The aerosol and cloud chamber AIDA (Aerosol Interactions and Dynamics in the Atmosphere) of the Karlsruhe Institute of Technology has been used to quantify the deposition mode ice nucleation ability of airborne crystalline sodium chloride dihydrate (NaCl ∙ 2H2O) particles with median diameters between 0.06 and 1.1 µm. For this purpose, expansion cooling experiments with starting temperatures from 235 to 216 K were conducted. Recently, supermicron-sized NaCl ∙ 2H2O particles deposited onto a surface have been observed to be ice-active in the deposition mode at temperatures below 238 K, requiring a median threshold ice saturation ratio of only 1.02 in the range from 238 to 221 K. In AIDA, heterogeneous ice nucleation by NaCl ∙ 2H2O was first detected at a temperature of 227.1 K with a concomitant threshold ice saturation ratio of 1.25. Above that temperature, the crystallized salt particles underwent a deliquescence transition to form aqueous NaCl solution droplets upon increasing relative humidity. At nucleation temperatures below 225 K, the inferred threshold ice saturation ratios varied between 1.15 and 1.20. The number concentration of the nucleated ice crystals was related to the surface area of the seed aerosol particles to deduce the ice nucleation active surface site (INAS) density of the aerosol population as a function of the ice supersaturation. Maximum INAS densities of about 6 ⋅ 1010 m−2 at an ice saturation ratio of 1.20 were found for temperatures below 225 K. These INAS densities are similar to those recently derived for deposition mode ice nucleation on mineral dust particles.
 Recently, Wise et al.  have unraveled a potential new mechanism for ice formation in the troposphere, namely depositional ice nucleation on crystalline hydrated sodium chloride particles. In the atmosphere, crystalline sodium chloride particles will form once an air parcel containing aqueous NaCl solution droplets is exposed to a relative humidity (RH) below 38–44%. The latter represents the range of the efflorescence relative humidity (ERH) of NaCl for temperatures between 249 and 298 K [Cziczo and Abbatt, 2000; Koop et al., 2000a]. Upon efflorescence, two crystalline species of NaCl might form, namely anhydrous NaCl and sodium chloride dihydrate (NaCl ∙ 2H2O). According to the bulk phase diagram for the sodium chloride-water system, NaCl ∙ 2H2O is the thermodynamically stable solid phase below 273.3 K, whereas it is anhydrous NaCl above 273.3 K [Martin, 2000]. Previous studies, however, have shown that down to 253 K homogeneous crystallization of aqueous NaCl only leads to the formation of anhydrous NaCl [Cziczo and Abbatt, 2000], and that nucleation of the thermodynamically stable sodium chloride dihydrate is only triggered by heterogeneous crystallization on available surfaces like ice or oxalic acid dihydrate at such elevated temperatures [Koop et al., 2000a; Wagner et al., 2011]. Wise et al.  have extended the temperature range for crystallization experiments with aqueous NaCl solution droplets down to about 221 K. The investigated droplets typically had a diameter of about 5 µm and were deposited onto a hydrophobic quartz disk that was placed into an environmental cell equipped with a combination of optical microscope and Raman spectrometer. The authors have detected that there is a transition regime between 252 and 236 K where the composition of the crystallized solution droplets changes from only anhydrous to only dihydrate NaCl particles, meaning that below a temperature of 236 K exclusively the formation of NaCl ∙ 2H2O was detected.
 Once crystalline anhydrous and dihydrate NaCl particles are formed in the atmosphere at low RH, they could act as heterogeneous ice nuclei in the deposition mode when again exposed to an increasing relative humidity. This requires that the heterogeneous ice nucleation onset is below the deliquescence relative humidity (DRH) of the crystals, as otherwise aqueous NaCl solution droplets would be formed which can only freeze by homogeneous nucleation at elevated supersaturation levels. The DRH of anhydrous NaCl is rather insensitive to a change in temperature between 298 and 249 K and amounts to about 75% [Koop et al., 2000a]. For NaCl ∙ 2H2O, the deliquescence curve is obtained from the sodium chloride-water phase diagram shown in the relative humidity – temperature space by connecting the NaCl/NaCl ∙ 2H2O peritectic at 273.3 K and 75% RH with the ice/NaCl ∙ 2H2O eutectic at 251.9 K and 81% RH [Koop et al., 2000a]. In a previous study [Wagner et al., 2011], we have detected a DRH value of 82% for crystalline NaCl ∙ 2H2O at 242 K which is in good agreement with the extrapolation from the phase diagram. Wise et al.  have found DRH values for NaCl ∙ 2H2O between 76.6 and 93.2% RH in the temperature range from 235 to 257 K.
 Concerning the competition between deliquescence and heterogeneous ice nucleation, Wise et al.  have shown that above a temperature of 239 K, only deliquescence of the deposited NaCl ∙ 2H2O crystals occurred, whereas below 235 K, the crystals only heterogeneously nucleated ice in the deposition mode. In the narrow temperature regime from 235 to 239 K, both processes were observed. The NaCl ∙ 2H2O crystals thereby proved to be remarkably efficient ice nuclei in the temperature range from 221 to 238 K, yielding an average value of 1.02 ± 0.04 for the ice saturation ratio, Sice, at the onset of ice formation. For anhydrous NaCl particles, a slightly higher average value for Sice of 1.11 ± 0.07 was found.
 The high ice nucleation ability of NaCl ∙ 2H2O might represent an example for the chemical bond requirement in heterogeneous ice nucleation [Pruppacher and Klett, 1997], meaning that the water molecules in the crystal are preferential sites for the further deposition of water vapor from the gas phase. Also a chemically different hydrate species, namely oxalic acid dihydrate, has recently been observed to be a partly very efficient ice nucleus in the deposition mode [Kanji et al., 2008; Wagner et al., 2010]. Additionally, the microscope images recorded by Wise et al.  show that the dihydrate particles have a higher degree of surface roughness compared to the anhydrous crystals which is another factor that could add to the particular ice nucleation efficiency of NaCl ∙ 2H2O. Based on a model simulation where the trajectories of air parcels with initially aqueous NaCl were tracked through the tropical upper troposphere, Wise et al.  have estimated that dihydrate crystals are present 40–80% of the time in the upper troposphere at temperatures below 220 K and could therefore play a role in cirrus cloud formation in view of their high ice nucleation ability.
 Motivated by the Wise et al.'s  study and based on previous experiments at our own facility with NaCl and internally mixed NaCl/oxalic acid particles [Wagner et al., 2011], we have performed a dedicated series of crystallization and ice nucleation experiments with airborne NaCl particles of median diameters between 0.06 and 1.1 µm in the temperature range from 235 to 216 K. The experiments were part of a measurement campaign that was conducted at the coolable aerosol and cloud chamber AIDA (Aerosol Interactions and Dynamics in the Atmosphere) of the Karlsruhe Institute of Technology. Each experiment at a given temperature was started with the addition of aqueous NaCl solution droplets to the AIDA chamber. Crystallization of the injected droplets was then monitored at constant temperature and relative humidity via in situ depolarization and infrared extinction spectroscopy measurements over a timescale of several hours. After crystallization was complete, the particle ensemble was probed on its deliquescence and ice nucleation ability in an expansion cooling experiment during which ice supersaturated conditions were established in the chamber interior. We chose to present the results from these measurements in two separate articles. In a previous article, we have already analyzed the temperature-dependent partitioning between anhydrous and dihydrate particles upon crystallization of the injected aqueous NaCl solution droplets [Wagner et al., 2012]. A brief summary of our major findings is given in the next paragraph. The present article describes the additionally performed ice nucleation experiments with the crystalline NaCl and NaCl ∙ 2H2O particles.
 For exploring the crystallization behavior of the aqueous NaCl solution droplets, we have quantitatively analyzed the infrared extinction spectra of the crystallized particles to infer the relative number fractions of NaCl and NaCl ∙ 2H2O particles. Extinction spectra of anhydrous NaCl particles at mid-infrared wavelengths solely reveal a structureless scattering signature, whereas NaCl ∙ 2H2O particles additionally give rise to characteristic absorption bands due to the vibrational modes of the water molecules in the crystals. As a prerequisite for the quantitative analysis of the temperature-dependent partitioning between the two solid phases of sodium chloride from the infrared spectra, we have additionally deduced the so far unknown infrared optical constants of NaCl ∙ 2H2O. In agreement with Wise et al. , the analysis of the AIDA experiments also showed that there is a narrow temperature range where the composition of the crystallized particles almost completely changes from anhydrous to dihydrate NaCl particles. Under our experimental conditions, this transition regime was however shifted to lower temperatures by an amount of about 13 K when comparing the temperatures where 50% of the particles have crystallized as NaCl ∙ 2H2O. This means, e.g., that at a temperature of 235 K where Wise et al.  have exclusively detected the formation of NaCl ∙ 2H2O, in AIDA only about 7% of the injected NaCl solution droplets have crystallized as the dihydrate. This different trend is in accordance with a former AIDA crystallization experiment with aqueous NaCl particles at 244 K [Wagner et al., 2011], in which dihydrate formation could not be detected at all. The temperature shift of the NaCl-NaCl ∙ 2H2O transition regime between the AIDA and the Wise et al.'s  experiments could be related to the difference in the size of the investigated particles or to the fact that the particles were either airborne or deposited onto a surface.
 Note that Wise et al.  in their article do not explicitly refer to sodium chloride dihydrate as the low-temperature phase of crystalline NaCl which forms below 236 K. Instead, they introduce the general term “hydrated form” of NaCl. This is because they have observed slight spectral discrepancies in their recorded Raman intensities in comparison with the spectrum from a preceding work which supposedly was due to NaCl ∙ 2H2O [Dubessy et al., 1982; Wise et al., 2012]. They also underline, however, that this might simply be due to the different temperatures at which the Raman spectra were collected. Our recorded infrared spectra showed good agreement with those previously published for NaCl ∙ 2H2O [Mutter et al., 1959; Schiffer and Hornig, 1961], indicating that it is indeed the dihydrate that forms in the investigated temperature range. It is therefore reasonable to assume that also Wise et al.  have observed the formation of NaCl ∙ 2H2O, and we therefore chose to use this assignment when referring to their results in our manuscript.
 Concerning the subject of the present article, namely the heterogeneous ice nucleation ability of the crystallized NaCl and NaCl ∙ 2H2O particles, both the experimental approach from Wise et al.  and the AIDA expansion cooling technique have their inherent advantages and can thus be considered as complementary methods to fully explore the ice nucleation behavior in this system. The appeal of the procedure by Wise et al.  lies in the chemical identification of the particle that has acted as the ice nucleus. The AIDA chamber experiments probe an ensemble of airborne particles with a well-characterized size distribution which allows a more quantitative analysis of the heterogeneous ice nucleation ability with respect to the surface area of the seed aerosol particles, as, e.g., in the framework of the ice nucleation active surface site (INAS) density concept [Hoose and Möhler, 2012]. Moreover, not only the onset of ice nucleation on the most ice-active seed aerosol particles can be determined, but the data can additionally be analyzed to yield spectra of the activated fraction of the aerosol particles as a function of the supersaturation at different temperatures. Our article will first introduce the key aerosol and ice particle measurement techniques of the AIDA chamber which are used to quantitatively deduce the depositional ice nucleation ability of crystalline NaCl and NaCl ∙ 2H2O (section 2). The results of this analysis are then presented in section 3 in terms of temperature-dependent nucleation onsets, activated fractions, and INAS densities, including a comparison with the ice nucleation ability of different atmospheric aerosol types which also promote ice nucleation in the deposition mode, like mineral dust and volcanic ash particles. A summary in section 4 concludes our article. Here we also discuss whether our results on pure sodium chloride particles can directly be transferred to predict the ice nucleation ability of actual sea salt aerosol (SSA) particles, which, apart from NaCl as their primary constituent, contain further ions like SO42− and Mg2+ as minor components [Koop et al., 2000a; Wise et al., 2012].
2.1 General Setup and Instrumentation
 The ice nucleation experiments were performed in the 84 m3 sized aerosol and cloud chamber AIDA at the Karlsruhe Institute of Technology (Figure 1) [Wagner et al., 2006b]. The aerosol vessel is located in an isolating housing and its temperature can be controlled between ambient and about 183 K. The temperature inhomogeneity is typically less than ±0.3 K throughout the chamber volume, as ensured by ventilating air around the aerosol vessel and continuously operating a mixing fan inside the chamber during the experiments. The chamber can be evacuated with two vacuum pumps, which allows an efficient cleaning of the vessel and yields background particle number concentrations of less than 1 cm−3. Purified water is evaporated into the evacuated chamber in such an amount that ice coverage of the inner chamber walls is achieved. Afterward, the chamber is refilled to ambient pressure with particle-free synthetic air and the NaCl seed aerosol particles, whose ice nucleation ability is to be investigated, are injected into the chamber as detailed in section 2.2.
 To probe heterogeneous ice nucleation by NaCl and NaCl ∙ 2H2O crystals in the deposition mode, the particles must be exposed to an environment that is supersaturated with respect to ice. The ice layer on the chamber walls controls the initial relative humidity at the start of an ice nucleation experiment to almost 100% with respect to ice. Ice supersaturated conditions are then established by expansion cooling, i.e., by lowering the pressure inside the aerosol vessel at controllable rates with the vacuum pumps. The relative humidity during expansion cooling as well as the microphysical properties of the nucleated ice crystals are measured by a comprehensive set of instruments [Wagner et al., 2009], whose characteristics are briefly described in the following.
 The relative humidity is measured in situ by tunable diode laser (TDL) absorption spectroscopy with a time resolution of 1 s. A selected ro-vibrational water vapor absorption line at 1.37 µm is scanned to deduce the water vapor pressure, pw(T), at the prevailing AIDA temperature with an estimated uncertainty of ±5%. For the same temperature, the saturation water vapor over ice, pw,ice(T), is computed [Murphy and Koop, 2005]. The quotient of pw(T) and pw,ice(T) then yields the ice saturation ratio, Sice(T). As a major advantage, this technique is applicable to both cloud-free and in-cloud conditions.
 Two optical particle counters (OPC1 and OPC2, type WELAS, Palas GmbH) are connected to the bottom of the AIDA chamber to measure time series of the number concentration of the nucleated ice crystals, Nice, during the expansion cooling experiments. The detection ranges of the instruments (for a refractive index of 1.33) are 0.7–46 µm (OPC1) and 5.0–240 µm (OPC2). This implies that, depending on their sizes (section 2.2), also a fraction of the seed aerosol particles might be detected by these instruments. The nucleated ice crystals, however, rapidly grow to much larger sizes compared to the NaCl and NaCl ∙ 2H2O crystals. Therefore, an optical threshold size can be applied to separately count the ice particles. Based on previous intercomparisons of the OPC1 and OPC2 records for the ice particle number concentration with the results from different instruments, we assume that the maximum uncertainty in Nice is ±20% [Möhler et al., 2006].
 The number concentration and size of the nucleated ice crystals can additionally be retrieved from their infrared extinction signatures, which are measured at 4 cm−1 resolution between 6000 and 800 cm−1 with an FTIR (Fourier transform infrared) spectrometer (IFS66v, Bruker) that is connected to an open-path multiple reflection cell inside the AIDA chamber [Wagner et al., 2006a]. Infrared extinction spectroscopy is also employed to identify the phase of the NaCl seed aerosol population. The same twofold area of application, i.e., characterization of both the seed aerosol particles and the ice crystals, also holds for in situ laser light scattering and depolarization measurements with the SIMONE instrument [Schnaiter et al., 2012]. This device measures the intensity of laser light that is scattered from the particles in 178° backward direction. Formation of large ice crystals on a subset of the aerosol population by deposition mode nucleation is detected by a strong increase in the back-scattering intensity. Additionally, the back-scattered laser light is polarization-resolved to determine the back-scattering linear depolarization ratio, δ, of the aerosol particles. As δ is zero for light scattering by spheres but different from zero when aspherical particles are present, the depolarization measurements are a powerful tool to detect phase transitions (deliquescence and efflorescence) in the NaCl particle ensemble.
2.2 Aerosol Generation and Characterization
 As outlined in section 1 and detailed in our previous article [Wagner et al., 2012], ensembles of crystalline NaCl and NaCl ∙ 2H2O particles were generated in situ by first injecting aqueous NaCl solution droplets into the AIDA chamber, and then observing their crystallization at constant temperature and relative humidity over timescales of several hours. An ultrasonic nebulizer (GA2400 Sinaptec), filled with a 5 wt% (weight percent) aqueous NaCl solution (NaCl, Merck, >99.5%), was typically used for aerosol injection. Four individual experiments with using the nebulizer for aerosol generation were conducted at 235.7, 230.7, 225.7, and 216.0 K (Table 1, Experiments 1 to 4). These values also correspond to the starting temperatures of the subsequent expansion cooling runs. In all experiments, the ultrasonic nebulizer was operated with identical settings and with a portion of the same NaCl solution.
Table 1. Parameters of Individual Ice Nucleation Experiments With Crystallized NaCl and NaCl ∙ 2H2O Aerosol Particlesa
aT: the mean AIDA gas temperature at aerosol injection and during the crystallization period until the start of pumping. For aerosol generation (third column), either an ultrasonic nebulizer (UN) or an atomizer (AM) filled with an aqueous NaCl solution of the specified wt % was used. : the number fraction of particles that have crystallized as NaCl ∙ 2H2O. Ntotal: the overall number concentration of the NaCl and NaCl ∙ 2H2O particles, dmean: the mean diameter of the number size distribution measured at ambient temperature, Atotal: the overall surface area concentration of the NaCl and NaCl ∙ 2H2O particles, including a size correction for the dihydrate particle fraction, : the surface area concentration of only the dihydrate particle fraction. See text for details.
UN/5 wt %
UN/5 wt %
UN/5 wt %
UN/5 wt %
AM/4 wt %
AM/1 wt %
 The gradual crystallization of the particle ensemble of initially spherical aqueous NaCl solution droplets was evidenced by a continuous increase in the depolarization ratio. After the entire particle population had crystallized, δ remained at a constant value. The infrared extinction spectra of the crystallized particle ensembles were then used in an optimization scheme to retrieve the number fractions of anhydrous and dihydrate NaCl crystals [Wagner et al., 2012]. The key step of this approach was to fit simulated spectra to the measured infrared extinction spectra. These simulated spectra were computed as a superposition of extinction due to NaCl ∙ 2H2O and scattering due to anhydrous NaCl particles. The size distribution parameters of both the dihydrate and the anhydrous particle fraction were optimized to get best agreement between the measured and the computed infrared spectra. The temperature-dependent partitioning between the two solid phases was then obtained from the optimized size distribution parameters. As summarized in Table 1, the percentage of dihydrate particles, , increased from 7% at 235.7 K to 88% at 216.0 K. Table 1 additionally includes the total particle number concentrations, Ntotal, and the mean diameters (arithmetic averages) [Hinds, 1999], dmean, of the number size distributions of the crystallized aerosol populations, as both measured immediately before the start of the expansion cooling experiments. Ntotal was measured with two condensation particle counters (CPC3010 and 3775, TSI), and dmean was calculated from the number size distributions that were measured by combining the size spectra from a scanning mobility particle sizer (SMPS, TSI, mobility diameter range: 0.014–0.82 µm) and an aerodynamic particle spectrometer (APS, TSI, aerodynamic diameter range: 0.523–19.81 µm). Two representative number size distributions from Experiments 1 and 4 are shown in Figure 2 as a function of the volume-equivalent sphere diameter, dp (solid lines). To convert the mobility diameter from the SMPS into dp, a dynamic shape factor, χ, of 1.08 as representative for cubic anhydrous NaCl particles was employed. The same value for χ as well as the particle density of anhydrous NaCl, ρ(NaCl) = 2.165 g cm−3, were chosen to transform the aerodynamic diameter from the APS into dp [Hinds, 1999].
 As already explained in Wagner et al. , anhydrous NaCl was chosen as the reference for the dp conversion because the size distributions were measured outside the isolating housing at ambient temperature by sampling air from the cold chamber interior. It is therefore reasonable to assume that the dihydrate particle fraction from AIDA will transform upon warming and be detected as anhydrous NaCl in the size distribution measurements. The actual size of the NaCl ∙ 2H2O crystals as present in AIDA at low temperatures will be larger, both due to the inclusion of the water molecules in the crystals and due to the lower particle density of NaCl ∙ 2H2O compared to NaCl [Light et al., 2009]. Therefore, a correction factor for the sizes of the dihydrate particle fraction has been applied according to equation (4) in Wagner et al.  for the accurate computation of the total surface area concentration, Atotal, of the particle ensemble, which is required as input for analyzing the ice nucleation ability within the INAS concept.
 The two dashed lines in Figure 2 show the estimated actual number size distributions of the NaCl and NaCl ∙ 2H2O particle ensembles as prevalent at AIDA conditions, i.e., after applying the size correction to the dihydrate particle fraction. These number size distributions were then converted into surface area size distributions and integrated to yield Atotal of the aerosol populations, as summarized in Table 1. In lack of any information about the actual particle habit of NaCl ∙ 2H2O, we assumed that the particles were spherical in shape. A further table entry shows the calculated values for , which represents the surface area concentration of the subset of particles that have crystallized as NaCl ∙ 2H2O. The distinction between Atotal and is necessary because it is the dihydrate particle fraction that triggers heterogeneous ice nucleation in the expansion cooling experiments.
 Table 1 additionally includes two experiments (Experiments 5 and 6) where an atomizer with a 4 and 1 wt % NaCl solution was used for aerosol generation, respectively. Those were dedicated experiments to generate smaller sized particle ensembles with mean diameters less than 0.1 µm (Experiment 6), which were explicitly needed to deduce the infrared optical constants of NaCl ∙ 2H2O [Wagner et al., 2012]. These smaller sized ensembles of crystallized NaCl and NaCl ∙ 2H2O particles, however, were probed on their heterogeneous ice nucleation ability as well and are therefore also included in our analysis. Only from their infrared extinction spectra, one cannot retrieve the number fractions of dihydrate and anhydrous NaCl particles as for Experiments 1–4. This is because for small particle sizes, the scattering contribution to extinction vanishes so that anhydrous NaCl crystals become invisible in the infrared spectra recordings, meaning that their contribution cannot be quantified. For our analysis, we have assumed that for Experiment 5 is identical to Experiment 3 because it was conducted at the same temperature. For Experiment 6 that was conducted at an intermediate temperature, the dihydrate particle fraction was estimated by a linear interpolation of the values from Experiments 3 and 4.
3 Results and Discussion
3.1 General Overview of the Temperature-Dependent Heterogeneous Ice Nucleation Ability
 Before turning to the quantitative analysis of the ice nucleation data in section 3.2, we want to give a general overview of the temperature-dependent heterogeneous ice nucleation ability of the crystallized NaCl and NaCl ∙ 2H2O particle ensembles. We present the AIDA measurements from three selected expansion cooling experiments which underline that there are two competing processes upon increasing relative humidity, namely deliquescence and heterogeneous ice nucleation. A particularly interesting example is the crystallization experiment at 230.7 K where anhydrous and dihydrate NaCl particles have formed with roughly equal number fractions (Experiment 2, Table 1).
 Figures 3a–3d show time series of relevant AIDA records for the expansion cooling cycle performed with the crystallized aerosol population from Experiment 2. The two arrows on the timescale labeled X and Y denote the times when the two infrared extinction spectra that are shown in Figure 3e were recorded. Spectrum X was monitored after the entire population of injected aqueous NaCl solution droplets had crystallized. As described in section 2.2, this was confirmed by reaching a constant level of the back-scattering linear depolarization ratio δ. The fraction of solution droplets that have crystallized as anhydrous NaCl contribute to the spectral habitus with a featureless scattering signature, whereas the water molecules in the NaCl ∙ 2H2O particle fraction give rise to prominent extinction bands in the O–H stretching (~3400 cm−1) and the O–H bending mode (~1600 cm−1) regime [Wagner et al., 2012]. As already indicated in section 1, spectrum X compares well with a literature spectrum for NaCl ∙ 2H2O. For comparison, the latter is also plotted in Figure 3e (gray lines). From top to bottom, Figures 3a–3d show the AIDA pressure and mean gas temperature (black and gray line in Figure 3a, respectively), the saturation ratios with respect to ice and liquid supercooled water (Sice, Sliq = RHw/100) as inferred from the TDL water vapor absorption measurements (black and gray line in Figure 3b, respectively), the number concentration of nucleated ice crystals, Nice, as measured with the optical particle counters (Figure 3c), and the back-scattering linear depolarization ratio δ (Figure 3d).
 Pumping is started at time zero and immediately leads to an increase in RHw. At RHw = 73%, as highlighted by the vertical gray line, δ drops from 0.31 to about 0.10, indicating a deliquescence transition in the aerosol population. In accordance with the deliquescence relative humidities reported in the literature, this is due to deliquescence of the fraction of particles that have crystallized as anhydrous NaCl. Because δ does not drop to the background value that is observed for an entirely liquid particle ensemble, aspherical crystals are still present, namely those particles that have crystallized as NaCl ∙ 2H2O. At the time when recording the extinction spectrum Y, we thus have a mixture of aqueous NaCl solution droplets and crystalline NaCl ∙ 2H2O particles. This is nicely reflected in the spectrum that shows the broad extinction signatures due to liquid water, with the characteristic narrow peaks due to NaCl ∙ 2H2O still superimposed on them.
 Before surpassing the deliquescence relative humidity of NaCl ∙ 2H2O at about 82% RHw, the NaCl ∙ 2H2O crystals act as heterogeneous ice nuclei in the deposition mode. As marked by the first vertical black line, ice formation initiates at a threshold ice saturation ratio of Sice = 1.25 and at a temperature of 227.1 K. We define the nucleation onset as the time when the ice particle number concentration has exceeded 1 cm−3, because modes with Nice ≤ 1 cm−3 could also be due to nucleation on impurities. During continued pumping, Sice reaches a peak value of 1.33 before the increasing number of nucleated ice crystals starts to deplete the excess of water vapor in the gas phase, leading to a decrease in Sice. As soon as Sice again drops below the nucleation threshold of 1.25 (second vertical black line), no further ice crystals nucleate and pumping is stopped shortly thereafter. The ice particle number concentration at Sice = 1.33 amounts to about 17 cm−3. In relation to the NaCl ∙ 2H2O seed aerosol number concentration, as obtained from the data in Table 1 and corrected for pumping dilution, this corresponds to an ice active fraction of 9.6%. This example underlines that knowledge of the partitioning between anhydrous NaCl and NaCl ∙ 2H2O is a prerequisite to predict the behavior of the aerosol population upon increasing relative humidity: a particle ensemble of only anhydrous NaCl crystals probed in an expansion cooling cycle started at 230.7 K would deliquesce and then homogeneously nucleate ice at elevated supersaturated levels. In contrast, crystalline NaCl ∙ 2H2O particles are able to act as heterogeneous ice nuclei in the deposition mode at that temperature before surpassing their deliquescence relative humidity, thereby inducing ice formation at a much lower supersaturation level compared to homogeneous freezing.
 The expansion cooling cycles can further be illustrated by plotting their experimental trajectories in the temperature-relative humidity space (Figure 4). For Experiment 2, it becomes obvious that the value for the initial deliquescence transition of anhydrous NaCl at 73% RHw, as shown by the square, is slightly below that expected from the extrapolation of the Tang and Munkelwitz's  parameterization, even when taking into account an uncertainty of ±5% for our humidity measurements. This can be explained, on the one hand, by the inhomogeneity in the gas temperature throughout the chamber volume. The value of 73% RHw refers to the mean gas temperature, but deliquescence will initiate in the coldest parts of the vessel with concomitantly locally higher RHw values compared to the mean. On the other hand, there are also differences between various formulations for the saturation water vapor pressure over supercooled water at such low temperatures [Murphy and Koop, 2005], which increases the uncertainty of the determined RHw value. Heterogeneous ice formation during Experiment 2, as marked by the circle, then starts clearly before surpassing the deliquescence relative humidity of NaCl ∙ 2H2O and well below the homogeneous freezing limit.
 We now turn to Figure 5 that shows the AIDA records during the expansion cooling cycles from Experiments 1 (235.7 K, left panel) and 5 (225.9 K, right panel). The respective experimental trajectories in the T-RHw space are plotted in Figure 4. In Experiment 1, the crystallized aerosol population is dominated by anhydrous NaCl particles. During the expansion cooling cycle, these particles first deliquesce at RHw = 73% (vertical gray line) and then homogeneously freeze at Sice = 1.38 (vertical black line). Temperature inhomogeneities, which could account for the difference between the measured and expected deliquescence relative humidity for anhydrous NaCl as outlined above, might explain the deviation of our recorded freezing onset from the Koop et al.'s [2000b] homogeneous freezing parameterization. The fate of the minor percentage of NaCl ∙ 2H2O crystals cannot be exactly traced. After the deliquescence step of the anhydrous crystals, the depolarization ratio almost drops to the background value that is observed in the presence of only spherical particles, meaning that a potential further deliquescence step due to NaCl ∙ 2H2O at about RHw = 82% cannot be resolved. This is because the anhydrous particles, in addition to their much higher number fraction compared to the NaCl ∙ 2H2O crystals, also have a larger size after deliquescence due to the water uptake and therefore dominate the intensity of the backscattering signal. Before the homogeneous freezing onset, a small heterogeneous ice mode with Nice ≤ 1 cm−3 is formed. This could be related to deposition mode nucleation on the minor percentage of NaCl ∙ 2H2O crystals, but as explained above, nucleation on impurities cannot be excluded.
 In a former AIDA expansion run started at 235.7 K, we have probed the ice nucleation ability of crystallized aqueous NaCl solution droplets that had contained a solid inclusion of oxalic acid [Wagner et al., 2011]. These solid inclusions have triggered the precipitation of sodium chloride dihydrate in a much larger subset of the NaCl solution droplets than observed in Experiment 1 from our present study. We cannot completely rule out that the presence of oxalic acid modifies the behavior of the particles compared to pure NaCl. The most likely morphology of the particles after crystallization, however, can be envisaged as a solid oxalic core surrounded by an almost pure matrix of solid NaCl or NaCl ∙ 2H2O, which is available for deposition mode ice nucleation. Only the dihydrate particle fraction clearly underwent a deliquescence transition in the course of this expansion cooling experiment at RHw = 82% (Sice = 1.21) and T = 232.9 K. Prior to deliquescence, heterogeneous ice formation by deposition mode nucleation on NaCl ∙ 2H2O was not observed. This experiment from our previous work in combination with Experiment 2 from our present study thus localizes the transition regime of the two competing processes deliquescence and heterogeneous ice nucleation for NaCl ∙ 2H2O. At and above 232.9 K, deliquescence occurs; at and below 227.1 K, heterogeneous ice nucleation in the deposition mode takes place.
 The AIDA records of our third selected expansion cooling cycle (Experiment 5, Figure 5) are representative for all experiments that were started below 226 K. At these lower temperatures, the crystallized particle ensembles are dominated by NaCl ∙ 2H2O. During Experiment 5, we observe heterogeneous ice nucleation starting at a threshold of Sice = 1.20 (RHw = 75%, vertical black line). Just before the strong increase in the ice particle number concentration, there is a slight decrease in the trace of the depolarization ratio, which indicates the deliquescence transition of the small number fraction of anhydrous crystals. This deliquescence step, however, is not as clearly resolved as at higher temperatures where larger fractions of anhydrous particles are present, and we therefore did not include this point into the trajectory of Experiment 5 in Figure 4. At starting temperatures even lower than in Experiment 5, heterogeneous ice formation initiates before the deliquescence relative humidity of anhydrous NaCl is surpassed. Therefore, we cannot unambiguously resolve whether ice formation is solely due to deposition mode nucleation on the NaCl ∙ 2H2O particle fraction or, at temperatures below 220K, additionally involves the minor percentage of anhydrous NaCl crystals that are also present in the particle ensemble. We will further discuss this issue in the next section.
3.2 Quantitative Analysis and Discussion
3.2.1 Onset Conditions for Ice Nucleation
 In Figure 6, we have summarized the temperature-dependent onsets for deposition mode nucleation on crystalline sodium chloride dihydrate particles from all conducted expansion cooling cycles. Different symbols are used to differentiate between the experiments where the ultrasonic nebulizer (squares) and the atomizer (circles) were used for aerosol generation, which strongly affects the mean diameter of the number size distribution of the aerosol population. Filled symbols denote the nucleation onset for the first expansion cooling cycle performed with the crystallized particle ensemble. In most experiments, we have probed the aerosol load again in a second expansion cooling cycle at the same starting temperature and initial relative humidity, which was conducted after the ice cloud from the preceding expansion run had sublimed. This was done to investigate whether the ice nucleation ability of the NaCl ∙ 2H2O particles changes after they have already been involved in ice cloud formation. The nucleation onsets from the second expansion cooling cycles are shown as the open symbols. As a comparison, the average nucleation onsets for anhydrous and dihydrate NaCl from the Wise et al.'s  study are also shown. Please note that we have defined the onsets as the times when 1 cm−3 ice particles have nucleated. For the experiments with the atomizer (AM, Experiments 5 and 6) where on the order of 104 cm−3 seed aerosol particles were present, we are thus sensitive to an about 0.01% number fraction of the aerosol particles becoming ice-active at the nucleation onset. For the experiments with the ultrasonic nebulizer (UN, Experiments 1–4) with a seed aerosol number concentration of the order of 102 cm−3, the nucleation onsets correspond to an ice-active fraction of already about 1%. At first glance, this seems to make the comparison between the results from the AM and the UN experiments less meaningful. The AIDA data for Experiment 5 (Figure 5), however, reveal that the ice particle number concentration rapidly increases after the nucleation onset from 1 to above 100 cm−3. This indicates that the nucleation onset for this AM experiment, corresponding to only a 0.01% active fraction, does not just represent a minor percentage of the aerosol population with particularly high ice nucleation ability. Instead, a larger fraction of the NaCl ∙ 2H2O particle ensemble starts to become ice active within a narrow range of Sice values above this nucleation threshold, justifying that the latter can indeed be compared to those inferred from the UN experiments. The application of the INAS concept (section 3.2.2) will further illustrate the good agreement between the AM and the UN experiments.
 The AIDA results for the nucleation onsets feature the following characteristics:
 The threshold ice saturation ratios are rather insensitive to temperature. The highest onset value of Sice = 1.25 is observed for the highest nucleation temperature of 227.1 K (Experiment 2). At colder temperatures, the Sice values vary between 1.15 and 1.20.
 The nucleation thresholds do not depend on the size of the seed aerosol particles. The smaller sized NaCl ∙ 2H2O particle ensembles generated with the atomizer (AM) show the same onset conditions for deposition mode nucleation as the larger sized aerosol populations generated with the ultrasonic nebulizer (UN).
 The ice nucleation ability does not notably change in repeated expansion runs performed with the same aerosol load. This excludes any type of preactivation mechanism, representing a memory effect by which seed aerosol particles that already have participated once in ice crystal formation maintain a higher ice nucleation ability for succeeding nucleation events [Pruppacher and Klett, 1997].
 Note again that only the nucleation onsets above 220 K can unambiguously be related to NaCl ∙ 2H2O, because the deliquescence transition of the anhydrous particle fraction was detected before the ice nucleation onset, and therefore this particle fraction can be excluded from contribution to heterogeneous ice nucleation. We assume, however, that this also holds for the experiments below 220 K, given that the aerosol population is dominated by the dihydrate particles and that Wise et al.  have always observed ice formation to initiate on NaCl ∙ 2H2O in a mixture where both anhydrous and dihydrate NaCl crystals were present. In comparison with the Wise et al.'s  study, the AIDA experiments yield higher nucleation onsets for deposition mode nucleation on NaCl ∙ 2H2O. In addition, the transition regime of the two competing processes deliquescence and heterogeneous ice nucleation is shifted to lower temperatures. In AIDA, we still have detected deliquescence of NaCl ∙ 2H2O to occur at 232.9 K, whereas exclusively heterogeneous ice nucleation was observed in the Wise et al.'s  experiments at that temperature.
 The ensemble of deposited NaCl ∙ 2H2O crystals in the Wise et al.'s  experiments thus contained at least one extremely efficient ice nucleus that triggered deposition mode nucleation at only 2% supersaturation. Such efficient ice nuclei were either not present in AIDA, or their number concentration was simply too low to become detectable as a distinct ice crystal mode by our instrumentation. At least for Experiments 5 and 6 with seed aerosol number concentrations of about 104 cm−3, we had the same sensitivity to freezing compared to the Wise et al.'s  study. The airborne particles probed in AIDA, however, had smaller sizes than those investigated by Wise et al. , thereby reducing the probability for the existence of an ice-active site that could promote ice formation at very low supersaturation levels. Moreover, the dihydrate crystals probed in AIDA could have had a lower degree of surface roughness compared to those investigated by Wise et al. , thereby featuring a lower ice nucleation ability. Concerning the Wise et al.'s  experiments, one cannot tell whether the low reported nucleation onsets are representative for a larger subset of the deposited NaCl ∙ 2H2O crystals because only the nucleation of the first ice crystal was monitored. Ice active fractions or INAS densities as a function of Sice could not be determined. Note that we also cannot completely exclude the possibility that Wise et al. , instead of the dihydrate, have probed a different hydrated species of NaCl. Another difference between both studies is the magnitude of the humidification rate, which in the AIDA expansion cooling experiments is above the upper threshold value of 10% RH per minute reported for the Wise et al.'s  experiments. The higher humidification rate in AIDA, however, would only affect our results if there is a significant time delay between the actual nucleation onset and the time when the pristine deposition ice germs have grown to ice crystals of sizes detectable by the optical particle counters. For the temperatures covered by our experiments, Möhler et al.  have estimated this time delay to be about 4 s (223 K). The absolute difference between the Sice values for our inferred nucleation onsets to those measured 4 s earlier is only about 0.01. This bias is much smaller than the difference between the Sice values for the dihydrate from the Wise et al.  and our study. We are therefore strongly inclined to rule out that the difference in humidification rate explains the different results for the freezing onsets.
 Notwithstanding the somewhat higher nucleation onsets for NaCl ∙ 2H2O as inferred from the AIDA experiments, the key finding from the Wise et al.'s  study remains unaffected: the dihydrate particles are efficient heterogeneous ice nuclei and promote ice formation by deposition mode nucleation at relative humidities well below the homogeneous freezing threshold. For assessing these nucleation onsets with respect to the ice nucleation ability of different classes of atmospherically relevant ice nuclei like mineral dust, we recommend the recent review by Hoose and Möhler , showing for example a compilation of the ice nucleation onset temperatures and saturation ratios for submicron Arizona Test Dust (ATD) and natural dust particles from a large set of laboratory experiments. In the temperature range from 213 to 227 K, our reported range of ice saturation ratios for NaCl ∙ 2H2O (Sice = 1.15–1.25) is lower than that inferred from the majority of the deposition mode nucleation experiments with natural dusts, whose derived onset Sice values typically range from about 1.25 to 1.45. Rather, the nucleation onsets for NaCl ∙ 2H2O are comparable to those for ATD (reported onsets at 213–227 K: Sice ≈ 1.02–1.20), which is frequently used as a surrogate for desert dusts although it is generally more ice active than natural soil samples [Möhler et al., 2006].
 As discussed above, a large fraction of the NaCl ∙ 2H2O aerosol particles is observed to become ice active within a narrow range of Sice values above the nucleation threshold (Figure 5, Experiment 5). One might argue that this supports the idea of the chemical bond requirement to be the dominant criterion for the heterogeneous ice nucleation ability of NaCl ∙ 2H2O, meaning that the water molecules in the crystal are the preferential sites for additional water molecules to make chemical bonds with. Since the presence of the water molecules is an intrinsic property of all seed aerosol particles, it could explain the rather uniform ice nucleation ability of the NaCl ∙ 2H2O particle ensemble. Due to the large number of ice crystals that are formed shortly after the nucleation onset, the peak saturation ratio is confined to a value of only a few percent above the threshold value because the ice crystals rapidly deplete the excess of water vapor in the gas phase. We therefore cannot tell whether there is an additional fraction of NaCl ∙ 2H2O particles which is less ice active and only triggers deposition mode nucleation at higher relative humidities. This issue could be resolved in future experiments with lower seed aerosol and concomitantly lower ice particle number concentrations, so that the peak ice saturation ratio is not limited and can be further raised above the threshold value during continued pumping. For this first study, it was essential to employ higher aerosol number and mass concentrations in order to obtain a good signal-to-noise ratio for the infrared spectroscopic analysis of the partitioning between NaCl and NaCl ∙ 2H2O.
3.2.2 Application of the INAS Concept
 The INAS concept represents a singular description which neglects the time dependence of the stochastic nucleation process [Pruppacher and Klett, 1997]. It assumes that there is a distribution of nucleation sites on the particle surface area which each possess a characteristic threshold Sice value for deposition mode nucleation at a given temperature and within the typical timescale of the experiment. In this approach, the so-called ice nucleation active surface site (INAS) density, ns(Sice, T), describes the cumulative number of nucleation sites per surface area which become ice active in the deposition mode up to the saturation ratio Sice at temperature T [Connolly et al., 2009; Hoose and Möhler, 2012; Murray et al., 2012; Vali, 2008]. In order to obtain a simple expression for ns(Sice, T) which also applies to aerosol particle ensembles of different sizes, one must assume that the INAS density is constant throughout the size distribution. It can then be calculated from the following equation:
 In equation (1), denotes the aerosol surface area per particle, Nae is the seed aerosol particle number concentration, and Nice(Sice,T) is the cumulative number concentration of ice crystals that have nucleated up to the saturation ratio Sice at temperature T. When having a closer look at the time series of Nice during Experiment 2 (Figure 3), there still seems to be a slight increase of about 20% in the ice particle number concentration in the short time period after the peak ice saturation ratio has been surpassed and before it again drops below the nucleation threshold (second vertical black line). This could represent the stochastic element to nucleation which is not covered by the singular approach where a specific nucleation site becomes instantaneously ice active as soon as its characteristic Sice threshold is surpassed. One should consider, however, that the temperature still decreases in this designated time period, which could trigger the further nucleation of ice crystals given that the INAS density is a function of both Sice and T. Alternatively, this might also be due to a measurement artifact because the number concentration of the nucleated ice crystals is inferred from the OPC records by employing an optical threshold size to distinguish them from the smaller sized seed aerosol particles. So there might be a small time shift before all nucleated ice crystals have grown large enough to exceed this applied size threshold.
 The heterogeneous ice nucleation experiments 2–6 from the present study were thus evaluated according to equation (1) and the derived ns(Sice, T) values are summarized in Figure 7 (black symbols). The data are shown from the nucleation onset until the peak ice saturation ratio was reached in the expansion run. We have assumed that the generated ice crystal modes are entirely due to deposition mode nucleation on the NaCl ∙ 2H2O particle fraction, meaning that Nae in equation (1) corresponds to and to (Table 1). As discussed above, this assignment is only uncertain for Experiments 4 and 6, where ice formation occurred before the deliquescence relative humidity of the anhydrous particle fraction was surpassed. As the dihydrate particle fraction, however, makes up more than 80% of the total aerosol surface area, the error of neglecting some potentially ice active anhydrous crystals will be small. Moreover, Wise et al.  have observed that the nucleation threshold for anhydrous NaCl is on average 0.09 higher than that for NaCl ∙ 2H2O. The peak ice saturation ratio during the expansion runs four and six, however, was confined to less than 0.09 above the threshold value for nucleation. This makes it unlikely that any anhydrous particles have contributed to ice formation. When evaluating ns(Sice, T), Ntotal was corrected for the dilution due to pumping. The assumption of a compact spherical particle shape to compute represents a lower limit of the actual surface area of the irregularly shaped crystals.
 Figure 7 additionally includes the INAS densities for deposition mode nucleation on five different aerosol types, namely volcanic ash, ATD, two natural dust samples from Sahara (SD2), and Takla Makan desert in Asia (AD1), as well as glassy aqueous citric acid aerosol particles (gray line and gray symbols). These data were selected as a comparison from the comprehensive set of laboratory studies recently analyzed by Hoose and Möhler , because they were also inferred from AIDA expansion cooling experiments using the same instrumentation as in the present study.
 The main characteristics of the reported INAS densities can be summarized as follows. For Experiments 3–6, the ns(Sice, T) data are a very steep function of Sice. This reflects the observed rapid increase in the ice particle number concentration as soon as the threshold ice saturation ratio is exceeded during the expansion runs (Figure 5, right). The starting temperatures of these expansion cycles cover the range from 216.0 to 225.9 K. Within this regime, the INAS densities for NaCl ∙ 2H2O show little sensitivity to temperature. Most noteworthy, the derived ns(Sice, T) data for both the experiments with the ultrasonic nebulizer and the atomizer closely agree, meaning that the surface properties indeed do not vary with size. On average, a maximum INAS density of about 6 ⋅ 1010 m−2 at Sice = 1.20 can be inferred. This is the same order of magnitude as derived for the volcanic ash and the natural dust particles at that saturation ratio. The ATD particles reveal a similarly steep increase of the INAS density with Sice as observed for the NaCl ∙ 2H2O particles, however shifted to lower ice saturation ratios. The data for volcanic ash and natural dust reveal a more gradual increase in ns, pointing to a higher diversity of the nucleation ability of the active sites. Due to the rapid depletion of the supersaturation and the concomitantly low accessible values of the peak ice saturation ratio, we cannot determine from the present experiments whether the INAS densities for NaCl ∙ 2H2O would further increase to values greater than 1011 m−2 at higher saturation ratios, as observed, e.g., for the highly ice active ATD particles.
 The INAS densities of Experiment 2 show a slightly different trend compared to those from Experiments 3–6. In addition to the higher nucleation onset, the slope with respect to Sice is different, and the maximum ns value is only 2.3 ⋅ 1010 m−2 at Sice = 1.33. A potentially further increase of the INAS density for higher Sice values can be ruled out for this experiment, because the peak ice saturation ratio already corresponds to the expected deliquescence relative humidity for NaCl ∙ 2H2O. NaCl ∙ 2H2O crystals that have been ice inactive up to this point are thus expected to transform into supercooled NaCl solution droplets instead of nucleating ice when further raising the relative humidity. The quotient Nice(Sice,T)/Nae that appears in the expression for the INAS density (equation (1)) is simply the ice active fraction of the aerosol population, fice. The lower ice nucleation ability of NaCl ∙ 2H2O as observed during Experiment 2 can therefore also be analyzed in terms of fice. As reported in section 3.1, fice is at most 9.6% for Experiment 2, whereas maximum values of 20.8 and 17.7% are obtained during Experiments 3 and 4, respectively.
 Notwithstanding the apparent differences to Experiments 3–6, the INAS densities for NaCl ∙ 2H2O from Experiment 2 still show a rather steep increase with respect to Sice, pointing to the uniform freezing ability of the nucleation sites of this particular aerosol type. In this context, it is interesting that just the glassy aqueous citric acid aerosol particles show the highest diversity with respect to the freezing ability among all aerosol types, as expressed by the rather smooth increase in ns over a wide range of Sice values. For a population of smooth aqueous glassy spheres composed of a single chemical component, one would rather expect a narrow dispersion of the ice nucleation ability with respect to Sice as observed for NaCl ∙ 2H2O. The experimental data, however, indicate that the opposite is true, underlining the complexity of the heterogeneous ice nucleation process.
 In this article, we have provided additional evidence that crystalline sodium chloride dihydrate particles, NaCl ∙ 2H2O, are able to act as heterogeneous ice nuclei in the deposition mode, a mechanism that has been first discovered by Wise et al. . In our experiments, heterogeneous ice nucleation by NaCl ∙ 2H2O was somewhat less efficient as detected by Wise et al. , as expressed by higher threshold ice saturation ratios and a lower temperature, at which heterogeneous ice nucleation first competed with deliquescence of the salt crystals. Nonetheless, the quantitative analysis of the ice nucleation experiments in the framework of the INAS concept has shown that NaCl ∙ 2H2O is approximately an equally good heterogeneous ice nucleus as mineral dust in the investigated regime of temperatures and ice saturation ratios.
 In order to explore the atmospheric significance of this process, we identify at least three key subjects for future chamber studies.
 Dedicated mechanistic studies of the heterogeneous ice nucleation process: As discussed above, expansion cooling cycles with lower NaCl ∙ 2H2O particle number concentrations would allow us to establish higher ice supersaturations to extend our present INAS analysis. Additionally, idealized isohumid experiments could be envisaged to investigate the time dependence in the nucleation process. In such experiments, the ice supersaturation would be maintained at a constant level by controlling the pumping speed, during which the time evolution of the number concentration of the nucleated ice crystals could be monitored.
 Crystallization and ice nucleation experiments with seed aerosol particles that more closely mimic the actual composition of sea salt aerosol (SSA) particles: As already mentioned by Wise et al. , this would not only involve mixtures with further inorganic ions contained in natural seawater but also the inclusion of organic material [Middlebrook et al., 1998; O'Dowd et al., 2004]. Of particular importance, it has to be investigated whether further constituents would change the low temperature partitioning between anhydrous and dihydrate NaCl crystals. Moreover, crystallization will not form a pure solid (NaCl ∙ 2H2O) phase, but the SSA particles might then consist of internally mixed phases (solid/liquid) due to the presence of many different ions [Koop et al., 2000a]. The solid phase might then no longer be available for deposition mode nucleation but can trigger ice formation only via immersion freezing.
 Competition between SSA and mineral dust in heterogeneous ice nucleation: One could imagine expansion cooling experiments with aerosol populations where both crystallized SSA and mineral dust particles are present as an external mixture. The nucleated ice crystals could be sampled with an ice-selective inlet to enable the mass spectrometric analysis of the chemical composition of the ice residue. If the SSA particles were indeed found to make up a substantial fraction of the ice residue, then this would have serious implications for the interpretation of atmospheric ice residue analyses: the presence of sea salt would then not necessarily be linked to a homogeneous freezing mechanism [Cziczo et al., 2004].
 We gratefully acknowledge the continuous support by all members of the AIDA staff. In particular, we thank Olga Dombrowski and Rainer Buschbacher for assistance with the aerosol generation and characterization, as well as Tomasz Chudy and Georg Scheurig for the technical maintenance of the chamber. The work has been funded by the Helmholtz-Gemeinschaft Deutscher Forschungszentren as part of the program “Atmosphere and Climate.”