A CCNC is a type of instrument frequently used for studying the cloud droplet activation potential of aerosol particles. In its simplest set-up, a CCNC consists of a saturator unit and an optical particle counter (OPC) frequently running in parallel with a condensation particle counter (CPC). For all measurements presented herein, the CCNC used was a commercially available instrument produced by Droplet Measurement Technologies, Inc. (DMT-CCNC), the basic principles of operation of which are described below.

Upon entering the measurement set-up, the aerosol flow is split into two sample flows, with the first flow leading to a CPC to determine the total particle number concentration, hereafter referred to as NCN. The second flow feeds the aerosol into the saturator unit of the CCNC, inside of which the conditions of supersaturation Seff with respect to water vapour down the centre of the column are established. Aerosol, flowing under laminar flow conditions, is subjected to these supersaturation conditions, during which particles with a critical supersaturation Sc smaller than Seff will grow by the condensation of water vapour and remain in stable equilibrium, i.e. activate as CCN. The residence time inside the saturator column (∼10 s) allows for the activated particles to grow to sizes larger than 1 µm in diameter; these particles are then counted by the OPC providing the number concentration of activated aerosol particles, a quantity hereafter referred to as NCCN. The described set-up is characteristic of polydisperse measurements; an inclusion of a drier, a neutraliser and a differential mobility analyzer (DMA; Knutson and Whitby, 1975) prior to the splitting of the flow into two parallel lines allows for the selection of a particular particle size, i.e. quasi-monodisperse measurements. Such measurements can be performed either by varying the particle size at a constant Seff (D-scan) or by varying Seff at a constant particle size (S-scan). Such a set-up, while more complex, provides activation spectra and allows for a direct calculation of the critical dry diameter of droplet activation Dc (in case of the D-scan) or the critical supersaturation Sc (in case of the S-scan). Typically, a CCNC operates at several different levels of Seff, most commonly ranging between 0.1 and 1.0 %; the deviations from the nominal assigned Seff values can be monitored and corrected by applying a standardised calibration procedure, as described in Sect. 2.3. A more detailed description of the general operating procedures of the CCNC can be found in Roberts and Nenes (2005); exact details of the measurement set-up at each of the locations described in the next section can be found in the respective published literature referenced throughout the text.

Data from a total of 14 EUCAARI locations have been provided for this analysis, including both long-term measurement stations and short-term campaigns (Fig. 1). As seen in the figure, data sets came from a wide variety of locations representing various environments, including marine and continental, urban and background, at altitudes ranging from the ground level to the free troposphere. The location and description of each measurement site are given in Table 1. All measurements presented herein were performed within the EUCAARI framework.

Hyytiälä Forestry Field Station in southern Finland is the location of the Station for Measuring Ecosystem–Atmosphere Relations (SMEAR II), operated by the University of Helsinki. Located on a flat terrain and surrounded by the boreal coniferous forest, mainly Scots pine, the station is well representative of the boreal environment (Hari and Kulmala, 2005). It is a rural background site, with the nearest city of Tampere (population 220 000) located 50 km to the south-west. Air masses at the site can be of both Arctic and European origin; however, aerosol particle number concentrations at this site are typically low (Sogacheva et al., 2005).

Vavihill in southern Sweden is a continental background site surrounded by grasslands and deciduous forest and operated by Lund University. The site is located 60–70 km NNE of the Malmö and Copenhagen urban area (population ∼2 000 000); however, it is considered to not be affected by the local anthropogenic sources (Tunved et al., 2003). Due to its location, the site is often used for monitoring the transport of pollution from continental Europe into the Nordic region (Tunved et al., 2003).

The Jungfraujoch is a high Alpine station in the Bernese Alps in Switzerland, where the aerosol measurements are performed by the Paul Scherrer Institute (PSI). Being located high in the mountains (3580 m a.s.l.), the station is far from local sources of pollution and is, in fact, in the free troposphere most of the time; hence, it is considered a continental background site and aerosol concentrations are very low (Collaud Coen et al., 2011). However, particularly during the summer months, the Jungfraujoch site is frequently influenced by the injections of more polluted air from the planetary boundary layer, driven by thermal convection (Jurányi et al., 2010, 2011; Kammermann et al., 2010a). The station is frequently inside clouds allowing for direct measurements of aerosol–cloud interactions.

Mace Head is a coastal marine site located on the west coast of Ireland and operated by the National University of Ireland, Galway. The distance to the nearest urban settlement of Galway City (88 km, population 65 000) renders Mace Head a clean background site; being on the coast, the station is directly exposed to the North Atlantic Ocean. Occasionally the station is subject to more polluted air masses originating from continental Europe and the United Kingdom (O'Dowd et al., 2014).

Pallas is a remote continental site in northern Finland located in the northernmost boreal forest zone in Europe; it is run by the Finnish Meteorological Institute (FMI). The station is situated on top of a treeless hill and, due to the frequent presence of clouds, is suitable for in situ measurements of aerosol–cloud interactions. The Pallas station is subject to both clean Arctic air masses, as well as to more polluted European air masses; regardless, absolute particle number concentrations are typically low (Hatakka et al., 2003).

Finokalia station is a remote coastal site located on the island of Crete and operated by the University of Crete. The station is located on top of a hill, and most frequently air masses arrive in Finokalia over the Mediterranean Sea (Stock et al., 2011). The station is representative of background conditions as there are no local sources of pollution present; the largest nearby urban centre of Heraklion (population 175 000) is 50 km to the west.

The Cabauw Experimental Site for Atmospheric Research (CESAR) is located in the central Netherlands, 44 km from the North Sea. The station is in a rural area; however, the big cities of Utrecht and Rotterdam are nearby; the station is subject to both continental and maritime conditions (Mensah et al., 2012). The station is operated by the Royal Netherlands Meteorological Institute (KNMI).

The University of Manchester conducted four short-term measurement campaigns utilising a CCNC: K-puszta, Chilbolton, COPS and RHaMBLe. K-puszta is a rural site surrounded by deciduous–coniferous forest located on the Great Hungarian Plain in central Hungary 80 km SE of Budapest. The site has no local anthropogenic pollution sources (Ion et al., 2005). Chilbolton is also a rural site, located in southern United Kingdom, 100 km WSW of London. The site is most frequently influenced by the marine air masses; a potential local source of anthropogenic pollution is the seasonal agricultural spraying (Campanelli et al., 2012). The Convective and Orographically-induced Precipitation Study (COPS) campaign took place at the top of the Hornisgrinde Mountain in the Black Forest region of south-west Germany. While this site is primarily surrounded by the coniferous forest, the close proximity to the Rhine Valley exposes the site to some anthropogenic pollution. Due to its elevation, the site is occasionally in the free troposphere (Jones et al., 2011). The Reactive Halogens in the Marine Boundary Layer (RHaMBLe) Discovery Cruise D319 campaign was a cruise conducted in the tropical North Atlantic between Portugal and Cabo Verde. The operational area can be described as a remote marine environment with few, if any, sources of anthropogenic pollution. Air masses can originate from both the ocean and from the African mainland (Good et al., 2010).

The Max Planck Institute for Chemistry (MPIC) also conducted four CCNC measurement campaigns within the EUCAARI framework: PRIDE-PRD2006, AMAZE-08, CAREBeijing-2006 and CLACE-6, with the last one having taken place at the previously described Jungfraujoch station. The PRIDE-PRD2006 campaign took place in southeastern China, in a small village ∼60 km NW of Guangzhou, in the vicinity of a densely populated urban centre. The wind direction during the campaign rendered the site a rural receptor of the regional pollution originating from the Guangzhou urban cluster (Rose et al., 2010). The AMAZE-08 campaign took place at a remote site in an Amazonian rainforest, 60 km NNW of Manaus, Brazil. During the campaign, the site experienced air masses characteristic of clean tropical rainforest conditions as well as air masses influenced by long-range transport of pollution (Gunthe et al., 2009; Martin et al., 2010). The CAREBeijing-2006 campaign was conducted at a suburban site in northern China, on the grounds of Huang Pu University in Yufa, ∼50 km south of Beijing. The site is subject to air masses originating both in the south and in the north; however, being located on the outskirts of a large urban centre, particle concentrations are generally high (Garland et al., 2009).

The measurement period for each location and a brief summary of available CCNC data are presented in Fig. 2 and Table 2, respectively. Available data range from mid-2006 to the end of 2012; the four long-term data sets all exceed one year in duration. As originally requested by the authors from the EUCAARI partners, some of the data were submitted in the NASA-Ames format with daily and monthly/campaign averages. Other data sets were submitted in the original time resolution and have been compiled accordingly for this overview study.

For the quality assurance of the CCNC data, data providers were requested to recalculate all values to correspond to the standard temperature and pressure and to utilise a consistent procedure for the CCNC calibration. Calibrations were asked to be performed as outlined in Rose et al. (2008) using nebulised, dried, charge-equilibrated and size-selected ammonium sulphate or sodium chloride aerosol particles. To predict Seff for instrument calibration, water activity was asked to be parameterised according to either the AIM-based model (Rose et al., 2008) or the ADDEM model (Topping, 2005); both of these models can be considered as accurate sources of water activity data, and the discussion about their associated uncertainties can be found in the corresponding references. As none of the participating data providers noted a deviation from the calibration procedure, it is assumed that the data were treated accordingly. However, deviations from the described procedure and from the target Seff levels may be possible and can potentially affect some of the conclusions presented in this paper. Uncertainties associated with deviations from the mentioned calibration procedure and parameterisation are discussed in great detail in Rose et al. (2008) and Topping (2005).

For some of the polydisperse data sets, where available, differential/scanning mobility particle sizer (DMPS/SMPS; Wang and Flagan, 1989; Wiedensohler et al., 2012) data were used in conjunction with the CCNC to derive the critical dry diameter Dc. The procedure was carried out by comparing NCCN to the DMPS/SMPS-derived number size distributions; these were integrated from the largest size bin until the cumulative NCN concentration was equal to NCCN. Dc was then calculated by interpolating between the two adjacent size bins (Furutani et al., 2008). Following the calculation of Dc, the hygroscopicity parameter κ was determined using the effective hygroscopicity parameter (EH1) Köhler model (Eq. 1) assuming the surface tension of pure water (Petters and Kreidenweis, 2007; Rose et al., 2008). Due to the surface tension of actual cloud droplets being lower than that of pure water droplets (Facchini et al., 2000), this assumption, although commonly used, typically leads to an overestimation of the NCCN (Kammermann et al., 2010b). S=Dwet3-Ds3Dwet3-Ds3(1-κ)exp4σsolMwRTρwDwet, where S is water vapour saturation ratio, Dwet is the droplet diameter, Ds is the dry particle diameter, which, as per Rose et al. (2008), can be substituted with Dc, κ is the hygroscopicity parameter, σsol is the surface tension of condensing solution (assumed to be pure water), Mw is the molar mass of water, R is the universal gas constant, T is the absolute temperature and ρw is the density of pure water.

For certain sites, total number concentrations of particles larger than 50 or 100 nm in diameter (N50 or N100) were calculated from the corresponding DMPS or SMPS data.

In order to compare the results from different stations, several interpolation/extrapolation techniques were used. All NCCN concentrations were averaged for each site for each Seff level and then recalculated to correspond to the target Seff levels suggested by the Aerosols, Clouds and Trace gases Research InfraStructure (ACTRIS) Network: 0.1, 0.2, 0.3, 0.5 and 1.0 %. Recalculation to the nearest target supersaturation was accomplished by a simple linear interpolation/extrapolation of NCCN as a function of Seff using the two adjacent/nearest Seff points. For the Jungfraujoch data, Dc at Seff of 0.12 and 0.95 % was recalculated to the corresponding Dc at the target Seff of 0.1 and 1.0 %, respectively, assuming a size-independent κ.

Table 3 presents CCN number concentrations NCCN at all 18 measurements locations and campaigns for five Seff levels mentioned in the previous section. First and foremost, since CCN are simply a fraction of the total aerosol population with their concentration depending on Seff, NCCN values at Seff of 1.0 % follow a similar pattern known from total particle number concentrations. The lowest NCCN values, thus, originate in remote and clean locations, such as Pallas, the Amazonian rainforest (AMAZE-08), Jungfraujoch and Chilbolton. The highest NCCN values are found in more polluted locations – CAREBeijing-2006 and PRIDE-PRD2006, both in China. At lower Seff levels, other effects, such as those of size distribution and hygroscopicity, become more pronounced. When examining NCCN at Seff of 0.1 %, the highest values are still found in China; similar to NCCN at Seff of 1.0 %, the lowest values are found in Pallas, the Amazonian rainforest (AMAZE-08), Jungfraujoch and also in south-west Germany (COPS).

In order to examine the CCN activation spectra in more detail, Fig. 3 presents cumulative NCCN concentrations shown as a percentage of the NCCN measured at the highest Seff of 1.0 %. One group of locations that can be pointed out in the figure is representative of the marine environment: Finokalia, Mace Head and the RHaMBLe campaign. At these marine locations the presence of large and hygroscopic sea salt particles is expected, and a large fraction of particles already activates at the lowest Seff, i.e. of the total NCCN measured at the highest Seff, about a third activates already at the lowest Seff. In the case of Mace Head, the observed behaviour is due to the presence of sea salt particles and a peculiar organic composition of the marine aerosol (Ovadnevaite et al., 2011). Additionally, both Finokalia and Mace Head have a large fraction of the long-range transported and aged aerosol (Bougiatioti et al., 2009; Ovadnevaite et al., 2011), which has been shown to increase particle hygroscopicity (Perry et al., 2004; Furutani et al., 2008). Chilbolton, being a continental background site representative of the regional aerosol properties, also belongs to this group; however, the NCCN concentrations at this location may be underestimated due to the aerosol not being dried prior to entering the CCNC (Whitehead et al., 2014).

Another group of locations with a different CCN activation pattern is represented by Pallas and Cabauw – at these locations very few particles activate at the lowest Seff, and the NCCN increases drastically when Seff changes from 0.5 to 1.0 %. This may indicate that the aerosol is dominated by the Aitken mode particles and, to a lesser extent, that the aerosol may be of low hygroscopicity. A high concentration of Aitken mode particles in the autumn and low aerosol hygroscopicity in Pallas have been previously reported by Tunved et al. (2003) and Komppula et al. (2006), respectively. The two measurement locations discussed here are interesting with regard to the ratio of presumed cloud droplet number concentration (CDNC) to the total aerosol particle number concentration. It has been reported that, although under the clean and convective conditions ambient Sc may reach as high as 1.0 %, in the polluted boundary layer Sc usually remains below 0.3 % (Ditas et al., 2012; Hammer et al., 2014; Hudson and Noble, 2014). If one assumes this value, a comparatively small fraction of aerosol in northern Finland and central Netherlands would potentially activate into cloud droplets if exposed to this Sc. This has direct implications for the cloud formation and, thus, local climate at these locations.

Another variable describing CCN activation properties of an aerosol population that was examined for the majority of locations is the activated fraction A calculated as a ratio of NCCN to NCN (Fig. 4). Each activation curve in Fig. 4 is based on the arithmetic mean values of A calculated from all available data for each station for each Seff level. Included in the figure is the overall fit shown with prediction bounds (95 % confidence level) based on most of the activation curves, except the outlying ones of Finokalia, COPS, Jungfraujoch and Pallas A, B and C. As can be seen in the figure from the similar shape and placement of the activation curves and in Table 4 from the similar slope and intercept values, for many locations there is no discernible difference in how A responds to changing Seff on an annual basis; this is further signified by the prediction bounds of the overall fit. Therefore, the average total number concentration NCN alone is sufficient in order to roughly estimate the annual mean NCCN at any given Seff, for example, using the overall fit parameters presented in Table 4. The appropriateness of the overall fit for estimating NCCN based on NCN alone was investigated for the whole Hyytiälä data set, by comparing the NCCN measured by the CCNC with the NCCN calculated using the NCN and the overall fit presented in Table 4. Such a comparison revealed that for Hyytiälä the overall fit leads to an annual median overestimation of NCCN of 49, 41, 33, 17 and 2 % for the Seff levels of 0.1, 0.2, 0.3, 0.5 and 1.0 %, respectively.

For Seff levels below 0.3 %, the variability of the overall fit, as shown by the prediction bounds, leads to the uncertainty of the predicted NCCN of up to an average of ∼45 %. This uncertainty decreases exponentially for Seff levels above 0.3 %. A global modelling study conducted by Moore et al. (2013) reported that CDNC over the continental regions is fairly insensitive to NCCN, where a 4–71 % uncertainty in NCCN leads to a 1–23 % uncertainty in CDNC. Since the overwhelming majority of measurements analysed in this paper were conducted on land, and the overall fit results in an uncertainty of the predicted annual mean NCCN of up to ∼45 %, for many sites the use of the overall fit would yield a deviation of the predicted average CDNC of approximately less than 10 %. CDNC, however, is more sensitive to NCCN in cleaner regions with low total particle number concentrations, such as the Alaskan Arctic and remote oceans (Moore et al., 2013). In such areas the use of the overall fit may not be appropriate.

Four locations stand out in Fig. 4, which were not included in the overall fit. A is visibly higher in Finokalia and during the COPS campaign than in other locations, with approximately 60 % of the total aerosol population at both locations activating into cloud drops at the Seff of ∼0.4 %. Reasons for the observed behaviour in Finokalia were discussed in the preceding Sect. 3.1. During the COPS campaign the size distributions varied greatly, and, as will be shown later, Aitken mode aerosol was more hygroscopic than accumulation mode aerosol, possibly explaining the behaviour of the COPS activation curve seen in Fig. 4 at least for higher Seff levels (Irwin et al., 2010; Jones et al., 2011). Another location with seemingly different activation curves is Pallas, where the activation spectrum changes throughout the year, and even at fairly high Seff level of 1.0 %, less than half of the total aerosol population activated into cloud drops. The long-term Jungfraujoch data set also exhibited comparatively low A values, lower than those presented by Jurányi et al. (2011) and those during the CLACE-6 campaign at the same location (Fig. 4). While the A values in the long-term Jungfraujoch data set were calculated with respect to CPC measurements of total particle number concentration, A values for the CLACE-6 campaign and those reported by Jurányi et al. (2011) were calculated with respect to integrated SMPS size distribution measurements with a higher size cut-off. While the aerosol hygroscopicity at these locations will be discussed later, the effect of the size distribution on the activation curves is evident.

The similarity in how A responds to Seff at the majority of studied locations is an interesting result. In other words, at any given Seff the annual mean fraction of aerosol that will activate into cloud drops is pretty much the same in many locations, a fact that was pointed out previously by Andreae (2009). This phenomenon can easily be illustrated using the example of the activation curve during the RHaMBLe cruise in the tropical North Atlantic. As will be discussed later, while the NCCN here is comparable to several other locations, the hygroscopicity of the aerosol is much higher, with the hygroscopicity parameter κ being just below unity across all studied sizes. Yet, the fact that the aerosol is so hygroscopic seems to affect the activation efficiency of the aerosol in a similar manner as, for example, during the PRIDE-PRD2006 campaign in southeastern China. During this campaign absolute NCCN was an order of magnitude higher than during the RHaMBLe cruise (Table 2), and the hygroscopicity was much lower (Rose et al., 2010). This order of magnitude difference in NCCN, a large difference in κ and at least some presumed difference in the shape of size distribution between the RHaMBLe cruise and the PRIDE-PRD2006 campaign seem to result in no apparent difference in the fraction of the aerosol that activates into cloud drops at any given Seff. For most of the continental locations the overall fit presented in Table 4 can provide a reasonable estimation of annual mean NCCN based on the NCN for any given Seff. It should be kept in mind, however, that the activation curves in Fig. 4 for the long-term data sets do not reflect the potential short-term or seasonal variability, which, as can be seen in the example of the three Pallas campaigns, can be rather high. This and the fact that the short-term campaigns have been conducted during different seasons mean that the overall fit represents the annual mean activation behaviour and does not capture the variability on the shorter timescales.

One important uncertainty associated with the comparison of the activation curves in Fig. 4 is the precise size range from which NCN is determined. In order for the activation curves to be directly comparable, the lower size limit of NCN must be the same for all locations. In this study, data of the lower limit of NCN for each location (NCN,Dmin) were unavailable and, hence, this parameter was likely to vary, complicating the comparison of activation curves in Fig. 4. To circumvent the problem, to conduct a more accurate comparison and to reveal more information about the effect of size distribution on CCN variability, N100 and N50 concentrations were used instead of NCN to calculate the effective activated fractions corresponding to a certain lower cut-off diameter A100 and A50, respectively. These were calculated for the four long-term measurement locations only (where the data were available), and the results of the comparison are depicted in Fig. 5. When N100 is used instead of NCN, the differences among locations described above almost disappear except for the lowest values of S. In general, the activation curve of A100 for Mace Head is similar to those for Hyytiälä, Vavihill and Jungfraujoch for Seff above 0.4 %. In other words, when one considers the fraction of only accumulation mode particles that activates into cloud drops at any given Seff, the difference in how Seff affects A at all examined locations diminishes. In Hyytiälä, Vavihill and Jungfraujoch, particles with a dry diameter of 100 nm activate at the Seff of slightly higher than 0.2 % assuming an internally mixed aerosol. Around this Seff Mace Head does exhibit a slightly higher A100 compared to other locations, possibly due to the increased CCN activity of the organically enriched Aitken mode aerosol (Ovadnevaite et al., 2011).

When A50 is examined in detail, the difference between Mace Head and other locations seen in Fig. 4 remains, with Mace Head exhibiting a higher activated fraction compared to the three other locations. In Hyytiälä, Vavihill and Jungfraujoch, particles with a dry diameter of 50 nm activate at a Seff of ∼0.7 %, while in Mace Head these same particles activate at a Seff of ∼0.55 %. Differences observed in Figs. 4 and 5 lead to the conclusion that A50 and A100 have a more stable dependence on S; i.e. the variability in the fraction of nucleation/Aitken mode particles among different locations is large. Consequently, when comparing data sets of activated fractions A from several locations with different expected concentrations of nucleation/Aitken mode particles and instrumental set-ups, a recommendation is made for the consideration of using N100 and/or N50 concentrations instead of NCN when calculating A coupled with A values derived from total number concentrations. Besides more systematic comparison of activation curves and, therefore, more accurate results, such an approach can provide additional information about the effect of size distribution and its variability, and hygroscopicity on CCN activation. The use of A100 and A50 also diminishes the effect of the spatial variability of the fraction of nucleation/Aitken mode particles, those less relevant for CCN activation at typical ambient Seff levels.

Critical dry diameter Dc and hygroscopicity parameter κ were provided for the majority of the presented locations, and the variation of κ with dry size is seen in Fig. 6 (the figure is split into four panels for better visual representation). The variation of κ with dry size is not the same everywhere, and three groups can be pointed out.

In the first group of locations κ clearly increases with size; this is the case for Hyytiälä, Vavihill, Jungfraujoch (Fig. 6, upper left panel), Pallas (Fig. 6, upper right panel), and for the four campaigns conducted by the MPIC (Fig. 6, lower right panel). At these locations accumulation mode particles have a higher hygroscopicity than the Aitken mode particles, likely due to cloud processing. The results of the Mann–Whitney U test (Mann and Whitney, 1947) for two populations that are not normally distributed (below and above 100 nm of dry size; Paramonov et al., 2013) reveal that in Hyytiälä, Vavihill, Jungfraujoch and Pallas A and C the difference in κ is statistically significant at the 5 % significance level; i.e. the median κ of Aitken and accumulation mode particles are significantly different (Table 5). Published data for the PRIDE-PRD2006, CAREBeijing-2006, CLACE-6 and AMAZE-08 campaigns have previously reported such a trend (Rose et al., 2010, Gunthe et al., 2011, Rose et al., 2013 and Gunthe et al., 2009, respectively). Data for Chilbolton (Fig. 6, lower left panel) also reveal an increase in κ with size, although absolute κ values at this site may be underestimated due to the aerosol sample not being dried before entering the CCNC (Whitehead et al., 2014). Such behaviour of κ leads to two implications. First, as already discussed in Su et al. (2010) and Paramonov et al. (2013), the hygroscopicity of the whole aerosol population can, and in some cases should, be presented as a function of size; this can be done by way of either separate κ values for the Aitken and accumulation mode aerosol or hygroscopicity distribution functions. Values of κ derived from the CCNC are frequently discussed in conjunction with the chemistry information obtained, e.g. from the aerosol mass spectrometer (AMS) measurements. The second implication here is that if, due to instrumental limitations, such measurements are representative only of the accumulation mode particles, κ values derived from such measurements should be extended to the Aitken mode particles with caution. The effect of extending the accumulation mode κ down to the Aitken mode was examined using detailed data from Hyytiälä as an example. NCCN was calculated using the median annual size distribution and Dc calculated with size-dependent and the assumed size-independent κ values. It was found that if κ of the accumulation mode is assumed to be the same for the Aitken mode, the NCCN, on average, is overestimated by 16 and 13.5 % for the Seff of 0.6 and 1.0 %, respectively.

The second group of locations, or in this case only one location, exhibits a decrease of κ with particle dry size, and such a trend exists only for the COPS campaign (Fig. 6, lower left panel). Apparently, at the mountainous site in the Black Forest region of south-west Germany the chemical composition of the accumulation mode aerosol makes it less hygroscopic compared with the Aitken mode at supersaturated conditions (Irwin et al., 2010). However, the same study reported that the measurements by the hygroscopicity tandem DMA (HTDMA) in a sub-saturated regime revealed an increase of κ with particle dry size.

The third group of locations, represented by the K-puszta station and RHaMBLe measurement campaign, is characterised by the absence of any dependence of κ on the particle dry size. Though quite different in magnitude (Fig. 6, lower left panel), κ values and, therefore, aerosol chemical composition seem to have no particular size dependence across the whole measured size range. Also of interest is the high aerosol hygroscopicity across the whole investigated aerosol size range (Aitken mode) during the RHaMBLe cruise – all κ values are just below unity (Good et al., 2010). The marine nature of the aerosol and clean background conditions of the remote tropical North Atlantic are likely responsible for high aerosol hygroscopicity.

Three of the four long-term data sets, excluding Mace Head, included Dc and κ data, making it possible to examine aerosol hygroscopicity both on the annual basis and diurnal basis separated by seasons. Figure 7 presents the annual variation of Dc for lowest and highest Seff levels in Hyytiälä, Vavihill and Jungfraujoch. As can be seen in the y axis of the upper panel, particles measured at the Seff of 0.1 % are in the accumulation mode, i.e. Dc is larger than 100 nm in diameter. Of the three stations presented, Dc has an annual pattern only in Hyytiälä, with a minimum Dc and an increased hygroscopicity in the winter and a maximum Dc and a decreased hygroscopicity in the summer, as previously reported by Paramonov et al. (2013). The likely reason for a decrease in the accumulation mode particle hygroscopicity in Hyytiälä in the summer is the increase in the emissions of the volatile organic compounds (VOCs), leading to an increase in secondary organic aerosol (SOA) formation and, thus, a higher organic fraction. The higher hygroscopicity in the winter can also be explained by a higher sulphate fraction, stronger aerosol oxidation and potentially other ageing processes that are known to increase particle hygroscopicity (Furutani et al., 2008). No annual pattern is present in the aerosol hygroscopicity of accumulation mode aerosol in Vavihill and Jungfraujoch. The lower panel in Fig. 7 depicts the annual variation of aerosol hygroscopicity for the Aitken mode aerosol, revealing no pattern for any of the three locations. The absence of a pattern coupled with the absence of an apparent difference among sites indicates that the aerosol hygroscopicity of Aitken, ∼50 nm aerosol is fairly similar and constant throughout the year at all three locations.

The diurnal patterns of aerosol hygroscopicity were analysed for Hyytiälä, Vavihill and Jungfraujoch on a seasonal basis. It was discovered that for the accumulation mode particles, those measured at the Seff of 0.1 %, no diurnal pattern was observed at any of the three locations in any of the seasons, indicating that throughout the day photochemistry does not have any apparent effect on the hygroscopicity of the accumulation mode particles. Diurnal patterns of aerosol hygroscopicity for Aitken mode particles can be seen in Fig. 8. In the winter no particular pattern is visible at any of the locations; it can, however, be seen that while the aerosol hygroscopicity is similar between Hyytiälä and Vavihill, the Aitken mode aerosol at the Jungfraujoch is less hygroscopic. In the spring both Hyytiälä and Vavihill exhibit a clear diurnal pattern, which extends also into the summer. A peak in aerosol hygroscopicity is observed around midday when Dc reaches its minimum. Several previous studies have reported such behaviour in Hyytiälä and have attributed it to the vegetation activity, photochemistry and the ageing of organics during sunlight hours (Sihto et al., 2011; Cerully et al., 2011; Paramonov et al., 2013). While no diurnal pattern of aerosol hygroscopicity is visible for Jungfraujoch for winter and spring, a very clear pattern does exist in the summer and autumn. In these seasons Aitken mode particles exhibit an obvious decrease in hygroscopicity in the afternoon shown by the peak in Dc during these hours. This phenomenon has also been previously reported and attributed to the daytime intrusions of air from the planetary boundary layer (PBL) injecting less hygroscopic particles into the free troposphere (Kammermann et al., 2010a). The discussion above demonstrates that diurnal patterns of hygroscopicity are not the same everywhere and vary by seasons; however, the environments of Hyytiälä and Vavihill are similar enough to result in similar diurnal patterns.