Contribution of new particle formation to the total aerosol concentration at the high-altitude site Jungfraujoch (3580 m asl, Switzerland)

2.1 Site Description and Instrumentation

The Sphinx laboratory of the high-alpine research station JFJ (3580 m asl; 46.55°N, 7.98°E) is located on a crest between the summits of two mountains, the Jungfrau (4158 m asl) in the southwest and the Mönch (4089 m asl) in the northeast (Figure 1a). Therefore, local wind observations show mainly northwesterly or southeasterly direction. The distribution of wind directions during the LOP can be found in Figure 1b. Due to its remote location, the JFJ is considered to be a continental background site [Nyeki et al., 1998].

The JFJ is classified as a global station in the Global Atmosphere Watch (GAW) program and is part of the Swiss National Air Pollution Monitoring Network (NABEL). The laboratory is equipped with a large number of instruments for monitoring aerosol properties as well as greenhouse and reactive gases [Bukowiecki et al., 2016]. Trace gas observations used in this study are carbon monoxide (CO) measurements made with Cavity Ring-Down Spectrometry [Zellweger et al., 2012] and the sum of oxidized nitrogen species (NO_y) measured after conversion of NO_y to NO by chemiluminescence [Pandey Deolal et al., 2012]. In addition, meteorological parameters such as temperature, pressure, global radiation, wind speed and direction, and relative humidity are monitored by MeteoSwiss. In winter, the JFJ has been found to be in the FT most of the time; however, a significant PBL influence is observed (up to 80% of the time in summer and as low as 30% in winter) [Herrmann et al., 2015].

For this work, the nano-SMPS and SMPS (both custom-built TSI-type [Wang and Flagan, 1990]) are used to derive the size distribution from 5 to 600 nm. The total aerosol concentration above a diameter of 10 nm is measured with a CPC 3772 and the concentration above a diameter of 3.2 nm with a CPC 3776 (both TSI). Aerosol size information throughout the text is given in mobility diameter, all times are UTC + 1, i.e., local time without daylight saving time in summer.

As we consider particles with diameters below 50 nm, the aerosol concentration is corrected for diffusion losses in the sampling lines. To keep the diffusion losses and thus the corrections small, the nano-SMPS and the CPC 3776 were installed with two individual sampling lines of only 65 cm length each. Additionally, core sampling probes were installed [Wimmer et al., 2015]. The core sampling probe uses a makeup flow of 5 L min⁻¹ to increase the flow in this 65 cm sampling line. Right before the instruments, the actual sample aerosol flow of 1 L min⁻¹ is extracted from the middle of the tubing using an isokinetic probe. In that way, we could decrease the residence time in the sampling line to one sixth of its original value. The diffusion losses were calculated for laminar flow conditions [Kulkarni et al., 2011] and determined to be less than 10% for 5 nm particles and around 15% for 3 nm particles. The nano-SMPS data were additionally corrected for diffusion losses in the neutralizer (Krypton-85) and the differential mobility analyzer column (DMA) assuming an effective length of 1 and 7.1 m, respectively [Wiedensohler et al., 2012]. The major part of the sampling line was outside the building and heated slightly above 0° to avoid clogging by frozen water. The rest of the sampling line and the instruments were at room temperature. The SMPS and the CPC 3772 were installed at the total aerosol inlet of the Sphinx [Weingartner et al., 1999]. Setup and SMPS data analysis were described in detail by Jurányi et al. [2011].

The SMPS and nano-SMPS agree within 10% for the overlapping size range 20–90 nm when corrected for diffusion losses, charging efficiency, and CPC detection efficiency (Figure S1 in the supporting information (SI)). Comparison between CPC 3772 and CPC 3776 at times when the cutoff difference is of minor influence indicates a reproducibility of the particle number concentration measurements of approximately ±20% (Figures S2 and S3).

Signals from pollution due to local activities were removed from the data set upon visual inspection of single size distributions. Local pollution can be identified by its short-lived and drastic fluctuations in the aerosol properties [Herrmann et al., 2015]. Sources of local pollution include helicopter engines, diesel generators, and construction work as well as smoking on the visitor terrace of the Sphinx laboratory [Fröhlich et al., 2015].

2.2 NPF Event Classification

2.3 Formation and Growth Rate Determination

2.3.1 Formation Rates

The formation (or nucleation) rate J_D corresponds to the number of particles crossing a certain diameter D per time unit and per unit volume. It is typically expressed in cm⁻³ s⁻¹. Assuming a concentration (in cm⁻³) in a certain size range from D_min to D_max, a change in this concentration during a nucleation event at the JFJ can result from the following sink and source terms: (1) particles growing into the detection range, which corresponds to the true formation rate ; (2) particles growing out of the detection range (R_growth); (3) particle coagulation (R_coag); (4) net transport of particles at the JFJ including dilution (R_NetTrans); and (5) local primary sources (R_primary). Thus, represents an apparent formation rate that consists of the following terms [Bianchi et al., 2016]:

(1)

As a basis for the calculation we used the CPC 3776 concentration N_{3.2 − 1000}. Due to the high cutoff of the CPC, it is very unlikely that aerosol particles grow out of the CPC upper detection limit, so that R_growth is virtually zero. During nucleation events, meteorological conditions at the JFJ were typically stable. The background aerosol concentrations were rather low, such that nucleation events typically cause a substantial increase of the total particle number concentration. This implies that the contribution of the background aerosol to the term R_NetTrans during air mass changes (e.g., after change of wind direction) and during vertical transport (e.g., appearance of particles below 10 nm) is typically negligible. Still, transport processes could influence the determined nucleation rates, but an exact quantification is not possible which increases the uncertainty of the determined formation rates. We further assume that NPF events are spatially homogeneously distributed over a large area, as is commonly done. With this approach, we neglect R_NetTrans in the calculation of J_D but consider it in the uncertainty estimation of the formation rate (see SI and below). Changes due to local sources such as cigarette smoke were identified and removed, as previously stated; thus, R_primary does not need to be considered. Therefore, only coagulation losses over the full size range need to be corrected for. The coagulation sink, S_coag (s⁻¹), for a particle with a certain diameter D_j is determined in the following way [Kulmala et al., 2012]:

(2)

where is the coagulation coefficient (cm³ s⁻¹) between two particles with diameters D_j and corrected for the transition regime using the Fuchs correction factor [Seinfeld and Pandis, 2006]. is the number concentration (cm⁻³) in a certain size range with being the diameter midpoint of this size bin (geometric mean of size bin boundaries). For the calculation of S_coag, the concentration of the combined nano-SMPS and SMPS aerosol size distribution from 10 to 600 nm was used. Below 10 nm, the nano-SMPS size distribution was not used due to high uncertainties in the diffusion loss correction of the nano-SMPS system and in the charging equilibrium. Thus, one additional size bin from 3.2 to 10 nm was inserted, which was derived from the difference N_{3.2 − 1000}−N_{10 − 1000} (CPC 3776 − CPC 3772). From 3.2 nm to 10 nm the coagulation rate decreases exponentially, so that using the diameter midpoint could be misleading. Therefore, 3.2 nm was chosen as the representative diameter for this lowest size bin, such that the calculation provides an upper limit of S_coag. The coagulation correction term was then determined in the following way [Kürten et al., 2015]:

(3)

with the factor 1/2 avoiding double counting of collisions due to the double summation [Seinfeld and Pandis, 2006]. The summation was done over all size bins j of the combined size distribution. J_3.2 is then calculated as follows:

(4)

In the end, J_3.2 and dN_3.2/dt were often within 5%, and thus, a correction would not have been needed for most events. However, all stated J_3.2 values do include a coagulation correction.

The CPC concentration accuracy is 20%, the coagulation correction adds an additional uncertainty of 5%, as we only estimate an upper limit. Uncertainties related to transport have been estimated to 25% (Figure S4). Formation rate uncertainties were determined to be 50% in total.

3.1 Event Statistics and Meteorology

Figure 3 shows the frequency of all event types during NUCLACE. An average event frequency of approximately 17–20% was observed for types 1A, 1B, and 2 events, which is comparable to an earlier study at the JFJ by Boulon et al. [2010]. During the NUCLACE campaign, the JFJ was in or below clouds approximately 50% of the time. This reduces global radiation and thus limits the photochemistry necessary to form low-volatility vapors, such as sulfuric acid and highly oxygenated molecules (HOMs). Additionally, in-cloud conditions increase the condensation sink prohibiting aerosol nucleation. The actual NPF frequency during sunny days is ∼40%. Off-site nucleation events were detected during approximately 55% of the days, whereby Aitken mode bands were also observed during below-cloud days, when in-situ nucleation events were not. In-situ and off-site events add up to >100% for some months, because they occasionally took place during the same day.

No clear seasonality was found for in-situ nucleation events. A clear seasonal pattern is observed for the off-site events. Nucleation mode events are mostly observed during warmer seasons (spring through fall), when vertical transport occurs more frequently. During these periods, vertical transport is likely to occur. Aitken mode events were mostly observed during colder periods, when stable conditions were present (winter and spring).

The formation rates J_3.2 varied between 0.2 and 7.5 cm⁻³ s⁻¹ with a mean formation rate of 1.8 cm⁻³ s⁻¹ for type 1A events and 0.7 cm⁻³ s⁻¹ for type 1B events. The growth rates varied between 1.6 and 10 nm h⁻¹ for type 1A events and between 0.9 and 11.9 nm h⁻¹ for type 1B events. The mean growth rates for the nucleated particles in the size range between 5 and 15 nm were 4 nm h⁻¹ for type 1A and 4.8 nm h⁻¹ for type 1B events. Both, nucleation and growth rates, are comparable to other FT sites, as summarized by García et al. [2014]. Further details can be found in Table 1.

In the following, meteorological properties of type 1 events are compared to those of sunny non-event days and the full LOP in order to determine meteorological factors that possibly favor nucleation. In summer, off-site nucleation events are very common, due to the favored vertical transport. Thus, no sunny non-event day was found during this season. As a consequence, the category “sunny non-event” could show a seasonal bias in the meteorological data. Therefore, we additionally show the category “sunny” for reference, which includes all days with a global radiation ≥90% of the daily, expected maximal global radiation in the category sunny. In a next step, we determined the average monthly values of each meteorological parameter for an equal representation to minimize the seasonal bias. In Figure 4, variability of temperature, relative humidity, absolute water content, and global radiation are shown as box plots. The typical time window for nucleation to occur is between 09:00 and 15:00. Therefore, the median (horizontal line), the 25th and 75th percentiles (boxes), and the 10th and 90th percentiles (whiskers) of sunny non-event days, all sunny days (monthly average), and all campaign days were calculated for this time window. For nucleation days, these values were taken during the time when nucleation and growth were observed, to get statistics of the conditions during NPF.

The temperatures during sunny non-events are slightly higher compared to type 1 events, all sunny days, and the campaign average, despite the exclusion of the summer months. A strong effect of the temperature, however, was not observed, although one might expect an influence of the temperature as observed in laboratory studies [Kirkby et al., 2011].

The global radiation is decreased for sunny non-event days compared to all sunny days, which is due to the seasonal bias of sunny non-events. Type 1 events show a slightly higher value compared to the campaign average. This is expected, as photochemistry is essential for the formation of low-volatility vapors, such as sulfuric acid and HOMs.

The relative humidity (RH) is considerably lower on sunny non-event days and sunny days compared to type 1 events and the campaign average, while the absolute water content is lower on sunny non-event days only. An increased nucleation probability in air masses with high RH is possibly related to increased abundance of condensable vapors through the following causalities [Kulmala et al., 2014]. First, the increased RH could be an indicator of PBL influence, which increases the concentration of precursor gases. Sunny conditions without recent PBL contact are often characterized by low RH (typical FT conditions). Second, a lower water content also leads to a lower OH radical concentration, which could lower the production of low-volatility vapors.

Figure 5 shows the average wind direction for nucleation events and sunny non-events. During nucleation events, the wind direction was mainly from southeast. Non-event days behave similar to the campaign average, with the wind coming mostly from the northwest (for campaign average see Figure 1). As mentioned previously, the approaching air masses are funneled by the surrounding mountains thus confining observed local wind directions to either northwest or southeast. Therefore, the wind direction alone cannot be used as indicator for the exact air mass origin. Bianchi et al. [2016] further investigated the origin of the air masses in detail using the Lagrangian particle dispersion model FLEXPART [Stohl et al., 2005] for the same nucleation and sunny non-events as presented here. The most pronounced ground contact before arrival at the JFJ occurred at middistance including regions in northern France and also southern Italy when averaged over all air masses of type 1A events. Type 1B events also showed the most pronounced ground contact at larger distance including northern France, northern Germany, and the Benelux. Sunny non-events on the other hand showed generally small ground contact at larger distances.

3.2 Factors Controlling Nucleation at the JFJ

A previous study indicated that NPF at the JFJ is linked to PBL influence [Bianchi et al., 2016]. During a large fraction of the year (especially in winter), the JFJ is considered to be in the lower free troposphere [Collaud Coen et al., 2011; Herrmann et al., 2015]. The JFJ is very rarely in the PBL, but PBL influence is observed throughout the year and is most pronounced in summer. Mountain ranges favor the transport of PBL air toward the FT, creating the so-called injection layer [Nyeki et al., 2000; Henne et al., 2005a; Nyeki et al., 2002]. This injection layer is decoupled from the PBL and comprises about 20–30% air that originated in the PBL [Henne et al., 2005b]. In summer, the JFJ can sit within this injection layer, which can reach up to 4000 m asl [Nyeki et al., 2000]. Aerosol particle, precursor and oxidant concentrations are elevated in the injection layer compared to clean FT air masses [Henne et al., 2005a].

The CO/NO_y ratio was found to be a proxy for the age of an air mass [Zellweger et al., 2003]. At the JFJ, the CO/NO_y ratio can be used to estimate the time since (and strength of) the last PBL contact of the air mass. Ratios of 6.25–10 are observed close to anthropogenic sources. Ratios >100 indicate either a strong contact with subsequent long-range transport and dilution or a weak, recent contact. For free tropospheric conditions, values between 200 and 500 are observed at the JFJ depending on the seasonality (500 in winter, 200 in summer). As a remote site, JFJ is characterized by values above 100, whereby lower values (down to 20) can be observed during strong PBL influence. After a PBL injection, CO/NO_y continuously increases with time, approaching clean FT conditions [Herrmann et al., 2015].

Figure 6 shows the median diurnal variation of CO/NO_y for (a) type 1, off-site, and sunny non-event days and (b) winter 2013/2014 and summer 2013 days. Sunny non-event days are characterized by a high ratio, indicating rather clean free tropospheric conditions (corresponding to N₉₀< 60 cm⁻³ according to Herrmann et al. [2015]). For the type 1 events, this ratio is lower and around 200 indicating a very recent weak PBL influence or a strong PBL influence several days before arrival at the JFJ. This is in agreement with Bianchi et al. [2016] who showed that nucleation at the JFJ is favored within a short time period (24–48 h) after the last PBL contact of the air mass, as determined with FLEXPART dispersion modeling. If the observed nucleation mode particles were due to transport of PBL air into the FT directly at the JFJ, the CO/NO_y ratio would be seen to reduce rapidly during the arrival of this air mass. As CO/NO_y ratios were rather constant throughout the nucleation event days, this scenario can be ruled out. This conclusion is further supported by Figure 6b. For winter months, the JFJ is expected to be most of the time in the free troposphere [Herrmann et al., 2015]. As a result, vertical transport is hardly observed and therefore CO/NO_y ratios do not show a diurnal pattern.

During summer, CO/NO_y started decreasing around noon and only increased again after sunset, reflecting the time window when the injection layer can expand up to 4000 m asl [Nyeki et al., 2000]. From the 25th and 75th percentiles it is visible that the CO/NO_y in the morning varies strongly in summer. For days with very low values, the decrease is not as dominant. This can happen when PBL air was entrained very recently (the day before), so that the air mass did not restore to FT conditions yet. Under such conditions, an additional injection only slightly decreases the ratio. When looking at the summer event days separately, they only show a slight but distinct decrease in the afternoon (thus not visible in the yearly median), so that it is difficult to differentiate if the nucleation was triggered by an injection on the same or on a previous day as suggested by Bianchi et al. [2016] (see Figure S5 and Supporting Information S1).

Off-site Aitken events showed higher CO/NO_y ratios than type 1 events. We hypothesize that in these cases the particle formation resulted from an injection that occurred more than two days before arrival at the JFJ. However, we cannot determine if nucleation happened in the PBL or in the FT. During off-site nucleation days, the CO/NO_y ratios were slightly elevated in the morning but decreased around noon, which is the typical onset time for vertical transport. This supports our hypothesis of local nucleation mode particle bursts during vertical uplifting.

3.3 Parametrization of Observed Formation Rates

Following the qualitative analysis of the PBL influence, we will now use additional factors such as the global radiation (Rad), temperature (T), relative humidity (RH), CO as indicator for the concentration of aerosol precursors, and condensation sink (S_cond), to quantitatively determine their influence on the formation rates. Only limited knowledge of the concentrations of the precursor species is available. The nucleating compounds have been identified by Bianchi et al. [2016] using APi-TOF measurements for the same events showing that sulfuric acid is of minor importance for nucleation at the JFJ. Furthermore, the SO₂ concentration is very low and below the detection limit most of the time (see FigureS6 in the SI). SO₂ is therefore not considered. Bianchi et al. [2016] showed that the identified HOMs differed from the ones found in the nucleation and initial growth of α-pinene oxidation products in the absence of sulfuric acid [Kirkby et al., 2016; Tröstl et al., 2016]. They therefore hypothesized that at the JFJ the nucleating HOMs originate in the PBL from anthropogenic rather than biogenic sources, even though the latter could not be excluded. Thus, we use the CO concentration as a tracer for PBL influence, which implicitly is an indicator for the organic precursors forming the HOMs.

We cannot determine the diffusion coefficient of the organic nucleating species, which is necessary to determine S_cond. Therefore, we use S_coag calculated at a fixed diameter of 3.2 nm as a proxy for S_cond. For such small diameters, S_cond and S_coag are highly correlated; thus, using S_coag implicitly considers the combined effect of precursor losses due to condensation and losses of newly formed particles due to coagulation.

We applied least square fits (minimizing χ²) to assess the importance of several parameters and variables or combinations of them as summarized in Table 2. This parametrization could only be done on a subset of 14 NPF events, for which all variables were available.

In an initial step, we tested how well single variables correlated with the formation rate J_3.2 (see Table 2). Temperature and RH show the highest χ² and are not further considered at this point. Although CO, CO/NO_y, S_coag, and the global radiation already show a low χ², one variable does not satisfactorily describe the observed formation rates. By combining these best correlations, χ² further decreases. Additionally, we tested different combinations of three variables including CO, CO/NO_y, S_coag, and global radiation. Combining CO, CO/NO_y, and the global radiation yielded the best result with χ²=1.37. However, already a combination of CO and radiation gives a good result with χ²=1.44. The addition of CO/NO_y thus only causes a modest improvement of the fit. Finally, a fit involving all four variables listed above did not lead to further improvement. Moreover, the fits always yielded a positive exponent for S_coag, which is physically unexpected. Adding more variables such as RH and temperature does not further improve the goodness of the fit. One also has to note that while introducing more variables and fit parameters could mathematically improve the fit, the interpretation would be quite difficult due to the limited number of data points. Figure 7 shows the best fit results of the parametrization using CO, global radiation, and CO/NO_y and the best single variable fit using CO.

From Table 2 it is not directly obvious to which factors the formation rate is most sensitive, as the used input variables have different units and magnitudes, so that the magnitude of the derived fit parameters does not directly show the importance of the input variable. Therefore, we used the best fit parametrization to perform a sensitivity analysis. For this we used the interquartile ranges and the median of the variables CO, CO/NO_y, and the global radiation. We compared the resulting formation rate using the median values to the values obtained by exchanging one median value either with the 25th percentile or the 75th percentile of one variable only, leaving the median values of the others fixed. Using the 25th/75th percentiles, the formation rate changed by +11.8%/−4.1% for CO, −7.8%/+6.6% for CO/NO_y and +22.4%/−15.8% for global radiation. The formation rate is thus most sensitive to global radiation and the CO concentration which directly corresponds to the available concentration of condensable vapors. The PBL influence as determined by the CO/NO_y ratio has the weakest impact of these three variables on the formation rates, probably because the effect is already mostly covered by CO.

Due to the small subset of observations and the lack of precursor quantification, this parametrization cannot provide a complete representation of the factors controlling nucleation at the JFJ. Nevertheless, it gives some further insight into nucleation criteria at the JFJ. Meteorological parameters such as temperatures and RH seem to have a minor impact on the formation rate at the JFJ in contrast to our expectations. A temperature dependence of formation rates has been observed previously in the laboratory [Kirkby et al., 2011]. This dependence could be obscured by other factors more important for nucleation at the JFJ. The RH was significantly higher on in-situ event days compared to sunny non-event days. The poor correlation of RH with the formation rate suggests that the elevated RH on in-situ event days mostly reflects the recent PBL influence. Further, the positive correlation with S_coag indicates that the coagulation/condensation sink is not a limiting factor for NPF at the JFJ due to the generally low values. S_coag and S_cond are even higher on events days compared to non-event days, as also reported by Boulon et al. [2010], so that this positive correlation is attributable to condensing vapors due to PBL injections. The variables used for our best estimate are indicators for the PBL influence (CO/NO_y), the required photochemistry (global radiation), and the proxy of aerosol precursor concentration (CO). In summary, the availability of condensable vapors and global radiation is more important than the condensation sink and meteorological parameters at the JFJ.

3.4 Particle Production Due To NPF

Figure 8 shows the seasonal diurnal pattern of the mean total aerosol number size distribution from 5 to 80 nm (all days). Figure 8a shows winter months and Figure 8b summer months during NUCLACE LOP. We compare these two seasons as the difference in the PBL influence is most prominent. Confirming findings from the previous section, the growth stops at ∼30 nm in winter, with a mode diameter of ∼20 nm, due to the shorter daytime (less radiation) and less available vapors (less photochemistry and/or less PBL influence). Nucleating particles can be observed until 18:00. A persistent band is visible around 10–30 nm. This band most likely originates from previous nucleation events, either at the JFJ or off-site. Growth stops as well around 18:00 and the concentration of nucleated particles decreases. This is most likely due to air exchange in the free troposphere, allowing undisturbed air masses to reach the JFJ.

For the summer months (Figure 8b), the particles reach ∼50 nm, with a mode diameter of ∼30 nm. This is most likely due to the longer sunshine duration and the higher intensity in summer which allows for more photochemistry and thus stronger oxidation of the aerosol precursors and also more time to grow the freshly nucleated particles. Nucleating particles can only be observed until 15:00, which is earlier than in winter. The stronger PBL influence in summer results in an enhancement of the aerosol precursor concentrations; however, the higher condensation sink (and possibly a higher temperature) could counteract nucleation at an earlier stage than in winter, whereby it was shown in the previous section that the condensation sink in general is not a limiting factor at the JFJ. A persistent band in the Aitken mode is observed similar to winter; however, with a mode diameter at ∼40 nm. This band likely originates from two processes. On the one hand NPF in the FT is followed by transport to the JFJ. On the other hand, a large contribution to the aerosol number size distribution from convective uplifting is observed. During day time, the concentration of this band, and also the concentration of larger sizes, increases, which is not observed in winter, as also reported by Herrmann et al. [2015]. We also see the appearance of particles <10 nm without growth due to off-site nucleation events. Growth of newly formed particles is mostly observed in the afternoon, whereby NPF events were sometimes observed before noon. Thus, there is a distinct difference between winter and summer NPF.

In Figure 9, we further investigate the direct impact of NPF on the particle number concentration in three different size ranges: (1) 10–50 nm (N_10–50), (2) 50–90 nm (N_50–90), and (3) >90 nm (N₉₀). The figure shows the median diurnal variation of the number concentration for these three size ranges for (a) type 1 events, (b) sunny non-events, and (c) off-site Aitken mode events. First, we consider the particle number concentration in the 10–50 nm size range, as the nucleated particles do not grow directly beyond this size during in-situ nucleation (as seen in Figure 8). During NPF events, the number concentration N_10–50 increases by an order of magnitude from 100 particles to over 1000 particles cm⁻³ during the typical time window (09:00–15:00) when NPF was observed. This number is very likely to come solely from the NPF events. This is supported by the flat CO/NO_y ratio shown in Figure 6, indicating that vertical transport during these events yielding additional particles was of minor importance. The later decrease around 18:00 in N_10–50 is explained by air exchange, allowing undisturbed air masses to reach the JFJ as discussed above, or potentially scavenging in clouds. During off-site nucleation events, N_10–50 is slightly increased compared to sunny non-event days and lower compared to type 1 nucleation days. This supports our hypothesis that these particles nucleated elsewhere and got diluted during the transit time between the location of the nucleation event and the JFJ.

At the JFJ, the mean CCN activation diameter is typically at about 90 nm [Jurányi et al., 2011; Hammer et al., 2014]. The nucleated particles did not reach this diameter within the accessible time frame of 1 day. Nucleation at the Jungfraujoch therefore does not directly contribute to the CCN number concentration (see Figure 8), while growth of preexisting particles due to condensation across the CCN size threshold will be addressed below.

The coagulation sink for particles >10 nm is typically below 10⁻⁶ s⁻¹ outside clouds as determined from the SMPS, resulting in a lifetime of several days, making in-cloud scavenging the likeliest sink. Thus, under cloud-free conditions, these particles can further grow during the following days, suggesting that a multistep growth process allows for particles nucleated at the JFJ to reach CCN size.

Next, we consider particles in the size range 50–90 nm and above 90 nm. During type 1 events, the number of particles with diameters between 50 and 90 nm and larger than 90 nm increases during the typical time window for nucleation events. This is due to the condensation of vapors on preexisting particles growing them into these size ranges. This is supported by Figure 10, which combines SMPS and nano-SMPS data for type 1 nucleation days when both instruments collected data (21 out of 36 event days without data gaps during the event days for both instruments). Preexisting particles grow simultaneously with NPF and reach diameters above 90 nm. Approximately 50–100 cm⁻³ particles (median) grew above 90 nm during NPF events. In principle, this is to be regarded as an upper limit, as part of these particles could originate from the PBL. However, this number is likely to be very close to the true number, as no decrease in the CO/NO_y ratio was observed on these days indicating stable conditions. The change of N₉₀ during nucleation events can thus likely be explained by the growth of preexisting particles in the Aitken mode.

To consider a possible multistep growth of nucleated particles, we further investigated the off-site nucleation events (see event description). These events can be observed at the JFJ during 55% of the days. Here we only consider Aitken mode events, as the growth is not fast enough to grow nucleation mode particles to diameters larger than 90 nm on the same day. Figure 11 shows four example observations, which support our proposed multistep mechanism. The Aitken mode bands can be observed over days during stable conditions and also during the night, as their lifetime can be rather long in the FT (see Figure 11c). These narrow bands appear frequently and look very similar to the bands formed in a nucleation event (see Figures 11a and 11b). However, their concentration is lower as dilution has already taken place. Event a (2 March 2014) is a particularly strong nucleation with subsequent growth over a period of 9 h. Event b (25 August 2013) was another strong nucleation event; however, the growth could not be fully observed. This interruption was caused by an air mass change evidenced by an abrupt change in wind direction. During event d (27 February 2013), a band appeared right before the nucleation event. A fraction of those particles that nucleated earlier could eventually reach CCN size via a multistep growth process in the FT. This mechanism was also observed at different high-altitude stations such as Pyramid station and Chacaltaya, as visible from the aerosol number size distribution [Venzac et al., 2008; Rose et al., 2015a], but it was not directly addressed. This multistep growth process could therefore be important at other high-altitude sites as well.

To further differentiate between NPF and PBL contribution to the CCN number concentration and to confirm the findings on a longer and thus more robust data set, we consider the diurnal variation of N_10–50, N_50–90, and N₉₀ for (a) winter and (b) summer, during a 6 year period in Figure 12 (the same data set that served as a basis for Herrmann et al. [2015]). In winter, a change in N_10–50 is visible, with ∼100 cm⁻³ additional particles in this size range appearing in the time window when NPF occurs. In comparison, Figure 9 showed that ∼1000 cm⁻³ are formed during an NPF event. To compare this to Figure 12, this number (∼1000 cm⁻³) needs to be scaled with the in-situ NPF event frequency of ∼20%. This yields 200 cm⁻³, so that the change in N_10–50 can be explained by in-situ NPF. It needs to be noted that the event frequency can vary significantly from year to year.

N_50–90 and N₉₀ slightly increase during the typical time of nucleation. As argued already above, this was most likely due to the growth of preexisting larger particles in the Aitken mode. These preexisting particles most likely resulted from NPF and indeed contributed to the CCN concentration via a multistep growth process as depicted in Figures 10 and 11d, so that NPF may be a main source of CCN in winter at the JFJ.

In summer, the background concentration N_10–50 is very similar to winter. The increase is approximately a factor of 2 higher compared to winter (∼200 cm⁻³ in summer and ∼100 cm⁻³ in winter). Also, the temporal evolution looks different, showing two peaks, a short one in the morning and a longer one starting around noon. This is in agreement with Figure 8b, where the first peak is due to arriving nucleation mode particles and the second peak due to NPF events. The background concentration and the increase in N_50–90 and N₉₀ in summer is much higher compared to winter and also much higher compared to Figure 9 and can therefore not be explained by the growth of preexisting particles. The concentrations also show a double-peak feature. The first peak starts around 08:00 in the morning and the second around noon. The first increase in N_50–90 and N₉₀ is most likely due to local upslope venting caused by heating of the mountain slopes. The second peak is observed in parallel to the decrease in the CO/NO_y ratio and is most likely a result of the approaching injection layer. The increase is in agreement with a previous study [Herrmann et al., 2015], where it is shown that N₉₀ increases by ∼170 cm⁻³ (median) during PBL injections at the JFJ.

To summarize, our data indicate that in winter nucleation at the JFJ is due to the oxidation of aerosol precursors entrained from the PBL (with a window of opportunity of approximately 24–48 h ago according to Bianchi et al. [2016]). In summer, nucleation is also accompanied by an increase of aerosol precursor concentrations due to PBL injections. NPF can be observed before noon, when it is likely that the precursors have been entrained before, similar to winter nucleation events, or by local upslope winds. Or it can be observed in the afternoon, when the entrainment occurred via the injection layer, immediately triggering nucleation and subsequent growth to larger sizes (see also Supporting Information S1 and Figure S5). Therefore, in summer it is not possible to clearly identify if the nucleation was triggered by an injection 24–48 h ago, as is typical for winter, or by an injection at the same day, because the PBL injections occur too frequently. Also, it cannot be determined if the JFJ was actually within the injection layer or not, for this additional measurements (e.g., lidar) would be required, which was not available during NUCLACE. The major fraction of N₉₀ at the JFJ is expected to be entrained from the PBL and did not form via condensation during NPF. In contrast, NPF could be a major source of CCN at the JFJ in winter, via a multistep growth process of nucleated particles.