LONG-TERM MEASUREMENT OF SUB-3NM PARTICLES AND THEIR PRECURSOR GASES IN THE BOREAL FOREST

The knowledge of the dynamics of sub-3nm particles in the atmosphere is crucial for our understanding of first 10 steps of atmospheric new particle formation. Therefore, accurate and stable long-term measurements of the smallest atmospheric particles are needed. In this study, we analyzed over five years of particle concentrations in size classes 1.1–1.3 nm, 1.3–1.7 nm and 1.7–2.5 nm obtained with the Particle Size Magnifier (PSM) and three years of precursor vapor concentrations measured with the Chemical Ionization Atmospheric Pressure Interface Time-of-Flight mass spectrometer (CI-APi-ToF) at the SMEAR II station in Hyytiälä, Finland. The results show that the 1.1–1.3 nm particle concentrations 15 have a daytime maximum during all seasons, which is due to increased photochemical activity. There are significant seasonal differences in median concentrations of 1.3–1.7 nm and 1.7–2.5 nm particles, underlining the different frequency of new particle formation between seasons. In particular, concentrations of 1.3 – 1.7 nm and 1.7 – 2.5 nm particles are notably higher in spring than during other seasons. Aerosol precursor vapors have notable diurnal and seasonal differences as well. Sulfuric acid and highly oxygenated organic molecule (HOM) monomer concentrations have clear daytime maxima, while 20 HOM dimers have their maxima during the night. HOM concentrations for both monomers and dimers are the highest during summer and the lowest during winter. Higher median concentrations during summer result from increased biogenic activity in the surrounding forest. Sulfuric acid concentrations are the highest during spring and summer, with autumn and winter concentrations being two to three times lower. A correlation analysis between the sub-3nm concentrations and aerosol precursor vapor concentrations indicates that HOMs, particularly their dimers, and sulfuric acid play a significant role in new 25 particle formation in the boreal forest. Our analysis also suggests that there might be seasonal differences in new particle formation pathways that need to be investigated further. https://doi.org/10.5194/acp-2020-719 Preprint. Discussion started: 27 July 2020 c © Author(s) 2020. CC BY 4.0 License.

scans) and three size bins (1.1 -1.3 nm 1.3 -1.7 nm, 1.7 -2.5 nm) to minimize the effect of noise on the analysis, but still retain a high enough time resolution for the analysis.
Recently, Cai et al. (2019) recommended another inversion routine for PSM data, the expectation-maximization algorithm (EM). However, as our data was already inverted with the kernel method and the EM method is computationally expensive, 130 we decided to stick with the kernel method. The two inversion methods produce similar concentrations and size distributions when both are optimized for the dataset in question (Cai et al., 2019). It remains future work to test the applicability of the EM algorithm for SMEAR II data and optimize it to conditions with rather low particle concentrations.

Effect of supersaturation and background counts 135
At optimal temperature settings the PSM should activate a large fraction of even the smallest particles around 1 nm, while still minimizing the effect of homogenous nucleation within the PSM. In practice, a small background from homogeneous nucleation needs to be tolerated at higher saturator flow rates in order to activate the smallest particles, especially organic clusters. The amount of homogeneously nucleated droplets can be taken as an indicator of the supersaturation level 140 (activation efficiency) (Jiang et al., 2011).
To monitor the instrument operation and supersaturation level, the background counts were automatically measured three times a day.. Due to changes in external conditions and the state of the instrument, the background varies and if the operator though it was too high (> ca. 50 cm -3 ) or too low (< ca. 1 cm -3 ), the temperature settings were adjusted to keep the cut-off 145 sizes same as before. During the whole measurement period, the daily averaged (over all saturator flow rates) homogenous background varied from less than 1 cm -3 to almost a 1000 cm -3 . The background counts were subtracted from the data during data processing.
We investigated the fraction of scans discarded during data quality control as a function of the background concentrations 150 (daily median) (Figure 1). The u-shape of the bad scan percentage clearly shows that the quality of the scans goes down if the PSM background is either too low (<0.1 -1) or too high (>10). If the supersaturation level and consequently the background level inside the PSM is too low, the smallest particles cannot be activated and there is no detectable signal, which leads to noisy scans. A high background, on the other hand, can indicate that the PSM is not functioning properly.
Thus, based on the quality of the scans alone, the PSM appears to work most stably when the background signal at the 155 highest saturator flow rate is between 1 and 10 cm -3 . However, in our measurements, the devices were never intentionally run at background levels higher than circa 50 cm -3 . For this reason, the PSM could be stable at higher background levels as well, but our data does not allow us to draw conclusions on that. Furthermore, this behavior seems to be uncorrelated with https://doi.org/10.5194/acp-2020-719 Preprint. Discussion started: 27 July 2020 c Author(s) 2020. CC BY 4.0 License. the measured relative humidity at the measurement location, although laboratory studies have shown that RH can affect the particle activation efficiency with DEG (Kangasluoma et al., 2013;Jiang et al., 2011). 160

165
To investigate the effect of the background level (supersaturation) on the activation of the smallest particles, we split the data points to groups where the background is below 1 cm -3 , above 10 cm -3 or between these two limits. The limits were chosen based on the observed fraction of bad scans in Fig. 1. We then studied 1.1 -1.3 nm concentration as a function of the PSM background ( Figure 2). When the background is under 1 cm -3 , the measured concentrations are on average notably lower than when the background level is above 1 cm -3 , indicating that we are not activating all of the 1.1 -1.3 nm particles at those 170 settings. However, if the background level rises to over 10 cm -3 , the variation becomes bigger and the median concentration smaller, underlining the various factors affecting the concentration at higher background. At very high background levels the PSM likely activates large vapor molecules or clusters whose concentrations are not stable, leading to larger variation in the concentration. When these species dominate the activated particles, the particle size distribution cannot be easily resolved from the scans. Other causes for higher variation at high background include but are not necessarily limited to faulty 175 instrumentation, dirty sample lines and dominant homogeneous nucleation of the working fluid.
The analysis described above lead to the conclusion that in the conditions of the SMEAR II station, the optimal settings for the PSM is found when the measured background is between 1 cm -3 and 10 cm -3 . As mentioned before, the PSM could be https://doi.org/10.5194/acp-2020-719 Preprint. Discussion started: 27 July 2020 c Author(s) 2020. CC BY 4.0 License. run stably at background levels above 10 cm -3 as well, but due to the results discussed above, we selected for further analysis 180 only PSM data with the instrument background between 1 and 10 cm -3 .

185
The whiskers mark the location of the 95th and 5th percentile data points and the red plusses are outliers beyond those percentiles.

Measurement uncertainties
Measuring sub-3nm particles involves notable uncertainties, as small particles are very difficult to detect. Both PSM calibration and measurement are sensitive to the chemical composition of the particles being measured. The activation 190 probability with DEG seems to be lower for organic particles than for inorganic particles. The cutoff size, the diameter at which 50% of the particles are activated in the PSM, can be over a nanometer larger for organic particles than for inorganic particles (Kangasluoma et al., 2014, Kangasluoma et al., 2016b, but there is no systematic studies for different kinds of ambient particles. Because we do not know the exact chemical composition of the particles in the ambient air, the sizing in the measurement contains uncertainties. In addition, the activation probability of particles is also slightly different for 195 charged and non-charged (neutral) particles (Kangasluoma et al., 2016b). https://doi.org/10.5194/acp-2020-719 Preprint. Discussion started: 27 July 2020 c Author(s) 2020. CC BY 4.0 License.
The PSM is calibrated by measuring particles from a known source. A certain particle size is selected with a differential mobility analyzer (DMA) and an electrometer is used as a reference instrument. This gives us the PSM detection efficiency for each selected size. The PSMs in this study were calibrated using charged tungsten oxide particles in the size range 200 between 1.0 and 3.2 nm in mobility diameter, as there is no good calibration method and reference instrument readily available for neutral particles. Therefore, the diameters given should be taken as activation-equivalent sizes (we assume that the particles would activate as charged tungsten oxide particles do). The PSM may also be sensitive to ambient conditions, mainly relative humidity (Kangasluoma et al, 2013;Jiang et al., 2011). More discussion on the uncertainties can be found in Kangasluoma et al. (2020) and references therein. 205 When dealing with long time series, an additional complication arises from changing and maintaining equipment. While the different PSMs used in the study over the years are essentially similar devices, they have slightly varying cutoff limits and detection efficiency curves, which has been taken into account during data processing, but which could still affect the final inverted concentrations. The data preprocessing and inversion method can also produce additional uncertainties which are 210 difficult to quantify (Lehtipalo et al., 2014;Cai et al., 2019).
To estimate the magnitude of error caused by the uncertainties related to PSM measurement, we compared the ion concentrations detected by the PSM to those from a Neutral Cluster and Air Ion Spectrometer (NAIS, described briefly in section 2.4), which is the only other instruments measuring in the same size range at SMEAR II. The ion concentrations 215 were acquired from a PSM with an ion trap inlet (Wagner et al., 2017, Kangasluoma et al., 2016a. The setup is otherwise similar to the PSM used in the rest of this study, but the ion trap is switched on every 8 minutes and then off again after 8 minutes. This allows us to differentiate between neutral particle and total particle concentrations and acquire the ambient ion concentration from the PSM.   nitrate ion-nitric acid dimers (HNO3NO3 − ) and nitrate ion-nitric acid trimers ((HNO3)2NO3 − ). The nitrate ion chemical ionization is a very selective method as nitrate ions react only with strong acids, such as malonic acid, sulfuric acid and methane sulfonic acid (Eisele and Tanner, 1993) and oxidized organic compounds that have at least two hydroperoxy (OOH) groups or other H-bond-donating groups (Hyttinen et al., 2015).
In the chemical ionization inlet ~20 liters per minute (lpm) of sheath flow is mixed with ~5 milliliters per minute flow of air 245 saturated with nitric acid (HNO3) and then guided to the ionization source. In the ionization source, nitric acid is ionized with The CI-APi-TOF measurements were conducted at a 35-meter altitude in the same area as the ground level particle measurements. The measurement height is above the forest canopy.  found that the HOM concentrations above and inside the canopy are similar when the boundary layer is well-mixed. The concentrations between these altitudes 255 may differ during a strong temperature inversion or a shallow surface layer in nighttime.
All the low-volatility vapor measurements were performed with the same instrument that was calibrated twice during this measurement period with sulfuric acid calibrator (Kürten et al., 2012). In the calibrations we achieved calibration coefficients 2.4e9 cm -3 for 2014-2018 and 4.6e9 cm -3 for 2019 onwards and used the same coefficient for all detected compounds. This assumption is valid for compounds that cluster with nitrate ions at the collision limit and have equal 260 collision rates. The collision rates of nitrate ions with SA and with HOMs are approximated to be very similar (Ehn et al., 2014). Mass spectra obtained from the instrument were analyzed using the "tofTools" program described in Junninen et al.
(2010) and unit mass resolution was used in peak integration. The uncertainty of the concentrations is estimated to be -50%/+100%.

Complementary data
The NAIS measures the mobility distribution of ions in the atmosphere between 0.8 and 40 nm and it can be used to measure either naturally charged ions or the particles can be charged with a corona discharge to measure total particle concentration . We used an automatic atmospheric NPF event classification algorithm developed by Dada et al.
(2018) to determine NPF event times during the investigated time span. The event classification algorithm provided the start, 270 peak and end times of NPF events using data from the NAIS. Relative humidity data was used from the Rotronic MP102H RH sensor in the measurement mast at the SMEAR II station, measured at 16 (before 2/2017) and 35-meter heights. The global radiation data was measured at the same measurement mast, with the Middleton SK08 pyranometer at 18-meter height (before 9/2019) and with the EQ09 pyranometer at 35-meter height.

Analysis methods for comparing PSM and CI-APi-TOF data 275
We used the time series of quality controlled and inverted sub-3nm particle concentrations to study the diurnal and seasonal patterns of sub-3nm aerosol particles. The same seasonal analysis was performed on the available CI-APi-ToF data.
Then, measured sub-3nm particle concentrations were then compared to the vapor concentrations to determine correlations between observed particle and vapor concentrations during NPF events. In order to ignore the effect of the diurnal cycles on the analysis, only events that occurred between 10:00 and 14:00 were included in the correlation analysis. Correlations were 280 also separately investigated for spring-and summertime NPF events. There were not enough data points for events during autumn and winter for separate analysis during those seasons.
We compared the particle concentrations with measured SA and HOM concentrations since they have been identified to participate in NPF in laboratory studies (Sipilä et al., 2010;Kirkby et al., 2016). The HOM molecules were divided to 285 monomers and dimers, as well as nitrates and non-nitrates according to their elemental composition. For each category, we summed up the concentrations of the selected peaks. For our purpose, it is not necessary to identify all possible peaks in each category, but to obtain the temporal variation of different types of particle precursor.
Sub-3nm particle concentrations were also compared to combinations of different precursor molecule concentrations since particle formation might involve several different vapor species. Laboratory experiments replicating boundary layer NPF in forested regions (Riccobono et al., 2014;Lehtipalo et al., 2018) and analysis of field data sets (Paasonen et al., 2010) have 305 shown that particle formation rates can be parametrized using a product of sulfuric acid concentration and organics concentrations. were selected based on correlation with the bolded mass peaks and summed together in order to increase the signal-to-noise ratio. HOM nitrate monomer peaks listed as "Several compounds" contain HOM nitrate monomers and radicals, but a single peak cannot necessarily be identified as the main compound.

Results
In the following section we present the 74-month time series of sub-3nm particle concentrations and the 31-month time 315 series of aerosol precursor vapors measured at the SMEAR II -station in Hyytiälä, southern Finland and their comparison for the overlapping time period.

Time series of particle concentrations
The entire time series of the particle concentrations are shown in Fig. 4. The concentrations show a clear seasonal pattern for all three size bins: 1.1-1.3 nm, 1.3-1.7 nm and 1.7-2.5 nm. We observe a clear annual maximum during late spring and early 320 summer, and we also observe the lowest concentrations during the winter months, consistent with earlier observations at the  slightly wider size range in that study (up to 3 nm, where the largest size bin was obtained from the difference between PSM and a differential mobility particle sizer) and because their data was not filtered to remove scans with too high background.
The two months of measurements from 2014 at the beginning of the time series show a higher median concentration than the rest of the data, which could be due to the difference in background removal, as the background was measured manually at that time. 340 https://doi.org/10.5194/acp-2020-719 Preprint. Discussion started: 27 July 2020 c Author(s) 2020. CC BY 4.0 License. The diurnal patterns of particle concentrations in three size bins are shown in Fig. 5. We observe two maxima for the 1.1-1.3 nm concentration: one around midday and another during the evening. In this size bin, the measured concentrations can consist of both very small particles, large gas molecules or molecular clusters (Ehn et al., 2014); the distinction between 350 them cannot be made based on the measurement. Consequently, the daytime maximum can result from a combined effect of the diurnal behaviors of large organic molecules and newly formed molecular clusters. The diurnal variation of organic compounds is discussed below. Similarly, the evening-time maximum can be due to organic molecules or molecular clusters, which have been observed to form during evening time by biogenic ion-induced mechanism . Both 1.3-1.7 nm and 1.7-2.5 nm particle concentrations exhibit a daytime maximum in the afternoon. However, we observe a second, 355 larger maximum for the 1.7-2.5 nm particle concentration in the evening as well.
During regional NPF events, we expect the sub-3nm size distribution to behave differently than when there is no event, in the case that the formation of small particles takes place at our measurement location. In Fig. 5, we present diurnal cycles for NPF event and non-event days separately. Even though the event classification algorithm gives us exact event times, we used entire event days in this part of the analysis. The most noticeable difference between NPF event and non-event days is the strong midday maximum for both 1.3-1.7 nm and 1.7-2.5 nm particle concentrations on NPF event days. This maximum 370 does not appear during non-event days, leading to the conclusion that the increase in midday concentrations can be attributed to regional NPF. The concentrations are also generally higher during NPF than non-NPF days, indicating that conditions are favorable for cluster/particle formation. However, the smallest size bin does not show a clear difference during NPF and non-NPF days. This could mean that the production and sinks of 1.1-1.3 particles are large enough that the enhanced growth into the 1.3-1.7 particle size range during event times is not visible in the concentration. It also confirms that there is a 375 constant concentration of small particles/clusters present in the atmosphere, much like ion clusters 2013;Kontkanen et al. 2017).
https://doi.org/10.5194/acp-2020-719 Preprint. Discussion started: 27 July 2020 c Author(s) 2020. CC BY 4.0 License. We also analyzed the diurnal behavior of the measured aerosol precursor vapors in the same fashion as the particle concentrations discussed above. The diurnal patterns of SA, HOM monomer (nitrate and non-nitrate) and HOM dimer 420 (nitrate and non-nitrate) concentrations for the entire dataset are shown in Fig. 8. SA concentration has a similar diurnal pattern to that of global radiation, which is expected as sulfuric acid is formed in the atmosphere mainly through photochemical oxidation (Lucas and Prinn, 2005;Petäjä et al., 2009). HOM non-nitrate monomer concentration has a minimum in the early morning, with the concentration rising throughout the day until the maximum is reached after 18. In https://doi.org/10.5194/acp-2020-719 Preprint. Discussion started: 27 July 2020 c Author(s) 2020. CC BY 4.0 License. contrast, HOM nitrate monomer concentration exhibits a single daytime peak around midday similar to the sulfuric acid concentration. HOM dimer (both nitrate and non-nitrate) concentrations have different diurnal cycles than the other vapors, exhibiting minima during daytime and an increased concentration at night. Similar patterns were found by Bianchi et al. 435 (2017) using CI-APi-ToF data from spring 2013 in Hyytiälä and Jokinen et al. (2017) during a solar eclipse.

2015.
During regional NPF event days, the concentrations of all analyzed aerosol precursor vapors are higher than during nonevent days (Figure 8). However, the diurnal patterns of the precursor vapors are otherwise similar on event and non-event days. Additionally, we observe that the aerosol precursor vapor concentrations rise earlier and the difference in 440 concentrations between the night-time and the daytime is larger on NPF event days. These observations suggest that during https://doi.org/10.5194/acp-2020-719 Preprint. Discussion started: 27 July 2020 c Author(s) 2020. CC BY 4.0 License. event days there is more photochemical production and potentially also higher emissions of biogenic vapors. This is further corroborated by the diurnal patterns of global radiation in Figures 5 and 8.
The diurnal behavior of SA has a similar daytime maximum as the 1.1-1.3 nm particle concentration when we compare the 445 diurnal behaviors of the entire particle and vapor concentration data sets. Additionally, the diurnal behavior of sulfuric acid matches that of the 1.3-1.7 nm and 1.7-2.5 nm particle concentrations during event days. This points to the importance of sulfuric acid in atmospheric cluster formation and in the initial stages of aerosol growth. The HOM non-nitrate monomer maximum in the evening coincides with the observed peak in 1.1-1.3 nm particle concentration, implying that this maximum is likely due to formation of organic clusters. 450 In Fig. 9, we show the diurnal patterns of precursor vapor concentrations separately for each season. During spring, the diurnal patterns of all studied vapors exhibit similar behavior when compared to the diurnal behavior in the entire dataset.
This underlines the effect of spring-and summertime data on the entire dataset because the diurnal cycles during these seasons are fairly similar and will therefore dominate the dataset. 455 In summer, the diurnal behaviors of sulfuric acid and HOM dimer concentrations are fairly similar to springtime behavior.
The HOM monomer concentration does not rise as strongly during the day as during spring, but we still observe a sharp increase in HOM non-nitrate monomer concentration during the evening. Also, in summertime HOM nitrate monomer concentration reaches a maximum earlier than in spring, around 9:00 This is most likely because there is solar radiation 460 available for a longer period of time during summer. Overall, the HOM monomer concentrations are as much as five times higher during the summer than during spring and the concentrations of other HOMs are higher as well. The observed high HOM concentrations during summer can be explained by high emissions of organic vapors from the surrounding forest (Hellén et al., 2018) and increased photochemical activity.

465
In autumn, sulfuric acid concentration shows a daytime maximum around midday, but the concentration rises later in the morning and decreases earlier in the afternoon than during spring and summer. Additionally, the daytime sulfuric acid peak is smaller. Similarly, HOM monomer concentrations begin to rise later during the morning than in spring or summer, most likely because of the seasonality of solar radiation's diurnal behavior. Overall, the median concentrations and diurnal behaviors during autumn are comparable to springtime diurnal behaviors of the vapors, with the exception that the 470 concentration of HOM nitrate monomers is lower and more comparable to winter concentrations. HOM dimer diurnal behaviors are similar to spring, but nighttime concentrations are higher and the period of decreased concentrations during the day is shorter. These are again likely due to the seasonal differences in global radiation.
https://doi.org/10.5194/acp-2020-719 Preprint. Discussion started: 27 July 2020 c Author(s) 2020. CC BY 4.0 License. In winter, the concentrations of precursor vapors are lower than during other seasons and their diurnal variation is smaller.
Sulfuric acid and HOM monomer concentrations begin to rise later in the morning, around 09:00-10:00, again likely due to the seasonality of solar radiation diurnal behavior. HOM nitrate dimer concentration behavior remains similar to other 485 seasons, but the median concentration is lower. Notably, the HOM non-nitrate dimer concentration has a barely detectable diurnal cycle during the winter.

The connection between precursor vapors and the formation of sub-3nm particles
We used correlation analysis to investigate the relationship between atmospheric sub 3nm-particle concentrations and the 490 selected atmospheric vapors. The data is from NPF event times as specified by the NPF event algorithm and only from events occurring between 10 and 14 as to diminish the effect of the diurnal cycles on the correlations. Limiting our data selection to this time range diminishes the effect of meteorological variables on our analysis and allows us to focus on daytime NPF events. Because absolute vapor concentrations are not needed for this analysis, we used the aerosol precursor vapor data without the calculated calibration coefficients to eliminate this source of uncertainty. The results are shown in 495 Table 2.
Particle concentrations during NPF events show clear correlations with sulfuric acid and with HOM dimer concentrations.
Notably, sulfuric acid concentration correlates particularly with 1.7-2.5 nm concentrations, underlining its importance in the initial growth of the newly formed particles or determining when NPF events happen. The smallest particles correlate better 500 with HOMs than SA, especially when looking at the whole data set, confirming that this size range is influenced by organic molecules or clusters. HOM monomers correlate only with the smallest size bin, while HOM dimers correlate with all sub-3nm concentrations measured with the PSM, especially during event times. This indicates that HOM dimers are important in the formation of the smallest atmospheric particles, a finding consistent with Lehtipalo et al. (2018). The best correlation is found between HOM nitrate dimer concentration and 1.3-1.7 nm particle concentrations. 505 The combination of HOM (nitrate or non-nitrate) dimer and sulfuric acid concentration correlates most strongly with the 1.3-1.7 nm and 1.7-2.5 nm particle concentrations. This is consistent with laboratory experiments in the CLOUD chamber, showing that particle formation rates at 1.7 nm correlate with the product of sulfuric acid, ammonia and 510 HOM dimers . However, we observe that HOM nitrate dimers have a better correlation with the particle concentrations than the HOM non-nitrate dimers. We would expect the opposite, as non-nitrate HOMs have lower volatility than nitrate HOMs (Yan et al., 2020). This discrepancy with the laboratory results could be explained by the nitrate HOMs being better correlated with global radiation (Figure 8), as NPF most frequently occurs at the SMEAR II station during the global radiation maximum. However, the difference in correlation coefficients is not large, so it could also mean 515 that at least some of the nitrate dimers already have low enough volatility to participate in NPF, especially together with SA.
The scatter plots for the best correlations between atmospheric vapors and sub-3nm concentrations are shown in Fig. 10. It is possible that both sulfuric acid and large organic molecules are required for the formation and growth of new particles, which would explain the observed correlations. However, the correlation can also point to two separate formation pathways, 520 organic and inorganic. HOM nitrate monomers, on the other hand, do not correlate with sub-3nm concentrations, although https://doi.org/10.5194/acp-2020-719 Preprint. Discussion started: 27 July 2020 c Author(s) 2020. CC BY 4.0 License. they also show a daytime maximum like sulfuric acid, and in some earlier studies they have been connected to cluster formation (Jokinen et al., 2017). This supports the concept that HOM nitrate monomers have higher volatilities than other HOMs, so they might participate in later growth of particles, but not in clustering and initial growth (Yan et al., 2020; NPF events occurring between 10 and 14. The correlations of sub-3nm particle concentrations and sulfuric acid are on the left, the correlations of sub-3nm particle concentrations and HOM nitrate dimers are in the middle and the correlations of sub-3nm particle concentrations and the product of sulfuric acid and HOM nitrate dimers are on the right. Due to the differences observed in the diurnal cycles of both sub-3nm particle concentrations and vapor concentrations 540 between different seasons, we investigated the correlation between sub-3nm particle concentrations and atmospheric vapors in different seasons. Because of the lack of data for both vapor and particle concentrations during winter NPF events, we https://doi.org/10.5194/acp-2020-719 Preprint. Discussion started: 27 July 2020 c Author(s) 2020. CC BY 4.0 License.
were only able to analyze spring, summer, and autumn events. The results of the seasonal NPF analysis are shown in Table   3.

550
The analysis reveals clear seasonal differences between correlations of precursor vapors and sub-3nm particle concentrations. During springtime NPF events, HOM dimers correlate with all size bins of measured sub-3nm particle concentrations. This is consistent with the results of Yan et al. (2018), who compared vapor concentrations with particle measurements performed with the NAIS. Additionally, sulfuric acid correlates well with 1.7-2.5 nm particle concentrations.
Interestingly, HOM monomers anticorrelate with 1.1-1.3 nm and 1.7-2.5 nm particle concentrations during NPF events in 555 spring, maybe because dimer formation is a sink of the monomers.
During summertime events, we do not observe as many statistically significant correlations. HOM monomers correlate with 1.7-2.5 nm particle concentrations while correlations HOM non-nitrate dimers anticorrelate with 1.1-1.3 nm particle concentrations. It is possible that this is caused by the higher evaporation rate of HOM dimers with the increased temperature (Donahue et al.,2011). However, due to the lack of vapor concentration data from summer months, the amount of data 560 available for analysis here is limited and thus correlations may not be representative.
During autumn, HOM monomers correlate with the 1.1-1.3 nm particle concentration, a notable difference from spring.
correlations with the 1.3 -1.7 nm particle concentration and precursor vapors. These differences in correlations between 565 particle and vapor concentrations point to an annual variation in the formation mechanisms of sub-3nm particles.
It should be noted that SA do not form particles by itself at concentrations relevant to atmospheric boundary layer (Kirkby et al. 2011). Rather, it needs ammonia (NH3) or amines to stabilize the forming clusters. It is yet unclear if SA can form stable clusters with HOMs, although SA-organics nucleation has been proposed (e.g. Riccobono et al. 2014). Lehtipalo et al. 570 (2018), showed that SA and HOMs do not to interact unless NH3 is present. As there is no continuous ammonia and amine measurements available at SMEAR II, we could not include those in the correlation analysis, although variations in these vapors can affect the NPF mechanism and thus our results, especially the seasonal variation. Hemmilä et al. (2018) showed that there is a weak positive correlation between 1-2 nm particles measured with the PSM and ammonia and dimethylamine concentrations. 575

Conclusions
In this study, we analyzed five years of sub-3nm particle concentration and aerosol precursor vapor concentration data from the SMEAR II station in Hyytiälä, southern Finland. The sub-3nm particle concentrations were measured with the PSM and the aerosol precursor molecule concentration data was measured with the CI-APi-ToF. 580 The analysis of the PSM background counts and stability shows that to operate the PSM at the SMEAR II station in such a way that it reliably activates sub-3nm atmospheric particles, the measured background in the PSM should be within 1 and 10 cm -3 . Too low a background, and consequently too low a supersaturation level in the PSM, results in poor activation of atmospheric sub-3nm particles. When the supersaturation is too high, the measurement becomes unstable and the observed 585 concentrations are affected by homogenic nucleation of the working fluid. The settings of the PSM indicated by this analysis are valid for the SMEAR II station and other similar boreal background stations, but when measuring in other environments, the optimal background level may be different.
The size distribution of sub-3nm particles shows a clear seasonal cycle. The 1.1-1.3 nm particle concentrations have the 590 highest concentrations during the summer, which coincides with increased summertime photochemical activity and biogenic activity in the surrounding forest. The 1.3-1.7 nm and 1.7-2.5 nm particle concentrations show a marked increase during springtime, coinciding with increased regional NPF frequency. The diurnal patterns of sub-3nm concentrations exhibit clear daytime maxima around midday. This maximum is the clearest during spring and autumn, during which regional NPF events are also most common. A second maximum in the evening is observed for the 1.1-1.3 nm particles during spring and summer, but not for the concentrations in bigger size ranges or during wintertime. This maximum may be linked to organic clusters that form but do not grow to larger particles in the atmosphere.
The precursor vapors also show seasonal variability. The concentrations of all selected precursor vapors are the highest during summer and the lowest during winter. This is attributed to increased biogenic activity in the surrounding forest during 600 the warmer periods of the year as well as increased photochemical production. Additionally, the concentrations of sulfuric acid and HOM monomers have seasonally changing diurnal behavior because of solar radiation.
When comparing sub-3nm particle concentrations with aerosol precursor vapors, we found that the smallest particles (1.1-1.3 nm) correlate with HOMs when looking at the whole time series, indicating their presence in this size range. The 1.3-1.7 nm 605 and 1.7-2.5 nm particles, which are more directly connected to NPF events, correlate with SA and HOM dimers (and the product of these) during NPF events, but not with HOM monomers. There was no significant difference between nitrate and non-nitrate HOMs regarding their correlations with sub-3nm particles. The seasonal analysis of the correlations reveals some differences between the seasons, which could be due to changes in the mechanism forming clusters. However, understanding the seasonal differences in the formation mechanisms of HOMs and sub-3nm particles in detail requires 610 further studies.
Data availability. The meteorological data from the SMEAR II station can be accessed from the smartSMEAR website: http://avaa.tdata.fi/web/smart/. The data is licensed under a Creative Commons 4.0 Attribution (CC BY) license. The particle and vapor concentration data measured with PSM, CI-APi-ToF and NAIS are available from the authors upon request. and 739530 (ACTRIS). SMEAR II staff is acknowledged for their help in running the measurements.