Cloud condensation nuclei in polluted air and biomass burning smoke near the mega-city Guangzhou , China – Part 1 : Size-resolved measurements and implications for the modeling of aerosol particle hygroscopicity and CCN activity

Cloud condensation nuclei in polluted air and biomass burning smoke near the mega-city Guangzhou, China – Part 1: Size-resolved measurements and implications for the modeling of aerosol particle hygroscopicity and CCN activity D. Rose, A. Nowak, P. Achtert, A. Wiedensohler, M. Hu, M. Shao, Y. Zhang, M. O. Andreae, and U. Pöschl Biogeochemistry Department, Max Planck Institute for Chemistry, Mainz, Germany Leibniz Institute for Tropospheric Research, Leipzig, Germany

pristine air.Nevertheless, the ratios between CCN concentration and total aerosol particle concentration (integral CCN efficiencies) were similar to the ratios observed in pristine continental air (∼6% to ∼85% at S=0.068% to 1.27%).Based on the measurement data, we have tested different model approaches for the approximation/prediction of N CCN,S .Depending on S and on the model approach, the relative deviations between observed and predicted N CCN,S ranged from a few percent to several hundred percent.The largest deviations occurred at low S with a simple power law.With a Köhler model using variable κ values obtained from individual CCN efficiency spectra, the relative deviations were on average less than ∼10% and hardly exceeded 20%, confirming the applicability of the κ-Köhler model approach for efficient description of the CCN activity of atmospheric aerosols.Note, however, that different types of κ-parameters must be distinguished for external mixtures of CCN-active and -inactive aerosol particles (κ a , κ t , κ cut ).Using a constant average hygroscopicity parameter (κ=0.3) and variable size distributions as measured, the deviations between observed and predicted CCN concentrations were on average less than 20%.In contrast, model calculations using variable hygroscopicity parameters as measured and constant size distributions led to much higher deviations: ∼70% for the campaign average size distribution, ∼80% for a generic rural size distribution, and ∼140% for a generic urban size distribution.These findings confirm earlier studies suggesting that aerosol particle number and size are the major predictors for the Published by Copernicus Publications on behalf of the European Geosciences Union.

Introduction
Atmospheric aerosol particles that enable the condensation of water vapor and formation of cloud droplets are called Cloud Condensation Nuclei (CCN).Elevated concentrations of CCN tend to increase the concentration and decrease the size of droplets in a cloud.Besides changing the optical properties and the radiative effects of clouds on climate, this may lead to the suppression of precipitation in shallow and short-lived clouds and to greater convective overturning and more precipitation in deep convective clouds (Rosenfeld et al., 2008).The response of cloud characteristics and precipitation processes to increasing anthropogenic aerosol concentrations represents one of the largest uncertainties in the current understanding of climate change.One of the crucial underlying challenges is to determine the ability of aerosol particles to act as CCN under relevant atmospheric conditions, an issue that has received increasing attention over the past years (McFiggans et al., 2006;IAPSAG, 2007;IPCC, 2007;Andreae and Rosenfeld, 2008, and references therein).
The Pearl River Delta (PRD) in southeastern China is one of the main centers of economic activity and growth in Asia.Due to strong anthropogenic emissions, the PRD region is often plagued with high aerosol concentrations that not only lead to low visibility, but can also impact the regional radiative balance, precipitation patterns and hydrological cycles (Hagler et al., 2006;Andreae et al., 2008;Fan et al., 2008;Wendisch et al., 2008;Zhang et al., 2008).
Within the "Program of Regional Integrated Experiments of Air Quality over the Pearl River Delta" intensive campaign in July 2006 (PRIDE-PRD2006), we measured and characterized the CCN properties of aerosol particles in polluted air and biomass burning smoke near the mega-city Guangzhou as a function of particle diameter (20-290 nm) and water vapor supersaturation (0.068-1.27%).In this manuscript, we focus on the results of the size-resolved CCN measurements and on the implications for different approaches of approximating and predicting CCN number concentrations.A follow-up study will address the relationships between aerosol chemical composition and CCN activity (Rose et al., 2010a).

Measurement location, meteorological conditions and supporting data
The measurements were performed during the period of 1-30 July 2006 in Backgarden (23.548056 • N, 113.066389 • E), a small village ∼60 km northwest of Guangzhou on the outskirts of the densely populated center of the PRD.Due to the prevailing southeast monsoon circulation at this time of year, the air masses came mainly from the south/southeast, making this site a rural receptor site for the regional pollution resulting from the outflow of the city cluster around Guangzhou.
A two-story building was used exclusively to house the measurement campaign, with most of the instruments placed in air conditioned rooms on the top floor and sample inlets mounted on the rooftop.The main aerosol inlet used in this study was equipped with a Rupprecht and Patashnick PM 10 inlet (flow rate 16.7 L min −1 ).The sample flow passed through stainless steel tubing (1.9 cm i.d., 5.1 m length) and a diffusion dryer with silica gel/molecular sieve cartridges (alternating regeneration with dry pressurized air, regeneration cycles 15-50 min, average RH=33±7%).After drying, the sample flow was split into separate lines.One led to the CCN measurement setup described below (0.9 cm i.d.stainless steel, ∼4 m length, flow rate 1.5 L min −1 ); another was used for aerosol particle size distribution measurements (3-900 nm) with a Twin Differential Mobility Particle Sizer (TDMPS).The inlet, dryer and size distribution measurements were operated by the Leibniz Institute for Tropospheric Research (IfT).
Besides aerosol particle size distribution and CCN activity, on which we focus in this manuscript, a wide range of other aerosol, gas phase, and meteorological parameters were measured to characterize local and regional air pollution (Garland et al., 2008;Hua et al., 2008a;Liu et al., 2008).These will be used in a follow-up study addressing the relationships between aerosol chemical composition and CCN activity (Rose et al., 2010a).

Instrumentation and measurement procedure
Size-resolved CCN efficiency spectra (CCN activation curves) were measured with a Droplet Measurement Technologies continuous flow CCN counter (Roberts and Nenes, 2005;Lance et al., 2006) coupled to a Differential Mobility Analyzer (DMA; TSI 3071) and a condensation particle counter (CPC; TSI 3762; Frank et al., 2006;Rose et al., 2008).
The CCN Counter (CCNC) was operated at a total flow rate of 0.5 L min −1 with a sheath-to-aerosol flow ratio of 10.For the campaign, the average sampling temperature and pressure as measured by the CCNC sensors were (23.7±1.4)• C and (1006±6) hPa (posititve deviation from ambient pressure due to measurement uncertainties; Rose et al., 2010b).The effective water vapor supersaturation (S) was regulated by the temperature difference between the upper and lower end of the CCNC flow column ( T ) and calibrated as described below and in Rose et al. (2008).
For each CCN measurement cycle, T was set to 5 different levels in the range of 1.98-16.9K corresponding to S values of 0.068% to 1.27%.For each T and S, respectively, the diameter of the dry aerosol particles selected by the DMA (D) was set to 9 different values in the range of 20-290 nm.At each D, the number concentration of total aerosol particles (condensation nuclei, CN), N CN , was measured with the CPC, and the number concentration of CCN, N CCN , was measured with the CCNC.The integration time for each measurement data point was 30 s, the recording of a CCN efficiency spectrum (N CCN /N CN vs. D) took ∼16 min (including 50 s adjustment time for each new particle size and 4 min for adjustment to the next supersaturation level), and the completion of a full measurement cycle comprising CCN efficiency spectra at 5 different supersaturation levels took ∼85 min (including 5 min for adjustment between the highest and lowest level of S).

Calibration of CCN counter
With respect to the effective water vapor supersaturation S, the CCNC was calibrated with ammonium sulfate aerosol as described by Rose et al. (2008).During the campaign, five calibration experiments were performed, and in each of these experiments multiple CCN efficiency spectra were recorded for 5 different T values.The midpoint activation diameter of each CCN efficiency spectrum was taken as the critical dry diameter for the CCN activation of ammonium sulfate particles, and the corresponding critical supersaturation was calculated with an activity parameterization Köhler model (AP3; Rose et al., 2008) that can be regarded as the most accurate reference available.Note that other frequently used Köhler models and the corresponding calibration lines would deviate by up to 20% or more, and care has to be taken when comparing the results of different CCN measurement and model studies (Rose et al., 2008).The calculated critical supersaturation was taken as the effective supersaturation at the given T value.
Figure 1 shows the average CCN efficiency spectra obtained from the 5 calibration experiments with ammonium sulfate aerosol, and the corresponding average calibration parameters are given in Table 1.A linear least-squares fit to the data pairs of S and T was taken as the CCNC calibration line for the entire campaign: S=k s T +S 0 with k s =0.08041%K −1 and S 0 =−0.09109%,R 2 =0.9929.It was applied to calculate S from the average value of T recorded during each measurement of a CCN efficiency spectrum of atmospheric aerosol.As detailed by Rose et al. (2008), variations in S are mostly due to variations of the CCNC inlet temperature.The standard deviations of the calibration data points and their maximum deviations from the calibration line ( S/S) as listed in

Correction of measured CCN efficiency spectra
The measured atmospheric CCN efficiency spectra were corrected for multiply charged particles as described by Frank et al. (2006) and for the DMA transfer function as described by Rose et al. (2008).For the multiple charge correction we used the total aerosol particle number size distributions measured in parallel with the TDMPS.For several days TDMPS data were not available, and no charge correction was performed.Nevertheless, the CCN data from these days remained comparable with the others, because the effects of the charge correction were generally small (<5% change in activation diameters and other parameters used for further analysis).
The CCN efficiency spectra were also corrected for the differences in the counting efficiencies of the CCNC and the CPC.If the CCNC and CPC counting efficiencies were the same, a maximum activated fraction of N CCN /N CN ≈1 would be expected for ammonium sulfate calibration aerosol particles at all supersaturation levels.As illustrated in Fig. 1, however, the measured maximum value of N CCN /N CN was close to one only for larger particles.For smaller particles the measured maximum levels of N CCN /N CN decreased with decreasing particle size, which can be attributed to a decrease in the counting efficiency of the CCNC (most likely due to wall losses in the tubing inside the instrument).To correct for this bias, we have fitted an asymptotic function to those data points of the calibration efficiency spectra that reached at least 95% of their respective maximum values (red line in Fig. 1): 3 with x 1 =1.00547, x 2 =0.26208, x 3 =0.98024,R 2 =0.70881.The inverse of this correction function was multiplied with all the atmospheric N CCN /N CN data points after the charge and before the transfer function corrections.In the preceding discussion paper, the counting efficiency correction was performed after the charge and transfer function correction.On average the change in the sequence of corrections had little effect on the results (deviations <1% for D a and MAF, ∼4% for σ ), but the data analysis with the counting efficiency before the transfer function correction was more robust (fewer cases where CDF fits did not converge).In the following, for simplicity, the corrected CCN efficiency spectra are referred to as the "measured" CCN efficiency spectra.Note that all CCN efficiency spectra presented in the figures of this paper show the corrected ones.
The scattering of the highest N CCN /N CN measurement values around the counting efficiency correction function determined in the calibration experiments (f corr , Fig. 1) indicate a relative uncertainty of ∼5% for the CCN efficiencies determined for atmospheric aerosols (corrected CCN efficiency spectra).For the period after 20 July the relative uncertainty increased to ∼10%, as indicated by a decrease in the observed maximum CCN efficiencies (offset in the CCNC flow rate).

Basic spectral parameters
The measured CCN efficiency spectra were fitted with a cumulative Gaussian distribution function (CDF; Rose et al., 2008): The following best-fit parameters were determined for each spectrum: the maximum activated fraction MAF f =2a, the midpoint activation diameter D a , and the CDF standard deviation σ a .In addition to the 3-parameter CDF fits with varying a, D a , and σ a , we have also performed 2-parameter CDF fits which were forced to MAF f =1 by fixing the parameter a at 0.5 and varying only D a and σ a .For the midpoint activation diameters and CDF standard deviations obtained from these fits we use the symbols D t and σ t .In addition to the above CDF fit-based parameters, the CCN efficiency measured at the largest diameter of each spectrum (D max ) was also taken for further analysis and discussion: Characteristic examples of atmospheric CCN efficiency spectra of atmospheric aerosols and the corresponding CDF fits and parameters are illustrated in Fig. 2. Figure 2a shows an "ideal" spectrum that is characteristic for internally mixed aerosols with homogeneous composition and hygroscopicity of the particles (similar to ammonium sulfate calibration aerosol).In this case, the observed CCN efficiencies reach up to one (MAF f ≈MAF m ≈1), and the activation diameters and standard deviations derived from the 3-parameter and 2parameter CDF fits are essentially the same (D a ≈D t ; σ a ≈σ t ).
At medium and high supersaturation (S=0.47-1.27%),most CCN efficiency spectra were qualitatively similar to the one in Fig. 2a.At low supersaturation (S=0.068-0.27%),however, most CCN efficiency spectra deviated from the ideal shape of a completely internally mixed aerosol and looked like the exemplary spectrum displayed in Fig. 2b, which is characteristic for externally mixed aerosols.
In these cases, the highest observed CCN efficiencies remain well below one (MAF<1), which indicates an external mixture of CCN-active particles with CCN-inactive particles, whereby the difference in CCN activity is due to chemical composition and hygroscopicity (not particle size).Test experiments with different CCNC flow rates yielded the same result, indicating that the observed deviations of MAF from unity were not governed by potential kinetic limitations of water uptake in the CCNC.
For CCN efficiency spectra with MAF<1 the activation diameters and standard deviations derived from the 3parameter and 2-parameter CDF fits are not the same: the 3-parameter fit results represent the average properties of the CCN active aerosol particle fraction, whereas the 2parameter fit results approximate the overall properties of the external mixture of CCN-active and CCN-inactive particles.
The difference between unity and the maximum observed CCN efficiency (1-MAF m or 1-MAF f , respectively) represents the fraction of externally mixed CCN-inactive particles at D max or averaged over the diameter range of D a to D max , respectively.The CDF standard deviations are general indicators for the extent of external mixing and heterogeneity of particle composition in the investigated aerosol: σ a characterizes the CCN-active particles in the size range around D a , and σ t characterizes the overall heterogeneity of CCNactive and -inactive particles in the size range around D t .Under ideal conditions, the CDF standard deviations should be zero for an internally mixed, fully monodisperse aerosol with particles of homogeneous chemical composition.Even after correcting for the DMA transfer function, however, calibration aerosols composed of high-purity ammonium sulfate exhibit small non-zero σ a values that correspond to ∼3% of D a and can be attributed to heterogeneities of the water vapor supersaturation profile in the CCNC or other non-idealities such as DMA transfer function and particle shape effects (Rose et al., 2008).Thus, normalized CDF standard deviations or "heterogeneity parameter" values of σ a /D a ≈3% indicate internally mixed CCN, whereas higher values indicate external mixtures of particles with varying chemical composition and hygroscopicity.
In addition to the activation diameters derived from the CCN efficiency spectra, we have also determined an apparent cut-off diameter of CCN activation D cut .It is the diameter above which the integral CN number concentration equals the observed CCN concentration (N CCN,S as detailed below).Note that unlike D a and D t , the determination of D cut requires knowledge of the CN size distribution and the assumption of a sharp cut-off (corresponding to σ t =0).A list of frequently used symbols is given in Table A1 at the end of the manuscript.

Effective hygroscopicity parameters
As proposed by Petters and Kreidenweis (2007), an effective hygroscopicity parameter κ can be used to describe the influence of chemical composition on the CCN activity of aerosol particles, i.e. on their ability to absorb water vapor and act as CCN.Based on Köhler theory, κ relates the dry diameter of aerosol particles to the critical water vapor supersaturation, i.e. the minimum supersaturation required for cloud droplet formation.For a given supersaturation, D. Rose et al.: CCN in polluted air and biomass burning smoke -Part 1 κ allows calculating the critical dry particle diameter, i.e. the minimum diameter required for the particle to be CCN-active.According to measurements and thermodynamic models, κ is zero for insoluble materials like soot or mineral dust, ∼0.1 for secondary organic aerosols, ∼0.6 for ammonium sulfate and nitrate, ∼0.95-1 for sea salt (obtained from ρ ion values in Niedermeier et al., 2008), and 1.28 for sodium chloride aerosols.The effective hygroscopicity of mixed aerosols can be approximated by a linear combination of the κ-values of the individual chemical components weighted by the volume or mass fractions, respectively (Kreidenweis et al., 2008;Gunthe et al., 2009).On average, continental and marine aerosols tend to cluster around κ=0.3 and κ=0.7, respectively (Andreae and Rosenfeld, 2008;Kreidenweis et al., 2009;Pöschl et al., 2009).Laboratory experiments with biomass burning aerosols yielded κ values ranging from 0.02 to 0.8 (Petters et al., 2009).For all data pairs of supersaturation and activation diameter derived from the CCN efficiency spectra measured in this study, κ parameters were calculated from the following Köhler model equation (equivalent to Eq. 6 of Petters andKreidenweis, 2007, andEq. A30 of Rose et al., 2008): For the temperature we inserted T =298.15K, the droplet surface tension was approximated by that of water (σ sol =0.072 J m −2 ), and the other parameters were set to R=8.315 J K −1 mol −1 (gas constant), ρ w =997.1 kg m −3 and M w =0.018015 kg mol −1 (density and molar mass of water).Note that κ values derived from CCN measurement data through Köhler model calculations assuming the surface tension of pure water have to be regarded as "effective hygroscopicity parameters" that account not only for the reduction of water activity by the solute ("effective Raoult parameters") but also for surface tension effects.For more information see Petters and Kreidenweis (2007); Gunthe et al. (2009); Mikhailov et al. (2009); Pöschl et al. (2009).The parameter κ a calculated from the data pairs of S and D a characterizes the average hygroscopicity of CCN-active particles in the size range around D a .κ t calculated from D t is an approximate measure (proxy) for the effective hygroscopicity of CCN-active and -inactive particles in the size range around D t .Accordingly, κ t is better suited for comparison with average κ values calculated from H-TDMA data and for the calculation of CCN number concentrations when CCN-active particles are externally mixed with CCN-inactive particles.On the other hand, κ a is better suited for comparison with κ values predicted from AMS measurements, because κ a is not influenced by CCN-inactive particles consisting mostly of insoluble and refractory materials like mineral dust and soot (or biopolymers that tend to char upon heating), which are not (or less efficiently) detected by AMS.The parameter κ cut calculated from the data pairs of S and D cut characterizes the effective average hygroscopicity of CCN-active particles in the size range above D cut .D cut and κ cut can also be determined from the results of integrated CCN concentration measurements of polydisperse aerosols, and may thus be useful for comparison with studies lacking size-resolved CCN data.

CCN size distributions and number concentrations
CCN size distributions (dN CCN /dlogD) were calculated by multiplying the CCN efficiency spectra (3-parameter CDF fits of N CCN /N CN ) with the total aerosol particle (CN) number size distributions measured in parallel (dN CN /dlogD).In these calculations, the fit parameter a was limited to a maximum value of 0.5 (MAF f =1), because CCN concentrations exceeding CN concentrations are physically not realistic.Additional test calculations with a not limited to 0.5 led to total CCN concentrations that were on average less than 1% higher.
Near the activation diameter, the size resolution of the CCN efficiency spectra was generally higher than that of the CN size distribution measurement data from the TDMPS (dlogD=0.083).Thus the CN size distributions were linearly interpolated on a grid with ten-fold smaller size steps.
Total CCN concentrations (N CCN,S ) were calculated by stepwise integration of the CCN size distributions with dlogD=0.0083from 3 to 900 nm.Note that insufficient sizeresolution near the activation diameter can lead to substantial deviations in the calculation of total CCN number concentrations (up to ∼10% at low S, up to ∼5% at high S with dlogD=0.083 vs. dlogD=0.0083in this study).

CCN efficiency spectra and related parameters
During the 30 day campaign period of PRIDE-PRD2006, we measured ∼2200 size-resolved CCN efficiency spectra (activation curves) for atmospheric aerosols at water vapor supersaturations in the range of 0.068% to 1.27%.Exemplary spectra are shown in Fig. 2, and the derivation and interpretation of characteristic parameters is explained in Sect.2.2.4.Fig. 3. CCN efficiency spectra at S=0.068%-1.27%averaged over the entire campaign.The data points are median values calculated from the CDF fits to all measured spectra at the particle diameters initially selected with the DMA (20-290 nm).The error bars extend from the lower to the upper quartile, and the lines are 3-parameter CDF fits to the data points (Sect.2.2.4).

Campaign averages
Figure 3 shows campaign averages of the atmospheric CCN efficiency spectra at the six investigated supersaturation levels.The average parameters derived from the CCN efficiency spectra are summarized in Table 2.
As expected, the midpoint activation diameters D a increased with S and were larger than the critical dry diameters for CCN activation of pure ammonium sulfate particles at the same supersaturation levels.At medium and high supersaturation (S=0.47-1.27%), the CCN efficiency spectra generally reached up to one (MAF f ≈1) and the relative standard deviations of the 3-parameter CDF fits were small (σ a /D a ≈10%), which implies that nearly all aerosol particles larger than the midpoint activation diameter (D>D a ) were CCN-active.At low supersaturation (S=0.068-0.27%),however, the maximum activated fractions remained on average well below one, which indicates a substantial portion (1−MAF f ) of externally mixed CCN-inactive particles with much lower hygroscopicity.At S=0.068%, the average MAF f was only ∼0.75 with minimum values as low as ∼0.4,i.e. even at diameters as large as ∼250 nm an average of ∼25% and up to ∼60% of the aerosol particles were not CCN-active.To our knowledge such high portions of externally mixed CCN-inactive particles have not been observed before in atmospheric aerosols.
Sensitivity tests with the κ-Köhler model described in Sect.2.2.4 (Petters and Kreidenweis, 2007) indicate that particles as large as ∼300 nm must have an effective hygroscopicity parameter κ<0.1 to not be activated at S=0.068%.On the other hand, S≈0.7% would be required to activate 300 nm particles that are wettable but completely insoluble and non-hygroscopic (κ=0).Most likely the CCN-inactive particles were freshly emitted (non-aged/non-coated) soot particles with κ≈0.01, which will be discussed further in a follow-up study based on Volatility Tandem DMA (VT-DMA) and chemical composition data (Rose et al., 2010a).
Other recent studies from PRIDE-PRD2006 (Garland et al., 2008) and a similar field campaign in the vicinity of Beijing (Cheng et al., 2009;Garland et al., 2009;Wehner et al., 2009;Wiedensohler et al., 2009) also indicate strong regional pollution with large proportions of externally mixed soot particles in the atmospheric aerosol near Chinese mega-cities and city-clusters.
www.atmos-chem-phys.net/10/3365/2010/Atmos.Chem.Phys., 10, 3365-3383, 2010 Figure 4a gives an overview of the maximum activated fractions (MAF f ) and normalized standard deviations (σ a /D a ) of the 3-parameter CDF fits as well as the normalized standard deviations of the 2-parameter CDF fits (σ t /D t ) to the measured CCN efficiency spectra.The average parameter values are plotted against the corresponding average midpoint activation diameters (D a , D t ) that have been observed at the six prescribed levels of water vapor supersaturation (S=0.068-1.27%).
As detailed in Sect.2.2.4,σ a /D a characterizes the heterogeneity of CCN-active particles in the size range around D a , whereas σ t /D t characterizes the overall heterogeneity of aerosol particles in the size range around D t .
For small particles in the nucleation or Aitken size range (∼30-70 nm), the heterogeneity parameters σ a /D a and σ t /D t were nearly identical and close to ∼10%.This is clearly higher than the ∼3% observed for aerosols of homogeneous chemical composition (e.g.pure ammonium sulfate), indicating that the particles in this size range were not fully internally mixed with respect to their solute content.For larger particles in the accumulation size range (∼70-200 nm), σ a /D a remained at ∼10% whereas σ t /D t increased strongly up to ∼25% at ∼200 nm.This confirms that the CCN-active particles in the accumulation size range had fairly homogeneous properties but were externally mixed with CCN-inactive particles.
Figure 4b gives an overview of the effective hygroscopicity parameters (κ a , κ t ) that have been derived from the midpoint activation diameters (D a , D t ) of the 3-parameter and 2parameter CDF fits, respectively.As detailed in Sect.2.2.4,κ a calculated from S and D a characterizes the average hygroscopicity of CCN-active particles in the size range around D a , whereas κ t calculated from S and D t is a proxy for the effective hygroscopicity of mixtures of CCN-active and -inactive particles in the size range around D t .
For small particles in the nucleation or Aitken size range (∼30-70 nm), κ a and κ t were nearly identical and close to ∼0.3.For larger particles in the accumulation size range (∼70-200 nm), however, κ a increased substantially to ∼0.4-0.5, whereas κ t increased only slightly to ∼0.33.
Overall, larger particles were on average more hygroscopic but also more heterogeneous than smaller particles.The observed values of κ a , κ t , σ a /D a , σ t /D t , and MAF f suggest that the particles in the nucleation or Aitken size range were composed mostly of organics and sulfate and were largely but not fully internally mixed, whereas the particles in the accumulation size range consisted mostly of an external mixture of soot particles (κ<0.1;∼25% at ∼200 nm) and sulfate-rich particles (κ≈0.4-0.5;∼75% at ∼200 nm).Note that the properties of large accumulation mode particles are not only important for cloud formation at low and medium supersaturation (low and moderate updraft velocities; Segal and Khain, 2006;Reutter et al., 2009) but also for aerosol optical properties and direct radiative effects on climate (Garland et al., 2008).These and other aspects of aerosol chemical composition and mixing state will be explored further and discussed in more detail in a followup study (Rose et al., 2010a).Averaged over all diameters, the mean hygroscopicity parameter values for the entire campaign were κ a =0.34 and κ t =0.30 (Table 2).The hygroscopicity parameter related to the cut-off diameter D cut (κ cut ) was on average 10% smaller than κ t and had an arithmetic mean value of 0.26.

Time series and biomass burning event
Figure 5 shows the time series of characteristic parameters (D a , D t , κ a , κ t , σ a , σ t , σ a /D a , σ t /D t , MAF f , MAF m ) derived from the atmospheric CCN efficiency spectra measured throughout the campaign.For clarity, the parameters in Fig. 5c-j are shown only for the smallest and largest supersaturations that were measured during the entire campaign (S=0.068%,S=0.87%).The temporal evolution of most parameters at S=0.068% was qualitatively similar to S=0.27%, and that at S=0.87% was representative for S=0.47-1.27%.
Most parameters exhibited pronounced diurnal cycles which are consistent with the results of other recent studies from PRIDE-PRD2006 (Garland et al., 2008;Hua et al., 2008b).The diurnal cycles in CCN properties will be described and discussed together with the variability of other aerosol properties including chemical composition, volatility, and optical parameters in a follow up study (Rose et al., 2010a).As illustrated in Fig. 5i and j, both the fitted and the measured maximum activated fractions (MAF f , MAF m ) dropped by ∼10% after 20 July, which is most likely due to a measurement artifact (offset in the flow rate of the CCNC).
In addition to the diurnal variability, several CCN parameters exhibited pronounced changes during high pollution events.Especially on 23-26 July, the midpoint activation diameters and standard deviations of the CDF fits increased and the hygroscopicity parameters decreased relative to the campaign average (Fig. 5, panels a-h).The changes indicate an increase in the portion of particulate matter with low hygroscopicity (organic substances) and in the heterogeneity of particles (external mixing), and they were most pronounced for small particles (∼30-80 nm; S≥0.27%).
The highly polluted period of 23-26 July 2006 was characterized by intense local biomass burning and very high aerosol mass concentrations (Garland et al., 2008).During this period, the source of the pollution was evident and unique: the burning of plant waste by local farmers was visible in the vicinity surrounding the measurement site, and it was the only time that such intense local biomass burning and pollution occurred during the campaign.The heavy biomass burning event started after a power outage in the evening of 22 July and ended by heavy rainfalls beginning at ∼13:00 on 26 July.Thus the period of 23 July, 00:00-26 July, 12:59 will be referred to as the "Biomass Burning Event (BBE)".The average CCN parameters for this period are summarized in  During the BBE, the CCN efficiency spectra were shifted towards larger particle sizes for all supersaturations, reflecting lower CCN activity than during the rest of the campaign.The increase of D a was most pronounced for S=0.67% (+22% during the BBE) and least different for S=0.068% and S=1.27% (+8% during the BBE).Moreover, for all supersaturations, except 0.068%, the standard deviations of the CDF fits and heterogeneity parameters (σ/D), increased by factors up to ∼2, indicating a strong increase in the heterogeneity of small particles (30-100 nm, Table 2).Only the maximum activated fractions did not change significantly during the biomass burning event, i.e. the externally mixed fraction of particles that could not be activated at low S remained the same.
Figure 6 gives an overview of the effective hygroscopicity parameters (κ a , κ t ) that have been derived from the midpoint activation diameters (D a , D t ) averaged over the entire campaign and over the biomass burning event.Figure 6a shows that during the BBE the average hygroscopicity of CCN-active particles was substantially reduced at all sizes.Averaged over all diameters, the mean value of κ a during the BBE was ∼30% lower than during the rest of the campaign: 0.24 vs. 0.34 (Table 2).
As illustrated in Fig. 6b, the average hygroscopicity of the total aerosol, including CCN-active and -inactive particles, was also strongly reduced for small particles (<100 nm) but not so much for large particles (∼200 nm).Averaged over all diameters, however, the mean value of κ t during the BBE was also ∼30% lower than during the rest of the campaign: 0.21 vs. 0.30 (Table 2).
To our knowledge, these are the first size-resolved CCN field measurement data and hygroscopicity parameters reported for freshly emitted biomass burning smoke in the atmosphere.They are consistent with earlier lab studies reporting low hygroscopicity of freshly emitted biomass burning particles (Rissler et al., 2006;Andreae and Rosenfeld, 2008;Petters et al., 2009).

CCN size distributions and number concentrations
Figure 7 shows total aerosol particle (CN) and CCN number size distributions averaged over the entire campaign and over the biomass burning event, respectively.The corresponding averages of the total CCN number concentrations (N CCN,S ) and of the total CCN efficiencies (N CCN,S /N CN,tot ) are summarized in Table 2.As illustrated in Fig. 7a, the average CN size distribution for the entire campaign was monomodal with a maximum at ∼70 nm, and the corresponding total particle number concentration was N CN,tot =1.8×10 4 cm −3 (Table 2).At S=0.068% the CCN size distribution accounted only for ∼5% of N CN,tot , because only a minor fraction of the CN were larger than the activation diameter (∼200 nm).At S=0.27-0.87% the CCN activation diameters were close to the maximum of the CN number mode and the integral CCN efficiencies were substantially higher (N CCN,S /N CN,tot =36-71%; Table 2).During the biomass burning event (Fig. 7b), the CN size distribution was broader and the maximum was shifted to larger sizes (∼120 nm, a value typical of biomass smoke; Reid et al., 2005).The average number concentration of CN was slightly smaller (∼1.5×10 4 cm −3 ), but due to the larger average particle sizes the CCN number concentrations at S=0.068% and 0.27% were higher (+90% and +8%, respectively).For S≥0.47%, however, N CCN,S was lower compared to the rest of the campaign (Table 2).The geometric mean diameters (D g ), standard deviations (σ g ), and integral number concentrations (N g ) of monomodal lognormal fits to the median CN size distributions were D g =70 nm, σ g =1.94, and N g =14 600 cm campaign, and D g =107 nm, σ g =2.09, and N g =14 200 cm −3 for the biomass burning event.

Prediction of CCN number concentration
In this section we compare different model approaches for the approximation/prediction of CCN concentration as a function of water vapor supersaturation, aerosol particle number concentration, size distribution and hygroscopicity: (1) the classical power law approach relating N CCN,S to N CCN,1 , i.e. to the CCN concentration at S=1%; (2) a modified power law approach relating N CCN,S to the concentration of aerosol particles with D>30 nm (N CN,30 ); and (3) the κ-Köhler model approach relating N CCN,S to the aerosol particle size distribution (dN CN /dlogD) and effective hygroscopicity.For all data points obtained during the campaign, the model results were compared with the observed values, and the mean values of the relative deviations are summarized in Table 3.

Classical power law
Figure 8 shows the campaign median values of N CCN,S plotted against S and a power law fit of the function N CCN,S =N CCN,1 •(S/(1%)) k (Pruppacher and Klett, 1997;Origin 6.1G software, Levenberg-Marquardt algorithm).The obtained fit parameter N CCN,1 ≈1.4×10 4 cm −3 is substantially higher than any previously reported value, and k≈0.65 is within the range of values reported for other continental locations (0.4-0.9; Pruppacher and Klett, 1997;Andreae, 2009).The mean relative deviations of the individual measurement data points from the average power law were in the , including systematic and statistical errors).CPL is the classical power law and MPL the modified power law approach, respectively.SD m is the campaign average CN size distribution (Fig. 7a), and SD u and SD r are the generic urban and rural size distributions as listed in Seinfeld and Pandis (2006, Table 8.3)

Table 4.
Fit parameter Q and q of the fit functions N CCN,S =N CN,30 •(S/(1%)) Q and N CCN,S =N CN,30 •s −q , respectively.The correlation coefficient R 2 is the same for both fits.n SD is the number of data points.range of 40-90% for S=0.27-1.27%,but as high as 310% for S=0.068% (Table 3).

Modified power law
Figure 9 shows all observed values of N CCN,S plotted against N CN,30 and power law fits of the form N CCN,S =N CN,30 •s −q with s=1+S/(100%) (Pruppacher and Klett, 1997; Origin 6.1G software, Levenberg-Marquardt algorithm).An overview of the best fit values obtained for the exponent q and the corresponding correlation coefficients is given in Table 4.In this approach, CN with D<30 nm were excluded, because they are generally not CCN-active and highly variable due to new particle formation (nucleation events).Moreover, the water vapor saturation ratio s was used instead of the supersaturation S, because the exponent varies more regularly with s than with S (Table 4: monotonous dependence of q on s vs. non-monotonous dependence of Q on S).At high supersaturations (S≥0.47%),N CCN,S was fairly well correlated to N CN,30 (R 2 =0.69-0.95), and the mean relative deviations between the power law fit and the observed values of N CCN,S were less than 40% ).The lines are modified power law fits of the function N CCN,S =N CN,30 •s −q with the parameter q as given in Table 4.
(Table 3).At S=0.27% the correlation was much worse (R 2 =0.49, mean deviation 54%), and at S=0.068% there was practically no correlation and the individual N CCN,S data points deviated by up to one order of magnitude from the power law fit (R 2 =0.04, mean deviation 64%).

κ-Köhler model with variable CN size distribution
In Fig.As illustrated in Fig. 10a, the predicted and observed values of N CCN,S are in very good agreement, when for each data point κ t was taken from the CCN efficiency spectrum measured in parallel to the CN size distribution.With this approach, the mean relative deviation averaged over all supersaturations was only 6%, and the overall mean bias of the model values was +4% (Table 3).The agreement demonstrates that κ t is a suitable proxy for the effective hygroscopicity and CCN activity of the investigated ensemble of aerosol particles including CCN-active and -inactive particles.The deviations and the bias increased with lower S, which can be explained by the decreasing MAF, indicating that the effect of external mixing between CCN-active and -inactive particles is not fully captured by κ t .Nevertheless, the results obtained with κ t were clearly better than with κ a , which represents the effective hygroscopicity of the CCNactive particles only and thus leads to higher mean deviation and bias (15%, +14%, Table 3).Only at S=0.47-1.27%,where MAF f ≈1 and κ a ≈κ t , were the results obtained with individual κ a values nearly the same as with individual κ t , and thus the overall mean bias and deviation were significantly higher (+14%, 15%, Table 3).
Fair agreement was also achieved when the campaign average value of κ t =0.30 and the corresponding constant activation diameters for the prescribed supersaturation levels (215, 86, 59, 47, 39, and 30 nm; S=0.068-1.27%)were used for the calculation of N CCN,S from the individual measured CN size distributions.With this approach, the overall mean relative deviation was more than twice as high, but the bias was hardly higher than when individual κ t values were used (14% and +7.2%, respectively; Table 3; Fig. 10b).Note that the campaign average value of κ t equals the average value of hygroscopicity parameters observed or inferred for other continental locations (Andreae and Rosenfeld, 2008;Pöschl et al., 2009;and references therein) in agreement with global model simulations (Tsigaridis et al., 2006;Kreidenweis et al., 2009).
The approach using a constant average value of κ=0.3 cannot fully account for the observed temporal variations in aerosol composition and CCN properties.It yields relative deviations in the range of −40% to +80% of N CCN,S .Under most circumstances, however, i.e. for 76% of all data points, the deviations were still less than ±20%, which appears quite reasonable for data that span more than two orders of magnitude.Even during the BBE, which was characterized by ∼30% lower hygroscopicity parameters (κ t ≈0.2; Sect.3.1.2),the average deviation between predicted and observed N CCN,S was only 21%.
With average or individual values of κ a , the positive bias was higher than with κ t (6-10%, Table 3).The approach using individual values of κ cut yielded the same concentrations as observed because N CCN,S,p was calculated from κ cut in the reverse way as κ cut from N CCN,S .With a constant average value of κ cut =0.26, the relative deviations were essentially the same as with constant κ t .

κ-Köhler model with constant CN size distribution
To test the relative importance of aerosol particle size distribution and hygroscopicity for the variability of N CCN,S , we performed additional Köhler model calculations in which the effective hygroscopicity parameters were allowed to vary (variable κ t from the fitting of measured CCN efficiency spectra) while the CN size distribution was kept constant.In analogy to Gunthe et al. (2009), we used the campaign average size distribution (Fig. 7a) and a generic urban and rural size distribution as listed in Seinfeld andPandis (2006, Table 8.3, based on Jaenicke, 1993) with three lognormal modes, respectively.
With the constant campaign average size distribution (Fig. 7a) and variable κ t , the mean relative deviation between observed and predicted CCN concentrations (∼70%) was higher by a factor of ∼5 than with the constant campaign average value of κ t =0.3 and variable size distribution (∼15%, Table 3).Assuming a constant generic size distribution for urban areas with variable κ t , the mean relative deviation between observed and predicted CCN concentrations (∼140%) was higher by a factor of 10 than under the assumption of an average value of κ t =0.3 with variable size distribution (14%, Table 3).Assuming a constant generic size distribution for rural areas with variable κ t , the mean relative deviation between observed and predicted CCN concentrations (∼80%) were only a little higher than when assuming the campaign average size distributions, but lead to an under-prediction of N CCN,S .
This demonstrates that the variability of CCN concentrations is much more strongly influenced by the variability of aerosol particle number concentration and size distribution than by the variability of aerosol composition and hygroscopicity.This applies for the temporal variations during the PRD 2006 campaign (factor of ∼5 between col.11 and 17, Table 3) as well as for spatial/geographic variations between the Pearl River Delta and other urban regions (factor of 2 between col.17 and 19, Table 3).

Summary and conclusions
The dry CCN activation diameters measured during PRIDE-PRD2006 at S=0.068-1.27%were in the range of 200-30 nm and the effective hygroscopicity parameters varied in the range of 0.1-0.5.The mean value of κ t characterizing the hygroscopicity of all aerosol particles averaged over the whole campaign and investigated size range was 0.3, which equals the average value of κ observed or inferred for other continental locations (Andreae and Rosenfeld, 2008;Pöschl et al., 2009).Particles in the Aitken size range (D≈30-70 nm, κ a ≈κ t ≈0.28) were on average less hygroscopic than particles in the accumulation size range (D≈70-200 nm, κ t ≈0.33, κ a ≈0.41).
During a strong local Biomass Burning Event (BBE) the aerosol particles were generally less CCN active (κ t ≈0.2).Due to the very intense local sources and high level of pollution, the κ values observed during the BBE can be regarded as characteristic for freshly emitted smoke from the open burning of agricultural waste.
At low S(≤0.27%), the maximum activated fraction remained generally well below one, which indicates substantial portions of externally mixed CCN-inactive particles with much lower hygroscopicity (κ≈0.01)-most likely soot particles.At S=0.068%, the average MAF f was only ∼0.75 with minimum values as low as ∼0.4,i.e. even at diameters as large as ∼250 nm an average of ∼25% up to ∼60% of the aerosol particles were not CCN-active.To our knowledge such high portions of externally mixed CCNinactive particles have not been observed before in atmospheric aerosols.Note, however, that these CCN-inactive particles contributed only around ∼3% to the total aerosol particle number concentration.The integral CCN efficiencies at moderate supersaturations (N CCN,S /N CN,tot ≈0.36-0.53 at S=0.27%-0.47%,Table 2) were even slightly higher than the global average value reported by Andreae (2009) (N CCN,S /N CN,tot ≈0.36 at S=0.4%).
From the measured CCN efficiency spectra and total aerosol particle (CN) size distributions, we derived CCN size distributions and total CCN number concentrations (N CCN,S ).On average, N CCN,S ranged from 1000 cm −3 at S=0.068% to 16 000 cm −3 at S=1.27%, representing ∼6% to ∼85% of the total aerosol particle number concentration (N CN,tot ).During the biomass burning event, the CN size distribution was broader and the maximum was shifted to larger sizes (from ∼70 nm to ∼120 nm).The average number concentration of CN was slightly smaller (∼1.5×10 4 cm −3 vs. ∼1.8×10 4 cm −3 ), but due to the larger average particle sizes the CCN concentrations at low supersaturation were substantially higher (+100% at S=0.068%, +10% at S=0.27%).For high supersaturations (S≥0.47%),however, N CCN,S decreased by up to ∼25% compared to the rest of the campaign.
Based on the measurement data, we have tested different model approaches (power laws and κ-Köhler models) for the approximation/prediction of N CCN,S as a function of water vapor supersaturation, aerosol particle number concentration, size distribution and hygroscopicity.Depending on S and on the applied type of power law or hygroscopicity parameter, the relative deviations between observed and predicted N CCN,S can range from a few percent to several hundred percent.The largest deviations occurred at low S with power laws based on particle number concentration without size information.Much better predictions could be made when using measured aerosol size distributions in combination with κ-Köhler models.
With variable κ values obtained from individual CCN efficiency spectra, the relative deviations between observed and predicted N CCN,S were on average less than ∼10% and rarely exceeded 20%.These results confirm the applicability of the κ-Köhler model approach for efficient description of the CCN properties of atmospheric aerosols.Note, however, that in the case of externally mixed CCN-active and -inactive aerosol particles, the use of κ parameters derived from different types of fits to the measured CCN efficiency spectra (2-or 3-parameter CDF) can lead to substantially different results -especially at low S (increase of deviations by up to a factor of ∼4).
Assuming a constant average value of κ=0.3, the deviations were on average still less than ∼20%, which confirms that κ=0.3 may be suitable as a first-order approximation for the effective hygroscopicity and CCN activity of continental aerosols.Model calculations assuming constant particle number concentrations and size distributions led to substantially larger deviations (∼70% to ∼140%).
These findings confirm earlier studies suggesting that aerosol particle number and size are the major predictors for the variability of the CCN concentration in continental boundary layer air, followed by particle composition and hygroscopicity as relatively minor modulators (Gunthe et al., 2009;Pöschl et al., 2009;and references therein).Thus the influence of aerosol chemical composition and hygroscopicity appears to be less variable and less uncertain than other factors that determine the effects of aerosols on warm cloud formation in the atmosphere (e.g., particle number concentration, size distribution, sources, sinks, and meteorological conditions).Depending on the required and applicable level of detail, the information and parameterizations presented in this study should enable efficient description of the CCN activity of atmospheric aerosols in detailed process models as well as in large-scale atmospheric and climate models (Heintzenberg and Charlson, 2009).

κ
was determined by inserting the observed activation diameter (D a , D t , or D cut ) for D and varying both κ and the droplet diameter D wet until the saturation ratio s was equivalent at the same time to the prescribed supersaturation S and to the maximum of a Köhler model curve of CCN activation (numerical minimum search for −s and for |s−(1+S/100%)| with Matlab "fminsearch" and start values of κ=0.2 and D wet =D).

Fig. 4 .
Fig. 4. Characteristic parameters derived from the CCN efficiency spectra averaged over the entire campaign: (a) maximum activated fractions (MAF f ) and heterogeneity parameters (σ a /D a , σ t /D t ); (b) hygroscopicity parameters (κ a , κ t ) plotted against the midpoint activation diameter (D a or D t , respectively).The data points are median values corresponding to a given level of supersaturation, and the error bars extend to lower and upper quartiles.

Fig. 5 .
Fig. 5. Time series of the characteristic parameters derived from the CCN efficiency spectra measured at different supersaturations plotted against the date in July 2006.

Fig. 6 .
Fig.6.Hygroscopicity parameters (a) for the CCN-active particles (κ a ) and (b) for the total aerosol (κ t ) averaged over different periods: the entire campaign, the Biomass Burning Event (BBE) and the campaign excluding the biomass burning event.The data points are median values corresponding to a given level of supersaturation, and the error bars extend to lower and upper quartiles.

Fig. 7 .
Fig. 7. Number size distributions of total aerosol particles (CN) and Cloud Condensation Nuclei (CCN) averaged over (a) the entire campaign and (b) the biomass burning event.The CCN size distributions were calculated by multiplying the median CN size distribution with the median CCN efficiency spectra from Fig. 3.For clarity and to avoid potential biases due to different averaging times, CCN size distributions are displayed only for the supersaturation levels covered throughout the campaign.

Fig. 8 .
Fig. 8. CCN number concentrations (N CCN,S ) averaged over the entire campaign and plotted against water vapor supersaturation (S).The data points are median values, and error bars extend to lower and upper quartiles.The red line is a classical power law fit of the function N CCN,S =N CCN,1 •(S/(1%)) k with the best fit parameters N CCN,1 =13 699 cm −3 and k=0.65 (R 2 =0.97, n=6).

Fig. 10 .
Fig. 10.Predicted CCN number concentrations (N CCN,S,p ) based on the κ-Köhler model approach (a) with variable values of κ t as derived from individual CCN efficiency spectra and (b) with a constant average value of κ=0.3 plotted against the observed CCN number concentrations (N CCN,S ).

Table 1 .
Characteristic parameters from the 5 calibration experiments performed during the campaign (arithmetic mean ± standard deviation).The last column shows the maximum relative deviation of individual calibration data points from the average calibration line (S vs. T ), indicating maximum relative uncertainties in S.
Table 1 indicate a relative uncertainty of less than ∼7% for S in the CCN measurements reported in this study.

Table 2 .
Characteristic average CCN parameters (arithmetic mean values ± standard deviation) for the entire campaign, for the Biomass Burning Event (BBE, 23-26 July) and for the campaign excluding the BBE for different S. Quantities are midpoint activation diameters (D a , D t , D cut ), maximum activated fractions (MAF f , MAF m ), CDF standard deviations (σ a , σ t ), heterogeneity parameters (σ a /D a , σ t /D t ), hygroscopicity parameters (κ a , κ t , κ cut ), number concentrations of total aerosol particles (3-900 nm, N CN,tot ) and of cloud condensation nuclei (N CCN,S ), and ratio of N CN,tot to N CCN,S as defined in Sect.2.2.4.n ES and n SD are the numbers of averaged CCN efficiency spectra and size distributions, respectively.

Table 3 .
Characteristic deviations between observed CCN number concentrations N CCN,S and CCN number concentrations predicted by different model approaches (N CCN,S,p ): arithmetic mean values of the relative bias ( b N CCN,S =(N CCN,S,p −N CCN,S )/N CCN,S ) and of the total relative deviation ( d N CCN,S =|N CCN,S,p −N CCN,S |/N CCN,S , respectively.n SD is the number of data points. 10, predicted CCN number concentrations (N CCN,S,p ) that were obtained with the κ-Köhler model and different hygroscopicity parameters are plotted against observed values of N CCN,S .For each data point, N CCN,S,p was calculated by integrating the measured CN size distribution above the critical dry particle diameter for CCN activation that corresponds to the given values of κ and S (Sects.2.2.4 and 3.2).