Atmospheric Chemistry and Physics Open Access

Abstract. Ambient aerosol, CCN (cloud condensation nuclei) and hygroscopic properties were measured with a size-segregated CCNC (cloud condensation nuclei counter) in a boreal environment of southern Finland at the SMEAR (Station for Measuring Ecosystem-Atmosphere Relations) II station. The instrumental setup operated at five levels of supersaturation S covering a range from 0.1–1% and measured particles with a size range of 20–300 nm; a total of 29 non-consecutive months of data are presented. The median critical diameter D c ranged from 150 nm at S of 0.1% to 46 nm at S of 1.0%. The median aerosol hygroscopicity parameter κ ranged from 0.41 at S of 0.1% to 0.14 at S of 1.0%, indicating that ambient aerosol in Hyytiala is less hygroscopic than the global continental or European continental averages. It is, however, more hygroscopic than the ambient aerosol in an Amazon rainforest, a European high Alpine site or a forested mountainous site. A fairly low hygroscopicity in Hyytiala is likely a result of a large organic fraction present in the aerosol mass comparative to other locations within Europe. A considerable difference in particle hygroscopicity was found between particles smaller and larger than ~100 nm in diameter, possibly pointing out to the effect of cloud processing increasing κ of particles > 100 nm in diameter. The hygroscopicity of the smaller, ~50 nm particles did not change seasonally, whereas particles with a diameter of ~150 nm showed a decreased hygroscopicity in the summer, likely resulting from the increased VOC emissions of the surrounding boreal forest and secondary organic aerosol (SOA) formation. For the most part, no diurnal patterns of aerosol hygroscopic properties were found. Exceptions to this were the weak diurnal patterns of small, ~50 nm particles in the spring and summer, when a peak in hygroscopicity around noon was observed. No difference in CCN activation and hygroscopic properties was found on days with or without atmospheric new particle formation. During all seasons, except summer, a CCN-inactive fraction was found to be present, rendering the aerosol of 75–300 nm in diameter as internally mixed in the summer and not internally mixed for the rest of the year.


Introduction
Aerosol particles are omnipresent in the Earth's atmosphere and are known to have an impact on climate (e.g.McCormick and Ludwig, 1967;Twomey, 1974;and Lohmann and Feichter, 2005), visibility (e.g.Jinhuan and Liquan, 2000;and Seinfeld and Pandis, 2006) and human health (e.g.Seaton et al., 1995;Finlayson-Pitts and Pitts, 1997;and Hinds, 1999).From a climate perspective, aerosol particles directly interfere with the incoming solar and outgoing terrestrial radiation, as well as influence the Earth's radiation balance through their interactions with clouds.Through a variety of microphysical processes atmospheric particles influence the albedo, lifetime and precipitation patterns of clouds in what is known as the indirect effects of aerosols on climate (Forster et al., 2007).As noted in the Fourth Assessment Report of the Intergovernmental Panel on Climate Change (IPCC), the quantification of these indirect effects presents a major uncertainty in the current modelling and understanding of the aerosol-climate interactions, signifying the inclusion of clouds in the triple aerosol-cloud-climate system.
The aerosol particles participating in cloud formation are commonly known as cloud condensation nuclei (CCN), and their relevant behaviour in the atmosphere is dictated by their CCN activation and hygroscopic properties, i.e. their ability Published by Copernicus Publications on behalf of the European Geosciences Union.and likelihood to attract and retain water vapour molecules by means of condensation.It is important to remember that the activation of an aerosol particle into a cloud drop depends on the multitude of processes and atmospheric conditions prevalent during the particle formation and lifetime (McFiggans et al., 2006;Reutter et al., 2009).Besides the particle properties themselves, such as the number concentration, size and chemical composition, whether a particle can act as a CCN also depends on the partial pressure of the water vapour in the atmosphere (Pruppacher and Klett, 1997).
Atmospheric CCN have been studied extensively, and a multitude of studies exists related to the activation of CCN into cloud drops.Atmospheric CCN concentrations and sizeresolved activation spectra have been presented in numerous studies (e.g.Gras, 1990;Gunthe et al., 2009;and Sihto et al., 2011).A number of CCN closure studies has been conducted (e.g.VanReken et al., 2003;Broekhuizen et al., 2006;McFiggans et al., 2006;and Rissman et al., 2006), and the effect of organics and aerosol mixing state on CCN has been investigated as well (e.g.Cruz and Pandis, 1997;andWang et al., 2008, 2010).From a theoretical perspective, the Köhler theory and droplet growth kinetics have also been examined (e.g.Kulmala et al., 1993;Asa-Awuku and Nenes, 2007;Engelhart et al., 2008;and Prisle et al., 2010).Due to logistical matters and limitations, various in situ in-cloud measurements of CCN activation present a challenge, and have only been conducted at a small number of sites (e.g.Baltensperger et al., 1997;Hatakka et al., 2003;Leskinen et al., 2009;and van Pinxteren et al., 2012).At the same time, surface measurements, while easier to implement and with many measurement options available, are mostly limited to shortterm campaigns (e.g.Gunthe et al., 2009;and Cerully et al. 2011).Therefore, there is most definitely a need for more comprehensive long-term measurements of CCN activation and hygroscopicity around the world in order to improve the understanding of cloud droplet activation, aerosol-cloud interactions and the modelling of the future climate.
This paper presents a unique data set of CCN activation and hygroscopic properties measured in the boreal environment of southern Finland with a cloud condensation nuclei counter (CCNC).The measurement setup utilised in this study allowed both to estimate the particle size at which the aerosol in a boreal environment becomes important for cloud formation, as well as to infer its chemical composition.Since measurements were conducted at different supersaturation levels, relevant physical and chemical properties of CCN were determined for particles of different sizes.Aerosol mixing state was also indirectly determined, dividing the whole aerosol population into the CCN-active and CCN-inactive fractions.The uniqueness of the data set stems from its duration -a total of 29 months of data are presented here, making it the longest data set of ambient CCN activation and hygroscopic properties measured with a size-resolved CCNC.The aim of this paper is to describe the CCN activation of ambient aerosol in a boreal environment, both with respect to its size and chemistry in order to (i) investigate the presence of temporal (viz.diurnal and seasonal) variations of these properties, (ii) to determine the aerosol mixing state and its seasonality, (iii) to compare the derived values and variations with previously published results, and, finally, (iv) to provide the comprehensive insight into cloud droplet activation in the boreal environment with the outcomes possibly applicable to other boreal regions of the world.

Theory
In the atmosphere the formation of water droplets occurs as a result of condensation during the conditions of supersaturation S of air with respect to water vapour.While homogeneous nucleation of supersaturated water vapour is possible in laboratory conditions, the S required for this process to onset is on the order of several hundred percent and does not occur in the atmosphere (Andreae and Rosenfeld, 2008).Aerosol particles, always present in the atmosphere, act as seeds for cloud droplet formation at S levels much lower than those required for homogeneous nucleation, i.e. just a few percent.Therefore, the condensation of supersaturated water vapour onto the aerosol particles is the only significant pathway of cloud droplet formation in the atmosphere (Pruppacher and Klett, 1997).These aerosol particles are known as cloud condensation nuclei.The number of CCN in any given air mass may vary by several orders of magnitude, depending on the origin, and the spatial and temporal variation of CCN concentrations around the globe has been highlighted in various previous studies (e.g.Twomey and Wojciechowski, 1969;Hobbs et al., 1980;Snider and Brenguier, 2000;and Roberts et al., 2006).
From a thermodynamic point of view, the number of CCN is directly related to the ambient supersaturation within the air mass; Köhler theory indicates that higher S results in higher CCN concentration, in other words, as S increases, smaller particles can activate as cloud droplets (Köhler, 1936;Seinfeld and Pandis, 2006;Reutter et al., 2009).It is, therefore, important to remember that varying degrees of S will affect the aerosol of different properties, i.e. its size, hygroscopicity and the mixing state.However, the concept of CCN is not to be confused with the cloud droplet number concentration (CDNC) -while the CCN concentration is not dependent on cloud dynamics, the CDNC depends strongly on both aerosol particle number concentration and supersaturation, which, in turn, is affected by updraft velocities, aerosol number size distribution and its chemical composition (Leaitch et al., 1992;Snider et al., 2003).It is assumed in this paper that the growth of cloud droplets in warm stratiform clouds occurs on a very short time scale at a constant S, and, therefore, the effect of aerosol number size distribution on ambient S is not considered in this paper.Besides the total aerosol number concentration and ambient S, the ability of aerosol particles to act as CCN is strongly linked to their physical and chemical properties, such as aerosol critical diameter D c and hygroscopicity parameter κ (Petters and Kreidenweis, 2007).A multitude of literature exists on the effects of physics and chemistry of aerosol particles on their ability to act as CCN (e.g.Roberts et al., 2002;Dusek et al., 2006;and McFiggans et al., 2006); aerosol mixing state has also been investigated with respect to its influence on CCN properties (e.g.Cubison et al., 2008;Su et al., 2010;and Ervens et al., 2010).
There exist several techniques and measurement setups aimed at measuring and describing the properties of CCN.Number concentration of CCN-sized aerosol can be indirectly estimated from any instrument measuring aerosol size distributions at relevant sizes, such as the differential mobility particle sizer (DMPS); however, defining a lower limit of particle diameter is important.Since ambient in-cloud levels of S may vary both spatially and temporally, several lower limits have been proposed, e.g.50 (Roberts et al., 2002;Lihavainen et al., 2003) and 80 nm (Komppula et al., 2005;Asmi et al., 2011b); Asmi et al. (2011a) and Kerminen et al. (2012) included several other lower limits in an attempt to cover a wide range of S encountered in various ambient environments.The direct measurement of ambient CCN concentrations requires a measurement setup where ambient particles are counted, subjected to a certain known S, and then counted again.For such measurements various chambers can be used, such as a thermal gradient diffusion cloud chamber (e.g.Covert et al., 1998), coupled with particle counters and, possibly, a differential mobility analyser (DMA) in case the knowledge about a particular particle size is required.One such thermal gradient diffusion chamber constitutes the main component of the CCNC, an instrument commonly used to measure ambient CCN concentrations and to derive various CCN properties (Roberts and Nenes, 2005).It has been used to measure both total (e.g.Sihto et al., 2011) and sizeresolved (e.g.Gunthe et al., 2009) CCN concentrations.
The critical diameter D c is typically defined as the smallest particle size at which particles are activated and can grow to cloud droplet size under certain ambient conditions; this size is directly related to S through the Köhler theory (Pruppacher and Klett, 1997).While in theory this size separates particles into completely non-activated and activated fractions, in practice however, due to ambient aerosol commonly being externally mixed, it is typically defined as the size at which 50 % of particles grow to cloud droplets.Sihto et al. (2011) estimated both the activated fractions and D c by comparing CCN number concentrations measured by the CCNC to the total number concentration (N CN ) measured by the DMPS.If size-segregated CCN measurements are available, such as when coupling CCNC with a DMA, D c can be directly estimated from activation spectra, i.e. by plotting the N CCN /N CN ratios as a function of size and defining the point on a curve at which a certain fraction of aerosol particles activates (typically, 50 %) (e.g. Rose et al., 2008).Such measurement setup has been utilised in numerous studies (e.g.Corrigan and Novakov, 1999;Petters et al. 2007;and Gunthe et al. 2009).
Coupling aerosol particle dry size and its chemical composition is imperative for understanding the aerosol-water interactions (McFiggans et al., 2006), and there exists a number of approaches aimed at describing them (e.g.Fitzgerald et al., 1982;Svenningsson et al., 1992;Rissler et al., 2005;Rissler et al., 2006;Khvorostyanov and Curry, 2007;and Padró et al., 2007).The hygroscopicity parameter κ (Petters and Kreidenweis, 2007) is a unitless parameter describing the relationship between particle dry diameter, its chemical composition and the ambient S. For ambient atmospheric aerosol κ values fall within the range of 0.1 < κ < 0.9, and owing to the simplicity of its computation, it has been widely used in numerous studies (e.g.Merikanto et al., 2009;Kammermann et al., 2010;Rose et al., 2010;and Sihto et al., 2011).There are several ways to determine κ using ambient measurements; these can be performed by various instruments under both subsaturated and supersaturated conditions.CCN activation spectra measured by the aforementioned CCNC can be used to determine κ by applying the effective hygroscopicity parameter (EH1) Köhler model (Rose et al., 2008).D c calculated from activation spectra paired up with the known prescribed CCNC supersaturation level can be used in this model to determine κ utilising Eq. ( 1): where S is water vapour saturation ratio, D wet is the droplet diameter, D s is the dry particle diameter, κ is hygroscopicity parameter, σ sol is the surface tension of condensing solution, M w is the molar mass of water, R is the universal gas constant, T is the absolute temperature and ρ w is the density of pure water.As per Rose et al. (2008), D s can be substituted with D c and σ sol is taken as that of pure water (0.072 J m −2 ); T is the CCNC column average temperature recorded for each spectrum.κ values can be determined by varying κ and D wet so that S is equal to the prescribed supersaturation inside the chamber S eff and to the maximum of the Köhler model curve of CCN activation.The hygroscopicity tandem differential mobility analyser (H-TDMA) is another common measurement setup used to determine D c , κ and other relevant CCN properties at subsaturated conditions (Ehn et al., 2007;Swietlicki et al., 2008).The growth factors measured by the H-TDMA can be used together with, e.g. the EH1 Köhler model equation to determine both D c and κ.Since both CCNC and H-TDMA provide information about the hygroscopic growth of aerosol particles and other CCN properties, it has been common to run these measurements in parallel to conduct the so-called CCN closure studies, in which the results of the two instruments are evaluated and compared.The summary of such studies can be found in McFiggans et al. (2006).
As stated above, due to the simplicity of estimating the κ of ambient aerosol as described by Petters and Kreidenweis (2007), numerous studies, based both on ambient measurements and model simulations, have reported κ values for various locations around the globe.Both Andreae and Rosenfeld (2008) and Pringle et al. (2010) studied the hygroscopicity of the aerosol on a global scale.Several measurement campaigns have also attempted to describe the hygroscopicity of European aerosol (e.g.Dusek et al., 2006;Kandler and Shütz, 2007;and Bougiatioti et al., 2009).Gunthe et al. (2009), Jurányi et al. (2011) andLevin et al. (2012) all used a similar measurement setup to describe the hygroscopicity of aerosol in an Amazonian rainforest, a high Alpine site Jungfraujoch in Switzerland and a forested mountainous site in Colorado, respectively.There also exists a number of CCN hygroscopicity studies based on short-term measurement campaigns in the boreal environment, namely Hyytiälä Forestry Field Station, which is also the location of long-term CCN measurements presented in this paper (e.g.Ehn et al. 2007, Sihto et al. 2011;and Cerully et al., 2011).Temporal patterns of aerosol hygroscopic properties have also been reported in existing literature; however, due to the majority of studies being short-term measurement campaigns, mostly diurnal patterns of D c and κ have been reported.Seasonal variation has been reported in a modelling study by Pringle et al. (2010) and in several experimental studies (Cerully et al., 2011;Jurányi et al., 2011;Sihto et al., 2011;Levin et al., 2012).

Measurement site
CCNC measurements have been conducted at the SMEAR II (Station for Measuring Ecosystem-Atmosphere Relations) site in Hyytiälä Forestry Field Station, and they form a part of the comprehensive network of aerosoland meteorology-related measurements in southern Finland (Hari and Kulmala, 2005).The station (61 • 50 50.685N, 24 • 17 41.206E, 181 m a.m.s.l.) is located 220 km northnorthwest of Helsinki on a flat terrain surrounded by a boreal coniferous forest, mainly Scots pine, and is, therefore, well representative of the boreal environment.The nearest city of Tampere (pop.220 000) lies 50 km southwest; the station may, therefore, be described as a rural background site.SMEAR II is subject to both clean maritime air masses, as well as to more polluted (and less frequent) continental air masses; however, most often both of these air mass types are characterised by low aerosol particle number concentrations (Sogacheva et al., 2005).

Instrumentation
The setup for size-resolved CCN measurements is a multicomponent system, consisting of the DMA, condensation particle counter (CPC) and the CCNC unit itself, which includes a saturator column and an optical particle counter (OPC).The CCNC is a commercially available instrument, distributed by Droplet Measurement Technologies (DMT), Inc., and its main component is a continuous-flow streamwise thermal-gradient diffusion chamber.At SMEAR II the CCNC model is DMT CCN-100, and it has been operating continuously since July 2008.A more detailed discussion, than what follows below, about the operating principles of the DMT-CCNC, can be found in Roberts and Nenes (2005) and in Rose et al. (2008).
As the aerosol is entering the measurement setup through an inlet mounted approximately 8 m above the ground, it is first dried with a Nafion drier, following which the particles are neutralised with a 370 MBq 14 C radioactive source in order to achieve an equilibrium charge distribution.Afterwards the dry aerosol with a known charge distribution enters the DMA system, which categorises the particles according to their physical dimensions based on the principle of the aerosol electrical mobility (Aalto, 2004).The specific DMA in question is a Hauke-type DMA with a concentric cylindrical configuration and a length of 28 cm.The DMA is a key component for size-segregated CCNC measurements, and it sorts the particles into 30 logarithmically distributed size bins, with size ranging from 20-300 nm.Following this, the flow of the monodisperse aerosol of a known size is split into two parallel lines, with the first leading to the laminar flow type CPC (model TSI 3772; sample flow 1 ± 0.015 L min −1 ; sheath flow 14 L min −1 ), which determines the total particle number concentration of that particular size.This quantity is hereafter referred to as N CN .The second parallel line leads to the CCNC unit itself.At SMEAR II the CCNC saturator unit is a vertical flow tube of cylindrical shape with an inner diameter of 2.3 cm and a length of 50 cm.Inside this chamber the aerosol flows from top to bottom with a flow rate of 0.45 ± 0.02 L min −1 , surrounded by the filtered sheath air with a sheath-to-aerosol flow ratio of 10; aerosol flow occurs under laminar conditions and near-ambient pressure.The thermal electric coolers (TEC) and thermocouples mounted at the beginning, middle and end of the outer tube wall create and monitor a near-linear positive temperature gradient within the tube.The inner walls of the flow tube are continuously wetted with liquid water.During the passage of the laminar flow through the tube, heat and water vapour are effectively transported from the inner walls to the centerline of the tube, thus creating a constant uniform water vapour supersaturation S eff along the centerline.The CCNC in question operates at five levels of S eff : 0.1, 0.2, 0.4, 0.6 and 1.0 %.As the monodisperse aerosol flows through the tube, those particles with smaller critical supersaturation S c than S eff will grow by the condensation of water vapour, i.e. activate as CCN.The typical residence time of the aerosol particles in the saturator unit is on the order of 10 s -sufficient enough for the resulting drops to grow to ∼ 1 µm in diameter.The activated particles are then counted by the OPC, and this quantity is hereafter referred to as N CCN .In order to assure the quality of the CCNC data, the CCNC is calibrated approximately every two months using nebulised, dried, chargeequilibrated and size-segregated ammonium sulfate aerosol, and the procedure is carried out as per Rose et al. (2008).These calibrations reveal a deviation of a presumed S eff of 0.1 %, 0.2 % and 0.4 % by up to 10 %, 12 % and 7 %, respectively.These deviations were not taken into account throughout the analysis, and the nominal values of S eff were used.
For a given S eff both prior to and after the measurement of each activation spectrum, the instrument also measures the total, non-size-segregated concentration of both N CN and N CCN , (DMA bypassed).The non-size-segregated CCNC measurements are not discussed in this paper.The full scan at one S eff takes, on average, 17 min 40 s, with activation spectrum taking 17 min.Prior to 25 January 2010, between 13 and 40 s were allocated for the saturator unit to adjust to a new, higher S eff ; 3 min 20 s were allocated for adjusting from the S eff of 1.0 to 0.1 %.From 25 January 2010 onwards these times have been increased to 1 min 10 s and 15 min for an upward and a downward shift in S eff , respectively.Therefore, slightly more than three full activation spectra per hour were measured by the instrument prior to 25 January 2010, and slightly less than three full activation spectra per hour after that date.The statistics about data coverage presented in the next section are based on these numbers.

Data description
This paper presents the analysis of size-segregated CCNC data from SMEAR II from 10 February 2009 until 30 April 2012.Despite continuous measurements, from 9 July 2010 until 9 May 2011 the instrument was operating at levels of S eff different from those indicated above.This occurred due to a miscalibration of the instrument, and the data from this period are not included in the analysis.The data set has undergone a rigorous procedure of cleaning and processing in order to remove bad data -due to the complexity of the measurement setup and absence of standardised procedures for working with size-segregated CCNC data, this was the most challenging and time consuming part of the analysis.The data set was scanned for unstable S eff levels, problems with CPC and OPC and the resulting counting errors of N CN , N CCN and A, and differences between set and measured sheath flow rates.Figure 1 presents the data availability for the whole measurement period.The results presented below are based on a total of 46 400 activation spectra.
For each spectrum the activated fraction A was calculated for each size bin by dividing N CCN by the corresponding N CN .The following cumulative Gaussian (normal) distribution function based on the non-linear least-squares fitting method was then fitted to each activation spectrum: where a is half the maximum value of A in each spectrum, erf is the error function, D is the fitted particle diameter, D a is the particle diameter at A = a, and σ is the standard deviation of the cumulative Gaussian distribution function, as described in Rose et al. (2008).In the function above the parameter D a is the critical diameter of dry aerosol particles D c , which in this study is defined as the diameter at which half of the incoming particles are activated at a certain S eff .Two methods were used to fit Eq. ( 2) to each activation spectrum.The first method closely follows that proposed by Rose et al. (2008), in which all A in each activation spectrum are normalised to unity, i.e. divided by the maximum A. In this method D a and σ are the fit parameters, where D a is the critical diameter D c at a = 0.5, and the maximum activated fraction (MAF) is 1 (Fig. 2, left panel).Hereafter, this method is referred to as normalised.
Since the measurements deal with ambient aerosol, which can be externally mixed, and since a significant variation was observed in A at larger sizes, a second fitting method was used, where no normalisation of A was carried out and a was the third fit parameter.In this method the function found the best fit to each spectrum regardless of where it levelled out at larger sizes (MAF and a are not specified and allowed to  vary), and D c was found as D a at a (Fig. 2, right panel).Hereafter, this method is referred to as non-normalised.As per Rose et al. (2010), the normalised method gives properties of the external mixture of both CCN-active and CCN-inactive aerosol, while the non-normalised method specifically considers only the CCN-active fraction of the aerosol population.The κ values were calculated according to Eq. ( 1) using nominal values of S eff and both sets of D c derived using normalised and non-normalised methods.It is important to note here that both fitting methods yielded physically unreasonable D c and κ values for a small amount of data, and 0.4 % of all activation spectra were excluded from the analysis.The activation spectra were not corrected for effects of the DMA transfer function, as these effects are assumed to be negligible in the described measurement setup.The effects of doubly charged particles were investigated for a subset of spectra at S eff of 0.1 and 0.2 % by determining the fraction of activated doubly charged particles and assuming that this fraction is constant over the whole size range (Rose et al., 2008).In the majority of investigated cases this fraction of activated doubly charged particles was small (< 0.1), and it was found that correcting the spectra for the doubly charged particles increases D c by an average of 2 %, rendering the implementation of correction procedure for the whole data set unnecessary.

Results and discussion
As mentioned previously, a total of 46 400 activation spectra were collected, processed and analysed in this study; this number was roughly equally distributed among the five levels of S eff .The median values and variability of the total particle number and CCN concentrations, as well as of the activated fraction for two levels of S eff are presented in Fig. 3.As expected, a larger number of particles activated at a higher S eff ; this is also signified by the s-shaped activation curve moving to the left in the right panel of the figure (i.e.smaller size).In the studied 20-300 nm size range, the highest concentrations were found in the upper end of the Aitken mode, around 80-100 nm.While the median concentrations in these size bins are approximately at 12 particles cm −3 , the percentiles indicate that a large variation in number concentrations is present.
Several things are of interest in Fig. 3. First, the largest size bin of 300 nm exhibited a fairly large variation of A values; the median A in this size bin at the S eff of 1.0 % is above unity.This, while physically unreasonable, is a direct consequence of very low particle number concentrations at this size, and may also stem from different counting efficiencies of the CPC and the OPC.The 300 nm particle size bin was not excluded from the data set due to its presumably low impact on the Eq. ( 2) fitting method, especially so for the non-normalised procedure.Second, the variability of A increased with an increase in particle size -this is also related to low particle number concentrations and counting statistics.Third, while certain activation spectra, especially at lower S eff , did exhibit a secondary plateau of doubly charged particles, this plateau is not visible in the left panel of Fig. 3, indicating its infrequent occurrence and further explaining the reasons behind the decision of not correcting the spectra for doubly charged particles.The results of the normalised and non-normalised fitting of Eq. ( 2), the related terms and the derived parameters of D c and κ are shown in Table 1; these are presented separately for each S eff , as well as for the whole data set.

Method comparison
While both normalised and non-normalised methods are intended to provide hygroscopic properties of an aerosol population, their differences in describing the respective entire aerosol population and only the CCN-active fraction of the population are clearly evident (Table 1).For each fitting of Eq. ( 2) to the activation spectrum, a root mean square error (RMSE) was calculated to estimate the appropriateness of the fit.While for the lowest S eff of 0.1 % the median RMSE of Eq. ( 2) fit was the same for both methods, for all other S eff the goodness of the fit was worse for the normalised method.This feature is a direct consequence of the inclusion of the presumably present CCN-inactive fraction in the fit of the normalised method, and the normalised method fit being largely dependent on one, maximum observed, A value in each spectrum.
For all levels of S eff , both separately and overall, D c values and their variation in the normalised method are both higher than those in the non-normalised method.In fact, 97.4 % of all D c from the normalised method are higher than the corresponding D c from the non-normalised method (Fig. 4).Logically, if one includes the CCN-inactive fraction in the Eq. ( 2) fit, the overall hygroscopicity of the aerosol population in question would decrease, increasing the critical diameter needed for CCN activation.The remaining few percent visible in Fig. 4 above the white 1 : 1 line result from those spectra at S eff of 0.1 % where the number of A points close to unity was very low and the resulting non-normalised Eq. ( 2) fit curve levelled out at values of MAF much higher than unity.
Following what has been stated above, κ values from the normalised method are generally lower than those from the non-normalised method (Table 1).The results of the comparison of methods presented above with respect to D c and κ agree well with those reported by Gunthe et al. (2009).The non-normalised method, besides describing only the CCNactive fraction of the aerosol, also provides insight into the aerosol mixing state by way of quantifying the CCN-inactive fraction of the aerosol population, a topic further examined in Sect.4.6.In order to carefully describe the CCN activity of an ambient aerosol population, and taking into account the fact that the non-normalised method demonstrates the properties of CCN-active aerosol fraction and provides insight into the aerosol mixing state, from this point forward only the D c and κ values from the non-normalised method are discussed, unless otherwise specified.

General patterns in data
It is clearly visible in Table 1 that the function fit, as shown by the RMSE, was best at S eff of 0.1 % and worsened as S eff increased.For the highest S eff the modelled A deviated by up to 15 % from observed A as seen by the 75th percentile of the RMSE.Having mentioned the decreasing absolute particle number concentrations at sizes larger than ∼ 100 nm and the resulting poor statistics of A, this increase in RMSE is not surprising.It was observed that while the function modelled A very well for small, non-activated particles and during the slope of the s-shaped curve, the largest differences in the observed and modelled A were found where the function reached MAF (Fig. 2).The increased RMSE at higher S eff is not expected to greatly affect the derived D c ; it does, however, call for a strict mathematical definition of the observed MAF.With respect to the mid-point a of the modelled A spectrum and MAF, there is no significant correlation with S eff .Both a and MAF are indicative of the CCN-active and -inactive fractions of the aerosol.The median critical diameter D c ranged from 150 nm at S of 0.1 % to 46 nm at S of 1.0 %.As expected from the Köhler theory, the median D c decreases with increasing S eff , and as expected from Eq. ( 1) this decrease is exponential (Fig. 5).From the lowest to highest S eff the D c decreases by over 100 nm from the lower end of accumulation mode to the midpoint of the Aitken mode.Considering that highest number concentrations within the 20-300 nm size range were found in the upper end of the Aitken mode, a change in ambient S has a strong impact on the number of particles that can act as CCN and, consequently, the number of cloud drops formed.When the quartiles of the median D c are examined, it is obvious that the variation in calculated D c also diminishes with increasing S eff .This may stem from the fact that in general at higher S eff more particles activate, as well as that there are more A points close to MAF at the larger end of the size spectrum, improving both the fit and statistics, and reducing the uncertainty and variation.The exponential decrease of both median D c and its variation with increasing S eff indicates that a change in ambient S at smaller values (e.g.0.05 % < S < 0.3 %) has a larger effect on the number of CCN than a change in ambient S at higher values (e.g.0.5 % < S < 1 %).This notion is important to keep in mind considering that the typical levels of ambient S found in warm stratiform clouds in the boreal environment of the measurement site are typically low (Seinfeld and Pandis, 2006;Lihavainen et al., 2008;Anttila et al., 2009).Assuming that the maximum ambient S typically encountered in warm clouds in the boreal environment is ∼ 0.3 %, the minimum particle size required for CCN activation would be ∼ 85 nm; air masses with predominant S of 0.1-0.15% would require particles as big as 120-150 nm in diameter for CCN activation.This notion needs to be kept in mind when estimating CCN concentrations from, e.g.DMPS data.
The median aerosol κ ranged from 0.41 at S of 0.1 % to 0.14 at S of 1.0 % (Table 1; Fig. 6, left panel).This supports the notion previously made by Birmili et al. ( 2009) that particle hygroscopicity is size-dependent in the region of interest.The variation of κ was also found to decrease with decreasing D c , as seen by the quartiles, which is in disagreement with the results presented by Swietlicki et al. (2008) and Jurányi et al. (2011).The relationship between κ, D c and S eff is complex (Eq.1), and D c may be used as an indicator of particle hygroscopicity as is; the parameter κ, however, allows for an easier comparison with previous studies, including those with different measurement setups.
Given the overall median κ of 0.22 reported in this study, it can be concluded that the aerosol in the boreal environment is slightly less hygroscopic than the global continental aerosol (an overall average of 0.27 as reported by Pringle et al., 2010).Figure 5 shows that while particles larger than 100 nm in diameter are slightly more hygroscopic than the global continental average, those smaller than 100 nm in diameter are less hygroscopic than the global continental mean of κ.Indeed, it seems as though the smaller the particle, the less hygroscopic it is comparative to the global continental mean of κ as seen by the deviation of the points away from the mean global κ line in Fig. 5.The size at which the aerosol particles in Hyytiälä are divided according to differences in CCN-relevant chemical composition coincides with the size typically defining the boundary between Aitken and accumulation modes, i.e. 100 nm (Fig. 5).This notion is of interest since accumulation mode particles are more likely to have been processed by clouds, and, hence, have a higher hygroscopicity.This issue is examined in more detail in the subsequent section.The aerosol at the background station in southern Finland is also quite less hygroscopic than the European continental aerosol (overall average of 0.36 reported by Pringle et al., 2010).Being located in a forested rural area in the boreal zone, this lower hygroscopicity is likely explained by the presence of a larger aerosol organic fraction when comparing to other locations within Europe (Zhang et al., 2007).Gunthe et al. (2009) utilised a similar CCNC measurement setup in the Amazonian forest for a month in February and March 2008.For the CCN-active fraction of the aerosol the study reported median D c values of 198 and 129 nm for S eff of 0.1 and 0.19 %, respectively, both of which are higher than their corresponding values from this study.Accordingly, κ values reported by the same study for same S eff levels were lower than in this study (Fig. 6, left panel).In fact, for five levels of S eff from 0.1 to 0.82 % Gunthe et al. (2009) reported an overall median value of κ of 0.15, attributing it to a very high organic mass fraction in the Amazonian aerosol, reported to be as high as 90 % for Aitken mode particles.Even though the measurements by Jurányi et al. (2011) were conducted at the high Alpine site, mostly in the free troposphere, their reported median κ of 0.20 is very similar to the one reported here, albeit a bit smaller -this is especially true for larger particles (Fig. 6, left panel).Measurements by Levin et al. (2012) in a forested mountainous site in Colorado resulted in an average κ value of 0.16, also smaller than 0.22 reported here.Comparing the hygroscopicity of aerosol among these sites reveals that while aerosol in Hyytiälä exhibits low hygroscopicity with a large organic fraction present, it is more hygroscopic than the aerosol in the Amazon, high Alpine or forested mountainous environments (Fig. 6, left panel).When the slope values of the linear regression fits in the left panel of Fig. 6 are examined, they point out that the rate of change of aerosol hygroscopicity with size in Hyytiälä is higher than in the Amazon, high Alpine or forested mountainous environments, revealing differences in the species of condensing material and the oxidation and aging processes.
The slopes of the linear regression fits in the right panel of Fig. 6 indicate that the increase in hygroscopicity with size in Hyytiälä is highest in the winter and lowest in the summer.One possible reason for this seasonal difference is the notion that in the summer more active SOA formation takes place, resulting in less hygroscopic particles, and in the winter a large fraction of aerosol contains sulfate, especially from the long-range transport.Particle growth is also slower in the winter compared to the summertime due to a smaller amount of condensable vapours.Sihto et al. (2011) studied aerosol hygroscopic properties for a year at SMEAR II using CCNC coupled with a DMPS and H-TDMA.The reported average D c values for S eff of 0.4 % derived from both of these setups were higher than 75 nm reported here, stemming from the fact that aerosol populations measured with these two setups include the CCN-inactive fraction, therefore decreasing its overall hygroscopicity.The same explanation applies to the overall average κ of 0.18 as reported by Sihto et al. (2011) for the aerosol of 35, 50, 75 and 110 nm in diameter compared to the overall median κ of 0.22 reported in this study.An average κ of 0.18 is similar to an average κ of 0.2 reported by Cerully et al. (2011) measured with a CCN chamber at the same location for particles of 40, 60 and 80 nm in diameter.In their study, however, κ did not seem to be dependent on the size.
Having mentioned the variation of D c and κ with S eff , it is important to note that the overall median values presented for all levels of S eff should be used with caution -they are representative only of the aerosol population which was measured at five specified levels of S eff .In other words, had the CCNC in question been operating at an additional S eff level of, for example, 1.2 %, the overall median values of both D c and κ would be lower.It is advised that for size-segregated CCNC measurements the presentation of averaged D c and κ values should always include the levels of S eff at which the instrument operated; the values may also, of course, be presented for each level of S eff separately.

The distributions of D c and κ
Figure 7 shows the distributions of κ as a function of D c .Several features are visible in this figure.For a given S eff , κ is related to D c with an approximate function of κ ∼ D −3 c , a direct consequence of Eq. ( 1) assuming that other factors in that equation do not significantly change.The ratio of these κ ∼ D −3 c relationships scales approximately with 1/ln 2 (S eff ) for different S eff levels, and there is very little variation in the κ ∼ D −3 c relationship due the functional interrelationship in the fitting procedure.
Perhaps the most interesting feature in Fig. 7 is the evolution of the κ distributions with decreasing S. The measured κ is probably mostly representative of the particle properties near the activation diameter D c , and thus different S eff values give indication of the frequency of different hygroscopicities as a function of particle diameter.Figure 7 includes the frequency distributions of both κ and D c , plotted separately on vertical and horizontal axes, and the 25th, 50th and 75th percentiles are plotted directly over the main figure.The κ distributions of the higher S eff measurements show very similar behaviour, with similar distribution width and the median κ close to 0.2.Between the S eff of 0.4-0.2%, the median κ changes significantly and increases to approximately 0.4, and the distribution gets much wider.This widening can be partially explained by decreasing instrumental accuracy at smaller S eff , but more likely it is a result of the much higher variability of the hygroscopicity of larger particles in general.The increase of κ with increasing (critical) diameter is consistent with the change of aerosol population from Aitken to accumulation modes around 100 nm particle diameter, also suggested earlier in Sect.4.2 (see e.g.Asmi et al., 2011a, for details of SMEAR II station size distributions).Accumulation mode particles are expected to have been activated at least once as cloud droplets, and thus been subject to cloud processing and other heterogeneous reactions in the aqueous phase.This should increase the contribution of soluble material, and, hence, the increase in κ should be expected.It is, however, clear that this kind of size-dependent differences in the κ distributions does not support the use of a single κ value regardless of the particle size.One should also notice that even for a single S eff , the κ distributions are close to lognormal in shape, which indicates that arithmetic mean values of κ are probably not the most representative values to use.
Another way to look at the κ-D c distributions is by observing the range of κ for a given critical diameter.For example, looking at the particles of near 100 nm critical diameter, one could get κ values from 0.01 to over 1, with the possibility of even higher hygroscopicities for S eff lower than the minimum used in this study.Thus, using a single κ value for a given particle size can severely oversimplify the hygroscopicity variability of an aerosol population.Due to the low number of sampled S eff , a 3D probability density map of κ for each D c , or even a reliable fit for any specific D c are not trivial to derive; however, these results show that even in generally relatively homogeneous regional background conditions the range of κ is large.One should note that the κ values given are not for individual particles, but they represent the aerosol population at sizes near the D c value, and thus the given variability should be considered to be the variability of aerosol populations, not of individual particles.

Temporal variation
As mentioned previously, the data set of aerosol CCN activation and hygroscopic properties presented here is unique in a way that it is the first multi-year long-term data set of ambient measurements in a boreal environment.Figure 8 presents the seasonal variation of D c and κfor two levels of S eff .The seasonal pattern of aerosol CCN activation and hygroscopicity is only pronounced for larger particles, similar to what has been reported by Levin et al. (2012) for a high-altitude forested site in Colorado.Ambient aerosol populations measured at S eff of 0.1 % seem to be more hygroscopic in winter and less hygroscopic in summer, with February and July being the months showing respective maximum and minimum aerosol hygroscopicity.A minimum of κ in July was also reported by Levin et al. (2012).
It is plausible that the more active SOA formation and the increased organic fraction being the result of increased emissions of the volatile organic compounds (VOCs) from the surrounding boreal environment in the summer are responsible for reducing aerosol hygroscopicity when compared to the wintertime.Considering that in Fig. 8 the D c and κ points for S eff of 0.1 % mirror each other, it is important to point out that the aerosol hygroscopicity can be inferred from the critical diameters alone.The highest median κ of 0.74 in the month of February for S eff of 0.1 % is very likely related to a higher mass fraction of sulfate within the aerosol mass and a generally slower growth of particles to larger sizes, which allows for a longer time for the oxidation and aging of particles.What is more important to remember is that this value of κ reflects the hygroscopicity only of larger particles, those with diameters of ∼ 150 nm.Naturally, as the size decreases, the hygroscopicity decreases (Table 1), and the seasonal pattern disappears.The winter peak in aerosol hygroscopicity presented here agrees well with seasonal patterns presented by Pringle et al. (2010) and Sihto et al. (2011) for sites in Germany and Hyytiälä, respectively.Both of these studies also reported aerosol hygroscopicity to be lowest in November, which differs from results presented here (Fig. 8).Since Sihto et al. (2011) employed an H-TDMA to derive hygroscopic properties, the difference in the lowest annual aerosol hygroscopicity may stem from the inclusion of the CCN-inactive fraction in the calculations in the aforementioned study.The pattern in Fig. 8 is clear, and represents the annual hygroscopic properties of the CCN-active fraction only.Also of interest is the fact that the seasonal pattern of D c presented in Sihto et al. ( 2011) is evident for all levels of S eff (at S eff of 0.1 % pattern is unclear) and, therefore, for all particle sizes; in the current study the pattern is obvious only for the larger particles and becomes insignificant towards the smaller sizes.There does not seem to be a distinct seasonal pattern for particles smaller than 50 nm -those measured at S eff of 1.0 %.Taking into account that typical levels of ambient S encountered in warm stratiform clouds are low (Pruppacher and Klett, 1997), and, therefore, that particles which activate into cloud drops are the larger ones (> 100 nm), the variation in their chemical composition throughout the year plays a crucial role in cloud formation.The exact reasons for decreased aerosol hygroscopicity during the summertime in Hyytiälä need to be further investigated by a careful comparison with other relevant variables, such as, e.g.concentrations of VOCs.
The analysis of the diurnal behaviour of the aerosol CCN activation and hygroscopic properties revealed that there is no clear diurnal variation of either D c or κ for the largest particles (those with ∼ 150 nm diameter measured at S eff of 0.1 %).When the smallest particles (< 50 nm) are examined, the absence of the variation is not as clear.When the median values alone are examined, it seems as if particle hygroscopicity is highest around noon and lowest around midnight.The variability and the absolute difference in monthly median values, however, render this variation insignificant and inconclusive.In order to examine the diurnal behaviour in more detail, it was separated by seasons (Fig. 9).Visibly there is most certainly no diurnal variation in aerosol CCN activation and hygroscopic properties in Hyytiälä in autumn and winter.
The spring and, especially, summer do exhibit more conclusive diurnal patterns, with both plots showing a decrease in D c after noon and an increase around midnight.Of interest also is the variation as demonstrated by the error bars -during the summertime variation in monthly median values is the smallest, pointing to a more significant pattern in data.Both Sihto et al. (2011) and Cerully et al. (2011) reported diurnal patterns of aerosol hygroscopic properties in Hyytiälä and attributed them to the influence of photochemical reactions and the aging of organics during the sunlight hours.The diurnal patterns presented here indicate that these processes may well be the reasons for the observed diurnal behaviour, especially considering that the variation is only observed in the spring and summer.The results here agree well with those presented by Cerully et al. (2011), which reported a similar diurnal pattern in κ for particles of 60 nm in diameter measured for two months during the spring in Hyytiälä.While the diurnal pattern presented here is also similar to the one reported by Sihto et al. (2011), the latter study pointed out that the diurnal variation is more pronounced for larger particles (∼ 175 nm in diameter), which is opposite to what has been seen in this analysis.Considering all that has been stated above, it seems that photochemical reactions, aging of the organics, temperature and, possibly, atmospheric nucleation all affect the diurnal behaviour of CCN activation and hygroscopic properties only for the small, Aitken size aerosol, with a diameter of ∼ 50 nm.For larger aerosol (> 100 nm in diameter), the occurrence of these and other processes taking place at different time scales results in no distinguishable effect of photochemistry and temperature on the diurnal behaviour of aerosol CCN activation and hygroscopicity.
Overall, the biogenic emissions and subsequent condensation in the boreal environment of southern Finland make ambient aerosol of > 100 nm in diameter less hygroscopic in the spring-and summertime, compared to other seasons; the participation of these emissions in photochemical reactions is responsible for introducing a diurnal pattern in the behaviour of aerosol hygroscopicity in the spring and summer for particles less than 50 nm in diameter.The diminished photochemistry, temperature and biogenic activity in autumn and winter result both in the highest seasonal hygroscopicity of larger ambient aerosol and in the absence of the diurnal variation of aerosol hygroscopicity for particles of any size.

CCN activation and hygroscopic properties of particles produced by atmospheric new particle formation (NPF)
It has already been shown for a number of locations that NPF locally increases the CCN number concentration (e.g.Kerminen et al., 2012).As a location where atmospheric new particle formation has been studied extensively (e.g.Dal Maso et al., 2005), of interest is to determine the potential difference in D c and κ between NPF and non-NPF days in Hyytiälä.For this purpose the nucleation event classification As mentioned in the previous section, the hygroscopicity of particles of ∼ 150 nm in diameter did not exhibit a diurnal variation in any of the seasons, and the same is true when only the NPF days are considered.During NPF days there is no diurnal variation of D c and κ, and, considering the time it takes for freshly nucleated particles to grow to ∼ 150 nm in diameter, similar CCN activation and hygroscopic properties are expected for NPF and non-NPF days.For smaller, ∼ 50 nm particles, the situation is more complicated.Since nucleation events in Hyytiälä occur primarily in the spring (Dal Maso et al., 2005), a subset of 29 spring Type I nucleation events and 53 spring non-event days was selected and analysed (Fig. 10).The diurnal pattern of D c during nonevent days roughly follows the spring pattern depicted in Fig. 9, with a trough observed around noon; there is, however, a large variation present.The diurnal pattern during the NPF days is also similar in that lowest D c values are found in the afternoon, and highest between 00:00 and 06:00 UTC+2 the following day.Nucleation during an event in Hyytiälä typically starts around 10:00 and it takes, on average, 15 h for particles to grow to 50 nm in diameter (Dal Maso et al., 2005).Therefore, if any difference is to be observed in D c values between NPF and non-NPF days, it should be visible at 01:00 the next day and later.Figure 10 shows that at this time the freshly nucleated particles have grown to be less hygroscopic than they would be in the absence of NPF.While it may be stated that particles produced by NPF and by subsequent growth by vapour condensation may be less hygroscopic due to an increased organic fraction within the aerosol mass, the variability seen in Fig. 10 indicates that this difference between NPF and non-NPF days is not well pronounced and not conclusive.Indeed, if Type II events are included in the analysis, the difference in D c between 01:00 and 06:00 the following day disappears.Similar is true if summer NPF and non-NPF days are included.Both Ehn et al. (2007) and Sihto et al. (2011) reported an absence of difference in diurnal patterns between NPF and non-NPF days.
What can be concluded from this analysis is that CCN activation and hygroscopic properties of ∼ 50 nm aerosol may be different between NPF and non-NPF days; however, this difference only becomes probable when strong spring nucleation events, for which the growth rates can be calculated with a certain degree of accuracy, are taken into account.Generally speaking, it seems as though particle CCN activation and hygroscopic properties are more affected by the photochemical reactions and atmospheric oxidation of aerosol chemical species, regardless of whether the particles have been produced by NPF or not.

The inference of ambient aerosol mixing state
Of interest in Table 1 is the overall median maximum activated fraction MAF of 0.95, indicating that for particles of larger sizes on average only 95 % of the total aerosol population were activating into cloud drops.Derived as a result of the non-normalised fitting procedure of Eq. ( 2), it represents the fraction of the aerosol population which can be considered CCN-active (Rose et al., 2010).The remaining fraction of 5 %, calculated as 1 -MAF, can then be considered as the CCN-inactive fraction or the fraction of the aerosol population that does not activate into cloud drops regardless of how big the particles are or how high the S inside the CCNC is.This partitioning into CCN-active and -inactive fractions may provide insight into the ambient aerosol mixing state; however, such derivation of the mixing state is size-dependent, since for different levels of S eff the function reaches MAF at different sizes.In other words, the point at which the function in Eq. ( 2) reaches MAF shifts to smaller sizes at increasing S eff .Even though the size range, for which the aerosol mixing state can be estimated, varied with S eff , the median MAF values and the quartiles indicate that there was no significant difference in MAF values among S eff levels.The absence of this difference points out that within the particle size range of ∼ 75-300 nm in diameter the nonnormalised method can provide an accurate estimate of the ambient aerosol mixing state.
The quartiles of the overall median MAF indicate that the CCN-inactive fraction may constitute as much as 10 % of the particle number within the size range of ∼ 75-300 nm in diameter or may not be present at all.A closer look at the monthly distribution of the CCN-inactive fraction revealed a seasonal pattern (Fig. 11).The smallest CCN-inactive fraction appears to occur in May, June and July, with the latter being a somewhat unphysical result with a median MAF higher than unity.While this may be explained by the nature of the non-normalised fitting of Eq. ( 2) and variation in number concentrations at larger sizes, it seems clear that aerosol in the boreal environment is internally mixed in May, June and July, and partially externally mixed throughout the rest of the year.In fact, for May, June and July the median CCN-inactive fraction comprises only 0.2 % of the aerosol population in the abovementioned size range, and for the rest of the year this fraction is 6.6 %.Logically, this seasonal difference also has implications for the choice of method for fitting Eq. ( 2) to each activation spectrum.If the CCNinactive fraction is very small, the difference in D c derived from two methods is not as pronounced.Indeed, the regression coefficient R 2 between D c derived from two methods in May, June and July is 0.86, and for the rest of year it is 0.57.If the aerosol population in question is internally mixed, it makes no difference whether activation spectra are normalised or not, as both methods produce similar results.If, however, aerosol population is externally mixed, the choice of the method is necessary and it depends on whether one is interested in the whole aerosol population or only in its CCN-active fraction.
It is commonly presumed that the CCN-inactive fraction of the aerosol population consists mainly of insoluble and refractory components, such as mineral dust and black carbon (Gunthe et al., 2009).It can, therefore, be concluded that in Hyytiälä in late spring and early summer the aerosol of ∼ 75-300 nm in diameter is internally mixed with low concentrations of CCN-inactive compounds.During the rest of the year the aerosol is not internally mixed and the insoluble and refractory material is present to a larger degree.With black carbon being the most important light-absorbing constituent of the aerosol, this notion is consistent with seasonal cycles of absorption coefficients presented by Virkkula et al. (2011); the study showed the lowest and the highest average absorption coefficients in July and February, respectively.Hyvärinen et al. (2011) andHäkkinen et al. (2012) also reported the lowest aerosol black carbon concentrations in Hyytiälä in the summertime.

Conclusions
CCN activation and hygroscopic properties of ambient aerosol have been measured in the boreal environment of southern Finland with a cloud condensation nuclei counter (CCNC), and this paper presents an in-depth analysis of 29 non-consecutive months of size-segregated measurement data, covering a particle size range of 20-300 nm in diameter and a supersaturation S range of 0.1-1 %.The median critical diameter D c ranged from 150 nm at S of 0.1 % to 46 nm at S of 1.0 %.For estimating the CCN number concentrations from ambient size distribution measurements in Hyytiälä, a lower limit of 85 nm particle diameter is suggested, assuming an ambient S of < 0.3 % in warm stratiform clouds.The median aerosol hygroscopicity parameter κ ranged from 0.41 at S of 0.1 % to 0.14 at S of 1.0 %, indicating that ambient aerosol in Hyytiälä is slightly less hygroscopic than the modelled global continental average and quite less hygroscopic than the modelled European continental average (Pringle et al., 2010).The lower hygroscopicity in Hyytiälä is likely a result of a large organic fraction present in the aerosol mass comparative to other locations within Europe (Zhang et al., 2007).At the same time, the hygroscopicity itself and its rate of increase with size are larger in the boreal environment than in the Amazon rainforest, high Alpine or forested mountainous environments, pointing to differences in the condensing species, aerosol oxidation and aging.A considerable difference in particle hygroscopicity was found between particles smaller and larger than ∼ 100 nm in diameter, possibly pointing to the effect of cloud processing increasing κ of particles > 100 nm in diameter.Considering that κ was found to vary with size and that κ distributions were found to be lognormal, the use of a single parameter, mean or median, κ for describing the hygroscopicity of an aerosol population is discouraged.
Throughout the year the hygroscopicity of the smaller, ∼ 50 nm particles did not change, indicating the homogeneity of the CCN-relevant chemical composition.Particles with a diameter of ∼ 150 nm showed a decreased hygroscopicity in the summer, which is possibly attributed to the increased VOC emissions of the surrounding boreal forest.This has direct implications for cloud formation, since typical levels of ambient in-cloud S are very low.For the most part, no diurnal patterns of aerosol hygroscopic properties were found; exceptions to this are weak diurnal patterns of small, ∼ 50 nm particles in the spring and summer.An increase in hygroscopicity around noon was observed for particles of this size, signifying the role of photochemistry and aging of the organics during the sunlight hours.No clear difference in CCN activation and hygroscopic properties was found for aerosol on days with or without atmospheric new particle formation.It was discovered that during all seasons, except summer, a CCN-inactive fraction is present, accounting for as much as 7 % of the total aerosol population, rendering the aerosol as www.atmos-chem-phys.net/13/10285/2013/internally mixed in the summer and not internally mixed for the rest of the year.The analysis and results presented in this paper concentrate primarily on what can be derived and deduced from CCNC data alone.Taking this analysis one step further by using data from other instrumentation at the same location would certainly provide more interesting results and a more detailed insight into aerosol hygroscopic properties in a boreal environment.The concentrations of non-refractory ambient material, such as black carbon, could be used to quantitatively correlate them with the CCN-inactive fraction.The source apportionment of differing aerosol hygroscopicity can be carried out by using the wind direction and backward trajectories data.To demonstrate whether the reduced aerosol hygroscopicity in Hyytiälä is, indeed, a result of heightened organic fraction and emissions, atmospheric mass spectrometry data can be used, and the same applies to sulfate, nitrate and chloride.The representativeness of Hyytiälä in the whole global boreal zone can also be investigated once other longterm data sets obtained with similar measurement setups become available.In general, the results presented here provide a detailed insight into ambient aerosol CCN activation and hygroscopicity in a boreal environment and may serve as a valuable input into the current modelling of cloud droplet activation and aerosol-cloud interactions.

Fig. 2 .
Fig. 2. A sample CCN efficiency spectrum measured on 24 June 2009 from 20:02:03 to 20:23:13 UTC (Universal Time Coordinated)+2, and two ways of determining D c .Left panel: normalised method; right panel: non-normalised method.Both panels show the calculated D c , a root mean square error (RMSE) of the fit, a and the maximum activated fraction MAF.Note, that in the normalised method a and MAF were assigned values of 0.5 and 1, respectively; in the non-normalised method these values were allowed to vary.

Fig. 3 .
Fig. 3. Median N CN , N CCN and A per size bin for the whole data set for two levels of supersaturation S eff .Left panel: S eff = 0.1 %; right panel: S eff = 1.0 %.Error bars show 25th and 75th percentiles.

Fig. 4 .
Fig. 4. A contour plot frequency distribution of the critical diameters calculated by two methods.Each grid cell is 1 nm × 1 nm.Included in the figure is the 1:1 line (white line).

Fig. 5 .
Fig. 5. Relationship between particle dry size (taken as D c ) and critical supersaturation S eff .The black and grey lines correspond to the global continental mean κ of 0.27 ± 0.21 as reported by Pringle et al. (2010), and the error bars are 25th and 75th percentiles of median D c .

Fig. 6 .
Fig. 6.Relationship between particle dry size (taken as D c ) and κ.Left panel shows the comparison among sites; right panel shows comparison among seasons for Hyytiälä.Both panels show the median values with error bars being 25th and 75th percentiles (for Gunthe et al., 2009, percentiles estimated from the original publication).Legend entries also indicate the slope of the linear regression y = ax + b fit.

Fig. 7 .
Fig. 7. Distributions of κ values as a function of dry particle diameter.The different colours indicate different S eff levels, and the sides of the figure show relative occurrence of κ (vertical axis) and D c (horizontal axis) calculated with log-equal bins.The dashed and the adjacent lines show the 25th, 50th and 75th percentiles of κ for each S eff level.The gray dotted lines show the expected κ ∼ D −3 c

Fig. 8 .
Fig. 8. Monthly median D c (top) and κ (bottom), shown for two levels of supersaturation S eff .Error bars are 25th and 75th percentiles.

Fig. 9 .
Fig. 9. Hourly median D c (black) and κ (red), shown for S eff of 1.0 % and separated by seasons.Error bars are 25th and 75th percentiles.

Fig. 10 .
Fig. 10.Diurnal variation of median D c on 29 spring Type I event (red) and 53 spring non-event (black) days.Shown from 06:00 UTC+2 of the day in question until 05:00 of the following day, for S eff of 1.0 %.Error bars are 25th and 75th percentiles.

Table 1 .
Number of spectra, the results of the normalised and non-normalised fitting of Eq. (2), the related terms and the derived parameters of D c and κ used in this study.For parameters shown are the median values, with 25th and 75th percentile values included in the square brackets.Data shown separately for each supersaturation S eff level, as well as for the whole data set.