Author's personal copy Remote Sensing: Cloud Properties P Yang, Texas A&M University, College Station, TX, USA BA Baum, University of Wisconsin–Madison, Madison, WI, USA Ó 2015 Elsevier Ltd. All rights reserved. Synopsis Clouds constitute a unique and important component of the atmosphere. This article briefly reviews the methods of inferring cloud-top height, determining cloud thermodynamic phase, and retrieving cloud microphysical and optical properties (specifically, the effective particle size and optical thickness). Some examples based on observations made by a passive spaceborne sensor (the Moderate Resolution Imaging Spectroradiometer) and an active spaceborne sensor (the Cloud– Aerosol Lidar with Orthogonal Polarization) are illustrated. Introduction On any given day, clouds cover about 65% of the planet. In a fairly stable atmosphere, clouds may be cellular in appearance (i.e., cumuliform) or may appear in sheets (i.e., stratiform) that may extend over large horizontal distances. While these clouds may extend over wide areas, their typical geometric thickness is less than 1 km. In unstable atmospheres, clouds may extend from near the planet’s surface to the upper troposphere. As most of the tropospheric water vapor resides near the surface, where temperatures tend to be relatively warm, low-level clouds tend to be composed of water droplets and are generally opaque to the viewer. The opacity is denoted in terms of a quantity known as optical thickness, or optical depth, and is a dimensionless measure of light attenuation caused by the scattering and absorption of energy by atmospheric particles. Clouds forming near the tropopause reside at very cold temperatures and are typically composed of ice particles. For clouds at intermediate heights between the planetary boundary layer (w1 km above the surface) and the middle troposphere, clouds may be composed of a mixture of supercooled water and ice particles. Water and ice clouds interact with solar radiation differently and have a large influence on the Earth’s radiative energy budget. The energy budget is composed of both solar and terrestrial radiation components. Solar radiation spans from ultraviolet (l < 0.4 mm, where l is the wavelength) to infrared (IR) wavelengths (l > 5 mm). A portion of the incoming solar radiation may be absorbed at the surface and within the atmosphere by clouds, aerosols, water vapor, and other trace gases such as carbon dioxide and methane. Subsequently, absorbed solar radiation is reemitted at longer wavelengths ranging from 5 to 100 mm. Data from operational polar-orbiting and geostationary meteorological satellites are analyzed routinely for global cloud macrophysical properties such as cloud height, phase (water, ice, or some mixture of both), and microphysical and optical properties such as optical thickness and the effective particle size. Global cloud observations based on satellite measurements serve many uses. In numerical weather models, where the time scale of interest is on the order of hours to days, satellite-derived cloud and clear-sky properties from the geostationary satellites can serve as initial conditions for the models, that is, where the clouds are at some given time, their 116 height, and other properties. Numerical weather models may be regional in extent, covering a specific area such as North America, or global, in which case global and near real-time clouds and clear-sky properties are required for initialization of the models. Monthly, annual, or decadal averages of satellite-derived cloud properties are also useful for comparing with results from global climate models where the time scale of interest is much longer than for weather prediction models. For this type of use, cloud properties need to be collected, analyzed, and ultimately reduced to a global-gridded and time-interpolated product. An example of such a product would be one where each of the cloud properties retrieved during the course of a month is reduced to a monthly average with a time resolution of every 3–6 h. One of the primary issues in building a decadal climatology based on satellite observations is that the satellite sensor calibration needs to be very accurate. Since the advent of meteorological satellites, beginning around 1980, a long line of weather satellites have come into or out of service. Once in space, the platforms are subject to very harsh environments that can modify the sensor calibration over time, and for polar-orbiting platforms, the orbit can degrade over time. The derivation of a decadal record of cloud properties requires constant attention to sensor calibration. To date, meteorological satellites have recorded information over the Earth at a limited number of wavelengths through the use of specially designed filter radiometers. The filters only allow radiation over a very narrow wavelength range to pass through to the detectors. Such narrowband wavelengths are typically chosen in atmospheric ‘windows,’ where the atmospheric constituents such as water vapor and carbon dioxide least attenuate the energy along the path to/from the surface, through the atmosphere, and finally to the satellite. At a minimum, operational satellite data are recorded at a visible (VIS) wavelength (e.g., 0.65 mm), a medium-wave-infrared (MWIR) wavelength (3.82 mm), and an IR wavelength (11 mm). Radiances at VIS and near-IR wavelengths are often converted to reflectances whose values range from 0 to 1. IR radiances are often converted to brightness temperatures (BTs) through application of the Planck function. Because of the huge volumes of data collected by satellites, the data reduction effort can become quite complex. In this article, we will discuss some of the available methods to infer cloud properties such as Encyclopedia of Atmospheric Sciences 2nd Edition, Volume 5 http://dx.doi.org/10.1016/B978-0-12-382225-3.00503-X Encyclopedia of Atmospheric Sciences, Second Edition, 2015, 116–127 Author's personal copy Satellites and Satellite Remote Sensing j Remote Sensing: Cloud Properties cloud-top pressure, phase, optical thickness, and the effective particle size. Cloud-Top Pressure–Height–Temperature Over the past several decades, a number of approaches have been developed to infer cloud-top heights from satellite multispectral data. Actually, the literature provides a wealth of different research-grade algorithms, but very few have been fully developed and adopted for routine, operational processing of global data. For operational data processing, the assumption is made that only a single cloud layer is present in any individual field of view (FOV). Both surface observations and spaceborne lidar or radar measurements indicate that multilayered clouds occur frequently. If the uppermost cloud layer is optically thick, then a passive satellite sensor cannot sense the presence of lower level cloud layers. If, however, the upper cloud layer is optically thin, such as cirrus, then there is some potential for the presence of a lower level cloud layer to modify the radiances observed by the satellite sensor, causing errors in the assessment of the cloud properties for that FOV. Another assumption generally made when inferring the cloud height is that there is a well-defined cloud-top boundary. For low-level water clouds, such as stratocumulus or cumulus, the cloud-top boundary is well defined. For high-level clouds, such as cirrus, this assumption is more problematic as the cirrus layer can be geometrically thick but with very sparse ice particles throughout the layer, which is another way of saying the cloud is optically thin. 117 The clouds that require the most attention in operational retrievals are those that reside either near the tropopause (highlevel clouds) or near the surface. Some low-level clouds occur in atmospheres with temperature inversions. Proper placement of cloud-top heights requires that there be some knowledge of the atmospheric temperature profile, and numerical models are somewhat deficient on this in many cases. Given the many assumptions that need to be made, e.g., that an FOV contains only a single-layered cloud, is not optically thin at the top of the cloud layer, and that the temperature profile contains no surprises, there are some general approaches to inferring cloud height that are in use. On many satellite platforms, measurements are obtained at wavelengths located in the 15-mm wavelength region, a region in which atmospheric transmission is dominated by atmospheric CO2. As the wavelength increases from 13.3 to 15 mm, the atmosphere becomes more opaque due to CO2 absorption, thereby causing each channel to be sensitive to a different portion of the atmosphere. This sensitivity is demonstrated in Figure 1, which shows weighting functions at several Moderate Resolution Imaging Spectroradiometer (MODIS) channels located at wavelengths ranging from 12 to 14 mm. Each channel has a peak in its weighting function that occurs at a different pressure level than the other channels. The 12-mm channel is shown for comparison – note that its weighting function peaks at the surface. This is a ‘window’ channel that is insensitive to CO2. In the 1970s, Moustafa Chahine, William Smith Sr., and Martin Platt developed a technique known as CO2 slicing to infer cloud-top pressure from radiances measured at wavelengths between 13.3 and 14.2 mm. In principle, the CO2 slicing method is based on the Figure 1 Weighting functions that are derived for MODIS wavelengths ranging from 12 to 14.2 mm. The weighting function is the derivative of the transmittance profile as a function of pressure. The peak in the weighting function provides an indication of what levels in the atmosphere provide most of the upwelling radiance that will be measured by a satellite. Encyclopedia of Atmospheric Sciences, Second Edition, 2015, 116–127 Author's personal copy 118 Satellites and Satellite Remote Sensing j Remote Sensing: Cloud Properties following relation derived from the theory of radiative transfer: RP N10 c GðP; n1 ÞfvB½TðPÞ; n1 =vPgdP Rðy1 Þ Rclear ðy1 Þ P ¼ ; [1] R P0 Rðy2 Þ Rclear ðy2 Þ N 0 c GðP; n2 ÞfvB½TðPÞ; n2 =vPgdP 2 P 0 where R(y1) and R(y2) are the radiances measured at two channels centered at wave numbers y1 and y2, whereas Rclear(y1) and Rclear(y2) are the corresponding clear-sky radiances. The terms G, T, and P indicate the transmissivity, temperature, and pressure, respectively. P0 and Pc indicate the pressure values at the surface and cloud top, respectively. N0 denotes the effective cloud amount that is the product of the cloud fraction and the cloud emissivity. If the two channels are selected to be sufficiently close in wave number, the corresponding effective cloud amount values are approximately the same. In this case, it is straightforward to find an appropriate value for the cloud-top pressure Pc by assuring the equality in eqn [1]. The pressure at cloud level is converted to cloud height and cloud temperature through the use of gridded meteorological products that provide temperature profiles at some nominal vertical resolution every 6 h. One benefit to this algorithm is that cloud properties are derived similarly for both daytime and nighttime conditions as the IR method is independent of solar illumination. This approach is very useful for the analysis of midlevel to high-level clouds and even optically thin clouds such as cirrus. The drawback to the use of the 15-mm channel is that the signal-to-noise ratio becomes small for clouds occurring in the lowest 3 km of the atmosphere, making retrievals problematic for low-level clouds. When low clouds are present, the 11-mm channel (also a window channel) is used to infer cloud height. Cloud Thermodynamic Phase While the cloud phase is extremely important in radiative transfer simulations of clouds and the retrieval of cloud properties, it is not always straightforward to determine a cloud’s phase. If the cloud is located in the upper troposphere where the temperatures are extremely cold, it is assumed to be composed of ice. Conversely, if the cloud is located in the boundary layer over warm surfaces, it is assumed to be water. The difficulty lies in the inference of phase when the cloud-top temperature lies between 233 and 273 K. If the cloud temperature is below 233 K, the homogeneous nucleation temperature, it will be composed of ice. If the cloud temperature is above 273 K, it will be composed of water. If the cloud has a temperature between 233 and 273 K, it could be ice, water, or some mixture of both. In the high-latitude storm tracks in either hemisphere, large-scale stratiform cloud decks tend to form with cloud-top temperatures in the 250–265 K range, and cloud phase is quite difficult to discern. At temperatures below 273 K, the supersaturation of ice is much higher than the supersaturation with respect to water. If water vapor is present in an atmospheric layer at a temperature in this range, say 260 K, and both water and ice particles are present in this layer, the water vapor will preferentially condense on the ice particles rather than the water particles. As the ice particles become larger, which occur over the course of seconds to minutes, the growing ice particles will begin to fall through the cloud layer. In this situation, the top of the cloud layer tends to be populated primarily by very small water droplets, while ice particles fall through the cloud base. The cloud layer may contain both ice and water particles, so inference of the cloud phase from satellite data under these conditions is quite challenging. Two simple approaches are discussed here to infer cloud phase from the radiometric observations made by a passive sensor. One method involves IR radiances measured at 8.5 and 11 mm. The radiances are converted to BTs through the Planck function, and the phase is inferred from the brightness temperature difference (BTD) between the 8.5 and 11 mm BTs (BTD[8.5–11]) as well as the 11 mm BT. Ice clouds exhibit positive BTD[8.5–11] values, whereas water clouds tend to exhibit highly negative values. There are three contributing factors to the behavior of the BTD[8.5–11] for ice and water clouds. First, the imaginary component of the index of refraction (mi) differs for ice and water at these two wavelengths. Second, while the atmosphere is relatively transparent to gaseous absorption, absorption by water vapor in the atmospheric column above the cloud can still exert a considerable effect on the BTD values. As most of the atmospheric water vapor resides in the lower layers of the atmosphere near the surface, the BTD[8.5–11] values will be most affected in moist atmospheres rather than high-level clouds that reside above most of the water vapor. Third, while a small effect, cloud particles scatter radiation even at the IR wavelengths, and clouds with smaller particles will tend to scatter more radiation than those with larger particles. Multiple-scattering radiative transfer calculations show that for ice clouds, the BTD[8.5–11] values tend to be positive in sign, whereas for low-level water clouds, the BTD[8.5–11] values tend to be very negative (<2 K). This simple BTD approach with IR channels can be improved for optically thin ice cloud discrimination by calculating cloud emissivity ratios. In the simplest terms, the cloud emissivity for a channel is based on three numbers: the measured cloud radiance, the black cloud radiance, and the calculated clear-sky radiance. The term ‘black’ here means that the cloud radiates as a blackbody, which implies that it is opaque at the wavelength of the observation. This is more complicated than a simple BTD approach above because it requires the use of a radiative transfer model (RTM) to provide the clear-sky and black cloud radiances. However, what this approach provides is much more sensitive to optically thin ice clouds. The IR methods are not very useful when supercooled water clouds are present, however, since it is problematic to discriminate between water and ice as discussed previously. One way to improve the discrimination between water and ice clouds is to analyze reflectances obtained at a VIS wavelength and a shortwave-infrared (SWIR) wavelength (e.g., 0.65 and 1.64 mm, respectively). At wavelengths less than about 0.7 mm, clouds composed of either liquid or ice tend to absorb very little solar radiation. However, at 1.64 mm (and 2.15 mm), the mi values for both water and ice increase in comparison with those at the VIS wavelength and diverge, with mi for ice being greater than the value of mi for water. From this line of reasoning, one might expect that for two different clouds (one ice and one water) of similar particle size and habit (or particle Encyclopedia of Atmospheric Sciences, Second Edition, 2015, 116–127 Author's personal copy Satellites and Satellite Remote Sensing j Remote Sensing: Cloud Properties shape) distributions, the cloud reflectance at 0.65 mm would not depend on thermodynamic phase, whereas the cloud reflectance at 1.64 mm would. In theory and in practice, the 1.64 mm (and 2.15 mm) reflectances are much lower for a cloud composed of ice than water particles. The observations made by an active spaceborne sensor, for example, the Cloud–Aerosol Lidar with Orthogonal Polarization (CALIOP) on the Cloud–Aerosol Lidar and Infrared Pathfinder Satellite Observations (CALIPSO) platform can be used to effectively determine the cloud thermodynamic phase. The CALIOP 532-nm channel measurements offer polarization capabilities. Two quantities, the layer-integrated backscatter (g0 ) and the layer-integrated depolarization ratio (d) can be employed to effectively discriminate cloud thermodynamic phase, which are defined as follows: 0 g ¼ Z cloud base h cloud top R d ¼ R b0t ðzÞ þ b0k ðzÞ i dz; [2] cloud base b0 ðzÞdz cloud top t ; cloud base 0 b ðzÞdz cloud top k [3] where b’t ðzÞ and b’k ðzÞ indicate the vertical backscatter profiles associated with the perpendicular and parallel components, respectively. For a given cloudy scene, the g0 –d relationship can be used to distinguish cloud phase. As illustrated in the right 119 panel of Figure 2 (the physical concept was originally developed by Yongxiang Hu at NASA Langley Research Center), water cloud pixels correspond to a g0 –d relationship with a positive slope, whereas a g0 –d relationship with a negative slope is related to ice cloud pixels. Furthermore, in the case of ice cloud pixels, the upper left branch of the g0 –d curve corresponds to ice clouds containing horizontally oriented ice crystals, whereas the lower right branch of the g0 –d curve is related to ice clouds composed of randomly oriented ice particles. The right panel of Figure 2 shows the frequency of occurrence of the g0 –d relations of ice clouds based on the CALIOP data collected from July through December 2006. Cloud Optical Thickness and Particle Size The basic retrieval methodology for inferring the optical thickness and effective particle size is to (1) employ a RTM to develop a lookup table (LUT) for a wide range of assumed cloud properties and viewing geometries and subsequently (2) compare the measured radiances for selected wavelength channels to values in the LUT. The RTM requires a set of singlescattering properties for the cloud layer, which includes the single-scattering albedo, the scattering phase function, the scattering–absorption–extinction efficiencies, and the asymmetry factor. These parameters essentially determine how much incident radiation is reflected or absorbed by the cloud. The single-scattering albedo is defined as the ratio of the portion of energy scattered by a particle to the total extinction Figure 2 Left panel: schematic diagram showing the g0 d relationships for water and ice cloud pixels. Right panel: g0 d relationships based on the CALIOP measurements in the case when the lidar beam was pointed within 0.3 from the nadir. Encyclopedia of Atmospheric Sciences, Second Edition, 2015, 116–127 Author's personal copy 120 Satellites and Satellite Remote Sensing j Remote Sensing: Cloud Properties (scattering þ absorption) of energy by the particle. The phase function specifies the percentage of radiative energy that is not absorbed but is instead redistributed by the action of scattering by cloud particles when radiation impinges on clouds. The asymmetry factor describes the ratio of forward scattered to backscattered energy and is a quantity often used in radiative flux calculations. In practice, the single-scattering albedo and the asymmetry factor are parameterized in terms of analytical functions (normally polynomials) that depend on particle effective size for both water and ice clouds. In many RTMs, the radiative properties of clouds are described in terms of particle effective size and either liquid or ice water content (LWC or IWC), depending on the cloud phase. Cloud optical thickness and particle effective size are critically dependent on the accurate determination of the cloud bulk radiative properties, and a focus of recent research has been to improve the description of ice clouds in RTMs. Various methods have been suggested to derive the optical thickness and effective particle size based on narrowband radiometer measurements by airborne- or satellite-based imagers. Operational methods tend to rely on IR bands or a combination of VIS and SWIR bands. The IR approach depends on the spectral information from thermal emission of clouds, whereas the VIS–SWIR approach is based on the reflection of solar radiation. Teruyuki Nakajima and Michael King were among the first to use reflected solar radiation to simultaneously retrieve cloud optical thickness and effective particle size for water clouds. The typical IR technique employs the BT or BTD values based on window channels at 8.5, 11, and 12 mm. Regardless of the detailed spectral information involved in these two methods, they are similar in that both depend on comparison of measured radiance data with simulated radiances derived for similar viewing and atmospheric conditions. The first step in this process is to discuss the generation of reliable libraries of simulated cloud and clear-sky radiances. Single-scattering calculations must be carried out regarding how individual cloud particles interact with incident radiation. For water clouds, the liquid droplets can be well approximated as spheres for light scattering. The scattering properties of an individual liquid sphere can be calculated by using the wellknown Lorenz-Mie theory that has been documented in many texts. James Hansen and Larry Travis have extensively discussed the effect of size distribution on single-scattering properties of spheres. Their work provides a theoretical framework for using and applying the bulk radiative properties of liquid droplet distributions which is briefly recaptured here. Within a given water cloud, liquid water droplets span a range of sizes that may be represented mathematically in terms of the Gamma distribution, given by V 1 V N0 reff Veff ð eff Þ= eff ð13Veff Þ=Veff r exp r=reff Veff ; nðrÞ ¼ G 1 2Veff Veff [4] where N0 is the total number of the droplets in a unit volume; reff and Veff are the effective radius and effective variance that are defined, respectively, as follows: R r2 3 r r nðrÞdr ; [5] reff ¼ R r12 2 r1 r nðrÞdr R r2 Veff ¼ r1 2 r reff r 2 nðrÞdr R : r reff2 r12 r 2 nðrÞdr [6] In a plot of the Gamma distribution, the peak of the distribution defines the reff, while Veff affects the width of the distribution. Typical values of the effective variance for water clouds range from 0.05 to 0.1. For a given size distribution, the bulk-scattering properties of cloud droplets may be calculated. For example, the phase function averaged over a size distribution is given by R r2 r ss ðrÞPðq; rÞnðrÞdr < PðqÞ > ¼ 1 R r2 ; [7] r1 ss ðrÞnðrÞdr where ss is scattering cross section of droplets and P(q,r) is the phase function for droplets with radii of r, which describes the angular distribution of scattered radiation versus scattering angle q. Figure 3 shows the phase functions averaged for size distributions for water clouds at wavelengths 0.65, 1.63, and 11 mm. For the 0.65-mm wavelength, the phase function displays scattering maxima at 140 and 180 . Physically, the two maxima are due to mechanisms associated with the rainbow and glory, both characteristic features of Mie scattering. The phase functions at the SWIR wavelength (1.63 mm) are similar to those at 0.65 mm, but the rainbow and glory maxima are somewhat reduced by absorption within the particle. At the IR wavelength of 11 mm, the scattering maxima of the phase function are largely smoothed out due to absorption within the water droplets. Another measure of the relative amounts of scattering versus absorption is provided by the single-scattering albedo. At 0.65 mm, the scattering of incident radiation by cloud droplets is conservative, meaning that energy may be scattered, but not absorbed, by the particles. Thus, the single-scattering albedo is unity at 0.65 mm but less than unity at 1.63 mm. The particle size also affects the single-scattering albedo at 1.63 mm. For example, for effective sizes 4 and 32 mm, the particle single-scattering albedo is unity at 0.65 mm, whereas the corresponding values at 1.63 mm are 0.9976 and 0.9824, respectively. Because of the difference in single-scattering albedo at the two wavelengths, reflection by an optically thick cloud at 0.65 mm is essentially a function of optical thickness. At 1.63 mm, however, cloud reflectance is sensitive to droplet effective size. This feature of cloud reflectance provides a mechanism to retrieve cloud optical thickness and particle sizes using two channels at VIS and SWIR wavelengths, as will be further explained later in this section. Ice clouds are almost exclusively composed of nonspherical ice particles with various sizes and habits (i.e., shapes). Ice particles can consist of relatively simple shapes such as bullet rosettes, columns, and plates or more complex shapes such as aggregates of columns or plates. Most of the columnar particles can have hollow intrusions at the ends, which is caused by preferential molecular deposition onto a growing particle. In an environment where supercooled water droplets are present, the ice particles can also become rimed, which increases an individual particle’s surface roughness. An increasing amount of research is showing that the consistency of inferred ice cloud properties improves between algorithms using solar, IR, or Encyclopedia of Atmospheric Sciences, Second Edition, 2015, 116–127 Author's personal copy Satellites and Satellite Remote Sensing j Remote Sensing: Cloud Properties 121 Figure 3 Scattering phase function of water droplets calculated at three wavelengths at 0.65, 1.63, and 11 mm for effective radii of 4, 8, and 16 mm. polarized measurements if an assumption of ice particle severe surface roughening is adopted. Research is underway to determine how to accurately calculate the single-scattering properties of a limited set of idealized ice habits. In practice, methods such as the discrete dipole approximation, finite-difference time domain technique, or the T-matrix method are used to calculate the scattering properties of a given habit for which the ratio of the particle circumference to the wavelength (also known as the size parameter) is small, i.e., less than 30. For ice particles with larger size parameters, scattering calculations are performed using a ray-tracing technique based on the principles of geometric optics. Figure 4 shows the phase matrices at 0.65-mm wavelength for two types of ice crystals: a solid column with smooth surfaces and aggregates of plates with rough surface. The phase function of smooth hexagonal columns displays a strong scattering peak at 22 and is produced by the hexagonal structure typical of ice crystals. In addition to the peak at 22 , the phase function of solid columns also displays a small peak Encyclopedia of Atmospheric Sciences, Second Edition, 2015, 116–127 Author's personal copy 122 Figure 4 Satellites and Satellite Remote Sensing j Remote Sensing: Cloud Properties The scattering phase matrices of hexagonal ice crystals with smooth surface and aggregates of plates with rough surfaces. corresponding to a 46 halo. Compared to the phase function for pristine crystal habits, the phase function for aggregates of plates is essentially featureless due to the severely roughened surface texture. The rougher the particle, the more featureless is the phase function. The other nonzero elements of the phase matrix are related to the polarization state of the scattered light. The impact of surface roughness on the polarization state is significant. Some recent studies have demonstrated that polarization measurements, for example, by the Polarization and Anisotropy of Reflectances for Atmospheric Sciences coupled with Observations from a Lidar (PARASOL) offer unprecedented capabilities to infer ice crystal habit and associated particle roughness. In particular, the comparison between the polarized reflectance observed by PARASOL and the relevant theoretical simulations illustrates that the closest match occurs when assuming the presence of ice crystals with severely roughened surfaces. In reality, ice clouds are composed of many different crystal habits. To derive the bulk radiative properties of cirrus clouds, we need to consider not only a particle size distribution but also the percentages of the various particle habits that comprise the cloud. For this reason, the derivation of accurate radiative transfer simulations of ice clouds is considered more difficult than for water clouds. For a given size distribution, a number of definitions have been suggested for the effective size. If the effective size is defined as the ratio of total volume to total projected area, however, the bulk optical properties are insensitive to the detailed structure of the size distribution. The effective radius is then RP fi Vi ðDÞnðDÞdD 3 i ; [8] reff ¼ R P 4 fi Ai ðDÞnðDÞdD i where D is the maximum dimension of an ice particle, fi is the habit fraction, V and A are the volume and projected area for individual particle, and n is the particle number concentration. Based on in situ measurements within ice clouds, a modified gamma distribution is used most often to describe the particle size distribution. In situ ice cloud measurements are now available from numerous field campaigns based at locations around the world. For example, Table 1 (data courtesy of Andrew Heymsfield, National Center for Atmospheric Research) lists a number of the particle size distributions obtained at various field campaigns and the instruments used for the microphysical measurements. This is by no means a complete list. A new generation of sensors is beginning to provide measurements of the smallest particles in a given particle population and even a sense of the particle roughening. In situ measurements indicate that the effective radius of ice crystals in cirrus clouds may range from about 5 mm (small ice particles near the tropopause) to more than 100 mm (deep convection). Larger particle radii might be expected for ice clouds formed in convective situations where the updraft velocity is much higher (m s1) than that found under conditions where optically thin cirrus tend to form (cm s1). The in situ measurements provide insight for the development of an appropriate ice cloud model in terms of the ice crystal habit and size distributions. As an example, the upper left panel of Figure 5 illustrates an ice model based on two habits (hexagonal columns and aggregates of plates) with surface roughness. The lower left panel of Figure 5 shows the comparisons of the computed medium mass diameter (where half the mass is in smaller particles and half in larger particles) versus in situ measurements, whereas the lower right panel shows the corresponding comparison for IWC. Apparently, the two-habit model can reasonably represent in situ microphysical measurements. The upper right panel of Figure 5 shows the phase function based on the two-habit model in comparison with the MODIS Collection 5 counterpart. Note that the asymmetry factors associated with the two Encyclopedia of Atmospheric Sciences, Second Edition, 2015, 116–127 Author's personal copy Satellites and Satellite Remote Sensing j Remote Sensing: Cloud Properties 123 Table 1 Number of the ice particle size distributions obtained during various field campaigns and the instruments for the microphysical measurements Field campaign Year Location Probes ARM-IOP TRMM KWAJEX CRYSTAL-FACE Pre-AVE MidCiX ACTIVE-Hector ACTIVE-Monsoon ACTIVE-Squall Line SCOUT TC-4 MPACE 2000 1999 2004 2004 2004 2005 2005 2005 2005 2006 2004 Oklahoma, USA Kwajelein, Marshall Islands Florida area, over ocean Houston, Texas Oklahoma Darwin Darwin Darwin Darwin, Australia Costa Rica Alaska 2D-C, 2D-P, CPI 2D-C, 2D-P, CPI CAPS, VIPS VIPS CAPS, VIPS CAPS CAPS CAPS FSSP, 2D-C CAPS, CPI 2D-C, 2D-P, CPI The data are filtered such that the in situ measurement occurs at a cloud temperature T 40 C. Notes: (1) The table is from: http://www.ssec.wisc.edu/ice_models/microphysical_data.html. (2) The data sets currently include a total of 14 406 particle size distributions and the list will increase over time. Figure 5 Upper left panel: a two-habit ice cloud model based on hexagonal columns and aggregate of plates in conjunction with the Gamma distribution. Lower left panel: comparison of the theoretical median mass diameter versus in situ measurements associated with the data sets listed in Table 1. Lower right panel: comparison of the theoretical IWC versus in situ measurements associated with the data sets listed in Table 1. Upper right panel: the phase function computed with the two-habit model in comparison with the MODIS Collection 5 phase function. phase functions are quite different; particularly, the asymmetry factor for the two-habit model is approximately 0.76, whereas the MODIS Collection 5 counterpart is 0.82. Given the single-scattering properties, radiative transfer computations can be carried out for various cloud optical thickness and effective particle sizes for a number of solar and viewing geometry configurations. To calculate the bidirectional radiance of clouds, one can use well-established discrete ordinate or adding–doubling methods. Figure 6 shows the correlation of 2.13-mm reflectance and 0.86-mm reflectance of cirrus Encyclopedia of Atmospheric Sciences, Second Edition, 2015, 116–127 Author's personal copy 124 Satellites and Satellite Remote Sensing j Remote Sensing: Cloud Properties Figure 6 The theoretical relationship between the reflection function at 0.86 and 2.13 mm for various values of cloud optical thickness and effective particle size. clouds for a number of optical thickness and effective sizes for a given incident-view geometry. At higher optical thicknesses (meaning the cloud is more opaque), there is a ‘quasi-orthogonality’ between the optical thickness and particle size curves. As we have mentioned previously, the cloud reflectance at 0.86 mm is primarily sensitive to cloud optical thickness, whereas the reflectance at 2.13 mm is sensitive to both the particle size and cloud optical thickness. This orthogonality forms the underlying principle for application of the twochannel correlation technique for retrieving cloud optical thickness and effective size. For example, assume the symbol ‘X’ in Figure 6 to represent the (0.86 and 2.13 mm) reflectivity pair. One may infer that the corresponding optical thickness is approximately 14, whereas the effective particle size is 25 mm. It should be pointed out that, in practice, the (0.86 and 2.13 mm) reflectivity combination is usually used for retrieval over ocean, the (0.64 and 2.13 mm) reflectivity combination is used for over land, and the (1.24 and 2.13 mm) reflectivity combination is used over snow or ice. In addition to the 2.13-mm channel, a channel located at 1.64 or 3.7 mm can be used as the SWIR or MWIR channel involved in the aforementioned bispectral method. As an alternative or as a complement to the VIS–SWIR bispectral retrieval algorithm, IR channels in the window region (8–12 mm) may be used for retrieving cloud properties. The window region is an important part of the IR spectrum because terrestrial thermal emission peaks within this spectral region. IR-based methods are useful because a single approach may be used for both daytime and nighttime conditions, thereby simplifying the data reduction effort and also the comparison between daytime and nighttime cloud properties. IR methods are insensitive to sun glint over water that is often present in operational data. Interpretation of data over reflective surfaces is often performed more readily using IR methods rather than those that involve VIS–SWIR wavelengths. The underlying principle for IR retrievals is based on the sensitivity of the BT or the cloud emissivity (related to blackbody or graybody emission) to cloud optical thickness and particle size. The BT is the temperature that, when applied to the calculation of Plank function for blackbody radiation, gives the same value as the satellite measured IR radiance. Figure 7 illustrates the sensitivity of the BTD between the 11- and 12-mm channels as a function of the BT at the 11-mm channel for various cloud optical thickness and the effective particle size. Evidently, comparing the measurements of the BTD–BT relation with the theoretical computations permit simultaneous retrieval of cloud optical thickness and the effective particle size. However, the IR technique is more sensitive to the atmospheric profile (particularly, the temperature profile) and the surface emissivity than the VIS–SWIR technique. In addition to the use of BDT and BT, a quantity known as the cloud emissivity has been widely used to infer cloud properties. In practice, the cloud emissivity can be calculated as follows: εðlÞ ¼ RðBÞ R ; RðBÞ RðCÞ [9] where R is the upwelling radiance at the cloud top, R(B) is the upwelling radiance at the cloud bottom, and R(C) is the upwelling blackbody radiance corresponding to the cloud temperature. In practice, for a given scene, the radiance at cloud base can be obtained by the noncloudy (i.e., clear sky) pixels. Furthermore, the IR techniques for retrieving ice cloud properties are less sensitive than their VIS–SWIR counterparts to ice crystal habits assumed in the forward light-scattering and radiative transfer simulations. To illustrate this point, panels (a) and (b) of Figure 8 show the phase functions of two ice crystal habits (hexagonal columns and hollow bullet rosettes) Encyclopedia of Atmospheric Sciences, Second Edition, 2015, 116–127 Author's personal copy Satellites and Satellite Remote Sensing j Remote Sensing: Cloud Properties Figure 7 125 The variation of the BTD between the 8.5- and 12-mm channels as a function of the BT at the 11-mm channel. Figure 8 Panel (a): bulk phase functions of solid columns and hollow bullet rosette with an effective particle size of 50 mm at 0.86 mm. The gamma distribution is used to simulate the size distribution. Panel (b): similar to panel (a) except for a wavelength of 11 mm. Panel (c): comparison of cloud optical thickness retrievals based on the VIS–SWIR retrieval on the basis of a solid column habit model and a hollow bullet rosette habit model. Panel (d): similar to panel (c) except that an IR technique is used. Encyclopedia of Atmospheric Sciences, Second Edition, 2015, 116–127 Author's personal copy 126 Satellites and Satellite Remote Sensing j Remote Sensing: Cloud Properties at wavelengths of 0.86 and 11 mm, respectively. Substantial differences are noticeable at the 0.86-mm near IR wavelength, whereas the two phase functions are quite similar at the 11-mm wavelength. Panel (c) of Figure 8 compares the optical thickness values retrieved with the VIS–SWIR bispectral method based on MODIS Band 2 (0.86 mm) and Band 7 (2.13 mm) measurements. The impact of the assumed ice crystal habit on the retrieval is obvious from panel (c). The optical thickness values retrieved from the IR technique are shown in panel (d) based on the MODIS Band 29 (8.5 mm), Band 31 (11 mm), and Band 32 (12 mm) observations. The effect of ice crystal habit on the IR-based technique is negligible. Cloud radiative and microphysical properties are cloud inherent properties that should be independent of a specific retrieval algorithm employed to infer the cloud properties. In this sense, the VIS–SWIR and IR retrievals should be consistent. The spectral consistency of cloud property retrievals is critical to some analyses, particularly, the study of the diurnal variations of cloud properties based on a VIS–SWIR algorithm for daytime and an IR algorithm for nighttime. In the case of ice cloud, recent studies have demonstrated that the ice cloud optical model involved in the forward radiative transfer simulation is essential for achieving spectral consistency. Future Challenges in Cloud Property Retrieval Current efforts to derive a global cloud climatology from satellite data generally do not account properly for multiple cloud layers in pixel-level imager data. To date, operational algorithms are designed to infer cloud properties for each imager pixel under the assumption that only one cloud layer is present. Climatologies of retrieved cloud properties do not address the effect of an optically thin upper cloud layer, such as cirrus, that may overlay a lower cloud layer such as a cumuliform cloud deck. Surface observations show that clouds often occur in multiple layers simultaneously in a vertical column, i.e., cloud layers often overlap. Multiple cloud layers occur in about half of all cloud observations and are generally present in the vicinity of midlatitude fronts and in the tropics where cirrus anvils may spread hundreds of kilometers from the center of convective activity. When multilayered clouds are present, the retrieval algorithms will generally place the cloud layer at a height between the two (or more) actual layers present in the FOV. Currently, available satellite cloud climatologies provide a horizontal distribution of clouds but need improvement in the description of vertical distribution of clouds. At this point, a reliable method has not been developed for the retrieval of cloud properties (optical thickness, cloud thermodynamic phase, and effective particle size) when multilayered, overlapping clouds are present. Even for a single-layered cloud, satellite retrieval algorithms do not account for the effect of a likely vertical variation of cloud microphysical properties, which in turn will decrease the ability of radiative transfer calculations to accurately simulate the cloud. It is unlikely that cloud particles are homogeneously distributed throughout any given cloud. For example, ice crystal size and habit are typically quite different for midlatitude cirrus at cloud top from at cloud base. A common assumption in satellite imager–based cirrus retrieval algorithms is that the radiative properties of a cirrus cloud may be represented by those associated with a specific ice crystal shape (or habit) and a single particle size distribution. However, observations of synoptic cirrus clouds with low updraft velocities have shown that pristine small ice crystals with hexagonal shapes having an aspect ratio close to unity (length and width are approximately equal) are predominant in top layers. The middle layers of cirrus are often composed of well-defined columns and plates, while irregular polycrystals or aggregates are dominant near cloud base. This picture is quite different from ice particles that form in deep convection; in this case, the population of ice particles may be dominated by complex aggregates. Another interesting area of complexity in satellite remote sensing is caused by mixed-phase clouds. Single-layered clouds composed of mixtures of supercooled water droplets and ice particles have been observed frequently during various field campaigns. Recent analyses of these data and MODIS satellite cloud property retrievals highlight the difficulty of ascertaining phase. If mixed-phase clouds are present in the data, one might expect larger errors in retrieved properties such as optical thickness and particle size than clouds that are primarily of a single phase. From the perspective of satellite remote sensing, the working assumption is that any imager pixel contains either ice or water but not a mixture. There is no rigorous method available for determining the single-scattering properties of mixed-phase clouds. From the microphysical cloud process perspective that is important for developing cloud model parameterizations, the presence of both ice particles and supercooled water droplets will affect cloud lifetime. Why? It is likely that the ice particles will grow much more quickly from vapor deposition than the water droplets as the environment may be supersaturated with respect to ice. The result of this process is that the ice particles will rime, grow quickly in size, and fall through the cloud, and the available water will be depleted quickly. The process of glaciation is very important for modelers because the water– ice conversion rates affect cloud lifetime. Details of cloud microphysics, such as cloud water amount, cloud ice amount, snow, graupel, and hail, are important for improving cloud retrieval. While approaches exist to retrieve a variety of cloud properties from satellite imager data, it is not an easy problem to compare the satellite retrievals with ground-based measurements of the same cloud. Comparisons are often attempted between a surface-based measurement at a fixed location over a long temporal period and satellite measurements that provide an instantaneous measurement over a wide area. While difficult and often creative, confidence in retrievals is often gained through painstaking comparison between the two. For some cloud properties, it may be possible to compare properties derived from two or more different satellite instruments. This will be one of the more active areas in future research. Acknowledgments The authors are grateful to several individuals for their assistance in the preparation of the diagrams in this article, particularly, Lei Bi (for Figure 4), Chao Liu (for Figure 5), Chenxi Wang (for Figures 6–8), and Chen Zhou (for Figure 2). Encyclopedia of Atmospheric Sciences, Second Edition, 2015, 116–127 Author's personal copy Satellites and Satellite Remote Sensing j Remote Sensing: Cloud Properties See also: Aerosols: Aerosol–Cloud Interactions and Their Radiative Forcing. Clouds and Fog: Classification of Clouds; Climatology; Contrails; Measurement Techniques In Situ; Lidar: Backscatter. Radiation Transfer in the Atmosphere: Cloud-Radiative Processes; Scattering. Satellites and Satellite Remote Sensing: Research. 127 Mishchenko, M.I., Hovenier, J.W., Travis, L.D. (Eds.), 1999. Light Scattering by Nonspherical Particles: Theory, Measurements, and Geophysical Applications. Academic Press, San Diego. Stephens, G.L., 1994. Remote Sensing of the Lower Atmosphere. Oxford University Press, Oxford. Wendisch, M., Yang, P., 2012. Theory of Atmospheric Radiative Transfer – A Comprehensive Introduction. Wiley-VCH Verlag GmbH & Co., KGaA, Weinheim, Germany. Further Reading Kidder, S.Q., Vonder Haar, T.H., 1995. Satellite Meteorology: An Introduction. Academic Press. Liou, K.N., 1992. Radiation and Cloud Processes in the Atmosphere. Oxford University Press, Oxford. Encyclopedia of Atmospheric Sciences, Second Edition, 2015, 116–127