RemoteSensingofEnvironment187(2016)30–48 ContentslistsavailableatScienceDirect Remote Sensing of Environment journal homepage: www.elsevier.com/locate/rse Dynamics of the Terra Nova Bay Polynya: The potential of multi-sensor satellite observations ThomasHollandsa,⁎,WolfgangDierkinga,b aAlfredWegenerInstitute,HelmholtzCentreforPolarandMarineResearch,27570Bremerhaven,Germany bArcticUniversityofNorway,9019Tromsø,Norway a r t i c l e i n f o a b s t r a c t Articlehistory: Researchonprocessesleadingtoformation,maintenance,anddisappearanceofpolynyasinthePolarRegions Received27April2016 benefitssignificantlyfromtheuseofdifferenttypesofremotesensingdata.TheSentinelsoftheEuropean Receivedinrevisedform13September2016 SpaceAgency(ESA),togetherwithothersatellitemissions,provideavarietyofdatafromdifferentpartsofthe Accepted2October2016 electromagneticspectrum,atdifferentspatialscales,andwithdifferenttemporalresolutions.Inacasestudy Availableonline10October2016 wedemonstratetheadvantageofmergingdatafromdifferentspaceborneinstrumentsforanalysingicecondi- tionsandicedynamicsinandaroundthefrequentlyoccurringTerraNovaBayPolynya(TNBP)intheRossSea Keywords: intheAntarctic.Startingwithalistofpolynyaparametersthataretypicallyretrievedfromsatelliteimages,we Seaice Polynya assesstheusefulnessofdifferentsensortypes.Onregionalscales(several100km),passivemicrowaveradiom- Multi-sensorsatelliteobservations etersprovideaviewonthemutualinfluenceofthethreeRossSeapolynyasonseaicedriftanddeformationpat- TerraNovaBay terns.Opticalsensorswithmeter-scaleresolution,ontheotherhand,allowverylocalizedanalysesofdifferent Microwaves polynyazones.Thecombinationofdifferentrangesoftheelectromagneticspectrumisessentialforrecognition ThermalIR andclassificationoficetypesandstructures.Radarimagestogetherwithdatafromthermalinfraredsensors,op- Opticalimages eratedattenstohundredsofmetersresolution,improvetheseparationoftheoutletzoneofthepolynyafromthe Icetypeclassification adjacentpackice.Thedirectcomparisonofradarandpassivemicrowaveimagesrevealsthevisibilityofde- Icedriftretrieval formedicezoneinthelatter.AsequenceofradarimageswasemployedtoretrieveicedriftaroundtheTNB, whichallowsanalysingthetemporalchangesofthepolynyaareaandtheextensionandstructureoftheoutlet zoneaswellasicemovementsanddeformationthatareinfluencedbythekatabaticwinds. ©2016TheAuthors.PublishedbyElsevierInc.ThisisanopenaccessarticleundertheCCBYlicense(http:// creativecommons.org/licenses/by/4.0/). 1.Introduction productionishigh,andatmosphereCO2issequesteredintotheocean byphysical-chemicalprocesses(Willmottetal.,2007). Inthispaperwedealwithconcurrentmulti-sensorsatelliteobserva- Polynyasoccurinice-coveredoceanregionsintheArcticandAnt- tionsofafrequentlyoccurringcoastalpolynyaintheTerraNovaBay, arctic,mostlyininaccessibleplaces.Hence,remotesensingprovides whichislocatedintheRossSea/Antarctica.Themotivationistoassess anessentialtoolforgatheringdataaboutpolynyas.Onemajorquestion thegainthatcanbeachievedintheresearchofpolynyaevolutionanddy- instudiesdealingwithpolynyadynamicsiswhichparameterscanbe namicswhencombiningdataofESA'sdifferentSentinelsatellitemissions providedbymeansofremotesensing?Hereweaddresstheuseofremote (e.g.,https://sentinels.esa.int),whichcarryvarioussensorssuchasimag- sensingdataforprocessstudiesandparameterretrievals,consideringvar- ingradar,multi-spectralinstruments,andthermalradiometers.Coastal ioussatellitesensors,which(a)coverawiderangeoftheelectromagnetic polynyasarehighlydynamicareasofopenwaterandrecentlyformed spectrumfromvisibletomicrowavefrequencies,(b)areoperatedondiffer- icethatdevelopbetweenthecoastandtheoffshorepackice.Froma entspatialscales,and(c)deliverdataatdifferenttemporalintervals. geoscienceandbiochemicalpointofviewtheyareoflargeinterestbe- Becauseoftheirindependencefromcloudcoverageandfrequent cause(a)theyarelocationsofstrongheatandmoistureexchangebe- dataacquisitionsoveragivenarea,passivemicrowaveradiometers tweenatmosphereandocean;(b)coolingeffectsandtheformationof (PMR)arepreferredsatellitesensorsformonitoringpolynyas(e.g., frazilicecauselocaldensitychangesandmixingofthewatervolume Kernetal.,2007;Kern,2009).Methodshavebeendevelopedtoesti- below, which are processes that may affect ocean stratification on matethepolynyaarea(e.g.,MarkusandBurns,1995;Hunewinkelet localandevenregionalscales;(c)indaylightthebiologicalprimary al., 1998), and the thickness of thin ice that forms in the polynya (Martinetal.,2004;Martinetal.,2005).Thepolynyaextentisdirectly determinedfrommeasuredmicrowaveintensityratios,usingthresh- ⁎ Correspondingauthor. oldsforseparatingopenwaterandthinicefromthickeroffshorepack E-mailaddresses:[email protected](T.Hollands),[email protected] (W.Dierking). iceandlandoriceshelves(e.g.,Willmesetal.,2010).Theicethickness http://dx.doi.org/10.1016/j.rse.2016.10.003 0034-4257/©2016TheAuthors.PublishedbyElsevierInc.ThisisanopenaccessarticleundertheCCBYlicense(http://creativecommons.org/licenses/by/4.0/). T.Hollands,W.Dierking/RemoteSensingofEnvironment187(2016)30–48 31 hastobecalculatedindirectlyusingempiricalrelationshipsthattypical- satellites,significantlyreducethetemporalgapsbetweendataacquisi- lyareestablishedbycomparisonwithcomplementarydata,e.g.,from tions.IthasalsotobenotedthattheswathwidthsofSARsystemsare infrared sensors. Considering the small width of several polynyas limitedbetweenafewtensand500km. (here,“small”meansb1–10km),inparticularintheirearlyevolution Inthestudiesmentionedaboveseveralproblemswererecognized phase,thedrawbackofsatellitePMRsistheircoarsespatialresolution regardinglimitationsofthesensor'scapacitytoprovidecertaininfor- (presentlybetween5and40km,dependentoninstrument,frequency, mationandparameters.Forrealisticsimulationof,e.g.,polynyawidth andimagingmode).Thisleadstocontaminationeffectsatthepolynya and ice production rate, the extent of the open water and thin ice edges(e.g.,signalmixturesoflandandopenwaterorthinandpack areasaswellasthethinicethicknessdistributionhavetobedetermined ice).TheadvantageofPMRsisthatvariationsofpolynyaextentand withhighaccuracy.Thisisdifficulttoachieveatcoarsespatialresolu- thin icethicknesscan be continuously monitored in thelong term tionand/orinsufficientsignalcontrastbetweenopenwateranddiffer- withonlysmalltemporalgapsofroughly24h,butoftenless.Recentsat- enticetypes.Oneexampleisthemisinterpretation ofPMRsignals ellite-bornePMRsaretheAdvancedMicrowaveScanningRadiometer sinceiceshelves,icebergs,fasticeandthinicehavesimilarmicrowave (AMSR-2)ontheJapaneseGCOM-W1satelliteandtheSpecialSensor characteristics,andtheirlocationsatagiventestsitechangecontinu- MicrowaveImagerSounder(SSMIS)carriedonboardthespaceborne ouslyduetocalving,breakupanddrifting.TIR-dataareoftenconsider- platformsoftheDefenseMeteorologicalSatelliteProgram(DMSP).Al- ablyinfluencedbyatmosphericconditions.Thephysicalpropertiesof thoughoperatedatsimilarcoverageandspatialresolution,datafrom seaicefoundinoneregionmaydifferfromthoseofseaiceinotherre- scatterometerssuchasQuikSCAThaveonlybeenusedoccasionally gions;henceautomatedalgorithmsforparameterretrievalfromsatel- (e.g.,Willmesetal.,2010).Thecoarse-resolutionradarbackscattering litedata,developedforonelocaltestsite,cannotbeappliedglobally coefficientsobtainedfromQuikSCATaremoredifficulttointerpretin ineverycase.Inthesingle-frequencysingle-polarizationSARimages termsofpolynyaextentandicethicknesssincetheyarealsoverysensi- usedhitherto,thehighlyvariablesignalsbackscatteredfromthinice tivetoicesurfaceroughnessvariations.Hence,QuikSCATdataareof (bare,rafted,coveredwithfrostflowers)complicateitsautomatedsep- minorimportanceforthedevelopmentofrobustretrievalalgorithms. arationfromopenwaterandpackice. Despitetheirsensitivitytothepresenceofclouds,theusefulnessof Basedonthereferencescitedabove,theconclusionsregardingfu- thermalinfrared(TIR)imagesfortheretrievalofpolynyasizeandthin turesatelliteobservationstrategiesoverpolynyasare: ice thickness has also been demonstrated in several studies (e.g., Willmes et al., 2010;Krumpen et al.,2011; Ciappa et al.,2012).In – Concurrentdataacquisitionsusingcomplementarysensors(optical, mapsofsurfacetemperature,coldiceandwarmeropen watercan thermal,radar)improvethesegmentationandclassificationofdif- oftenbeeasilyseparated.Fortheestimationofthinicethickness,ather- ferentzonesinandoffshorefromapolynya. modynamicicegrowthmodelisemployedforwhichthesurfacetem- – Regionalandlocalinteractionmechanismsatsitesoffrequentpo- peraturehastobeprovidedasinputparameter(e.g.,YuandLindsay, lynyaoccurrencesrequiresatelliteimageswithdifferentspatialcov- 1996;Krumpenetal.,2011).Majordifferencesbetweenthetempera- erageandresolution(e.g.,PMRversusSAR). tureintheTIR-andthemicrowave-regimearisebecausetheformeris – Ahigherdataacquisitionfrequencyisneededforsensorsproviding influencedbyaverythinskinlayer(forsalinewater,e.g.,thethickness imageswithhighspatialresolutioninordertoresolvethedynamics variesbetween1μmand1mmforwavelengthsbetween2and16μm), oflocalprocesses. whereasthelatterisdeterminedbylayersof0.04–0.5cminthickness (frequenciesbetween1and20GHz).OneadvantageofTIR-datais thattheyareavailableathighspatialresolution,typicallyontheorder Inthisstudy,weanalyseexamplesdemonstratingthepotentialofthe of1kmandbetter,andthattheyareoperatedatlargeswathwidths, mostrecentandfuturesatellitemissionsforimprovedmonitoringofpo- whichdecreasesthetimebetweenacquisitionsoveragivenpolynya lynyasandfortheretrievalofparameterscharacterizingpolynyaevolution. (e.g., the swath width of the Moderate Resolution Imaging Otherexamplesfortheuseofdatafrommultiplesensorsforstudiesof Spectroradiometer MODIS is 2330 km compared to 1445 km for polynyadynamicsareprovided,e.g.,inCiappaandPietranera(2013), AMSR-E). Willmesetal.(2010),orDruckeretal.(2003).Theyfocusinparticular AscomplementarydatasourcetoTIR-imaging,syntheticaperture ontheretrievalofgeophysicalparameters.Inourstudy,wealsopayat- radar(SAR)sensorsprovideanevenbetterspatialresolutionbetween tention on the usefulness of image processing techniques that can be b10mand150m,dependentonradarfrequencyandimagingmode. regardedaspreparatorysteptoincreasetherobustnessandreliabilityofre- Sincetheyareoperatedatmicrowavefrequencies,theiradvantageis trievalalgorithms.InSection2wegiveabriefoverviewaboutpolynya thattheydeliverdataindependentofcloudcoverageanddaylight–in parametersthatcanberetrievedfromremotesensingdata.Weintro- contrasttooptical(“visiblerange”)sensorswhichotherwisecanbeop- ducethebackgroundabouthowthoseparametershavebeenusedto eratedatcomparablespatialresolutions.Atpresent,EarthObservation analysepolynyaprocesses,tosimulatetheiropeningandclosing,and (EO)satelliteSARsystemsoperateatfrequenciesbetween1.2GHz(L- toestimateiceproductionrates.Themotivationistoprovidethereader, band)and10GHz(X-band),andusedifferentpolarizationcombina- whoisnotfamiliarwiththerequirementsofpolynyaresearch,withthe tions(mostlyoflineartype:HH,HV,VH,VV,withH-horizontal,V-ver- informationnecessarytoassesstheprosandconsofdifferentsatellite tical,thefirstletterindicatingthetransmitted,thesecondthereceived sensorsandretrievalmethods.Section3dealswiththelocalandregion- polarization).SARimageshavebeenappliedforvalidatingalgorithms alenvironmentalconditionsinandaroundourtestsite,theTerraNova derivedforPMR-andTIR-sensors(e.g.,Willmesetal.,2010;Ciappa Bay,andSection4providesinformationaboutthesatellitedataweused andPietranera,2013;MorelliandParmiggiani,2013).Alsoopticalim- forouranalyses.Wedemonstratethegainofcombiningdifferentimage agesoraerialphotographyareusefulinthiscontext(e.g.,Willmeset typesforaqualitativeanalysisoficeconditionsinSection5.Multi-sen- al.,2010).Haarpainteretal.(2001),whodevelopedamodelforsimulat- soriceclassificationisdiscussedinSection6,andthedeterminationof ingtheevolutionoftheStorfjordenpolynya(Svalbard),tooktimeseries icedriftanddeformationpatternsisdealtwithinSection7.Atthe ofSARimagesacquiredbytheEuropeanRemoteSensing(ERS-2)satel- endweprovideadiscussionofadditionalaspectstobeconsidered, liteformanualclassificationofseaicetypes.Besidesvariationsofpolyn- followedbytheconclusions. yashapeandsize,alsoicedriftcanbedeterminedfromsequencesof opticalorSARimages(e.g.,Druckeretal.,2003).Themajorproblem 2.Polynyaparametersobtainedfromremotesensing withpastsatelliteSARmissionswasthatdataacquisitionsoverpolyna siteswereonlyirregularandwithlargetimegaps.Constellationmis- Inthissectionweprovideexamplesofpolynyaparametersthathave sions such as Sentinel-1 and Sentinel-2, consisting of two or more been directly retrievedor indirectly determined from observations 32 T.Hollands,W.Dierking/RemoteSensingofEnvironment187(2016)30–48 usingremotesensingtechniques.Thepurposeofretrievalsistoprovide between different ice growth stages or ice types. Williams et al. parametersfordescribingthelong-termdynamics,climatology,andas- (2007),theirFig.1)distinguishfourdifferentmeasuresofthepolynya sociatedchangesofapolynya(e.g.,Kern,2009),tocompareobserva- width,namely(1)theregionofopenwater,(2)theregionofopen tions with model simulations (e.g., Hollands et al., 2013) or to andpartiallyopenwater,(3)the“practicalwidth”whichencompasses determineparametersthatareneededforrunningthemodels(e.g., allicecoverwhichistoothinforsafetravel,andfinally(4)the“full Druckeretal.,2003). width”whichincludesallicethatisthinnerthantheoffshorepackice. AccordingtoWillmottetal.(2007)onedistinguishestwomajorap- The degree of details that can be achieved in separating different proaches for modelling the evolution of a polynya. These are flux zonesdependsnotonlyonspatialresolutionbutinthecaseofSAR, modelsandgeneralcirculationmodels.Theformerarebasedontheas- e.g.,alsoonthefrequencyandpolarization. sumptionthatawind-generatedcoastalpolynyaattainsamaximum size,whichisdeterminedbythebalancebetweentheiceproduction intheopenwaterzoneandthefluxoftheoffshorepackiceoutofthe 2.2.Polynyaandcoastlineshape polynya.Theadvantageoffluxmodelsisthatitiseasiertoidentifysin- gledominantprocessesthatinfluencepolynyadevelopment.However, Amorecomprehensivelookatpolynyaevolutionrequiresconsider- itisextremelydifficulttoincorporateallpossiblefeedbackmechanisms, ingtheshapeofthecoastlineinfrontofwhichthepolynyaislocated. whicharemoreorlesswellintegratedinoceancirculationmodelsin- Besides of the orientation of single coastline segments, the steady cludingdynamic-thermodynamicseaicecomponents(e.g.,Hollands stateshapeofacoastalpolynya(morepreciselyofitsoffshoreopen- etal.,2013).Iceconcentration,thickness,anddriftvelocityareamong wateredge)isinfluencedbymagnitudeuanddirectionαofthemove- theprognosticvariablesofsuchmodels.Thepolynyaisthendefined mentoffree-driftingfrazilice,andbymagnitudeUanddirectionθofthe asanareaforwhichtheiceconcentrationisbelowacertainthreshold consolidatedthiniceinthepolynya(e.g.,Darbyetal.,1995;Krumpenet (Willmottetal.,2007). al.,2011,theirFig.2).Darbyetal.defineanalong-shorelengthscaleL C onwhichthepolynyaadjuststoitssteady-statewidth(seetheirFig.2 2.1.Polynyaarea explainingthepracticaldeterminationofL ).Thisparameterdepends C bothonθandα.CoastlinevariationsthatareshorterthanL arenotmir- C Themostobviousvariablethatcanberetrievedfromsatelliteimag- roredintheshapeoftheopen-wateredge.ThemagnitudeofL canvary C eryistheextentofthepolynya.Besidesbeinganessentialparameterre- betweenzeroandtensofkilometres(Darbyetal.,1995).Hencesatellite latedtopolynyaformation,evolutionanddecay,thisinformationis imageswithspatialresolutionsontheorderof100mandlessandwith neededtovalidatesimulationsbothwithfluxandcirculationmodels, aclearaccentuationofthecoastlineandtheopen-wateredgeareneed- andfortuningcoupledpolynya-atmosphericmodelsornumericalsim- ed.Darbyetal.(1995)presentedsimulationsfortheTerraNovaBayPo- ulationsofthethermohalinecirculationinducedbypolynyas(Ciappaet lynyawiththeactualcoastlinegeometry,whichrevealedarealistic al.,2012).Withknowledgeofthepolynyaextentitisinprinciplepossi- shapeoftheopen-wateredge. bletoquantifyheatlosses,newiceproductionandsaltfluxes.When usingdatafromPMR,thesimplestmethodistodefineathresholdfor theiceconcentration,belowwhichthecorrespondingresolutioncell isregardedpartofapolynya(e.g.,MorelliandParmiggiani,2013). Thiscorrespondstothedelimitationofpolynyasinsimulationscarried outwithcirculationmodels(seeabove).Anotherpossibilityistoem- ploy the polynya signature simulation method (PSSM) to separate thinice,openwater,andthickice(MarkusandBurns,1995;Kernet al.,2007;Kern,2009;Willmesetal.,2011;Adamsetal.,2013).Thepo- lynyaextentisthenthesumofthethiniceandopenwaterareas.Kern etal.(2007)comparedresultsobtainedfromthePSSMappliedtoSSMI data(89GHzand37GHzchannelswithspatialresolutionsof15km× 13kmintheformerand37km×28kminthelattercase)withicecon- centrationestimatesfromAMSR-E89GHzdata(spatialresolution6 km×4km).TheyfoundthatAMSR-Eiceconcentrationsof25–40% correspondedtoareasclassifiedasopenwaterusingthePSSM,andcon- centrationsof65–80%werefoundforareasofPSSMclass“thinice”. (Notethatthetheoreticalconcentrationsare0and100%,respectively). Kernetal.(2007)explainedthiswiththedifferentfrequenciesandspa- tialresolutionsusedfortheestimationandclassification.Theresult demonstratestheneedofachievingmoredetailedspatialinformation abouttheiceconditionsinandaroundapolynya. Becauseoftheirhigh-resolutionmodes,SAR,opticalandTIRsensors enablemoresubtleanalysesoficeconditionsespeciallyforsmallerpo- lynyas(widthsb10km)andtheseparationofopenwater,fraziland greaseice,closedthinice,andtheoffshorepackice(Dokkenetal., 2002;Willmesetal.,2010;CiappaandPietranera,2013;Hollandset al.,2013).Examplesfromourstudywillbepresentedbelow.Inthe caseofSARimages,visualclassificationisachievedbyconsideringthe backscattered radarintensityandimagetexture,whicharerelated mainlytotheicesurfacestructureonscalesbetweenafewmillimeters tocentimeters(“small-scale”surfaceroughness)and10sofmeters (e.g.,iceridges,raftingzones).Inthecaseoflowsalinityice,theinflu- enceofvolumeinclusionsandstructurehasalsotobetakenintoac- Fig.1.TNBPandadjacentregions,showingmajorseaicezoneswithdominantdrift count. Temperature variations are related to thickness changes directionsandlocationsoftheoutletglaciersmentionedinthetext. T.Hollands,W.Dierking/RemoteSensingofEnvironment187(2016)30–48 33 Fig.2.SatelliteimagesshowingthelocalenvironmentaroundtheTNBP.Therespectivesensorwithacquisitiondateandtimingisgivenintheupperrightcornerofeachimage,the coverageisabout150km×150km. 2.3.Iceproductionrate inthemeteorologicaldata(Willmesetal.,2010).WhenusingPMR,the distinctionofdifferenticethicknessclassesishamperedduetothe Fromageophysicalpointofview,iceproductionratesinpolynyas coarsespatialresolutionofthosesensors(Willmesetal.,2010,their arethemostinterestingparametersincetheyhavetobetakenintoac- Fig.7).Thevalidationofthicknessretrievalsoverthinpolynyaiceisex- countforestimatingvariationsbetweeniceproductioninagivenarea tremelydifficult,evenwithairborneinstrumentssuchaselectromag- andicetransportoutofthisarea.Theestimationoficeformationre- netic probes since their measurement uncertainty is too large quirestodeterminethesizeofthepolynya(or–better–ofthedifferent (Willmesetal.,2010).Insummary,theestimationoficethicknessand waterandicezonesinthepolynya,seeabove)andtoretrievetheice iceproductionratesfromremotesensingdataandtheirvalidationre- thickness.Thesurfaceheatlossandicevolumeproductioncanbecalcu- mainverychallenging. latedifcorrespondingmeteorologicaldataareavailable(e.g.,Willmeset al.,2010,Krumpenetal.,2011).Underlyingassumptionsarethatthe 2.4.Iceformation,drift,anddeformation sumofradiativeandturbulentfluxesattheicesurfaceisbalancedby theconductiveheatfluxintheice,thatallheatlossatthesurfaceis Asmentionedabove,driftvelocityanddirectionoffrazilandconsol- usedforiceformation,andthattheoceanicheatfluxissmall.Thelatter idatedpolynyaicearerequiredinfluxmodels(e.g.,Darbyetal.,1995; isoftenvalidinpolynyasbecausethewholewatercolumnisclosetothe Krumpenetal.,2011).Theicedriftisaprognosticvariableofcirculation freezingpointduringwinterinthecontinentalshelfregion(Tamuraet models(e.g.,Hollandsetal.,2013).Whilethedriftoffraziliceinthe al.,2008).TheuseofTIRandPMRforicethicknessretrievalandtheir openwaterareaofpolynyascannotberetrievedwithrecentsatellite prosandconswerementionedintheintroductionabove.UsingTIR, systems,thisisoftenpossibleforconsolidatedpolynyaandoffshore goodresultscanbeobtainedforicethicknessesbelow0.5m,ifthever- packice.Differentretrievalalgorithmsareavailablethatcanbeapplied ticaltemperatureprofileintheiceislinearandtheiceissnow-free onsequencesofsatelliteimages(SAR,TIR,VIS).Oneofthemostpopular (Druckeretal.,2003).Alargesourceoferrorsarisesfromuncertainties approaches is a multi-scale multi-resolution pattern matching 34 T.Hollands,W.Dierking/RemoteSensingofEnvironment187(2016)30–48 algorithmrequiringanimagepairasinput,whichwasusedbyHollands complexmixtureofdifferentbranchesfromLarsen,Reeves,Priestly etal.(2013)andresultsinacceptableaccuraciesof3to5pixelsforthe andDavidGlacier.TheDrygalskiIceTongue,whichisanextensionof retrieveddisplacementvectorsduringfreezingconditions(Hollands theDavidGlacier,blocks northwarddriftingseaicethatoriginates and Dierking, 2011). However, the approach requires recognizable fromMcMurdoSoundandthecoastofRossIsland.Hence,thelength radarsignaturevariationsondifferentspatialscalesthatcanbefound and orientation of the ice tongue influences shape and size of the inboththefirstandsecondimage(seenextparagraph).Therelatively TNBP.Ruscianoetal.(2013)separatetwodifferentstagesofpolyniaac- largetemporalgapbetweenbothimages(rarelylessthanoneday tivities:increasinganddecreasingiceproductioninMarch–Juneand withmostrecentSARsystemswhenoperatedinhigh-resolutionimag- December–February,respectively,andaperiodofmaximumefficiency ingmode)isoftenaproblemconsideringthefastchangesoficecondi- inseaiceproductionfromJulytoNovember.Theopenwaterareaof tionsoffshorefromevolvingpolynyas.Thetemporalacquisitionrateis theTNBPduringwintertimeestimatedfromTIR-data rangesfrom higherwithrecentlylaunchedorplannedsatelliteconstellations(such 1000to1300km2(butpeaksof8500km2wereobserved).Thecom- asCosmoSkymed,Sentinel-1,RadarsatConstellation). binedopenwaterandthiniceareaobtainedfromPMRislargerbyafac- IntheiranalysisofthePeasemodel,e.g.,Druckeretal.(2003)em- tor of about 2 (Ciappa et al., 2012). Ciappa and Pietranera (2013) ploySARdatatodeterminetheadvectionofthepackiceawayfrom characterizethedifferentphasesofpolynyaevolution.Whentheiceis thepolynyaregion.AlsousingSARimagery,Hollandsetal.(2013)re- pushedawayfromthecoast(openingphase),frazilicethatisorganized trievedthedriftvelocitiesofconsolidatedthiniceandofpackicein asicestreaksparalleltothewinddirectionoccursintheopenwater theRonnePolynyaarealocatedintheWeddellSea.Theynotethatthe area. Using high-resolution SAR imagery, Ciappa and Pietranera accuracydependscriticallyonthepresenceofrecognizableicestruc- (2013)measureddistancesbetweenthestreaksrangingfrom300to tures,whichmaybeburiedunderwetsnowduringmeltingconditions, 800m.Theyalsorecognizedwavesoflengthsbetween30and70m, ormaychangeveryquicklybecauseofstrongmovementsanddeforma- whichmodulatedthestreaksandthefrazilandgreaseiceaccumulated tionsofnewlyformediceinthepolynya.Inbothcases,theretrievedice attheoffshorepolynyamargin.Whenthewindspeeddecreases,the velocitiesarelessreliablethanfortheoffshorepackiceunderfreezing gapsbetweenthestreaksbecomesmaller.Progressivefreezingfrom conditions. theoffshoreedgeofthepolynyatothecoastcharacterizestheclosing Gallée(1997)usedacoupledatmosphere–polynyamodeltostudy phase.ThissuggeststhatthedetectionofthepolynyaareawithIR-sen- theair-seainteractionsovertheTerraNovaBayPolynyainwinter.He sorsorPMRismorereliableduringthegrowingphase,whenlargeareas pointedoutthatthereisaneedforhavingabetterknowledgeoffrazil ofunfrozenwaterarepresent,thanduringtheclosingphasewhenlarge iceevolution(consolidationoffrazilintopancakes,frazilherding).In areasofthinsea-iceprevail. hismodelheusedasimplecavitatingfluidbehaviourfortheseaicerhe- Gallée(1997)statesthatthepolynyasizeispoorlycorrelatedwith ologyinthepolynya,whichmeansthatthe“modelice”revealsdiver- thelarge-scalewindforcing,suggestingthatitsmainforcingisthekat- gence or shear but does resist convergence. Hence the effects of abaticwind.Regionaliceconditionsareinfluencedbywindsblowing ridgingandraftingarenotincludedinthesimulations. acrosstheRossIceShelf(VanWoert,1999).Gallée(1997)foundthat theopenwarmwaterareasignificantlyinfluencestheatmosphericcir- 3.TerraNovaBay culationinthecoastalzone.MorelliandParmiggiani(2013),whocom- binedsatelliteobservationsandmodelsimulations,obtainedasimilar 3.1.Largescalesituation result,whichrevealsthattheheatingoftheairassociatedwiththepo- lynyaincreasesthespeedofthekatabaticwindoverthepolynyaafterit IntheAntarcticOcean,theRossSeaPolynya(RSP)isthelargestpo- hascrossedthecoastline.This,inturn,isinlinewiththestudybyVan lynyawithawinterareaofarond20,000km2(BarberandMassom, Woert(1999)whoshowedthatchangesofsensibleandlong-wave 2007,theirTable23).Twosmallerpolynasarelocatedinthewestern heatfluxescanexplainuptoabout50%oftheobservedvariancein partoftheRossSea,oneinTerraNovaBay(theTerraNovaBayPolynya, thepolynyaopenwaterfraction.Differentstagesofcloudinesscanex- inthefollowingdenotedasTNBP,withameanareaof1300km2and plainanother8–10%ofthevariance.Ciappaetal.(2012)foundmost maxima up to 5000 km2 (Van Woert, 1999)) and the other in the cloudyperiodsduringphasesoflimitedpolynyaactivitiesatlowwind McMurdo Sound (MSP, with an area about 2/3 of TNBP, see Kern conditions.Fluctuationsoftheopenwaterareacanoccurwithinvery (2009),Table2).Tamuraetal.(2008)foundthatthehighesticeproduc- shorttimeintervalsintheorderofhours,andpeaksofarealgrowing tionoftheAntarcticOceanoccursintheRossSea.Fromthe1990stothe rate may exceed 300 km2 during extreme wind gusts (Ciappa and 2000s,theiceproduction(involume)decreasedbyabout30%duetoa Pietranera,2013). decreaseinpolynyaareasandduetoatmosphericwarming.Tamuraet TIRimagescanbeusedtoanalysethevariabilityofstrongsurface al.(2008)supposethatthenegativetrendiniceproductionisonerea- winds(Bromwich,1989)oversnow-coveredregions.Theimagesmay sonfortherecentfresheningoftheAntarcticBottomWater.Druckeret revealdelimitedwarmerareasindicatingthepresenceofstrongkata- al.(2011)reportthattheiceproductionintheRossSeaapproximately baticairstreams.Thereasonwhythoseairstreamsmanifestthemselves equalstheiceexport.Theformerwasestimatedfrom36GHzAMSR-E inthethermalsignatureistheintenseverticalmixingoftheairand data,thelatterwascalculatedforfluxgatesparalleltothe1000-m driftingsnow.InlandfromthecoastofTerraNovaBayCiappaetal. isobaths,usingmapsofdailyicemotionretrievedfrom89GHzAMSR- (2012)recognizedthesignaturesofkatabaticwindsdescendingfrom Edata.From2003to2008,theaverageannualiceproductionofthe theglaciersflowingtowardsthebay.Theyfoundthatwhenthepolynya RSPamountedto510–730km3,oftheTNBP70–111km3,andofthe isopen,airflowfromtheReevesGlacierisalwaysstrong.Theopeningis MSP11–80km3(numbersaregivenperfreezingseason).Thetotalice largewhensurfacewindsoftheotherthreeglacierscontributetothe production in these three polynyas accounts for 20% to 50% of the katabaticwindfloworiginatingfromtheReevesGlacier.Becauseof totalseaicevolumeintheRossSea. thiscomplexairflowstructure,itisdifficulttoexplaintheTNBPevolu- tion using a one-dimensional flux model, instead, a 2-dimensional 3.2.LocalconditionsinTerraNovaBay modelisneeded. ThecoastalpolynyainTerraNovaBayisorientedineast-westdirec- 4.Data tion.ItisboundedbytheDrygalskiIceTongueinthesouthandbythe CampbellIceTongueinthenorth(seeFig.1).Thepolynyaisgenerated ThedatathatweusedforouranalysiswereacquiredovertheTNBP andmaintainedbypersistentkatabaticwindswithspeedsofuptoN inSeptember2009(Table1).Thedatasetconsistsofhigh-andcoarse- 40m/s,whichpushthebayiceoffshore.Thekatabaticwindfieldisa resolution imagery. The former includes SAR data (ALOS PALSAR, T.Hollands,W.Dierking/RemoteSensingofEnvironment187(2016)30–48 35 EnvisatASAR),opticaldatafromEO1-ALI(EarthObserving1Mission – separationoficetypesandstructuresinandadjacenttothepolynya AdvancedLandImager,inthefollowing‘optical’isusedsynonymously on localscale employingnear-simultaneously acquiredimagery with‘visiblerange’),MODISopticalandnear-infrared(NIR),AVHRR withhighspatialresolutionfromSAR,TIR,andopticalsensors, (TIR),andAATSRthermal(TIR)images.Coarse-resolutionproducts – interpretationofregionalicecovercharacteristicsafew100km arefromapassivemicrowaveradiometer(PMR),namelyAMSR-E.De- aroundthepolynya,basedonPMR-data,andlinkinglocaltoregional tailsaboutsensorsandimagesareprovidedinTable2.Inparticular conditions, forSeptember10,2009,favouriteconditionsforinter-comparisonsof – monitoringofthepolynyaevolutionandofvariationsofpolynyapa- differentsensorsaregivensincePALSAR,ASAR,AATSR,MODIS,and rameterssuchasitssize,shape,orrelativearealfractionsofdifferent EO1-ALIdatawereallacquiredwithinashorttemporalwindow. zones(i.e.openwater,fraziliceherding,accumulatedand/orconsol- TheASARandPALSARdataweregeocodedandcalibratedusingthe idatedice). commercial SARscape software. The high-resolution data were re- projectedtoanAntarcticPolarStereographicProjectionwithaCentral Longitudeof180°EandaStandardLatitudeof71°S.Ifcomparisonsof Whencombiningdatafromdifferentrangesoftheelectromagnetic fixedtargets(e.g.,coastlinesandislands)revealedslightdiscrepancies, spectrumonehastoconsiderthesensitivitiesoftherespectivemea- therespectiveimageswereco-registeredtothecorrespondingEnvisat suredquantity(brightnesstemperatureT ,backscatteringcoefficient B ASARscene.Allimageswereresampledtoaresolutionof100m× σ0,temperatureT,reflectivityR)tothephysicalpropertiesoftheim- 100m. agedarea.Sensitivitiestosensorconfigurations,suchase.g.,frequency, BothAVHRRandAATSRseasurfacetemperature(SST)temperatures polarization,ordirectionofmeasurement,havetobetakenintoaccount are based on the combination of the 11 μm- and 12 μm-band. The aswell. AVHRRSSTwasevaluatedfollowingKeyetal.,1997.Forthecalculation of the AATSR SST the ESA VISAT Software was used. In contrast to 5.1.Localanalysis AVHRR,the AATSRsensoracquiresoneimagein forwarddirection (lookingslantedthroughtheatmosphere)andasecondonenadir- High-resolutionSARandAVHRRimagesare,e.g.,usedinthestudy looking,inordertocompensateforatmosphericeffectsinthecalcula- byMassometal.(2001)foranalysingtheiceconditionsaroundthe tionsoftheSST(ESA,2002;Corlettetal.,2006).Thecomparisonof MertzGlacierPolynya.InFig.2,multi-sensordataacquiredoverthe theAVHRRandAATSRdatashowedvariabledifferencesonapoint-to- TerraNovaBayareshown:A–EnvisatASAR,B–ALOSPALSAR,C– pointscale(whichispartlycausedbyicemovementsinthetimeinter- MODIS Band1,andD–EnvisatAATSR,covering an area of 22,500 valof2hbetweendataacquisitions).Onaveragethedatacompared km2, with spatial resolutions between 100 m (PALSAR) and 1 km well,allowingsolidconclusionsintheframeofouranalysis. (AATSR).Thetemporaldifferencebetweenthefirstandlastdatatake Thehigh-resolutionimagesweresupplementedbyAMSR-E89GHz is1:39h.Majorstructuresoftheicecovercanbeidentifiedintheim- datafortheperiodSeptember5–12,bothatoriginalandatspatiallyen- ages,e.g.,thecoastline,thelandice,theicetongue,theopenwater hancedresolution(thedatasetthatweusedisdescribedinLongand areaofthepolynya,andsinglesmallerandlargericefloes(compare Stroeve,2011).Incontrasttothespatialresolutionof5.4×5.4km/ alsotoFig.1).Theiceformedinthepolynyaisfirstexportedtowards pixelintheoriginal89GHzdata,thespatiallyenhancedversionhasa eastbutthenisdeflectednortheast.Wetermthisbandoficeexport resolutionof2.2×2.2km.Inenhancementalgorithms,theantennapat- “outletzone”.Inthevisiblerange(MODISBand1),majorpartsofthe ternisde-convolvedforreconstructionoftheunderlyingbrightness outletzonearecoveredbyclouds.Icefloesappearlightgrey,andthin- temperaturedistributiononahigher-resolutiongrid.Resolution-en- ner ice areas dark grey. In particular the belt starting south of the hancementtechniques,however,provideimprovedresolutionatthe Drygalskiicetongueandthengraduallyturningtowardsnortheastis expenseofanincreasednoiselevel(LongandDaum,1998).Neverthe- characterizedbyheavilybrokenice,consistingofsmallerandlarger less,theenhancedimagesallowedtheidentificationofmoredetailsin thickerfloes(lightgrey)andthinice.Wetermthiszone“deformation theareaaroundthepolynya.Weusethebrightnesstemperaturedata belt”.Itiscarriedawaytotheeastduetoshearingforcesimposedby asobtained,withoutconsideringanyatmosphericeffectsthatinfluence theiceintheoutletzoneoftheTNBPandpushedtothenorthbythe highfrequencybandsliketheemployed89GHzchannel. iceoriginatingfromtheRSPandtheMSP(compareFig.1).Smaller clouds are visible over different parts of the scene. The dark area abovetheDrygalskitongueistheopenwaterzoneoftheTNBP,with 5.Qualitativeanalysesofpolynyaconditions streaksoffraziliceweaklyshowinguptowardstheseaiceedgetoits right.Thethermalimage(Fig.2D)revealsacoarserspatialresolution Withourdatasetitispossibletocarryoutqualitativeanalysesin- butresemblestheMODISsceneinmajorparts.Thecloudcovercan cludingthefollowingelements: lessclearlybedistinguished.Mostpresumablythedistributionofclouds changedinthe90minbetweentheacquisitionsofMODISandAATSR data.Theopenwaterzoneappearsverybright.Itistheareawiththe Table1 SatelliteimagesusedforthecasestudypresentedinSections5–7.Sensorabbreviations highesttemperatureintheregion.Thestreaksoffrazilicearehardlyvis- areexplainedinthetext. ible,whichmaybepartlyattributabletothecoarserspatialresolutionof thethermaldata,butmostprobablyindicatesthatthetemperaturedif- Date Time(UTC) Sensor ferencebetweentheformingicecrystalsandthewateratthesurfaceis 06.09.2009 19:41:47 EnvisatASARWSScene verylow.TheiceintheoutletzoneoftheTNBPiswarmerthaninthe 07.09.2009 19:10:28 EnvisatASARWSScene adjacenticezones,whichmeansthatitisrelativelythin.Alsothethin 07.09.2009 21:19:36 EO1ALI iceareasinthedeformationbeltappearbright.Comparedtothereflec- 09.09.2009 19:47:34 EnvisatASARWSScene 10.09.2009 12:39:57 EnvisatAATSRScene tance(Fig.2C)orradarbrightness(Fig.2,AandB),thetemperaturevar- 10.09.2009 19:16:19 EnvisatASARWSScene iationsseemtobemostsuitablefordelimitingtheoutletzone.Inthe 10.09.2009 19.39:20 ALOSPALSARScanSAR SARimages,alsotheiceunderthecloud-coveredareasisvisible.Thein- 10.09.2009 20:35:00 MODIS formationcontentoftheC-andL-bandimageissimilar,buttheice 10.09.2009 20:55:37 EnvisatAATSRScene 11.09.2009 20:19:36 ALOSPalSARScanSAR coverstructureismorepronouncedatL-band.Thisismoreaconse- 11.09.2009 20:24:20 EnvisatAATSRScene quenceofthesignaturecontrastsofdifferenticetypesandstructures Daily(5.–12.9.2009) AMSR-E89GHz–Horizontal thanofthedifferentspatialresolutions(PALSAR100m,ASAR150m). Daily(5.–12.9.2009) AVHRR Theappearanceoftheopenwaterzonedependsontheradarfrequency 36 T.Hollands,W.Dierking/RemoteSensingofEnvironment187(2016)30–48 Table2 SpecificationsforthesensorslistedinTable1. Sensor Wavelength Frequency Swath Spatialresolution Parameter width EnvisatASARWS 5.62cm 5.33GHz 400km 150×150m Backscatteringcoefficient (C-Band) [dB] ALOSPALSARScanSAR 23.61cm 1.27GHz 250–350 100×100m Backscatteringcoefficient (L-Band) km [dB] AMSR-E89GHz, 89GHz(W-Band) 1445km 5.4×5.4km(2.2×2.2kmafter Brightnesstemperature[K] H-polarization enhancement) ModisBand1 620–670nm 2330km 250×250m Reflectance ModisBand2 841–876nm 2330km 250×250m Reflectance AATSR 3.7,10.8, 512km 1×1km SST[K] 12μm AVHRR 0.58–12.5μm 2900km 1.1×1.1km Radiance/SST[K] EO1ALI 0.433–2350μm(7bands) 37km 30×30m(10×10mpanchromatic) Reflectance 0.48–0.69μm panchromatic andonwindspeedanddirection.AtL-band(Fig.2B),arimisclearly andtheirdistancetooneanotherincreasewiththedistancefromthe visibleattheseaiceedgeeast(totheright)oftheopenwaterareaof coast.MorelliandParmiggiani(2013)pointoutthatthewindspeedis theTNBP,whichismoredifficulttoidentifyintheotherimages(A,C, largerwhenmovingawayfromthecoastbecauseoftheheatreleased D).Weinterpretthisrimasanarrowzoneofaccumulatingfrazilice. bytheopenwaterzoneofthepolynya.Consideringtheresultsofexper- Inbothradarscenes,thestreaksoffraziliceintheopenwaterzoneof imentalstudiesonLangmuircirculationjustmentionedabove,thein- theTNBParerecognizable.Thinsmoothice(darkerintheradarimag- creasingdistancebetweenstreaksandtheirbroadeningconformsto ery)andthickericefloes(brighter)canbebetterdistinguishedatL- thewindspeedincreasewithdistancefromthecoast. band. The outlet zone reveals alternating bright, partly banded ice InFig.4,theiceandwatersurfacetemperaturesintheTerraNova zones,anddarkzonesconsistingofbrokenlevelice.Fromthisstructure BayforSeptember10,2009,derivedfromAVHRRdata,arepictured. andtheavailablesequenceofSARimagesweconcludethatrelatively WehereselectedAVHRRandnotAATSRdatasinceforthatdatethefor- smoothlevelicezonesdevelopinthepolynyaareaatphasesofcalm merprovideabetterspatialcoverageofthesouthwestpartoftheRoss wind and break due to external forces exerted by the adjacent ice Sea,whichwetakeadvantageofintheanalysispresentedinSection5.2. masses, while being pushed to the northeast when the katabatic Thelocationsofautomatedweatherstationsareindicatedbyredtrian- windsincreaseinstrength.Thebrightzones,withlargepartsoffine- gles.ForthedayoftheAVHRRdataacquisition,thelocalairtempera- grainedimagetexture,arecharacteristicforroughicesurfaces.Based tures measured at 2 m height above surface at the stations were: on pairs of SAR images (revealing similar grey-tone patterns) and LaurieII:−44.6°C,CapeBird:−27.5°C,Manuela:−26.5°C,Eneide: high-resolution optical images acquired on other days, we assume −23°C.Adetailedcomparisonbetweenthestationdataandthetem- thatthesebrightareasconsistofaccumulatedfrazilandgreaseiceat peraturesretrievedfromAVHRRthermalimageryisbeyondthescope theopenwatermarginthatisconsolidatedatlargerdistancesfrom ofourstudyhere,butwenotethataone-to-onecorrespondencecan theopenwaterzone,eventuallyformingaheavilyraftedicecoverand hardlybeexpectedforanumberofreasons.Nevertheless,thestation brokenicefloes.Thebandingisthenduetodifferentstagesofcompac- dataprovideanadditionalpossibilitytojudgethetemperaturecondi- tionand/orsurfaceroughness.Thedarkiceareaintheupperleftcorner tionsintheregionshowninFig.4.Thezonesofhighertemperatures ofbothSARimagesisfastice(confirmedbyouranalysisofasequenceof (yellow)delimittherangeofinfluencefromtheRSP(lowerright),the SARimagesdiscussedbelow). MSP(inasmallpartnorthwestfromstationCapeBird),andtheTNBP. Anotherexample is shownin Fig. 3,in which ahigh-resolution TheopenwaterzoneoftheTNBPisindicatedbythebrownishcolour. (10m)EO1ALIimageissuperimposedonanASARscene.Theformer Verywellreflectedisalsothedeformationbeltwithsmallerandlarger wasacquired2:09hlaterthanthelatter.Theopticaldatarevealtheor- colder(thicker)icefloesandwarmer(thinner)icebetweenthem.The ganizationoffrazilandgreaseiceasstreaks.Windgeneratedwaves darkblueareaintheupperleftcorneroftheimageisafasticezone. travelthroughthestreaks(zoom-in,Fig.3),whichwasalsorecognized Thesouthernpartofthiszoneisbreakingoff.Thisprocessmanifestsit- by Ciappa and Pietranera (2013) in high-resolution (5 m) COSMO selfinthecrackofhighertemperature. SkyMedSARimages.Inourwide-swathASARscenewithaspatialreso- Thejointanalysisofoptical(near-infrared),thermal,andradarim- lutionof150m,theicestreaksarealsovisible,althoughmoredifficultto ageryacquiredwithinshorttimeintervals(about2handless)demon- identify.Thelocation,width,andshapeofindividualstreakscanchange strates the great advantage that multi-sensor data offer for the withinminutesdependentonlocalwindconditions.Hence,theydonot qualitativeretrievaloficeconditionsandanalysisofcertainprocesses. matchperfectlybetweentheALIandtheASARimageconsideringthe Weregardthecombinationofthermalandradardatauseful,sincethe timeintervalbetweentheiracquisitions. formerpermitarelativelyclearseparationofthinnerandthickerice AccordingtoCiappaandPietranera(2013)thegapsbetweenthe (withthedisadvantageofbeinghamperedbycloudcover),whereas streaksdependonwindspeed.Lackingsatellitephotostakenfordiffer- thelatteremphasizethesurfaceroughnessandnarrowdeformation entwindspeeds,wecannotexaminethisdependency.Theicestreaks structures.Inourcase,thelowerfrequencyL-bandimageofferssome areunveiledbyLangmuircirculation.Thorpe(2004)reportsthatthe advantages with respect to the identification of ice structures and separationscalesandlengthsofLangmuircellsincreasewithwind types.However,werefrainfromrecommendingL-bandimageryfor speed.Inexperimentalstudiesithasbeenobservedthatthereisaten- allcases.Anygeneral(global)preferenceofeitherlow-(L-band)or dencyforlargerspacingbetweenthestreakswithhigherwindspeeds higherfrequencyradar(C-orX-band)forthediscriminationofthin (Plueddemannetal.,1996,theirFig.5),buttherearealsocleardevia- icetypeshasyetnotbeenformulated(Dierking,2010;Dierkingand tionsfromthisrelationship.OnereasonisthatLangmuircirculation Busche,2006).If,e.g.,multi-yeariceoccursinicefieldsaroundapolyn- doesnotonlydependonthewindstressbutalsoontheStokesdrift(av- ya(morefrequentlyintheArctic,butmulti-yearicepersistsalsointhe eragevelocityof afluidparcel)ofsurfacewaves. Inourscene,the easternRossSea),C-bandmaybeabetterchoicefordiscriminationand streaksinthenorthernpartarebroader.Thewidthofindividualstreaks classificationoficetypes(e.g.,Dierking,2013).Opticalimagesareuseful T.Hollands,W.Dierking/RemoteSensingofEnvironment187(2016)30–48 37 Fig.3.EO1ALIopticalimageoverlaidonanEnvisatASARWSscene.BothwereacquiredonSep.7,2009,theformerat19:10:28UTC,thelatterat21:19:36.Thescenecoversanareaof about60kmineast-westand100kminnorth-southdirection.Thezoom-inatthebottom(right)demonstratesthatoceanwavescanwellberecognizedintheicestreaks. toseparatesnow-coveredandsnow-freeice,andtheyimprovethereli- RossSea(Section3andFig.1).Fig.5showsanimagesequenceacquired abilityoftheretrievaloficeconditionsfromSARimages.Theexample withtheAMSR-Efrom5thto12thofSeptember2009,representingthe showninFig.3alsodemonstratesthatprocessesshapingtheicecondi- brightnesstemperatureT thatwasmeasuredat89GHz,H-polariza- B tionsinandaroundpolynyas(inthiscasethewindshapingtheice tion,andmappedwithaspatialresolutionof2.2km.Onecanidentify streaks)canonlybefullyconceivedifadditionalmeteorologicaland theRossIceShelfatthebottomoftheindividualimages(whichisin oceanographicdataareavailable. thesouth),andVictoriaLandwiththeDrygalskiicetongueontheleft (west).TheRSPcanbespottedattherightbottom,theMSPleftofthe 5.2.Regionalanalysis bottomcenter,andtheTNBPabove(northof)theDrygalskiicetongue. InthemiddleoftheimageabowshapedfeatureoflowerT -values B Ontheregionalscale,meteorologicalandiceconditionscausea showsup,whichseparatesthesouth-easticecoverinfluencedby(and complexinteractionofthethreepolynyas(RSP,MBP,TNBP)inthe partly originating from) the RSP and the north-west sea ice cover 38 T.Hollands,W.Dierking/RemoteSensingofEnvironment187(2016)30–48 surface,andinthecaseofseaiceonsalinity,brinevolumefraction, brinepocketshapeandspatialdistribution(ShokrandSinha,2015). At90HGHz,εofopenwateris0.528,fornewice0.573,andforfirst- year ice 0.886 (Shokr and Sinha, 2015, Table 8.10). Their Fig. 8.32 showsthatεincreasessignificantlyduringthefirst10mmgrowthof ice,butthenreachingvaluestypicalforfirst-yearice.Hence,weexpect largervariationsofεonlyinareasofopenwater,greaseandthinice,e.g., intheoutletzoneandinleads.Fig.7showsagraphrelatingT andT , ice B whichrevealsslightlygrowingT -valueswithincreasingT forthe B ice zonesoffastice,smoothicefloes,androughfloes(blueinFig.4).For thezonesofaccumulatedice(yellowandorangeinFig.4),aclearrela- tionshipbetweenT andT wasnotfound:theintervalofT ismainly ice B ice clusteredbetween−20°Cand−27°C,thecorrespondingrangeofT B from218Kto239K,whichresultsinarangeofεbetween0.86and 0.97, typical for first-year ice of varying roughness (see Shokr and Sinha,2015,Fig.8.37demonstratingtheeffectofsurfaceroughness). ThehighestvaluesofT (−4to−5°C)arefromthefraziliceand ice openwaterzoneofthepolynya,inwhichT variesbetween200and B 220K.Here,wedidnotcorrectfortheemittinglayerthickness.There- sultindicatesthepresenceofraftedandbrokenice.Forthefastice,level ice,andthinicefloes(bluishcolorsinFig.4)wealsoobtainemissivities offirst-yearicethickerthanafewcentimeters. ThewarmeroutletzonesoftheRSPandtheTNBPinFig.5(covering theopenwaterzoneandthepolynyaicepushedoffshore)arecharac- terizedbyspatialbrightnesstemperaturevariationsbetween215K Fig.4.IcesurfacetemperaturederivedfromAVHRR4and5(11and12μm).Theimage and235KonSep.10(thedateisselectedwithreferencetoFig.4). wasacquiredonSep. 10,2009. Thespatial resolution is 1.1km. Average surface ThecolderdeformationbeltrevealsvariationsofT between195K B temperaturesfordifferenticetypesare:1fastice(darkblue) −40°C±2.9°C2level and215K(blue-yellow)withinterspersedwarmer(thinner)ice(or- icefloes(blue) −35°C±1.8°C3thinicefloes(lightlue) −31°C±1.8°C4 ange)overtheperiodfromSep.5to12.OnSep.12,however,itde- accumulationzone −24°C±1.8°Candrefrozenleads(lightyellow)Thepartially openwaterzoneofthepolyna(brown)revealedameantemperatureof−5°C. creasedinwidth.Thedeformationbeltasazoneofshearbetweenthe southeastRSPiceandthenorthwestTNBPicecanbeidentifiedover thewholeimagesequenceshowninFig.5.Itscontrastinbrightness movingawayfromtheTNBP(seealsoFig.1).Fromacomparisonofthe temperaturemayberelatedtothemagnitudeoftheforcesexertedon AMSR-EdatawiththecorrespondingSARimagesavailableforthisperi- it,whichcausesstrongericedeformation.Unfortunately,wedonot od(seeexampleshowninFig.6),wefoundthatthisfeaturecorre- havethenecessarydata(iceandwindconditions,oceancurrents)to spondstothedeformationbeltthatisalsovisibleinFigs.2and4. provethishypothesis. Themicrowavebrightnesstemperatureistheproductoftheemis- TheT -patternobservedintheAMSR-E89GHzdatacanalsoberec- B sivityεandthephysicaltemperatureToftheimagedarea.Theemitted ognizedinthe36GHzchannel,althoughthesignaturecontrastisless. radiationisfromtheuppericelayer.Itsthicknessisdeterminedbythe Thismaybecausedbythelowerspatialresolutionofthe36GHzchan- penetrationdepth,whichvariesbetweenafewmillimeterstodecime- nel.Thedeformationbeltcanbeobservedregularlyovertheyears. ters,dependentonicephysicalpropertiesandmicrowavefrequency. Forafrequencyof89GHz,Mathewetal.(2008)givevaluesof28cm 6.Separationoficezones fordrysnow,4cmformulti-yearice,and0.9cmforfirst-yearice.The averagephysicaltemperatureoftheemittinglayerdiffersfromtheair Inthissectionwedealwiththesegmentationandclassificationof orsurfacetemperature.Itdependsonthemicrophysicalpropertiesof distincticezonesintheTerraNovaBay.Inandaroundapolynyadiffer- thesnowandiceandvarieswithmicrowavefrequency(duetothedif- entstagesofnewiceformationarefound,e.g.,accumulationsoffrazil ferentpenetrationdepths).Mathewetal.(2008)determinedthe“emit- andgreaseiceduetocompressionalforcesexertedbythekatabatic tinglayertemperature”asalinearfunctionoftheairtemperaturefor wind,andconsolidatedthinlevelicethatmayrevealdeformationstruc- theArctic.Forfirst-yearice,theirresultsrevealthattheformerranges tures(rafting,ridging).Inourstudy,theaimofsegmentationandclas- from-19°Cto−8°Cat89GHzforairtemperaturesbetween−40°C sification (the latter is linking segments and actual ice conditions) and−10°C.Weassumethattheseresultsareapproximatelyvalid mustbetoreconstructdifferentstepsintheevolutionofapolynya. alsoforiceconditionsintheTerraNovaBay.Iftheicesurfacetempera- Theindividualicezonesmayhavecompletelydifferentpropertiesre- tureattwodifferentfirst-yearicelocationsdiffersby30°C,thecorre- gardingsaltrelease,heatexchange,anddeformation.Foranexample spondingchangeofT wouldhenceonlybeabout10K,ifεremains ofclassificationwecombinethenear-simultaneousmulti-sensorsatel- B constant.ThevariationsoftheicesurfacetemperatureT onSeptem- liteacquisitionsovertheTNBPshowninFig.2,whichhavecomparable ice ber10,showninFigs.2and4fortheTerraNovaBay,rangefromap- spatialresolutions(here,wedonotconsiderthecoarse-resolutionPMR proximately−45°Cto−15°C(−2to−4°Cinthepolynyazoneof data). partiallyopenwater,seeFig.7below).Thebrightnesstemperatures Inafirststep,welookedatthesegmentationpotentialofdifferent T inthecorrespondingareaarebetweenapproximately195Kand setsofdiscriminationrules(supervisedhierarchicalapproach)andof B 235K.Thepolynyazoneisnotrecognizableinthemicrowaveimage, unsupervisedclusteringalgorithms(suchasISODATA).Inbothcases, whichweattributetothecoarseresolutionandthemixed-pixeleffect. spatiallyvaryingmeanvaluesandvariancesofthedirectlymeasured Consideringthattherangeoftheemittinglayertemperatureissmaller quantities(reflectance,temperature,backscatteringcoefficient)arede- thantheoneoftheicesurface,thevariationsofT mustalsobeinflu- termined.Afterdividingtheimagesintodifferentsegments,theyhave B encedbytheemissivity.Theemissivityεoficedependsontheicetem- tobelinkedwithactualiceclassesinasecondstep.Forsupervisedclas- perature,surfaceroughness,porosity,snowwetness,snowgrainsize, sificationsanoperatoridentifiesareasofinterestintrainingdatathat icelayersinthesnow,presenceofslushorsuperimposediceonthe mirrorthestatisticalcharacteristicsofsingleisolatedicezones.Several T.Hollands,W.Dierking/RemoteSensingofEnvironment187(2016)30–48 39 Fig.5.SequenceofimagesfromAMSR-E89GHz,H-polarization,fortheperiodfromSep.5to12,showingthebrightnesstemperaturesTBatanenhancedresolutionof2.2km.Bluecolors indicatelowervaluesofTB,redcolorshighervalues. approachesareavailable,spanningtherangefromvariousdistance allowsagooddifferentiationofthedifferenticetypesandfloes,L- measuresthatareusedtoassociateeachpixelwithacertainsegment, bandrevealsastrongerbackscatterforhighlydeformedregions(e.g. tohighlycomplexsystemssuchasneuralnetworksorsupportedvector deformedpolynyaiceordeformationssuchasraftingandridges).SAR machines.Themanualidentificationoftrainingclassesistimeconsum- imagesacquiredunderfreezingconditionsshowmanymoredetails ingandambiguous(duetothesubjectivecomponentofhumaninter- abouticestructure(floes,cracks,brashiceetc)thanopticalandthermal ference). An alternative is the use of unsupervised techniques. A sensors,becausetheradarsignalspenetratethroughdrysnow.The popularapproachinthisgroupistheISODATAalgorithm.Suchap- thermalsensoreasestheseparationofopenwater,thinandthickice proachescanhandlemultidimensionaldataandclassifymulti-sensor withoutbeinghamperedbytoomanysmallstructuraldetails. data.Theadvantageisthattheyoftendetectsegmentsthatmightbedif- TheidealclassificationintheareaofTNBPshouldcomprisethefol- ficulttorecognizebyvisualanalysis.Whilevisualseparationofseg- lowingelements. mentsisstillpossiblewhencombiningdataoftwoorthreedifferent channels,sensormodes,orsensors,itbecomesimpossibleinfouror 1. Thepackicezoneconsistsofdifferentlysizedthickericefloeswith moredimensions.Thedisadvantageofunsupervisedmethodsisthat interspersedthinice.Thickericefloesreveallowersurfacetempera- theirresultsareoftendifficulttolinktoaclassificationschemethatis turesandmoderatetohighbackscatteringintensityduetodifferent optimallyadaptedtotheactual(ice)conditions.ForourISODATAclas- stages of deformation and fracturing. Thin ice manifests itself sificationwefoundthatasupervisedpost-adjustmentwasnecessaryto throughhighertemperaturesandlowerbackscatteringcoefficients. adapttheautomatedsegmentationtotheactuallypresenticecondi- 2. Theoutleticerevealsasequencewithalternatingaccumulated,part- tions,whichdevaluatestheadvantageofunsupervisedtechniques.In lybandediceandbrokenicefloesthatpartlyrevealsignsofridging thissection,wefocusthereforeexclusivelyonthesupervisedsegmenta- andrafting.Thesurfacetemperatureishighanddoesnotshowany tion.Landandcloud-coveredareasareexcludedfromtheclassification differencesbetweenaccumulatediceandicefloezones.Itgradually procedure. decreaseswithdistancefromthepolynya.Thebackscatterintensity SincetheMODISNIRreflectanceimage(Fig.2)ismoredifficulttoin- ishighoveraccumulatedice(indicatingaroughsurface)andlower terpretthanthethermalAATSRdataandmoreseverelyaffectedby overtheicefloezones. clouds, we decided to include only the latter into a segmentation 3. Theopenwaterpolynyazonecanbeeasilyrecognizedbecauseofits scheme,togetherwiththeC-andL-bandradarimages.WhileC-band veryhightemperature.Thebackscatteredintensitydependsonthe