24
1 AMSR2 SEA ICE ALGORITHM THEORETICAL BASIS DOCUMENT Walter N. Meier Thorsten Markus Josefino Comiso Alvaro Ivanoff Jeffrey Miller NASA Goddard Space Flight Center Cryospheric Sciences Lab Code 615 Greenbelt, MD 20771 12 July 2017

AMSR2 SEA ICE ALGORITHM THEORETICAL BASIS DOCUMENT

  • Upload
    others

  • View
    8

  • Download
    0

Embed Size (px)

Citation preview

Page 1: AMSR2 SEA ICE ALGORITHM THEORETICAL BASIS DOCUMENT

1

AMSR2

SEAICEALGORITHMTHEORETICALBASISDOCUMENT

WalterN.Meier

ThorstenMarkus

JosefinoComiso

AlvaroIvanoff

JeffreyMiller

NASAGoddardSpaceFlightCenter

CryosphericSciencesLab

Code615

Greenbelt,MD20771

12July2017

Page 2: AMSR2 SEA ICE ALGORITHM THEORETICAL BASIS DOCUMENT

2

TableofContents

1. Overview 3

2. Background 3

2.1Experimentalobjective 3

2.2Historicalprospective 3

3. TheoreticalBasisofAlgorithm 4

3.1Physicsoftheproblem 4

3.2Mathematicaldescription 5

3.2.1Seaiceconcentration 5

3.2.1.1Implementationoftheconcentrationalgorithm 6

3.2.1.2Landspillovercorrection 11

3.2.1.3Reductionofatmosphericeffects 11

3.2.2Snowdepthonseaice 14

3.2.3Seaicedrift 15

3.2.4Adjustmentstobrightnesstemperatures 17

4. VarianceandUncertaintyEstimates 19

5. ImplementationIssues 19

5.1Computation/programming/proceduralconsideration

5.2Qualitycontrolanddiagnostics

5.3Constraints,limitationsanddiagnostics

6. References 22

Page 3: AMSR2 SEA ICE ALGORITHM THEORETICAL BASIS DOCUMENT

3

1.Overview

TheAMSR2seaicestandardlevel3productsincludeseaiceconcentration,snowdepthonseaice,andseaicedrift.ItissimilartotheAMSR-Eproducts,withadjustmentsforthechangeinsensor.MuchofthecontenthereisadaptedfromtheAMSR-EATBD.TheAMSR2standardseaiceconcentrationproductisgeneratedusingtheenhancedNASATeam(NT2)algorithmdescribedbyMarkusandCavalieri(2000,2009),thesnowdepthisproducedfromthealgorithmdescribedbyMarkusandCavalieri(1998)forbothhemispheres,butexcludingtheArcticperennialiceregions,andtheseaicedriftisproducedfromanalgorithmdescribedbyLiuandCavalieri(1998).Additionally,thedifferencebetweentheAMSR2Bootstrap(ABA)andtheNT2retrievedconcentrations(ABA-NT2)arearchived.TheseproductstogetherwithAMSR2calibratedbrightnesstemperatures(TBs)aremappedtothesamepolarstereographicprojectionusedforSSMIdatatoprovidetheresearchcommunityconsistencyandcontinuitywiththeexisting32-yearNimbus7SMMRandDMSPSSMIseaiceconcentrationproducts.TheTBgridresolutionsareasfollows:(a)TBsforallAMSR2channels:25-km,(b)TBsforthe18,23,36,and89GHzchannels:12.5-km,(c)TBsforthe89GHzchannels:6.25-km.AlloftheseTBproductsarestoredasacompositeof(i)daily-averagedascendingorbitsonly,(ii)daily-averageddescendingorbitsonly,and(iii)allorbitscreatingafulldailyaverage.Seaiceconcentrationsareproducedat12.5-kmand25-kmresolutionsandstoredasacompositeofdaily-averagedascendingorbits,daily-averageddescendingorbits,andallorbitsforafulldailyaverage,similartotheTBproducts.Snowdepthonseaiceisproducedasa5-dayaverageataresolutionof12.5km.Seaicedriftisalsoafive-dayproductcomputedataresolutionof6.25-km,butmappedataresolutionof100-km.

ThealgorithmsdescribedbelowareessentiallythesameasthosedevelopedforAMSR-E.TheprimarychangeforAMSR2isanadjustmenttotheAMSR2brightnesstemperaturestobeconsistentwiththeAMSR-Evalues.ThisadjustmentissummarizedinSection3.2.4

2.Background

2.1Experimentalobjective

TheobjectiveoftheAMSR2seaiceproductsistoprovidehigh-qualityestimatesofimportantgeophysicalseaiceparameters,including:concentration,snowdepthonseaice,anddrift.ThesefieldswillcontinuethetimeseriesthatbeganwithAMSR-Ein2002.WiththeadditionofcalibratedAMSR2parameters,theAMSR(includingbothAMSR-EandAMSR2)willextentatleast15years,providingahigherqualitycomplementtothelong-termnearly40-yearrecordofpassivemicrowaveseaiceestimatesfromSMMR,SSMI,andSSMIS.

2.2Historicalprospective

Seaiceisakeyclimateindicatorduetoitshighalbedothatreflectsincomingsolarradiation,itsroleasaphysicalbarriertoheatandmoisturetransferbetweentheoceanandatmosphere,and

Page 4: AMSR2 SEA ICE ALGORITHM THEORETICAL BASIS DOCUMENT

4

itsroleinbiogeochemicalcycles.ItisalsoplaysamajorroleinpolarecosystemsandinlivesofindigenouspeoplesoftheArctic.ChangesinArcticseaiceoverthepastnearlyfourdecadeshavebeendramatic,withalossofoverathirdofsummericecover,alargedeclineinicethickness,andanearly-completelossofoldice.Satelliteremotesensinghasbeentheprimarytoolformappingseaiceconcentration,extent,andagesincethelate1970s.Theconsistentandcontinuouslong-termmultichannelmicrowaveradiometerseaicerecordbeganinOctober1978withtheScanningMultichannelMicrowaveRadiometer(SMMR)ontheNASANimbus-5satellite,followedbyaseriesofSpecialSensorMicrowaveImagers(SSMI)andSpecialSensorMicrowaveImager/Sounders(SSMIS)instrumentsonU.S.DefenseMeteorologicalSatelliteProgram(DMSP)platforms.TheSSMIandSSMISrecordbeganinJuly1987andcontinuestoday(throughmid-2017).Thesesensorsprovidethecontinuousandconsistentlong-termrecordofseaicethatisnownearly40yearslong.Startingin2002,theAdvancedMicrowaveScanningRadiometerfortheEarthObservingSystem(AMSR-E),aJAXAsensorlaunchedontheNASAAquasatellite,providesimprovedspatialresolutionandotheradvancesovertheSMMR-SSMI-SSMISrecord.TheAdvancedMicrowaveSoundingRadiometer2(AMSR2)onboardthefirstGlobalChangeObservationMission–Water(GCOM-W1)satelliteincludessimilarcharacteristicsasAMSR-Ewithslightlybetterspatialresolutionandotherimprovements.TheAMSR-EandAMSR2sensorprovidea15+yearrecordofenhancedseaiceretrievalsthatcomplementthelongerrecord.

3.TheoreticalBasisofAlgorithm

3.1Physicsoftheproblem

PassivemicrowaveradiationisnaturallyemittedbytheEarth’ssurfaceandoverlyingatmosphere.Thisemissionisacomplexfunctionofthemicrowaveradiativepropertiesoftheemittingbody(HallikainenandWinebrenner,1992).However,forthepurposesofmicrowaveremotesensing,therelationshipcanbedescribedasasimplefunctionofthephysicaltemperature(T)oftheemittingbodyandtheemissivity(ε)ofthebody.

TB=ε*T(1)

TBisthebrightnesstemperatureandistheparameter(aftercalibrations)retrievedbysatellitesensorsandistheinputparametertopassivemicrowaveseaiceconcentrationalgorithms.

Themicrowaveelectromagneticpropertiesofseaiceareafunctionofthephysicalpropertiesoftheice,suchascrystalstructure,salinity,temperature,orsnowcover.Inaddition,openwatertypicallyhasanelectromagneticemissionsignaturethatisdistinctfromseaiceemission(Eppleretal.,1992).Thesepropertiesformthebasisforpassivemicrowaveretrievalofseaiceconcentrations.

Specifically,theunfrozenwatersurfaceishighlyreflectiveinmuchofthemicrowaveregime,resultinginlowemission(Figure5).Inaddition,emissionfromliquidwaterishighlypolarized.

Page 5: AMSR2 SEA ICE ALGORITHM THEORETICAL BASIS DOCUMENT

5

Whensaltwaterinitiallyfreezesintofirst-year(FY)ice(icethathasformedsincetheendofthepreviousmeltseason),themicrowaveemissionchangessubstantially;thesurfaceemissionincreasesandisonlyweaklypolarized.Overtimeasfreezingcontinues,brinepocketswithintheseaicedrain,particularlyiftheseaicesurvivesasummermeltseasonwhenmuchofthebrineisflushedbymeltwater.Thismulti-year(MY)icehasamorecomplexsignaturewithcharacteristicsgenerallybetweenwaterandfirst-yearice.Othersurfacefeaturescanmodifythemicrowaveemission,particularlysnowcover,whichcanscattertheicesurfaceemissionand/oremitradiationfromwithinthesnowpack.Atmosphericemissionalsocontributestoanysignalreceivedbyasatellitesensor.Theseissuesresultinuncertaintiesintheretrievedconcentrations,whicharediscussedfurtherbelow.

Becauseofthecomplexitiesoftheseaicesurfaceaswellassurfaceandatmosphericemissionandscattering,directphysicalrelationshipsbetweenthemicrowaveemissionandthephysicalseaiceconcentrationarenotfeasible.Thus,thestandardapproachistoderiveconcentrationthroughempiricalrelationships.Theseempirically-derivedalgorithmstakeadvantageofthefactthatbrightnesstemperatureinmicrowavefrequenciestendtoclusteraroundconsistentvaluesforpuresurfacetypes(100%wateror100%seaice).Concentrationcanthenbederivedusingasimplelinearmixingequation(Zwallyetal.,1983)foranybrightnesstemperaturethatfallsbetweenthetwopuresurfacevalues:

TB=TICI+TO(1-CI)(2)

WhereTBistheobservedbrightnesstemperature,TIisthebrightnesstemperaturefor100%seaice,TOisthebrightnesstemperatureforopenwater,andCIistheseaiceconcentration.

Inreality,suchanapproachislimitedbythesurfaceambiguitiesandatmosphericemission.Usingcombinationsofmorethanonefrequencyandpolarizationlimitstheseeffects,resultinginbetterdiscriminationbetweenwateranddifferenticetypesandamoreaccurateconcentrationestimate.

3.2Mathematicaldescription

3.2.1Seaiceconcentration

ThetworatiosofbrightnesstemperaturesusedintheoriginalNASATeamalgorithm(Cavalierietal.1984;GloersenandCavalieri1986;Cavalierietal.1995)aswellasintheNT2approacharethepolarization

PR(ν)=[TB(νV)−TB(ν)]/[TB(νV)+TB(νH)] (3)

andthespectralgradientratio

GR(ν1pν2p)=TB(ν1p)−TB(ν2p)]/[TB(ν1p)+TB(ν2p)] (4)

Page 6: AMSR2 SEA ICE ALGORITHM THEORETICAL BASIS DOCUMENT

6

whereTBisthebrightnesstemperatureatfrequencyνforthepolarizedcomponentp(vertical(V)orhorizontal(H)).

Figure1(top)showsatypicalscatterplotofPR(19)versusGR(37V19V)forSeptemberconditionsintheWeddellSea.TheNTalgorithmidentifiestwoicetypesthatareassociatedwithfirst-yearandmultiyeariceintheArcticandicetypesAandBintheAntarctic(asshowninFigure1(top)).TheA-Blinerepresents100%iceconcentration.Thedistancefromtheopenwaterpoint(OW)tolineA-Bisameasureoftheiceconcentration.Inthisalgorithm,theprimarysourceoferrorisattributedtoconditionsinthesurfacelayersuchassurfaceglazeandlayering(Comisoetal.1997),whichcansignificantlyaffectthehorizontallypolarized19GHzbrightnesstemperature(Matzleretal.1984)leadingtoincreasedPR(19)valuesandthusanunderestimateoficeconcentration.Inthefollowingdiscussion,wewillrefertotheseeffectsassurfaceeffects.InFigure1(top),pixelswithsignificantsurfaceeffectstendtoclusterasacloudofpoints(labeledC)awayfromthe100%iceconcentrationlineA-BresultinginanunderestimateoficeconcentrationbytheNTalgorithm.Theuseofhorizontallypolarizedchannelsmakesitimperativetoresolveathirdicetypetoovercomethedifficultyofsurfaceeffectsontheemissivityofthehorizontallypolarizedcomponentofthebrightnesstemperature.

WemakeuseofGR(89V19V)andGR(89H19H)toresolvetheambiguitybetweenpixelswithtruelowiceconcentrationandpixelswithsignificantsurfaceeffects.AplotofthesetworatiosarefoundtoformnarrowclustersexceptforareaswheresurfaceeffectsdecreaseTB(19H)andconsequentlyincreaseGR(89H19H)(Figure1(bottom)).ValuesofhighGR(89V19V)andhighGR(89H19H)areindicativeofopenwater.TherangeofGR(89H19H)valuesislargerbecauseofthegreaterdynamicrangebetweeniceandwaterforthehorizontallypolarizedcomponents.Withincreasingiceconcentration,thetworatioshavemoresimilarvalues.ThenarrowclusterofpixelsadjacenttothediagonalshowninFigure1(bottom)represents100%iceconcentrationwithdifferentGRvaluescorrespondingtodifferenticetypes.Whensurfaceeffectscomeintoplay,pointsdeviatefromthisnarrowclustertowardsincreasedGR(89H19H)values(cloudofpointstotherightofthediagonal)whileGR(89V19V)changeslittleorremainsconstant.ThiscloudofpointslabeledCinFigure3(bottom)alsocorrespondstotheclusterofpointslabeledCinFigure3(top).Thedifference,therefore,betweenthesetwoGRs(ΔGR)isusedasameasureofthemagnitudeofsurfaceeffects.BasedonthisanalysisweintroduceanewicetypeC,whichrepresentsicehavingsignificantsurfaceeffects.ForcomputationalreasonswerotatetheaxesinPR-GRspace(Figure1(top))byanangleφsotheA-Blineisvertical.ThismakestherotatedPRs(PRR(19)andPRR(89))independentoficetypesAandB(first-yearandmultiyearfortheArctic).Theuseofthe89GHzdatarequiresacorrectionforatmosphericeffects.ThisisaccomplishedthroughanadditionalAMSR2variable,PR(89).

Theresponseofthebrightnesstemperaturestodifferentweatherconditionsiscalculatedusinganatmosphericradiativetransfermodel(Kummerow1993).Inputdataintothemodelaretheemissivitiesoffirst-yearseaiceunderwinterconditionstakenfromEppleretal.(1992)with

Page 7: AMSR2 SEA ICE ALGORITHM THEORETICAL BASIS DOCUMENT

7

modificationstoachieveagreementbetweenmodeledandobservedratios.Atmosphericprofilesusedasanotherindependentvariableinthealgorithm,havingdifferentcloudproperties,specificallycloudbase,cloudtop,cloudliquidwateraretakenfromFraseretal.(1975)andaverageatmospherictemperaturesandhumidityprofilesforsummerandwinterconditionsaretakenfromAntarcticresearchstations.Theseatmosphericprofilesarebasedonclimatologyandareassumedvalidforbothhemispheres.

Figure1:Top:GR(37V19V)versusPR(19)fortheWeddellSeaonSeptember15,1992.ThegraycirclesrepresentthetiepointsfortheicetypesAandBaswellasforopenwaterasusedbytheNTalgorithm.LabelCindicatespixelswithsignificantsurfaceeffects.Φistheanglebetweenthey-axisandtheA-Bline.Bottom:GR(85V19V)versusGR(85H19H).TheicetypesAandBareclosetothediagonal.TheamountoflayeringcorrespondstothehorizontaldeviationfromthislinetowardslabelC.TakenfromMarkusandCavalieri[2000].

TheplotsofΔGRversusPRR(19)(Figure2a)andΔGRversusPRR(89)(Figure2b)illustratethealgorithmdomain.Thegraysymbolsindicatethetie-pointswiththedifferentatmospheresforthethreesurfacetypes(A,C,andOW).Theyalsoillustratethattheeffectofweatheriswellmodeled.Forexample,theclusterofopenwatervaluesismainlytheresultofchangingatmosphericconditions.ThemodeledatmospheresadequatelyspanthelengthsoftheOWclusters.AcomparisonofFiguresB4aandB4balsoshowshowmuchmorethe89GHzdataareaffectedbytheatmospherecomparedtothe19GHzdata.

Page 8: AMSR2 SEA ICE ALGORITHM THEORETICAL BASIS DOCUMENT

8

Figure2(a)ΔGRversusPRR(19)and(b)ΔGRversusPRR(89)forSeptember15,2008.ThegraysymbolsrepresenttheNT2tie

We,then,calculatebrightnesstemperaturesforallpossibleiceconcentrationcombinationsin1%incrementsandforeachofthosesolutionscalculatetheratiosPRR(19),PRR(89),andΔGR.Thiscreatesaprisminwhicheachelementcontainsavectorwiththethreeratios(Figure3).ForeachAMSR2pixelPRR(19),PRR(89),andΔGRarecalculatedfromtheobservedbrightnesstemperatures.Next,wemovethroughthisprismcomparingtheobservedthreeratioswiththemodeledones.Theindiceswherethedifferencesaresmallestwilldeterminethefinaliceconcentrationcombinationandweatherindex.Thenextsectionwillprovidedetailedinformationabouttheimplementation.

Becauseoftheuniquesignatureofnewiceinthemicrowaverange,wesolvefornewiceinsteadoficetypeCforselectedpixels.UsingaGR(37V19V)thresholdof–0.02weeitherresolveicetypeC(forpixelswhereGR(37V19V)isbelowthisthreshold)orthinice(forpixelswhereGR(37V19V)isabovethisthreshold).AreasoficetypeCandthinicearemutuallyexclusivebecausethinicehaslittle,ifany,snowcover.Alimitation,ofcourse,isthatwecannotresolvemixturesofthiniceandthickericewithlayeringinitssnowcover.

3.2.1.1Implementationoftheconcentrationalgorithm

Incontrasttootheroperationalseaiceconcentrationalgorithmsusingdailyaveragedbrightnesstemperaturesasinput,theAMSR-ENT2concentrationsarecalculatedfromindividualswath(Level2)datafromwhichdailymapsaremadebyaveragingtheseswathiceconcentrations.UsingswathbrightnesstemperaturesisparticularlycriticalfortheNT2algorithmanditsatmosphericcorrection.Theatmosphericinfluenceonthebrightness

Page 9: AMSR2 SEA ICE ALGORITHM THEORETICAL BASIS DOCUMENT

9

temperaturesisnonlinearandbyusingaveragebrightnesstemperatureswewoulddilutetheatmosphericsignal.Theiceconcentrationalgorithmisimplementedasfollows:

1.Generatelook-uptables:ForeachAMSRchannelwithfrequencyνandpolarizationpcalculatebrightnesstemperatureforeachiceconcentration-weathercombination(usingTBow,TBA/FY,TBC/thinasgivenintheAppendixofMarkusandCavalieri(2009)):

TBca,cc,wx(νp)=(1−CA−CC)•TBow(νpWx)+CA•TBA/FY(νpWx)+CC•TBC/thin(νpWx) (5)

whereCAreferstotheicetypeA/Bconcentration(FY/MYforArctic),CCtoicetypeCconcentration,andWxtotheweatherindex.Iceconcentrationsarebetween0and100in1%increments,weatherindicesarebetween1and12correspondingtothetablesintheAppendixofMarkusandCavalieri(2009).

Figure3.FlowdiagramoftheNT2algorithm(fromMarkusandDokken2002).

Page 10: AMSR2 SEA ICE ALGORITHM THEORETICAL BASIS DOCUMENT

10

2.FromtheseTBscalculateratioscreatingthelook---uptables,e.g.LUTPR19(CA,CC,Wx)etc.:

LUTPR19(CA,CC,Wx)=[TBca,cc,wx(37V)−TBca,cc,wx(19V)]xsinφ19/[TBca,cc,wx(37V)+TBca,cc,wx(19V)]+[TBca,cc,wx(19V)−TBca,cc,wx(19H)]xcosφ19/[TBca,cc,wx(19V)+TBca,cc,wx(19H)] (6)

LUTPR89(ca,cc,wx)=[TBca,cc,wx(37V)−TBca,cc,wx(19V)]xsinφ89TBca,cc,wx(37V)+TBca,cc,wx(19V)+[TBca,cc,wx(89V)−TBca,cc,wx(89H)]xcosφ89/[TBca,cc,wx(89V)+TBca,cc,wx(89H)] (7)

IfGR(37V19V)<-0.02wesolveforicetypeCusingΔGRasourthirdvariable,i.e.,

LUTdGR(ca,cc,wx)=[TBca,cc,wx(89H)−TBca,cc,wx(19H)]/[TBca,cc,wx(89H)+TBca,cc,wx(19H)] −[TBca,cc,wx(89V)−TBca,cc,wx(19V)]/[TBca,cc,wx(89V)+TBca,cc,wx(19V)] (8)

WhereasforpixelswhereGR(37V19V)>-0.02wesolveforthiniceusingthestandardGR(37V19V)assuggestedbyCavalieri(1994),i.e.,

LUTdGR(ca,cc,wx)=[TBca,cc,wx(37V)−TBca,cc,wx(19V)]/[TBca,cc,wx(37V) +TBca,cc,wx(19V)] (9)

Eachofthesearrayshasthedimensionsof101x101x12where,ofcourse,thetotaliceconcentration(ca+cc)cannotexceed100.

3.ForeachpixeliwehavetheactualmeasuredAMSR-Ebrightnesstemperatures(TBi(νp))

4.Calculatesameratiosfromthesebrightnesstemperaturesasinstep2(PRi(19),etc.).

5.Comparetheseobservedratioswitheachoftheratiosinthelook-uptablesloopingthroughalliceconcentration-weathercombinations,i.e.,

δ=(PRi(19)−LUTPR19(ca,cc,wx))2+(PRi(89)−LUTPR89(ca,cc,wx))2 +(ΔGRi−LUTdGR(ca,cc,wx))2 (10)

6.Theindicesca,cc,wxwhereδisminimaldeterminetheiceconcentration(andweatherindex),i.e.:

CT=CAminδ+CCminδ (11)

Page 11: AMSR2 SEA ICE ALGORITHM THEORETICAL BASIS DOCUMENT

11

3.2.1.2LandSpilloverCorrection

Althoughalandmaskisappliedtotheiceconcentrationmaps,landspilloverstillleadstoerroneousiceconcentrationsalongthecoastlinesadjacenttoopenwater.Thismakesoperationalusageofthesemapscumbersome.Therefore,weapplyalandspillovercorrectionschemeonthemaps.Thedifficultyistodeleteallclearlyerroneousiceconcentrationwhileatthesametimepreservingactualiceconcentrations,asforexample,alongthemarginsofcoastalpolynyas.Weapplyafive-stepprocedure.

1.Classifyallpixelsofthepolar-stereographicgridwithrespecttothedistancetocoast.Oceanpixelsdirectlyalongthecoastareclassifiedby1,whereaspixelsfartherawayare2and3.Openoceanpixelsarezero.Landpixelsdirectlyalongthecoastareclassifiedas4andpixelsfartherawayhaveincreasingvalues.

2.Allpixelswithclasses1or2willbeassessedforerroneousseaiceconcentrationsduetolandspilloverbyanalyzingthe7by7pixelneighborhood.Theareaoftheneighborhood(7pixelsor87.5km)needstobegreaterthantheAMSR2antennapattern.Pixelswithvaluesof3and0willnotbechanged.

3.Checkwhetherallclass3pixelsin7-pixelneighborhoodareopenwater(ifso,seticeconcentrationto0).

4.Calculateanaverageseaiceconcentrationforthe7by7pixelboxassumingalloceanpixelshavezeroiceconcentrationandalllandpixelshaveaniceconcentrationof90%.Thisapproximatesatheoreticalconcentrationcausedbylandspilloveronly.

5.IftheAMSR2iceconcentrationislessthanorequaltothisvalue,setpixelatcenterofboxtoopenwater.

Figure4showsanexampleiceconcentrationwithandwithoutthelandspillovercorrection.

3.2.1.3ReductionofAtmosphericEffects

TheNT2algorithmhasanatmosphericcorrectionschemeasaninherentpartofthealgorithm.Itprovidesweather-correctedseaiceconcentrationsthroughtheutilizationofaforwardatmosphericradiativetransfer(RT)model.However,toeliminateremainingsevereweathereffectsoveropenocean,twoweatherfiltersbasedonthespectralgradientratioareimplementedusingthresholdvaluessimilartothoseusedbytheNTalgorithm(GloersenandCavalieri1986;Cavalierietal.1995).However,theadvantageoftheRTatmosphericcorrectionisthatnotonlyarespuriousiceconcentrationsovertheopenoceanremoved,butatmosphericcorrectionsareappliedtoicecoveredportionsoftheocean.

Page 12: AMSR2 SEA ICE ALGORITHM THEORETICAL BASIS DOCUMENT

12

Figure4.Mapoficeconcentrationwithandwithoutlandspillovercorrection.

Figure5showsAMSR-EseaiceconcentrationmapsfortheSeaofOkhotsk.Figure5ashowstheiceconcentrationmapifPRR(19),PRR(89),andΔGRareusedwithoutanyweathercorrection.Figure5bshowstheiceconcentrationmapwiththeNT2weathercorrection.ThedifferencesbetweenFigure5aand5bareshowninFigure5dandillustratetheeffectoftheweathercorrectionnotonlyovertheopenocean,butalsoovertheseaice.Moresevereweathereffectsovertheopenocean(forexample,inthebottomrightcorner)arefinallyremovedbytheNTweatherfilters(Figure5c).ThethresholdfortheGR(37V19V)NTweatherfilter(GloersenandCavalieri1986)is0.05,wherethethresholdfortheGR(22V19V)NTweatherfilter(Cavalierietal.1995)is0.045.IftherespectiveGRvaluesexceedthesethresholds,theseaiceconcentrationsaresettozero.Figure5eshowsthedifferenceiniceconcentrationsbetweentheretrievalsusingonlytheNT2weathercorrectionandtheretrievalsusingboththeNT2correctionandtheNTfilters.Aslightchangealongtheiceedgeisobserved.

EvenwithboththeatmosphericcorrectionschemeandtheGRfilters,westillhadproblemswithresidualweathercontaminationparticularlyatlowlatitudes.Afilterbasedonmonthlyclimatologicalseasurfacetemperatures(SSTs)fromtheNationalOceanicandAtmosphericAdministration(NOAA)oceanatlas,usedearlierbyCavalierietal.(1999),wasemployedtoeliminatetheselow-latitudespuriousiceconcentrations.IntheNorthernHemisphere,anypixelwherethemonthlySSTisgreaterthan278K,theiceconcentrationissettozerothroughoutthemonth;whereasintheSouthernHemisphere,whereverthemonthlySSTisgreaterthan275K,theiceconcentrationissettozerothroughoutthemonth.ThehigherSSTthreshold

Page 13: AMSR2 SEA ICE ALGORITHM THEORETICAL BASIS DOCUMENT

13

valueintheNorthernHemisphereisneededbecausethe275KisothermusedintheSouthernHemisphereistooclosetotheiceedgeinthenorth.Theclosestdistancethethresholdisothermsaretotheiceedgeismorethan400km(Cavalierietal.1999).

Insummary,theorderofprocessingisasfollows:

1.Calculateseaiceconcentrationswithatmosphericcorrection.2.ApplyGRfilters.3.ApplySSTmask.4.Applylandspillovercorrection.

Figure5.AMSR-EseaiceconcentrationsforMarch1,2007.(a)IceconcentrationscalculatedusingPRR(19),PRR(89),andΔGRwithoutapplyinganatmosphericcorrection;(b)iceconcentrationwithatmosphericcorrection;(c)finaliceconcentrationwithadditionalclean-upovertheopenoceanbyapplyingthestandardNASATeamGRweatherfilters;(d)differencebetween(a)and(b);(e)differencebetween(b)and(c).Differencesgreaterthan10%havebeentruncatedfortheerroneousseaiceconcentrationsinthelowerrightcorner.

Page 14: AMSR2 SEA ICE ALGORITHM THEORETICAL BASIS DOCUMENT

14

3.2.2Snowdepthonseaice

TheAMSR-Esnow-depth-on-sea-icealgorithmwasdevelopedusingDMSPSSMIdata(MarkusandCavalieri1998)toestimatesnowdepthonseaicefromspace.Thesnowdepthonseaiceiscalculatedusingthespectralgradientratioofthe18.7GHzand37GHzverticalpolarizationchannels,

hs=a1+a2GRV(ice) (12)

wherehsisthesnowdepthinmeters,anda1=2.9anda2=-782arecoefficientsderivedfromthelinearregressionofinsitusnowdepthmeasurementsonmicrowavedata.GRV(ice)isthespectralgradientratiocorrectedfortheseaiceconcentration,C,asfollows

GRV(ice)=[TB(37V)-TB(19V)-k1(1-C)]/[TB(37V)+TB(19V)-k2(1-C)] (13)

withk1=TBO(37V)-TBO(19V)andk2=TBO(37V)+TBO(19V).Theopenwaterbrightnesstemperatures,TBO,areaveragevaluesfromopenoceanareasandareusedasconstants.TheprincipalideaofthealgorithmissimilartotheAMSR-Esnow-on-landalgorithm(Kellyetal.2003)utilizingtheassumptionsthatscatteringincreaseswithincreasingsnowdepthandthatthescatteringefficiencyisgreaterat37GHzthanat19GHz.Forsnow-freeseaice,thegradientratioisclosetozeroanditbecomesmoreandmorenegativeasthesnowdepth(andgrainsize)increases.Thecorrelationofregionalinsitusnowdepthdistributionsandsatellite-derivedsnowdepthdistributionsis0.81(Figure6).Theupperlimitforsnowdepthretrievalsis50cm,whichisaresultofthelimitedpenetrationdeptha19and37GHz.

Thealgorithmisapplicabletodrysnowconditionsonly.Attheonsetofmelt,theemissivityofboththe19GHzandthe37GHzchannelsapproachunity(thatofablackbody)andthegradientratioapproacheszeroinitiallybeforebecomingpositive.Thus,snowdepthisindeterminateunderwetsnowconditions.Snow,whichiswetduringtheday,frequentlyrefreezesduringthenight.Thisrefreezingresultsinverylargegrainsizes(Colbeck1982)whichleadstoareducedemissivityat37GHzrelativeto19GHztherebydecreasingGRV(ice)andthusleadstoanoverestimateofsnowdepth.Thesethaw-freezeevents,therefore,causelargetemporalvariationsinthesnowdepthretrievals.Thistemporalinformationisusedinthealgorithmtoflagthesnowdepthsasunretrievablefromthoseperiodswithlargefluctuations.

Asgrainsizeinsitumeasurementsareevenlessfrequentlycollectedthansnowdepthmeasurements,theinfluenceofgrainsizevariationscouldnotbeincorporatedintothealgorithm.Becauseoftheuncertaintiesingrainsizeanddensityvariationsaswellassporadicweathereffects,AMSR-Esnowdepthproductswillbe5-dayaveragessimilartothesnow-on-landproduct.

Page 15: AMSR2 SEA ICE ALGORITHM THEORETICAL BASIS DOCUMENT

15

SnowdepthsareretrievedfortheentireSouthernOcean,butonlyfortheseasonalseaicezonesintheArctic,becausethemicrowavesignatureofsnowisverysimilartothemultiyearicesignaturesothatsnowdepthonmultiyearicecannotberetrievedunambiguously.Tothisend,weuseadynamicmultiyearicemaskbasedonathresholdinGRwhichevolvesonaday-to-daybasisstartingfromOctober1ofeachyearuntiltheonsetofmelt.

Figure6:Comparisonofin-situandSSM/I-derivedsnowdepthdistributions[fromMarkusandCavalieri1998].

3.2.3Seaicedrift

ThemethodtoestimateseaicedriftistheMaximumCross-Correlationfeature-trackingalgorithmdevelopedattheUniversityofColorado(CU)(Emeryetal.,1995).Thebasicmethodologyofthealgorithmisfairlysimple.Twospatiallycoincidentimagesareobtained,separatedbysomeperiodoftime.Atargetarea,whichmaybedefinedbyapixeloragroupofpixels,ischoseninthefirst(older)image.Then,asearchareasurroundingthetargetareaischoseninthesecond(newer)image(Figure1).Correlationswiththetargetareainthefirstimagearecomparedwithallregions(ofthetargetarea’ssize)inthesearchareainthesecondimage.Theregionwiththehighestcorrelationisdeterminedtobethelocationwherethe

Page 16: AMSR2 SEA ICE ALGORITHM THEORETICAL BASIS DOCUMENT

16

targetmoved(Figure2).Afilteringalgorithmisthenemployedtoremoveatleastsomeofthequestionablematches.

GriddedLevel3dailycompositebrightnesstemperaturesat12.5kmonthepolarstereographicgridareusedasinputintothemotionalgorithm.Icemotionsarecalculatedfromtwocompositeimagesusingtheslidingwindowtofindthecorrelationpeak,whichdeterminesthedistanceafeaturehasmoved.Thedriftisthencomputedbydividingthedistancebythetimeseparation(24hours).Aseaicemaskisappliedtoonlyretrieveicemotionwhereconcentrationisabovethestandardseaiceextentthresholdof15%concentration.Falsecorrelationscanoccurduetocloudsorvariabilityoficesurfacefeatures.Toeliminatethis,firstaminimumcorrelationthresholdof0.7isappliedtoeliminateweakmatches.Nextapost-processingfilterprogramisruntoremoveatleastsomequestionableanderroneousmotions.Thisusesthefactthatmotionishighlyspatiallycorrelatedandrequiresthateachvectorbereasonablyconsistentinspeedanddirectionwithatleasttwoneighboringmotionestimates.

TheCUalgorithmissimpletoimplementandwidelyvalidated.Ithasbeenwidelyusedonavarietyofimagery,includingAMSR-E(e.g.,MeierandDai2006)andiscurrentlythebasisforalong-termseaicedriftproduct(Tschudietal.2016)attheNationalSnowandIceDataCenter.Intheorytheaccuracyofthealgorithmislimitedbythegriddedspatialresolutionoftheinputdata–i.e.,aseaiceparceleitherstaysinitscurrentgridcellormovestoasurroundinggridcell.Thiswouldlimittheaccuracyto~12.5kmperday.Therearealsoadditionalerrorsourcessuchasfalsecorrelationsfromcloudsoricesurfacechanges,sotheuncertaintywouldbeevenmorethan12.5kmperday.However,thealgorithmincludesa4Xoversamplingroutinetoyieldatheoreticalaccuracyofupto3.125kmperday.ValidationstudiesofAMSR-Emotionsincomparisonwithbuoy-derivedmotions(MeierandDai2006)showanRMSerrorof4.5–5.0kmperdayforeachvelocitycomponentwithnegligiblebias.ThisisconsistentwithpreviousstudiesusingSSMI-derivedmotions,giventheincreasedresolutionofAMSR-E(Meieretal.,2000).Errorsarelowestinthecentralicepackduringwinter.Highererrorsoccurneartheiceedgeduetogrowthandmeltoficeandthinicecondition,duringsummermeltduetoliquidwateronthesurfaceobscuringpossiblecorrelation,andnearthecoastduetothespatialresolutionlimitations.

3.2.4Adjustmentofbrightnesstemperatures

Asnotedabove,theseaicealgorithmsforAMSR2arethesameasthoseusedforAMSR-E.TheonlysubstantialdifferenceisthatAMSR2brightnesstemperaturesareadjusted–i.e.,intercalibrated–tomatchAMSR-EsothatthealgorithmcoefficientscanremainthesameandobtainconsistentgeophysicalestimatesfrombothAMSR-EandAMSR2.

Page 17: AMSR2 SEA ICE ALGORITHM THEORETICAL BASIS DOCUMENT

17

Mostintercalibrationeffortsuseoverlapbetweensensorstomakeadjustmentstothesensordata.ThiswasnotpossibleherebecausetherewasnooverlapbetweenAMSR2andthefully-operationalAMSR-E.However,AMSR-Edidoperateinaslowrotationmode.ThisallowedanintercalibrationacrosscommonpointsofAMSR2andtheslow-rotationAMSR-Ebrightnesstemperatures.RegressionswereperformedforthesecommonpointsoverafullyearandaveragedtodevelopregressionequationstoadjustAMSR2brightnesstemperaturesinto“pseudo-AMSR-E”brightnesstemperaturesthatwouldbeconsistentwiththeAMSR-Ealgorithms;regressionsweredoneindependentlyforboththeArcticandtheAntarctic.ThevaluesusedintheAMSR2processingareprovidedinTable1below.TheseregressionequationsareappliedtotheAMSR2brightnesstemperaturesbeforetheyarefedintothesubsequentgeophysicalalgorithms.

TableIAverageregressionandcorrelationcoefficientsforAMSR2Tbs,basedoncomparisonwith2rpmAMSR-ETbsforJanuary–December2013(FromMeierandIvanoff2017)

NorthernHemisphere SouthernHemisphere

Channel Slope Intercept Correlation Slope Intercept Correlation

18V 1.031 -9.710 0.9984 1.032 -10.013 0.9982

18H 1.001 -1.104 0.9985 1.000 -1.320 0.9983

23V 0.999 -1.706 0.9979 0.993 -0.987 0.9976

36V 0.997 -2.610 0.9916 0.995 -2.400 0.9919

36H 0.996 -2.687 0.9938 0.994 -2.415 0.9932

89V 0.989 0.677 0.9766 0.975 4.239 0.9619

89H 0.977 3.184 0.9621 0.969 4.935 0.9549

Overall,theregressionsshowedverygoodconsistencybetweenAMSR2andtheslow-rotationAMSR-E.Themostnotabledifferenceisfoundinthe18GHzVpolarizationchannel,whichhasslopesslightlyofffrom1andhigherinterceptvalues.The89GHzchannelsalsoshowlessagreementthantheotherchannels,likelyduetothegreateratmosphericemissionat89GHz.

Forseaice,theadjustmentsareevaluatedintermsofminimizingthedifferenceinextent,area,andconcentration.ThiswasnotexplicitlypossibleherebecausethereisnotcompletecoveragetocalculatetheseparametersfromtheAMSR-Eslowrotationdata.Inordertoexaminethese

Page 18: AMSR2 SEA ICE ALGORITHM THEORETICAL BASIS DOCUMENT

18

variables,abridgeconcentrationdatasetwasused,basedonSSMISbrightnesstemperatures.ThisbridgedatasetspannedafullyearofAMSR-EandAMSR2(2010–2013).DifferencesinextentandareabetweeneachsensorandSSMISwerecalculated.Thesetwofieldswerethendifferencedtoobtainadouble-differenceAMSR2-AMSREextentandareadifference.Fromthis,andexaminationoftheconcentrationfields,itwasfoundthatanadjustmentintheGR3619weatherfilter,from0.050to0.046,wasneededtooptimizetheseaiceconcentrationfields.

Thedouble-differencefieldsandsubsequentGR3619adjustmentdecreasedtheextentbiasbetweenAMSR2andAMSR-Efrom~200,000squarekilometersto<10,000squarekilometers(Figure7).Thisisasgoodorbetterthanpreviousseaiceextentsensorintercalibrations(e.g.,Cavalierietal.2012;Meieretal.2011).ThefullmethodologyforderivingtheregressioncoefficientsandtheadjustmenttoAMSR2brightnesstemperaturesfortheseaicealgorithmsisdescribedinMeierandIvanoff(2017).

4.VarianceandUncertaintyEstimates

TheseaiceproductshavebeenthoroughlyvalidatedinpreviousstudiesusingAMSR-E(e.g.,MarkusandCavalieri2000;MarkusandCavalieri2009;MarkusandCavalieri1998;LiuandCavalieri1998).FurthercomparisonshavebeenmadeforAMSR2seaiceconcentrationinMeieretal.(2017).ThesecomparedAMSR2concentrationestimateswithconcentrationsderivedfromvisibleimageryobtainedfromtheSuomiVisibleandInfraredImagingRadiometerSuite(VIIRS).

Thisvalidationshowedaconcentrationbiasof3.9%andanRMSEof11.0%intheArctic(Figure8),andabiasof4.45%andanRMSEof8.8%intheAntarctic.ThesevaluesareconsistentwithpreviousvalidationstudiesofAMSR-Eseaiceconcentrations.ThereisconsiderablevariationthoughintheaccuracyofAMSR2concentration,dependingonseaiceconditions.Errorsarelowforhighconcentrationregions,butarehigherinregionswithlowerconcentrations,suchasthemarginalicezone.Again,thisperformanceistypicalforpassivemicrowaveretrievalsandhasbeenfoundinnumerousotherstudies.

5.Implementationissues

5.1Computation/programming/proceduralconsideration

5.2Qualitycontrolanddiagnostics

5.3Constraints,limitationsanddiagnostics

Page 19: AMSR2 SEA ICE ALGORITHM THEORETICAL BASIS DOCUMENT

19

ThebasiclimitationsoficeconcentrationdatafromAMSR2includerelativelycoarseresolutionandtheinabilitytounambiguouslyidentifyareascoveredbythinice,pancakeiceandmeltpondedice.Thinice,pancakeiceandmeltpondedicecanhaveemissivitiesintermediatetothoseofopenwateranddrythickiceandcouldcontributetosignificantuncertaintiesintheretrievaloficeconcentration.Agoodschemethatenablesclassificationofeachdataelementintodifferentsurfacetypeswouldhelpminimizeuncertainties.

Fig.7.(a)Arcticand(b)Antarcticextentdoubledifferences,(AMSR2–SSMIS)–(AMSR-E–SSMIS)fororiginalAMSR2(green),regressedAMSR2(purple),andregressedAMSR2withGRadjustment(blue).FromMeierandIvanoff(2017).

!0.4%

!0.2%

0%

0.2%

0.4%

0.6%

0.8%

1%

1.2%

J% F% M% A% M% J% J% A% S% O% N% D%

Extent%Differen

ce%(1

06%km

2 )%

Date,%2010%&%2013%

ArcCc%NT2%AMSR2!AMSRE%Extent%%Double%Difference,%2010%&%2013%

AMSR2%NT2(Original)!AMSRE%NT2%

AMSR2%NT2(Regressed)!AMSRE%NT2%

AMSR2%NT2(GR%Adjusted)!AMSRE%NT2%

!0.2%

0%

0.2%

0.4%

0.6%

0.8%

1%

1.2%

J% F% M% A% M% J% J% A% S% O% N% D%

Extent%Differen

ce%(1

06%km

2 )%

Date,%2010%&%2013%

AntarcCc%NT2%AMSR2!AMSRE%Extent%%Double%Difference,%2010%&%2013%

AMSR2%NT2(Original)!AMSRE%NT2%

AMSR2%NT2(Regressed)!AMSRE%NT2%

AMSR2%NT2(GR%Adjusted)!AMSRE%NT2%

(a)

(b)

Page 20: AMSR2 SEA ICE ALGORITHM THEORETICAL BASIS DOCUMENT

20

Fig.8.ComparisonofAMSR2minusVIIRSiceconcentrationsfordifferentAMSR2iceconcentrationranges/binsintheArctic.TheAMSR2concentrationiscomputedwiththeNASATeam2algorithm.Notethatthey-axisrangeisdifferentfor"All","90-100%",andtheotherplots.FromMeieretal.(2017).

Page 21: AMSR2 SEA ICE ALGORITHM THEORETICAL BASIS DOCUMENT

21

6.References

Bhatt,U.S.,etal.2010.CircumpolarArctictundravegetationchangeislinkedtosea---icedecline.EarthInteractions14:1-20.doi:10.1175/2010EI315.1.

Cavalieri,D.J.,P.Gloersen,andW.J.Campbell.1984.DeterminationofseaiceparameterswiththeNIMBUS7scanningmultichannelmicrowaveradiometer.J.Geophys.Res.89:5355-5369.

Cavalieri,D.J.,K.St.Germain,andC.T.Swift.1995.ReductionofWeatherEffectsintheCalculationofSeaIceConcentrationwiththeDMSPSSM/I.J.Glaciology41:455-464.

Cavalieri,D.J.,C.L.Parkinson,P.Gloersen,J.C.Comiso,andH.J.Zwally.1999.DerivingLong-TermTimeSeriesofSeaIceCoverfromSatellitePassive---MicrowaveMultisensorDataSets.J.Geophys.Res.104:15,803-15,814.

Cavalieri,D.J.,C.L.Parkinson,N.DiGirolamo,andA.Ivanoff,2012.IntersensorcalibrationbetweenF13SSMISandF17SSMISforglobalseaicedatarecords,IEEEGeosci.Rem.Sens.Lett.,vol.9,no.2,pp.233-236,doi:10.1109/LGRS.2011.2166754.

Colbeck,S.C.1982.Anoverviewofseasonalsnowmetamorphism.Rev.Geophys.SpacePhys.20:45-61.

Combes,J.M.,A.Grossmann,andPh.Tchamitchian.1989.Wavelet:timefrequencymethodsandphasespace.InProceedingsoftheInternationalConferenceonWavelet,Marseille,France.NewYork:Springer-Verlag.

Comiso,J.C.,D.J.Cavalieri,C.L.Parkinson,andP.Gloersen.1997.Passivemicrowavealgorithmsforseaiceconcentration-Acomparisonoftwotechniques.RemoteSens.Environ.,60:357-384.

Comiso,J.C.2012.LargedecadaldeclineintheArcticmultiyearicecover.J.Climate25(4):1176-1193.doi:10.1175/JCLI-D-11-00113.1.

Comiso,J.C.andF.Nishio.2008.TrendsintheseaicecoverusingenhancedandcompatibleAMSR-E,SSM/I,andSMMRdata.J.Geophys.Res.113(2):C02S07.doi:10.1029/2007JC004257.

Comiso,J.C.2009.EnhancedseaiceconcentrationandiceextentfromAMSR---EData.J.RemoteSensingSoc.ofJapan29(1):199---215.

Comiso,J.C.,D.J.Cavalieri,andT.Markus.2003.Seaiceconcentration,icetemperature,andsnowdepth,usingAMSR-Edata.IEEETGRS41(2):243---252.

Page 22: AMSR2 SEA ICE ALGORITHM THEORETICAL BASIS DOCUMENT

22

Comiso,J.C.2010.PolarOceansfromSpace.NewYork:Springerverlag.

Comiso,J.C.1986.Characteristicsofwinterseaicefromsatellitemultispectralmicrowaveobservations.J.Geophys.Rev.91(C1):975-994.

Comiso,J.C.andK.Steffen.2001.StudiesofAntarcticseaiceconcentrationsfromsatellitedataandtheirapplications.J.Geophys.Res.106(C12):31361-31385.

Comiso,J.C.andC.L.Parkinson.2008.ArcticseaiceparametersfromAMSR-Eusingtwotechniques,andcomparisonswithseaicefromSSM/I.J.Geophys.Res.113(C2):C02S05.doi:10.1029/2007JC004255.

Comiso,J.C.1995.SSM/IConcentrationsusingtheBootstrapAlgorithm.NASARP1380.GoddardSpaceFlightCenter:NationalAeronauticsandSpaceAdministration.

Emery,W.,C.Fowler,andJ.Maslanik.1995.SatelliteRemoteSensingofIceMotion,inOceanographicApplicationsofRemoteSensing,ed.MotoyoshiIkedaandFredericW.Dobson.CRCPress,BocaRaton.

FraserR.S.,N.E.Gaut,E.C.Reifenstein,andH.Sievering.1975.InteractionMechanisms------WithintheAtmosphere.InManualofRemoteSensingeditedbyR.G.Reeves,A.Anson,D.Landen,181-223.FallsChurch,VA:AmericanSocietyofPhotogrammetry.

Gloersen,P.andD.J.Cavalieri.1986.Reductionofweathereffectsinthecalculationofseaiceconcentrationfrommicrowaveradiances.J.Geophys.Res.91:3913-3919.

Kelly,R.E.J,A.T.C.Chang,L.Tsang,andJ.L.Foster.2003.AprototypeAMSR---Eglobalsnowareaandsnowdepthalgorithm.IEEETransactionsonGeoscienceandRemoteSensing41(2):230-242.

Kummerow,C.1993.OntheaccuracyoftheEddingtonapproximationforradiativetransferinthemicrowavefrequencies.J.Geophys.Res.98:2757-2765.

Lindsay,R.W.andJ.Zhang.2005.ThethinningofArcticseaice,1988-2003:Havewereachedthetippingpoint?J.Climate18:4879-4894.

Markus,T.andD.J.Cavalieri.1998.SnowdepthdistributionoverseaiceintheSouthernOceanfromsatellitepassivemicrowavedata.InAntarcticSeaIce:PhysicalProcesses,InteractionsandVariability,19-39.Washington,DC:AmericanGeophysicalUnion.

Markus,T.andD.J.Cavalieri.2000.AnenhancementoftheNASATeamseaicealgorithm.IEEETrans.Geosci.andRemoteSensing38:1387-1398.

Page 23: AMSR2 SEA ICE ALGORITHM THEORETICAL BASIS DOCUMENT

23

Markus,T.andD.J.Cavalieri.2009.TheAMSR---ENT2seaiceconcentrationalgorithm:itsbasisandimplementation.JournalofTheRemoteSensingSocietyofJapan29(1):216-225.

Markus,T.andS.T.Dokken.2002.EvaluationofArcticlatesummerpassivemicrowaveseaiceretrievals.IEEETrans.GeoscienceRemoteSensing40(2):348-356.

Matzler,C.,R.O.Ramseier,andE.Svendsen.1984.Polarizationeffectsinsea---icesignatures.IEEEJ.OceanicEng.9:333-338.

Meier,W.N.,J.A.Maslanik,andC.W.Fowler.2000.ErroranalysisandassimilationofremotelysensedicemotionwithinanArcticseaicemodel,J.Geophys.Res.,105(C2),3339–3356,doi:10.1029/1999JC900268.

Meier,W.N.,andM.Dai.2006.High-resolutionsea-icemotionsfromAMSR-Eimagery,Ann.Glaciol.,44,352-356.

Meier,W.N.,S.J.S.Khalsa,andM.H.Savoie.2011.IntersensorcalibrationbetweenF-13SSM/IandF-17SSMISnear-real-timeseaiceestimates.IEEETrans.Geosci.Rem.Sens.,vol.49,no.9,pp.3343-3349.

Parkinson,C.L.andJ.C.Comiso.2008.AntarcticseaicefromAMSR---EfromtwoalgorithmsandcomparisonswithseaicefromSSM/I.J.Geophys.Res.113(C2):C02S06.doi:10.1029/2007JC004253.

Shibata,A.,H.Murakami,andJ.Comiso.2010.AnomalousWarmingintheArcticOceanintheSummerof2007.J.RemoteSensingSocietyofJapan30(2):105-113.

Sigmond,M.andJ.C.Fyfe.2010.HastheozoneholecontributedtoincreasedAntarcticseaiceextent?Geophys.Res.Lett.37(L18):L18502.doi:10.1029/2010GL044301.

Svendsen,E.,C.Matzler,andT.C.Grenfell.1987.Amodelforretrievingtotalseaiceconcentrationfromaspacebornedual-plarizedpassivemicrowaveinstrumentoperatingnear90GHz.Int.J.Rem.Sens.8:1479-1487.

Swift,C.T.,L.S.Fedor,andR.O.Ramseier.1985.Analgorithmtomeasureseaiceconcentrationwithmicrowaveradiometers.J.Geophys.Res.90(C1):1087-1099.

Tschudi,M.,C.Fowler,J.Maslanik,J.S.Stewart,andW.Meier.2016.PolarPathfinderDaily25kmEASE-GridSeaIceMotionVectors,Version3.Boulder,ColoradoUSA.NASANationalSnowandIceDataCenterDistributedActiveArchiveCenter.doi:http://dx.doi.org/10.5067/O57VAIT2AYYY.

Page 24: AMSR2 SEA ICE ALGORITHM THEORETICAL BASIS DOCUMENT

24

Turner,J.,J.C.Comiso,G.J.Marshall,T.A.Lachlan---Cope,T.Bracegirdle,T.Maksym,M.Meredith,andZ.Wang.2009.Non---annularatmosphericcirculationchangeinducedbystratosphericozonedepletionanditsroleintherecentincreaseofAntarcticseaiceextent.Geophy.Res.Lett.36(L8):L08502.doi:10.1029/2009GL037524.

Wang,M.andJ.Overland.2009.AseaicefreesummerArcticwithin30years?Geophys.Res.Lett.36(L7):L07502.doi:10.1029/2009GL037820.

Yaguchi,R.andK.Cho.2009.ValidationofseaicedriftvectorextractionfromAMSR---EandSSM/IdatausingMODISdata.J.RemoteSens.Japan29(1):242-252.

Zwally,H.J.,J.C.Comiso,C.Parkinson,D.Cavalieri,P.Gloersen.2002.VariabilityoftheAntarcticseaicecover.J.Geophys.Res.107(C5):1029-1047.

Zwally,H.J.,J.C.Comiso,C.L.Parkinson,W.J.Campbell,F.D.Carsey,andP.Gloersen.1983.AntarcticSeaIce1973---1976fromSatellitePassiveMicrowaveObservations.NASASpec.Publ.459.Washington,DC:NASA.