Effects of atmospheric dynamics and aerosols on the thermodynamic phase of cold clouds

Effects of atmospheric dynamics and aerosols on the thermodynamic phase of cold clouds Jiming Li, Qiaoyi Lv, Min Zhang, Tianhe Wang, Kazuaki Kawamoto and Siyu Chen Key Laboratory for Semi-Arid Climate Change of the Ministry of Education, College 5 of Atmospheric Sciences, Lanzhou University, Lanzhou, China Graduate School of Fisheries Science and Environmental Studies, Nagasaki University, Nagasaki, Japan


Introduction
Clouds play an important role in regulating the Earth's radiation budget and 85 global hydrological cycle (Stephens, 2005).However, because observations are lacking and understanding of the physical processes involved in cloud formation is insufficient, clouds are also regarded as the greatest uncertainties in climate change predictions made by various climate models (Williams et al., 2003;Zhang et al., 2005;Klein et al., 2013).One of the primary challenges in better understanding the role of 90 clouds in climate forcings and feedbacks involves determining how to more accurately define the cold cloud phase (cloud top temperature<0ºC) composition between 0ºC and -40ºC, with unsophisticated cloud phase schemes in GCMs (general circulation models; Li and Le Treut, 1992;Morrison et al., 2003;Tao, 2003;Tsushima et al., 2006).Currently, many models specify the fraction of liquid-phase clouds 95 solely as a function of temperature (Doutriaux-Boucher and Quaas, 2004;Storelvmo et al., 2008;Song et al., 2012), related ice heterogeneous nucleation processes are not considered in some models because of the poor understanding of aerosol particles' ice nucleation ability, coating conditions and nucleation modes (e.g., deposition, immersion freezing, contact or condensation freezing) (Lohmann and Feichter, 2005).

100
In view of the entirely different radiative and microphysical properties of ice and liquid particles, changes in the liquid-ice phase transition will significantly affect the Earth's radiation budget and precipitation efficiency (Fu et al., 1999;Fu, 2007;Sassen and Khvorostyanov, 2007;Sun et al., 2004Sun et al., , 2015)).Thus, the oversimplification of cloud phases in climate models inevitably leads to large biases in the study of various 105 climate feedbacks and the sensitivity of these models.
The Clausius-Clapeyron theory and laboratory results have indicated that liquid water particles can exist at a temperature threshold as low as -38ºC to -40ºC before homogeneous nucleation occurs (Roger and Yau, 1989).Studies based on Lidar data and satellite observations have further verified the existence of liquid water at 110 temperatures as low as -30ºC to -40ºC (e.g., Intrieri et al., 2002;Naud et al., 2006;Shupe et al., 2006;Morrison et al., 2011).For example, using un-polarized, ground-based Lidar data from Chilbolton in Southern England, Hogan et al. (2003) Atmos.Chem. Phys. Discuss., doi:10.5194/acp-2016-147, 2016 Manuscript under review for journal Atmos.Chem.Phys.Published: 25 February 2016 c Author(s) 2016.CC-BY 3.0 License.have found that 27% of clouds between -5ºC and -10ºC in Chilbolton contain a supercooled liquid-water layer, and this percentage falls steadily with temperature and 115 reaches approximately zero at temperatures below -35ºC.Giraud et al. (2001) have used the Along-Track Scanning Radiometer (ATSR)-2 infrared data from the ERS-2 satellite to analyze the relationship between cloud phase and cloud top temperature.
Their results have indicated that the probability of ice phase clouds decreases quasi-linearly with cloud top temperature from nearly 100% at around -33ºC to close 120 to 0% at -10ºC.By using polarimetric satellite data, Doutriaux-Boucher and Quaas (2004) have also derived a global lower limit of -32°C for 100% of ice phase clouds.
However, the lowest temperature thresholds at which liquid water particles can exist within various climate models vary dramatically from -15ºC (Smith,1990;Doutriaux-Boucher and Quaas, 2004) to -23ºC (Weidle and Wernli, 2008) to -40ºC 125 (Del Genio et al., 1996;Collins et al., 2004).In addition, the relationship between the supercooled liquid cloud fraction (SCF) and cloud top temperature (CTT) in some models and reanalysis datasets is fixed with an exponent of 1.7 (Doutriaux-Boucher and Quaas, 2004) or 2 (Smith, 1990;Weidle and Wernli, 2008).The unique temperature thesholds and relations for the entire globe, regardless of their geographic 130 or temporal variations, eventually lead to the SCF at a given cloud top temperature express considerable differences among GCMs.For example, the liquid water cloud fraction at -15ºC varies from 12% to 83% in six single column models (SCMs) used in a model comparison study of Arctic mixed-phase clouds (Klein et al., 2009;Morrison et al., 2009).The geographic and temporal variations of SCF at a given 135 temperature are further complicated by several factors, including ice nuclei (IN) concentrations and dynamic conditions such as vertical motion (Naud et al., 2006;Choi et al., 2010;Zhang et al., 2015).Combined satellite observations and reanalysis datasets have the potential to yield global cloud phase statistics and to clarify the relationship between cloud phase and microphysical/dynamic processes.This 140 information would aid in the design and evaluation of more physically based cloud phase partitioning schemes, improve calculations of clouds' radiative effects, and reduce uncertainties in cloud feedbacks within GCMs.
The millimeter-wavelength cloud-profiling radar (CPR) on CloudSat (Stephens et al., 2002) and the cloud-aerosol Lidar with orthogonal polarization (CALIOP) 145 (Winker et al., 2007) on CALIPSO (launched in late April 2006) can provide more accurate data related to the vertical structure of clouds, along with cloud phase information on a global scale (Hu et al., 2010;Li et al., 2010Li et al., , 2015)).The depolarization ratio and layer-integrated backscatter intensity measurements from CALIOP can help distinguish cloud phases (Hu et al., 2007(Hu et al., , 2009)).Using combined

Meteorological reanalysis dataset
In this study, the temperature profiles and surface temperatures (that is, skin temperature) used in our analysis are taken from the ECMWF-AUX product (Partain, 2004), which is an intermediate product that contains the set of ancillary ECMWF state variable data interpolated to each CloudSat cloud profiling radar (CPR) bin.In 190 addition to this information, the collocated vertical velocity parameter from the ERA-Interim daily dataset (Dee et al., 2011) is also extracted and used in our analysis.
Here, the temperature profile is used to identify supercooled water clouds from all water clouds, determine the aerosol and cloud layer top temperatures, and calculate the lower-tropospheric static stability (LTSS), which is defined as the difference in 195 potential temperature between 700 hPa and the surface (Klein and Hartmann, 1993),

or
, where p is pressure, T is temperature, R is the gas constant of air, and C p is the specific heat capacity at a constant pressure.A high LTSS value represents a stable atmosphere, whereas a low LTSS value represents an unstable atmosphere.In Section 3.4, we will discuss the effects of 200 vertical velocity, LTSS and skin temperature on cloud phase in detail.

Cloud phase product
/ / 700 700 1000 1000 CALIOP and the temperature profile from the ECMWF-AUX product to identify three different cloud phases (ice, mixed and liquid).However, the Lidar-only phase algorithm only distinguishes the water and ice phases of a cloud by using the Lidar depolarization ratio and layer integrated attenuated backscattering coefficient (IBC) (Hu et al., 2007(Hu et al., , 2009)).Due to the strong multiple scatter effect in the Lidar 215 depolarization measurements, as well as Lidar's limited ability to penetrate optically thick clouds, CALIPSO's Lidar-only algorithm is restricted in its ability to identify mixed-phase clouds; in particular, it is unable to penetrate the supercooled liquid layer to detect the ice layer (Zhang et al., 2010), and it is unable to distinguish pure liquid clouds from mixed-phase clouds.Nevertheless, only cloud top information is needed 220 in this study.Therefore, the differences between these two algorithms should not result in abrupt or obvious changes in cloud phase fractions.Given the importance of multilayered cloud systems (Huang et al., 2005(Huang et al., , 2006a;;Lv et al., 2015), we obtain the cloud phase information of every cloud layer in each sample profile and further group every cloud layer into separate temperature bins (1ºC interval) according to the those supercooled water phase clouds with a CTT between -40ºC and 0º C are further analyzed in this study.Here, we define the supercooled water cloud fraction (SCF) in a given temperature bin as the ratio of the number of liquid phase samples and the 235 total (liquid+ice) samples gathered in a 2 º ×6 º grid box.

Aerosol types and relative frequency
Aerosol data are obtained from the CALIPSO level 2, 5 km aerosol layer product.
Using scene classification algorithms (SCA), CALIPSO first classifies the atmospheric feature layer as either a cloud or aerosol by using the mean attenuated 240 backscatter coefficients at 532/1064 nm, along with the color ratio (Liu et al., 2009).
A confidence level for each feature layer is also reported in the level 2 products.
Using the surface type, lidar depolarization ratio, integrated attenuated backscattering coefficient and layer elevation, aerosols are further distinguished as desert dust, smoke, polluted dust, clean continental aerosol, polluted continental aerosol, and 245 marine aerosol (Omar et al., 2009).Mielonen et al. (2009) have used a series of Sun Photometers from the Aerosol Robotic Network (AERONET) to compare CALIOP and AERONET aerosol types and have found that 70% of the aerosol types from these two datasets are similar, with the closest similarities occurring in dust and polluted dust types.Mineral dust from arid regions has been widely recognized as an important 250 source of ice nuclei in mixed-phase clouds because of its nucleation efficiency and abundance in the atmosphere (Richardson et al., 2007;DeMott et al., 2010;Atkinson et al., 2013).In addition to dust, some studies have also verified the potential ice nucleation ability of polluted dust and smoke at cold temperatures (Niedermeier et al., 2011;Cziczo et al., 2013;Zhang et al., 2015).For example, by using satellite lidar  (Choi et al., 2010).We remove those aerosol layers with low confidence values (that is, those with an absolute value lower than 50) from the dataset (approximately 6.5% 275 of all aerosol layers).

Cloud phase partitioning schemes in GCMs
Recently, several ice nucleation processes based on theoretical and empirical studies have been developed to more explicitly represent these processes in certain 280 climate models.These new schemes have indicated that the liquid cloud fraction should depend not only on temperature but also on the presence of aerosols that have undergone ice nucleation (Storelvmo et al., 2008;Gettelman et al., 2012).However, (1) For scheme 2, the liquid fraction f can be expressed as: (2 where T is temperature, , T w is the temperature above which all clouds are liquid, T ice is the temperature below which all clouds are ice, and n is a shape 295 parameter that controls the slope of f(T) between -40º and 0º.Based on table 1 of Choi et al. ( 2014), we select several models and list the values of these parameters in Table 1.Obviously different thresholds of these parameters and different cloud phase schemes in climate models indicate that an inability for models to accurately separate the cloud phases, and the large biases and inconsistency between the models may be

Global distributions of T w , T ice and n
Based on the processes outlined in Section 2.2, the SCF for each temperature bin (1K) of every grid can be derived.The value of T w in each grid equals the temperature above which all SCFs equal 1, whereas the value of T ice in each grid equals the temperature below which all SCFs equal 0. After obtaining values for T ice and T w , the 310 value of n can be further determined by performing nonlinear fitting to T ice , T w , f and T using Eq.( 1).In the following analysis, we determine that scheme 1 (Eq.( 1)) better simulates the variation of SCF with temperature than scheme 2 (Eq.( 2)), and thus only the distributions of n for scheme 1 are provided in this section.indicate missing data or areas where the supercooled water cloud fraction doesn't reach 1 between -40 ºC and 0 ºC.Those grids in which T w is higher than 0 ºC are Atmos. Chem. Phys. Discuss., doi:10.5194/acp-2016-147, 2016 Manuscript under review for journal Atmos.Chem.Phys.Published: 25 February 2016 c Author(s) 2016.CC-BY 3.0 License.
excluded in this study.These grids almost all located within typical subsidence regions (e.g., stratocumulus regions) where strong subsidence favors low cloud 320 formation and suppresses ice or mixed-phase cloud generation (Yuan and Oreopoulos, 2013).Fig. 1 clearly illustrates that the value of T w ranges from -2ºC to -6ºC across the majority of the globe; moreover, no clear seasonal variations are found in our results for T w .At high latitudes, T w ranges from -2ºC to -3ºC; this value decreases from the high latitudes to the tropics.Our analysis indicates that the current T w value 325 used in CAM3 may be too low and may result in the overestimation of supercooled water clouds at lower altitudes (see Fig. 7), whereas the T w value used in CAM5 is consistent with the distribution of T w for most regions across the globe.For the ERA40 reanalysis dataset (Weidle and Wernli, 2008) and other models such as the LMDZ (Doutriaux-Boucher and Quaas, 2004), a high threshold is generally adopted, cloud top temperature T, T ice , and T w .This relationship is showed in Fig. 3, which depicts the apparent decreases of grid-mean SCF from 0.6 to 0.3 with corresponding 350 increases in parameter n from 0.5 to 5.5.Here, the grid-mean SCF is the averaged value of the supercooled water cloud fraction from all cloud top temperature bins ranging from -40ºC to 0ºC within a given grid cell, and the color bar represents the number of grid cells within a 4 year period.Further, Fig. 4 shows the clear geographic and seasonal variations of parameter n.Based on Fig. 4, we find that n is 355 approximately 1 at 60º poleward and varies strongly from 0 to 5 throughout a majority of the globe.Larger values (equal to or greater than 3) locate at the mid-latitudes of the northern hemisphere, South America and the mid-latitude oceans of the southern hemisphere.Especially at the mid-latitudes of the northern hemisphere, the value reaches even 4 or 5. Given the values of n used in the CAM 3 (Collins et al., 2004) 360 and CAM 5 (Song et al., 2012), it is clear that the CAM 3 and CAM5 better simulate the relationship between temperature and SCF at the high-latitudes (60º poleward); meanwhile, the values of n adopted in the ERA40 (Weidle and Wernli, 2008) and LMDZ (modified version) (Doutriaux-Boucher and Quaas, 2004) are consistent with observed results in only some regions during certain seasons (e.g., the Pacific Ocean 365 during all seasons except summer, and the high-latitudes of the northern hemisphere during summer and Russia year-round).For other regions such as Asia, South America and the North Pacific, large n values in these models definitively indicate the models' inability to accurately simulate the relationship between temperature and SCF locally.

Evaluation of cloud phase partitioning schemes
Following the process used to evaluate scheme 1, we also derive the parameter n for scheme 2 (not shown).After inputting the dynamic thresholds of T ice , T w and n into Eq.(1) and Eq. ( 2), we are able to calculate the SCF of each cloud top temperature bin within each geographic grid for the different schemes, and we further evaluate which 375 scheme is better able to simulate the variation of SCF with temperature in each grid.
Here, we define the grid mean absolute value of the difference between calculated and observed SCF (absolute difference) at each temperature bin as follows: (3) Following the same logic, the grid mean relative difference can be written as: where T is the cloud top temperature, and and are the observed and calculated SCFs from scheme 1 and 2, which are determined by inputting the dynamic (or fixed) thresholds of T ice , T w and n into Eq.( 1) and Eq.( 2), respectively.In addition, 41 is the number of cloud top temperature bins from -40ºC to 0ºC.Figs. 5 385 and 6 compare the geographic distributions of absolute and relative differences (annual means) for different schemes, respectively.Compared with scheme 2, which is used in the GISS model (see Fig. 5b), the cloud phase partitioning scheme (scheme 1) is used in CAM 3 (or 5) better simulate the variation of SCF with temperature almost everywhere, especially at the mid-and high-latitudes (see Fig. 5a).For 390 example, the absolute difference for scheme 1 (Fig. 5a) is smaller than 0.08 at 40º poleward, with a large value (0.12) only apparent in the oceanic regions of the subtropics.However, in scheme 2 (Fig. 5b), the differences across most regions of globe still exceed 0.16, even when dynamic thresholds are inputted.Figs.5c and 5d further illustrate the absolute difference between CAM3 (and 5) calculated SCFs and 395 observed SCFs.At present, CAM 3 and CAM 5 still rely on unique temperature thresholds and the n value identified in scheme 1 for the entire globe, which has led to considerable variations in absolute difference values compared with those shown in Fig. 5a.By comparing Figs.5b and 5d, we find that the distributions and magnitudes of the absolute differences are very similar, and even the high-latitude values from 400 CAM5 are smaller than those results derived from scheme 2. These figures further verify the importance of the cloud phase partitioning scheme in general circulation models.Although the schemes used in CAM 3 and 5 are similar, the difference is more apparent in CAM 3 than CAM 5 (see Fig. 5c).In fact, based on Eq.( 1) and Table 1, it is clear that the difference between Fig. 5c and 5d is mainly caused by the  and 6d), whereas SCF is overestimated significantly (14%) in other regions by the 415 models, especially CAM 3 (Fig. 6c).However, the relative difference for scheme 1, which consider the dynamic thresholds of T ice , T w and n, ranges from only -0.04 to 0.02 (Fig. 6a).
For studying the vertical distribution of zonal mean SCF with temperature, and further evaluating the absolute and relative differences between calculated and 420 observed SCFs for scheme 1 are mainly from which temperature bin, the Fig. 7 give us some new insights.For example, Fig. 7a  is more apparent than those results for the same range derived from CAM3.Fig. 6 allows us to infer that the clear bias in relative differences poleward of 60°S produced by CAM5 is primarily caused by the underestimation of SCFs in the -20ºC to -25ºC temperature range, whereas the apparent overestimation of SCF from -10ºC to -20ºC contributes most of the bias at other latitude belts. 455

Effects of atmospheric dynamics and aerosols on cloud thermodynamic phase
The above analysis demonstrate the inability of current models to accurately simulate the vertical and geographic variations of the supercooled cloud fraction, due to incomplete knowledge of the underlying physical processes related to cloud phases.
In fact, several factors in addition to cloud top temperature, including IN and vertical 460 motion, probably together affect the distribution of the supercooled cloud fraction, especially the relationship between temperature and supercooled water clouds.For example, the distribution of parameter n at the mid-latitudes of the northern hemisphere may be largely related to dust aerosols that can serve as IN and thus enhance the glaciation occurrences at lower relative SCFs than in other regions of the globe at a given cloud top temperature, e.g., -20ºC (Fig. 8).For example, as showed in Fig. 8, the SCF at -20ºC is only approximately 0.12 (especially at spring) at the mid-latitudes of the northern hemisphere, whereas SCF reaches 0.7 at the high-latitudes of the southern hemisphere (that is, poleward of 60°S).Our results are consistent with those from previous studies from Choi et al ( 2010) and Tan et al 470 (2014), which have verified that the regional differences in the supercooled water cloud fraction at -20ºC are highly correlated with the dust frequency above the freezing level.However, by collocating this variable with the geographic and seasonal distributions of relative aerosol occurrence frequency (RAOF) at -20ºC (see Fig. 9), we find that the SCF still has a low value at the mid-latitudes of the northern crystals (e.g., Pruppacher and Klett, 1978).The value of T 50 tends to increase when of SCF at -20ºC as the RAOF and 500 hPa vertical velocity increase.The logarithmic intervals of RAOF (see the X-axis of Fig. 12c) indicate that there is a semi-logarithmic relationship between SCF and RAOF (Choi et al., 2010).For relatively clean air without IN aerosols, the SCF at a cloud top temperature of -20 ºC exceeds 40% when the 500 hPa vertical velocity is smaller than 25 hPa/day.This 580 value gradually decreases to 15% under high aerosol loading conditions when the 500 hPa velocity is also high (>50 hPa/day).The clearly decreasing trend in SCF with increasing RAOF is concordant with previous studies' conclusions based on CALIPSO measurements (Choi et al., 2010;Tan et al., 2014).These distinctly T 50 values at different surface temperatures reaches 10K without aerosol loading and gradually decreases to approximately 3K with high aerosol loading.On average, the different surface temperatures lead to 25% SCF differences, and the bias is larger than the difference caused by different vertical motions (10%).This is also comparable to the effect of aerosols on cloud phase changes when the vertical velocity is limited to 605 the same level.High LTSS represents a stable atmosphere, whereas low LTSS represents an unstable atmosphere.In Fig. 14, a stable atmosphere (LTSS>=19K) can be seen to inhibit ice nucleation and enhance the SCF, and it is associated with colder T 50 values than an unstable atmosphere (LTSS<=14K).Although the effects of LTSS on these parameters are not as apparent as those of surface temperature and vertical 610 motion, some interesting results still are captured.Naud et al. (2006) have found that outside the frontal ascent zone, T 50 is not uniformly warm everywhere the mean strength of the 500 hPa vertical velocity is high.They have suggested that vigorous updrafts either suppress ice formation or advect supercooled water to the colder cloud to glaciate as ice crystals (Bower et al., 1996).In fact, these two opposite mechanisms may correspond to storms of different intensities, different cloud systems within different atmospheric stability levels (convective cloud or stratiform frontal cloud) or different locations with the cloud (cloud top or inside the cloud) (Bower et al., 1996;Naud et al., 2006).In our results, we find that Naud's results outside the frontal ascent 620 zone may be interpreted using the LTSS and surface temperature.For the same vertical motion, high LTSS tends to reduce the T 50 to a cold temperature.In addition, stratiform clouds can be generated more easily within a stable atmosphere; thus, we further find that the effect of LTSS on cloud phase is obvious at middle and high 630 latitudes, particularly in the northern hemisphere, where shallow stratiform clouds such as altostratus, stratus and nimbostratus clouds are frequent (Wang and Sassen, 2001;Sassen and Wang, 2008;Li et al., 2015).Due to the wide distribution of land at the middle and high latitudes of the northern hemisphere, seasonal variations in surface temperature result in significant differences in LTSS during different seasons, In the cloud phase partitioning schemes of models, we primarily focus on the effects of aerosols and dynamic factors on the parameter n.However, we also perform a similar analysis for T w with n, and we find that the value of T w systematically

Conclusions and Discussion
Changes in cloud phase can significantly affect the Earth's radiation budget and global hydrological cycle (Sassen and Khvorostyanov, 2007;Choi et al., 2010).Based

Figure Captions
150 CALIOP/IIR/MODIS measurements, Hu et al. (2010) have compiled global statistics on the occurrence, liquid water content and fraction of supercooled liquid clouds; and they have further developed a new cloud thermodynamic phase parameterization.Cheng et al. (2012) have examined the effect of this new cloud phase parameterization within a climate simulation by replacing the default parameterization 155 in the CAM4 with this new one.In addition, Choi et al. (2010) and Tan et al. (2014) have utilized the vertically resolved observations of clouds and aerosols from CALIPSO to analyze cold cloud phase changes and possible aerosol impacts at given temperatures.However, systematic studies of the statistical relationship between cloud phase and IN aerosol properties under different dynamic conditions on a global 160 scale have received far less attention.In this study, we combine cloud phase information from CloudSat and CALIPSO, aerosol data from CALIPSO, and dynamic parameters from the ECMWF-AUX and ERA-interim reanalysis datasets to investigate the geographic and seasonal variations of different parameters' thresholds used in the cloud phase partitioning schemes of climate models.We also perform a 165 preliminary evaluation of how well different cloud phase partitioning schemes can characterize the variation of the SCF at cloud top temperatures from -40ºC to 0ºC; and we further evaluate and discuss the effects of atmospheric dynamics and aerosols on cloud phase at a given temperature.This paper is organized as follows: a brief introduction of all datasets used in this 170 study is given in Section 2. Section 3 outlines the global distributions of several important cloud phase parameters used in the models, evaluates the performance of different cloud phase partitioning schemes and discusses the effects of atmospheric Atmos.Chem.Phys.Discuss., doi:10.5194/acp-2016-147,2016 Manuscript under review for journal Atmos.Chem.Phys.Published: 25 February 2016 c Author(s) 2016.CC-BY 3.0 License.dynamics and aerosols on a cloud's thermodynamic phase.Important conclusions and discussion are presented in Section 4.

175 2 .
Datasets and methodsIn the following study, 4 years(2007)(2008)(2009)(2010) of data from the latest release of the CloudSat 2B-CLDCLASS-LIDAR (version 1.0) product (e.g., radar-LiDAR cloud classification), the ECMWF-AUX product and the CALIPSO level 2, 5 km aerosol layer product are collected to analyze the effects of atmospheric dynamics and 180 aerosols on the thermodynamic phase of cold clouds on a global scale.To analyze the regional variability of the studied parameters, we divide the globe into 2 º ×6 º grid boxes and collect a valid sample set from each grid box.Only those results and findings derived from daytime data are provided in this study in order to support the analysis of the radiative effects of different cloud phases in parallel studies. 185 . Phys.Discuss., doi:10.5194/acp-2016-147,2016 Manuscript under review for journal Atmos.Chem.Phys.Published: 25 February 2016 c Author(s) 2016.CC-BY 3.0 License.Naud et al. (2006) have indicated that cloud-radiation interactions are most sensitive to various parameters near the cloud top.Thus, we focus on the cloud top phase (CTP) and temperature (CTT) information in this analysis.Cloud phase 205 information is derived from the CloudSat 2B-CLDCLASS-LIDAR (version 1.0) product.Compared with the CALIPSO phase identification (lidar-only alogorithm), the 2B-CLDCLASS-LIDAR product utilizes cloud boundaries retrieved from combined CPR and CALIOP measurements, the cloud layer maximum Ze identified with CPR, the layer integrated attenuated backscattering coefficient (IBC) from 210

225ECMWF-
AUX temperature profiles and 2B-CLDCLASS-LIDAR cloud layer top heights.For mixed-phase clouds, we define the cloud as ice-topped or liquid-topped based on the "water_layer_top" information.If the temperature of the water layer in a mixed-phase cloud is equal to or lower than the mixed-phase cloud top temperature, it is classified as a liquid-topped cloud.Otherwise, it is classified as ice-topped.230 Furthermore, liquid phase clouds are divided into warm water-phase (CTT≥0) and supercooled water-phase (CTT<0) according to their cloud top temperature.Only Atmos.Chem.Phys.Discuss., doi:10.5194/acp-2016-147,2016 Manuscript under review for journal Atmos.Chem.Phys.Published: 25 February 2016 c Author(s) 2016.CC-BY 3.0 License.

255observations,
Tan et al (2014) have found negative temporal and spatial correlations between the supercooled liquid cloud fraction and the polluted dust and smoke aerosol frequencies at the −10ºC, −15ºC, −20ºC, and −25ºC isotherms, although those correlations are weaker than those found between dust frequencies and the supercooled liquid cloud fraction.As a result, we combine the dust, polluted dust and 260 smoke information from CALIPSO to further analyze the relationship between aerosols and the SCF in this study.Given the difficulty of quantifying the Atmos.Chem.Phys.Discuss., doi:10.5194/acp-2016-147,2016 Manuscript under review for journal Atmos.Chem.Phys.Published: 25 February 2016 c Author(s) 2016.CC-BY 3.0 License.concentration of IN aerosols (here, IN aerosols are the sum of dust, polluted dust and smoke), this study utilizes the relative occurrence frequency of IN aerosols to quantify this variable instead.We first group every IN aerosol sample from each observation 265 profile into a different temperature bin (1ºC interval) according to the ECMWF-AUX temperature profiles and CALIPSO aerosol layer top height measurements.Then, following Choi et al. (2010), we define the frequency of IN aerosols within a given temperature bin as the ratio of the number of IN aerosol samples to the total number of observation profiles in the same temperature bin and grid.Finally, we calculate the 270 relative occurrence frequency of IN aerosols with respect to the highest IN aerosol frequency.The relative occurrence frequency of aerosols is indicative of the temporal and spatial variability of IN aerosols compared to the maximum occurrence frequency schemes that are widely used in present models.Scheme 1 can be written as: 290

300
because these thresholds are based on aircraft observations or field experiments from different regions.Thus, a constant global threshold would probably introduce large uncertainty in the simulation of cloud feedbacks, and the spatial and temporal variations of these parameters should be considered in the future parameterization of cloud phase partitioning. 305

Fig. 1
Fig.1shows the geographic and seasonal variations of T w values across the 2 º ×6 º 330 which probably results in the underestimation of supercooled water clouds at lower altitudes.Fig. 2 illustrates the geographic and seasonal variations of T ice at the 2 º ×6 º grid box scale.The warmest T ice values for each season generally occurs in typical stratocumulus regions and in northern Africa; these values are warmer than -26ºC, indicating that supercooled water clouds in these regions are restricted to warmer 335 atmospheric levels than are found throughout the majority of the world, where T ice values are almost always below -30ºC.Although the differences between these observations and the models are clear, the thresholds used in the CAM3 (or 5) cloud phase scheme and GISS are relatively reasonable compared with those values used in the ERA40 reanalysis dataset (Weidle and Wernli, 2008) and other models.Generally, 340 T ice values can reach lower temperatures in a clean or IN-poor environment.However, our results show that the geographic and seasonal variations in T ice are also negligible under different aerosol loading conditions, indicating that the combination of several factors, such as IN, vertical motion or other dynamic parameters affect the distribution of T ice .345 In each model's cloud phase scheme, the shape parameter n controls the slope of the curve between temperature and the supercooled water cloud fraction.For example, in scheme 1, a large n value corresponds to a low liquid cloud fraction at a given Atmos.Chem.Phys.Discuss., doi:10.5194/acp-2016-147,2016 Manuscript under review for journal Atmos.Chem.Phys.Published: 25 February 2016 c Author(s) 2016.CC-BY 3.0 License. 370 Atmos.Chem.Phys.Discuss., doi:10.5194/acp-2016-147,2016   Manuscript under review for journal Atmos.Chem.Phys.Published: 25 February 2016 c Author(s) 2016.CC-BY 3.0 License.
.Phys.Discuss., doi:10.5194/acp-2016-147,2016   Manuscript under review for journal Atmos.Chem.Phys.Published: 25 February 2016 c Author(s) 2016.CC-BY 3.0 License.unreasonablethreshold of T ice in the CAM3, which yields an additional 17% difference in values compared to the CAM5.However, the limits of CAM 5 are still apparent in the northern hemisphere, as seen by comparison of Figs.5a and 5d, especially at mid-latitudes where the effect of IN aerosols is important (the difference can reach 0.2).Thus, scheme 1's ability to consider the dynamic thresholds of T ice , T w 410 and n may further improve the prediction of the supercooled water cloud fraction at different temperatures, particularly poleward of 40°, thus making it a preferred option in general circulation models.For relative difference values (Fig.6), a clear underestimation (exceeding -6%) occurs poleward of 60°S over the ocean(Figs.6b depicts the vertical distribution of zonal mean SCFs (annual mean) with temperature based on the observation.Here, the SCF at each temperature bin of each latitude belt is the averaged value of SCFs of all grids at this temperature in this latitude belt.Fig.7afurther illustrates that nearly all of the 425 SCF values are close to 1 or 0 at temperatures above -5ºC and below -35ºC, respectively.Thus, the T ice and T w temperature thresholds used in CAM5 are probably more reasonable values than those in CAM3, at least at the zonal mean level.In addition, the meridional asymmetry (or hemispheric asymmetry) of the SCF values is apparent, with southern hemispheric values exceeding those in the northern 430 hemisphere.This clear difference is particularly prominent at the temperature range from -15ºC to -30ºC poleward of 50°S, where the difference between SCFs in the southern and northern hemispheres sometimes exceeds 20% and reaches 40% at -25ºC in the polar region (figure not shown).Given the similar vertical distributions of the SCF values, Fig.7bprovides a typical example of the SCF profile differences between the dynamic thresholds derived by scheme 1 and the observed values.Overall, the difference is small in most temperature bins (with bias ranging from -9% to 9%) and is primarily concentrated in the southern hemisphere.A clear underestimation (or overestimation) exceeding 12% is produced by scheme 1 for the temperature range -20ºC to -25ºC (or about -10ºC) at 40°S to 40°N, and especially 440 approximately 40°S; however, CAM3 generally produces a substantial overestimation almost everywhere except for the temperature range -20ºC to -25ºC poleward of 60°S, where a slight underestimation occurs (from 9% to18%).The significant overestimation of SCF by CAM3 is particularly prominent in the lower atmosphere at temperatures between -10ºC to -20ºC for almost every latitude belt (where the bias 445 reaches up to 45%).For those regions where temperatures drops below -20ºC, the SCF bias decreases to approximately 30%; nevertheless, SCF values are still significantly overestimated.Compared with CAM3, the bias is smaller in CAM5 (Fig.7d), although the patterns of differences are very similar.At the temperature range -20ºC to -25ºC poleward of 60°S, the underestimation of values (approximately 30%) 450

475
hemisphere during the summer season, even though the IN aerosol loading is insignificant at -20ºC in these regions.As in the findings ofChoi et al (2010), our results also show that the persistent low SCF (or large n) throughout the year at −20ºC (see Fig.8) could not be explicitly related to IN aerosol frequency because the RAOF is significantly lower in central South America.These results indicate that aerosols' 480 effect on nucleation cannot fully explain all changes of cold cloud phase in our study;In other words, there is no evidence to suggest that its effect is always dominant at each altitude of each region.Besides aerosol effect, what is the role of meteorological effect in determining cloud phase change, especially at those regions which aerosol effect on nucleation 485 isn't a first-order influence due to low IN aerosol frequency To further discuss this question, we analyze the seasonal and zonal variations of SCF and RAOF at -20ºC, LTSS and 500 hPa vertical velocity (see Fig.10).For the tropical region (from 20°S to 20°N), the RAOF is very low (approximately 0.005) and the seasonal variation is negligible.However, the corresponding SCF undergo a clear seasonal change.For 490 example, opposing SCF distributions are found during the summer and winter, and the maximum difference reached 15%.Opposite distributions are primarily linked to distinctly different atmospheric vertical motions, whereas surface temperature (figure not show) and LTSS contribute minimal amounts, due to their weak seasonal variation above this region.During the summer, the inter-tropical convergence zone (ITCZ) is correlated with deep convective clouds shift to the northern hemisphere and has strong vertical velocity.In contrast, the strength of atmospheric vertical motion in the northern hemisphere tropics decreases during the winter season, due to the ITCZ shift to the southern hemisphere.Poleward of 40°N, the inconsistency of seasonal variations in SCF and RAOF is particularly apparent.For example, maximum RAOF 500 and SCF are visible during winter at the middle and high latitudes of the northern hemisphere, whereas these values both decrease during the summer.Another apparent phenomenon is that winter RAOF values are larger than spring values poleward of 40°N.We recalculate the RAOF by considering only IN dust aerosols, and find that this phenomenon still exists.These results suggest that the trend is real and is not fully 505 caused by combined polluted dust and smoke frequencies, thus verifying the importance of meteorological effects on cloud phase changes.Figs.10a and 10bshow how the differences between winter and summer SCF may be linked to the seasonal variations in surface temperature and atmospheric stability (LTSS), whereas seasonal changes in vertical motion are weak and probably minimally affect SCF at the mid-510 and high-latitudes of the northern hemisphere.The results also show that the effects of different meteorological factors on cold cloud phase have regional characteristics.That is, low surface temperature and high LTSS (or a stable atmosphere) inhibit ice nucleation and push supercooled water to colder temperatures at mid-and high-latitudes, whereas strong vertical motion enhances ice nucleation in the tropics.515Tofurther quantify the effects of aerosol and meteorological factors on cold cloud phase, we group the RAOFs of grids into several RAOF bins within each specified vertical velocity, surface temperature or LTSS bin in order to analyze the relationship between the studied parameters (SCF at -20ºC, T 50 and n) and aerosol loading under different meteorological conditions.Here, T 50 is defined as the cloud top temperature 520 for exactly 50% of supercooled water clouds(Naud et al. 2006) and can be derived by inputting the T ice , T w , n and f (50%) thresholds into Eq.(1).Fig.11gives the seasonal and geographic variations of this parameter.Based on Fig.11, we find that T 50 in the high-latitude regions is lower (around -20ºC) than that in other regions, especially the middle latitudes of the two hemispheres, where T 50 reaches -10ºC.The apparent difference (approximately 10K) in T 50 values between the high and middle latitudes reflects the general tendency for supercooled water clouds to persist at colder temperatures in the high latitude regions, and this result based on global observation also further supports the findings ofNaud et al. (2006), who have analyzed MODIS data collected over the North Atlantic and Pacific Ocean basins.Naud et al. (2006) 530 have also found that the warmest T 50 values in each studied subregion generally occur in areas of ascent and heavy precipitation.However, by analyzing the distributions of T 50 on a global scale, our results indicate that strong subsidence areas also tend to generate warm T 50 values.This phenomenon is particularly apparent in Fig.12, which depicts the dependences of T 50 , n and SCF on the RAOF and 500 hPa vertical velocity.535Notably,the temperature bins used to calculate the RAOF differ for T 50 , n and SCF, based on their global distributions(See Fig. 4, Fig. 8 and Fig. 11).For example, parameter n reflects the relationship between SCF and CTT at the temperature ranges from -40ºC to 0ºC; thus, the calculation of RAOF for n considers aerosols at all temperature bins from -40ºC to 0ºC.However, only the aerosol samples at ±2 bins 540 around -20ºC are used to calculate the RAOF for SCF at -20ºC.For T 50 , the calculation of RAOF is primarily based on aerosol samples from the -20ºC to 0ºC temperature bins.We separate the relationship between T 50 and RAOF into three groups based on the strength of the 500 hPa vertical velocity (e.g., 0<|vertical velocity|<=25 hPa/day, 25<|vertical velocity|<=50 hPa/day and |vertical velocity|>50 545 hPa/day).Such grouping ensures a sufficient number of samples available in each bin (at least hundreds of samples in each bin) to ensure statistical significance.The error bars correspond to the ±5 standard error.Here, the standard error (SE) is computed as:, where SD is the standard deviation of the data falling in an RAOF bin and vertical velocity bin, and N is the sample number in each bin.As RAOF 550 increases (Fig. 12a), T 50 gradually tends to increase from around -20ºC (IN aerosols absent) to approximately -12ºC (high aerosol loading).The continuously increasing trend verifies that the existence of IN aerosols can hasten the glaciation of supercooled droplets through the Bergeron-Findeisen growth mechanism of ice / SE SD N = Atmos.Chem.Phys.Discuss., doi:10.5194/acp-2016-147,2016 Manuscript under review for journal Atmos.Chem.Phys.Published: 25 February 2016 c Author(s) 2016.CC-BY 3.0 License.

555
Fig.12billustrates the relationships between parameter n and aerosol loading for different vertical velocities.As with Fig.12a, the n value gradually increases as 570 Atmos.Chem.Phys.Discuss., doi:10.5194/acp-2016-147,2016   Manuscript under review for journal Atmos.Chem.Phys.Published: 25 February 2016 c Author(s) 2016.CC-BY 3.0 License.separate curves verify that the vertical velocity indeed significantly affects ice 585 nucleation even in the case of high aerosol loading.In fact, the effect is particularly important in those regions without IN aerosols where the SCF bias can reach 15%, such as in the tropics.For the same RAOF bin, a large vertical velocity can enhance the glaciation process and reduce the SCF, which partly explains why the persistent low SCF at -20ºC across central South America throughout the year could not be 590 explicitly related to aerosol frequency.Overall, different vertical motions lead to 10% mean SCF differences, and the bias is comparable to the effect of aerosols on cloud phase changes when the vertical velocity is limited to the same speed.Similarly, Figs. 13 and 14 show how T 50 , n and SCF depend on aerosols, surface temperature and atmospheric stability, respectively.Here, surface temperature is 595 classified into three levels: high, where the surface temperature>285 K; medium, where 270 K<surface temperature<285 K; and low, where surface temperature<270K.Generally, high surface temperatures enhance ice nucleation and reduce SCF values (or have large n values).As with strong vertical velocity, warm surface temperatures can also force supercooled water to glaciate at warmer temperatures.The difference in 600 Atmos.Chem.Phys.Discuss., doi:10.5194/acp-2016-147,2016 Manuscript under review for journal Atmos.Chem.Phys.Published: 25 February 2016 c Author(s) 2016.CC-BY 3.0 License.top level because vigorous updrafts do not leave enough time for supercooled droplets 615 different LTSS values are linked to different cloud systems.Compared with a convective cloud system, which requires a warmer surface temperature and lower 625 LTSS, stratiform clouds have a weak precipitation rate and inhibit the exhaustion of supercooled water.Thus, different T 50 values within and beyond the frontal ascent zone for similar vertical motions inNaud et al. (2006) actually reflect different surface temperatures and LTSS.By performing a similar analysis at different latitudes,

635
which further causes a difference in cloud types and amounts over this region.By comparing the different cloud types and covers at different seasons using the 2B-CLDCLASS-LIDAR, we find that shallow stratiform cloud covers such as, altostratus, stratus and nimbostratus clouds indeed are greater during winter than summer, providing a reason for why the SCF and RAOF are both larger during the 640 winter than the summer.
Atmos.Chem.Phys.Discuss., doi:10.5194/acp-2016-147,2016 Manuscript under review for journal Atmos.Chem.Phys.Published: 25 February 2016 c Author(s) 2016.CC-BY 3.0 License.increases with increases in IN aerosols from -9ºC to -4ºC, indicating that high aerosol 645 loading can enhance the temperature of glaciation.However, the effect of dynamic factors on T w is negligible.In addition, the stable relationship between aerosols (or dynamics) and T ice is not evident in our results, indicating the complexity of the distribution of T ice .Based on the above analysis, the current model (e.g., CAM5) already provides relatively reasonable T ice and T w values compared with other models.650 Its bias toward SCF is primarily caused by the unreasonable presentation of parameter n, which is closely linked to aerosol loading and meteorological factors.These results thus suggest that the effects of dynamics and aerosols on the parameters (especially for parameter n) in the studied cloud phase partitioning schemes are very important and should be considered in the parameterization of cloud phase in future studies in 655 order to further improve the calculation of cloud radiative effects related to cloud phase changes.

660 on 4
schemes, regardless of their geographic or temporal variations, which may result in considerable differences regarding the estimation of the supercooled liquid cloud 670

Fig. 1 .
Fig.1.The geographic and seasonal variations of T w value over 2º×6º grid boxes based 1070

Fig. 2 .
Fig.2.The geographic and seasonal variations of T ice value over 2º×6º grid boxes based on the 2B-CLDCLASS-Lidar product.

Fig. 6 .
Fig.6.The geographic and seasonal variations of the grid mean value of relative difference (annual mean) between calculated and observed SCFs for different schemes, respectively.(a) for the scheme 1 used the dynamical thresholds of T ice , T w and n; (b) 1090

Fig. 7 .
Fig.7.(a) The observed vertical distribution of zonal mean SCF with temperature; and the difference of vertical distribution between calculated and observed SCFs, (b) for 1095

Fig. 9 .
Fig.9.The geographic and seasonal variations of relative aerosol occurrence frequency (RAOF) at -20ºC over 2º×6º grid boxes based on the CALIPSO level 2 5 km aerosol level product. 1105

Fig. 12 .
Fig.12.The dependences of T 50 , n and SCF at -20ºC on the RAOF and 500hPa vertical velocity.The error bars correspond to the ±5 standard error.

Fig. 13 .
Fig.13.The dependences of T 50 , n and SCF at -20ºC on the RAOF and surface 1115

Fig. 14 .
Fig.14.The dependences of T 50 , n and SCF at -20ºC on the RAOF and LTSS.The error bars correspond to the ±5 standard error.

Fig. 6 .
Fig.6.The geographic and seasonal variations of the grid mean value of relative difference (annual mean) between calculated and observed SCFs for different schemes, respectively.(a) for the scheme 1 used the dynamical thresholds of T ice , T w and n; (b) for the scheme 2 used the dynamical thresholds of T ice , T w and n; (c) for the CAM3; and (d) for the CAM5.

Fig. 7 .
Fig.7.(a) The observed vertical distribution of zonal mean SCF with temperature; and the difference of vertical distribution between calculated and observed SCFs, (b) for the scheme 1 used the dynamical thresholds of T ice , T w and n; (c) for the CAM3; and (d) for the CAM5.

Fig. 12 .
Fig.12.The dependences of T 50 , n and SCF at -20º C on the RAOF and 500hPa vertical velocity.The error bars correspond to the ±5 standard error.

Fig. 14 .
Fig.14.The dependences of T 50 , n and SCF at -20º C on the RAOF and LTSS.The error bars correspond to the ±5 standard error.