3.2.1 Particle number size distribution and concentration

Particle size is one of the most important parameters to characterize the atmospheric aerosol. Figure 3 presents contour plots for PNSDs of supermicron particles in Fig. 3a and submicron particles in Fig. 3b. In order to show details of supermicron particles, we adopted different color bar scales for submicron and supermicron particles. Most of the time, two submicron modes (Aitken and accumulation mode) and one supermicron mode (coarse mode) are observed. The Aitken mode is observed from ∼10 to ∼80 nm, and the accumulation mode is observed from ∼80 to ∼1000 nm. However, from 03:30 to 20:00 21 September and from 09:30 28 September to 18:30 30 September, the submicron particles only exhibited a unimodal distribution. The supermicron particles exhibited a high concentration during those periods. N_total was changed significantly, from ∼200 to ∼1500 cm⁻³ with a median of ∼700 cm⁻³, shown as black line in Fig. 3b.

Figure 3Contour plots for PNSDs of 1000 nm to 10 µm in panel (a) and 10 to 1000 nm in panel (b). The color scale indicates dN∕dlogD_p in cubic centimeters (cm⁻³). Time series of N_total is shown by the black line in panel (b).

Download

Particles of different sizes have different formation routes, sources and behaviors. To better define the modes of our data, we fitted the PNSDs to three lognormal functions. The lognormal distribution was expressed by Seinfeld and Pandis (2016), and the parameterization function is as follows:

where N_i is the total number concentration of the ith mode, D_i is the geometric mean diameter of the ith mode and σ_i is the geometric standard deviation of the ith mode distribution. Every PNSD was individually parameterized by a trimodal distribution, where the number of i, $i=\mathrm{1},\mathrm{2},\mathrm{3}$, stands for Aitken, accumulation or coarse mode, receptively. For each PNSD, we searched for an optimal fitting function, until the coefficient of determination (R²) was larger than 0.97.

Time series of PNC in Aitken mode (N_Aitken), accumulation mode (N_accumulation) and coarse mode (N_coarse), together with sum of N_i and measured N_total, are shown in Fig. 4. Due to the unimodal distribution of submicron particles from 03:30 to 20:00 21 September and from 09:30 28 September to 18:30 30 September, the trimodal lognormal fitting did not work properly, so we did a bimodal lognormal fitting instead, with one submicron mode (N_submicron, as shown by purple dots) and one coarse mode.

Figure 4Time series of N_Aitken, N_accumulation, N_coarse, N_submicron, sum of N_i and N_total at CVAO. The different shading colors indicate different aerosol type periods.

Download

PNC showed large variability during our measurement. N_Aitken and N_accumulation varied from 41 to 789 and 89 to 639 cm⁻³, with medians of 244 and 354 cm⁻³, respectively. Generally, Aitken mode particles are produced by homogeneous and heterogeneous nucleation processes, formed during natural gas-to-particle condensation. Aitken mode particles are transferred to the accumulation mode through cloud processing (Hoppel et al., 1994). And accumulation mode particles are furthermore formed by coagulation of smaller particles or condensation of vapors onto existing particles, during which they grow into that size range (Seinfeld and Pandis, 2016). Therefore, when N_accumulation is higher than N_Aitken, this indicates long-range transport and a more aged aerosol. N_coarse varied from 3 to 71 cm⁻³, with a median of 21 cm⁻³. Coarse mode particles are mostly emitted to the atmosphere from natural sources, e.g., marine aerosol, mineral dust or biological materials.

3.2.2 Particle classification and origins

To better understand the particle sources and features, we divided the data from the campaign into different periods. An overview of the classification criteria and features of the different resulting aerosol types are summarized in Table 2. Details about the classification criteria are discussed in the Supplement. Classification results are shown as different background colors in Fig. 4. Note that from 00:00:00 27 September to 00:00:00 28 September, N_total suddenly decreased. This might due to the wet deposition that happened before the air masses arrived at the measurement site. The precipitation is an output parameter of the calculated NOAA HYSPLIT backward trajectories. From 00:00:00 27 September to 00:00:00 28 September, the total precipitation (sum of precipitation of 144 segment endpoints) exceeded a value of 7 mm in the past 144 h (corresponding to 6 d) of each backward trajectory history. Therefore, this period was not included in the aerosol classification.

Table 2Classification criteria and features of four different particle types.

^* 1 standard deviation

Download Print Version | Download XLSX

Figure 5 shows the median of PNSDs of the four different aerosol types, with a linear (top) and a logarithmic (bottom) scaling on the y axis. The error bar indicates the range between 25th and 75th percentiles. PNSDs which have N_Aitken larger than N_accumulation and N_coarse<25 cm⁻³ are attributed to the “marine type” in this work. PNSDs resembling those show three modes, i.e., Aitken, accumulation and coarse modes, which can be clearly distinguished, as shown in blue lines in Fig. 5. For the separation of this marine type, trajectories were additionally examined. The marine type featured the lowest N_Aitken, N_accumulation and N_coarse median values of 189, 143 and 7 cm⁻³, respectively. The minimum between the Aitken and accumulation mode of PNSDs (Hoppel minimum; see Hoppel et al., 1986) at roughly 70 nm indicates the sizes above which particles had previously been activated to cloud droplets during the history of the air mass at least once. When passing through a cloud, soluble material is added to the activated particles by aqueous-phase chemistry, increasing particulate mass and hence the size of those particles. The coarse mode particles can also be assumed to be sea spray aerosol (SSA) during the marine type period, as discussed in previous studies (Modini et al., 2015; Wex et al., 2016). A decent correlation (R²=0.69, p<0.01) was found between SSA number concentration and wind speed (Supplement, Fig. S6). Modini et al. (2015) also observed that SSA number concentration correlated with local wind speed, which is consistent with the fact that SSAs are generated from the process associated with the agitation of the sea surface by air moving above it. The SSA accounted for 1.1 % to 4.4 % of N_total at CVAO (wind speed from 4 to 10 m s⁻¹), which is relatively low when comparing to Wex et al. (2016), who found the SSA particles contributed 4 % to 10 % of N_total (wind speed up to 14 m s⁻¹) for the marine aerosol on Barbados. Figure 6 shows the 6 d backward trajectories with 1 h time resolution ending at 200 m above CVAO. Looking at Fig. 6a, which displays the marine periods, the backward trajectories clearly featured paths over the Atlantic Ocean and traveled to Cabo Verde. None of the backward trajectories touched the European or African continents.

Figure 5The median of PNSDs of marine type (blue), mixture type (green), dust type1 (purple) and dust type2 (black), with a linear (a) and a logarithmic (b) scaling on the y axis. The error bar indicates the range between 25th and 75th percentiles.

Download

Figure 6The 6 d backward trajectories arriving at CVAO at a height of 200 m with 1 h resolution for marine type (a), mixture type (b), dust type1 (c) and dust type2 (d). Each calculation is shown as a separate dot, which is separately visible when air masses moved fast.

PNSDs that have a larger N_accumulation than N_Aitken are attributed to the “mixture type” in this work, shown as green lines in Fig. 5, with three modes, i.e., Aitken, accumulation and coarse modes, which can be clearly distinguished. N_Aitken, N_accumulation and N_coarse have median values of 247, 405 and 20 cm⁻³, respectively. The Hoppel minimum of the mixture type is at roughly 80 nm. The respective backward trajectories, colored in green in Fig. 6b, showed that the related air mass came from the north Atlantic Ocean and spent some days over southern Europe and northern Africa. Anthropogenic aerosol and mineral dust may be incorporated into air parcels and transported to Cabo Verde, causing higher levels of Aitken, accumulation and coarse mode particles than in the marine type.

PNSDs with larger N_Aitken than N_accumulation and N_coarse>25 cm⁻³ are attributed to the “dust type1” in this work, shown as red lines in Fig. 5. PNSDs attributed to this type show three modes, i.e., Aitken, accumulation and coarse modes, which can be clearly distinguished. N_Aitken, N_accumulation and N_coarse had median values of 556, 312 and 39 cm⁻³, respectively. The Hoppel minimum of the mixture type is at roughly 100 nm. The respective backward trajectories, colored in red in Fig. 6c, featured two pathways. One air mass group originated from the north Atlantic Ocean and stayed a few days over southern Europe and northern Africa. Another air mass group came from the Sahara.

It is interesting to note that the Hoppel minimum is at the lowest diameter for the marine air mass (∼70 nm), compared to all other air masses. This suggests that the supersaturation in the clouds forming in the clean marine air masses is highest, as there is less surface area for the water vapor to condense onto during cloud formation.

PNSDs which featured a single mode in the submicron size range are attributed to “dust type2”, shown as black lines in Fig. 5. No visible Hoppel minimum can be seen. The dust type2 featured highest N_total and N_coarse median values of 994 and 44 cm⁻³, respectively. It is worth mentioning that previous field measurements at the Sahara found similar PNSDs to what we observed in this study (Kaaden et al., 2009; Kandler et al., 2009; Weinzierl et al., 2009). We assumed that dust type2 is the heaviest dust plume period during this campaign. The respective backward trajectories, colored in black in Fig. 6d, showed that related air masses originated from the Sahara.

The higher N_coarse during dust type1 and type2 periods is due to the direct dust aerosol from the Sahara. Schladitz et al. (2011b) also found that the higher coarse mode number concentration at Cabo Verde originated from the Sahara. Besides, a very high concentration of Aitken mode particles was observed during dust type1 and dust type2 periods. A previous study also found that an African-influenced period showed a great enhancement in the Aitken mode particles and an overall increase in the number of particles of all sizes (Allan et al., 2009). Nie et al. (2014) found that new particle formation and growth happened in the remote ambient atmosphere during the strongest observed dust episodes. Both the formation and growth rates of particles in the diameter range of 15–50 nm were enhanced during the dust episodes. In our data, we found that backward trajectories often traveled from the upper troposphere down to the marine boundary layer during dust periods, which means that Aitken mode particles could have been transported from the upper troposphere. Therefore, there are different factors contributing to the observed high N_Aitken and N_accumulation during dust plumes, such as direct transport of particles from the desert and Sahel region, and additional new particle formation and growth in the vicinity or in the upper troposphere.

To summarize Sect. 3.2, based on number concentrations in different aerosol modes, an aerosol classification was done, and four well-separable types of PNSDs were found, i.e., the marine type, mixture type, dust type1 and dust type2. Marine type particles are mainly from the Atlantic Ocean, while dust type particles are mainly from the Sahara. Mixture type particles are a combination of marine, anthropogenic and dust particles. Backward trajectories support this classification and analysis. The marine, mixture, dust type1 and dust type2 in this study are comparable to type A, D, C and B in Fomba et al. (2014), respectively, who characterized particle chemical composition at CVAO over a time period of 4 years.

3.3.1 Comparison of PNC and PNSD

PNSDs from 10 to 800 nm were measured by MPSS and a bimodal lognormal parameterization was adopted to calculate N_Aitken and N_accumulation at MV. Figure 7 shows the time series of PNC in the size range between 10 and 800 nm at CVAO ($N_{\text{10–800\hspace{0.17em}nm}}^{\mathrm{CVAO}}$) and MV ($N_{\text{10–800\hspace{0.17em}nm}}^{\mathrm{MV}}$) in Fig. 7a, PNC of accumulation mode at CVAO ($N_{\mathrm{accumulation}}^{\mathrm{CVAO}}$) and MV ($N_{\mathrm{accumulation}}^{\mathrm{MV}}$) in Fig. 7b, and PNC of Aitken mode at CVAO ($N_{\mathrm{Aitken}}^{\mathrm{CVAO}}$) and MV ($N_{\mathrm{Aitken}}^{\mathrm{MV}}$) in Fig. 7c. The variation in $N_{\text{10–800\hspace{0.17em}nm}}^{\mathrm{CVAO}}$ and $N_{\text{10–800\hspace{0.17em}nm}}^{\mathrm{MV}}$ was similar sometimes, e.g., from 23 to 25 September. However, sometimes the concentrations at MV were obviously lower than the respective values at CVAO, at least for $N_{\text{10–800\hspace{0.17em}nm}}^{\mathrm{MV}}$ and $N_{\mathrm{accumulation}}^{\mathrm{MV}}$, as, for example, from 5 to 9 October. Such a decrease was sometimes, but not always, also observed for $N_{\mathrm{Aitken}}^{\mathrm{MV}}$. This is a typical observation for cloudy air, in which particles from the accumulation mode and maybe also some from the Aitken mode are activated to cloud droplets which are then removed in the aerosol inlet on MV. When the ratio of $N_{\mathrm{accumulation}}^{\mathrm{MV}}$ to $N_{\mathrm{accumulation}}^{\mathrm{CVAO}}$ was lower than 0.85, we assumed that MV is in the cloud. When the trimodal fitting function did not work for the CVAO data set (from 03:30 to 20:00 21 September and from 09:30 28 September to 18:30 30 September), a slightly different approach was needed. For that, we used the ratio of PNC in the size range between 80 and 800 nm at MV ($N_{\text{80–800\hspace{0.17em}nm}}^{\mathrm{MV}}$) to that at CVAO ($N_{\text{80–800\hspace{0.17em}nm}}^{\mathrm{CVAO}}$) (replacing the ratio of $N_{\mathrm{accumulation}}^{\mathrm{MV}}$ to $N_{\mathrm{accumulation}}^{\mathrm{CVAO}}$). When this ratio was lower than 0.75, we assumed that MV is in the cloud. It is described in more detail in the Supplement how this ratio was derived separately for cases with trimodal and bimodal fitting. The time for cloud events is shown as red shading in Fig. 7. As outlined above in the meteorology part, we observed RH =100 % at MV. Figure S8 shows the time series of RH at MV together with the time for cloud events as red shading. It is clear that times with RH =100 % are consistent with cloud events identified as described above, which verifies our identification of cloud events.

Figure 7Time series of $N_{\text{10–800\hspace{0.17em}nm}}^{\mathrm{CVAO}}$ and $N_{\text{10–800\hspace{0.17em}nm}}^{\mathrm{MV}}$ in panel (a), $N_{\mathrm{Aitken}}^{\mathrm{CVAO}}$ and $N_{\mathrm{Aitken}}^{\mathrm{MV}}$ in the panel (b), and $N_{\mathrm{accumulation}}^{\mathrm{CVAO}}$ and $N_{\mathrm{accumulation}}^{\mathrm{MV}}$ in panel (c). The times of cloud events are shown by red shading.

Download

To better understand the cloud effect of PNSDs, we compared the PNSDs at CVAO and MV during cloud events and noncloud events of different aerosol types. Figure 8 shows the median PNSDs of different particle types during cloud events and noncloud events. During noncloud events, PNSDs at CVAO (${\mathrm{PNSD}}_{\text{non-cloud}}^{\mathrm{CVAO}}$) and MV (${\mathrm{PNSD}}_{\text{non-cloud}}^{\mathrm{MV}}$) were similar for marine, mixture or dust type1 periods. During dust type2, there is only a very short period of noncloud event with 15 PNSDs observed. Therefore, we did not include the comparison of ${\mathrm{PNSD}}_{\text{non-cloud}}^{\mathrm{CVAO}}$ and ${\mathrm{PNSD}}_{\text{non-cloud}}^{\mathrm{MV}}$ during the dust type2 period in Fig. 8.

During noncloud events, PNSDs at CVAO and MV were the same, as shown in Fig. 8. For periods with clouds, PNSDs in the size range >80 nm at MV are lower than that at CVAO for all the particle types. For dust type1 and dust type2, depending on the clouds, i.e., the highest supersaturation the particles encounter, particles in Aitken mode were also activated to cloud droplets. For particles in the size range <40 nm, PNSDs are similar during cloud and noncloud events. This is because the particle size is not large enough to activate to cloud droplets. Furthermore, it also indicates that PNSDs are similar at CVAO and MV during cloud events, at least in the size range <40 nm. For a more detailed comparison of PNSDs at CVAO and MV, contour plots for PNSDs can be found in Fig. S9 in the Supplement.

Figure 8The median of PNSDs for four different particle types during cloud events and noncloud events at CVAO and MV.

Download

During the campaign, a decoupled marine boundary layer was observed with our balloon measurements in three cases, i.e., 10:30 to 11:00 16 September, 16:00 to 16:30 5 October and 17:20 to 17:50 12 October (shown as red dots in Fig. 1). Only for the first decoupling case (10:30 to 11:00 16 September) was MV cloud free; otherwise, PNSDs were similar at CVAO and MV (Fig. S10). Therefore, the MBL may be generally well mixed, maybe still from times before the decoupling of the layers formed. On the other hand, lifting of the air masses over the mountain might also partially explain this observation. However, due to the fact that there is only this one decoupled case, a thorough analysis of the influence of coupling and decoupling can not be done.

3.3.2 Comparison of N_CCN

Figure 9 shows the scatter plot of N_CCN at CVAO ($N_{\mathrm{CCN}}^{\mathrm{CVAO}}$) against that at MV ($N_{\mathrm{CCN}}^{\mathrm{MV}}$) during cloud events (green dots) and noncloud events (red rectangles) at different supersaturations. During cloud events, large particles that had been activated to cloud droplets were removed by the aerosol inlet on MV. Therefore, $N_{\mathrm{CCN},\mathrm{cloud}}^{\mathrm{CVAO}}$ is larger than $N_{\mathrm{CCN},\mathrm{cloud}}^{\mathrm{MV}}$ at each supersaturation. During noncloud events, all the data points are close to the 1:1 line (for the slopes see Fig. 9), and R² values between $N_{\text{CCN,non-cloud}}^{\mathrm{CVAO}}$ and $N_{\text{CCN,non-cloud}}^{\mathrm{MV}}$ are all above 0.90, indicating N_CCN is similar at CVAO and MV. Although there were slight differences in supersaturation at CVAO and MV due to the CCNC calibration, the similarity between N_CCN at the two stations conveys the same message as what was discussed before concerning the comparison of PNSDs at CVAO and MV, i.e., particles are generally well mixed in the MBL.

Figure 9Scatter plots of N_CCN at CVAO against those at MV at different supersaturations. Slope and R² values for these fits are given in the legend.

Download

To summarize Sect. 3.3, cloud events were observed at MV and can be identified based on the integrated concentrations between ground and cloud level. During the cloud events, larger particles (mainly accumulation and coarse mode) are activated to cloud droplets. Aitken mode particles starting with sizes of roughly 40 nm can also be activated to cloud droplets if the cloud is strong enough. During noncloud events, PNC, PNSD and N_CCN are similar at CVAO and MV. The aerosol particles measured at ground level (CVAO) can represent the aerosol particles at the cloud level (MV).

3.4.1 Statistical analysis of N_CCN, d_crit and κ

Figure 10a shows the time series of N_total and N_CCN, d_crit in Fig. 10b and κ in Fig. 10c, with different colors for different supersaturations. The error bars of d_crit show 1 standard deviation (SD), and error bars of κ show 1 geometric standard deviation (geoSD). Explanation of error bars can be found in Sect. 2.4 as well as in the Supplement. N_CCN shows large variability, e.g., N_CCN,0.30 % varied from 106 to 884 cm⁻³, with a median of 509 cm⁻³. We observed highest N_CCN,0.30 % of 503 cm⁻³ (median) during dust type2 periods, and lowest N_CCN,0.30 % of 109 cm⁻³ (median) during marine periods. N_CCN,0.30 % during different aerosol type periods are summarized in Table 2. Figure 11a shows the boxplot of N_CCN at different supersaturations during the whole campaign. As can be seen, N_CCN increases towards higher supersaturation, which is expected. The median of N_CCN at different supersaturations also exhibited large variability, varying from 327 (median) at a supersaturation of 0.15 % to 652 cm⁻³ (median) at a supersaturation of 0.70 %. Table 3 summarizes those numbers and shows additional details.

Figure 10Time series of N_CCN in (a), d_crit in (b) and κ in (c). All of those are measured at CVAO. Error bars of d_crit and κ show 1 standard deviation and 1 geometric standard deviation. The color bar in (c) indicates the times of different aerosol type periods.

Download

d_crit at supersaturations of 0.15 %, 0.20 %, 0.50 % and 0.70 % were almost constant throughout the campaign, as shown in Fig. 10b. The mean value of d_crit and its SD are summarized in Table 3. For the supersaturations of 0.70 % and 0.50 %, d_crit is below 80 nm, i.e., inside the Aitken mode. However, for the lower supersaturations of 0.15 % and 0.20 %, d_crit is located in the accumulation mode. Consequently, hygroscopicities derived at these supersaturations can be assumed to be representative of the Aitken (at supersaturations of 0.70 % and 0.50 %) and the accumulation modes (at supersaturation of 0.10 % and 0.20 %), respectively. d_crit at a supersaturation of 0.30 % (d_{crit,0.30 %}) is not as constant as it is at other supersaturations, and it is larger during the marine type period than during other periods. With a median of 79.7 nm, it is close to the Hoppel minimum. Therefore, the hygroscopicity derived at a supersaturation of 0.30 % can be assumed to be representative of the mixture of Aitken and accumulation particles.

Figure 11Boxplots of N_CCN (a), d_crit (b) and κ (c) at different supersaturations (SS). Whiskers show the 10th to 90th percentiles. Circles show the outliers (1 %). (d) κ as a function of d_crit. Error bars of d_crit and κ show 1 standard deviation and 1 geometric standard deviation, respectively.

Download

Table 3Median and mean values of N_CCN, d_crit, κ, 1 standard deviation of d_crit and 1 geometric standard deviation of κ at different supersaturations.

^* 1 geometric standard deviation

Download Print Version | Download XLSX

The particle hygroscopicity, expressed as κ, can be seen as a measure for average particle chemical composition. κ values at different supersaturations show little variability over time (lower panel in Fig. 10), with geoSD lower than 0.12, i.e., there is no clear trend in κ over time during the campaign. A slightly increasing trend of κ was observed with decreasing supersaturations, as shown in Fig. 11c. At supersaturations of 0.70 % and 0.50 %, i.e., for Aitken mode particles, κ values are 0.18 and 0.25 (geomean), respectively. At the lowest supersaturation of 0.15 % and 0.20 %, i.e., for accumulation mode particles, κ values are 0.32 and 0.34 (geomean). Table 3 summarizes those numbers and shows additional details.

Figure 11d shows κ as a function of d_crit and error bars of κ and d_crit show geoSD and SD, respectively. A slightly increasing trend of κ over increasing d_crit is observed. It suggests that the soluble, likely inorganic, material added during cloud processing increases κ of the originally very organic-rich particles, which has also been observed in previous studies (Kalivitis et al., 2015; Kristensen et al., 2016). Overall, κ averaged 0.28. Pringle et al. (2010) used an atmospheric chemistry model to derive global distributions of effective particle hygroscopicity κ. For CVAO, this model resolved an annual cycle of monthly-mean κ values ranging from 0.25 in February to 0.60 in April. This annual circle of κ likely originated in a change of chemical composition of the aerosol throughout the year, related to different precursors and a higher organic content during times with higher algal activity. For September and October, the period of this study, values of 0.35 and 0.30 were reported, respectively, which is consistent with what we obtained during this campaign.

The Hoppel minimum diameter range of 70 to 100 nm for different aerosol types (mentioned in Sect. 3.2.2), together with the average κ of 0.28, can be used to obtain a rough estimate of maximum supersaturations present in trade wind clouds along the path of the sampled air masses. Resulting values are roughly 0.22 % to 0.37 % for dust type2 and marine air masses, respectively. These are close to an earlier estimate given in Clarke et al. (1996) of 0.35 %, and they can be interpreted as typical values for trade wind cumuli.

3.4.2 Dust and marine comparison

In this section, we will focus on examining the difference between the cleanest periods (marine type) and heaviest observed dust pollution periods (dust type2). Therefore, we compared N_CCN and κ during marine type and dust type2 periods. Figure 12a shows the boxplot of N_CCN as a function of supersaturation. As outlined above, during dust periods, the aerosol shows a great enhancement in the Aitken, accumulation and coarse mode particles; therefore, overall N_CCN increases at different supersaturations. It is clear that N_CCN during dust type2 periods is much higher than that during marine periods. For example, N_CCN,0.30 % median values were 503 and 190 cm⁻³ during dust type2 and marine periods, respectively. During marine periods, N_coarse, i.e., SSA particles, accounted for roughly 3.7 % of N_CCN,0.30 %, for the range of wind speeds from 4 to 10 m s⁻¹ that were present during this study. This is relatively low compared to Wex et al. (2016), who found that the SSA particles accounted for up to 15 % of N_CCN,0.30 % for wind speeds up to 14 m s⁻¹ for the marine aerosol on Barbados, and Modini et al. (2015), who found that SSA particles accounted for up to 16 % to 28 % (wind speeds up to 16 m s⁻¹) and 5 % to 10 % (wind speed from 4 to 10 m s⁻¹) of N_CCN,0.30 %. However, these fractions not only depend on the concentrations of SSA but also on those of particles in the accumulation mode, which have other sources. Still, the respective accumulation modes and related particle concentrations in Modini et al. (2015), Wex et al. (2016) and the present study resemble each other. Therefore the lower fractions of SSA particles in our study are likely connected to the low wind speeds (lower SSA number concentration) or, to some extent, to different accumulation mode particle number concentrations.

Figure 12(a) Boxplots of N_CCN as a function of κ for marine type (blue) and dust type2 (black). Whiskers show the 10th to 90th percentiles. The predicted N_CCN based on median PNSD and different κ values (0.1, 0.2, 0.3, 0.4 and 0.5) are shown in solid (during dust type2 period) and dashed lines (during marine period). (b) κ as a function of d_crit for marine type (blue) and dust type2 (black). Error bars of d_crit and κ show 1 standard deviation and 1 geometric standard deviation, respectively.

Download

κ as a function of d_crit is shown in Fig. 12b. The error bars of d_crit and κ show SD and geoSD, respectively. During dust type2, slightly increasing κ with increasing d_crit was observed, similar to the overall trend we described above. κ featured lower values from 0.13 to 0.31 for Aitken mode particles, while higher values from 0.26 to 0.45 were found for accumulation mode particles. Until now, the only field measurement of particle hygroscopicity during a dust plume at Cabo Verde was carried out by Schladitz et al. (2011a), who found that hygroscopic particles featured a κ value (based on hygroscopic growth factor of particles) from 0.35 to 0.65. Our CCN-derived κ values in this study for the aerosol influenced by dust are therefore comparable to the values reported by Schladitz et al. (2011a).

No clear trend of κ with d_crit was observed during marine type periods (as shown in Fig. 12), during which κ averaged 0.30. Larger error bars of κ and d_crit at the supersaturation of 0.30 % were observed, as in this case the d_crit is close to the Hoppel minimum, where a small change in N_CCN causes a comparably large change in d_crit (explained in the Supplement). Kristensen et al. (2016) found that, for the marine aerosol on Barbados in June and July 2013, values for κ of 0.2 to 0.5 were derived, which are consistent with this study. When considering the scatter observed in κ (see the error bars in Fig. 12), κ during the dust type2 period still agreed with that of the marine period within uncertainty. Therefore, no distinguishable differences in κ during marine and dust periods in the size range from 40 to 140 nm were found during this campaign.

We additionally derived N_CCN based on PNSDs. For that, we assumed values for κ of 0.1, 0.2, 0.3, 0.4 or 0.5, and we calculated the corresponding d_crit at different supersaturations. The integrated particle number concentrations in the size range larger than d_crit were derived from the median PNSDs during dust type2 and marine periods. These particle number concentrations also can be treated as the predicted N_CCN at different supersaturations, as shown in solid (dust type2) and dashed (marine type) lines with different colors (indicating different κ) in Fig. 12a. As expected, the thusly derived N_CCN values were within the measured N_CCN range. Comparing the solid and dashed lines, it can be seen that different aerosol types, i.e., different PNSDs, played an important role in N_CCN variation. When looking at the results from the different κ values, we found that the particle chemical composition did not control N_CCN, especially when the particle number concentration was very low. These colored solid and dashed lines connected the κ and N_CCN, which can be helpful for N_CCN predictions in modeling studies.

To summarize Sect. 3.4, overall there is a slight increase in κ with particle size, indicating the addition of soluble, likely inorganic, material during cloud processing. κ values in this study are comparable to previous model work and field measurements. N_CCN during the heaviest observed dust periods is much higher than that during marine periods, while κ values for these two periods show no significant difference.