2.1 Measurement site and period

The NASCENT campaign took place in the Svalbard archipelago (also known as Spitsbergen) close to the small seaside settlement of Ny-Ålesund from October 2019 to October 2020. Svalbard itself has very limited anthropogenic aerosol emissions, but it can be influenced by North Atlantic stormy air masses. In order to limit the influence of locally produced sea spray aerosols and make the results more regionally representative, a measurement station located on top of Mount Zeppelin (475 m a.s.l.; 78^∘54^′ N, 11^∘53^′ E), approximately 2 km south of Ny-Ålesund, served as a sampling site for all data presented in this study, unless explicitly stated otherwise.

2.2 Instrumentation

3.1 Particle number size distribution, composition and aerosol hygroscopicity

Based on the comparison between both DMPSs given in Sect. 4.1, we utilized data from both DMPSs in the analysis of this study so that gaps in DMPS_1 data are filled with data from DMPS_2. All figures displaying aerosol number concentrations thus include data from both DMPSs.

Organic mass concentration was provided by the ACSM, while organic carbon concentration was measured by a high-volume sampler. The organic carbon concentration given by the filter analysis from this instrument was multiplied by a factor of 2.2 to obtain an estimation of the organic mass concentration, following the recommendations of Turpin and Lim (2001).

Among the species measured by the ACSM, chloride and sodium are assumed to be the only ones predominantly present in the coarse mode. However, the two DMPSs only measured in the submicron range, and the ACSM does not measure sea salt. This is an issue because the inputs of the cloud droplet parameterization should consider the size distribution and hygroscopicity of the same aerosol population. For that reason, chloride and sodium were not included in the calculation of aerosol hygroscopicity. However, we know that these compounds can contribute to the fine mode and increase submicron hygroscopicity at Zeppelin, mostly in winter, as shown by Zieger et al. (2010) and Adachi et al. (2022). To estimate the potential effect of these compounds on our droplet formation results, we performed a sensitivity study, detailed in Sect. 4.3.

Both the ACSM and the filter pack allow for the retrieval of inorganic nitrate, sulfate and ammonium mass concentration. Volume fractions of neutral salts and their hygroscopicity parameters are used as inputs to the volumetric mixing rule required to calculate the total hygroscopicity of the aerosol. For both datasets, we used the simplified ion-pairing scheme detailed in Gysel et al. (2007) to calculate the concentration of neutral salts from that of ions, using daily averaged data. From there, the total hygroscopicity parameter κ was introduced by Petters and Kreidenweis (2007) to describe aerosol hygroscopicity based on a semi-empirical parameterization of the Raoult effect. Knowing the hygroscopicity value of each neutral compound – we utilized values listed in Petters and Kreidenweis (2007), Carrico et al. (2010), and Zieger et al. (2017) – the mass-mixing rule enables us to estimate the overall aerosol hygroscopicity. We utilized filter-pack-derived hygroscopicity values as input to the cloud droplet parameterization described in the next section because the filter pack shows fewer data gaps than the ACSM over the duration of the NASCENT campaign. ACSM-derived hygroscopicity is thus only used as a validation of the filter-pack-derived hygroscopicity values retrieved (see the comparison of both in Fig. S3 in the Supplement).

3.2 Cloud droplet number concentration and cloud supersaturation

Knowledge of particle number size distribution and overall hygroscopicity parameter κ as well as of air temperature and pressure allows for the determination of the potential cloud droplet number concentration, N_d; the maximum available cloud supersaturation, S_max; and the minimum diameter required for an aerosol particle to activate to a cloud droplet, D_act, using a cloud droplet formation parameterization. Note that the term “potential” is used to describe these outputs because the parameterization results include periods when no clouds were effectively present at Zeppelin. Based on cloud parcel theory, this parameterization was initially developed by Nenes and Seinfeld (2003) and improved with new implementations successively by Fountoukis and Nenes (2005), Barahona et al. (2010), and Morales Betancourt and Nenes (2014). Results of N_d and S_max are then constrained by updraft velocity measurements, here given by a uSonic and a wind lidar (Sect. 2.2.3). The activation parameterization is based on cloud parcel theory and solves the equations that describe droplet formation in an ascending air parcel containing aerosols and water vapor, specifically at the point where supersaturation becomes equal to S_max; N_d is then equal to the number of CCN with critical supersaturation less than S_max. The parameterization uses as inputs the observed pressure and the temperature, the aerosol size distribution data, the hygroscopicity parameter κ, and the updraft velocities.

We extracted wind lidar data corresponding to the updraft velocity at 500 m above ground level to make them comparable to uSonic data from the Zeppelin station. We noticed surprisingly high values of vertical velocity measured by the uSonic during northern wind conditions (Fig. S1a and b), which we attribute to the presence of winds orographically lifted by Zeppelin Mountain. Based on Fig. S1a and b, we decided to discard uSonic data when the wind direction was between 335 and 15^∘ so that any droplet calculation made is more representative of regional conditions than specific conditions at Zeppelin during strong orographically driven updrafts.

The high resolution of wind lidar and ultrasonic anemometer data reveals the highly variable nature of vertical velocity; to calculate droplet numbers relevant for the average cloud, we use the probabilistic approach detailed in Georgakaki et al. (2021): wind lidar data are grouped by hour, and each block of 1 h data is fitted to half-Gaussian probability density functions (PDFs) with a zero mean and a standard deviation σ_w. The cloud droplet formation parameterization is then applied for a characteristic velocity, $w*=\mathrm{0.79}$ σ_w, as this provides the average droplet number over the distribution of positive velocities in the domain. A comparison of the results of this analysis derived from the uSonic and the wind lidar is shown in Fig. S2. Several studies performed using this approach gave successful droplet closures for N_d and S_max in various types of clouds (e.g., Fountoukis et al., 2007; Kacarab et al., 2020; Georgakaki et al., 2021). Note that this approach is not valid for boundary layers that undergo deep convection but only for low vertical velocities typical of boundary layer ascent and descent over a diurnal cycle, which is the case for Zeppelin (Di Liberto et al., 2012; Collaud Coen et al., 2018).

4.1 Particle number concentration and size distribution

Figure 1a presents N_aer time series, with values that show a minimum in October 2019 (20–60 cm⁻¹) followed by a relatively constant increase until an upper plateau (100–1000 cm⁻¹) reached between May and August 2020. A sharp decrease in N_aer is then observed towards the October minimum. These measurements are in good agreement with the annual cycle of integrated particle concentration at Zeppelin reported by Tunved et al. (2013), who measured particle number size distribution between 20 and 630 nm before 2005 and between 10 and 790 nm after and averaged the results over 10 years from 2000 to 2010. However, we note a 1-month lag in the appearance of the high-concentration plateau (reported from April to July, whereas it appears from the beginning of May to the end of August in the present study). Aerosol levels measured at Zeppelin during the NASCENT campaign can thus be considered typical of this site. Overlapping periods of both DMPSs allow for a comparison of N_aer (see Fig. 1b) and confirm the strong correlation between both instruments.

Figure 1(a) Air temperature measured by a probe (red curve), periods of precipitation (vertical blue bars) and integrated particle number concentration N_aer (green dots) as measured by two DMPS systems at the Zeppelin station, displayed as time series. It should be noted that the two DMPSs have different size ranges: 20 to 775 nm for DMPS_1 and 10 to 945 nm for DMPS_2. (b) Comparison of integrated particle number concentration N_aer as measured by two differential mobility particle sizer (DMPS) systems at Zeppelin, with the date as the color scale.

Download

4.2 Aerosol hygroscopicity parameter κ

The time series of κ values derived from both the filter pack and the ACSM data is shown in Fig. S3a. Both instruments, despite being based on techniques with a different aerosol size cutoff (PM_2.5 for the ACSM and PM₁₀ for the filter pack), generally provide κ values that agree to within 50 % for the majority of data points (Fig. S3b). The mean hygroscopicity parameter κ value over the whole campaign was 0.40 when derived from the ACSM and 0.32 from the filter pack. The temporal trend shows a relatively constant κ value of around 0.3 in fall, winter and spring but a slightly lower value in summer, dropping to approximately 0.2, although rather variable.

A small number of field campaigns at Zeppelin have investigated aerosol hygroscopicity. In summer 2008, Zábori et al. (2015) utilized both bulk chemical composition and size-resolved CCN measurements, retrieving κ values of 0.5 and 0.3–0.4, respectively, slightly higher than our results. Zieger et al. (2010) measured wet and dry aerosol scattering from July to October 2008 using humidified and dry nephelometers. Using Mie theory, they were able to retrieve the hygroscopicity parameter κ and found a mean value of 0.57. This is also higher than the values we report in Fig. S3a, but the techniques used and the seasons studied are different. The median κ value of 0.23, retrieved by Jung et al. (2018) and based on 5 years of CCN-scanning mobility particle sizer (SMPS) measurements, however agree with our results. Year-round hygroscopic growth measurements by Rastak et al. (2014) led to the conclusion that the influence of hygroscopic growth on the direct radiative effect was higher in summer than during the Arctic haze period.

The hygroscopicity parameter has also been constrained in other locations in the Arctic, all of them combining a CCN counter and an instrument measuring particle number size distribution, either an ultra-high-sensitivity aerosol spectrometer (UHSAS), an SMPS or a DMPS, to provide input data for κ–Köhler theory calculations. Moore et al. (2011) and Herenz et al. (2018) both characterized springtime Arctic aerosol. The former reported values of around 0.4 for background air masses and of slightly below 0.6 for the Arctic boundary layer in aircraft flying over northern Alaska (USA); the latter retrieved a κ value of 0.23 at a station located in the Inuvik region in Canada. Martin et al. (2011) and Lathem et al. (2013) ran measurements in the summer season and found similar values (averages of 0.33 and 0.32, respectively), on an icebreaker on its way from Svalbard to the vicinity of the North Pole and on aircraft flights between Alberta (Canada) and Greenland. Also in summer but at a ground-based station in northern Sweden (68^∘ N), Kammermann et al. (2010) retrieved lower κ values, between 0.07 and 0.21, but the close presence of the Stordalen mire, known to emit organic precursors, could have caused a local reduction in aerosol hygroscopicity. At the Villum station in northeastern Greenland, Massling et al. (2023) reported κ values very close to the present study using CCN SMPS measurements: 0.28–035 in spring and 0.23–0.35 in summer.

4.3 Potential cloud supersaturation, droplet number concentration and activation diameter

The cloud droplet formation parameterization outputs S_max, N_d and D_act are displayed in terms of their frequency of occurrence in Fig. 2 and as a time series in Fig. S4 with the measured values of σ_w, which are used to constrain the parameterization. During periods of rain (noted in Fig. 1a), the aerosol load was strongly reduced, directly implying sharp decreases in N_d and D_act, which impact the other parameters. As expected, S_max and σ_w covaried during the whole year, since turbulence and vertical velocity are primary drivers of cloud supersaturation generation. Values of S_max were highest in fall, centered at around 1 % and reaching up to 4 % (Fig. 1b). Except for a drop in early August, S_max values ranged between around 0.5 % and 1.5 % in the second half of spring and in summer.

Together with Bougiatioti et al. (2020), Kacarab et al. (2020) examined the conditions for which cloud formation is insensitive to any increase in N_aer (i.e., updraft-limited regime) and found that it corresponds to when S_max is below 0.1 %. Subsequent studies (e.g., Georgakaki et al., 2021) also supported this criterion, which is the one we decided to use here as well. Figure S4b indicates that such conditions only occurred over short periods in winter and during the first half of spring. This shows that updraft-velocity-limited conditions can be found, although very rarely, in the Svalbard archipelago when aerosol concentrations approach the Arctic haze maximum in winter and early spring (Figs. 2a and S4b). This is an important result because when these conditions occur, cloud formation is not linearly influenced by these large aerosol loads but only until a certain threshold, which will be discussed in the following. The presence of very large particles, larger than the maximum diameter that can be detected by the DMPSs, could potentially scavenge water vapor and cause drops in S_max that are as low as those we observed. However, the mode diameter measured during the campaign was consistently between about 30 and 300 nm, and particles larger than 500 nm were very rare (not shown), in agreement with previous literature about the same site (Tunved et al., 2013; Pasquier et al., 2022b). We thus consider it very unlikely that low values of S_max are caused by this phenomenon.

N_d, however, is not clearly linked to S_max but follows the trend of N_aer shown in Fig. 1a (this is confirmed by the scatterplots in Fig. 3) because a higher number of aerosol particles goes with a higher number of CCN on which water vapor can condense, as shown at Zeppelin by Jung et al. (2018). On the other hand, higher N_aer also results in more intense competition for water vapor, leading to a decrease in S_max, which in turn tends to limit N_d. Measuring the annual cycle of cloud residual number concentration at the same site from November 2015 to February 2018 using a ground-based counterflow virtual impactor (GCVI) inlet, Karlsson et al. (2021) reported a similarly high and variable plateau between April and August as well as a minimum in fall. They however reported very low concentrations in January and February, followed by an extremely sharp increase from March to April. Our results show a steadier increase over the whole winter, in agreement with the increase in N_aer (Fig. 1a).

Clear seasonal variations in D_act can also be seen: the pristine-like conditions in fall and early winter associated with high cloud supersaturations led to low values of D_act, with frequent occurrences below 20 nm (Figs. 2c and S4d). These are even lower diameters than the values reported by Koike et al. (2019) – minimum D_act of around 30 to 50 nm – using air parcel model calculations. Together with relatively high S_max values during this period of the year, the very low aerosol concentrations, mainly concentrated in the accumulation mode (Tunved et al., 2013), lead to the efficient activation of the aerosol particles to cloud droplets, in agreement with previous results from Siegel et al. (2022). Although anthropogenic pollution transported from lower latitudes during the Arctic haze period (late winter and spring; Rahn, 1981; Hirdman et al., 2010) controls the CCN and droplet population, only a fraction of it was activated to cloud droplets, as larger D_act values, centered around 50–100 nm, were in the range of accumulation-mode particles typically linked to this type of atmospheric transport. Such D_act values persisted over the summer with a particularly high peak between 100 and 200 nm at the beginning of August. Similarly, the dominant Aitken mode, originating from the intense new particle formation activity, likely stayed to a very large extent in the interstitial (unactivated) phase.

The comparison between uSonic-derived and wind-lidar-derived updraft velocity shown in Figs. S1 and S2 provides relatively accurate but not very precise results, which could be due not only to fine-scale variability in vertical motion but also to larger-scale differences related to the horizontal distance between both instruments. However, Figs. 2 and S4 provide a good indication of the second-order influence of the updraft velocity on the outputs of the cloud droplet formation parameterization, since the N_d, S_max and D_act results based on the wind lidar and on the uSonic strongly agree.

With a view to consistency with data from the DMPSs, the cloud droplet parameterization outputs shown in Figs. 2 and S4 are based on hygroscopicity calculations that do not include sodium and chloride, under the assumption that submicron aerosol particles do not contain these compounds. According to Adachi et al. (2022), this assumption may not be fully correct, as particles of 1 µm in winter are found to be composed of 50 % by mass of sea salt, with a decreasing proportion with decreasing particle diameter. To address any effects of sea salt on droplet formation, we performed a sensitivity analysis assuming the extreme case for which half of the measured aerosol mass was sea salt and repeated the analysis detailed above. Using a κ value of 1.1 for sea salt, as suggested by Zieger et al. (2017), the overall hygroscopicity shifted from values of 0.2–0.3 (see Fig. S3) to around 0.7. Figure S5 shows the seasonal percent change such an increase in aerosol hygroscopicity has on N_d, S_max and D_act. The two former parameters are very slightly affected in fall, winter and spring (up to ∼5 % for N_d). In summer they undergo a 20 % change, but this is the season when the effective sea salt fraction is the lowest, as shown by Adachi et al. (2022), making such an effective change particularly unlikely. Values of D_act are more affected, although moderately, with a reduction of around 30 % regardless of the season. Overall, this sensitivity analysis shows that aerosol hygroscopicity effects from sea salt may have a second-order influence on the cloud droplet parameterization outputs and, thus, would not significantly affect the results and conclusions based on the base-case hygroscopicity value.

Figure 2Normalized number of occurrences of (a) S_max, (b) N_d and (c) D_act during the NASCENT campaign. The data are divided per season: fall (green), winter (light blue), spring (orange) and summer (red).

Download

4.4 Limiting droplet number concentration

During updraft-limited cloud formation conditions, values of N_d reach an upper limit, independently of N_aer, which Kacarab et al. (2020) suggested naming the limiting droplet number concentration, $N_{\mathrm{d}}^{\lim }$. This upper limit can be visualized as a plateau for which S_max drops below 0.1 % when plotting N_d as a function of N_aer. This is shown in Fig. 3, where as a sensitivity analysis, we prescribed three different σ_w values, 0.1, 0.2 and 0.3 m s⁻¹, for running the cloud droplet parameterization. Note that the three values of σ_w chosen here are representative of the stratiform cloud conditions that typically prevail in the Arctic (Shupe et al., 2008b). Based on the results in Sect. 4.3, updraft-limited conditions were observed in winter and early spring. Extracting mean plateau values corresponding to these seasons from Fig. 3, we retrieved $N_{\mathrm{d}}^{\lim }$ values of 173 cm⁻³ for winter and 128 cm⁻³ for spring, when an updraft velocity σ_w=0.1 m s⁻¹ is prescribed. These results are in good agreement with the 2-year-averaged peak concentration of cloud residuals measured by Karlsson et al. (2021) for April and May. This implies that when updraft velocity was low at the Zeppelin station in winter and early spring, only the fraction of the aerosol number concentration corresponding to these seasonal plateau values formed cloud droplets; any surplus of aerosol (very likely during Arctic haze conditions) remained in the interstitial phase.

Adding to the different regimes of cloud formation distinguished in Sect. 4.3., the $N_{\mathrm{d}}^{\lim }$ plateau was almost never reached in fall and in summer. Short drops in S_max below 0.1 % occurred, e.g., at the beginning of December and August as shown in Fig. 1b, but over periods that are too short for them to be associated with updraft-velocity-limited conditions. For that reason, we do not consider the summer S_max values below 0.1 % in Fig. 3 relevant $N_{\mathrm{d}}^{\lim }$ plateau values.

In addition, it is worth noting that when applying the same analysis with an assumption about σ_w being larger than 0.1 m s⁻¹ (i.e., 0.2 or 0.3 m s⁻¹; middle and right panels in Fig. 3) and even higher (not shown), the 0.1 % S_max threshold is not reached at all, neither in winter nor in spring, indicating that vertical-velocity-limited conditions cannot be found if the turbulence of the boundary layer is not extremely low.

Observing the shape of the plots displayed in Fig. 3 also provides information on the efficiency of cloud droplet formation and corroborates the results discussed above. The high S_max and corresponding low D_act in fall and winter are associated with the droplet activation of the whole aerosol population, leading $N_{\mathrm{d}}^{\lim }$ to match N_aer; these two parameters are thus represented in a scatterplot as a narrow band close to the 1:1 relationship. On the contrary, the scatterplot for the summer season shows a much larger spread with an offset from the 1:1 line, in agreement with a large proportion of Aitken-mode aerosol particles that did not activate to cloud droplets. Interestingly, the sharp transition between these two activation behaviors, also observed by Engvall et al. (2008) and which we have shown to occur in the middle of spring (Figs. 2a–c and S4b–d), results in a scatterplot where both staggered and matching N_d–N_aer relationships can be observed.

Extracting the plateau values of $N_{\mathrm{d}}^{\lim }$ from Fig. 3, we investigate their relationship with σ_w in different types of clouds located in various geographic locations using previous studies in which the same probabilistic analysis was performed (Fig. 4). Kacarab et al. (2020) measured from an aircraft in summer in the marine boundary layer over the southeastern Atlantic, a region that can be strongly influenced by biomass burning fires from the African continent. They retrieved values of σ_w of between 0.30 (relatively clean air mass) and 0.56 m s⁻¹ (more polluted) for which $N_{\mathrm{d}}^{\lim }$ was reached. Georgakaki et al. (2021) also reported σ_w values as high as 0.5 m s⁻¹ at two mid-altitude stations located in the Swiss Alps in central Europe. The fact that they had winter measurements allowed them to reach values as low as 0.1 m s⁻¹, making their results directly comparable to those shown here. They could link this value of σ_w to an $N_{\mathrm{d}}^{\lim }$ plateau of 108.1 cm⁻³. At the other end of the spectrum, Bougiatioti et al. (2020) reached an $N_{\mathrm{d}}^{\lim }$ plateau for σ_w values as high as 1.2 m s⁻¹ thanks to late-spring and summer flights in the boundary layer over the southeastern United States. They proposed the following linear correlation between $N_{\mathrm{d}}^{\lim }$ [cm⁻³] and σ_w [m s⁻¹]: $N_{\mathrm{d}}^{\lim }=\mathrm{1033.9}{\mathit{\sigma }}_w+\mathrm{112.28}$ (R²=0.92). Georgakaki et al. (2021) updated this relationship, including the results from Kacarab et al. (2020) and their own, $N_{\mathrm{d}}^{\lim }=\mathrm{1137.9}{\mathit{\sigma }}_w-\mathrm{17.1}$, and proved very strong agreement between all reported data (R²=0.94). The two additions from the present study, associated with winter and spring plateau values for σ_w=0.1 m s⁻¹, also agree well with these previous results, although they were retrieved in a very different environment, i.e., relatively clean with clouds mainly originating from maritime air masses and uplifted in mountainous terrain. This provides another confirmation of the robustness of the empirically demonstrated σ_w–$N_{\mathrm{d}}^{\lim }$ relationship regardless of the environment, type of clouds and aerosol sources.

Figure 3Predicted potential droplet number concentration N_d compared to integrated particle number concentration N_aer. Data are shown in 12 panels corresponding to four seasons and three different assumptions about updraft velocity σ_w: 0.1, 0.2 and 0.3 m s⁻¹. N_d is an output of the cloud droplet formation parameterization, and N_aer is measured by the DMPSs. The color scale is the maximum predicted cloud supersaturation S_max; note that data points for which S_max<0.1 % are shown in black. Plateau values of $N_{\mathrm{d}}^{\lim }$, calculated as the average value of data points for which S_max<0.1 %, are displayed as dashed grey lines.

Download

Figure 4Predicted limiting cloud droplet number concentration plotted against the corresponding assumption about the updraft velocity, comparing the present study with the existing literature. The linear fit displayed in black is the one given by Georgakaki et al. (2021); the red one includes results from the current study.

Download

4.5 Droplet closure

N_d is one of the outputs provided by the cloud droplet formation parameterization utilized in the present study (see Figs. 2b and S4c). The availability of instruments measuring this parameter facilitates the validation of the cloud droplet parameterization by comparing in situ data with the parameterization output. This was achieved several times in past studies through successful droplet closures (e.g., Fountoukis et al., 2007; Kacarab et al., 2020) and can also be performed here using measurements from HOLIMO taken on nine HoloBalloon flights during the NASCENT campaign (Fig. 5; see Pasquier et al., 2022a for detailed descriptions of the atmospheric conditions during the flights; we note that, based on the radar measurements they performed, the clouds studied on 10, 11 and 12 November 2019 were monolayer clouds, but the 1 April case is a typical seeder–feeder configuration, with a synoptic cloud above the sampled cloud). The closure, which is assessed to be attained when predictions were within a factor of 2 of observations, appears successful for N_d>8–10 cm⁻³, thereby validating the use of the cloud droplet formation parameterization in the Arctic environment. Values of N_d below 8–10 cm⁻³ approach the minimum threshold concentration for which a cloud can still be defined as such, potentially including periods when droplet nucleation is not effectively occurring or has been followed by out-of-cloud scavenging.

Mixed-phase clouds tend to rapidly glaciate (i.e., convert to pure ice clouds) due to the Wegener–Bergeron–Findeisen (WBF) process or riming (Korolev et al., 2017), which can effectively transfer mass from the liquid to the ice phase. Nevertheless, Fig. 5 provides evidence that even for a degree of glaciation (i.e., the fraction of cloud water that is in the form of ice) as high as 90 %, cloud parcel activation theory can predict N_d to within 50 % of observations. This implies that significant amounts of glaciation over the duration of the HoloBalloon flights, and possibly over spring and fall, may have occurred through processes that do not deplete droplet number, e.g., WBF that is promoted by secondary ice production (SIP) through ice–ice collisions or droplet shattering (Field et al., 2017; Korolev and Leisner, 2020). SIP through rime splintering is unlikely as it would have reduced the available N_d. This hypothesis is in line with the findings of Pasquier et al. (2022a), where the effect of SIP was inferred in about 40 % of the in-cloud measurements.

Previous observations (Borys et al., 2003; Lance et al., 2011; Norgren et al., 2018) reported that large aerosol loadings could hamper the efficiency of riming in mixed-phase clouds. Here we show that even in the very pristine conditions during winter and spring, the amount of riming does not seem to affect the droplet number concentrations significantly compared to what is expected from warm-cloud activation theory.

Figure 5Scatterplot of predicted against measured cloud droplet number N_d during nine HoloBalloon flights on 10, 11 and 12 November 2019 and 1 April 2020.

Download