Multitemporal Repeat Pass SAR
- 格式:pdf
- 大小:560.90 KB
- 文档页数:10
IEEETRANSACTIONSONGEOSCIENCEANDREMOTESENSING,VOL.43,NO.6,JUNE20051219MultitemporalRepeatPassSARInterferometryofBorealForests
JanAskne,LifeFellow,IEEE,andMaurizioSantoro,Member,IEEE
Abstract—MultitemporalinterferometricEuropeanRemoteSensing1and2satellitetandempairsfromaforesttestsiteinFinlandareexaminedinordertodeterminethestemvolumeretrievalaccuracy.Aformofmultitemporalfilteringisintroducedtoinvestigatewhatforeststandsshowamultitemporalconsistencyincoherence.Itisfoundthatalargestandsizeisamajorfactortoobtainaccurateretrievals.Theeffectofheterogeneityofforeststandsisalsodiscussed.Basedonthestandsshowinghighestmultitemporalconsistencydifferentmodelsforscatteringandcoherencearecompared.Theinterferometricwatercloudmodelischosenforstemvolumeretrieval.Thevariationofthemodelparameterswithmeteorologicalparametersisinvestigatedandtheresultsillustratethatthebestimagingconditionsareobtainedforsubzerotemperaturesandwindyconditions.Itisshownthatforthe20standsshowinghighestmultitemporalconsistencythestemvolumecanberetrievedwitharelativeerrorof21%,deterioratingwhenthenumberoftestingstandsisincreased,e.g.,for80standstheerroris48%.For37largeforeststandsrepresenting48%oftheinvestigatedareatherelativestemvolumeerroris26%.WithexperiencefromanothersiteinSwedenwemayconcludethattheerrorlevelforamultitemporalinterferometricsyntheticapertureradarevaluationofstemvolumeforlargeforeststands(2ha)inawellmanagedandhomogeneousborealforestmaybeexpectedtobeinthe15%to25%range,deterioratingforsmallandheterogeneousstandsandforimagesacquiredundernonwinterconditions.
IndexTerms—Borealforest,EuropeanRemotesensingSatel-lite1and2(ERS-1/2),interferometricsyntheticapertureradar(InSAR),multitemporal,stemvolume.
I.INTRODUCTION
THEREisaneedfromaforestmanagementpointofview
aswellasfromanenvironmentalpointofviewtoderiveforestbiomassorstemvolume.ForthispurposewewillanalyzepropertiesofC-bandsyntheticapertureradar(SAR)images.Generally,C-bandSARbackscatterisconsideredunsuitableforretrievingforestbiomassorstemvolumesincethebackscattersaturatesatalowbiomassorstemvolumelevel.However,aswillbediscussedfurtherbelow,ithasbeenfoundthatinter-ferometricsyntheticapertureradar(InSAR)fromtheEuropeanRemoteSensing1and2(ERS-1/2)satellitetandemperiodcanprovidecoherenceinformationwhichgiveshighstemvolumeretrievalaccuracyundersuitableconditions.Someinvestiga-tionsshowveryhighretrievalaccuracy[1]–[4],whileothers
ManuscriptreceivedAugust2,2004;revisedDecember26,2004.ThisworkwasinpartsupportedinpartbytheSwedishNationalSpaceBoard.J.AskneiswiththeDepartmentofRadioandSpaceScience,ChalmersUniversityofTechnology,SE-412Gothenburg,Sweden(e-mail:askne@rss.chalmers.se).M.SantoroiswiththeInstituteofGeography,Friedrich-Schiller-University,D-07737Jena,Germany(e-mail:maurizio.santoro@geogr.uni-jena.de).DigitalObjectIdentifier10.1109/TGRS.2005.846878
[5]–[7]showmorelimitedpotentialalthoughstillmuchbetterthanthosederivedfromtheSARbackscatter.Importantques-tionsareunderwhatconditionsacceptableaccuracycanbeob-tained,towhatextentresultsaresitedependent,andtowhatextenttheyaremethodrelated.Theimportantaspectwithspacebornemicrowaveremotesensingistheindependenceofclouds.However,thisdoesnotmeanthattheobservationsareweatherindependent.Forstemvolumeretrievalwefindthatmicrowavebackscatterandrepeat-passcoherenceareverysensitivetoweatherconditions.Theindependenceofcloudsontheotherhandyieldspossibilitytoregularlycoveracertainsiteandthepossibilityforamulti-temporalretrievalcancompensatethenoiseassociatedwiththeindividualacquisitions.ThemultitemporalaspectisoneofthethemesofthispaperforwhichamainobjectiveistoanalyzetheaccuracyofstemvolumeestimatesfromaFinnishtestsite,Tuusula,usingC-bandInSAR.Forapplicationofaremotesensingtechniqueitisimpor-tanttoevaluateitspropertiesusinganumberofwellcharacter-izedtestsites.C-bandrepeatpassinterferometryobservationsfromTuusula,Finland,indicatearelativelyhighstemvolumeretrievalerror.48%relativeerrorwasobtainedin[6]for134foreststandsusing14imagepairsand34%in[5]for37standsusingfiveimagepairs.Aninvestigationofasitewithman-agedandhomogeneousforeststandsinSwedenonthesamelatitude,Kättböle,resultedinmuchhigheraccuracy.In[2]–[4]InSARobservationsfromKättbölehavebeenanalyzedshowingarelativeerrorinstemvolumeestimatesof13%(uncorrectedforerrorsinsitumeasurementslikeotherresultsreportedinthispaper)fortheinvestigatedstandswithstemvolumesupto335m/haandameanstemvolumeof140m/ha.InTu-
usula,ontheotherhand,wefindstemvolumesupto539m/ha,whilethemeanstemvolumeof158m/haisnotmuchdifferentfromKättböle.Asimilarsituationregardingstemvolumere-trievalaccuracywasfoundconcerningL-bandbackscatterfromtheJapaneseEarthResourceSatellite1[8],wheredatafromKättböleresultedin25%relativeerroranddatafromTuusularesultedin40%.Thedifferencesinaccuracyarepartlyassoci-atedwiththeweatherconditionsduringtheacquisitionsbutareprobablymainlycausedbystructuraldifferencesbetweentheforestconditionsonthetwositesandthelargerrangeofstemvolumesinTuusula.Theaccuracyoftheinsitudataisalsoofimportance.Anotherobjectiveofthispaperistoanalyzedifferentmodelsforthedependenceofbackscatterandcoherenceonstemvolume,inparticularthemodelsproposedin[9]andappliedtotheTuusulatestsitein[6]andthemodelsusedin[1],[3],and[4]andappliedin[5]totheTuusulatestsite.Thegoalistodeterminewhichmodelbestdescribesthedataand