Most global climate models generally prescribe the partitioning of condensed water into cloud droplets and cloud ice in mixed-phase clouds according to a temperature-dependent function, which affects modeled cloud phase, cloud lifetime and radiative properties. This study evaluates a new mixed-phase cloud microphysics parameterization (for ice nucleation and water vapor deposition) against the Atmospheric Radiation Measurement (ARM) Mixed-phase Arctic Cloud Experiment (M-PACE) observations using the NCAR Community Atmospheric Model Version 3 (CAM3) single column model (SCAM). It is shown that SCAM with the new scheme produces a more realistic simulation of the cloud phase structure and the partitioning of condensed water into liquid droplets against observations during the M-PACE than the standard SCAM. A sensitivity test indicates that ice number concentration could play an important role in the simulated mixed-phase cloud microphysics, and therefore needs to be realistically represented in global climate models.
 Mixed-phase clouds are composed of a mixture of liquid droplets and ice crystals. The two-stage (nucleation and growth) process of ice formation in mixed-phase clouds has been known to be important since Bergeron  and Findeisen . The partitioning of condensed water into cloud droplets and cloud ice in most current climate models is specified as a function of temperature, but in nature the partitioning varies with time and cloud properties besides temperature. The liquid/ice partitioning in mixed-phase clouds can significantly impact the cloud optical depth, cloud radiative forcing, and cloud coverage [Fowler and Randall, 1996; Gregory and Morris, 1996].
 Ice nucleation involves a variety of mechanisms, each of which involves aerosol or cloud particles. There are two main classes of mechanisms: homogeneous freezing of supercooled cloud or haze droplets and heterogeneous freezing (including deposition, contact, immersion nucleation processes, etc.). After primary ice has formed in a cloud, the concentration of cloud ice can be increased through secondary multiplication mechanisms. Ice nuclei (IN) play a critical role in converting liquid to ice in mixed-phase clouds, which can have important climate consequences [e.g., Prenni et al., 2007].
 Arctic clouds play a central role in the Arctic climate feedbacks and radiative budget, but are not well represented in current models [Curry et al., 1996]. To provide a physically-consistent treatment of mixed-phase clouds in climate models, ice nucleation parameterizations [e.g., Liu and Penner, 2005] and a water vapor deposition growth scheme [Rotstayn et al., 2000] have both been implemented in the NCAR Community Atmosphere Model (CAM, Version 3) [Liu et al., 2007]. This paper documents the performance of these schemes against ARM Mixed-phase Arctic Cloud Experiment (M-PACE) observations [Verlinde et al., 2007; McFarquhar et al., 2007] using the single column model version of CAM (SCAM).
2. NCAR CAM3
 The NCAR CAM3 is the atmospheric component of the Community Climate System Model (CCSM) [Collins et al., 2006]. The treatment of cloud condensation and microphysics in CAM3 [Boville et al., 2006] is based on Rasch and Kristjánsson  as updated by Zhang et al.  with separate prognostic equations for the liquid and ice-phase condensate. Even though each phase of water is transported separately, after advection, convective detrainment, and sedimentation, the liquid and ice are repartitioned according to a temperature dependent fraction of ice in total water (fi)
where the bounds Tmin and Tmax are adjustable within a narrow range. The values in CAM3 are −40° and −10°C, respectively. In this way the CAM3 neglects the representation of a variety of processes (e.g., the Bergeron-Findeisen process, ice nucleation, etc.) in mixed-phase clouds.
 A two-moment ice-phase microphysics scheme has been introduced to the CAM3 [Liu et al., 2007], in which cloud ice number concentration is predicted. Thus the effective radius for cloud ice which is used in the radiation and gravitational settlement calculations is now calculated from model predicted mass and number of cloud ice rather than diagnosed as a function of temperature. Ice nucleation mechanisms include homogeneous ice nucleation on sulfate aerosol and heterogeneous immersion nucleation on soot particles in ice clouds with temperature less than −35°C [Liu and Penner, 2005], contact freezing of cloud droplets through Brownian coagulation with insoluble IN (assumed to be mineral dust), and deposition/condensation nucleation [Meyers et al., 1992]. The aerosol climatology prescribed in the CAM3 [Collins et al., 2001] is used.
 We added the Rotstayn et al.  scheme in CAM3 to represent the conversion of cloud liquid water to ice in the mixed-phase clouds via the Bergeron-Findeisen process (in which cloud ice grows through vapor deposition at the expense of evaporating cloud droplets). The deposition growth of cloud ice is proportional to its number density. The vapor deposition, combined with a treatment of riming of cloud droplets on snow in CAM3 [Rasch and Kristjánsson, 1998], permits explicit treatment of the Bergeron-Findeisen process, rather than simply diagnosing the condensate phase from temperature.
3. M-PACE Case Study
 The Mixed-Phase Arctic Cloud Experiment (M-PACE) was conducted from late September through October 2004 in the vicinity of the Department of Energy North Slope of Alaska (NSA) field site [Verlinde et al., 2007]. Single-layer boundary-layer mixed-phase clouds were present in the period 9–14 October 2004. The microphysical properties of this cloud system, such as the vertical profiles of cloud liquid water content (LWC), ice water content (IWC), droplet number concentration, ice number concentration, and effective sizes of droplets and ice crystals, were sampled with instruments on the University of North Dakota Citation aircraft [McFarquhar et al., 2007]. The bulk cloud properties sampled by the two flights on October 10 are used to validate model simulations in this study.
 The NCAR SCAM is used in this study to evaluate the mixed-phase cloud parameterizations which we implemented in CAM3 [Liu et al., 2007], as discussed in section 2. Two sets of simulations with the SCAM are performed: one with the standard CAM3 microphysics (standard SCAM) and another with the modified ice microphysics (SCAM-ICE). The period of our simulations is from 17Z October 9 to 8Z October 10. The lower boundary condition is an ocean with an air temperature of 274 K and pressure of 1010 hPa. The initial condition has an inversion with the top pressure at 850 hPa (S. Klein, personal communication, 2007). The large-scale forcing and vertical pressure velocity were derived from an analysis of the ECMWF data for the ocean region adjacent to the NSA [Xie et al., 2006] and are constant in time in the simulations. The surface turbulent heat fluxes are also specified with the ECMWF data (S. Klein, personal communication, 2007).
 Ice formation mechanisms in mixed-phase clouds are very complicated [Fridlind et al., 2007]. The observed ice number is also highly uncertain due to potential contamination by shattering on the probes. Thus a sensitivity test for modified ice microphysics (simulation SCAM-ICE-p2) is performed to examine the impact of cloud ice number density on cloud microphysics. We use in situ out-of-cloud observations of IN number concentrations obtained on October 8 and 10 from the Continuous Flow Diffusion Chamber (CFDC) [Prenni et al., 2007] aboard the Citation aircraft. These measurements represent the total number concentration of active IN that have diameters less than 2 μm acting in deposition, condensation, and immersion-freezing modes. The measured mean concentration of about 0.2 L−1 is used to represent the aforementioned nucleation modes in simulation SCAM-ICE-p2. Contact IN, which are not measured by the CFDC, are represented in the model as described in section 2. IN number concentration (∼0.2 L−1) measured by the CFDC was considerably lower than the averaged in-cloud number concentration of ∼2.0 L−1 for ice crystals with maximum dimension larger than 53 μm observed on 10 October [McFarquhar et al., 2007]. Considering the lower vertical resolution of the SCAM standard configuration, sensitivity tests are conducted in which the vertical resolution is significantly increased from the standard 26 levels to 60 levels for SCAM, SCAM-ICE and SCAM-ICE-p2.
4. Comparison of SCAM Simulations Against M-PACE Observations
 Aircraft observations of single layer mixed-phase clouds present in the period 9–14 October 2004 show that the liquid water content increases with height above cloud base as would be expected for a stratocumulus cloud at the top of a well-mixed boundary layer [McFarquhar et al., 2007]. In general, liquid dominated near the tops of the clouds with precipitating ice (sometimes heavy) near the base, although considerable variability exists between spirals. The LWC increased from cloud base with an average temperature of −12.7 °C to a peak value of 0.3 g m−3 near cloud top with a mean temperature of −16.0 °C for the second flight on October 10. Although the IWC varied throughout cloud, it generally increased with decreasing normalized cloud altitude zn (zn = 1 at cloud top zt and zn = 0 at cloud base zb) with values of about 0.005 g m−3 near zb and larger values of about 0.01 g m−3 beneath zb corresponding to precipitating ice. Neither droplet concentration (Nw) nor ice concentration (Ni) with maximum dimension larger than 53 μm exhibited strong dependence on zn, with Nw having an averaged value of 26 cm−3 and Ni of 2 L−1. Mean LWP values derived from the ARM surface Microwave Radiometer (MWR) using the retrieval algorithm described by Turner et al.  and another algorithm by Wang  are 174.8 and 160.9 g m−2, respectively averaged over the last 6 hours of the simulation (between 9 hour and 15 hour). IWP derived from the ARM cloud radar and lidar measurements [Wang, 2007] is 23.9 g m−2 over the same period.
Figure 1 shows the LWC and IWC as a function of altitudes and time from standard and modified SCAM simulations with the standard 26 vertical levels for the period from 17Z October 9 to 8Z October 10. The standard SCAM produces LWC and IWC profiles overlapping with each other in the mixed-phase clouds. Model calculated maximum LWC and IWC occur in the middle of cloud (∼1100 m) with values of 0.3 and 0.036 g m−3, respectively. LWP and IWP averaged over the last 6 hours of the simulation are 179.2 and 22.8 g m−2, respectively (Table 1). Note that modeled IWC and IWP here and hereafter include snow component since the observations cannot separate snow from ice. The standard SCAM with prescribed partitioning of phase with temperature reproduces the observed high liquid water fraction during the M-PACE by coincidence.
Table 1. Model Simulated Parameters Averaged Over the Last 6 Hours (Between 9 and 15 Hour) Compared with M-PACE Observations
Model calculated peak values in clouds.
Averages and standard deviations over spirals through single-layer mixed-phase clouds by flight 9a on October 10 [McFarquhar et al., 2007].
Averages and standard deviations over spirals through single-layer mixed-phase clouds by flight 9b on October 10 [McFarquhar et al., 2007].
 The SCAM with explicit mixed-phase microphysics (SCAM-ICE) effectively separates the LWC and IWC in clouds with liquid water dominating the water amount in the upper portion of mixed-phase clouds, while the clouds in the bottom portion are purely ice phase with ice precipitating beneath (see Figure 1). Model calculated maximum LWC occurs at ∼1100 m with a value of ∼0.3 g m−3, while the IWC maximum is at the bottom of liquid water with a value of 0.023 g m−3. Thus SCAM-ICE with our physically-based treatment for mixed-phase clouds reproduces the observed cloud phase structure in the M-PACE [McFarquhar et al., 2007]. Modeled Ni is about 3 L−1, which is very close to the averaged in-cloud number concentration of cloud ice (∼2 L−1) with maximum dimension larger than 53 μm [McFarquhar et al., 2007]. We note that the modeled Ni also includes particles smaller than 53 μm. However, cloud ice number was shown to be dominated by particles larger than 53 μm from the explicit bin microphysics model simulation for M-PACE [Fridlind et al., 2007]. Thus we do not expect that changing the cloud ice number concentration due to these smaller particles will significantly impact our analysis here. Of the total cloud ice, ∼0.2 L−1 is formed from the contact nucleation with the rest from the deposition/condensation nucleation. The LWP and IWP averaged over the last 6 hours are 176.0 and 26.5 g m−2, respectively. The standard SCAM and SCAM-ICE predict similar LWP and IWP which are close to measurements (Table 1). As a result surface radiative fluxes are also similar between the two runs (results not shown).
 In a sensitivity simulation with the CFDC measured IN concentration of 0.2 L−1 representing deposition and condensation nuclei (SCAM-ICE-p2), model predicted Ni is much less (∼0.4 L−1) compared to ∼3 L−1 from the SCAM-ICE. This results in a significant reduction in modeled IWC. Here the contact freezing of cloud droplets by mineral dust is treated in the same way as in the SCAM-ICE. The peak LWC occurs at the same altitude (∼1100 m) as the SCAM-ICE, but with a higher value of 0.35 g m−3 due to less liquid converted to ice (Figure 1). IWC is uniformly distributed at the cloud bottom and beneath with values of 0.004–0.006 g m−3, somewhat lower than the observation (0.006–0.015 g m−3) and considerably less than 0.023 g m−3 from the SCAM-ICE simulation. Thus ice number concentration plays an important role in the cloud microphysics through impacting the conversion of liquid mass to ice which needs to be well represented in global climate models. The LWP and IWP averaged over the last 6 hours are 202.1 and 6.6 g m−2, respectively. The SCAM-ICE-p2 performs worse in predicting IWC and IWP than the SCAM-ICE simulation with much higher IN concentrations. Fridlind et al.  suggested that other ice formation mechanisms (e.g., droplet freezing during evaporation) not included in CFDC measurements could be strong enough to account for the M-PACE observations of ice number concentrations [McFarquhar et al., 2007].
 The liquid water fraction (fl) from aircraft observations increased with normalized cloud altitudes zn with fl averaging 0.96 ± 0.13 near zt and 0.70 ± 0.30 near zb [McFarquhar et al., 2007]. Figure 2 shows model derived fl as a function of normalized height in cloud as compiled from SCAM, SCAM-ICE, SCAM-ICE-p2 simulations with 60 vertical levels compared against M-PACE aircraft observations. We use 60 vertical level results here since the standard SCAM with 26 levels (4 vertical levels beneath 850 hPa) has too few cloud points for our statistics analysis. Cloud liquid water content (10−3 g m−3) is used to determine the cloud base/top. The standard SCAM using a prescribed fl increasing with temperature produces an opposite trend of fl with altitudes, while the SCAM with explicit mixed-phase cloud microphysics (SCAM-ICE and SCAM-ICE-p2) reasonably captures the observed trends in the vertical variability of fl. The SCAM-ICE produces too little liquid and thus too high an ice mass fraction compared to the observations, while SCAM-ICE-p2 with the lower ice number density has a better simulation of liquid fraction. Figure 3 shows fl in clouds as a function of temperature. The standard SCAM gives a liquid fraction decreasing with lower temperature, an opposite trend compared with the observation (another branch of fl decreasing with higher temperature is caused from snow around cloud base). The modified SCAM-ICE and SCAM-ICE-p2 reasonably capture the variation of liquid water fraction with temperature compared with observations.
 Successful modeling of Arctic mixed-phase clouds and their microphysics structure has been a challenge [e.g., Morrison and Pinto, 2005; Prenni et al., 2007]. The partitioning of condensed water into cloud droplets and cloud ice in most current climate models is based on parameterization schemes [e.g., Bower et al., 1996; Boudala et al., 2004] derived from past measurements in mixed-phase clouds combined from many geographical locations where liquid water fraction fl increases with temperature. Our results reinforce the conclusion that these conventional parameterization schemes are not applicable to the M-PACE observations of Arctic clouds which showed the opposite trend [McFarquhar et al., 2007]. In order to realistically simulate the vertical variability of fl in mixed-phase clouds for large-scale models, explicit treatment of ice nucleation and liquid conversion to ice via Bergeron-Findeisen process should be considered. In this study the SCAM with our physically-based mixed-phase scheme reasonably reproduces the cloud structure and partitioning of total water as observed in the M-PACE. Ice number concentration from ice nucleation is found to be important to the simulated cloud phase partition through impacting the conversion of liquid to ice and thus needs to be well represented in the climate models. As reported by Xie et al. , the improved scheme also shows promising features in the simulated cloud microphysical properties for M-PACE when it is tested under the U. S. Department of Energy (DOE) CCPP-ARM Parameterization Testbed (CAPT) framework, which is used to run climate models in numerical weather forecast mode. The global radiative forcing and climate feedbacks of these treatments are also under investigation.
 The authors would like to acknowledge the support from the Department of Energy (DOE) Environmental Science Division Atmospheric Radiation Measurement (ARM) program. The Pacific Northwest National Laboratory is operated for the DOE by Battelle Memorial Institute under contract DE-AC06-76RLO 1830. This work at LLNL was performed under the auspices of the U. S. Department of Energy by the University of California, Lawrence Livermore National Laboratory, under contract W-7405-Eng-48.