Cloud droplet activation of secondary organic aerosol



[1] Measurements of hygroscopicity and cloud condensation nuclei (CCN) activity were conducted on secondary organic aerosol (SOA) formed in a smog chamber. SOA precursors included α-pinene, β-pinene, Δ3-carene, and toluene, representative of both naturally and anthropogenically emitted organic species. Measured CCN activation was comparable for all of the species studied and occurred at humidity conditions which are readily attained in the atmosphere. Further, there was little variation in hygroscopic growth between compounds. However, measured droplet activation conditions were inconsistent with hygroscopicity measured below water saturation and Köhler theory expressions based on Raoult’s law for several parameterizations for water activity. In the atmosphere, SOA may compose a large fraction of atmospheric particulate matter and will often exist internally mixed with inorganic species. Using the current results, we compare SOA to insoluble organic species to calculate CCN activation from mixed organic-sulfate particles for a range of atmospheric conditions. We find that droplet activation behavior of mixed particles containing SOA is the same as that of mixed particles for which the organic component is nonhygroscopic, except for cases in which there are low particle concentrations, low updraft velocities, and the aerosol composition is dominated by organics.

1. Introduction

[2] Organic moieties are commonly found in ambient aerosol, often accounting for 20–50% of fine particle mass over the continental midlatitudes [Pöschl, 2005; Saxena and Hildemann, 1996; White, 1990] and up to 90% in tropical forests [Andreae and Crutzen, 1997; Artaxo, 2001]. Primary organic aerosol is emitted directly in the condensed phase, while secondary organic aerosol (SOA) forms when volatile organic compounds are oxidized in the atmosphere to form highly oxygenated, lower vapor pressure products which partition into the particle phase [Seinfeld and Pankow, 2003]. Terpenes, emitted naturally by forests, undergo oxidation by ozone, hydroxyl, and nitrate radicals to produce low volatility compounds such as pinic acid, which can directly condense, or more volatile multifunctional compounds that apparently become incorporated into the particles through condensed-phase reactions that form oligomers [Docherty et al., 2005; Gao et al., 2004; Tolocka et al., 2004; Yu et al., 1999a]. The products of terpene oxidation have been identified in ambient aerosol [Kavouras et al., 1998; O’Dowd et al., 2002; Tunved et al., 2006; Yu et al., 1999b ] and appear to provide a substantial contribution to SOA. Global SOA formation from biogenic precursors has been estimated from 18.5 [Griffin et al., 1999] to 30–270 Tg yr−1 [Andreae and Crutzen, 1997], although recent modeling studies suggest that SOA may be highly underestimated [Heald et al., 2005; Henze and Seinfeld, 2006]. Although biogenic sources dominate global SOA formation, anthropogenic sources such as gasoline may dominate in more polluted areas. In this case, the aromatic content of the fuel appears to control SOA formation [Odum et al., 1997], although recent studies have indicated that other components may also contribute [de Gouw et al., 2005; Volkamer et al., 2006]. The SOA formed from aromatics oxidation is composed of a large variety of multifunctional ring-opened products and oligomers [Kalberer et al., 2004; Kleindienst et al., 2004].

[3] Atmospheric aerosols play an important role in the Earth’s radiative budget. Aerosol effects on climate are complex and generally are separated into the broad categories of direct and indirect effects. The direct effect derives from the extinction properties of the particles themselves, whereas indirect effects arise from the ways in which aerosols affect the microphysical and radiative properties of the clouds they nucleate, including enhancements in cloud reflectivity [Twomey, 1974] and suppression of drizzle [Albrecht, 1989] or cloud formation [Petters et al., 2006d]. Numerous studies have explored the cloud condensation nuclei (CCN) activity of pure organic compounds [Broekhuizen et al., 2004; Corrigan and Novakov, 1999; Cruz and Pandis, 1997; Giebl et al., 2002; Huff Hartz et al., 2006; Kumar et al., 2003; Petters et al., 2006a, Petters et al., 2006d; Prenni et al., 2001]. Although single component studies have provided much insight into the role of organics in the atmosphere, each component often represents less than 1% of ambient particle mass. Investigations of aerosol-water interactions of SOA formed under controlled conditions in a smog chamber represent a step closer to ambient compositions. Water uptake of monoterpene-derived SOA has been studied previously, with reported growth factors falling between 1.06 and 1.11 near 85% relative humidity (RH) [Saathoff et al., 2003; Varutbangkul et al., 2006; Virkkula et al., 1999]. CCN activity also has been reported for monoterpene-derived SOA [Hegg et al., 2001; Huff Hartz et al., 2005; VanReken et al., 2005]. Experimental differences limit the ability to compare many of the CCN measurements; however, studies by VanReken et al., [2005] and Huff Hartz et al., [2005] of SOA formed from α-pinene ozonolysis showed broadly consistent results for similar experimental conditions, discussed further below (section 3.2).

[4] Here we add to these studies by simultaneously measuring water uptake below water saturation and CCN activation above water saturation from SOA generated in a smog chamber. The SOA studied is formed from the ozonolysis of three naturally emitted monoterpenes (α-pinene, β-pinene, and Δ3-carene) and from the photooxidation of toluene. The three monoterpenes represent ∼60% of the estimated global monoterpene emissions [Griffin et al., 1999]. Toluene, a component of gasoline emitted in vehicle exhaust and through fuel evaporation, is used as a proxy for SOA formation in anthropogenically modified air sheds. We begin by comparing our measurements to previous measurements. In an effort to describe composition-dependent water activity for SOA over a range of humidities, we then attempt to link hygroscopicity to CCN activation using several existing expressions of Köhler theory. We finish by considering the potential role of SOA in cloud formation in the atmosphere.

2. Experiment

2.1. SOA Formation

[5] Experiments were performed in a ∼7 m3 polytetrafluoroethylene environmental chamber [Lim and Ziemann, 2005] operated at room temperature (25°C) and pressure (970 hPa). Prior to measurements, the chamber was flushed and filled with dry (RH <1%), hydrocarbon-free (<5 ppbv) air (Aadco Instruments, Inc.), with typical background conditions, 20 < [CN] < 60 cm−3, [NOx] < 40 ppbv, and [O3] < 1 ppbv. Background particles are not expected to influence results, as SOA particle concentrations were in excess of 5 × 104 cm−3. A scanning mobility particle sizer (SMPS) monitored the particle size distribution inside the chamber throughout the experiment. An experimental schematic is shown in Figure 1.

Figure 1.

Schematic for SOA generation and measurements of size distribution, water uptake and CCN activity. SMPS, scanning mobility particle sizer; DMA, differential mobility analyzer; CPC, condensation particle counter; CCN, cloud condensation nuclei.

[6] For studies involving SOA formation from monoterpenes, the chamber first was filled with >3 ppmv ozone. Clean air then was flushed through a glass bulb containing 20 μL of the monoterpene and into the chamber. After a short delay, ranging from a few seconds up to 10 min, particles formed via nucleation. Following the nucleation event, the particle size distribution evolved. The mode of the size distribution increased for the first 15–30 min and stabilized subsequently. The particle number concentration and the particle surface area decreased thereafter because of coagulation and losses to the walls. Typical aerosol mass loadings in the chamber decreased during the experiment and ranged between 10–300 μg m−3. It should be noted that high precursor concentrations are necessary to produce sufficient particle numbers for the experiments. These concentrations do not reflect average ambient concentrations, and the resulting chemistry, and SOA chemical composition, may be affected.

[7] For studies involving SOA formation from toluene, clean air was again flushed through a glass bulb containing 20 μL of the precursor. Then 450 ppbv of methyl nitrite were added to the chamber, together with 240 ppbv of NO, followed by 6 min of illumination with the blacklights (λmax ∼ 360 nm). This led to the formation of OH radicals, which initiated oxidation of the toluene and, in turn, led to particle formation via nucleation. The added NO scavenged any O3 and NO3 radicals (forming NO2) that formed during the reaction. The chemicals used in these experiments and their sources were as follows: α-pinene (99+%), β-pinene (99+%), Δ3-carene (99%), and toluene (99+%) were from Sigma-Aldrich, NO (99%) was from Matheson Tri-Gas, 2% O3 in O2 was from a Welsbach T-408 ozone generator, and methyl nitrite was synthesized according to the procedure of Taylor et al., [1980]. A Thermo Environmental Instruments Inc. 42C NO–NO2–NOx Analyzer and a Dasibi Model 1003-AH Ozone Monitor also were used in the smog chamber.

2.2. HTDMA

[8] Hygroscopic growth factors were measured at 30°C using a humidified tandem differential mobility analyzer (HTDMA) [Brechtel and Kreidenweis, 2000b]. Particles sampled from the environmental chamber passed through two bipolar charge neutralizers (Aerosol Dynamics Inc.), each containing four 2-month old 210Po strips (NRD Staticmaster 2U500). The particles were passed through a delay volume and subsequently were size-selected using a Differential Mobility Analyzer (DMA 1, TSI 3071, sheath-to-monodisperse flow ratio of 10:2 LPM). The delay volume was included to allow for possible dilution of particle concentrations; dilution was not needed for any of the SOA measurements. Relative humidity in the environmental chamber was low (RH <1%), and so it is expected that all measurements, both HTDMA and CCN, are for initially dry aerosol. The near-monodisperse aerosol flow was split, with equal amounts being sent to the CCN counter (discussed below) and to the humidified section of the HTDMA. In the HTDMA, particles were exposed to a controlled RH (5, 80, 85, 90, or >90%) using a Nafion (Perma Pure Inc.) humidifier. The resulting wet size distribution was then measured with a second DMA (DMA 2; sheath-to-monodisperse flow ratio of 5:1 LPM) humidified to the same RH as the sample stream, coupled to a condensation particle counter (CPC, TSI 3010). The total time that aerosol was exposed to the enhanced humidity was ∼6 seconds. Size distributions were fit to a theoretical model accounting for instrument transfer functions using the inversion algorithm of Zhou et al., [2002]. From these fits, we determined the hygroscopic growth factor, HGF, the ratio of the equilibrium geometric mean diameter at the indicated RH to that measured at RH ≈ 5%. This method yields an overall uncertainty in HGF of ±0.02 in absolute units [Carrico et al., 2005].

[9] The humidity of DMA 2 was measured for the sample and sheath flows using ROTRONIC HygroClip sensors (type S). The sensors were inserted into the flow through a Swagelok tee and are accurate to ±1.5% RH. Near the middle of the study, we discovered that the sensors were partially recessed from the flow, and, as such, the uncertainty in the RH measurement was greater than that of the sensors themselves. We compared HGF measurements of 100 nm ammonium sulfate particles taken throughout the study to predicted growth factors [Clegg et al., 1998] to determine the extent to which the placement of the sensors affected our measurements. Prior to repositioning the sensors, we determined the standard deviation (σ), 2σ = 4.7% RH; after repositioning, 2σ = 1.2% RH. Although only about half of the data were collected with the sensors recessed, we use the upper value (2σ = 4.7% RH) for uncertainty in our analysis.

[10] Note that RH, and not water activity (aw), is measured in the HTDMA. For very large droplet sizes, where the Kelvin effect is negligible, aw and RH are equivalent. For the small particles studied here, however, we must account for the Kelvin effect. The Köhler equation is employed to convert measured RH to aw for the discussion below:

equation image

where the Kelvin term is given in the exponential, S is the saturation ratio, MWw is the molecular weight of water, σsol is the surface tension of the solution (assumed to be equal to the surface tension of water, σw = 0.0725 J m−2), R is the universal gas constant, T is the temperature, ρw is the density of water, and D is the droplet diameter.

2.3. CCN

[11] A CCN instrument (Droplet Measurement Technologies, Inc., Model CCN-2) was used to measure CCN concentration as a function of dry particle size (Ddry) and processing supersaturation (s = RH − 100%); it is a streamwise gradient, continuous flow, single supersaturation instrument [Roberts and Nenes, 2005]. Total particle concentration entering the CCN counter was monitored using a CPC (TSI 3010). In the CCN counter, the particles are exposed to controlled supersaturations. Those particles that activate as cloud droplets grow to supermicron sizes. Upon exiting the processing region, the particles are sized and counted with an optical particle counter. All droplets larger than a certain diameter, typically chosen to be 1 μm, are considered CCN at a given supersaturation. The ratio of CCN to CN defines the activated fraction. Calibrations using size-selected ammonium sulfate particles were performed on an almost daily basis (variable but typically one calibration point per day). These data were used to calculate instrument supersaturation and the repeatability in the resulting critical supersaturations: 2σ = 0.042%. Our calculation of the supersaturation required for ammonium sulfate particle activation is described further in section 3.2.

[12] Because aerosol size distributions were quite narrow (shown below), the contribution of larger, multiply charged particles transmitted through the DMA was appreciable for diameters on the ascending branch of the size distribution. Accounting for these larger particles is necessary in order to unambiguously identify the critical supersaturation for a given particle diameter. For ∼80% of the measurements presented here, data were corrected for multiply charged particles by fitting to a model describing the transfer of mobility-selected, charge-equilibrated particles through an ideal DMA/CCN counter setup, based on measured size distributions [Petters et al., 2006b].

2.4. Experimental Procedure

[13] A key parameter when comparing these studies is time, as both hygroscopicity and CCN activity of SOA can change with time, with changes attributed to the formation of oligomers [Baltensperger et al., 2005; VanReken et al., 2005]. Variations in SOA mass concentration may also alter the partitioning of semivolatile species [Donahue et al., 2006]. Here we outline the general timeline for experiments. SOA was generated using the procedures noted above. After the size distribution stabilized, HTDMA measurements were conducted while simultaneously stepping through a range of supersaturations (S-scan) at fixed dry diameter in the CCN instrument; these measurements lasted 1–2 hours. Next, we stepped through a range of dry diameters (D-scan) at fixed supersaturation in the CCN instrument. Stepping diameter was considerably faster, and higher resolution activation curves could be achieved using D-scans. Scans were conducted over the range of s = 0.13 to 0.92%, with each scan lasting ∼20 min. Approximately 10 measurements were made for each supersaturation or size selected. The supersaturations explored often were limited by the low number concentrations of particles at smaller and larger sizes. The total length of each experiment was 3–5 hours.

3. Results and Discussion

3.1. HTDMA Measurements

[14] Results from the HTDMA measurements are shown in Figure 2, with hygroscopic growth factor presented as a function of water activity. Data are shown for each of the precursors, indicating a consistent HGF-aw relationship for all of the SOA formed. Water uptake from toluene-derived SOA is similar to that of the monoterpenes for aw < 0.87 but is somewhat larger at higher aw. It has been observed previously that growth factor is inversely proportional to precursor molecular weight [Varutbangkul et al., 2006]. The range of measured growth factors at 85 ± 1% RH is 1.01 to 1.07 ± 0.02, in agreement with a previous measurement of 1.07 ± 0.01 at 84% RH for SOA formed from ozonolysis of α-pinene [Virkkula et al., 1999]. Our measurements are slightly lower than measurements by Saathoff et al., [2003] of 1.106 ± 0.002 at 85% RH, taken 6 hours after introduction of α-pinene. This difference simply may be the result of the uncertainty of the measurement or differences in experimental procedure which may affect the nature of the SOA. In particular, previous authors [Baltensperger et al., 2005; Saathoff et al., 2003; Varutbangkul et al., 2006] have noted that growth factors of SOA can increase over the first few hours of the experiment. We did not observe any temporal trends in the data after the particle size distribution stabilized. However, our measurements were not aimed at identifying time-dependent processes, and all but one of the HTDMA experiments were completed within 2 hours of SOA formation.

Figure 2.

Hygroscopic growth as a function of water activity for all of the compounds studied. Representative error bars are shown for Δ3-carene at aw = 0.92, based on the worst case uncertainty for RH. The best fit to the data is shown as a solid line, following Kreidenweis et al., [2005], and the upper and lower limits of the fit are shown as the shaded regions, based on the uncertainty of the RH and HGF. Also shown is an extrapolation of the CCN measurements based on κ = 0.1 (dotted line) and the best fit to CCN and HTDMA data using FHK theory (dashed-dotted line).

3.2. CCN measurements

[15] We now explore the potential impact of time-dependent changes in CCN activity for these measurements using data from one α-pinene experiment. Particle size distributions are shown in Figure 3a, with geometric diameter increasing from 0.072 to 0.107 μm during the experiment, while geometric standard deviation increases from 1.29 to 1.35 concurrently. An S-scan was first conducted for 100 nm particles, starting ∼30 min after introduction of α-pinene to the chamber and lasting ∼65 min. Results are shown in Figure 3b. The data are fit with a cumulative normal distribution to determine the critical supersaturation, sc, which is represented by the supersaturation at which 50% of the particles activate as cloud droplets for a given dry diameter. For these measurements, sc = 0.38% for 100 nm particles. Alternatively, the dry diameter at which 50% of the particles activate as cloud droplets, D50, can be determined for a given supersaturation. A D-scan was carried out at s = 0.4% from 2:17 to 2:34 (hours:minutes) after introduction of α-pinene. Results from these measurements are shown in Figure 3c. As can be seen in the figure, there is a shoulder apparent in the data in an otherwise sigmoidally shaped curve. This shoulder is due to multiply charged particles and was accounted for as discussed in the previous section. For the D-scan, D50 is determined to be 90 nm at s = 0.4%. Although the D-scan and S-scan are not carried out at identical conditions, the measured CCN activity is consistent. Finally, a D-scan was repeated at s = 0.4% from 4:07 to 4:23 after introduction of α-pinene. Data are shown in Figure 3d, D50 = 89 nm, nearly identical to the earlier measurements. Contrary to previous results [VanReken et al., 2005], we observed no significant temporal changes in CCN activity within the timeframe of our experiments.

Figure 3.

Size distributions and CCN activation curves for α-pinene over the course of 1 experiment. (a) Size distributions as a function of time, starting 15 min after introduction of α-pinene. Size distributions were charge corrected up to +6 charges. White color indicates spectral densities <103 cm−3. (b) S-scan for 100 nm dry particles, 0:30–1:35 (hour:minutes) after introduction of α-pinene; (c) D-scan for s = 0.4%, 2:17–2:34 after introduction of α-pinene; (d) D-scan for s = 0.4%, 4:07–4:23 after introduction of α-pinene. CCN activation in Figure 3b and 3d are fit with a cumulative normal distribution. CCN activation in Figure 3c is fit to a model that accounts for the contribution of multiply charged particles.

[16] Results for the CCN measurements of SOA are summarized in Figure 4, and data are tabulated in Table 1. Data are shown as open symbols, with sc plotted as a function of Ddry. Also, shown are data from the ammonium sulfate calibrations. As was the case for the HTDMA, CCN activation is similar for all of the SOA studied. However, contrary to the HTDMA measurements, which showed very little water uptake at RH < 95%, these results suggest that SOA is fairly hygroscopic and can serve as CCN at accumulation mode particle sizes (0.06 μm < Ddry < 0.2 μm) and typical atmospheric supersaturations (0.1% < s < 2%). The relationship between the HTDMA and CCN data, and atmospheric implications, are discussed below.

Figure 4.

Critical supersaturation as a function of dry particle diameter for all SOA measurements (large symbols) and the ammonium sulfate calibration data from this study. The uncertainty in the measurement (2σ = 0.042%) is shown for two α-pinene data points (Ddry = 0.056 μm, sc = 0.919%; Ddry = 0.223 μm, sc = 0.125%). The error bar shown at high supersaturation is roughly equal size of the symbol, while the error bar at low supersaturation appears much larger, because of the fact that the data are on a log plot. Also shown is the theoretical limit for insoluble, but wettable particles. A best fit to the data using equation 2 is shown as a thick dotted line, for κ = 0.1. Data from the work of VanReken et al., [2005] are shown as small open symbols with solid lines for measurements taken within the first 5 hours of the experiment (comparable to this study) and dashed lines for measurements taken more than 12 hours later. Data from the work of Huff Hartz et al., [2005] are shown as filled gray symbols.

Table 1. CCN Measurements of SOA
SOA precursorDdry, micronsSc, %Start Timea (hour:minute)Scan Type
  • a

    Time after introduction of SOA precursor or, for toluene, time after lights turned off.


[17] The data can be fit using an alternative form of equation (1) and the parameter, κ, which provides a quantitative comparative measure of CCN activity [Petters and Kreidenweis, 2006]:

equation image

[18] The κ term characterizes aw as a function of aerosol composition, assuming volume additivity. Values range from κ = 0.6 for ammonium sulfate to κ = 0 for insoluble, but wettable, particles. Critical supersaturation is calculated as the maximum of the S(D) curve. Equation (2) predicts a slope of −3/2 when log(sc) is plotted as a function of log(Ddry). We model our data using equation (2). Minimizing the χ2 statistic yields a best fit for κ = 0.10, shown as the thick dotted line in Figure 4, with more than 95% of the observations falling between κ = 0.10 ± 0.04. The good agreement between the SOA formed from the different precursors and the absence of a strong temporal trend suggests, at least initially, that Köhler theory with a simple one-parameter representation of aw can be used to describe aerosol-water interactions for the multiple organic components found in SOA. This simplified treatment will be explored further in section 3.3.

[19] Our data can be compared with previous measurements of CCN activity of SOA formed from α-pinene, β-pinene, and Δ3-carene. Data from the SOA experiments of VanReken et al., [2005] are shown in Figure 4 for measurements taken within the first 5 hours of the experiment (comparable to this study) and taken more than 12 hours later. It should be noted that in the VanReken et al. study, D50 was determined for an integrated polydispersed distribution of particles, instead of from a scan with quasi-monodisperse aerosol as in this study; manifestations of any size-dependent differences in particle composition may differ for the two studies. Although the earlier measurements from VanReken et al. are comparable to the data presented here, measurement uncertainties for the two studies are insufficient to explain some of the differences between the two data sets. Further, the CCN activity reported by VanReken et al. was dependent on the SOA precursor, whereas we saw no such dependence. Data from VanReken et al. from all times also follow a significantly steeper slope than measured in this study; that is, sc is much more sensitive to Ddry. Finally, the SOA monitored near the end of each of their experiments was considerably less CCN active. VanReken et al. suggested that gradual oligomerization of the SOA could account for some of this behavior. VanReken et al. did not report hygroscopic growth data for their SOA, but it has been suggested that the steep slopes could result from deliquescence-limited CCN activation [Kreidenweis et al., 2006]. However, with increasing evidence that SOA takes up water at relative humidities below 100% (this study, [Saathoff et al., 2003; Varutbangkul et al., 2006; Virkkula et al., 1999]), deliquescence-limited activation seems unlikely to explain the VanReken et al. slopes.

[20] CCN activation results from Huff Hartz et al., [2005] also are shown in Figure 4. Data shown are limited to experiments in which 2-butanol was not used as an OH radical scavenger, as it has been shown that the presence and type of scavenger may affect SOA composition [Docherty et al., 2005; Docherty and Ziemann, 2003]. By culling the data in this way, the remaining data shown in the figure are mostly for α-pinene, although all three monoterpenes were measured in their study [Huff Hartz et al., 2005]. CCN measurements from Huff Hartz et al. were made between 30 min and 5 hours after SOA formation, comparable to this study. However, these authors used a method of self-seeding to control particle size. Although their measurements are generally consistent with our observations, particularly at s = 1 and 0.3%, differences exist for some of the measurements which fall outside of the stated uncertainties of the two studies. Further, the slope in the data from Huff Hartz et al. from s = 1% to s = 0.6% is relatively shallow, much less than the −3/2 slope expected from Köhler theory. In contrast, from s = 0.6% to s = 0.3%, the slope is quite steep, comparable to that observed by VanReken et al., [2005]. Such behavior may result for particles which have size-dependent composition or for time-dependent processes which affect CCN activity over the course of the measurements.

3.3. Linking HTDMA Data and CCN Data

[21] For fully dissolved, single-phase aqueous solutions, aerosol-water interactions can be described fully by the Köhler equation [equation (1)], both for hygroscopic growth below water saturation and CCN activation above water saturation. Empirical [Tang and Munkelwitz, 1994], semiempirical [Brechtel and Kreidenweis, 2000a; Koehler et al., 2006; Kreidenweis et al., 2005], and theoretical [Clegg et al., 1998, 2001] descriptions have been put forth to parameterize aw for a variety of atmospheric solutes, including polymers [Petters et al., 2006a] and for multicomponent particles where some components are only sparingly soluble in the aqueous solution [Raymond and Pandis, 2002; Shulman et al., 1996]. Here we use several of these parameterizations in an attempt to link measured hygroscopicity and CCN activation. We also explore some of the physical processes represented in these formulations. Table 2 provides a list of the parameterizations to aid in the discussion.

Table 2. Linking HTDMA and CCN Data From This Study
Approachaw ParameterizationσνResultFit Parameters
equation imageequation image0.0725 N/mN/AOverpredicts HGFequation image = 0.1
HGF fitequation image0.0725 N/mIn polynomialSc overpredicteda = −0.3745, b = 0.9678, c = −0.5857
HGF fitequation image0.0725 N/mMax ΔνSc overpredicted 
HGF fitequation imageVariableIn polynomialUnrealistic values of σ required 
Limited solubilityequation image0.0725 N/m1Incorrect shape of curveγ = 1; ɛ ≤ 1a
FHKequation image0.0725 N/mN/AModels HTDMA and scψ = 0.0196, β = −0.3883, νFHK = 0.8942b

[22] We begin with the parameterization described in equation (2). Using κ = 0.1, determined from the CCN measurements, we calculate water uptake for conditions in the HTDMA. These calculations are shown as a dotted line in Figure 2. As can be seen in the figure, there is poor agreement between the predicted growth factors and the measurements, suggesting that equation (2) does not accurately characterize SOA throughout the aw range of interest. We also note that the growth factors predicted for κ = 0.1 are greater than those measured in previous studies [Saathoff et al., 2003; Virkkula et al., 1999].

[23] Kreidenweis et al., [2005] and Koehler et al., [2006] have used HTDMA hygroscopicity data to parameterize water activity:

equation image

where a, b, and c are adjustable parameters. Here we fit all of the HTDMA data using this parameterization, shown in Figure 2 as a solid line. The shaded region represents the range of the fit, determined by shifting all of the data points to lower aw/higher HGF and higher aw/lower HGF, based on the stated uncertainties, and using equation (3) to fit the shifted data. From this relationship, sc is then calculated as the maximum of the S(D) curve (sc = [S(D)max − 1] × 100) described by equation (1). The sc-Ddry relationship thus obtained is shown in Figure 5 as a thick solid line; the shaded regions indicate the upper and lower limits from the fit. For comparison, the CCN data from this study also are plotted. Clearly, the calculated critical supersaturations do not accurately represent the measurements, even given the relatively large uncertainty noted for the HTDMA measurements. Specifically, the SOA particles are significantly more active as CCN than is predicted from the HTDMA measurements. Thus, equation (3) also appears insufficient in describing aw throughout the range of interest.

Figure 5.

SOA and ammonium sulfate measurements from Figure 4. Shown as a thick solid line is an extrapolation of the HGF data, following the method of Kreidenweis et al., [2005], with the shaded regions indicating the upper and lower bounds from the fit. The thin dashed line is calculated from Köhler theory using the assumptions of Huff Hartz et al., [2005] for limited solubility of the organic. Finally, the best fit to the CCN and HTDMA data using FHK theory is shown as a dashed-dotted line.

[24] Equations (2) and (3) can both accurately model the HTDMA data or the CCN data, but neither of these parameterizations can predict water uptake for both sets of measurements. The problem is rooted in the inability to describe aw as a function of particle composition throughout the aw range of both instruments. To examine this issue further, in Figure 6, we plot aw versus solute volume fraction, ϕ = HGF−3 (assuming volume additivity and particle sphericity). Solute volume fraction is defined as the ratio of solute volume to total volume of the particle and can be calculated from the best fit to the HTDMA data [equation (3)] and the best fit to the CCN data [equation (2)]. In both cases, the data are plotted as solid lines over the aw range at which the measurements were made, and as dotted lines were extrapolated outside of the range of the measurements. Clearly, the two fits do not create a continuous relationship. This discontinuity is reflected in the inability to use either aw expression to describe CCN activation and hygroscopic growth. We now examine several possible reasons why equations (2) and (3) fail to accurately capture the aw-ϕ relationship and explore other possible parameterizations.

Figure 6.

Water activity dependence on solute volume fraction (ϕ). The best fits to the HTDMA data [equation (3), thin line] and the CCN data [equation (2), thick line] are plotted as solid lines over the aw range at which the measurements were made, and as dotted lines where extrapolated outside of the range of the measurements. Note that the fit for the CCN data are limited to aw > 0.996 and ϕ  < 0.06, shown in the upper left hand corner of the figure. Other aw parameterizations include that for limited solubility (dashed line) [Huff Hartz et al., 2005] and for FHK theory (dashed-dotted line) [Petters et al., 2006a].

[25] The polynomial in equation (3) represents the relationship between aw and particle composition. For dissociating species, this includes a dependence on the number of ions present in solution, ν.  In equation (3), it is assumed that variations of ν with composition are captured by the polynomial and can be extrapolated to higher aw. In equation (3), the polynomial is related to ν as [Kreidenweis et al., 2005]:

equation image

where Φ is the molal osmotic coefficient, MWs is the molecular weight of the solution, and ρs is the density of the solution. It is unclear if variations in concentrated (HTDMA) solutions can accurately describe variations for more dilute (CCN) solutions. We explored a range of possible dissociation constants (K = 10−5 to 1) to determine ν as a function of composition to examine whether this could explain the discontinuity evident in Figure 6. For these calculations, we assumed that the compound dissociates into two species. Even for species which are only slightly dissociated at the more concentrated conditions of the HTDMA and almost completely dissociated at the dilute concentrations of the CCN counter, that is, for a maximum change in the number of ions in solution over the measurement range (Δν = 0.7), such a change cannot explain the apparent inconsistency between the HTDMA and CCN measurements.

[26] The inconsistency between HTDMA and CCN measurements may not be related to poor parameterizations of aw but rather may result from assumptions regarding the Kelvin term in equation (1). Specifically, this discrepancy may be related to the assumption that the organic solution surface tension is equal to the surface tension of pure water. Atmospherically relevant organic species have been shown to lower the surface tension of water [Dinar et al., 2006; Facchini et al., 1999; Shulman et al., 1996], reducing the Kelvin term and leading to lower sc. Using equation (3) to parameterize aw from the HTDMA measurements, we explored a range of values for σsol. To predict the measured CCN activity from the HTDMA data requires σsol = 0.030 J m−2. Accounting for surfactant partitioning [Sorjamaa et al., 2004] may require even lower bulk surface tensions. Bulk measurements of relevant organic solutions [Dinar et al., 2006; Huff Hartz et al., 2006; Shulman et al., 1996] suggest that σsol is not likely to be this low, and so it does not appear that surface tension effects can explain the data.

[27] The previous expressions of aw have assumed that all of the solute is fully dissolved in the droplet, which may not be representative of aqueous solutions of organic species in SOA. Huff Hartz et al., [2005] used a modified version of Köhler theory [Raymond and Pandis, 2002], which allows for substances with limited solubility. Using this modified theory and the assumptions given in the work by Huff Hartz (saturation concentration, Csat = 0.11 g SOA/g H2O; ρSOA = 175 g mol−1), the aw-ϕ relationship was determined and is plotted in Figure 6 as a dashed line. Although their model does create discontinuity in the aw-ϕ relationship, the shape of the curve does not represent the data. Further, no water uptake is predicted at HTDMA humidities, consistent with deliquescence-limited activation. Critical supersaturation also was determined for a range of Ddry from their model and is plotted as a thin dashed line in Figure 5, showing poor agreement with the data. Deliquescence-limited activation seems unlikely to cause the discrepancy in the measurements.

[28] All of the expressions described thus far are based on Raoult’s law, with the basic assumption that aw scales with mole fraction of water in solution. However, this assumption fails for cases in which the solute and solvent have very different molecular sizes [Prausnitz et al., 1999]. Recent studies [Baltensperger et al., 2005; Docherty et al., 2005; Gao et al., 2004] have presented evidence for the presence of oligomers in SOA formed from monoterpene ozonolysis in the absence of seed particles, similar to the conditions employed for our experiments. As such, we attempt to model the aw-ϕ relationship using Flory-Huggins-Köhler (FHK) theory, developed for modeling CCN activation from polymer solutions [Petters et al., 2006a]. FHK theory parameterizes water activity as follows:

equation image

where ϕ is the volume fraction of the polymer, f is the chain segment number and is defined as the ratio of molecular volumes of the polymer and solvent, and χ is the semiempirical Flory-Huggins interaction parameter. In this expression, χ is dependent on solution composition and can be parameterized as a function of ϕ [Petters et al., 2006a; Wolf, 2003] using HTDMA and CCN measurements. To do so requires knowledge of the chain segment number. We evaluated a range of values for f to determine a best fit for the χ-ϕ relationship, using data from the HTDMA and CCN measurements. The best fit yields f = 10. For comparison, Docherty et al., [2005] suggested that reactions of α-pinene, β-pinene, and Δ3-carene with O3 form peroxyhemiacetal oligomers, with an average oligomer molecular weight of ∼300 g mol−1. Using this value, an assumed density of 1.2 g cm−3 for the aerosol [Bahreini et al., 2005] and the molar volume of water of 18 cm3 mol−1 for the solvent yields f = 14, generally consistent with the best fit to our data. We note here that to use FHK theory, we must assume that all of the solute molecules are oligomers, which may not be the case. Nonetheless, using f = 10, and the related χ-ϕ relationship, we characterize ϕ, and consequently HGF, as a function of aw. The ϕ-aw relationship obtained is plotted in Figure 6 as a dashed-dotted line. Interestingly, FHK theory is able to capture the seemingly discontinuous relationship apparent in the data. This fit also is plotted for hygroscopic growth data (Figure 2) and CCN activation data (Figure 5) as dashed-dotted lines. The obtained fit shows excellent agreement with the data. However, we stress that both HTDMA data and CCN data were used to constrain the fit, and, as such, the good agreement simply may reflect the flexibility of equation (5) and the associated χ-ϕ relationship. Therefore FHK theory had no predictive value to obtain CCN activity from the HTDMA measurements. Nevertheless, this is the only parameterization we found that could adequately describe both the HTDMA and CCN measurements. Petters et al., [2006a] also suggested the possibility of a miscibility gap for polymerized organic aerosol, which could explain such a step function. Unfortunately, such a transition cannot be detected using the current methods because wet diameters in the 0.92 < aw < 0.99 range are not available.

[29] We have explored quantitatively a number of possibilities to explain the inconsistency between the HGF and CCN data. At least two other explanations exist which we summarize qualitatively here. SOA is composed of many products with varying solubilities and volatilities. Multicomponent aerosols can exhibit hygroscopicity which is discontinuous, with nondeliquescent species absorbing water prior to deliquescence of other components [Choi and Chan, 2002]. Such changes in water uptake behavior, particularly if they occur at water activities greater than those measured in the HTDMA and lower than those measured in the CCNC, could explain the discontinuity in our data. Alternatively, the extent of partitioning of each semivolatile organic species may depend on the amount of water in the condensed phase [Seinfeld and Pankow, 2003], with a greater fraction of water-soluble species present at higher humidities. Such partitioning might also result in the discontinuity observed in Figure 6. These explanations are difficult to examine without measurements in the range aw = 0.92–0.99. In any case, efforts to use HGF data to predict CCN activity, and vice versa, appear problematic for the complex mixtures of organic species expected in SOA.

4. Atmospheric Implications

[30] The relatively hygroscopic CCN behavior observed for SOA is drastically different than has been observed for primary organic aerosol and primary organic aerosol which has undergone oxidation [Petters et al., 2006d], which behave more like nonhygroscopic species. As such, we expect that the two would behave markedly different in the atmosphere. To explore atmospheric implications, we also must consider that the less volatile species which compose SOA often condense onto existing particles, and so we expect SOA to exist as internally mixed organic-inorganic particles, with a wide range of organic fractions. In Figure 7, we compare hygroscopicity of internally mixed particles containing ammonium sulfate (κAS = 0.6) and organic species for the full range of organic volume fractions, ɛorg. The organic species include SOA (κorg = 0.1) and organics which are insoluble but wettable (κorg = 0). The hygroscopicity of the mixed particles, κmixed, can be calculated [Petters and Kreidenweis, 2006] as:

equation image

Also shown in the figure are calculations based on a parcel model [Cohard et al., 1998; Petters et al., 2006d] assuming a unimodal lognormal size distribution (Dg = 0.1 μm, and σg = 1.5) for a range of updraft velocities (1–10 m s−1) and particle number densities (400–4000 cm−3) (shaded region) with two limiting cases shown as dotted lines, low updraft velocity (w = 1 m s−1) and low particle concentrations (N = 400 cm−3), and much higher updraft velocity (w = 10 m s−1) and high particle concentrations (N = 4000 cm−3). The change in fraction of particles activated as cloud droplets is shown, calculated as the ratio of activated droplets from the model output for two different mixing assumptions (κorg = 0 versus κorg = 0.1 for the organic fraction). For all cases, it appears that the hygroscopicity of the organic is unimportant for ammonium sulfate dominated aerosol (ɛorg < 0.6). Further, for the convective case (w = 10 m s−1), the hygroscopicity of the organic has little effect on the fraction of particles which activate. It is only for the low updraft/low particle numbers with large organic volume fractions that the hygroscopicity of the organic plays a role. This results because the relative importance of the hygroscopicity of the organic is increased for large organic volume fractions. This effect becomes even more important for low updraft velocities, when the supersaturation balance is more sensitive to the condensational sink. For example, for the case ɛorg = 0.8 with w = 1 m s−1 and N = 400 cm−3, 20% fewer droplets are expected to activate for the insoluble organic-sulfate aerosol than for the SOA-sulfate aerosol. For the low number concentrations in this case, such a reduction in cloud droplet number can alter both the microphysical [Twohy et al., 2005] and radiative [Platnick and Twomey, 1994] properties of the cloud.

Figure 7.

Aerosol hygroscopicity parameter, κmixed, as a function of organic volume fraction for an insoluble organic (κorg = 0, dashed line) and SOA (κorg = 0.1, solid line) internally mixed with ammonium sulfate. Also shown is the change in fraction of particles that are activated as cloud droplets assuming κorg = 0 versus κorg = 0.1 for the organic fraction for a range of updrafts and aerosol number densities (shaded), with two limiting cases shown as dotted lines, (1) constant updraft velocity, w = 1 m s−1 and N = 400 cm−3, and (2) w = 10 m s−1 and N = 4000 cm−3. Calculations are based on a parcel model assuming a unimodal lognormal size distribution (Dg = 0.1 μm and σg = 1.5).

5. Summary and Conclusions

[31] Measurements of water uptake and CCN activity were conducted on SOA formed from α-pinene, β-pinene, Δ3-carene, and toluene. Measured CCN activation was comparable for all of the species studied and occurred for particle sizes and supersaturations commonly attained in the atmosphere. Water uptake also was invariant between compounds and was generally consistent with previous measurements. However, there was an apparent discrepancy between the measured CCN activity and the HTDMA measurements, based on aw expressions grounded in Raoult’s law. It seems unlikely that a mole-fraction-based aw representation can capture the full range of behavior observed, unless either a phase separation occurs or other species go into solution at higher aw. Otherwise, the polymer aw expression, which is volume fraction based, seems more appropriate for capturing this behavior. Composition-dependent water activity measurements which bridge those used in HTDMA and CCN studies (aw = 0.92–0.99) are needed to further elucidate water uptake of aerosols containing complex mixtures of organics such as SOA.

[32] SOA is significantly more hygroscopic than primary organic aerosol. However, we show that for mixed organic-inorganic aerosol which contain a substantial inorganic component (ɛinorganic > 0.4), the hygroscopicity of the organic component is unimportant in determining cloud droplet activation for a range of atmospheric conditions. We find that droplet activation behavior of mixed particles containing SOA differs from that of mixed particles for which the organic component is nonhygroscopic only for cases with low updraft velocities in which the aerosol composition is dominated by organics.


[33] We thank Timothy VanReken for sharing his CCN data. This research was supported by the Office of Science (BER), US Department of Energy, and US National Science Foundation grant ATM-0436196.