Novel method of generation of Ca ( HCO 3 ) 2 and CaCO 3 aerosols and first determination of hygroscopic and cloud condensation nuclei activation properties

Atmospheric mineral aerosols contain CaCO 3 as a reactive component. A novel method to produce CaCO 3 aerosol was developed by spraying Ca(HCO 3)2 solution, which was generated from a CaCO 3 suspension and CO 2. By aerosol mass spectrometry the freshly sprayed and dried aerosol was characterized to consist of pure Ca(HCO 3)2 which under annealing in a tube furnace transformed into CaCO3. Transmission Electron Microscopy demonstrated that the particles produced were spherical. The method was able to generate aerosol of sufficient concentration and proper size for the study of physiochemical properties and investigations of heterogeneous reactions of mineral aerosol. The dried Ca(HCO3)2 particles were somewhat more hygroscopic than CaCO 3 particles. However, during humidification a restructuring took place and ∼2/3 of the Ca(HCO3)2 was transformed to CaCO 3. The mixed Ca(HCO3)2/CaCO3(s) particles were insoluble with a growth factor of 1.03 at 95% (hygroscopicity parameter κ=0.011±0.007) relative humidity. This compares to a corresponding growth factor of 1.01 for CaCO 3(s) (κ=0.0016± 0.0004). Mass spectrometric composition analysis, restructuring, and insolubility of the mixed particles suggested that solid Ca(HCO3)2(s) was observed. This would be in contrast to the current belief that Ca(HCO 3)2(s) is thermodynamically instable. The CCN activity of Ca(HCO 3)2(s) aerosol Correspondence to: Th. F. Mentel (t.mentel@fz-juelich.de) (κ ≈ 0.15) is remarkably higher than that of CaCO 3 aerosol (κ=0.0019±0.0007) and less than that of Ca(NO 3)2. The noticeable but limited solubility of Ca(HCO 3)2 of ≈ 0.01 mol/l explains limited hygroscopic growth and good CCN activity. Experiments in the Large J ülich Aerosol Chamber indicated that Ca(HCO 3)2(s) could exist for several hours under dry atmospheric conditions. However, it was likely buried in a protective layer of CaCO 3(s). We conclude that Ca(HCO3)2 may be formed in the atmosphere in cloud roplets of activated mineral dust by reaction of CaCO 3 with CO2 and H2O. The presence of Ca(HCO 3)2 and as a consequence an enhanced CCN activity may alter the influence of mineral aerosol on global climate.


Introduction
Mineral aerosol is one of the most abundant components of aerosols in the atmosphere. It mainly arises from wind-blown soil in the deserts or semiarid areas, and also from volcanic dust, road dust and some industrial and agricultural processes. It was estimated that approximately 1000-3000 Tg of mineral aerosols are annually emitted into the atmosphere (Jonas et al., 1995). While mineral aerosol is typically considered as having a coarse mode type of distribution with particle diameters >2.5 µm, a significant number of mineral particles are also found in the fine mode with diameters in a range of 0.1-2.5 µm (Usher et al., 2003;Song et al., 2005). D. F. Zhao et al.: Novel method of generation of Ca(HCO 3 ) 2 and CaCO 3 aerosols On the regional scale, mineral dust has an impact on the visibility, air quality, and human health. On the global scale, it can not only influence radiative transfer by absorption and scattering of solar radiation, but also affect the cloud formation and cloud optical properties. Thus, mineral aerosols influence the global climate ultimately by direct and indirect radiative forcing. During long range transport, mineral aerosols can undergo various heterogeneous reactions and their deposition has an impact on the ocean ecosystem and the overall biogeochemical cycling. CaCO 3 is a common component of mineral dust, especially in East Asia and is generally regarded as the most reactive component in mineral particles (Krueger et al., 2003;Laskin et al., 2005;Sullivan et al., 2007;Usher et al., 2003). It is present at significant concentrations in both mineral aerosols (Okada et al., 2005;Zhang et al., 2003) and in surface soils (at least 5-10% by mass) of most of the major global dust sources (Claquin et al., 1999). It was found that CaCO 3 is among the few "volatile" components in mineral aerosols which can undergo heterogeneous reactions with atmospheric trace acids to form CO 2 (Krueger et al., 2003;Laskin et al., 2005, Santschi andRossi, 2006). Many researchers have investigated the heterogeneous reactions of CaCO 3 as a surrogate for mineral dust and its impact on the physiochemical properties of mineral particles (Krueger et al., 2003;Liu et al., 2008a, b;Gibson et al., 2006). To study the physiochemical properties of mineral aerosols, the generation of surrogates of mineral aerosol components is often helpful. At present, there are two main methods to generate CaCO 3 aerosols, dry dispersion of solid powders (Sullivan et al., 2009a, b;Prince et al., 2007) and atomizing aqueous suspension of CaCO 3 powder (Gibson et al., 2006(Gibson et al., , 2007. Both methods are not easy to handle and it is difficult to generate a stable output of proper size and sufficient mass concentration. Moreover, it was observed by several groups that mineral particles generated by spray-drying of suspension are more hygroscopic than those from dry beds (Herich et al., 2009, Koehler et al., 2007, Sullivan et al., 2010. Sullivan et al. (2010) reported that for sub-100nm, wet-generated CaCO 3 particles the hygroscopicity parameter κ is 100 times larger than that from the dry bed generated CaCO 3 particles.
Compared to the method of dry dispersion of solid powders, generating aerosol by atomizing solutions does not need complex instrumentation and is easier to operate and control. However, most of the mineral dust components are characterized by very low solubility which makes atomization from solution difficult. Vlasenko et al. (2006) have used atomization of a saturated solution of CaCO 3 to produce CaCO 3 aerosol. However due to the low solubility of CaCO 3 in water, the number concentration of the obtained aerosol is often not high enough and the size distribution has its maximum at too small diameters considering the need of studies of physiochemical properties, such as hygroscopicity using HTDMA and CCN activation.
In order to study chemical reactions on CaCO 3 particles and to determine microphysical properties of the processed aerosols it would be advantageous to have at disposal an easy to handle and reproducible method for generation of CaCO 3 aerosol with sufficiently high particle concentration at proper size. Here we describe a novel method to generate CaCO 3 aerosol by atomization of solutions with enhanced amounts of dissolved Ca(HCO 3 ) 2 . The basic idea is to make use of the equilibrium of CaCO 3 and the water-soluble form Ca(HCO 3 ) 2 in the presence of CO 2 (Reaction R1). This equilibrium is ubiquitous in natural waters when dissolved CO 2 is in contact with rocks containing CaCO 3 . The process of dissolution of CaCO 3 leads e.g. to the formation of stalactites, stalagmites, columns, and other speleothems within caves when CO 2 in the water is lost and CaCO 3 sediment is produced (Fairchild et al., 2006;Murray, 1954).
At high CO 2 concentrations (R1) will proceed in the forward direction and cause the dissolution of the solid CaCO 3 . On the other hand, when CO 2 is decreasing, the backward reaction will produce solid CaCO 3 . In this study, pure CO 2 was bubbled into a slurry of CaCO 3 powder in order to promote the dissolution of CaCO 3 and formation of a Ca(HCO 3 ) 2 solution. We will show in the following that atomizing Ca(HCO 3 ) 2 solutions, drying and annealing of the aerosol promote CaCO 3 formation. The aerosols produced were characterized by Scanning Mobility Particle Sizing, Aerosol Mass Spectrometry, and Transmission Electron Microscopy in order to investigate the size distribution, typical concentrations, chemical composition and morphology of the aerosol. At the same time, the aerosols before und after annealing were characterized with respect to their hygroscopicity and cloud condensation activity (CCN) activity using a Hygroscopicity Tandem Differential Mobility Analyzer (HTDMA) and a Cloud Condensation Nuclei Counter (CCNC). Here we took advantage of the different solubilities of CaCO 3 (0.00015 mol/l) and Ca(HCO 3 ) 2 (0.008 mol/l) at room temperature (Gmelin, 1961).
To our best knowledge this method of generating CaCO 3 aerosols from a CO 2 containing solution of Ca(HCO 3 ) 2 has not been described before. Moreover, not much attention has been paid to the possible importance of Ca(HCO 3 ) 2 which could be formed in the atmosphere in cloud droplets of activated mineral dust containing CaCO 3 by reaction with CO 2 and H 2 O. 20 mbar from the cylinder into 1 L of water (Milli-Q, 18.2 M cm, TOC <5 ppb) containing 2 g CaCO 3 powder (pro analysis, ≥99%, Merck). During CO 2 bubbling the slurry was stirred using a magnetic stirrer to promote the suspension of solid CaCO 3 in the water. After bubbling, the slurry was allowed to settle for 5 min and the supernatant clear Ca(HCO 3 ) 2 solution was decanted.
In order to characterize the reproducibility of the method and the stability of the solution for aerosol generation, different CO 2 bubbling durations were applied and the solutions were atomized over longer periods. The aerosol was dried and the number density and the count median of the size distribution were determined. It was found that stirring the slurry helped to reach the equilibrium faster.
In order to avoid introducing possible contaminants to the aerosol, careful measures in preparing solutions and subsequent spraying were taken into account. This included using high purity agents, gas, water and careful cleaning of bottles, tubes and atomizer to our best effort.

Generation of CaCO 3 particles
The Ca(HCO 3 ) 2 solution was sprayed by a constant output atomizer (TSI, Model 3076) using synthetic air (Linde LiPur 6.0, purity 99.9999%). The generated aerosol passed through a diffusion drying tube filled with silica gel (final RH < 5%) and a delay tube to equilibrate charges (Dinar et al., 2006), as shown in Fig. 1. The number size distribution of the resulting aerosol was measured by an SMPS system (TSI, DMA 3071, CPC 3022A).
At room temperature Ca(HCO 3 ) 2 should be thermodynamically unstable as a solid and only exist in solution as Ca 2+ and HCO − 3 (Keiser and Leavitt, 1908;Miller, 1952;House, 1981). Upon drying, it is supposed to convert to CaCO 3 very easily at room temperature. However, as shown in the following, the aerosols generated from spraying Ca(HCO 3 ) 2 solutions still contained a significant amount of Ca(HCO 3 ) 2 after drying. We therefore used a tube furnace at 300 • C in order to promote the transformation of Ca(HCO 3 ) 2 to CaCO 3 aerosol. The temperature of the tube furnace was set to 300 • C because at this temperature Ca(HCO 3 ) 2 should already decompose to form CaCO 3 (Reaction R2) while CaCO 3 is still stable (Keiser and Leavitt, 1908;Sanders and Gallagher, 2002).
The whole setup as used is shown in Fig. 1. After the delay tube the aerosol was neutralized (TSI, 3077) and size selected by a DMA (TSI, DMA 3071). The size-selected aerosol passed the tube furnace which was either at room temperature or set to 300 • C. After the furnace the aerosol flow was split and directed to the different instruments. A Quadrupole Aerosol Mass Spectrometer (Q-AMS, Aerodyne Research Inc., Jayne et al., 2000) was used to characterize the composition of aerosols. HTDMA and CCNC (Droplet Measurement Technologies, DMT-100) were used to investigate the hygroscopicity and CCN activity.

Characterization of chemical composition and morphology of the aerosol
Chemical characterization of the aerosols was provided by the Q-AMS by measuring H 2 O and CO 2 evaporating off the particles. In the Q-AMS a focused particle beam is generated by an aerodynamic lens. This beam impacts on the heated surface of a vaporizer which is located in the ionization region of a Quadrupole Mass Spectrometer. The ionization of the vapor is achieved by electron impact ionization with electrons of 70 eV. The Q-AMS is routinely operated in two alternating modes. In the so called MS-mode, the size integrated particle composition is measured. In the so called pToF-mode (particle time of flight mode), the size resolved composition is observed for a set of selected mass to charge ratios (m/z). From the combination of the MS-mode and the pToF-mode data, which were measured alternating with 5 min time resolution, the size dependent chemical composition over the accessible particle size range is retrieved. Details of the mode of operation of the Q-AMS are described in Jayne et al. (2000). Water, mainly detected on the m/z 18 (H 2 O + ), and CO 2 , mainly detected at m/z 44 (CO + 2 ), served to determine the bicarbonate and carbonate content. Water and CO 2 were quantified from the MS spectra as described in Allan et al. (2004). The relative ionization efficiency (RIE) of CO 2 was set to 1.4, for H 2 O we applied an RIE of 2 which was recently determined by Mensah et al. (2010). Blank measurements with a HEPA filter in line were conducted before each experiment and the m/z 18 and m/z 44 were lower than 0.3 µg m −3 and 0.04 µg m −3 , respectively. During experiments the value of m/z 18 and m/z 44 for the aerosols was up to tens of µg m −3 .
The detection and discrimination of Ca(HCO 3 ) 2 and CaCO 3 by the Q-AMS was accomplished by tuning the temperature of the vaporizer to 300 • C or 900 • C according to the thermal properties of Ca(HCO 3 ) 2 and CaCO 3 . When the temperature of the vaporizer was set to 300 • C, Ca(HCO 3 ) 2 in the aerosol decomposed at the vaporizer according to (Reaction R2) and was detected in the form of CO 2 and H 2 O. When the temperature of the vaporizer was set to 900 • C, both Ca(HCO 3 ) 2 and CaCO 3 are decomposed Reactions (R3, R4), whereat Ca(HCO 3 ) 2 should deliver two CO 2 and one 8604 D. F. Zhao et al.: Novel method of generation of Ca(HCO 3 ) 2 and CaCO 3 aerosols H 2 O per formula unit, whereas CaCO 3 is supposed to evaporate only one CO 2 per formula unit.
By interplay of the annealing temperature of the tube furnace and the temperature of the vaporizer of the Q-AMS the content of CaCO 3 and Ca(HCO 3 ) 2 in the aerosol was determined according to Reactions (R2-R4). (Defacto we bypassed the tube furnace for the measurements at room temperature, i.e. we measured before and after annealing in the tube furnace.). Because of the gas-phase suppression factor of 10 7 of the Q-AMS, the contribution of H 2 O(g) and CO 2 (g) evaporating from Ca(HCO 3 ) 2 in the furnace on m/z 18 and m/z 44 is negligible.
The particle mass distribution as function of the vacuum aerodynamic diameter was observed in the pToF mode of the Q-AMS. By comparing the modal positions of this particle mass distribution and of the particle volume size distribution calculated from the number size distribution measured by the SMPS, the effective particle density was obtained (DeCarlo et al., 2004).
In order to get information about the morphology of the particles, aerosol was sampled before and after annealing on a copper grid using a home-built aerosol sampling device (Marquardt et al., 1992) and characterized by Transmission Electron Microscopy (TEM). In the sampling device, a high voltage was applied and aerosol particles were deposited in the electric field onto the copper grid which was positioned in the centre of the gas flow.

HTDMA operation
The hygroscopic growth of the particles was measured with a home built HTDMA (Buchholz, 2007). With the first DMA (TSI 3071) particles were selected from the dried polydisperse aerosol with an electromobility diameter of 150 nm. The second DMA (TSI Model 3071) was operated in a scan mode and the particles were detected by a condensation particle counter (TSI CPC 3022A). Both DMA were operated with 0.3 lpm aerosol sample flow and 3 lpm sheath air flow.
The size selected aerosol and the sheath air of the second DMA were humidified at a temperature of 27 • C to almost the same relative humidity (RH) with the sheath air being at slightly higher RH. The second DMA is placed in an insulated box and cooled to 20 • C. Before entering the second DMA both aerosol and sheath air were cooled down to the same temperature and thus the RH increased to its final value. The particles remained for approximately 20 s in contact with the final humidity before they enter the SMPS which was operated with sheath air of the same humidity. To determine the RH inside the second DMA the absolute humidity was calculated from RH and T data of two Vaisala HMP235 sensors before and after the insulated area. In addition the temperatures of the air flows inside the insulated area were measured with PT100 sensors directly before or after they entered or left the DMA. For calibration ammonium sulfate aerosol was used.

CCN operation and evaluation
To determine the CCN activity, the polydisperse aerosol was size selected by scanning a DMA between 10 and 450 nm. For each size bin the total number of particles (CN) was detected with an ultrafine water CPC (UWCPC, TSI 3786) and the number of activated particles (CCN) at different supersaturations (SS) was measured in parallel with a continuous flow CCNC (Droplet Measurement Technologies, DMT-100). The ratio of CCN to CN is called activated fraction. Since a DMA selects electromobility diameters, multiple charged particles with accordingly larger diameters entered the CCN counter, too. The fraction of these particles contributes to activation at the small diameters. For a correct interpretation of activation diameters the fraction of multiple charged particles was calculated according to a natural charge distribution (Wiedensohler, 1988). The activated fraction was then determined for each charge class separately and an error function was fitted to the data. The dry activation diameter (or critical dry diameter, D crit ) is the turning point of this distribution.
For each SS at least three scans were performed and the resulting D crit were averaged. For calibration D crit of ammonium sulfate was measured for five different SS.

Aerosol chamber experiments
The long term behavior of the carbonate aerosols was studied in the Jülich Large Aerosol Chamber (Mentel et al., 1996). Before any experiment with Ca(HCO 3 ) 2 /CaCO 3 we determined the background concentration of HNO 3 in the chamber by sampling a continuous air stream of 1 lpm through a stripping solution (1% CaCl 2 w/w) in a washing bottle. The stripping solution was backed-up by a second washing bottle with the same solution. The nitrate content of the stripping and the backup solution were analyzed by ion chromatography. The background HNO 3 concentration was estimated to be 1.1 ± 0.5 ppb at a RH of 30%.
In order to generate the aerosols the solution containing Ca(HCO 3 ) 2 was prepared as described above and sprayed by ten two-fluid nozzles against the walls of a small prechamber to impact larger droplets (Wahner et al., 1998). The generation air stream of 140 lpm served to flush the submicron portion of the Ca(HCO 3 ) 2 (aq) into the large chamber. This air stream is saturated with water vapor and the aerosol is conditioned in the large chamber by mixing with the chamber air. Nevertheless, the RH rises in the large chamber by the generation air stream; typically a flushing for 20 min raised the RH by less than 2% at 290 K. solutions with a TSI3076 constant output nebulizer. The aerosols were dried to RH < 5% in a diffusion drier using Silica gel as desiccant.

Fig. 2.
Typical dry size distribution of aerosols generated by atomizing the Ca(HCO 3 ) 2 solutions with a TSI3076 constant output nebulizer. The aerosols were dried to RH <5% in a diffusion drier using Silica gel as desiccant.
We performed three types of experiments. In the first experiment the large aerosol chamber was flushed for several hours with dry synthetic air (Linde LiPur 6.0, purity 99.9999%) without CO 2 (final RH < 3%) and only the aerosol was added in one step. The experiment lasted 36 h and was repeated once wherein the aerosol was monitored for 24 h. In the second experiment the chamber was flushed with particle free outside air leading to a RH of 40%. The outside air contained about 400 ppm CO 2 . The aerosol was added to the chamber in two steps with 3 h in between. In the third experiment the chamber was flushed with dry synthetic air without CO 2 , but then humidified by evaporating water (MilliQ) to approximately 40% RH. Initially 370 ppb NO, 100 ppb NO 2 , and 370 ppb O 3 were filled into the chamber and after 2 h the aerosol was added. In all cases we measured the size distribution, CCN activation, hygroscopic growth, and the ratio Ca(HCO 3 ) 2 /CaCO 3 .

Size distribution and stability of aerosol generation
Different CO 2 bubbling durations were applied at room temperature to determine the time required for formation of sufficient Ca(HCO 3 ) 2 and for the dissolving process to reach equilibrium. For this, the total number concentration N tot of the dried aerosol generated from the respective solutions served as a measure. N tot increased fast with increasing bubbling duration but leveled off after >2 h of bubbling. After 1 h of CO 2 treatment the aerosol number concentration has reached 80% of the maximum concentration.
To ensure equilibrium, CO 2 was bubbled through the stirred, aqueous CaCO 3 suspension for 3 h to prepare the Ca(HCO 3 ) 2 solutions. A typical size distribution of aerosol  generated from Ca(HCO 3 ) 2 solution is shown in Fig. 2. The number concentration peaked at ∼120 nm with a geometric standard deviation of ∼1.8.
The stability of the solution was tested by continuous spraying of the same solution over a longer period. N tot and the count median diameter of the number size distribution served as a measure of stability (Fig. 3). First the number concentration of the aerosol decreased with spraying time and then leveled off after 100 min. Meanwhile, the median size of aerosol decreased gradually from 120 nm to ∼80 nm over 2 h. During spraying of the solution a white deposit was formed in the storage bottle and the connecting tubes. This can be explained, since the solution was sprayed with synthetic air, free of CO 2 . The solution lost excess CO 2 and solid CaCO 3 was formed. As CaCO 3 deposits from the solution, the concentration of Ca(HCO 3 ) 2 aerosol was decreased resulting in a decrease of both the number concentration and median size of the particles.

Particle density and morphology
The particles were size selected at an electromobility diameter of 150 nm (Fig. 4). The peaks at 150 nm, 233 nm, and 300 nm corresponded to the single, double and triple charged particles. After passing the tube furnace at 300 • C, the size of the particles decreased to 138 nm, 216 nm, and 279 nm, respectively.
A corresponding decrease of the vacuum aerodynamic size was observed in the pToF mode of the Q-AMS as can be recognized in Fig. 5. From the mode positions of the mass size distribution (D max va ) measured by Q-AMS and of the volume size distribution (D max em ) derived from the SMPS data, an effective density (ρ eff ) was obtained according to (1)  The respective ρ eff were 1.83 ± 0.1 g/cm 3 before the furnace and 1.79 ± 0.1 g/cm 3 after the furnace, thus the same within the errors. The effective density of the generated aerosol is lower than the bulk density of calcite or aragonite of 2.71 and 2.83 g/cm 3 , respectively, which implies that it is not compact crystalline CaCO 3 . TEM was used to obtain images of the particles as shown in Fig. 6. Aerosol particles generated from the Ca(HCO 3 ) 2 solution were spherical with a heterogeneous distribution of round structures, some of which look like hollow pits on the surface or spherical structures inside. After annealing in the furnace, these round structures became much finer and evenly distributed on the surface implying that the particles restructured during the annealing process.

Particle chemical composition
The composition of the aerosols was retrieved from the Q-AMS data. Remember, that we alternated the combination of two heated devices: the tube furnace at 300 • to anneal the aerosols before the Q-AMS measurement and the vaporizer of the Q-AMS to flash-evaporate the particles for detection. (We will strictly use these notations furnace and vaporizer in the following.) We compared the following four combinations (see also Table 1): Q-AMS vaporizer at 300 • C probing before (case 1) and after (case 2) tube furnace at 300 • C, Q-AMS vaporizer at 900 • C probing before (case 3) and after (case 4) the tube furnace at 300 • C.
The Q-AMS data were converted from mass to mole numbers and normalized to the Ca content retrieved from the respective SMPS data as follows. To get the aerosol mass concentrations, the volume size distributions were integrated from 60 nm-700 nm and multiplied with ρ eff . Dividing by the molecular weights of Ca(HCO 3 ) 2 and CaCO 3 gives the number of moles of the salts thus of Ca per unit volume of    air before and after the furnace respectively. This is an anticipation of the following results, thus a working hypothesis at this stage. The normalization to Ca as a conserved quantity is very helpful but not crucial for the following analysis. (The normalization to SMPS data compensated for day to day variations in the aerosol concentrations. As will be shown below an analysis based on normalization to the particle mass instead of Ca led to the same results.) Of course it would have been easier to normalize to a Q-AMS internal quantity. However, neither Ca nor total mass of the carbonates can be measured quantitatively by the Q-AMS since the CaO resulting from Reactions (R3 and R4) is not flashevaporating completely at Q-AMS vaporizer temperatures of 300 • C or 900 • C. However we can also benefit from the normalization to SMPS based Ca since it allows for testing the collection efficiency (CE) of the Q-AMS. Note already here, that the CE was ∼50% in case 1 and 3. The CE was close to 100% in case 4 and cannot be determined for case 2. This will be discussed in more detail below.
n CO 2 and n H 2 O : detectable mole numbers of CO 2 and H 2 O per formula unit of xCa(HCO 3 ) 2 · yCaCO 3 · zH 2 O The molar ratios CO 2 /Ca, H 2 O/Ca based on Q-AMS/SMPS and the molar ratio of H 2 O to CO 2 based on the Q-AMS data only are shown in Fig. 7a, b and c, respectively. The error bar represents the standard deviation of repeated measurements. As mentioned above, at the Q-AMS vaporizer temperature of 300 • C we expect to detect Ca(HCO 3 ) 2 by one CO 2 and one H 2 O by the Q-AMS, because of the incomplete thermal decomposition (R2). At the vaporizer temperature of 900 • C thermal decomposition (R3) is complete and two CO 2 and one H 2 O should be detected. CaCO 3 is not detected at 300 • C but at 900 • C where one CO 2 is formed.
For the aerosol before annealing significant amounts of CO 2 were detected at 300 • C Q-AMS vaporizer temperature (case 1, Fig. 7a). The measured amount doubled when the vaporizer was turned to 900 • C (case 3, Fig. 7a). The detected amount of H 2 O was the same for both vaporizer temperatures (Fig. 7b). The molar ratio of H 2 O to CO 2 was close to 1 at 300 • C and close to 0.5 for 900 • C, respectively (Fig. 7c). This means that before annealing the aerosol consisted of nearly pure Ca(HCO 3 ) 2 with no additional water in the particles and that all Ca(HCO 3 ) 2 was destroyed on the vaporizer of the Q-AMS at 300 • C. Nevertheless, the ratios CO 2 /Ca and H 2 O/Ca in cases 1 and 3 are only half as large as expected from stoichiometry indicating a reduced CE of the Q-AMS.
For particles that were annealed in the tube furnace to 300 • C neither CO 2 nor H 2 O was detected at 300 • C vaporizer temperature (case 2), because all thermally available CO 2 and H 2 O were already driven out in the tube furnace ( Fig. 7a and b). This also means that after annealing in the furnace, the aerosol consisted of nearly pure CaCO 3 . In case 2 small values near the detection level were leading to the large errors bars in Fig. 7c. At 900 • vaporizer temperature (case 4) we observed only a very small amount of H 2 O, but a large signal of CO 2 (Fig. 7b and 7a).
If the particles consisted of pure Ca(HCO 3 ) 2 before annealing and of pure CaCO 3 thereafter, one CO 2 is lost by conversion of Ca(HCO 3 ) 2 to CaCO 3 (Reaction R2) in the tube furnace. Therefore we should observe for Q-AMS vaporizer temperatures of 900 • C only half of the CO 2 after annealing. However, comparing the cases 3 and 4 the CO 2 signals before and after annealing have the same magnitude. This cannot be due to incomplete conversion of Ca(HCO 3 ) 2 to CaCO 3 because only little water is observed in case 4. Also large shape factor effects can be excluded since the particles kept their overall spherical shape after passing the furnace (Fig. 6). The observed as expected mole ratios of H 2 O/CO 2 in cases 1 and 3 (Fig. 7c) and the observed as expected doubling of the CO 2 amount going from case 1 to case 3 excludes simple effects of the different vaporizer temperatures.
The observed CO 2 /Ca ratios (Fig. 7a) and H 2 O/Ca ratios (Fig. 7b) indicate that the CE of the Q-AMS must be differed between the cases. The data are consistent with a CE of about 50% in cases 1 and 3 and near 100% in case 4. This is dissatisfying. Collection efficiencies for solid particles below 100% have been observed before. This is explained by solid particles bouncing off the vaporizer surface before full evaporation (Matthew et al., 2008). Note in Fig. 5 that our mass distributions were within the transmission window of 60-600 nm, and the little tailing at large diameters into the fall off range of the aerodynamic lens cannot cause effects of a factor of two. Unfortunately, we intrinsically cannot observe a signal in case 2. This prevents testing if annealing of the particles at 300 • C and conversion from Ca(HCO 3 ) 2 to CaCO 3 has led to higher CE. However, higher CE for tempered solid particles compared to freshly dried particles seems to be somewhat anti-intuitive. We repeated the experiments with virtually the same results, i.e. enhanced CO 2 and CE in case 4 is not a singular artifact. It remains unexplained for the moment, but adds only little uncertainty to our findings. All other evidence supported that we produced pure Ca(HCO 3 ) 2 aerosol by spraying the solution and drying the aerosol to RH < 5% and that we were able to convert it to CaCO 3 by annealing it at 300 • C. Moreover, because of the H 2 O/CO 2 stoichiometry in cases 1 and 3 (Fig. 7c) and the absence of excess water (Fig. 7b, c), we can rule out inclusions of aqueous phase and the formation of metastable CaCO 3 hydrates as described e.g. by Brecevic and Kralj (2007).  The general composition of the generated carbonate aerosols can be written as xCa(HCO 3 ) 2 · yCaCO 3 · zH 2 O. The combination of furnace temperature and the Q-AMS vaporizer temperature and the corresponding species detected are listed in Table 1. If the total calcium salt is normalized to unity, then y equals 1 − x. The parameters x, y, z then can be obtained from the Eqs. (2-7): Herein n nCO 2 (900 • ) is the mole number of CO 2 in the aerosol normalized to the aerosol mass concentration before annealing (22 • C) detected at 900 • C in the AMS and n nH 2 O (300 • ) and n nCO 2 (300 • ) are the mole numbers of H 2 O and CO 2 normalized in the same way but detected at 300 • C in the AMS. The composition of the aerosol before and after annealing was determined this way and is shown in Table 2. For the aerosol before the furnace, x was near unity and y and z were The measured mass concentrations of CO 2 and H 2 O were normalized to the aerosol mass concentration derived from SMPS volume and effective density ρ eff . Calc. refers to the compositions obtained in Table 2. zero within the error limits (the value of x should be less than or equal to 1 and x, y, z should be positive) which means almost the whole particle consisted of Ca(HCO 3 ) 2 . The ratio of z to x was near zero indicating there is no additional water. In contrast, after the furnace, the particle consists of almost pure CaCO 3 with very little Ca(HCO 3 ) 2 being left. This is consistent with the analysis discussed above.
From the composition of the aerosol in Table 2, the mass concentration of CO 2 , H 2 O normalized to the aerosol mass concentration were calculated as a theoretical value and compared with the measured value (Table 3). The ratio of measured to calculated concentrations gives an estimate of CE supporting the above mentioned values of ∼50% in cases 1 and 3 and of 100% in case 4.
Based on the analysis of the composition of the aerosol above, we will refer to the dried aerosol before annealing in the furnace as Ca(HCO 3 ) 2 aerosol and to the aerosol after annealing as CaCO 3 aerosol in the following. We hypothesize that the Ca(HCO 3 ) 2 particles were solid, although it is known that solid Ca(HCO 3 ) 2 is thermodynamically instable at ambient conditions. The CaCO 3 particles were pure solids within the errors of the analysis but may have contained traces of Ca(HCO 3 ).

Hygroscopicity of the aerosol
The humidograms of the laboratory generated aerosols are shown in Fig. 8. CaCO 3 (s) aerosol (red dots) took up only very little water at very high RH and deliquescence was not observed up to RH of 97.4(±1.4)%. We calculated hygroscopicity parameters κ according to the one parameter representation of the Köhler equation proposed by Petters and Kreidenweis (2007). By averaging κ in the range 97-97.4% RH we obtained a κ GFCO3 of 0.0016 ± 0.0004.
Dried Ca(HCO 3 ) 2 aerosols (blue triangles) showed a different growth behavior. Up to RH of 71(±1.4)% the particle diameter increased by 1%. Between 71% and 78% the growth factor (GF) decreased and reached a minimum of 0.95 at RH 89.5(±1.4)%. Thereafter the particles grew continuously with increasing RH without showing a distinct deliquescence point. Condensation of small amounts of water (GF = 1.01 at 71% RH) obviously caused shrinking of the particles probably due to some water aided restructuring of the particles (compare the TEM image in Fig. 6a, which indicates inhomogeneities on or in the particles.). Shape effects can be excluded since the dried particles were already spherical.
To test the effect of wetting the dry Ca(HCO 3 ) 2 particles, the aerosol was humidified to almost 100% and then dried again before entering the instruments. After the preconditioning (pc) the minimum in the growth curve disappeared and the particles did not grow until ca. 92% (Fig. 8, green squares). The uptake of water already occurred at lower RH and was higher than for CaCO 3 . No distinct deliquescence point was observed. We calculated κ in the range 97-97.5%RH and obtained an average κ GFpc of 0.011 ± 0.007 for the pre-conditioned Ca(HCO 3 ) 2 aerosol.
When the growth curve of the non-conditioned (nc), dry Ca(HCO 3 ) 2 aerosol was shifted using the diameter of the minimum at RH 89.5% as reference diameter (Fig. 8, black triangles), the high humidity branches at RH > 93% for the particles with and without pre-conditioning agree within the errors, thus should have the same composition and morphology. We calculated an average κ GFnc = 0.008 ± 0.003 between 97-97.5% RH for the non-conditioned Ca(HCO 3 ) 2 aerosol from the shifted humidogram. This is smaller than κ GFpc but still agrees within the uncertainties.
The Q-AMS analysis showed that the pre-conditioned and then dried again particles consisted of ∼67% CaCO 3 and ∼33% Ca(HCO 3 ) 2 . The conditioning process can be imagined as dissolving some bicarbonate in the water layer, where it decays and CO 2 evaporates. This is the same process as occurring in weathering of carbonate rocks. In comparison to the pure CaCO 3 , the presence of Ca(HCO 3 ) 2 increased the growth factors and thus shifted the growth curve to lower RH. The moderate growth at RH < 71 % of the nonconditioned, dried Ca(HCO 3 ) 2 particles and the restructuring supports our hypothesis that these particles were solid. It must be a metastable solid phase; the sudden evaporation of water in the drying process seems to hinder the formation of thermodynamically stable CaCO 3 . Also the pre-conditioned, mixed particles did not show growth at RH < 92% although they contained 33% Ca(HCO 3 ) 2 . This indicates that these mixed particles were also solid. Ca(HCO 3 ) 2 (s) seemed to become unstable at RH > 71% even on the short time scales of the HTDMA measurement. It is likely that with increasing humidity the composition of the particles shifted towards ∼67% CaCO 3 and ∼33% Ca(HCO 3 ) 2 , as indicated by the similarity of κ GFpc and κ GFnc . Because of this instability the hygroscopic growth behavior of pure Ca(HCO 3 ) 2 could not be determined. However, it should be much more hygroscopic than CaCO 3 (s) and the CaCO 3 /Ca(HCO 3 ) 2 (s) mixed salt as indicated by the GF of 1.01 at RH 71% (see Fig. 8, blue triangles).
We can estimate κ HCO3 for Ca(HCO 3 ) 2 under the assumptions of constant composition of the mixed CaCO 3 /Ca(HCO 3 ) 2 (s) aerosol over duration of the succeeding humidogram and a homogenous distribution of Ca(HCO 3 ) 2 and CaCO 3 within the particles, thus equal availability for the condensing water: Herein f x are the volume fractions of CaCO 3 and Ca(HCO 3 ) 2 , respectively. For simplicity we assume that CaCO 3 and Ca(HCO 3 ) 2 feature the same density, in agreement with the equal ρ eff of dried Ca(HCO 3 ) 2 and annealed CaCO 3 particles. Moreover, volume additivity (ideality) is intrinsically assumed in the κ approach. Assuming the value of κ GFCO3 = 0.0016 is applicable to CaCO 3 , we obtained  κ HCO3 = 0.030 from Eq. (8). If we apply κ CO 3 = 0.0011 ± 0.0004 reported by Sullivan et al. (2009b), we yield only a little larger κ HCO3 = 0.031. However, κ HCO3 = 0.03 is probably only a lower limit, since our experiments in the dry Large Aerosol Chamber indicated the preferential formation of CaCO 3 (s) in the outer layer of the particles (see below).

CCN activity of aerosol
We determined the dry diameter of those particles which are activated at a given supersaturation (SS) with a CCN counter. We will call this the critical dry diameter (D crit ) which is different from another often used definition of D crit as diameter of the droplet at the maximum of the Köhler curve. The measured D crit are shown in Table 4 and in Fig. 9. CaCO 3 particles (red filled squares) activated at significantly larger diameters than the Ca(HCO 3 ) 2 (s) particles (blue dots) and the CaCO 3 /Ca(HCO 3 ) 2 (s) mixed particles (green dots). The D crit of our spray-dried-annealed CaCO 3 agreed well with observations of Sullivan et al. (2009b), who measured CaCO 3 aerosol generated by dispersing dry powder (open red squares). The red solid line represents one parameter Köhler calculations with the average κ CNNCO3 of 0.0019 ± 0.0007 for CaCO 3 from this study. The values of κ GFCO3 and κ CNNCO3 determined by GF and CCN measurements D. F. Zhao et al.: Novel method of generation of Ca(HCO 3 ) 2 and CaCO 3 aerosols 8611 agree well within the error limits. Sullivan et al. (2009b) determined the corresponding average κ CO 3 = 0.0011 ± 0.0004 (red dotted line). The results of our study feature systematically larger κ CO 3 , thus somewhat easier droplet activation. However, the average values of both studies agree within the error limits. The red dashed line in Fig. 9 represents the average κ CO 3 based on the combination of results of this study and of Sullivan et al. (2009b).
Wet generated water-insoluble or mineral particles often exhibit a severely enhanced hygroscopicity and cloud droplet activation compared to dry generated particles (Herich et al., 2009;Koehler et al., 2007Koehler et al., , 2009Sullivan et al., 2010), which was suggested to be due to the soluble materials on mineral particles introduced by atomizing aqueous suspensions. Impurities also have been suggested to be one explanation for inconsistencies among CCN measurements for adipic acid aerosols generated by atomization of aqueous solutions (Hings et al., 2008). Sullivan et al. (2010) reported a κ of 0.1 for wet-dispersed CaCO 3 compared to κ CO 3 = 0.0011 of drydispersed CaCO 3 aerosols. Considering this huge difference of a factor of 100 in κ, the small difference within the errors between the CaCO 3 aerosol generated by our method and the dry-dispersed method of Sullivan is negligible. Both results agree very well, although trace amounts of impurities may also be present in this study, either from the solution or imported with the mineral particle that is concentrated when the droplets dry. This indicates that impurity interference is not important in this study. Moreover, the impurity content in the particles was estimated roughly. To prepare the Ca(HCO 3 ) 2 solutions, 2 g of CaCO 3 with a purity of ≥99% were added to 1 L of Milli-Q-water (total organic carbon <5 ppb, total ion concentration <0.05 ppb), i.e. the maximum impurity in the solution is 0.002%. The number mean droplet diameter of the atomizer specified by the manufacturer is 350 nm for water, and the mean particle diameter after drying is 120 nm, therefore the concentrating factor is ∼25. If the same density is assumed for the impurity and CaCO 3 /Ca(HCO 3 ) 2 , it would account for a maximum of 0.05% of the particle mass. Therefore, we conclude that our CaCO 3 aerosol is of sufficient quality as required by hygroscopicity and CCN studies.
To our best knowledge there are no studies of the CCN activation of Ca(HCO 3 ) 2 as it was always thought to be too unstable under ambient atmospheric conditions. We are confident that we delivered pure Ca(HCO 3 ) 2 (s) particles (non-conditioned, nc) and CaCO 3 /Ca(HCO 3 ) 2 (s) particles (pre-conditioned, pc) to the CCN spectrometer. The pre-conditioning leading to the CaCO 3 /Ca(HCO 3 ) 2 (s) mixed salt did not cause large effects on the CCN activity at SS > 0.25% (compare blue and green dots in Fig. 9). Obviously, the presence of 33% Ca(HCO 3 ) 2 (s) in the mixed particles was sufficient to cause onset of CCN activation at almost the same critical parameters as in the pure Ca(HCO 3 ) 2 (s) particles. The blue and the green solid lines are Köhler calculations with κ CCNpc = 0.14 and κ CCNnc = 0.16. With exception of one point at the lowest SS all data points fall on both κ-lines within their errors. In any case dried Ca(HCO 3 ) 2 (s) and mixed CaCO 3 /Ca(HCO 3 ) 2 (s) particles were a factor of 60-70 better CCN than CaCO 3 particles.
Both, κ CCNpc = 0.14 and κ CCNnc = 0.16 are much larger than the corresponding κ GFpc and κ GFnc of 0.011 and 0.08. This can be explained by the limited solubility of Ca(HCO 3 ) 2 described in the literature. The solubility of CaCO 3 is ≈0.015 g/l = 1.5·10 −4 mol/l in a CO 2 free atmosphere at 25 • (Gmelin 1961, p. 933). Raising the CO 2 pressure up to 1 bar (e.g. by bubbling CO 2 to CaCO 3 slurry) increases the solubility by a factor of 50 to about 0.008 mol/l (Gmelin 1961, p. 933). The additionally dissolved CaCO 3 with increasing CO 2 pressure exists as HCO − 3 in the solutions (Gmelin 1961, p. 927). We approximate the saturation concentration of Ca(HCO 3 ) 2 by about 0.01 mol/l at room temperature and the according saturation mole fraction by 0.01/(0.01+55.5) ≈0.0002. The corresponding saturation ratio for an ideal solution would be 0.9998. We thus estimate a deliquescence humidity of Ca(HCO 3 ) 2 around 99% RH. This estimate is reconcilable with our observations of minimal hygroscopic growth for the mixed salt below 92%RH and the absence of deliquescence below 97.5%RH (Fig. 8, green squares). Ca(HCO 3 ) 2 appears to be a non-hygroscopic salt below 97.5%RH, because of its high deliquescence point, but a good CCN because of its still notable solubility. Ca(HCO 3 ) 2 is somewhat less CCN active than Ca(NO 3 ) 2 (black open squared symbols in Fig. 9, taken from Sullivan et al., 2009b), but it is a better CN than natural mineral dust samples. This is shown in Fig. 9 where the shaded area comprises the range of expectation for CCN activation of atmospheric mineral dust and the black dashed line the median κ of 0.03 suggested by Koehler et al. (2009). In addition κ is given for Arizona Test Dust (κ = 0.025, blue dashed line) and Saharian Dust (κ = 0.054, brown dashed line) (Koehler et al., 2009). Ca(HCO 3 ) 2 is the expected product of the reaction of CaCO 3 with CO 2 and H 2 O, which have high concentrations in the atmosphere. Thus, especially in cloud droplets of activated mineral dust some Ca(HCO 3 ) 2 may be formed. Several studies on the water soluble inorganic ions in aerosols conducted during ACE-Asia have suggested the co-existence of Ca and carbonate/bicarbonate based on the high correlation between Ca 2+ and the difference of total cations and anions (Maxwell-Meier et al., 2004;Song et al., 2005). Unfortunately, carbonate and bicarbonate were not measured and could not be differentiated in the particle-into-liquid sampler coupled to a dual-channel ion chromatograph (PILS-IC). The significant enhancement of CCN activity going from CaCO 3 to Ca(HCO 3 ) 2 indicates that CaCO 3 could become much more CCN active by natural processes especially in remote area when the heterogeneous reactions with pollutant trace gases are negligible.

Long term experiments in the Large Jülich Aerosol Chamber
Ca(HCO 3 ) 2 (aq) was introduced into the Large Aerosol Chamber and kept there for up to 36 h. CCN activation, hygroscopic growth factors, and chemical composition by Q-AMS were monitored to capture the long time evolution of the carbonate aerosols under dry and humid conditions. Table 5 provides an overview of the experimental conditions. The critical diameters D crit at 0.2% SS of the carbonate particles are shown in Fig. 10. In the two dry experiments D crit at 0.2% SS was 127 nm (κ ≈0.17) directly after generating the Ca(HCO 3 ) 2 aerosol, which is about the expected value of κ ≈0.15 for Ca(HCO 3 ) 2 (s) and CaCO 3 /Ca(HCO 3 ) 2 (s) according to our laboratory scale study above. Thus we assume that we formed some Ca(HCO 3 ) 2 (s). During the first 3.5 h the activation diameter increased to 150 nm at 0.2% SS (κ ≈0.1) and stayed constant thereafter for the next 5 h. The results of the two experi-ments differ in that the red curve approaches the final value of 150 nm at 0.2% SS within 2 h and stays constant thereafter, whereas the black curve took somewhat longer to reach the same value. After 10 h the black curve started to rise again and reached values of 170 nm at 0.2% SS (κ ≈0.07) within 6 h. We have no explanation for this different behavior, which is not related to changes of RH or T in the chamber. However, we can state that after 22 h/36 h the hygroscopicity parameters κ of 0.1/0.07 were smaller than κ ≈0.15 observed for the mixed CaCO 3 /Ca(HCO 3 ) 2 (s) from the laboratory study. Still, the chamber particles were much better CCN than pure CaCO 3 particles with κ = 0.0019 ± 0.0007.
Since the tendency for droplet activation was lower than that of the mixed CaCO 3 /Ca(HCO 3 ) 2 (s) one would expect that the hygroscopic growth was also smaller. However, as is shown in Fig. 11 the aerosol from the chamber (blue triangles, black rhombs, and turquoise triangles) had significantly higher GF than the mixed CaCO 3 /Ca(HCO 3 ) 2 (s) after preconditioning (green dots). Still, at 95% RH the GF is only 1.165 which translates into κ = 0.18.
The Q-AMS data showed that the aerosol in the dry chamber after 14 h contained ∼40% Ca(HCO 3 ) 2 , whereas the mixed CaCO 3 /Ca(HCO 3 ) 2 (s) aerosol after pre-conditioning contained only 33% bicarbonate. We detected some nitrate uptake of background HNO 3 , which was initially about 2% of the CO + 2 mass and increased slowly to about 4% of the CO + 2 within 14 h. Therefore, besides residual Ca(HCO 3 ) 2 , also Ca(NO 3 ) must have contributed to the GF and the CCN activity of the particles after 14 h.
These observations can be rationalized by the following considerations. Upon spraying the Ca(HCO 3 ) 2 solution and drying it by mixing with the dry air in the chamber primarily Ca(HCO 3 ) 2 (s) aerosol was formed. Favored by the low RH and CO 2 free synthetic air in the chamber the particles lost water and CO 2 over several hours and were slowly converted to the thermodynamically stable CaCO 3 . However, even after 36 h the particles were much better CCN (κ of 0.1 or 0.07, respectively) than CaCO 3 (κ = 0.0019 ± 0.0007). Therefore a large portion of Ca(HCO 3 ) 2 was still left in the particles. This is supported by Q-AMS data that showed that after 14 h the Ca(HCO 3 ) 2 accounted still for ∼40% of the total calcium salt. Very little gas-phase HNO 3 (2%) was taken up instantaneously by the fresh aerosols. Due to the low HNO 3 background and the residual 5% RH, we exclude the formation of significant amounts of surface adsorbed carbonic acid as observed by Al-Hosney and Grassian (2004) at much larger HNO 3 partial pressures. Thereafter nitrate increased only slowly in the dry experiments and had obviously no or small influence since the D crit increased with time. We suggest that the surface of the particles reacted quickly according to (Reaction R2), protecting the core of the particles from further conversion. Thus after the relatively fast initial rise the activation diameter increased only slowly. Such a surface to bulk kinetic constrain was suggested before by Vlasenko et al. (2009) for the reaction of nitric acid dust with Arizona Test Dust. Overall Ca(HCO 3 ) 2 (s) existed much longer under dry conditions and at room temperature than commonly expected from its thermodynamic stability. Based on the significant difference in CCN activity between Ca(HCO 3 ) 2 (s) and CaCO 3 (s), a longer existence of Ca(HCO 3 ) 2 (s) could have important implications for the influence of mineral aerosol on climate.
In the humid chamber (flushed with particle free outside air) D crit at 0.2% SS reached 135 nm (κ = 0.14) within the first 15 min (blue curve in Fig. 10) which agreed well with Ca(HCO 3 ) 2 (s) from our laboratory study (see above). Thereafter we observed a fast decrease of D crit . A second addition of fresh aerosols induced the same behavior. After 20 h activation diameters of 92(±3) nm at 0.2% SS  (κ = 0.45) were reached which were only slightly higher than the 87.5(±3) nm for pure Ca(NO 3 ) 2 (κ = 0.5) according to Sullivan et al. (2009b) (black dashed line in Fig. 10).
In the second humid experiment NO, NO 2 and O 3 were added on purpose to the aerosols. In this mixture HNO 3 was formed by heterogeneous reaction of NO 2 and N 2 O 5 on the particle surfaces and on the chamber walls. The alkaline Ca(HCO 3 ) 2 then obviously reacted with NO 2 and HNO 3 to Ca(NO 3 ) 2 . D crit showed the same fast decay and approached the Ca(NO 3 ) 2 values as described above (green curve in Fig. 10).
Also the GF at 90% RH increased rapidly with time and then levelled off ( Table 6). The water uptake represented as growth curves (Fig. 11, red and orange) was significantly lower than the water uptake measured for pure Ca(NO 3 ) 2 by Gibson et al. (2006), who find a GF already of 1.77 at 85% RH.
In both humid experiments increasing nitrate was detected at m/z = 30 and m/z = 46 by the Q-AMS, which replaced HCO − 3 . This is demonstrated in Fig. 12 for the experiment with NO, NO 2 , and O 3 addition. However, even after 23 h, still nominal 6% (mole) Ca(HCO 3 ) 2 and 30% CaCO 3 can be detected by switching the vaporizer temperature of the AMS between 900 • C and 300 • C. Together the carbonate (including bicarbonate) contributed about ∼36% of total calcium salt. Ca(NO 3 ) 2 was determined to account for ∼64% (molar content) of the total calcium salt. The total volume of the aerosol increased during the first seven hours. Nitrate and water uptake over-compensated the decrease of aerosol volume concentration due to losses to the chamber walls (Fig. 12). These observations demonstrated that a substantial heterogeneous conversion into the more soluble Ca(NO 3 ) 2 occurred. We suggest that in early stages of the experiment this took only place on the surface of the particles. But since the nitrate is much more hygroscopic and much better soluble than both, carbonate and bicarbonate, the particles took up water forming a concentrated solution on the surface (Krueger et al., 2003). This further increased the uptake of HNO 3 which is larger on wet particles (Liu et al., 2008a) and also dissolved the upper layers of carbonate. The conversion of Ca(HCO 3 ) 2 to Ca(NO 3 ) 2 thus was accelerated and the reaction of bicarbonate to carbonate was promoted as no protective layer of CaCO 3 (s) could be formed as suggested for the dry chamber case. Carbonate/bicarbonate decreased as the reaction proceeded but were not completely depleted over the time period of the chamber experiment. Therefore the hygroscopic growth was lower than that of pure nitrate. The κ = 0.45 for droplet activation of chamber particles at the final stages was only 10% less than that of pure Ca(NO 3 ) 2 as the reaction to nitrate was much faster than that conversion to carbonate.

Conclusions
In this study, a novel, easy to operate method was developed using atomization of a Ca(HCO 3 ) 2 solution to produce CaCO 3 aerosol, one of the most reactive components in mineral aerosol. In this procedure CO 2 was bubbled through a slurry of CaCO 3 dissolving it as bicarbonate. The supernatant solution was decanted, sprayed with an atomizer and dried. Annealing the aerosol in a tube furnace at 300 • C promoted the conversion back to CaCO 3 . Q-AMS measurements confirmed that the particles after annealing consisted of pure CaCO 3 . By hygroscopic growth measurements we derived a hygroscopicity parameter κ(97%RH) of 0.0016 ± 0.0004 for the CaCO 3 particles, which agreed with the κ of 0.0019 ± 0.0007 obtained by measurements of CCN activation. The κ from both methods agree very well within the errors and with the κ of 0.0011 ± 0.0004 reported by Sullivan et al. (2009b) for drydispersed CaCO 3 powder. The new method is suited to generate CaCO 3 aerosol of sufficiently high concentration and proper size for studies of hygroscopicity, CCN activity and aerosol chemistry studies in simulation chambers.
Q-AMS measurements further confirmed that the particles obtained by spraying Ca(HCO 3 ) 2 solutions and drying the aerosol to RH<5% consisted of pure Ca(HCO 3 ) 2 . This finding could be highly interesting as we have indication that these particles were solid, although Ca(HCO 3 ) 2 (s) is thermodynamically unstable under ambient conditions. TEM showed that the Ca(HCO 3 ) 2 particles were solid and spherical like the CaCO 3 particles, with Ca(HCO 3 ) 2 (s) particles having some internal structures, bubbles or hollow pits. Since in the Q-AMS study pure Ca(HCO 3 ) 2 was observed without excess water, we can rule out inclusions of Ca(HCO 3 ) 2 solution or the formation of metastable hydrates. In addition, in HTDMA measurements Ca(HCO 3 ) 2 (s) featured restructuring at 71% RH, which should not be observed in case of liquid particles. Processing the Ca(HCO 3 ) 2 (s) particles at 100% RH and drying it again to <5% RH produced mixed CaCO 3 ) 2 /Ca(HCO 3 ) 2 (s) particles. These mixed particles must have been solid too, since they did not show noticeable hygroscopic growth below 92% RH and GF at 95% was only 1.03. The mixed particles and pure Ca(HCO 3 ) 2 (s) were much better CCN (κ = 0.15 ± 0.02) than pure CaCO 3 (κ = 0.0019±0.0007). The difference in CCN activation and hygroscopic growth can be deduced to the limited solubility of Ca(HCO 3 ) 2 (s) and a deliquescence point larger than 97.5% RH, which was not assessable by our HTDMA. We therefore hypothesize that we have observed a new, solid phase of Ca(HCO 3 ) 2 (s). This must have been a metastable phase and the formation of CaCO 3 according the back Reaction (R1) may be hindered when water of the solution evaporates rapidly in the diffusion dryer.
In the dry Large Aerosol Chamber Ca(HCO 3 ) 2 (s) was converted over hours to a mixed salt containing ≈40% Ca(HCO 3 ) 2 (s) and ≈60% CaCO 3 . The uptake of HNO 3 was slow, probably due to a similar kinetic hindrance suggested before on the solid Arizona Test Dust particles (Vlasenko et al., 2009). In the dry chamber Ca(HCO 3 ) 2 (s) could probably only persist, because it was protected by a CaCO 3 coating. Ca(HCO 3 ) 2 (s) was shorter lived at 40% RH in ambient Jülich air in the presence of HNO 3 (≈1 ppb) as well as in synthetic air in presence of several hundred ppb NO x and O 3 . In the latter case, carbonate contributed still about ∼36% (including 6% bicarbonate) after about one day and nitrate accounted for ∼64% (molar content) of the total calcium salt.
The chamber experiments showed that Ca(HCO 3 ) 2 can exist longer under atmospheric conditions than conventionally expected from its thermodynamic stability, especially under dry conditions. In the atmosphere Ca(HCO 3 ) 2 can be formed through the reaction of CaCO 3 with CO 2 and H 2 O, in particular in cloud droplets of activated mineral. A significant increase of CCN activity in going from CaCO 3 to Ca(HCO 3 ) 2 by natural processes may alter the role of mineral aerosols in cloud formation, therefore eventually modifying the impact of mineral aerosol on global climate, especially in remote areas where competing heterogeneous reactions with pollutant trace gases are missing. Ca(HCO 3 ) 2 may be present in the atmosphere for longer times, and likely is faster converted into water soluble salts (e.g. nitrates) than CaCO 3 .