remote sensing Article New Scheme for Validating Remote-Sensing Land Surface Temperature Products with Station Observations WenpingYu1,*,MingguoMa1,* ID,ZhaoliangLi2,JunleiTan3andAdanWu3 1 ChongqingEngineeringResearchCenterforRemoteSensingBigDataApplication,SchoolofGeographical Sciences,SouthwestUniversity,No.2TianshengRoad,BeibeiDistrict,Chongqing400715,China 2 ICube,Uds,CNRS(UMR7357),300BldSebastien-Brant,CS10413,67412Illkirch,France;[email protected] 3 HeiheRemoteSensingExperimentalResearchStation,NorthwestInstituteofEco-Environmentand Resources,ChineseAcademyofSciences,320DonggangWestRoad,Lanzhou730000,China; [email protected](J.T.);[email protected](A.W.) * Correspondence:[email protected](W.Y.);[email protected](M.M.) Received:27September2017;Accepted:20November2017;Published:24November2017 Abstract: Continuousland-surfacetemperature(LST)observationsfromground-basedstationsare an important reference dataset for validating remote-sensing LST products. However, a lack of evaluationsoftherepresentativenessofstationobservationslimitsthereliabilityofvalidationresults. In this study, a new practical validation scheme is presented for validating remote-sensing LST products that includes a key step: assessing the spatial representativeness of ground-based LST measurements. Threeindicators,namely,thedominantland-covertype(DLCT),relativebias(RB), and average structure scale (ASS), are established to quantify the representative levels of station observationsbasedontheland-covertype(LCT)andLSTreferencemapswithhighspatialresolution. WevalidatedMODISLSTsusingstationobservationsfromtheHeiheRiverBasin(HRB)inChina. Thespatialrepresentativeevaluationstepsshowthattherepresentativenessofobservationsgreatly differsamongstationsandvarieswithdifferentvegetationgrowthandotherfactors.Largedifferences inthevalidationresultsoccurwhenusingdifferentrepresentativelevelobservations,whichindicates alargepotentialforlargeerrorduringthetraditionalT-basedvalidationscheme. Comparisonsshow thatthenewvalidationschemegreatlyimprovesthereliabilityofLSTproductvalidationthrough high-levelrepresentativeobservations. Keywords: spatialrepresentativeness;heterogeneity;validation;land-surfacetemperatureproducts (LSTs);observations;HiWATER;remotesensing 1. Introduction Land-surfacetemperature(LST)isanimportantparameterrelatedtothesurfaceenergyandwater balanceatlocalandglobalscalesandhasprincipalsignificanceforapplicationssuchasmonitoringthe climate,hydrologicalcycle,andvegetation[1]. Satelliteremotesensingprovidesarepetitivesynoptic view in short intervals of the global land surface and is a vital tool for monitoring the LST of the Earth. Withthedevelopmentofremote-sensingtechnology,manyLSTproductshavebeenprovided bydifferentgroupsbasedonretrievalfromdifferentsatellitedata[2–5]. Thefirstlong-termglobal sensingLSTdataset,theNOAA/NASAPathfinderAVHRRLanddataset(PAL)[2],wasreleasedin 1994. ThesecondgenerationAVHRRLandPathfinderΠ(PALΠ)wasarefinementproductfromthe PALreleasedin2000[3]. SunandPinkerestimatedLSTproductsfromaGeostationaryOperational EnvironmentalSatellite(GEOS)in2003[4]. TheLSTsfromSpinningEnhancedVisibleandInfrared Imager (SEVIRI) onboard the Meteosat Second Generation (MSG) was retrieved in 2008 [6]. As a partofNASAEarthObservingSystem(EOS)project,MODISLSTshaveplayedanimportantrolein recentstudies,especiallyinregionalstudies,becauseofthesuitabletemporalandspatialresolution, RemoteSens.2017,9,1210;doi:10.3390/rs9121210 www.mdpi.com/journal/remotesensing RemoteSens.2017,9,1210 2of24 acceptableaccuracy,andaccessibilityoftheseLSTs. Therefore,MODISdailyLST/LSEproductswith 1-kmspatialresolutionarevalidatedinthisstudy. Remotely sensed LSTs must be appropriately and precisely evaluated to ensure effective application [7]. Mainly two types of methods exist for validating LST products retrieved from thermal-infraredsatellitedata: temperature-basedmethods(T-based)andradiance-basedmethods (R-based)[8–14]. ThemainadvantageofR-basedmethodsisthattheyworkduringboththedaytime andnighttimebecauseinsituLSTobservationsarenotrequired,andfindingvalidationsiteswith smallspatialvariationsinland-surfaceemissivityisrelativelyeasy[14]. However,theatmospheric andwatervaporprofilesatvalidationsitesfromradiosondeballoonsthataresynchronouslylaunched with the satellite are a necessary dataset, which limits the implementation of this method for long-termandlarge-regionvalidation.Therefore,T-basedmethodsremaincommon,andground-based measurements are still the primary source of datasets to directly validate remotely sensed LSTs. However,wecannotperformadirectcomparisonwithapixelgrid,especiallyforacoarse-resolution productoverheterogeneousareas,becauseofthespatialheterogeneityanddifferentscalesbetween ground-basedobservationsandremotelysensedLSTpixels. Agenerallyacceptedmethodisasystematicsite-to-networkmethod,whichdeeplydevelops an in situ sampling strategy and upscaling approaches to acquire the truth at the pixel scale over a heterogeneous surface based on multiscale, multi-platform and multi-source observations [15]. Thisapproachemploysbothfieldmeasurementsfromnodesofawirelesssensornetwork(WSN)and high-resolutionremote-sensingdatafromsynchronoushigh-resolutionsatellitesorairbornesensors to establish a site-specific relationship and generate high-resolution LST reference maps over the validationarea[15]. TheseLSTreferencemapsarethentreatedasbenchmarkstoobtainmultiscale validationdatasetsbyupscalingmethods[16–18]. However,onlyafewhigh-resolutionLSTreference mapscanbesynchronouslyobtainedforagivenregionbecauseofcostlimitationsandtherevisiting cyclesofsatelliteswithhighresolution,whichisthegreatestchallengetowardstheglobalvalidation ofLSTproducts,especiallyintermsoftemporalconsistencyinproductvalidation. IncontrasttomorecomplicatedR-based,site-to-networkmethodswithlimitedLSTreference maps, simple T-based methods are directly based on existing global, long-term ground LST measurementsandareanimportantsupplementforvalidation. SimpleT-basedmethodshavebeen widelyusedtovalidateremotelysensedLSTproductsathomogeneousstations[8,13,19].Whendirectly validating LST products with spatial resolutions above hundreds or even thousands of meters by ground-basedmeasurements,theerrorfromthescalemismatchchangeswiththeland-covertype(LCT) andtheproportionsofmixturesinpixelgridsreducethereliabilityofthevalidationandhinderthe applicationofground-basedLSTmeasurementsduringthevalidationofremotelysensedLSTproducts. Colletal. havepointedoutthatduringtheday, LSTcanvaryby10Kormoreoverafewmeters inaheterogeneoussurface[9]. Ground-basedLSTmeasurementsfromtwotypesLSTobservation instrumentswithdifferentfieldofview(FOV)wereselectedtodiscussthescalemismatchimplications forvalidationofremotesensingLSTproductsinthestudybyYuetal.,andthevalidationresultsshow thatthereisanextra26.9%intheerror>3Krangecausedbythe41.5FOVdifference[20].Therefore, wemustassessthespatialrepresentativenessofstationobservationsatagivenspatialresolutionto reliablyvalidateremotelysensedLSTs. Recently,severalmethodshavebeenusedtoassessthespatial representativenessofdifferentland-surfaceparameters,suchastheleafareaindex[21],surfacesolar radiation[22],bidirectionalreflectancedistributionfunction(BRDF)/albedo[23],airtemperature[24] and air quality [25], which are observed by ground stations. These methods are based on two factors: thepoint-to-areaconsistencyandthespatialheterogeneity[21]. Thepoint-to-areaconsistency indicatorcanbecalculatedthroughtwomethods. Thefirstinvolvesdirectlycomparingthefootprint of the ground-station observations to the corresponding product pixels [26] or the average value of the corresponding area [27]. In the second approach, the observational representativeness is determinedbytheaveragedifferencebetweenagivenstationanditsneighboringstationsbasedon multi-temporalobservationsfrommultiplestations[9,28]. Thesemi-varianceisusuallyselectedto RemoteSens.2017,9,1210 3of24 describethespatialrepresentativenessbyanalyzingthespatialheterogeneityaroundthestations[29]. Afirst-orderstatisticalalgorithmisanimportantspatialheterogeneityindicator,forexample,using window-size analysis to assess the spatial variation of the landscape around a given station [30]. Spatialrepresentativenessassessmentshavebeenwidelyimplementedtovalidatesatellite-albedo, evapotranspiration,andLAIproducts[21,23,26,31,32]. However,fewrepresentativenessassessments Remote Sens. 2017, 9, 1210 3 of 24 existforstationLSTobservations,whichincreasestheuncertaintyofthevalidationofLSTproducts, heterogeneity around the stations [29]. A first-order statistical algorithm is an important spatial particularlyforsimpleT-basedimplementations,andhinderstheapplicationofstationobservations. heterogeneity indicator, for example, using window-size analysis to assess the spatial variation of the ThispaperpresentsanewmethodologyforvalidatingLSTproductsthatfocusesonquantifying landscape around a given station [30]. Spatial representativeness assessments have been widely the spatiaiml prelepmreensteend tatot ivvaelnideastes soafteslltiatet-iaolbnedoob, seevravpaottriaonnsspirtaotioimn, parnodv LeAtIh peroadcuccutsr a[2c1y,23a,n26d,31r,e3l2i]a. bility of LST prodHuocwtevvaerl,i dfeawt iroepnr.eseTnthaetivteenremss a“sssepssamtieanltsr eexpisrte fsoer nsttaatitoinv eLnSTe sosb”serrveafteiorsns,t wohmiche ainscureraesmes ethnet s of the degree touwncheritcahingtyr oofu tnhed v-balaidsaetdiono bofs LeSrTv aptrioodnusctcs,a pnarrteicsuolalvrlye ftohr esismuprlreo Tu-bnadseidn gimLpSleTmebnytaetixontes,n adnidn g to the hinders the application of station observations. satellite footprint. This validation technique assesses the spatial characteristics of the LST, and This paper presents a new methodology for validating LST products that focuses on quantifying the seasonal representativeness changes within a statistical framework. A scheme that is based the spatial representativeness of station observations to improve the accuracy and reliability of LST on spatiaplrroedpurcet sveanlidtaattiiovne. nTehses teirnmd i“csaptaotirasl riesprperseensteantitveednesisn” trhefiesrsp tao pmeera,saunredmetnhtes nof tthhee dgergardeei ntog criteria are outlinwehdicihn gdroeutnadil-.baAseldl othbseersvtaattioionns scafnr ormesoltvhee thHe esiuhreroWunadtinegrs LhSeTd bAy lelxietedndTinegle tmo ethtrey saEtexllpiteer imental footprint. This validation technique assesses the spatial characteristics of the LST, and the seasonal Research (HiWATER) [33] are selected for applying the validation strategy. The study area and representativeness changes within a statistical framework. A scheme that is based on spatial data-processing procedure are introduced in Section 3. In Section 4, the representativeness of the representativeness indicators is presented in this paper, and then the grading criteria are outlined in givenstatdioetnaiol. bAsellr vthaet iostnastioisnsa sfsroemss etdhet oHveaihleid aWteateMrsOheDd ISAlVlie5dd TaeillyemLeStrTy pErxopderuimctesn(taMl OReDse/aMrchY D11A1). Therepres(HeniWtaAtiTvEeRn)e [s3s3]a asrsee sseslemcteedn tfoarn adppLlySinTgp threo dvaulicdtavtiaonli dstarattieogny.a Trheea sltsuodya naraelay aznedd daatnad-prdoicsecsusisnsge dinthis Section. Fpinroacleldy,utrhe earceo innctrloudsuiocends ianr eSescutimonm 3a. rIinz eSdec.tion 4, the representativeness of the given station observations is assessed to validate MODIS V5 daily LST products (MOD/MYD11A1). The representativeness assessment and LST product validation are also analyzed and discussed in this 2. Methodology Section. Finally, the conclusions are summarized. 2.1. NewValidationScheme 2. Methodology TheLSTisaland-surfaceparameterwithgreatspatialandtemporalheterogeneity,whichcreates 2.1. New Validation Scheme manychallengesfor“point-to-pixel”comparisons. Localchangesinthesurfacetemperaturewithin The LST is a land-surface parameter with great spatial and temporal heterogeneity, which andbetweendifferentecosystemsintroducescalemismatcherrors. Moreover,thesepatternschange creates many challenges for “point-to-pixel” comparisons. Local changes in the surface temperature seasonallyandareparticularlydifficulttoidentifyduringperiodsofrapidlychangingsurfaceconditions. within and between different ecosystems introduce scale mismatch errors. Moreover, these patterns Thereforec,h“apnogein set”asmonealalys uanredm aree nptasrtaicluolnarelya drieffincuoltt stou ifdfeicniteifny tdtuorinvga pliedriaotdes soaf trealpliidtely- dchearnivgiendg sLuSrfTacree trievals, especiallycornedmitiootnes-. sTehnesreinfogreL, “SpToinptr”o mdeuacsutsrewmeitnhts maloondee arrae tneoat nsudffilcoiewnt rtoe svoalliudtaitoe nsa(teilllliutes-tdrearitveedd iLnSTF igure1). retrievals, especially remote-sensing LST products with moderate and low resolution (illustrated in Thistemporalmismatchcanbesolvedbyincreasingtheobservation-acquisitionfrequencyofstations. Figure 1). This temporal mismatch can be solved by increasing the observation-acquisition frequency Theschemethatisdevelopedheretovalidateremote-sensingLSTproducts(seeFigure1)attempts of stations. The scheme that is developed here to validate remote-sensing LST products (see Figure 1) to solve sapttaetmiapltsm tois smolvaetc shpaetifafle mctissmdautcrhin egffevctasl diduaritnigo nv.aliTdahteionk.e Tyhein ketyh iins tshcihs escmheemies isto toa asssseessss tthhe e spatial representasptiavteianl eresps,rewsehnitcahtiviesnbesass, ewdhoicnh rise mbaosetde -osne nresminogted-saetnasiwngi tdhatha iwghiths phiagthia slpraetisaol lruestiooluntitohna tthaatr eclosely related toatrhe ecloLsSelTy. reInladteidc atot othres LaSrTe. pInrdoicpaotosresd arteo prqoupaosnetdif tyo qthuaenstipfya tthiae lspreatpiarle rseepnretsaetnivtaetinveensse,ssa, nandd then the then the grading criteria are designed before selecting appropriate ground-based measurements for gradingcriteriaaredesignedbeforeselectingappropriateground-basedmeasurementsforvalidation. validation. Figure1.FNigeuwres 1c. hNeemw escfhoermlea nfodr lsaundrf saucrefatceem tepmepreartauturree ((LLSSTT) )vvalaidliadtiaotni obnasebda soend thoe nastshesesmasesnet sosfm loecanl toflocal spatialrepspreatsiealn rteaptrievseennteastisve(snietses l(esivtee lle)v.eIln). Itnh ethsec shcheemmee,, LLCCTT isi sthteh aebabrbebvriaetvioina toifo lnanodf-cloavnedr -tycpoev, earndty pe,and NDVI is the abbreviation of normalized difference vegetation index. NDVIistheabbreviationofnormalizeddifferencevegetationindex. RemoteSens.2017,9,1210 4of24 2.2. IndicatorsforAssessingSpatialRepresentativeness Three indicators are proposed to describe the spatial characteristics for a specific parameter, mainlyincludingtheconsistencyfrompointstopixelsandthespatialheterogeneitywithinpixels. These indicators are calculated on high-resolution images, which are much easier to access, thus simplifyingtheprocess. TheLSTisadirectparameterforassessingtherepresentativenessofagiven station’sLSTobservations. However,obtainingtemporalhigh-spatial-resolutionLSTmatchesforLST productswithlowerresolutionisdifficult. Therefore,high-resolutionland-covertype(LCT)andthe normalizeddifferencevegetationindex(NDVI),whichcanindicatethesurfaceconditionsandtheir changes,arechosenasadditionalsupportingparameterstoevaluateground-basedLSTobservations inadditiontoLSTdatawithahighspatialresolution. IftheLCTsobservedbystationsdonotmatchthedominantLCTsinthepixels,thesestationLST observationscannotrepresenttheLSToftheLCTsinthepixels. Thus,thedominantLCT(DLCT), whichisgivenbythepercentageoftheobservedLCTthroughoutthepixel’sarea,isdefinedas M(s) DLCT= ×100 (1) N(s) where s is the product pixel, M(s) is the area with the LCT that is observed by the given station, andN(s)isthetotalareaoftheLCTsintheLSTproduct’spixelgrid. Whenusingahigh-resolution LCT map, the DLCT can also be described as the percentage of fine pixel numbers covered by station-observed LTC to total LCT pixel numbers in the LST product’s pixel range. A high DLCT indicatesthattheLCTobservedbyastationisconsistentwiththatintheLSTproduct’spixelandlow heterogeneitywithintheproductpixelbecausethemixingrateofLCTsinthepixelisnotlarge. Wedevelopedarelativebias(RB)indicatortoassesshowcloseaground-basedLSTmeasurement istothevalueofthecorrespondingpixelarea. Accordingtothehigh-spatial-resolutionLSTreference images, therelativebiasisusedtodescribethedifferencebetweentheLSTvalue T(s) atastation andtheaverageLSTvalueT(s)intheproductpixel’sarea. Ifweconsiderthecomparabilitybetween differentrangesofLSTvalues,theRBisdefinedas |T(s)−T(s)| RB= ×100 (2) T(s) wheresistheproductpixelandRBdependsonbothitsresolutionandtheresolutionofthereference LSTimage. Thisindicatorcanquantifythecertaintyoftheground-basedmeasurementstotheproduct pixel area LST values. A smaller RB indicates more spatially representative observations for the correspondingpixelatthespecificspatialresolutionofs. Thetwoaboveindicatorsmainlymeasurethepoint-to-areavalueconsistency,sotheheterogeneity ofthespatialdistributionoftheLSTwithinapixel,whichiscorrelatedwiththevegetationgrowth,must beseasonallyquantified. IntermsofthespatialautocorrelationofLSTparameters,semivariogramis oneofthemostcommonlyusedandefficientgeostatisticalanalysistoolsforquantitativelyevaluating spatial variations. Semi-variance and related geostatistical kriging were developed from mining researchduringthelate1950sandhavebeenwidelyusedafterapublicationbyJournelandHuijbregts in1978[34,35]. Thesegeostatisticaltechniqueshavebeenusedinmanyscientificprojects,suchasin describingthedistributionanddensityofplantsandanimals[36,37]andindeterminingthespatial scalesofvariationandsamplingstrategiesinremotesensing[38–40]. Inregionalizedvariabletheory, thesemi-variancemeasuresthedissimilarityofaspatialvariableobservedatdifferentlocations. The semi-varianceiscalculatedbytheaveragesquareddifferencebetweenobservationsZ(x )andZ(x ), i j whichareseparatedbydistanceh,asdescribedbelow: r(h) = 1 ∑ (Z(x )−Z(x ))2 (3) 2N(h) i j ||xi−xj||=h RemoteSens.2017,9,1210 5of24 where2N(h)isthenumberofobservationpairs,whichareseparatedbyadistanceh,orlag,asintervals tocalculatethesemi-variance. Inthisstudy,thevariogramestimatorr(h)iscomputedondiscretized point values from high-spatial resolution LST pixels, and then a variogram model is established asaparametricfunctionalapproximationbasedonthesesemi-variancevalues. Severaltheoretical variogrammodeltypesexist, includinglinearmodels, sphericalmodels, exponentialmodels, and Gaussianmodels. Amongthesemodels,sphericalmodelsarethemostwidelyusedvariogrammodels fortheirstrongfittingandgeneralizationcapabilitiesandarerecommendedforassessingthespatial representativenessofobservations[23,32]. Theisotropicsphericalvariogramthatisusedtoestimate thevariogramisasfollows: (cid:18) (cid:16) (cid:17)3(cid:19) c +c× 1.5× h −0.5× h for0≤ h ≤ a r (h) = 0 a a (4) sph c +c forh > a 0 In Equation (4), it is obvious that the r (h) generally increases from a nonzero value to a sph relativelystableconstantvaluewithh. Whenh =0,ther (h)valueisanonzerovaluec ,namely, sph 0 r (0) = c , and c isthenuggetofthevariogram. Thestableconstantvalueisthesill (c +c) of sph 0 0 0 thevariogram. Whenr (h)reachesthesill,thevalueofthevariablehisa,namely,therangeofthe sph variogram. Thekeyparametersrange(a),nugget(c ),andsill(c +c)canbeobtainedfromfitting 0 0 Equation(4). aisthemaximaldistancebetweenthetwocorrelatedpointsandindicatestheaverage structural scale (ASS) of the given area. The nugget c indicates the level of Z(x )’s randomness, 0 i which may be caused by internal variations in Z(x ) over a smaller distance h than the sampling i distanceormaybederivedfromthesamplingerror. Thesill(c +c)representsthelargestextentof 0 theregionalizedvariation. LiandReynolds[41]introducedtheproportionofstructuralvariation, whichisbasedonsubtractingthevariogramnugget(c )fromthesill(c)andthendividingbythesill, 0 todiscussthedefinitionandquantificationofecologicalheterogeneity. However,theheterogeneity ofLSTparametersconsiderablyvariesovertimewhencomparedtothatofecologicalparameters, andreferencehigh-resolutionLSTmapsmaynotbecompletelysynchronouswiththevalidatedLST products. Therefore,theASSisintroducedbasedontherangeofthevariogrammodel,whichreflects theaveragestructuralscaleofthegivenareaandthesizeofthehomogeneousarea. AlargeASSvalue indicatesthatthestationobservationsrepresentalargehomogeneousarea. 3. DataInstructionandPreparation 3.1. Ground-BasedLSTMeasurements Inthisstudy,weselectedtheHeiheRiverBasin(HRB),whichisthesecondlargestinlandvalley in China’s arid regions, to evaluate MODIS LSTs with 1-km resolution. The HRB is located in the northernaridregionwithin97.1◦E–102.0◦Eand37.7◦N–42.7◦N.Glaciers,frozensoils,alpinemeadows, forests, irrigated crops, riparian ecosystems, deserts, and gobi are distributed from upstream to downstreamregions(seeFigure2). TheHRBwasselectedasanexperimentalwatershedtorevealthe processesandmechanismsoftheecohydrologicalsysteminaninlandriverbasin. Alliedtelemetry experimentssuchastheHeiheBasinFieldExperiment(HEIFE)[42]andWatershedAlliedTelemetry ExperimentalResearch(WATER)[43]havebeenconductedintheHRB,andtheHiWATER[33]project has been ongoing since 2012. The stations that collect watershed hydrological observations cover a wider range than those in previous studies and provide a large amount of ecohydrological data forevaluation. EighteenatmosphericstationsfromHiWATERarescatteredaroundtheHRBregion. Longwave-radiationdataforeighteenstationsfrom2013to2014wereselectedtoobtainground-based LSTstoevaluatetheMODISLSTs. ThelocationsofthestationsareshowninFigure2. Theinformation for the stations is listed in Table 1, and environmental photos of these sites are shown in Figure 3. TheARC,ARS,ARY,DSL,JYL,HZS,HCG,andEBZstationsarelocatedintheupstreamarea;the DMZ,GBZ,HZZ,SDZ,andSSWstationsarelocatedinthemidstreamarea;andthedownstreamarea RemoteSens.2017,9,1210 6of24 containstheSDQ,HJL,HYZ,NTZ,andLTZstations. TheLCTsofthesestationsarealltypicaltypesin thethreesignificantlydifferentareas. Remote Sens. 2017, 9, 1210 6 of 24 FiFgiugruere2 .2.S Stutuddyya arreeaa aanndd llooccaattiioonnss ooff tthhee vvaalliiddaatitoionn stsatatitoionns.s . TTabablele1 1..I nInffoorrmmaattiioonn ffoorr tthhee ssttaattiioonnss iinn tthhisis sstutuddyy. . SSttaattiioonn NNaammee LLoonnggiittuuddee//◦°EE LLataittiutudde/e◦/N°N AAltlittiutudde/em/m HHeeiigghhtt//mm FFoooottpprriinntt 11/m/m LLaannddssccaappeess A’rou superstation (ARC) 100.46 38.05 3033 5 37.32 Alpine meadow A’rousuperstation(ARC) 100.46 38.05 3033 5 37.32 Alpinemeadow A’rou sunny slope station 100.52 38.09 3559 6 44.78 Alpine grassland (AA’rRoSu) sunnyslope 100.52 38.09 3559 6 44.78 Alpinegrassland station(ARS) A’rou shade station (ARY) 100.42 37.99 3538 6 44.78 Alpine grassland DA’arsohuaslohnagd estsatatitoionn (D(ASRLY) ) 19080.9.452 373.89.983 33573785 66 4444.7.788 ASwlpainmepg mraesasldaonwd JDinayshanalgolninggs statatitoionn( D(JSYLL)) 9180.19.511 383.78.385 33777050 66 4444.7.788 SAwlpaimnep mmeeaaddooww Huangzangsi station (HZS) 100.19 38.23 2660 6 44.78 Cropland (wheat) Jinyanglingstation(JYL) 101.11 37.85 3700 6 44.78 Alpinemeadow Huangcaogou station (HCG) 100.73 38.00 3186 6 44.78 Alpine grassland Huangzangsistation(HZS) 100.19 38.23 2660 6 44.78 Cropland(wheat) E’bo station (EBZ) 100.94 37.96 3407 6 44.78 Alpine grassland DHaumanagnc asuogpoeurstsatatitoionn (D(HMCZG)) 110000..7337 383.80.086 31158169 162 8494.5.778 AClrpoipnleangdra (smslaanizde) GE’obboi ssttaattiioonn ((EGBBZZ)) 110000..9340 373.89.689 31450771 66 4444.7.788 AGlopbini Degesraesrts land HDaumazahnasizuip deerssteartti ostnat(iDoMn Z) 110000..3372 383.88.677 11571296 21.25 1889.6.567 CDreospelratn sdte(pmpaei ze) (HZZ) Gobistation(GBZ) 100.30 38.89 1571 6 44.78 GobiDesert Wetland station (SDZ) 100.45 38.97 1460 6 44.78 Wetland SHhueanzshhaaiwziod deseesretrst tsattaiotinon(H ZZ) 100.32 38.77 1726 2.5 18.66 Desertsteppe 100.49 38.79 1582 6 44.78 Desert (WSSetWla)n dstation(SDZ) 100.45 38.97 1460 6 44.78 Wetland PShoepnuslhuas wfooredsets setrattion (HYZ) 101.12 41.99 927 24 179.14 Populus forest 100.49 38.79 1582 6 44.78 Desert Cstraotipolnan(SdS sWta)tion (NTZ) 101.13 42.01 919 6 44.78 Cropland Barren-land station (LTZ) 101.13 41.99 931 6 44.78 Bare soil Populusforeststation(HYZ) 101.12 41.99 927 24 179.14 Populusforest Euphrates poplar Croplandstation(NTZ) 101.13 42.01 919 6 44.78 Cropland Sidaoqiao station (SDQ) 101.12 41.99 935 10 74.64 olive and Tamarix Barren-landstation(LTZ) 101.13 41.99 931 6 44.78 Bmairxeedso fiolrest EEuupphhrraatteess ppooppllaarr MSidixaeodq ifaooresstta tsitoantio(SnD (QH)JL) 110011..1123 414.19.999 993259 2104 17749..6144 oolliivvee aanndd TTaammaarriixx mmiixxeedd ffoorreesstt 1 “Footprint” refers to the diameter of the footprint; “Height” indicates the installationE huepihgrahtte.s poplar Mixedforeststation(HJL) 101.13 41.99 929 24 179.14 oliveandTamarix mixedforest Eighteen meteorological towers are located in the HRB region, consisting of three superstations 1“Footprint”referstothediameterofthefootprint;“Height”indicatestheinstallationheight. and fifteen ordinary stations. Pyrgeometers are deployed at these 10-m to 35-m high meteorological station towers to measure longwave radiation (see Figure 3 and Table 1). At least two pyrgeometers EighteenmeteorologicaltowersarelocatedintheHRBregion,consistingofthreesuperstations are positioned on a single tower: one facing upward and the other facing downward. The field-of- andfifteenordinarystations. Pyrgeometersaredeployedatthese10-mto35-mhighmeteorological view (FOV) of the upward-facing pyrgeometer is nearly 180°, while that of the downward-facing stationtowerstomeasurelongwaveradiation(seeFigure3andTable1). Atleasttwopyrgeometers pyrgeometer is 150°. Therefore, the effective diameter of the FOV of the pyrgeometers on a 10-m to arepositionedonasingletower: onefacingupwardandtheotherfacingdownward. Thefield-of-view 35-m tower is approximately 2.5–24 m with a 6-m average mounting height, and the diameters of the ground-observation footprints are shown in Table 1. The pyrgeometers are sensitive to the spectral range from 4.5 to 42 µm in the longwave band. All the instruments at each station were calibrated RemoteSens.2017,9,1210 7of24 (FOV) of the upward-facing pyrgeometer is nearly 180◦, while that of the downward-facing pyrgeometeris150◦. Therefore,theeffectivediameteroftheFOVofthepyrgeometersona10-mto 35-mtowerisapproximately2.5–24mwitha6-maveragemountingheight,andthediametersofthe ground-observationfootprintsareshowninTable1. Thepyrgeometersaresensitivetothespectral rangefrom4.5to42µminthelongwaveband. Alltheinstrumentsateachstationwerecalibrated beforeandafterfielddeployment. Field-routingexamswereimplementedoncepermonth. Assurance andqualitycontrolprovidedthebestpossibledataforthelevel-2dailydata. Allthesedataandrelated informationcanbefoundattheHiWATERwebsite[44]. Alltheground-basedmeasureddatafromthe eighteenRHemioWte SAenTs.E 20R17,s 9i,t 1e2s10w ere10-minaveragedvalues. Thelongwaveradiationdataw7 oef r2e4 selected accordingtothefieldviewingtimefortheMODISLST.TheLSTisrelatedtoland-surfacelongwave before and after field deployment. Field-routing exams were implemented once per month. radiationAascscuorarndcien agntdo qtuhaelitSyt ecofanntr-oBl oplrtozvmidaedn nthlea bwes[t1 p,4os5s]i:ble data for the level-2 daily data. All these data and related information can be found at the HiWATER website [44]. All the ground-based measured data from the eighteenL ↑H=iWεAbTδETR4 s+ite(s1 w−erεeb )10×-mLin↓ averaged values. The longwave (5) radiation data were selected according to the field viewing time for the MODIS LST. The LST is where L↑reilsatethd eto slaunrdfa-scuerfaucpe wlonegllwinavge lroadnigatwioanv aeccorraddiniagt tioo nth,e εSbtefiasnt-Bhoeltzsmurafnanc leawb r[1o,4a5d]:b and emissivity, δ isStefan-Boltzmann’sconstant(5.67 ×(cid:1838)10=−8(cid:2013) W(cid:2012)(cid:1846)·(cid:2872)m+−(12·−K(cid:2013)−)4)×,(cid:1838)an d L↓ istheatmosphericdo(5w) nwelling ↑ (cid:3029) (cid:3029) ↓ longwaveradiationatthesurface. Therefore,theground-measuredLSTcanbeestimatedfromstation where (cid:1838) is the surface upwelling longwave radiation, (cid:2013) is the surface broadband emissivity, (cid:2012) is longwave-radiat↑ionobservationsbythefollowingequat(cid:3029)ion: Stefan-Boltzmann’s constant (5.67 × 10−8 W∙m−2∙K−4), and L is the atmospheric downwelling ↓ lloonnggwwaavvee -rraaddiiaattiioonn aotb tsheer svuartifoances. T bTyh =ethree(cid:20)f ofoLrle↑l,o t−whein( gg1r oe−uqnuεdabt-)imo×ne:a sLu↓re(cid:21)d14 LST can be estimated from station (6) s εbδ (cid:2869) InEquation(6),L↑andL↓areobtain(cid:1846)e(cid:3046)d=fr(cid:3428)o(cid:1838)↑m−t(h1e(cid:2013)−g(cid:2012)(cid:2013)r(cid:3029)o)u×n(cid:1838)d↓(cid:3432)-(cid:2872)b asedmeasurements. Sevenn(6a) rrowband (cid:3029) emissivities exist in MOD/MYD11B1 LST/LSE products. ε can be estimated from these MODIS In Equation (6), (cid:1838) and (cid:1838) are obtained from the gbround-based measurements. Seven ↑ ↓ narrowbannadrroewmbiasnsdi vemitiiesssiv[i4ti5e]s: exist in MOD/MYD11B1 LST/LSE products. (cid:2013) can be estimated from these (cid:3029) MODIS narrowband emissivities [45]: ε =0.2122×ε +0.3859×ε +0.4029×ε (7) b (cid:2013) =0.2122×2(cid:2013)9 +0.3859×(cid:2013) 31+0.4029×(cid:2013) 32 (7) (cid:3029) (cid:2870)(cid:2877) (cid:2871)(cid:2869) (cid:2871)(cid:2870) whereεbisthebroadbandemissivity,andε29,εε31,aεndε32aεrethenarrowemissivityproductsofMODIS bands29,w3h1e,rea n(cid:2013)(cid:3029)d i3s 2thteh bartoaadrbearnedt reimevisesidviftyro, amndt he29M, O31D, aInSd da3y2 /arnei gthhet nLarSrTowa legmoisrsiitvhimty p(rModOucDts/ oMf YD11B1 MODIS bands 29, 31, and 32 that are retrieved from the MODIS day/night LST algorithm LST/LSE). (MOD/MYD11B1 LST/LSE). Figure3.Cont. RemoteSens.2017,9,1210 8of24 Remote Sens. 2017, 9, 1210 8 of 24 Figure3.FiPghuroet o3.s Pohfottohse of1 8theH 1e8i hHeeiWhe aWteartserhsehdedA Alllliieedd TTeelelmemetreyt rEyxpEexripmeernitmal eRnetsaealrRche s(HeaiWrcAhT(EHR)i WATER) stations.Tsthateiotnws.o Thpey trwgoe opmyrgeetoemrsettehras tthwate wreerue suesdedt too rreeccoorrdd thteh leonlognwgawvea rvadeiartaiodni awtieoren dwepeloreyedde apt laony edatan average height of 6 m, with one facing upwards and the other facing downwards. averageheightof6m,withonefacingupwardsandtheotherfacingdownwards. 3.2. Remote-Sensing Data 3.2. Remote-SensingData 3.2.1. MODIS Data 3.2.1. MODISADs aat acomponent of NASA’s Earth Observing System (EOS) project, two MODIS instruments were placed onboard the Terra and Aqua satellite platforms to provide information for global AsacomponentofNASA’sEarthObservingSystem(EOS)project,twoMODISinstrumentswere atmosphere-, land- and oceanic-process studies [46]. MOD/MYD22_L2, MOD/MYD11A1, placedonMboOaDr/dMtYhDe1T1Be1rr aanadn MdOADq/MuaYDsa0t7e_lLl2it earpe laaltl fdoarimly sLtSoT pprroodvuidctes ibnafsoedrm ona ttihoenrmfoalr-ignflroabreadl adtamta osphere-, land- andfroomc eMaOnDicI-Sp. rMoOceDs/sMYstDu1d1_ieLs2 w[4a6s] r.etrMievOedD b/yM a YgeDne2r2a_lizLe2d, sMpliOt-wDin/dMowY Dalg1o1rAith1m, wMitOh D1-/kmM YD11B1 and MOD/MYD07_L2 are all daily LST products based on thermal-infrared data from MODIS. MOD/MYD11_L2 was retrieved by a generalized split-window algorithm with 1-km spatial resolution [47]. MOD/MYD11A1 is tile-based and gridded in the sinusoidal projection from MOD/MYD11_L2. MOD/MYD11B1wasobtainedusingtheday/nightLSTalgorithmat5-kmspatial resolution[48]. MOD/MYD07_L2wasretrievedbytheatmosphericteamusingstatisticalregression methods [49]. In this study, we focus on the collection of 5 MOD/MYD11A1 products, which are morewidespread. Uncertaintiesfromthesatellitemeasurementsandimprovementsintheoriginal RemoteSens.2017,9,1210 9of24 MODISLSTsforcloudydaysarebeyondthescopeofthispaper.Toeliminateeffectsfromtheinversion algorithmandclouds,onlypixelswithhigh-qualityMODISLSTswereselectedfortheevaluation basedonaqualitycontrolflagvalueof0. ThenarrowemissivityproductsfromMOD/MYD11B1 LST/LSEwereselectedtoestimatetheland-surfacebroadbandemissivityε andobtainground-based b LSTs[45]. ThenarrowemissivitiesfromMOD/MYD11B1LST/LSEat5-kmresolutionwereresampled to1-kmspatialresolutiontomatchtheevaluatedMODISLSTs. 3.2.2. High-Spatial-ResolutionImages Land-cover images with a spatial resolution of 30 m (data doi:10.3972/hiwater.155.2014.db, downloadedfromtheHiWATERlandcoverdatasets[50],whichwereproducedbyZhongetal.[51,52], wereselectedtoobtaintheDLCTina1-kmLSTproductpixel. Thisdatasetwasmainlybasedon charge-coupled device (CCD) data from the Huan Jing 1 (HJ-1) satellite, which was launched on 6 September 2008, by the China Center for Resources Satellite Data and Application (CRESDA). HJ-1/CCDhasthreevisiblebandsandonenear-infraredband[53]. Thisdatasetprovidesmonthly land-covermapsoftheHRBfrom2011to2015with30-mspatialresolution. LCTschangewiththe seasons,andthesechangesaresimilaracrossconsecutiveyears,sotheDLCTproductsin2013were collectedtocalculatetheDLCTindicator. Landsat8,whichiscalledtheLandsatDataContinuityMission,isextendingthedistinguished 40-yearrecordsofLandsat-seriessatellitesandhasenhancedcapabilities,suchasaddingnewspectral bandsinthevisibleandthermal-infraredwavelengthsandimprovingthesignal-to-noiseratioand radiometric resolution of the sensor [54]. The Landsat 8 satellite includes two instruments: an Operational Land Imager (OLI) and Thermal Infrared Sensor (TIRS). High-resolution LST maps wereretrievedfromtheLandsat8TIRSdataandOLIdata. BeforeretrievingtheLSTmaps,theLandsat OLIandTIRSimageswerepreprocessed,includingradiometriccalibrationandatmosphericcorrection basedonthecorrectionmodelintheENVIsoftware. MonthlyNDVImapswereobtainedtoassessthe relationshipsbetweenthemonthlychangesintheindicatorsandvegetationgrowth. TheNDVImaps werebasedonthevisibleredband(R,band4)andthenear-infraredband(NIR,band5)accordingto thefollowingequation: NIR−R NDVI= (8) NIR+R Inthisstudy,theLSTdatawereestimatedfromtheTIRSaboardLandsat8basedonapractical split-window (SW) algorithm developed by Du et al. [55]. The SW algorithm can be expressed asfollows: T = b +(cid:18)b +b 1−ε +b ∆ε(cid:19)Ti+Tj +(cid:18)b +b 1−ε +b ∆ε(cid:19)Ti−Tj +b (cid:0)T −T(cid:1)2 (9) 0 1 2 ε 3 ε2 2 4 5 ε 6 ε2 2 7 i j whereTisLST,T andT aretheTOAbrightnesstemperaturesinthethermal-infraredchannelsiand i j j,respectively; εistheaverageemissivityofthetwochannels(i.e.,ε = 0.5(+ε )); ∆εisthechannel j emissivitydifference(i.e.,ε =0.5(ε −ε ));andb (k =0, 1, ... 7)arethealgorithmcoefficientsfrom i j k the simulated dataset. In this algorithm, the coefficients were determined based on atmospheric water-vaporsubranges,whichwereobtainedthroughamodifiedsplit-windowcovariance-variance ratiomethod. Thechannelemissivitieswereacquiredfromnewlyreleasedgloballand-coverproducts at30mandafractionofthecalculatedvegetationcoverfromvisibleandnear-infraredimagesthat wereobtainedbyLandsat8. Theeffectofheterogeneitychangesdependingontheseason,soweselectedLandsat8datafrom September2013toAugust2014witha16-daytemporalresolutiontocalculatetheRB,andASS.Atotal of92Landsat8imagesforallthestationsintheHRBweredownloadedfromthefollowingUSGS website[56]. Thestatisticalresultswerebasedonper-monthaveragestoeliminateinvaliddatafrom cloudcover. RemoteSens.2017,9,1210 10of24 4. ResultsandDiscussion 4.1. SpatialRepresentativenessClassification SincetheDLCTandRBcanmeasurepoint-to-pixelvalueconsistency,andASScanassessthe spatial patterns for a given station, respectively. Therefore, all three indicators were combined to describe the representativeness including point-to-pixel consistency and spatial heterogeneity. Thespatialrepresentativenesswasalsoclassifiedbasedonthesethreeindicators. TheDLCTindicator determines the representativeness of the station’s LCT in the product pixel. When the LCT in the viewfootprintofastationisnotdominantwithintheproductpixel,thestationobservationscannot berepresentativeofthepixel,andtheLSTvalueofallothervegetation-covertypesmaybeignored, evenifthepoint-to-areaLSTconsistencyishighatthestationsometime. Whenapixelhasalarge DLCTvalue,theRBandASSsubsequentlywoulddeterminethespatialrepresentativenesstogether. Presumably,thestation-observedLSTsrepresentidealdataforLSTproductvalidationiftheRBvalueis smallandtheASSvalueislarge. IftheRBandareASSvaluearebothsmall,thestationmayhavesome spatialrepresentation. Insomeextremecases,thestationobservingareaisanaverageheterogeneity sub-areasintheproductspixel,whichmeansthestationobservationsarerepresentativeforpixels, althoughthesurfaceisheterogeneousintheseproductspixels. Finally,iftheRBislargebuttheASSis small,theobservationsprobablydifferfromthevaluesofthepixel. ReasonablethresholdsfortheDLCT,RBandASSarerequiredtodeterminetherepresentativeness levelofagivenstation’sobservations. Theemissivityproductsoftheversion5collectionofMODIS LST/LSEproductsareretrievedbasedontheLCTofapixel,andtheLCTisclassifiedasthelandcover ofthispixelbasedontheclassificationruleforMODISland-coverproducts(MCD12Q1)ifthearea percentofoneLCTinapixelishigherthan60%[57]. Thus,60%wasdefinedasthethresholdofthe DLCTinthisstudy. TheRBwasusedtoevaluatethedifferencebetweentheland-basedmeasurements withintheviewfootprintateachstation(inTable1,theviewfootprintsareshowninthesixthcolumn) andthemeanpixelvalueatthestationlocations. TheidealRBvalueisclosetozero. However,the thresholdoftheRBisnotuniqueanddependsonthespatialresolutionoftheLSTproducts,theview footprintofthestationmeasurements,andtheretrievalaccuracyofthehigh-spatial-resolutionLST mapsthatareusedtoevaluatetherepresentativeness. Thus,areasonablethresholdfortheRBwas 0.5%inthisstudyforMODISLSTproductswith1-kmspatialresolution,astationviewfootprintabove 30manda1-Kretrievalerrorfromthehigh-resolutionLSTmapitself[55]. TheASSiscalculatedfrom variogrammodelsbasedonthesemi-varianceina3-km×3-kmareathatiscenteredatagivenstation andcanindicatethegreatestdistanceoverwhichthevalueatapointonthesurfaceisrelatedtothe valueatanotherpoint. TheASSdefinesthemaximumneighborhoodoverwhichcontrolpointsshould beselectedtoestimateagridnodetotakeadvantageofthestatisticalcorrelationamongobservations andcandescribethespatialdistributionofLSTsandquantifytheaverageLSTspatialstructuresin thegivenarea. Inthisstudy,themeasurementsfromstationswereusedtoevaluatetheMODISLSTs witha1-kmspatialresolution. Therefore,areasonableASSindicatorshouldbelargerthanthespatial resolutionoftheLSTproducts,thatis,1kminthisstudy. Thespatialrepresentativenessofthestation’sLSTobservationswasclassifiedintofivedifferent levels based on the difference-constraining degrees of the three indicators and their thresholds. ThelevelsandtheirdescriptionsarepresentedinTable2.
Description: