ 0.003 s −1 with 240 decay integrations), a setup with VACES and an auto-switching aerosol filter were used before LFP-LIF. The VACES was built according to Sioutas et al. (1999), and the ambient air sample was drawn into a tank (containing ultrapure water heated up to ∼ 32 • C) of VACES through a PM 2.5 cyclone at a flow rate of over ∼ 100 L min −1 , where the ambient air stream was saturated and subsequently cooled down in a condenser connected above the tank (with a temperature of 2 • C). During this process, the water droplets with diameters > 2 µm formed on the collected ambient aerosols, which were then enriched by a virtual impactor (with a 50 % cutoff point less than 1 µm) and dried by passing through a diffusion dryer connected right after the condenser in sequence. The concentration enrichment of the ambient aerosols can be estimated using the total intake flow of VACES and the minor output flow of the virtual impactor that connected to the aerosol instrumentations (more details are given in the Supplement: the enrichment of the ambient aerosols). Wang et al. (2013Wang et al. ( , 2014 claimed that when using the same technique as VACES for the online measurement of copper in ambient aerosols, equivalent copper concentrations were obtained compared to those measured by an inductively coupled plasma mass spectrometer (ICP-MS) for both total and water-soluble components, which indicates that the impact of the VACES system on the solubility of Cu contained in ambient aerosol is negligible. Furthermore, previous studies found that the liquid-liquid phase separation RH ranged from 60 % to 100 % in atmospherically relevant particles consist-ing of organic species and inorganic salts (Yu et al., 2014), and the organic component appears to be the most useful parameter for estimating the liquid-liquid phase separation, which was always observed for oxygen-to-carbon elemental ratio (O : C < 0.5 and was never observed for O : C ≥ 0.8 (Bertram et al., 2011). In this study, the ambient aerosol O : C ranged from 0.1 to 0.7, with the RH changing from ∼ 80 % (in ambient air) to > 100 % (in the water tank) and then to ∼ 75 % (in the reaction cell), suggesting that the phase separation may have already happened before entering the VACES system, and thus we assume that the morphology of the ambient aerosols did not change during the concentration enrichment process. Unfortunately, we did not measure the chemical composition after the VACES, and thus we are not able to compare the chemical composition of the post-VACES aerosols to ambient aerosol. However, previous tests using the ambient aerosol fractions including coarse and fine PM indicated that the enrichment process of the VACES system does not differentially affect the chemical composition of ambient PM (Kim et al., 2001), and thus we assume that the chemical composition changing due to the enrichment process of the VACES can be neglected. The enrichment factor of the ambient aerosol surface area (E) was calculated from the difference between the surface areas measured before and after VACES by two scanning mobility particle sizers (SMPSs).

Aerosol physical properties and the enrichment factor of VACES
The mass concentration and surface area of ambient aerosols (before VACES) were determined using a SMPS 1 (model 3936L72, TSI, measure particle size distribution at 14.1-736.5 nm, 5 min intervals). The mass concentration of PM 2.5 was measured using a palm-sized optical PM 2.5 sensor (Nakayama et al., 2018). In order to test the enrichment factor of the VACES, a SMPS 2 (model 3936L75, TSI, measure particle size distribution at 14.6-661.2 nm, 5 min intervals) was used to measure the enriched mass concentration and surface area of ambient aerosols (after VACES) for ∼ 2 h every day for 6 d. The enrichment factor of VACES for the surface area was estimated as 12.5 ± 2.5 from the ratio between S 2 and S 1 , where S 2 and S 1 are the averaged surface areas measured by SMPS 2 and SMPS 1 of each day, respectively. According to the test from previous study of the VACES system, there is no distortion of the size distribution of the original ultrafine aerosols as the particle concentration enrichment occurs without any coagulation (Sioutas et al., 1999); here, we listed the mean radius and geometric standard deviation (Geo. SD) of the ambient aerosols before and after VACES during the enrichment factor measurement periods, as shown in Table 1. We could see that the mean radii before and after VACES are not statistically different within the standard deviation. 111.40 ± 8.21 1.01 ± 0.02 115.32 ± 6.26 1.01 ± 0.03 * These represent the time period of the enrichment factor measurements; ± represents the standard deviation of the averaged values of mean radius and Geo. SD.
The enriched surface area of ambient aerosols with aerodynamic diameter < 0.74 µm (PM 0.74 ) was calculated from the surface area of ambient aerosol measured by SMPS 1 and the enrichment factor. The enriched surface area of PM 2.5 was then calculated by multiplying the enriched surface area of PM 0.74 by the mass ratio between PM 2.5 and PM 0.75 (∼ 1.1), where we assume that the surface areas are increased in proportion to the mass concentration. However, as the larger particles (here referring to particles ranging from 0.74 to 2.5 µm) tend to have a lower surface area than the smaller particles, we consider the obtained enriched surface area of PM 2.5 to be the upper-limit value. More details can be found in the Supplement.

HO 2 uptake kinetics
After passing through the VACES system, the ambient air was sampled using a three-port valve (Bolt, Flon Industry Co., Ltd) and injected into the LFP-LIF system. The valve was switched automatically between two sampling lines, one with the aerosol filter on and the other one with the aerosol filter off, and HO 2 reactivities in ambient air caused by two modes were measured: (a) the gas-phase mode with the aerosol filter on, where the HO 2 reactivities are represented as k g , and (b) the gas + enriched aerosol-phase mode with the aerosol filter off, where the HO 2 reactivities are represented as k g + Ek a , where E represents the enrichment factor of k a , Ek a represents the total HO 2 reactivities caused by enriched ambient aerosols, and the usage of Ek a is based on the assumption that HO 2 uptake with aerosol particles follows the pseudo-first-order rate law. We modeled k g in both modes using a theory identified previously (see Supplement: HO 2 reactivity of the ambient gas phase) (Zhou et al., 2019b) and compared it with the measured values. The differences between measured and modeled k g in mode (a) enabled us to establish their interrelationship and to check instrument stability. The differences between (k g +Ek a ) and the modeled k g in mode (b) are considered the enriched aerosol-phase HO 2 reactivity (Ek a ). The total HO 2 reactivity decay profile follows single-exponential decay: where k bg denotes the zero air background obtained by injecting zero air with the same RH as the real-time ambient value into the reaction cell every 24 h for 30 min. The RH was controlled by passing some of the zero air through a water bubbler. The value of k bg was subtracted separately on each day. The variability of k bg (i.e., the reproducibility of the laser system) was calculated as the standard deviation of the response of repeated measurements on different days. It was found to be ∼ 4 %, which is slightly higher than the instrument precision (3 %). A 30 min average calculation was applied to the data to reduce data fluctuation. The observed HO 2 uptake coefficients onto ambient aerosols (γ obs ) can be calculated from the dependence of Ek a on γ obs : where ES and ω HO 2 represent the enriched surface area of ambient aerosol after VACES and the mean thermal velocity of HO 2 (∼ 437.4 m s −1 ), respectively. The uncertainty of the enriched surface area was estimated from the instrument systematic error of SMPS (∼ 8 %) and the uncertainty of the enrichment factor (± 2.5), which are shown in Fig. 1b (see Supplement). The HO 2 reactivity of ambient aerosol (k a ) can be obtained from Ek a by dividing by the enrichment factor E.

High resolution-time of flight-aerosol mass spectrometry (HR-ToF-AMS)
A field-deployable HR-ToF-AMS (Aerodyne Research Inc.) (DeCarlo et al., 2006) was used for the characterization of the non-refractory aerosol mass with a time resolution of ∼ 3 min. The HR-ToF-AMS measured the total organic aerosol (OA), sulfate (SO 2− 4 ), nitrate (NO − 3 ), ammonium (NH + 4 ), chloride (Cl − ), and two most dominant oxygencontaining ions in the OA spectra, i.e., mass-to-charge ratios of m/z = 44 (Org44, mostly CO + 2 ) and m/z = 43 (Org43, mainly C 2 H 3 O + for the oxygenated OA and C 3 H + 7 for the hydrocarbon-like OA) (Ng et al., 2011). The fractions of Org44 and Org43 in OA are represented as f 44 and f 43 , respectively. Ambient air was sampled through a critical orifice into an aerodynamic lens, which efficiently transmitted particles between 80 nm and up to at least 1 µm. Particles were flash-vaporized by impaction on a resistively heated surface (∼ 600 • C) and ionized by electron ionization (70 eV). The m/z values of the resulting fragments were determined using a ToF mass spectrometer. Data were analyzed using the ToF-AMS software SQUIRREL and PIKA. Data were not corrected for lens transmission efficiency. Standard relative ionization efficiencies (RIEs) were used for organics (RIE = 1.4), nitrate (RIE = 1.1), chloride (RIE = 1.3), sulfate (RIE = 1.12), and ammonium (RIE = 4). Concentration data were obtained from background-subtracted stick-mass data (low-mass-resolution-base mass concentration data, which are calibrated using ammonium sulfate particles) and determined assuming a collection efficiency (CE) of 0.5.

Filter-based photometer
Real-time measurement of the equivalent black carbon (eBC) was performed using a five-wavelength dual-spot absorption photometer (MA300, AethLabs, San Francisco, CA, USA), which performed an online correction for possible artifacts resulting from filter loading and multiple scattering (Drinovec et al., 2015). In this study, eBC data obtained from light attenuation at a wavelength of 880 nm were used to avoid possible contributions from brown carbon; the time resolution was ∼ 1 min.

Trace elements
Fourteen trace elements (Al, V, Cr, Mn, Co, Ni, Cu, Zn, As, Se, Sr, Cd, Ba, and Pb) were measured using an offline method at 2 d intervals from 21 July to 5 August 2019. The suspended particulate matter (SPM) was collected onto 623.7 cm 2 -sized quartz fiber filters (Pallflex Tissuquartz 2500QAT-UP), which had an available collecting area of 405.84 cm 2 using a high-volume sampler (1000 L min −1 ). Approximately 2 cm 2 of each filter was cut into pieces and placed into a polytetrafluoroethylene (PTFE) pressure digestion tank with 1 mL 49 % hydrofluoric acid (HF) and 5 mL 69 % nitric acid (HNO 3 ). A Thermo Fisher X2 Series ICP-MS was then used to determine metal concentrations. By assuming that the metal fractions were the same in SPM and PM 1 (aerosol particles with aerodynamic diameters less than 1 µm), the concentrations in PM 1 were estimated according to the tested metal concentrations in SPM and the ratio between SPM and PM 1 measured in situ.

Water-soluble inorganic species
NR-PM 1 water-soluble inorganic species (including Na + , SO 2− 4 , NH + 4 , NO − 3 , Cl − , Ca 2+ , K + , and Mg 2+ ) used for the ISORROPIA-II model were also measured using the offline method, as described above. For extraction, we cut one-fourth of a 47 mm filter punched from the original collected filter and placed it in 10 mL of ultrapure water (18.2 MW cm −1 ) in a centrifuge tube. This was followed by 15 min of ultrasonication in a 30 • C water bath. The solution was then vortexed (Vortex Genie 2, Scientific Industries, USA) for 1 min to ensure homogeneity and filtered through a syringe filter with a pore size of 0.45 µm (Advantec Dismic-25, PTFE). An Ion Chromatograph (IC, ICS1600, DIONEX, USA) was employed to determine the concentrations of these inorganic ions in the extracted solution.
2.2.9 Gas-phase monitors NO 2 was measured by cavity-attenuated phase shift (CAPS, Aerodyne Research, USA, at 1 s intervals), NO y -NO by chemiluminescence (model 42i-TL, Thermo, at 10 s intervals), CO by a Thermo CO analyzer of nondispersive infrared spectroscopy (model 48i-TLE, Thermo Scientific, USA, at 10 s time intervals), and O 3 by UV absorption (model 1150, Dylec, AMI Co., Ltd, at 10 s time intervals). HCHO was determined by high-performance liquid chromatography (HPLC; 1260 Infinity, Agilent Technologies Inc, USA) from 14:00 29 July to 12:00 3 August 2019. An average value of ∼ 2 ppb was used for the data analysis.

ISORROPIA-II model
NR-PM 1 water-soluble inorganic species (including Na + , SO 2− 4 , NH + 4 , NO − 3 , Cl − , Ca 2+ , K + , and Mg 2+ ) and meteorological parameters including temperature and RH were used to calculate the aerosol pH and liquid water content based on the ISORROPIA-II model (Fountoukis and Nenes, 2007). We ran ISORROPIA-II in "reverse" mode, and the particles were assumed to be deliquescent, i.e., in metastable mode (Hennigan et al., 2015). The thermodynamic equilibrium of the NH + 4 -SO 2− 4 -NO − 3 system case was used for modeling.

Results and discussion
3.1 The HO 2 uptake kinetics onto ambient aerosols The measured total HO 2 reactivities were compared against the modeled gas-phase HO 2 reactivity under the experimental conditions both with and without the aerosol phase. Without the aerosol phase, the modeled k g values are calculated according to the description in Sect. 2.2 and are not statistically different with the measured k g values (Fig. 1a second panel, t test, p = 0.49, with inspection level = 0.05), indicating that HO 2 loss in the reaction cell was dominated by its reaction with NO 2 in the LFP-LIF system. With the aerosol phase, the measured (Ek a + k g ) and modeled values (≈ k g ) were significantly different (see Fig. 2b, first panel, t test, p = 0.04, with inspection level = 0.05). The differences were considered to be the HO 2 reactivities of enriched ambient aerosols (Ek a ). Ek a ranged between 0.015 s −1 (25th percentile) and 0.097 s −1 (75th percentile), with a median value of 0.059 s −1 , the corresponding k a , calculated by dividing Ek a by E, ranging between 0.001 s −1 (25th percentile) and 0.008 s −1 (75th percentile), with a median value of 0.005 s −1 and an average value of 0.005 ± 0.005 s −1 .
The error for Ek a was estimated to be ∼ 0.05 s −1 , calculated as the propagated error from k g +Ek a (i.e., the systematic error of the instrument, ∼ 0.05 s −1 ) and the modeled k g in mode (b) (∼ 0.001 s −1 ). The error for k a was then estimated as ∼ 0.004 s −1 by dividing the error of Ek a by the enrichment factor E. The corresponding γ , calculated from Eq.
(2), ranged from 0.05 (25th percentile) to 0.33 (75th percentile), with a median value of 0.19 (with an average value of 0.23 ± 0.21). The mean diameters of ambient particles ranged from 0.1 to 0.46 µm (with a median value of 0.25 µm), and the gas-phase diffusion effects on γ were estimated to be ∼ 6.6 % (further details are given in the Supplement). The absolute increase in γ due to the gas-phase diffusion is 0.03 on average, which is negligible compared to γ uncertainty (∼ 0.21 on average). Therefore, we ignored the gas-phase diffusion effects on γ .
Statistical significance analysis showed that the average γ value of group i (0.35 ± 0.28) is significantly higher than that of group ii (0.21 ± 0.16) (calculated p = 4.9E-5; Mann-Whitney), indicating that the air masses from the ocean yield higher γ values than the air masses from mainland Japan. The difference in γ values between group i and group ii may be due to the different chemical components contained in the ambient aerosols arriving from the ocean or mainland, which we will discuss in the following sections. The average value of k a at Yokohama (0.005 ± 0.005 s −1 ) was much higher than that at Kyoto (0.0017 ± 0.0015 s −1 ) (with calculated p < 0.05; Mann-Whitney); this may be due to the different aerosol properties in Kyoto and Yokohama. We list some of them as follows. (1) Mass composition: the aerosols at the coastal city (Yokohama) tend to contain more sea salts and thus increased k a . (2) Particle size distribu-tion: smaller particles tend to yield higher γ values than larger particles owing to the depleting species (e.g., transition metal ions) being mostly distributed in the accumulation mode of aerosol. (3) The water content and the metal concentrations will highly influence the HO 2 uptake capacity of the ambient aerosols. However, the average value of the HO 2 uptake coefficient onto ambient aerosols (γ ) at Yokohama is ∼ 0.23, which is comparable with previous measured (∼ 0.24-0.25) (Zhou et al., 2019b;Taketani et al., 2012) and modeled (∼ 0.20) values (Stadtler et al., 2018;Jacob, 2000). The large standard deviation (± 0.21, 1σ ) of the values along with the measurement time may be due to the instantaneously changed chemical and physical properties of ambient aerosols, indicating that a large bias may exist if a constant γ value is used for modeling. Figure 2a shows the time series of the mass concentrations of OA, SO 2− 4 , NO − 3 , NH + 4 , Cl − , and eBC in PM 1 in Yokohama from 24 July to 2 August 2019, which is ∼ 1.5 d less than for the LFP-LIF data. During this period, PM 1 ranged from ∼ 1 to 35 µg m −3 (average ≈ 13 µg m −3 ) and was dominated by OA, SO 2− 4 , and NH + 4 , with contributions of 39 ± 11 %, 30 ± 12 %, and 12 ± 4 %, respectively; these were followed by eBC and metals, with contributions of 10 ± 7 % and 8 ± 8 %, respectively. Cl − contributed < 1 % in both groups, which is similar to that reported for an urban area in winter in Bern (Switzerland) (Zhou et al., 2019a). However, NO − 3 contributed much less (∼ 2 ± 0.7 %) compared with that reported for Bern (∼ 19 ± 4 %), which may be due to the reverse reaction of NH 4 NO 3 converting to HNO 3 . Since Yokohama is a coastal city and HNO 3 is easily vaporized in summer, gaseous HNO 3 may sink with sea salt particles by forming NaNO 3 through heterogeneous reactions (Finlayson-Pitts and Pitts, 2000). Figure 2b shows the average contribution fractions of different components of group i and group ii. The main differences in the components between these two groups are the fractions of OA, BC, SO 2− 4 , and metals. The OA fraction was ∼ 1.8 and ∼ 8.4 times higher than that for the metals in groups i and ii, respectively. As OA can cover the surface of the particles and thereby decrease γ (Lakey et al., 2016a;Takami et al., 2013), the difference between the OA and metal fractions in these two groups may partially explain the much higher γ values of group i (vs. group ii). Previous studies have shown a quite low HO 2 uptake coefficient on BC (∼ 0.01) (Saathoff et al., 2001;Macintyre and Evans, 2011), which is different from the result obtained here. This may be due to the much higher fraction of BC in group i (vs. group ii) providing active sites for HO 2 self-reaction or its reaction with the H atom from the abstraction reaction from hydrogen-containing functional groups and produce H 2 O 2 (Bedjanian et al., 2005), or BC can be coated with additional materials (e.g., sulfate and organic carbon), thus influenc- Figure 1. Temporal variation of parameters under different experimental conditions. (a) Without aerosol phase: first panel: measured NO 2 concentrations (ppb); second panel: measured (red line) and modeled (black line) k g ; third panel: fitting residues of modeled k g values, ranging from −0.04 (25th percentile) to 0.05 (75th percentile). Therefore, we consider the systematic error of the LFP-LIF instrument to be ∼ 0.05 s −1 . (b) Gas + aerosol phase: first panel: measured total HO 2 reactivity (k g +Ek a ) and modeled k g ; second panel: Ek a , calculated from the difference between the measured and modeled values from the first panel; the gray shadow area represents the uncertainty of Ek a ( Ek a ), propagated from the error of (k g +Ek a ) and modeled k g ; third panel: the upper-limit surface area of the enriched ambient aerosols (ES): the gray shadow area represents the uncertainty of ES ( ES), propagated from the systematic errors of the SMPS instrument (∼ 8 %), and the uncertainty of the enrichment factor; fourth panel: γ calculated from Ek a and ES according to Eq. (2). The errors for γ were propagated from Ek a and ES, γ = γ × Eka Eka 2 + ES ES 2 . The blue shaded area represents the air masses from group i (from the coast), and the remainder is from group ii (from the mainland).

Bulk chemical composition of ambient aerosols
ing HO 2 uptake (Schwarz et al., 2008). We also observed a slightly higher Cl − and BC fraction in group i (from the ocean) than that in group ii (from the mainland), which may be due to the effects of sea salt and the ship emissions in the air mass from the ocean, respectively. From the average diurnal patterns (Figs. S5 and S6), the trends in k a follow the trends in chemical composition, whereas γ shows a contrasting trend with both variables in both groups. For group ii, SO 2− 4 and OA exhibited higher values, whereas γ exhibited lower values during the daytime than those during the nighttime, indicating that secondary aerosol formation resulting from photochemical reactions may decrease γ . To identify the influence of each individual chemical component of ambient aerosol on k a and γ , we further performed correlation matrix analysis.

Influence of individual chemical components of ambient aerosol on k a and γ
For multiple-component ambient aerosol, k a and γ are influenced by different chemical components, and these chemical components may also have mutual effects on each other. For example, the transition metals Cu and Fe contained in ambient aerosols can be chelated by organics (Lakey et al., 2016b). Therefore, we produced a Pearson correlation matrix of all the testing factors at Yokohama, including different chemical components, k a and γ . Here we note that the different chemical components were measured using HR-ToF-AMS for ambient aerosols with aerodynamic diameters < 1 µm, while k a and γ were measured using the VACES-LFP-LIF system for ambient aerosols with aerodynamic diameters < 2.5 µm, but due to most "fine-mode" aerosols having a mean diameter ranging from 0.09 to 0.47 µm (with a median value of 0.25 µm, measured by SMPS), we assume that the chemical components of ambient aerosols with the aerodynamic diameter ranging between 1 and 2.5 µm have a negligible impact on the Pearson correlation matrix result. However, the present results do not include the effects of coarse particles (with aerodynamic diameters > 2.5 µm) on the HO 2 uptake kinetics, and we may partially miss measuring sea spray (with diameters ranging from ∼ 0.05 to 10 µm) effects. When Cl − measured by AMS increased, coarse particles may exist and our results may not represent the real ambient conditions. Consequently, we consider our results to be the lower limit of the HO 2 uptake kinetics onto real ambient aerosols.
To exclude the effects of the different fractions of chemical components in groups i and ii, the bootstrap method, which is based on the creation of replicating the inputs by perturbing the original data through resampling, was employed. The resampling was performed by randomly reorganizing the rows of the original time series such that some rows of the original data were present several times while other rows were removed. The final results were obtained by running the data for 1000 bootstrap replicates. The average values of these 1000 bootstrap replicates are listed in Fig. 3.
Most of the chemical components had strong or moderate Pearson correlations with each other (Fig. 3), although k a and γ showed only moderate correlations with each other (0.56). As γ can be correlated with the qualitative, rather than quantitative, properties of the aerosols, we further investigated Pearson's r values between the chemical composition fractions and γ . The results show that k a was positively correlated with total mass and the individual chemical components, whereas γ showed only a weak positive correlation with f metals (0.30) and f eBC (0.18). According to previous studies, metals may act as a catalyst, thus accelerating the depletion of HO 2 (Mao et al., 2013a), and BC can provide active sites or can be coated by other chemical components, thus facilitating the HO 2 uptake (Bedjanian et al., 2005;Schwarz et al. (2008), as described in Sect. 3.2. The very weak correlation of γ with f − Cl (0.04) may be related to Cl − chemistry: for example, HO 2 (g) can react with NaCl(g), produce NaOH and Cl 2 (g), and thus cause a decrease in the HO 2 concentration and indirectly affect γ (Remorov et al., 2002). Interestingly, when considering the Org44 and Org43 fractions in total OA separately, γ is positively correlated with f 43 (0.18) but negatively correlated with f 44 (−0.24). This is consistent with the previous conclusion that more oxidized organic aerosols tend to be highly viscous and thus decrease HO 2 uptake coefficients (Lakey et al., 2016b). In summary, γ was dominated by the free forms of transition metals that can act as catalysts of HO 2 uptake onto ambient aerosols and was indirectly affected by chemical components that might alter the properties of ambient aerosols; e.g., oxygenated OA can cover the aerosol surface and alter the viscosity of ambient aerosols, thereby decreasing γ (Lakey et al., 2016a, b;Takami et al., 2013), whereas BC may provide active sites or be coated by other chemical components, thereby increasing γ . This is further confirmed by the classification of the air masses; i.e., the air mass from the ocean (group i), which contained less OA and more metals than that from mainland Japan (group ii), had a higher HO 2 uptake capacity. We further compared the measured γ values with the modeled γ values using previously proposed mechanisms, as shown below.

Possible mechanism of HO 2 uptake onto ambient aerosols
Two mechanisms of HO 2 uptake onto aqueous ambient aerosols have been proposed, for which equations have been derived from a previous study (Thornton et al., 2008), (i) HO 2 -only chemistry and (ii) chemistry with transition metals playing a role. In this study, the liquid content of the total ambient aerosol mass ranged from 70 % to 88 %, as obtained from the ISORROPIA-II model. As the solubility of Fe is rather small in ambient aerosol, the reaction rates of Fe/Mn for liquid-phase HO 2 in aerosol are ∼ 100 times slower than they are for Cu, and thus the influence of Fe and Mn on HO 2 uptake can be neglected compared to Cu or scaled as equivalent [Cu 2+ ] (Fang et al., 2017;Hsu et al., 2010;Baker and Jickells, 2006;Oakes et al., 2012;Song et al., 2020); therefore, we use the soluble Cu as a surrogate for transition metals in ambient aerosols to assess their influence on γ . The Cu solute mass fraction in the liquid content of the ambient aerosols was estimated as (3.5-30) × 10 −4 mol L −1 according to our offline filter test (Sect. 2.2), and to get the effective concentrations of Cu ions which can participate in the reaction of the destruction of peroxy radicals, the activity coefficient for total Cu was assumed to be 0.1 (upper limit) based on a study of (NH 4 ) 2 SO 4 particles at 68 % RH (Ross and Noone, 1991;Robinson and Stokes, 1970). Using copper ions as a surrogate metal for transition metal ions (TMIs), the potential HO 2 loss onto aqueous ambient aerosols via mechanisms involving TMIs was estimated as (Hanson et al., 1994) 1 where α HO 2 is the mass accommodation coefficient of HO 2 , ω is the mean HO 2 molecular speed (cm s −1 ), H eff is the effective Henry's law coefficient, R is the gas constant (J K −1 mol −1 ), and T is the temperature (K). k I is the pseudo-first-order rate constant equal to k II TMI [TMI], where k II TMI is the second-order rate constant for aqueous-phase re-action with HO 2 / O − 2 and TMI. Q accounts for aqueousphase diffusion limitations and is expressed as (4) Table S1 shows more details of the parameters used for modeling. Previous laboratory studies suggest the mass accommodation coefficient for various single-component aerosols doped with Cu(II) is commonly > 0.2 (Taketani et al., 2008(Taketani et al., , 2009Mozurkewich et al., 1987;Thornton and Abbatt, 2005;George et al., 2013;Lakey et al., 2016a, b), and organics substantially reduce HO 2 uptake onto aerosols containing TMI (Lakey et al., 2016b). Here we calculated γ TMI with α HO 2 = 0.2 using Eq. (3), which are plotted in Fig. 4a along with the measured γ values; the much lower variation of the modeled values may be due to the low time resolution (∼ 2 d) of [Cu]. The measured γ values (∼ 0.23 on average) are significantly higher than the modeled γ TMI with α HO 2 = 0.2 (∼ 0.16 on average), with calculated p < 0.05 (t test), which may be due to the TMI contained in the ambient aerosol. However, when using the upper limit of the mass accommodation value for modeling (with α HO 2 = 1), the measured γ values are significantly lower than the modeled γ TMI (averaged value: ∼ 0.50), these results indicating that the chemical components may be internally mixed, as proposed by Takami et al. (2013), which influences the aerosol surface tension and the activity of the free form of the copper ion (i.e., OA and BC) to constrain γ TMI . We suggest that the additional collective effects of different chemical components on α HO 2 and the bulk reactions should be involved in the γ TMI modeling to get accurate estimation. No linear correlation was found between γ TMI and γ . Further classification of measured γ ≥ 0.4 and γ < 0.4 shows that γ TMI has a weak correlation with measured γ values when γ ≥ 0.4 (Fig. S7), which may be due to the higher fraction of metals in the total mass at measured γ ≥ 0.4 (∼ 12 %) than at < 0.4 (∼ 7 %); therefore, the impact of the other chemical components is much lower. The γ values obtained here are comparable with those in previous ambient aerosol studies (Taketani et al., 2008;Zhou et al., 2019b) (Fig. 5b). When compared with single-compound aerosols obtained from laboratory studies, γ values were generally higher than the HO 2 uptake coefficients onto organic species (Lakey et al., 2015), soot particles (Bedjanian et al., 2005), and the dry state of inorganic aerosols (i.e., (NH 4 ) 2 SO 4 , NaCl, and H 2 SO 4 ) but comparable to or lower than aqueous and copper-doped aqueous phases of inorganic species (Fig. 4b) (George et al., 2013;Lakey et al., 2016b;Taketani et al., 2008;Thornton and Abbatt, 2005). This may indicate the collective effects of the individual chemical components of ambient aerosols on γ and the significant influence of RH on aerosol states of singlecomponent particles and thus their HO 2 uptake coefficients.
Other studies have shown that γ is strongly negatively temperature dependent (Remorov et al., 2002;Mao et al., 2010;Cooper and Abbatt, 1996;Hanson et al., 1992;Thornton and Abbatt, 2005;Gershenzon et al., 1995). Here, RH and T were stabilized by the VACES-LFP-LIF system (in the reaction cell) as compared with those in ambient air (Fig. S8), and statistical analysis indicates the RH and T in the reaction cell were on average decreased by 3.3 % (t test, p < 0.05, with inspection level = 0.05) and 2.3 • C (t test, p < 0.05, with inspection level = 0.05) compared to that in ambient air, respectively, which is much smaller than the standard deviation of T and RH (which are ∼ 3.7 • C and 16.4 %, respectively) along with the measurement time. We noticed that k a and γ showed no dependence on RH and T in the reaction cell (see Fig. S9), indicating that the instantaneous change in RH and T may not be the dominating factor in terms of the variation of k a and γ with measurement time, and the stabilizations of RH and T by the VACES-LFP-LIF system have negligible effects on k a and γ . This suggests that the individual chemical components and physical properties of ambient aerosols may dominate γ variation during the field campaign; both the metal-catalyzed reactions and the chemical components and their states should be considered to yield more accurate γ values. Results obtained here are in accordance with previous results on correlations between particulate H 2 O 2 (which can be formed by the uptake of HO 2 ) and coarse particulate transition metals (Wang et al., 2010). Using an offline methodology to investigate the influence of RH and T by maintaining constant experimental conditions or chemical compositions will be the subject of future work.
3.4 Influence of the physical properties of ambient aerosols on k a and γ HO 2 heterogeneous loss rates are driven by the different particle sizes of different aerosol types (i.e., urban ambient aerosols and marine aerosols) (Morita et al., 2004;Guo et al., 2019;Jacob, 2000). In this study, k a and γ showed no linear dependence on the mean ambient particle diameters (see Fig. S10). Identifying the fractional contributions of aerosols in different particle size ranges to k a and γ is highly desirable in terms of understanding their influence. However, it seems that high γ values (> 0.8) occur when the surface area is < 2 × 10 −6 cm 2 cm −3 and the mean particle diameter is < 110 nm. This is in accordance with a previous study showing that aerosols yield the highest fractional contribution to the total heterogeneous loss rate of HO 2 radicals of size < 0.1 µm (Morita et al., 2004) and that the mass accommodation process plays the determining role for smalland medium-sized aerosols in controlling HO 2 uptake. Guo et al. (2019) state that the HO 2 radicals experience less loss upon their diffusion into larger droplets than their diffusion into small droplets due to dilution effects making the larger aerosols have lower depleting species concentrations (Cu 2+ ). However, this was based on the assumption that the total mass of Cu 2+ is constant during the hygroscopic growth of  Thornton and Abbatt (2005).
particles, which is not always true in the ambient conditions. Further studies about Cu 2+ content in particles with different sizes are needed to fully understand the result here.

Significance of k a to O 3 formation potential
In the urban atmosphere, XO 2 (=HO 2 + RO 2 ) fate is important to the photochemical production of ozone (P (O 3 )).
Here, the loss rates of XO 2 due to three factors were compared. (i) Uptake onto the ambient aerosols (L P−XO 2 in Eq. 5); since no experiment or reference is available for RO 2 uptake onto ambient particles, we assume that the RO 2 reactivities caused by its interaction with ambient aerosols were the same as k a . (ii) XO 2 self-reactions (L R−XO 2 in Eq. 6).
(iii) Reaction with NO (L N−XO 2 in Eq. 7), which can produce NO 2 , a precursor of O 3 ; therefore, Eq. (7) can also be regarded as P (O 3 ).
where k HO 2 −HO 2 and k HO 2 −RO 2 are the second-order rate constants of HO 2 self-reaction and its reaction with RO 2 , respectively. k NO−HO 2 is the second-order rate constant of the reaction of HO 2 with NO. The HO 2 concentration was estimated from O 3 concentration using the method described by Kanaya et al. (2007a). The RO 2 concentration is then estimated by assuming a steady state of HO 2 in the HO x cycle; the reaction rates of HO 2 radicals are approximated as 0: where k CO−OH and k H 2 CO−OH are the second-order rate constants of the reactions of CO and H 2 CO with OH, respectively. The different XO 2 loss rates described in Eqs. (5)-(7), along with the measurement times, are shown in Fig. 5a. Generally, L P−XO 2 is much greater than L R−XO 2 , indicating that the XO 2 taken up by ambient aerosols will compete with the XO 2 self-reaction, thus influencing XO 2 concentration. However, such an influence may have a negligible impact on P (O 3 ) because L P−XO 2 is tens of thousands of times lower than L NO−XO 2 owing to the relatively high NO x concentration at Yokohama. We further tested the impact of L P−XO 2 on the ozone formation sensitivity regime, according to the method proposed by Sakamoto et al. (2019), in which L N /Q is used as a new indicator: where k OH−VOCs and k OH−NO 2 are the second-order rate constants of the reactions of OH with VOCs and NO 2 , respectively. k NO−HO 2 is the second-order rate constant of the reaction of NO with HO 2 . α is the proportion of RO 2 in XO 2 . L N is the OH radical loss rate through its reaction with NO 2 (= k OH−NO 2 [OH] [NO 2 ]), and Q is the total loss of the HO x radicals in the HO x cycle reaction (= L N + L P−XO 2 + L R−XO 2 ). The regime transition point can be expressed as where χ = L P−XO 2 /(L P−XO 2 + L R−XO 2 ). The results indicate that both L N /Q and L N /Q_without_aerosol (calculated with and without including k a in Eq. 9, respectively) were higher than L N /Q_transition, indicating that ozone formation was VOC sensitive throughout the campaign and that the aerosol uptake of XO 2 (ka') showed no impact on the O 3 formation regime (see Fig. 5; here we only consider the daytime, when photochemical reactions occur). The plots of L N /Q and L N /Q_transition as a function of NO concentration show the values were closer to L N /Q_transition (∼ 0.4) at lower NO concentrations (Fig. S11); therefore, aerosol uptake may play a more important role in the O 3 formation regime at NO levels lower than those observed in this study. The temporal variations in key factors used in this section are shown in Fig. S12.

Conclusions
This study used a reliable online methodology to investigate HO 2 uptake kinetics onto in situ ambient aerosols (i.e., HO 2 reactivity of ambient aerosols k a and HO 2 uptake coefficients γ ) and discussed the impacting factors on such processes, i.e., chemical compositions and physical properties of ambient aerosols and experimental conditions. k a ranged between 0.001 s −1 (25th percentile) and 0.005 s −1 (75th percentile), with an average value of 0.005 ± 0.005 s −1 . The corresponding γ ranging from 0.05 (25th percentile) to 0.33 (75th percentile), with a median value of 0.19 and an average value of 0.23 ± 0.21, is comparable with previous measured (∼ 0.24-0.25) (Zhou et al., 2019b;Taketani et al., 2012) and modeled (∼ 0.20) values (Stadtler et al., 2018;Jacob, 2000). However, the k a and γ values obtained here are considered to be the lower-limit values for real ambient aerosols, as the coarse particles were not measured in this study. We noticed that k a and γ showed no dependence on RH and T in the reaction cell in this study, indicating that the instantaneous changes in RH and T may not be dominating factors in terms of the variation of k a and γ with measurement time, and the large stan-dard deviation of the γ values along with the measurement time (± 0.21, 1σ ) may be due to the instantaneously changed chemical and physical properties of ambient aerosols; a large bias may exist if a constant γ value is used for modeling. We found that the individual chemical components of ambient aerosols may have collective effects on γ through the analyses of (1) separating the air masses into two groups, group i from the ocean and group ii from mainland Japan; (2) the average diurnal patterns; (3) the correlation matrix analysis of each individual chemical component of ambient aerosol with k a and γ ; and (4) the modeling studies using previously proposed mechanisms. All these efforts clearly indicate that the transition metals contained in ambient aerosols may act as a catalyst, thus accelerating the depletion of HO 2 ; however, they can be chelated by OA. OA can also cover the aerosol surface and alter the viscosity of ambient aerosols, thereby decreasing γ , and more oxidized organic aerosols tend to be highly viscous, thus decreasing HO 2 uptake coefficients. Results obtained here are in accordance with previous laboratory and modeling studies (Mao et al., 2013a;Lakey et al., 2016, b;Takami et al., 2013;Thornton et al., 2008;Hanson et al., 1994). The chemical components of ambient aerosols may be internally mixed, as proposed by Takami et al. (2013), which influences the aerosol surface tension and the activity of the free form of the copper ion (i.e., OA and BC) to constrain γ . In contrast to previous studies saying that BC may shrink HO 2 losses onto ambient aerosols (Saathoff et al., 2001;Macintyre and Evans, 2011;Bedjanian et al., 2005), we found BC positively correlated with HO 2 uptake coefficients (0.18); this may be due to BC being able to provide active sites or being coated by other chemical components, thus facilitating the physical uptake of HO 2 . Here, we observed higher γ values (> 0.8) when the mean particle diameter is < 110 nm, identifying the fractional contributions of aerosols in different particle size ranges to k a and γ as highly desirable in terms of understanding their influence.
In summary, the chemical components and physical properties of ambient aerosols may dominate γ variation during the field campaign; to yield a more accurate γ value, total suspended particles in ambient air should be measured, and the metal-catalyzed reactions, chemical components, and aerosol states should be considered. Also, improvements to the time resolution of metal measurements are needed for more precise analysis. For more detailed investigation of HO 2 uptake mechanisms, an offline methodology that can maintain constant chemical compositions and experimental conditions (such as RH and T ) will be useful. The HO 2 loss onto ambient aerosols was identified as having a negligible impact on the O 3 production rate and formation regime owing to the high NO x concentrations at Yokohama. This process may play a more important role in O 3 formation under low NO x concentrations and high aerosol loading conditions. The results help us to understand the impacts of HO 2 uptake kinetics on chemical transformations in the troposphere. Data availability. Data supporting this publication are available upon request to the corresponding author (junzhou@jnu.edu.cn).
Author contributions. JZ, KM, YS, and YoK designed the experiments and JZ and YB carried them out. JZ did the data analysis and prepared the manuscript with contributions from all the co-authors.
Competing interests. The authors declare that they have no conflict of interest.
Disclaimer. Publisher's note: Copernicus Publications remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.