Comparison of the seasonal change in cloud-radiative forcing from ...

JOURNAL

OF GEOPHYSICAL

RESEARCH,

VOL. 102, NO. D14, PAGES 16,593-16,603, JULY 27, 1997

Comparison of the seasonal change in cloud-radiative forcing from atmospheric general circulation models and satellite

observations

R. D. Cess,1 M. H. Zhang,• G. L. Potter,2 V. Alekseev,3 H. W. Barker,4 S. Bony? R. A. Colman,6 D. A. Dazlich,? A.D. Del Genio,8 M. Ddqu6,9 M. R. Dix, TM V. Dymnikov,3 M. Esch,TML. D. Fowler,7 J. R. Fraser,6 V. Galin,3 W. L. Gates,2 J. J. Hack,TMW. J. Ingram,13J. T. Kiehl,TMY. Kim,8 H. Le Treut,s X.-Z. Liang,TM B. J. McAvaney,6 V. P. Meleshko,ls J. J. Morcrette,•6 D. A. Randall,7 E. Roeckner,TM M. E. Schlesinger, •7 P. V. Sporyshev, •s K. E. Taylor,2 B. Timbal,9 E. M. Volodin? W. Wang,•7 W. C. Wang,TMand R. T. Wetheraid•8 Abstract. We compareseasonalchangesin cloud-radiativeforcing(CRF) at the top of the atmospherefrom 18 atmosphericgeneralcirculationmodels,and observations from the Earth Radiation BudgetExperiment(ERBE). To enhancethe CRF signaland suppressinterannualvariability,we consideronly zonal mean quantitiesfor which the extrememonths(Januaryand July), as well as the northern and southernhemispheres, have been differenced.Since seasonalvariationsof the shortwavecomponentof CRF are causedby seasonalchangesin both cloudiness and solarirradiance,the latter wasremoved.In the ERBE data, seasonalchangesin CRF are driven primarily by changesin cloud amount.The sameconclusionappliesto the models.The shortwavecomponentof seasonalCRF is a measureof changesin cloud amount at all altitudes,while the longwave componentis more a measureof upper level clouds.Thus important insightsinto seasonal cloudamountvariationsof the modelshavebeen obtainedby comparingboth components,as generatedby the models,with the satellitedata. For example,in 10 of the 18 modelsthe seasonaloscillations of zonalcloudpatternsextendtoo far polewardby one latitudinalgrid.

•Institutefor Terrestrialand PlanetaryAtmospheres, Marine Sci-

1.

Introduction

encesResearchCenter, State Universityof New York at StonyBrook.

Three-dimensionalgeneralcirculationmodels(GCMs) are the most comprehensiveclimate modelsfor projectingclimate 3Department of NumericalMathematics, RussianAcademyof Sci- changecausedby human activities.One of the greatestuncerences, Moscow. tainties associatedwith these models,however,is their ability 4CanadianClimateCentre,Downsview,Ontario. to simulate how climate-inducedchangesin cloudinesswill SLaboratoire de Mdtdorologie Dynamique, Paris. 6Bureauof MeteorologyResearchCentre,Melbourne,Victoria, impact a climate change projection; i.e., cloud feedback in whichcloudinesschangesmight amplify(positivefeedback)or Australia. 7Department of Atmospheric Science, ColoradoStateUniversity, diminish (negative feedback) a model's simulated climate Fort Collins. change. A broad range of cloud feedbackswas noted in a 8NASAGoddardInstitutefor SpaceStudies, NewYork. comparisonof 19 atmosphericGCMs [Cesset at., 1990], while 9Maltrio-France, CentreNationalde Recherches Mdtdorologiques, a more recent comparison[Cesset at., 1996] showeda more Toulouse, France. •øDivision of Atmospheric Research, Commonwealth Scientific and narrow difference,with most modelsproducingmodestcloud 2program for ClimateModelDiagnosis andIntercomparison, Law-

rence Livermore National Laboratory,Livermore, California.

Industrial ResearchOrganisation,Aspendale,Victoria, Australia.

•Max PlanckInstitutefor Meteorology, Hamburg,Germany. •2National Centerfor Atmospheric Research, Boulder,Colorado. •3HadleyCentrefor ClimatePrediction andResearch, U. K. Mete-

feedback. There were, however, substantial differences in the

longwaveand shortwavefeedbackcomponents,indicatingthat the modelsstill have physicaldisagreements.Clearly, there is a needto improveour understanding of cloud-climateinteractions. orologicalOffice, Bracknell,England. •4Atmospheric Sciences ResearchCenter,StateUniversityof New Althoughnot an analogfor long-termclimatechange,seasonal York at Albany. variationsof cloud-radiative forcingconstituteone meansof test•SVoeikov Main Geophysical Obseratory, St. Petersburg, Russia. ing cloud-climate interactions in GCMs and, perhaps more •6European Centrefor Medium-Range WeatherForecasts, Readimportantly,of providingphysicalinsightsinto suchinteractions. ing, England. •7Department ofAtmospheric Sciences, University of Illinois,Urbana. In this study, 18 atmosphericGCMs are compared with 18Geophysical FluidDynamics Laboratory, NationalOceanic andAt- seasonalvariations of cloud-radiativeforcing as determined mosphericAdministration,PrincetonUniversity,Princeton,New Jersey. from Earth Radiation Budget Experiment (ERBE) satellite data. Seasonal variations of the shortwave component are driven by seasonalchangesin both cloudinessand solar irraCopyright1997 by the American GeophysicalUnion. diance, and the latter is removed so as to isolate the impact of cloudinessvariations.The goal of this particularcomparisonis Paper number 97JD00927. 0148- 0227/97/97JD-00927509.00 to demonstratewhat canbe learnedby comparinga numberof 16,593

16,594

CESS ET AL.: CLOUD-RADIATIVE 3O

ß

i

ß

a

i

i

ß

i

',

COMPARISONS

a CrF = arc - av

i

(3)

1985

.... 1986

while for the SW component

/'_• --- 1987 ._,,:/'\\.......

. ... 1%

ß

--

LONGWAVE

2O

10

ß

JANUARY

FORCING

SW ACRF = (ac-

a)AS + (Aac-

+ (Sac-

S&-

Aa)S

Sac + Sa)

-10

It is important to recognizethat averagingis performed on fluxes and not albedos,so that the definitions of • and •c

-2O

followfrom 5• = aS and 5c• = acS. Thusthe above expressionreducesto

i 30b', ß, ßi ß, ßi ß JANUARY

20

A,,_ ,f,"

o

-90 •

• f i • /•':: . i

40N

'":..'X? /i

20N

0

20S

seasonalcloudinessvariability. Thus we delete this term and definethe SW ACRF as [Cesset al., 1992a]

"V. -..

,,..y % "

(4)

The first term in (4) is related solelyto seasonalvariabilityof the solar irradianceand containsno informationconcerning

-10

60N

SW ACRF = (ac- a)AS + (Aac- Aa)S

SHORTWAVE

-

40S

SW ACRF = (Aac- Aa)S

(5)

Note that S representsthe monthly mean solar irradiancefor the month in question, rather than the annual mean solar

60S

LATITUDE (DEGREES)

irradiance

S.

Figure l. Zonal mean and area-weighted(a) LW and (b) SW ACRF for Januaryof eachof the 4 years(1985 to 1988).

All quantitiesin (3) and (5) are availablefrom the ERBE monthlymean processeddata [Harrisonet al., 1990].The conventional data set consistsof roughly2 initial years of combineddata from the Earth RadiationBudgetSatellite(ERBS) GCMs with seasonalcloud-radiativeforcing data; it is not to whichwas in a 57ø orbit relativeto the equator,and NOAA 9, determine which models are the "best models." Models that whose orbit was Sun synchronouswith a 1430 LT equator maybe superiorto otherswhencomparedwith seasonalcloud- crossingtime. The NOAA 9 scannerfailed about 2 months radiativeforcingdata might not showsimilarsuperioritywhen after the launch of NOAA 10, whoseorbit was also Sun syncomparedwith other typesof data. chronousbut with an 0730 LT equatorcrossingtime. So there

2.

are roughlytwo monthsof combineddata from ERBS, NOAA 9, and NOAA 10, followedby about 2 yearsof combineddata

Seasonal Cloud-Radiative Forcing

The term "cloudy" is used to denote a domain containing both overcast-sky and clear-skyregions,followingRamanathan et al. [1989], while the term clear refers to an average of clear-sky regions within that domain. We employ monthly meantop-of-the-atmosphere (TOA) reflectedshortwave(SW) and emittedlongwave(LW) radiativefluxesasprovidedby the ERBE for 2.5ø longitude and 2.5ø latitude grids and for both cloudy and clear designations[Ramanathanet al., 1989;Harrisonet al., 1990]. With H representingthe net TOA radiative heating of the climate system

from ERBS and NOAA

10, at which time the NOAA

10

scannerfailed. The last (fifth) year of data is solelyfrom ERBS. All this raisesthe possibilityof artificial "interannual variability" causedby changesin satellite combinations.To avoidthis,we employdata solelyfrom ERBS. This imposesa restrictionto latitudeslessthan 60ø as dictatedby the ERBS orbit. However, even if this were not the case, the ERBE

overbar(e.g.,AS = S - •). The evaluation of the LW

clear-skysceneidentificationis not reliable over snowand ice [Nemesure et al., 1994],sothat highlatitudesshouldbe excluded. Regionalplotsof ACRF exhibitsubstantialinterannualvariability, a problem that is reduced by addressingonly zonal mean ACRF. However, care must be exercisedin performing the zonal averagingbecauseof missingclear-skygrid points that are due to cloudiness persisting over some regions throughoutan entire month.This will resultin missingACRF valuesfor those grid points, so that if ACRF is zonally averaged, the missinggrids result in biasesbecausethose grids containlarge amountsof clouds.A more accurateprocedureis to first evaluatezonal meansof F and a, noting for the latter casethat averagingthe albedo is equivalentto averagingthe flux becausethe monthly mean TOA insolationis effectively constantin the zonal direction. Unlike the clear quantities, there are no missinggrid values.Next, Fc and ac are zonally averagedwith missingclear-skygrids not being counted in either the numerator or denominatorwhen performing the averaging.This removesthe aforementionedbias associated with averagingACRF, becauseenhancedcloudinessover the missingclear-skygrid pointswould not bias the clear-skyaverages.Zonal averagesof LW and SW ACRF are then evalu-

componentof ACRF is straightforward,giving

ated from (3) and (5) usingthe zonalmeaninput.

H= (1-

a)S-F

(1)

where a, S, and F denote the albedo, solar irradiance, and

emitted LW radiation at the TOA, respectively. Cloudradiative forcing (CRF) refers the cloudy-skyH to that for clear skies,so that [Ramanathanet al., 1989]

CRF = (ac- a)S + (Fc- F)

(2)

where the subscriptc is used to denote clear-skyquantities. Positivevaluesof CRF indicatethat cloudsradiativelyheat the climate system,while negativevalues correspondto cooling. Since(Fc - F) is generallypositive,this representsthe LW greenhousewarmingcausedby clouds.Conversely,(ac - a) is generallynegative,and so the first term in (2) is a cooling due to SW reflection by clouds. To investigatethe seasonalvariationof CRF, we follow Cess et al. [1992a]and let A denotethe seasonalvariationof a given quantity about its annual mean value which is denotedby an

CESS ET AL.' CLOUD-RADIATIVE

FORCING

COMPARISONS

To be consistentwith the GCM simulations, as will be dis-

6o

cussedin the followingsection,only 4 of the 5 yearsof ERBE data are used,1985 through1988.JanuaryLW and SW ACRF resultsfor each of theseyears are shownin Figure la and lb, respectively.The ACRF valueshave been weightedby latitudinal area throughmultiplicationby the cosineof latitude.The

3o

latitudinal

variations

of LW

and SW ACRF

seasonal shifts of cloudiness with latitude.

16,595

ß

ß

are the result of

The features

that

dominatetropical latitudesare causedby the migrationof the intertropicalconvergence zone (ITCZ), relativeto the annual mean, into the summer hemisphere.This migration of enhancedcloudinessis responsiblefor the respectivepositiveand negativepeaksin LW and SW ACRF at roughlylatitude 10øS. There are two important pointsto note concerningFigure 1. The interannualvariability is substantial,even after zonal averaginghas been performed, and the magnitudeof the zonal

w

I-"T'

w

reversal of seasons. Themaximum signals, roughly60W m-2, arewellabovethe +_5.5 W m-2 (SW) and+_3.2W m-2 (LW)

ALL LATITUDES R = -0.86

w

-60 6O

,

I

,

I

'

I

ß

i

3O

signals aresmall,roughly 20W m-2. Differencing theextreme months(JanuaryminusJuly) and the southernand northern hemispheres(SH minus NH) amplifiesthe signaland suppressesinterannualvariability,asFigures2a and 2b show.This doubledifferencingis adoptedfor comparisonwith the GCMs. The JanuaryminusJuly differencingclearly representsa seasonalchange,while the SH minus NH differencingis to some extent an amplificationof the seasonalchangebecauseof the

-30

ee•R =-0.96

R=-0.98 %' ©•

w

I-"T' W

-3O

--e-- < 35DEGREES

•©

--o--ß35DEGREES

W

-60 -60

,

I

-30

,

I

0

•

I

30

oNq •

60

AREA-WEIGHTED LWbCRF(W m'2)

Figure 3. Scatterplotsof the SW ACRF versusLW ACRF resultsshownin Figure 2c, (a) for all latitudesand (b) subdi-

8O

vided at latitude

35 ø. R denotes the correlation

coefficient.

4O

-4O

-8O ". E

80

u.

40

,,,

0

w

-40

ß

I

ß

i

ß

i

'

b

i

ß

i

ß

SHORTWAVE

w

n- -•0 •0

.

i

,

i

.

I

,

I

,

I

ß

I

ß

I

ß

I

'

I

ß

I

C

_

ß

FOUR-YEAR MEANS

4O

uncertainties in the ERBE monthly mean regional data [Wielickiet al., 1995],andtheseuncertainties shouldbe evenless in the double-differencedzonal mean data we have adopted. The stronganticorrelationbetweenLW and SW ACRF (Figure 2c) suggests that both are governedprimarilyby seasonal changesin cloudamount.An increasein cloudamount,suchas the migration of the ITCZ into the summer hemisphere,simultaneouslyincreasesLW ACRF and decreasesSW ACRF. Figure 3a clearlydemonstratesthis anticorrelation.Each point representsthe SW and LW ACRF pairs,for 2.5ølatitudezones, from Figure 2c. This correlation is considerablyincreased whenthe data are separatedfor latitudesbelowand above35ø, as shownin Figure 3b. Thus with one exceptionthat will be discussed later, ACRF servesmainly as a measureof seasonal changesin cloud amount. There are, however,two caveatsthat applyto the discussion above.Correlatedopposite-sign changesin SW andLW ACRF couldbe causedby changesin cirrusoptical depth becausethe emissivityof cirrus cloudsis generallylessthan unity. An increasein cirrus optical depth would thus increasethe magnitudes of both SW and LW ACRF. Moreover, LW ACRF de-

ducedby ERBE dependsalso on changesin cirrus structure. For example, if the degree of horizontal variability of two cirrus clouds differed, but cloud fraction and optical depth --o-- SHORTWAVE were equal, the scenescould still yield significantlydifferent -80 0 10 20 30 40 50 60 valuesof LW CRF [Barkeret al., 1993]. Since GCMs assume LATITUDE(DEGREES NORTH & SOUTH) that all clouds are homogeneous,and therefore overlook Figure 2. (a) Zonal-meanand area-weightedJanuaryminus changesto LW CRF due to changesin cloud structure,it may Julyand SH minusNH LW ACRF for eachof the 4 years.(b) at times be incorrect to attribute differences between GCM The sameasFigure2a but for the SW. (c) Four-yearmeansof and ERBE valuesof LW ACRF simplyto differencesin cloud the LW ACRF from Figure2a andthe SW ACRF from Figure2b. fraction. -4O

ß

I

,

I

,

I

,

I

,

I

,

16,596

CESS ET AL.' CLOUD-RADIATIVE

FORCING

COMPARISONS

Table 1. List of GCMs Used in the PresentStudy Model

Investigators

Bureau of MeteorologyResearchCentre, (BMRC) CanadianClimate Centre (CCC) NCAR CommunityClimate Model, Version 2 (CCM2, CCM2A) NCAR CommunityClimate Model, Version 1; LawrenceLivermoreNational Laboratory(CCM/LLNL) Centre National de RecherchesM6t6orologiques(CNRM) CommonwealthScientificand IndustrialResearchOrganisation(CSIRO) ColoradoState University(CSU 95) Department of NumericalMathematicsof the RussianAcademyof Sciences(DNM)

McAvaney, Fraser, Colman Barker

Kiehl, Hack, Zhang, Cess Taylor Timbal and D6qu6 Dix

Randall, Dazlich, Fowler

Galin, Dymnikov, Volodin, Alekseev

GENESIS/State Universityof New York at Albany Max PlanckInstitutefor MeteorologyHamburg(ECHAM) EuropeanCentre for Medium-RangeWeather Forecasts,Cycle36 (ECMWF) GeophysicalFluid DynamicsLaboratory(GFDL) NASA Goddard Institute for SpaceStudies(GISS) Laboratoirede M6t6orologieDynamique(LMD) Main GeophysicalObservatory(MGO) Universityof Illinois at Urbana-Champagne(UIUC) Hadley Centre for Climate Predictionand Research,United KingdomMeteorologicalOffice(UKMO)

3.

GCM

Liang and W. C. Wang Roeckner

and Esch

Morcrette, Potter, Gates Wetheraid Del Genio

and Kim

Bony and Le Treut Meleshkoand Sporyshev Schlesingerand W. Wang Ingram

the form of GCM versus ERBE differences, and the root-

Simulations

mean-square(RMS) of thesedifferencesare shownin Figure6 The GCMs used in the present study are summarizedin for eachmodel. In general,with this signconventionthe modTable 1, and descriptions of mostof thesemodelsare provided els tend to underestimateLW ACRF in the tropicsand overby Phillips[1994],althoughin manyinstancesthe modelshave estimate it in midlatitudes(Figure 5a). Only CSU 95 and sincebeen updated. Two versionsof CCM2 have been used, GENESIS producepositivedifferencesfrom ERBE throughthe standard version (CCM2), and a modified version (CCM2A) in which cloud SW absorptionwas increasedas suggestedby some recent observationallybased studies[RalOO

manathan et al., 1995; Cesset al., 1995; Pilewskieand Valero, 1995] by modifyingthe cloud single-scatteringalbedo as de-

scribedby Kiehl et al. [1995]. The model-generatedLW and SW ACRF were from the last 4 years(1985-1988) of 10-year simulationsperformed as part of the AtmosphericModel IntercomparisonProject(AMIP), whichusedprescribedseasonally varyingsea surfacetemperaturesover the 10-yearperiod [Gates,1992]. For the purposeof comparingthe GCM results to ERBE, the ERBE LW and SW ACRF were interpolatedto the latitudinal grids of each GCM by using a cubic spline interpolation. The procedure for calculatingLW and SW ACRF is the sameas hasbeen describedfor the ERBE data, exceptfor the clear-skyfluxes.Method II [Cessand Potter,1987]wasadopted for the clear-skyflux evaluation,by which the clear fluxesare diagnosticallyevaluated at each grid by settingcloud amount to zero. Therefore there are no missingclear-skygrid pointsin the models. This samplingdiscrepancybetween the models and the ERBE data couldimpactthe interpretationof ACRF. Several studies have been conducted

to examine

alternate

•

60

20

-20

.

[

.

.........

i

ß

i

ß

i

.:::":::::... ß

,

ß

i

GCMs

ERBE

....•:•:;;::...........

f

-_. -.. -_-_. •_ ?_ -__.

,.... ...... -.-. '.'..........

-60 [

-lOO

?'"

a

,

f

,

f

,

i

,

ß

i

,

i

ß

i

ß

lOO

LONGWA I

,

i

i

ß

i

E 6O