2.1.

CALIPSO Data Set

The payload of CALIPSO satellite consists of three coaligned, near-nadir viewing instruments: a dual-wavelength Cloud Aerosol Lidar with Orthogonal Polarization (CALIOP), an imaging infrared radiometer, and a high-resolution wide-field camera.²⁴ The CALIOP uses an Nd:YAG laser that transmits pulses at both 532- and 1064-nm wavelengths, with pulse energy of 110 mJ, and a pulse length of 20 ns. The lidar pulse repetition at the rate of 20.16 Hz allows for a sampling of complete atmospheric column every $\sim 333\mathrm{m}$ on the ground. To reduce the bandwidth of the downlinked data, the resolution of the data decreases with altitude. The vertical and horizontal resolutions of the downlinked data are 30 and 333 m between the ground and 8.2 km, 60, and 1000 m between 8.2 and 20.2 km, respectively. The CALIOP measurements are calibrated and geolocated to produce attenuated backscattering profiles that are made available to users as level-1 data product. Feature detection algorithms and cloud properties retrieval techniques are applied to the level-1 attenuated backscatter profiles to yield level-2 cloud data products.

We use four years (December 1, 2006 to November 30, 2010) of CALIOP measurements in this study. We use the 5-km cloud layer data product (i.e., CAL_LID_L2_05kmCLay-Prov-V3), which provides information on all clouds detected from various horizontal signal-averaging schemes. The CALIPSO feature detection algorithm uses several resolutions (i.e., 5, 20, and 80 km) and signal averaging to increase the signal-to-noise ratio of the measurements and detect even, thin clouds and aerosol layers in the atmosphere.²⁵ All the clouds that are detected are provided in the 5-km data product along with the horizontal averaging used in identifying those clouds.

2.2.

Cloud Classification Algorithm

We summarize our cloud classification algorithm in Fig. 1. We first limit our attention to tropospheric clouds. Our algorithm employs a maximum allowable cloud top height as 1 km above the local tropopause level to avoid the inclusion of stratospheric clouds. We used a maximum allowable height of 12 km for the regions poleward of 60° N and 60° S as the tropopause altitude is often ambiguous in the polar regions. We then identify all clouds with cloud top altitudes less than 2 km above the ground as low-level clouds (L). We classify the clouds with top altitudes greater than 2 km of the ground level into six distinct types based on the cloud top temperature, cloud base temperatures, and optical depth. We classify the clouds with top temperature above $-38°\mathrm{C}$ as midlevel clouds (M) and below $-38°\mathrm{C}$ as high-level (H, i.e., cirrus) clouds. Studies show that as the cloud temperature falls below $-38°\mathrm{C}$ supercooled liquid droplets freeze homogeneously to initiate the formation of ice particles.^26–28 We use the transparency flags of the cloud layers provided by the CALIPSO data product to classify the clouds as attenuated or transparent. The CALIOP data processing algorithm characterizes the identified cloud layers as either attenuated or transparent and calculates the optical depth of transparent (i.e., optically thin) clouds with visible optical depths less than 3 to 4.²⁹ Finally, we use the cloud base temperature of $-10°\mathrm{C}$ to distinguish between ice- and water-dominated clouds. Studies have shown that clouds have a minimum ice particle concentration of 1000 particles/m at temperatures less than $-10°\mathrm{C}$.³⁰ Using these rules, we categorize all the tropospheric clouds into seven types (i.e., L1, M1, M2, M3, H1, H2, and H3). We present these categories of tropospheric clouds with their descriptions in Table 1. The M1 clouds are the ice dominated optically thin midlevel clouds that are the primary focus of this study.

Fig. 1

Logic-based rules used to categorize tropospheric clouds detected by CALIPSO.

Table 1

Description of cloud types based on logic-based rules.

We show an example of our cloud classification scheme on CALIPSO data segment drawn from over both northern and southern hemispheres in Fig. 2. Figure 2(a) shows CALIPSO measurements on December 27, 2006, over the Arctic Ocean from Bathurst Island (75° N, 99° W) to the New Siberian Islands (75° N, 143° E). We detect six of our seven cloud types (i.e., L1, M1, M2, M3, H1, and H3). Several layers of optically thin high-level clouds (H1) and optically thin midlevel clouds (M1) dominate the clouds in this observation. The M1 cloud observed over the New Siberian Islands (75° N, 143° E) spans a horizontal scale of 135 km (27 profiles). In Fig. 2(a), the temperature remains less than $-10°\mathrm{C}$ at the surface, thus there is no contour plotted at $-10°\mathrm{C}$. Figure 2(b) shows CALIPSO measurements on May 18, 2007, over the Southern Ocean (45° S, 91° E to 57° S, 85° E). This example clearly illustrates the occurrence of optically thin midlevel ice cloud, where the lidar signals could penetrate through the cloud and detect a cloud layer lying below. The M1 cloud observed here spans a horizontal scale of 110 km (22 profiles).

Fig. 2

(Upper) CALIPSO backscattering (lower) and cloud type derived from our logic-based rules from an overpass (a) over the Arctic region on December 27, 2006 and (b) over the Southern Ocean on May 18, 2007. The dotted line indicates the 2-km altitude level above the surface. The temperature contours at $-10°\mathrm{C}$ and $-38°\mathrm{C}$ are also plotted.

Having first classified all tropospheric clouds into seven types, we investigate the horizontal scale of the clouds using a continuity analysis. We conclude that a cloud has a scale of $n$ profiles if a given cloud type is present at the same altitude over $n$ successive profiles. We choose $n$ as 2, 5, 10, 20, and 100 corresponding to 10, 25, 50, 100, and 500 km, respectively. The use of continuity over increasing numbers of successive profiles not only provides information on the spatial scales of the cloud, but also eliminates isolated cloudy profiles from the sample (e.g., M2 and H3 in Fig. 2). After counting the total number of valid clouds in each type, we calculate the frequency of occurrence and fraction of clouds based on the total observations (profiles) divided by the value of $n$. We then bin the types of clouds obtained from the entire data set into a $2.5°\times 5°$ latitude–longitude grid. We derive cloud statistics for an average of four years of binned data. We define the frequency of occurrence of a cloud type as the number of cloudy profiles to the total number of profiles (clear plus cloudy). We define the cloud type fraction for a given cloud type as the ratio of the number of clouds identified as that particular type to the total number of clouds identified.

3.1.

Global Distribution of Tropospheric Clouds

We first investigate the CALIPSO observations at 1-profile resolution or a 5-km spatial scale. Over the four-year period, there were 149.5 million profiles, where clouds are absent in 46.0 million profiles. This indicates that 31% of the Earth’s surface is under clear skies, while 69% is cloudy. In the cloudy profiles, 28% of the clouds are low-level clouds (L1), 36% are midlevel clouds (M1, M2, and M3), and 36% are high-level clouds (H1, H2, and H3). We find that these results are insensitive to changes of $\pm 2°\mathrm{C}$ in the cloud top and cloud base temperature thresholds of our classification scheme. In summary, we find that M1 clouds have the following characteristics, optical depths between 0.1 and 6.0 with a mean optical depth of 0.8, cloud top heights between 2.0 and 12.0 km with mean height of 6.0 km, cloud top temperatures between $-38°\mathrm{C}$ and $-10°\mathrm{C}$ with a mean temperature of $-27°\mathrm{C}$, cloud base heights between 0.1 and 11.3 km with a mean height of 4.7 km, and cloud base temperatures between $-37°\mathrm{C}$ and $-10°\mathrm{C}$ with a mean temperature of $-21°\mathrm{C}$.

As seen in Fig. 2, clouds of one type can be found as single profiles embedded in a larger cloud of another type (e.g., M2 in M1). We assume this may reflect a classification error. We thus present our results based on the CALIPSO observations at 2-profile resolution or a 10-km spatial scale. We tabulate the frequency of occurrence and the fraction of clouds for each category from the four years (December 1, 2006 to November 30, 2010) in Table 2. Over these four years, the total number of profiles is divided by two and amounts to 74.7 million profiles. We find a total of 45.3 million valid cloudy profiles, indicating a global cloud cover of 61% for clouds of a horizontal size of 10 km or greater.

Table 2

Cloud detection statistics for CALIPSO observations at 10-km scale.

We present the global distribution of cloud frequency of occurrence and fraction at 2-profile resolution in Fig. 3. The observations show that L1 and H1 clouds (low-level and cold thin cirrus clouds) occur most commonly and represent the highest fraction of the cloud types. The L1 clouds occur more commonly over oceans than land. The lower occurrence of L1 clouds over Southern and South East Asia might be due to the presence of H3 or M3 clouds (thick high-level and midlevel clouds) that prevent the detection of underlying L1 clouds. M1 clouds (cold thin midlevel clouds) form most commonly in the extratropics. In the northern hemisphere, the M1 clouds represent a larger fraction of the clouds over land than ocean, which reflects the increased occurrence of L1 clouds over the oceans. M2 clouds (warm thin midlevel clouds) primarily occur in the tropics. M3 clouds (thick midlevel clouds) occur more commonly over ocean than land and extend into the polar regions over the North Atlantic and Southern Oceans. H1 clouds (cold thin cirrus clouds) are found primarily in the tropics and also over Antarctica. The detection of cirrus clouds over the Antarctic may reflect the formation of polar stratospheric clouds that extend into the upper troposphere. H2 clouds (warm thin cirrus clouds) are largely nonexistent. H3 clouds (optically thick high-altitude clouds) are most common in the tropics, least common in the subtropics, and also form in the midlatitudes and polar regions.

Fig. 3

Global (a) cloud frequency of occurrence and (b) cloud type fraction derived from CALIPSO observations at 10-km resolution.

3.2.

Spatial Scales of Clouds

We have determined the number of clouds present at different scales in the CALIPSO data set by detecting clouds that are continuous over increasing larger scales. The total number of clouds detected decreases with increasing scale from $4.53\times {10}^7$ clouds at 2-profile (10 km) resolution to $1.15\times {10}^6$ clouds at 20-profile (100 km) resolution to $1.70\times {10}^4$ clouds at 100-profile (500 km) resolution. The global mean fraction of clouds derived from CALIPSO observations at different spatial scales is shown in Fig. 4. We see that with increasing spatial scale, the relative fraction of the different cloud types changes. The global mean fraction of clouds for L1 clouds is found to increase from $\sim 30\%$ at 10 km to $\sim 50\%$ at 500 km. The fraction for H1 and H3 clouds is found to stay relatively constant at $\sim 10\%$ and $\sim 30\%$, respectively. The fraction of midlevel clouds decreases with increasing spatial scales. The fraction of M3 clouds decreases less rapidly, from $\sim 20\%$ at 10 km to $\sim 10\%$ at 500 km, than the fraction of M1 and M2 clouds. For M1 clouds, the fraction of clouds is 7.4%, 4.2%, 2.0%, and $<0.1\%$ at 2-profile (10 km), 5-profile (25 km), 10-profile (50 km), and 20-profile (100 km) resolution, respectively. At the largest spatial scale of 500 km, we find only five M1 clouds in this four-year data set. These five clouds are found in the Arctic. We will use these five large M1 clouds later to investigate the radiative effects of M1 clouds.

Fig. 4

Global mean fraction of clouds derived from CALIPSO observations plotted as a function of horizontal spatial scale. 1 profile corresponds to 5 km.

3.3.

Seasonal and Geographic Distributions of M1 Thin Midlevel Ice Clouds

We show the seasonal variations in M1 cloud occurrence and cloud type fraction in Fig. 5. These thin midlevel ice clouds occur up to 15% over the polar regions and extend into the midlatitudes. M1 clouds occur most commonly during the northern winter [December, January, February (DJF)] and least during the northern summer [June, July, August (JJA)], in both hemispheres with an exception of the two high-altitude regions of Greenland and Tibet. These clouds appear along the midlatitude storm tracks in all seasons but are quite low in northern summer (JJA). No seasonal symmetry in M1 cloud occurrences is observed. M1 clouds occur most commonly over both the Arctic and the Southern Ocean during northern winter (DJF) indicating that M1 clouds form through a variety of processes. Over the tropics, M1 cloud occurrences are found to be minimum (5%) and are concentrated along the convergence zone, resembling the altocumulus cloud distribution.³¹ The fraction of clouds clearly shows that M1 clouds are more abundant in the northern hemisphere than in the southern hemisphere. This may be related to the availability of ice nuclei for the formation of M1 clouds. Studies have shown that dust particles from the arid regions act as effective ice nuclei.^32,33 The occurrence of high fraction of M1 clouds over arid regions in Africa, Asia, and Australia supports the assertion that such nucleation processes contribute to the formation of M1 clouds. In the northern hemisphere, M1 clouds are most common along the western Arctic extending into the midlatitudes and less common in the synoptically active North Pacific and Atlantic Oceans, where cloud droplets grow to large sizes and produce precipitation. Over the southern hemisphere, M1 clouds are found over the oceans surrounding Antarctica.

Fig. 5

Global seasonal variation in frequency of occurrence and cloud type fraction of M1 clouds at 10-km resolution.

We show the zonally averaged frequency of occurrence of M1 clouds with altitude in Fig. 6. The M1 clouds occur at highest altitudes in the tropics and decrease in altitude toward the poles. The clouds are found at lowest altitudes in the wintertime at the poles. The clouds occur most often in the polar regions and least often in the subtropics. The occurrence of M1 clouds follows the atmospheric circulation, where the clouds are less common in the descending branch of the Hadley-cell in the subtropics. In the tropical regions, M1 clouds are found at altitudes of 7- to 10-km altitude, while the tropopause level varies between 16 and 18 km. This observation supports the view that tropical M1 clouds are formed through detrainment of convective towers that extend into the upper atmosphere.^34,35

Fig. 6

Zonally averaged frequency of occurrence of M1 clouds plotted as a function of latitude and altitude.

We tabulate the macrophysical properties of M1 clouds over different latitude regions in Table 3. Here, the relative occurrence is calculated as the ratio of M1 clouds that occur in each latitudinal belt to the global M1 cloud amounts. M1 clouds are more abundant in the northern hemisphere ($\sim 47\%$) than in the southern hemisphere ($\sim 38\%$) extratropics (poleward of 30°). As mentioned earlier, this may be related to the availability of ice nuclei for the M1 cloud formation. Further, the average M1 cloud thickness increases from tropical to polar regions in both hemispheres. The average M1 cloud thickness is higher in the Arctic (1.4 km) than any other region. This observation is consistent with the thin midlevel ice cloud climatology derived from ground-based lidar observations that reported average thin midlevel ice cloud thickness as 1.7 km over the sub-Arctic region.³⁶

Table 3

Macrophysical properties of M1 clouds observed by CALIPSO.

We highlight the occurrence of M1 clouds in the polar regions in Fig. 7. We see distinct seasonal asymmetry between the occurrence of these clouds in the Arctic and the Antarctic. These clouds occur most commonly in the wintertime (DJF) in the Arctic and the summertime (DJF) in the Antarctic. These clouds occur least commonly in the summertime (JJA) in the Arctic and the wintertime (JJA) in the Antarctic. We also see that the clouds occur more commonly over low altitudes than high altitudes, being less common over Greenland and the Antarctic plateau than the surrounding regions. The M1 clouds are less common over the North Atlantic, where M2 clouds are more common.

Fig. 7

Frequency of occurrence of M1 clouds as a function of season in the Arctic and the Antarctic.

4.1.

Radiative Transfer Model

We use the library for radiative transfer (libRadtran) package to estimate the radiative properties of M1 clouds.³⁸ LibRadtran is a software package designed to calculate the radiative quantities such as radiances, irradiances at user-defined altitudes within the atmosphere using input data of the extraterrestrial radiation, the atmosphere (pressure, temperature, humidity, ${\mathrm{O}}_3$, ${\mathrm{N}}_2$, O, and ${\mathrm{CO}}_2$ profiles), and surface properties. The complete details of the libRadtran model along with examples on how to run simulations are provided online.³⁹ This package is publicly available and suitable for computing radiative effects of aerosols and clouds under user-defined boundary conditions.⁴⁰ We use the meteorological data (i.e., pressure, temperature, relative humidity, and ozone number density) obtained from the CALIPSO data product as inputs to the model to represent the atmospheric conditions for the observed clouds. We choose the standard subarctic winter profiles for the atmospheric data above 30-km altitude.⁴¹ The aerosol microphysical properties of maritime type for the winter season represent the background aerosol conditions for the simulations. At $1\mu \mathrm{m}$, the maritime aerosol has an aerosol optical thickness of 0.11 and an angstrom exponent of 0.63.⁴² The surface spectral albedo plays a crucial role in the simulation of radiative transfer in the solar spectral range. The libRadtran software provides an option to input any measured surface albedo or to use a standard library of spectral albedo values. These standard values are taken from International Geosphere–Biosphere Program database that provides spectral albedo of these surface types derived from satellite measurements.⁴³ However, for our simulations, we use a modified spectral albedo of the surface-type permanent ice/snow that we have validated with published studies.^21,37 The simulations in the solar spectral range (290 to 4000 nm) are performed with the discrete ordinate radiative transfer solver DISORT version 2.0.⁴⁴ We use the extraterrestrial solar irradiance spectrum in the solar spectral range.⁴⁵ We use the LOWTRAN band model from SBDART to account for the gaseous absorption.⁴⁶ We use the two-stream radiative solver in the infrared range (4 to $100\mu \mathrm{m}$).⁴⁷

We estimate the cloud microphysical properties based on previous studies that reported in-situ measurements of optically thin midlevel clouds.^20,21 These studies reported thin altocumulus clouds with small water droplets not exceeding $20\mu \mathrm{m}$ in size and thin altostratus clouds consisting solely of ice particles (columns with rounded edges) with maximum size up to $100\mu \mathrm{m}$. Further, the measured liquid/ice water content for these clouds was reported to have an average value of $0.01\mathrm{g}{\mathrm{m}}^{-3}$ with a maximum value of $0.45\mathrm{g}{\mathrm{m}}^{-3}$. With this information, we have chosen effective radii of 30 and $60\mu \mathrm{m}$ for ice clouds and 10 and $20\mu \mathrm{m}$ for water clouds, respectively. For the case of mixed phase clouds, we choose clouds composed of $20\text{-}\mu \mathrm{m}$ water drops and $60\text{-}\mu \mathrm{m}$ ice columns and then consider four cases where the ice fraction of the total optical depth is 20%, 40%, 60%, and 80%. We have performed radiative transfer simulations using these eight scenarios of microphysical properties and the cloud altitudes and optical depths derived from CALIPSO data.

4.2.

Results of Case Studies

We find five large M1 clouds that each extended over 500 km. These five clouds are all found in the Arctic and we tabulate their locations, solar zenith angle, and optical depth in Table 4. Of these five clouds, four are observed when the sun was below the horizon and one, case 5, is observed when the sun was just above the horizon. We cite the median optical depth rather than the mean value, as the median is more representative of the whole cloud. The mean values are higher than the median values and reflect the influence of a few high values. We tabulate the range and average of the optical depth, cloud top altitude, cloud base altitude, cloud top temperature, and cloud base temperature in Table 5. The median value of the optical depth is presented in parenthesis next to the average value. The synoptic meteorological conditions for these days show that these clouds are formed along the frontal zones with weakly ascending air masses.⁴⁸

Table 4

Distinct large-scale M1 clouds identified by CALIPSO.

Table 5

Characteristics of Arctic M1 clouds.

For each of these five cases, we calculate the cloud radiative forcing (CRF) as the difference between the net downward irradiance under cloudy and clear sky conditions. A positive value of the CRF indicates that the cloud has a warming effect on the surface. For example in case 1, the clear sky has a net longwave (LW) irradiance of $-55\mathrm{W}/{\mathrm{m}}^2$ indicating cooling. In case 1, with the M1 cloud present, the net irradiance is $-12\pm 1\mathrm{W}/{\mathrm{m}}^2$. The value of $-12\mathrm{W}/{\mathrm{m}}^2$ with an uncertainty of $1\mathrm{W}/{\mathrm{m}}^2$ is the average and standard deviation in the values determined from the eight cloud scenarios. The small standard deviation reflects the fact that the different microphysical characteristics have little effect on the radiative properties of these thin clouds. Thus, in case 1, we estimate that the M1 has a CRF of $43\pm 1\mathrm{W}/{\mathrm{m}}^2$ and a net warming effect on the surface. We tabulate our results for all five case studies in Table 6. We estimate the LW radiative forcing for all cases and the shortwave (SW) forcing for case 5, where the cloud is illuminated by the sun. We find that the LW CRF is positive in all cases with values ranging between 28 and $48\mathrm{W}/{\mathrm{m}}^2$, with a mean value of $36\mathrm{W}/{\mathrm{m}}^2$. We conclude that M1 cloud has a net warming effect in the LW. In case 5, we see that the cloud has an SW CRF of $-7\mathrm{W}/{\mathrm{m}}^2$ indicating that the cloud has a net cooling effect in the SW. However, the combined SW and LW effect of the cloud is to yield a CRF of $23\mathrm{W}/{\mathrm{m}}^2$, indicating a net warming. To further explore the balance of SW and LW radiative forcing, we conducted the CRF when changing the solar zenith angle from 85 deg to 80 deg and 70 deg. While the LW warming effect of the cloud remains constant, the SW cooling effect increases and the total CRF reduces to $10\mathrm{W}/{\mathrm{m}}^2$ at 80 deg and to $-6\mathrm{W}/{\mathrm{m}}^2$ at 70 deg.

Table 6

Radiative impact of M1 clouds.