7
ATMOSPHERIC SCIENCE LETTERS Atmos. Sci. Let. 10: 115–121 (2009) Published online 20 April 2009 in Wiley InterScience (www.interscience.wiley.com) DOI: 10.1002/asl.218 A new IR technique for monitoring low cloud properties using geostationary satellite data Qingyuan Han, 1, * Jean-Louis Brenguier, 2 Kuo-Sen Kuo, 3 and Aaron Naeger 1 1 Department of Atmospheric Sciences, University of Alabama Huntsville, Huntsville, Alabama, USA 2 Meteo-France, CNRM-GAME, GMEI, Toulouse, France 3 Caelum Research Corporation and NASA Goddard Space Flight Center Greenbelt, Maryland, USA *Correspondence to: Qingyuan Han, Department of Atmospheric Sciences, University of Alabama Huntsville, Huntsville, Alabama, USA. E-mail: [email protected] Received: 4 September 2008 Revised: 16 January 2009 Accepted: 6 February 2009 Abstract A new technique of using satellite infrared radiance data for retrieving cloud properties is developed and applied to SEVIRI data, which is based on direct radiative transfer calculations, not on the emissivity approximation as used by other satellite IR only techniques. Instantaneous atmospheric profiles are used in the new technique for improving the accuracy of retrievals. Comparison of the retrieved results with coincident observations of CloudSat and CALIPSO shows excellent agreement for low clouds. This study shows that, using only IR radiances, the single layer assumption would significantly underestimate cloud optical depth when multilayered cloud system is presented. Copyright 2009 Royal Meteorological Society Keywords: low cloud; infrared; SEVIRI 1. Introduction Cloud feedback has long been recognized as the source of the largest uncertainty in climate sensitivity among Global Climate Models (GCMs). For example, the National Center for Atmospheric Research (NCAR) model and the Geophysical Fluid Dynamics Labora- tory (GFDL) model differ not only at current cloud distribution but also at cloud amount change for a warmer climate, leading to negative and positive cloud feedbacks, respectively (Bretherton et al., 2004). The GCM simulated cloud feedbacks are highly depen- dent on low-level cloud responses, which differ sig- nificantly among models (Bony and Defresne 2005; Wyant et al., 2006). Hence, improvement of cloud parameterizations has to address questions about how clouds form, evolve, and affect climate, which are dif- ficult to answer by the current global observation data (Stephens, 2005). Monitoring cloud evolution process requires continuous daytime and nighttime observa- tions over a certain region. However, over the past decades, most retrieval efforts have been focused on daytime clouds while the nighttime algorithms have received much less attention because the reflectance of visible channel is more sensitive to the cloud optical thickness than brightness temperature of IR channel. Among the major satellite cloud projects, the Clouds and the Earth’s Radiant Energy System (CERES) and the National Polar-Orbiting Operational Environmen- tal Satellite System (NPOESS) have nighttime algo- rithm developed for cloud property retrievals other than cloud masks (Minnis et al., 1998; Wong et al., 2007). Both algorithms are based on the concept of effective emissivity that was developed first for aircraft measurements of cirrus cloud height (Kuhn, 1963) and later developed for case study of cirrus cloud emit- tance (Platt and Stephens, 1980) and optical depth (Liou et al., 1990) using satellite data. The maximum retrievable optical thickness is about 6–8 (Liou et al., 1990; Minnis et al., 1998) because of the exponen- tial relationship between optical depth and emissiv- ity. The contribution of the atmosphere above cloud layer is neglected, which is valid for satellite retrieval of very high clouds (z > 10 km). If applied to low- level clouds, the retrieved effective emissivity often becomes unrealistic (>1) due to the significant con- tribution of the atmosphere above clouds. To char- acterize the contribution of atmospheric temperature and humidity above clouds, surface temperature and cloud-top temperature were included into the optical depth and emissivity relationship by multiple regres- sion analyses over different atmosphere profiles (Min- nis et al., 1998). The regression process is by nature an average analysis so that the coefficients in the formula, and the retrieval algorithm, are based on an implicitly averaged atmospheric profile. The limited range of optical thickness retrieval and the implicit assumption of an averaged atmospheric profile in the current IR retrieval algorithms introduce bias and uncertainty if applied to individual cloud pro- cess retrievals. The potential problem of multilayered cloud system is not addressed in these algorithms. We developed a new IR technique based on direct radia- tive transfer calculations to include the contribution of the atmosphere above clouds, which extends the range of the retrievable optical depth and enables the use of instantaneous atmospheric profiles for improv- ing retrieval accuracy. Comparison with coincident Copyright 2009 Royal Meteorological Society

A new IR technique for monitoring low cloud properties using geostationary satellite data

Embed Size (px)

Citation preview

Page 1: A new IR technique for monitoring low cloud properties using geostationary satellite data

ATMOSPHERIC SCIENCE LETTERSAtmos. Sci. Let. 10: 115–121 (2009)Published online 20 April 2009 in Wiley InterScience(www.interscience.wiley.com) DOI: 10.1002/asl.218

A new IR technique for monitoring low cloud propertiesusing geostationary satellite data

Qingyuan Han,1,* Jean-Louis Brenguier,2 Kuo-Sen Kuo,3 and Aaron Naeger1

1Department of Atmospheric Sciences, University of Alabama Huntsville, Huntsville, Alabama, USA2Meteo-France, CNRM-GAME, GMEI, Toulouse, France3Caelum Research Corporation and NASA Goddard Space Flight Center Greenbelt, Maryland, USA

*Correspondence to:Qingyuan Han, Department ofAtmospheric Sciences, Universityof Alabama Huntsville,Huntsville, Alabama, USA.E-mail: [email protected]

Received: 4 September 2008Revised: 16 January 2009Accepted: 6 February 2009

AbstractA new technique of using satellite infrared radiance data for retrieving cloud propertiesis developed and applied to SEVIRI data, which is based on direct radiative transfercalculations, not on the emissivity approximation as used by other satellite IR onlytechniques. Instantaneous atmospheric profiles are used in the new technique for improvingthe accuracy of retrievals. Comparison of the retrieved results with coincident observationsof CloudSat and CALIPSO shows excellent agreement for low clouds. This study showsthat, using only IR radiances, the single layer assumption would significantly underestimatecloud optical depth when multilayered cloud system is presented. Copyright 2009 RoyalMeteorological Society

Keywords: low cloud; infrared; SEVIRI

1. Introduction

Cloud feedback has long been recognized as the sourceof the largest uncertainty in climate sensitivity amongGlobal Climate Models (GCMs). For example, theNational Center for Atmospheric Research (NCAR)model and the Geophysical Fluid Dynamics Labora-tory (GFDL) model differ not only at current clouddistribution but also at cloud amount change for awarmer climate, leading to negative and positive cloudfeedbacks, respectively (Bretherton et al., 2004). TheGCM simulated cloud feedbacks are highly depen-dent on low-level cloud responses, which differ sig-nificantly among models (Bony and Defresne 2005;Wyant et al., 2006). Hence, improvement of cloudparameterizations has to address questions about howclouds form, evolve, and affect climate, which are dif-ficult to answer by the current global observation data(Stephens, 2005). Monitoring cloud evolution processrequires continuous daytime and nighttime observa-tions over a certain region. However, over the pastdecades, most retrieval efforts have been focused ondaytime clouds while the nighttime algorithms havereceived much less attention because the reflectance ofvisible channel is more sensitive to the cloud opticalthickness than brightness temperature of IR channel.

Among the major satellite cloud projects, the Cloudsand the Earth’s Radiant Energy System (CERES) andthe National Polar-Orbiting Operational Environmen-tal Satellite System (NPOESS) have nighttime algo-rithm developed for cloud property retrievals otherthan cloud masks (Minnis et al., 1998; Wong et al.,2007). Both algorithms are based on the concept ofeffective emissivity that was developed first for aircraft

measurements of cirrus cloud height (Kuhn, 1963) andlater developed for case study of cirrus cloud emit-tance (Platt and Stephens, 1980) and optical depth(Liou et al., 1990) using satellite data. The maximumretrievable optical thickness is about 6–8 (Liou et al.,1990; Minnis et al., 1998) because of the exponen-tial relationship between optical depth and emissiv-ity. The contribution of the atmosphere above cloudlayer is neglected, which is valid for satellite retrievalof very high clouds (z > 10 km). If applied to low-level clouds, the retrieved effective emissivity oftenbecomes unrealistic (>1) due to the significant con-tribution of the atmosphere above clouds. To char-acterize the contribution of atmospheric temperatureand humidity above clouds, surface temperature andcloud-top temperature were included into the opticaldepth and emissivity relationship by multiple regres-sion analyses over different atmosphere profiles (Min-nis et al., 1998). The regression process is by nature anaverage analysis so that the coefficients in the formula,and the retrieval algorithm, are based on an implicitlyaveraged atmospheric profile.

The limited range of optical thickness retrieval andthe implicit assumption of an averaged atmosphericprofile in the current IR retrieval algorithms introducebias and uncertainty if applied to individual cloud pro-cess retrievals. The potential problem of multilayeredcloud system is not addressed in these algorithms. Wedeveloped a new IR technique based on direct radia-tive transfer calculations to include the contributionof the atmosphere above clouds, which extends therange of the retrievable optical depth and enables theuse of instantaneous atmospheric profiles for improv-ing retrieval accuracy. Comparison with coincident

Copyright 2009 Royal Meteorological Society

Page 2: A new IR technique for monitoring low cloud properties using geostationary satellite data

116 Q. Han et al.

CloudSat and CALIPSO measurements show excellentagreement for low-level cloud-top heights. The resultsindicate that significant underestimation of opticaldepth could occur if multilayered clouds are retrievedas single layer cloud.

2. Method

The radiative transfer model used in this study isdescribed in Han et al. (1994). Radiance calculationsare performed for the four IR bands (3.9, 8.7, 10.8,and 12.0 µm) of the Spinning Enhanced Visible andInfrared Imager (SEVIRI) weighted by the relativespectral response function of each channel. To min-imize possible errors due to interpolation, 25 opticalthickness values at 0.55 µm ranging from 0.2 to 64were calculated for 15 vertical levels from 1 to 15 km.For low-level clouds, droplet size distributions are thesame as described in Han et al. (1994) with 14 effec-tive radius values from 2 to 32 µm. For high clouds,only one effective size of 37.3 µm corresponding tothe −40◦ cirrus (Han et al., 2005) was used to repre-sent ice clouds because the main focus of the study

is not on high cloud microphysics. Although the tech-nique is applied to 3 days of SEVIRI data at 15 minseparation over the west coast of central Africa, theresults shown later are based on the data of July 31,2006 0200UTC for comparison with the coincidentmeasurements of CloudSat and CALIPSO that passesthe same region at 0154UTC. The atmospheric pro-file over that region is from NCEP/NCAR reanalysisdata. One example of calculated brightness tempera-ture differences of different channel combinations asfunctions of optical thickness and cloud-top height isshown in Figure 1. The thickness of the horizontal barsat the left end represents the noise level of each band.Apparently, the retrievable optical depth is not lim-ited to 6 or 8 as the case by the approach of effectiveemissivity. The range and resolution of the retrievableoptical thickness are dependent on cloud height, view-ing zenith angle, particle sizes, and instrument noise.For example, in the case shown in Figure 1, the resolv-able optical depths for clouds at 1 km are: 0, 3.5, 14,and 64; for clouds at 2 km are: 0, 0.2, 0.5, 1, 4, 5, 6,7, 8, 10, 12, 15, 17, 20, 24, and 64; for clouds at 3 kmare: 0, 0.2, 0.5, 1, 1.5, 3, 3.5, 4, 5, 6, 7, 8, 9, 10, 11,13, 18, 22, and 64; for clouds at 4 km are: 0, 0.2, 0.5,

Figure 1. Example of calculated differences of brightness temperature (�BT) for the four IR channels of SEVIRI as functions ofoptical thickness and cloud-top heights.

Copyright 2009 Royal Meteorological Society Atmos. Sci. Let. 10: 115–121 (2009)DOI: 10.1002/asl

Page 3: A new IR technique for monitoring low cloud properties using geostationary satellite data

New IR technique for monitoring low cloud properties 117

1, 1.5, 2, 3, 3.5, 4, 5, 6, 7, 8, 9, 10, 11, 12, 14, 16, 20,and 28; for clouds at 5 km are: 0, 0.2, 0.5, 1, 1.5, 2,2.5, 3, 3.5, 4, 4.5, 5, 5.5, 6, 6.5 7, 7.5, 8, 8.5, 9, 9.5, 10,10.5, 11, 11.5, 12, 13, 14, 16, 20, and 24. The exam-ple shows that for clouds at 1 km, the resolution ofretrieved optical thickness is very low, which can onlyclassify cloud qualitatively as thin (τ = 3.5), medium(τ = 14), and very thick (τ = 64). For clouds at 2 kmand higher, the resolution improved significantly. Gen-erally, the maximum range of retrievable cloud opticaldepth using this technique is about 20–24 for cloudsabove 2 km with reasonable resolutions. If a globallyaveraged atmospheric profile is used in this case, theretrieved cloud amount would be significantly underes-timated, the cloud-top height would be underestimatedand the optical depth would be overestimated.

In this technique, the cloud-top height, optical thick-ness, and particle size are determined by minimizingthe differences between model and satellite values of�BT10.8–12.0 and �BT10.8–3.9, or �BT10.8–8.7 throughan iterative process. The instrument noise and theassumption of single layer cloud may introduce biasesand uncertainties in the retrieved results, which shouldbe evaluated by coincident, independent observations.

CloudSat and CALIPSO cloud profile data overthe same region at UTC 0154 on July 31, 2008,6 min earlier than the SEVIRI observation, were usedfor validation. The CloudSat data were ordered fromthe CloudSat Data Processing Center at the URLhttp://cloudsat.atmos.colostate.edu/data. The CALI-PSO data were obtained from NASA Langley ASDC.It is important to keep in mind the differences ofinstrument footprint sizes. The footprint of SEVIRIis a 2.8 × 2.8 km2, of CloudSat is an area of 1.4 kmwide and 2.5 km along the satellite subtrack, and ofCALIPSO is a 0.09 km diameter circle.

3. Results

Examples of the original 10.8 µm channel image, andretrieved cloud-top height and cloud optical depthare shown in Figure 2(a)–(c), respectively and theA-Train subtrack is marked by dashed white lines.Figure 2 shows a general trend of high-level clouds atthe northern part and low-level clouds at the southernpart of the scene. The results of optical depth areabnormally small at high cloud covered regions due tothe impact of multiplayer cloud that will be discussedin the next section.

For comparison, Figure 3 shows coincident observa-tions of cloud vertical profile over the same region byCloudSat (Figure 3(a)) and CALIPSO (Figure 3(b)).The general trend of cloud distribution shown inFigure 2(b) is also seen from the profiles of theseinstruments, suggesting qualitative agreement betweenresults from CloudSat/CALIPSO measurements andthe SEVIRI retrieval. However, the high cloud-tops(Z ∼ 15 km) around latitude −2◦ shown on the Cloud-Sat/CALIPSO data are retrieved as ∼5 km by this

technique. To reveal the details of the differences,point-by-point comparisons are conducted. Since thefootprint size of SEVIRI is the largest, the CloudSatand CALIPSO data were downgraded. Specifically, thecloud-top heights of CloudSat and CALIPSO are aver-aged if several pixels are within one grid cell of aSEVIRI data point so that for each SEVIRI data point,there is only one cloud-top height value for CALIPSOand for CloudSat.

Figure 4 shows the results of the cloud-top heightsby CloudSat (Figure 4(a)), CALIPSO (Figure 4(b)),and SEVIRI (Figure 4(c)) with 608 data points eachalong the A-Train subtrack. It is apparent that highcloud-tops around latitude −2◦ found by CloudSatand CALIPSO was retrieved by this technique as∼5–6 km, a significant underestimation. In order tounderstand the physical reason, the SEVIRI 10.8 µmchannel temperature T108 along the A-Train subtrackis also plotted on Figure 4(d). It reveals that, thehigh cloud-tops detected by CloudSat and CALIPSOmostly occur at scenes with warmer 10.8 µm temper-ature, which could be caused by multilayered cloudsand from horizontal inhomogeneity caused by smallerfootprints of the CALIPSO and CloudSat. The corre-lation coefficients between T108 and cloud-top heightsZCALIPSO, ZCloudSat, and ZSEVIRI are −0.21, 0.01, and−0.92, respectively, suggesting thin cirrus over warmsurface along the track. The existence of multiplayerclouds is indicated by small particle sizes retrieved(not shown) at regions where ZCALIPSO, ZCloudSat arelarge and T108s are warm.

To compare retrievals of low clouds, a simplethreshold of T108 > 290 K is applied to the SEVIRIdata. The difference of T108, �T108 < 1 K for adjacentpixels is used as a filter to remove scattered contami-nation of thin cirrus. Figure 5 shows the comparison ofcloud-top heights for the 74 data points along the trackcovering ∼200 km. The identical mean and standarddeviation of ZCALIPSO and ZSEVIRI are fortuitous con-sidering their differences at each data point. However,the excellent agreement between SEVIRI retrievedcloud-top heights and the A-Train measurements sug-gests that accurate low cloud property retrieval is pos-sible during cloud process if instantaneous (not globalmean) atmospheric profiles were used in the radiativetransfer calculations.

4. Problem of multiplayer clouds

The commonly used single layer assumption in satel-lite remote sensing may cause retrieval errors whenoverlapping multilayered clouds are present. Baumet al. (1994) pointed out that such assumption mightlead to the retrieved cloud heights being locatedbetween the high and low cloud layers. If visible bandis used (e.g., Huang et al., 2005), the retrieved opticaldepth by single layer assumption is close to the totaloptical depth of the two cloud layers. No error estima-tion of retrieved optical depth using only IR radiances

Copyright 2009 Royal Meteorological Society Atmos. Sci. Let. 10: 115–121 (2009)DOI: 10.1002/asl

Page 4: A new IR technique for monitoring low cloud properties using geostationary satellite data

118 Q. Han et al.

Figure 2. (a) SEVIRI 10.8 µm image of 2006 July 31, 0200UTC, (b) retrieved cloud-top height, and (c) retrieved optical depth.The white dash line is the ground track of CALIPSO and CloudSat at 0154UTC on the same day.

has been made for multilayered cloud system so far.In the following, we show that the retrieved opticaldepth would be significantly underestimated by usingonly IR radiances.

Figure 6 shows 10.8 µm radiances for three singlelayer high clouds (z = 10, 12, and 15 km represented

by green, blue, and purple curves, respectively) anda two-layer cloud system (red curve). The two-layercloud system is composed of a low-level cloud (1 km)with various optical depth covered by a high cloud at12 km with optical depth 3. By single layer assump-tion, the radiance of a low cloud with optical depth

Copyright 2009 Royal Meteorological Society Atmos. Sci. Let. 10: 115–121 (2009)DOI: 10.1002/asl

Page 5: A new IR technique for monitoring low cloud properties using geostationary satellite data

New IR technique for monitoring low cloud properties 119

Figure 3. Coincident observations of cloud profiles at UTC 0154, July 31, 2006 along the subtrack shown in Figure 2 by(a) CloudSat and (b) CALIPSO.

Figure 4. Cloud-top heights from (a) CloudSat, (b) CALIPSO, (c) SEVIRI, and (d) SEVIRI 10.8 µm band temperature along theA-Train subtrack shown on Figure 2.

Copyright 2009 Royal Meteorological Society Atmos. Sci. Let. 10: 115–121 (2009)DOI: 10.1002/asl

Page 6: A new IR technique for monitoring low cloud properties using geostationary satellite data

120 Q. Han et al.

Figure 5. Comparison of low cloud retrievals whenT108 > 290 K and �T108 < 1 K were applied to SEVIRI data(see text).

Figure 6. Radiances of three single layered clouds with differentaltitudes and a two-layer cloud system composed of a 1 kmcloud layer covered by a high cloud at 12 km with opticalthickness of 3.

10 overlapped by a high cloud with optical depth 3would be retrieved as optical depth of 4.9, 3.2, or 2.4for cloud at 9, 12, or 15 km, respectively. The finalsolution of specific cloud altitude and optical depth(and particle size) is determined by the best radiancefit in three IR bands. No matter what the final solu-tion could be, it is apparent that the retrieved opticaldepth would significantly underestimate the total opti-cal depth of the two-layer cloud system, which is 13in the case shown in Figure 6.

The impact of multiplayer clouds on retrievedoptical depth is apparent in Figure 2(c), in which theoptical depth values are suspiciously small at the vastregions with high cloud cover (cloud-top height largerthan 4 km in Figure 2(b)). This is consistent with themultiplayer cloud distribution shown in Figures 3 and4 and with the explanation shown in Figure 6. Thedifficulty of multiplayer cloud in satellite retrievals

has been addressed by using CO2 slicing techniqueto obtain information of upper cloud height (Baumet al., 1994) or by using cloud vertical profiles derivedfrom ground-based radar data (e.g., Huang et al.,2005). However, the spectral band for CO2 slicingis not available on GOES data that is necessary formonitoring cloud process.

The multiplayer cloud problem can be alleviated bycontinuously monitoring cloud properties, such as withGOES or SEVIRI data. For example, properties ofa single layer boundary cloud or a high-level cloudcould be retrieved accurately using this technique. If asingle layer high-level cloud gradually moves towardand finally overlap with a boundary layer cloud, thealtitudes of the two-layer cloud could be put intomodels to estimate cloud property change during theevolution of the two-layer cloud system. Even thougherrors may still exist, they would be much less thanthat by the single layer assumption.

5. Conclusions

A new IR technique suitable for monitoring low cloudprocess is developed and applied to SEVIRI data.The technique is based on accurate radiative transfercalculations with instantaneous atmospheric profilesat the time and region where cloud properties areretrieved. The retrieval results agree very well withcoincident CloudSat and CALIPSO measurementsfor low cloud heights. The new technique can beused to monitor low cloud formation and evolutionprocess, which is needed for improving low cloudparameterizations.

Multilayered cloud system may cause large errorsin retrieved cloud properties using only IR radiances.Continuous monitoring of cloud properties may helpto reduce the errors in the retrieval results.

Results of this paper are from applying the newtechnique on 3 days of SEVIRI data over ocean nearAfrica. More results and validation over other regions(e.g. over land) will help us to better understandthe capability and limitation of the new technique.Currently, the technique is applied to low clouds overARM site and the results will be reported later.

AcknowledgementsWe thank two anonymous reviewers for their very helpful com-ments and suggestions. A. Naeger is supported by the Atmo-spheric Radiation Measurement Program, Office of Science,U.S. Department of Energy, Grant No. DE-FG02-08ER64669.K. S. Kuo is supported by NASA TC4 program for thisstudy. Thanks for the support of Meteo-France during Q.Han’s sabbatical when the work was started. The SEVIRIdata were kindly supplied by Marcel Derrien. NCEP reanal-ysis data provided by the NOAA/OAR/ESRL PSD, from URLhttp://www.cdc.noaa.gov are acknowledged.

References

Baum BA, Arduini RF, Wielicki BA, Minnis P, Tsay SC. 1994.Multilevel cloud retrieval using multispectral HIRS and AVHRR

Copyright 2009 Royal Meteorological Society Atmos. Sci. Let. 10: 115–121 (2009)DOI: 10.1002/asl

Page 7: A new IR technique for monitoring low cloud properties using geostationary satellite data

New IR technique for monitoring low cloud properties 121

data: nighttime oceanic analysis. Journal of Geophysical Research99: 5499–5514.

Bony S, Dufresne J. 2005. Marine boundary layer clouds at the heart oftropical cloud feedback uncertainties in climate models. GeophysicalResearch Letter 32: L20806. DOI:10.1029/2005GL023851.

Bretherton CS, Ferrari R, Legg S. 2004. Climate process teams: a newapproach to improving climate models. U.S. CLIVAR Variations 2:1–6.

Han Q, Rossow WB, Lacis AA. 1994. Near-global survey of effectivedroplet radii in liquid water clouds using ISCCP data. Journal ofClimate 7: 465–497.

Han Q, Zeng J, Kuo K-S, Chen H, Smith E. 2005. Effect of particlesize distributions on the retrieval of ice cloud properties. GeophysicalResearch Letter 32: L13818. DOI:10.1029/2005GL022659.

Huang J, et al. 2005. Advanced retrievals of multilayered cloudproperties using multispectral measurements. Journal GeophysicalResearch 110: D15S18, DOI:10.1029/2004JD005101.

Kuhn PM. 1963. Measured effective long-wave emissivity of clouds.Monthly Weather Review 91: 635–640.

Liou KN, Ou SC, Takano Y, Valero FPJ, Ackerman TP. 1990. Remotesounding of the tropical cirrus cloud temperature and optical depthusing 6.5 and 10.5 µm radiometers during STEP. Journal of AppliedMeteorology 29: 716–726.

Minnis P, et al. 1998. Parameterizations of reflectance and effectiveemittance for satellite remote sensing of cloud properties. Journalof Atmospheric Science 55: 3313–3339.

Platt CMR, Stephens GL. 1980. The interpretation of remotelysensed high cloud emittances. Journal of Atmospheric Sciences 37:2314–2322.

Stephens GL. 2005. Cloud feedbacks in the climate system: a criticalreview. Journal of Climate 18: 237–273.

Wong E, Hutchison KD, Ou SC, Liou KN. 2007. Cirrus cloud toptemperatures retrieved from radiances in the NPOESS-VIIRS 8.55and 12.0 µm bandpasses. Applied Optics 46: 1316–1325.

Wyant MC, Khairoutdinov M, Bretherton CS. 2006. Climate sensitiv-ity and cloud response of a GCM with a superparameterization. Geo-physical Research Letter 33: L06714. DOI:10.1029/2005GL025464.

Copyright 2009 Royal Meteorological Society Atmos. Sci. Let. 10: 115–121 (2009)DOI: 10.1002/asl