The characteristics of ice clouds with a wide range of optical depths are studied based on satellite retrievals and radiative transfer modeling. Results show that the global-mean ice cloud optical depth, ice water path, and effective radius are approximately 2, 109 g m−2, and 48 , respectively. Ice cloud occurrence frequency varies depending not only on regions and seasons, but also on the types of ice clouds as defined by optical depth values. Ice clouds with different values show differently preferential locations on the planet; optically thinner ones ( < 3) are most frequently observed in the tropics around 15 km and in midlatitudes below 5 km, while thicker ones ( > 3) occur frequently in tropical convective areas and along midlatitude storm tracks. It is also found that ice water content and effective radius show different temperature dependence among the tropics, midlatitudes, and high latitudes. Based on analyzed ice cloud frequencies and microphysical properties, cloud radiative forcing is evaluated using a radiative transfer model. The results show that globally radiative forcing due to ice clouds introduces a net warming of the earth–atmosphere system. Those with < 4.0 all have a positive (warming) net forcing with the largest contribution by ice clouds with ~ 1.2. Regionally, ice clouds in high latitudes show a warming effect throughout the year, while they cause cooling during warm seasons but warming during cold seasons in midlatitudes. Ice cloud properties revealed in this study enhance the understanding of ice cloud climatology and can be used for validating climate models.
Clouds play an important role in modulating the Earth radiation budget and global hydrological cycle (e.g., Stephens et al. 1990; Chen et al. 2000; Haynes et al. 2013). They impact Earth radiation via their albedo and greenhouse effects, that is, cooling the Earth by reflecting solar incident radiation and at the same time, warming it by blocking longwave radiation. The net effect of the competition between these two processes depends on cloud properties, for example, cloud phase, water amount and thickness, and so on (Stephens and Webster 1981). For most mid- and low-level clouds, the solar albedo effect usually dominates so that a net cooling to the earth–atmosphere system because of these clouds is observed (Stephens 2005). In contrast, high and thin ice clouds are known to have a warming net effect since they are nearly transparent to solar incident radiation but opaque to longwave radiation. As a result, their greenhouse effect outweighs their solar albedo effect (Liou 1986). As optical depth increases, solar albedo effect would dominate and therefore the magnitude and sign of the ice cloud radiative forcing would rely largely on its optical thickness. Where this warming–cooling separation occurs in natural ice clouds is an interesting scientific question.
Classifications have been made for ice clouds to better study their properties. For instance, Sassen and Cho (1992) defined three categories of cirrus based on visible optical depth in their lidar observation studies, that is, subvisual cirrus ( < 0.03), thin cirrus (0.03 < < 0.3), and opaque cirrus 0.3 < < 3.0. Based on radiative transfer calculations with inputs of in situ measurement data, subvisual cirrus clouds were found to have a positive radiative forcing of approximately 1.6 W m−2 in the tropics (McFarquhar et al. 2000), while tropical cirrus clouds with 0.02 < < 0.3 tend to have a small influence on shortwave radiation (instantaneous forcing < 2 W m−2) but significantly impact longwave instantaneous emission (~20 W m−2) based on CloudSat measurements (Haladay and Stephens 2009). For deep convections, although their overall albedo effect usually offsets greenhouse effect (Rajeevan and Srinivasan 2000), their radiative forcing produced by the upper level of ice clouds has not been adequately investigated. Although optically thin ice clouds have received much more attention, the radiative effects of ice clouds over the entire spectrum of optical depth have not yet been well documented. Characterization of ice cloud properties over the entire optical depth spectrum is necessary in order to fully understand the radiative effects of all naturally occurring ice clouds.
Moreover, in studying the “climatology” simulated by various global climate models (GCMs), precipitation, cloud fraction, and precipitable water are reasonably well compared among different models. However, large discrepancies of global-mean atmospheric ice water amount exist with model-to-model results differing in the orders of magnitudes. The reasons for these discrepancies are that there are no robust global ice cloud observational data to constrain the GCM models and that only suspended clouds are considered in most GCMs (Waliser et al. 2009; Li et al. 2012). While there is clearly an urgent need for observed global ice water amount, substantial disagreements exist in current satellite retrievals, which are primarily because of the sensitivity of various observing techniques to different part of clouds (Eliasson et al. 2011). For example, passive sensors have difficulties in obtaining vertical ice water profiles. Although active sensors resolve the vertical structures of clouds, a single active remote-sensing technique merely detects a segment of clouds. Thin clouds are invisible to microwave, while visible and infrared are limited to thin clouds or the topmost part of clouds. As a result, a combination of multiple sensors/techniques becomes necessary to get a whole spectrum of atmospheric ice water (Wu et al. 2006, 2009).
New observations in recent years provide unprecedented possibility for studying and improving the understanding of ice clouds. The Cloud Profiling Radar (CPR) on CloudSat and the Cloud–Aerosol Lidar with Orthogonal Polarization (CALIOP) on Cloud–Aerosol Lidar and Infrared Pathfinder Satellite Observations (CALIPSO) provide global ice cloud retrievals for the first time with vertically resolved values of ice water content (IWC) (Stephens et al. 2002, 2008; Winker et al. 2003). For the CloudSat radar, thin ice clouds with IWC smaller than approximately 0.4 mg m−3 are invisible (Wu et al. 2009), whereas the CALIPSO lidar can only detect ice clouds with optical depth smaller than 3.0 (Liu et al. 2005). With respect to the total mass of ice water, thin ice clouds make very little contributions. However, when it comes to radiative effects, the role of thin ice clouds becomes vitally important because of their potentially strong warming effects (Liou 1986, 2002; McFarquhar et al. 2000). The joint observations by the CloudSat radar and the CALIPSO lidar make it possible to investigate ice clouds with a wide range of optical depths, as the former can detect thick clouds while the latter is better suited to the thin ones.
The approach of combining data from the CloudSat radar and the CALIPSO lidar has been taken by several investigators in cloud climate studies. For example, Sassen et al. (2008, 2009) studied global cirrus cloud properties and the cirrus’s connection to deep convections following the Sassen and Cho (1992) definition of cirrus ( < 3.0 and cloud-top temperature < −40°C). Using CloudSat and CALIPSO data, Schwartz and Mace (2010) found that cirrus clouds in the tropopause layer are generally independent to lower clouds, which provides insight into the formation and dissipation mechanisms of tropopause layer cirrus. Moreover, Mace (2010) surveyed cloud properties and radiative forcing over storm tracks of the Southern Ocean and North Atlantic using the CloudSat radar and the CALIPSO lidar observations, and they found a mean ice water path (IWP) of ~100 g m−2 as well as a strong net top of the atmosphere (TOA) cooling effect mainly caused by boundary layer clouds. While the above studies targeted a particular type or types of ice clouds in certain regions, a global climatology of the whole spectrum of ice clouds, including both optically thin and thick ice clouds and in both high and low altitudes, seems to still be missing.
The purpose of this study is to investigate cloud properties over the broad spectrum of ice clouds globally, which include 1) the magnitude of global ice cloud optical depth, IWP, and effective radius re; 2) global ice cloud spatial and seasonal variations; 3) the differences between optically thin and thick ice cloud properties; 4) the temperature dependence of ice microphysical properties; and 5) the ice cloud radiative effects over the entire spectrum of ice clouds. To achieve this goal, ice cloud–retrieved properties produced from the combination of the CloudSat radar and the CALIPSO lidar observations are analyzed in this study, and radiative transfer modeling is performed to estimate ice cloud radiative effects.
2. Data and methodology
Many satellite ice cloud retrieval products have been currently derived from a series of passive or active sensors (Waliser et al. 2009; Wu et al. 2009; Eliasson et al. 2011, 2013). Using active sensor measurements, CloudSat level-2B radar-only cloud water content (2B-CWC-RO; Austin 2007) provides vertically resolved retrievals of IWC and effective radius, but this product does not contain those thin ice clouds undetected by the CloudSat radar. To reach a wider range of IWP, a combination of the CloudSat radar and the CALIPSO lidar is used to derive ice cloud properties. These products include the CloudSat level-2C ice cloud property product (2C-ICE) (Deng et al. 2010) published by the CloudSat Data Processing Center and radar/lidar product (DARDAR) developed at the University of Reading (Delanoë and Hogan 2008, 2010).
Both DARDAR and 2C-ICE products derive ice cloud properties including ice water content, effective radius, and the extinction coefficient via a synthesis of both CloudSat radar reflectivity Ze and CALIPSO lidar attenuated backscattering . During the period of our study, CloudSat was 17.5 s ahead of CALIPSO. The CloudSat radar operates at 94 GHz with a minimum sensitivity of −30 dBZ and has a vertical resolution of 500 m (Stephens et al. 2002). The CALIPSO lidar is operating at 532 and 1064 nm with a vertically averaging resolution of 60 m in the troposphere (Winker et al. 2003). Although IWC, re, and are included in both DARDAR and 2C-ICE, these two products are distinct from one another by making different assumptions and using varied techniques in their reversion algorithms (Li et al. 2012). Deng et al. (2013) found IWC and of 2C-ICE agree well with DARDAR, especially in radar and radar–lidar regions. While in lidar-only region, DARDAR algorithms produce larger IWC and values than 2C-ICE, but the re values are still in reasonably good agreements. In this study, we choose one of these two agreeable ice retrieval products, DARDAR, to investigate global ice cloud properties. A brief description of the DADAR algorithm is given as follows.
According to the different sensitivities of lidar and radar to cloud liquid drops and ice particles, one advanced feature of the DARDAR algorithm is to distinguish between supercooled liquid water and ice by taking the advantage of lidar backscattering signals. Lidar produces strong echoes when meeting liquid cloud drops because of the large liquid particle number concentrations, while CloudSat radar signals from liquid drops are usually quite weak (Delanoë and Hogan 2008, 2010). Supercooled liquid water is identified by the CALIPSO lidar backscatter in regions of wet-bulb temperature lower than 0°C and air temperature higher than −40°C (Delanoë and Hogan 2008, 2010). Liquid water is assumed when wet-bulb temperatures are greater than 0°C. When temperatures are lower than −40°C, all hydrometeors are assumed to be ice. Lidar signals in and below the liquid water layer are discarded once a supercooled liquid layer is detected. Instead, radar signals are used to derive ice cloud properties in these regions by assuming that radar echoes are predominated by ice particles because of their large sizes compared to liquid water drops (Delanoë and Hogan 2008, 2010).
In performing the retrievals, the DARDAR algorithm adopts an optimal estimation formulation starting with a first guess of a state vector (extinction coefficient, number concentration, and extinction to backscatter ratio) for a forward model to predict observational parameters. The prediction is then compared to true observations to get the best estimates of the state vector by minimizing a cost function using Gauss–Newton iteration (Delanoë and Hogan 2008). Lidar multiple scattering is estimated by the fast multiple-scattering model of Hogan (2006). Even when only observations from one sensor are available, retrievals can still be performed by adopting empirical methods in the literature. For example, when only lidar data are available, IWCs can still be obtained by introducing a mass–size relationship of Brown and Francis (1995). In the absence of lidar observations, the area–size relation taken from Francis et al. (1998) is used to derive the extinction coefficient (Delanoë and Hogan 2010), and IWCs are then derived as a function of the radar reflectivity factor Ze and temperature. This method enables us to derive retrievals without gaps between optically thin and thick ice clouds.
Effective radius is derived by Foot (1988), which is related to the ice water content and extinction coefficient by
where is the density of ice and is the extinction coefficient at wavenumber . For the accuracy of retrievals, it is estimated that the extinction coefficient and IWC have significant uncertainties up to 60%, and those of effective radius are about 30% depending on the mass–size relationship (Delanoë and Hogan 2010). Nevertheless, it is believed that the DARDAR product provides the arguably best estimates of ice properties currently available with a wide range of ice water paths (Stein et al. 2011; Eliasson et al. 2013).
As a demonstration of how cloud forcing depends on cloud properties, we evaluated the ice cloud radiative effects on the earth–atmosphere system using radiative transfer modeling. In this evaluation, ice cloud radiative effects are estimated over the whole ice cloud spectrum using 4-yr-averaged ice cloud properties. The meteorological conditions (temperature, pressure, water vapor, and ozone) are averaged in the same period from European Centre for Medium-Range Weather Forecasts (ECMWF) reanalysis data (Dee et al. 2011). Ice cloud optical properties are derived from their microphysical properties using parameterizations of Yang et al. (2000, 2005). The radiative transfer model libRadtran, developed by Mayer and Kylling (2005), is used in this study, which has been employed for studying ice cloud radiative effects (Wendisch et al. 2005, 2007; Ehrlich et al. 2009). The discrete ordinate radiative transfer (DISORT) model, version 2.0 (Stamnes et al. 1988), with six streams is applied, and the correlated-k distribution band parameterization suggested by Fu and Liou (1992) is used for broadband absorption parameterizations. Tropospheric aerosols in spring–summer conditions with a visibility of 50 km below 2 km have been used as background aerosols (Shettle 1989). For global-mean simulations, earth surface albedo is set to be 0.15, while the mean solar zenith angle is calculated to be approximately 67.5°. When considering ice cloud forcing in different regions and seasons, calculations are performed at different solar zenith angles, and surface albedo is correspondingly set for different regions based on MODIS MOD43B3 (Moody et al. 2005) surface albedo product.
DARDAR data from 2007 to 2010 are used in this study. The retrieved variables have a horizontal resolution of 1.4 km and a vertical resolution of 60 m. To identify ice clouds, we used DARMASK_Simplified_Categorization, an index for scene categorizations included in the DARDAR product. In our data analysis, we only used those data where DARMASK_Simplified_Categorization is 1, which indicates no supercooled water was included.
a. Occurrence frequency of ice cloud properties
Statistics of ice cloud properties over the entire globe have been analyzed using the 4-yr-long DARDAR products. The global ice cloud frequency is approximately 53%; that is, ice clouds are observed in 53% of all samples. Figure 1 displays the frequency and accumulative frequency distributions of optical depth and ice water path. Ice cloud optical depths range from smaller than 0.1 up to over 100 as indicated by Fig. 1a. Thin clouds are usually related to nonprecipitating clouds, while those with large values of optical depth may be associated with precipitation. The results indicate that a global, conditional-mean optical depth of ice clouds is around 4, whereas a mean value is about 2 if calculated by including nonice samples (Fig. 1a). The mode value is around 1.2 on a logarithmic scale. Ice clouds with optical depth smaller than 0.03 account for about 9% of the total ice cloud population, while those with optical depth smaller than 0.3 and 3.0 make up about 40% and 79%, respectively, as denoted in Fig. 1b. In other words, subvisual ( < 0.03), thin (0.03 < < 0.3), and opaque (0.3 < < 3.0) ice clouds compose approximately 9%, 31%, and 39%, respectively, among the whole spectrum of ice clouds. Classification for the above three types of ice clouds originates from Sassen and Cho (1992) for cirrus. The term cirrus is not adopted in this paper since cloud height or temperature criteria are traditionally set for cirrus (Rossow and Schiffer 1991; Sassen et al. 2008), while we use optical depth alone. The portion of thick ice clouds is quite small. Ice clouds with > 20 account for approximately 5%, and those with > 100 account for merely 0.5%.
Similarly, IWP frequency and accumulative frequency distributions are shown in Figs. 1c and 1d. The frequency distribution has a mode value at approximately 30 g m−2 on a logarithmic scale. The IWP conditional-mean value is 205 g m−2, and the all sample mean is 109 g m−2. The cumulative frequency distribution shows that ice clouds with IWP larger than the mode value account for 59%, while those with IWP values smaller than the mean (~100 g m−2) are about 76%. In other words, the percentage of ice clouds with IWP greater than the mean value is only 24%, implying thick ice clouds contribute significantly to the global-mean ice amount even though their frequencies are small. Meanwhile, the frequency distribution of effective radius has a mode value around 33 , and the global-mean value is around 48 (Fig. 1e).
A common feature of ice cloud property distributions in terms of , re, and IWP is that the distributions are skewed with their mean values greater than their modes. In other words, a majority of ice clouds have an optical depth or an IWP value smaller than their mean values, indicating that most of the ice clouds are thinner than “a global-mean ice cloud.” Additionally, most ice crystals have a smaller effective radius than the global mean. Therefore, a mean cloud state cannot represent global ice cloud properties. Accordingly, applying a global-mean ice cloud state with the mean IWP and the mean effective radius to climatic models cannot evaluate the mean radiative properties of these clouds because of the nonlinear relationship between these parameters and radiative effects (Eliasson et al. 2011). Particularly, Chen et al. (2000) indicated that calculating radiative fluxes using preaveraged inputs would cause biases in cloud forcing. However, some earlier studies have used mean profiles to obtain qualitative analysis of cloud forcing (Schweiger and Key 1994; Zhang et al. 1995). Although a more adequate estimate based on profile-by-profile calculations should be carried out, we attended to interpret ice cloud forcing in a qualitative fashion in this study, so climatological properties of ice clouds were adopted as inputs to the radiative transfer model (section 3f).
b. Spatial variations of ice cloud properties
Cloud frequency distributions are important in determining the actual cloud radiative importance (Chen et al. 2000). To investigate the spatial distributions of ice cloud amount, we calculate ice cloud occurrence frequency, which is defined as
where is the number of ice cloud profiles, that is, IWP > 0, and refers to the number of all profiles in a longitude–latitude space (2.5° latitude by 2.5° longitude in a grid box). In a latitude–altitude space, refers to the number of IWC > 0, and is all cases in a 2.5° latitude by 60-m grid box. The number of Nall includes both clear and cloudy (ice and water) conditions. Using DARDAR data from 2007 to 2010, the global-mean ice cloud occurrence frequency is approximately 53%, as stated in section 3a. In Fig. 2, we show the horizontal and vertical distributions of ice cloud occurrence frequencies. As indicated by the horizontal distributions (Fig. 2a), large ice cloud occurrences are closely related to climate regimes, for example, frequently in tropical deep convection regions, including the Pacific warm pool, South America, and central Africa. They are also broadly distributed in mid- and high latitudes, such as near 60° in both hemispheres. The patterns of ice cloud occurrence frequency distributions in northern midlatitudes are modulated by land, whereas those in southern midlatitudes are quite zonally continuous. Tropical ice clouds are commonly located above 10 km but are mostly below 5 km in mid- and high latitudes, as shown by the vertical distributions in Fig. 2b. Moreover, the frequency of occurrence in Southern Hemisphere midlatitudes is generally greater than those in Northern Hemisphere midlatitudes. In particular, high frequencies around 1–3 km in southern midlatitude regions (Fig. 2b) agree with findings in Haynes et al. (2011), who found a high occurrence of low clouds over the Southern Ocean, with 79% of clouds having tops below 3 km.
The horizontal distributions of IWPs are shown in Fig. 3a. The global-mean IWP is approximately 109 g m−2. They generally display a similar pattern with their occurrence frequency distributions, in which large values are shown over equatorial convective regions and along storm tracks, while small values are mostly in the subtropics. Vertically, large ice water contents occur in several kilometers above the freezing levels (Fig. 3b). However, a high frequency of ice cloud occurrence does not necessarily collocate a large amount of ice water. For example, the mean IWPs over northern Eurasia are generally lower than 100 g m−2 (Fig. 3a), but the corresponding occurrence frequencies over the same area are relatively large (Fig. 2a). Similarly, tropical ice clouds frequently occur above 10 km (Fig. 2b), but IWCs are quite small at such high altitudes, suggesting that ice clouds in those places frequently occur but are usually thin. Meanwhile, the vertical distributions of effective radii (Fig. 3c) show that re continually decreases with altitudes. Large effective radii collocate with large IWCs in the tropics. Small re values (<20 ) happen above 15 km in the tropics and above 12 km in South Pole, while the most common tropical ice clouds located at 10 km through 15 km have re ranging from 20 to 50 .
c. Seasonal variations of ice cloud properties
Figure 4 corresponds to global ice cloud occurrence frequency distributions in four seasons, that is, March–May (MAM), June–August (JJA), September–November (SON), and December–February (DJF). Across all seasons, ice clouds are frequently observed in deep convective regions, storm tracks, and midlatitudes along 60°N/S. Seasonal variation of ice clouds in the tropics is closely associated with the movement of the intertropical convergence zone (ITCZ) from north of the equator during JJA to the south during DJF but is located almost symmetrically around equator during MAM and SON. These features are also evident in the vertical distributions, in which large occurrence frequencies in the tropics shift from north to south of the equator from JJA to DJF (Figs. 4f,h).
Additionally, fewer ice clouds occur in northern mid- and high latitudes during JJA but can be much more frequently observed during DJF, especially over the ocean (Figs. 4b and 4d). In contrast, southern mid- and high latitudes have consistently large ice cloud frequencies over four seasons, with the frequencies slightly larger during JJA than DJF (Figs. 4b and 4d). These features suggest that more ice clouds are produced in cold seasons for extratropical regions. Another region with a large number of ice clouds is coincident with the Asian summer monsoon, which moves following the seasonal march of monsoonal clouds. The vertical distributions reveal high frequencies of ice cloud occurrences in the stratosphere during JJA in the southern polar region (Fig. 4f). These are presumably polar stratospheric clouds, which have been observed during winter and spring times (Adhikari et al. 2010).
Seasonal variations of atmospheric ice water and effective radius are shown in Figs. 5 and 6. The distributions of IWPs (Fig. 5) in four seasons show similar patterns to those of ice cloud occurrences; that is, regions with high IWP values follow the movement of tropical convections and the seasonal march of the Asian monsoon. Large IWPs are observed in tropical deep convective areas and in midlatitude storm-track regions, while in other places, IWPs are relatively small. Global-mean IWP path values are about 114, 116, 115, and 111 g m−2 in MAM, JJA, SON, and DJF, respectively, being quite constant across all seasons. Vertically, the seasonal variations of IWC and re (Fig. 6) in the tropics are both associated with the movement of ITCZ. Effective radii smaller than 20 are consistently observed above 15 km in the tropics, while in the South Pole these small res occur with polar stratospheric clouds during JJA and SON.
d. Ice cloud properties across the spectrum of optical depth
Analyses in the previous sections have revealed great variability of ice clouds with respect to locations and seasons. In the following, we investigate how these features vary with different optical depths. In other words, we investigate the ice cloud properties across the spectrum of ice clouds.
To obtain a complete view of the whole spectrum of ice cloud distributions, five groups of ice clouds are considered in this study, that is, < 0.03, 0.03 < < 0.3, 0.3 < < 3, 3 < < 20, and > 20. The first three groups correspond to the three types of ice clouds, that is, subvisual, thin, and opaque, according to Sassen and Cho (1992). The last two groups are constructed with the aim of investigating the properties of thick ice clouds. According to section 3a, the five groups of ice clouds account for approximately 9%, 31%, 39%, 16%, and 5%, respectively, among all ice cloud samples, and their global occurrence frequencies among all observed samples are approximately 5%, 16%, 21%, 9%, and 3%, respectively.
The occurrence frequency distributions of the five ice cloud types as defined above are shown in Fig. 7. The frequency of subvisual ice clouds (Fig. 7a) is generally low but is widely spread over the globe. This type of cloud locates preferentially in the midlatitudes, especially in latitudes between 40° and 60° in both hemispheres. The patterns of frequency distributions for thin (Fig. 7b) and opaque (Fig. 7c) ice clouds are similar, with high frequencies primarily occurring over tropical deep convective regions and in mid- and high latitudes around 60°N/S, although thin ice clouds extend more in tropical oceans, while the opaque ones are more concentrated over land and in the warm pool regions. Ice clouds with 3 < < 20 are mainly observed in midlatitudes, and those with > 20 occur mostly in deep convective and storm track regions.
The vertical distributions of occurrence frequency for the above five groups of ice clouds are shown in Fig. 8, averaged respectively for the following five regions: tropics (TRP), northern midlatitudes (NML), northern high latitudes (NHL), southern midlatitudes (SML), and southern high latitudes (SHL). We define tropics, midlatitudes, and high latitudes using latitudinal boundaries of 23.5° and 66.5° in both hemispheres. Frequency here is calculated by the number of ice cloud observations (IWC > 0) divided by all measurements at a given layer.
Vertical profiles of subvisual ice clouds (Fig. 8a) show that they are primarily located in the midlatitudes at low altitudes of 1–2 km, especially in the southern midlatitudes. Another mode in the midlatitude profiles locates around 10 km. In the tropics, the relatively high frequency is at 10–17 km. The thin ice clouds (Fig. 8b) are frequently observed in the tropics above 10 km, and the peak frequency occurs near 16 km. In mid- and high latitudes, their occurrences are less frequent but with peak frequency in altitudes near 1–2 km. Opaque ice clouds (Fig. 8c) most frequently appear near 15 km in the tropics. Their patterns of frequency profiles in mid- and high-altitude regions are similar, uniformly occurring between 2 and 10 km in midlatitudes and with a bell-shaped profile for high latitudes. The frequency in Northern Hemisphere is larger than that in Southern Hemisphere. For thick ice clouds with 3 < < 20 (Fig. 8d), the maximum frequency is in the northern midlatitudes at an altitude around 5 km. The peak occurrence in the vertical frequency profile in the tropics is located near 12 km. For even thicker ice clouds with > 20 (Fig. 8e), the altitude of peak frequency is around 5 km in midlatitudes and slightly higher in the tropics. They are almost nonexistent in high latitudes. Results that subvisual and thin ice clouds dominate low altitudes in midlatitudes are consistent with the findings of Mace (2010) and Huang et al. (2012a,b) that boundary clouds are common in the Southern Ocean and North Atlantic. A summary of the above ice cloud distribution properties is given in Table 1.
Apart from the occurrence frequency, ice cloud microphysical properties also vary with optical depths. Figure 9 shows IWC and effective radius frequency distributions over the five groups of ice clouds. IWC frequency distributions are nearly Gaussian on a logarithmic scale, especially when < 3.0. Subvisual ice clouds have the narrowest IWC distributions (Fig. 9a). As optical depth increases, IWC distributions become broader and their modes increase as well. That is, smaller IWCs dominate in ice clouds with < 3.0, while thick ice clouds ( > 3.0) contain a wide range of IWCs. Compared to IWC, the widths of effective radius frequency distributions (Fig. 9b) are similar among all types of ice clouds but are more complicated than a Gaussian distribution. Subvisual ice clouds manifest three modes, that is, 19, 33, and 58 . The mode of 33 is also observed in all other types of ice clouds. Bimodal distributions are shown in distributions of thin ice clouds as well as of those with > 3.0, while opaque ice clouds have merely one mode. The mode values in these re distributions are closely related to ice cloud vertical distributions. For example, three modes of subvisual ice clouds are in fact corresponding to those ice clouds around 15 km in the tropics and 10 km and below 2 km in midlatitudes (Fig. 8a), where mean re values are about 19, 33, and 58 as shown in Fig. 3c. Similarly, the mode of 78 of ice clouds with > 20 corresponds to ice crystals at the base of thick ice clouds. More details on IWC and effective radius distributions are also summarized in Table 1.
Modes in the effective radius frequency distributions indicate that re are highly correlated with altitudes, while IWC frequencies show nearly Gaussian distributions, indicating IWC are related to other factors. These features can be more clearly seen in Fig. 10, in which the global-mean IWC and re are shown as a function of altitude and optical depth. The results indicate that ice clouds with large IWC and re mainly occur at low altitudes when optical depth is large. Generally speaking, IWC follows mainly the change of optical depth, while re is more closely related to altitude than optical depth, in particular where the altitude is above 5 km.
e. The dependence of microphysical properties on temperatures
In many climate and weather prediction models, temperature is often treated as an independent variable to parameterize microphysical properties such as the number of ice particles. In the following, we investigate the temperature dependence of ice water content and effective radius and their seasonal and regional variations. For convenience, in the following we define the warm season as March through August in the Northern Hemisphere and September through February in the Southern Hemisphere, while the cold season is March through August in the Southern Hemisphere and September through February in the Northern Hemisphere. Thus, we divide ice clouds into five groups according to regions and seasons: TRP, midlatitude warm seasons (ML-W), midlatitude cold seasons (ML-C), high-latitude warm seasons (HL-W), and high-latitude cold seasons (HL-C). Based on this grouping, occurrence frequency is examined to explore the variations of ice cloud microphysical properties over different regions and seasons with regards to the change of temperature.
Figures 11 and 12 show the occurrence frequencies of ice clouds (donated by the occurrence number in this section) in IWC–temperature space, which indicates how IWC and re are related to temperature. The most evident features shown in these two figures are that IWC and re grow as temperature increases, which agrees with previous findings (e.g., Stephens et al. 1990; Ou and Liou 1995). However, the results also indicate that there are visible differences of the relation among ice clouds in different regions. Tropical ice clouds are more frequently observed with relatively small IWC (<0.01 g m−3), low temperature (190–230 K), and small re (20–40 ). In midlatitudes, ice clouds have a high frequency of occurrence at relatively high temperature (210–255 K), large re (30–60 ), and high IWC (up to 0.1 g m−3). High-latitude ice cloud occurrence frequencies are similar to those in midlatitudes but with some additional occurrences at very low temperatures (~180 K) with small IWCs (Fig. 11e) and res (Fig. 12e) in cold seasons. These clouds are most likely to be polar stratospheric clouds (Sassen et al. 2008; Adhikari et al. 2010). The above analysis of ice cloud occurrence frequency reveals that the temperature dependence of ice cloud microphysical properties varies with regions and seasons, which implies a cloud parameterization in GCMs should treat ice clouds differently according to climate regimes.
As discussed in section 1, GCMs-produced ice water contents have large discrepancies among models. Besides the lack of observational constraints and insufficient understanding of ice cloud properties, most GCMs do not consider falling ice particles/snow, which would also lead to disagreements between these models (Waliser et al. 2009; Li et al. 2012). However, the CloudSat radar and the CALIPSO lidar are sensitive to both suspending and falling particles, which might offer implications for improving parameterization schemes among GCMs. As a utility of satellite ice cloud observations for model verification, we provide in the following mean IWC and re states in a thermodynamic (pressure–temperature) diagram. Figures 13 and 14 show the global conditional means of IWC and re from the DARDAR retrievals as a function of temperature and pressure for five groups: tropics, midlatitude warm seasons, midlatitude cold seasons, high-latitude warm seasons, and high-latitude cold seasons. Figure 13 displays that large IWC values are produced in high temperatures of ~270 K around a level of 700 hPa in both midlatitudes and the tropics. In contrast, large IWCs in high latitudes are formed in two areas: high temperature (~270 K) near the surface and low temperatures (<220 K) with pressure ranging from 400 to 800 hPa. In the former situation, IWCs grow as temperatures increase, which are the same as those in the tropics and midlatitudes. This positive correlation has been adopted in climate models (Stephens et al. 1990; Ou and Liou 1995). In the latter situation, large IWCs (e.g., >0.1 g m−3) exist in temperatures as low as 220 K between layers of 400–800 hPa. These cases exist in high-latitude warm and cold seasons. However, their samples are relatively small, especially in warm seasons (Figs. 11d,e). In some degree, it is the extremely cold temperatures in the polar region that allow relatively large IWCs to exist. When investigating these large IWCs in low temperatures (not shown), we found these cases mostly exist in the southern polar region in the winter season, during when Antarctica suffers from very low temperatures.
Unlike IWCs, effective radii tend to be uncorrelated with pressure, which are almost exclusively determined by temperature, especially in mid- and high latitudes (Fig. 14). In other words, IWC–temperature relationships are more complicated than those of re temperature. The former depends on thermodynamic conditions including both pressure and temperature, while the latter is nearly always positively correlated with temperature for all pressure levels.
f. Ice cloud radiative forcing
In this section, we demonstrate how ice cloud radiative effects vary with optical depth values. To do so, we use the seasonally and regionally dependent cloud occurrence frequencies and cloud microphysical properties derived from previous sections as input for radiative forcing computations. The radiative transfer model libRadtran (Mayer and Kylling 2005) is used to calculate radiative fluxes. The ice cloud forcing is calculated by
where C is cloud forcing, and F is radiative flux with upward and downward arrows indicating the fluxes’ directions, and subscripts denote shortwave (SW) and longwave (LW) radiation. The “cloudy” and “clear” subscripts respectively indicate conditions with and without ice clouds.
The radiative forcing induced by different types of ice clouds is examined at the TOA, at the surface, and in the atmosphere. Figure 15 displays the SW, LW, and net radiative forcing at both the TOA and the surface, which have been weighted by ice cloud optical depth frequency distributions (Fig. 1a) to obtain a global-mean forcing. It is shown that ice clouds with ~ 1.2 contribute the most to the forcing in both the LW and the SW, while contributions from thick ice clouds are insignificant on average as these types of ice clouds occur less frequently. The magnitude of SW forcing at the TOA is very close to that at the surface (Fig. 15a). In the LW, the forcing is much larger at the TOA than that at the surface. The net forcing, that is, the sum of the SW and the LW forcing, reveals that solar albedo effect dominates the greenhouse effect at the surface, whereas at the TOA, the greenhouse effect outweighs the solar albedo effect when optical depth values are smaller than approximately 4.0 (Fig. 15c). The mean SW forcing at the TOA and at the surface is −51 and −49 W m−2, respectively, while the mean LW forcing is 55 and 33 W m−2 at the TOA and at the surface, respectively. As a result, a global net mean forcing at the TOA is 4 and −16 W m−2 at the surface, implying that ice clouds keep the earth–atmosphere system warming although they cool Earth’s surface. This warming effect is generated by ice clouds with < 4.0 with peak contribution being produced by those with ~ 1.2.
For ice cloud forcing within the atmosphere, the difference of heating rates (K day−1) between cloudy and corresponding clear-sky conditions is applied as the indicator of the impact of ice clouds. Vertical heating structures induced by ice clouds are shown in Fig. 16. Generally, within the atmosphere the ice cloud–induced heating rate change is much larger in the LW than in the SW. Ice clouds in the SW cause temperatures to increase for the atmosphere above 10 km and cool below. Conversely, temperatures are decreased by ice clouds in the LW above 12 km and increased below. The net heating rate change shows that the atmosphere is warmed by ice clouds below 12 km and cooled above. Along the whole spectrum of ice clouds, strong cooling or warming is primarily because of ice clouds with 0.2 < < 10.
To study ice cloud forcing in different regions and seasons, ice cloud forcing as a function of optical depths is studied for ice clouds for five groups as defined in section 3e. Results are displayed in Fig. 17. Ice cloud forcing at the TOA shows a warming effect when optical depth is small ( < 2.0) and a cooling effect as optical depth increases in the tropics (Fig. 17a). It is, however, always positive in high latitudes at both the TOA and the surface, implying ice clouds keep these regions warming over the whole year (Figs. 17a,b). In midlatitudes, ice clouds in warming seasons have a stronger SW forcing but weaker LW forcing than those in cold seasons, causing a cooling effect during warm seasons but a warming effect during cold seasons at both the TOA and the surface (Figs. 17a,b).
The above cloud forcing evaluation demonstrates the complexity of the radiative effects of ice clouds on the earth–atmosphere system, that is, their dependence on regions, seasons, and optical depths. To fully understand the ice cloud radiative effects, their spatial–temporal distributions and microphysical properties must be characterized over a global scale. The results presented in the previous sections serve as one of steps toward that goal.
4. Summary and conclusions
The objective of this research aims at enhancing our understanding of ice cloud properties using state-of-the-art observations from the CloudSat radar and CALIPSO lidar observations. Emphases are placed on 1) global and zonal distributions of ice cloud occurrences and ice water mass; 2) geographical and seasonal variations of ice cloud properties; 3) differences of ice cloud properties across a wide range of optical depths; 4) microphysical properties in relation to thermodynamic variables, with implications for GCM model verifications; and 5) ice cloud radiative effects as a function of optical depth and over different regions and seasons.
Ice clouds have a wide range of optical depths with peak frequency around ~ 1.2. The global-mean optical depth and effective radius are about 2 and 48 , respectively. The mean global IWP is assessed to be approximately 109 g m−2, which is larger than those produced by most GCMs based on the Intergovernmental Panel on Climate Change Fourth Assessment Report (IPCC AR4) (Waliser et al. 2009). Most GCMs do not include falling particles and convective cores of mass, which contributes uncertainties to ice cloud simulations (e.g., Waliser et al. 2009; Li et al. 2012).
Ice clouds frequently occur in deep convective areas in the tropics as well as in mid- and high latitudes. Their seasonal variations are associated with seasonal shift of climate regimes, for example, ITCZ, monsoon, and midlatitude storm tracks. Also, more ice clouds are formed during cold seasons than during warm seasons.
Macroscale distributions and microphysical properties are also analyzed over the entire spectrum of ice clouds. Ice clouds with < 0.03 (subvisual) have a low occurrence frequency (~5%) but widely spread over the globe. Subvisual ice clouds occur more frequently in the southern midlatitude ocean with altitudes near 1–2 km as well as in the tropics above 15 km. The observed occurrence frequency of both thin and opaque ice clouds (0.03 < < 3.0) is approximately 37%, which accounts for 70% within the ice cloud spectrum. Ice cloud occurrence frequency peaks around 15 km in the tropics. Thick ice clouds ( > 3.0) mainly exist in the tropics and over the ocean in midlatitudes associated with deep convections. Thick ice clouds are observed to occur 11% globally, which is much greater than that of subvisual ice clouds. However, the former is absent in the subtropics, while the latter is distributed more broadly.
IWC and effective radius reveal clear differences of microphysical properties between the five types of ice clouds defined in this study. For a given ice cloud type, IWC displays a near-Gaussian distribution on a logarithmic scale, while the effective radius often shows multiple modes. A mode of 33 dominates in effective radius frequency distribution. As optical depth increases with the change of ice cloud type, mode in IWC distribution increases accordingly. However, there is no clear trend of variation for the mode of effective radius. Rather, its mode values are closely related to most probable altitudes where ice clouds occur.
As for the relation between temperatures and ice cloud microphysical properties, tropical ice clouds mostly occur in low temperatures (190–230 K), with small effective radii (20–40 ) and low IWCs (0–0.01 g m−3), while mid- and high-latitude ice clouds have large portions with high temperatures (210–255 K), large effective radii (30–60 ), and large IWCs (up to 0.1 g m−3). The effective radius is positively correlated to temperature but almost uncorrelated to pressures. On the other hand, the relationships between IWC and temperature vary with pressures. They are negatively correlated in high latitudes when temperature is low but positively correlated in other situations. Cases in high latitudes with temperature < 220 K and IWC > 0.1 g m−3 mostly occur in polar winter seasons, especially in Antarctica.
As a demonstration of how ice cloud radiative effect varies with cloud properties, radiative transfer modeling has been performed using analyzed results from this study as model inputs. On average, ice clouds as a whole have a positive net forcing at the top of atmosphere, mostly contributed by those with optical depth ~ 1.2. Ice clouds with < 4.0 show a warming effect, while those with higher τ show a cooling effect on the earth–atmospheric system. Ice clouds in high latitudes keep the earth–atmosphere system warming through the year. The strongest cooling effect over the globe occurs in midlatitude warm seasons when the solar albedo effect outweighs the greenhouse effect. In the tropics, ice clouds with < 2.0 display a warming effect. While these findings revealed some characteristics of ice cloud radiative forcing, its precise climatological value requires aggregating individual ice cloud forcing calculated from observational profiles, which will be our future research topic.
This study has presented a comprehensive analysis of the properties of all types of ice clouds with respect to their horizontal distributions, vertical structures, regional and seasonal variations, the dependence on temperature and pressure, and a qualitative analysis of radiative forcing over the whole spectrum of ice clouds. These results can, on one hand, advance our understanding of ice cloud climatology. On the other hand, they are useful for validating microphysical parameterizations in global climate models.
We acknowledge members of the DARDAR project in the University of Reading for providing DARDAR data. DARDAR data were obtained from the data archive at the University of Reading. The ECMWF and MODIS MOD43B3 datasets are obtained respectively from archives at ECMWF and at NASA. This research has been supported by NASA Grants NNX13AQ39G and NNX13G34G.