Comparison between Snow Albedo Obtained from Landsat TM, ETM+ Imagery and the SPOT

19 Pages·2016·3.39 MB·English

hydrology
Article
Comparison between Snow Albedo Obtained from
Landsat TM, ETM+ Imagery and the SPOT
VEGETATION Albedo Product in a Mediterranean
Mountainous Site
RafaelPimentel1,*,CristinaAguilar2,†,JavierHerrero2,†,MaríaJoséPérez-Palazón1,† and
MaríaJoséPolo1,*
1 FluvialDynamicandHydrology,AndalusianInstituteforEarthSystemResearch,UniversityofCordoba,
ÁreadeIngenieríaHidráulica,EdificioLeonardodaVinci,CampusRabanales,14071Cordoba,Spain;
[email protected]
2 FluvialDynamicandHydrology,AndalusianInstituteforEarthSystemResearch,UniversityofGranada,
CentroAndaluzdeMedioAmbiente(CEAMA),Av.delMediterráneos/n,18006Granada,Spain;
[email protected](C.A.);[email protected](J.H.)
* Correspondence:[email protected](R.P.);[email protected](M.J.P.);Tel.:[phone](R.P.)
† Theseauthorscontributedequallytothiswork.
AcademicEditor:JurajParajka
Received:1November2015;Accepted:16February2016;Published:23February2016
Abstract:Albedoplaysanimportantroleinsnowevolutionmodelingquantifyingtheamountofsolar
radiationabsorbedandreflectedbythesnowpack,especiallyinmid-latituderegionswithsemiarid
conditions.Satelliteremotesensingisthemostextensivetechniquetodeterminethevariabilityof
snowalbedoover mediumtolarge areas; however, scaleeffectsfrom thepixelsizeof thesensor
source may affect the results of snow models, with different impacts depending on the spatial
resolution. This work presents the evaluation of snow albedo values retrieved from (1) Landsat
images,L(16-dayfrequencywith30ˆ30mpixelsize)and(2)SPOTVEGETATIONalbedoproducts,
SV(10-dayfrequencywith1ˆ1kmpixelsize)intheSierraNevadamountainrangeinSouthSpain,a
Mediterraneansiterepresentativeofhighlyheterogeneousconditions. Dailysnowalbedomapseries
werederivedfrombothsources,andusedasinputforthesnowmoduleintheWiMMed(Watershed
IntegratedManagementinMediterraneanEnvironment)hydrologicalmodel,whichwasoperational
at the study area for snow monitoring for two hydrological years, 2011–2012 and 2012–2013, in
theGuadalfeoriverbasininSierraNevada. Theresultsshowedsimilaralbedotrendsinbothdata
sources,butwithdifferentvalues,theshiftbetweenbothsourcesbeingdistributedinspaceaccording
tothealtitude. ThisdifferenceresultedinlowersnowcoverfractionvaluesintheSV-simulations
thataffectedtherestofsnowvariablesincludedinthesimulation. Thisunderestimation,mainly
duetotheeffectsofmixedpixelscomposedbybothsnowandsnow-freeareas,producedhigher
divergencesfrombothsourcesduringthemeltingperiodswhentheevapo-sublimationandmelting
fluxesaremorerelevant. Therefore, theselectionofthealbedodatasourceintheseareas, where
snowevapo-sublimationplaysaveryimportantroleandthepresenceofsnow-freepatchesisvery
frequent,canconditionthefinalaccuracyofthesimulationsofoperationalmodels;Landsatisthe
recommendedsourceifthemonitoringofthesnowpackisthefinalgoalofthemodeling,whereas
theSVproductmaybeadvantageouswhenwaterresourceplanninginthemediumandlongterm
isintended. Applicationsoflargepixelsizealbedosourcesneedfurtherassessmentforshort-term
operationalobjectives.
Keywords: snowalbedo;Landsat;SPOTVEGETATION;Mediterraneanregions
Hydrology2016,3,10;doi:10.3390/hydrology3010010 www.mdpi.com/journal/hydrology
Hydrology2016,3,10 2of19
1. Introduction
Albedoplaysinimportantroleintheenergyandwaterbalanceinareaswiththepresenceofsnow,
sinceitaffectstheshortwaveradiativeflux(0.3–2.5µm),quantifyingtheamountofsolarradiation
reflectedandthenabsorbedbythesnowpack. Specifically,itisoneofthemainfactorsinvolvedin
the melting processes due to the dominant importance of the radiation components during these
periods[1,2]. Thisinfluenceisenhancedinsemiaridareas,duetoboththesignificantlyhighersolar
energyfluxesthroughouttheyear[3]andthesuccessivemeltingcyclesthatusuallytakeplaceduring
thesnowseason.
AlthoughsnowfallisnotthemostcommonformofprecipitationinMediterranean-climateregions,
waterresourceavailabilityduringthewarmseasonandfluvialregimearehighlydependentonthe
snowoccurrenceandevolutioninmountainousareassuchastheAtlasMountains(Morocco), the
southernAlps(Italy),MountArarat(Turkey),MountLebanon(Lebanon),SierraNevada(California),
the Andean (Chile), and Sierra Nevada Mountains (Spain). Recent data show the importance of
thissolidwaterresourceinCaliforniaandtheimpactofclimatechangefordroughtfrequencyand
snowretrievalinthemediumterm[4]. Theuseofenergyandwaterbalancemodelstocharacterize
andpredictthesnowpackdynamicsisthemostrecommendedapproachunderthehighlyvariable
conditionsdominantintheseregions[5–7],togetherwiththeadoptionofaspatialresolutionhigh
enoughtorepresentthehydrologicalresponseofthewatershed[8]. Thisisespeciallyhighlighted
during the snowmelt periods to be capable of predicting the snow contribution to runoff and,
potentially,tospringfloods.
However,longandhigh-frequencyseriesofalbedovaluesarenotalwaysavailable. Measuring
andmonitoringsnowalbedopresentdifficulties,whichincreasewhenthemodelsrequirespatially
distributeddata. Satelliteremotesensingisthemostextendedtechniqueusedtomonitorsnowalbedo
atmediumandlargescales,sincethissourcepotentiallycapturesthevariabilityofsnowalbedoineach
ofthethreedomains: time,spaceandfrequency[9,10]. However,theinclusionofthesemeasurements
inmodelsissomehowlimitedbythefixedresolutionofsatelliteimages(e.g.,NOAA,dailyimages
with1kmˆ1kmcellsize;MODIS,dailyimageswith250mˆ250mcellsize;Landsat,16-dayimages
with30mˆ30mcellsize;SPOT,10dayssceneswith1kmˆ1km)andtheirspectralwavelength
range,aseachsatelliteprovidesinformationincertainbandsoftheelectromagneticspectrum. Some
otherconsiderationsthatmaycomplicatetheacquirementofsnowalbedofromremotesensingmust
betakenintoaccount. First, theatmosphericeffectsbetweenthesensorandtheobservedsurface,
such as the scattering produced by water vapour and aerosols, or the presence of clouds, make it
difficult to obtain information during long time periods over the snow season. Additionally, the
degreeofanisotropyofthemeasuredsurfaceconditionstherelationbetweentheincidentandreflected
angle, andthenarrowbandtobroadbandconversion[11,12]. Furthermore, inroughmountainous
areastopographyusuallyproducesanintenseshadowingeffect,whichneedstobeconsideredinthe
modellingprocessbutwhichalsomodifiesalbedoduringtheday,andhasasignificantinfluenceon
theeffectivevaluetobeincludedinmodels. Shadedareasshowlessreflectancethanexpectedfrom
thesurfacematerials,whereassunnyconditionsexhibittheoppositebehaviour,witharesultinghighly
variableresponsefromsnowwiththesameagebutdifferentlocation[13].
Moreover,itisessentialtoaccuratelyapproximatethecorrectdistributionofthesnowduringthe
wholeseason,inordertodistinguishsnowfromotherlandcoverspresentinthearea. Theoccurrence
of small but abundant patches of snow during the melting cycles involves working with high
frequency/spatialresolutionimages,whichcannoteasilybeachievedfromasinglesatellite;terrestrial
photographyhasbeenusedtoanalysethesemicroscaleeffects[14].Givenalltheseconsiderations,
LandsatTM,ETM+andOLIproducethemosthighlyrecommendedimagestostudysnowalbedo
evolutioninmountainousareasinsemiaridregions,duetotheirspatialscale(30mˆ30m);however,
therevisitingtimeis16days,whichconstrainstheobtainmentofinformationabouttheevolution
duringquickprocessesandgeneratestimegapslongerthanonemonthduringwetperiods(cloudy
conditions). Moreover,albedomustberetrievedbytheuserfromeachimagethroughasequential
Hydrology2016,3,10 3of19
processinvolvingcorrections,snowdetectionandbandintegration. Asanalternative,MODISdaily
snow products provide direct estimations of albedo and snow cover fraction, and they have been
widelyusedtomonitorthesnowpackevolutioninsnow-dominatedregions;however,theircellsize
(500mˆ500mforsnowproducts)oftenincludeslargefractionsofno-snowareas,whichreduces
thenumberofpixelsintheimagewith“puresnow”conditionsanddecreasestheaccuracyofthe
modelling results [15]. This dilemma between time and spatial accuracy is also recurrent in other
applications of these data sources, such as vegetation cover fraction studies, evapotranspiration
calculations,etcetera[16],beingLandsattheselectedsourcewhenthespatialresolutionisarestriction
toachievethedesiredaccuracy[17]. Differentapproachestoderivefusionalgorithmsthatbenefit
frombothsources’advantageshavebeenperformed,suchastheDisALEXIS[18]modeltoincrease
theoriginalLandsatfrequencybydownscalingMODIS[19],orthesemi-physicalfusionapproachfor
MODISBRDF/AlbedoandLandsatETM+forsnowcoverfractionretrieval[20]. Additionally,other
satellitesprovidedirectproductswithevenhighercellsizes. ThisisthecaseforSPOTVEGETATION,
with similar revisiting time to Landsat, 10 days, and a spatial resolution of 1 km ˆ 1 km, which
providesbroadbandalbedoproductintheshortwaveregionofthespectrum. Largerscaleeffectsare
expectedifthepixelsizeisincreased,althoughnon-linearitymightdamperthefinalimpactonthe
calculations. Ontheotherhand,duringhighlyvariableseasons,theavailabilityofonlyafewbutwith
highresolutionimagesmaybiasthesimulationresults,over-orunderestimatingboththeextentand
amountofsnow,dependingonhowrepresentativeofthewholeperiodtheconditionsofeachisolated
imageare.
TheSierraNevadaMountainRangeinsouthernSpainisarepresentativeexampleofthedescribed
conditions. Thewiderangeofhabitatsrunningfromitsvalleystothetopsmerititsinclusioninthe
NationalParksnetwork,anditssouthernlatitudeledtoitsselectionasoneoftheClimateChange
ObservatoriesinSpain,sincetheevolutionofthesnowregimewillnotonlyimpacttheassociated
ecosystemsandwaterresourcesbutalsoserveasreferenceforfuturebehaviourofsnowsystemsin
regionsfarthernorthinEurope. Theradiativeprocesses,amongothers,conditionahighvariabilityin
theannualregimeofsnowrelatedtothedry-wetsequencesduringthesnowseasonandthespring.
This work analyses the albedo spatial resolution effect on snow modelling in these environments
from two remote sensing sources with similar revisiting time but significantly different pixel size;
2-year coincident series available for this area from Landsat (30 m ˆ 30 m, 16 days) and SPOT
VEGETATION images (1 km ˆ 1 km, 10 days) were used to quantify the scale effects and assess
selectioncriteriaforsnowmodelling. Thedatasources’influenceontheaccuracyoftheresultingsnow
distributionmodellingisevaluatedtogetherwiththeirimpactonthewaterfluxesassociatedwiththe
snowevolution.
2. MaterialandMethods
2.1. StudySite
ThisstudywascarriedoutintheSierraNevadaMountains,SouthernSpain,whereoneofthe
greatestwealthsofbiodiversityinfaunaandflorainEuropecanbefound[21,22]. Thehighestregion
oftherangewasdeclaredaUNESCOBiosphereReservein1986,NaturalandNationalParksin1989
andin1999,respectively). TheGuadalfeoriverbasinupstreamtheRulesdamwasselectedforthis
study(1058km2)(Figure1). Thearea,withanaverageelevationof1418.5m.a.s.l.,ischaracterized
byhightopographicgradientsduetoitsproximitytothecoastline,only40kmsouth. Therefore,a
combinationofalpineandMediterraneanclimateconditionscanbefound.Snowfallappearseveryyear
ataltitudesover2000m.a.s.l. butitspresenceanddistributionishighlyvariablebetweenconsecutive
years depending on the meteorological conditions. Precipitation increases with elevation and the
meantemperatureisusuallyhigherthaninalpineclimates. The40-yearmeanannualvaluesofthe
precipitationanddailytemperatureare630mmand12.2˝C,respectively[23]. Globalradiationis
highthroughouttheyearduetoitssouthernaspectandusuallackofcloudcover,evenduringwinter,
Hydrology2016,3,10 4of19
despitethecoldtemperaturesandthepresenceofsnow. Theroughtopographyleadstoimportant
differencesintheinstantaneousglobalradiationbetweentheeastandwest-facingmountainsslopes
and,consequently,inthesnowalbedovalues[3].
Figure1.LocationoftheGuadalfeoRiverbasinupstreamtheRulesdam(blackline).Thestudyareais
theregionabove1500m.a.s.l.andisdividedintofouraltitudinalregions:R1(3500–3000m.a.s.l.),R2
(3000–2500m.a.s.l.),R3(2500–2000m.a.s.l.)andR4(2000–1500m.a.s.l.).Bluedotsindicatethelocation
ofthesixweatherstationsusedinthestudy;thereddotshowsthelocationofthealbedoinsitudata
monitoringstation.
2.2. RemoteSensingAlbedoData
TwosourcesofremotesensingdatawereusedinthisstudyfortheperiodOctober2011–August
2013. Thefirstsourceconsistsofthe21cloud-freemultispectralimagesfromLandsatTM5andETM+
availablefromthe2yearsanalysed;thesecondsourcewastheSPOTVEGETATIONalbedoproduct
with52availableimagesintheareaduringthestudyperiod(Figure2).
Figure2.Acquisitiondatesoftheavailableimagesfrombothalbedosourcesusedinthestudy.
2.2.1. LandsatAlbedo
TwentyoneLandsatTM5andETM+sceneswereselectedwithinthestudyperiodforthecloud
freescenesavailableinordertominimizeuncertaintiesduetoheterogeneousatmosphericconditions
Hydrology2016,3,10 5of19
(Figure2). Landsatsatellitespresentatemporalresolutionof16daysand30ˆ30mspatialresolutions.
Allscenesweresubsettoformtheimageofthestudyarea. Bothsensorsanalysed,TMandETM+,
havethesamemultispectralwavebandswithsimilarspectralcoverage. Thereforealltheimageswere
processedinthesameway.
Priortothesnowalbedocomputation,apre-processingoftheimagesisrequired[24]. Figure3a
shows the pre-processing applied in this study that includes the radiometric calibration and the
atmospheric, saturation and topographic corrections; in this study, the surface is considered to be
Lambertian,thatis,thedegreeofanisotropyofthelandsurfaceisnotconsideredand,therefore,the
angularcorrectionisnotapplied[12]. SnowisnotaLambertiansurface;nevertheless,someworks
showthisassumptioncanbeadoptedwithoutsignificanterrors. Dumondetal. (2010)[25]showed
that,forsnowsurfaces,theanisotropyfactorforwavelengthsunder1.2nmisnearlyconstantandclose
totheunit;thus,forthosewavelengths,theLambertianassumptiondoeshold. Forwavelengthsabove
this 1.2 nm threshold, the anisotropy factor varies depending on the incident/viewing directions;
as the TM sensor is for nadir viewing, the direction is always fixed, and the incident direction is
approximatelythesamebecauseofthesensoroverpassforthearea,andforthat,approximately30˝,
theanisotropicfactorisalsocloseto1.
Figure3. Flowchartofpre-processingandtimealbedo30ˆ30mmapsretrievalfrombothdata
sources(a)LandsatTMandETM+sceneand(b)SPOTVEGETATIONdatasets.
RadiometriccalibrationwasachievedthroughtheequationsandrescalingfactorsinChanderetal.
(2009)[26].
Atmosphericcorrectionwasconductedtoretrievesurfacereflectancebyuseofthedarkobject
subtraction(DOS)method,whichrelatesthesatelliteradianceandthesurfacereflectance[27,28]by
assumingaLambertiansurfaceandacloudlessatmosphere. Moreover,allthescatteringeffectsare
consideredtobethesameasthoseofablackbodywithinthescene[29]. Thisdarkobjectisselected
from the minimum radiance value of the histogram with at least 200 pixels [30], excluding those
pixelsinthescenethatlieintheshadowsproducedbytheterrainroughness[31]. Thus,fixedvalues
for the downwelling transmittance parameters of the atmosphere for each band are adopted [32]
and the upwelling atmospheric transmittance and the diffuse irradiance are both neglected. DOS
methods constitute a good alternative when the atmospheric optical properties needed by more
complexradiativetransferencecodes(RTCs)aretotallyorpartiallyunavailableorareofquestionable
quality[33].
Snow in the Landsat scene where the study site is included constitutes less than 5% in the
maximumsnowextensionandbecomesinsignificantattheendofthesnowseason. Forthisreason,
severalbandsareradiometricsaturated,especiallyinwinterwhenthedifferencebetweensnowand
therestoflandcoversinthesceneisgreater. Tocorrectthissaturation,amultivariablecorrelation
analysisbetweenbandsisemployedtorecoverthesnow-saturatedpixels[34].
Regardingthetopographiccorrection,aCcorrectionwithlandcoverseparationalgorithmwas
applied[31]. Again,undertheLambertiansurfaceassumptionalinearfitbetweentheillumination
angleandthedifferentbandsreflectanceisestablished. Additionally,thediffuseirradianceistaken
intoaccountviaasemi-empiricalestimationoftheCfactor. Inordertoconsiderthemultiplereflective
propertiesofthedifferentvegetativesoilcovers,thepixelswereclassifiedintobaresoilandvegetated
areasbyusingtheNormalizedDifferenceVegetationIndex(NDVI)[35].
Hydrology2016,3,10 6of19
Once the images are corrected, the discrimination of snow-covered areas and the
narrowband-to-broadbandconversionareappliedtogeneratethesnowalbedomaps. Theseparation
of snow-covered and snow-free areas was carried out with the normalized difference snow index
(NDSI)[15]andthehighreflectanceofsnowinthevisiblepartofthespectrum(Band1inLandsat
images). Inthiswork,thetwothresholdsusedtodiscriminatethesnowwereNDSI>0.4andBand
1>0.1[36,37]. Thenarrowbandtobroadbandconversionwasperformedfollowingthemethodof
BrestandGoward(1987)[38],basedonthenumericalintegrationofthesnowspectralalbedocurve
throughouttheshortwavespectrum. Intheirwork,thehomogeneouspartsinthesnowspectralalbedo
curvewereidentifiedandthenagivenpercentageofthetotalspectrumwasassignedtoeachidentified
homogeneousintervalportion. Thedifferentspectralbandsofthesensorwereusedasrepresentative
of each one of such parts. This method can be applied to different kinds of mixes surfaces, as the
authorsshowed,includingsnowareas. Followingthis,4bands(TM2,TM4,TM5,andTM7)ofthe6
bandsoftheTMsensorwereusedinthisstudy,eachoneassumedtoberepresentativeofthefollowing
segments,s1=300–725nm,s2=725–1000nm,s3=1000–1400nm,s4=1400–2500nm(eachoneof
themwithapproximatelyconstantspectralalbedo). TheweightingfunctionshowninEquation(1)
givesanestimationofthebroadbandalbedoinsnowsurfaces[38,39].
Albedo“0.493 ¨pBand2q ` 0.203 ¨pBand4q ` 0.150 ¨pBand5q ` 0.154¨pBand7q (1)
2.2.2. SPOTVEGETATIONAlbedo
Sixty-eightGeoland-2SPOT/VGTAlbedoproductsfromCopernicusLandMonitoringService
were analyzed along the study period (Figure 2). The temporal and spatial resolutions of the
products were 10 days and 1 km ˆ 1 km respectively. From the two albedo variables available
intheGMES/Copernicusdatasets,thehemisphericalalbedo(AL-BH)wastheoneusedinthisstudy.
Itprovidestheintegrationofthedirectionalalbedoovertheilluminationhemisphereandassumes
a complete diffuse illumination. Moreover, the FLAG sub-product was employed to discriminate
snow-coveredandsome-freepixels. Thealgorithmusedtomakethisdiscriminationisalsobasedon
theNDSIindex,butinthiscase,adaptedtotheavailablebandinSPOTVEGETATION[40].
Thepre-processingappliedtoSPOTVEGETATIONdataisshowninFigure3b. First,allproducts
weresubsettoformtheimageofthestudyarea. Then,astheprojectionsystemsusedbythemodel
and the SPOT VEGETATION data are different, an automatic routine to re-project the data was
implementedinMATLAB.Finally,inordertomatchthedifferentcell-sizesofbothremotesensing
datasources,areshapeinthecellsizeoftheavailableproductsfrom1kmˆ1kmto30mˆ30mwas
applied. Anearestneighborinterpolationalgorithmwasthemethodemployed.
Followingthisprocess,thealbedomapsarederivedforthesnowcoveredpixelsdeterminedin
theFLAGproduct.
2.2.3. InSituAlbedoMeasurements
In situ snow albedo data are only available from two stations within the study area, but only
oneofthem(ColladoBuitreras;seeFigure1)hasbeenoperativeduringthewholestudyperiod. Itis
locatedat2867m.a.s.l. andisbeingmeasuringalbedofromthewinterof2008on;thesecondonehas
justbeenoperativefromthelatespringof2013,outofthestudyperiodinthiswork,andthuswas
notused.
The in situ albedo measurements were used to validate the albedo obtained from the remote
sensing sources. Only 7 of the 21 Landsat scenes and 35 of the 52 SPOT VEGETATION scenes
could be used in the validation process, since the in situ albedo measurement was out of range
intherestofthem. Forcomparison,theinstantaneousin-situalbedovaluemeasuredat10:40a.m.
(GMT+1),approximatelytheacquisitiontimeofLandsatscenes,wasselected;inthecaseoftheSPOT
VEGETATIONvalidation,anaveragingoftheinstantmeasurementscorrespondingtothe10-days
windowsofthesensor’sscenewasobtained.
Hydrology2016,3,10 7of19
2.3. MeteorologicalData
Dailymeteorologicalobserveddataofprecipitation,temperature,windvelocity,radiationand
relativehumidityrecordedatsixweatherstationscoveringthealtitudinalgradientinthearea(Figure1)
wereusedinthisstudy. Thesemeteorologicaldatawereusedasinputstothehydrologicalmodel
WiMMed(WathershedIntegratedModelinMediterraneanEnvironments),whosesnowmoduleis
operationaloverthisareafrompreviousworks[9,14,37,41]. Specificinterpolationalgorithmswere
used to distribute the point measurements of the meteorological variables over the whole study
area. Precipitation,temperature,humidityandwindspeedareinterpolatedbymeansofasquared
inverseofdistanceweightmethod(IDW2);acorrectionbasedontheobservedaltitudinalgradientsis
includedfortemperatureandprecipitation. Thesealgorithmsweretestedandvalidatedinthestudy
area[42]andarecurrentlyusedintheoperationalversionofthesnowandhydrologicalmodelinthe
associatedwatersheds. Moreover,forradiation,aphysicalmodelbasedontheclearnessindexandthe
diffuse/directfractionwithinclusionofthedifferenttopographiceffectsisapplied[3]. Finally,forthe
atmosphericemissivity,aspecificformulationformountainousareasthatwaslocallyvalidatedatthe
studysiteisused[43].
Thestudyperiodcoverstwohydrologicalyears,fromSeptember2011toAugust2013. Toanalyse
the most vulnerable changes in snow albedo, only the area usually covered by snow within the
watershed,locatedabove1500m.a.s.l.,wasselectedforthestudy. Moreover,thisstudyareahasbeen
dividedintofourregionsrelatedtothealtituderange: 3500–3000m.a.s.l. (R1),3000–2500m.a.s.l. (R2),
2500–2000m.a.s.l. (R3)and2000–1500m.a.s.l. (R4)(Figure1). Table1showsthemaintopographicand
meanannualclimaticfeaturesdescribingthestudyareaandeachaltitudinalregion,Region3being
representativeoftheaveragefeaturesofthestudyarea.
Table1.Maintopographicandannualclimaticfeaturesofthestudyareaandeachaltitudinalregions
calculatedona10mˆ10mDEMsuppliedbytheRegionalAndalusianGovernment.
Area Altitude Slope Precipitation Snowfall Tmean Tmax Tmin
(km2) (m.a.s.l.) (mm´1) (mm) (mm) (˝C) (˝C) (˝C)
REGION1 15.8(5%) 3110.8 0.22 935 759 2.9 16.685 ´13.95
REGION2 100.6(33%) 2727.5 0.19 834 565 4.8 18.3 ´11.5
REGION3 124.8(41%) 2249.4 0.21 735 323 7.4 20.8 ´8.15
REGION4 151.8(51%) 1744.6 0.22 637 129 10.3 24.0 ´4.5
TOTAL 393.0 2211.4 0.21 731 328 7.7 21.1 ´7.8
2.4. SnowModelling
Thesnowmelt-accumulationmodeldevelopedbyHerreroetal. (2009)[5]isaphysicalmodel
basedonapointmassandenergybalance,spatiallydistributedbymeansofdepletioncurves[37,41].
ThismodelwascreatedandmainlytestedinMediterraneanandsemiaridhighmountains.
Thepointmodelassumesaone-layereduniformsnowpack. Inthecontrolvolumedefinedby
thesnowcolumnperunitarea,whichhastheatmosphereasanupperboundaryandthegroundasa
lowerone,thewatermassandenergybalancecanbeexpressedbyEquations(2)and(3):
dSWE
“ R´E`W´M (2)
dt
dpSWE¨uq dU
“ “K`L`H`G`R¨u ´E¨u `W ¨ u ´M¨ u (3)
R E W M
dt dt
whereSWE(snowwaterequivalent)isthewatermassinthesnowcolumn,anduistheinternalenergy
perunitofmass(Ufortotalinternalenergy). Inthemassbalance,Rdefinestheprecipitationrate;Eis
watervapourdiffusionrate(evaporation/condensation);Wrepresentsthemasstransportratedueto
wind;andMisthemeltingwaterrate. Ontheotherhand,regardingenergyfluxes,Kisthesolaror
shortwaveradiation;Lthethermalorlong-waveradiation;Htheexchangeofsensibleheatwiththe
Hydrology2016,3,10 8of19
atmosphererate;Gtheheatexchangewiththesoilrate;andu ,u ,u andu aretheadvectiveheat
R E W M
ratetermsassociatedwitheachoneofthemassfluxesinvolvedinEquation(2).
InthemassbalanceEquation(2),Wisdisregardedduetothequicksnowmetamorphosisobserved
atthestudyarea,thatcompactsthesnowpackandreducesitsmobility. Gisnotconsideredeither
intheenergybalanceequation,beingusuallyconsideredasecondarytermperseinthisbalance[44]
Shortwaveradiation(K)iscalculatedthroughthemeasureddownwellingshortwaveradiationfluxand
thealbedo.Longwaveradiationiscalculatedbymeansofanempiricalformulationfortheatmospheric
longwaveemissivity,locallyvalidatedinthearea[43].Finally,Hismodelledasadiffusionprocess[45].
Thepointmodelisthendistributedoverthestudyareausinga30mˆ30mrectangulargrid
wherethesnowenergybalanceisperformedovereachcell. Theparameterizationofsubgridprocesses
iscarriedoutbydefiningdepletioncurves,whichareempiricalfunctionsthatrelatethefractionofcell
areacoveredbysnowwithsomeothersnowstatevariable. Theyallowquantificationofthedecrease
insnowcoverareawithinthecelltotakeintoaccountthereductionofthesurfaceareaimpliedinthe
energy-massbalance. Becauseoftheannualoccurrenceofdifferentsnowmeltcyclesduringthesnow
seasonandtheirvariabilitybetweenyears,differentdepletionscurveswereproposedforeachkind
ofmelting/accumulationcycleidentifiedinthearea. Themodelhasbeenpreviouslycalibratedand
validatedovertheareaintermofsnowcoverfraction(SCF)obtainedfromLandsatTMandETM+
usingaspectralmixture(Figure4)[41].
Figure4.Comparisonbetweensimulatedsnowcoverfractionvalues(SCF)(WiMMedsnowmodel)
and“measured”snowcoverfractionvalues(fromLandsatspectralmixtureanalysisovertheGuadalfeo
watershed,Figure1)duringthe2004–2013period.RMSE=0.05m2¨m´2,associatedwithaglobalvalue
of50km2overthewholewatershed.
2.5. ComparisonbetweenLandsatandSPOTVEGETATIONAlbedoValues
Snowalbedovaluesretrievedfrombothdatasourceswereusedtoobtaindistributeddailyseries
ofsnowalbedoof30mˆ30m-cellsizefromOctober2011toAugust2013. Forthis,timeintervals
betweentwoconsecutiveimageswerefilledbyassumingtheacquisitiondatevalueduringNdays
beforeandafterwards;then,linearinterpolationwasappliedfortherestoftheintermediateperiod.
Nwasfixedas10and5daysforLandsatandSPOTVEGETATION,respectively.
TheresultingalbedomapswereusedasinputvaluesinthesnowmoduleofWiMMedmodel,
operationalatthestudysite,toassessalbedoresolutioninfluenceonthesnowregime.Twosimulations
wereperformed:Simulation1and2,usingalbedovaluesobtainedfromLandsatTM,ETM+,andSPOT
VEGETATIONdataset,respectively. Themodelresultsanalyzedineachsimulationwerethedaily
valuesofthefollowingsnowvariables: meansnowcoverfraction(SCF),meansnowwaterequivalent
(SWE),snowmelt(Melt)andevapo-sublimation(Ev).Theiraveragevaluesoverthewholestudyarea
arepresentedinthiswork,togetherwiththeiraveragevaluesovereachoneofthefouraltitudinal
regionspreviouslydefined.
TheSCFresultsshownforeachsimulationandareacorrespondtothemodelestimations;the
indirectinformationaboutSCFvaluesthatbothremotesensingsourcesprovidedhasnotbeenusedin
thiswork,sincethemaingoalwasassessingthealbedosourceinfluence. However,somedisagreement
mayappearbetweenthemodeldetectionofsnow(SCFdifferentfromzero)andtheimagedetection
(albedodifferentfromzero); insuchcases,theSCFestimatedvaluewasmaintained. Toavoidthe
Hydrology2016,3,10 9of19
associatednon-continuityinthecalculations,thefollowingcriteriawerefollowed: whenSCFresults
differentfromzero-whereaswhenthesnowalbedoisequaltozero,aminimumalbedovalueof0.4is
assignedbythemodel,providedSCFishigherthan0.5;similarly,whenthecalculatedSCFisequal
tozerobutthesnowalbedoisdifferentfromzero,thesnowalbedodataisignoredandazerovalue
isadopted.
Finally,toassesstheinfluenceofthefractionofmixedpixels(SCF<1)ontheresultsduringthe
meltingperiods,theSCFvaluesretrievedbyPimenteletal. (2014)[41]fromLandsatimagesbymeans
ofspectralmixtureanalysiswereusedfortheLandsatacquisitiondates(Figure2).
3. Results
3.1. SnowAlbedoTimeSeriesduringtheStudyPeriod
Figure5showsbothLandsat(L)andSPOTVEGETATION(SV)snowalbedodailyseriesusedas
inputinSimulation1andSimulation2,respectively,averagedoverthewholestudysite(Figure5a)
andeachoneofthefouraltitudinalregionsidentifiedinthiswork(Figure5b),duringthestudyperiod
(October 2011-August 2013); the periods with zero albedo correspond with no-snow conditions
in the study area (2012/05/28–2012/09/16; 2013/03/28–2013/05/31; 2013/06/07–2013/06/16;
2013/06/23–2013/08/31). Bothseriesexperimentasimilarevolutionwhenaveragedoverthewhole
studyarea(Figure5a),withcoincidentincreasinganddecreasingphasesduringthesnowseasons.
L-albedoisgenerallylowerthanSV-albedowhenaveragedoverthestudyarea,withameanvalue
forthestudyperiodof0.11against0.21;thedifferenceishigherduringwarmperiodsandincreases
fromthebeginningofspringtosummer. L-albedoexhibitsamaximumaveragevalueof0.36and
aminimumof0.01inMarch2013,respectively;forSV-albedo,thesevaluesare0.34and0.13,inthe
sameperiods.
Figure5.AveragedailysnowalbedovaluesretrievedfromLandsat(solidlightblueline)andSPOT
VEGETATION(soliddarkblueline)sourcesoverthetotalstudyarea(a),andeachaltitudinalregion
(b);seeregionclassificationinTable1.Dashedlinesrepresentthemaximumdailysnowalbedovalues
measuredthroughoutthestudyperiod,Landsat(grayline)andSPOTVEGETATION(blackline).The
areabetweendashedandzerolinelimitstherangeofcellvaluescorrespondingtoeachdayforeach
albedosource.
The regional analysis (Figure 5b) shows a clear decrease of the average snow albedo with
the elevation, with mean average L-albedo values of 0.38, 0.27, 0.08, 0.01 for Region 1, 2, 3 and
4,respectively,and0.38,0.35,0.18,0.14meanSV-albedovaluesforeachregiontoo. Althoughmean
L-albedo values are always lower than those from SV-albedo, the difference increases as altitude
Hydrology2016,3,10 10of19
decreases,ascanbeeasilyobservedfromtheFigure5. Again,thedifferenceishigherduringspring
andsummer,wherethepresenceofmixedpixelsismoresignificant.
ThevalidationofthealbedovaluesobtainedfromremotesensingsourcesisshowninFigure6
foreachoneoftheavailabledates. Ascanbeseen,theobtainedvaluesfallwithintheobservedrange,
taking into account the fact that the sensor-albedo values correspond to 1 ˆ 1 km and 30 ˆ 30 m
pixel areas from SPOT VEGETATION and Landsat respectively, whereas the in situ albedo values
arepointmeasurements. Themixednatureoftheareasrepresentedbyeachpixelinthestudysite
duringlongperiodsjustifiestheselowervalues,sinceitisasurfacevalueagainstasnowpointvalue,
andthepixelisnotjustasnowsurface,butaheterogeneousone. Onlypuresnowpixelscouldbe
approximateinthiscomparison. ARMSEof0.17and0.30wasfoundbetweeninsituvaluesandeach
sourcerespectively.
Figure6.Comparisonbetweenthesnowalbedoin-situmeasurementsthroughoutthestudyperiod
andthealbedoobtainedfrombothremotesensingsourcesatthemonitoringpointinColladoBuitreras
(Figure1).Dispersiongraphsforbothalbedoproductsareincluded.
3.2. SimulatedSnowCoverFraction
TheevolutionofthedailymeansnowcoverfractionsimulatedfrombothLandSV-albedovalues
(Simulation1and2,respectively)throughoutthestudyperiodisshowninFigure7. AsinSection3.1,
theresultsarerepresentedastheaveragevaluesoverthewholestudyarea(Figure7a)andovereach
region(Figure7b). Veryclosevaluesforbothsimulationsarefoundinallthecases,withgenerally
similarorlowerSV-SCF(Simulation2),withameandifferencefromL-SCFof0.02m2¨m´2(Figure7a),
andmaximumdifferencesbetween0.38and0.33associatedtotheoccurrenceofisolatedsnowfall
eventsinspring(March–April2012andMay2013)andduringsomeoftheintermediatefall-winter
meltingcycles(December2011,November2012)(Figure7).
Figure 8 shows the difference between the simulated daily L and SV-SCF values against the
differencebetweenthedailyLandSV-albedovalues,bothvariablesaveragedoverthewholestudy
area(Figure8a). FromFigure8b–e,itcanbeobservedthatthemeandifferenceofLandSV-albedo
valuesshiftsfromapositivevalueinRegion1(0.034)tonegativevaluesintherestofregions(´0.062,
´0.102,´0.127respectively),withanincreasingtrendintheirabsolutevalueasthealtitudedecreases
(redaxesinthegraphs),whereasthemeandifferencesfoundbetweenL-SCFandSV-SCFvaluesare
alwayspositive,alsoclearlydependentonthealtituderegion,withhighervaluesinRegions1and2,
wheresnowfallhasastrongerpresenceandpersistence,andageneraldecreasingtrendasaltitude
islower. ThemeanSCFdifferencefoundrangesfrom0.086inRegion1to0.020inRegion4. Thisis