SEARCH

SEARCH BY CITATION

Keywords:

  • cloud;
  • lidar;
  • CALIPSO

Abstract

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. Particle Type Discrimination From CALIOP Data
  5. 3. Results and Discussions
  6. 4. Conclusions
  7. Acknowledgments
  8. References
  9. Supporting Information

[1] A method for discriminating cloud particle types was developed using lidar backscattering copolarization and cross-polarization channel measurements from Cloud-Aerosol Lidar With Orthogonal Polarization (CALIOP) on board Cloud-Aerosol Lidar and Infrared Pathfinder Satellite Observation (CALIPSO). In spaceborne lidar measurements, significant multiple scattering effects discriminate between cloud water and ice difficult using the depolarization ratio (δ). We theoretically estimated the relationship between δ and cloud extinction on the basis of the backward Monte Carlo method. Cloud particle type was determined by the combined use of δ and the ratio of attenuated backscattering coefficients for two vertically consecutive layers. Ice particles were further classified into two categories: randomly oriented ice crystals (3-D ice) and horizontally oriented plates (2-D plate). The method was applied to CALIOP data for September–November 2006. We found that 3-D ice generally occurred colder than −20°C, whereas 2-D plate occurred between −10°C and −20°C, with high-occurrence frequency in high-latitude regions. We compared the results to those obtained using the vertical feature mask (VFM). The VFM tended to show a homogeneous cloud type through the entire cloud layer in vertical directions and misclassified 2-D plate as water. The ratio of water particles relative to ice particles decreased with decreasing temperature. By the proposed method, water cloud occurrence in subtropical and high-latitude regions was greater (up to 20%) than in the other regions below −10°C; however, the VFM results did not show such dependence on latitude. Comparison of ice and water cloud between our results and Moderate Resolution Imaging Spectroradiometer (MODIS) products showed better agreement for water cloud than for ice cloud.

1. Introduction

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. Particle Type Discrimination From CALIOP Data
  5. 3. Results and Discussions
  6. 4. Conclusions
  7. Acknowledgments
  8. References
  9. Supporting Information

[2] Information on cloud particles, such as their phase, shape, and orientation, is crucial for understanding cloud-generation mechanisms and estimating the global radiation budget distribution as affected by clouds. Discrimination of cloud particle type is a necessary prerequisite for retrieving cloud microphysics from remotely sensed data. Active remote sensors such as lidar and cloud-profiling radar enable us to obtain the cloud vertical structure. Thus, spaceborne active sensors are expected to provide information on the global distribution of three-dimensional cloud particle types.

[3] The polarization capability of lidar provides the depolarization ratio, defined as a ratio of backscatters in the planes of polarization perpendicular and parallel to the linearly polarized incident beam. The depolarization ratio (δ) has been used to infer ice particle habits (see Takano and Liou [1989] for visible data and Okamoto [2002] and Sato and Okamoto [2006] for 95 GHz cloud radar). Furthermore, use of δ is the most established way to infer the cloud phase and the nature of ice crystals from ground-based lidar measurements [Sassen, 1991]. Noel et al. [2004] developed a technique to classify ice particles into different categories on the basis of δ and compared the results to in situ observations. Meanwhile, Hogan et al. [2003] presented an algorithm that uses integrated backscatter for a layer to identify liquid water clouds. The algorithm was applied to data obtained by the Lidar In-Space Technology Experiment (LITE) that flew on the Space Shuttle Discovery in September 1994 [Hogan et al., 2004]. Latitudinal supercooled water occurrence and its temperature dependence were derived. However, LITE was operational for only 53 h in a latitude range limited within ±60°. Much longer statistics around the globe can be derived from the Cloud-Aerosol Lidar and Infrared Pathfinder Satellite Observations (CALIPSO) satellite mission that has been operating since June 2006. CALIPSO lidar can measure depolarization, which is expected to be useful for inferring cloud particle phases. However, water clouds tend to exhibit significant δ owing to multiple scattering effects [Hu et al., 2001], which may create difficulty when discriminating the cloud phase by δ alone. The ice water algorithm (IWA) for the CALIPSO Lidar Level 2 Vertical Feature Mask (VFM) cloud phase product primarily uses integrated δ over a cloud layer [Liu et al., 2005]. Layer integration, however, reduces the vertical resolution of the phase product relative to the original resolutions. For near nadir-pointing lidar observation, horizontally oriented plates (the 2-D plate type) produce δ values close to zero, and 2-D plate can be misclassified as water particles by the VFM method. Hu [2007] introduced a new technique involving the layer-integrated depolarization ratio and effective lidar ratio. The method discriminated cloud phase and horizontally oriented particles. However, some problems remained in the vertical resolution of phase discrimination because the method used the layer-integrated values.

[4] The newly developed method presented in this paper enables discrimination of vertically resolved cloud particle type and requires no layer integration of CALIPSO lidar data. In addition to δ, we introduce a new parameter estimated from two attenuated backscattering coefficients in neighboring layers. This parameter enables discrimination between water and ice particles. Ice is further categorized into randomly oriented ice crystals (3-D ice) and 2-D plate.

[5] In section 2, we describe the CALIPSO lidar data and cloud mask scheme. The method to discriminate cloud particle type is also described with theoretical estimation of δ in relation to the extinction coefficient for CALIPSO lidar observations. Section 3 presents case studies of the particle type discrimination; we also present global statistics for the discrimination results and compare our results with VFM results. Section 4 presents a comparison with the MODIS cloud phase product. Finally, section 5 gives conclusions.

2. Particle Type Discrimination From CALIOP Data

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. Particle Type Discrimination From CALIOP Data
  5. 3. Results and Discussions
  6. 4. Conclusions
  7. Acknowledgments
  8. References
  9. Supporting Information

2.1. Data and Cloud Detection

[6] The Cloud Aerosol Lidar with Orthogonal Polarization (CALIOP) instrument is a spaceborne lidar carried by the CALIPSO satellite, which has been operating since June 2006. CALIOP pointed in a near-nadir direction (∼0.3°) before 28 November 2007. Since then, to avoid specular reflection, the instrument has been pointed 3° off-nadir. CALIOP observational data include the total attenuated backscattering coefficients for the 532 nm (β532) and 1064 nm (β1064) channels, and the attenuated backscattering coefficient for the 532 nm perpendicular channel (β532,⊥) as a function of altitude, longitude, and latitude. β532 is composed of parallel and perpendicular components (β532 = β532,∥ + β532,⊥). Therefore, δ is calculated for the 532 nm channel. The depolarization ratio is defined as δ = β532,⊥/β532,∥. The lidar has fundamental sampling resolutions of 30 m in the vertical and 333 m in the horizontal. CALIPSO, CloudSat, Aqua, and other instruments are part of a compound satellite observation system called A-train [Stephens et al., 2002]. CloudSat carries the nadir-pointing 94 GHz Cloud Profiling Radar (CPR), and Aqua carries the Moderate Resolution Imaging Spectroradiometer (MODIS), a 36 band spectroradiometer.

[7] We modified cloud mask schemes originally developed for the ground-based radar and lidar systems by Okamoto et al. [2007] and applied them to CloudSat and CALIOP data. This study uses only the cloud mask results for CALIOP. The scheme has two main steps. First, masking of the candidate cloud is conducted using a threshold backscattering coefficient at 532 nm. The criterion was developed for the ground-based lidar system described by Okamoto et al. [2007] for the midlatitudes and was further tested in the tropical western Pacific by Okamoto et al. [2008]. The threshold depends on the altitude, the molecular signal derived from European Centre for Medium-Range Weather Forecasts (ECMWF) data, and the background noise signal estimated from signals at altitudes of 19–20 km. The second step is a spatial continuity test. For each candidate cloud pixel, the surrounding 5 × 5 pixel box is considered. If more than half of the 25 pixels in the box exceed the criterion from step 1, the central candidate cloud pixel is set to 1 (i.e., cloud); otherwise, the pixel is set to 0 (i.e., clear-sky pixel). The cloud mask results for the original resolution are then averaged vertically over 240 m and horizontally over 1.1 km in order to have the same resolutions as CloudSat. Consequently, the averaged cloud mask values for the CloudSat grid can take values between 0 and 1. To avoid spurious signals caused by noise, we applied a further threshold: if the average mask value exceeded 0.5, we considered the pixel to be cloud.

[8] It should be noted that the official CALIPSO Lidar Level 2 VFM cloud mask product [Vaughan et al., 2005] occasionally made a false detection through the scheme's horizontal averaging procedure, and also seemed to misclassify noise or aerosols as cloud (and vice versa), as noted by Marchand et al. [2008] and Holz et al. [2008]. Hagihara et al. [2010] also revealed that our cloud mask results have less contamination by noise or aerosol signals compared to the VFM. We also compared the zonal mean cloud coverage for topmost layer detected by sensors by our CALIOP scheme, the VFM, our combined CloudSat/CALIOP scheme, and collocated Aqua MODIS results [Menzel et al., 2008]. The CALIOP, CloudSat/CALIOP, and MODIS results were similar for total cloud coverage, but the VFM result was different. Because of possible misclassification at low levels, the VFM showed the largest cloud coverage in the middle and low latitudes.

2.2. Method of Cloud Particle Type Discrimination

[9] We used the backscattering coefficient and depolarization ratio at the 532 nm wavelength measured by CALIOP for cloud type discrimination. When a single-scattering process is dominant, spherical particles in water clouds do not show δ > 0. In contrast, the randomly oriented ice crystals in ice clouds produce large δ (comparable to or larger than 40%). Therefore, δ has been used for cloud phase discrimination in cases of ground-based polarization lidar measurements [Sassen, 1991]. However satellite-borne lidar instruments, such as CALIOP, observe much larger footprints (e.g., 90 m) on Earth's surface than does ground-based lidar (generally less than 10 m). Similar discrimination is thus not straightforward for satellite observations because of large δ values caused by multiple scattering in water clouds.

[10] To estimate the multiple scattering effects on δ quantitatively in water clouds, we adopted a backward Monte Carlo (MC) method developed by Ishimoto and Masuda [2002]. The MC method was developed to solve the Stokes vector of multiple scattered light for an optional scattering medium with arbitrary boundaries. This method enables estimation of δ and the backscattering coefficient at 532 nm as a function of the penetration depth for a given extinction coefficient. CALIOP has a 0.13 mrad field of view (FOV) and orbits Earth at a distance of ∼700 km above the ground. The scattering matrix is calculated from Mie theory. We assumed a plane-parallel homogeneous cloud model for which the microphysics was given by the water cloud model introduced by Hansen [1971] and Deirmendjian [1975] and the cloud geometrical thickness was assumed to be 0.6 km. The simulated depolarization ratio δ for CALIOP data was plotted for given cloud extinction coefficients for the 532 nm wavelength (σ532) as a function of the penetration depth (Figure 1), using 10 m bin ranges for the calculation. Increases in the penetration depth and extinction coefficient lead to larger δ, which can reach 100% when σ532 is larger than 10 (/km). A lidar beam might be completely attenuated in a cloud of τ > 3, where τ is the optical thickness for the lidar wavelength. Therefore, in actual CALIOP observations, the observable penetration depth is less than 0.15 km for σ532 = 20/km and less than 0.30 km for σ532 = 10/km. Within the observable penetration depths, we found that the value of δ for water clouds was sometimes larger than 40% and comparable to or larger than values for randomly oriented ice crystals. Therefore, we can conclude that there is a fundamental difficulty in discriminating water from ice when we only use the measured δ value from CALIOP.

image

Figure 1. Simulated depolarization ratio δ of water clouds by the backward Monte Carlo method for CALIOP at the 532 nm wavelength. Plane-parallel cloud with geometric thickness of 600 m was considered. The distance between the CALIPSO lidar and the cloud was assumed to be 700 km. The field of view was set to 0.13 mrad, and σ532 values were set at 1.0, 2.5, 5.0, 10.0, and 20.0 per kilometer.

Download figure to PowerPoint

[11] For ground-based lidar observations, we performed similar theoretical estimations using the MC method; for example, we assumed the same extinction coefficient and penetration depth as in the CALIOP simulations but the FOV was set to 1 mrad and the distance between the cloud and lidar was 4 km. We found that the depolarization ratio rarely exceeded 10% even for optically thick clouds (not shown).

[12] To overcome the above problem, we introduced an additional parameter for the cloud particle type. According to Okamoto et al. [2003], the attenuated backscattering coefficient with finite geometrical thickness for the 532 nm wavelength of layer i can be written as

  • equation image

where β532 is the backscattering coefficient for the 532 nm wavelength, ΔR is the vertical resolution (here equal to 240 m), Ri denotes the distance to the center of layer I, and Ri−1/2 is the top boundary of layer I viewed from the satellite. The geometrical thickness of a layer is 240 m. The cloud microphysics of the layer was assumed to be homogeneous. Equation (1) is a discrete expression corresponding to the finite thickness of lidar observations. The ratio of the attenuated backscattering coefficient of layer i to that of the next layer (i + 1) is expressed as

  • equation image

Assuming extinction coefficient and backscattering coefficient are the same in the successive two layers; that is, the same size distribution and number concentration in the two layers,

  • equation image

we then obtain

  • equation image

where Δτ532 is the optical thickness of the layer for the wavelength of 532 nm. Thus, the ratio β532(Ri)/β532(Ri+1) might correspond to the optical thickness of layer i. Because water clouds generally have larger extinction coefficients compared to ice clouds at 532 nm, the ratio is expected to be effective for cloud phase discrimination. Note that there is an ambiguity in the discrimination of the particle type in a boundary of ice and water clouds. Here we use the common logarithmic expression x as follows:

  • equation image

in which x(Ri) is proportional to Δτ532 when the two neighboring vertical layers are homogeneous. Note that we do not assume homogeneity in the two successive layers in the following analyses and we do not use equations (3) and (4) in the analyses but equation (5) is used.

[13] We estimated two-dimensional frequency distributions of x and δ for all cloud observations for three different temperature (T) categories bounded at 0, −20, and −40°C during October 2006 (Figure 2). The ECMWF numerical atmospheric model was used for T and interpolated to the CALIOP grid box. Water cloud occurrence was expected to be dominant for the warm temperature category (T > 0°C) (Figure 2a). The results showed a positive correlation between x and δ, with δ reaching up to 40% when x was about 1.5. High δ values seemed to be produced by the multiple-scattering process, as expected from the theoretical estimation in section 2.1. In Figure 2a, the white circles with solid lines indicate water cloud values simulated by the MC method. Different lines correspond to different extinction coefficients. In estimating these lines, simulated δ values were averaged over 240 m to fit the vertical resolution of CALIOP and are given as a function of the penetration depth (Figure 1). Here, the cloud-top layer with 240 m thickness was defined as the first layer. Each extinction coefficient was converted to x by the following equation, derived from equations (4) and (5):

  • equation image

Thus, simulated δ was plotted as a function of x for the first, second, and third cloud layers. Observed values were within the range of the simulated results. Thus, we concluded that the warm temperature category actually represents water clouds. A large portion of the clouds in this temperature range had values of x > 0.5 and δ > 10%. Negative x was found for a small fraction of the data (Figure 2a). This might be explained by the occurrence of vertical inhomogeneity of the cloud microphysics in the neighboring layers.

image

Figure 2. Two-dimensional frequency distributions of clouds in terms of the ratio of attenuated backscattering coefficients for successive layers, x, and δ for all temperature ranges in October 2006: (a) only for T ≥ 0°C with simulated results by the Monte Carlo method (white solid lines), (b) only for −20°C ≤ T < 0°C, and (c) only for T < −20°C.

Download figure to PowerPoint

[14] When the temperature ranged from −20 to 0°C, in addition to the water cloud signals with δ > 10%, very low δ close to zero also appeared frequently (Figure 2b). Ice clouds containing horizontally oriented plates exhibit low δ and high backscatter owing to specular reflection. Thus, in this temperature range, we inferred that ice clouds containing horizontally oriented plates would occur frequently. Sassen and Benson [2001] also reported that specular reflection occurred on horizontally oriented plates observed by zenith-pointing ground-based lidar in this temperature range. In their study, the minimum value of δ was located at −17.5°C in the observation of midlatitude ice clouds. Note that CALIOP pointed along track at an angle of 0.3° from true nadir in this observation period, and plate-like particles might be better observed in this situation than in the case of off-nadir pointing observations.

[15] The frequency of water clouds might be small when the temperature is below −20°C (Figure 2c). Indeed, high values of δ (about 30%) were dominant. Such high δ values are a characteristic of ice clouds composed of randomly oriented particles (3-D ice).

[16] These analyses suggest that it might be possible to relate the lidar-observed values (i.e., x and δ) and cloud particle types. Here, we considered the following cloud particle types: warm water, supercooled water, 3-D ice, 2-D plate, “unknown1,” and “unknown2” (Table 1). An initial discrimination was made on the basis of the temperature and empirically derived lines in the xδ plane (Figure 3). Each line corresponds to one estimated by the MC simulation and also derived from the analyses of the observations described above.

image

Figure 3. Cloud particle diagram on the x and δ planes. The colors denote occurrence frequencies from observations during October 2006 at all temperatures.

Download figure to PowerPoint

Table 1. Cloud Particle Types
TypeDescription
Warm waterliquid droplets with T ≥ 0°C
Supercooled waterliquid droplets with T < 0°C
3-D icerandomly oriented ice crystals
2-D plateice crystals containing horizontally oriented plates with specular reflection
Unknown1likely ice crystals containing horizontally oriented plates with weak specular reflection
Unknown2liquid droplets or randomly oriented ice crystals

[17] In Figure 3, the frequency of occurrence was also plotted from observations in October 2006 at all temperatures. Cloud particles were assumed to be warm water for T ≥ 5°C, regardless of the values of x and δ. This temperature criterion (T ≥ 5°C) was selected because falling ice particles might exist at T < 0°C but do not exist for T ≥ 5°C [Liu, 2008].

[18] The cloud particle type was discriminated on the basis of the observed x and δ when T < 5°C. Each area in the xδ plane corresponds to each particle type (Figure 3). For warm water and supercooled water (later referred to simply as “water”), x was assumed to be larger than 0.5, corresponding to Δτ532 of 0.58 (equation (6)). When x < 0.5 and δ > 10%, 3-D ice was selected. Ray tracing simulations of 3-D ice shows that δ is around 30% or larger [Takano and Liou, 1995] and 2-D plate does not produce any depolarization. When δ was smaller than 3%, the grid box was assumed to contain 2-D plate. Considering the mixture of 3-D ice and 2-D plate, δ > 10% indicates the dominance of 3-D ice particles and δ < 3% indicates 2-D plate is dominant [Sassen and Benson, 2001]. When δ > 10%, the separation of 3-D ice from water is done by using δ < 60x2 + 10 [%]. This criterion is determined on the basis of the results of Monte Carlo simulation (as in Figure 2a and also the observational results for T > 5°C). Unknown1 had δ between 3 and 10%, although most of the signals in this area contained 3-D ice crystals with horizontally oriented plates. Unknown2 was assigned in a relatively small area in the xδ plane where liquid droplets or randomly oriented ice crystals might exist. The discrimination of water from unknown1 for x > 0.5 and discrimination between unknown1 and unknown2 are performed by using δ = 7.5 exp{−4.0(x − 0.2)2} + 2.5 for x > 0.5 in order to connect the δ = 10% for 3-D ice and δ = 3% for 2-D plate.

[19] After the initial discrimination was performed, the following second classification (spatial consistency test) was applied. For each discriminated pixel, the surrounding 3 × 5 pixels (i.e., 3 pixels in the vertical direction and 5 pixels in the horizontal direction) were considered. First, the major particle type was estimated from the results for these 15 pixels. Then, if the particle type of the central pixel differed from the major particle type, the central pixel was forced to be the major type. The spatial consistency test also reduced uncertainties owing to the fixed thresholds introduced here. It is worth noting that most of the unknown2 pixels selected by the first classification were actually changed to water or 3-D ice on the basis of the consistency tests of surrounding pixels; only a minor number of unknown2 particles remained.

3. Results and Discussions

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. Particle Type Discrimination From CALIOP Data
  5. 3. Results and Discussions
  6. 4. Conclusions
  7. Acknowledgments
  8. References
  9. Supporting Information

3.1. Case Study

[20] We first examined cloud particle types from CALIOP observations taken on 8 October 2006 in the area 33°S–58°S, 114°W–123°W over ocean (Figure 4). We also examined the CloudSat radar reflectivity, CALIOP backscattering coefficient at 532 nm, depolarization ratio, the resulting cloud particle type by our method (referred to as TU), and the VFM cloud phase product. The VFM cloud phase is derived from the ice water algorithm (IWA) [Liu et al., 2005] and distributed by the CALIPSO team based at the National Aeronautics and Space Administration (NASA), Langley.

image

Figure 4. Case study example of (a) a latitude-height cross section of CloudSat radar reflectivity from 8 October 2006 at 33°S–58°S, 114°W–123°W; (b) the CALIPSO lidar backscattering coefficient at 532 nm; (c) the depolarization ratio at 532 nm; (d) the cloud particle type determined by the method presented in this study (TU method); and (e) cloud phases from the VFM product.

Download figure to PowerPoint

[21] The VFM product generally showed more cloudy pixels than were found in our product (Figures 4d and 4e). The differences reflect differences in the cloud mask schemes [Okamoto et al., 2007, 2008; Hagihara et al., 2010]. Temperature from the ECMWF model was also overlaid in Figures 4d and 4e. In Figures 4d and 4e, CloudSat and CALIOP detected southern midlatitude storm tracks. These were typical storm tracks in that latitude zone and are known to be associated with extratropical cyclones. CloudSat CPR detected clouds with precipitation having top heights of ∼10 km and horizontal scales larger than 2000 km (Figure 4a). Owing to attenuation, the CALIOP signal did not reach the surface or the cloud base detected by CloudSat (Figure 4b). However, CALIOP detected more optically thin clouds than did CloudSat around 30°S above 8 km, as shown in Figure 4b.

[22] We found frequent occurrence of specular reflections characterized by large backscattering coefficients close to zero δ around the center of the convection zone, at altitudes above the melting layer around 2 km as observed by CloudSat and below 5 km (e.g., at 53, 46, 44, 42, and 35°S; see Figures 4b and 4c). Cloud particles in this area were the 2-D plate type according to our classification method (Figure 4d).

[23] The upper parts of the clouds were mostly 3-D ice, with unknown1 found at altitudes between those of 2-D plate and 3-D ice. Unknown1 might have consisted of ice crystals that exhibited intermediate δ between values for 2-D plate and 3-D ice.

[24] By our method, we could infer the vertically resolved particle type inside the same cloud layer. In contrast, the VFM cloud phase results present a vertically homogeneous type; for example, all liquid water particles between 3 and 10 km at 46°S. This was because the IWA primarily uses a layer-integrated depolarization ratio defined as

  • equation image

which assigns one cloud phase for an estimated cloud layer. Because of the integration, this method seemed to misclassify 2-D plate and part of 3-D ice as water associated with very low δ owing to specular reflections. Consequently, the VFM tended to overestimate low-level cloud occurrences. In contrast to the VFM method, our method could provide vertically resolved cloud types.

[25] The altitude of 2-D plate occurrences corresponded to temperatures between −10 and −20°C. Supercooled water was found below −10°C, with maximum occurrence located at around 35°S. Significant attenuation of the lidar backscatter was observed when supercooled water was identified by the TU method, suggesting stratiform cloud. Although the VFM method tended to show greater detection near the bottom edges of stratiform clouds than the TU method did, overall agreement was obtained for supercooled water.

3.2. Latitudinal Statistics for September–November 2006

[26] We examined the zonal mean properties of cloud particle type occurrence from global analyses of CALIOP data from September to November 2006 (SON06). CALIOP had been pointing at 3° off nadir during 6–15 November 2006, but the change in its pointing direction had only very minor effects on the zonal mean results for the 3 months.

[27] Figure 5 shows the latitudinal distributions of the cloud fractions for each particle type during SON06 as a function of altitude (Figures 5a, 5c, and 5e) and temperature (Figures 5b, 5d, and 5f). The latitudinal resolution was 2.0°, and the altitude and temperature intervals were 240 m and 2.0°C, respectively. The major part of water in water clouds (warm water and supercooled water) was observed in low-altitude ranges below 3 km at any latitude (Figure 5a). The maximum altitude at which water could be observed depended on the latitude; that is, the altitude became higher as the latitude decreased, so that altitude was ∼10 km at the equator and 5 km at 60° in both hemispheres. At any latitude, the water cloud fraction became close to zero (<2%) below −40°C (Figure 5b).

image

Figure 5. Latitudinal distributions of the cloud fraction for each particle type from September to November 2006. The vertical resolution is 240 m or 2°C, and the horizontal resolution is 2° latitude. (a) Water as a function of cloud altitude. (b) The same as Figure 5a but for temperature. (c) The same as Figure 5a but for 3-D ice. (d) The same as Figure 5c but for temperature. (e) The same as Figure 5a but for 2-D plate. (f) The same as Figure 5e but for temperature.

Download figure to PowerPoint

[28] A strong peak of the cloud fraction of 3-D ice was found at around 14 km near the equator, with a value larger than 0.25 (Figure 5c). These clouds represented the upper part of deep convective clouds and cirrus clouds associated with a tropical weather system. Beyond the subtropical high, midlatitude storm-track regions were found in both hemispheres, and the 3-D ice cloud fraction was ∼0.15 (Figure 5c).

[29] From the analysis of the relationship between temperature and cloud particle types, we found that most 3-D ice clouds occurred at temperatures colder than −20°C (Figure 5d). There are two maxima of 3-D ice fraction; for example, one at −60°C and the other at about −85°C near the Equator. It is found that the upper part of the 3-D ice near 15 km, shown in Figure 4c, actually belongs to very cold temperature (T < −80°C) (Figure 5d).

[30] The results clearly show that 2-D plate cloud occurrence was limited to −20°C < T < −10°C (Figure 5f), as seen in the midlatitude case study illustrated in Figure 4. The 2-D plate type occurred in most of the latitudes, and the cloud fraction increased at high latitudes of >50° in both hemispheres.

[31] We also examined the occurrence ratio of each particle type relative to the total cloud occurrence and its dependence on latitude during SON06 (Figure 6). In addition to the analyses for water, 3-D ice, and 2-D plate shown in Figures 6a, 6c, and 6e, respectively, for comparison purposes, we also present the occurrence ratios for water and ice particles determined by the VFM in Figures 6b and 6e, respectively. Latitude and temperature intervals were 2° and 2°C, respectively. As expected, the ratio of water cloud occurrence decreased with decreasing temperature (Figure 6a). However, somewhat surprisingly, the ratio of water occurrence varied with latitude symmetrically on both sides of the equator. In subtropical (20–30°) and high-latitude (60°) areas of both hemispheres, the ratio of water was often close to 0.2, which is larger than in other latitude ranges, at temperatures below −10°C.

image

Figure 6. The latitude-altitude distribution of the occurrence ratios of each particle type from September to November 2006. The vertical resolution is 2°C, and horizontal resolution is 2° latitude. (a) Water by the TU method, (b) water from the VFM product, (c) 3-D ice by TU, (d) ice from VFM, and (e) 2-D plate by TU.

Download figure to PowerPoint

[32] In contrast, the VFM results showed nearly uniform temperature dependence on latitude. This finding can be explained as follows: the IWA of the VFM used the cloud-top temperature to determine the probability of water (and also ice) occurrence [Liu et al., 2005]. Thus, the VFM results showed uniform temperature dependence. In addition, the VFM tended to misclassify 2-D plate and a part of 3-D ice as water, as seen in Figure 4. Therefore, the ratio of water clouds by the TU method was also smaller than that for the VFM.

[33] The temperature at which the water cloud occurrence ratio reached 0.5 by the TU method was ∼−10°C, whereas the same cloud occurrence ratio was found in a colder environment (i.e., at around −25°C) in the VFM result. At T < −35°C, 3-D ice and ice dominated in cloud occurrences by the TU and VFM methods, respectively (Figures 6c and 6d).

[34] Contrary to the water and 3-D ice cases, the occurrence ratio of 2-D plate clouds by the TU method showed nearly uniform temperature dependence on latitude. The ratio became larger than 10% between −20°C < T < −10°C and almost zero in other temperature ranges. Occurrences of plate-like ice crystals in this temperature range were also reported from sampling in natural clouds by Ono [1970] and from laboratory experiments on ice crystal growth such as the study by Kumai [1982]. From zenith-pointing ground-based lidar observations, Hogan et al. [2003] also reported the occurrence of specular reflection from horizontally oriented plates in these temperature ranges.

[35] Figure 7 presents the zonal mean properties of the ratio of each cloud particle type determined by the TU method for four latitude categories bounded at 15°, 35°, and 65°, mostly in the Northern Hemisphere, as a function of temperature with intervals of 2°C. That is, the latitude categories were high (65°N), middle (35°N–65°N), subtropical (15°N–35°N), and tropical (15°S–15°N) (Figure 7). Figure 7 also indicates the cloud fractions for all cloud particle types for each latitude category.

image

Figure 7. Zonal mean properties of the occurrence ratios of each particle type in four latitude bands from 15°S to 90°N; bounded at 15, 35, and 65°N; as a function of temperature with intervals of 2°C.

Download figure to PowerPoint

[36] The cloud fraction between 0°C and −40°C was larger in the middle- and high-latitude regions than in the tropics and subtropics. The ratio of water cloud occurrence in the high-latitude category was larger than those in other regions, as illustrated in Figure 6a. In the subtropical region, this tendency was not clear and showed latitudinal broadening. Except for this tendency, the temperature dependences of each type in the latitude categories were similar to each other; for example, in all latitude categories, 50% water occurrence was observed at about −10°C.

[37] Only very small fraction of water clouds was identified at T < −40°C. The rates of water cloud occurrence between −40°C and −42°C for each latitude categories are 0.014 in high, 0.014 in middle, 0.017 in subtropical and 0.013 in tropical regions, respectively. The 2-D plate occurrence ratio exhibited its maximum (0.5) between −12°C and −14°C in any latitude category.

[38] The cloud fractions of total clouds were found to increase in association with 2-D plate occurrences between −10°C and −20°C. Around this temperature range, a change in the gradient of water clouds was also found. Unknown1 occurrence was associated with 2-D plate occurrence and increased at the lower temperatures of the 2-D plate peak. Thus, most unknown1 consisted of ice crystals, with only a small part being water particles. As a result of the consistency test step, the fraction of unknown2 was close to zero at any temperature.

3.3. Comparison With the MODIS Global Distribution

[39] We also compared the particle type occurrence found for CALIOP with MODIS water and ice cloud data. Monthly averaged global distributions of water and ice cloud by Terra/MODIS were available from monthly level 3 data products (MYD08_M3). The MODIS cloud phase product is produced by a combination of two methods, one using the brightness temperature difference of two thermal infrared bands, and the other using the ratio for the near-infrared and visible bands [Chylek et al., 2006]. We compared the global distribution of averaged daytime coverage by water and ice clouds between CALIOP (by the TU method) and MODIS for the SON06 period (Figure 8). The grid boxes had 5.0° latitude by 5.0° longitude resolution. Note that ice cloud cover for CALIOP was calculated as the sum of 3-D ice and 2-D plate cover (Figure 8c).

image

Figure 8. Global distribution of averaged daytime cloud coverage for water and ice clouds from CALIPSO and MODIS from September to November 2006. The resolution is 2.0° latitude by 2.0° longitude. (a) Cloud cover by water clouds from CALIPSO (TU). (b) The same as Figure 8a but from MODIS. (c) Cloud cover of ice clouds from CALIPSO (TU). (d) The same as Figure 8c but from MODIS.

Download figure to PowerPoint

[40] Both CALIOP and MODIS revealed large coverage by water clouds over the west coast of California, Peru, and Namibia, where low-level clouds are known to occur (Figure 8a for CALIOP and Figure 8b for MODIS). The global distribution of ice clouds was characterized by occurrence in the Intertropical Convergence Zone (ITCZ) (Figures 8c and 8d). Both water and ice clouds were frequently observed at latitudes above 30° in both hemispheres. Although these features were also seen in the MODIS results (Figures 8b and 8d), MODIS tended to show smaller cloud cover compared to CALIOP, and the tendency was especially significant for ice clouds.

[41] Figure 9 shows a one-to-one comparison of CALIOP and MODIS results for water and ice cloud coverage. The correlation coefficient for ice was smaller than that for water (0.917 for water, 0.764 for ice). The difference in the correlation coefficients might be due to differences in cloud detection sensitivity between CALIOP and MODIS. Lidar can detect optically thin cloud such as subvisible cirrus that is often difficult to detect by MODIS. Ice clouds generally have smaller optical thickness compared to water clouds. Thus, the difference in cloud detectability between CALIOP and MODIS was larger for ice clouds. In spite of bias and sensitivity differences, similar global patterns for ice and water cloud distribution were obtained from CALIOP and MODIS data.

image

Figure 9. One-to-one comparisons of cloud cover from CALIPSO and from MODIS. The data are the same as in Figure 8. (a) Water cloud. (b) Ice cloud.

Download figure to PowerPoint

4. Conclusions

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. Particle Type Discrimination From CALIOP Data
  5. 3. Results and Discussions
  6. 4. Conclusions
  7. Acknowledgments
  8. References
  9. Supporting Information

[42] We have developed a new method to discriminate the cloud particle type, using backscatter and depolarization measurements by CALIOP lidar. Unlike the currently available IWA scheme used for the VFM product, our method enables discrimination of vertically resolved cloud particle types. A unique feature of this method is its use of a ratio of attenuated backscattering coefficients for two vertically consecutive cloud layers (x), in addition to the depolarization ratio δ. On the basis of criteria for xδ relationships estimated from statistics for observations and theoretical simulations by the MC method, our method enabled discrimination of cloud particle types: warm and supercooled water, 3-D ice, and 2-D plate particle types.

[43] The IWA used in the VFM cloud phase product did not provide a vertically resolved product and also tended to misclassify 2-D plate and part of 3-D ice as water owing to specular reflections and layer integration. The misclassification possibly contributed to overestimation of water cloud occurrence in the VFM cloud phase product. Our method overcomes these problems. The cloud mask results were also larger in the VFM product compared with our product owing to noise and aerosol contamination in low-level clouds, as shown by Hagihara et al. [2010]; these larger values led to the overestimation of water cloud occurrence than our results.

[44] Global analyses were carried out by applying our method to CALIOP data. Our results indicate the following. Water cloud occurrence decreased as temperature decreased at any latitude, and almost no water clouds were observed below −40°C. In subtropical and high-latitude regions of both hemispheres, the ratio of water cloud occurrence to total cloud occurrence was 0.2, larger than that in other latitude regions at temperatures below −10°C. In contrast, the VFM cloud phase product did not show this feature.

[45] The 2-D plate type occurred at temperatures between −10°C and −20°C and was more frequently observed in high latitudes. The temperature dependence of the 2-D plate occurrence ratio did not differ in latitude; the maximum fraction of 0.5 was found between −12°C and −14°C.

[46] The 3-D ice type tended to occur at temperatures below −20°C. Water and ice cloud cover were compared with those from the MODIS cloud phase product, revealing similar global distributions of water and ice cloud coverage. CALIOP cloud coverage was larger, especially for ice clouds, because of its high sensitivity to optically thin clouds. Note that our results were not validated and we need more comparison study with other independent data sources.

[47] Understanding the global occurrence and temperature dependence of each type of cloud particle is crucial for understanding cloud generation mechanisms and cloud radiative properties. Using the information on cloud particle types presented by this study, a retrieval algorithm for cloud microphysics was applied to CloudSat/CALIPSO synergy data (Okamoto et al., 2009, submitted to Journal of Geophysical Research). Together with the cloud particle type results, such cloud microphysics information should help advance evaluations of cloud parameterization in climate models.

Acknowledgments

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. Particle Type Discrimination From CALIOP Data
  5. 3. Results and Discussions
  6. 4. Conclusions
  7. Acknowledgments
  8. References
  9. Supporting Information

[48] CALIPSO data were obtained from the Atmospheric Sciences Data Center at NASA's Langley Research Center. CloudSat data were obtained from the CloudSat Data Processing Center, and MODIS monthly Level 3 data were acquired from the Level 1 and Atmosphere Archive and Distribution System website of NASA's Goddard Space Flight Center (http://ladsweb.nascom.nasa.gov/). We would like to thank the CALIPSO, CloudSat, and MODIS science teams. This work was partly supported by the Ministry of Education, Culture, Sports, Science, and Technology of Japan through a grant-in-aid for scientific research (B) (19340132), by a grant-in-aid for scientific research (B) (22340133), and also by Special Coordination Funds for Promoting Science and Technology, Japanese Cloud Seeding Experiments for Precipitation Augmentation.

References

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. Particle Type Discrimination From CALIOP Data
  5. 3. Results and Discussions
  6. 4. Conclusions
  7. Acknowledgments
  8. References
  9. Supporting Information

Supporting Information

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. Particle Type Discrimination From CALIOP Data
  5. 3. Results and Discussions
  6. 4. Conclusions
  7. Acknowledgments
  8. References
  9. Supporting Information
FilenameFormatSizeDescription
jgrd15724-sup-0001-tab01.txtplain text document0KTab-delimited Table 1.

Please note: Wiley Blackwell is not responsible for the content or functionality of any supporting information supplied by the authors. Any queries (other than missing content) should be directed to the corresponding author for the article.