SO2 oxidation products other than H2SO4 as a trigger of new particle formation ? Part 2: Comparison of ambient and laboratory measurements, and atmospheric implications

Atmospheric new particle formation is generally thought to occur due to homogeneous or ion-induced nucleation of sulphuric acid. We compare ambient nucleation rates with laboratory data from nucleation experiments involving either sulphuric acid or oxidized SO 2 . Atmospheric nucleation occurs at H 2 SO 4 concentrations 2–4 orders of mag- 5 nitude lower than binary or ternary H 2 SO 4 nucleation. In contrast, the atmospheric nucleation rates and H 2 SO 4 concentrations are very well replicated in the SO 2 oxidation experiments. We explain these features by the formation of free HSO 5 radicals in pace with H 2 SO 4 during the SO 2 oxidation. We suggest that at temperatures above ∼ 250 K these radicals produce nuclei of new aerosols much more e ﬃ ciently 10 than H 2 SO 4 . These nuclei are activated to further growth by H 2 SO 4 and possibly other trace species. However, at lower temperatures the atmospheric relative acidity is high enough for the H 2 SO 4 –H 2 O nucleation to dominate.


Introduction
The formation of aerosol particles in atmospheric nucleation events is a source of cloud 15 condensation nuclei (CCN) (Laaksonen et al., 2005;Komppula et al., 2005) missing from most current climate models, and constitutes therefore a considerable uncertainty in climate change predictions. The new particle formation is generally thought to occur due to homogeneous or ion-induced nucleation of sulphuric acid (H 2 SO 4 ) (Kulmala, 2003) formed in oxidation of sulphur dioxide (SO 2 ). However, it has so far remained 20 unclear why the binary classical nucleation theory (CNT) seems to explain ambient observations of nucleation above the altitude of 4 km (Weber et al., 1999) as well as laboratory nucleation experiments with vaporized sulphuric acid (Ball et al., 1999;Zhang et al., 2004), but severely underpredicts nucleation below 4 km (Weber et al., 1999) and in laboratory experiments involving oxidized SO 2 (Friend et al., 1980;  A further puzzle related to boundary layer nucleation events was noticed by Weber et al. (1996), who compared particle formation rates with the measured gas-phase concentrations of sulphuric acid, and deduced that the dependence is consistent with a mechanism in which the rate of appearance of new particles is proportional to the rate of sulphuric acid -sulphuric acid collisions in the gas. However, in the boundary layer, 5 clusters containing just two H 2 SO 4 molecules should not be stable even if the possibility of immediate association of water molecules is allowed for. As a solution to the problem it has been suggested that other associated molecules such as ammonia (Weber et al., 1996;Ball et al., 1999;Kulmala et al., 2000) or organic acids (Zhang et al., 2004) may have a stabilizing effect on the clusters. However, Berndt et al. (2005Berndt et al. ( , 2006 recently 10 presented measurements on particle formation in a flow chamber in which mixtures of SO 2 gas and water vapour were irradiated with UV light. The nucleation occurred at calculated sulphuric acid concentrations approximately three orders of magnitude lower than nucleation from H 2 SO 4 itself. Experiments in the absence of UV light (dark reaction) showed exactly the same results indicating that irradiation itself does not 15 influence new particle formation. This observation is qualitatively in good accord with the similar, long overlooked experiments of Friend et al. (1980).
In this paper, we compare atmospheric observations to various laboratory datasets of nucleation involving sulphuric acid or its precursors, and other substances. We suggest, in line with the companion paper (Berndt et al., 2008), a chemical mechanism 20 that could explain the seemingly conflicting experimental results, and examine an atmospheric nucleation dataset for possible influences by other than sulphur-containing compounds. Finally, we synthesize our results, and offer a potential solution as to why the binary CNT produces better predictions of upper tropospheric than boundary layer nucleation. Introduction Atmospheric nucleation rates were measured during 10 nucleation events during the spring 2003 QUEST measurement campaign in Hyytiälä, Finland. The atmospheric nucleation rate is obtained from particle size distribution measurements that were car-5 ried in 10 min intervals during the campaign. The primary quantity we obtain is the 3 nm particle appearance rate (J 3 ), determined from the increase of 3-6 nm particle concentrations between the measurement cycles, taking account of coagulational losses, and the flow of particles out of this size range as they grow by condensation of supersaturated vapors (Dal Maso et al., 2005). The actual nucleation rate, on the other hand, is the rate of critical cluster (diameter ∼1.5 nm) formation. We therefore also calculated the 1.5 nm particle formation rates (J 1.5 ) from the J 3 data by accounting for the observed particle growth rates and coagulation of 1.5-3 nm particles with larger aerosols (Kerminen et al., 2002). The 1.5 nm particle formation rate is taken to represent atmospheric nucleation rate.

Laboratory experiments
Laboratory experiments were carried out in order to extend the results of Berndt et al. (2005;  generation outside the flow tube. SO 2 was taken from a 1ppmv calibration mixture in N 2 (Messer). The total gas flow was 3.6 l min −1 resulting in a bulk residence 5 time of 420 s. Furan was added for OH radical titration. With knowledge of the amount of reacted furan and the initial concentrations for SO 2 and furan, the concentration of reacted SO 2 and the formed H 2 SO 4 (assuming a H 2 SO 4 yield of unity) can be easily calculated (Berndt et al., 2005). The measured temperature dependent data was used to obtain estimates for the 10 corresponding particle formation rates at atmospheric temperatures. This was done by fitting a line to the (T , logJ) -data (see Fig. 1), where J refers to the particle formation rates observed in the measurements. The slope of the fitted curve (k=-0.2217 K −1 ) was used to scale the measured particle formation rates to T =273 K. In our extrapolation we assumed an exponential relationship between the particle formation rates 15 and temperature. The physical reasoning behind this assumption is that the formation rates are likely to depend on the saturation vapour pressures of the condensing species which show an exponential relationship with temperature.

Comparison atmospheric and laboratory nucleation rates
To investigate how the laboratory data compare with atmospheric nucleation rates we 20 plotted the nucleation rate as a function of H 2 SO 4 concentration in Fig. 2. It can be seen that the atmospheric data points (Sihto et al., 2006) are situated more or less in between the Friend et al. (1980) and Berndt et al. (2005Berndt et al. ( , 2006 laboratory data sets. Uncertainties in the laboratory data on SO 2 induced nucleation include possible errors in calculated H 2 SO 4 concentrations due to uncertain chemical mechanisms and not 25 well defined wall losses of H 2 SO 4 ; uncertainty in nucleation rate due to uncertain effective volume inside the flow chamber where nucleation occurred; and possible effects from gaseous impurities. We note that the nucleation rates of Berndt et al. (2005Berndt et al. ( , 2006  are minimum values as nucleation has been assumed to occur effectively throughout the irradiated region of the flow tube, and it is not impossible that the true rate is by up to an order of magnitude higher. On the other hand, the laboratory measurements have been performed at 293 K and the ambient measurements between 263-281 K. Taking into account laboratory data regarding the temperature dependence of nucleation 5 ( Fig. 1), we extrapolated the laboratory nucleation rates measured at 28% relative humidity (RH) to 273 K obtaining very good compatibility with the ambient data (see blue line in Fig. 1). It should be noted, however, that the slope of the extrapolated laboratory data appears somewhat steeper than that of the ambient dataset. Figure 2 also shows two datasets of measured nucleation rates for the H 2 SO 4 -H 2 O system both with and without a third species, which in the case of the Ball et al. (1999) measurements is ammonia and in the case of the Zhang et al. (2004) measurements is organic acids. It is clearly seen that although these species enhance the nucleation rate somewhat, the enhancement is completely inadequate to explain the atmospheric nucleation rates. This conclusion holds even if the temperature dependence of the 15 ternary nucleation is accounted for: using the classical nucleation theory temperature dependence calculated by Napari et al. (2002) for sulphuric acid-ammonia-water, and extrapolating at constant nucleation rate from room temperature (corresponding to the ternary experiments in Fig. 2) down to around 270 K, it can be seen that the ternary curves will not come down more than two orders of magnitude in sulphuric acid con-20 centration (besides, the experiments of Ball et al. (1999) were carried out at higher ammonia concentrations (80-170 ppt) than are usually found in Hyytiälä). All of this is in accord with the conclusion of Yu (2006), who noted that the classical ternary nucleation predicts close to 30 orders of magnitude too high nucleation rates compared with those measured by Ball et al. (1999). 25 The remaining dataset in Fig. 2, by Sorokin and Arnold (2007) shows results of ion-induced nucleation of sulphuric acid at 295 K (the different lines are for different sized ions). As with ammonia and organic acids, it appears that ions enhance the nucleation but not enough to explain atmospheric nucleation. The puzzling feature with Interactive Discussion the Sorokin-Arnold data is that the sulphuric acid in their experiments was produced by oxidizing SO 2 , hence one would expect closer agreement with the Friend et al. (1980) and Berndt et al. (2005Berndt et al. ( , 2006 experiments rather than with the Ball et al. (1999) and Zhang et al. (2004) experiments. We show below that this apparent paradox can be resolved by accounting for the low total pressure (80 hPa) employed by Sorokin and 5 Arnold (2007).

Analysis of atmospheric nucleation data
The scatter of the atmospheric nucleation rates in Fig. 2 originates from variations in temperature and RH and from measurement uncertainties (e.g. variations in aerosol size distributions caused by other factors than nucleation). Additionally, unknown species participating in the nucleation with varying contributions may have an effect.
To this end, we analyzed our ambient nucleation rates using multivariate regression analysis to determine if various trace species participate in the nucleation. The starting point of our analysis is the nucleation theorem, which can be written as (∂ln(J)/∂ln(A i )) T,Aj =n i +∆ i , where J denotes nucleation rate, A i is the gas-phase 15 activity (partial pressure divided by saturation vapour pressure) of species i , n i is the number of molecules in the critical nucleus, and ∆ i is a small term originating from kinetic terms describing vapour molecule collisions with the nucleus (the sum of the individual ∆ i 's is between 0 and 1; Oxtoby and Kaschiev, 1994). It has been shown using both thermodynamic (Oxtoby and Kaschiev, 1994) and statistical mechanical 20 (MacDowell, 2003) arguments that the nucleation theorem is a very general relation that extends down to the smallest cluster sizes and holds independently of any specific nucleation theories.
In order to apply the nucleation theorem, one should determine the slope of nucleation rate as a function of the gas-phase concentration of the species in question at 25 constant temperature and gas phase concentrations of other species participating in the nucleation process. However, we do not have enough data of atmospheric nucleation rates to obtain meaningful correlations at narrow temperature and gas concentra-9680 Introduction Interactive Discussion tion intervals. Furthermore, it is not certain that we know all of the species participating in the nucleation. We therefore apply nonlinear regression analysis based on nucleation theory in order to obtain information on the nucleating species. The nucleation rates J 1.5 and J 3 were the dependent variables in the regression models. The set of independent variables consisted of gaseous H 2 SO 4 , NH 3 , H 2 O and NO x 5 (other meteorological and chemical variables were also tested but no improvement in statistical significance was found). The values of the nucleation rate J 3 were advanced in time by the amount of estimated growth time between 1.5 and 3 nm in comparison to the independent variables. The same model was computed for both J 1.5 and J 3 .
The multiple regression equation used in our analysis is where β 0 is the intercept, β 1 ,. . . , β 4 are the regression coefficients and ε describes the zero-mean normally distributed and uncorrelated error terms. NO x denotes gas phase concentration, and the other quantities inside the logarithms are gas-phase activities of the respective species. As suggested by a self-organising map analysis in which 15 the data was seen to separate into two clusters distinguished by temperatures below and above freezing, respectively, the regression coefficients β 1 and β 3 were taken to depend on the temperature such that the slopes are different for temperatures of below and over 0 • C. This division improved the statistical significance of the model.
The regression coefficients, standard errors of the coefficients, t-statistics and asso-20 ciated p-values are shown in Table 1 for J 1.5 . With the exception of NO x , the regression coefficients of the model for J 1.5 should represent the apparent numbers of molecules contained by the critical nucleus. As can be seen from Table 1, the regression model suggests that the critical nucleus contains possibly one sulphuric acid related molecule (the regression coefficient is only 0.3 at subzero temperatures, but its statistical signif- 25 icance is somewhat lower than that of the higher temperature regression coefficient) and no ammonia. Additionally, water vapour has a negative and NO x a positive influ- Interactive Discussion ence on nucleation. The regression coefficients of the J 3 -model are qualitatively similar although small quantitative differences are seen. Note, that the negative influence of water vapour on new particle formation is in line with previous observations that nucleation events occur more frequently at dry conditions (Boy and Kulmala, 2002). It has been shown that water vapour inhibits the 5 formation of condensable organic species formed in ozonolysis (Bonn and Moortgat, 2002, Bonn et al., 2002 that may be involved in new particle formation. In order to see whether monoterpene oxidation products (MTOP) participate in the nucleation, we repeated the regression modelling with just two independent variables, H 2 SO 4 and MTOP (Sellegri et al., 2005) concentrations. (Here, we were not able to use activity for MTOP, and for consistency we chose to use concentration for sulphuric acid as well. Note that we did not include MTOP in the above multiple regression analysis because both water vapour and NO x appeared to be correlated with MTOP, thus all three would not have been independent variables. Because there was much less data of MTOP compared with the other gases, we left MTOP out.) The results are shown in 15   Table 2. H 2 SO 4 now shows statistically significant regression coefficients both below and above freezing, with values somewhat above unity. This is consistent with studies which have indicated one to two sulphuric acid molecules in the critical cluster (Weber et al., 1996;Sihto et al., 2006;Riipinen et al., 2007). At subzero temperatures MTOP shows a statistically significant regression coefficient close to unity, indicating possi-20 ble organic contribution to nucleation. However, at higher temperatures this indication disappears.

Chemistry
Taken together, the results from Sect. 2 above indicate that ammonia and monoterpene Introduction Interactive Discussion tainly cannot enhance binary nucleation enough to explain them. On the other hand, the SO 2 induced laboratory nucleation appears to be in quite close agreement with the atmospheric data. Here we provide a possible explanation for this agreement.
To explain their observations, Friend et al. (1980) suggested in the beginning of the 1980's a mechanism involving the formation of the free radicals HSO 3 and HSO 5 as 5 SO 2 is oxidized by OH radicals in the presence of water vapor. Collisions of these sulphur-containing radicals would then result in the formation of stable clusters consisting of H 2 S 2 O 6 or H 2 S 2 O 8 probably associated with water molecules. The currently accepted mechanism of atmospheric SO 2 oxidation is as follows (Finlayson-Pitts and Pitts Jr., 2000): 10 OH + SO 2 → HSO 3 (1) The reaction of HSO 3 with O 2 is very fast (Gleason et al., 1987) resulting in atmospheric steady state concentrations for HSO 3 of ∼0.05 molecule cm −3 ((OH)=10 6 and 15 (SO 2 )=10 11 molecule cm −3 ). In competition to the stated reaction products of pathway (2), SO 3 and HO 2 , also the formation of the addition product HSO 5 has to be considered: Stockwell and Calvert (1983) found as a result of their experiments that more than 20 80% of initial OH radicals are regenerated as HO 2 radicals in pathway (2), whereas the work of Schmidt et al. (1985) indicates that at atmospheric pressure, the loss of OH radicals via pathway (2a) is less than 10%. That points at a dominance of pathway (2) over (2a) but does not exclude that HSO 5 is produced in substantial fractions. HSO 5 is a peroxy radical and its hydrated form could initiate the nucleation process (Wayne,25  Interactive Discussion 1991) rather than HSO 3 . This fact is especially caused by the expected higher atmospheric concentration of HSO 5 (note that gas-phase HSO 5 has not been directly observed). In the atmosphere, HSO 5 can react with other atmospheric species such as other peroxy radicals, and we suggest that these reaction products act as the critical clusters triggering atmospheric nucleation. Gaseous H 2 SO 4 is formed in the SO 2 5 oxidation process via pathways (2) and (3) in pace with the sulphur-containing radicals via pathway (2a) or pathway (1), partially explaining the apparent dependence of the nucleation rate on H 2 SO 4 concentration. The companion paper (Berndt et al. 2008) provides experimental support for reaction (2a) to be operational and offers further discussion of the chemical mechanisms possibly contributing to new particle formation.

10
As an alternative explanation of the results of Friend et al. (1980), McGraw (1982 suggested that the data might be an artifact produced by sulphuric acid vapour nucleating inside the particle detector (where the RH was 300%). However, this cannot have been the case with the Berndt et al. (2005Berndt et al. ( , 2006) results (where the particle detector used butanol as working fluid). As the Friend et al. (1980) and Berndt et al. (2005, 15 2006) data are in rather good agreement especially at high RH, it is likely that the free radical nucleation mechanism is the correct explanation of the experimental results.
To explain the experimental result of Sorokin and Arnold (2007) who found nucleation rates consistent with sulphuric acid nucleation rather than the radical nucleation despite the fact that they produced the H 2 SO 4 by oxidizing SO 2 , we note that reaction (2a) 20 is possible only in the presence of sufficient concentration of inert molecules whose collisions with the reactants ensure that excess energy is carried away and the product is stable. Apparently the 80 hPa total pressure employed by Sorokin and Arnold (2007) is too low for the HSO 5 radical production to lead to nucleation, and new particles can only form once enough H 2 SO 4 has accumulated in the air to nucleate without 25 the radicals. In the atmosphere, the efficiency of the radical production will decrease as a function of altitude (the 80 hPa pressure corresponds roughly to an altitude of 17 km) while the efficiency of H 2 SO 4 nucleation will increase (because for fixed acid concentration, the relative acidity increases as temperature decreases). Note that the  Gleason et al. (1987) indicates that the reaction pathway (2a) is negligible at pressures below 10 hPa.

Activation and growth
It should be noted that regardless of their exact chemical composition, the reaction products from sulphur radical recombination are too small to be directly detected by 5 the aerosol instruments used by Berndt et al. (2005Berndt et al. ( , 2006 which have a lower detection limit at roughly 3 nm in particle diameter. Hence, the molecules need to grow into detectable sizes to be counted as particles. Our nucleation experiments at different temperatures indicate that this growth is due to condensation of sulphuric acid (probably together with water vapor), and that the growth is activated, i.e. triggered by 10 heterogeneous nucleation . The reason for this conclusion is that if the new particle formation was governed solely by gas-phase reactions and molecular collisions of the reaction products, the nucleation rate should increase as a function of temperature, but we observe a decreasing trend. This decreasing trend is explained by the fact that the probability of heterogeneous nucleation of sulphuric acid on sulphur 15 radical reaction products decreases as a function of H 2 SO 4 relative acidity, which in turn decreases as a function of temperature at fixed H 2 SO 4 concentration. Note that it is possible that in the atmosphere also other vapors than H 2 SO 4 and water contribute to the heterogeneous nucleation. In order to analyze the water content of critical nuclei in the laboratory experiments, 20 we examined the slope of the nucleation rate versus RH on logarithmic scales, at fixed H 2 SO 4 concentration. According to the nucleation theorem this slope very nearly gives the molecular content of the critical nucleus. The resulting numbers of water molecules were 2 for the experiments of Friend et al. (1980) and 4 for the experiments of Berndt et al. (2005Berndt et al. ( , 2006. This difference causes the two experimental data sets to diverge 25 more at drier conditions (see Fig. 1). A likely explanation for the overall discrepancies in the two data sets is the difference in particle detection efficiences. Friend et al. (1980)  Interactive Discussion grow by condensation of water vapour at 300% RH. It is conceivable that at such high supersaturation, already the sulphur radical reaction products may grow to detectable sizes. As noted above, Berndt et al. (2005Berndt et al. ( , 2006 were not able to detect all of the sulphur radical reaction products, but only the fraction of them which was activated to grow to 3 nm by condensation of H 2 SO 4 and water vapors.

4 Atmospheric consequences
To summarize the suggested new particle formation mechanism, we propose that (1) sulphur radicals are formed in parallel with H 2 SO 4 in the SO 2 oxidation process; (2) the sulphur radicals react (possibly in hydrated form) with as yet unspecified atmospheric species; (3) H 2 SO 4 vapour nucleates heterogeneously on the reaction products, starting condensational growth of the new particles. Note that it depends on the relative acidity of the H 2 SO 4 vapour whether just a fraction or all of the reaction products nucleate heterogeneously (or "activate"). Thus, at constant H 2 SO 4 vapour concentration, there will be some temperature at which all available nuclei will be activated, and at colder conditions the nucleation rate should thus no more depend on temperature.
15 Figure 3 shows calculations for classical homogeneous H 2 SO 4 -H 2 O nucleation, and for nucleation of H 2 SO 4 on pre-existing nuclei. The latter is described using so called activation nucleation mechanism , where the activation coefficient contains information of the nucleus concentrations and their sizes. The production of nuclei from the sulphur radicals followed by their activation by H 2 SO 4 is consistent with 20 this theory; however, since we do not know the sulphur radical production rate or the size of the nuclei, we fixed the activation coefficient to fit atmospheric nucleation rate data (Sihto et al., 2006). As shown in Fig. 3, the activation mechanism is more efficient than binary homogeneous nucleation at low sulphuric acid concentrations and at high temperatures. This is consistent with practically all available nucleation data both from Introduction above the altitude of about 4 km, but not below.

Conclusions
By comparing an ambient nucleation dataset (particle formation rate vs. sulphuric acid concentration) with several different laboratory datasets, we come to the conclusion that sulphuric acid vapour produced by evaporating liquid H 2 SO 4 is not capable of apparent paradox is the role of sulphur radicals produced during SO 2 oxidation. In particular, we hypothesize that HSO 5 radicals are formed in a reaction between HSO 3 and O 2 in the presence of sufficient concentration of inert molecules whose collisions stabilize the reaction product. Experimental support for such a mechanism is given in the companion paper (Berndt et al., 2008). We further suggest that the HSO 5 radicals 15 will react with other trace species, and that the resulting molecules act as nuclei for heterogeneous nucleation of H 2 SO 4 vapour which will initiate the growth of the new particles. The suggested new particle formation mechanism is consistent with the socalled activation type nucleation , and may provide an explanation for the finding that binary sulphuric acid nucleation seems to explain upper tropospheric 20 particle formation but not the boundary layer nucleation events.