Sensitivity of cirrus and mixed-phase clouds to the ice nuclei spectra in McRAS-AC: single column model simulations

The salient features of mixed-phase and ice clouds in a GCM cloud scheme are exam-ined using the ice formation parameterizations of Liu and Penner (LP) and Barahona and Nenes (BN). The performance of LP and BN ice nucleation parameterizations were assessed in the GEOS-5 AGCM using the McRAS-AC cloud microphysics framework 5 in single column mode. Four dimensional assimilated data from the intensive observation period of ARM TWP-ICE campaign was used to drive the ﬂuxes and lateral forcing. Simulation experiments where established to test the impact of each parameterization in the resulting cloud ﬁelds. Three commonly used IN spectra were utilized in the BN parameterization to described the availability of IN for heterogeneous ice nucleation. 10 The results show large similarities in the cirrus cloud regime between all the schemes tested, in which ice crystal concentrations were within a factor of 10 regardless of the parameterization used. In mixed-phase clouds there are some persistent di ﬀ erences in cloud particle number concentration and size, as well as in cloud fraction, ice water mixing ratio, and ice water path. Contact freezing in the simulated mixed-phase clouds 15 contributed to transfer liquid to ice e ﬃ ciently, so that on average, the clouds were fully glaciated at T ∼ 260K, irrespective of the ice nucleation parameterization used. Comparison of simulated ice water path to available satellite derived observations were also performed, ﬁnding that all the schemes tested with the BN parameterization predicted average values of IWP within ± 15% of


Introduction
The role of atmospheric aerosols in modulating the atmospheric radiative balance, by directly scattering solar radiation, or indirectly, modifying cloud optical and microphysical properties, has received considerable attention during the last couple of decades. Soluble and insoluble aerosol species provide nucleation sites for the atmospheric wa-curate and physically based schemes a high priority (Intergovernmental Panel on Climate Change, 2007).
Aerosol indirect effects (AIE) in warm clouds have been long studied and implemented in atmospheric models (e.g., Penner et al., 2006), but less has been accomplished for cold clouds. Modifications to the number density and sizes of ice crystals 10 not only strongly affect the radiative properties of ice-bearing clouds, but also impacts the development of precipitation (e.g., Lohmann and Diehl, 2005;Lohmann, 2002). The complexities associated with cold and mixed-phase clouds (due in part to the concurrent action of different freezing mechanisms, the high selectivity of the IN process, and the theoretical uncertainties associated with their description) have challenged the 15 representation of such clouds in GCMs, most of which lack explicit ice microphysics (Lohmann and Feichter, 2005). As a result, even the sign of the radiative effects of aerosol-ice cloud interactions remains uncertain in climate simulations. Important steps to improve the simple treatments of cold and mixed-phase cloud microphysics originally included in GCMs have been undertaken in recent years. For 20 example, the partitioning of cloud condensate between ice and liquid water in mixedphase clouds (235 K ≤ T ≤ 273 K) was typically represented by a temperature-only approach (e.g., DelGenio et al., 1996;Rasch and Kristjánsson, 1998). This approach has been progressively replaced by a less empirical and more physically-based representation, in which the deposition growth of cloud ice at the expense of the liquid water, the Introduction ice content q i , and ice crystal concentration, N c , has been adopted by a variety of GCMs (Sud and Lee, 2007;Liu et al., 2007;Salzmann et al., 2010). Another advancement in GCM cloud schemes is the implementation of two-moment cloud microphyisics, which include prognostic equations for the mass as well as the the number concentration of different hydrometeor categories (e.g., Beheng, 5 2001, 2006;Morrison and Gettelman, 2008). This has permitted the prognostic computation of cloud particles sizes and ice deposition rate (e.g., Salzmann et al., 2010;Muhlbauer and Lohmann, 2009).
Estimates of the AIE on ice-bearing clouds require an adequate description of the aerosol-cloud coupling through the nucleation process. This is, the prognostic calcu- 10 lation of hydrometeor sizes should be done in a manner consistent to aerosol load changes and aerosol characteristics. However, an efficient and comprehensive representation of the current understanding of the ice nucleation process in the framework of a GCM has proven difficult. Most ice nucleation parameterizations rely on simple functions to determine how many ice crystals will be heterogeneously nucleated at 15 a given set of environmental conditions. These relations describing the availability of IN, termed IN spectra, exhibit different level of complexity, ranging from saturationdependent schemes Phillips et al., 2007) to IN spectra with aerosol-dependent parameters derived empirically . Theory-based approaches have also lead to formulations of IN spectra with explicit dependence on 20 aerosol number concentration, aerosol size distribution, and aerosol surface properties (Khvorostyanov and Curry, 2009;Barahona and Nenes, 2009b;Barahona, 2012). Most GCM microphysical schemes that account explicitly for aerosol effects represent ice nucleation assuming that there is no variation in ice nucleation properties within an aerosol species. In reality, there is large variability in the ice nucleation properties 25 of aerosol populations, which contributes to the large uncertainty in the predicted IN concentrations.
Homogeneous freezing of solution droplets (i.e., without the presence of a solid aerosol phase) may occur only at temperatures below 235 K, the homogeneous Introduction  (Pruppacher and Klett, 1997). For temperatures higher than T hom , in which mixed-phase clouds typically exist, the presence of a solid phase is necessary for ice formation, and therefore only heterogeneous ice nucleation is active. Below T hom , where ice-only clouds form, the supersaturation with respect to ice is the result of the competition between the rate of cooling of the cloud parcel and the 5 condensation on the nucleated ice crystals. Therefore, it varies dynamically given the amount of IN present and the dynamical forcing available. Furthermore, since homogeneous and heterogeneous ice nucleation may occur simultaneously, the competition from both mechanisms and their impact on supersaturation further complicate the calculations. For this reason Lagrangian simulations have been used to develop solutions 10 to the variable supersaturation problem (e.g., Lin et al., 2002), and fits to these numerical solutions have been used to develop ice nucleation parameterizations. A few such parameterizations have been developed (Kärcher andLohmann, 2002, 2003;Liu and Penner, 2005), and have been implemented in GCM models (Hoose et al., 2010). Analytical solutions to this problem have been developed in which any IN spectra can 15 be used (Barahona and Nenes, 2009b). For the case of mixed-phase clouds, liquid water, water vapor, and ice are simultaneously present, and can exhibit complex dynamics (e.g., Korolev, 2007). For the coarse resolution of GCM cloud schemes, the simplifying assumption that the water vapor is saturated with respect to liquid water is sometimes made. The supersaturation with 20 respect to ice, S i , is therefore constrained by thermodynamic equilibrium rather than by the competition of cooling and condensation. With this assumption, it is sufficient to know the availability of IN (given by an IN spectrum) at S i to compute the nucleation rate of ice crystals.
A number of studies have focused on the implementation and evaluation of new 25 microphysical schemes in GCM simulations, including prognostic calculation of the ice fraction in mixed phase clouds, and using more physically-based ice nucleation schemes (Storelvmo et al., 2008;Sud and Lee, 2007;Liu et al., 2007;Salzmann et al., 2010). Curry and Khvorostyanov (2012)  In this study, we use the parameterization of Barahona and Nenes (2009b), BN hereafter, in which the ice nucleation problem is treated in a general framework that admits 5 the use of any IN spectra, empirical or theoretical. Barahona et al. (2010) used the BN parameterization to compare common formulations of the IN spectrum in a chemical transport model, finding that the 2 to 3 orders of magnitude variation in the IN concentrations among different schemes would lead to up to a factor of 20 variation in N c in cirrus clouds. The sensitivity can be even larger in mixed-phase clouds where only heterogeneous ice nucleation is active, and competition for water vapor does not buffer the response of crystal number to IN concentration changes.
Testing the impact of IN spectra in a comprehensive cloud microphysical framework would provide valuable information on how the uncertainties associated with ice nucleation are reflected on the cloud field variables when coupled to other cloud processes.

15
In this study, we report the implementation of the BN ice nucleation scheme into the Microphysics of Clouds with Relaxed Arakawa-Schubert and Aerosol-Cloud interaction (McRAS-AC) (Sud and Lee, 2007) driven by the Goddard Earth Observing System Model, version 5 (GEOS-5). The flexibility provided by the BN ice nucleation parameterization is ideal for testing the sensitivity of the simulated cloud properties to the rep-20 resentation of IN spectra in the McRAS-AC framework. To isolate the response from the underlying physical parameterization, all the simulations were performed in the Single Column Model version of GEOS-5. This is a common test of GCM microphysics since the SCM configuration contains the same physical parameterizations as the host GCM model, with the advantage of a much smaller computational burden, and the laterally

Model description and simulation set-up
A detailed description of the McRAS-AC microphysics can be found elsewhere (Sud and Walker, 1999;Sud and Lee, 2007;Bhattacharjee et al., 2010). Here we will primarily focus on describing the treatment of cold and mixed-phase clouds microphysics in McRAS-AC.

Ice nucleation in McRAS-AC
McRAS-AC has the option to invoke the Liu and Penner (2005), (LP), or the Barahona andNenes (2008, 2009a,b) parameterizations to describe the ice nucleation process. This possibility was used to assess and compare the performance of the two schemes. The LP parameterization was originally designed to describe the nucleation process at temperatures typical of cirrus cloud formation, i.e., for temperature less than the homogeneous freezing threshold (T hom = 235 K). It is based on numerical correlations derived from statistical fits to a large number of Lagrangian parcel model simulations, in which homogeneous and heterogeneous freezing mechanisms were explicitly accounted for. The homogeneous freezing of deliquesced sulfate aerosol was 15 approached using an effective freezing temperature. Immersion freezing on soot particles was included in the parcel model simulations using a classical nucleation theory description, in which a fixed aerosol size distribution and freezing characteristics were assumed. Deposition freezing is calculated using the Meyers et al. (1992)  well when extrapolated beyond the curve-fit data domain, nor be applied to aerosols that does not follow the prescribed freezing properties used on the simulations. The implementation of LP in McRAS-AC for mixed-phase clouds (T hom < T < 273 K) follows closely that of Liu et al. (2007), and was described and tested in a SCM framework (Bhattacharjee et al., 2010). In this regime, N c,nuc is calculated by adding the 5 contributions from the the numerical correlations described above and the contribution from deposition freezing as given by a modified version of the Meyers et al. (1992) formula, where N id is the number concentration of ice crystals due to deposition nucleation 10 in m −3 , N 0 = 10 −3 m −3 , a = −0.639, and b = 0.1296, and f (z) is an empirical height correction factor, given by f (z) = 10 (z 0 −z)/δz , with z 0 = 1 km, δz = 6.7 km, and f (z) ∈ [0.12, 1.0]. This decay factor was derived from observations by Minikin et al. (2003) during the INCA (Interhemispheric Differences in Cirrus Properties from Anthropogenic Emissions) campaign, to augment the formula by Meyers et al. (1992) that was derived 15 from ground-level observations. In the present work we implemented and tested the BN parameterization in McRAS-AC. BN is based on an analytical solution of the governing equations of a cooling air parcel in which deliquesced aerosol and heterogeneous IN are allowed to freeze and grow by water vapor deposition (Barahona andNenes, 2008, 2009a,b). Accord-20 ingly, BN circumvents the need for curve-fitted equations, and holds for a wide range of configurations encountered in the physical system. The availability of IN in the BN parameterization can be described, in principle, with any heterogeneous nucleation parameterizations. Here we use the correlations of Meyers et al. (1992) Phillips et al. (2008) (PDA08), and the semi-empirical spectra derived from classical nucleation 25 theory of Barahona and Nenes (2009b) CNT BN is based on an approximation of classical nucleation theory to scale observed IN concentration as a function of supersaturation. For temperatures T < T hom , BN calculates the competing effects of homogeneous nucleation on deliquesced aerosol and the heterogeneous freezing for the availability of water vapor in a forming cirrus cloud. The maximum supersaturation with respect to ice attained in the ascending parcel, S i,max , is 5 calculated by balancing the depletion effect from deposition growth of ice crystals and the availability of water vapor from cooling. In this way, S i,max is given by the dynamics of cooling and ice nucleation. BN then uses the maximum saturation to calculate N c,nuc . The application of BN in the mixed-phase cloud regime differs slightly from that of LP. In the absence of any liquid water, the maximum supersaturation in the parcel would 10 be dictated dynamically by expansion cooling and by the IN concentration. However, in McRAS-AC, any initial condensate is considered to be liquid (Rotstayn, 1997) and its then partitioned following Rotstayn et al. (2000). Therefore, in practice, ice nucleation above T hom is assumed to occur in an environment saturated with respect to water. Under this circumstance, S i,max is fixed by the assumption of water saturation, equal 15 to S i,max = e sw (T )/e si (T ), i.e., the ratio of the saturation vapor pressure over water and over ice, and it is therefore independent of the dynamic forcing, w, or aerosol loading. The number concentration of nucleated ice crystals, N c,nuc , is then calculated by direct application of the IN spectra at the given S i,max . 20 The cloud microphysics in McRAS-AC include balance equations for the mixing ratios of liquid water, q l , and cloud ice q i . The precipitation microphysics are described by Sud and Lee (2007), which recast Seifert and Beheng (2006) to apply it to the thicker clouds of a coarse resolution GCM. The activation of aerosol to cloud droplets follows the parameterization of Fountoukis and Nenes (2005). Aerosol mass concentrations are taken from the Goddard Chemistry Aerosol Radiation and Transport (GOCART), and log-normal size distribution for each species are prescribed. The partitioning of cloud condensate between ice and liquid in mixed-phase clouds is prognostic, and takes into account the Bergeron-Findeisen (BF) process, by which cloud droplets evaporate and the resulting water vapor deposits, to ice crystals. The process is represented following Rotstayn et al. (2000), which explicitly accounts for the dependence of the ice deposition rate on crystal number concentration, N c . 5 Ice crystal number concentration N c is determined in McRAS-AC by the processes of ice nucleation, contact freezing, and by melting of cloud ice. The ice nucleation term is calculated with the LP and the BN parameterizations as explained in Sect. 2.1. Contact freezing of supercooled cloud droplets through Brownian coagulation with insoluble IN (mineral dust) is included as given by Young (1974). 10 Aerosol input for ice nucleation is also based on GOCART aerosol climatology. A single mode log-normal size distribution was assumed for black carbon, with geometrical mean diameter, d g = 0.04 µm, and a geometric standard deviation σ g = 2.3 (Jensen and Toon, 1994). Similarly, sulfate aerosol size distribution is assumed log-normal with d g = 0.14 µm and σ g = 1.5 (Pueschel et al., 1992). The density of black carbon was 15 assumed equal to 1 g cm −3 while for sulfates, we assumed the density of sulfuric acid (1.84 g cm −3 ). A probability distribution function of cloud scale vertical velocity, w, was used to represent the local variations of velocity at scales relevant for nucleation. The distribution was assumed to be a normal distribution with a fixed standard deviation of 0.25 ms −1 , consistent with observations in the INCA campaign (Kärcher and Ström,

Forcing data
The SCM configuration consists of an isolated column of a global circulation model, and is therefore, a 1-dimensional time-dependent atmospheric model. The lateral forcing fields to the 72 pressure levels in the atmospheric columns of GEOS-5 are pre- 25 scribed from assimilated 4-D observational data. For the purpose of this study, we used the forcing from the TWP-ICE intensive observation period (IOP), derived by the Atmospheric Radiation Measurement (ARM) program. It includes data from 17 January to 12 February 2006. This data set has been previously utilized in forcing SCM simulations with the intent of testing ice microphysics for GCMs (Wang et al., 2009a), as well as for comparing simulations produced with bulk microphysical schemes of varying complexity in a cloud resolving model with observational data (Wang et al., 2009b;Lee 5 and Donner, 2011). The TWP-ICE data is ideally suited for testing the representation of cold and mixed-phase clouds in models and is an often used test case that allows comparison with other existing studies (e.g., Varble et al., 2011;Fridlind et al., 2012). It includes periods dominated by deep convective clouds and by persisting layers of cirrus clouds.  Table 1 summarizes the simulations considered in this study. The objective of the simulation experiments is to evaluate the sensitivity of the cloud fields to the treatment of ice nucleation. A "control" simulation was performed with the LP ice nucleation parameterization as described above, which has been used in the McRAS-AC framework 15 before (Bhattacharjee et al., 2010). Other simulation experiments were carried out with the BN parameterization, utilizing three different IN spectra. Two additional simulations were considered, in which the contribution to N c from contact freezing was neglected (LP-NoFrzc and BN-PDA08-NoFrzc). All the simulations share the same lateral forcing fields, surface fluxes, and aerosol input, and they only differ on the treatment of ice 20 formation. The time-height distributions of the total cloud fraction, CF, exhibits the basic features observed during the TWP-ICE campaign (Fig. 1). In the first period of the intensive observation period (IOP), prior to 25 January 2006, the region was influenced by an active monsoon period characterized by considerable convective activity. From 25 26 January to 2 February, the monsoon was suppressed, and little convective activity was observed, but high clouds persisted through the period. In the final part of the IOP (3-13 February) the region was increasingly impacted by continental storms, reflected in a renewed increase in the convective activity. The simulated cloud fields show some differences in the CF, particularly the simulation with the BN-PDA08, which shows higher frequency of high CF cells. Common to all the simulations with the BN scheme is an increase in the CF for the mixed-phase 5 regime as compared with the LP-CTRL simulation, particularly in the convectively active periods, as shown for two of the simulations in Fig. 3. The resulting simulated q i fields are shown in Fig. 2. Ice mixing ratios generally reach a maxima in the layer extending from the 0 • C to the −38 • C levels. The overall ice mixing ratios encountered in the LP-CTRL simulation are generally higher than for the BN cases, the difference 10 being more pronounced for the mixed-phase regime.
The temperature dependence of N c,nuc and the total ice crystal concentration N c was calculated from the model output for each one of the simulations as a function of temperature (Fig. 4). The impact of the nucleation scheme in the partitioning of condensate was investigated through the ice fraction, f c , defined as (2) Figure 5 shows the temperature dependence of the condensate partitioning, for the BN-PDA08, LP-CTRL, BN-PDA08-NoFrzc, and LP-NoFrzc. Attention was given to variables affecting the radiative properties of the ice clouds. The size of ice particles would be among the most directly affected variables with changes in crystal concentrations. 20 The behavior of effective radius for ice particles as a function of temperature is shown in Fig. 6 for two of the simulations. Figure 7 shows a time series of IWP from different simulation experiments, together with IWP derived from satellite retrievals using the Visible Infrared Shortwave-Infrared Split-Window Techinique (VISST), described in Fridlind et al. (2012).

Discussion of the results
Since the lateral forcing and surface fluxes were prescribed identically in all simulation experiments, any differences in the cloud fields can be attributed to the interaction of the ice nucleation scheme with the BF process and the cloud microphysical response that follows. Some such differences are encountered between the fields produced with 5 LP and BN parameterization, respectively. N c,nuc calculated with BN is systematically lower for the mixed-phase cloud regime irrespective of the heterogeneous nucleation scheme used, however, the difference is greater between LP-CTRL and BN-PDA08, for which the maximum difference in the predicted N c,nuc can be considerable (Fig. 4). The low concentration of IN predicted by the PDA08 spectrum, typically two orders of magnitude lower than produced by the other spectra, explains part of this difference. However, the systematic discrepancy between LP and BN in the mixed-phase regime is likely due to the different implementation of the two nucleation schemes. As described in Sect. 2.1, the LP scheme adds the contributions from immersion freezing (given by the numerical correlations of Liu and Penner, 2005, and from deposition, as given by 15 Eq. 1). In the BN schemes the availability of IN in the mixed-phase regime is dictated by the IN spectrum alone, which consider deposition and condensation freezing. The large differences in predicted N c,nuc are also noticeable in the resulting N c fields, but the magnitude of the difference is significantly lower. In the range of temperatures where contact freezing is active (270.15 K > T > 235 K) this mechanism was found to 20 contribute, on average, between 10 −4 cm −3 and 10 −3 cm −3 to the ice crystal concentration, thereby effectively providing a lower bound for N c (Fig. 4). This contribution is significant only for IN spectrum predicting very low N c,nuc (such as PDA08), or for the temperatures above T ∼ 260 K, in which the other IN spectrum (MY92 and BN-CNT) predicts very small N c,nuc . 25 It is expected that the differences in N c would significantly impact other cloud microphysical variables, particularly through the modification of the rate of the BF process. Lower ice crystal concentrations should result in lower rates of conversion of liquid water to ice because the surface area for vapor-ice mass transfer is low (Rotstayn et al., 2000). Such behavior, in which low aerosol concentrations are associated with low f c , has been observed in satellite retrievals (Choi et al., 2010). However, the ice fraction exhibits little to no change across simulations even for the cases where N c differ by a factor of 100 (Fig. 5a, b). This diminished sensitivity of f c to ice crystal con-5 centration seems to be caused by the action of the contact freezing mechanism. To verify this, two simulations in which this mechanism was neglected were performed LP-NoFrzc and BN-PDA08-NoFrzc, (Fig. 5c, d). LP-NoFrzc shows that the transition from pure liquid to pure ice cloud occurs over a larger temperature interval as compared to simulations in which contact freezing is allowed to occur. However, because 10 LP predicts relatively large crystal concentrations in the entire range of supercooling temperatures, the BF process is always fast, resulting in a rather similar dependence of f c on temperature. This is not the case for the simulations with BN-PDA08, in which the low N c severely limits the rate of conversion of liquid water to ice by water vapor deposition, which is evidenced when contact freezing is turned off (Fig. 5d). The narrow temperature range associated with the transition from f c = 0 at 273 K to f c = 1 at 260 K when contact freezing was included, is consistent with other studies with the same partitioning scheme, as well as with available cloud observations of f c (Rotstayn et al., 2000;Liu et al., 2007). The transition to a fully glaciated state which sometimes produces excessive ice cloud at low temperatures could arise from the assumption that 20 the BF mechanism dominates this regime (Korolev, 2007). The cloud amount in the mixed-phase regime was affected by the crystal ice nucleation parameterization used in the simulations. As shown in Fig. 3, there is an increase in the frequency of occurrence of cloudy cells with CF > 0.5 when the BN-PDA08 parameterization is used instead of LP. This is true for the three IN spectra utilized in this 25 study, with CF being 49 % larger for BN-PDA08, and ∼ 25 % for MY92 and BN-CNT, as compared to simulations with LP.
In the cirrus cloud regime, the difference in N c,nuc between LP and BN is less pronounced than in the mixed-phase regime. For this temperature range, crystal concentrations calculated with LP and BN are within one order of magnitude irrespective of the IN spectrum used, which is consistent with the variability reported in previous studies (Barahona et al., 2010). N c,nuc for LP-CTRL and BN-PDA08 are in close agreement, specially for temperatures above 200 K, however, the predicted mechanism of freezing is different for both parameterizations. Due to the very low IN number pre-5 dicted with PDA08, the contribution of heterogeneous freezing to N c,nuc in BN-PDA08 is negligible, and the process is dominated by homogeneous freezing. The opposite behavior is observed when the LP parameterization is used, in which homogeneous freezing only contributes significantly to the N c,nuc at extremely low temperatures. When BN-CNT or MY92 are used instead, the lower N c,nuc is the result of the depletion of water vapor from the more numerous IN, and homogeneous freezing is triggered only at temperatures between 200 K and 220 K (Fig. 4). Finally, even though the impact of N c on the simulated condensate partitioning is small, the different ice crystal concentration predicted with the parameterizations considered in this study considerably impact the cloud radiative properties. For instance, Figs. 2 and 3 shows the ice mixing ratio for different simulation scenarios. The differences observed translate also into ice water path differences, as well as of the hydrometeor sizes. Figure 6 shows the temperature dependence of the median values of the calculated ice effective radius for BN-PDA08 and LP-CTRL. The inset shows a histogram of the frequency distribution of the effective radius for BN-PDA08 and LP 20 only for the rage of mixed-phase temperatures. Due to the much lower N c predicted by PDA08, the effective radius is shifted from a median of 45 µm for BN-PDA08, to a smaller size with a median of 32 µm in the LP-CTRL simulation.
The changes induced in the cloud microphysics by the different IN spectra consequently modify the overall column integrated properties of the cloud fields. Figure 7 25 illustrate the tendencies in the simulated IWP for the different parameterizations. It is clear that IWP for LP-CTRL is higher than for any simulation with BN, with differences being larger in the periods of convective activity. In fact, the average IWP in the active monsoon period for the LP simulation was found to be 0. 30  Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | of 0.22 kg m −2 for BN-PDA08, and 0.28 kg m −2 for BN-CNT and BN-MY92. In the suppressed period, IWP averaged ∼ 0.04 kg m −2 in all the simulation experiments. This simulation results compare qualitatively well to the available data of IWP as retrieved from VISST. However, the lower bound in the VISST observations tends to be much lower than simulated IWP, while the peaks during the convective events often exhibit 5 higher values than simulated fields. Ice water path from VISST during the active monsoon period averages 0.25 kg m −2 and 0.04 kg m −2 for the suppressed period.

Summary and conclusions
The ice nucleation parameterization of Barahona and Nenes (2009b)  other, but generally ∼ 100 times larger than PDA08 at any given temperature. These simulation experiments were compared to a control simulation using the LP parameterization, which was found to predict the highest ice crystal concentrations across the simulations. It was shown that the different schemes used in this study often predicted IN con-20 centrations differing by up to three orders of magnitude. Despite these important differences in IN availability, ice crystal number concentration for cirrus cloud temperatures predicted in all the simulations were found to agree within a factor of 10. However, the mechanism by which these ice crystal are produced is considerably different; In the regime of mixed-phase clouds, the variations in N c among simulations with the different nucleation schemes was considerably larger than for the ice-only clouds, with the largest variations being within a factor of ∼ 100 in some cases. This larger variability is not surprising, since in the absence of homogeneous freezing, the nucleation schemes strongly depend on the IN nucleation spectra. However, the contribution to 5 N c from contact freezing of cloud droplets with dust particles of ∼ 10 −3 cm −3 provided a lower bound on N c , and was effectively the largest contributor to crystal concentration when the PDA08 scheme was used. This contribution to N c also acted to counteract the very large variations in predicted IN concentrations. Similarly, it was also found that the action of contact freezing efficiently transforming 10 cloud water into cloud ice buffered the impact of the large variations of N c seen across the different simulation experiments on the partitioning of cloud condensate. Ice mixing ratios, however, where strongly affected by the ice nucleation scheme. Accordingly, cloud microphysical variables relevant to radiative properties, such as the effective radius of ice crystals and the ice water path, were impacted by the wide range of N c 15 predicted. It was observed that nucleation schemes that predict lower N c lead to lower in-cloud ice mixing ratios and ice water path, and considerably larger crystal sizes. This study highlights the need for detailed cloud microphysical observations to constrain the large uncertainties associated with the ice nucleation process which limit the ability of GCM models to make accurate estimates of the contribution of cold clouds to 20 the overall aerosol indirect effects. Continued development and refinement of ice nucleation schemes capable of accounting correctly for different freezing mechanisms is needed; using the approaches used here will help accomplish this. Introduction

Tables Figures
Back Close

Full Screen / Esc
Printer-friendly Version