Jan 27, 1997 - tially to the lack of surface forcing data to drive the models so these processes .... fractions, cloud-top heights, and ozone amounts. Clouds are .... personal communication, 1995)), and the carbon dioxide volume mixing ratio is ...
JOURNAL OF GEOPHYSICAL RESEARCH, VOL. 102, NO. D2, PAGES 1795-1806, JANUARY 27, 1997
A method to derive downwellinglongwavefluxesat the Arctic surface from TIROS operational vertical sounder data Jennifer A. Francis Instituteof Marine andCoastalSciences, RutgersUniversity,New Brunswick,New Jersey
Abstract. The dominantcomponentof the polar surfaceenergybudgetduringhalf the year is the ,-.,1 ..-.11:.,. ..... J: 1 .-.. ...I.,. ..+ •.,.,,.,+.; ,., •' .•1 JL,•x I'/, .... yet ':"• IlLtiC is Known aoOtlt its tClllpOlal' uuw,wcnmg nun m tongwaveradiation•,I. ;•,T ' ' :•pattaI and ....... variabilityexcepton monthlytimescales.As surfacemeasurements will alwaysbe sparse,the most promisingopportunityfor diagnosing. the DLF is providedby satellitedata.Estimatingthis flux from space,however,presentschallengesoverall surfacetypesandparticularlyin polar environmentswhereclouddetectionandcloudfractionestimationare lesscertain.A new methodis presentedto estimateDLF from measurements by the TIROS-N operationalverticalsounder(TOVS). Temperatureprofiles,humidityestimates,andcloudcoverareretrievedfrom TOVS radiancesusing the improvedinitializationinversionalgorithm,whichhasbeenmodifiedto producemore accurate resultsoversnowand seaice. This informationis combinedwith brightnesstemperaturedifferences from pairsof infraredandnear-infraredTOVS channels.Thesedifferencesare usedto infer cloud phaseandgeometricthickness.Longwavefluxesarethencalculatedusinga forwardradiativetransfer model. Resultsduringwinter 1988 and spring1992 are comparedwith hourlyradiationmeasurementsfrom the CoordinatedEasternArctic Experimentin the easternArctic basinand from the
LeadExperiment intheBeaufort Sea.Erroranalyses yieldabiasofapproximately 3 W m-2,a standarddeviation of23W m-2,andacorrelation coefficient ofabout 0.75.These errors arecomparable to resultsfrom similar studiesovermidlatitudeland and oceanareaswhere cloudsare more easily identified.
1. Introduction
mean fieldshavebeen derivedfrom availabledata (e.g., Vowinkel and Orvig, 1966, 1967; Khrol, 1992; Chernigovskiy,1964], but, to date, little is known of the spatialvariability on meso-or synoptic scales.Schweigerand Key [1994] computedsurfaceradia-
One of the highestpriorities in Earth sciencetoday is to developmodelsto simulatethe globalclimateand how it will changein responseto increasinggreenhouse gasconcentrations. tion fluxes in the Arctic from International Satellite Cloud Many global circulationmodels(GCMs) of varyingsophisticaClimatology Project (ISCCP) cloud data, but the accuracyof tion now exist,but theyexhibitlargedifferencesin theirrepresenthesefluxesin polar regionsis questionableowing to significant tationsof the present,let alonefuture,climate.One of the major
disagreementsbetweenISCCP cloud retrievalsand surface-based climatologies[Schweigerand Key, 1992]. If we are to improve our understandingof air-sea-iceinteractionprocessesand seaice evolution, with the ultimate goal of providing more realistic parameterizations for GCMs, estimatesof the surfaceenergybaltially to the lack of surfaceforcingdata to drive the modelsso anceare requiredon shortertime- and spacescales. theseprocesses can be studiedand morerealisticparameterizaThe net longwaveflux is the differencebetweenthe upwelling tionsdeveloped. broadbandinfrared flux from the surfaceand the downwelling During 5 monthsof the year when thereis little or no solar longwave flux (DLF) from the atmosphere.The upwelling flux radiationin the centralArctic basin,the surfaceenergybudgetis dependsonly on the surfaceskin temperatureand infrared emisdominatedby the net longwaveradiation(Figure 1). Recentsensivity of the surface according to the Stefan-Boltzmannlaw. sitivitystudies,e.g.,FischerandLemke[1994],Ebertand Curry While the ice surfacetemperatureis not easilymeasuredeitherin [1993], Thorndike [1992], and Makshtasand Timachev[1990], situ or from space,promisingtechniquesto providethis informasuggestthat annual-average ice thicknessin thermodynamic sea tion from satellitedataareunderdevelopment[e.g.,KeyandHaeice modelsis more sensitiveto perturbationsin longwaveradiafiiger, 1992; Yu et al., 1995]. The DLF, on the other hand, is tion fluxes than in shortwavefluxes. Measurementsof longwave affected by a number of atmosphericvariables:the temperature fluxes over sea ice, however,are limited to pointsin spaceor to and moisture contentof the atmosphere;cloud amount, optical shortperiodsin time, and mostrepresentonly the easternArctic [e.g., Marshunova,1961; Doronin, 1963; Fletcher,1965; Badg- depth,and baseheight;clear-airprecipitationor "diamonddust;" the almostubiquitousnear-surfacetemperatureinversions;and to ley, 1966; Donn and Shaw, 1966; Gavrilova, 1978]. Monthly a lesser degree, aerosols and greenhousegas concentrations. Problemsassociatedwith estimatingsurfaceradiationfluxesfrom satelliteobservationsare reviewedby Schmetz[ 1989, 1991], and Copyright1997by theAmericanGeophysical Union. for polar regionsin particularby Raschkeet al. [ 1992]. A model sensitivitystudyby Frouinet al. [ 1988] investigates the effectsof Papernumber96JD03002. 0148- 0227/97/96JD-03002509.00 perturbationsin variousatmosphericvariableson the DLF. Their sourcesof thesedifferencesis believedto be the parameterization schemesfor sea ice and its interactionwith the atmosphereand ocean.Theseparameterizations differ widely, partially owing to our deficientunderstanding of air-ice-oceaninteractionsandpar-
1795
1796
FRANCIS: LONGWAVE FLUXES AT THE ARCTIC SURFACE 6O
......
•olar' Flux//
40
/
•,
E
Sensible
20
•
Heat
0 ....:•,--•...a
/
....
/
---•.......•:•.........
• - 20'-'"'-'"•• Infrared Latent Hea(. _
-40
•
,
.....
Flux
ice or snow-covered surface.Frouin et al. [1988], for example, focuson oceanconditionsoff the Californiacoast.They testfour methodsfor estimatingcloudthickness thatemployvaryinglevels of sophistication; daylightis requiredfor all but one method. The four methodsperformsimilarly(error statisticsare shownin Table 1), probablybecausein this region, abundantmoisturein theboundarylayeris the dominantfactorcontrollingthe DLF. In an earlier studyby Darnell et al. [1986], operationalNational Oceanic and Atmospheric Administration (NOAA)/National Environmental
Satellite Data and Information
Service TOVS
retrievalsprovidetemperatureprofiles,moistureestimates,cloud Sep Oct Nov Dec Jan Feb Mor Apr May Jun fractions,cloud-top heights, and ozone amounts.Clouds are Figure 1. Estimatesof thenetsurfaceenergyfluxesin thecentral assumedto have a constantthicknessof 50 mbar, but later, this Arcticduringthemonthswhenthesurface is notmelting.(Radia- techniqueis modifiedto includeclimatologicalcloudthicknesses tion fluxes are from Marshunova[1961], turbulentfluxes are [Gupta et aL, 1992]. Resultsare validatedwith measurements from Doronin [1963], interpolatedto valueson the firstof each from four land sitesduringJuly 1981 to June1982. Table 1 commonthby Maykut [ 1978].) paresresultsfrom thesetwo studies,as well as statisticsfor fluxes
computedin the presentinvestigation. Finally,Wu and Cheng [1989] computeglobal DLF usingTOVS soundings from the temperature profiles(3 W m'2rms)andmoistureamount(_+40%) GoddardLaboratoryfor Atmospheres physicalretrievalalgofrom the TIROS operationalvertical sounder(TOVS) [Francis, rithm.Cloudbasesareestimated severalways,depending on the 1994] translateto flux errorsthat are within the requiredaccu- cloudtype.Their fluxesarenot comparedwith groundtruthdata; racy,ß +10 W m-2,statedby theWorldClimateResearch Program consequently, the errorsare unknown.The resultspresentedin (WCRP) [Raschke et al., 1990]. Our own model calculations Table 1 showthat fluxescomputed in the Arcticusingthe meth(Figure 2) demonstratethat fluxes computedfor the sub-Arctic odologydescribedin this paperhave similar,and often better, resultssuggestthat the typical errors over sea ice in retrieved
winter atmosphereare most sensitiveto errorsin cloud fraction and cloud optical thicknessbut are relatively insensitiveto the cloud-baseheight.Validationof TOVS-retfievedcloudamountin the Arctic is difficult, however,becausereliable ground-based measurementsof cloud amount are scarce.Uncertaintyis also introducedin comparingpointobservations to 100-kmareaaverages,as well as surface-observed fractionalcloud coverto satellite-retrieved effective cloud amount (cloud fraction times
emissivity).Nevertheless,we have comparedretrievedeffective cloud amount to available observations of cloud cover, and we
accuracythanthosecomputedunderlessambiguous cloudconditions.
Becauseof theneedfor informationaboutthesurfaceenergy budgetof the Arctic for climatechangeand air-sea-iceprocess studiesandbecauseof theuniquedifficultiesencountered in estimatingcloudparameters overa snow-icebackground, thisstudy was undertakento investigatenew methodsto derive the DLF
fromsatellitedataovertheArctic.Temperature andwatervapor retrievalsfrom TOVS havesufficientaccuracyfor computing
clear-skyfluxes[Francis,1994],butchallenges remainasto estimatingthe cloudamountandthickness. We attemptto charactereffectsof varying emissivity.Model calculationssuggestthat an ize bulkcloudproperties fromTOVS brightness temperatures Ts error of this magnitudewould resultin unacceptably largeerrors and to combinethis informationwith temperature andmoisture in DLF. Becauseof thesedifficultiesin validatingcloudamount retrievalsin a forwardradiativetransfermodelto computeArcdirectly,a moreusefulevaluationis achievedby comparing more tic-wideradiationfluxesat synopticspacescales andtimescales. reliablemeasurements of DLF at the surfaceto that computed accordingto the methoddescribedin thispaper,aserrorsin DLF will be largelydueto errorsin retrievedcloudamountandoptical 2. Validation Data estimate errors of _+30%. Note that this does not take into account
thickness.
Several investigatorshave computedinstantaneous surface
A recurringproblemin any studyof the Arctic environment, whetherit be ice surfacetemperatures or cloudcharacteristics, is open-oceanareas,but nonehasaddressed the problemovera sea the lack of groundtruthdata.At thistime thereare few measureradiation fluxes from satellite data over midlatitude land and
60
• 60 'a' i , , , i . . , i , , . I , , .
•, 60
E
[ 50 =
[ 50
• 40
40
x
=
40
(•
30
o
•
'-
30
30
o
o
c) 20
,
20
o
10
,T 10
0
0
,
20
o
10
x
o
0.0
o
0.2
0.4 0.6 0.8 Cloud Fraction
1.0
400
o
600
800
1000
CloudBase Height[mb]
o
2
4
6
8
lO
Cloud OpticalDepth
Figure2.Modeled sensitivity ofdownwelling longwave flux(DLF)to(a)cloud fraction, (b)cloud-base height, and(c)cloud opticaldepth. Cloudfraction andoptical depth fluxes arecompared toclear-sky values, andcloud-base height fluxes arecompared toa cloud with its base at 1000 mbar.
FRANCIS'
LONGWAVE
FLUXES
AT THE ARCTIC
SURFACE
1797
Table 1. DownwellingLongwaveFlux Computedin PresentandPreviousStudies Number of
Reference
Root-mean-
squareerror,
computationsWm'2
Frouin et al. [1988]
Bias,
Correlation
Wm-2
Coefficient
-- 100
Comments
Over ocean in autumn
Method A Method B Method C Method D Observations from
23 24 27 25 9
-6.5 - 13.3 -4.3 - 1.2 4.6
0.69 0.74 0.55 0.58 0.96
1027
32
-
0.78
122
21
-3
0.78
Oct.-Dec.
98
26
9
0.74
March-April 1992
off California coast, 1983 -- Mixed
Layer Dynamics Experiment
two ships Darnell et al. [1986]
Over land
Presentstudy
Over Arctic sea ice
Winter, no insolation
Spring,insolation
1988
mentsavailableof DLF alongwith the corresponding atmo- meter often missedlow-level ice cloudsall together,or else the sphericconditions fromthe centralArctic,especially in winter retrievedcloud-baseheightwas significantlydifferentfrom that whenDLF is mostimportant.Validationdatausedin this investi-
gationare from the followingtwo field experiments: the drift phaseof theCoordinated Eastern ArcticExperiment (CEAREX) andfromthe LeadExperiment(LeadEx).Duringthe CEAREX drift phase(mid-October throughDecember1988),rawinsondes
derived from the sounding.This uncertainty,combinedwith errorsintroducedby comparinga point,bottom-upmeasurement with a 100-km-average,top-downestimaterenderedthesedataof limited
value
for
validation
of
satellite-derived
cloud-base
heights.
werelaunched every12 hours,approximate cloud-base height (high,middle,low) and cloudtype were observed,and surface
meteorological conditions were recorded. Downwelling long- 3. Methodology wave fluxes were measured either on the ice surface or on the
researchvesselPolarbjOrn,driftingwith the seaice northeastof
Figure4 presentsa flow chartof the stepsfollowedin calculating the DLF. To computeDLF with an atmospheric radiative
Svalbard (approximately 82øN,35øE).An Eppleypyrgeometertransfer model, one requires numerouspieces of information: measuredbroadbandlongwaveradiationbetween4 and 50 gm. The datawere correctedfor dometemperatureand averagedover 10-minute intervals. Snow and frost were manually removed from the dome [CEAREXDrift Group, 1990]. For this study,an attemptwas madeto use only thosemeasurements immediately followinga cleaning.In somecases,suchasthatshownin Figure 3 for November29, the effectsof cleaningthe dome everyhour are significant and conspicuous.This source of error could account for as much as 20 W m '2 difference between individual
calculatedvaluesand observations, which are elevatedby dome frost. Instrument
• 300 • 05
from TOVS/3I'
Temperatureprofile Water vapordistribution Effective
cloud fraction
Cloud-toppressure Cloud "type"andphase
Cloud thickness
Mean cloudtemperature in 1 or 2 layers LWC/IWC o• cloudmeantemperature
L•ngwove Irrodionce 1
250
0g
Acquireretrievedvariables
errors other than those due to frost are estimated
to be approximately5 W m-2 [Lackmanet al., 1989]. Similar measurements were madeat the LeadEx ice campin the Beaufort Sea (approximately200 km northeastof Barrow, Alaska) from March24 to April 21, 1992 [LeadExGroup,1993].In addition,a ceilometermeasuredcloud-baseheightevery30 s; statisticswere compiledevery 15 min. Thesecloud-baseheightswerecompared with estimatesfromradiosonde dataat theLeadExbasecamp(O. Persson,personalcommunication,1996). He foundthat the ceilo-
E
atmospherictemperatureand humidity profiles,ozone amount,
Radiative transfer model
0•
09
12
15
18
21
00
71me
Figure 3. DLF measuredduringthe CoordinatedEasternArctic Experiment(CEAREX) on November29, 1988. Note the hourly fluctuationscausedby the cleaffingof frost from the radiometer dome.
DLF
Figure 4. Stepsin the procedure for determining DLF from TIROS operational verticalsounder (TOVS)retrievals andhigh resolution infraredsounder (HIRS) brightness temperatures.
1798
FRANCIS: LONGWAVE FLUXES AT THE ARCTIC SURFACE
cloud-topand cloud-baseheight,cloudphase,cloudliquid water content,droplet number and size distribution,aerosoloptical depth, and carbon dioxide concentration.In this study,ozone amounts
are taken from
the standard sub-Arctic
winter
atmo-
sphereprofilesby McClatcheyet al. [ 1971], the effectsof aerosols on infrared fluxes are neglected (maximum errors are estimatedto be 5 W m-2for a summerArctic clearsky (J. Key, personalcommunication,1995)), andthe carbondioxidevolume mixing ratio is constantat 350 ppm. 3.1 Information
From
Standard
TOVS/3I
Retrievals
Temperatureprofiles,layer-averagehumidityestimates,cloud amount,and cloud-toppressuresare retrievedfrom TOVS radiancesusingthe improvedinitializationinversion("3I") algorithm developedby the AtmosphericRadiationAnalysisgroupat the Laboratoire de M6t6orology Dynamique in Palaiseau,France [Chddin et al., 1985; Escobar, 1993; Stubenrauchet al., 1996]. The algorithm has been modified to improve resultsover snow
zontalresolutionof approximately100km x 100 km, includinga temperatureprofile, layer-averagedmoisturecontent,surface temperatureand microwaveemissivity,effectivecloudamount, cloud-topheight,andsurfacealbedo. In the first stepof the procedureto obtainDLF, retrievedtemperatureprofiles are read directly into a two-streamradiation model [Key, 1996]. Moisturecontent,which is retrievedin three thicklayers(1000 to 800, 800 to 500, and 500 to 300 mbar),is linearly interpolatedto the 39 layer averagesof the temperature profile.It is assumedthatthe retrievedvaluesrepresent the pressure-weightedcentersof their respectivethick layer. Layers above300 mbar are assignedmoisturevaluesaccordingto the standardsub-Arcticwinterprofile[McClatcheyet al., 1971]. It is assumed
that
variations
in retrieved
effective
cloud
amountcanbe interpretedas a completelyovercastsky,with all the variationoccurringin the emissivityof the cloud.Without ad:litional information, such as high-resolutionimagery, it is impossibleto separatethe contributionsto the variationfrom thesetwo factors.This shouldnot introducesignificanterrorsin and ice surfaces[Francis, 1994]. the Arctic, however,as cloudsare generallystratiformin nature. The TOVS sensorsystemhas flown continuouslyon NOAA This would reduce possible nonlinear relationshipsbetween polar-orbitingsatellitessince1978 and consistsof threeradiomebrightnesstemperatureand cloud fractionthat are a concernin ter arrays: high-resolutioninfrared radiation sounder(HIRS), sceneswith cumuluscloud types. microwavesoundingunit (MSU), andstratospheric sounding unit Retrievedcloud-toppressures areinputdirectlyinto the radia(SSU). Data from the SSU are not usedin this study.The HIRS tive transfermodel unlessfurther analysis(discussedin section measuresradiancesin 19 infraredand near-infraredwavelength 3.2.1) suggests the cloudis very low, in whichcasethe cloud-top bands(3.7 to 15 gm) andonevisibleband(0.7 gm) at a nadirresis set at 900 mbar for cloudsthat appearwarmerthanthe ice surolution of 17 km. The MSU measuresradiancesat four frequenface and at 950 mbar for situationswhich are identifiedas having ciesin the oxygenabsorptionbandbetween50 and 58 GHz at a diamond dust or blowing snow.These modificationsare made nadir resolutionof approximately110 km. The instrumentsscan becausethe 3I-retrieved cloud-top pressurefor low cloud is to 58ø on either side of nadir, creating a swath approximately sometimestoo low (cloudtoo high). Finally, the minimumpres2200 km wide, and fly over the Arctic basinabout 14 times per sureof retrievedcloud-topsis setto 300 mbar. day per satellite.Channelsfor TOVS were selectedaccordingto the absorptionand emissionpropertiesof particulargasesin the 3.2 Additional Cloud Characteristics From HIRS atmosphere. Becausepropertiesof thesegasesare relativelyconBrightnessTemperatures stantin spaceandtime (exceptwatervaporandozone),variations Oncethe appropriate retrievalsareobtainedfrom3I, thecloud in the signalcan be related,throughinverseradiativetransfer,to the averagetemperatureof the layer from which mostof the sig"type," phase, and geometricthicknessare estimatedusing nal emanates(assumingno clouds).The emittinglayer for each observedbrightnesstemperaturesin certain HIRS channels (Table 2). Clear air and cloud particlesinteractdifferentlywith channelis defined by the weighting function. See Smith et al. infrared radiation, and these interactionsalso vary with wave[ 1979] for more details. length,particlephase,andparticlesize.SeveralphysicalproperThe 3I retrievalsystemusesradiancesmeasuredby the HIRS for the varyinginteractions of cloudparticles andMSU to produceestimatesof geophysical variablesat a hori- tiesare responsible
Table 2. High ResolutionInfraredRadiationSounderChannelsUsedto Estimate Cloud Characteristics
Pair ofHIRS ChannelsWavelengths, gm
Night only;clear-skycheck,cloudtempera-
Windowpair, HIRS19
- HIRS8
3.7 - 11.1
ture structure,detect thin cirrus
Determinephaseof cloud,(cannotusewith
Phasepair, HIRS10
Purpose
- HIRS8
8.3 - 11.1
NOAA I 1)
Phasepair, HIRS 19 -HIRS
18
3.7 - 4.0
Estimate thickness of clouds
Thicknesspair, HIRS6 - HIRS15
13.7 - 4.47
HIRS7
with topsabove750 mbar Estimate thickness of clouds
Thicknesspair, HIRS14-
Day only,detectwaterclouds
4.52 - 13.4
with topsbelow750 mbar
FRANCIS: LONGWAVE FLUXES AT THE ARCTIC SURFACE
1799
with radiation, and these can be exploitedto infer certain bulk cloud characteristics from the HIRS measurements in the infrared
and near-infraredwavelengths.Figure 5 illustrateshow the absorptioncoefficientk•,s for ice and water varies with wavelength. The central wavelengthsof selectedHIRS channelsare superimposed. 3.2.1 Cloud type. Cloud type in this studyrefersto whether the cloud has a positivelapserate (temperatureincreaseswith increasingpressure)or whetherthe cloudhasan invertedtemperature structure(temperaturedecreaseswith pressure),as would be the casefor a cloudbelow the near-surfacetemperatureinversion.Figure 6 is a schematicof thesetwo cloud types.By using tllC the differencein TB betweentwo window cnmlnmb, .......... one in '"near-infrared(HIRS19, 3.7-gin wavelength)and one in the thermal infrared(HIRS8, 11.1 gin), the signof the cloudlapserate can be inferred. Hereafter, the differencebetween this pair of channelswill be referredto as the "window pair."Thesechannels were selectedbecausethey havenearlythe sameweightingfunctions(peaksat the surface)yet their sensitivityto cloudparticles is significantlydifferent.The differencein absorptioncoefficients (Figure5) resultsin cloudsbeingmoretransparent at 3.7 gm than at 11 gin, so that cloudswith a positivelapserate will contribute to a positivedifferencewhile cloudswith a negativelapse rate will havea small or negativedifference.If the window pair >1.5 K, the cloud is declaredto have a normal temperaturestructure. Becausethin cloudsproducelargepositivedifferencesin the window pair, this thresholdis alsousedas an addedtestfor cirrus.If the windowpair
