Kaurene ozonolysis experiments were conducted in a 2 m3 Teflon (FEP, supplied by Vector Foiltec, Germany) chamber, the COALA

Comprehensive molecular characterization of secondary Organic AerosoL formation in the Atmosphere

All instruments sampled the chamber air continuously, and the input into the chamber was varied (Fig. 1a) to achieve different concentrations and ratios of kaurene and O3. The O3 concentrations ranged from ∼70 to ∼110 ppb for different experiments, and the kaurene concentration was at most ∼20 ppb, though most of the time well below 1 ppb (Fig. 1a). Seed aerosols were added at different times (the gray shading in Fig. 1) in order to initiate condensation onto aerosols rather than chamber walls, which is the dominant sink for low-volatility vapors. The total particle mass reached its maximum value of more than 200 µg m-3 in the experiment on the 31 January. The injection of kaurene was one of the most challenging parts of the experiments due to its low volatility. Kaurene is a solid at room temperature, and the kaurene in this study was purchased from OIChemIm s.r.o. Gas-phase kaurene was introduced into the chamber by flushing nitrogen (N2) through a vial containing the solid kaurene, and the flow was subsequently directed into the chamber. In addition, a heater was placed under the vial, heating the air around it to promote the evaporation of kaurene. During the experiments, the vial's bottom temperature stayed below 45 ∘C, except for the last experiment day when the vial reached a temperature around 60 ∘C. Already at low heating, the solid kaurene had melted to liquid form, but we did not see indications of decomposition of the kaurene from the heating in any of our mass spectra. Variable levels of kaurene in the chamber were achieved, either by turning the heater on and off or changing the flow rate of the N2 carrier gas, which was controlled with a mass flow controller (MKS, G series, Andover, MA, USA). Due to the low volatility of kaurene, residual amounts were introduced into the chamber also when no active addition was performed, presumably due to evaporation from surfaces in the tubing and chamber.

We deployed a suite of online instrumentation measuring gas- and particle-phase species. The Vocus PTR-TOF (Tofwerk AG/Aerodyne Research, Inc.) (Krechmer et al., 2018) was applied to determine the reactant kaurene concentration and some of its oxidation products. The instrument allows for fast and continuous measurements of VOCs with sub-ppt detection limits (Krechmer et al., 2018). Based on the previous evaluation, the Vocus can also detect various oxidation products of monoterpenes containing up to around six oxygen atoms (Riva et al., 2019). In this study, the instrument was run with the axial and radial voltages being 350 and 400 V, respectively. The pressure in the ionization regions (drift tube) was set to 1.4 mbar, giving an electric field strength / number density of gas E/N of 120 Td (townsend). However, as discussed in more detail in Sect. 2.3, the pressure drifted during the experiments, resulting in significant changes in the absolute sensitivity of the sampled molecules. Around 4.5 L min-1 of the chamber air was directed into the Vocus through 1 m long PTFE tubing (6 mm o.d., 4 mm i.d.) for further analysis. A water flow (HPLC grade or 18 MΩ Milli-Q water) at the rate of 15 sccm (standard cubic centimeters per minute) was utilized to produce reagent ions in a discharged ion source. The mass-resolving power of the long TOF mass analyzer was 12 000–13 000, and data were recorded with a time resolution of 5 s.

For measuring HOM and other more oxygenated products from kaurene ozonolysis, a chemical-ionization atmospheric-pressure-interface time-of-flight mass spectrometer was used (CI-APi-TOF, Tofwerk AG/Aerodyne Research, Inc.) (Jokinen et al., 2012). We used nitric acid (HNO3) as the reagent and an X-ray source to produce nitrate reagent ions (NO3-). The sample molecules could be charged by collision with nitrate ion clusters (NO3-, HNO3NO3-, and (HNO3)2NO3-). After collisions, the sample molecules were either ionized via a direct proton transfer and detected as a deprotonated ion or, more typically, via clustering with reagent ions and detected as adducts (Jokinen et al., 2012). During the experiments, the instrument, equipped with a long TOF mass analyzer (mass-resolving power 13 000–14 000), was configured to measure ions up to 960 Th with a time resolution of 10 s. The sampling flow rate was around 10 L min-1 for all experiments.

A custom-built scanning mobility particle sizer (SMPS) and a long time-of-flight aerosol mass spectrometer (L-TOF-AMS, hereafter AMS, Aerodyne Research Inc.) were deployed to measure the particle phase. The L-TOF-AMS is similar to the high-resolution TOF-AMS (HR-TOF-AMS) described in DeCarlo et al. (2006) but has a longer TOF chamber for increased mass-resolving power.

Data of both the Vocus and the nitrate CI-APi-TOF were analyzed with a MATLAB tofTools package (version 607) (Junninen, 2014), while the TOF-AMS analysis software packages SQUIRREL (version 1.63H) and PIKA (version 1.23H) (Sueper et al., 2011) within Igor Pro (version 6.37, WaveMetrics Inc.) were used for the AMS data. The SMPS data were analyzed with MATLAB and Igor Pro.

A system for total O3 reactivity measurement (TORM) was also deployed during the experiments. The TORM consists of a reactor (three 2 L borosilicate glass bottles) in which sampled air reacts with a set concentration of O3 (150 ppb in this study) produced by an O3 generator. A modified O3 analyzer (model 49i, Thermo Scientific, Waltham, MA, USA) operating in differential mode recorded directly the difference in O3 before and after the reactor, which is used to calculate the total O3 reactivity in the COALA chamber. The kaurene concentration could be estimated based on the calculated O3 reactivity data at a given rate coefficient of kaurene–O3 reaction, with the assumption that kaurene was the dominant sink for O3. A detailed explanation of the instrument and calculation can be found elsewhere (Helmig et al., 2021). During the experiments, the total flow through the reactor was 6.0 L min-1 with 5.0 L min-1 sampling flow from the COALA chamber and 1.0 L min-1 coming from the O3 generator (dilution factor f=0.83). The residence time in the reactor was estimated to be 60 s.

Multiphase adsorbent tubes (Tenax TA/Carbopack B, 60–80 mesh) (o.d. 1/4 in. × 3 1/2 in.; PerkinElmer Inc., Waltham, MA, USA) were utilized as one way to quantify diterpene concentration in the chamber due to the lack of any authentic diterpene standards for Vocus. We collected four Tenax tubes during the experiments, and details about the sampling parameters can be found in Table S1 in the Supplement. After sampling, a thermal desorption gas chromatograph (TD-GC, TurboMatrix 350 automatic TD unit, Clarus 680 GC, PerkinElmer Inc.) coupled with a quadrupole mass spectrometer (MS, Clarus SQ 8 T, PerkinElmer Inc.) was applied to determine and quantify kaurene in the Tenax samples. More details about the methodology and Tenax tubes have been reported by Helin et al. (2020).

In Vocus, the sampled air is charged in a focusing ion–molecule reactor (FIMR) by combining with the reagent ions produced from a low-pressure discharge reagent ion source. The FIMR is a glass tube with four quadrupole rods installed outside in a radial pattern. The FIMR pressure is controlled by a valve between the reactor and a mechanical pump. The FIMR pressure typically remains constant due to the control where additional pumping is applied to it to keep a set pressure. However, in our study, the pinhole through which the sample air entered the FIMR became partially blocked by seed aerosol particles. When sufficiently blocked, the pressure control could no longer be maintained since even without additional FIMR pumping, the pressure dropped below the desired 1.4 mbar, as the flow through the pinhole had decreased. As the pinhole clogging was not noticed during the experiments, the FIMR pressures kept decreasing in a stepwise pattern from the initial ∼1.4 mbar at the beginning of the experiments to ∼0.9 mbar in the end. The changes in FIMR pressure are expected to impact both the sensitivity and the fragmentation behavior of the Vocus by affecting the frequency and energy of ion–molecule collisions. Hence, the raw signal intensity of kaurene could not directly depict the actual changes of kaurene in the chamber over the course of the experiments. Therefore, we needed to correct the sensitivity changes as far as possible to simulate the actual variation of kaurene using a simple model. For this purpose, we utilized the observed changes in chamber O3 during kaurene additions. The details are described in the following.

The O3 concentration was monitored online for all experiments. When we injected kaurene into an O3-stable chamber, the O3 loss rate increased due to reactions with kaurene (in addition to ever-present flush-out). For any time point, we can write the change in O3 concentration in the chamber as 1dO3dt=QO3,in-O3×[kaurene]×k1-O3×kflushout, where QO3, in is the injection rate of O3; [kaurene] and [O3] are the current kaurene and O3 concentrations, respectively; kflush out is the rate at which air is flushed out of the chamber; and k1 is the kaurene–O3 reaction rate coefficient.

For every experiment starting with a stable O3 condition, we simulated a kaurene time series. We solve Eq. (1) for [kaurene], and take it as the expected kaurene concentration at time i (kau_expi). The rate dO3dt was estimated based on the average change of O3 from time point (i-1) to the time point (i+1) (time resolution: 15 min = 900 s), giving the expression finally as 2kau_expi=QO3, in-O3,i-1-O3,i+12×900s-O3,i×kflushoutO3,i×k1. The average residence time in the COALA chamber concerning flush out was 56 min; thus, the reciprocal kflush out=3×10-4 s-1. QO3, in (unit cm-3 s-1) was estimated for each experiment from the steady-state periods before kaurene injection, when dO3dt=0. The kaurene–O3 reaction rate coefficient k1 was set to be 5×10-16 cm3 s-1 based on the model results of the rate coefficient tests of kaurene and O3 performed in the beginning of our experiments (see the Supplement). This value is fairly close to the only value (1.4 ± 8.2 ×10-15 cm3 s-1) so far reported in the literature (Helin et al., 2020).

The simulated kaurene concentrations were only reliable when kaurene concentrations were high enough (>∼1 ppb) to make a noticeable impact on the O3 concentrations. We used these periods to scale the measured Vocus raw kaurene signal (kau_sigi) to the simulated values (kau_expi) by a correction factor, cor_faci: 3cor_faci=kau_expikau_sigi. We further assumed that this factor stayed constant as long as the pressure in the FIMR did not change; thus, we also extrapolated the correction factors to periods with low kaurene concentrations. In this way, a complete and consecutive time series of correction factors for kaurene was achieved (see the blue line in Fig. 2a).

All the above corrections bring with them considerable uncertainty and potential error sources. In order to assess the reliability of the correction method, we compared our acquired kaurene time series with those measured by TORM and the four collected Tenax samples. First, quantified kaurene concentrations by Tenax tubes are shown in Table S1. The average Vocus signal intensity of kaurene during the sampling time for each Tenax tube was calculated, and the corresponding correction factors required by the Vocus to match the Tenax concentrations can be found in Fig. 2a. The correction factors we derived above were in the same order of magnitude as those deduced from the Tenax samples, although the Tenax samples consistently suggested lower kaurene concentrations. However, the third Tenax sample (Table S1) on 28 January was 2 orders of magnitude lower, which leads us to expect that this sample was unsuccessful for unknown reasons. For comparison with TORM, we scaled the raw Vocus signal intensity data of kaurene to match the estimated kaurene based on TORM. The required scaling factor time series is shown in Fig. 2a for periods when the kaurene was above the detection limit (∼1 ppb) of TORM. Overall, the TORM calibration factors compared well with those from the O3 depletion method above. The good agreement becomes evident when plotting the final kaurene time series from TORM and the corrected Vocus data (Fig. 2b), lending confidence to the Vocus correction method. The figure also shows the importance of utilizing the Vocus data, as the TORM was not able to quantify kaurene concentrations below ∼1 ppb due to background. Despite this, the correlation between the instruments was quite good (R2=0.87, Fig. 2c).

Based on the comparison of our simulated kaurene results with these two independent methods, we conclude that our correction approach was successful. However, the uncertainties remain large due to the needed corrections, the lack of authentic standards, and the dependence on needing to know the kaurene reaction rate coefficient with O3. We estimate that the absolute concentrations have an uncertainty of at least a factor of 3 (+200 %/-67 %), while the uncertainty in the relative changes over the course of our study is closer to a factor of 2 (+100 %/-50 %).

The change rate of HOM concentration in our chamber was determined by the production and loss rates according to Eq. (5): 5dHOMdt=ProductionHOM-LossHOM. HOM can be formed from reactions with both O3 and OH. Most of the time, the kaurene concentration was very low in the chamber and may not have been the major sink for the formed OH. Instead, OH was likely lost to reactions with O3 and potential contaminant molecules in the chamber. In addition, it is also hard to estimate the relative importance of the two oxidants, because neither the O3 nor OH reaction rate coefficients with kaurene are well known. Thus, for simplicity, the production of HOM in the equation above will be written as purely the ozonolysis reaction, i.e., k1γ [kaurene][O3], where k1 is the kaurene–O3 reaction rate coefficient, and γ is the HOM molar yield from the kaurene–O3 reaction. Note that γ will be overestimated if the kaurene–OH reactions contribute substantially to the measured HOM. The kaurene–O3 reaction rate coefficient k1 was assumed to be 5×10-16 cm3 s-1 as shown in Sect. 2.3.

The loss of HOM mainly consists of three parts: chemical reactions, flush out from the chamber, and the condensation onto the walls and potential aerosol particles; they can be written as kloss[HOM], where kloss is the total loss coefficient of HOM. Under steady-state conditions, the concentration of HOM is constant, and we can write it as 6dHOMdt=k1γkaureneO3-klossHOM=0. Thus, 7γ=klossHOMk1kaureneO3. All data used to estimate HOM molar yield were selected from the periods when there were no seed particles in the chamber. According to an earlier study in the COALA chamber by Peräkylä et al. (2020), the lifetime of HOM in the gas phase under a typical situation without seed particles could be assumed to be 200 s, primarily controlled by losses to chamber walls. Therefore, 0.005 s-1 was used for kloss here. The flush-out rate is over an order of magnitude slower, and considering the low oxidant concentrations in our chamber, it is unlikely that the chemical loss of HOM would be faster than the wall loss rate.

The HOM quantification alone had a large uncertainty, which we can only estimate to be at least +100 %/-50 %. Also the kaurene quantification required several assumptions, as did k1. As for HOM quantification, we can only make estimations on the uncertainties of these parameters. If we assume that these three are the major error sources, and each of them have the same uncertainty of +100 %/-50 %, the final uncertainty (via error propagation) becomes +173 %/-87 %.

We observed the oxidation products of kaurene ozonolysis in the gas phase using Vocus and nitrate CI-APi-TOF. Product distributions are shown in Fig. 4 including both less oxidized species and HOM, and the temporal behavior of some important oxidation products are shown in Fig. 1. When discussing detected compounds (M) in the following sections, they were either detected after protonation (MH+) in the Vocus or as clusters with nitrate ions (M⋅NO3-) in the CI-APi-TOF. In other words, the identified species are presented after omitting the reagent ions, but if the specific ions need to be addressed, they will contain the charging ion as well as the label for the charge (+ or -). When masses are discussed, they will refer to the mass-to-charge ratio in the spectra where the compound was identified, including the corresponding reagent ion.

We first compared the spectra of the reactant-free chamber (no kaurene or O3 injection) with that after injecting ∼5 ppb kaurene. As shown in Fig. 4b, a predominant peak of kaurene, C20H32, along with a series of hydrocarbon ions was detected by the Vocus. These hydrocarbons with carbon numbers from 10 to 16 were observed between 130 and 230 Th. The time series of those hydrocarbons showed high similarities to the kaurene time series with very high correlation coefficients (R2>0.95) at different FIMR pressures, indicating that they most likely are fragments of kaurene formed in the Vocus. More details about the fragments of kaurene in the Vocus will be discussed in Sect. 3.4.

After introducing O3 (∼70 ppb), several less oxidized kaurene oxidation products (C18-20H26-34O1-3, complete peak list in Table S3.) were observed between 270 and 330 Th in the Vocus spectra (Fig. 4b). The highest concentrations were observed from compounds identified as C19H30O, C19H32O2, and C20H32O3, of which the time series is shown in Fig. 1. Compounds with identical elemental compositions have been observed in the headspace samples of heated pine needles and spruce twigs (Helin et al., 2020), but to the best of our knowledge, these compounds have not been reported in the ambient air. The ions containing an N atom are believed to be adducts with NH4+, as will be discussed in Sect. 3.4.

In the nitrate CI-APi-TOF, various HOM monomers (C18-20H24-34O5-13) and dimers (C37-40H58-68O5-12) were detected in the mass-to-charge ratio (m/z) ranges 400–550 and 650–750 Th, respectively (Fig. 4c and d). Several radicals with C19 and C20 carbon skeletons were observed as well (Table S4). Overall, the spectra contain a wide range of molecules with different compositions, and some speculation about possible formation pathways is given below. However, we stress that our data are not optimal for deducing detailed oxidation mechanisms, and in some cases, several pathways may lead to the same product, while for some other products, no viable formation pathways were identified. Among the closed-shell HOM monomers, C19H28O6, C19H28O8, C19H30O7, C20H30O5, C20H30O7, and C20H32O7 were the most prominent signals. C19H28O6,8 and C19H30O7 could be explained by the primary ozonolysis RO2 (C19H29O3) undergoing autoxidation and then terminating either in a unimolecular way (for the former) or via reaction with HO2 (for the latter). Corresponding highly oxidized RO2 species were detected (C19H29O7,9, Table S4). However, HO2 is expected to be low in our chamber, so other formation pathways (e.g., RO2+ RO2 reactions) may be involved in forming C19H30Ox species. Among the OH-initiated HOM, we detected C20H30O5-8 as the dominant group, and related RO2 were also identified (C20H31O7-8, Table S4). C20H32O6-9 HOM presumably initiated from OH-addition reactions were observed with decent signal intensities, with C20H32O7 as the most abundant. However, signals of the corresponding C20H33Ox radicals were very low, and only noisy signals were detected even at high kaurene concentrations.

The signal intensities of HOM dimers were about 1 to 2 orders of magnitude lower than those of monomers (Fig. 4d). The highest peaks in the dimer range were observed at 674, 688, 690, 706, and 722 Th corresponding to the formulae as C38H60O6, C39H62O6, C38H60O7, C38H60O8, and C39H64O8, respectively. C38H60O6,7,8 dimers are most likely produced from the cross-reactions of C19H29Ox and C19H31Oy. Similarly, C20H33O2-8 radicals could take part in different bimolecular reactions with C19H29O2-6 and/or C19H31O2-8 forming C39H62O6 and C39H64O8 dimers, respectively. While we do not have suggestions on the formation mechanisms of C19H31Ox radicals, both these dimer signals and the monomeric C19H30Ox species would most easily be explained by the involvement of these radicals. Another unanticipated finding was that we did not detect any dimers with oxygen numbers above 12, and the most abundant dimer signals were identified as compounds with 6–8 oxygen atoms. This is similar to the detected monomers, which is in contrast to earlier findings of monoterpenes and sesquiterpene (Quéléver et al., 2019; Jokinen et al., 2016; Li et al., 2021; Barreira et al., 2021; Kirkby et al., 2016; Rissanen et al., 2014), where dimers typically contained clearly more oxygen than the monomers. We can only speculate on the reasons for this, and a possible explanation is that the most oxygenated RO2 species have very short lifetimes and thus terminate through unimolecular channels before colliding with another RO2. It is also possible that the nitrate CI-APi-TOF might be more sensitive towards less oxygenated dimers if there are 40 C atoms rather than 20 in their skeleton, if a larger molecule increases the likelihood of forming a stable cluster with NO3-. Mass-dependent transmission may also play a role, as the detected dimers were around 700 Th and above, where sensitivity is most likely decreasing (Heinritzi et al., 2016).

Interestingly, a group of HOM with 21–29 carbon atoms was observed in the m/z range from 500 to 560 Th (Fig. 4c), and they exhibited similar temporal behaviors to those of other HOM (Fig. 1). The dominant peaks were identified as C24,26,27,28H36,38,40,42O6-8 species. We conclude that the most probable explanation for these HOM is that they were the dimers formed via the reactions of kaurene-derived RO2 with smaller RO2 derived from various C2-9 organic contaminants in the chamber. Both Vocus and nitrate CI-APi-TOF observed such contaminants during the experiments (Fig. S2), which is the likely source being fairly high-loading monoterpene experiments performed in the chamber prior to the kaurene experiments. Low-volatility gases and particles deposited on the chamber walls can then lead to compounds slowly off-gassing from the walls. In reactions with OH, these C2-9 precursors can then form RO2 that terminate by reacting with the C19-20 radicals, ending up forming the observed C21–C29 HOM dimers. As most of our experiments were done with very low kaurene concentrations (<1 ppb), the contaminants could amount to an equally high (or even higher) OH sink than the kaurene itself. By comparing the most abundant observed contaminants and the C24–C28 dimers, we also tried to assess which RO2 species from kaurene were the most important. In most cases, C19H29O3 and C19H31O2 seemed to be the most plausible radicals to explain the most abundant observed dimers. These were also thought to be the main contributors to C38–C39 dimers of kaurene oxidation. Unfortunately, due to their low oxygen content, we cannot directly detect these radicals with nitrate CI-APi-TOF.

During most of the experiments, all HOM detected by nitrate CI-APi-TOF exhibited similar temporal behaviors as the kaurene time series (Fig. 1), as expected due to the low variability in O3 concentrations. The products measured by the nitrate CI-APi-TOF reacted rapidly to changes in the injected kaurene, while the less oxidized products detected by the Vocus showed a slower response to kaurene changes, especially when kaurene was removed.

We calculated the molar yield for HOM from kaurene ozonolysis (without OH scavenger) as described in Sect. 2.5. Figure 5 shows the measured HOM concentration as a function of the kaurene ozonolysis rate (k1×kaurene×O3). In the kaurene ozonolysis experiments at low oxidation rates (Fig. 5b), the total HOM concentrations had a near-linear dependence on the amount of kaurene reacting with O3. We can estimate a molar HOM yield of ∼2 % in this range, though keeping in mind the large uncertainties in the quantification of kaurene and HOM. At higher oxidation rates (Fig. 5a), the increase in total HOM concentration exhibited a logarithmic dependence. The slower growth of HOM concentration at higher oxidation rates might be explained by increasing sinks in the chamber, e.g., due to particle formation. The RO2 lifetimes will also decrease, which may hamper the autoxidation when bimolecular reactions become faster. When the entire data range was included, the molar HOM yield ranged between 0.1 % and 10 %.

HOM yields from kaurene ozonolysis, as for any other system, will always depend on the atmospheric conditions, in particular the type and concentration of bimolecular reaction partners (mainly NO, RO2, HO2). The reaction partner will determine the possible branching pathways, while their absolute concentrations will affect RO2 lifetimes and thereby the potential to undergo autoxidation. According to previous studies, the reported kaurene concentrations in the atmosphere range from ppq to a few ppt (Li et al., 2020; Yee et al., 2018). In our study, concentrations were higher (Fig. S3a and c), but the ∼2 % HOM yield was estimated based on periods when kaurene concentrations were <50 ppt (Fig. S3b). As the atmosphere is much more complex, with various other species present, it is possible that the RO2 concentrations in our experiments are, in fact, lower than in atmosphere, despite the kaurene concentrations being higher. Nevertheless, we think that the oxidation conditions in our experiments are quite close to atmospheric conditions and that the potential bias in HOM yields due to this is marginal when compared to the overall uncertainty of HOM quantification (Sect. 2.4.2). There are two main exceptions to this. Firstly, if NO concentrations would be very high, it could lead to very different oxidation pathways, with unknown effects on the HOM yields. Secondly, in the atmosphere it would be much less likely for two kaurene-derived RO2 species to react, as the majority of the RO2 would come from different precursors; thus, the C37-40 dimers would be unlikely to form in the atmosphere.

Although large uncertainties are included in the derivation of our HOM yield from kaurene ozonolysis, yields of a few percent have been reported for a variety of other BVOCs earlier (Table 1). More studies are needed to better quantify the HOM formation in this system and get further chemical insights into the detailed oxidation pathways. Although the global emission budgets of diterpenes remain unquantified, when combining the limited emission rates reported previously (Matsunaga et al., 2012; Yáñez-Serrano et al., 2018) with the HOM yield estimated here, it is possible that HOM formation from diterpenes, as well as their influence on SOA formation, are of larger importance than previously thought.

To deduce the ability of the oxidation products from kaurene ozonolysis to contribute to aerosol formation, we investigated their behavior as a function of the available condensation sink (CS) in the chamber. This method has been applied to study the volatilities of the α-pinene and cyclohexene oxidation products by Peräkylä et al. (2020) and Räty et al. (2021). In this methodology, we assume that the condensation onto seed particles is the main driver of the concentration changes of oxidation products during the experiment, although other changes (e.g., the changes of precursors and intermediates due to seed injection) can have a non-negligible influence (Peräkylä et al., 2020). This approach does not provide direct estimates of volatilities, but it does provide a good separation between readily condensing vapors (accommodation coefficient near unity) and the more volatile ones that are unaffected by the condensation sink (i.e., with very low accommodation coefficients).

In this experiment, kaurene, O3, and seed particles were injected into the chamber until it reached a steady state (referred to as S1 in Figs. 1c and S4b). The injection of seed particles was then stopped, causing a decrease in the CS and a corresponding increase in the concentrations of several compounds, ultimately reaching a new steady-state condition (referred to as S2 in Figs. 1c and S4b). For each observed compound, we then calculated the concentration ratio at S1 and S2, henceforth termed the fraction remaining (FR) (Fig. 6).

From Fig. 6a, we found a clear transition of volatilities when the molecular mass increased, in line with previous findings for α-pinene and cyclohexene systems (Peräkylä et al., 2020; Räty et al., 2021). Interestingly, all species detected by the Vocus had an FR around 1, i.e., not affected by the CS, suggesting that the Vocus primarily detects VOCs and IVOCs (intermediate-volatility organic compounds). Kaurene (and all its fragments) were in this range, as expected. Signals with FR much above unity may either be semivolatile compounds that never reached a steady state during the experiments or molecules that were preferentially formed in the particle phase before evaporating into the gas phase. A sharp drop in the FR, from ∼1 to ∼0.1, was found in the mass range 300–350 Th, and the oxidation products with more oxygen tended to have lower values. This trend is consistent with that in the mass range above 350 Th (Fig. 6b), where most HOM were found. This indicates that these species condensed efficiently and were mainly LVOCs (low-volatility organic compounds) or ELVOCs (extremely low-volatility organic compounds). Similar results were reported for HOM formed from α-pinene ozonolysis by Peräkylä et al. (2020). The relatively large spread of the FR above 350 Th may indicate that the seed particles also impacted the formation pathways, in addition to providing a larger CS. For example, some longer-lived RO2 might also have been lost to the particles, as discussed by Peräkylä et al. (2020).

In the mass range 200–300 Th, the FR of nitrate CI-APi-TOF species were also around one even though some have quite high oxygen content (Fig. S5a). These compounds were primarily contaminants from the chamber walls, typically with less than 10 carbon atoms (Fig. S5c). While the high O-atom content might suggest lower volatility, the FR around unity suggests that the CS was not a noticeable sink for these species. On the one hand, this is credible since their existence in the chamber during these experiments means that they evaporate efficiently from the chamber walls. On the other hand, it is also possible that there is large storage on the walls, and they have reached a balance between the gas-phase and wall concentrations. When these species condensed onto the seed particles, more came out from the chamber walls, keeping the measured gas-phase concentrations stable. Either way, although not of particular importance for this study, the mere existence of these compounds in the continuously flushed chamber strongly suggests that they are SVOCs (semivolatile organic compounds).

In summary, these results show that the oxidation products of kaurene ozonolysis with approximately five or more oxygen atoms were mainly LVOCs or ELVOCs, and will thus efficiently condense to form organic aerosols. In line with earlier studies (Räty et al., 2021), we also found that the oxygen content is the most important parameter affecting the volatility. Despite the longer carbon chain, thus higher mass, of the diterpenes compared to, for example, monoterpenes or cyclohexene, the transition from condensing to non-condensing was not dramatically different. For those compounds (C6 and C10), similar experiments indicated that approximately seven O atoms were needed, whereas for the C20 kaurene the number was around five.

Fragmentation is a ubiquitous process when measuring VOCs with PTR instruments, complicating the identification and interpretation of the mass spectra. Previous studies have shown that the settings of PTR-MS (in essence E/N), the types of VOCs, and environmental conditions (e.g., RH) can all affect the fragmentation process and change the fragment distributions (Tani et al., 2003; Demarcke et al., 2010; Kim et al., 2010; Rimetz-Planchon et al., 2011; Gueneron et al., 2015). Within the standard E/N settings (80–140 Td) of PTR instruments, a decrease in E/N may decrease the absolute sensitivity, although the fraction of signal detected at the parent ion may become larger due to decreased fragmentation (Kim et al., 2009; Demarcke et al., 2010). Various terpenes, including monoterpenes and sesquiterpenes, have been observed to undergo different degrees of fragmentation within the PTR instruments (Tani et al., 2003; Kim et al., 2009; Rimetz-Planchon et al., 2011; Misztal et al., 2012). However, potential fragmentation patterns of diterpenes have not been reported before this study. We find several fragments of kaurene in the Vocus, and by exploiting the decreases in FIMR pressures, we were able to investigate the fragmentation process of kaurene as a function of this pressure as well.

As can be seen in Fig. 4a, a total of seven main kaurene fragments with carbon numbers exceeding 10 were observed in the Vocus spectra. Also, some peaks smaller than 130 Th in the spectra were seen. However, as the difference in their signals before and after kaurene injections are very small, we focus on the seven main fragments in this section. All fragments were detected as protonated ions (MH+). We classified these fragments into two groups (Group 1: C10H15+, C11H17+, C12H19+, C13H21+, C14H23+; Group 2: C15H23+, C16H25+), where each included fragments separated by a CH2 unit (14 Th) based on a previous classification method reported by Sovová et al. (2011). The fragment peaks were dominated by Group 1, and the ratio of Group 1 signals to Group 2 signals was around 6 for all the experiments. When FIMR pressure was ∼1.4 mbar (original setting in this study), m/z 191, C13H21+, was the highest signal (Fig. 7h), and the ratios of C13H21+ signal to C11H17+, C12H19+, and C14H23+ were about 1.6, 1.3, and 3, respectively. m/z 217, C16H25+, was the dominant signal in Group 2, and it was 2-fold higher than that of C15H23+. All these fragments of kaurene were much larger than those of monoterpenes (Maleknia et al., 2007; Tani et al., 2003; Misztal et al., 2012), because the precursor kaurene is twice as large as monoterpenes (C10H16) in terms of molecular mass. However, the fragment m/z 149, C11H17+, was observed as the most abundant fragment of many sesquiterpenes in previous studies (Kim et al., 2009, 2010; Demarcke et al., 2010). Table 2 summarizes fragmentation patterns for kaurene (in this study) and sesquiterpenes from previous studies (Kim et al., 2009). Although only two fragments (C10H15+ and C11H17+) were identical between kaurene and the sesquiterpene, we found that the lost neutral fragments were surprisingly similar between the diterpenes and sesquiterpenes.

Figure 7a–g show the relationship between kaurene (C20H33+) signals and its fragments under different FIMR pressures. From these figures, we can see that the signals of fragments at a FIMR pressure of 0.9 mbar increased more than 1 order of magnitude relative to C20H33+ compared to those at 1.4 mbar for C11H17+, C12H19+, and C15H23+. The increases in C13H21+ and C14H23+ were smaller, but they still doubled when the pressure dropped from 1.4 to 0.9 mbar. In addition, the ratio of the sum of all fragments to C20H33+ increased from ∼0.4 to ∼3 during the experiments. Thus, it is clear that the fragmentation increases with decreasing FIMR pressures, as can be expected since the electric field strength stayed constant, thus leading to a net increase in E/N.

Changes in FIMR pressure also impacted the distribution of the fragments (Fig. 7h). When Vocus was run at the FIMR pressure of 1.4 mbar, C13H21+ was the most abundant fragment signal. However, when the pressures decreased, the fractions of C13H21+ started to decrease, and at the end of the experiments (FIMR pressure =0.9 mbar), C12H19+ and C11H17+ had become the dominant fragments. The overall distribution of fragments for Group 1 clearly transitioned towards smaller fragments, with all the largest ones decreasing while the smaller ones increased. Though with fewer ions, the same trend was also visible in Group 2.

In addition to the protonated molecular ions and fragments typically observed in PTR instruments, some compounds were also found to be charged by ammonium, NH4+, which can be intrinsically formed in the ion source (Norman et al., 2007; Müller et al., 2020) or be introduced as impurities either in the water or sample air. Although typically not a major signal, in this study, we observed the clusters of kaurene and two oxidation products with NH4+ at surprisingly high amounts (Fig. 4b). As shown in Fig. 8, the temporal behavior of the NH4+ adducts was in good agreement with those of the corresponding protonated ions (R2>0.9). The ratio between the NH4+ cluster signal and the protonated signal was less than 2 % for C20H32, whereas the ratios could be more than 50 % (ranging from 15 % to 70 %) for the two oxidation products (C19H30O and C19H30O2). These ratios decreased over time, as the FIMR pressure decreased, and subsequently the collision energies and thus the fragmentation increased.

We cannot say if NH4+ production in our Vocus during this study was higher than typical, as the signal is buried under the large water signals. However, it is possible that the large size of the diterpene and its oxidation products may help to stabilize the clusters formed with NH4+, thus increasing the survival probability compared to NH4+ adducts with smaller molecules. In addition, if protonation of these large molecules would lead to considerable fragmentation, NH4+ adducts, as a softer ionization method (Chen and Her, 1993; Harrison, 2018), may again increase the probability of the adduct compared to the protonated molecule. Thus, it is possible that an instrument with NH4+ ionization could be better suited for the detection of diterpenes and their oxidation products, but further studies are needed to validate this hypothesis.