Table Of ContentRemoteSensingofEnvironment152(2014)467–479
ContentslistsavailableatScienceDirect
Remote Sensing of Environment
journal homepage: www.elsevier.com/locate/rse
An approach for the long-term 30-m land surface snow-free albedo
retrieval from historic Landsat surface reflectance and MODIS-based
a priori anisotropy knowledge
YanminShuaia,⁎,JeffreyG.Masekb,FengGaoc,CrystalB.Schaafd,TaoHee
aEarthResourcesTechnology,Inc.,atBiosphericScienceLaboratory(Code618),NASA/GSFC,Greenbelt,MD20771,USA
bBiosphericScienceLaboratory(Code618),NASAGoddardSpaceFlightCenter,Greenbelt,MD20771,USA
cHydrologyandRemoteSensingLaboratoryUSDA,AgriculturalResearchService(ARS),Beltsville,MD20705,USA
dSchoolfortheEnvironment,UniversityofMassachusettsBoston,Boston,MA02125,USA
eDepartmentofGeographicalSciences,UniversityofMaryland,CollegePark,MD20742,USA
a r t i c l e i n f o a b s t r a c t
Articlehistory: LandsurfacealbedohasbeenrecognizedbytheGlobalTerrestrialObservingSystem(GTOS)asanessentialclimate
Received21May2014 variablecrucialforaccuratemodelingandmonitoringoftheEarth'sradiativebudget.Whileglobalclimatestudies
Receivedinrevisedform16July2014 canleveragealbedodatasetsfromMODIS,VIIRS,andothercoarse-resolutionsensors,manyapplicationsinhetero-
Accepted17July2014 geneousenvironmentscanbenefitfromhigher-resolutionalbedoproductsderivedfromLandsat.Wepreviously
Availableonlinexxxx developeda“MODIS-concurrent”approachforthe30-meteralbedoestimationwhichreliedoncombiningpost-
2000LandsatdatawithMODISBidirectionalReflectanceDistributionFunction(BRDF)information.Herewe
Keywords:
Albedoalgorithm presenta“pre-MODISera”approachtoextend30-msurfacealbedogenerationintimebacktothe1980s,through
Landsat anapriorianisotropyLook-UpTable(LUT)builtupfromthehighqualityMCD43ABRDFestimatesoverrepresen-
MODISBRDF tativehomogenousregions.EachentryintheLUTreflectsauniquecombinationoflandcover,seasonality,terrain
Forestdisturbance information,disturbanceageandtype,andLandsatopticalspectralbands.AninitialconceptualLUTwascreated
forthePacificNorthwest(PNW)oftheUnitedStatesandprovidesBRDFshapesestimatedfromMODISobserva-
tionsforundisturbedanddisturbedsurfacetypes(includingrecoverytrajectoriesofburnedareasandnon-firedis-
turbances).ByacceptingtheassumptionofagenerallyinvariantBRDFshapeforsimilarlandsurfacestructuresasa
prioriinformation,spectralwhite-skyandblack-skyalbedosarederivedthroughalbedo-to-nadirreflectanceratios
asabridgebetweentheLandsatandMODISscale.Afurthernarrow-to-broadbandconversionbasedonradiative
transfersimulationsisadoptedtoproducebroadbandalbedosatvisible,nearinfrared,andshortwaveregimes.We
evaluatetheaccuracyofresultantLandsatalbedousingavailablefieldmeasurementsatforestedAmeriFluxsta-
tionsinthePNWregion,andexaminetheconsistencyofthesurfacealbedogeneratedbythisapproachrespective-
lywiththatfromthe“concurrent”approachandthecoincidentMODISoperationalsurfacealbedoproducts.Using
thetowermeasurementsasreference,thederivedLandsat30-msnow-freeshortwavebroadbandalbedoyieldsan
absoluteaccuracyof0.02witharootmeansquareerrorlessthan0.016andabiasofnomorethan0.007.A
furthercross-comparisonoverindividualscenesshowsthattheretrievedwhiteskyshortwavealbedofromthe
“pre-MODISera”LUTapproachishighlyconsistent(R2=0.988,thescene-averagedlowRMSE=0.009and
bias=−0.005)withthatgeneratedbytheearlier“concurrent”approach.TheLandsatalbedoalsoexhibits
moredetailedlandscapetextureandawiderdynamicrangeofalbedovaluesthanthecoincident500-mMODIS
operationalproducts(MCD43A3),especiallyintheheterogeneousregions.Collectively,the“pre-MODIS”LUT
and“concurrent”approachesprovideapracticalwaytoretrievelong-termLandsatalbedofromthehistoric
Landsatarchivesasfarbackasthe1980s,aswellasthecurrentLandsat-8mission,andthussupportinvestigations
intotheevolutionofthealbedoofterrestrialbiomesatfineresolution.
©2014ElsevierInc.Allrightsreserved.
1.Introduction TerrestrialObservingSystem(GTOS)asoneoftheessentialclimatevar-
iables governing Earth's surface energy budget (Pinty et al., 2008;
Surfacealbedo,definedastheratioofradiantfluxreflectedfromthe Schaaf, Cihlar, Belward, Dutton, & Verstraete, 2009; Schaaf et al.,
Earth'ssurfacetotheincidentflux,hasbeendocumentedbytheGlobal 2008).Theradiativeforcinginterceptedbythelandsurfaceisperhaps
themostimportantinitialenergysourceforbiophysicalprocesses,
⁎ Correspondingauthor. through a further conversion into latent, sensible, and stored heat
E-mailaddress:[email protected](Y.Shuai). terms and input to the soil–vegetation biophysical system (Betts,
http://dx.doi.org/10.1016/j.rse.2014.07.009
0034-4257/©2014ElsevierInc.Allrightsreserved.
468 Y.Shuaietal./RemoteSensingofEnvironment152(2014)467–479
2000;Lyone,Jin,&Randerson,2008;Ollingeretal.,2008;Peckham,Ahl, Whileglobalclimatestudiescanutilizethecoarse-resolutionsurface
Serbin,&Gower,2008;Randersonetal.,2006;Sellers,Los,etal.,1996; albedodatasetsdescribedabove,thereremainsaneedforconsistent,
Sellers,Randall,etal.,1996;Zhangetal.,2009).Studieshaveshown fine-resolutionalbedoproductsforspecificapplications.Severalpubli-
thatlandcoverchange(andecosystemdisturbance)mayhaveasignifi- cationshavehighlightedtheimportanceoflandcoverchange,including
cantinfluenceonregionalalbedo,andhencelong-termclimateforcing deforestation,afforestation,agriculturalexpansion,urbanization,and
(Balaetal.,2007;Betts,2000;Claussen,Brovkin,&Ganopolski,2001; otherhuman-inducedlandsurfacealteration,totheterrestrialcarbon
Randersonetal.,2006).Terrestrialalbedovariesenormouslyinspace cycle and climate changes (Goward et al., 2008; Masek & Collatz,
andtimeasaresultofbothnaturalevents(e.g.weatherdisaster,insect, 2006;Panetal.,2011;Randersonetal.,2006).However,spatialresolu-
disease,wildfire,season-shifts,andvegetationphenologicalphase)and tionscoarserthan250-mmaybeinsufficienttocapturepatch-scale
humanactivities(e.g.forest-thinning&clearing,crops-sowing&harvest- vegetationchangesassociatedwithhumanlanduseandforestdistur-
ing,urbanization,andotherlandusemanagementmethods)(Jin&Roy, bance(TownshendandJustice1988;Maseketal.,2013).Fineresolution
2005;Ju,Roy,Shuai,&Schaaf,2010;O¡¯Halloran,etal.,2011;Shuaiand imagery(~30morbetter)canmoreaccuratelyquantifytheareasand
Schaaf,2010;Shuai,Schaaf,etal.,2013;Shuai,Xie,Wang,&Wu,2013; rates of these anthropogenic land changes. In addition, for climate
Xuetal.,2013).Asstrategiesemergeformanagingecosystemcarbonin changeinvestigations,longtimeseriesofalbedoproductsarerequired.
ordertomitigateglobalwarming,severalstudieshavepointedoutthe Althoughoperationalalbedodatasetscoveringthelast30yearshave
potentialriskofignoringthephysicalconsequencesoflandcoverchange, been assembled from different sensors covering different periods,
includingchangestolandsurfacealbedo(Betts,2000;Lyoneetal.,2008; themergingofmultiplerecordsraisesissuesofdataconsistencyand
Peckhametal.,2008;Randersonetal.,2006). quality.Becauseofthedifferencesamongsensors(wavelengthofspectral
Albedodatasetshavebeenderivedfromexistingcoarse-resolutionsatellite bands,orbitgeometry,spatialresolution,andgeographicregion),thede-
sensorstoparameterizegloballandsurfaceandclimatemodels.Compared rivedalbedoproductsmaydifferdependingonthespecificproduct,the
withprevioussingle-anglemodels,modernalbedoalgorithmsrelyonmultiple datasource,andtheproductionstrategies(Schaafetal.,2009).Therefore,
directionalreflectancemeasurementstofirstestimateaBi-directionalReflec- datasetsderivedfromasinglecontinuousacquisitionprogramoffersa
tanceDistributionFunction(BRDF)modelofthetarget,thenintegrateoverin- greaterpotentialforconsistencyindataquality.Despitedifferencesin
cidentandviewhemispherestocalculatealbedo.Studieshaveconcludedthat sensordesignovertime,theLandsatprogramhasacquireda42-yearre-
relativeerrorscanreachupto45%withouttheconsiderationofdirection/ cordofEarthObservationsthatcapturedgloballandconditionsanddy-
angleeffectsinthealbedoestimation(Kimes&Sellers,1985;Kimes,Sellers, namicsthroughsixsuccessfulmissionssince1972.Withthelaunchof
&Newcomb,1987).Becausemostsatellitesensorscannotcollectmultipleob- Landsat-8inFebruary2013(Loveland&Dwyer,2012),thisrecordhas
servationsofatargetinasinglepass,thesequentialaccumulationofdataover thepotentialofreaching50years.TheopeningoftheLandsatar-
multipledays(forsun-synchronousorbit)ormultiplehours(geostationary chiveforfreedistributioninlate2008hasinvigoratedthepushfor
orbit),maybeadoptedasarelevantsolutiontoachievemulti-anglemeasure- creatinglong-termbiophysicalandlandcoverproductsfromnew
mentssamplingthefullsun–target–sensorgeometry.Globalsurfacealbedo andarchivedLandsatdata(Woodcocketal.,2008;Wulder,Masek,
hasbeenmappedfromtheAdvancedVeryHighResolutionRadiometer Cohen,Loveland,&Woodcock,2012).Itincludesthisefforttodevel-
(AVHRR)(Csiszar&Gutman,1999;Key,Wang,Stroeve,&Fowler,2001), op the long-term, consistent surface albedo products from the
EarthRadiationBudgetExperiment(ERBE)radiometerdata(Li&Garand, Landsatprogram.
1994),andtheAlongTrackScanningRadiometer(ATSR).Withtheadvent Inapreviousstudy,wedevelopeda“concurrent”approachforgen-
ofroutinealbedoproductsretrievedfromPolarizationandDirectionalityof erating30-mresolutionalbedoproductsforthepost-2000(MODIS)era
the Earth's Reflectances (POLDER-I and II) (Bicheron & Leroy, 2000; bycombiningLandsatsurfacereflectancewithMODISsurfaceanisotro-
Hautecoeur&Leroy,1998;Leroyetal.,1997;Maignan,Breon,&Lacaze, pyinformation(Shuai,Masek,Gao,&Schaaf,2011).Inthisstudy,we
2004),Multi-angleImagingSpectroRadiomenter(MISR)(Martonchik,Pinty, proposeandvalidateanewapproachtogenerateLandsatalbedoprod-
& Verstraete, 2002; Martonchik et al., 1998), Clouds and the Earth's uctsforthepre-MODISera,byusingalbedo-to-nadirreflectanceratios
RadiantEnergySystem(CERES)(Rutanetal.,2009),MeteosatVisi- (Shuaietal.,2011)andanapriorianisotropyLook-UpTable(LUT)
ble and Infrared Imager (MVIRI)/Meteosat and Meteosat Second thathasbeenbuiltupfromthehighqualityMCD43ABRDFretrievals
Generation (MSG) (Carrer, Roujean, & Meurey, 2010; Geiger, over representative homogeneous regions. This approach yields
Carrer,Franchisteguy,Roujean,&Meurey,2008;Pintyetal.,2000), bothspectralandbroadbandalbedos,andaqualityassessment(QA)
SPOT4/VEGETATION (Franchistéguy, Geiger, Roujean, & Samain, mapbasedonthequalityofMODISanisotropyandLandsatsurfacere-
2005),andtherecentlylaunchedVisibleInfraredImagerRadiometer flectance.Inthispaper,wefirstaddressthetheoreticalbasisofthe
Suite(VIIRS)(Justiceetal.,2013;Liang,Yu,&Defelice,2005),albedo “pre-MODIS-era”LUTapproach,creationoftheBRDF-LUT,andthen
mapswithspatialresolutionsof500-mtotensofkilometerandtempo- demonstrateitsapplicationovermorethan100Landsatscenesinthe
ralfrequenciesofdailytomonthlyarenowavailabletoserveforclimate PacificNorthwest oftheUnitedStateswheresimultaneous ground
modelrefiningandinter-annualexploration(Schaafetal.,2008). measurementsareavailableforvalidation.
2.Albedodefinition
ThespectralDirectional–HemisphericalReflectance(DHR)ofaplanesurfaceisdefinedastheratioofradiantenergyscatteredupwardfromthe
surfaceinalldirectionstothedown-wellingincidentirradianceonthesurfacewithinthetargetspectrumregime(λ ,λ ).Itequalstheintegralofthe
1 2
BRDFovertheviewhemisphereforanincidentbeamatagivenwavelength,asshowninformula(1).Undertheextremeconditionthatnodiffuse
radiationbutonlythedirectbeamarrivesfromthesolarincidenceangle(θ,φ)definedbyzenithangleθ,andazimuthangleφ(L(θ,φ)),thealbedois
referredtoas“Black-SkyAlbedo”(BSA)Rðθ;φ;λÞintheMODISproductseries(Lucht,Schaaf,&Strahler,2000;Strahleretal.,1999).Undertheas-
i i
sumptionthatallirradianceisisotopic(purelydiffuseskylight),afurtherintegraloverilluminationhemisphereprovidestheBi-HemisphericalRe-
flectance(BHR)RðλÞ,or“White-SkyAlbedo”(WSA)formulae(2)and(3)(Luchtetal.,2000;Strahleretal.,1999).ThespectralBHRunderactual
atmosphericconditions(knownasthe“blue-skyalbedo”,or“actualalbedo”)canbeapproximatedthroughalinearcombinationofBSAandWSA,
weightedbythefractionofactualdirecttodiffuseskylight(Lewis&Barnsley,1994;Luchtetal.,2000;Románetal.,2010).Becausetheupwelling
radiancedependsonnotonlytheBRDFpropertiesoftheobservedsurface,butalsoatmosphericconditions,RðλÞmaychangewiththevariation
oftheinstantaneouscloudcoverandaerosolloading,aswellasoverthecourseofthedayasthesolargeometrychangesevenforconstantatmo-
sphericandsurfaceconditions(Luchtetal.,2000).Inaddition,multiplescatteringbetweensurfaceandatmosphereaffectstheangulardistribution
Y.Shuaietal./RemoteSensingofEnvironment152(2014)467–479 469
ofskyradiance.Therefore,bi-hemisphericreflectance(i.e.albedo)isnotatruesurfaceproperty,butratherafunctionofsolarbeamdirection,
atmosphericstate,andsurfaceanisotropicfeatures.
Z2ππZ=2
Rðθ;φ; λÞ¼ f ðθ;φ;θ ;φ ;λÞcosθ sinθ dθ dφ
i i r i i v v v v v v
0 0 ð1Þ
Z2ππZ=2
1
¼ Rðθ;φ;θ ;φ ; λÞcosθ sinθ dθ dφ
π i i v v v v v v
0 0
whereRðθ;φ; λÞ=Spectralblack-skyalbedo(Directional–HemisphericalReflectance,DHR)asafunctionofthesolarincidenceangle(θ,φ)
i i i i
(Strahleretal.,1999),andf(θ,ϕ;θ,ϕ;λ)=BidirectionalReflectanceDistributionFunction(BRDF)describingthebehaviorofsurfacescattering
r i i v v
asafunctionofaparallelincidentbeamfromonedirection(θ,ϕ)intheilluminatinghemisphereintothereflecteddirection(θ,ϕ)intheviewing
i i v v
hemisphere,ataparticularwavelengthλ.FurtherelaborationispresentedinNicodemus,Richmond,Ginsberg,andLimperis(1977)andSchaepman-
Strub,Schaepman,Painter,Dangel,andMartonchik(2006).Theterms“BRDF”and“anisotropy”inthispaperrefertothisunderlyingproperty.
Z Z (cid:2)Z Z (cid:3)
2π π=2 2π π=2
f ðθ;φ;θ ;φ ;λÞcosθ sinθ dθ dφ Lðθ;φ;λÞcosθ sinθdθdφ
r i i v v v v v v i i i i i i i
RðλÞ¼ 0 0 0 0 Z Z
2π π=2
Lðθ;φ;λÞcosθ sinθdθdφ
i i i i i i i
Z Z 0 0 ð2Þ
2π π=2
Rðθ;φ;λÞLðθ;φ;λÞcosθ sinθdθdφ
i i i i i i i i i
¼ 0 Z0 Z
2π π=2
Lðθ;φ;λÞcosθ sinθdθdφ
i i i i i i i
0 0
Z2ππZ=2
1
R ðλÞ¼ Rðθ;φ;λÞcosθ sinθdθdφ ð3Þ
WSA π i i i i i i
0 0
Strictlyspeaking,fornaturaltargets,BRDForanyothernominaldirectional-relatedmetricisnotameasurablequantity,asitrequiresperfectly
collimatedbeamsofilluminationandobservation,whileactualsunlightispartlydiffuseandthemeasurementsinvolveconicalgeometries.Thus,in-
dividualsatellitemeasurementprovidesonlyanapproximationofthedirectionalreflectance.
Formostoftheapplicationsinvolvingenergybalance,thereflectancequantityofinterestisnotthespectralreflectancebutratherreflectancein-
tegratedoverabroadspectralinterval(λ ,λ ),seeformula(4),tocapturetheoverallradiativeforcing.Thespectralintegralsforthehemispherical
1 2
reflectancearefunctionsofthedown-wellingsolarspectrumasdefinedintheaboveformulae.Thevisibleregime(0.3–0.7μm)knownas
photosynthetically-activeradiation(PAR)isofspecialinteresttocarboncyclemodelersfortheestimationofcarbonfixationviaphotosynthesis
(Dorman&Sellers,1989).Incontrast,thetotalshortwaveregime(0.3–3.0μm),aswellasvisibleandnear-infraredbands,aretypicallyrequired
bysurfaceenergybalancestudies.Notethatthegenericterm“albedo”,withoutanyspecificationofthesun-viewgeometryandintegralwavelength,
oftenimpliesthebi-hemisphericbroadbandalbedoofthewholesolarirradiancedomain.
Z (cid:2)Z Z (cid:2)Z Z (cid:3) (cid:3)
λ2 2π π=2 2π π=2f ðθ;φ;θ ;φ ;λÞcosθ sinθ dθ dφ Lðθ;φ;λÞcosθ sinθdθdφ dλ
r i i v v v v v v i i i i i i i
Rðλ →λ Þ¼ λ1 0 0 0 0Z (cid:2)Z Z (cid:3) ð4Þ
1 2 λ2 2π π=2Lðθ;φ;λÞcosθ sinθdθdφ dλ
i i i i i i i
λ1 0 0
3.Algorithm Landsat Thematic Mapper and Enhanced Thematic Mapper Plus
(TM/ETM+) L1T, and remove pixels contaminated by cloud and
Theinitialimpetustodevelopa“pre-MODISera”approachflowed snow from further analysis. The aim of “BRDF-LUTs creation” is to
fromthedesiretounderstandalbedoconsequencesofspecifictypes build up the a priori anisotropy information for each defined land
offorestdisturbanceandrecoveryatafineresolution.Toextend30-m surfacecategoryfromtheoperationalMODIS500-mhighqualityan-
surfacealbedogenerationintimebacktothe1980s,wefirstbuild isotropyproducts(i.e.MCD43A)overrepresentativehomogeneous
an a priori anisotropy Look-Up Table (LUT) from the high quality landsurfacestructureregions.Theaimof“Landsatalbedogeneration”
MCD43A BRDF estimates over representative homogenous regions, istoobtainthenarrow-bandspectralalbedobycombiningLandsatdi-
thencalculate thealbedo-to-nadirreflectance ratios foreachentry rectionalsurfacereflectancewiththespecificapriorianisotropyinfor-
andapplytheseratiostothe30-mLandsatnadirreflectance.Finally, mationstoredintheBRDF-LUTs,andthenconvertnarrowtobroad
weusenarrow-tobroad-bandconversionfactorstoderivebroadband bandalbedosforthevisible(0.3–0.7μm),NIR(0.7–3.0μm),andshort-
Landsatalbedos.Fig.1outlinestheoverallworkflowofthisapproach wave(0.3–3.0μm)regimes.
intothreemainfunctionalcomponents:surfacereflectancecalculation
andassessment(Fig.1A),BRDF-LUTcreation(Fig.1B),andLandsatsur- 3.1.Surfacereflectanceassessment
facealbedogeneration(Fig.1C).Theaimof“surfacereflectancecalcula-
tionandassessment”istoretrieveterrainandatmospherecorrected LandsatsurfacedirectionalreflectanceateachspectralbandofThe-
surface reflectance(thatisdefined in Masek etal.,2006)fromthe maticMapperandEnhancedThematicMapperPlus(TM/ETM+)has
470 Y.Shuaietal./RemoteSensingofEnvironment152(2014)467–479
Fig.1.Flowchartofthe“pre-MODISera”LUTapproachcomposedofthreefunctionalcomponents(A.surfacereflectanceassessment;B.BRDF-LUTcreation;andC.Landsatalbedo
generation).
been produced from orthorectified Landsat level 1 T raw images directlyfrommultipleLandsatdirectionalreflectanceobservations.In-
downloadedfromUSGSEROS,usingtheLandsatEcosystemDisturbance stead,weneedtoobtaintargetBRDFestimatesfromothersources,
AdaptiveProcessingSystem(LEDAPS)(Maseketal.,2006).Thelevel1 suchasMODIS,orMISR.
rawradiometrydataingestedbyLEDAPSwerecalibratedtoat-sensor Forthisstudy,theCollectionV005MODIS8-dayanisotropydataset
radiance,convertedtotop-of-atmospherereflectance,andthenatmo- (MCD43A)wasusedtocreatetheBRDF-LUTbecauseofitswiderange
sphericallycorrectedtosurfacereflectanceusingthesecondsimulation ofsunandviewangles,thebroadspectralcoverageofMODISforsimul-
ofsatellitesignalinthesolarspectrum(6S)model(Maseketal.,2006; taneousatmospherecorrection,frequentacquisitionforthepotential
Vermote,Saleous,&Justice,2002;Vermoteetal.,1997).LEDAPSdem- dailyadjustmentofBRDFretrieval,the500-mmoderateresolution,
onstratedgoodperformancethroughcomparisonswithground-based andespeciallythecontinuityofglobalproductssince2000.Theopera-
AERONETopticalthicknessmeasurements(Maseketal.,2006),concur- tionalMODISalbedoandreflectanceanisotropyproductsmakeuseof
rentMODISTerrareflectance(Fengetal.,2012;Maseketal.,2006),and thekernel-driven,linearalgorithmthatreliesontheweightedsumof
otherapproachesforLandsatsurfacereflectancegeneration(Ju,Roy, anisotropicandtwoadditionalkernels(respectivelycalledRoss-thick
Vermote,Masek,&Kovalskyy,2012).Tomitigatethecloudeffecton andLi-sparse-reciprocalmodels,RTLSR)ofviewingandilluminationge-
thesurfaceradiometricaccuracy,pixelscontaminatedbycloud,cloud ometrytoestimatetheBRDFmodel(Li&Strahler,1992;Luchtetal.,
shadow,andadjacentcloudswerescreenedfromthisstudyusingthe 2000;Ross,1981;Roujean,Leroy,Podaire,&Deschamps,1992).There-
LEDAPS-derivedcloudmask.Anadditionalscreeningforsnowwasper- trievedkernelweights(alsocalledBRDFmodelparameters)arethose
formedbasedontheoperationalMODISsnowmappingalgorithm(Hall, thatbestfitanadequateangularsampleofthehighqualitycloud-
Riggs,Salomonson,DiGirolamo,&Bayr,2002),throughtheNormalized cleared,atmosphericallycorrectedsurfacereflectancesavailablefor
DifferenceSnowIndex(NDSI)calculatedfromreflectanceatLandsat eachpixelovera16-dayperiod(Luchtetal.,2000;Schaafetal.,
green(0.53–0.61μm)andshortwaveinfra-red(1.55–1.75μm)bands. 2002,Schaaf,Liu,Gao,&Strahler,2011;Shuai,Schaaf,Strahler,Liu,
Furtherthresholdsforgreenbandreflectance(N0.10)andNDVIwere &Jiao,2008;Shuai&Schaaf,2010).Thismodelcombinationhas
applied to reduce the erroneous classification of very dark targets beenshowntobewell-suitedtodescribethesurfaceanisotropyof
(suchasblackspruceforests),aswellasthethermalmasktoeliminate thevarietyoflandsurfacesdistributedworldwide(Privette,Eck,&
thespurioussnowcoverpossiblyinducedbyresidualcloudcover,aero- Deering,1997).TheabsoluteaccuracyofMCD34Aalbedoatlocal
sol effect and snow/sand confusion on coastlines (Hall, Riggs, & solarnoon(LSN)derivedfromtheestimatedBRDFmodelhasbeen
Salomonson,1995;Halletal.,2002). establishedbycomparisonwithgroundmeasurementsfromavail-
ableinternationalBaselineSurfaceRadiationNetwork(BSRN)and
3.2.BRDFLook-UpTables Fluxnetsites(Cescattietal.,2012;Románetal.,2009;Wangetal.,
2014).Thisalgorithmassumesthatthelandsurfacedoesnotexperi-
Themostdirectwaytoobtainanisotropyinformationofanyland encesignificantstructuralchangesduringthe16-dayobservation
surfacetargetatthepixelscaleistocollectarepresentativesampleof period,whichisreasonableexceptincircumstancesofabruptdistur-
reflectanceobservationsatmultipledirections,overashortintervalof banceorconversion.
time.However,becauseofthenarrowfieldofviewofLandsat(±7.5de- ThecreationofaBRDFLUTisbasedontheidentificationoflandsur-
grees)andthelimitednumberofacquisitionsofferedbythe16-dayre- faceintrinsicanisotropicfeatureswhichmakeoneobjectdistinguishable
peatcycle,itisnotfeasibletoobtaintargetanisotropyinformation fromothers.Numerousstudieshavedemonstrateduniqueanisotropic
Y.Shuaietal./RemoteSensingofEnvironment152(2014)467–479 471
Table1
StructureoftheBRDFLUT.
Type Landcoverclass# Disturbanceage Disturbanceseverity Month DEM QA Bands(1–5,7) RTLS-Rparameters
Rangeofvalue NLCDclassificationscheme 0–30 Low,medium,orhigh 1–12 Flatormountainous 0–5 Landsat1–5,7 Isotropic,volumetric,and
geometrickernelweights
Un-disturbed ✓ NA NA ✓ ✓ ✓ ✓ ✓
Fire-disturbed ✓ ✓ ✓ ✓ NAa ✓ ✓ ✓
Nonfire-disturbed ✓ ✓ NAa ✓ NAa ✓ ✓ ✓
a Limitedbythelackofcurrentancillarydatacommunity.
featuresamongdistinctlandscapeattributes(Bacour&Bréon,2005; timing(fromtheNationalInteragencyFireCenter),theNAFDdataset
Bicheron&Leroy,2000;Lovell&Graetz,2002;Maignanetal.,2004; mayhaveoneortwoyearbiasonthetimingofdisturbanceifacloud-
Shuai & Schaaf, 2010; Strugnell & Lucht, 2001), biome components freeimageorcompositewasnotproducedforagivenyear.Sinceboth
(Chen & Leblanc, 1997), vegetation life-cycle and seasonal stages datasetscovertheperiodsince1984,anyfireandnon-firedisturbance
(Kimes,1983;Shuaietal.,2011),andclassesofdisturbanceinducedby eventsencounteredbeforetheLandsatTM/ETM+era(1980s)arenot
naturalorhumanactivity,aswell assignificanteffectsfromterrain bedefinedinthemaps.Inaddition,theShuttleRadarTopographyMis-
(Schaaf,Li,&Strahler,1994).Thus,theattributesdefinedinTable1are sion(SRTM)DEMdata(Farretal.,2007)wasutilizedtodifferentiatethe
adoptedtobuildtheoverallconceptualstructureofBRDF-LUT.Each qualityofBRDFshapesaffectedbymountainousterrain.Duetothe
entryintheLUTreflectsauniquecombinationoflandsurfacetype,ter- studyinSchaafetal.(1994)interraineffectsontheanisotropyfeature,
rain,timeofyear,limiteddisturbanceageandtype,andLandsatspectral theMODISestimatedBRDFswerequalitativelygradedintotwostrata
bands. (flat tomoderatewithslope≤15°;andsteepermountainouswith
Severalancillarydatasetsprovidedthebasisforthisstratification. slopeN15°).AnexampleisshowninFig.2forindividualundisturbed
First,the30-m2006NLCD(NationalLandCoverDatabase,Vogelmann, evergreenneedleleafforestpatchesrespectivelyinaflatregionanda
Sohl,&Howard,1998)classificationmapswithhighoverallanduser's high-slopemountainousregion.
accuracy(Wickhametal.,2013)wereusedtodeterminelocallandscape ThustheaprioriBRDFLUTsforthePacificNorthwest(PNW)region
attributes,andtoidentifyrepresentativehomogenouslandsurfacere- ofUnitedStateswerecreatedfromMCD43Aproductsandancillary
gions when aggregated to the MODIS 500-m resolution. Then, two datasets(seemoduleBinFig.1)intermsoftheaboveconceptualstruc-
datasetsgivingthetimingandlocationofecosystemdisturbancewere tureoftheBRDFLUT.ThePNWwasselectedforthisinitialprototype
usedtoquantifytheBRDFevolutionofdisturbedlandscapes.Theannual duetoitsrangeofecosystems,prevalenceofbothfire-andnon-fire
30-mMonitoringTrendsinBurnSeverity(MTBS)(Eidenshinketal., forestdisturbances,andrangeoftopography.Inordertominimizethe
2007) dataset has mapped the low/medium/high burn severity of effectofbiomemixtures,MODIS500-mpixelswerelabeledasrepre-
fires(greaterthan1000acresinthewestand500acresintheeast) sentative“pure”pixelsiftheywerecomposedofatleast85%ofasingle
thathaveoccurredsince1984acrossalllandsoftheUnitedStates.The landsurfacetypewhenaggregatedfromthe30-meterNLCDlandcover
30-m NAFD (North American Forest Dynamics, Masek et al., 2008; map.InadditiontobeingstratifiedbyNLCDlandcover,theLUTofBRDF
Maseketal.,2013;Huangetal.,2010)datasetidentifiedotherforest wasalsostratifiedbydisturbancetype(“undisturbed”,“firedisturbed”,
non-firedisturbanceevents(suchasharvest,stormdamage,ordisease) “non-firedisturbance”),disturbanceseverity(fromtheMTBSfiredis-
overthesametimeperiod.WhiletheNAFDdatasettargetsrapiddistur- turbanceproduct),topographicslope(greaterorlessthan15°),time
banceeventsthatremovesubstantialcanopycover,moresubtleorgrad- sincedisturbance(0–26yearscorrespondingtotheNAFDandMTBS
ualdeclinesinlivebiomass(e.g.selectivetreeremoval,gradualinsect coverageof1984–2010),andmonthoftheyear(Table1).Foreachcom-
outbreaks)maynotbecaptured.BothMTBSandNAFDdatasetsaregen- binationoftheseattributes,theBRDFshapesforLandsat(andMODIS)
eratedfromLandsatspectralsignaturesbeforeandafterthedisturbance reflective bands were extracted from the operational V005 8-day
events.WhiletheMTBSdatasetusesindependentconfirmationoffire MCD43A1(BRDFparameters)andMCD43A2(QAflags)11-yearprod-
uct (Schaaf et al., 2002; Shuai et al., 2008). The time dimension
(monthfortheundisturbedLUT,andageofdisturbance),wasusedto
depicttheseasonality,growthphase,andgrowthevolutionsincedistur-
bance,intheBRDFshapesoverforagivenlandsurfacescenario.Ifno
highqualityBRDFwasavailableforagivenmonth(foraseasonalchar-
acterization)oryear(forcharacterizingpost-disturbanceevolution),a
backupBRDFshapewasestablishedthroughlinearinterpolationof
theBRDFmodelparametersfromavailabletimeperiods.Todocument
thequalityofBRDFshapesintheLUT,eachwasassignedaqualityflag
denoted as “high quality” for the original MCD43A estimation and
“lowquality”forthoseinterpolatedones.
Asanexample,Fig.3showstheBRDFshapesintheprincipleplane
withsolarincidentat30°zenithangle,averagedoverdisturbedever-
greenforestregionsinthePNW.Thesnow-freetimeseriesofBRDF
shapesfromSeptemberillustratetheevolutionofevergreenforestsig-
natureovertwodecadesinthegreen,NIR,andSWIRbands.Itisseen
thatBRDFsofbothfireandnon-firedisturbancetypeshavesystematic
temporalvariationsinshapesandmagnitudes.Theevolutionofthis
generalizedBRDF-shapemaybeassociatedwithregrowthandrecovery
Fig.2.ExampleofthedifferenceinMODISBRDFshapeestimatedfornon-disturbedever- ofcanopygreennessandstructureforthedisturbedforestland,indicat-
greenforest(inprincipleplaneat30°solarzenithangle)obtainedfromamountainousre- edbythegradualsharpeningorflatteningofthehot-spot.Thereare
gion(slopeN15°,dot-line)andarelativelyflatregion(slope≤15°,solidline)atNIR
(upper),SWIR(middle),andRed(lower)bandsfromthePacificNorthwestregionof strongtemporalsignaturesofgreenvegetationinbothexamplesof
theUnitedStates. thedisturbancetypes,firstlydisplayedasaclearenhancedhot-spotin
472 Y.Shuaietal./RemoteSensingofEnvironment152(2014)467–479
Fig.3.TwodecadesofBRDFevolutionfollowingnon-firedisturbance(harvest,thinningdominated,toppanels)andhigh-severityfiredisturbance(bottompanels)inthePacificNorthwest
oftheUnitedStates,forgreen(left),near-infrared(middle),andshortwave-infrared(right)bands.TheoriginalMCD43ABRDFshapeswereretrievedfromSeptemberintheprinciple
planewithsolarincidentat30°zenithangle.TheBRDFshapesshowastronghot-spotinthebackward(showingaspositiveviewzenithangle—VZA)direction,andsystematicchanges
inmagnitudeandshapefollowingdisturbanceevents.
theNIRbandwiththeincreasingofgreenness,andasuppressionof 3.3.Surfacealbedodetermination
the hot-spot at the SWIR with the augmentation of canopy water
contentaccompanyingtheforestregrowth.Incontrasttothemono- OncetheBRDFshapeisdetermined,surfacealbedoat30-meter
tonic decrease in brightness following non-fire disturbance, the resolutioncanbecalculatedfromthealbedo-to-nadirreflectanceratio
high-burn-severityfiredisturbedforestpresentsacomplicatedtrajec- (A/N)andLandsatsurfacereflectanceasdetailedforthe“concurrent”
toryofanisotropydevelopmentinthegreenband,withamultiple- approachinShuaietal.(2011).Thismethodassumesthatagivensur-
year(~7–12yearssincedisturbance)reductionduringtheincreaseof facetypehasthesameBRDFshapeatMODISorLandsatresolution,
thehot-spot.Itmaybeexplainedbythedifferentrecoverytrajectories andcanbescaledtoalbedousingthe30-meterdirectionalreflectance
ofthepost-fireresidualstructures.Thesepost-fireresidualtransition fromLandsatasshownin(5),withR andR denotingthecorre-
lnd mod
rateswillvaryamongfires,withahighrateinthefirsttwoyearsfrom spondingspectralreflectancefromLandsatandMODIS,respectively.
treetosnag(i.e.treemortality),andalatepeakafterseveralyears Then,theLandsatblack-skyalbedowithasolarzenithangleatthe
laterforthetree-to-downedwoodandsnag-to-downedwoodchange Landsatoverpasstimeandwhite-skyalbedowerecomputedrespec-
dependingonthespeciesandtreesizeoftheburnedforestregion. tivelyforthesixnon-thermalLandsatbands.Thebroadbandalbedos
Oncethegreensignaturefromthere-grownforestandunderstoryveg- forvisible(0.3–0.7μm)α ,nearinfrared(0.7–3.0μm)α ,andshort-
vis nir
etation(suchasgrassorshrub)becomesdominant,acontinuousgrad- wave(0.3–3.0μm)α bandswereproducedbyafurtherconversion
short
ualincreasingcanbecapturedgenerally10yearsafterseverefires,as fromnarrowspectralbandalbedovalues(α)usingnewconversionco-
i
showninFig.3.Somesmallfluctuationsfoundinthegradualevolution efficientsforLandsat5TM(6–8)andLandsat7ETM+(9–11).Theseco-
ofeachBRDFshapecouldbeduetouncertaintiesinthemappedtiming efficientswerederivedfromradiativetransfersimulationsusing245
ofdisturbance,poorerqualityBRDFestimation,variationsinatmo- surfacespectrarepresentingdifferentsurfacetypes(He,Liang,Wang,
sphericconditions,andresidualcloudandsnoweffects. Shuai,&Yu,2013;Liang,2000).Finally,aqualityassessment(QA)
layer constructed into a 16-bit word was stored for each pixel
(seeTable2)totrackthequalityofinputdata,andestimateerrorprop-
agationthroughthefusionofmultipledatasources.
(
Table2 R ðθ ¼θφ ¼φ;θ ;φ ;λÞ≈R ðθ ¼θ;φ ¼φ;θ ;φ ;λÞ
mod mod mod i v v lnd lnd i lnd i v v
Segmentsofthepixel-based16-bitQAwordforeachLandsatalbedomaptoindicatethe f ðθ;ϕ;θ ;ϕ ;λÞ¼f ðθ;ϕ;θ ;ϕ ;λÞ
performanceofalbedoretrieval. 8r−mod i i v v r−lnd i i v v
Bit Meaning >>>><R ðλÞ¼R ðθ;φ;θ ;φ Þ(cid:2) RmodðλÞ ð5Þ
lnd lnd i i v v R ðθ;φ;θ ;φ Þ
b15 Fillvalue(1=fill-value) ⇒ mod i i v v
bb1134 SCnloouwdflflaagg((11==scnloouwdccoonnttaammiinnaattiioonn)) >>>>:Rlndðθi;φi;λÞ¼Rlndðθi;φi;θv;φvÞ(cid:2)RRmðoθdð;θφi;φ;θi;;λφÞ Þ
b12 Disturbanceflag(0=undisturbed;1=disturbed) mod i i v v
b11-10 Disturbancetype(00=fire;01=non-fire;10and11=reserved)
b9–8 Firedisturbanceseverity(00=reserved;01=low;10=medium;
11=high)
b7 BRDFQA(0=original;1=backup/interpolation)
b6–0 Disturbanceagefordisturbedpixelorlandcoverclassfor α ¼0:3206α þ0:1572α þ0:3666α þ0:1162α
un-disturbedpixel short þ0:04571α −0:00633 4 5 ð6Þ
7
Y.Shuaietal./RemoteSensingofEnvironment152(2014)467–479 473
Fig.4.Examplesofthe“pre-MODISera”approachgeneratedfromscene(path/row:45/29)onday2007-08-29.(A)Thespectralblack-skyalbedocompositeofLandsat-5bands5,4,and3,
(B)thebroadbandblack-skyalbedocompositeofvisible,nearinfrared,andshortwavebands,(C)theblack-skyalbedofortheshortwaveband,and(D)thequalityassessmentmaps.
α ¼0:6000α þ0:2204α þ0:1828α −0:0033 ð7Þ agriculturalfieldsinthelower-centralregionandforeststandsinthe
vis 1 2 3
middle-easternregioninFig.4A.Inthethree-broadbandcomposite
image(Fig.4B),however,thealbedointhevisibleregimehashigher
α ¼0:6646α þ0:2859α þ0:0566α −0:0037 ð8Þ valuesthantheothertwobandsandshowsupasbrown-redinthecor-
nir 4 5 7
respondingareas.Fortheretrievalofeachpixel,onecorrespondingQA
map(Fig.4D)providesthepossiblecloudandsnowcontamination,and
α ¼0:3141α þ0:1607α þ0:3694α þ0:1160α detailsofundisturbedordisturbedinformation.
short 1 3 4 5
þ0:0456α −0:0057 ð9Þ
7
4.AccuracyassessmentoftheLandsatalbedoproducts
α ¼0:5610α þ0:2404α þ0:2012α −0:0026 ð10Þ Threeapproacheshavebeenusedtoevaluatetheaccuracyofalbedo
vis 1 2 3
productsgeneratedbythe“pre-MODISera”LUTapproachpresentedin
thispaper.Oneisthedirectvalidationofshortwavealbedowithactual
αnir¼0:6668α4þ0:2861α5þ0:0572α7−0:0042 ð11Þ groundmeasurements.Theothertwomethodsarecross-comparisons
ofsurfacealbedomapsgeneratedby(1)the“concurrent”approachof
Shuaiet al.(2011) that uses coincident MODIS products to retrieve
3.4.CentralOregonexampleforthederivedAlbedoandQAmaps Landsat-scalealbedo,and(2)thecoincidentoperationalMODISalbedo
productsthemselves.Comparisonwithgroundmeasurementsisaninde-
Fig.4showsmapsofthe30-mLandsatalbedoproductsgenerated pendentandoptimalapproachforproductvalidation,butsuffersfrom
fromthe“pre-MODISera”LUTapproachforasceneincentralOregon thelimitedavailabilityofgroundalbedo-metermeasurements.Cross-
(path/row: 45/29) on August 29, 2007. Spectral black-sky albedo comparisonwithotherproductscanbeperformedonalargevolumeof
estimatesareprovidedasthecompositeofshortwaveinfrared,nearin- MODISimages,butdoesnotprovidearobustestimateofabsoluteaccura-
frared,andredbands(wavelengthcentered1.65μm,0.83μm,and cy.Utilizationofthesemultiplevalidationmeansmayincreasetheability
0.66 μm) (Fig. 4A). Broadband black-sky albedos are available for toevaluatethealgorithmperformancethoroughlyandobjectively.
thevisible(0.3–0.7μm),near-infrared(0.7–3.0μm),andshortwave
(0.3–3.0μm)bands(Fig.4BandC).Atthedatecorrespondingtothese- 4.1.Validationwithgroundmeasurements
lectedsamplecase,alargepartoftheCentralandEasternregionwas
dominatedbysparseshrubsorbarrenland.Comparedwiththeforest Independentgroundortoweralbedomeasurementsaregenerally
regioninthecentral-westpart,theseareasappearashighvaluesin consideredtobemoreaccuratethansatelliteretrievals,andareoften
theSWIRandRedbands,lowervaluesintheNIRband,withscattered takenasareferenceforthevalidationofsatelliteproducts.However,
474 Y.Shuaietal./RemoteSensingofEnvironment152(2014)467–479
Table3
ForestedgroundstationsinthePacificNorthwestregion.Acquiredfromthenetwork-wideAmeriFluxdatabase.
Sitename Vegetationtype Locationa Towerheight(m) Canopyheight(m) Footprintof Dataperiod Landsat
observation(m)b retrieval#
US-Me2 ENFc 44°27′8.28″N,121°33′25.92″W 32.0 ~22 228.6 2005–2007;2009–2011 32
US-Me3 ENFc 44°18′55.68″N,121°36′28.29″W 18.0d ~3.11 342.9 2004–2009 4
US-Me6 ENFc 44°19′23.43″N,121°36′15.69″W 18.6e ~7.0 265.2 2010–2011 7
US-NR1 MFf 40°1′58.31″N,105°32′49.09″W 26 11.5 331.5 2006–2011g 24
US-GLE Subalpine,alpine 41°21′59.51″N,106°14′23.82″W 23/30h 12.1 249.2/409.2 2004–2011 24
US-Blk Conifer 44°09′01″N,103°38′24″W 24 13–15 251.5–205.7 2004–2009 24
a LocationofeachsiteisconfirmedbytheirPIsviaprivatecommunication.
b Diameterofgroundmeasurementsfootprintinthehorizontalplaneatcanopyheight.
c Evergreenneedleleafforest.
d Towerheightis18m,instrumentCNR-1ismountedat14m.
e Towerheightis18.6m,instrumentmountedat17.7m.
f Subalpinemixedconiferousforest.
g Currentlyonlypost2005grounddatatobeusedintermsofdataprocessor'ssuggestionviapersonalcontact.
h Towerheightis30mduring1999–2006,andadjustedto23msince2006.
thevalidationofsatellite-derivedproductsisdifficultbecausethefoot- 4.1.1.Surfacealbedogroundmeasurements
printofsatelliteobservationsdifferssignificantlyfromthatofin-situin- Tower-basedsurfacealbedomeasurementswereacquiredfromsix
struments.Onlymeasurementsspatiallyrepresentingthesurrounding availableforestedsitesofAmeriFluxnetworkinthePacificNorthwest
landscapeatbothin-situandsatellitescalescanprovideacomparable region of the United States (Table 3; Ruehr, Martin, & Law, 2012;
basisforvalidation(Románetal.,2009). Thomasetal.,2009;Vickers,Thomas,Pettijohn,Martin,&Law,2012;
Fig.5.DistributionofAmerifluxvalidationsitesinthePacificNorthwestregionoftheUnitedStates.Foreachsite,agroundphoto(Upper-left),photooftowersurroundings(lower-left),
andhigh-resolutionsatelliteimage(right)areshown.Note:imageoftowersurroundingsforthecurrentlydeactivatedUS-Blksiteisnotavailable.
Y.Shuaietal./RemoteSensingofEnvironment152(2014)467–479 475
Wilson&Meyers,2007).Forestedsiteswereofparticularinterestsince
oneoftheoverallobjectiveswastounderstandhowforestdisturbance
andrecoveryinfluencedthealbedotrajectories.Thefieldsitessample
forestecosystemswithdifferentspeciescomposition,ageanddistur-
banceregimes(seethedistributionandlandscapesinFig.5),including
sub-alpineforestolderthan400yearswithdispersedyoungertreesat
theUS-GLEsite,subalpinemixedconiferousforestnaturallyregrown
fromextensiveloggingduring1900–1910atUS-NR1,coniferforest
withscatteredherbsandshrubsrecoveringfromloggingactivityin
theearly1900satUS-Me2,veryyoungponderosapinestanddisturbed
byfireandharvestinthe1980satUS-Me3,andareforested20-yearold
ponderosa pine site following fire and salvage cutting at US-Me6.
Both upward and downward broadband shortwave solar radiation
(0.3–2.8 μm) were measured via Kipp and Zonen (CNR1, CM-3, or
CM-6b),orEppley-PSPtoweralbedo-meterswith170°effectivefield
ofview.Dataseriescollectedfromtheindividualsiteswereprocessed
intothe30-minutestandardvalues,andobtainedfromtheAmeriflux
website:http://ameriflux.ornl.gov.Forthisstudy,dailytoweralbedo
valueswereretrievedcorrespondingtotheLandsatimagingtimeof
10:30AM,aswellaslocalsolarnoon(LSN),thetimecorrespondingto
Fig.7.Comparisonofactual(alsocalledblue-sky)shortwavealbedobetweenground
theMODISMCD43Aproductsuite.Inaddition,thesurfacealbedodata
measurements(Y-axis)andsatelliteretrievals(X-axis)at10:30AMfromLandsatre-
series(level2)forsiteUS-NR1werereviewedforinconsistencyduring
trievalsandatlocalsolarnoonfromoperationalMCD43A3(V005)productsoversix
periodpre-andpost-2005asanewCNR1sensorwasinstalledinthe AmeriFluxforestedsitesinPacificNorthwestregionofUnitedStates,bothLandsatand
fallof2005.Thepost-2005datawhichweremeasuredwiththenew MODISmeetthenominal0.02accuracyrequirementinrootmeansquareerror(Sellers
well-calibrated sensors were recommended for use by the data etal.,1995).Thedashedlinesrepresentanabsoluteaccuracyof0.03comparedtothe
grounddata.
provider-(S.Burns,personalcommunication).Anefforttoestablish
sensor-to-sensor cross-calibration is underway, and may provide
correctedpre-2005datasoonforfurthervalidationactivitiesatUS_NR1. α wasobtainedasthesumofavailable30-mretrievals
towerfootprint
(α (i))weightedbycos(θ),whereθ istheviewanglebetweenthe
lnd i i
4.1.2.AggregationfromLandsatscaletotowermeasurementfootprint towertopandthecenterofpixeli(Fig.6)foralltheNpixelsthatfallen
Thedisparatespatialscalebetweensatelliteandin-situmeasure- inthefootprintofgroundmeasurements(Eq.12).
mentsisoneofthebarrierstovalidatingsatellite-derivedproducts.
X
Severalstudieshaveconcludedthatdirect“point-to-pixel”comparison, N ðcosðθÞ(cid:2)α ðiÞÞ
withoutconsideringspatialscales,isnotsufficientforalbedoproduct αtowerUfootprint¼ i¼1XN ciosðθÞlnd ð12Þ
validation,unlessthevalidationfocusesonalargeandhomogenousre- i¼1 i
gions(Liangetal.,2002;Románetal.,2009).Thetowerbasedinstru-
mentpyranometerisinfluencedbythe“cosine-law”oftheresponse
(cid:4)dire(cid:5)ctionandhasa170°effectivefieldofview.Anareaof 2h(cid:2) tan 4.1.3.Comparisonwithgroundmeasurement
85(cid:3) diameterinthehorizontalplaneatforestcanopyheightisthende- Wederivedthesurfacealbedofromthe30-minutetowermeasured
2
finedbythedownward-lookingsensormountedonatower(hmeters downwellingandupwellingradiationat10:30AMforLandsatandat
above canopy). The calculated diameter of the tower footprint for localsolarnoon forMODISovereachsite.Notethat retrievalsfrom
eachsiteislistedinTable3.Tofacilitatethecomparisoninthisstudy, LandsataswellasMODIScalculateintrinsicsurfacealbedoundertwoex-
acosine-law-basedup-scalingmethodwasappliedtoaggregatethe tremeincidentradiationsituations(“black-skyalbedo”correspondingto
30-mLandsatalbedotothetowerfootprintforindividualsites(Shuai purelydirectsolarilluminationand“white-skyalbedo”corresponding
etal.,2011).Thesurfacealbedocorrespondingtothetowerfootprint topurelyisotropicillumination),whilethefieldmeasurementsrecord
theactualilluminationcorrespondingtoamixtureofbothdirectanddif-
fuseradiation.Toobtaincomparablemetricswithfieldmeasurements,
wecalculatetheactualalbedo(alsocalled“blue-skyalbedo”)viathein-
terpolationbetweenblack-skyandwhite-skyalbedosweightedbythe
ratioofdirectordiffusetothetotaldownwellingradiation(Luchtetal.,
2000;Románetal.,2011;Schaafetal.,2002).Sincethein-situdatasets
lackinformationondirect/diffuseratios,wesimulatedthedirect/diffuse
ratiosforrequiredsolarzenithanglesusing6Sbasedonthesimultaneous
MODISTerraatmosphereopticaldepthsatthe550nmband.Errorsin-
ducedbythedifferenceofdefinedwavelengthintervalfortheshortwave
band(ground0.3–3.0μm,Landsat0.3–3.0μm,and0.3–5.0μmforMODIS)
arenegligiblebecausethesolarirradiancebeyond2.5μmaccountsforless
than1.8%ofthetotalbetween0.3and14.3μm(Hulstrom,Bird,&Riordan,
1985).
Thescatterplot(Fig.7)comparestheLandsatblue-skyalbedoaggre-
gatedtothetowerfield-of-viewwiththein-situmeasuredalbedoat
10:30AMintheshortwaveforthesixAmeriFluxnetworksites.Re-
trievalswithsnowandcloudcontaminationwereremovedfromthe
analysisusingthesnowandcloudflagsintheQAwordofthesatellite
Fig.6.IllustrationoftheaggregationfromLandsat30-mpixels(dottedgraygrids)intothe
footprintprojectedonthegroundbyalbedometer(FOV=α)mountedonthetowerh products.TheLandsatretrievalsareinverygoodagreementwiththe
metersabovecanopy. tower-basedalbedo,witharootmeansquareerror(RMSE)lessthan
476 Y.Shuaietal./RemoteSensingofEnvironment152(2014)467–479
Fig.8.Cross-comparisonofshortwavewhite-skyalbedomaponday2007-08-29generatedfrom“pre-MODISera”LUTapproach(left)with“concurrent”approach(middle),andthere-
latedscatterplotsoverallavailablepixels(right).
0.016andabiasnomorethan0.007.DiscrepancybetweentheLandsat 4.2.Cross-comparisonwiththe“concurrent”approach
andgroundalbedosisconfinedtowithin±0.03albedo(dottedlinein
Fig.7),whichsupportstheabsoluteaccuracyrequirement(0.02–0.05) Asaninitialvalidation,wecomparedalbedomapsgeneratedbythe
establishedbytheclimatemodelingcommunity(Sellersetal.,1995). “pre-MODISera”LUTapproachtothosegeneratedbythepreviously
ComparedwiththeoperationalMODIS(V005)shortwavealbedore- published“concurrent”approach,whichhasbeenvalidatedpreviously
trievalsatlocalsolarnoonviathegroundmeasurementsasabridge, (Románetal.,2013;Shuaietal.,2011).Fig.8showstheshortwave
theLandsatretrievalsareslightlyhigher,exceptfortheUS-NR1site. broadbandwhite-skyalbedomapsderivedfrombothapproachesfor
Thismakessenseifweconsiderthedefinitionofblack-skyalbedoasde- theidenticaldate2007-08-29(fillvaluesexcludedintheanalysis).
scribedpreviously.Becausevaluesofblack-skyalbedodependclosely The“pre-MODISera”LUTapproachderievedalbedo(left)isconsistent
onthedirectionofsolarillumination(i.e.solarzenithangleortiming withthatfromthe“concurrent”approach(right)forthespatialvaria-
ofobservation),andblack-skyalbedoiscommonlyobservedtode- tionofalbedovalues,fromlowinthePNWforestregiontothehighin
creasefromsunrisetonoon,thenincreasefromnoontosunset,asval- the eastern barren land. In general, albedo value extracted by the
idatedforMODISinLiuetal.,2009. “pre-MODISera”LUTapproachisslightlyhigherthanthe“concurrent”
Fig.9.Illustrationoftheconsistencybetweenthe“pre-MODISera”LUTapproachandthe“concurrent”approachofShuaietal.(2011).Shortwavewhite-skyalbedomapsgeneratedre-
spectivelyfrom“pre-MODISera”LUTapproach(panelA)and“concurrent”approach(panelB),overanundisturbedforestregioninMontana(fillvalueordisturbedregionsinblack).The
absolutedifferencemapsofwhite-skyalbedobetween“concurrent”and“pre-MODISera”albedo(panelC,fillvalueordisturbedregionsinwhite)fortheoverlappingyears2001–2011.
Dayofyearisindicatedforeachalbedomap(YYYY-DOY).