Measurements of cloud droplet activation of aerosol particles at a clean subarctic background site

Transcription

1 JOURNAL OF GEOPHYSICAL RESEARCH, VOL. 110,, doi: /2004jd005200, 2005 Measurements of cloud droplet activation of aerosol particles at a clean subarctic background site Mika Komppula, Heikki Lihavainen, and Veli-Matti Kerminen Research and Development, Finnish Meteorological Institute, Helsinki, Finland Markku Kulmala Department of Physical Sciences, University of Helsinki, Helsinki, Finland Yrjö Viisanen Research and Development, Finnish Meteorological Institute, Helsinki, Finland Received 6 July 2004; revised 4 October 2004; accepted 22 December 2004; published 22 March [1] Two years of continuous aerosol particle size distribution measurements provided the basis for this cloud droplet activation study. The cloud droplet activation of aerosol particles was studied in Pallas, a clean background site in northern Finland. A slightly different approach compared with traditional methods is presented by measuring simultaneously the cloud interstitial particle size spectrum and a nearby out-of-cloud particle size spectrum. The main advantage of this approach is that one can determine the activated fraction of different-size particles over the whole submicron-size range and that a large number of cloud activation events can be analyzed. The number of cloud droplet activation days peaked in late autumn and was lowest in summer. The annual variation of cloud droplet activation events followed the annual pattern of low clouds (those below 1000 m). A relation was found between the total particle number concentration outside the cloud and the number of activated particles. A larger particle concentration led to a higher number of activated particles, a lower activation percent, and a larger activation diameter (D 50 ). D 50 was on average 80 nm and varied from 50 to 128 nm. The average fraction of activated particles during the cloud events was 47% and varied from 9 to 86%. The cleaner and colder air masses from the northern Atlantic or the Arctic Ocean had, on average, 15 nm lower D 50 than the more polluted air masses from south and east containing more particles. The annual variation of the number of the activated particles and other variables were also closely related to the annual variation of particle concentration (high in summer and low in winter). On average 87 and 30% of accumulation and Aitken mode particles, respectively, were activated. Aitken mode particles were observed to cover up to 55% of the total number of activated particles (i.e., number of formed droplets), which demonstrates that they have a significant effect on cloud droplet activation and must be taken into account when estimating the aerosol indirect climate effects. Citation: Komppula, M., H. Lihavainen, V.-M. Kerminen, M. Kulmala, and Y. Viisanen (2005), Measurements of cloud droplet activation of aerosol particles at a clean subarctic background site, J. Geophys. Res., 110,, doi: /2004jd Introduction [2] Atmospheric aerosol and clouds have a mutual influence on each other. Aerosol particles modify many climatically important cloud properties, including the cloud reflectivity and lifetime [Rosenfeld, 2000; Norris, 2001; Harshvardhan et al., 2002; Krüger and Grassl, 2002]. Clouds, in turn, affect aerosol particle populations by altering their chemical composition, by enhancing their activation efficiencies in subsequent cloud encounters, by transferring ultrafine particles into optically more active Copyright 2005 by the American Geophysical Union /05/2004JD sizes, and by removing particles from the atmosphere by precipitation [Choularton et al., 1998; Flynn et al., 2000; Krämer et al., 2000, O Dowd et al., 2000; Kerminen, 2001; Andronache, 2003; Kerkweg et al., 2003]. [3] Aerosol-cloud interactions are crucially dependent on which fraction of the aerosol population activates into cloud droplets and which fraction remains as cloud interstitial particles. Despite the recent theoretical advances made in understanding and simulating cloud droplet activation [Kulmala et al., 1993; Chuang et al., 1997; Liu and Daum, 2000; Charlson et al., 2001; Abdul-Razzak and Ghan, 2002; Feingold and Chuang, 2002; Nenes and Seinfeld, 2003], we are still unable to predict the exact division between cloud droplets and cloud interstitial particles in the real atmosphere 1of10

2 Figure 1. (left). Location of Pallas GAW station (bottom right) and location of measuring stations at Pallas [Snider et al., 2003]. One reason for this is the lack of accurate field data on cloud droplet activation, especially what it comes to ultrafine particle size range that is relevant to clouds forming in clean or moderately polluted environments. [4] In field experiments the most common way of investigating cloud droplet activation has been to measure the number concentration and/or size distribution of cloud droplets and their residual particles, and to compare this with some total measure of the aerosol number concentration [Anderson et al., 1994; Gillani et al., 1995; Twohy et al., 2001; Glantz et al., 2003]. In many cases, like in all aircraft measurements, this comparison is made difficult by the fact that some of the relevant quantities have been measured at different times and typically well apart from each other. Here we approach the problem in a slightly different way by measuring simultaneously the cloud interstitial particle size spectrum and nearby out-of-cloud particle size spectrum. The main advantage of this approach is that one can determine the activated fraction of differentsize particles over the whole submicron size range. Another very useful feature is that our measurements are part of a continuous monitoring program. As a result, we are able to capture a large set of cloud events to analyze statistical dependences between different aerosol populations, their activation fractions at different size ranges and various meteorological factors. 2. Site Description and Methods 2.1. Site Description [5] Measurements were done in Finnish Meteorological Institute s Pallas-Sodankylä Global Atmosphere Watch (GAW) station located in northern Finland (Figure 1). The two measuring sites, Sammaltunturi ( N, E) and Matorova ( N, E), are 6 km apart from each other and at different altitudes (Figure 1). The higheraltitude station, Sammaltunturi, is inside the cloud during 10% of all days (5% of the time), while the lower-altitude Matorova station is practically always outside the cloud. This provides an opportunity to investigate the cloud droplet activation of aerosol particles. [6] The Sammaltunturi station resides on a top of the second southernmost fjeld, a round topped treeless hill, in a 50-km-long north and south chain of fjelds at an elevation of 565 m above sea level (a.s.l.). The vegetation on the fjeld top is sparse, consisting mainly of low vascular plants, moss and lichen. The Sammaltunturi station is about 300 m above surrounding area and the timberline lies about 100 m below the station. The surrounding forest consists of mixed pine, spruce and birch. The highest fjelds in the chain are 600 to 800 m a.s.l. Otherwise the region is hilly ( m a.s.l.), forested and partly swampy with some rather large lakes (ca 250 m a.s.l.). The Matorova station lies six km east-northeast of Sammaltunturi. It is situated on top of a small hill (340 m a.s.l.) in the middle of a ca. 100 m 100 m clearing inside a coniferous forest. [7] Both the stations are located inside the Pallas-Ounastunturi National Park. The area is inside subarctic region at the northernmost limit of the northern boreal forest zone. The average temperature at Sammaltunturi between years was 1.1 C, varying from a daily minimum of 32.7 C to a daily maximum of 20.7 C. The average wind speed was 7.0 m s 1 and the average relative humidity was 85%. The area has no significant local or regional pollution sources. The distance from Sammaltunturi to the nearest town, Muonio, with some 2500 inhabitants, is 19 km to the west. The closest major sources of pollution are the smelters 2of10

3 Nikel and Montshegorsk in Russia, located about 350 km away from Pallas, Nikel to the northeast and Montshegorsk to the east. A detailed description on the site has been given by Hatakka et al. [2003]. The Pallas area can be considered as a remote continental site, since it has an annual average particle concentration of cm 3. Daily particle concentration averages vary from 40 to 3500 cm 3, with low values in the winter and high in the summer Measurements [8] Similar DMPS (Differential Mobility Particle Sizer) systems were used at both sites for particle sizing, measuring the dry particle diameter in the size range nm. The size range is divided into 30 discrete size bins. DMPSs are built up with 28-cm-long Hauke-type differential mobility analyzer (DMA) [Winklmayr et al., 1991] with a closed loop sheath flow arrangement [Jokinen and Mäkelä, 1997] and a condensation particle counter (CPC), TSI model The used sheath air flow rate is 11.0 liters min 1 and sample aerosol flow rate is 1.0 liter min 1. The DMPS systems are equipped with sheath air temperature, pressure and relative humidity sensors. The sheath air is kept at relatively low and constant humidity (less than 20%) with a silicagel dryer. On days with no low clouds present, the size distribution is usually similar in both the measurement sites [Komppula et al., 2003]. At Sammaltunturi the visibility and rain intensity is measured with a Vaisala FD12P weather sensor and relative humidity with a Vaisala HMP35 sensor. [9] At Sammaltunturi station all the air samples into the instruments are taken from one main sampling line. The inlet of the main sampling line is about 7 m above the ground and about 3 m above the roof of the station building. The inner diameter of the stainless steel inlet nozzle and main sampling line is 56 mm. The flow rate in the main sampling line was about 90 m 3 h 1. The inlet nozzle is H-shaped, leading to two downward and two upward orientated inlets with a flow rate of about 22.5 m 3 h 1 through each. The outer surface of the inlet is heated about 1 2 C above zero to avoid the build up of ice and snow. Because the residence time of the sample air in the H-shaped nozzle is <0.1 s, the effect of heating on the temperature of the sample air or size of aerosol particle is negligible. There is no special rain protection for the inlet due to the problem with ice and snow. The nozzle is followed by about a 5-m-long vertical line. After this there is a T connector, in which the sample air makes a 90 turn into the station building. The other branch, directing downward and sealed at the end, collects possible condensed water. This is also the point where almost all the particles larger than 10 mm are separated from the sample air by inertia. The water collector is heated to avoid freezing. This heating does not affect the sample air. The connector is followed by about 4 m of horizontal tube with two 90 turns before the sample air arrives to the point, from which it is drawn into the aerosol instruments. The tubing inside the station building is insulated to avoid the condensation of water during summer. The diameter of 50% particle transmission efficiency with this arrangement was calculated to be about 7 mm for ambient air conditions and with an average wind speed (6.4 m s 1 ) in the data analyzed here. This diameter was calculated to vary slightly, by 1 2 mm, with wind speed extremes. The calculations were made according to the equations presented by Baron and Willeke [2001]. In the analysis described below it was assumed that all particles larger than 7 mm were cloud droplets. At Matorova station the air sample was drawn straight to the DMPS through stainless steel tube with 4 mm inner diameter about half a meter away form the station building outer wall and about 2 m above the ground. [10] At both stations the sample air was heated to stations indoor temperature (about 20 C) prior to its entering the aerosol instruments. The average outdoor temperature was about 5 C in these measurements, the highest temperature being about 9 C. In addition, the sheath air of the DMPS was kept below 20% relative humidity (16% on the average) and temperature above 20 C (21.2 C on the average). These facts ensured that the particle dry diameter was measured at both the sites. [11] An uncertainty estimate was derived in order to help the selection of days when the data at the two sites were comparable enough for a further analysis. A maximum uncertainty of 30% was used as an upper limit. One critical assumption that was made was that all particles larger than 7 mm in diameter are cloud droplets. This had to be assumed because of the aerosol inlet cutoff diameter of 7 mm. It should be noted that we did not measure the cloud in a lagrangian sense. The cloud was moving and the measurements were made in one fitted location on the ground. This forced us to make the assumption that the cloud was homogenous by its physical properties. The drift of the cloud makes it difficult to estimate any changes that the cloud droplet population, or the cloud droplet residual particle size distribution, would experience during the cloud evolution. This estimation would also require a smaller size bin interval in DMPS measurements. These issues will probably be included in our future activities along with droplet size distribution measurements and collecting/ analyzing the cloud water Data Analysis [12] The analyzed data covered a period of 22 months, from April 2000 to February The activation of aerosol particles into cloud droplets was identified as a sudden drop in the particle number concentration corresponding to the largest size classes of the DMPS (Figures 2a and 2b). Simultaneously, a drop in the aerosol light scattering coefficient was seen (Figure 2c). To verify the existence of a cloud, weather and other related data were examined. A decrease in the visibility (Figure 2d) and increase in the relative humidity were clear indications of the presence of a cloud. A Web camera, taking still photos in hour intervals, was also helpful in determining the cloud. Ceilometer measurements, giving the height of the lowest clouds, were used to verify the annual pattern of low clouds. Ceilometer measurements, although done at Pello some 130 km away from Pallas, provide a good view of the annual pattern of clouds. The presence or absence of rain was identified from weather data. [13] The selection procedure for cloud events was started by checking the DMPS data day by day. If a loss of largest particles was observed at Sammaltunturi the day passed the first stage. After that a comparison with the reference DMPS data at Matorova was done to verify that a difference 3of10

4 Figure 2. Example of a cloud droplet activation event, 15 October 2000: (a) in-cloud particle size distribution, (b) total concentration in and out of cloud, (c) aerosol scattering coefficient in three wavelengths, and (d) visibility. in the size spectra between these two sites existed. From weather data it was then checked whether a rapid increase in the relative humidity close to 100% (>98%) and a sudden drop in the visibility close to zero (<200 m compared with a typical value of about 50 km) was found at Sammaltunturi. Finally, the Web camera photo was checked to ensure the cloud occurrence. If all these criteria were fulfilled, the day was classified as a cloud event day. After this all rainy days were left out. This was done because during the rain, falling raindrops remove unactivated aerosol particles, which would cause errors in activation calculations. Out of the remaining days, all days with fluctuations in the DMPS data during the cloud time were ruled out and only the most obvious cloud days were taken into a more detailed inspection. We are confident that this procedure ruled out the days with thin clouds (very probably with fluctuations in the DMPS data) and mixed-phase clouds with very rough and varying interstitial particle size distribution as seen in work by Henning et al. [2002]. A good measure of the very critical selection procedure is that after the first procedure stage we still had close to 200 days left and only 50 passed to this study. [14] The size distribution of activated particles (cloud residual particles) was deduced indirectly by subtracting the in-cloud (Sammaltunturi station) DMPS size spectra from the corresponding out-of-cloud (Matorova station) size spectra (Figures 3a and 3c). From the same data, the fraction of activated particles as a function of size could be calculated (Figures 3b and 3d), including the diameter of 50% activation efficiency (D 50 ). D 50 can also be obtained visually from the size spectra as the point where in-cloud spectrum and the spectrum of activated particles cross each other (e.g., Figure 3a). The number of activated particles and the activated fraction were also estimated separately for different modes. All these comparisons and estimations were done for average values during the whole cloud event. An error estimate was derived for each day from the concentration difference in the half-hour average values between the sites on the moment just before the cloud appeared. Numerous weather-related and other parameters (temperature, temperature lapse rate, wind, sun radiation, relative humidity, absolute water vapor concentration, visibility, rain, pressure, air mass origin and concentrations of SO 2, NO, NO 2 and O 3 ) were studied for correlation analysis to see their effects on cloud droplet activation. 3. Results and Discussion [15] After a careful classification and selection, described in section 2.3, a total of 50 cloud droplet activation days 4of10

5 Figure 3. Two example cases of the data analysis procedure: (a b) typical continental air masses on 18 September 2001 and (c d) typical arctic marine air masses on 6 November Figures 3a and 3c give the size spectra out-of-cloud (stars) and in-cloud (circles) and the spectra of activated particles (triangles) (out-of-cloud minus in-cloud size spectra), and Figures 3b and 3d give the activated fraction in 30 size classes. were left for a further inspection. The number of these days peaked in late autumn (October November) and was lowest in summer (May August). In November, cloud droplet activation was observed during one third of the days. In summer only 0 2 events were observed per month. The annual variation of cloud droplet activation events followed the annual pattern of low clouds (those below 1000 m). On average the cloud base was lowest during late autumn (October November) and highest in summer (June). The average depth of the cloud was smallest in summer and highest in late autumn/winter (October January). Orographic clouds were not observed, but orography may have some effect on the cloud layer. The annual average probability for a cloud droplet activation day was 10%. Cloud events were observed to start throughout the day. The duration of cloud events varied from one hour to 47 hours being 11 hours on average. [16] As mentioned earlier, an uncertainty estimate was calculated from the concentration difference between the sites before the cloud appeared. This estimate helped us to select the days when the data at the two sites were comparable enough for a further analysis. Particle size distributions and concentrations on these two sites are usually very similar. Size distributions are almost equally shaped and total particle number concentrations are within 10% of each other during the time without low clouds [Komppula et al., 2003]. Despite this one has to be careful when calculating averages for shorter periods, as was done in this study. For the whole 50-day data set, this uncertainty was 22% on average and varied between 1 and 64%. This uncertainty was also translated into the uncertainty in the activated fraction, though the values did not change significantly. To lower the possible error, but to keep a reasonable number of cases for further analyses, a maximum uncertainty of 30% was chosen. This uncertainty limit ruled out 11 days. Tighter uncertainty limits were tested but their effect on the final results was found to be negligible. Some basic results with different uncertainty limits are shown in Table 1. Results with a lower uncertainty do not differ significantly from those with a 30% uncertainty. At 10% uncertainty limit the number of days is getting so low that it is hard to get any reliable statistical results Activation Diameter (D 50 ) [17] The diameter of 50% activation efficiency, D 50,isa useful quantity especially for modeling purposes, since it 5of10

6 Table 1. Comparison of Some Basic Results With Different Uncertainty Levels All Cases Uncertainty Limit of 30% Uncertainty Limit of 20% Uncertainty Limit of 10% Average uncertainty, % Number of data days A tot, a cm D 50, b nm A tot, c % A ait, d % A acc, e % a Average number of activated particles. b Diameter corresponding to 50% activation efficiency. c Average activation percent of particles. d Average activation percent for the Aitken mode particles. e Average activation percent for accumulation mode particles. provides a simple way to divide the particle population into two classes: those that activate into cloud droplets (particles with a dry diameter larger than D 50 ) and those that remain as cloud interstitial particles. In our measurements, D 50 varied between 50 and 128 nm with an average of 80 nm. The value of D 50 increased typically with increasing preexisting particle concentration, as can be seen from Table 2. A moderate correlation between D 50 and the number of accumulation mode particles (N acc ) could be seen, R 2 being equal to 0.23 and 0.27 for the data sets with the 30% and 10% uncertainty limit, respectively. The cleaner and colder air masses from northern Atlantic or the Arctic Ocean (see Figures 3c and 3d) had an average D 50 of 76 nm, while the more polluted air masses from south and east (see Figures 3a and 3b) had an average D 50 of 91 nm. Figure 4 shows the average activation spectra for both clean marine and polluted continental air masses. It is clearly seen that in clean marine air smaller particles are activated than in continental air. The variation in sizes below about 40 nm is partly due to measurement uncertainties caused by low particle concentrations in this size range, partly due to coagulation of the smallest particles with cloud droplets. [18] For D 50 a slight correlation was also found with ambient temperature (R 2 = 0.23) and absolute water content in the air (R 2 = 0.24), but these parameters are related to the time of the year and the air mass origin. The annual variation of D 50 seemed to be related to the air mass origin. In summer, Pallas experiences more eastern (i.e., continental) air, while in winter western (i.e., marine) air masses dominate [Hatakka et al., 2003]. The average D 50 for spring/summer and autumn/winter were 94 and 78 nm, respectively. Although all rainy days were taken out of the final data set, D 50 was also calculated for these 23 days for a general interest. On rainy days particles will experience also below cloud scavenging additional to in-cloud scavenging/activation. For these rainy days D 50 was on average 92 nm and varied between 49 and 137 nm Number of Activated Particles [19] The pre-existing particle concentration varied between 60 and 1330 cm 3 with an average of 410 cm 3 during the cloud events. The average number of activated particles was 154 cm 3 and varied between 30 and 610 cm 3. The resulting percent of activated particles varied from 9 to 86% with an average of 47%. [20] A clear relation between the total pre-existing particle number concentration (N tot ) and the number of activated particles (A tot ) was found. Also the percent of activated particles and D 50 were related to total number concentration. On average, a higher pre-existing particle number concentration was associated with a higher number of activated particles, lower activation percent and larger D 50. The correlation coefficient (R 2 ) between N tot and A tot was 0.47 and 0.90 for the 30 and 10% uncertainty limits, respectively, while the correlation coefficient between N tot and activation percent was 0.32 and 0.24 for 30 and 10% uncertainty limits, respectively. [21] The cleaner and colder air masses from the Northern Atlantic or Arctic Ocean had, on average, 120 cm 3 activated particles with an activation percent of 42%. In the more polluted air masses from south and east, the average A tot was 200 cm 3 and the activation percent was 53%. The slightly lower activation percent in clean marine air compared with continental air is somewhat surprising and not consistent with the general trend shown in Table 2. The most likely reason for this behavior can be found in the different shape of the particle number size distribution between these two air mass types: marine air masses had typically a more pronounced Aitken mode and thereby a larger fraction of particles that are too small to act as cloud condensation nuclei (see Figure 3). Another factor that affects the activation percent is the chemical composition of particles, which might be different between the marine and continental air masses. [22] The correlation coefficient (R 2 ) of A tot with the ambient temperature was 0.20 and also with the absolute water content in the air 0.20, but, as mentioned earlier, these parameters are more related to the time of the year. The annual variation of the number of the activated particles was associated with the annual variation of particle concentration (high in summer, low in winter). In spring/summer the average A tot was 340 cm 3 and in autumn/winter it was 130 cm 3. The average activation percent for spring/summer and autumn/winter were 41 and 48%, respectively Activation of Different Particle Modes [23] The modal cloud activation of particles was calculated for three fixed modes: the nucleation mode (7 25 nm), Aitken mode (25 95 nm) and accumulation mode ( nm) (see Figure 5). These fixed modes have been found statistically relevant for Pallas [Komppula et al., 2003]. On average, the total activation percent of particles was 47% during the 39 selected cloud events observed at Pallas. The relation between the total particle concentration and number of activated particles reported earlier was also found when Table 2. Pre-Existing Particle Concentration (Concentration Outside the Cloud) Compared With Cloud Droplet Activation (Number and Percent) and Activation Efficiency (D 50 ) a Total Pre-Existing Particle Number, cm 3 A tot,cm 3 A tot,% D 50,nm < > a Average values are given in the last row. 6of10

7 Figure 4. masses. Average activation spectra for clean marine (stars) and polluted continental (triangles) air looking at modal observations. For Aitken and accumulation mode particles, the annual variation of the number of activated particles as well as that of the activation percent was found similar to that of the total particle number concentration. [24] In the accumulation mode, 87% of the particles were activated on average (Table 3). The activation percent varied from 71 to 97%. The average number of activated accumulation mode particles was 110 cm 3 varying between 20 and 310 cm 3. A relation between the number of activated accumulation mode particles (A acc ) and N tot was found (Table 3), as expected, the correlation coefficient (R 2 ) being equal to If the nucleation mode was left out of N tot and the correlation for A acc was done with N ait+acc (the number concentration of particles larger than 25 nm in diameter), R 2 increased to A very high correlation (R 2 = 0.97) was found between N acc and A acc, as one might expect. The air mass origin influenced accumulation mode activation, partly differently to total activation. The cleaner and colder marine air masses from northern Atlantic or Arctic Ocean had on average 80 cm 3 activated accumulation mode particles with an activation percent of 89%. In the more polluted air masses from south and east the average A acc was 140 cm 3 and the activation percent 84%. [25] Even in Aitken mode notable activation was found to take place, since on average 30% of the Aitken mode particles were activated. The average number of activated Aitken mode particles was 35 cm 3 varying from zero to 200 cm 3. The number of activated Aitken mode particles (A ait ) correlated moderately with N tot (R 2 = 0.41), as well as with N ait (R 2 = 0.34) and N ait+acc (R 2 = 0.28). Aitken mode particles were observed to cover up to 55% of the total number of activated particles (i.e., number of formed droplets), 22% on average. All this demonstrates that Aitken mode particles have a significant influence on cloud droplet activation, being important when estimating the indirect climatic effects by aerosols. This observation agrees nicely with the earlier model predictions by Kulmala et al. [1996]. The air mass origin did not have such a large effect on the activation of Aitken mode particles as it did for the accumulation mode. In clean and cold marine air masses from northern Atlantic or Arctic Ocean, there were on average 35 cm 3 activated particles, which is equal to that in more polluted air masses from south and east. The average activation percentages differed, however, being 36% for marine and 23% for continental air masses. [26] Not much attention was paid to the nucleation mode because of the low concentrations in the smallest size classes. This, together with large variations and sudden changes in the concentration (nucleation events) made a reliable statistical analysis difficult. Furthermore, cloud droplets provide a high coagulation sink for the smallest nucleation mode particles. As a result, the smallest particles may have been removed by cloud droplets either within the current cloud or during some earlier cloud encounter. This makes the estimation of nucleation mode activation even more difficult. Despite all this, strong indications of coagulation losses of the smallest nucleation mode particles onto cloud droplets were found on a few days (see, e.g., Figure 3d). On these days, on average 50% of particles smaller than 13 nm were lost, probably due to coagulation onto cloud droplets. On some days all particles up to 11 nm in diameter were lost. While these results are not statistically significant because of the uncertainties mentioned above, they suggest that clouds can act as a sink for small nucleation mode particles in the atmosphere. [27] In order to avoid the difficulties caused by nucleation mode particles, the analysis was also done for the combined Aitken and accumulation mode. This total concentration of particles larger than 25 nm in diameter (N ait+acc ) showed less variation than the original N tot. On average the activation percent of N ait+acc was 58% varying from 12 to 93%. 7of10

8 Figure 5. Cloud activation event at Pallas, 11 January On the left are (top) in-cloud and (bottom) out-of-cloud size distribution contour plots, and on the right are modal concentrations. The clouds were present from 1630 to The number of activated N ait+acc particles (A ait+acc ) varied from 30 to 370 cm 3 being on average 140 cm 3. The correlation coefficient (R 2 ) between N ait+acc and A ait+acc was 0.47 and 0.72 with 30 and 10% uncertainty limits, respectively Comparison With Earlier Measurements [28] A compilation of cloud droplet measurements done in marine and continental clouds has been presented by Miles et al. [2000]. They found significant differences between these two cloud types, with average cloud droplet number concentrations being equal to 74 and 288 cm 3 for marine and continental clouds, respectively. These results are in qualitative agreement with our values of 120 cm 3 and 200 cm 3 for marine and continental air masses, respectively. The slightly higher droplet concentration in marine air masses measured by us may be due to the fact that these air masses have traveled over the continent prior to arrival at Pallas. The relatively clean continental air at Pallas explains the lower number of activated particles in Table 3. Pre-Existing Particle Concentration (Concentration Outside the Cloud) Compared With Cloud Droplet Activation (Number and Percent) of Different Modes and Combined Aitken Plus Accumulation Mode a Total Pre-Existing Particle Activated Particles Number, cm 3 A nuc,cm 3 A nuc,% A ait,cm 3 A ait,% A acc,cm 3 A acc,% A ait+acc,cm 3 A ait+acc,% < > a Average values are given in the last row. 8of10

9 continental air masses compared with the results in Miles et al. [2000]. The range of the number of activated particles in Pallas, cm 3, was in the range of the average values reported by Miles et al. [2000]. Droplet concentrations up to 2500 cm 3, observed in polluted conditions [Bower et al., 2000] were not reached at Pallas site. On the contrary, the extremely low average cloud droplet concentration of cm 3 measured in the Southern Ocean by Yum and Hudson [2004] is in the range of the lowest values (30 cm 3 ) found at Pallas. The average value at Pallas, 154 cm 3, could be classified as clean continental or slightly polluted marine clouds. [29] Data on the activated fraction of aerosol particles is available almost solely for the accumulation mode. In these studies, the activated fraction of accumulation mode particles has been found to vary from <0.2 to values close to unity, with higher fractions typically encountered in cleaner air and for higher cloud liquid water contents [Anderson et al., 1994; Hallberg et al., 1994, Gillani et al., 1995; Schwarzenboeck et al., 2000; Henning et al., 2002; Glantz et al., 2003]. The activated fractions measured by us for the accumulation mode are at the upper end of those found in the earlier studies, averaging 93% for the cleanest air mass category and 83% for the most polluted air mass category (Table 3). [30] Glantz et al. [2003] found a poor correlation between the number of cloud droplets and the mass (volume measured) of cloud residual particles for clean air, on the basis of which they suggested that a major fraction of particles determining the cloud droplet number concentration were below the accumulation size range (<100 nm). In our measurements the contribution of Aitken mode particles to the cloud droplet population was found to be significant, being 22% on average. The activated fraction of combined Aitken and accumulation mode particles ranged from 12 to 93% with an average of 58%. These values are larger than those measured by Schwarzenboeck et al. [2000] but comparable to those measured by Lowenthal et al. [2002] at a high-altitude site (3210 m a.s.l.) in northwestern Colorado. Direct evidence of the activation of Aitken mode particles at clean conditions has also been reported by Henning et al. [2002] at a high-alpine site Jungfraujoch in Central Europe. [31] Concerning the nucleation mode, Lowenthal et al. [2002] found that roughly half of the 10 nm particles were removed by the clouds, while Twomey [1977] estimated that all particles smaller than 10 nm would be removed by clouds within one hour. These two works give support to our finding that occasionally a significant fraction of nucleation mode particles were removed in clouds by coagulation. 4. Summary and Conclusions [32] The cloud droplet activation of aerosol particles over a period of 22 months was studied. The study was done in Pallas, a clean background site in northern Finland. A slightly different approach compared with traditional methods was presented by measuring simultaneously the cloud interstitial particle size spectrum and nearby out-of-cloud particle size spectrum. The main advantage of this approach is that one can determine the activated fraction of differentsize particles over the whole submicron size range, and that a large number of cloud events could be analyzed. [33] Statistically, a clear relation between the number of pre-existing particles (N tot ), the number of activated particles (A tot ) and the diameter of 50% activation efficiency (D 50 ) was found. Larger values of N tot were associated with larger A tot, larger D 50 and lower percent of activated particles. The values of D 50 were on average 80 nm and varied from 50 to 128 nm. The average activation percent during the cloud events was 47% and varied from 9 to 86%. The air mass origin had an effect on droplet activation because of its effect on the aerosol concentration. For example, the cleaner and colder air masses from northern Atlantic or the Arctic Ocean had on average 15 nm lower D 50 than the more polluted air masses from south and east containing more particles. The annual variation of D 50 could be related to the air mass origin, since in summer Pallas experienced more eastern (i.e., continental) and in winter more western air masses (i.e., marine). The annual variation of the number of the activated particles and other variables were also closely related to the annual variation of particle concentration (high in summer, low in winter). [34] In the accumulation mode, 87% of the particles were activated on average. Even in the Aitken mode the activation was found notable, since on average 30% of the Aitken mode particles were activated. Aitken mode particles were observed to cover up to 55% of the total number of activated particles (i.e., number of formed droplets), 22% on average. A combined Aitken and accumulation mode (N ait+acc ) showed less variation than the original N tot.on average the activation percent of N ait+acc was 57% reaching values up to 93%. Some evidence was found that clouds act as a sink for small nucleation mode particles in the atmosphere. It has to be noted that in this study we used fixed mode limits which were found statistically relevant for Pallas, for example Aitken mode is particles from 25 to 95 nm in diameter. [35] Our measurements demonstrate that Aitken mode particles have a significant influence on cloud droplet activation under clean and moderately polluted conditions, being therefore important when estimating the indirect climatic effects by aerosols. Our earlier measurements have shown that atmospheric new particle formation can be a significant source of Aitken mode particles at the same site [Lihavainen et al., 2003]. Since atmospheric new particle formation is observed all over the world [Kulmala et al., 2004], it is also important to find out its contribution to indirect climatic effects. [36] Acknowledgments. This work has been supported by the Academy of Finland (project numbers and ), by the European Union, and by the Maj and Tor Nessling Foundation. References Abdul-Razzak, H., and S. J. Ghan (2002), A parametrization of aerosol activation: 3. Sectional representation, J. Geophys. Res., 107(D3), 4026, doi: /2001jd Anderson, T. L., D. S. Covert, and R. J. Charlson (1994), Cloud droplet number studies with a counterflow virtual impactor, J. Geophys. Res., 99(D4), Andronache, C. (2003), Estimated variability of below-cloud aerosol removal by rainfall for observed aerosol size distributions, Atmos. Chem. Phys., 3, Baron, P. A., and K. Willeke (2001), Aerosol Measurement, 2nd ed., John Wiley, Hoboken, N. J. 9of10

10 Bower, K. N., et al. (2000), ACE-2 HILLCLOUD: An overview of the ACE-2 ground-based cloud experiment, Tellus, Ser. B, 52, Charlson, R. J., J. H. Seinfeld, A. Nenes, M. Kulmala, A. Laaksonen, and M. C. Facchini (2001), Reshaping the theory of cloud formation, Science, 292, Choularton, T. W., K. N. Bower, K. M. Beswick, M. Parkin, and A. Kaye (1998), A study of the effects of cloud processing of aerosol on the microphysics of cloud, Q. J. R. Meteorol. Soc., 124, Chuang, P. Y., R. J. Charlson, and J. H. Seinfeld (1997), Kinetic limitations on droplet formation in clouds, Nature, 390, Feingold, G., and P. Y. Chuang (2002), Analysis of the influence of filmforming compounds on droplet growth: Implications for cloud microphysical processes and climate, J. Atmos. Sci., 59, Flynn, M. J., K. N. Bower, T. W. Choularton, W. Wobrock, J. M. Mäkelä, B. Martinsson, G. Frank, H.-C. Hansson, H. Karlsson, and P. Laj (2000), Modelling cloud processing of aerosol during the ACE-2 HILLCLOUD experiment, Tellus, Ser. B, 52, Gillani, N. V., S. E. Schwartz, W. R. Leaitch, J. W. Strapp, and G. A. Isaac (1995), Field observations in continental stratiform clouds: Partitioning of cloud particles between droplets and unactivated interstitial aerosols, J. Geophys. Res., 100(D9), 18,687 18,706. Glantz, P., K. J. Noone, and S. R. Osborne (2003), Scavenging efficiencies of aerosol particles in marine stratocumulus and cumulus clouds, Q. J. R. Meteorol. Soc., 129, Hallberg, A., et al. (1994), Phase partitioning of aerosol particles in clouds at KleinerFeldberg, J. Atmos. Chem., 19, Harshvardhan, S. E. Schwartz, C. M. Benkovitz, and G. Guo (2002), Aerosol influence on cloud microphysics examined by satellite measurements and chemical transport modeling, J. Atmos. Sci., 59, Hatakka, J., et al. (2003), Overview of the atmospheric research activities and results at Pallas GAW station, Boreal Environ. Res., 8(4), Henning, S., E. Weingartner, S. Schmidt, M. Wendisch, H. W. Gäggeler, and U. Baltensberger (2002), Size-dependent aerosol activation at the high-alpine site Jungfraujoch (3580 m asl), Tellus, Ser. B, 54, Jokinen, V., and J. M. Mäkelä (1997), Closed-loop arrangement with critical orifice for DMA sheath/excess flow system, J. Aerosol Sci., 28, Kerkweg, A., S. Wurzler, T. Reisin, and A. Bott (2003), On the cloud processing of aerosol particles: An entraining air-parcel model with two-dimensional spectral cloud microphysics and a new formulation of the collection kernel, Q. J. R. Meteorol. Soc., 129, Kerminen, V. (2001), Relative roles of secondary sulfate and organics in atmospheric cloud condensation nuclei production, J. Geophys. Res., 106(D15), 17,321 17,334. Komppula, M., H. Lihavainen, J. Hatakka, P. P. Aalto, M. Kulmala, and Y. Viisanen (2003), Observations of new particle formation and size distribution at two different heights and surroundings in subarctic area in Northern Finland, J. Geophys. Res., 108(D9), 4295, doi: / 2002JD Krämer,M.,N.Beltz,D.Schell,L.Schütz, C. Sprengard-Eichel, and S. Wurzler (2000), Cloud processing of continental aerosol particles: Experimental investigations for different drop sizes, J. Geophys. Res., 105(D9), 11,739 11,752. Krüger, O., and H. Grassl (2002), The indirect aerosol effect over Europe, Geophys. Res. Lett., 29(19), 1925, doi: /2001gl Kulmala, M., A. Laaksonen, P. Korhonen, T. Vesala, T. Ahonen, and J. C. Barrett (1993), The effect of atmospheric nitric acid vapor on CCN activation, J. Geophys Res., 98(D12), 22,949 22,958. Kulmala, M., P. Korhonen, T. Vesala, H.-C. Hansson, K. Noone, and B. Svenningsson (1996), The effect of hygroscopicity on cloud droplet formation, Tellus, Ser. B, 48, Kulmala, M., H. Vehkamäki, T. Petäjä, M. Dal Maso, A. Lauri, V.-M. Kerminen, W. Birmili, and P. H. McMurry (2004), Formation and growth rates of ultrafine atmospheric particles: A review of observations, J. Aerosol Sci., 35, Lihavainen, H., V.-M. Kerminen, M. Komppula, J. Hatakka, V. Aaltonen, M. Kulmala, and Y. Viisanen (2003), Production of potential cloud condensation nuclei production associated with atmospheric new-particle formation in northern Finland, J. Geophys. Res., 108(D24), 4782, doi: /2003jd Liu, Y., and P. H. Daum (2000), Spectral dispersion of cloud droplet size distributions and the parametrization of cloud droplet effective radius, Geophys. Res. Lett., 27, Lowenthal, D. H., R. D. Borys, and M. A. Wetzel (2002), Aerosol distributions and cloud interactions at a mountaintop laboratory, J. Geophys. Res., 107(D18), 4345, doi: /2001jd Miles, N. L., J. Verlinde, and E. E. Clothiaux (2000), Cloud droplet size distribution in low-level stratiform clouds, J. Atmos. Sci., 57, Nenes, A., and J. H. Seinfeld (2003), Parametrization of cloud droplet formation in global climate, J. Geophys. Res., 108(D14), 4415, doi: /2002jd Norris, J. R. (2001), Has northern Indian Ocean cloud cover changed due to increasing anthropogenic aerosol?, Geophys. Res. Lett., 28, O Dowd, C. D., J. A. Lowe, N. Clegg, M. H. Smith, and S. L. Clegg (2000), Modeling heterogeneous sulphate production in maritime stratiform clouds, J. Geophys. Res., 105(D6), Rosenfeld, D. (2000), Suppression of rain and snow by urban and industrial air pollution, Science, 287, Schwarzenboeck, A., J. Heintzenberg, and S. Mertes (2000), Incorporation of aerosol particles between 25 and 850 nm into cloud elements: Measurements with a new complementary sampling system, Atmos. Res., 52, Snider, J. R., S. Guibert, J. L. Brenguier, and J.-P. Putaud (2003), Aerosol activation in marine stratocumulus clouds: 2. Köhler and parcel theory closure studies, J. Geophys. Res., 108(D15), 8629, doi: / 2002JD Twohy, C. H., J. G. Hudson, S.-S. Yum, J. R. Anderson, S. K. Durlak, and D. Baumgardner (2001), Characteristics of cloud-nucleating aerosols in the Indian Ocean region, J. Geophys. Res., 106(D22), 28,699 28,710. Twomey, S. (1977), Atmospheric Aerosols, Elsevier, New York. Winklmayr, W., G. P. Reischl, A. O. Linder, and A. Berner (1991), A new electromobility spectrometer for the measurement of aerosol size distribution in the size range 1 to 1000 nm, J. Aerosol Sci., 22, Yum, S. S., and J. G. Hudson (2004), Wintertime/summertime contrasts of cloud condensation nuclei and cloud microphysics over the Southern Ocean, J. Geophys. Res., 109,, doi: /2003jd V.-M. Kerminen, M. Komppula, H. Lihavainen, and Y. Viisanen, Research and Development, Finnish Meteorological Institute, P. O. Box 503, FI Helsinki, Finland. (mika.komppula@fmi.fi) M. Kulmala, Department of Physical Sciences, University of Helsinki, P. O. Box 64, FIN Helsinki, Finland. 10 of 10