<>Introduction The large 84m³ expansion cloud chamber, AIDA, (Möhler et al., 2003, 2004) was used to conduct controlled studies of heterogeneous ice nucleation of desert dusts. Desert dusts were injected into the chamber and exposed to different temperature-supersaturation environments with respect to ice by quasi-adiabatic expansion of the chamber volume in order to simulate cloud formation. The ice nucleation properties for dust samples taken from the two different Saharan desert locations, the Takla Makan (China) desert region. Experiments with Arizona dust were also made. The aim was to gain an improved understanding of how ice forms heterogeneously in atmospheric ice clouds. Desert dust was studied as it has recently been inferred though measurements of the particle mass fragments in ice crystal residuals that mineral particles and fly ash are predominant in the atmosphere as primary ice nuclei (IN) (Cziczo et al., 2004; DeMott et al., 2003). A large number of experiments has been performed at AIDA, and are currently being assessed. Initial assessment of these cases shows that they include ice formation by: 1) the deposition nucleation mode, 2) the condensation freezing mode, 3) immersion freezing and 4) homogeneous freezing. The experiments were designed in order to isolate a particular nucleation mode. In this paper we present results obtained of the ice number fraction, N_ice, nucleated as a function of temperature and supersaturation for these different dust types.

  <>Method The nucleated ice fraction, N_fraction, the ratio of the number of ice particles activated to become ice to the initial seeding dust particle concentration requires accurate measurements of N_ice. Several methods are available using a variety of instrumentation in the AIDA chamber to determine N_ice. For this study we have used two approaches. The first method uses direct measurements of N_ice from the Small Ice Detector (SID) probe, e.g. Hirst et al., (2001), which measures particle concentration and asphericity thus discriminating between water and ice, in the size range (0.5<D_p<30 mm).  For ice particles that grow outside the accurate range of SID, (10<D_p<1000 mm) concentrations were determined using a combination of measurements from a Cloud Particle Imager (CPI) and theoretical models. Corrections were developed for the CPI sample volume and for mis-sizing for ice crystals as a function of habit for sizes < 40 mm based on laboratory and field studies, Connolly et al. (2006). Theoretical considerations were required to calculate ice number concentrations at the point of primary nucleation as, quite often, ice particles formed from seeds or droplets that were smaller than the low size range of the CPI. Firstly, a time-series of ice and liquid (where droplets were used) number concentration (also segregated with size) was calculated according to Connolly (2005). This data was then used with the measurements of chamber temperature, relative humidity and pressure to retrieve a number concentration.  The growth rate equation for ice particles documented by Chen and Lamb (1994) was used in order to retrieve the true ice concentration at the time of nucleation by integrating backwards in time starting with the size distribution measured by the CPI. Growth trajectories were obtained for size-concentration categories of particles and these were used to infer number concentrations. The point of nucleation and concentrations agreed well with other probes including SID. <> 

Results

            Figure 1                                                                       Figure 2

            Figure 3                                                                       Figure 4 Figures 1-3 show examples of how the fraction of activated ice crystals depended on the variation of relative humidity with respect to ice (RH_ice) for three different dust types, Saharan (SD1 and SD2) and Asian Dust (AD1) at a given temperature, T. The experiments consistently showed a bimodal behaviour in N_fraction with two distinct activation modes whose relative magnitudes were a function of T and RH_ice. Figure 4 shows a surface plot of N_ice as a function of RH_ice and T. Note that through either an increase in temperature or RH_ice, the second mode can become active. At the critical -38^oC level (not shown) homogenous nucleation then became the dominant nucleation mode.

The two modes appeared to behave very differently with one activating below water saturation and the other at water saturation. The high RH mode nucleated on average between five and ten times the fraction of the low RH mode for SD. This behaviour is not accounted for in numerical cloud models.

The magnitude of the second activation mode was found to be similar for both SD and AD dusts, highlighting that parameterisation of ice nucleation on desert dust may be simplified in some cases. Interestingly, the active fraction of AD was observed to increase dramatically when water saturation was approached, up to eight-fold in going from typically 5% sub-saturated with respect to liquid water to just above water saturation. In another experiment, Arizona test dusts (ATD) were examined and they were found to be significantly more active with a high fraction of dust particles being activated as ice nuclei at considerably lower saturation ratios than for AD. These results may have implications for explaining some observations of high ice number concentrations in clouds; however, further work is needed in this area to confirm this hypothesis.

Non-linear regression models were applied to the data to provide a functional description of the bimodal behaviour and were of the form

where N_frac,1 is the fraction of dust activated as IN due to the deposition mode, N_frac,2 is the fraction activated due to condensation freezing and N_frac,3 is the temperature dependence of the condensation freezing mode, N_F is the total dependence of the activated fraction due to both deposition and condensation, s_crit is the critical supersaturation, s_i is the supersaturation with respect to ice and T is the absolute temperature. The parameters, b₁-b₉, were obtained by solving with a non-linear regression technique using the Gauss-Newton method.

            The results will be described in greater detail and the consequences for its inclusion in cloud resolving models discussed.