Vous êtes sur la page 1sur 129

RFF REPORT

Measurementand Monitoringofthe WorldsForests


AReviewandSummaryofRemoteSensing TechnicalCapability,20092015
MatthewFaganandRuthDeFries

DECEMBER 2009

Preface
ThisreportispartofaseriesofstudiescarriedoutunderaninitiativeatResourcesfortheFuture (RFF)toconsidertheeconomic,technical,andinstitutionalissuesassociatedwiththegoalof improvingglobalforestmeasurementandmonitoring.Companionstudiesunderthisinitiative includeanRFFdiscussionpaperdocumentinguncertaintiesanddiscrepanciesinexistingforest measurement(Waggoner2009)andanotherRFFreportprovidingperspectivesonhowgoodis goodenoughinimprovingmeasurementandmonitoring(Macauleyandcoauthorsforthcoming). ThefinancialsupportoftheAlfredP.SloanFoundationisgratefullyacknowledgedforallwork underthisinitiative.

Acknowledgements
SpecialcontributionsweremadebyMollyMacauley,PaulWaggoner,RogerSedjo,Michael Obersteiner,CraigDobson,JosefKellndorfer,andMiriamMarlier.SteffenFritz,RalphDubayah, StevenSesnie,andBonnieTicealsohelped.

TableofContents
ExecutiveSummary..................................................................................................................................................i Scope..........................................................................................................................................................................i CurrentandEmergingSatelliteTechnologies........................................................................................ii RemoteSensingandGlobalForestMeasurement................................................................................iii SummaryandConclusions...................................................................................................................................v Chapter1.Forests,TheirMeasurement,andtheNeedforGlobalRemoteSensing....................1 1.1CurrentNeedsforForestMonitoring..................................................................................................1 1.2ABriefHistoryofGlobalForestMeasurementandRemoteSensing....................................4 1.3DiversityofForestTypes..........................................................................................................................5 1.4ForGlobalForestMeasurement,WhyUseSatelliteImagery?.................................................6 1.5ScopeandPurpose......................................................................................................................................6 Chapter2.RemoteSensingandForestMeasurement:anOverview.................................................7 2.1ComparingSatelliteandAirborneRemoteSensing......................................................................7 2.2PassiveandActiveRemoteSensing:AnOverview.....................................................................11 2.3PuttingitAllTogether:RemoteSensingFusion..........................................................................21 2.4AnOverviewofCurrentandNearTermEarthObservingTechnology............................24 Chapter3.TechnicalConsiderationsinGlobalForestMonitoring..................................................32 3.1MeasuringForestsGlobally..................................................................................................................32 3.2CriteriaforaGlobalForestMonitoringProgram........................................................................35 3.3DevelopingandEvaluatingReferenceForestMaps ...................................................................35 3.4SurveyMethodsforDeterminingForestAreaandType.........................................................38 3.5PriceandLogistics....................................................................................................................................40 Chapter4.AnOverviewofRemoteSensingCapabilitiesforForestMeasurement:Current andNearTermTechnologies...........................................................................................................................42

4.1ForestArea:CurrentMeasurementMethods...............................................................................42 4.2ForestStructure:CurrentMeasurementMethods.....................................................................52 4.3EstimatingForestVolumeandBiomassusingRemoteSensing...........................................57 4.4EstimatingForestCarbonStocksfromRemotelySensedData .............................................63 Chapter5.ForestsfromSpace:UnparalleledMeasurements,TimelyMoment........................68 5.1ImprovingForestBiomassandCarbonEstimates.....................................................................68 5.2InConclusion:TowardImprovedMeasuresandMonitoring................................................71 Satellite,Sensor,andRemoteSensingAcronymsandAbbreviations............................................73 Acronyms..................................................................................................................................................................75 Appendix.CurrentandNearTermEarthObservingSatelliteTechnology:AnOverview....77 Bibliography............................................................................................................................................................97

FAGAN AND DEFRIES

MeasurementandMonitoringoftheWorlds Forests:AReviewandSummaryofRemote SensingTechnicalCapability,20092015


MatthewFaganandRuthDeFries* Executive Summary
orestsareecosystemsoffundamentalimportancetohumanity,yetweknowlittleaboutthe globalstatusofforests.WecanmakemorecurrentandinformativemapsofMarsthanofthe AmazonbasinortheRussianborealforest.Thisgapinourknowledgearisesfromseveralsources, includingahistoricalshortageoftoolstoobservetheentireglobe;thelackofconsistentglobal initiativesonforests;balkanized,inconsistentrecordkeeping;andtheabsenceofaconcertedand systematicefforttoinventoryandmonitortheworldsforests. Tounderstandtheplanetarycarbonbudget,itisimperativetogenerateaccurateandreliable estimatesofglobalforestcoverandtheamountofbiomassandcarbonharboredbytheplanets forests.Yetwidespreaduncertaintiesinforestmeasurementshavehamperedeffortstoobtainthis basicscientificdata.Indeed,themostsignificantweaknessesinestimatesoftheplanetarycarbon budgetderivefromuncertaintiesaboutterrestrialecosystems.Satellitebasedestimatesofforest coverandbiomasshavebeguntofillthisneed. Tomeasureforestsworldwide,satelliteimageryisapracticalnecessity.Aerialobservationsare expensiveatpresentandonlycoversmallareasatatime.Groundmeasurementsarealsoexpensive andarelogisticallychallengingandspatiallyrestricted.Neitheraerialnorgroundobservationsare wellsuitedtocontinuousmeasurementoftheentireglobalforest.Satellitemappingisnecessaryto detectdeforestationandregrowthinremotetropicalforestsandtotrackthenorthernexpansionof borealforestsinawarmingworld.Thegreateststrengthsofsatellitebasedmeasurementsaretheir unparalleled,unbiasedmeasurements,theirmonthlytodailyfrequency,andabovealltheir synopticnature.Satellitesprovideageneralviewofthewholeearththatisnotpossiblewithany otherforestmeasurementmethod.

Scope
Weevaluatecurrentandnearterm(20092015)technologiesformeasuringandmonitoring globalforests.Wefocusprimarilyonremotesensing(definedinthisreportastheanalysisof satelliteandaerialimagery),becausethistechnologymeetsthesteeplogisticalchallengeof measuringtheworldsforestsinanaccurate,repeatable,andinexpensivemanner.Weemphasize theobservationsneededtoprovideaccurate,basicmeasuresofforestattributesforusebyforest scientists,decisionmakers,andthepublic.Wealsoillustratehowtheresultingdatawouldaddress

*MatthewFagan(mef2153@columbia.edu)isagraduatestudentintheDepartmentofEcology,Evolution,andEnvironmental

BiologyatColumbiaUniversity.RuthDeFries(rd2402@columbia.edu)istheDenningProfessorofSustainableDevelopmentin theDepartmentofEcology,Evolution,andEnvironmentalBiologyatColumbiaUniversity.

FAGAN AND DEFRIES

someofthemostcompellingquestionsthatpersistaboutourforests.Remotesensingcan ultimatelyhelptoanswertheseimportantquestions: Whereareforestsbeinglostandgained? Howarebiodiversityandregionalclimatesrespondingtoforestlossandregrowth? Howdoforestsaffecttheglobalcarbonbudget? Howcanforestmanagementbeimprovedbycarbonmarketsandsatellitetechnology? Howwillforestsrespondtoclimatechange?

Inthisreport,weevaluatethetechnicalcapacityofsatelliteimagerytomeasureandmonitor globalforests.Inparticular,weexaminesatelliteobservationsofforestarea,volume,biomass,and carbon.Thesemeasurementsarefundamentaltoourunderstandingofthestatusandtrendsof forestsaroundtheglobeandtoourgraspoftherolethatforestsplayinmodulatingglobalclimate. Weevaluatecurrentcapacityandpredictfuturecapacityfromexpectedsatellitelaunchesand technologicaldevelopmentsbetween2009and2015.Oursummarizedareinthetablebelow.

Current and Emerging Satellite Technologies


Satellitesensorsareeitherpassiveoractive.Passivesensorsreceivereflectedopticaland thermalradiationfromtheEarthssurface.Althoughtheyarecapableofobtainingveryhigh resolutionimagesanddistinguishingamongtypesoflandcover(suchasvegetation,baresoil,and snow),theycannotpenetratecloudcover.Activesensorsarelesschallengedbycloudcover becausethesesensorsemitradiationandmeasurethereflectionsfromtheEarthssurface.Active sensorscanprovideathreedimensionalpictureoftheground,althoughtheycannotreliably renderlandscapeswithsharpchangesinelevation,suchasmountainousregions. Ingeneral,satellitetechnologyinvolvesacombinationofpixelresolutionandimagesize,with anincreaseinonedimensionrequiringareductionintheotherdimension.Coarseresolution satellitescangenerateimagesofverylargeareas,butthepixelresolutionmaybe1,000meters. Conversely,therearesensorsthatcanrenderextremelyhighresolutionpictureswithpixelsizes lessthanhalfameterbutonlyoverlimitedareas.Moderateresolutionsatellites,suchasthose withapixelsizeof30meters(m)andanimagewidthof185kilometers(km),areoftenseenasa goodcompromiseforlandmapping.TheLandsatsystem,sponsoredbytheUnitedStates,isthe oldestandmostpopularmoderateresolutionsensor,butbecauseoftechnicalfailuresinthe operatingsatellites,Landsatimagerycoverageiscurrentlynotglobal.Alternativesensorsexist,but theplannedlaunchoftheLandsatDataContinuityMission(LCDM)andtheEuropeanSentinel2 seriesofremotesensingsatellitesareimportantforcontinuedgloballandmapping. TherecentavailabilityoffreeimageryfromLandsatandtheChinaBrazilEarthResources Satellite(CBERS)missionshasrevolutionizedglobalforestmappingandwillcontinuetodosointo thefuture.Evenwithfreeorlowcostimagery,however,costremainsacriticalparameterfor worldwidemapping.Forexample,therequiredprocessingcapabilityislargebecausethevolumeof imageryisenormous. Satellitegroupsorconstellationsimprovecoveragebyenablingmorefrequent,cloudfree observationsofanygivenpointonEarth.Severaloftheseconstellationsarecurrentlyinorbitand moreareplanned.Geostationarysatellites,stationedatsome22,000milesaboveearth(most

FAGAN AND DEFRIES

ii

Earthobservingsatellitesareonlyafewhundredmilesaboveearth)canproviderapidand repeatedscansofaverylargearea.Thepixelsizeobtainedatageostationaryaltitudeisonlyabout 2.5km;by2014,however,theresolutionisexpectedtoimprovetoapproximately1kmpixelsize, makingitpossibletomonitorglobalvegetationatcoarseresolutioninrealtime.Highresolution (0.310mpixel)sensorswillcontinuetoincreaseinnumber;thesesensorsdonotacquire continuous,cloudfree,worldwidecoverageforforestmapping,buttheyareidealforvalidating estimatesfrommidresolutionsensors. Activesensors(syntheticapertureradar,knownasSAR;andlightdetectionandranginglaser, knownasLIDAR)arecurrentlyasmallproportionofthesatellitefleet,butwilldramatically increaseinnumberandcomplexityinthenextfewyears,makingnewtypesofanalysispossible. ThefirstLIDARsatellitetheIce,Cloud,andLandElevationSatellite(ICESat)willbefollowedby twoothersthatcanaddressissuesinforestmanagement.TheseothersatellitesareICESatIIand theU.S.Deformation,EcosystemStructure,andDynamicsofIceSatellite(DESDynI).TwonewSAR satellitesthatarelikelytolaunchbetween2009and2015willrevolutionizemappingofforest biomass.DESDynIwillbethefirstcombinedSARLIDARmission,anditwillintegratetherelative strengthsofSARandLIDARtocreateunparalleledmeasurementsofecosystemstructureandforest height.TheEuropeanSpaceAgencysplannedBIOMASSsatelliteisalongwavelengthSAR, specificallydesignedtopenetrateforestsandmeasureforestbiomassinrealtime. Althoughnewsatellitetechnologycanrevolutionizescientificmonitoringandmeasurementof theEarth,itisnotfailsafe.Satellitesandsatellitelaunchescan,anddo,fail.Globalforest measurementsneedtobeaccurateandrepeatedoverlongtimeperiods.And,attheendoftheir operatinglifetimes,usuallyafterfivetotenyearsorso,thesatellitesandsensorsneedtobe replacedtoenablecontinuedmeasurementandobservationovertime.

Remote Sensing and Global Forest Measurement


Foranyworldwideforestmonitoringefforttosucceed,theremustbeconsensusonforest definitions,pastreferencemaps(sothatchangecanbedetected),andselectedforestmetrics.We havechosentheForestIdentity(Kauppietal.2006)asanorganizingprincipleforthecentral metricsofthisreport.TheIdentityrelatesfourforestattributes(area,volume[densityofgrowing stock],biomass,andsequesteredcarbon)thatprovideausefulstartingpointforglobalforest monitoring. Currentmapsofforestareahavemediumtohighaccuracy.1Monitoringvolume,biomass,and carbononaregionaltoglobalscaleispossiblewithcurrenttechnologybutaccuracyislower(see tablebelow).Similarly,wecandeveloppastreferencemapsforforestarea(mapsofwhatanarea oncelookedlike),butpastreferencemapsforvolume,biomass,orcarbonwillrequireinnovative reprocessingofoldimagery.Forforestarea,accuracyisroughlydefinedasthepercentageof pixelsintheremotesensingimagerythatcorrectlyidentifylandcovertype.Forforestvolume, biomass,andcarbon,accuracyreferstothematchbetweenpredictionsfromremoteimageryand observedgroundmeasurements.

Inthisreport,theoverallaccuracyofsatelliteimageryanalysisisdesignatedbyfivestandardadjectives:veryhighaccuracy (>90percent),highaccuracy(>80percent),acceptableaccuracy(>70percent),lowaccuracy(5070percent),andpoor accuracy(<50percent).Wherealternativequalifiers(forexample,mixed)areused,weprovidetheactualaccuracy(say,67 percent).

FAGAN AND DEFRIES

iii

Designingasatellitebased,worldwideforestmonitoringsystemrequireschoicesinbudgeting, processinglogistics,samplingframeworks,andthecollectionofvalidation(orgroundtruth)data fromforestinventoriesandhighresolutionimagery.Collectionofgroundtruthdataistypically necessaryasameansofdeterminingtheaccuracyofremotesensing.Thesedataareparticularly essentialwhenattemptingtoestimateforestvolume,biomass,andcarbonusingremotesensing technology.Archivingandstandardizingglobalgroundtruthdataforforestswouldbeasignificant contributiontoglobalforestscience.Grounddata,aerialimagery,andhighresolutionsatellite imageryareexpensiveandrequirecoordinationinasamplinghierarchyforefficiency. Incurrentcoarseresolutionworldforestmaps,forestareaismeasuredwithmediumaccuracy astwoclasses(forest/nonforest)orcategorizedwithlowaccuracyintohomogenousforesttypes basedonleafpersistence(forexample,evergreenforest).Recentimprovementsinclassification techniquesandthecombinationofdistincttypesofsatelliteimagery(calledimageryfusion)have allowedmoderateresolutionmappingofforesttypeswithhighaccuracy(8090percent). Currently,completeforestclearingcanbedetectedwiththehighestaccuracy. Withcurrenttechnology,itremainsdifficulttodistinguishprimaryforestsfromtree plantationsandoldersecondaryforestsinremotesensingimages.Itisalsochallengingtodetect forestdegradationinwhichaforestispartiallyclearedbyhumanactivity.Significantprogresson theseproblemshasbeenmadeincertaingeographicregions,butaccurateglobalforestmapswith multipleclassesremainelusive.Intheyearsbetween2009and2015,wecanexpecttosee numerousimprovementsthatpromisetoaddressmanyofthesechallenges.Someoftheanticipated advancesinclude: hyperspectralsatellitesthatproduceimagerywithgreatsensitivityacrossthe electromagneticspectrum; improvedrevisittimesfromopticalandSARsatelliteconstellations; improvedalgorithmsforanalyzinglargeamountsofremoteimagery;and additionalhighresolutionandactiveimageryavailableforfusions.

Activeandpassivesatelliteimageryissensitivetoforeststructure(bothverticaland horizontal)andforeststructurecanbeusedtoestimateforestvolume,biomass,andaboveground carbon.BothSARandLIDARaredirectlysensitivetoforestvolume.SARimagestendtosaturate (orfailtopenetrate)indenseforestsbuttheycancoverlargeareas.Conversely,LIDARdatadonot saturatebutcanonlymeasuresmallareas.Inopenforests,stereoandhighresolutionimagerycan alsomeasureforestheightandcanopystructureandhavethepotentialtoaidLIDARandSAR measurementsofforestvolumeandbiomass. Denseforestsrepresentachallengetosatelliteestimationofforestvolumeandbiomass. Scientistsseektoovercomethischallengebyseveralmeans.Threeapproacheshaveshown mediumtohighaccuracywhenusedindenseforests: longwavelengthSAR; measuringforestcanopyheightusingLIDARorinferometricSAR(InSAR)sensors;and fusionsofactiveandpassiveimagery.

FAGAN AND DEFRIES

iv

FuturesatellitelaunchesofLIDAR,longwavelengthSAR,andInSARsensorswillsignificantly improveestimatesofbiomass,forestvolume,andcarbonintheneartermandmayprovide informationcrucialtothedevelopmentofaglobal,groundlevelelevationmodel.Ifsuchamodel wereavailable,scientistscouldcreateaccurate,worldwidemapsofforestheightandinturn, generateglobalreferencemapsthatestimatehistoricalforestbiomassasfarbackasthemid 1990s.

Summary and Conclusions


Itispossibletoimproveglobalmeasurementsofforestarea,structure,biomass,andcarbon usingremotesensingtechnologiesthatarecurrentlyavailableorthatareexpectedtobeinuse between2009and2015(seetableatendofexecutivesummary).Wearecapableofgenerating highlyaccuratemeasurementsofforestareanow;technicaldevelopmentsoverthenextsixyears willincreasethefrequencyofmoderateresolutionforestimageryandimproveourabilityto analyzyethedatagenerated.Thesedevelopmentswillfacilitatediscriminationbetweenforest typesandthedetectionoftemporalchanges.Forestareameasurementscouldbeimprovedby refiningtheaccuracyofcoarseresolutionmapsanddevelopinganaccuratemoderateresolution, globalforestmapoverthenextseveralyears.Suchaneffortwouldimprovethequalityofworld forestmapstenfold;wecouldcreatemoderateresolutionworldforestmapsgoingbackto1975,as wehavealreadydonefortheU.S.andothercountries.Historicalmapswouldserveasbaselines fromwhichtomeasureandmonitorchangesinforestarea. Today,forestvolume,biomass,andcarbonstockscanonlybeestimatedconservativelyandthe accuracyisextremelyvariable(seetablebelow)dependingonthestudyandmethodsused althoughrecentimageryfusionsshowpromise.Integrationwithhighqualityforestinventoryand LIDARdataisessentialforimprovingbiomassestimatesfromsatellites.Worldwidecollectionof forestinventorydataisnecessarytoachieveacceptablyaccurateglobalestimatesofforestvolume, biomass,andcarbonfromcurrentsatelliteimagery.FutureSARandLIDARsatellitelaunches,such asDESDynI,BIOMASS,andSARconstellations,areexpectedtoachievegoodtohighaccuracyin mappingforestvolumeandbiomass.Howeveritwillbenecessarytovalidatetheseestimateswith extensivegroundtruthdataandsupplementthemwithhistoricalreferencemapsofbiomass.A coordinatedglobaleffortisneededtoachievetheseobjectives. Fordecades,theonlyinstitutionalefforttogenerateglobalforestmeasurementswasthe inventoryassembledbytheFoodandAgricultureOrganizationoftheUnitedNations(FAO). DespitethebesteffortsoftheFAO,theseestimatesarewidelyrecognizedasinaccurate.Nations selfreporttheirforestinventoriesanddifferinhowforestsaredefinedaswellasthetechnical means,funding,andprioritygiventothemeasurement.Tofillgapsindataorreconcileconflicting data,theFAOoftenmustdependuponuntestedassumptions,thechanceofdataaggregationerrors, andchangesinthedefinitionofwhatconstitutesaforest.Ifwearetounderstandforest ecosystemsandifwehopetounderstandtheplanetarycarbonbudgetandtherolethatforests playinmodulatingclimatewemustimproveourabilitytomeasureandmonitorforestvolume, biomass,carbon,andtrends.Remotesensinghasastrongtrackrecordinglobalforest measurementandplannedsensorlaunchesofferevengreaterpotential.Atechnicalapproachthat integratessatelliteandgroundtruthdataandaninstitutionalmeanstoimplementitwillenable accurate,globalmonitoringoftheworldsforestsforthefirsttimeinhistory.

FAGAN AND DEFRIES

ExpectedImprovementsinAccuracyinRemoteSensingofGlobalForestIdentity
Identity Sensors Current(2009) Capability >80%accuracy for forest/nonforest mapsat moderate resolution(~30 50mpixels) >90%accuracy for forest/nonforest maps.2 Limitations Expected(20092015) Capability >80%accuracyfor globalmapsof nonforestandseveral foresttypes,at moderateresolution(30 m). Aglobalsetofhigh resolutionimageswill begatheredby2015. Limitations

Area

Opticaland Radar, moderate resolution

Lackofdetailedglobal foresttypes.Current coarseresolutionmaps havemoredetail,but ~65%accuracy.1

Availabilityoffree, moderateresolution imagerydependsonafew keysatellites(e.g.LDCM).

Optical, high resolution

Manyimagesareneed tomaplargeareas3: imagesaredifficultto standardizefor analysis.

Difficultiesin standardizationarelikely topersist,limitingusefor globalmapping. Seeabove.

Volume

Optical, high resolution SAR (Radar)

4090%accuracy Seeabove.Accuracyis forforestvolume lowinclosedforests withtreecanopy estimates. overlap. 5095% accuracy;>80% iscommonfor forestvolume. 3080%accuracy forforestvolume estimates(from forestheight). 4597% accuracy;>80% accuracyis commonfor forestvolume. Limitedtolowbiomass forests;higherbiomass decreasesaccuracy. Lackofground elevationdata preventsglobalforest height/volume estimation. LIDARsamplingis spatiallylimited,data intensive,and expensive.

Seeabove.Regional equationscorrectingfor canopyoverlapmaybe developed. >80%accuracyindense, highbiomassforests.4

Accuratevolume estimatesrequirethe launchofafewkey satellites.5 Aglobalgroundelevation mapmaybedifficultto develop.

inSAR (Radar derived height) LIDAR (Laser derived height)

Theamount/diversityof inSARdatawill increase.4Processing innovationsmaycreate groundelevationmaps. Globalsamplingof forestandground heightwillcomefrom newsatellitesensors.7

Satellitesensorswillbe spatiallylimited;global LIDARcoveragerequires expensiveaerial platforms.

Biomass Samesensorsandaccuracyasvolume;estimatedthroughcorrelationwith"groundtruth"points.6 Carbon Sameasbiomass;estimatedthroughastandardconversionfrombiomass,withminorinaccuracy(+/8%max.)

Notes:1Foresttypesaregeneral(e.g.,deciduousforest)inglobalcoarseresolutionmaps(>200mpixels).Theirforest/nonforestaccuracy is7080%. 2Currently,foresttypemappingaccuracyforhighresolutionimageryissimilartothatofmoderateresolutionimagery. 3Averylargenumberofexpensiveimagesareneededtocreateaglobalmap(seeTable5).Globalimagecoveragedoesnotcurrentlyexist. 4AdvancesinsatellitetechnologyandimageprocessingwillallowingfusingofinSARandSARforsynergyinvolume/biomassestimation. 5IncludingDESDynIandBIOMASS.InnovationsinprocessingSARimageryfrommultiplesatellitesmayalsoimprovevolumeestimates. 6Accurate"groundtruth"pointsfromforestinventorydataarecriticaltoanyefforttomeasureforestsusingremotesensing(seeSection 5.1.5). 7GlobalLIDARsamplingofforestvolumewillallowsynergywithforesttypemapsforregionalforestvolumeestimation.

FAGAN AND DEFRIES

vi

Chapter 1. Forests, Their Measurement, and the Need for Global Remote Sensing
Sometimes,ifyoustandonthebottomrailofabridgeandleanovertowatchtheriverslipping slowlyawaybeneathyou,youwillsuddenlyknoweverythingthereistobeknown. 1.1 Current Needs for Forest Monitoring
Forestecosystemsarefundamentallyimportanttohumanity.Theymoderatetheclimate, protectstreamsandsoil,generateoxygen,supplywoodandotherproducts,providebeautyand recreation,andhouseatreasuretroveofbiodiversity.Wedependonforestsfortheirecosystem servicesandhavedonesosincethedawnofcivilization.Itissurprising,then,thatweknowso littleabouttheglobalstatusofforests.WecanmakemorecurrentandinformativemapsofMars thanoftheAmazonbasinortheRussianborealforest.Thereareseveralreasonsforthegapsinour knowledge: adearth(untilrelativelyrecently)oftoolstoobservetheworldonaglobalscale; ahistoricallackofinterestinforestsbythepoliticalandeconomiccommunities(withthe exceptionoftimbercompanies); poorandinconsistentrecordsofglobalforestcover;and anabsenceofaconcertedefforttosystematicallyinventoryandmonitortheworldsforests.

WinniethePooh

Maturingobservationtechnologiesandintensepublicinterestinprotectingandmanaging forestsmakethisthetimetoexploreandbetterunderstandtheworldsforests. Inthisreportweevaluatecurrentandupcoming(20092015)technologiesformeasuringand monitoringglobalforests.Wefocusprimarilyonremotesensing(whichwedefineastheanalysisof satelliteandaerialimages)becausethistechnologymeetsthesteeplogisticalchallengeof measuringtheworldsforestsinanaccurate,repeatable,andinexpensivemanner.Remotesensing extendsexistinggrounddatatoexpandourunderstandingofforestsbeyondgeographically localizedareas.Remotesensingisarapidlydevelopingfielddrivenbytechnologicaladvancements indatagatheringandprocessingandhasalreadyyieldedmanyimportantdiscoveriesinrecent decadesaboutthechangingEarth. Manycriticalquestionsabouttheworldsforestsremainunderstudiedbyforestscientists.We focusontheobservationsneededtoprovideaccurate,basicmeasuresofforestattributestoinform forestscientists,decisionmakers,andthepublic.Wealsoillustratehowtheresultingdatawould addresscompellingquestionsthatpersistaboutourforests.

Whichcountriesaregainingandlosingforestsandwhy?
Webeginwiththissimplequestionasitisoftenattheheartofpublicunderstandingofthe statusoftheworldsforests.Forinstance,lossofrainforestintropicalregionsisofteninthenews.

FAGAN AND DEFRIES


Yetatpresent,nationalstatisticstrackonlynetchangesinforestcoverandhideforestlossin countriesundergoingreforestationandforestregrowth(Grainger2008).Inaddition,country measuredstatisticsonforestattributeshaveoftenbeeninconsistentandunreliable.Asaresult, longtermchangesaredifficulttotrack(Waggoner2009).Itisclearthataforesttransition(an increaseinforestcoverafteralongperiodofdeforestation)occurredintheUnitedStatesand Europeinrecentcenturies(Rudeletal.2005).Morerecentforesttransitionsintemperateand tropicalcountriesaredisputedbecauseofuncertainforeststatistics(Grainger2008;Rudeletal. 2005).Improvedandcontinuedglobalforestmonitoringandcarefulanalysisofarchivalremote sensingimagerywouldallowmoreaccurateforestmeasures. Howareforestloss,forestregrowth,andthereplacementofforestsbytreeplantationsaffectingforest biodiversity? Remotesensingmeasurementsofforestareaindicateadeclineintropicalforestareafromat leastthe1980sonwardandforestregrowthhascompensatedforonlysomeoftheforestloss (Mayauxetal.2005;DeFriesetal.2002;Hansenetal.2008).Characterizingtheresponseoftropical biodiversitytodeforestationandregrowthhasbeendifficultbecausewelackextensivedataonthe movementofforestspeciesacrossfragmentedlandscapesandontheconservationvalueofhuman modifiedhabitats(Chazdonetal.2009;Gardneretal.2008).Eventheareaoftropicalforest regrowthtodayisdebated(Grainger2008;Asneretal.inpress)becauseofdifficultiesin distinguishingregrowthandtreeplantations(SanchezAzofeifaetal.2009).Inothertypesof forests,evidencesuggeststhatspeciesrichmature(oldgrowth)temperateforestsareindecline, butthereislimitedglobalmeasurementoftherateofthisdeclineorofloggingdisturbancein matureforests(FAO2006;GOFCGOLD2008).Improvedandsystematicallycollected,well calibratedobservationsaboutforestswillenhanceourabilitytomonitorandunderstandthese changes.Newremotesensinganalysesofhumanmodifiedlandscapescouldsetconservation prioritiesandimprovepredictionsofspeciespersistenceinmodifiedlandscapes. Howwillchangesinforestcoveraffectclimatepatternsandhowwillforestsrespondtoclimate change? Theideathatdeforestationdecreasesrainfallisanoldone(Marsh1878),butitwasdifficultto scientificallytesttheideauntiltheadventofclimatecirculationmodelsandregionalforestmaps (e.g.,Malhietal.2009;RamosdaSilvaetal.2008).InBrazil,modelspredictthatdeforestationof approximately40percentoftheAmazonrainforestwouldresultinalargedeclineinrainfall throughoutthebasinandhavenotableconsequencesforagricultureintheeasternportion(Ramos daSilvaetal.2008).InCanada,increasesinborealforestcovermayactuallyincreaseregional warming(Balaetal.2007).Evaluatingtheeffectsofdifferentpatternsofdeforestationon precipitationrequiresdetailed,updatedforestmapsinareasthatoftenhavehighcloudcoverand rapidlandusechange. Forestswillbeakeyindicatorofchangeasclimatechangeadvances.Forestswillcontinueto undergochangesinphenology(thetimingofleafoutandleaffall),productivity,andflammability (Goetzetal.2005;Phillipsetal.2009).Coupledclimatecarbonmodelspredictamarkedincrease inrespirationintropicalforestswithwarminganddrying,whichmaysetinplaceapositive feedback(Tianetal.1998;Fieldetal.2007).Inthecaseofborealandtemperateforests,climate changeisexpectedtodrivesignificantrangeshiftsandforestexpansioninhighlatitudes(Loehle 2000;MacDonaldetal.2008).Coordinatedmonitoringofforestswouldcreateavaluablerecordof

FAGAN AND DEFRIES


theeffectofclimatechangeonnaturalsystemsandprovideawarningindicatorofsuddenshifts, suchaschangesinleafwatercontentbeforeforestfires(Chuvieco2008). Howdoforestscontributetotheglobalcarbonbudgetandthemissingsinkandhowcanforest carbonbemeasuredandvaluedtomitigateclimatechange? Terrestrialecosystemsabsorbcarbondioxidethroughvegetationgrowthandemitcarbon dioxidethroughmetabolismanddecay(respiration)andlandcoverchange(e.g.,deforestation). Theamountofcarbonstoredinvegetation(itscarbonstock)isroughlyequivalenttotheamount storedintheatmosphere,andabout87percentofthecarboninabovegroundvegetationisstored inforests(Houghton2007;LeToanetal.2008).Strongevidencesuggeststhattheterrestrial ecosystemhasfunctionedasacarbonsinkforthelast25years,takingupalmostathirdof anthropogenicemissions(Canadelletal.2007).However,theexactsizeandcauseofthismissing sinkisnotwellknownbecausetheuncertaintyincarbonsinkandemissionsestimatesisveryhigh (Canadelletal.2007;LeToanetal.2008).Althoughevidencesuggestsalargecarbonsinkinthe regrowingforestsintheNorthernHemisphere,sinksinthetropicsmayalsobeimportant(Fanet al.1998;Houseetal.2003).Recentevidencesuggeststhatthisterrestrialbrakingonaccelerating humanfossilfuelemissionsmaybedecreasing(Canadelletal.2007),butwithoutsoundestimates ofitsoriginalsize,itisdifficulttoprojectwhatimpactlosingthemissingsinkwillhaveonglobal climatechange(LeToanetal.2008). Arangeof7percentto30percentappearstocharacterizetheroleofforestdestructionin anthropogenicemissions(Canadelletal.2007;IPCC2007).Thereisintenseinterestinquantifying forestcarbonforaninternationaltradingframeworkonreducedemissions(HeroldandJohns 2007;Olanderetal.2008;Gibbsetal.2007).Extrapolatingfromgroundbasedplots underestimatesspatialvariationinforestcarbon(Houghtonetal.2001;Houghton2005),and grounddatamustbesupplementedbyaccurate,realtimemapsofforestareaandbiomass (Waggoner2009;Gibbsetal.2007). Whatistheimpactofforestcoveronstreamflowsanderosionprotection? Althoughitiscommonlyacceptedthatforestsplayasignificantroleinprotectingwatersheds, thereisrelativelylittleevidencerelatingforestcovertoflooding(Laurence2007).Forestmapping andmonitoringiscriticaltorelatinglandusechangetoongoingmeasurementsofwaterquantity andquality.Activeareasofresearchincludetheeffectivewidthandcontinuityofriparianbuffers andthebestconfigurationofregionalforestcovertominimizesoilerosion(Allan2004).To improveourunderstandingofhydrology,erosion,andforests,weneedtoobtainmoredetailed measuresofforestattributesandwemustmonitorforestsinasystematicwaytodetectchanges. Howcanrapidfiredetectionandotherforestobservationsbysatelliteimproveforestmanagement andcertification? Satellitetechnologyiscapableofrevolutionizingthemanagementofforestareas.Realtimefire detectionsystemscouldhelpcombatwildfireandillegalclearinginremoteforests,andhigh resolutionimagerycouldimprovemonitoringandverificationofforestmanagementfor sustainableharvestandforcarbonsequestration(Daviesetal.2009;SouzaandRoberts2005).

FAGAN AND DEFRIES

1.2 A Brief History of Global Forest Measurement and Remote Sensing


Terrestrialecosystemsarethesinglegreatestsourceofuncertaintyintheglobalcarbonbudget (IPCC2007).Thetotalareaoftheglobalforest,itstrendinrecentyears,theamountofbiomassand carbonlockedupinourforestsalloftheseestimates,whichwouldseembasicscientific knowledge,arepoorlyknown(Grainger2008;Houghton2005). Thereisstrongevidencethatforestsoncecoveredmanymodernagriculturalandarid landscapes,butregularmonitoringofglobalforestareadidnotbeginuntilthe20thcentury (Williams2008).TheUnitedNationsFoodandAgricultureOrganization(FAO)startedreporting informationaboutglobalforestsin1948,publishinganinventoryeveryfiveyearsuntil1963when theeffortwashaltedbecauseofpoortropicalforestdata(Grainger2008).TheFAOpublishedtwo assessmentsofglobalforestresourcesinthe1970s(Mayauxetal.2005),thenresumedregular reportingin1981.TheorganizationpublishedForestResourceAssessments(FRA)for1980,1990, 2000,and2005(Grainger2008).FAOreportshavebeenwidelycriticizedforinaccuracy(for example,seeWaggoner2009);errorsinestimatingforestareaareattributedtochangesin definitionsofforests,revisionsofestimatesbasedonconflictingdata,unreliablenationalinventory estimates,anddataaggregationerrors(Grainger2008;Houghton2005).HistoricalFAOestimates offorestareaaredemonstrablyassumptiondependentanddonotexistforsomecountries (Grainger2008;Houghton2005).AsWaggoner(2009)pointsout,itisunclearfromFAOstatistics whetherglobalforestareaisdecliningorgrowing. Remotesensingofforestsbeganin1972withthelaunchofLandsat,thefirstinaseriesofEarth observationssatellitesintheLandsatprogram(DeFries2008).Thefirstcontinentalscalemapsof landcoverwereproducedbyTuckeretal.(1985)andTownshendetal.(1987)fromfourkilometer (km)resolutionimageryfromtheAdvancedVeryHighResolutionRadiometer(AVHRR)satellite, followedbythefirstgloballandcovermapfromDeFriesandTownshend(1994)atonedegree resolutionusingAHVRRimagery.Lovelandetal.(1999)produceda1.1kmresolutiongloballand covermapfromAVHRRsatelliteimagery,followedbyseveralgloballandcovermapsat0.51km resolution(reviewedinMayauxetal.2005andAchardetal.2007).Thehighestresolutionglobal landcovermaptodatewasproducedin2007usingdatafromEuropesMediumResolution ImagingSpectrometer(MERIS)ontheEnvisatspacecraft(Bicheronetal.2008).Thatmaphasa resolutionof300meters(m). Inadditiontotheseglobalmappingefforts,numeroussatelliteimagerystudieshavemonitored globalforestarea,especiallyintropicalareas(e.g.,DeFriesetal.2002;Hansenetal.2008a;Achard etal.2007;Mayauxetal.2005).AllsatellitebasedestimateshavesupportedtheFAOsassertion thatnettropicalforestareahasbeendecliningforthelastthreedecades(Mayauxetal.2005; DeFriesetal.2002;Hansenetal.2008a).Forthe2000FRAreports,theFAOsampled10percentof theglobalforestusingsatelliteimagery,butthesedatawereinsufficientfornationalestimationin manycountries(TuckerandTownshend2000).Forthe2010FRA,thesatelliteobservationeffort hasbeenexpandedgloballytosampleforestcoveratlatitudinalgridintersections(Mayauxetal. 2005).Becausetropicaldeforestationisspatiallyconcentratednearroadsandagricultural frontiers,thisanalysiswillonlycapturenationaldeforestationtrendsinlargercountries(Achardet al.2007;TuckerandTownshend2000).

FAGAN AND DEFRIES

1.3 Diversity of Forest Types


Theboreal,temperate,andtropicalforestsareestimatedtocontainabouthalfoftheterrestrial carbonwiththemajorityofforestcarbonstoredinforestsoils.Theamountofcarbonstoredin forestsoilsvariesbyregionbecauseoftemperature(seeTable1;Malhietal.1999).Inboreal forests,abovegroundforestcarbonisthetipofanicebergofsoilcarbon;lowsoiltemperaturesand forestshadeslowdecayandthereleaseofsoilcarbon.Theopenborealforest,whichiscomposedof conifersandafewspeciesofdeciduoustrees,stretchesacrossthenorthernhighlatitudes(Malhiet al.1999).Clouds,longnorthernwinters,andshortgrowingseasonsmakesometypesofsatellite observationofhigherlatitudesdifficult(KasischkeandFrench1997). Table1.ComparisonofEstimatedAreaandCarbonStocksofForestsatDifferentLatitudes
Latitude(foresttype) Area(Mha) CDensity: vegetation(t/ha) 65 57 121 CDensity: soils(t/ha) 343 96 123 CStock: vegetation(Gt) 88 59 212 359 CStock: soils(Gt) 471 100 216 787

High(borealforest) Middle(temperateforest) Low(tropicalforest) Sumtotal


Source:Dixonetal.1994.

1372 1038 1755 4165

Thetemperateforests,withshorterwinters,encompassdeciduousforests,coniferdominated forests,andforestsofbroadleafevergreensindrierareas(Malhietal.1999).Widespreadhistorical clearingoftemperateforestsemittedsignificantamountsofcarbondioxideintotheairfrom biomassburninganddisturbancetocarbonrichforestsoils(Houghton2007).Thetermsecondary forestdescribesforeststhatareregrowingafteradisturbance,suchasloggingorfire.Muchofthe recentcarbonuptakeinnorthernlatitudesisbelievedtoresideinsecondaryforests,whichare growinglargerandstoringsoilcarbon(Houghton2005;Fanetal.1998).Satelliteobservations havedifficultydetectingsmallchangesingrowingforests,intreeheightanddiameter(Houghton 2005). HighlybiodiversetropicalforestscoveralargebandaroundtheEarthsrainyequatorandcan beclassifiedbyelevation(montaneforests),tolerance(floodedforests,mangroves),andthelength ofthedryseason,whichcanbenonexistent(evergreentropicalrainforest)orseveralmonthslong (deciduoustropicaldryforest).Themostcommonforesttype,tropicalrainforest,ischaracterized bytalltrees(>30m),densecanopies,andveryhighbiodiversity(Malhietal.1999).Tropicalforests havebeenrapidlyclearedinrecentdecadesandabouthalfoftheircarbonisstoredintheirliving mass,whichisemittedwhentheyarecleared(Malhietal.1999;DeFriesetal.2002).Tropical forestsevaporate(ortranspire)immenseamountsofwaterfromtheirleaves.Anestimated50 percentoftherainintheAmazonbasinisfromtranspiration(SalatiandVose1984).Theintense, dailycloudinessoftropicalforestsblockstheviewofmanytypesofsatellitesanddecreasesthe frequencyofclear,cloudfreeimages(Asner2001;Olanderetal.2008).

FAGAN AND DEFRIES

1.4 For Global Forest Measurement, Why Use Satellite Imagery?


Tomeasureforestsglobally,satelliteimageryisapracticalnecessity.Bothaerialandground observationsareexpensiveatpresentandonlycoversmallareasatatime;groundmeasurements aresimplyimpossibleoversuchalargelandarea(Houghton2005;Patenaudeetal.2005). Detectingrapidchangesintheareaofremotetropicalforests(bothdeforestationandregrowth) requiressatellitemapping(Achardetal.2007),aswilldetectingthenorthernexpansionofboreal forestsinawarmingworld(MacDonaldetal.2008).Thegreateststrengthsofsatellitebased measurementsaretheirunparalleled,unbiasedmeasurements,theirmonthlytodailyfrequency, andabovealltheirsynopticnature.Satellitesprovideageneralviewofthewholeearththatis unavailabletoanyotherforestmeasurementmethod.Forexample,satelliteimageryiswellsuited fordetectingcountryleveldeforestationinaREDD(reducingemissionsfromdeforestationand forestdegradation)agreement(Olanderetal.2008),butalsodetectsforestinvasionsofrangeland, forexample,orforestregrowthaftertheabandonmentoffarmland(Houghton2005). Satellitescanmeasureforestquantities,likecanopychemistryordailyleafphenology,thatare arduousorimpossibletocollectwithgroundcrews(Ustinetal.2004;DeFries2008).Although handcollectedforestdataareoftenmoreaccuratethansatellitemeasurementsatthepointand timethedataareobtained,satellitescollectdataacrossbroadareas,samplingthefullrangeof variationinforestmetricsandcapturingbroadtrendsanddynamicchangeintheworldsforests (Houghton2005).Assuch,satellitedataallowforintegrationacrossgroundmeasurements, extendingthemtotheglobalforest.Putsimply,satelliteimagerymeasuresforestsoncontinental scales,detectschangesinforeststhatwedonotexpectorcouldnotmeasure,anddetectsthemin realtime.

1.5 Scope and Purpose


Inthisreport,weevaluatethecurrentandfuturetechnicalcapacityofsatelliteimageryto measureandmonitorglobalforests.Otherresearchershaveprovidedextensivereviewsofcurrent capacity(Anderssonetal.2009;SanchezAzofeifaetal.2009;Heroldetal.2008;Olanderetal. 2008;Achardetal.2007;DeFriesetal.2007;Patenaudeetal.2006;Mayauxetal.2005;Rosenqvist etal.2003)andwesummarizetheirfindingshere.Wealsopredictfuturecapacityfromexpected satellitelaunchesandtechnologicaldevelopmentsduring20092015.Thistimewindoweffectively capturesindevelopmentsatellitetechnology(althoughsatellitesrarelylaunchontime)andallows anindepthlookathowwellwewillbeabletomeasuretheworldsforestsinthecomingyears. Thewidespreaduncertaintiesincurrentforestmeasurementmakeimprovementsinglobal forestmonitoringascientificimperative.Thisreportfocusesonobservationsofforestarea, biomass,andcarbon,addressingthecapacityofsatelliteimagerytoanswerquestionsposedabove (Section1.1).Therearemanycharacteristicsofforeststhatarenotwellestimatedgloballyandare worthmonitoring(Chuvieco2008;Mynenietal.2007;Gardneretal.2007;Phillipsetal.2009), includingseveralthatareimportantforansweringindividualquestionsabove(e.g.,fragmentation andforestbiodiversity).Forestarea,biomass,andcarbonarebasicforestmeasurementsthatare fundamentaltoansweringscientificquestionsaboutforests(Kauppietal.2006).Assuch,they representacriticalfirststepinglobalforestmeasurement.

FAGAN AND DEFRIES

Chapter 2. Remote Sensing and Forest Measurement: an Overview


Thischapterisintendedasabriefintroductiontoremotesensingandanoverviewofcurrent andemergingsensortechnologiesthosealreadyfamiliarwiththetopicshouldbeginwithsection 2.4.Remotesensingisgenerallydefinedasobservingandanalyzingadistanttargetusing informationfromtheelectromagneticspectrum(Jensen2007).Remotesensingforforest measurementismainlyconductedfromairborneandsatellitemissions(Jensen2007;CEOS2009; Patenaudeetal.2005;Olanderetal.2008).Weuseremotesensinginthisrestrictedsenseas observationsobtainedfromaircraftandsatellites.Inremotesensingterminology,thehardware thatobtainstheobservationsisreferredtoasaninstrument,orasensor,andismountedonaplane orsatellite.RemotesensingproducesimagesoftheEarthssurfaceforanalysisandinterpretation. Analysismayrangefromqualitativeobservationsfromtrainedimageinterpreterstoquantitative measurementsfromcomplicatedcomputeralgorithms.Byanalyzingmultipleimagesofthesame location,remotesensingdetectslandcoverchange(modificationsinthebiophysicalcoverofthe earth)overtime.

2.1 Comparing Satellite and Airborne Remote Sensing


Satelliteandairbornemountedsensorsdifferdramaticallyintheiraltitude,imagecoverage, costs,andspatialresolution.Spatialresolutionisdefinedastheeartharearepresentedinonepixel ofaremotesensingimage(seeFigure1foracomparisonofdifferentresolutionsensors).Airborne sensorsfunctionatamuchloweraltitudethansatellitesensorsandasaresulthaveamuchsmaller imagecoverage,orswath,thansatellitemountedsensors.Forexample,aeriallymountedsensors haveaswathof511km,whereasthecoverageofsatellitemountedsensorsis153,000km (Appendix).Airbornesensorsarealsolimitedbyfuelcostandrange,crewlogistics,weather,and theneedfornationalpermissiontoflyandtakeimages.Satellitesorbitrepeatedlyaroundthe entireglobeandtheirsensorsimageregardlessofweather.Becauseofprovisionsinthe1967U.N. OuterSpaceTreaty(UNOOSA2009),itisnotnecessarytoobtainpermissionbeforetakingsatellite imagesofanycountry.Bycontrast,overflightpermissionsarerequiredforaerialsensors.Asthere arenoexpensesforcrewandfuel,satelliteshavemuchloweroperationalcostsandcanimage muchlargerareasforafractionofthepriceofairbornemissions(Patenaudeetal.2005;seeTable2 foracostcomparison). Airbornemissionsdohavesomeadvantagesoversatellitemissions.Airbornemissionsare muchcheapertolaunch,becausethesensoristhemainexpenseindevelopingthemission.Satellite missions,bycontrast,costhundredsofmillionsofdollarstoconstructandlaunchintoorbitand unsuccessfullaunchesdooccur(Table2).Inaddition,airbornemissionsoftenprovidehigher qualitydatathansatellitemissions.Becauseasensormountedonaplaneisclosertotheearth,the identicalsensorwillhavehigherresolutionthanifitweremountedonasatellite.Airborne missionsarealsomoreflexibleintheirflightpathandinstrumentationthansatellites.Satellites mustfollowafixedorbitalpath(seeFigure2a),butplanescantimetheirmissionstoavoidcloud cover.Updatinginstrumentssay,tousenewtechnologyisalsoeasierforairbornemissionsthan onorbitingsatellites.Asaresult,planesoftencarrynewersensortechnologythansatellitesdo. Interesthasbeengrowinginunmannedaerialsystems(UAS)foractivesensors(seenextsection);

FAGAN AND DEFRIES


unmannedsystemshavelongerflightdurationpermissionandcanflyatnightbutaremorecostly topurchaseandoperatethanequivalentsatelliteimagery(Swansonetal.2009). Figure1.AnImageTakenwithDifferentSpatialResolutionSensors

Source:Short2009.

Insummary,airbornemissionsarerelativelyexpensivetoconductanddonotcoverabroad areabutareflexibleandcanprovidehighqualitydata(Olanderetal.2008;Anderssonetal.2009). Satellitescanmonitorabroadarealessexpensivelyandthusareviewedasthebestwaytomeasure globalforestsremotely(Olanderetal.2008;Anderssonetal.2009;DeFries2008;SanchezAzofeifa etal.2009).Withapproximatelyfourbillionhectaresofforestarea,anyglobalefforttomeasurethe worldsforestswillhavetorelyheavilyonsatellitebasedremotesensing,withairbornedataused asasupplementinareaswheresatellitedataarelackinginquality,coverage,oraccuracy.

FAGAN AND DEFRIES


Table2.CostComparisonofAirborne,Ground,andSatelliteDataforForestInventoryovertheUK
Monitoring fromremote sensing Forest area (km2) Ground surveyb Opticalc ApproximateCostsa Radar LiDAR

Satellite

Airborne

Satelliteor Shuttlec ~0.10/km2

Airborne ~300/km2 ~8,400,000

~0.10/km2 ~250/km2 AGCClossfrom Article3.3D activities 28,000 TotalUK forestarea plusARsince 1990 AllUKforest areaplanted after December 1989 4,500 ~3,500,000 Lessthan ~3,000

~7,000,000 Lessthan ~3,000

AGCCgains fromArticle3.3 ARactivities (forestgrowth only)

~307,000

Lessthan ~1,000

~615,000

Lessthan ~1,000

~738,000

AllUKforest AGCCgains fromArticle3.4 areaplanted after1940e FMactivities (forestgrowth only)d

19,245

~1,822,000 Lessthan ~2,000

~3,644,000 Lessthan ~2,000

~4,373,000

FAGAN AND DEFRIES


Figure2a.PointableSatellitesCanImageManyTargetsoffTheirOrbitbyTiltingTheirSensor

(c)2008CNES.DistributionSPOTImageCorp.,USA.Allrightsreserved. Note:ThePleiadessatellite(20kmswath)isshownhere.

Figure2b.SatelliteConstellationovertheEarth

Source:ImagecourtesyofNASA.Credit:EdHanka.

FAGAN AND DEFRIES

10

2.2 Passive and Active Remote Sensing: An Overview


Remotesensinginstrumentscomeintwomainvarieties:passiveandactive.Passivesensors, likecameras,receivereflectedandemittedelectromagneticenergyfromtheearth,andinclude opticalandthermalsensors(Jensen2007;seeFigure3.Activesensorsaimelectromagnetic radiation,suchaslaserlightormicrowaves,towardtheearthandreadthereflectedradiation, operatingmuchlikealaserrangefinderorsonar(Jensen2007,seeFigure4). Figure3.ComparisonofSpectrumsofVegetation,BareSoil,Snow,andWater

Note:Notethestrongincreaseinreflectanceofnearinfraredradiationbythepinesandgrasses:thisisdiagnosticof vegetation. Source:Short2009.

Figure4.PolarizationofSARWaves(HVexample)

Source:NaturalResourcesCanada2005.

FAGAN AND DEFRIES

11

2.2.1PassiveRemoteSensors:ConsiderationsandLimitations
Passivesensorscancaptureawiderangeoftheelectromagneticspectrumthatcomesfromthe earth,fromreflectedvisiblelight(0.40.7micrometers,abbreviatedasm)toreflectednear infraredradiation(0.71.3m),reflectedshortwaveinfrared(1.33m),andemittedthermal infrared(314m).Theelectromagneticradiationthatsensorsreceivehasalreadyinteractedwith theearthssurface(primarilythroughchemicalabsorptionandphysicalscattering),soitcontains informationaboutchemicalandphysicalpropertiesofthesurface(Figure3;Jensen2007).For example,passivesensorscandistinguishvegetation,baresoil,snow,andwaterquiteeasily,all basedontheirrelativereflectionacrosstheelectromagneticspectrum(seeFigure3forapartial comparison).Asaresult,passivesensorsexcelatdetectingchangesinlandcover(Hansenetal. 2008a;CohenandGoward2004;Asneretal.2005),vegetationphenology(Mynenietal.2007; Soudanietal.2008),anddifferencesinecosystemtype(Pateneudeetal.2005). Currentpassivesensorsarenotuniformlysensitivetoallregionsofthespectrum;theyreceive moststronglyincertainbandsofthespectrum,whichmaybenarrow(e.g.,0.4150.42m)or broad(e.g.,0.40.7m).Forexample,abroadbandsensorthatreceivedallvisiblelight(0.40.7 m)equallywouldimagetheearthinblackandwhite.Bycontrast,theLandsatmultispectral sensorhassixbandsofintermediatewidththatarespreadfromthevisible(blue,green,red) throughnearandthermalinfraredwavelengths(seeFigure5foracomparisonofthebandsof differentsatellites).Hyperspectralsensors,arelativelynewtechnology,haveveryhighspectral resolutionandmeasurehundredsofnarrowbandsthatarequitesensitivetochemicalabsorption signatures(Figure5). Figure5.BandwidthofMultispectralversusHyperspectralSatellites

Note:Themultispectralbandsindicatedhereareonlyshownascenterpoints,buttheyareactuallyquitewide.Hyperspectral sensorshavetenstohundredsofnarrowbandsandareabletoproduceacontinuousspectrum. Source:NASA2009b.GraphbyRobertSimmon.

FAGAN AND DEFRIES

12


Passiveopticalsensorsvarydramaticallyintheirspatialresolutionaswellastheirspectral resolution.Thesesensorsareclassedashighresolution(imageswithpixels0.410minwidth), moderateresolution(10200m),orcoarseresolution(>200m)(Figure1,Appendix).Becauseof technologicaltradeoffsthatareunlikelytochangeinthenearfuture,thespatialresolutionof passivesensorsisinverselyrelatedtothedimensionoftheswaththeycancapture(Anderssonet al.2009;Rosenqvistetal.2003;CEOS2009;seeFigure6.Atoneextremearethecommercialhigh resolutionsatelliteslikeGeoEye1,whichhasamaximumresolutionof0.41mandanimagewidth of15.2km.AttheotherextremearethecoarseresolutionsensorslikeAVHRR,whichhasa resolutionof1.1kmandanimagewidthof3000km.Somegeostationarysatellitescanimagethe entireearthdiscat1kmresolution.(Geostationarysatellitesorbitatahighaltitudeabovethe equatorandsynchronizedwithearthsrotation,allowingthemtooperateasiffixedoveronepoint ontheearth.)Moderateresolutionsensorssplitthedifferencebetweenthetwoextremes.Landsat 5,forexample,hasaresolutionof30mandanimagewidthof185km.ObservingtheentireEarth usingahighresolutionsensorislikeusingatelescopetoobservetheentirelandscapebelowa promontorydetailishighbutittakesalongtimetoseeeverything.Inthesameanalogy,using coarseresolutionsensorstoobservetheearthislikeviewingalandscapethroughbinoculars. Figure6.SpatialResolution,Cost,andSwathforCurrentOpticalSatelliteSensors

Notes:AlldataisfromtheAppendix.Duetochangingprices,costissimplynotedasfreeornotfree(cost).Bothaxesare logarithmicfordisplaypurposes.Theredlinehighlightsgoodperformanceamongfreesensors.Higherresolutionsensorsface steeptradeoffsinsmallerimageswath.

Forsatellites,thistradeoffinspatialresolutionandimagesizealsoextends,withexceptions,to thetimeittakesforasatellitetoimagethesamelocationonEarth(thereturntime).Polarorbiting satelliteshaveanorbitthatcrossesboththeEarthspolesallowingthespacecrafttoimagethe earthinsuccessivenorthorientedstripsastheplanetturnsunderneaththecraftsorbit(Figure 7a).Byorbitingatdifferentspeedsandangles,satellitesgenerallyspacetheirobservationsto matchthewidthoftheirimagingwindow(Figure7b).Coarseresolutionsatellites,therefore,can

FAGAN AND DEFRIES

13


imagetheentireEarthquicklyandrevisitthesamelocationeveryfewdays.Moremoderate resolutionsatellitescantake10to20daystorevisitalocation(seeAppendix). Figure7a.ExampleofaFixedPolarOrbitLandsatsatellite

Source:Short2009.

Figure7b.SatelliteOrbitsareSpacedtoFittheImageSwath

Source:Short2009.

Theexceptionstotheimagesizereturntimetradeoffincludeunusualorbits,pointable satellites,andsatelliteconstellations.Somesatelliteshaveunusualorbitsthatallowfrequent revisitsofcertainlocations;forexample,geostationarysatellitesobservethesamelocationinreal time.Pointablesatellitescanimagethesamespotfromseveralorbitsawaybyturningontheir sides(Figure2a).Satelliteconstellationsaregroupsofsatellitesthat,whenentrainedinthesame orbit,flyoveralocationwithgreatfrequency(Figure2b).

FAGAN AND DEFRIES

14


Returntimeisveryimportanttopassivesensors,becausecloudsinterferewiththereflectionof radiation.Theonlywayforpassive,satellitemountedsensorstoacquireacloudfreeimageisto returnfrequentlyenoughtocaptureallofthelandscapewithoutclouds.Thisisespeciallytruein tropicalareas,whicharecoveredbycloudsformostoftheyear(Asner2001;Olanderetal.2008).

2.2.2ActiveRemoteSensors:ConsiderationsandLimitations
ActivesensorsemitelectromagneticradiationtowardtheEarth,typicallymicrowaves(known assyntheticapertureradar,orSAR)orlasers(lightdetectionandranging,knownasLIDAR),and thenreceivethereflectionoftheiremittedenergy.Unlikepassivesensors,activesensorsareableto penetrateclouds(althoughLIDARiscloudsensitive)andimageduringbothdayandnight(before dailyconvectivecloudbuilding).Activesensorsthushaveamajoradvantageintropicalareas (Rosenqvistetal.2003;Olanderetal.2008).Ingeneral,activesensorsexplicitlymeasurethethree dimensionalshapeandtextureofecosystemsbuthavedifficultywithtopography.Inaddition, imagesfromactivesensorsaretypicallymorechallengingtoanalyzethanimagesfromoptical images(Kasischkeetal.1997).Orbitingactivesensorsareincreasinginnumberandcomplexity andmostexpertsexpectthatanalysisofSARimages,inparticular,willbecomeeasierandmore commoninthecomingdecade(Rosenqvistetal.2003;CEOS2009).

SyntheticApertureRadar(SAR)
SARsystemsemitmicrowavesandmeasurethestrengthofsignalreturn,alsoknownas backscatter.ThestrengthoftheSARbackscatterdependsuponhowitinteractswiththesurface vegetationandthetopographyofthelandscape.Inforestedareas,mostoftheSARsignalreflectsoff watercontainingbiomass(Kasischkeetal.1997)beforereachingtheground.Unliketwo dimensionalpassivesensors,inwhichthereflectionfromabaremountainappearsthesameasthe reflectionfromaflatdesert,theemittedenergyofactivesensorsrespondstothethreedimensional shapeoflandformsandecosystems(Rosenqvistetal.2003;Figure8.Inmountainousareas,SAR signalreflectionisdominatedbyinteractionswithtopographyandthevegetationsignalisharder toanalyzeduetoblackshadowsandareasofstrong,falsereturnsfromtopography(Jensen2007; Rosenqvistetal.2003;Figure8). ThereareseveraladjacentbandsofSARmicrowaveradiationinuse,andthereislargevariation incenterwavelengthbetweentheXband(~3centimeters,orcm),Cband(~5.6cm),Lband(~25 cm),Pband(~65cm),andVHF(115m)microwaveradars(Kasischkeetal.1997;Patenaudeetal. 2005).ThesedifferencesinwavelengthaffecthowtheSARradiationinteractswithsoilmoisture andthewatercontentofvegetation.Shorterwavelengths(e.g.,CandXband)tendtobemore stronglyreflectedbysmallerstructuresliketwigsandthinbranchesatthetopofaforestcanopy,so themajorityoftheshorterwavelengthbackscatterdoesnotpenetrateveryfarintoaforest.Longer wavelengths(e.g.,L,P,andVHFband)tendtobemorestronglyreflectedbylargerstructureslike trunks.ThestrengthofSARbackscatterfromvegetatedareasisoftencorrelatedwiththeamountof woodybiomassinthatarea(Dobsonetal.1995b;Kasischkeetal.1997).However,SARbandsX throughPalltendtoachievemaximumsignalreturn(saturate)atincreasinglevelsofbiomass (Figure9).TheCandXbandssaturateatlowerthresholdsofwoodybiomassbecausemostofthe shorterwavelengthSARradiationisreflectedfromthetopofdensecanopies.

FAGAN AND DEFRIES

15


AdvancedSARsensorscanemitandreceiveradiationwithdefinedpolarizations.Although naturalmicrowavesundulateinalldirectionsastheypropagate,itispossibleforSARsystemsto emitandreceivemicrowavesthatonlypropagateinverticalorhorizontalangles(Figure4).This increasestheabilityofSARsensorstomeasurethesurfaceroughnessandstructure,ortexture,of thelandsurface(SARtextureismorestrictlydefinedasthevariabilityinbackscatterreturnovera definedarea).Forexample,openfieldsmayreflectmorehorizontallypolarizedmicrowavesthanan irregularforestcanopy(Jensen2007).NewerSARsensorscanhaveveryhighspatialresolution (e.g.,1m),butitisoftennecessarytodegraderesolutiontoremoverandomspeckleerrorsby averagingacrossfourormorepixels.Practically,thismeansthatSARresolutionwillalwaysbe betweenthreeandfourtimeslowerthanspecified(Rosenqvistetal.2003).

Figure8.DifferencebetweenPassive(top)andActive(bottom)SensorSensitivity toTopographyonaMountainIsland

Note:NotetheshadowsandresponsetotopographyintheblackandwhiteSARimage. Source:Short2009.

FAGAN AND DEFRIES

16

Figure9.SaturationofSARatHighForestBiomass

Note:ThisgraphshowsthesaturationofSARbackscatterfromtheLband(darkline,top),Pband(gray,bottom),andVHFband (lightgray,middle)overaforestinLandes,France.Inthisstudy,LandPbandsensitivitytoincreasingbiomassislimitedafter 100Mg/ha;otherstudieshaveachievedhighersensitivitybycombiningpolarizations. Source:LeToanetal.2004.

ItispossibletoobtainexactmeasurementsofdistancetotheEarthbyusingtwoormoreSAR sensorsthatreceivesignalsfromthesametarget.Thisisknownasinterferometryanditoperates onthesameprinciplesthatanowlusestodeterminesounddirectionanddistance.Withtwo separatedSARsensors,thebackscattersignaldoesnotarriveattheexactsametimeorstrengthat eachsensor.DifferencesbetweenthepairedSARsignalscanbeusedtodeterminetheangleand distanceofthetargetrelativetothesensor(Balzter2001).Therearetwotypesofinterferometry: repeatpassinterferometry,bywhichtwosensorsimagethetargetatdifferenttimesandangles; andspatialinterferometry,bywhichtwosensorsseparatedbyabaselinedistanceimagethetarget atthesametime.InterferometricSAR(InSAR)datacanbegatheredfromtwoimagesondifferent days(e.g.,fromsatellitesflyinginaknownorbitorformation)orfromasensorspecificallybuiltto containtwoSARsensors. UsingSARinterferometry,passivestereoimaging,orLIDAR(discussedbelow),itispossibleto produceathreedimensionalmapoftheEarthssurface,calledadigitalelevationmodel,orDEM, thatisaccuratetolessthanameteroferror.ThefirstglobalDEMwasproducedin2000bythe ShuttleTopographicRadarMission(STRM),inwhichaCandXbandInSARsensorwasmounted onthespaceshuttle(Jensen2007).ThisDEMisdegradedformilitaryreasonsfrom30to90m pixelsoutsidetheUnitedStates,andinforestswithmoderateorhigherbiomass,itconfusestree canopyheightfortheground(BourgineandBaghdadi2005;Kellndorferetal.2004).

FAGAN AND DEFRIES

17


CorrectingSARimageswithaccurate,groundlevelDEMsgreatlyimprovesbackscatter interpretationandallowsdirectforestheightmeasurementusinginterferometry(Ticehurstetal. 2004;Kellndorferetal.2004,).Thedevelopmentofaglobal,groundlevelDEMforforestedregions wouldenableasignificantadvanceintheabilityofSARtomeasureforests(Kellndorferetal.2004; BourgineandBaghdadi2005).

LightDetectionandRanging(LIDAR)
LIDARsystemsemitcoherentlightandmeasurethetimingofsignalreturnfromtheEarths surface.Theyareessentiallylaseraltimeters,orrangefinders,thatmeasurethedistancetotargets withgreatprecisionbytiminghowlonganemittedlasertakestoreflectoffthegroundtarget (Dubayahetal.2000).ThetargetsthatLIDARsamplesaresmallareascalledfootprints,andthe lasersreflectfromfoliageandwoodybiomassaswellasfromthegroundsurface.CurrentLIDAR systemsdifferfromSARsystemsinthattheysamplefootprintsatdiscontinuousintervals,rather thanreceivewavereflectionsfromentirelandscapes(Figure10).InterpolationbetweenLIDAR samplingfootprintsisnecessarytogeneratevegetationcanopymapsorDEMs(Lefskyetal.2002b; Figure11a).TheresolutionofthistechnologydependsonthesamplingdensityofLIDARfootprints. Figure10.PotentialLIDARsamplingfootprintsfortheDESDynImission

Note:Examplefootprintsaredotsindicatedontheinset.Footprintsnotto30kmscale.LIDARfootprintswillbetakenjust alongtheredlines. Source:Obertoetal.2008.CourtesyofNASA.

LIDARsystemsdifferintheirfootprintsize,firingfrequency,andtheinformationrecordedfrom thelaserpulse.TherearecurrentlytwomaintypesofLIDARsystems:smallfootprint,discrete postingsystems,andlargefootprint,waveformrecordingsystems(Lefskyetal.2002b).Athird experimentaltype,flashLIDAR,scanslargeareasandreceivesreturnsinasimilarfashiontoSAR

FAGAN AND DEFRIES

18


(Swansonetal.2009).Becausethistypeisexperimental,wewillnotdiscussitfurtherinthis report. SmallfootprintLIDARssamplesmallareas(submeterwidth)athighdensity(multiple footprintspermeter,Figure11a).Theyrecordsingleormultiplereturntimes,orposts,fromthe returnedlaserpulse(e.g.,firstreturn,lastreturn,and/orthepeaklaserreturns,Figure11a).A largenumberofsmallfootprintLIDARmeasurementsallowcreationofveryrealisticmapsofbare areasandthetopsofvegetatedcanopies(Lefskyetal.2002b;Figure11b).SmallfootprintLIDAR pulsesareabletopenetratetothegroundthroughdensevegetatedcanopiesonlyiftheyarespaced veryclosely(Clarketal.2004b).Thelargenumberofcloselyspacedfootprintsrequiredtoimagea forestcanopycurrentlylimitssmallfootprintLIDARtoaerialsurveyingofsmallswaths(e.g.,5km; Swansonetal.2009;Dubayahetal.2000). Figure11aandb.ReturnsfromSmallFootprintLIDAR

Source:CourtesyofImageTreeCorporation.

FAGAN AND DEFRIES

19


Figure11c.ExampleWaveformfromaLargeFootprintLIDAR

Note:Thegraphontheleftindicatestheactualwaveformdatacollectedbythesensor,andtheimageisillustrativeofthe interactionstakingplace. Source:CourtesyoftheJetPropulsionLaboratory,NASA.

Itispossibletorecordmoreinformationfromeachlaserpulse,andlargefootprintLIDARsuse laserpulseslargeenoughtopenetrateindividualtreecanopies(e.g.,1030metersindiameter, Figure11c).Fromeachfootprint,thefirstlaserreturnindicatesthetopofthevegetatedcanopy,the lastlaserreturnindicatesthegroundelevation(Lefskyetal.2002b),andtheamountofreturned energypertimeindicatestheheightandareaofthereflectingsurfaces.Bydisplayingthe distributionofsignalreturnsbyreturntime,LIDARsystemscreateawaveformimageofthe vegetationstructureandtopographyatthesamplingpoint(Figure11c).LIDARwaveformsareused tocreatehighlyaccurate,threedimensionalmodelsofactualforeststructure;forestheight, abovegroundbiomass,andotherquantitiescanbeestimatedfromlargefootprintLIDAR(Lefskyet al.2002b;Dubayahetal.2000). TheutilityofLIDARforforestmappingisonlylimitedbythewidthanddensityof footprints.Atwideintervals,significantinterpolationisnecessarytoestimateforestheightor underlyingtopography.Accuracyislowerinheterogeneousareas(Hudaketal.2002).Whilesmall footprint,aerialLIDARtypicallysamplesasmallareaveryintensively,largefootprint,aerialLIDAR maysample25mfootprintsspacedat50mintervals.LIDARmissionsmustbalancefootprint width,number,andspacingtomeasurelargeareasaccurately.AerialLIDARsystemsarecommonly usedinbothscientificandcommercialforestmeasurement;theseprimarilyaresmallfootprint sensors,whichdespitetheirlimitedimageswathhaveimpressivedetailoverimagedforests (Figure11b). Becauseofthedistanceofsatellitesfromtheearthandtheamountofareathesatelliteare designedtocover,satellitebasedLIDARmissionsarelargefootprintandwidelyspacedbetween orbitalswaths(e.g.,50mfootprintsspacedat13kmintervals)(Obertoetal.2008;NASA2009a; Figure10).ThereiscurrentlyonlyonesatellitemountedLIDARsensor,ICESatGLAS,andalthough itslargefootprint(~65m)wasnotexplicitlydesignedforforestmeasurement,themissionisbeing

FAGAN AND DEFRIES

20


usedtoestimateforestbiomassandimproveDEMs(NASA2009a;Nelsonetal.2009).Followon missionstothisLIDARmissionarebeingplanned,butthemissionscanbeexpectedtohaveshort lifetimesofonlythreetofiveyearsbecausethelasersburnoutafterasetnumberofuses(NASA 2009a).

2.3 Putting it All Together: Remote Sensing Fusion


Remotesensingforforestmeasurementusesmeasuredquantities(e.g.,canopygreenness andradarbackscatter)toestimateotherquantitiesofinterest(leafwatercontent,forestbiomass). Incorporatingmorethanonesourceofremotesensingdata,knownasimagefusion,typically improvesestimatesofforestproperties.Theseestimatescanalsobeimprovedbyincluding informationfromgeographicallyreferenceddatasetsorganizedusinggeographicinformation systems(GIS;Figure12a).Forlackofabetterterm,wedefinetheintegrationofremotesensingand GISdataasGISfusion.GISfusionisstandardpracticeinremotesensing.Imagefusionhasonly recentlyincreasedinprevalence.Bothpracticesgreatlyincreasethepowerandutilityofremotely senseddata. Figure12a.ExampleofGISRemoteSensingFusion

Note:Undisturbedforests(greengrid)canbemappedbyoverlayingGISdataonroadnetworks(blue)ontoasatellitederived forestmap. Source:GOFCGOLD2008.

FAGAN AND DEFRIES

21


Figure12b.ExampleofImageFusion

CoarsefineImageFusion

Note:Inthiscase,alowerresolutioncolorSPOTimageisbeingfusedwithahigherresolutionSPOTpanchromaticimage,ina typeofcoarsefineimagefusionknownaspansharpening.Noticetheboatwakesthatappearonlyinthehigherresolution panchromaticandfusedimages. Source:Short2009.

FAGAN AND DEFRIES

22

2.3.1ImageFusion
Satelliteimageryfromtwoormoresensorscanbecombinedtocreateasingleimage(or dataset)withhigherresolutionand/orutility(Treuhaftetal.2003;Walkeretal.2007;Royetal. 2008;Dalponteetal.2008;Saatchietal.2007).Alltypesofimagescanbefused.Fusionrefersto thecreationofnewmapsbycombiningdatafromoverlappingimages(Figure12b).Thetwomost commontypesoffusionsareactivepassiveandcoarsefine,butthepotentialforimagefusionis limitedonlybyavailabledata. Passiveandactivesensorsmeasureverydifferentquantitiesabouttheearthssurfaceanddiffer inuses.Becauseapassivesensormeasuresreflectancefromsurfaceswithdifferentphysicaland chemicalproperties,thesensormeasureslightabsorptionbygreen,photosynthesizingleaves.The degreeoflightabsorptionbyphotosynthezingleavesisaproxyforestimatingforestproductivity (therateofbiomassproduction).Anactivesensorcanestimatesecondaryforestbiomassby measuringthethreedimensionalstructureofforests.Becauseproductivityispositivelycorrelated withbiomassinlowbiomassforests(KeelingandPhillips2007),fusingapassiveimagewithan activeSARimagewouldyieldbetterestimatesofproductivityandbiomassthaneitherimageby itself. Forpassivesensors,currentlimitsondatastorageandthecapabilityofsensorsmeanthatno onesensorcansimultaneouslymaximizepixelresolution,spectralresolution(numberofbands), andareaofcoverage(Anderssonetal.2009;Jensen2007).Thus,itisdifficulttocoverlargeareas withhighresolutionimagerywithouthavingtomosaic(orcombine)thousandsofdiverseimages takenatdifferenttimes(Anderssonetal.2009;Figure13).Eachimagemustbecorrectedfor differencesinsolarilluminationandatmospherichaze,andsuchcorrectionisoftendifficult.One waytoimprovetheresolutionofwideswathsensorsistofusehighresolutionimagerywithcoarse resolutionimagery.Inpansharpeningfusion,thehighresolutionimageryisusedtoincreaseblack andwhite(panchromatic)detailinthecoarsercolorimage(Figure12b;Qianetal.2007).In trainingfusion,highresolutionimagesareusedtocorrecttheestimatesofthecoarserresolution imageryinaniterativeprocess.Bothofthesemethodscandramaticallyimprovetheresolutionand utilityofcoarserresolutionimagery(Qianetal.2007;e.g.,Hansenetal.2003).

2.3.2GISFusion
FusingGISdatawithremoteimagesisroutine(e.g.,Jensen2007).UseofGISallows organizationanddisplayofdatasetswithexactlyknowngeographiclocations(georeferenced data).Theselocationscanbegroundderived(e.g.,citybuildingdatasets)ortakendirectlyfrom remotelysensedimagery(e.g.,road,landcover,orDEMdatasets).IntegratingGISdatawithremote sensingimagerymakesnewanalysespossible(e.g.,roadsandlogging;Asneretal.2006)andcan improveestimatesfromremotesensingdata(e.g.,climateandbiomass;Baccinietal.2004).Geo referenceddataareusedregularlytoquantifyandimprovetheaccuracyofremotesensinganalyses andpredictionsinaprocessknownasgroundtruthing.Groundtruthdataarehighaccuracydata usedtoevaluate,orvalidate,theaccuracyofmapsderivedfromsatelliteimagery.Thesedataarea crucialcomponentofremotesensing(seeSection3.1.3).TheincreasingintegrationofGISand remotesensinghasfacilitateddisplayandcommunicationofsatelliteimagerybetweenscientists andforthepublic,aswitnessedbytheexplosivegrowthinmappinginsocietyandthewidespread useofGoogleEarth,MicrosoftsBingMaps,andotherproducts.

FAGAN AND DEFRIES

23


Figure13.MosaicofModerateResolutionLandsat7Images

Source:ImagecourtesyofU.S.GeologicalSurveyGlobalVisualizationViewer(GLOVIS).

2.4 An Overview of Current and Near-Term Earth-Observing Technology


Therearemanyearthobservingsatellites(EOS)currentlyinorbit,bothactiveandpassive, includingapproximately110nonmilitarysatellites(UCS2009).Thereareabout75distinctsensor typesfocusedonobservingtheearthssurfacewithcoarseorhigher(<1.5km)spatialresolution, andthesearethesubjectofthissection. Between2009and2015,weexpectapproximately87additionalsurfacefocusedEOStobe launched,manyofthembearingnewtypesofsensors.ThemajorityofEOSarenotcommercially owned,butmanynationalagenciesselltheirimageryatcommerciallycompetitiveprices.The Appendixsummarizesthecharacteristicsofcurrentandexpectedearthobservingsensors.We dividethesensorsintofivecategories:highresolutionpassivesensors(<6m),moderateresolution passivesensors(6150m),coarseresolutionopticalsensors(>150m),activesensors,andfuture (20092015)satellitesensors.WediscussothermiscellaneousEOSandnotableairbornesensors

FAGAN AND DEFRIES

24


onlybriefly.Whenanindividualsensorspanscategoriesdependingonitsmodeorband,weplaceit inthemoredominantcategoryforthatsensor.Wegatheredallofthedataandspecificationson satellitesfromthesourcesnotedintheAppendixandpubliclyavailablewebsites.

2.4.1CurrentSensorTechnology
MultispectralpassivesatellitesmakeupthevastmajorityofEOSandhavebeenincreasingin spatialresolution,spectralresolution,andreturnfrequencyinrecentyears.Spatialresolutionhas increasedinmoderatespatialresolutionsatellites,buttheadventofhighspatialresolution, pointablecommercialsatelliteshaschangedthecostandavailabilityofhighresolutionimagery. Theseincreasesinspatialresolutionhaveoftencomeatthecostofimageswath(seeFigure6)but technologyisimprovingoverall,asillustratedbythewideswathbutmoderateresolutionofthe DMCconstellation.Thereareafewhyperspectralsatellitescurrentlyinoperation(e.g.,theU.S. ModerateResolutionImagingSpectroradiometer(MODIS),theHyperioninstrumentontheU.S. EarthObservingMission1,andtheCompactHighResolutionImagingSpectrometer(CHRIS)onthe EuropeanSpaceAgencysProjectforOnboardAutonomy(PROBA)spacecraft),butingeneralmost satellitesmeasurebetweenoneand12bands(Appendix).Severalnotablecoarseresolution satelliteconstellationshavebeeninoperationfordecades,includingAVHRRandGOES,theU.S. launchedGeostationaryOperationalEnvironmentalSatellites.Therecentlaunchingofmore moderatetohighresolutionconstellations,suchasMODIS,theDMCseries,andtheRapidEye series,isdramaticallychangingreturntimesforforestmeasurement(Appendix).WhereasLandsat imagesonceevery16days,andcloudfreeimagesinsomeareasonlyoccuronceayear(Olanderet al.2008;Asner2001),theDMCconstellationhasdailyimagingcapability(seetheAppendix). RecentandupcomingSARsensorlauncheswillincreasethenumberofactivesensorsinorbit. ModernSARsatelliteshavehigherandmoreflexiblespatialresolution,greaterimageswath flexibility,andmorepolarizationmodesthanprevioussatellites(seetheAppendix).Japans AdvancedLandObservingSatellite(ALOS),whichincludesanLbandsensorwithfourpolarization modes,hasbeenusedforglobalforestcovermappingsinceitslaunchin2007(Olanderetal.2008). ThereisonlyonefunctioninglandobservingLIDARsatellite.Thebiggestbarriertowidespread usageofactivesensorsremainsthetechnicaldifficultyofprocessingSARdata,butthecostofSAR data,althoughequaltosomeopticalsensors,isalsoabarriertoglobalmapping(Patenaudeetal. 2005). Fromthestandpointofglobalforestmeasurement,themostsignificantrecentadvancein remotesensinganalysisiseconomic,nottechnological.Asofearly2009,theU.S.Landsatarchive (19722009)becamefreelyavailable,thusenablingglobalandtemporalanalysisonmoderate resolutionimagery(Olanderetal.2008).TheregionalBrazilianCBERSarchive(19992009)of imageryisalsonowfreeinmostdevelopingcountries(GOFCGOLD2008;Powelletal.2007). Currently,thesetwofreeimageryseriesofferthehighestresolutioncoverageperdollar,andthey haveagoodtemporalrecord(Figure6).InconjunctionwithotherfreeMODISandAVHRRdata,itis nowpossibletosurveyglobalforestsdailywithcoarseresolutionimagesandcreatemoderate resolutionforestmapswithmonthlytoannualfrequency.TheLandsatDataContinuityMission (andtheupcomingLandsat7imagerymosaics,seebelow)willcontinuethisfreedatapolicyand makeitpossibletodevelopalongterm,globalrecordoflanduseandlandcover(Lovelandetal. 2008).

FAGAN AND DEFRIES

25

CoarseResolutionPassiveSensors
Coarseresolutionpassivesensorshaveapixelsizeof2001,500m,receiveradiationinthe opticalthroughthermalrange,andhaveimageswathsrangingfrom360kmtothefullearthdisk (Figure6,Appendix).Itwouldtakebetween14and40AVHRRor2580MODISimagestocoverthe 13.4billionhectare(ha)globallandarea(estimatesareinexactforcoarseresolutionimagesthat includelandandwater;Table3).Becausecoarseresolutionisusefulformonitoringweatherand climate,manycoarseresolutionsensorsarepartsoflargeconstellations.Forexample,theAVHRR sensorconstellationincluded12satellitesbetween1978and2009,includingseveralsatellites operatingsimultaneously(Jensen2007).Althoughmanycoarseresolutionsensorsonlyhaveafew visibleornearinfraredbandsatresolutionstoocoarsetodetectmostlandcoverchange,thevery highrepeattimesmakethesesensorsidealformonitoringphenology,fire,andotherephemeral events.Theadventofmoderatecoarsesensors(e.g.,MODIS,MERIS,250300m)hasimproved globallandcovermapping(Heroldetal.2008).Arecentlylaunchedhyperspectralcoarse resolutionsensormountedontheIndianMicrosatellite(IMSHySI)haspotentialforlandcover mappingaswell(Appendix). Coarseresolutionsensorsaremountedonbothpolarorbitingsatellitesandgeostationary satellites.Althoughgeostationarysatellitesimagetheentireearthdiskinrealtime,currentsensors areultracoarse(>1.5km)inresolutionandonlytheGOESseriesisdescribedhereasanexamplein theAppendix.Thereareseverallongterm,similargeostationarysatellitesseriessupportedbythe UnitedStates(GOES),Europe(Meteostat),Russia(Elektro),andChina(FY3and4). Table3.CostsofImageryforWalltoWallSamplingoftheGlobalLandArea
SatelliteSensor MODIS Landsat7ETM GeoEye1 ALOSPALSAR SensorType(pixel size) Coarseresolution (2501000m) Moderate resolution(28.5m) Highresolution (1.65m) Coarsestresolution mode(100m) NumberofImages Required 2580 3921 580,787 1096 Priceperkm2(in US$,2009) Free Free,formerly $0.02in2008 $25(forabasic imageofEurope) $0.0017 EstimatedTotal Cost $0 $0,formerly $2,683,700 $3,354,625,000 $230,075

Note:CreatedfromAppendix1forselectedsensors;thenumberofrequiredimagesisanestimatebasedonthe134,184,000 km2globallandarea(seetext).Priceschange;theseestimatesarefor2009.ALOSPALSARwouldlikelybedegradedto400m resolutionforaccuracy.

ModerateResolutionPassiveSensors
Moderateresolution(10200mpixel)passivesensorscanreceiveopticalthroughthermal radiationandhaveimageswathsrangingfrom60890km(Figure6,Appendix).Itwouldtake approximately3,921LandsatimagestocoverthegloballandareaTable3.Therearecurrentlyfive majormoderateresolutionsatelliteseries:Landsat,theSatellitePourLObservationdelaTerre (SPOT),CBERS,theIndianRemoteSensingsatellite(IRS),andtheDMCseries(Achardetal.2007; Powelletal.2007).Inaddition,China,Argentina,Thailand,andRussiaallsupportindividual

FAGAN AND DEFRIES

26


moderateresolutionsatellites,butitisunclearwhetherChinasdataareavailabletothepublic. Additionally,thereareseveralexperimentalmoderateresolutionsatellites,mostnotablythe hyperspectralHyperionandCHRISsensors,andthesehaveshownpromiseforhyperspectral imagingasameansofdistinguishinglandcovers(Ustinetal.2004). TheLandsatseries(satellites17,1972present;Jensen2007)istheworkhorseforgloballand coveranalysis.Unfortunately,Landsat5isonlytransmittingdatainregionswithgroundstations, andLandsat7hadascanlinefailurein2003thatcausedblackstripestoappearinthesidesofits images.TheUnitedStatesGeologicSurvey(USGS)isworkingtomosaic(thatis,combine)Landsat7 imagestocorrectthescanlineproblem,butthenextLandsatisnotduetobelauncheduntil2012 (Lovelandetal.2008).Acontinualrecordofgloballandcoverrequiresfillingthe20032012 Landsatdatagap.Tthereareseveralpossibleapproaches.TheAdvancedSpaceborneThermal EmissionandReflectionRadiometer(ASTER,launchedin1999)isamoderatecostsatellitethathas nearglobalcoverage(Jensen2007).SPOT(15,1986present;Jensen2007)isasatelliteseries withglobalcoveragebutdataarerelativelyexpensivetopurchase(Figure6).TheChina/Brazil EarthResourceSatelliteseries(CBERS,threesatellites,1999present;Jensen2007)isagrowing constellationofsatelliteswithfreeimagery,buttheimageryisfocusedoverBrazil,China,andmost recently,Africa.TheIndianResourceSatelliteseries(IRS,sevensatellites,1988present;Jensen 2007)isalarge,longrunningconstellationwithglobalimagecoverageavailableatsomecost. Finally,theDMCseriesisaconstellationofsatellitesfromdifferentcountriesinthesameorbit; imagesfromdailyoverflightsareprovidinganimportantcommercialbackuptoLandsat(GOFC GOLD2008).

HighResolutionPassiveSensors
Highresolution(0.310mpixel)passivesensorsreceiveopticalthroughnearinfrared radiationandhaveimageswathsrangingfrom11to70km(Figure6,Appendix).Therehavebeena numberofcommercialandnationallaunchessincethefirsthighresolutionsatellite,IKONOS, launchedin1999,andtheseadditionstothefleetofEOShaveledtotheformationofcommercial highresolutionconstellations.Theseconstellationsincludehighresolutioninstrumentsoperated byseveralU.S.companies,includingGeoEyeandDigitalGlobe.Oneofthechallengesofusinghigh resolutionimageryforglobalforestmappingisthatapproximately580,787GeoEye1imagesare requiredtocoverthegloballandarea(Table3).Mostofthehighresolutionsatelliteshavefive bands;fournarrowbandsinthevisibleandnearinfrared,andonehigherresolutionpanchromatic bandthatcanhavesubmeterresolution(fornationalsecurityrequirements,panresolutionis actuallydegradedto0.5monU.S.commercialsatellites).Therehavebeenseveralrecentlaunches ofsolelypanchromaticsensorswithprimaryapplicationsindefenseandmapping.Mosthigh resolutionsensorsarepointable,decreasingtheireffectivereturntimefrom,forexample,26days to3days.Thisresultsinspottyglobalcoverage,particularlyforforests,asthecoverageisoften concentratedindenselypopulatedandlesscloudyareas.Therecentdevelopmentofcheap microsatellitesthatcanbelaunchedasaconstellation(e.g.,RapidEye)orasanondemandsensor (e.g.,theU.K.smicrosatellite,TopSat)coulddramaticallyincreasetheglobalcoverageandreturn timeofhighresolutionsensors(KramerandCracknell2008).

FAGAN AND DEFRIES

27

ActiveSensors
Activesensors(SARandLIDAR)currentlymakeupasmallproportionoflandobserving satellites;thereareapproximatelynineSARsensorsandoneLIDARsensorinorbit(Appendix). Activesensorshaveresolutionsrangingfrom11,000mandimageswathsrangingfrom10500 km;itwouldtakeapproximately1,096ALOSimagestocoverthegloballandarea.Inthelast decade,SARsatellitetechnologyhasimprovedtofourpolarityoptionsandvariableresolution (withsmallerswathsathigherresolutions)indifferentimagingmodes.Until2006,allpublicly availableSARsatellitescarriedCbandsensors(Jensen2007)exceptfortheLbandJERS1mission thatranfrom19921997,theshortlivedSeasat(Lband)andAlmaz(Sband)missions,andthe briefSRTMandSIRA,SIRB,andSIRC/Xshuttlemissions.TheCanadianRadarsatserieshasbeen gatheringCbandSARimagerysince1995,andtheEnvisatASARhasbeengatheringCbandaswell since2002.Themostrecentsatellitelaunches,ALOS(LBand),TerraSARX(Xband),andCOSMO SkyMed(Xband),haveincreasedthediversityofSARbandsinorbit.ALOSprovidesglobal coverage,TerraSARXisthefirstsatelliteinaninterferometricconstellation,andCOSMOSkyMedis amicrosatelliteSARconstellationwithavarietyofmodes.Whenevaluatingthespatialresolutionof SARsatellites,givenintheAppendix,itisimportanttorememberthattheireffectiveresolution afterspeckleaveragingwillbethreetofourtimesmorecoarse,approximately50100m. ThefirstgroundmeasuringLIDARaltimeter,ICESat,waslaunchedin2004andafollowonis planned.WhiletheswathsICESatmeasuresdonotcovertheentirelandsurface,theinstrumenthas providedusefulinformationformodelingice,elevation,andforeststructure(NASA2009a).

2.4.2ExpectedSensorTechnology,20092015
Futuresatellitelauncheswillfillcurrentdatagaps,providefollowonmissions,populate constellations,andprovidenewfunctionality.Thediversityofcountrieswithsatellitesinorbitwill increasewithnewsatellitesfromSpain,Turkey,Taiwan,theUnitedArabEmirates(UAE),Sweden, Ukraine,Malaysia,andNigeria.ThemajorityoflauncheswillcomefromChina,theUnitedStates, countriesintheEuropeanUnion,Argentina,andIndia,withnotablesmallercontributionsfrom Brazil,Russia,Canada,andJapan(Appendix). Amongfollowonmissions,theLDCM,CBERS3,andtheultrahighresolutionGeoEye2merit mentionfortheiradvancedtechnology.Bothnationalandcommercialhighresolutionoptical missionswillcontinuetoincreaseinnumberasmicrosatellitetechnologylowersthecostsof production(CEOS2009;Olanderetal.2008;Kramer2008).SeveralnewSARconstellationsare planned(TerraSAR,Sentinel1,Radarsat,ALOS,andSAOCOM1series).TheSentinel2series,a moderateresolutionopticalconstellation,isalsoplanned.Therewillbeseveralnewmoderate resolutionhyperspectralsensors(EnMap,HyspIRI,TESHYS,SACF,andPRISMA).Mostnotably,of the87orsolandobservingmissionsplannedbetween2009andlate2015,25willmountSAR sensors,twowillmountLIDAR,andonewillmountbothadramaticincreaseinthenumberand diversityoffunctioningactivesensors. Overall,itislikelythatspatialresolution,returntimes,andSARcapabilitieswillimproveover the20092015period,withsignificantnewhyperspectral,InSAR,andLIDARfunctionality.Despite theunpredictablenatureofgovernmentdatapoliciesandimageryprices,aswellasuncertainties

FAGAN AND DEFRIES

28


relatingtolaunchschedulesandsuccess,thenextsixyearswilllikelybringasignificantexpansion inthecapacityofremotesensingtomonitortheearthssurface.

ExpectedCoarseResolutionPassiveSensors
Approximately27coarseresolutionsensorsareplannedforlaunchbetween2009and2015, butitisunclearifdatawillbepubliclyavailablefromninesensorstobelaunchedbyChina.Most launcheswillcontinueconstellations(e.g.,GOES,FY3),withtheAVHRRsensorcontinuingon MetOpAandMetOpB.TheUnitedStatesplanstolaunchanewsatelliteseries,NPOESS,asa replacementtotheMODISandAVHRRsatellites,butNPOESShaslowermaximumresolutionthan MODISandmaybemoresuitableasanAVHRRfollowup(TownshendandJustice2002).Two sensorsthatmaybemoresuitablesuccessorstoMODISaretheGCOMC1(with11bandsat250m resolution)andtheSentinel3series(with21bandsat300mresolution).Mostexpertsagreethat itwillbedifficulttomatchthecombinationofresolution,revisittime,andfreeimagerythatMODIS currentlyoffersforgloballandcoveranalysis. InadditiontoGCOMC1andtheSentinel3series,therewillbeafewtechnicalimprovementsin coarseresolutionsensors.GISATwillmountacoarsesensor,amoderatesensor,andacoarse resolutionhyperspectralsensor.SACDwillbeacoarseresolutionthermalinfraredsensorto measurethetemperatureandenergyreleaseoffires.Mostsignificantly,realtimemonitoringof vegetationandproductivitywillbecomepossibleafter2014,withthelaunchofvisibleandnear infrared(VNIR)geostationarysensorswitharesolutionof1km.Thesesensors,startingwithGOES R,offerasignificantimprovementovercurrenttechnology.

ExpectedModerateResolutionPassiveSensors
Atleastfifteensatellitesbearingmoderateresolutionsensorsareplannedforlaunch.Theseare NigeriaSat2,Amazonia1,RASAT,SACF,ARGO,CBERS3,CBERS4,KanopusV,Resourcesat2, Resourcesat3,LDCM(Landsat8),GISAT,HyspIRI,Sentinel2A,andSentinal2B.Themultispectral ARGO,LCDM,Resourcesat2,GISAT,andSentinel2sensorsareexpectedtocollectglobaldata. Severalhyperspectralmoderateresolutionsensorsareunderdevelopment.Ofthesixplanned hyperspectralsensors(EnMap,HyspIRI,TESHYS,GISAT,PRISMA,andSACF),onlyHyspIRIhasa swathsizesuitedtofrequentglobalcoverage.Iflaunchingagenciessupportglobaldatacollection, thisincreasingdiversitywillsafeguardglobalcoverageagainstthelossofasinglesatellite. ResolutionintheSPOTserieshasimprovedtothepointwhereSPOT5andSPOT6areconsidered wideswath,highresolutionsensors.TheLDCMwillhaveninebandsandmaintaincontinuityina criticalimageryseries,IndiasResourcesat3willhave23mresolutionandanextrawideswathof 700km,andtheSentinel2serieswillbeanadvanceinsensortechnology,withhighresolution(10 m),13bands,awideswath(240km),andthreededicatedbandsforatmosphericcorrection.Ifthe LDCMfailstolaunchoroperateproperly,aninterruptioninfreeaccesstoglobalimageryislikely.

ExpectedHighResolutionPassiveSensors
Althoughitisdifficulttoanticipatecommerciallaunches,thenextseveralyearswillseean increaseinthenumberofstandard(fourbandVNIR,resolution110m)highresolutionsensors (Appendix).Severalcountrieshavetakenadvantageofthelowercostofmicrosatellitestocreate highresolutionsensorswithtechnicalassistancefromaerospacecompanies(e.g.,DubaiSat(UAE)

FAGAN AND DEFRIES

29


andMalaysiasRazakSat).MicrosatellitetechnologyhasledtothehighresolutionRapidEye constellationofsixsatellites.Theconstellationwillbecompletedbytheadditionofthesatellite ARGOinmid2009.Withhighrevisittimesandalargerswaththanmanyhighresolutionsatellites (78km),RapidEyehasthepotentialtodevelopglobalcoveragequickly(RapidEye2009). Someimprovementsinhighresolutionsensorcapabilitiesandlowerdatacostsareexpected. WorldView2willhavesubstantialspectralandspatialresolutionwitheightbandsintheVNIRand 1.8mresolution.VENUSwillalsoofferhighspectralandspatialresolution,with12bandsinthe VNIRwitharesolutionof5.3m.Pleiades1andPleiades2areextremelypointable(+/50)and willaddflexible,highresolutioncapacitytotheSPOTseries.SPOT6willalsocontinuetheSPOT series,withawideswath(60km)andhigherresolution(8m,2mpan).GeoEye2willhavethe highestspatialresolution(0.25m)inthe20092015interval.PRISMAwillintegratea2.5m resolutionpancamerawithhyperspectralimagery,andtheSentinel2series(discussedabove)will havehighresolutionimagery(10m)intheVNIRwithaverywideswath.Finally,CBERS3and CBERS4willcombinehighresolutionimagerywithamoderateresolutionimagerandacoarse resolutionthermalimager.

ExpectedActiveSensors
SARandLIDARsensorswillmakeupathirdofnewlaunchesfrom20092015,withanincrease inthediversityofSARbands(C,X,S,L,andpossiblyP),functions(InSARanddualInSAR/LIDAR), andnumberofSARconstellations.TheTerraSARX(Xband)constellationwillflyininterferometric formationtoproduceahighresolutiondigitalelevationmodelforcommercialpurposes.The resolutionwillbe1mbeforespeckleaveraging.TheCOSMOSkyMEDconstellation(Xband)willbe integratedwiththeplannedSAOCOM1constellation(Lband)fortwicedailyoverflightsintwo bands.TheRadarsatConstellation(Cband)willcontinuethelongrunningRadarsatseriesof imageryanddecreaserevisittime(eoPortal2009).TheSentinel1constellation(Cband)willhave aninterferometricmodeandvariableswathwidth.Finally,thelaunchofALOS2willcreateahigh quality,Lbandconstellation. Thereareseveraladvancesinactivesensortechnologyplannedforlaunch,includingavariable C/Xbandsensor(DMSAR),interferometricsensors(InSAR,severalsatellites),aLIDARfollowon mission(GLAS),ahighresolutionSbandsensor(HJ1C),aLIDARInSARintegratedLbandsensor (DESDynI),andpotentiallyaPbandinterferometricsensor(BIOMASS).Ofthese,DESDynI,GLAS, andBIOMASSmeritspecialmentionfortheirpotentialtoinformforestmeasurementand monitoring.DESDynIwouldbethefirstintegratedInSARLIDARmission,anditwouldcombinethe relativestrengthsofInSARandLIDARtocreateunparalleledmeasurementofecosystemstructure andforestheight.GLASwouldcontinueglobalLIDARmeasurement,thusincreasingtheaccuracyof globalDEMestimates.BIOMASSiscurrentlyasensorconceptunderconsiderationforlaunchinlate 2015orearly2016andwouldconsistofaPband,interferometricSARspecificallydesignedto measureforestbiomassinrealtime.

2.4.3ACautionaryNoteonPlanningandSatelliteTechnology
Althoughplannedlaunchesaregoodindicatorsoffuturepossibilitiesforearthobservation, successfulsatellitelaunchandoperationareneverguaranteed.Satellitelaunchesareusually delayedbybudgetaryandconstructionexigencies.Satellitesregularlyfailtolaunch(examples

FAGAN AND DEFRIES

30


includethelaunchfailuresofLandsat6,IKONOS1,andtheOrbitalCarbonObservatory,in1993, 1999,and2009,respectively).Onceinorbitandfunctioning,satelliteshaveexpectedoperatinglife spansandoftenoutlivethese,buttheiractualoperatinglifespanisamatterofchance.An instrumentonLandsat7hadamajorproblemfouryearsafterlaunch,whileinstrumentson Landsat5arestillfunctioning24yearsafterlaunch(Powelletal.2007;Lovelandetal.2008). Satellitebornesensorscandegradeandshiftinqualityovertime,aswell,andtheirimageproducts requireregularcalibrationandvalidation.Processinganddistributingthedatageneratedbythe sensorsareasimportantashavingthesatelliteinspace;manydataaredifficulttoobtainorprocess orarenotmadewidelyavailable.Asaresult,longterm,globalforestmonitoringplanscannotrely onsinglesatellitestocarryoutthebulkofdatacollection.

FAGAN AND DEFRIES

31

Chapter 3. Technical Considerations in Global Forest Monitoring


Theuseofremotesensingtoprovideaccurate,longtermglobalforestmeasurementsrequires severaltechnicalandlogisticchoices.Agreementisrequiredaboutwhatactuallyconstitutesa forestandwhichforestquantitiestomeasure.Theselectionofsatelliteimageryandimageanalysis methodsmustbeguidedbytheforestmeasurementsofinterestandbybasicprinciplesofscientific inquiry,suchasaccuracy,repeatability,andlongevity.Todetectchanges,itisnecessarytocreate referenceforestmaps.Aconsistentsurveymethodmustbeselectedthatbalancesanalysistime, coverage,andaccuracy.Finally,thelogisticsofimageprocessing,cost,andgroundtruthingmustbe considered(SanchezAzofeifaetal.2009;Achardetal.2007;Gibbsetal.2007;Olanderetal.2008).

3.1 Measuring Forests Globally


Anyglobalefforttomonitortheworldsforestsmustbeginwithcommondefinitionsand commonmeasurements.Definingaforestisnotassimpleataskasitmightseem,giventhewide variationinforestecosystemsfromcountrytocountryandthedynamicnatureofforestregrowth afterhumandisturbance.Forexample,atwhattreedensityshouldanopenecosystemofshort treesandgrassbeclassifiedasaforest?Isaclearcutdeforestation?Isanabandonedtree plantationaforest? Tomeasureforestseffectively,wemustbeabletocomparegroundmeasurementstakenby forestersandforestecologiststothosetakenbyremotesensingmeasurements.Todothis,wemust beclearaboutwhichforestcharacteristicsweareinterestedinmeasuring.

3.2DefiningForests
Theexactdefinitionofforestversusnonforesthasimportantconsequencesforthemonitoring ofdeforestationandforestdegradationandpoliticaleffortstoreducecarbonemissions.Currently, onlytheFAOusesasingleglobalstandarddefinitionofforest.In2005,theFAOdefinedaforestasa minimumcoverof10percent,heightof5m,andareaof0.5m(FAO2006).UndertheKyoto Protocol,participatingcountriesdefineforestswithintheirbordersbyselectingonevalueof minimumarea,treeheight,andcanopycoverfromthefollowingrange:0.051hectare(ha)area, 25mminimumheightatmaturity,andminimumcrowncoverof1030percent(UNFCC2001).2 DeforestationisdefinedbytheUnitedNationsFrameworkConventiononClimateChange (UNFCCC)asasemipermanentconversionofforestedlandtootheruses.Thisdefinitionassumes thatregrowthwillnotoccurquicklyandexcludestemporarydeclinesincanopycoverdueto logging(UNFCC2001).Remotesensinghasdifficultydistinguishingareasofdeforestationfrom intactforestwhentheintactforestdefinitionincludesaverylowcanopycoverthreshold(e.g.,10 percent).Theoppositeistrueforforestdegradation,whichisdefinedasadecreaseinforestcanopy coverwithoutdeforestation;lowercanopycoverthresholdsincreasethelikelihoodofdetecting degradation(e.g.,adecreasefrom90percentto10percentcover). Thelackofaglobalstandarddefinitionforforestcovermakesitdifficulttocreateforestmaps thatcanbereadilyusedbyallcountries,buttheFAOstandardmaybeausefulstartingpoint.

Anupdatedlistofforestdefinitionsbycountrycanbefoundathttp://cdm.unfccc.int/DNA/index.html.

FAGAN AND DEFRIES

32


Ideally,globalforestmonitoringeffortswillcreatemapsthatcanberesampledbyinterested countriestofittheirforestdefinitions,suchasasetofprocesseddataknownasthevegetation continuousfieldsproductfromMODIS(Hansenetal.2003). 3.1.2ForestMeasurementandtheForestIdentity Historically,forestmeasurementhasbeengroundbased,asdescribedinChapter1.Foresters andforestecologistshavemeasuredseveralforestquantitiesincludingthese: biodiversityspeciesrichnessandevennessinanarea; canopyqualitiesopenness,leafareaindex(LAI),leafwatercontent,andphenology); structuralvariablestreedensity,basalarea,treeheight,andtreediameteratbreastheight (DBH); forestfloorqualitieslitterfall,soilcarboncontent,andseedlingdensity;and ecophysiologicalcharacteristicstranspirationrate,treerespiration,waterstress,and nitrogenflux.

Anorganizingprinciple,theForestIdentity(Kauppietal.2006),isacentralthemeinthe remainderofthisreport.TheIdentityrelatesfourforestattributes(area,volume(densityof growingstock),biomass(themassoflivingmaterial),andsequesteredcarbon)andissummarized indetailbyWaggonner(2009).TheForestIdentityattributesarerelated:forestarea(ha)canbe convertedintovolume(cubicmeters,orm3)usingstemdensity(m3/ha);volume(m3)canbe convertedtobiomass(kilograms,orkg)usingwooddensity(kg/m3);andbiomass(kg)canbe convertedtocarbon(kgC)usingcarbondensity(kgC/kg).Asdiscussedearlier,anysuccessful globalforestmonitoringprojectwillhavetointegrateremotesensingandgrounddata.The simplicityoftheForestIdentitymakesitstraightforwardtoconvertbetweenground measurementsandremotesensing(Table4;Kauppietal.2006).ThecomponentsoftheForest Identitytietogetherfourscientificallyandeconomicallyimportantforestcharacteristicsand provideausefulstartingpointforglobalforestmonitoring.

3.1.3ComparingForestInventoryandRemoteSensingMeasurement
Thesameforestquantities(e.g.,biomass)areestimateddifferentlybygroundforestinventory andbyremotesensing(Table4).Forestinventorytypicallymeasurestreeabundance,diameter, crownwidth,species,andheight(Song2007;Chaveetal.2005).Althoughitispossibletomeasure allofthequantitiesoftheForestIdentitydirectlythroughdestructiveharvesting,foresters generallyuseallometricequationstoestimatevolume,biomass,andcarbonfromtreediameter, speciesspecificwooddensity,and/orheight,andthenextrapolatetotheentireforest(Chaveetal. 2005;Gibbsetal.2007,seeTable4).Allometricequationshavebeendevelopedfortropicalforests (Chaveetal.2004)andhigherlatitudeforests(TerMikaelianandKorzukhin1997)through exhaustivemeasurementandselectiveharvestingandweighingoftrees.Someallometricequations estimatetotalforestbiomassdirectlyfromtreediameterbutgivenlargespeciestospecies variationinwooddensity,theseequationscanbemarkedlyinaccurate(Bakeretal.2004;Chaveet al.2005).Thereissignificantvariationbetweenthebiomassestimatesofdifferentallometric equationsandthisvariationcanbeasignificantsourceoferrorinstandbiomassestimates(Baker etal.2004;Chaveetal.2005).Scalingupfromstandlevelestimatesofbiomasstoregional

FAGAN AND DEFRIES

33


estimatesoftenhaslowaccuracyduetolocalvariabilityinforestcoveranddensity(Houghtonetal. 2001;Houghton2005). Inremotesensing,satellitemeasurementsofforestareaandstructureareusedtoestimate volume,biomass,orcarbon(Table4).Theseestimatescanbesupplementedbyothermeasuresof forestleafarea,productivity,andflammabilityandincorporatedintoaforestmodelthatcanyield improvedestimates(vanderWerfetal.2006;Table4).Satellitederivedforestmetricsaretaken acrossanentireregionmakingextrapolationunnecessary.Twoimportantconsiderations pertainingtoremotesensingmeasurementsofforestsaretheimportanceofcorrelationandthe necessityofaccurategrounddataforoverallaccuracy. Table4.HowForestInventoryandRemoteSensingEstimatetheForestIdentity
ForestIdentity Area ForestInventory Measuredlocally,atonetime,in onetoafewforesttypes. RemoteSensing Measuredregionallyandrepeatedly, distinguishingforesttypesandages (Optical,SAR). Estimatedfrommeasuresofforest height/structure(SAR,LiDAR).

Volume

Estimatedusingdiameteratbreast height,treeheight.

Biomass

Estimatedfromvolumeandwood Estimatedfromarea&forest densitymeasurements.Extrapolated structure.Estimatesareimprovedby regionally measuresofforestflammability, productivity,leafarea,phenology,and gasflux. Estimatedfrombiomassandcarbon Sameasbiomass.Estimatesaresite densitymeasurements.Extrapolated specific,acrossentireregions. regionally.

Carbon

Remotesensingmeasuresreflectedspectra;forestareaandthehorizontalandvertical structureofforestscanbemeasureddirectlyfromthesereflectedspectra.Fieldworkorhigher resolutionimagerycanbeusedtogenerategroundtruthdatatoassesstheaccuracyoftheseforest areaandstructuremeasurements(Jensen2007).Remotesensingenablesestimatesoftheother metricsintheForestIdentity(volume,biomass,andcarbon)by: obtainingforestinventorymeasurements, correlatingspectrawiththosemeasurements,and extrapolatingfromthesecorrelationstotherestoftheforest.

Forexample,existingfieldmeasurementsofbiomassarerequiredtocreateforestwide predictionsofbiomassfromsatelliteimagery;anotherindependentsetoffieldmeasurementsis requiredtoassesstheaccuracyofthesepredictions. Fieldestimatesofvolume,biomass,andcarbonarederivedfromallometricequations.These equationscanbeinaccurateorbiased(e.g.,bynotincludingspeciesspecificwooddensity;Bakeret al.2004).Remotesensingcanonlybeasaccurateasitsgroundtruthdata.Itispossibleforremote sensingtoaccuratelypredictbiasedestimatesofbiomassfromstandardallometricequationsthat,

FAGAN AND DEFRIES

34


forexample,donotincorporatespeciesspecificdifferencesinwooddensity.Thus,itisdifficultto assessthereportedaccuracyofremotesensingpredictionsofforestvolume,biomass,andcarbon. Becauseestimatingthetruevalueoftargetsisquitedifficult,andonlythecorrelationsbetween remotesensingandgroundbasedestimatesarereportedintheliterature,wehavechosento reportthepublishedaccuracyofremotesensingestimates.Theiraccuracyissummarizedhereas eithercorrelationcoefficientsorthepercentageofvarianceexplainedbyregressionequations. Furtherrefinementsingroundbasedmeasurementofforestbiomasswillleadtomoreaccurate, unbiasedestimatesoftruevaluesfrombothforestinventoryandremotesensing.

3.2 Criteria for a Global Forest Monitoring Program


Severalprinciplesguideconsiderationindesigningaforestmonitoringprogramthataimsto measuretheworldsforestsonaregularbasis.Oneprincipleisaccuracy,discussedmorefully below.Anydatageneratedbyaforestmonitoringprogramwillhavepoliticalandscientific consequencesandshouldmeetorexceedstandardsofaccuracyforremotesensing.Another principleisrepeatability,requiringconsistencywithearlierscientificefforts(e.g.,forest definitions),transparencywithscientificmethodology,andpublicavailabilityofsourcedata(e.g., imagery).Thedataandconclusionsfromaforestmonitoringeffortwillbemostusefultoscientists, politicians,andforestersifthedataareverifiable,extendable,andrelevant.Athirdprincipleis longevity;reliable,longtermdataonforestsarerareandthecreationofaremotesensingbased, globalrecordofforestchangewouldbeasignificantscientificaccomplishment(Grainger2009). Thislastguidingprinciplehassignificantconsequencesforprogramdesign. Satellitefailureisunpredictable,sodesigningaforestmonitoringprogramthatdependsonthe successfullaunchoroperationofasinglesatelliteisrisky.Thebeststrategyforlongterm continuousmonitoringistouseastandardspatialandspectralresolutioncollectedbyseveral differentsatelliteserieslaunchedcontinuouslyoverdecades.Forexample,the30mresolution,six bandLandsatsensordataformathasbeencollectedforafewdecadesanditcanbeapproximated byseveralsatelliteseries(e.g.,SPOT,Avnir,andIRS)(Powelletal.2007).The1.1kmresolution, multibandAVHRRsensordataformatisalsosupportedlongtermandcanbeapproximatedbya numberofsatelliteconstellations,includingSPOT,thegeostationaryconstellations,andtheChinese FYseries(seetheAppendix).TheSARCbandhasbeentakenat30mresolutionbyseveral differentsatellitesoverthelasttwodecades,andCbandsensorsareplannedfornewlaunches (Appendix).Allofthesedataformatswouldbegoodcandidatesforlongtermdataseriesonforest characteristics,andifnecessarytheycouldbefusedwithotherdatatoincreasetheirutility(e.g., Hansenetal.2008b). Thesedataformatscurrentlyhavetheadvantageofbeingfreelyavailable,whichfacilitates programfinanciallongevityandincreasestransparency.Althoughseveralcommercialsatellite series(SPOT,GeoEye)alsohavelongterm,highqualityimageryarchives,therightstorelease theseimagestothepublicarelimited.Theroleofthesedatainaglobalmonitoringprogram requiresfundingandameansofallowingthedatatobesharedtomaximizetheirusefulness.

3.3 Developing and Evaluating Reference Forest Maps


Todetectchangeinforestcharacteristicsofinterest,baselinereferencemeasurementsare required.Biomassestimatesderivedfromremotesensingarerareformanyregionsandaccurate

FAGAN AND DEFRIES

35


historicalreferencemapsofforeststructureorbiomasscurrentlydonotexist(Olanderetal.2008; Gibbsetal.2007;and,e.g.,Baccinietal.2008).Bycontrast,itispossibletodevelopmoderate resolutionreferencemapsofforestareadatingto1972(Olanderetal.2008).Formoredetailon thistopic,seeChapter4.

3.3.1AccuracyinForestMapping
Accuracyconsiderationsareparamountindetectingchangesintheworldsforests(Grainger 2008;Gibbsetal.2007).Therecanbeerrorsofomission(notdetectingorlocallyunderestimating forestquantities)andcommission(falsepositivedetectionsorlocaloverestimatesofforest quantities).Wereportherethegreaterofthetwotypesoferrorwhenassessingprediction accuracy.Forforestarea,accuracyisdefinedasthepercentageofpixelsintheremotesensing imagerythatarecorrectlyidentifiedwithrespecttolandcovertype,sothelowestoverallaccuracy (meanomissionorcommissionerroracrosslandcoverclasses)isusedwherepossible.Forforest volume,biomass,andcarbon,wereportthematchbetweenpredictionsfromremoteimageryand observedgroundmeasurements(i.e.,thepercentageofvarianceingroundtruthdataexplainedby regressionequations).Ingeneral,remotesensingestimatesofforestareahavehighaccuracy3while estimatesofforeststructureandbiomassarelessaccurate(DeFriesetal.2007;Olanderetal.2007; Rosenqvistetal.2003).

3.3.2ReferenceMapsforCarbon
Approximately50percentofvegetationbiomassiscarbonbutthereisgreatuncertaintyin estimatesofglobalbiomassstocks(seeSection4.4).Theguidelinesissuedbythe IntergovernmentalPanelonClimateChange(IPCC)describethreeacceptabletiersofforest measurement.Eachtierisincreasinglyaccurateandprecisebutalsomorecomplexandexpensive inthemonitoringrequirements(GOFCGOLD2008;seeTable5foranoverview).Conservative principlesofestimationcanprovideusefulestimatesofforestareaandbiomassforpolicymakers (Achardetal.2007;Grassietal.2008;GOFCGOLD2008).Conservativeprinciplesdictatethat,a tierlowerthanrequiredcouldbeusedoracarbonpoolcouldbeignoredifitcanbedemonstrated thattheoverallestimateofreducedemissionsare[sic]likelytobeunderestimated(GOFCGOLD 2008,12).Wherelargelevelsofuncertaintyexistinbiomassestimates,discountingtheestimated forestcarbonwouldbeanappropriateconservativeresponse.Forexample,ifestimatesofforest carbonvarybetween60and90tonsperhectare,60couldbeusedasadefaultregionalvalue(Tier 1)orcouldbefurtherdiscountedto40mt/hatoaccountforuncertainty.

3.3.3ForestInventoryandReferenceMaps
Becauseforestinventorydatacancurrentlyestimatelocalbiomassmoreaccuratelythan remotelysenseddata,referencemapsforforestbiomassandotherquantitiescouldbedeveloped fromgrounddatabyinterpolatingbetweenmeasurementplots(Gibbsetal.2007).Thisapproach

Inthisreport,theoverallaccuracyofsatelliteimageryanalysisisdesignatedbyfivestandardadjectives:veryhighaccuracy (>90percent),highaccuracy(>80percent),acceptableaccuracy(>70percent),lowaccuracy(5070percent),andpoor accuracy(<50percent).Wherealternativequalifiers(e.g.,mixed)areused,weprovidetheactualaccuracyisprovided(e.g., 67percent).

FAGAN AND DEFRIES

36


hasmanyadvantageswhereforestinventorydataexistanditcanbereadilyintegratedwith satellitedataonforestarea(Gibbsetal.2007). Table5.IPCCTierDescriptionforCountryLevelCarbonReporting
Tiersforemissionsfactors:ChangeinCStocks 1.IPCCdefaultfactors. 2.Countryspecificdataforkeyfactors. 3.DetailednationalinventoryofkeyCstocks,repeatedmeasurementsof keystocksthroughtimeormodeling.
Note:Tier3isthemostdetailed. Source:GOFCGOLD2008.

Figure14.ExampleForestTypeMapthatCouldBeUsedforStratificationofForestInventoryPlots

Source:GOFCGOLD2008.

FAGAN AND DEFRIES

37


Strictplottoplotinterpolationacrosslargeareasisinaccurateasitignoresspatialvariationin foresttypesandages(Houghtonetal.2001;Gibbsetal.2007).Stratifyingforestinventoriesacross arangeofforesttypesandages,andtheninterpolatingonlywithinthemeasuredforesttypesand ages,increasestheaccuracyofforestinventoryreferencemaps(Gibbsetal.2007;seeFigure14). Thestratificationapproach,althoughmoreaccurate,requiresregionalmapsofforesttypeand forestagefromremotesensinganalyses.Remotesensingprovidesestimatesofforestareaquite well;however,distinguishing(orclassifying)differentforesttypesandageshashadvariable accuracy.Imagefusionsandnewimageanalysismethodsmayincreasetheaccuracyofforest classifications.ThesemethodsarediscussedfurtherinSection4.1.

3.4 Survey Methods for Determining Forest Area and Type


Threedistinctstrategiescanbeemployedinusingremotesensingtomeasureforestsglobally: gridsampling,changestratifiedrandomsampling,andwalltowallsampling(Olanderetal.2008; Achardetal.2007;DeFriesetal.2007).Ingridsampling,moderateresolutionimagesamplesare analyzedatregularspacingacrossalargearea(seeFigure15).ExamplesincludetheFAOs2010 ForestResourcesAssessmentwhichwillsampletheglobeatonedegreeintervals.Theadvantageof gridsamplingisthatitgivesrepresentativeestimatesoflandcoveroverlargeareaswithrelatively littleimageprocessing.Themaindisadvantageisthattheapproachmayundersampledeforested areasbecausesuchareasarenotrandomlydistributedandarelikelytobeconcentratedinjusta fewareas(Achardetal.2007).Areastratifiedsamplingisrandomsamplingthatisstratifiedby foresttype,forestcharacteristics,orlandcoverchange.Inchangestratifiedsampling,moderate resolutionsamplingisheaviestinareasthatcoarseresolutionimageryhasindicatedischanging rapidly.AnexampleistheNASALandCoverandLandUseChangeprogramatSouthDakotaState University.TheprogramusesfrequentMODISimagerytodetectchangeandthenstratifiesLandsat samplestomakeaccurateestimatesoflandcoverchange(Hansenetal.2008a;seeFigure16a). MODIScannotbeuseddirectlytoestimatelandcoverchangesincemostchangeoccursatsubpixel scales(Achardetal.2007;SanchezAzofeifaetal.2009).Theadvantagesofchangestratified randomsamplingarethatitmaximizeschangedetectionforagivensetofimagingprocessing resourcesandyieldsunbiasedmeasurementsofforestchange.Thisapproachislesssuitedfor samplingcharacteristicsofstandingforests. Thethirdapproach,walltowallsampling,completelymapsaregionwithimagery.Examples includeBrazilsPRODESprogram(seeFigure16b)andIndiasbiennialforestsurveyaswell ongoingeffortsintheUnitedStates(NLCD),Europe(CORINE),Canada,NewZealand,Australia,and SouthAfrica(DeFriesetal.2007;Heroldetal.2007).Walltowallsamplingisidealfordetecting landcoverchangeandshiftsindeforestation(oftenreferredtoasleakage)betweencountries,but themethodislogisticallychallengingintheneedtoacquire,process,andanalyzethenumberof imagesrequiredtocoverthegloballandarea.

FAGAN AND DEFRIES

38


Figure15.GridSamplingwithModerateResolutionImages

Note:EachdotisthecenterofasingleIRSimage. Source:GOFCGOLD2008.

Figure16a.ExampleofResolutionStratifiedSampling

Note:Moderateresolutionsamplesareconcentratedinareasofchange(red)determinedfromcoarserresolutionimagery. Source:GOFCGOLD2008.

FAGAN AND DEFRIES

39


Figure16b.ExampleofWalltoWallSamplingwithModerateResolutionImageryinBrazil

Note:Noticethedifferenceinilluminationandcloudcoverbetweenmoderateresolutionimages. Source:GOFCGOLD2008.

3.5 Price and Logistics


Thelargequantitiesofimageryassociatedwithglobalforestmeasurementwillnecessitate budgetarychoices,automatedprocessingchains,anddeliberationonselectionofresampling intervals(Achardetal.2007;DeFriesetal.2007;GOFCGOLD2008;Olanderetal.2008;Rosenqvist etal.2003).Table3summarizesthecostsofimageryrequiredtodowalltowallsamplingofglobal forestsatdifferentresolutions.Highresolutionimagerywouldbeveryexpensiveatcurrentprices, asdiscussedindepthinSection4.2.2forLIDAR.Automatedanalysisofimageryisquickand standardizesresults,butthelargenumberofimagesrequiredtocovertheworldsforestsat moderateresolutionwillrequirelargeprocessingresourcesforautomatedclassification (Anderssonetal.2009;DeFriesetal.2007;Mayauxetal.2005).Althoughmoderncomputersare capableofprocessingthisamountofimagery,eithersignificantinvestmentorcomputing partnershipsarerequired(GOFCGOLD2008).Theresamplingintervalcanbeaffectedbythe amountoftimerequiredtoprocesseachyearsimagery;theBrazilianNationalInstituteforSpace Research(INPE),forexample,prioritizeshighchangeareasforearlyprocessingtoensuretimely revisits(Achardetal.2007). Decisionsonhowoftentosampletheworldsforestsshouldtakeintoaccountprocessingtimes andtheavailabilityofcloudfreepassiveimagery(GOFCGOLD2008).Cloudfree,globalLandsat mosaicscontainimageryeverythreetofiveyearsbutcloudfreeMODISmosaicscanbecreated seasonallyduetotheirhighreturntime(e.g.,Mortonetal.2005).Cloudspersistinsomeareaseven inostensiblycloudfreeimages,necessitatingfusionwithSARimagerytopenetratecloudsin certaintropicalareas(Olanderetal.2008;Lindquistetal.2008;Rosenqvistetal.2008).TheALOS forestmapperprogram,notably,willproduceglobal,cloudfreeforestcovermapsannuallyfrom 2007onward(J.Kellndorfer,pers.comm.).Withprocessingandclouddelays,aglobalforest monitoringeffortmaybecapableofproducingannualorbiannualforestmapsatcoarseresolution; withouttheuseofSAR,moderateresolution,cloudfreemapsmaytakeseveralyearstoproduce.

FAGAN AND DEFRIES

40


Itisexpensivetointegrateremotesensingestimateswithforestinventorydata,LIDARdata, andhighresolutionimageryyetselectivegroundtruthingisabsolutelycriticaltoevaluateaccuracy (Gibbsetal.2007;GOFCGOLD2008;Olanderetal.2008;Rosenqvistetal.2003).Highresolution imagerycollections,includingGoogleEarth,canbeusedassubstitutesforgroundtruthingland coverestimates(e.g.,Gibbsetal.2007;Readetal.2003;Helmeretal.2009;Bicheronetal.2008), butgrounddataarerequiredforbiomassandcarbonestimates(Gibbsetal.2007). Incontrasttolocallyaccurateanalyses,forwhichconclusionsaregeographicallyrestricted, globalremotesensinganalysessacrificelocalaccuracyforgreaterglobalaccuracy.Aglobal, dispersednetworkofgroundtruthdataisnecessarytoquantifythisaccuracy(e.g.,Bicheronetal. 2008).Largeamountsofgroundtruthdataalreadyexistinsomeregionsandforestersandforest ecologistscontinuetocollectmoreinmanyorganizedforestsamplingprograms(Herold2009). Collectingandcoordinatinggroundtruthdataforglobalforestmeasurementwouldbeauseful scientificundertaking(Heroldetal.2008).

FAGAN AND DEFRIES

41

Chapter 4. An Overview of Remote Sensing Capabilities for Forest Measurement: Current and Near-Term Technologies
Intheprecedingchapterswenotedseveralkeylimitationsofremotesensing:(a)tradeoffsin swathandresolutionincurrentsensortechnology(Section3.2),(b)thesmallnumberofactive sensorscurrentlyinorbit(Section2.4.1),(c)therelevanceofinexpensiveimageryforfrequent globalcoverage(Section3.5),(d)thewisdomofredundancyinsatelliteselection(Section2.4.3), and(e)theneedforgroundtruthingtovalidatetheaccuracyofconclusionsfromremoteimagery (Section3.5).Inlightoftheselimitations,wenextconsiderthecapabilityofremotesensingto measuretheglobalforestspecifically,globalforestarea,foreststructure,andotherforest propertiesinTable4.

4.1 Forest Area: Current Measurement Methods


Globalforestareaisoftenmeasuredastwoclasses(forest/nonforest)orbinned(thatis, categorized)intohomogenousforesttypesthatdonotdistinguishtreeplantationsordisturbed forests(Hansenetal.2008a;Bicheronetal.2008).Forestsarequitedistinctfrommostnonforest typesoflandcover,soforest/nonforestareacanbemeasuredwithopticalandSARsensorswitha highdegreeofaccuracy(from80>90percent).Distinguishingmorethantwolandcoverclasses, suchasforestsofdifferentagesorcomposition,canstillresultinhighaccuracy,butclassification accuracyusuallydecreaseswithanincreasingnumberofforestclasses(Foody2002)orwhena largerareaisexamined(Olanderetal.2008).Theseresultsmeanthatremotesensingisideally suitedtodetectforestremoval(deforestationorclearcutting)butlesswellsuitedtodetectforest thinningorforestreplacementbyindustrialtreeplantations(DeFriesetal.2007;Sanchezetal. 2009).Challengesremainindistinguishingprimaryforestsfromtreeplantationsandolder secondaryforests(Kimesetal.1999;Sanchezetal.2009,Songetal.2007;Thenkabailetal.2004), andindetectingforestdegradation,whichisareductioninforestcanopycoverorbiomassthat doesnotresultincompleteforestclearing(DeFriesetal.2007).Weexplorethesetopicsfurther below.

4.1.1SensorTypesandMeasuringForestArea
Eachsensortypehasdifferentstrengthsandweaknessesformeasuringforestarea.Coarse resolutionsensorshavethegreatestcapabilityforglobalcoverageandhighreturntimes(Achardet al.2007;Rosenqvistetal.2003),buttheirpixelsizecausesthemtomissthemajorityof deforestationevents(Olanderetal.2008;DeFriesetal.2002;Sanchezetal.2009;Mortonetal. 2005).Highresolutionsensorsenablemeasurementoftreecrownareaandfinedelineationof forestcompositionanddisturbance,buttheirsmallswathslimittheirutilityforrepeated,cloud freeglobalmeasurement(Anderssonetal.2009).Bybalancingalargeswathsizewithspatial resolutionthatisabletodetectthemajorityoflandusechange,moderatepassivesensorsare consideredthebestcompromiseforregionallandcovermonitoring(Achardetal.2007;Olanderet al.2008;Anderssonetal.2009).Finally,hyperspectralandpolarizedSARsensorshaveimproved theabilitytodistinguishamongforesttypesandmapforestcover(DeFries2008;Thenkabailetal. 2004;HoekmanandQuinones2000).Analyzingtheirimagerycanbetechnicallydifficult,however (Kasischkeetal.1997;Ustinetal.2004).Futurerefinementsinimagefusiontechniquesandnew satellitetechnologycanbeexpectedtoimprovemeasurementsofforestareaandtype.

FAGAN AND DEFRIES

42

4.1.2ForestAreaversusMeasuringChangeinForestArea
Measuringforestareaisdistinctfrommeasuringchangesinforestarea,forbothpracticaland quantitativereasons.Practically,increasesinforestareaoftenresultfromlandcovertypesthatare quitespectrallydistinctfromtheoriginalforestandwouldnotbeclassifiedasforestarea.In temperateandtropicalareas,woodyencroachmentintograsslandscreatesspectrallyand structurallydistinctforestsinunexpectedareasandforestregrowthonabandonedfarmscreates distinctsecondaryforests(Houghton2005).Deforestationresultsintheconversionofforeststoa varietyofagriculturallandcovers,includingspectrallysimilartreeplantations.Tropicalforest regrowthisveryrapid,makingtentotwentyyearoldforestsdifficulttodistinguishfromprimary forestonasatelliteimage(seeSection4.1.5). Fromaquantitativestandpoint,measuringchangeinforestareaisdistinctfromclassifyingtwo landcovermaps.Toillustratewhy,considerthemostbasiclandcoverchangemethod,post classification;quantifyingthechangesbetweentwoclassifiedlandcovermaps.Inthismethod,high classificationaccuracyiscriticalbecausequantifyinglandcoverchangebycomparingtwoland covermapsmultipliestheirrespectiveerrors(Luetal.2004).Amultitudeofothermethodsof quantifyinglandcoverchangeovertimearereviewedatlengthbyLuetal.(2004).Thereare severalgeneralmethodsforgloballandcoveranalysis.Imagedifferencingsubtractstwoimages andthenclassifiestheareasthathaveexperiencedchangesinspectralcharacteristics(e.g., greenness).Combinedanalysisfusesbothimagesinthesamedataset,andthenclassifiesthe combineddataset,labelingthechanges.Hybridanalysisidentifieswhichpixelshavechanged betweendatesandclassifiesonlythechangedareas(Luetal.2004). Inthediscussionofmeasuringforestareabelow,wefocusprimarilyontheaccuracyof classifiedlandcovermapsbecausedistinguishinglandcovertypesisanecessaryfirststepin accuratelyclassifyingchangesinforestarea.

4.1.3CoarsePassiveSensorsandForestArea
Therehavebeennumerousglobalandregionaleffortstomapforestsusingcoarseresolution sensors,andtheseareablyreviewedbyothers(seeTable4inHeroldetal.2008;Table1inMayaux etal.2005;andTable2inAchardetal.2007).Effortstocreategloballandcovermapshaveused foursatellites:AVHRR,MODIS,SPOTVegetation,andEnvisat/MERIS.Twomaintypesofproducts haveresultedfromthesemappingprograms:landcoverclassificationsandtreecovercontinuous fields(AVHRRandMODISonly).Alloftheseproductshavehistoricallybeenfreelyavailablefor download,increasingtheiruseingloballandcoveranalysis. Thetreecovercontinuousfieldsmapsareapartofanefforttoproducecontinuous,subpixel estimatesofvegetationtraitsfromAVHRRandMODISdata(Hansenetal.2003).Theseeffortsuse higherresolutionLandsatandIKONOSdatatotrainanalgorithmthatcalculatespercenttreecover foreachcoarseresolutionpixel.Thiscoarsefinefusionmethodyieldsestimatesofforestcoverthat canbeadaptedforanyforestdefinitionandiscapableofcoarsedistinctionsofforestcoverbasedon leaftypeanddensity(Hansenetal.2003). Landcoverclassificationsassignonelandcoverclassperpixelandhavebeenongoingsince 1992(AVHRR),2000(MODIS,SPOTVegetation,and2005(MODIS,SPOTVegetation,and Envisat/MERIS)(Heroldetal.2008).CurrentlytheEnvisat/MERIS300mresolutionGLOBCOVER

FAGAN AND DEFRIES

43


productisthehighestresolutiongloballandcovermap.Mapsoftropicalforestshavealsobeen producedat250mresolutionfromMODIS(Bicheronetal.2008;Hansenetal.2008a).Evenjustas staticsnapshots,theseglobalmapshavebeenimportantinclimateandcarbonmodeling(DeFries 2008;Achardetal.2007). Thesecoarseresolutionmappingeffortshaveseverallimitations.Thenumberoflandcover classesdiffersbetweenmappingeffortsandlandcoverclassificationsareoftendifficulttoassignin disturbed,heterogeneous,ortransitionalecosystems(Heroldetal.2008;Mayauxetal.2006). Simple,coarseresolutionforest/nonforestclassificationshavebetweenacceptableandhigh accuracy(LatifovicandOlthof2004;Hansenetal.2003).Coarseresolutionclassificationswith multiplelandcovercategorieshavehadpoortolowaccuracyinrepresentingactuallandcover (e.g.,GLOBCOVERhas67percentaccuracy)(Bicheronetal.2008;LatifovicandOlthof2004; Mayauxetal.2006).Theselandcoverclassifications(e.g.,GLC2000;GLOBCOVER)donotalways agreewitheachother(Girietal.2005).Figure17indicatestwodifferentlandcovermaps,the GLC2000andMODIScontinuousfields,whichdisagreemarkedlyinCentralAmerica.Thescientific effortGeowiki.orgistryingtousevolunteerinputandhighresolutionimagerytoreviseconflicts betweengloballandcovermapsandimprovetheirclassificationaccuracy(IIASA2009). Thecapacityofcoarseresolutionsensorstodetectforestclearingandforestfiresinalmostreal timeisparticularlyimportanttoslowingdeforestationinthetropics,whereitisdifficulttomonitor remoteforests.Weexaminethislaterinthechapter(Section4.4.2).Thefrequentreturntimeof MODISpermitsphenologybasedmappingoftropicaldeforestationwith89percentaccuracy,for example(Mortonetal.2005).Withonedayimageprocessing,INPEusesMODISimageryinareal timedeforestationpreventionprogramcalledDETEXtopreventlargescale,illegalclearinginthe BrazilianAmazon(DeFries2008;Herold2009).Coarseresolutionimagerycanonlydetect completeclearingandonlydetectsalowpercentageofactualdeforestationthatisgreaterthan10 20hainsize(Sanchezetal.2009;Achardetal.2007).InCostaRica,forexample,only10percentof deforestationwasdetected(Sanchezetal.2009).Coarseresolutionimageryisbestused,therefore, asadetectionorsamplingtooltotargethigherresolutionsensors(i.e.,thestratifiedsampling programmentionedinSection3.4)(DeFriesetal.2007). Theavailabilityoffrequent,free250300mresolutionimageryhasimproveddeforestation detectionandlandcovermappingbecausetheimagerymatchesthescale(6.25to9ha)oflarge scaleforestclearingsforfarmingandranchingandtheapproachcandetectchangesofaround10 20hainsize(Figure18,Achardetal.2007).TheNPOESSsatelliteseriesisintendedtoreplacethe MODISandAVHRRsatellites,butNPOESSwillhave400mto1kmresolution(Townshendand Justice2002)limitingittodetectonlyextremelylargeclearingsbetween16hato100hainsize. TheJapaneseGCOMC1(with11bandsat250mresolution)andtheEuropeanSentinel3series (with21bandsat300mresolution)maybebettersuitedtocontinuetheMODISand Envisat/MERISrecord(Appendix).Itwillbedifficulttoequalthecombinationofresolution,revisit time,andfreeimagerythattheMODISTerraandAquaconstellationcurrentlyofferstogloballand coveranalysis.

FAGAN AND DEFRIES

44


Figure17:CoarseResolutionClassificationDisagreements

Notes:ForestcoverfromMexicotoPanamaclassifiedbyGLC2000andbyMODIS(A)andthedisagreementbetweenthem(B). TheredcircleidentifiesahotspotofdisagreementinGuatemalaandElSalvador.Thewebtoolallowsonetovalidatethese disagreementsusinghighresolutionsatelliteimageryandGeoTaggedPictures(forexample,Panoramio)(C). Source:CreatedusingGeoWiki,2009.

FAGAN AND DEFRIES

45


Thefrequentrevisittimeofcoarseresolutionsensorsallowsforneardailymonitoringof phenology,importantforlandcoverclassificationandfordetectingfutureclimatechangeas describedinChapterOne(Mortonetal.2005;Goetzetal.2005).Repeatedpassesthatcapture seasonalleafdynamicscanimprovedetectionofseasonaltropicalforests(e.g.,Mortonetal.2005) andcoarseresolutionimagerywillremainusefulforforestmeasurementwellintothefuture. Giventheusefulnessofthevegetationcontinuousfieldsandthepotentialforautomatedsubpixel detectionsofdeforestation(e.g.,DETER),coarsefineimageryfusionsareanexcellentmethodto increasetheresolutionofcoarseresolutionimagery(e.g.,Hansenetal.2003).Nextgeneration geostationarysatellites(GOESR,FY4)areexpectedtoimproveto1kmresolutionintheVNIR sometimearound2014.Thischangewouldenablerealtimemonitoringoflargescale deforestation,fires,andphenologyandcouldprovidedramaticallyimprovedinputstocarbonand climatemodels(Chuvieco2008;DeFries2008).

4.1.4ModeratePassiveSensorsandForestArea
Moderatepassivesensorshaveresolutiononascaleappropriateforforestmanagement(Figure 18).Thesensorsareabletocaptureforestchangesatthescaleofonehectareinlocalanalysesand twotofivehectaresinregionalanalyses(Olanderetal.2008;Achardetal.2007).Forest/nonforest covercanbedistinguishedbymoderatesensorswithgreaterthan90percentaccuracyinlocal areas.Overlargerareasaccuracydeclinestobetween85and90percent(Olanderetal.2008). Severalforestmappingeffortsusemoderateresolutionimages;theseincludesubnationaland nationalwalltowallmapping(Achardetal.2007;Heroldetal.2008)andglobalsamplingefforts (Mayauxetal.2005).Herold(2009)reviewsnationalmappingeffortsextensively. Globalwalltowallcollectionsofmoderateresolutionimagerycouldbeconstructedat considerableexpensefromIRS,SPOT,orDCMdata,buttheLandsatglobalmosaicsaretheonly free,publiclyavailableglobalimagery(Achardetal.2007).Asof2009,cloudfree,globalLandsat mosaicsfor1975,1990,2000,and2005havebeencreatedandeffortsareunderwaytocreatean errorfreeLandsat7mosaicfor2010(Lindquistetal.2008;Hansenetal.2008b).TheLandsat mosaicscontaindatemetadataforeachcomponentimagetoallowforcorrectinterpretationfor examplewhencalculatinglocaldeforestationrates.TheseglobalLandsatmosaicshavebeen suggestedbyseveralauthorsasexcellentsourcesforforestareareferencemapsforREDD(Gibbset al.2007;Olanderetal.2008;Achardetal.2007). AlthoughtheglobalLandsatmosaicsofferopportunitiesformappingglobalforests,numerous challengesariseinanalyzingthemforforestarea.Becauseofobstructionsincoverageincloudy areas,mostnonconstellationmoderatesensors(i.e.,allexcepttheIRSandDCMseries)havede factoreturntimesofayearormoreintropicalareas(Asner2001;Olanderetal.2008).Thislimit canseriouslycompromiseeffortstomapseasonalforests,especiallyintropicalareas,becauseof thedifficultyindistinguishingleaflessforestsduringdryseasonswhenmostimageryisacquired (Sanchezetal.2009;Asner2001).TodeveloptheglobalLandsatmosaics,scientistswereforcedto knittogetherimagestakenoverseveralyears(e.g.,19861991forthe1990mosaic).Eventhen unacceptablelevelsofcloudcoverpersistedonabout2530percentoftheLandsatmosaicimages inEcuador(Olanderetal.2008)andin16percentofmosaicimagesintheCongoBasin(Lindquist etal.2008).Asidefromthedetectionofseasonalforests,automatedclassificationofLandsat

FAGAN AND DEFRIES

46


mosaicsrequirescorrectionforthevarietyofsunilluminationangles(knownasradiometric correction)andthecreationofglobalorregionalalgorithmstoclassifydifferentlandcovers(Luet al.2004;Schroederetal.2006;Bicheronetal.2008).Landsatclassificationalgorithmsdevelopedin oneregionareunlikelytobeaccurateinanotherregioneveniftheregionsaresimilar(e.g.,Foody etal.2003). Figure18.ResolutionofSensorsRelativetoEcologicalMeasurements

Notes:Thedottedboxesindicatethetemporalandspatialscaleofecologicalprocessesandhumandisturbances.Thesolid coloredboxesshowthespatialscaleandtemporalrecordlengthofselectedsatelliteandairbornesensors(GOESdullred; MODISgreen;LiDAR/Radarpinkpurple;Landsat/EO1yellow;IKONOS/Quickbird/Airbornelightblue). Source:Chambersetal.2007.

Regionalwalltowallmappingefforts,globalsamplingefforts,andoneglobal,moderate resolutionlandcovermap(GeoCoverLC;Cunninghametal.2002)indicatethatthesechallenges couldbeovercome.ThecurrentmoderateresolutionGeoCoverLCmaphasmixedglobalaccuracy (72percent)andverybroadforestclasses(Cunninghametal.2002).Anaccurate,detailed30m resolutionforestmapwouldbeasignificant,tenfoldimprovementingloballandcovermapping (Bicheronetal.2008).Reprocessingtheglobal30mLandsatarchive(1983present)tocreatea seasonappropriate,cloudfree,andradiometricallycorrectedimagerytimeseriesisfeasiblewith currenttechnology,althoughtheprojectwouldbecomputationallyintensive.Algorithmshavebeen developedrecentlytoeliminatecloudsandcloudshadowsandimprovetheLandsatmosaicsby fusingtogether,orcompositing,timeseriesofLandsatandMODISimages(Lindquistetal.2008; Royetal.2008;Hansenetal.2008b).Inaddition,automatedradiometriccorrection,whichisoften difficultinallforestimages,hasbeendemonstratedonaseriesofLandsatimages(Schroederetal. 2006;Xianetal.2009).

FAGAN AND DEFRIES

47

4.1.5ModerateResolutionImageryandLandCoverClassification
Theliteratureonlandcoverclassificationwithmoderateresolutionimageryisextensive (partiallyreviewedinSanchezetal.2009;Patenaudeetal.2005;Olanderetal.2008;andCoppinet al.2004).Becauseclassificationaccuracygenerallydecreaseswithanincreasingnumberofclasses (Foody2002),weusethelandcoverclassificationhierarchyofAndersonetal.(1976)when referringtotheaccuracyofdifferentclassifications.Inthishierarchy,aLevelIclassification distinguishesbasicclasses(forest,nonforestvegetation,agricultural,urban)andaLevelII classificationdistinguishesmorelandcoverclasses,includingforesttypes.Asmentionedabove, LevelIaccuracyinexcessof85percentisusuallyobservedwithmoderateresolutionimagery,with higheraccuracyathigherresolution(SalajanuandOlsen2001;Petersonetal.2009).Generally, accuracydecreaseswithanincreasingnumberofclasses.PublishedLevelIIclassificationshave rangedfrom65to85percentaccuracy(referencesinPatenaudeetal.2005).Ingeneral,extending LevelIIclassificationalgorithmsdevelopedinoneregiontootherregionsisatechnicalchallenge thatisnotalwaysmet(Sanchezetal.2009;Lovelandetal.2002;Xianetal.2009;Roganetal.2008; Foodyetal.2003). Distinguishingamongforesttypes,forestages,degradedandintactforests,andtreebased agroecosystemscanalsobechallenging(Sanchezetal.2009;Olanderetal.2008;Patenaudeetal. 2005).Intemperateforests,SalajanuandOlsen(2001)wereabletoclassifyaforestedareain MichigantoLevel1with8991percentaccuracy,butaccuracydecreasedto7784percentfor distinguishing10foresttypes.InSiberiasBaikalregion,Petersonetal.(2009)wereableto distinguishfourborealforestclasseswith8098percentaccuracy.Intropicalforests,Thenkabailet al.(2004)foundthatmoderateresolution,multispectralsensorshadpoor(4050percent) accuracyindistinguishingnineforesttypesandages.Bycontrast,Sesnieetal.(2008)wereableto discriminate17forestclasseswith93percentaccuracyusingafusionofspectral,DEM,andclimatic GISlayers.Sesnieetal.(2008)had69percentaccuracywithspectralclassesalone,indicatingthat detailedLevelIIclassificationintropicalareasmayrequireGISfusionorhyperspectraldata.We explorethisfurtherbelow(Section4.1.6). Inbothtropicalandtemperatezones,moderateresolutionimagerycannotdistinguishforest regrowthfrommatureforestafter1530yearsofgrowth(Steininger1996;Nelsonetal.2000; Songetal.2007;FiorellaandRipple1993).Partiallyloggedorclearedforestshavelongbeen difficulttodistinguishfromintactforests(Saderetal.2003;WilsonandSader2002;Achardetal. 2007),butrecently,promisingspectralanalysistechniqueshavebeendevelopedtomapsubpixel forestdegradationintheBrazilianAmazonusingmoderateresolutionimagery(Asneretal.2005; Souzaetal.2005b). Inthetropics,distinguishingforestsfromtreebasedagriculture(i.e.,agroforestry,tree plantations)isdifficultusingmoderateresolutionimagery;theplantationsoftenappearsimilarto secondaryforests(Sanchezetal.2009).InMalaysia,BabanandYosuf(2001)wereableto distinguishrubberplantationswith74percentaccuracyandfoundthatplantationsweremost oftenconfusedwithintactforests.Sesnieetal.(2008)wereabletodistinguishtreeplantationswith approximately90percentaccuracyinCostaRica,butaccuracydecreasedto55percentwhenthe entireimagewasexaminedacrossdiverseplantationtypes.Kuplichetal.(2000)wereableto distinguishplantedEucalyptusandPinusforestswith89percentaccuracyinafragmented

FAGAN AND DEFRIES

48


Brazilianagriculturallandscape.IneasternEcuador,Santosetal.(2008)usedLandsat7todoa LevelIIclassificationofanoilpalmsecondaryforestlandscapewith77percentaccuracy. DespitethesechallengesinLevelIIclassificationofforests,therehasbeensignificantprogress inclassificationofmoderateresolutionimageryinrecentyears(Coppinetal.2004).Onepromising approachistheclassificationoflandcovertrajectoriesovertime(ratherthanstaticimages)made possiblebytherecentopeningoftheLandsatimagearchive(Kennedyetal.2007;Maseketal. 2008;Coppinetal.2004;Songetal.2007).Forexample,Helmeretal.(2009)mappedageclassesof tropicalforestregrowthwithveryhighaccuracybyusingtheentire,37yearLandsatarchivetoage regrowthonpreviouslyclearedpixels.Thisnewtrajectoryclassificationmethod,inadditiontothe compositingmethodsmentionedabove(Hansenetal.2008b),underscorestheimportanceoflong, continuoustimeseriesofmoderateresolutionimagerytolandcoveranalysis(andthecritical natureofthesuccessoftheLDCM).

4.1.6HyperspectralSensors:ForestAreaandBeyond
Hyperspectralimages,althoughtheyhavelimitedglobalcoverage,aremarkedlybetter(e.g., 4070percentbetter)thanmultispectralmoderateresolutionimagesindistinguishingand mappingdiverseforesttypesandotherlandcovers(Thenkabailetal.2004;UstinandXiao2001; Goodenoughetal.2003)..Eventhoughthepotentialisstillbeingexplored,hyperspectralimages alreadyhaveawidediversityofapplicationsduetotheabilitytomeasuretheabsorptionoflightby chemicals.Thisabilityinturnallowsdirectmeasurementofleafchemistryandthequantificationof photosyntheticandnonphotosynthetic(woody)cover(AsnerandMartin2009;Chambersetal. 2007;Ustinetal.2004).HyperspectralimageryimprovesmeasurementofLAI(Leeetal.2004; Schlerfetal.2005),forestproductivity(viacanopynitrogen;Smithetal.2002;OllingerandSmith 2005),canopystructure(ArroyoMoraetal.2009),anddroughtstress(Asneretal.2004).The imageryalsomakesnewmeasurementsofforestdiversitypossibleincludingcanopyleaf ecophysiology(Asneretal.2009b)andthediscriminationofindividualspeciesandspeciesgroups (Goodwinetal.2005;Asneretal.2008;Clarketal.2005).Moderateresolutionhyperspectral imagery,whichwillbecomewidespreadinthenextfewyearsastheEnMap,HyspIRI,andseveral othersatellitesarelaunched,hasthepotentialtorevolutionizeforestbiodiversitymeasurementsin anewscienceofremotespectranomics(AsnerandMartin2009).Althoughmanynew hyperspectralsensorswillhaverelativelynarrowswathsandmaybeunsuitedforglobalforest measurement,theHyspIRIsensorwillhavea90kmswath(Appendix)andislikelytomarkedly improvetheaccuracyoflandcovermapping.

4.1.7HighResolutionPassiveSensors:ForestAreainFocus
Becausehighresolutionpassivesensorshaveanarrowswathandpointable,interrupted coverage,therearecurrentlynocontinuousgloballandcovermapsorcloudfreeimagemosaicsat highresolution(Loarieetal.2007).GeoEyesconstellationwillbemappingtheentireglobeinthe nextfewyears(MarkBrender,pers.comm.),butcloudfreeimageryisunlikelytoresult(Asner 2001;Olanderetal.2008).SPOT5hascontinuousglobalcoverageat10mresolutionbutthe imageryhasnotbeenanalyzed.RapidEye,ahighresolutionconstellationwithdailyrepeatcapacity (launchedin20082009),hasthepotentialtodevelopcloudfree,continuousglobalmaps (RapidEye2001).

FAGAN AND DEFRIES

49


Currently,GoogleEarthhasthelargestcollectionofglobalhighresolutionimagery,butthe imageryisfromavarietyofsatelliteandairbornesourceswithdifferingresolutionsand undiscloseddates(Potere2008).GoogleEarthmaynotyetbeusefulforstrictchangedetectionona globalscale,buttheapproachisveryusefulforlandcovervalidationandvisualization(e.g.,IIASA 2009;Bichersonetal.2008;Helmeretal.2009;Olanderetal.2006).Thesameresultistruefor highresolutionimageryingeneral:theimageryisaninefficientwaytomapglobalforestsdueto irregularspatialsampling(Loarieetal.2007;Anderssonetal.2009),relativelyhighcost(Olander etal.2008;Anderssonetal.2009),datasharingrestrictions,andradiometriccorrectionchallenges (e.g.,variablecanopyshadowing)(Goetzetal.2003;Anderssonetal.2009).Currentpoliciesabout accesstodatarestrictthesharingoforiginaldatabutallowsharingofdataproducts,suchasbasic maps,foralmostallhighresolutionimagery.However,theimageryisusefulforidentifyinglocal landcovers(Wulderetal.2004)andforestdisturbance(Thompsonetal.2008;Wulderetal.2008; deSouzaandRoberts2005),andevenfollowingthepopulationdynamicsofindividualtreespecies (Clarketal.2004a).Highresolutionimageryhasimmenseutilityforgroundtruthinglower resolutionimagery,bothasafinescalevalidationtoolandastrainingdataforalgorithms(e.g., Hansenetal.2003;Wangetal.2005). Althoughhighresolutionimagerydoesnothavethespectralresolutionofothersensors(most notablymissingtheSWIRbands),theimagerycontainsadditionalinformationontheshapeand textureofobjectsinthelandscapethatlowerresolutionimagerylacks(Goetzetal.2003).The accuracyofhighresolutionimageryindistinguishingLevelIlandcoverclassesinforested landscapesisveryhighintherange95percentandgreater(Townsendetal.2009).Assuch,the imageryisoftenusedasatreecovervalidationforlowerresolutionimageryeithervisually(e.g., Kozaketal.2008;Neighetal.2008)orquantitatively(Knornetal.2009lHansenetal.2003lWang etal.2005).Becausespectralinformationislimited(oftenfourbands)andspatialalgorithmsare stillunderdevelopment,highresolutionimageryislessaccurateatLevelIIclassifications(~7090 percent;Oumaetal.2008;Mallinisetal.2008;Gergeletal.2007;andMoralesetal.2008).The highresolutionsensorsarecapableofdistinguishingsmallscalelandcoversanddisturbancesthat donotappearinmoderateresolutionimagery,suchasriparianstrips(Gergeletal.2007),pest outbreaks(Wulderetal.2008a),andlogging(deSouzaandRoberts2005),andthevisual interpretationofthehighresolutionimageryisstraightforward(Wulderetal.2004).Automated crowndelineationtechniqueshavebeendevelopedrecently(Broadbentetal.2008;Palaceetal. 2008),andhighresolutionimageryshowspotentialforusingcrownsizetoclassifysecondary foreststhatareolderthan20years(Clarketal.2004a;Kayitakireetal.2006).

4.1.8SARSensorsandPassiveSARFusionsforForestArea
SARhaspotentialforglobalforestmappingandSARcapabilitieshavebeenextensively reviewedelsewhere(Patenaudeetal.2005;Rosenqvistetal.2003;Lu2006;Balzter2001;and Kasischkeetal.1997).Here,wesimplysummarizethemainconclusionsofpriorresearchandfocus onrecentdevelopmentsinSARresearch.SARdistinguisheslandcoverbasedonthestrengthof backscatter,thevariationinbackscatteroverspace(texture),and,dependingontheanalysis, temporalvariationinbackscattersignatures.Regionalmapsofforestcoverhavebeenderivedfrom SARsensorsincludingtheGlobalRainforestMap(JERSsensor;Rosenqvistetal.2000)andthe CentralAfricaMosaicprojectmaps(ERSsensors;DeGrandietal.1999).TheALOSPathfinder

FAGAN AND DEFRIES

50


missionhascreatedthefirstglobalLbandSARforestmapat50mresolutionandthemissionwill continuetodosoannually(J.Kellndorfer,pers.comm.). ScientistsfullyappreciatethepotentialofsatelliteSARsensorstodistinguishlandcover;their capabilitiesarewellknownfromanalysisofdatafromadvancedairborneSARsensorsandthe multipolar,multibandSARsensor(C,X,andLband)aboardthe1994SIRC/Xshuttlemission (Kasischkeetal.1997;vanderSandenandHoekman1999).Ingeneral,singlebandSARsensors withonepolarization(e.g.,ERS1,JERS1)havevariableaccuracy(lowtoveryhigh)inLevelI classificationsdependingontopography,surfacemoisture,andthestructuralcomplexityofthe landscape(Patenaudeetal.2005;Kellndorferetal.1998;PodestandSaatchi2002).Asthenumber ofpolarizationsorSARbandsincreases,LevelIaccuracycanbeveryhigh(>90percent),rivalingor exceedingpassivesensors(Dobsonetal.1995a;Kellndorferetal.1998;Patenaudeetal.2005). LevelIIaccuracyisalsoverydependentonthenumberofpolarizationsandbandsemployed, achievingaccuraciesbetween7090percent(Kasischkeetal.1997;Kellndorferetal.1998). Dobsonetal.(1995a),notably,haveveryhighLevelIandIIregionalaccuracywithpolarized, multibanddatainMichiganforests.Saatchietal.(1997),withmultipolarSIRC/Xdata,areableto measureAmazonforestswithveryhigh(92percent)LevelIaccuracyandacceptable(72percent) LevelIIaccuracy.TheLbandJERS1andALOSPALSARsensorshavebeenshowntohaveveryhigh LevelIaccuracywhenclassifyingseveraldifferentforestedregions(Thieletal.2006;Saatchietal. 2000;Kellndorferetal.2008).BecauseofthehighSARreflectionfromwater,longerwavelength sensorshavecreatedhighlyaccuratemapsofmangrovesandfloodplainforests(Costa2004). TheaccuracyofSARsensorsisaffectedbyecosystemstructure(Almeidaetal.2007; Kellndorferetal.1998,2004),variabilityinsoilandvegetationmoisture(Kasischkeetal.1997), andtopography(Kellndorferetal.1998).Highresolutiondigitalelevationmapsarenecessaryto correctfortopographicaldistortions;accuracydecreasesmarkedlywithouteffectivecorrection (Kellndorferetal.1998;Ticehurstetal.2004).Becauseofitshighmoisturesensitivity,theCband istheleastusefulforlandcoverclassifications,althoughitsutilityforLevelIclassificationincreases whenmultiplepolarizationsareused(Saatchietal.1997;vanderSandenandHoekman1999; Kasischkeetal.1997). RecentresearchinSARlandcovermappinghasinvestigatedthepotentialofopticalSAR fusions(Saatchietal.2007;Santosetal.2008)andInSARforimprovedlandscapeclassification (Balzter2001;EngdahlandHyyppa2003;ParkandChi2008).InterferometricSARcanimprove classificationsbyaddingvegetationheightinformationtootherSARmeasuresliketextureand backscatter(Balzter2001).OpticalSARfusionsshowspromiseforsynergisticlandscape classification(Kasischkeetal.1997;Kuplichetal.2000)InIndiandryforests,fusionofmoderate resolutionpassiveandEnvisatASARCbanddataimprovedLevelIIclassification(>90percent) (ChandandBadarinath2007).InAmazonia,Kuplichetal.(2000,2006)foundLandsatSARfusions improvedLevel2classificationsoverLandsatorSARalone,andSaatchietal.(2007)usedaSAR opticalfusiontoderivevegetationandbiomassclassesfortheentirebasin.Santosetal.(2008) foundthataLevelIIclassificationofoilpalmplantationsandforestswasimprovedbyapassive SARfusion,to90percentfrom76percent(SAR)and77percent(Landsat).IncentralAfrica, Mayauxetal.(2000)usedAVHRRandERSdatafusiontoclassifyforests,savannas,mangroves, swampforests,andtreeplantationswith75percentaccuracy.

FAGAN AND DEFRIES

51


GiventhelargenumberofSARconstellationsplannedforlaunchinthe20092015period,itis likelythatInSAR,multibandSARfusions,andopticalSARfusionswillbecomeincreasingly importanttechniquesinlandcoverclassification.

4.2 Forest Structure: Current Measurement Methods


Wenextreviewtheuseandrelativeaccuracyofdifferentremotesensingmethodstomeasure verticalandhorizontalforeststructure,withanemphasisonforestheight.Remotesensorsare sensitivetoforeststructure,which,inconjunctionwitharea(Table4),isusedtoestimateforest volume(Section4.3),biomass(Section4.4),andcarbon(Section4.5). Forestsarehighlycomplexthreedimensionalstructures.Forestheightcanvarymarkedlyover adefinedarea(i.e.,atree,plot,pixel,orstand).Inthisreport,wedefinethehorizontalandvertical complexityofforestsurfacesasforeststructure;bothactiveandpassivesensorscanmeasure variationinforeststructure.Verticalforeststructurecanbecharacterizedbyforestheight,canopy depth,leafdensityprofiles,andothermeasurements.Horizontalforeststructurecanbe characterizedbycanopywidth,treedensity,andtreearchitecture. Becauseofgeneralitiesintreestructurefromhydraulicconstraints(Westetal.2009),forest heightallowsforaccurateestimationofvolume,biomass,andotherforestproperties(Lefskyetal. 2002b),alldiscussedfurtherbelow.Forestheightcaneitherbemeasureddirectlybyremote sensingorestimatedindirectlyfromallometricrelationshipswithotherremotelysensed measurements(e.g.,canopywidth;Song2007).Exactmeasurementsofforestheightmustderive twokeyparametersforagivenarea:groundelevationandcanopytopheight.Therearemultiple approachestomeasuringcanopytopheightandgroundelevation.Theseapproachesinclude passivestereo,LIDAR,InSARgrounddifferencing,andpolarimetricInSAR(Sextonetal.2009; Balzter2001;StOngeetal.2008).

4.2.1StereoMeasurementsofForestStructure
Multiangle(stereo)passivesensorsareusedtoestimatethethreedimensionalstructureof objectsfromtwoormoreoffsetobservations,muchlikeSARinterferometry(Jensen2007).Multi anglesensorsincludeMISR;ASTER;PROBACHRIS;anyofthepointable,highresolutionsatellites (e.g.,GeoEye,Spot5);andaeriallymountedstereodigitalcameras.Highresolutionstereo photographyhasalonghistoryofmeasuringobjectheightfromaerialimages(Jensen2007).High resolutionstereoimageshavebeeneffectivelyusedtomeasurecanopytopheightinforestswith highaccuracy(Brownetal.2005;StOngeetal.2008)andtodevelopanewglobal,canopytopDEM frommoderateresolutionASTERimages(ERSDAC2009).Althoughtheaccuracyofstereoimagery inestimatingcanopytopheightisexceededbyLIDARandsomeInSARmeasurements,the approachhasbeenproposedasameanstoextendLIDARresultstolandscapes(St.Ongeetal.2008; Dineretal.2005).HighresolutionstereoimageryislikelytobecheaperthansmallfootprintLIDAR onaglobalscaleyetithasseverelimitationsintermsofcosts,datausagerights,andcloudfree coverage(seeSection4.1.6),andcouldonlyprovidevalidationdataformonitoringtheglobal forest. Coarserresolutionmultianglesensors(e.g.,MISR,PROBACHRIS,POLDER,repeatpassMODIS) requirelocallyfittedalgorithmstomodelhowreflectionandabsorptionfrommultipleanglesrelate

FAGAN AND DEFRIES

52


tosubpixelcanopyheightandstructure(Dineretal.2005).Simplemetricsofvarianceinreflection havebeenusedtosuccessfullyestimatetreedensityandheightinopenforests(Sedanoetal.2008; Heiskanen2006),andMISRestimatesofheightandroughnesshavebeenusedtoimproveLevelII classificationaccuraciesinforests(Galvaoetal.2009;LiuandKafatos2007).Choppingetal.(2008) recentlyusedacomplexcanopymodeltoestimatetreeheightandbiomassinArizonawith acceptabletohighaccuracy(r2valuesof0.69and0.81,respectively)fromMISRdata,andKimeset al.(2006)foundthatlargefootprintLIDARestimatesofforestheightwerecloselyrelated(r2=0.89) toAirMISRheightpredictionsinNewHampshire.Althoughcoarseresolution,multiangleimage analysisisintheexperimentalstage,theanalysishaspotentialforrelativelylowcost,global estimatesofforeststructure(Choppingetal.2008;Dineretal.2005;Kimesetal.2006).Theneed forlocalizedmodelcreationandextensivevalidation,andpossibleinaccuraciesinclosedcanopy forests,makeitunclearwhethercoarseresolution,multianglesensorswillyieldglobalforest structuremeasurementsinthenextdecade(Kimesetal.2006;Choppingetal.2008;Schulletal. 2007).

4.2.2LIDARMeasurementsofForestStructure
BothsmallfootprintandlargefootprintLIDARprovideexcellentmeasurementsofvertical foreststructurethatexceedtheaccuracyofotherremotesensingmeasurements(Sextonetal. 2009;Hydeetal.2006;Nelsonetal.2006;StOngeetal.2008;Balzteretal.2007;Wolteretal. 2009;Hyyppaetal.2008).LIDARmeasurementsmeetstandardsofaccuracyforgroundbased measurementsforforestcanopytopheight(Deanetal.2009;Naessetetal.2004),andLIDAR measurementsaredistinctivebecausetheyalsomeasuregroundelevationatthesametime, allowingaccuratecalculationsofforestheight.LargefootprintLIDARshavemeasuredforest canopyheightwithgreaterthan75percentaccuracyacrossawidevarietyofforesttypes(Lefskyet al.2002a,b;Meansetal.1999;Hydeetal.2005;Lefskyetal.2007).Airplanemounted,small footprintLIDARshavehighaccuracy(>80percent)inmeasuringcanopyheight(Balzteretal.2007; Clarketal.2004b)andcanmeasurehorizontalcanopystructureandothercharacteristicsaswell duetotheirhighsamplingdensity(Lefskyetal.2002b;Hyyppaetal.2008).SmallfootprintLIDARs makeexcellentgroundtruthingdatasetsbuttheirsmallswathmakesglobalmeasurement challenging(Dubayahetal.2000;Balzteretal.2007).Althoughlargeareascouldbesurveyedvia aerial,smallfootprintLIDAR,itwouldtakeafleetoffiveplanes1063cloudfreedaystosurvey Brazils8.5millionsquarekilometers(Swansonetal.2009).Giventhecloudinessofthetropics,real flighttimemightbesixyearsforcompletecoverage. Global,walltowallcoveragebyLIDARsensorsisprobablyunlikelyinthenearfuture;even withlongrange,unmannedaerialsystems(UAS),walltowallglobalLIDARcoveragewouldbe logisticallydifficult.Obstaclesincludetherequirementofobtainingpermissiontoflyover countries,theamountofflyingneededforglobalcoverage,andhighcostsofpurchasingand runningmultipleUASsensors(roughly$460millionforfoursensorsoverfiveyears[Swansonet al.,2009]).However,threemethodsofglobalsamplingbyLIDARarepossible.Thefirstiscoarse resolutionsamplingbyLIDARsatellitesdescribedabove(Section2.2.2,Figure10).Thesecond, whichwouldcostapproximately$80million,couldbeaccomplishedbyusingasingleUASwith LIDARsensorstoconducttargetedsamplingofadiversityofforesttypesinselectedcountries.The thirdismountingaLIDARsensoronasatellite.

FAGAN AND DEFRIES

53


ThelargefootprintICESatGLASistheonlyLIDARsensorinorbitandtheglobal,wideinterval coverageisbeingusedforforestmeasurement(Figure19).ThewidespacingbetweenLIDARshots makesorbitalLIDARslikeGLASunsuitedforDEMcreationorwalltowallsamplingofforest height,butGLASLIDARfootprintsarewellcorrelatedgloballywithforestheight(r2=0.83,Lefskyet al.2007).Theseapproachesprovideanimportantsourceofdataforvalidationandfusionwith otherimagery(Lefskyetal.2002b;Baccinietal.2008;Nelsonetal.2009).Forexample,GLASdata havebeencombinedwithMODIS(Nelsonetal.2009;Boudreauetal.2008)andLandsat(Helmeret al.2009)datatocreateaccurateregionalbiomassandheightestimates. Figure19.GLASFootprintSpacinginAfrica

Notes:Eachblackdotrepresenta65mwidthLIDARfootprint;only30percentofallfootprintsarerepresentedhere. Source:Baccinietal.2008.

Inthe20092015timeframe,ICESatwillreachtheendofitsoperatinglifetimeandisexpected tobefollowedbytheICESatIIandDESDynILIDARsatellites.DESDynIisasatellitemissionplanned forlaunchin2014andcarryingalargefootprintLIDARaswellastwoLbandInSARsystemswith multiplepolarimetry.TheLIDARcapabilitywillbeusedinconjunctionwithInSARtoachieve accurate,moderateresolutionforestheightmeasurementsonaglobalscale(Donnellanetal. 2008).Accuratemeasurementsofgroundelevationmayalsoresultdependingonadvancesin processingInSARdata(ESA2009).

FAGAN AND DEFRIES

54

4.2.3InSARMeasurementsofForestHeight
SeveraldifferentmethodsusingSARinterferometrycanderiveforestheight.(Balzter2001; Sextonetal.2009).CanopytopheightfromshortwavelengthInSAR(e.g.,theSRTMmission)can besubtractedfromindependentmeasuresofthegroundelevation(Kellndorferetal.2004;Walker etal.2007).ShortwavelengthInSARestimatesofcanopytopheightcanbesubtractedfromlong wavelengthInSARestimatesofgroundelevation(Neeffetal.2005).Polarimetricinterferometric SAR(POLinSAR)usesinterferometric,polarizedSARandmultiplemeasurementstomodelthe verticalstructureandheightofforestsfromSARscatteringmodels(Treuhaftetal.2000;LeToanet al.2008).POLinSARisstillunderdevelopment,althoughthetechniquehasshownpromisefor heightestimationacrossadiversityofforesttypes(LeToanetal.2008;ESA2009). Sexton(2009)comparestheaccuracyofdifferentInSARmeasurementtechniques.Ingeneral, allInSARmeasurementsunderestimateforestheightbecausetheymeasuretothephasescattering center,whichisusuallybelowcrownlevel.Thisbiascanbecorrected(Saatchietal.,inreview).Ina forestinNorthCarolina,Sexton(2009)foundthatshortlongwavelengthdifferencing(XandP) wasmoreaccuratethanSRTMgrounddifferencingbutbothInSARmethodswerelessaccurate thanLIDAR.HealsofoundthatSARdifferencingwasleastaccurateinshorterstands(dueto groundreturnofcanopytopwavelengths)especiallyfortheSRTMgroundmethod.Regional calibrationwouldbenecessarytocorrectforthisbiasinheightestimation(Sextonetal.2009). GiventhenumberofSARsatellitestobelaunchedinthenextsixyears,InSARmeasurementof canopytopheightwillincreaseinfrequencyandglobalcoverage(Zinketal.2007).Current satellitesonlyhavethecapacitytodotemporalinferometryandseveralupcomingSAR constellationsandfollowonsatelliteswillcontinueandextendthiscapacity(Section2.4.2).Future capacityforspatialinterferometrywillcomefromanInSARsatellite(DESDynI)andseveral formationSARconstellations(e.g.,TerraSARX;seetheAppendix).Ultimately,theutilityof interferometricSARinmeasuringglobalforestheightwilldependonthecreationofahigh resolution,groundlevelglobalDEM(Walkeretal.2007)orthedevelopmentofmultiband,multi polarizedimagedatasetsthatallowtheestimationofforestheightthroughPOLinSARorshortlong wavelengthsubtractionalgorithms(Neeffetal.2005;Sextonetal.2009;ESA2009).TheGLASand DESDynILIDARswillimproveglobalDEMs(butnotatahighspatialresolution)andthelarge numberofSARsatellitestobelaunchedwillallowforInSARsynergiesbetweenC,X,L,and possiblyPbandsensors(NASA2009a;ESA2009;Saatchietal.inreview;Patenaudeetal.2005). Developmentofanaccurate,highresolutionglobalDEMwouldallowforthecreationof referencesmapsofforestheightusingCbandInSARfromtheERS,RADARSAT,orSRTMmissions (Kellndorferetal.2004).AgroundlevelDEMcouldbeachievedbyalongband(VHF)SARsensor, LIDARsaturationoftheearthssurfacefromsatelliteandairbornesensors(viatheproposedLIST satellite,orusinghighelevation,unmanneddronescalledUAVs),oraformationInSAR constellationwithaspecialvariablebaselineorbit(LeToan2008;ESA2009;NRC2007).Eachof theseoptionsisintheexperimentalstages.TanDEMXandTerraSARXwillcreateavariable baselineInSARforDEMmapping,butaccuracyinforestedareasisstilltobedetermined(Zinketal. 2007).Inaddition,atomographymodefortheproposedPbandBIOMASSsensorandavariable baselineTandemLcompanionsatelliteforDESDynImaybelaunchedinthe20092015period(Le Toan2008;Obertoetal.2008;ESA2009).

FAGAN AND DEFRIES

55

4.2.4FusionsofRemotelySensedDataonForestStructure
Inanyimageryfusion,LIDARprovidesoutstandinginterpolatedinformationonground elevationandcanbeusedtoincreasetheaccuracyofforestheightestimatesfromstereoopticalor interferometricSAR(Balzteretal.2007;Hydeetal.2006).Thisapproachisthemainconcept behindDESDynI,theInSARLIDARsatellitemissiondescribedabove.Toprovidemoreaccurate, spatiallyextensiveestimatesofforestheightandstructure,LIDARcanbefusedwithInSAR(Balzter etal.2007;Slattonetal.2001),passiveforeststructureestimates(Wolteretal.2009,Dubayahetal. 2000;Hydeetal.2006;Hudaketal.2002),stereomeasurementsofheight(Hydeetal.2006),or forestclassificationsfrompassiveoractivesources(Dubayahetal.2000;Helmeretal.2009; McCombsetal.2003).

4.2.5EstimatingForestStructurefromAllometricRelationships
Horizontalandverticalforeststructurecanbederivedfromthreedimensionalstereoimaging aswellasfromthetextureofhighresolutionsatelliteandaerialpassiveimagery(Song2007; Wolteretal.09).Analysisofcanopytexturefromopticalimagerytakesadvantageoftree illuminationgeometriesinheterogeneouscanopiestoestimatecanopywidthandotherstructural variables(Wolteretal.2009;Kayitakireetal.2006;Hyyppaetal.2000;Song2007).Canopy structure(e.g.,width)canthenbeusedtoestimatetreeheight,basalarea,andstemvolumevia allometricequations(Song2007;Lefskyetal.2002b).Canopystructurehaslongbeenderivedfrom remotelysensedimagery(thisisreviewedinmoredepthbySong[2007];Wolteretal.[2009];and Palaceetal.[2008]).Ingeneral,highresolutionimageryismoreaccuratethanmoderateresolution imageryatderivingcanopystructure(Hyyppaetal.2000;Song2007;butseeMcRoberts2008). Recenteffortshavefocusedonautomateddelineationofcanopystructuresusingavarietyof differentalgorithms.Successhasbeenmixedandstudieshavelargelybeenrestrictedtosmallareas (Song2007;Wolteretal.2009;Palaceetal.2008). Song(2007)achievedacceptableaccuracyincanopydelimitationinconiferforests(72percent variationincanopywidth),but,likeseveralothers,hadpooraccuracyindeciduousforestswith overlappingcanopies.Innorthernforestswithnonoverlappingcanopies,Wolteretal.(2009)found thatestimatesofcanopywidthandheightrivaledLIDARaccuracies(acceptabletohigh). AutomatedcanopydelimitationeffortsintheAmazonbasin,usingextensivegrounddata,observed clumpingofoverlappingtrees,underestimationofunderstorytreecrownarea,andoverestimation oftotalforestbiomass(Palaceetal.2008;Broadbentetal.2008).Whiletheseeffortshavenotbeen extensivelyvalidated,biomassestimateswerequitecomparabletogroundestimates. Ingeneral,texturederivedcanopywidthandheightmeasurementsarelessaccuratethan LIDARmeasurementsbuttheycanbecalculatedfromexistinghighresolutionimageryoverlarger areas(Wolteretal.2009;Broadbentetal.2008).Althoughthistechniquemaybeunsuitedfor globalforestheightmeasurementbecauseofcostanddatavolumes(DeFries2008),theapproach maybesuitableforextendingLIDARestimatesandvalidatingcoarserresolutionimagery(DeFries 2008;Wolteretal.09).

FAGAN AND DEFRIES

56

4.3 Estimating Forest Volume and Biomass using Remote Sensing


Boththevolumeandtheabovegroundbiomass(AGB)offorestscanbeestimatedfrom allometricrelationshipswithcanopywidth,structure,and/orheight,theintensityofSAR backscatter,correlationswithpassivespectra,andvariousfusionsoftheabove(Luetal.2006; Balzteretal.2007;Rosenqvistetal.2003).Wereviewthislargebodyofliteraturebriefly,witha focusonglobalmeasurementofforestbiomass.AmorethoroughreviewofthistopicisinLu (2006),Pateneudeetal.(2005),Rosenqvistetal.(2003),andLefskyetal.(2002b).

4.3.1ForestVolumeandRemoteSensing
Forestvolumeandgeometrydirectlyinfluencethereturnstosatellitesensors(Dobsonetal. 1995b,Dubayahetal.2000)butwiththeexceptionofseveralstudiesinplantationsandconifer dominatedforeststhatestimatebolevolume,manyremotesensingstudiesfocusonestimating forestbiomass(g/m2)ratherthanvolume(m3/m2)(e.g.,Drakeetal.2002a;Saatchietal.2007).The sameallometricequationsthatareusedtoestimategroundtruthbiomasscouldbeusedto estimatebolevolume(P.Waggoner,pers.comm.),butbolevolumeisoftennotreported.Asaresult, wewillfocusinthefollowingsectionsonestimatingbiomass,ratherthanvolume,fromremote sensing. Forestvolumeestimatesfromremotesensinghaveoftenbeenmadeinspeciespoorsystems withregular,distinctcanopiesinsupportofforestinventories.Activesensorsystemshavebeen frequentlyemployedtoestimateforestvolume.Patenaudeetal.(2005)reviewseveralstudiesin temperateconiferdominatedandborealforeststhatestimatedstemvolumeusingSAR(5383 percentaccuracy)andLIDAR(4697percentaccuracy).Hyyppaetal.(2008)reviewtheabilityof smallfootprintLIDARtoestimatestemvolumefromcanopyheightmeasurements.Wolteretal. (2009)reviewseveralLIDARstudiesthatachieved8591percentaccuracyforbasalarea.Meanset al.(1999),inthewesternUnitedStates,predict96percentofthevarianceinbasalareausinglarge footprintLIDAR.Notably,Andersonetal.(2008)onlypredicts25percentofthevarianceinbasal areawithLIDARinadeciduousforestinthenortheasternUnitedStates.Maltamoetal.(2006),on theotherhand,predictsstemvolumefromLIDARwithlessthan14percenterrorinaborealforest. AskneandSantoro(2009)areabletoestimateborealforeststemvolumewithlessthan20percent errorusingCbandInSAR. Itisunclearhowmuchthevariationinaccuratelypredictingvolumeisduetodifferencesin LIDARorSARsensortypes,processingchoices,orvariabilityamongregions.Expertshave hypothesizedthatLIDARhasmoredifficultyinpredictingforestvariablesindeciduousareasdueto irregularcanopystructure(Nelsonetal.2007)butthishypothesishasnotbeenthoroughlytested. Passiveopticalsensors,aloneandfusedwithactivesensors,havebeenusedtoestimateforest volumewithmixedresults.Forexample,nearestneighboralgorithms(kNN)haveprocessed moderateresolutionimagery,oftenintegratedwithforestinventorydata,toproduceestimatesof forestvolumeandbasalarea(McRobertsetal.2008).FrancoLopezetal.(2001)usednearest neighbortechniques,Landsatimagery,anddatafromtheUSForestServicesForestInventoryand AnalysisNationalProgram(FIA)toestimatebasalareaandvolumewithapproximately50percent accuracy,whileMcRobertsetal.(2008)estimatesbasalareawith66percentaccuracyusingsimilar techniques.Halletal.(2006)usesmoderateresolutionimagery,forestinventorydata,andan

FAGAN AND DEFRIES

57


allometricforeststructuremodeltopredictvolumewith69percentaccuracy.Thesemoderate resolutiondataalgorithmsmaybetoocomputerintensiveforlargeareaestimation. Highresolutionimageryhasalsobeenusedtoestimateforestvolume,oftenwithgreater success.Hyyppaetal.(2000)findthathighresolutionopticalsensorsoutperformedsome,butnot all,typesofSARsensors,aswellasmoderateresolutionpassivesensorsinpredictingbasalarea andstemvolume,withmaximumaccuracyof5558percent.Wolteretal.(2009)usehigh resolutionSPOTimageryinanopentemperateforesttoestimatebasalareawith71percent accuracy.Kayitakireetal.(2006)usetextureanalysisof1mresolutionIKONOS2imageryto estimatebasalareaofamixedconiferdeciduousforestwith35percentaccuracy.Inanopenpoplar plantation,Wangetal.(2007)useallometricrelationshipsbetweencanopyheightanddiameterto estimatestemvolumewith8792percentaccuracy.Inaborealforest,WallermanandHolmgren (2007)estimatestandvolumewithlessthan20percenterrorbyfusingsmallfootprintLIDARwith textureanalysisofSPOTHRGimages.Popescuetal.(2004)fusesmallfootprintLIDARwithhigh resolutionopticalimagerytoestimateforestvolumewithhighaccuracy(83percentofvariation)in coniferousforestsandpooraccuracyindeciduousforests(39percent).Aswithactivesensors,the variationinmethodsandforesttypesmakesitdifficulttodrawdefinitiveconclusions,butitisclear thatfusionsimproveestimates(e.g.,Popescuetal.2004)andthatreadilydistinguishabletree canopiesimproveallometricestimatesofstandvolumefromcanopywidth(Song2007). Becauseofthesimilarityofallometricequationsthatcalculatebiomassandstemvolume,the methodsdetailedinthefollowingsectionscouldeasilybeappliedtoestimationofstemvolume. Popescuetal.(2004)andMeansetal.(1999),forexample,foundverycloseagreementbetween theirestimatesofstemvolumeandbiomassfromremotesensing.Becausevariationinwood densityisimportantinglobalestimationofbiomassacrossregionsandforesttypes(Bakeretal. 2004)estimationofvolumemaybemoreaccuratethanbiomassinsomeecosystemsoratregional scales.Morestudiesareneededontherelativeaccuracyofvolumeandbiomassestimatesacross diverseforestecosystemstoevaluatethispossibility.

4.3.2ForestHeightandBiomass:aStrongPositiveCorrelation
Treeheightand/ordiameter,becauseoftheuniqueconstraintsofplantstructure,ispositively correlatedwithtreebiomasswithinaspecies(Chaveetal.2005;Westetal.2009;Dubayahetal. 2000).Usingwellestablishedallometricrelationships,biomasscanbecalculatedfromtree diameter,height,and/orwooddensity(Chaveetal.2005;Schroederetal.1997).Remotesensing cannotdirectlymeasurewooddensity,butcorrelativeforestinventorydatacanusespeciesspecific orregionspecificallometricequationstoprovideaccurateestimatesofbiomass(seeSection3.1.3). Forestheightcanbemeasuredfromavarietyofremotelysenseddata(seeabove)andusedto estimatebiomass(Kellndorferetal.2004;Palaceetal.2008,Pflugmacheretal.2009).Although diameter,height,andwooddensityarecentralvariables,biomassestimatescanbeimprovedby usingadditionalforeststructurevariables(e.g.,canopywidth,canopyvolume)(Dubayahetal. 2000;Palaceetal.2008;Popescuetal.2003).

4.3.3EstimatingBiomassusingSARBackscatter
ThemagnitudeandpolarityofSARbackscatterissensitivetoforeststructure,withshorter wavelengths(C,Xbands)interactingprimarilywithcanopyelements,andlongerwavelengths

FAGAN AND DEFRIES

58


interactingprimarilywithbranch,trunk,andgroundelements(L,P,VHFbands).Backscatter respondstothedensityofbiomassatdifferentheights,anditcanbeusedtoestimatebiomass. However,backscatterintensityfromCandXbandSARtendstosaturateatrelativelylowlevelsof forestbiomass(2550megagramsperhectare[Mg/ha]),Rosenqvistetal.2003).LbandSAR saturatesatvaryinglevels(60150Mg/ha;Patenaudeetal.2005),dependingonspecies compositionandforeststructure.PbandSARsaturatesatahigherrange(100200Mg/ha),again dependentonforestbiotaandstructure(Patenaudeetal.2005),andVHFbandSARisquite sensitivetoforestbiomassanddoesnothaveappreciablesaturationasbiomassincreases (Franssonetal.2000;Patenaudeetal.2005).PbandSARandexperimentalVHFSARsensorsare mountedinaerialsystemsnow,butneitherbandwillberepresentedinorbitinthe20092015 windowunlessthePbandBIOMASSsatelliteproposalisapprovedbytheEuropeanSpaceAgency (LeToan2008). Pateneudeetal.(2005)reviews11studiesandfindsthataccuracyofbiomasspredictionsfrom SARvarybetween50and96percentofobservedvariation,acrossavarietyofforestsandbands. SARbackscatterreturnsaresensitivetotopography,variationinsoilandcanopymoisture,and changesinforeststructure(Kellndorferetal.1998;Kasischkeetal.1997).Inhillyareas, backscattermustbecorrectedbyahighresolutionDEMbecauseofshadowing,layover,andother angulareffects.IfahighresolutionDEMisnotavailable,theaccuracyofbiomassestimatesfrom SARdecreases(Kellndorferetal.1998;Ticehurstetal.2004). PolarizedSARandmultibandSARprovidemoredetailaboutforeststructure,andimprove biomassestimates(Dobsonetal.1995b;Rosenqvistetal.2003).Askneetal.(2003)useshortand longwavelengthInSARtoestimatebiomassinaborealforestwithin15percentofground estimates,whileBalzteretal.(2007),usingasimilarmethod,isabletoestimatebiomassina deciduousforestwithin36percentofestimates.Recentresearchhasmeasuredtropicalforest biomasswith+/15percentaccuracyusingLbandpolarizedSAR(upto150mg/ha)andwithP bandpolarizedSAR(upto300+mg/ha)(Saatchietal.inreview). Polarizedbackscatterestimatesofbiomassmaybecombinedwithheightbasedestimatesof biomassfromInSARtoincreasetherangeofforestbiomassthatcanbeaccuratelymeasuredfrom SAR(Saatchietal.inreview).Theseestimatesarecomplementary:estimatesofbiomassfromtree heightaremoreaccurateastreeheightincreases,whilebackscatterislessaccuratewithincreasing height(Sextonetal.2009,Saatchietal.inreview;LeToan2008).InCentralAmerica,aSARInSAR combinationmethodhasbeenfoundtoincreasethesaturationpointofLband(to200t/ha)and increasetheaccuracyofLandPband(Saatchietal.inreview).TheBIOMASSandDESDynI missions(PbandandLband,respectively)willusethiscombinationapproachtoglobalbiomass estimation,whichmayleadtofrequent,100mresolutionestimatesofabroadrangeofforest biomass(LeToan2008;Donnellanetal.2008).

4.3.4IndirectlyEstimatingBiomassfromForestSpectra
Anumberofstudieshavecorrelatedpassiveremotesensingmeasurementstonetprimary productivityandbiomass(Lu2006;DeFries2008;Goetzetal.2005;Baccinietal.2008).Several vegetationindicesderivedfrompassiveremotesensingarewellcorrelatedwithleafareaindex (LAI),fractionofabsorbedphotosyntheticallyactiveradiation(fAPAR),andothermeasurementsof

FAGAN AND DEFRIES

59


productivity(Gaoetal.2000).NetprimaryproductivityismodeledfromeitherLAIestimatesor fAPARestimates;bothapproacheshavebeenshowntoberelativelyaccurate(LuusandKelly2008) eventhoughLAIestimatessaturateat24leavesperunitarea(Gaoetal.2000).Mostpassive sensors(MODIS,Landsat,GeoEye)canestimatefAPARandLAI,andproductivityislinearly correlatedwithbiomassatlowlevelsofbiomass(<200Mg/ha,KeelingandPhillips2007;Luusand Kelly2008).Inlowbiomasssystems,includingmanytemperate,boreal,anddrytropicalforests, passivesensorsarecapableofestimatingbiomassdirectlyfromvegetationindices(Lu2006; MuukonenenandHeiskanen2007). Houghtonetal.(2001)comparetwobiomassmapsoftheAmazonbasinfromremotesensing (MODISandAVHRR)andfindlittlespatialagreementbetweenthembutoverallagreementontotal biomass(~10percentdifference).Piaoetal.(2009)modelproductivityandbiomassacrossChina andfindgoodagreementbetweenremotelysensedbiomassandproductivityestimatesandtwo otherindependentestimationmethods.Baccinietal.(2008)useacombinationofMODISand grounddatatomodelAfricasbiomasswith8090percentaccuracy.Houghtonetal.(2007)use MODIStomapbiomassacrossRussiaandareabletopredict60percentofthevariation,although withinsiteerrorwashighlyvariable.Zhengetal.(2004)employsLandsat7anddataonforestage andcompositiontopredictabovegroundbiomass(AGB)with67percentaccuracy.Blackardetal. (2008)useacombinationofgrounddata,coarseandmoderateresolutionimagery,andGISlayers togenerateabiomassmapoftheUnitedStateswithvariablepredictability(0.310.73across regions).Foodyetal.(2003)useLandsatspectraldataandaneuralnetworkalgorithmtopredict biomassatthreetropicalforestsiteswith7184percentofthevarianceexplained.However,inthe studybyFoodyetal.(2003),vegetationindicesoptimizedtopredictbiomassatonesitehavevery poorperformanceatothersites,underscoringthedifficultyofdevelopinggeneralspectral relationshipsacrossmoderateresolutionimagerytakenindifferentseasons,cloudlevels,and lightingconditions(Schroederetal.2006,butseeWoodcocketal.2001). Thebiomassproductivityrelationshipsaturatesintropicalandhighbiomasstemperateforests (>200mg/ha,KeelingandPhillips2007),soattemptstoestimatebiomassthereoftenuseother metricsbesidesspectralresponse,includingimagetextureandforestage(Lu2006).Asreviewed above(Section4.2.5),highresolutionimagetexturecanbeusedtoderiveforeststructuremetrics (e.g.,canopywidth,height)(Wolteretal.2009),whichthencouldbeusedtoestimatebiomass throughallometricrelationships(Palaceetal.2008;Broadbentetal.2008).However,Song(2007) showsthattheaccuracyofthiscanopydelimitationapproachislowinforestswithoverlapping canopies.Indensetropicalforests,Broadbentetal.(2008)alsofindthatbiomassestimateshadlow accuracy(excludingemergenttrees)andemergenttreescoveredsmallertreecanopiesfromview andmadedetailedgrounddatacorrectionnecessaryforanybiomassestimates.Inmoreopen, lowerbiomassforests,texturemetricscanaidspectralestimatesofforeststructureandbiomass (Lu2006;Wolteretal.2009).Althoughtexturemetricsbasedonhighresolutionimageryare limitedinarealcoverageanddependentonlocalgrounddataandradiometriccorrection,the metricscouldbeavaluableadjuncttomoreaccurate,spatiallycoarseLIDARmeasurementsof foreststructure(Wolteretal.2009). Forestbiomasscanbeindirectlyestimatedbyusingremotesensingtomapforestageclasses (Castroetal.2003;Chambersetal.2007).Ifaccuratelocalrelationshipscanbedevelopedbetween biomassanddetectableforestageclasses(e.g.,Zhengetal.2004),thedistributionofforestage

FAGAN AND DEFRIES

60


classesacrossthelandscapecanbeusedtoestimatebiomassandmodelfutureproductivity (SanchezAzofeifaetal.2009;Castroetal.2003;Lu2006;Zhengetal.2004;Helmeretal.2009). However,secondaryforestsover1530yearsofagearedifficulttodistinguishfrommatureforests usingforestspectra(Steininger1996;Nelsonetal.2000;Songetal.2007).Secondaryforestalso variesmarkedlyinitsgrowthrateduetoavarietyoffactors,includingspeciescomposition, climate,landusehistory,soilfertility,humandisturbance,andregionaldifferences(Nelsonetal. 2000;Castroetal.2003;SanchezAzofeifaetal.2009).Bothofthesefactspresentseverechallenges tousingforestageclassesasapredictorofforestbiomass,inadditiontotheintensivefieldwork requiredtoestablisharelationshipbetweenforestageandbiomass(Castroetal.2003;Uhletal. 1988;Zhengetal.2004;butseeHelmeretal.2009).Sanchezetal.(2009)arguethatimprovingthe detectionofforestageclassesisacriticalgoalforforestcarbonmeasurement.However,continuing humandisturbancefromloggingandfires(e.g.,Alencaretal.2006;Asneretal.2006;Uhletal. 1988),aswellasthevariabilityinforestregrowthbiomasswithage(e.g.,Helmeretal.2009)make forestageametricofquestionableutilityinmeasuringglobalbiomass.Directestimationofforest regrowthbiomassandproductivityisamoreparsimoniousgoal,anditmaybeachievablewith expectedadvancesinthespectralandspatialresolutionofsensors,polarizedandinferometricSAR satellitesandconstellations,increasedLIDARcoverage,andimprovementsinmodelingforest reflectance.

4.3.5LIDAREstimationsofBiomass
LIDARisconsideredapromisingapproachtoestimatelocalforestbiomass(Lu2006)because directmeasurementofforestcanopystructureandheightcanbeconvertedtobiomassusing allometricequations(Dubayahetal.2000;Broadbentetal.2008;Chaveetal.2005).Earlierinthis report,wereviewedthebasiccapabilitiesofLIDARtechnology(Section2.2.2),thehighaccuracyof LIDARinmeasuringforestheight(usuallywithheighterrorsoflessthan1m,Section4.2.2),and thevariablebutoftenhighaccuracyofLIDARinestimatingforestvolme(Section4.3.1).The accuracyofLIDARinestimatingbiomassisalsooftengoodbutmethodsdifferamongLIDAR technologies. SmallfootprintLIDAR,takenfromaerialsensors,excelsatmeasuringforestheightand delimitingcanopywidthandoverallstructure.Withthistechnology,biomassisusuallyestimated byitscorrelationwithforestheightand/orcanopywidths.Itischallengingtodistinguishground returnsindenseforestsbutalgorithmsexistforefficientprocessingofsmallfootprintdataevenin thesechallengingconditions(e.g.,Clarketal.2004).SmallfootprintLIDARhasachievedpoorto veryhighaccuracyinestimatingforestbiomasswithmanystudieshavingaccuracyinexcessof80 percent(Patenaudeetal.2005).Patenaudeetal.(2005)reviewseveraltemperateforeststudies withpredictiveaccuraciesbetween4591percent,andPopescuetal.(2004)observepredictive accuraciesbetween3281percentindeciduousandpineforests,respectively.Asneretal.(2009) areabletousesmallfootprintLIDARtopredictbiomassintropicalHawaiianforestswith78 percentaccuracy,Nelsonetal.(2007)estimatebiomassintemperateNorthCarolinapineforests with93percentaccuracy,andNaessetandGobakken(2008)estimatebiomassinborealforests with88percentaccuracy.Zhaoetal.(2009)useLIDARheightdistributionstoestimateforest biomassineastTexaswithupto95percentaccuracy.Similarly,vanAardtetal.(2008)usesmall footprintLIDARtobothdistinguishtemperateforesttypesandestimatetheirbiomasswithgreater

FAGAN AND DEFRIES

61


than95percentaccuracy.Aswithestimationofforestvolume,itisdifficulttoassesswhetherthe observedvariabilityinaccuracyarisesfromthemethodsused,fromregionalvariabilityamong forestsingrowthform,orfromregionalvariabilityinspecieswooddensity. LargefootprintLIDAR,withitsfullwaveformdescribingthreedimensionalcanopystructure,is wellcorrelatedwithbiomass.Intemperateandborealforests,predictiveaccuraciesofbiomass fromaerial,largefootprintLIDARhaverangedbetweenalowof61percent(northeastUnited States,deciduous;Andersenetal.2006)andahighof95percent(Pacificnorthwestconifers,Means etal.1999).InstudyingthreeU.S.biomes,Lefskyetal.(2002a)achieve84percentaccuracy. Nilsson(1996)predictsbiomassforScotspinewith78percentaccuracy,andHydeetal.(2005) predict86percentofmountainconiferbiomassinaU.S.study.InnorthernChina,Sunetal.(2007) predictbiomasswith73percentaccuracyandBordreauetal.(2008)achieve81percentaccuracy inborealRussia.Intropicalforests,predictiveaccuraciesofbiomassfromlargefootprintLIDAR havebeenbetween66and93percent(Drakeetal.2003,2002a,b).Drakeetal.(2002b)findthat oneparameter(HOME)fromtheLIDARwaveformwashighlypredictiveofforestbiomassacross sites.Baccinietal.(2008)usethatrelationshiptovalidatearegionalmapofAfricanbiomasswith ICESatGLASsampling. Becauseoftheclosecorrelationwithglobalforestheightandbiomass(Lefskyetal.2007; Helmeretal.2009),LIDARpulsesfromtheorbitingICESatGLASsatellitehaveprovidedawealthof dataforforestbiomassmappingandvalidation(Baccinietal.2008;Helmeretal.2009;Bordreauet al.2008;Nelsonetal.2008;Phlugmacheretal.2008).ThelaunchofDESDynIpromisestoincrease thisnewsourceofglobalbiomassdata,butbothICESatandDESDynIwillcollectdataatcoarse spatialresolution:measurementsiteswillbeseparatedbyseveralkilometers(Obertoetal.2008; NASA2009a).Forthenearfuture,allLIDARsatellitesareexpectedtosampleonlyatcoarse resolution. Overall,LIDARhasthepotentialforgoodtohighaccuracyandcanmeasurebothforestcanopy heightandgroundelevation,thusovercomingakeylimitationofcurrentSARandInSAR approachestoestimatingbiomass.ThegreatestlimitationofLIDARmappingofbiomassisthe coarsespatialresolutionofsatelliteLIDARsensorsandthenarrowswathofaerialLIDARsensors. TheselimitscanpotentiallybeovercomebyintegratingLIDARdatawithotherremotedatasetsthat havegreaterresolutionorswaths(e.g.,InSAR,passive;seebelowformoredetail).FusionofLIDAR datawithothertypesofimagerywillallowaccurateestimatesofbiomassonaglobalscale(Baccini etal.2008;Andersonetal.2008;Donellanetal.2008).

4.3.6FusionApproachestoEstimatingForestBiomass
Threemaindatasourcesplayaroleinestimatingbiomass:GISdatalayers,georeferenced grounddata,andsatelliteimagery(Gibbsetal.2007;DeFriesetal.2007).Itispossibletofuseall three.Biomassestimatesfromremoteimagerycanbeimprovedbyusinganynumberofgeo referenceddatalayers:bioclimaticsurfaces,roads,generalhabitatclassifications,populationlayers, and/orgrounddata(Baccinietal.2004;GOFCGOLD2008;Saatchietal.2007;HeroldandJohns 2007;Blackardetal.2008).ThevolumeofglobalGISdataisgrowing;coarseresolution,global analysesofhumanimpactonforestsarenowpossible(Sandersonetal.2002).Itispossibleto discountforestcarbonestimatesbymeasuresofhumanimpact,includingroaddensityand

FAGAN AND DEFRIES

62


population(GOFCGOLD2008;Gibbsetal.2007).Saatchietal.(2007)useelevationdataanda varietyofremotesensingdatatoestimatebiomassacrosstheAmazonbasin;comparisonwith bioclimaticlayersindicatesthatdryseasonrainfallmightbeagoodadditionalpredictorofbiomass. Baccinietal.(2004)useacombinationoflandcoverclassificationfromMODISdata,ground inventorydata,andclimateandtopographicGISlayerstopredictbiomassinCaliforniawith73 percentaccuracy.Baccinietal.(2008)useasimilarapproachwithgrounddatainAfricato accuratelypredictbiomassbutfindthatGISlayersdonotimprovetheanalysis.Walkeretal.(2007) usegrounddata,nationalGISlayers,andbothpassiveandactiveremotesensingdatatocreatean accurate(r=0.88)mapofforestbasalareaforcentralUtah. Fusionwithgrounddatacanprovideadditionalattributesandaccuracytomappedforest classes(Baccinietal.2004).Thisapproachcanalsoprovidecontinuousdatathatcanbeintegrated acrossregionstreediameterdata,forexample,hasimprovedallometricestimatesofbiomass fromLIDAR(Meansetal.1999).Boudreauetal.(2008)useacombinationofgrounddata,aerial LIDARdata,satelliteLIDARdata,elevationdata,andmoderateresolutionlandcoverclassification toestimatebiomassfortheborealforestsofQuebec.Similarly,Nelsonetal.(2008)usegrounddata toparameterizebiomassestimatesfromGLAS,andthenextendtheGLASestimatesregionallyusing MODISlandcoverclassifications.Andersonetal.(2006,2008)findthatbyincludinggrounddataon forestcomposition,GLASLIDARestimatesofforestbiomassaresignificantlyimproved.The authorsalsoshowthatthisforestcompositiondatacouldbeestimatedaccuratelybyhigh resolutionhyperspectraldata. Thefusionofremotesensingdatatopredictgroundvalidationdatasetsisanincreasingly commonpracticewithgreatpotential(Rosenqvistetal.2003).Saatchietal.(2007)integratea varietyofSARandpassivesensorstoestimateAmazonbiomassin16forestclasseswith8191 percentaccuracy.Moghaddametal.(2002)findthataSARLandsatfusionistwiceasaccurateas Landsataloneinestimatingconiferfoliagebiomass.Treuhaftetal.(2003)usehyperspectraland InSARdatatoestimateforestbiomassinOregonwith+/16percenterror.Hydeetal.(2006)fusea varietyofdatasources(SAR,InSAR,LIDAR,highandmoderateresolutionpassive)andfindthata LIDARLandsatfusionpredictsbiomassthemostaccurately(80percentofthevariance);anall sensorfusionisonlyslightlymoreaccurate(83percent).Helmeretal.(2009)useglobal relationshipsbetweenGLASdataandforestbiomasstoestimatebiomassdirectlyinBrazil,then extendthoseresultstoLandsatderivedregionalforestageclasses.CombiningtheglobalGLAS datasetswithexistinglandcovermapsisonepromisingnewfusionmethod,outofmany,for improvingglobalforestbiomassestimatesinthe20092015timeperiod.

4.4 Estimating Forest Carbon Stocks from Remotely Sensed Data


Satelliteimagingcantellusmuchaboutglobalcarbonstocks,buttherearelimitstoits accuracy.Drybiomassisapproximately4755percentcarbonbyweight(IPCC2006),so abovegroundbiomassestimatesfromremotesensingcanbesimplyconvertedintoaboveground carbon(AGC)stockestimates(Gibbsetal.2007).Dependingonthemethodused(e.g.,SARvs. spectra),however,remotesensingmaynotincludedeadtreesinitsestimateofAGCstocks.In addition,satelliteandaerialestimatesofforestcarbonarefundamentallylimitedtoaboveground carbon.Remotesensingcannotmeasurelitter,root,andsoilcarbonstocks,whichcanbe considerable(GOFCGOLD2008;Malhietal.1999;Pageetal.2002;butseeFosteretal.2002). Belowground(root)carbonaveragesaround2030percentofAGCglobally(Cairnsetal.1997),but

FAGAN AND DEFRIES

63


variesfrom9to26percentinforestsacrosstheAmazonbasin(Houghtonetal.2001).National estimatesofcarbonstocksoftensimplyadjustabovegroundcarbonstockestimatesbyuniform percentages(e.g.,30percent)toestimatetotalnonsoilforestcarbonstocks(Gibbsetal.2007; Houghtonetal.2001).Differingmethodsofestimatingdeadandbelowgroundcarbonleadtomuch oftheobservedvariationinforestcarbonstockestimates(Houghtonetal.2001). Modelingsoilcarbonemissionsisimportantinforests:theamountofcarboninlitter,nontree vegetation,andsoilstocksvarymarkedlywithlatitude,foresttype,andsoiltype,butsoilsalone makeup,onaverage,4984percentoftotalforestcarbonstocks(Table1;Malhietal.1999;GOFC GOLD2008).Notallofthesoilcarbonisemittedwithdeforestation,butemissionsfromsoilcarbon stockscanaccountforasignificantpercentage(10percentorgreater)ofemissionsfrom deforestationanddegradation,especiallyinnorthernforestsandpeatswamps(Houghton1999; Pageetal.2002).Forexample,onestudyestimatesthatthatthe1997burninganddeforestationof Indonesianpeatswampsreleased253tonsofcarbonperhectare,whichconstitutedbetween13 and40percentoftotalglobalcarbonemissionsthatyear(Pageetal.2002).Althoughglobalmaps ofsoilcarbondoexist,theaccuracyofthemapsisdifficulttoassesswithavailabledata(GOFC GOLD2008).Intheparagraphsbelowwereviewthedifferentmethodsofestimatingforest abovegroundcarbonstocksonaglobalscale.

4.4.1AbovegroundCarbonStocksEstimates:InventoryandRemoteSensing
Forestinventoriesareastartingpointforestimatingforestabovegroundcarbon(AGC)stocks byprovidingmeasuresoftreediameter,treeheight,speciescomposition,anddensity.Applying generalallometricequationsthenallowsestimationoftreevolume,biomass,andcarbon(Chaveet al.2005;seeTable4).Allometricequationsarederivedfromharvesting,drying,andweighingtrees inforestplotsandcanbesiteandspeciesspecificifenoughsamplingisdone(Gibbsetal.2007). GeneralallometricequationsdevelopedacrossforesttypesarehighlyaccurateatpredictingAGC stocksinlocalforestplots(Chaveetal.2005;Schroederetal.1997)incombinationwithspecies specificwooddensitydata(Bakeretal.2004).Inthetropics,crossspecies,standlevelestimatesof wooddensityleadtodecreasedaccuracyinbiomassestimation(Bakeretal.2004). Onceabovegroundtreecarbonisaccountedfor,carboninothervegetation,litterandsoilscan beinventoriedaswelltoimproveestimationaccuracyandcarbonintreerootsisusuallyestimated asafunctionoftreesize(2030percent;Gibbsetal.2007;Cairnsetal.1997;GOFCGOLD2008).If soilinformationisnotincludedintheforestinventorythedecreaseintotalcarbonestimation accuracycanbemarked(GOFCGOLD2008;Malhietal.1999)butaccurateAGCestimatescanbe combinedwithglobalsoilcarbonmapstogetestimatesfromexistingforestinventorydata(GOFC GOLD2008). Incomparisontoremotesensingestimates,forestinventoryestimatesoftotalAGCare consideredmoreaccuratealthoughtheytoohaveestimationinaccuracies(Gibbsetal.2007; Saatchietal.2007;Chaveetal.2005;Bakeretal.2004).Ontheglobalscale,however,plotlevel measurementsofAGCareessentiallypointmeasurements.Simpleinterpolationbetweenplotsfails tocapturethefullrangeofspatialvariationinbiomassacrosslandscapes(Houghtonetal.2001; Gibbsetal.2007).Asaresult,thereisawiderangeofmethodstoestimateAGCstocks;Goetzetal. (2009)reviews(andnames)severalcommonmethodsforestimatingregionalAGCstocks, including:

FAGAN AND DEFRIES

64


StratifyandMultiply:assigninginventoryderivedbiomassvaluestolandcoverclasses, CombineandAssign:usingGISdatalayersandlandcoverclassestomodifybiomass estimateswithinclasses(aversionoftheabove),and DirectRemoteSensing:usingfielddatatotrainarulebasedalgorithmuntilithasan optimalpredictionofbiomassfromremotelysenseddata.

Allofthesemethodsinvolvebothremotesensingandgroundinventorydata.Inventorydata arenecessarytoestimateforestcarbonstocksbutthesedataalonearenotsufficient.Goetzetal (2009)discussestheadvantagesanddisadvantagesofeachapproachandwesummarizethat analysishere.Useofdiscretelandcoverclassesanddiscretebiomassvalues(oneperlandcover class)isaStratifyandMultiplyapproach.CombineandAssigngeneratesdiscretelandcoverclasses andcontinuousbiomassvalues(varyingwithGISlayers).TheDirectRemoteSensingapproach generatescontinuouslandcovervaluesandcontinuousbiomassestimates(whichstarts,inthe trainingdata,fromcontinuousspectraldataanddiscretelocalbiomassvalues). Eachoftheseapproachescanbeuseful(Goetzetal.2009).Forexample,existinglandcover maps(e.g.,GLOBCOVER)couldbeusedinaStratifyandMultiplyapproachasaTierI approximationofglobalAGCstocks(GOFCGOLD2008).Todetectforestdegradationnearroads, roadGISdatacouldbeusedtolowerbiomassestimatesinaCombineandAssignapproach(GOFC GOLD2008).Althoughitwasnotoriginallyconceivedassuch,theDRSapproachinGoetzetal. (2009)couldutilizeGISdataandalsoperformrulebasedclassificationoflandcoversintobiomass categoriesatthepixelscale.Wheresufficientdataareavailable,DRSisarguablytherightapproach fortworeasons:theapproachusestheoriginaldataandyieldscontinuousestimatesofbiomass. AsGoetzetal.(2009)illustrate,theuseofbroadlandcoverclassestodetectchangeinbiomass overtimeignoreswithinclassheterogeneityandcompoundsclassificationerrorbetweenland covermaps.Usingrawremotesensingdataandcontinuousestimateseliminateserrorsfromthe previousclassificationandhomogenizationofcomplexlandscapes.AsaresulttheDRSapproach willhavehigherresolutionandlikelyhigheraccuracywhenestimatingcarbonstocks. EvenwiththeadvantagesofDRS,discretelandcovermappingwillremainessential. ContinuousmapsofbiomassandproductivityfromDRSwouldbemostusefuliftheywere integratedwithdetailedlandcovermapsfortworeasons.First,researchintodistinguishingland covers(e.g.,Sesnieetal.2008)identifiestechniquesandremotesensingdatathatwouldimprove therulebasedclassificationalgorithmsthatproducecontinuousmaps.Second,landcover classificationisimportantforlanduseplanningandglobalmodeling.Currentcarbonandclimate modelsrequirelandcoverinformationtomakeaccurateprojections(DeFries2008).Humanland useintroducesdiscreteboundaries,distinctlanduses,andlegacyeffectstoforeststhatvaried continuouslyinbiomass(Asneretal.2009;Uhletal.1988;Uriarteetal.2004).Discretelandcover classificationallowsevaluationofhumanlandcoverdecisions(e.g.,determiningifbiomassis decreasingduetooilpalmplantationsreplacingintactforest,orduetologgingfollowingroad construction).Landcoverinformationalsocanalterfutureprojectionsforbiomassrecovery;in Brazil,forestregrowthhassimilarbiomasstointactprimaryforestonsandysoil,butregrowthhas averydifferentrangeofexpectedgrowthrates(Castroetal.2003;Uhletal.1988).

FAGAN AND DEFRIES

65

4.4.2FromStockstoFluxes:IntegratingCarbonPoolsandFlows
Modelsarerequiredtointegratecarbonfluxesandstocksonaglobalscaleandtocalculate carbonemissionsfromforests(vanderWerfetal.2006).AsnotedinSection4.4,aboveground carbonstocksrepresentlessthanhalfoftotalcarboninforestsonaverage(althoughthisvaries greatlybetweenforests;seeTable1).Inthisreportwehavenotdealtwithdynamiccarbongas fluxesfromlandusechange,whichcanbefastorslow(e.g.,treestumpdecayindeforestedpasture; CrutzenandAndreae1990),orwithchangesinnonAGCstocks(e.g.,lossoflitterinburned forests).Dynamiccarbonfluxes,especiallyinbelowgroundcarbon,aredifficulttomeasuredirectly andareusuallymodeled(e.g.,vanderWerfetal.2006).Wehavefocusedinsteadonhowremote sensingcandirectlymeasuretwoimportantdatainputstocarbonmodels:abovegroundcarbon stocks(areaxbiomass),andchangesinthosestocksovertime(deforestation,reforestation,net growth)(DeFriesetal.2007).Modelestimatesofforestcarbonfluxescanbealsobeimproved, however,bysatellitemeasurementsofcarbongasuptakeandrelease,vegetationproductivity,and theextentandfrequencyofforestfires(Palmer2008;Ramankuttyetal.2007;Rosenqvistetal. 2003;Royetal.2005;Alencaretal.2006;Gibbsetal.2007;Chuvieco2008;Heinschetal.2006;Liu etal.2002). Thetopicofcarbongasfluxesisreviewedextensivelyelsewhere(Chuvieco2008;Palmer2008) andisdealtwithonlybrieflyhere.Theworldwidenetworkofeddycovariancetowers, supplementedbybottledataandorbitingcarbongasmeasuringsatellites,samplecarbongasfluxes overtimefromvegetation(Palmer2008;Baldocchi2008).Forexample,theIBUKIsatellite measurescarbondioxideandmethaneat10.5kkmresolutiongloballyeverythreedaysandis improvingourunderstandingofglobalcarbonsourcesandsinks(Palmer2008).Thisdata collectionnetwork,bymeasuringcarbongasfluxes,isabletocapturecarbonstockchangesfrom forestdegradation(asexposedlitterandwooddecomposetocarbongases)andvegetation productivity(asvegetationuptakescarbondioxidetogrow).Althoughthemeasurementsdepend onmodelingtoseparateabovegroundfluxesfrombelowgroundfluxes,carbongasflux measurementsprovideanindependent,ultracoarseresolutionmetricbywhichtoevaluatethe higherresolutionestimatesofaboveground,netprimaryproductivityderivedfrompassivesensors likeMODISandAVHRR(Heinschetal.2006;Xaioetal.2008). Biomassburning,inadditiontodepletingcarbonstocks(Chuvieco2008),hasanumberof effectsonregionalairquality,hydrologicalcycles,andevensoilfertility(CrutzenandAndreae 1990).Firesaredetectablebyanumberofhighrevisit,coarseresolutionpassivesensorswith thermalbands(e.g.,MODIS,AVHRR,ATSR),byGOESseveralotherultracoarseresolutionsensors, andbyseveralmoderateresolutionsensors(e.g.,ASTER,Landsat)(Chuvieco2008;Schroederetal. 2008b).Realtime,globalfiremappingisongoingat1kmresolutionusingMODISandfuture satelliteswillcontinuethiseffort(e.g.,NPOESS)(Daviesetal.2009).Globalfiremonitoringhas beeninvaluableinaidingfiredetectioninwildernessareasandincontributingtocarbonandair qualitymodels(Daviesetal.2009;Chuvieco2008).Althoughmodelingcanovercomesomeofthe problemsincurrentglobalfiredetection,shortcomingsincludelowerchancesoffiredetectionin cloudyareas,difficultyindetectunderstoryfiresindenseforests,andtheinabilityofcoarser resolutionsensorstodetectthemajorityofsmallerfires(Schroederetal.2008a,b).Evenwhen firescannotbedetectedusingdirectmeans,bothpassiveandSARsensorsareabletomapburn scars(Giglioetal.2005;SiegertandRuecker2000).Modelingcarbonemissionsfromfiresfor

FAGAN AND DEFRIES

66


carbonemissionrequiresdetailedinformationonvegetationbiomassineitheractivefirelocations orburnscars(Chuvieco2008).Accurateestimatesoffireradiativepower(FRP)arealsoagood proxyforbiomasslost,buttheyrequirefrequentmeasurementsbyapassivethermalsensorwitha largedynamicrangetopreventsaturation(Giglioetal.2006).Giglioetal.(2006)mapglobal patternsinFRPusingMODIS,andRobertsetal.(REF2005)areabletopredict90percentof variationinfirebiomassemissionsinsouthernAfricawiththegeostationarySEVIRIsensor. Currently,onlyMODISandsomegeostationarysatellitesmeettherevisitandsensitivity requirementsforcalculatingFRP(Giglioetal.2006),butseveralfiredetectingcoarseand moderateresolutionfiresensorsareplannedforlaunchinthenextsixyears,includingtheNPOESS series,thehigherresolutionGOESABIsensors,andtheSACDfiresatellite(Appendix).Asglobal biomassmapsandsatellitemeasurementsofFRPimprove,modeledestimatesofbiomass emissionsfromfirewillalsocontinuetoimprove. Carbonmodelingiscriticaltoanyglobalforestmonitoringeffortasamethodofintegrating abovegroundcarbonstockmeasurementswithbelowgroundcarbonestimatesandimportant carbonfluxes(e.g.,theeffectoffires,orsoiluptakeofcarbon)(e.g.,vanderWerfetal.2006).Such modelingcanalsomakepossibleindependenttestsofbiomassandproductivityestimatesusing grounddata(e.g.,eddycovariancemeasurementsofcarbondioxide).Futureadvancesinremote sensingwillhelptodecreaseuncertaintyincarbonmodelsbyimprovingtheaccuracyof measurementsoflandcover,standingbiomass,vegetationproductivityanddecomposition,and emissionsfrombiomassburning.

FAGAN AND DEFRIES

67

Chapter 5. Forests from Space: Unparalleled Measurements, Timely Moment


WithdevelopmentsinsatellitetechnologyandtheglobalizationofGISlayerswecanviewthe wholeearthwithmoredetailthaneverbefore.Thesedevelopmentsleadtoatimelymomentfor globalforestmonitoring,aneffortfurtherenhancedbytechnologicalinnovationandan accumulationofforestinventorydata(Herold2009).Withthetoolsnowathandaglobalforest monitoringeffortcanbeguidedbyprinciplesofrepeatability,longevity,andaccuracy.Tobe successful,monitoringwillhavetoreducethegreatestsourcesoferrorinmeasuringforest characteristics.Currentlyforestareacanbemeasuredquiteaccuratelybutonlyatacoarse,two classresolution.Forestbiomassandcarbonstockscannotbeestimatedathighresolutionbut conservativeestimatesarepossiblebycombiningforestinventoryandLIDARdatawithmore spatiallyextensivesatelliteimagery.Thisconcludingchapterexamineshowcurrentsatellitedata andexpectedsensorlaunchesbetween2009and2015couldbeusedinaglobalforestmonitoring effort.

5.1 Improving Forest Biomass and Carbon Estimates


Inglobalforestmeasurement,estimatesofforestbiomassandcarbonarethemostvariable becauseitisdifficulttoestimateexistingstocksreliablyanddetectwhenthosestockshave decreased(Houghton2005).Ingeneral,remotesensingismuchbetteratdetectinglarge,canopy levelchangesinbiomassthansmall(subpixel),belowcanopychangesinbiomass(Peresetal. 2006;DeFriesetal.2007;Broadbentetal.2008).Variableconditionsofillumination,elevation,and moisturemakedetectionoftemporalchangemorechallenging(SanchezAzofeifaetal.2009;Asner 2001;Kellndorferetal.1998;Schroederetal.2006).Fromthestandpointofcost,image availability,andrepeattime,globaleffortstomapforestswillnotbeabletouselargeamountsof highresolutionimagerytodetectsmallscaleforestchanges(e.g.,logging;Achardetal.2007; DeFriesetal.2007;Rosenqvistetal.2003).AsstatedinChapter4,itismoredesirabletouseother typesofsensorstodetectchangesinabovegroundforestbiomassandstructure.Severalstepscan betakentoreduceuncertaintyinarea,biomass,andcarbonestimatesfromsatellitesensors.

5.1.1FirstOrderUncertainties:ForestArea
Becauseevendegradedforestscontainalargeamountofbiomassrelativetoagricultural ecosystems,measuringintactanddegradedforestareaisthefirststepinforestbiomassestimation (DeFriesetal.2007;Ramankuttyetal.2007;Houghton2007).Forest/nonforestarea(aLevelI classification)canbemappedwithveryhighaccuracy,buttheonlyglobalforestmapsareata coarsescaleandmissthemajorityofdeforestation.Global,moderateresolutionestimateshave beenmadebysubsampling(Achardetal.2008).Further,itisnotpossibletodistinguishtree plantations,agroforestry,andforestregrowthfrommatureforestusingcurrentcoarseresolution forestmaps(SanchezAzofeifaetal.2009;Hansenetal.2008).Giventheincreaseintropicalforest regrowthindeforestedareas(Asneretal.inpress)andtherapidexpansionoftreeplantations globally(FAO2006),theseshortcomingshavelargeandgrowingimportanceforglobalbiomass estimates.

FAGAN AND DEFRIES

68


Moderateresolutionopticaltimeserieshavebeenusedinseveralnationalmappingeffortsand haveenabledmappingofforesttypesanddeforestationwithmediumtogoodaccuracy(Section 4.1.5).ThisapproachcouldbeusedtomapglobalforestareatoLevelIIforesttypes,although achievinggoodglobalaccuracywouldbechallenging.Asanaddedbenefit,theseseriesextendback tobefore1980. 5.1.2ForestAreaandthePromiseofNewTechnology Between2009and2015,globalhighresolutioncoveragewilllikelybeachievedbytheGeoEye andRapidEyeconstellations.Thisglobaldatasetwillbeinvaluableinvalidationandalgorithm trainingforforestmonitoringefforts.ThegeostationaryGOESsatelliteconstellation,amongothers, shouldimproveto1kmresolutionatnadirandmakerealtime,coarseresolutiondeforestation detectionpossible.Thelaunchofseveralmoderateresolutionsatelliteswouldcontinuethese imageryseries,increasethenumberofbandsavailableformultispectrallandcovermapping,and makepossiblehighlyaccurate,hyperspectralgloballandcovermaps.Thedramaticincreaseinthe numberofSARsatellites,availablebands,andconstellationsshouldmakeSARpassivefusiona commontechniqueandallowmultipolarandInSARmappingoflandcoverandstructure.The creationofannualforestmapsfromALOSisafirsteffortinthisdirection(Kellndorferetal.2008). OpeningtheLandsatarchivehasenabledseveralnewefforts,includingthereanalysisofpast imagery,theuseoftemporalinformationinclassifyingmoderateresolutionimages(e.g.,Helmeret al.2009),andthecreationofglobal,moderateresolutionlandcoverclassifications(e.g., Cunninghametal.2002).Allofthesedevelopmentsshouldhelptoincreasetheaccuracyofland covermapping,bothpastandfuture.

5.1.3SecondOrderUncertainties:ForestBiomass,Degradation,andGlobalChange
Forestbiomassandcarboncanbeestimatedusingfusionsofpassive,sensors,active sensors,andfielddataalthoughtheaccuracyofremoteestimatesdependsonaccuratefielddata (seeSection3.1.3).TheglobalavailabilityofLIDARdatafromICESatGLAS,whencombinedwith MODISimagery,SARsensors,andlargegroundtruthdatasets,hasmaderegionalbiomassmapping andvalidationfeasible(e.g.,Saatchietal.2007;Baccinietal.2008;Boudreauetal.2008).Unlike forestareahowever,historicalreferencemapsforglobalforestbiomassdonotcurrentlyexist. Conservativeestimatesofbaselineforestbiomasswillhavetobegeneratedfromcurrentremote sensingandforestinventorydata(Gibbsetal.2007;GOFCGOLD2008;Rosenqvistetal.2003; Olanderetal.2008).Thelackofreferencebiomassmapsbefore20052009impedesboth internationalpolicyandscientificinquiry(Gibbsetal.2007;Grainger2008).Temporalchangein biomassisimportantforcarbonmodeling,monitoringforestdegradation,characterizingforest regrowth,andpredictingforestcarbonstocks.Rosenqvistetal.(2003)arguethatitmaybe possibletousetheJERS1datatodevelopreferenceglobalbiomassmaps,butthisismaybe unlikelygiventhelowbiomasssaturationofLbandSAR.Giventhenumberandvarietyofexpected activesensorsinthenextfewyears,ahindcastingapproachthatusesmoderndatatocreatepast biomassestimatesmightbepossible(seebelow). Largescalehumanandnaturaldisturbanceslogging,fires,pestoutbreaks,andglobal warmingalsocauseuncertaintyinforestbiomassestimates.Thisuncertaintyexistsbecausethese disturbancesareeitherdifficulttodetect(logging,somepestoutbreaks,firesincloudyareas)or becauseeffectsaredifficulttopredictandpervasive(firesintropicalfrontiers,globalwarming).

FAGAN AND DEFRIES

69


Progressistakingplaceindetectingloggingusingmoderateresolutionimagery(Asneretal.2005) andtheincreasednumberofhighresolutionandSARsensorsinorbit(e.g.,theRapidEyeor Sentinel1constellations)arelikelytomaketheglobalmappingofstructuralchangesinforests possible(althoughatlargercostthanpreviously).Determininghowglobalforestbiomasswill respondtotheeffectsoftropicalfires,deforestationclimateinteractions,andincreasesinforest respirationratesinawarmerworldwillrequireaccurateinputstocarbonandclimatemodelsfrom remotelysenseddata,includinglandcover,firefrequency,carbongasfluxes,phenologydata,and carbonstocks.Tocreatetheseestimates,concertedglobalforestmonitoringusingbothremote sensingandgrounddataisrequired.

5.1.4ForestBiomassandthePromiseofNewTechnology
WiththeplethoraofSARbands,polarities,andconstellationsexpectedtolaunchbetween 2009and2015,anumberofmultibandfusionpossibilitieswillarise.TheaccumulationofLIDAR databynewsatellitesandaerialsensorswillrefinebiomassmapsbasedonopticalandSAR sensors.Itislikelythatbigstepsforwardwillbetakeninmeasurementofforestheight,structure, biomass,andgroundelevationinthenextfewyearsbecauseoftheICESat2mission,theDESDynI andBIOMASSmissions,andthewealthofSARdata.Thedevelopmentofforestbiomassmapsis verylikelyintheneartermparticularlywiththeexpectedlaunchofDESDynIin2015andthe potentiallaunchofBIOMASSinlate2015orearly2016.Estimationofbiomassfromremotesensors facestwokeychallenges,however.Thefirstchallengeistocollectadequatereference,orground truth,dataweaddressthisissuebelow.Thesecondchallengeistocreateaccuratereference biomassmapsfor2000or1990,likethoserequestedbypoliticalagreementsandscientists(Gibbs etal.2007;Rosenqvistetal.2003).Thischallengecanonlybesolvedbyconservativeestimates and/orhindcasting. Hindcastingpastforestbiomassfromthecurrentrelationshipofbiomasswithremote sensingdataispossibleusing,forexample,eithertheDirectRemoteSensing(DRS)approachorthe StratifyandMultiply(SM)approach(Section4.4.1).OnepossibleDRSapproachwouldcombinea stilltobedevelopedgroundlevelDEMwithreferenceInSARdatafromthe1994and2000shuttle missions,creatingglobalforestheightmaps(e.g.,Walkeretal.2007;Kellndorferetal.2004).These historicheightmapscouldbeconvertedintobiomassestimatesusingcurrentrelationships betweenInSARheightsandbiomass(LIDARandgroundderived).Thesebiomassestimatescould evenbeimprovedbyfusingthemwithspectralinformationfromtheLandsatarchive.Alternatively, anSMapproachcoulduseground,SAR,andLIDARdatatodeveloprelationshipsbetween moderateresolution,LevelIIlandcovermapsandcurrent,conservativebiomassestimates(e.g., Helmeretal.2009).Historic,moderateresolutionLevelIIlandcovermapswouldbeusedto estimatereferenceforestbiomass. Inthenextsixyears,multipolar,repeatpassInSARandtheGLASandDESDynIfootprint archiveswillmarkedlyimproveDEMandBIOMASSmeasurements(ESA2009;NASA2009a,Saatchi etal.inreview).TheeightyearMODISrecordofvegetationproductivityanditscontinuationbythe NPOESS,GCOM,andSentinel3serieswillcontinuetobeusefulinmonitoringlandcoverand vegetationresponsetoclimatechange.TherecordofMODISindailyphenologyandfire measurementcouldbeimprovedtorealtimebycoarseresolutiongeostationarysatellites.Global modelingofcarbonstocksandfluxeswillbeaidedbyallofthesesensors.Globalcarbongas

FAGAN AND DEFRIES

70


monitors(e.g.,IBUKI)andhyperspectralrefinementofgloballeafareaestimateswillalsobe beneficial(DeFries2008;Palmer2008).

5.1.5ImprovedEstimatesRequireImprovedGroundTruthing
Despitethepromiseofnewtechnology,highqualitygeoreferencedgroundtruthdatawill continuetobeessentialinvalidatingandimprovingestimatesfromsatelliteimagery(Gibbsetal. 2007;Goetzetal.2009).Forexample,oneofthebiggestcausesofuncertaintyinestimating biomassintheAmazonandCongobasinsisthelackofgroundtruthforestinventorydataover largeareastotrainremotesensingalgorithms(Saatchietal.2007;Baccinietal.2008).Larger, higherqualitygroundtruthdatasetsimproveestimatesfromremotesensingimagery(Foody 2002).Toachieveaccurateestimates,aglobalforestmeasurementeffortwillhavetocollatealarge numberofexistinggroundtruthdatasetsandcoordinateadditionaldatacollectioningapareas (e.g.,centralAfrica;Herold2009)andneededmeasurements(regionalwooddensitypatterns; Bakeretal.2004).Ideally,groundtruthingandhighresolutionLIDARandopticalimagery (satelliteorairborne)shouldbothbeusedbutcollectionofforestinventorydataandaerialLIDAR datahaslaggedbehindsatellitedatacollectioninmanycountries(Herold2009). Thedependenceofaccurateremotesensingestimatesonlargeamountsofgroundtruthdata hasimportantimplicationsforanyglobalforestmeasurementeffort.AsnotedinSection3.5, groundtruthingisusuallythemostexpensivepartoflargescaleremotesensinganalysis.Different typesofgroundtruthing(e.g.,forestinventory,LIDAR,highresolutionoptical)wouldneedtobe coordinatedinasamplinghierarchyforefficiency(Table3;Patenaudeetal.2005;DeFriesetal. 2007).Inaglobalforestmeasurementeffort,thelogisticaldifficultiesofairandgroundaccessare nottrivialandcarefulcoordinationofaerialandgrounddatawithsatelliteimageryisapractical necessity(Achardetal.2007;GOFCGOLD2008;Rosenqvistetal.2003;DeFriesetal.2007).

5.2 In Conclusion: Toward Improved Measures and Monitoring


Accurateglobalestimatesofforestarea,foreststructure,forestbiomass,andcarbonare achievableusingremotesensingtechnologiesthatareavailablenoworthatwillcomeonlineby 2015.Forestbiomassandforestcarbonstockscannotbemeasureddirectlybutconservative estimatesarepossibleusingfusionsofcurrentforestinventorydataandsatelliteimagery.Global measurementofbiomassshouldbepossiblewithnewremotesensingtechnologyespeciallyfrom 2015onward. Anyglobalefforttomeasureforestsusingremotesensingtechnologieswillfaceseveral technicalchallenges.Groundtruthdatamustbeacquiredacrossdiversecountriesandecosystems andsharedglobally.Creatingglobalforestmapswithhigh,regionallyconsistentaccuracyis demanding.Choicesarenecessaryinselectingimagerywiththeappropriatetradeoffincost, resolution,swath,andrepeattime.Challengesalsoremaintoprocessandfuselargeamountsof imageryinagloballyconsistentmannerandtousenewremotesensingtechnologiestomeasure biomassaccuratelyoverlargeareas. Surmountingthesechallengeswillrequireplanning,globalcoordination,andunderstandingof theadvantagesandlimitationsofindividualremotesensingtechnologies.Giventhenumberof inaccuraciesincurrentglobalforestmeasurementbytheFAO,countrylevelcensuses,andother bodies(Waggoner2009),thereisaneedforimprovedmonitoringoftheworldsforests.Remote

FAGAN AND DEFRIES

71


sensinghasastrongtrackrecordinglobalforestmeasurementandplannedsensorlaunchesinthe nextfewyearshaveevengreaterpotential.Institutionalarrangementsthatimplementasystematic approach,integratingsatelliteandgroundtruthdata,wouldallowus,forthefirsttimeinhistory,to conductaccurate,globalmonitoringoftheworldsforests.

FAGAN AND DEFRIES

72

Satellite, Sensor, and Remote Sensing Acronyms and Abbreviations


AirMISR
ABI ALOS ASTER ATSR AVHRR Avnir2 CBERS CHRIS COSMOSkyMed DESDynI DETEX DMC DMSAR EnMAP Envisat EO1 EOS ERS1 FY3 FY4 GCOMC1 GLAS GLC2000 GOES AirborneMultiangleImagingSpectroRadiometer AdvancedBaselineImager AdvancedLandObservationSatellite AdvancedSpaceborneThermalEmissionandReflectionRadiometer AlongTrack ScanningRadiometer AdvancedVeryHighResolutionRadiometer AdvancedVisibleandNearInfraredRadiometerType2 ChineseBrazilianEarthResourcesSatellite CompactHighResolutionImagingSpectrometer ConstellationofSmallSatellitesforMediterraneanBasinObservation, Deformation,EcosystemStructureandDynamicsofIcesatellite DetectionandMonitoringofSelectiveLoggingActivities DisasterMonitoringConstellation DisasterManagementSAR EnvironmentalMappingandAnalysisProgram EnvironmentalSatellite EarthObservingSatellite EarthObservingSatellites EuropeanRemoteSensingSatellite FengYun3 FengYun4 GlobalChangeObservationMission GeoscienceLaserAltimeterSystem GlobalLandCover2000 GeostationaryOperationalEnvironmentSatellite

FAGAN AND DEFRIES

73


HJ1C HRG HySPIRI HySI ICESat IMS InSAR IRS JERS1 LDCM LIDAR MERIS Meteosat MetOp MISR MODIS NPOESS PALSAR POLDER PolinSAR PROBA Radarsat SAOCOM SAR SEVIRI SIR SPOT5 HuanJing(environment) Highresolutiongeometrical HyperspectralInfraredImager HyperspectralImager Ice,Cloud,andLandElevationSatellite InteractiveMultisensorSnowandIceMappingSystem InterferometricSyntheticApertureRadar IndianRemoteSensing JapaneseEarthResourcesSatellite LandsatDataContinuityMission LightDetectionandRanging MediumResolutionImagingSpectrometer MeteorologicalSatellite MeteorologicalOperationalSatellite MultiangleImagingSpectroRadiometer ModerateResolutionImagingSpectroradiometer NationalPolarorbitingOperationalEnvironmentalSatelliteSystem PhasedArraytypeLbandSyntheticApertureRadar PolarizationandDirectionalityoftheEarth'sReflectances PolarimetryandInferometry(toanalyzeimagesfromatleasttwoimagesfrom) SyntheticApertureRadars ProjectforOnBoardAutonomy RadarSatellite SatliteArgentinodeObservacinConMicroondas SyntheticApertureRadar SpinningEnhancedVisibleandInfraredImager ShuttleImagingRadar SatellitePourl'ObservationdelaTerre

FAGAN AND DEFRIES

74


SPOTHRG SRTM TanDEMX TerraSARX TESHYS UAS VENUS VNIR SatellitePourl'ObservationdelaTerreHighResolutionGeometric ShuttleRadarTopographyMission TandemDigitalElevationMappingSatellite TerraSyntheticApertureRadarSatellite TechnologyExperimentalSatelliteHyperspectral UnmannedAerialSystems VegetationandEnvironmentMonitoring VisibleandNearInfrared

Acronyms
AGB AGC DBH DEM DRS EOS FAO fAPAR FIA FRA FRP GeoCoverLC GIS INPE IPCC LAI abovegroundbiomass abovegroundcarbon treediameteratbreastheight digitalelevationmodel directremotesensing earthobservingsatellites UnitedNationsFoodandAgricultureOrganization photosynthetically activeradiation U.S.ForestServicesForestInventoryandAnalysisNationalProgram ForestResourceAssessments fireradiativepower global,moderateresolutionlandcovermap geographicinformationsystems BrazilianNationalInstituteforSpaceResearch IntergovernmentalPanelonClimateChange leafareaindex

FAGAN AND DEFRIES

75


REDD SM UAE UNFCCC USGS reducingemissionsfromdeforestationandforestdegradation stratifyandmultiply UnitedArabEmirates UnitedNationsFrameworkConventiononClimateChange UnitedStatesGeologicSurvey

FAGAN AND DEFRIES

76

Appendix. Current and Near-Term Earth-Observing Satellite Technology: An Overview


Thisguidetocurrentandfuture(2009toearly2016)earthobservingsatellitesisnotan exhaustivelist.Theauthorsattemptedtocoverallsatellitesthatarelikelytocreateregionalto globalcoverage,butanticipatethatsomesatelliteswereaccidentallymissed.Mostultracoarse(2+ kmspatialresolution),military,andnonlandimagingsatelliteswereexcluded. TheinformationprovidedinthesectiononFutureSatellitesisascurrentaspossible,but improvementsintechnology,delaysinsatellitelaunching,missioncancellations,andoutdated informationsourcesareinevitable.Theauthorswelcomecorrectionsandupdatesfromreaders. Thesatellitesspecificationsinthisappendixcomeprimarilyfrompubliclyavailable governmentalandcommercialwebsites.Thefollowingarethemainsourcesusedtocompilethe appendix.

CommitteeonEarthObservationSatellites.2009.CommitteeonEarthObservation Satellites.http://www.ceos.org/(accessedMarch28,2009). eoPortal(sponsoredandrunbytheEuropeanSpaceAgency).2009.ListofEOandnonEO SatelliteMissions. http://directory.eoportal.org/missions_all_list.php?filter=&view_all&order=start_ date&dir=ASC(accessedJuly23,2009). GOFCGOLD.2008.Reducinggreenhousegasemissionsfromdeforestationanddegradation indevelopingcountries:asourcebookofmethodsandproceduresformonitoring, measuring,andreporting.GOFCGOLDReportversionCOP132,editedbyGOFC GOLDProjectOffice.Alberta,Canada:NaturalResourcesCanada. Kramer,HerbertJ.2002.ObservationoftheEarthandItsEnvironment:SurveyofMissions andSensors,4thedition.NewYork:SpringerVerlag. Powell,S.L.,D.Pflugmacher,A.A.Kirschbaum,Y.Kim,andW.B.Cohen.2007.Moderate resolutionremotesensingalternatives:areviewofLandsatlikesensorsandtheir applications.JournalofAppliedRemoteSensing1.

FAGAN AND DEFRIES

77

CurrentModerate(200to15m)ResolutionOpticalSensorsRelevanttoForestMapping
Satellite& Sensor Landsat5TM LaunchDate Source Spectrum Resolution, ReturnInterval 30m,16days Image Width 185km Notes&GlobalImagery(GI)Extent CostforData Acquisition Free.Formerly600 US$/scene,0.02 US$/km2.

1984

USA

VTIR,7 bands

Imagesevery16daystoanysatellite receivingstation.Operatingbeyond expectedlifetime.ContinuousGIto 1985,spotty1985topresent. OnApril2003thefailureofthescan linecorrectorresultedindatagaps outsideofthecentralportionof images,seriouslycompromisingdata quality.Stilloperating.ContinuousGI. Dataisacquiredonrequestandisnot routinelycollectedforallareas.Spotty GI,pointable. Anexperimentalsensor,ALI'slifetime wasextendedattherequestof researchers.SpottyGI. Asanexperimentalsensor,Hyperion hashadissueswithsignaltonoise ratio.SpottyGI. Constellationmember,hoststwo moderateresolutionsensors. ContinuousGI.

Landsat7ETM+

1999

USA

VTIR,8 bands

30m(15pan), 185km 16days

Free.Formerly600 US$/scene,0.02 US$/km

TerraASTER

1999

USA/Japan VTIR,14 bands USA VSWIR,10 bands

1530m,16days

60km

60US$/scene,0.02 US$/km Free

EO1ALI

2000

30m(10pan), 37km 16days

Eo1Hyperion

2000

USA

VSWIR, 220bands. VSWIR,4 bands

30m,16days.

7.5km

Free

Resourcesat1 LISSIII&AWIFS

2003

India

23.5m,24days &56m,5days

142km& 730km

IRS1C/1DLISSIII &WIFS

1995,1997

India

VSWIR,4 bands&2 bands

23.5m,24days, &188m,5days.

142km& 810km

860US$/scene, 0.05US$/km,& 1790US$/scene, 0.004US$/km Constellationtwinsatellites,eachhosts 860US$/scene, twomoderateresolutionsensors. 0.05US$/km,& ContinuousGI? TBD

78


Satellite& Sensor LaunchDate Source Spectrum Resolution, ReturnInterval Image Width Notes&GlobalImagery(GI)Extent CostforData Acquisition

CBERS2,2B HRCCD

1999,2003

China/ Brazil

VNIR,5 bands

20m,26days

113km

CBERS/IRMSS

1999,2003

China/ Brazil,

VTIR, 4 bands

80160m,26 days

120km

Experimental;Brazilusesondemand imagestobolstertheircoverage. ContinuousoverBrazil,Africa,and Chinaonly. Mountedonthesamesatellitesasthe HRCCDsensor,above.Continuous overBrazil,Africa,andChinaonly.

FreeinSouth America,China, andAfrica FreeinSouth America,China, andAfrica

DMCseries

20022008.

Algeria/ VNIR,3 China/ bands Nigeria/ Turkey/UK. France France VNIR,4 bands VSWIR,5 bands VNIR,63 bands(18 full) VNIR,1 band VNIR,3 bands VNIR,4 bands

2232m,daily possible

160600 km

Commercial;Brazilusesalongside 585/scene,0.048 Landsatdata.Consistsofmultiple /km satellitesinthesametrack.Continuous GI. Twosensorscanexpandwidthof image.Pointable,continuousGI. Twosensorscanexpandwidthof image.Pointable,continuousGI. Experimentalhyperspectralsensor withextendedlifespan.Ithastwo modes:fullresolutionhasfewer bands.SpottyGI. Thispanchromaticcamerahasspotty GI,withaconcentrationoverSouth America. Twosensingmodes,higherorlower resolution. Thisminisatellite'simagerywillbe providedtodevelopingcountries.Itis unclearwhetherthissatellite'sdatais availabletothepublic.
79

SPOT2HRV SPOT4HRVIR

1990 1998

20m(10pan),2 60117km 3days 20m(10Pan), 60km 23days 34m(17mfull), TBD 17km

2000/scene,0.5 /km 2000/scene,0.5 /km

ProbaCHRIS

2001

Europe

SACCHRTC

2000

Argentina

35m,TBD

90km

Freeto researchers,TBD Free

MonitorERDSA IMS1MxT

2005 2008

Russia India

20or40m,TBD 37m,TBD

160or 890km 151km

CostTBD. Free


Satellite& Sensor THEOSMS LaunchDate Source Spectrum Resolution, ReturnInterval 15m,15days Image Width 90 km Notes&GlobalImagery(GI)Extent CostforData Acquisition CostTBD.

2008

Thailand

VNIR,4 bands VNIR,4 bands

Pointable,withapanimagerdetailed inthehighresolutionsection.

HJ1A/1B

2008

China

30m,4days

360or 720km

Pointable,twosensors.Onboard CostTBD. hyperspectralimagerdetailedinthe coarseresolutionsection.Itisunclear whetherthissatellite'sdataisavailable tothepublic. Thiscloudimageriscombinedwitha LIDARinstrumentforanalysisof clouds. CostTBD.

WFC(Calipso)

2006

USA

Vis,1band

125m,TBD

60km

80

CurrentCoarseResolution(>200mpixel)OpticalSensorsRelevanttoForestMapping
Satellite& Sensor AVHRR LaunchDate Source Spectrum Resolution, ReturnInterval 1.1km,Bidaily Image Width 3000km Notes&GlobalImageryExtent(GI) CostforData Acquisition Free

19782005

USA

VTIR,46 bands. VTIR,36 bands

Alongtermsatelliteconstellation,themost recentisAVHRR/3.ContinuousGI. Mountedontwosatellites.Vegetation bandsare250mresolution.Thermalbands areusedtodetectfires.ContinuousGI.

MODISTerra& Aqua

2000,2002

USA

2501000m,2 days

2330km

Free

GOES1013 Imager

19752006

USA

NTIR,5 bands

1kmvisible,5km Fullearth IR,realtime disc.

GOESisincludedasanexampleof Free geostationarymeterologicalimagersandis ultracoarse(5kmpixel)outsidethesingle visibleband.India,Russia,China,andEurope allhavegeostationarymeteorological constellationsaswell.Itisnotincludedin Figure13. ContinuousGI. Stereoimagingfromnineangles.Ittakes6 imagesonLocalsetting(275m)eachday, andtherestonGlobal. Thiscoarseresolutionsensorisoptimized foroceanremotesensing.ContinuousGI. Thiscoarseresolutionsensorcanmonitor seasurfacetemperatureandfires. ContinuousGI. Bandprogrammable,canbeselectedacross VNIRrange.ContinuousGI. Thermalbandsareusedtodectectfires. ContinuousGI.
81

Spot4/5 Vegetation MISR

1998 1999

France USA

VNIR,4 bands VNIR,4 bands

1.1km,Bidaily 275mOR550 1100m(Global), 9days 1.13km, daily

2200km 360km (Local)

CostTBD. Free

Orbview2 SeaWIFS ERS2ATSR2

1997

Comm. (USA) ESA

VNIR,8 bands. VTIR, Microwave, 10bands VNIR,15 bands VTIR,7 bands

1500km

CostTBD.

1995

1km,TBD

500km

CostTBD.

EnvisatMERIS EnvisatAATSR

2002 2002

ESA ESA

260x300m,3 days 1km,TBD

1150km 500km

CostTBD. CostTBD.


Satellite& Sensor ADEOSIIGLI LaunchDate Source Spectrum Resolution, ReturnInterval 2501000m,4 days 1km,TBD 1km,TBD 236360m,daily 175m(300pan), TBD 100m,TBD 505.6m Image Width 1600km Notes&GlobalImageryExtent(GI) CostforData Acquisition CostTBD.

20022003

Japan

VTIR,36 bands VTIR,6 bands. VTIR,10 bands. VNIR,8 bands VSWIR,5 bands VNIR,110 bands VNIR,64 bands VSWIR,20 bands,&V TIR,10bands VTIR,6 bands VTIR,10 bands VNIR,8 bands VTIR,3 bands VNIR,2 bands

Thisshortlivedsatellitewasverysimilarto MODIS,butwith29bandsintheVNIR. Thissatellitehasaprogrammable orientationwithinitswidefieldofview. Thesesatellitesareoptimizedforocean observation,with6bandsinthevisible. Thisgeostationaryoceancolorobservation satellitewillobservealimitedarea. Thissatellitemountsapanchromaticfire andlightningimagerandalandimaging sensor. Thishyperspectralsatellite'simagerymay notbeavailableforpublicpurchase. Thisexperimentalhyperspectralsatellitewill beoperatedoverIndiaonly. TheFY3constellationwillhosttheMERSI andVIIRinstrumentsforcoarseresolution earthobervation. Thistemperatureandcloudmapping instrumentcanalsotakelandcoverdata.

Insat2E,3A HY1B,1C,1D COMS1GOCI SACCMMRS, HTSC HJ1A/1B IMS1HySI FY3series: MERSI,VIRR

1999 2007 2009 2000

India China Korea Argentina

6000km 3083km. 1440km 360km (700pan) 50km 125.5km

CostTBD. CostTBD. CostTBD. CostTBD.

2008 2008 2008

China India China

CostTBD. CostTBD. CostTBD.

250m1km&1.1 2800km km,TBD

MSUMR (MeteorMN1) MVISR(FY1D)

2008

Russia

1km,TBD

2800km

CostTBD.

2002

China

1.1km,TBD

2800km.

Itisunclearifthisland andoceanobserving CostTBD. satellite'simageryisavailableforpurchase. Thesesatellitesareprimarilyoceancolor monitorsandtakedataprimarilyaround India. Thiscloud andcloudtemperature monitoringsensorcanalsodetectfires. Thissensorisdesignedtodetectvegetation.
82

OCM (Oceansat1/2) OLS(DMSP series) WFI(CBERS 2/2B)

1999,2008

India

236360m,2 days 560m,TBD 258m,TBD

1440km

CostTBD.

19972009 2003,2007

USA Brazil/ China

3000km 890km

CostTBD. CostTBD.

CurrentHighResolutionOpticalSensorsRelevanttoForestMapping
Satellite& Sensor SPOT5HRG Launch Date 2002 Source Spectrum Resolution, ReturnInterval Image Width NotesandGlobalImagery(GI) Extent Commercial;Indonesia& ThailandusealongsideLandsat data.Pointable. Commercial.Uniqueorbit:early arrivalatequator,GIisselected areasinstrips Pointable,stereocapability. SpottyGI,developing continuousGI. Pointable,panimageryreleased at0.5m.SpottyGI,developing continuousGI. Pointable.SpottyGI,developing continuousGI. Constellation,allowsfrequent imaging.Designedforglobal cropmonitoring. Thishighresolutionsensor operateswiththeALOSPALSAR sensor,allowingimageryfusion. CostforData Acquisition 2700 /scene,0.5 /km CostTBD.

Comm. (France) Comm. (Taiwan) Comm. (USA) Comm. (USA)

VSWIR,5 bands VNIR,5bands

10m(2.55pan), 60120 23days km 8m(2pan),1 day. 4m(1pan),3 days 24km

Formosat2

2004

IKONOS

2000

VNIR,5bands

1114km

CostTBD.

Geoeye1

2008

VNIR,5bands

1.65m(0.41pan), 15.2km 3days

380 US$/scene, 25US$/km CostTBD. CostTBD.

Quickbird RapidEye15

2001 2008

Comm. (USA) Comm. (Germany) Japan

VNIR,5bands VNIR,5bands

2.4m(0.61pan,3 days 5m,Daily

16.5km 78km

ALOSAvnir2

2006

VNIR,4bands

10m

70km

CostTBD.

Resourcesat1 LISS4

2003

India

VNIR,4bands

5.8m,5days.

70km

Pointableandmountedwithtwo 1070 moderateresolutionsensors.2 US$/scene, launched? 0.3US$/km SouthAfrica'sfirstsatellite,this smallsatelliteisnotlikelyto developcontinuousGI. CostTBD.

SumbandilaSat Imager

2009

South Africa

VNIR,6bands

6.25m,TBD

45km

83


Satellite& Sensor Launch Date Source Spectrum Resolution, ReturnInterval Image Width NotesandGlobalImagery(GI) Extent CostforData Acquisition

ResursDK1 GeotonL1

2006

Russia

VNIR,5bands (pan)

2.53.5m(1m pan),TBD

30km

Kompsat2 TopSAT Telescope Cartosat1,2 IRS1C/1DPAN ALOSPRISM

2006 2005

Korea UK

VNIR,5bands Vis.,4bands

2005, 2007 1995, 1997 2006

India India Japan

VNIR,1band (pan) VNIR,1band (pan) VNIR,1band (pan) VNIR,1band (pan) VNIR,1band (pan) VNIR,1band (pan) VNIR,1band (pan) VNIR,1band (pan) Vis.,1band (pan)

4m(1pan),28 15km days(max) 5.6m(2.8mpan), 17km TBD (pan),12 km 2.5m,TBD 30km 6.5m,24days (max) 2.5m,TBD 70km 70km (35km stereo) 14km 7km 17.5km (110x60 poss.) 24km 90km Not available.

Thiscommercialsatellitehasa variableorbitaldistance, affectingitsswathand resolution. Pointablesensor,commercially available. Experimental,lowcost microsatellite. Panchromaticcamera. Pointable,panchromatic camera. Panchromaticcamerawith triplestereooption. Highlypointable,panchromatic camera. Pointable,panchromatic camera. Pointable,panimageryreleased at0.5m.SpottyGI,developing continuousGI. Panchromaticcamera,unclearif availableforpublicpurchase. Panchromaticcommercial camera. Thismilitarypanchromatic cameraisnotincludedinFigure 13.

CostTBD.

CostTBD. CostTBD.

CostTBD. CostTBD. CostTBD.

EROSA EROSB Worldview1

2000 2000, 2006 Comm. (USA), 2004 2005 2005 2001

Comm. (Israel) Comm. (Israel) Comm. (USA),2004 China Russia India

1.9m 0.7m 0.5mpan,2days

CostTBD. CostTBD. CostTBD.

BJ1PAN MonitorEPSA TESPAN

4mpan,TBD. 8m,TBD. 1m,TBD

CostTBD. CostTBD. Notavailable.

84

CurrentCoarsetoModerateResolutionActiveSensorsRelevanttoForestMapping
Satellite& Sensor SRTM Launch Date 2000(one time) Source Spectrum Resolution, ReturnInterval 30x30m,N/A Image Width Notes&Global Imagery(GI)extent CostforData Acquisition FreeDEMand data.

USA

InSAR.C andXband. Quadpolar(C bandonly).

50225km Resultofa10day shuttlemission.Used tocreatea30mDEM (90moutsideUS). ContinuousGI between60Nand 56S. Resultofashuttle mission.Continuous GI. 75km. ContinuousGI,with temporalinferometry inselectareas. 100kmx ContinuousGI,with 100km. temporalinferometry inselectareas. 50500km Pointable.Continuous x50500 GI,withtemporal km. inferometryinselect areas. 20500km Pointable.Continuous x20500 GI,withtemporal km. inferometryinselect areas. 5150km Pointable.Continuous x5150 GI,withtemporal km. inferometryinselect areas. 1590km

SIRC/X mision JERS1

1994(one time) 1992 (launch, 1997(end) 1991,1995

USA

Japan

InSAR.X,C, Lband.Quad polar. SAR.Lband, HHpolarity. SAR.Cband. VVpolar.

30x1030m, N/A 18x18m,44 days 30x26m,TBD

Free

Free

ERS1,2

ESA

Free

Radarsat

1995

Canada

SARCband. 8100mx8100 HHpolar. m,TBD

Free

Radarsat2

2006

Canada

SAR.C band.Quad polar. SAR.C band.Quad polar.

3100mx2.4 100m,TBD

CostTBD.

Envisat ASAR

2002

EU

301000mx30 1000m,TBD.

CostTBD.

85


Satellite& Sensor Launch Date Source Spectrum Resolution, ReturnInterval Image Width Notes&Global Imagery(GI)extent CostforData Acquisition

TerraSARX

2007

Comm. (Germany)

SAR,Xband. Quadpolar.

118mx118 m,2.511days

ALOSPalsar

2006

Japan

SAR,Lband. Quadpolar.

7100m,TBD

10100km Commerical,variable x5150 resolution,andfirstin km. asatellite constellation. 40350km Resolutionandswath varydependingon mode.ContinuousGI. 10200km Varietyofscanning modes.Firstthree membersofSAR constellation,tandem orbitforDEM. Footprint centers 170m apart. N/A Tracksarecurrently spacedat15km (equator)and2.5km (80degreeslat.) 10.5kmfootprint, operatesincloudfree areasglobally.Takes instantaneous measurementsofCO2 andCH4.

CostTBD.

COSMO SkyMed

20072009

Italy

SAR,Xband. Quadpolar.

330m,26days

210.04 US$/scene, 0.017 US$/km. CostTBD.

ICESATGLAS

2004

USA

LIDAR

70mfootprint, 183days

Free

IBUKI (GOSAT)

2009

Japan

Gassampling, NIRTIR

10.5km,TBD

CostTBD.

86

ExpectedSensorsRelevanttoForestMapping(20092016)
Satellite&Sensor WorldView2 ARGO(RapidEye6) 2009 RazakSat DubaiSat ALSAT2A,2B EROSC TanDEMX 2009 2009 2009 2009 2009 LaunchDate 2009 Source Comm.(USA) Taiwan/ Germany Malaysia UAE Algeria Israel Comm. (Germany) India Spectrum VNIR,9bands. VNIR,5bands. VNIR,5bands. VNIR,5bands. VNIR,5bands. VNIR,bandsTBD. (5?) SAR,Xband.Quad polar. VSWIR,4bands. Resolution,Return Interval 1.8m(0.5mpan),1 4days. 5m,daily. 5m(2.5pan),TBD. 5m(2.5pan),TBD. 10m(2.5m),TBD. 2.8m(0.7mpan). 118mx118m, 2.511days. 23.5m,24days ImageWidth 16.4km 78km 20km 20km 17km 11km 10100kmx 5150km. 141x141 km 30x30km 20km&300 km 10200kmx 10200km. Notes Eightbands(+pan)intheVNIR,pointable. Commercial,partoftheRapidEye constellation. Pointablehighresolution.Maynotdevelop continuousGI. Pointablehighresolution.Maynotdevelop continuousGI. Pointablehighresolution,notlikelyto developcontinuousGI. Pointable,highresolution.Maynot developcontinuousGI. Commercial,variableresolution.Atwin satelliteorbitinginformationwith TerraSARXtocreateaglobalDEM. Thissatellitewillbeverysimilarto Resourcesat1,withitsmoderateandhigh resolutionimagerycapability. Stillunderdevelopment. Thissatellitewillmounttwosensors.The highresolutionsensorhasasmallerimage sizebutmorebands. Varietyofscanningmodes,lastsatellite additiontoCOSMOSARconstellation.

RESOURCESAT2 (ISRO) RASAT(Tubitak) NigeriaSat2 (NASRDA) COSMOSkyMed4 (ASI/MiD(Italy))

2009

2009 2009

Turkey Nigeria

Vis.,4bands. VNIR,5 &4bands.

15m(7.5mpan), TBD. 5m(2.5pan)&32 m,TBD. 330m,TBD.

2009

Italy

SAR,Xband.Quad polar.

87


Satellite&Sensor MeteorMN1 (Roshydromet/ Roscosmos) LaunchDate 2009 Source Russia Spectrum 1)SAR,Xband,VV polarized.2)VTIR,6 bands.3)VNIR,36 bands. Resolution, ImageWidth Notes ReturnInterval 1)4001000m. 1)450600 Thisinnovativesatellitewillhostthree distinctsensors,numberedhere.Thelast 2)1000m.3) km.2)2800 60120m. km.3)960km moderateresolutionwillcombinethree cameraimagesfor60mresolutionin threebands.Thesatellitewillproduce regionalimageryoverRussia.Firstina series. 1)4001000m, 1)450600 Thisinnovativesatellitewillhostthree distinctsensors,numberedhere.Thelast 2)1000m,3) km.2)2800 moderateresolutionwillcombinethree 60120m km,3)960 cameraimagesfor60mresolutionin km. threebands.Thesatellitewilldevelopa regionalGIonly,overRussia. 10.526m(2.1 pan)TBD. 7.8m(pan higher?) 340m. 20m. 1.21.5m. 0.7m,TBD. 250km Sourcesconflictontheexactresolution orbandsofthissmallRussiansatellite. Pointablehighresolutionsensor.

MeteorMN2 (Roshydromet/ Roscosmos)

2009

Russia

1)SAR,Xband,VV polarized.2)VTIR,6 bands.3)VNIR,36 bands.

KanopusVN1 (Roshydromet/ Roscosmos) Sich2

2009

Russia

VNIR,TBD.

2009

Ukraine

VNIR,4bands.

46.6km

RISAT1 HJ1C(SbandSAR) Svea Pleiades1(CNES)

2009 2009 2010 2010

India China Sweden France

SAR,Cband,quadpolar. SAR,Sband. TBD,passivehigh resolution. VNIR,5bands.

30240km 100km 812km 20kmx20 km

HighresolutionSARsatellite. UnusualSbandSAR. Underdevelopment. Extremelypointablesatellitegives+/50 degreesofftrack.

88


Satellite&Sensor TESHYS(ISRO) LaunchDate 2010 India Source Spectrum VSWIR,200bands. Resolution, ReturnInterval 15m,TBD. ImageWidth 30kmx30 km TBD. Notes Moderateresolutionhyperspectral imager,under development/experimental. Proposedgeostationarysatellitewith twocoarseimagers(onehyperspectral) andonemoderateimager. ProposedsatellitewithaSARcomponent andhighresolutionopticalcameras. Amultisensorsatellite,butthefire sensorissummarizedhere.Itwill measuretemperatureandreleased energyoffiresandotherhotevents. Twoinstruments:CoastRegionImager andcoarserOceanColorand Temperaturescanner.Constellation. ThefirstinanewNASAseriesofcoarse resolutionsatellitesintendedtofollow uptheMODISandAVHRRmissions. Asuperspectralcameraintendedtotake highresolution,multiplebandimages forlandcoverapplications. Thissatellite,thefirstoftwo,hostsfour sensorswithregionalcoverage.Thehigh andmoderateresolutionsensorsare pointable(35dayreturntime).Italso hasamoderateresolutionfireimager andacoarseresolutionsensor.

GISAT(ISRO)

2010

India

VTIR,3,4,&150bands.

SARE1(CONAE) SACD(CONAE/ NASA)

2010 2010

Argentina Argentina/USA

SAR. TIR,3bands.

1.5km,50m,& 192320m, TBD. TBD. 350m,TBD.

TBD. 1821000km

HY1C(NSOAS/ CAST) NPP(NASA/NOAA/ DoD(USA)) VENUS(CNES/ISA)

2010

China

VTIR,4&10bands.

250m&1.1 km. 400m1.6km, TBD. 5.3m,TBD.

500km& 3083km 3000km

2010

USA

VTIR,22bands.

2010

IsraelFrance

VNIR,12bands.

27km

CBERS3(CRESDA/ INPE)

2010

Brazil/China

1)NIR,4bands.2)NIR, 4bands.3)VTIR,4 bands.4)VSWIR,4 bands.

1)20m.2)10 m(5pan),326 days.3)4080 m,26days.4) 73m,5days.

1)120m.2) 60km.3)120 km.4)866 km.

89


Satellite&Sensor HY1D(NSOAS/ CAST) FY3D(NRSCC/ CMA) LaunchDate 2010 2010 Source China China Spectrum Vis.,4bands. Resolution, ReturnInterval 250m. ImageWidth 500km 2800km Notes Coastalregionimager.Constellation. AsanadditiontotheFY3constellation, itwillhosttheMERSIandVIIR instrumentsforcoarseresolutionearth observation. Geostationarysatellite,secondina series. Thiscoarseresolutionsatelliteismainly designedforcloudandicemeasurement. Thishighresolutionsatellitewillhave regionalcoverage. Commercialhighresolutionsatellite. Thissatelliteisunderdevelopment. Thehighestresolutionsatellitetodate.It isstillunderdevelopment. FirstinaSARconstellation.Variable resolution,stillunderdevelopment.With COSMOSkyMed,thisjointconstellation willprovidetwicedailycoverageintwo SARbands. Highresolutionpassivesatellite,under development. Highresolutionpanchromaticimager.

VSWIR,20bands;VTIR, 250m1km& 10bands. 1.1km,TBD.

ElektroLN2 (Roshydromet/ Roscosmos) MeteorMN3 (Roshydromet/ Roscosmos) SSOT Kompsat3 Kompsat5 GeoEye2 SAOCOM1A(CONAE /ASI)

2010

Russia

VTIR,10bands.

1 kmvisible4 kmIR,daily. 1)4001000m. 2)7001400m. 5.8m(1.45 pan),TBD. 2.8m(0.7m pan). TBD. 1.65m,(0.25 pan). 10100m,TBD.

Fullearthdisc imaged. 1)450600 km.2)2600 km. TBD. TBD. TBD. TBD. 40320km

2010

Russia

1)SAR,Xband.2)Vis.,1 band. VNIR,5bands. VNIR,5bands. SAR,XBand. VNIR,TBD. SAR,Lband.

2010 2010 2011 2011 2011

Chile SouthKorea SouthKorea Comm.(USA) Argentina/Italy

ASNARO CARTOSAT3(ISRO)

2011 2011

Japan India

Passivehighres. Vis.,1band(pan).

1m,TBD. 0.3m,TBD.

TBD. 6km

90


Satellite&Sensor MetOpB (EUMETSAT) LaunchDate 2011 EU Source Spectrum VTIR,6bands. Resolution, ImageWidth ReturnInterval 1.1km,twice 3000km daily. Notes ThisisthesecondintheMetOp constellation,carryinganAVHRR/3 instrument.Itwillcovertheentireglobe twicedaily. Thesecondinaconstellation.Pointable satellitegives+/50degreesofftrack. Thisproposedsatelliteiscurrentlyan airbornesystemwithhighresolutionand variablemodes. Thiswillcarryamoreadvancedsensor thanpreviousResourcesats,withan atmosphericcorrectioninstrument. Thishyperspectralimagerwillhavea narrowimageswathbuthighspectral andspatialresolution. LandsatDataContinuityMission,free imagery,globalcoverage. VariableresolutionandswathwidthSAR, firstinaconstellation.

Pleiades2(CNES) DMSAR(ISRO)

2011 2011

France India

VNIR,5bands. SAR,C/Xband.

0.7m,TBD. TBD.

20km TBD.

RESOURCESAT3 (ISRO) PRISMA(ASI)

2011

India

VSWIR,4bands.

23m,26days.

700km

2011

Italy

VSWIR, 200bands.

30m(2.5pan), TBD.

30km

LDCM Sentinel1A(ESA/ EC)

2011 2011

USA EU

VSWIR,9bands. SAR,Cband,quadpolar.

30m(15pan), 16days. 5100m,TBD.

185km 8400km

AMAZNIA1(INPE)

2011

Brazil

VNIR,4bands.

40m&12m,5 days&TBD.

800km&110 km

Thissatellitewillhostthemodres.AWFI sensorandthehigherres.RALCam3 sensor.

91


Satellite&Sensor SAOCOM1B(CONAE /ASI) Astroterra(Spot6) SACE/SABIA/mar (CONAE/INPE) Ingenio(SEOSAT) (CDTI/ESA) EnMAP(DLR) HY3A(NSOAS/ CAST) SABRINA(ASI) Sentinel2A(ESA/ EC) LaunchDate 2012 Source Argentina/Italy Spectrum SAR,Lband. Resolution, ReturnInterval 10100m,TBD. ImageWidth 40320km Notes SecondinaSARconstellation.Variable resolution,stillunderdevelopment. ThenextinSPOTserieshasstereo capabilityandhighlypointable. Underdevelopment.

2012 2012

Comm.(France) Argentina/Brazil

VNIR,5bands. VTIR,15bands.

8m(2pan),15 days. TBD.

60 TBD

2012

2012,Spain/EU

VNIR,5bands.

10m(2.5pan), TBD. 30m 110m,TBD. TBD. 10m(VNIR) and2060m (higher),TBD. 520m(extra wide25x100 m),TBD. 300m,TBD.

60km

Highresolutionsatellite,maynot developGI. Moderateresolutionhyperspectral imager. HighresolutionSARimager. Underdevelopment. Firstinaconstellation,thishighquality moderateresolutionsensorwillhave dedicatedatmosphericcorrectionbands andawideswath. Secondinaconstellation,thisSARsensor hasaninterferometricmode,awide fieldmode,andanextrawidemode. Thisdualswathinstrumentwillbea MODIS,AATSR,andMERISfollowup. Thefirstinaseries. ALOSfollowon,underdevelopment, likelytobehigherresolution.

2012 2012 2012 2012

Germany 2012 Italy EU

VNIR,TBD. SAR,Xband,TBD. SAR,Xband,Quad polar. VSWIR,13bands.

30km 40150km TBD. 290km

Sentinel1B(ESA/ EC) Sentinel3A(ESA/ EC) ALOS2

2012

EU

SAR,Cband,Quad polar. VSWIR,21bands.

80400km

2012

EU

7501675km

2012

Japan

SAR,Lband,Quadpolar.

TBD.

TBD.

92


Satellite&Sensor FY3E(NRSCC/CMA) LaunchDate 2012 Source China Spectrum VSWIR,20bands,&V TIR,10bands. Resolution, ImageWidth ReturnInterval 250m1km& 2800km 1.1km,TBD. Notes AsanadditiontotheFY3constellation, itwillhosttheMERSIandVIIR instrumentsforcoarseresolutionearth observation. Thefirstinanewgeostationaryseries, thissensorwillhavehigherVNIR resolutionthanpreviousgeostationary imagers,andthusbemoresuitablefor realtimevegetationmonitoring. ThelastofathreesatelliteSAR constellationforderivingahighquality DEM,thissatellitemaybejoinedor replacedbyanothersatellite,TandemL. ThesecondinanewNASAseriesof coarseresolutionsatellitesintendedto followuptheMODISandAVHRR missions. Thissatellite,thesecondoftwo,hosts foursensorswithregionalcoverage.The highandmoderateresolutionsensors arepointable(35dayreturntime).It alsohasamoderateresolutionfire imagerandacoarseresolutionsensor. ThishighresolutionLbandSARsatellite maypotentiallyacquirecontinuousGI. FirstinaSARconstellation,thissatellite hasawiderangeofresolutionmodes. Highresolution,pointablesensors;still underdevelopment.Itmaybejoinedby ahyperspectralsensorandathermal sensor.

FY4O/A(NRSCC/ CMA)

2012

China

VTIR,12bands.

1kmVisible,2 Fullearthdisc kmIR,4kmTIR, imaged. daily.

TerraSARX2(DLR)

2013

Comm. (Germany)

SAR,Xband.Quad polar.

118m,2.511 days.

10100kmx 5150km

NPOESS1(NOAA)

2013

USA

VTIR,22bands.

400m1.6km, TBD.

3000km

CBERS4(CRESDA/ INPE)

2013

Brazil/China

1)NIR,4bands.2)NIR, 4bands.3)VTIR,4 bands.4)VSWIR,4 bands.

1)20m.2)10 m(5pan),326 days.3)4080 m,26days.4) 73m,5days. 320m,weekly revisit. 3100m,TBD.

1)120m.2) 60km.3)120 km.4)866 km.

MAPSAR(INPE/DLR)

2013

Brazil/Germany

SAR,Lband,Quadpolar.

3055km

RADARSAT CONSTELLATION1 (CSA) SACF(CONAE)

2013

Canada

SAR,Cband,Quad polar. VSWIR,5bands.

20500km

2014

Argentina

10m(5pan).

60117km

93


Satellite&Sensor SAOCOM2A (CONAE) COMS2 GCOMC1(JAXA) Sentinel2B(ESA/ EC) LaunchDate 2014 2014 2014 2014 Source Argentina Japan Japan/USA EU Spectrum SAR,Lband. VNIR,8bands. VTIR,35bands. VSWIR,13bands. Resolution, ImageWidth ReturnInterval 10100m,TBD. 40320km 236x360m. 2501km,TBD. 10(VNIR)and 2060m (higher),TBD. TBD. 3100m,TBD. 1440km 11501400 km 290km Notes ThirdinaSARconstellation.Variable resolution,stillunderdevelopment. GeostationaryOceanColorImager. ThisMODISlikesensorwillhave11 bandsat250mresolutionintheVNIR. Secondinaconstellation,thishigh qualitymoderateresolutionsensorwill havededicatedatmosphericcorrection bandsandawideswath. Underdevelopment. SecondinaSARconstellation,this satellitehasawiderangeofresolution modes. ThelatestinthegeostationaryGOES constellation,thisimprovedABIsensor hasa0.51kmresolutionforrealtime vegetationmonitoring. AsanadditiontotheFY3constellation, itwillhosttheMERSIandVIIR instrumentsforcoarseresolutionearth observation. Geostationarysatellite;thirdinaseries.

RISATL(ISRO) RADARSAT CONSTELLATION3 (CSA) GOESS(NOAA)

2014 2014

India Canada

SAR,Lband. SAR,CbandQuadpolar.

TBD. 20500km

2014

USA

VTIR,26bands.

FY3F(NRSCC/CMA)

2014

China

VSWIR,20bands,&V TIR,10bands.

0.5kminthe visible,1km NIRSWIR,2km TIR. 250m1km& 1.1km,TBD.

Fullearthdisc imaged.

2800km

ElektroLN3 (Roshydromet/ Roscosmos) SAOCOM2B (CONAE) ICESatII(NASA)

2014

Russia

VTIR,10bands.

1kmvisible4 kmIR,daily. 10100m,TBD. 70mfootprint, 183days.

Fullearthdisc imaged. 40320km

2015 2015

Argentina USA

SAR,Lband. LIDAR

FourthinaSARconstellation.Variable resolution,stillunderdevelopment. Footprint Amissionfollowon.Tracksarecurrently centers170m spacedat15km(equator)and2.5km apart. (80degreeslat.).

94


Satellite&Sensor LaunchDate Source Spectrum Resolution, ReturnInterval 3100m,TBD. ImageWidth Notes

RADARSAT CONSTELLATION2 (CSA) GOESR(NOAA)

2015

Canada

SAR,Cband,Quad polar.

20500km

SecondinaSARconstellation,this satellitehasawiderangeofresolution modes. ThelatestinthegeostationaryGOES constellation,thisimprovedABIsensor hasa0.51kmresolutionforrealtime vegetationmonitoring. Thisdualswathinstrumentwillbea MODIS,AATSR,andMERISfollowup. Thesecondinaseries. ThissatellitewillcarrytheAVHRR/3 sensor. Thesecondinanewgeostationary series,thissensorwillhavehigherVNIR resolutionthanpreviousgeostationary imagers,andthusbemoresuitablefor realtimevegetationmonitoring. Thethirdinanewgeostationaryseries, thissensorwillhavehigherVNIR resolutionthanpreviousgeostationary imagers,andthusbemoresuitablefor realtimevegetationmonitoring. Ahighquality,moderateresolution hyperspectralsensor. Stillunderdevelopment.Thiscombined InSAR(2SARsensors)andLIDARsensor willprovideanunparalleledglobalDEM andaccuratemeasurementofecosystem 3Dstructure.

2015

USA

VTIR,26bands.

Sentinel3B(ESA/ EC)

2015

EU

VSWIR,21bands.

0.5kminthe visible,1km NIRSWIR,2km TIR. 300m,TBD.

Fullearthdisc imaged.

7501675km

MetOpC (EUMETSAT) FY4O/B(NRSCC/ CMA)

2015 2015

EU China

VTIR,46bands. VTIR,12bands.

1.1km,Bi daily.

3000km

1kmVisible,2 Fullearthdisc kmIR,4kmTIR, imaged. daily.

FY4O/C(NRSCC/ CMA)

2015

China

VTIR,12bands.

1kmVisible,2 Fullearthdisc kmIR,4kmTIR, imaged. daily.

HypsIRI(NASA)

2015

USA

VTIR,210bands&5 bands(TIR). LIDAR&InSAR,LBand, Quadpolar.

DESDynI(NASA)

2015

USA

19m(VSWIR), 90m(TIR),930 days. 25mLIDAR footprint,SAR resolution variable,8days.

90km(SWIR) &400km (TIR) >340km

95


Satellite&Sensor LaunchDate Source Spectrum Resolution, ReturnInterval 400m1.6km, TBD. 250m1km& 1.1km,TBD. ImageWidth Notes

NPOESS2(NOAA)

2016

USA

VTIR,22bands.

3000km

Thethirdinaseries,thisMODIS/AVHRR followonwillprovideimportant continuityfortheAVHRRtimeseries. AsanadditiontotheFY3constellation, itwillhosttheMERSIandVIIR instrumentsforcoarseresolutionearth observation. Stillinconceptcompetitionandunder development,thismissionwouldbe focusedonmappingforestbiomass.It wouldhaveinSARcapabilityandhavean experimental,50daytomographic mode.

FY3G(NRSCC/ CMA)

2016

China

VSWIR,20bands,&V TIR,10bands.

2800km

BIOMASS(ESA)

2016?

EU

InSAR,Pband,Quad polar.

50m,27days (4daysexptal).

102km

96

Bibliography
Achard,F.,R.DeFries,H.Eva,M.Hansen,P.Mayaux,andH.J.Stibig.2007.Pantropical monitoringofdeforestation.EnvironmentalResearchLetters2(4). Alencar,A.,D.Nepstad,andM.D.V.Diaz.2006.ForestunderstoryfireintheBrazilian AmazoninENSOandnonENSOyears:Areaburnedandcommittedcarbon emissions.EarthInteractions10. Allan,J.D.2004.Landscapesandriverscapes:Theinfluenceoflanduseonstream ecosystems.AnnualReviewofEcologyEvolutionandSystematics35:257284. Almeida,R.,A.Rosenqvist,Y.E.Shimabukuro,andR.SilvaGomez.2007.Detecting deforestationwithmultitemporalLbandSARimagery:Acasestudyinwestern BrazilianAmazonia.InternationalJournalofRemoteSensing28(6):13831390. Andersen,H.E.,S.E.Reutebuch,andR.J.McGaughey.2006.Arigorousassessmentoftree heightmeasurementsobtainedusingairbornelidarandconventionalfieldmethods. CanadianJournalofRemoteSensing32(5):355366. Anderson,J.E.,L.C.Plourde,M.E.Martin,B.H.Braswell,M.L.Smith,R.O.Dubayah,M.A. Hofton,andJ.B.Blair.2008.Integratingwaveformlidarwithhyperspectralimagery forinventoryofanortherntemperateforest.RemoteSensingoftheEnvironment 112(4):18561870. Andersson,K.,T.P.Evans,andK.R.Richards.2009.Nationalforestcarboninventories: Policyneedsandassessmentcapacity.ClimaticChange93(12):69101. Askne,J.,M.Santoro,G.Smith,andJ.E.S.Fransson.2003.MultitemporalrepeatpassSAR interferometryofborealforests.IEEETransactionsonGeoscienceandRemote Sensing41(7):15401550. Askne,J.I.H.,andM.Santoro.2009.Automaticmodelbasedestimationofborealforeststem volumefromrepeatpassCbandInSARcoherence.IEEETransactionsonGeoscience andRemoteSensing47(2):513516. Asner,G.P.2001.CloudcoverinLandsatobservationsoftheBrazilianAmazon. InternationalJournalofRemoteSensing22(18):3,8553,862. Asner,G.P.,E.N.Broadbent,P.J.C.Oliveira,M.Keller,D.E.Knapp,andJ.N.M.Silva.2006. ConditionandfateofloggedforestsintheBrazilianAmazon.Proceedingsofthe NationalAcademyofSciencesoftheUnitedStatesofAmerica103(34):12,947 12,950. Asner,G.P.,R.F.Hughes,T.A.Varga,D.E.Knapp,andT.KennedyBowdoin.2009. Environmentalandbioticcontrolsoverabovegroundbiomassthroughoutatropical rainforest.Ecosystems12(2):261278.

FAGAN AND DEFRIES

97

Asner,G.P.,M.O.Jones,R.E.Martin,D.E.Knapp,andR.F.Hughes.2008.Remotesensingof nativeandinvasivespeciesinHawaiianforests.RemoteSensingoftheEnvironment 112(5):1,9121,926. Asner,G.P.,D.E.Knapp,E.N.Broadbent,P.J.C.Oliveira,M.Keller,andJ.N.Silva.2005. SelectiveloggingintheBrazilianAmazon.Science310(5747):480482. Asner,G.P.,andR.E.Martin.2009.Airbornespectranomics:Mappingcanopychemicaland taxonomicdiversityintropicalforests.FrontiersinEcologyandtheEnvironment 7(5):269276. Asner,G.P.,D.Nepstad,G.Cardinot,andD.Ray.2004.Droughtstressandcarbonuptakein anAmazonforestmeasuredwithspaceborneimagingspectroscopy.Proceedingsof theNationalAcademyofSciencesoftheUnitedStatesofAmerica101(16):6,039 6,044. Asner,G.P.,T.K.Rudel,T.M.Aide,R.DeFries,andR.Emerson.2009.Acontemporary assessmentofglobalhumidtropicalforestchange.ConservationBiology, forthcoming. Baban,S.M.J.,andK.W.Yusof.2001.Mappinglanduse/coverdistributiononamountainous tropicalislandusingremotesensingandGIS.InternationalJournalofRemoteSensing 22(10):1,9091,918. Baccini,A.,M.A.Friedl,C.E.Woodcock,andR.Warbington.2004.Forestbiomassestimation overregionalscalesusingmultisourcedata.GeophysicalResearchLetters31(10). Baccini,A.,N.Laporte,S.J.Goetz,M.Sun,andH.Dong,2008.AfirstmapoftropicalAfricas abovegroundbiomassderivedfromsatelliteimagery.EnvironmentalResearch Letters3:19. Baker,T.R.,O.L.Phillips,Y.Malhi,S.Almeida,L.Arroyo,A.DiFiore,T.Erwin,etal.2004. VariationinwooddensitydeterminesspatialpatternsinAmazonianforestbiomass. GlobalChangeBiology10(5):545562. Bala,G.,K.Caldeira,M.Wickett,T.Phillips,D.Lobell,C.Delire,andA.Mirin.2007.Combined climateandcarboncycleeffectsoflargescaledeforestation.Proceedingsofthe NationalAcademyofSciencesoftheUnitedStatesofAmerica104(16):6,550. Baldocchi,D.D.2003.Assessingtheeddycovariancetechniqueforevaluatingcarbon dioxideexchangeratesofecosystems:past,presentandfuture.GlobalChange Biology9(4):479492. Balzter,H.2001.Forestmappingandmonitoringwithinterferometricsyntheticaperture radar(InSAR).ProgressinPhysicalGeography25(2)159177. Balzter,H.,C.S.Rowland,andP.Saich.2007.Forestcanopyheightandcarbonestimationat MonksWoodNationalNatureReserve,UK,usingdualwavelengthSAR interferometry.RemoteSensingofEnvironment108(3):224239. Bicheron,P.,P.Defourny,C.Brockmann,L.Schouten,C.Vancutsem,M.Huc,S.Bontemps,M. Leroy,F.Achard,M.Herold,F.Ranera,andO.Arino.2008.GLOBCOVERProducts

FAGAN AND DEFRIES

98

DescriptionandValidationReport.MEDIASFrance, ftp://uranus.esrin.esa.int/pub/globcover_v2/global/. Blackard,J.A.,M.V.Finco,E.H.Helmer,G.R.Holden,M.L.Hoppus,D.M.Jacobs,A.J.Lister,et al.2008.MappingUSforestbiomassusingnationwideforestinventorydataand moderateresolutioninformationRemoteSensingofEnvironment112(4):1,658 1,677. Boudreau,J.,R.F.Nelson,H.A.Margolis,A.Beaudoin,L.Guindon,andD.S.Kimes.2008. RegionalabovegroundforestbiomassusingairborneandspaceborneLiDARin Quebec.RemoteSensingofEnvironment112(10):3,8763,890. Bourgine,B.,andN.Baghdadi.2005.AssessmentofCbandSRTMDEMinadense equatorialforestzone.ComptesRendusGeoscience337(14):1,2251,234. Broadbent,E.N.,G.P.Asner,M.PenaClaros,M.Palace,andM.Soriano.2008.Spatial partitioningofbiomassanddiversityinalowlandBolivianforest:Linkingfieldand remotesensingmeasurements.ForestEcologyandManagement255(7):2,602 2,616. Brown,S.,T.Pearson,D.Slaymaker,S.Ambagis,N.Moore,D.Novelo,andW.Sabido.2005. Creatingavirtualtropicalforestfromthreedimensionalaerialimagerytoestimate carbonstocks.EcologicalApplications15(3):10831095. Cairns,M.A.,S.Brown,E.H.Helmer,andG.A.Baumgardner.1997.Rootbiomassallocation intheworld'suplandforests.Oecologia111(1):111. Canadell,J.G.,C.LeQuere,M.R.Raupach,C.B.Field,E.T.Buitenhuis,P.Ciais,T.J.Conway, N.P.Gillett,R.A.HoughtonandG.Marland,2007.Contributionstoaccelerating atmosphericCO2growthfromeconomicactivity,carbonintensity,andefficiencyof naturalsinks,ProceedingsoftheNationalAcademyofSciencesoftheUnitedStatesof America,104(47),1886618870. Castro,K.L.,G.A.SanchezAzofeifa,andB.Rivard.2003.Monitoringsecondarytropical forestsusingspacebornedata:ImplicationsforCentralAmerica.International JournalofRemoteSensing24(9):1,8531,894. CEOS(CommitteeonEarthObservationSatellites).2009.CommitteeonEarthObservation Satellites.http://www.ceos.org/(accessedMarch28,2009). Chambers,J.Q.,G.P.Asner,D.C.Morton,L.O.Anderson,S.S.Saatch,F.D.B.EspiritoSanto,M. Palace,andC.Souza.2007.Regionalecosystemstructureandfunction:Ecological insightsfromremotesensingoftropicalforests.TrendsinEcology&Evolution22(8), 414423. Chand,T.R.K.,andK.V.S.Badarinath.2007.AnalysisofENVISATASARdataforforest parameterretrievalandforesttypeclassificationacasestudyoverdeciduous forestsofcentralIndia.InternationalJournalofRemoteSensing28(22):4,9854,999. Chave,J.,C.Andalo,S.Brown,M.A.Cairns,J.Q.Chambers,D.Eamus,H.Folster,etal.2005. Treeallometryandimprovedestimationofcarbonstocksandbalanceintropical forests.Oecologia145(1):8799.

FAGAN AND DEFRIES

99

Chazdon,R.L.,C.A.Harvey,O.Komar,D.M.Griffith,B.G.Ferguson,M.MartinezRamos,H. Morales,etal.2009.BeyondReserves:AResearchAgendaforConserving BiodiversityinHumanmodifiedTropicalLandscapes,Biotropica,41(2),142153. Chopping,M.,G.G.Moisen,L.H.Su,A.Laliberte,A.Rango,J.V.Martonchik,andD.P.C.Peters. 2008.Largeareamappingofsouthwesternforestcrowncover,canopyheight,and biomassusingtheNASAMultiangleImagingSpectroRadiometer,RemoteSensingof Environment112(5):2,0512,063. Chuvieco,E.2008.SatelliteObservationofBiomassBurning:ImplicationsinGlobalChange Research.Dordrecht:SpringerNetherlands. Clark,D.B.,J.M.Read,M.L.Clark,A.M.Cruz,M.F.Dotti,andD.A.Clark.2004.Applicationof1 Mand4Mresolutionsatellitedatatoecologicalstudiesoftropicalrainforests. EcologicalApplications14(1):6174. Clark,M.L.,D.B.Clark,andD.A.Roberts.2004.Smallfootprintlidarestimationofsub canopyelevationandtreeheightinatropicalrainforestlandscape.RemoteSensing ofEnvironment91(1):6889. Clark,M.L.,D.A.Roberts,andD.B.Clark.2005.Hyperspectraldiscriminationoftropicalrain foresttreespeciesatleaftocrownscales.RemoteSensingofEnvironment96(34): 375398. Cohen,W.B.,andS.N.Goward.2004.Landsat'sroleinecologicalapplicationsofremote sensing.Bioscience54(6):535545. Coppin,P.,I.Jonckheere,K.Nackaerts,B.Muys,andE.Lambin.2004.Digitalchange detectionmethodsinecosystemmonitoring:Areview.InternationalJournalof RemoteSensing25(9):1,5651,596. Costa,M.P.F.2004.UseofSARsatellitesformappingzonationofvegetationcommunitiesin theAmazonfloodplain.InternationalJournalofRemoteSensing25(10):1,8171,835. Crutzen,P.J.,andM.O.Andreae.1990.Biomassburninginthetropicsimpacton atmosphericchemistryandbiogeochemicalcycles.Science250(4988):1,6691,678. Cunningham,D.,J.Melican,E.Wemmelmann,andT.Jones.2002.GeoCoverLCAmoderate resolutiongloballandcoverdatabase.PaperpresentedattheESRIInternational UserConference.July2002,SanDiego,CA.. Dalponte,M.,L.Bruzzone,andD.Gianelle.2008.FusionofhyperspectralandLIDARremote sensingdataforclassificationofcomplexforestareas.IEEETransactionson GeoscienceandRemoteSensing46(5):1,4161,427. Davies,D.K.,S.Ilavajhala,M.M.Wong,andC.O.Justice.2009.Fireinformationforresource managementsystem:ArchivinganddistributingMODISactivefiredata.IEEE TransactionsonGeoscienceandRemoteSensing47(1):7279. DeGrandi,G.,J.P.Malingreau,andM.Leysen.1999.TheERS1CentralAfricamosaic:Anew perspectiveinradarremotesensingfortheglobalmonitoringofvegetation.IEEE TransactionsonGeoscienceandRemoteSensing37(3):1,7301,746.

FAGAN AND DEFRIES

100

Dean,T.J.,Q.V.Cao,S.D.Roberts,andD.L.Evans.2009.Measuringheightstocrownbaseand crownmedianwithLiDARinamature,evenagedloblollypinestand.ForestEcology andManagement257(1):126133. DeFries,R.2008.Terrestrialvegetationinthecoupledhumanearthsystem:contributions ofremotesensing.AnnualReviewofEnvironmentandResources33:369390. DeFries,R.,F.Achard,S.Brown,M.Herold,D.Murdiyarso,B.Schlamadinger,andC.de Souza.2007.Earthobservationsforestimatinggreenhousegasemissionsfrom deforestationindevelopingcountries.EnvironmentalScienceandPolicy10(4):385 394. DeFries,R.S.,R.A.Houghton,M.C.Hansen,C.B.Field,D.Skole,andJ.Townshend.2002. Carbonemissionsfromtropicaldeforestationandregrowthbasedonsatellite observationsforthe1980sand1990s.ProceedingsoftheNationalAcademyof SciencesoftheUnitedStatesofAmerica99(22):14,25614,261. DeFries,R.S.,andJ.R.Townshend.1994.NDVIderivedlandcoverclassificationatglobal scales.InternationalJournalofRemoteSensing15(17):3,5673,586. Diner,D.J.,B.H.Braswell,R.Davies,N.Gobron,J.N.Hu,Y.F.Jin,R.A.Kahn,etal.2005.The valueofmultianglemeasurementsforretrievingstructurallyandradiatively consistentpropertiesofclouds,aerosols,andsurfaces.RemoteSensingof Environment97(4):495518. Dixon,R.K.,S.Brown,R.A.Houghton,A.M.Solomon,M.C.Trexler,andJ.Wisniewski.1994. Carbonpoolsandfluxofglobalforestecosystems.Science.263:185190. Dobson,M.C.,F.T.Ulaby,andL.E.Pierce.1995a.Landcoverclassificationandestimationof terrainattributesusingsyntheticapertureradar.RemoteSensingofEnvironment 51(1):199214. Dobson,M.C.,F.T.Ulaby,L.E.Pierce,T.L.Sharik,K.M.Bergen,J.Kellndorfer,J.R.Kendra,E. Li,Y.C.Lin,A.Nashashibi,K.Sarabandi,andP.Siqueira.1995b.Estimationofforest biophysicalcharacteristicsinnorthernMichiganwithSIRC/XSAR.IEEE TransactionsonGeoscienceandRemoteSensing33(4):877895. Donnellan,A.,P.Rosen,J.Graf,A.Loverro,A.Freeman,R.Treuhaft,R.Oberto,M.Simard,E. Rignot,andR.Kwok.2008.Deformation,ecosystemetructure,anddynamicsofice (DESDynI).PaperpresentedattheESRIInternationalUserConference.April2008, Washington,DC., Drake,J.B.,R.O.Dubayah,D.B.Clark,R.G.Knox,J.B.Blair,M.A.Hofton,R.L.Chazdon,J.F. Weishampel,andS.D.Prince.2002a.Estimationoftropicalforeststructural characteristicsusinglargefootprintlidar.RemoteSensingofEnvironment79(23): 305319. Drake,J.B.,R.O.Dubayah,R.G.Knox,D.B.Clark,andJ.B.Blair.2002b.Sensitivityoflarge footprintlidartocanopystructureandbiomassinaneotropicalrainforest.Remote SensingofEnvironment81(23):378392.

FAGAN AND DEFRIES

101

Drake,J.B.,R.G.Knox,R.O.Dubayah,D.B.Clark,R.Condit,J.B.Blair,andM.Hofton.2003. Abovegroundbiomassestimationinclosedcanopyneotropicalforestsusinglidar remotesensing:factorsaffectingthegeneralityofrelationships.GlobalEcologyand Biogeography12(2):147159. Dubayah,R.,R.Knox,M.Hofton,J.Blair,andJ.Drake(eds.).2000.LandSurface CharacterizationusingLIDARRemoteSensing.London:CRCPress. Engdahl,M.E.,andJ.M.Hyyppa.2003.LandcoverclassificationusingmultitemporalERS 1/2InSARdata.IEEETransactionsonGeoscienceandRemoteSensing41(7):1620 1628. eoPortal.2009.ListofEOandnonEOSatelliteMissions.EuropeanSpaceAgency.July23, 2009.http://directory.eoportal.org/missions_all_list.php?filter=&view_all&order=st art_date&dir=ASC(accessedJuly23,2009). ERSDAC(EarthRemoteSensingDataAnalysisCenter).2009.ASTERGlobalDigital ElevationModel.http://www.ersdac.or.jp/GDEM/E/4.html(accessedJune22, 2009). ESA.2009.SummariesandRecommendationsofthePOLinSAR2009Workshop.January 2009,Frascati,ItalyESRIN. Fan,S.,M.Gloor,J.Mahlman,S.Pacala,J.Sarmiento,T.Takahashi,andP.Tans.1998.Alarge terrestrialcarbonsinkinNorthAmericaimpliedbyatmosphericandoceaniccarbon dioxidedataandmodels.Science282(5388):442446. FAO(FoodandAgricultureOrganization).2006.GlobalForestResourcesAssessment2005. FAOForestryPaper147.Rome:Ed.byFoodandAgricultureOrganizationofthe UnitedNations. Field,C.B.,D.B.Lobell,H.A.Peters,andN.R.Chiariello.2007.Feedbacksofterrestrial ecosystemstoclimatechange.AnnualReviewofEnvironmentandResources32:1 29. Fiorella,M.,andW.J.Ripple.1993.Determiningsuccessionalstageoftemperateconiferous forestswithLANDSATsatellitedata.PhotogrammetricEngineeringandRemote Sensing59(2):239246. Foody,G.M.2002.Statusoflandcoverclassificationaccuracyassessment.RemoteSensing ofEnvironment80(1):185201. Foody,G.M.,D.S.Boyd,andM.E.J.Cutler.2003.Predictiverelationsoftropicalforest biomassfromLandsatTMdataandtheirtransferabilitybetweenregions.Remote SensingofEnvironment85(4):463474. Foster,J.,C.Kingdon,andP.Townsend.2002.PredictingtropicalforestcarbonfromEO1 hyperspectralimageryinNoelKempffMercadoNationalPark,Bolivia.Paper presentedatIEEEInternationalGeoscienceandRemoteSensingSymposium.July 2008,Boston,MA.

FAGAN AND DEFRIES

102

FrancoLopez,H.,A.Ek,andM.Bauer.2001.Estimationandmappingofforeststand density,volume,andcovertypeusingtheknearestneighborsmethod.Remote SensingofEnvironment77(3):251274. Fransson,J.E.S.,F.Walter,andL.M.H.Ulander.2000.Estimationofforestparametersusing CARABASIIVHFSARdata.IEEETransactionsonGeoscienceandRemoteSensing 38(2):720727. Galvao,L.S.,F.J.Ponzoni,V.Liesenberg,andJ.R.d.Santos.2009.Possibilitiesof discriminatingtropicalsecondarysuccessioninAmazoniausinghyperspectraland multiangularCHRIS/PROBAdata.InternationalJournalofAppliedEarthObservation andGeoinformation11:814. Gao,X.,A.R.Huete,W.G.Ni,andT.Miura.2000.Opticalbiophysicalrelationshipsof vegetationspectrawithoutbackgroundcontamination.RemoteSensingof Environment74(3):609620. Gardner,T.A.,J.Barlow,L.W.Parry,andC.A.Peres.2007.Predictingtheuncertainfutureof tropicalforestspeciesinadatavacuum.Biotropica39(1):2530. Gergel,S.E.,Y.Stange,N.C.Coops,K.Johansen,andK.R.Kirby.2007.Whatisthevalueofa goodmap?Anexampleusinghighspatialresolutionimagerytoaidriparian restoration.Ecosystems10(5):688702. Gibbs,H.K.,S.Brown,J.O.Niles,andJ.A.Foley.2007.Monitoringandestimatingtropical forestcarbonstocks:MakingREDDareality.EnvironmentalResearchLetters2(4). Giglio,L.,I.Csiszar,andC.O.Justice.2006.Globaldistributionandseasonalityofactivefires asobservedwiththeTerraandAquaModerateResolutionImaging Spectroradiometer(MODIS)sensors.JournalofGeophysicalResearchBiogeosciences 111(G2). Giglio,L.,G.R.v.d.Werf,J.T.Randerson,G.J.Collatz,andP.Kasibhatla.2005.Global estimationofburnedareausingMODISactivefireobservations.Atmos.Chem.Phys. Discuss.5:11,09111,141. Giri,C.,Z.L.Zhu,andB.Reed.2005.AcomparativeanalysisoftheGlobalLandCover2000 andMODISlandcoverdatasets.RemoteSensingofEnvironment94(1):123132. Goetz,S.J.,A.Baccini,N.T.Laporte,T.Johns,W.Walker,J.Kellndorfer,R.A.Houghton,andM. Sun.2009.Mappingandmonitoringcarbonstockswithsatelliteobservations:A comparisonofmethods.CarbonBalanceandManagement4(2). Goetz,S.J.,A.G.Bunn,G.J.Fiske,andR.A.Houghton.2005.Satelliteobservedphotosynthetic trendsacrossborealNorthAmericaassociatedwithclimateandfiredisturbance. ProceedingsoftheNationalAcademyofSciencesoftheUnitedStatesofAmerica 102(38):13,52113,525. Goetz,S.J.,R.K.Wright,A.J.Smith,E.Zinecker,andE.Schaub.2003.IKONOSimageryfor resourcemanagement:Treecover,impervioussurfaces,andriparianbuffer

FAGAN AND DEFRIES

103

analysesinthemidAtlanticregion.RemoteSensingandEnvironment88(12):195 208. GOFCGOLD(GlobalObservationofForestandLandCoverDynamics).2008.Reducing GreenhouseGasEmissionsfromDeforestationandDegradationinDeveloping Countries:ASourcebookofMethodsandProceduresforMonitoring,Measuring,and Reporting.GOFCGOLDReportversionCOP132.Ed.byGOFCGOLDProjectOffice. Alberta,Canada:NaturalResourcesCanada. Goodenough,D.G.,A.Dyk,O.Niemann,J.S.Pearlman,H.Chen,T.Han,M.Murdoch,andC. West.2003.ProcessingHyperionandALIforforestclassification.IEEETransactions onGeoscienceandRemoteSensing41(6):1,3211,331. Goodwin,N.,R.Turner,andR.Merton.2005.ClassifyingEucalyptusforestswithhigh spatialandspectralresolutionimagery:Aninvestigationofindividualspeciesand vegetationcommunities.AustralianJournalofBotany53(4):337345. Grainger,A.2008.Difficultiesintrackingthelongtermglobaltrendintropicalforestarea ProceedingsoftheNationalAcademyofSciencesoftheUnitedStatesofAmerica 105(2):818823. Grassi,G.,S.Monni,S.Federici,F.Achard,andD.Mollicone.2008.Applyingthe conservativenessprincipletoREDDtodealwiththeuncertaintiesoftheestimates. EnvironmentalResearchLetters3(3). Hall,R.J.,R.S.Skakun,E.J.Arsenault,andB.S.Case.2006.Modelingforeststandstructure attributesusingLandsatETM+data:Applicationtomappingofaboveground biomassandstandvolume.ForestEcologyandManagement225(13):378390. Hansen,M.C.,R.S.DeFries,J.R.G.Townshend,M.Carroll,C.Dimiceli,andR.A.Sohlberg. 2003.Globalpercenttreecoverataspatialresolutionof500meters:Firstresultsof theMODISvegetationcontinuousfieldsalgorithm.EarthInteractions Hansen,M.C.,D.P.Roy,E.Lindquist,B.Adusei,C.O.Justice,andA.Altstatt.2008.Amethod forintegratingMODISandLandsatdataforsystematicmonitoringofforestcover andchangeintheCongoBasin.RemoteSensingofEnvironment112(5):2,4952,513. Hansen,M.C.,S.V.Stehman,P.V.Potapov,T.R.Loveland,J.R.G.Townshend,R.S.DeFries,K.W. Pittman,B.Arunarwati,F.Stolle,M.K.Steininger,M.Carroll,andC.DiMiceli.2008. Humidtropicalforestclearingfrom2000to2005quantifiedbyusingmultitemporal andmultiresolutionremotelysenseddata.ProceedingsoftheNationalAcademyof SciencesoftheUnitedStatesofAmerica105(27):9,4399,444. Heinsch,F.A.,M.S.Zhao,S.W.Running,J.S.Kimball,R.R.Nemani,K.J.Davis,P.V.Bolstad,et al.2006.EvaluationofremotesensingbasedterrestrialproductivityfromMODIS usingregionaltowereddyfluxnetworkobservations.IEEETransactionson GeoscienceandRemoteSensing44(7):1,9081,925. Heiskanen,J.2006.TreecoverandheightestimationintheFennoscandiantundrataiga transitionzoneusingmultiangularMISRdata.RemoteSensingofEnvironment 103(1):97114.

FAGAN AND DEFRIES

104

Herold,M.2009.AnAssessmentofNationalForestMonitoringCapabilitiesinTropicalNon AnnexICountries:RecommendationsforCapacityBuilding. http://unfccc.int/files/methods_science/redd/country_specific_information/applic ation/pdf/redd_nat_capacity_report_herold_july09_publ.pdf. Herold,M.,andT.Johns.2007.Linkingrequirementswithcapabilitiesfordeforestation monitoringinthecontextoftheUNFCCCREDDprocess.EnvironmentalResearch Letters2(4). Hoekman,D.H.,andM.J.Quinones.2000.Landcovertypeandbiomassclassificationusing AirSARdataforevaluationofmonitoringscenariosintheColombianAmazon.IEEE TransactionsonGeoscienceandRemoteSensing38(2):685696. Houghton,R.A.1999.Theannualnetfluxofcarbontotheatmospherefromchangesinland use18501990.TellusSeriesBChemicalandPhysicalMeteorology51(2):298313. Houghton,R.A.2005.Abovegroundforestbiomassandtheglobalcarbonbalance.Global ChangeBiology11(6):945958. Houghton,R.A.2007.Balancingtheglobalcarbonbudget.AnnualReviewofEarthand PlanetarySciences35:313347. Houghton,R.A.,D.Butman,A.G.Bunn,O.N.Krankina,P.Schlesinger,andT.A.Stone.2007. MappingRussianforestbiomasswithdatafromsatellitesandforestinventories. EnvironmentalResearchLetters2(4). Houghton,R.A.,K.T.Lawrence,J.L.Hackler,andS.Brown.2001.Thespatialdistributionof forestbiomassintheBrazilianAmazon:Acomparisonofestimates.GlobalChange Biology7(7):731746. House,J.I.,I.C.Prentice,N.Ramankutty,R.A.Houghton,andM.Heimann.2003.Reconciling apparentinconsistenciesinestimatesofterrestrialCO2sourcesandsinks.Tellus SeriesBChemicalandPhysicalMeteorology55(2):345363. Hudak,A.T.,M.A.Lefsky,W.B.Cohen,andM.Berterretche.2002.Integrationoflidarand LandsatETMplusdataforestimatingandmappingforestcanopyheight.Remote SensingofEnvironment82(23):397416. Hyde,P.,R.Dubayah,B.Peterson,J.B.Blair,M.Hofton,C.Hunsaker,R.Knox,andW.Walker. 2005.Mappingforeststructureforwildlifehabitatanalysisusingwaveformlidar: Validationofmontaneecosystems.RemoteSensingofEnvironment96(34):427 437. Hyde,P.,R.Dubayah,W.Walker,J.B.Blair,M.Hofton,andC.Hunsaker.2006.Mapping foreststructureforwildlifehabitatanalysisusingmultisensor(LiDAR,SAR/InSAR, ETMplus,Quickbird)synergy.RemoteSensingofEnvironment102(12):6373. Hyyppa,J.,H.Hyyppa,M.Inkinen,M.Engdahl,S.Linko,andY.H.Zhu.2000.Accuracy comparisonofvariousremotesensingdatasourcesintheretrievalofforeststand attributes.ForestEcologyandManagement128(12):109120.

FAGAN AND DEFRIES

105

Hyyppa,J.,H.Hyyppa,D.Leckie,F.Gougeon,X.Yu,andM.Maltamo.2008.Reviewof methodsofsmallfootprintairbornelaserscanningforextractingforestinventory datainborealforests.InternationalJournalofRemoteSensing29(5):13391366. IIASA(InternationalInstituteforAppliedSystemsAnalysis).2009.GeoWikiProject:Help ImproveGlobalLandCover.InternationalInstituteforAppliedSystemsAnalysis. http://geowiki.org/(accessedJuly23,2009). IPCC(IntergovernmentalPanelonClimateChange).2006.IPCCGuidelinesforNational GreenhouseGasInventories.,ed.byI.P.o.C.Change,Japan,Hayama.http://www.ipcc nggip.iges.or.jp/public/2006gl/index.html. 2007.ClimateChange2007:ThePhysicalBasis.WorkingGroupIContributiontothe FourthAssessmentReportoftheIntergovernmentalPanelonClimateChange.New York,NY:CambridgeUniversityPress.. Jensen,J.R.2007.RemoteSensingoftheEnvironment:AnEarthResourcePerspective,2nd Edition.UpperSaddleRiver,NJ.:PrenticeHall. Kasischke,E.S.,J.M.Melack,andM.C.Dobson.1997.Theuseofimagingradarsforecological applicationsareview.RemoteSensingofEnvironment59(2):141156. Kauppi,P.E.,J.H.Ausubel,J.Y.Fang,A.S.Mather,R.A.Sedjo,andP.E.Waggoner.2006. Returningforestsanalyzedwiththeforestidentity.ProceedingsoftheNational AcademyofSciencesoftheUnitedStates103(46):17,57417,579. Kayitakire,F.,C.Hamel,andP.Defourny,2006.Retrievingforeststructurevariablesbased onimagetextureanalysisandIKONOS2imagery.RemoteSensingofEnvironment 102(34):390401. Keeling,H.C.,andO.L.Phillips.2007.Theglobalrelationshipbetweenforestproductivity andbiomass,GlobalEcologyandBiogeography16(5):618631. Kellndorfer,J.,W.Walker,D.Nepstad,C.Stickler,P.Brando,P.Lefebvre,A.Rosenqvist,and M.Shimada.2008.ImplementingREDD:ThePotentialofALOS/PALSARforForest MappingandMonitoring.PaperpresentedattheSecondGEOSSAsiaPacific Symposium.April2008,Tokyo,Japan. Kellndorfer,J.,W.Walker,L.Pierce,C.Dobson,J.A.Fites,C.Hunsaker,J.Vona,andM. Clutter.2004.Vegetationheightestimationfromshuttleradartopographymission andnationalelevationdatasets.RemoteSensingofEnvironment93(3):339358. Kellndorfer,J.M.,L.E.Pierce,M.C.Dobson,andF.T.Ulaby.1998.Towardconsistentregional toglobalscalevegetationcharacterizationusingorbitalSARsystems.IEEE TransactionsonGeoscienceandRemoteSensing36(5):1,3961,411. Kennedy,R.E.,W.B.Cohen,andT.A.Schroeder.2007.Trajectorybasedchangedetectionfor automatedcharacterizationofforestdisturbancedynamics.RemoteSensingof Environment110(3):370386.

FAGAN AND DEFRIES

106

Kimes,D.S.,R.F.Nelson,W.A.Salas,andD.L.Skole.1999.Mappingsecondarytropicalforest andforestagefromSPOTHRVdata.InternationalJournalofRemoteSensing20(18): 3,6253,640. Kimes,D.S.,K.J.Ranson,G.Sun,andJ.B.Blair.2006.Predictinglidarmeasuredforest verticalstructurefrommultianglespectraldata.RemoteSensingofEnvironment 100(4):503511. Knorn,J.,A.Rabe,V.C.Radeloff,T.Kuemmerle,J.Kozak,andP.Hostert.2009.Landcover mappingoflargeareasusingchainclassificationofneighboringLandsatsatellite images.RemoteSensingofEnvironment113(5):957964. Kozak,J.,C.Estreguil,andK.Ostapowicz.2008.Europeanforestcovermappingwithhigh resolutionsatellitedata:TheCarpathianscasestudy.InternationalJournalofApplied EarthObservationandGeoinformation10(1):4455. Kramer,H.J.,andA.P.Cracknell.2008.Anoverviewofsmallsatellitesinremotesensing. InternationalJournalofRemoteSensing29(15):4,2854,337. Kuplich,T.M.2006.ClassifyingregeneratingforeststagesinAmazoniausingremotely sensedimagesandaneuralnetwork.ForestEcologyandManagement234(13):19. Kuplich,T.M.,C.C.Freitas,andJ.V.Soares.2000.ThestudyofERS1SARandLandsatTM synergismforlanduseclassification.InternationalJournalofRemoteSensing 21(10):2,1012,111. Latifovic,R.,andI.Olthof.2004.Accuracyassessmentusingsubpixelfractionalerror matricesofgloballandcoverproductsderivedfromsatellitedata.RemoteSensingof Environment90(2):153165. Laurance,W.F.2007.Environmentalscience:Forestsandfloods.Nature449(7161):409 410. LeToan,T.,H.Baltzer,P.Paillou,K.P.Papathanassiou,S.Plummer,S.Quegan,F.Rocca,and L.Ulander.2008.BIOMASS:ToObserveGlobalForestBiomassforaBetter UnderstandingoftheCarbonCycle.ReportforAsssessment2.ed.byM.Davidson,A. Thompson,C.C.Lin,andP.Clissold.E.S.Agency, http://esamultimedia.esa.int/docs/SP13132_BIOMASS.pdf. LeToan,T.,S.Quegan,I.Woodward,M.Lomas,N.Delbart,andG.Picard.2004.Relating RadarRemoteSensingofBiomasstoModellingofForestCarbonBudgets.Climatic Change67(23):379402. Lee,K.S.,W.B.Cohen,R.E.Kennedy,T.K.Maiersperger,andS.T.Gower.2004.Hyperspectral versusmultispectraldataforestimatingleafareaindexinfourdifferentbiomes. RemoteSensingofEnvironment91(34):508520. Lefsky,M.A.,W.B.Cohen,D.J.Harding,G.G.Parker,S.A.Acker,andS.T.Gower.2002a.Lidar remotesensingofabovegroundbiomassinthreebiomes.GlobalEcologyand Biogeography11(5):393399.

FAGAN AND DEFRIES

107

Lefsky,M.A.,W.B.Cohen,G.G.Parker,andD.J.Harding.2002b.Lidarremotesensingfor ecosystemstudies.Bioscience52(1):1930. Lefsky,M.A.,M.Keller,Y.Pang,P.B.deCamargo,andM.O.Hunter.2007.Revisedmethodfor forestcanopyheightestimationfromGeoscienceLaserAltimeterSystem waveforms.JournalofAppliedRemoteSensing1. Lindquist,E.J.,M.C.Hansen,D.P.Roy,andC.O.Justice.2008.Thesuitabilityofdecadalimage datasetsformappingtropicalforestcoverchangeintheDemocraticRepublicof Congo:implicationsforthegloballandsurvey.InternationalJournalofRemote Sensing29(24):7,2697,275. Liu,J.,J.M.Chen,J.Cihlar,andW.Chen.2002.NetprimaryproductivitymappedforCanada at1kmresolution.GlobalEcologyandBiogeography11(2):115129. Liu,X.,andM.Kafatos.2007.MISRmultiangularspectralremotesensingfortemperate forestmappingat1.1kmresolution.InternationalJournalofRemoteSensing28(1 2):459464. Loarie,S.R.,L.N.Joppa,andS.L.Pimm.2007.Satellitesmissenvironmentalpriorities. TrendsinEcologyandEvolution22(12):630632. Loehle,C.2000.Forestecotoneresponsetoclimatechange:Sensitivitytotemperature responsefunctionalforms.CanadianJournalofForestResearchRevueCanadienne DeRechercheForestiere30(10):1,6321,645. Loveland,T.R.,M.A.Cochrane,andG.M.Henebry.2008.Landsatstillcontributingto environmentalresearchResponse.TrendsinEcologyandEvolution23(4):182 183. Loveland,T.R.,T.L.Sohl,S.V.Stehman,A.L.Gallant,K.L.Sayler,andD.E.Napton.2002.A strategyforestimatingtheratesofrecentUnitedStateslandcoverchanges. PhotogrammetricEngineeringandRemoteSensing68(10):1,0911,099. Loveland,T.R.,Z.L.Zhu,D.O.Ohlen,J.F.Brown,B.C.Reed,andL.M.Yang.1999.Ananalysis oftheIGBPgloballandcovercharacterizationprocess.Photogrammetric EngineeringandRemoteSensing65(9):1,0211,032. Lu,D.,P.Mausel,E.Brondizio,andE.Moran.2004.Changedetectiontechniques. InternationalJournalofRemoteSensing25(12):2,3652,407. Lu,D.S.2006.Thepotentialandchallengeofremotesensingbasedbiomassestimation. InternationalJournalofRemoteSensing27(7):1,2971,328. Luus,K.A.,andR.E.J.Kelly.2008.AssessingproductivityofvegetationintheAmazonusing remotesensingandmodeling.ProgressinPhysicalGeography32(4):363377. MacDonald,G.M.,K.V.Kremenetski,andD.W.Beilman.2008.Climatechangeandthe northernRussiantreelinezone.PhilosophicalTransactionsoftheRoyalSocietyB BiologicalSciences363(1501):2,2852,299.

FAGAN AND DEFRIES

108

Malhi,Y.,L.Arago,D.Galbraith,C.Huntingford,R.Fisher,P.Zelazowski,S.Sitch,C. McSweeney,andP.Meir.2009.Exploringthelikelihoodandmechanismofa climatechangeinduceddiebackoftheAmazonrainforest.Proceedingsofthe NationalAcademyofSciencesEarlyEditionPublishedonlinebeforeprintFebruary 13,2009,[doi:10.1073/pnas.0804619106] http://www.pnas.org/content/early/2009/02/12/0804619106.full.pdf+html Malhi,Y.,D.D.Baldocchi,andP.G.Jarvis.1999.Thecarbonbalanceoftropical,temperate andborealforests.PlantCellandEnvironment22(6):715740. Mallinis,G.,N.Koutsias,M.TsakiriStrati,andM.Karteris.2008.Objectbasedclassification usingQuickbirdimageryfordelineatingforestvegetationpolygonsina Mediterraneantestsite.ISPRSJournalofPhotogrammetryandRemoteSensing63(2): 237250. Maltamo,M.,K.Eerikainen,P.Packalen,andJ.Hyyppa.2006.Estimationofstemvolume usinglaserscanningbasedcanopyheightmetrics.Forestry79(2):217229. Marsh,G.P.1878.TheEarthasModifiedbyHumanAction:ANewEditionofManandNature. ProjectGutenbergRelease#6019.July2004. Masek,J.G.,C.Q.Huang,R.Wolfe,W.Cohen,F.Hall,J.Kutler,andP.Nelson.2008.North AmericanforestdisturbancemappedfromadecadalLandsatrecord.Remote SensingofEnvironment112(6):2,9142,926. Mayaux,P.,G.DeGrandi,andJ.P.Malingreau,2000.CentralAfricanforestcoverrevisited:A multisatelliteanalysis.RemoteSensingofEnvironment71(2):183196. Mayaux,P.,P.Holmgren,F.Achard,H.Eva,H.StibigandA.Branthomme.2005.Tropical forestcoverchangeinthe1990sandoptionsforfuturemonitoring.Philosophical TransactionsoftheRoyalSocietyBBiologicalSciences360(1454):373384. Mayaux,P.,H.Eva,J.Gallego,A.H.Strahler,M.Herold,S.Agrawal,S.Naumov,E.E.De Miranda,C.M.DiBella,C.Ordoyne,Y.Kopin,andP.S.Roy.2006.Validationofthe globallandcover2000map.IEEETransactionsonGeoscienceandRemoteSensing 44(7):1,7281,739. McCombs,J.W.,S.D.Roberts,andD.L.Evans.2003.Influenceoffusinglidarand multispectralimageryonremotelysensedestimatesofstanddensityandmeantree heightinamanagedloblollypineplantation.ForestScience49(3):457466. McRoberts,R.E.2008.Usingsatelliteimageryandtheknearestneighborstechniqueasa bridgebetweenstrategicandmanagementforestinventories.RemoteSensingof Environment112(5):2,2122,221. Means,J.E.,S.A.Acker,D.J.Harding,J.B.Blair,M.A.Lefsky,W.B.Cohen,M.E.Harmon,and W.A.McKee.1999.Useoflargefootprintscanningairbornelidartoestimateforest standcharacteristicsintheWesternCascadesofOregon.RemoteSensingof Environment67(3):298308.

FAGAN AND DEFRIES

109

Moghaddam,M.,J.L.Dungan,andS.Acker.2002.Forestvariableestimationfromfusionof SARandmultispectralopticaldata.IEEETransactionsonGeoscienceandRemote Sensing40(10):2,1762,187. Morales,R.M.,T.Miura,andT.Idol.2008.AnassessmentofHawaiiandryforestcondition withfineresolutionremotesensing.ForestEcologyandManagement255(7):2,524 2,532. Morton,D.C.,R.S.DeFries,Y.E.Shimabukuro,L.O.Anderson,F.D.B.EspiritoSanto,M. Hansen,andM.Carroll.2005.Rapidassessmentofannualdeforestationinthe BrazilianAmazonusingMODISdata.EarthInteractions9(8):122. Muukkonen,P.,andJ.Heiskanen.2007.Biomassestimationoveralargeareabasedon standwiseforestinventorydataandASTERandMODISsatellitedata:Apossibility toverifycarboninventories.RemoteSensingofEnvironment107(4):617624. Myneni,R.B.,W.Z.Yang,R.R.Nemani,A.R.Huete,R.E.Dickinson,Y.Knyazikhin,K.Didan,et al.2007.LargeseasonalswingsinleafareaofAmazonrainforests.Proceedingsof theNationalAcademyofSciencesoftheUnitedStatesofAmerica104(12):4,820 4,823. Naesset,E.,T.Gobakken,J.Holmgren,H.Hyyppa,J.Hyyppa,M.Maltamo,M.Nilsson,H. Olsson,A.Persson,andU.Soderman.2004.Laserscanningofforestresources:The Nordicexperience.ScandinavianJournalofForestResearch19(6):482499. Naesset,E.,andT.Gobakken.2008.Estimationofaboveandbelowgroundbiomassacross regionsoftheborealforestzoneusingairbornelaser.RemoteSensingof Environment112(6):3,0793,090. NASA(NationalAeronauticsandSpaceAdministration),2009a.IceSatHomepage. http://icesat.gsfc.nasa.gov/index.php(accessedJuly21,2009). .2009b.TheEarthSensingLegacy.EarthObservatory. http://earthobservatory.nasa.gov/Features/EO1/eo1_2.php. NaturalResourcesCanada.2005.HorizontaltransmitVerticalreceivepolarization. Glossaryofremotesensingterms.CanadaCentreforRemoteSensing. http://www.cct.nrcan.gc.ca/glossary/index_e.php?id=1560. Neeff,T.,L.V.Dutra,J.R.dosSantos,C.D.Freitas,andL.S.Araujo.2005.Tropicalforest measurementbyinterferometricheightmodelingandPbandradarbackscatter. ForestScience51(6):585594. Neigh,C.S.R.,C.J.Tucker,andJ.R.G.Townshend.2008.NorthAmericanvegetationdynamics observedwithmultiresolutionsatellitedata.RemoteSensingofEnvironment 112(4):1,7491,772. Nelson,R.,K.J.Ranson,G.Sun,D.S.Kimes,V.Kharuk,andP.Montesano.2009.Estimating SiberiantimbervolumeusingMODISandICESat/GLAS.RemoteSensingof Environment113(3):691701.

FAGAN AND DEFRIES

110

Nelson,R.F.,P.Hyde,P.Johnson,B.Emessiene,M.L.Imhoff,R.Campbell,andW.Edwards. 2007.InvestigatingRaDARLiDARsynergyinaNorthCarolinapineforest.Remote SensingofEnvironment110(1):98108. Nelson,R.F.,D.S.Kimes,W.A.Salas,andM.Routhier.2000.Secondaryforestageand tropicalforestbiomassestimationusingthematicmapperimagery.Bioscience 50(5):419431. Nilsson,M.,1996.Estimationoftreeweightsandstandvolumeusinganairbornelidar system.RemoteSensingofEnvironment56(1):17. NRC(NationalResearchCouncil),2007.EarthScienceandApplicationsfromSpace:National ImperativesfortheNextDecadeandBeyond.Washington,DC:NationalResearch Council,TheNationalAcademiesPress. Oberto,B.,A.Loverro,S.Hu,andB.Blair.2008.MissionandSpacecraftConfiguration Studies(PriorityTasks1,2,&4).DESDynIScienceStudyGroupMeeting.June2008, Greenbelt,MD. Olander,L.P.,H.K.Gibbs,M.Steininger,J.J.Swenson,andB.C.Murray.2008.Reference scenariosfordeforestationandforestdegradationinsupportofREDD:Areviewof dataandmethods.EnvironmentalResearchLetters3(2). Ollinger,S.V.,andM.L.Smith.2005.Netprimaryproductionandcanopynitrogenina temperateforestlandscape:Ananalysisusingimagingspectroscopy,modelingand fielddata.Ecosystems8(7):760778. Ouma,Y.O.,J.Tetuko,andR.Tateishi.2008.Analysisofcooccurrenceanddiscretewavelet transformtexturesfordifferentiationofforestandnonforestvegetationinvery highresolutionopticalsensorimagery.InternationalJournalofRemoteSensing 29(12):3,4173,456. Page,S.E.,F.Siegert,J.O.Rieley,H.D.V.Boehm,A.Jaya,andS.Limin.2002.Theamountof carbonreleasedfrompeatandforestfiresinIndonesiaduring1997.Nature 420(6911):6165. Palace,M.,M.Keller,G.P.Asner,S.Hagen,andB.Braswell.2008.Amazonforeststructure fromIKONOSsatellitedataandtheautomatedcharacterizationofforestcanopy properties.Biotropica40(2):141150. Palmer,P.I.2008.Quantifyingsourcesandsinksoftracegasesusingspaceborne measurements:Currentandfuturescience.PhilosophicalTransactionsoftheRoyal SocietyaMathematicalPhysicalandEngineeringSciences366(1885):45094528. Park,N.W.,andK.H.Chi.2008.Integrationofmultitemporal/polarizationCbandSARdata setsforlandcoverclassification.InternationalJournalofRemoteSensing29(16): 4,6674,688. Patenaude,G.,R.Milne,andT.P.Dawson.2005.Synthesisofremotesensingapproachesfor forestcarbonestimation:ReportingtotheKyotoProtocol.EnvironmentalScience andPolicy8(2):161178.

FAGAN AND DEFRIES

111

Peterson,L.K.,K.M.Bergen,D.G.Brown,L.VashchukandY.Blam.2009.Forestedland coverpatternsandtrendsoverchangingforestmanagementerasintheSiberian Baikalregion.ForestEcologyandManagement257(3):911922. Pflugmacher,D.,W.Cohen,R.Kennedy,andM.Lefsky.2008.Regionalapplicabilityofforest heightandabovegroundbiomassmodelsforthegeosciencelaseraltimetersystem. ForestScience54(6):647657. Phillips,O.L.,L.Aragao,S.L.Lewis,J.B.Fisher,J.Lloyd,G.LopezGonzalez,Y.Malhi,etal. 2009.DroughtsensitivityoftheAmazonrainforest.Science323(5919):1,344 1,347. Piao,S.L.,J.Y.Fang,P.Ciais,P.Peylin,Y.Huang,S.Sitch,andT.Wang.2009.Thecarbon balanceofterrestrialecosystemsinChina.Nature458(7241):1,0091,082. Podest,E.,andS.Saatchi.2002.ApplicationofmultiscaletextureinclassifyingJERS1radar dataovertropicalvegetation.InternationalJournalofRemoteSensing23(7):1,487 1,506. Popescu,S.C.,R.H.Wynne,andR.F.Nelson.2003.Measuringindividualtreecrowndiameter withlidarandassessingitsinfluenceonestimatingforestvolumeandbiomass. CanadianJournalofRemoteSensing29(5):564577. Popescu,S.C.,R.H.Wynne,andJ.A.Scrivani.2004.Fusionofsmallfootprintlidarand multispectraldatatoestimateplotlevelvolumeandbiomassindeciduousandpine forestsinVirginia,USA.ForestScience50(4):551565. Potere,D.2008.HorizontalpositionalaccuracyofGoogleEarth'shighresolutionimagery archive.Sensors8(12):7,9737,981. Powell,S.L.,D.Pflugmacher,A.A.Kirschbaum,Y.Kim,andW.B.Cohen.2007.Moderate resolutionremotesensingalternatives:AreviewofLandsatlikesensorsandtheir applications.JournalofAppliedRemoteSensing1. Qian,D.,N.H.Younan,R.King,andV.P.Shah.2007.Ontheperformanceevaluationofpan sharpeningtechniques.IEEEGeoscienceandRemoteSensingLetters4(4):518522. Ramankutty,N.,H.K.Gibbs,F.Achard,R.Defriess,J.A.Foley,andR.A.Houghton.2007. Challengestoestimatingcarbonemissionsfromtropicaldeforestation.Global ChangeBiology13(1):5166. RamosDaSilva,R.,D.Silva,andR.Avissar.2008.Regionalimpactsoffuturelandcover changesontheAmazonbasinwetseasonclimate.JournalofClimate21:1,153 1,170. RapidEye,A.,2009.RapidEye:GeofactsTurnedIntoKnowledge.http://www.rapideye.de/ (accessedJuly23,2009). Read,J.M.,D.B.Clark,E.M.Venticinque,andM.P.Moreira.2003.Applicationofmerged1m and4mresolutionsatellitedatatoresearchandmanagementintropicalforests. JournalofAppliedEcology40(3):592600.

FAGAN AND DEFRIES

112

Roberts,G.,M.J.Wooster,G.L.W.Perry,N.Drake,L.M.Rebelo,andF.Dipotso.2005. Retrievalofbiomasscombustionratesandtotalsfromfireradiativepower observations:ApplicationtosouthernAfricausinggeostationarySEVIRIimagery. JournalofGeophysicalResearchAtmospheres110(D21). Rogan,J.,J.Franklin,D.Stow,J.Miller,C.Woodcock,andD.Roberts.2008.Mappingland covermodificationsoverlargeareas:Acomparisonofmachinelearningalgorithms. RemoteSensingofEnvironment112(5):2,2722,283. Rosenqvist,A.,A.Milne,R.Lucas,M.Imhoff,andC.Dobson.2003.Areviewofremote sensingtechnologyinsupportoftheKyotoProtocol.EnvironmentalScienceand Policy6(5):441455. Rosenqvist,A.,M.Shimada,B.Chapman,A.Freeman,G.DeGrandi,S.Saatchi,andY.Rauste. 2000.TheGlobalRainForestMappingprojectareview.InternationalJournalof RemoteSensing21(67):1,3751,387. Roy,D.P.,Y.Jin,P.E.Lewis,andC.O.Justice.2005.Prototypingaglobalalgorithmfor systematicfireaffectedareamappingusingMODIStimeseriesdata.RemoteSensing ofEnvironment97(2):137162. Roy,D.P.,J.Ju,P.Lewis,C.Schaaf,F.Gao,M.Hansen,andE.Lindquist.2008.Multitemporal MODISLandsatdatafusionforrelativeradiometricnormalization,gapfilling,and predictionofLandsatdata.RemoteSensingofEnvironment112(6),3:1123,130. Rudel,T.K.,O.T.Coomes,E.Moran,F.Achard,A.Angelsen,J.C.Xu,andE.Lambin.2005. Foresttransitions:Towardsaglobalunderstandingoflandusechange.Global EnvironmentalChangeHumanandPolicyDimensions15(1):2331. Saatchi,S.S.,R.A.Houghton,R.Alvala,J.V.Soares,andY.Yu.2007.Distributionof abovegroundlivebiomassintheAmazonbasin.GlobalChangeBiology13(4):816 837. Saatchi,S.S.,M.Marlier,R.L.Chazdon,D.B.Clark,andA.E.Russell,inreview.Impactof SpatialVariabilityofforeststructureonRadarEstimationofAbovegroundBiomass inTropicalForests.RemoteSensingofEnvironment. Saatchi,S.S.,B.Nelson,E.Podest,andJ.Holt.2000.MappinglandcovertypesintheAmazon Basinusing1kmJERS1mosaic.InternationalJournalofRemoteSensing21(67): 1,2011,234. Saatchi,S.S.,J.V.Soares,andD.S.Alves,1997.Mappingdeforestationandlandusein AmazonrainforestbyusingSIRCimagery.RemoteSensingofEnvironment59(2): 191202. Sader,S.A.,M.Bertrand,andE.H.Wilson.2003.Satellitechangedetectionofforestharvest patternsonanindustrialforestlandscape.ForestScience49(3):341353. Salajanu,D.,andC.E.Olson.2001.ThesignificanceofspatialresolutionIdentifyingforest coverfromsatellitedata.JournalofForestry99(6)3238.

FAGAN AND DEFRIES

113

Salati,E.,andP.B.Vose.1984.Amazonbasin:Asysteminequilibrium.Science225(4658): 129. SanchezAzofeifa,G.A.,K.L.CastroEsau,W.A.Kurz,andA.Joyce.2009.Monitoringcarbon stocksinthetropicsandtheremotesensingoperationallimitations:Fromlocalto regionalprojects.EcologicalApplications19(2):480494. Sanderson,E.W.,M.Jaiteh,M.A.Levy,K.H.Redford,A.V.Wannebo,andG.Woolmer.2002. Thehumanfootprintandthelastofthewild.Bioscience52(10):891904. Santos,C.,andJ.P.Messina.2008.MultisensordatafusionformodelingAfricanpalminthe EcuadorianAmazon.PhotogrammetricEngineeringandRemoteSensing74(6):711 723. Schlerf,M.,C.Atzberger,andJ.Hill.2005.Remotesensingofforestbiophysicalvariables usingHyMapimagingspectrometerdata.RemoteSensingofEnvironment95(2): 177194. Schroeder,P.,S.Brown,J.M.Mo,R.Birdsey,andC.Cieszewski.1997.Biomassestimationfor temperatebroadleafforestsoftheUnitedStatesusinginventorydata.ForestScience 43(3):424434. Schroeder,T.A.,W.B.Cohen,C.H.Song,M.J.Canty,andZ.Q.Yang.2006.Radiometric correctionofmultitemporalLandsatdataforcharacterizationofearlysuccessional forestpatternsinwesternOregon.RemoteSensingofEnvironment103(1):1626. Schroeder,W.,I.Csiszar,andJ.Morisette.2008a.Quantifyingtheimpactofcloud obscurationonremotesensingofactivefiresintheBrazilianAmazon.Remote SensingofEnvironment112(2):456470. Schroeder,W.,E.Prins,L.Giglio,I.Csiszar,C.Schmidt,J.MorisetteandD.Morton.2008b. ValidationofGOESandMODISactivefiredetectionproductsusingASTERandETM plusdata.RemoteSensingofEnvironment112(5):2,7112,726. Schull,M.A.,S.Ganguly,A.Samanta,D.Huang,N.V.Shabanov,J.P.Jenkins,J.C.Chiu,A. Marshak,J.B.Blair,R.B.Myneni,andY.Knyazikhin.2007.Physicalinterpretationof thecorrelationbetweenmultianglespectraldataandcanopyheight.Geophysical ResearchLetters34(18). Sedano,F.,D.Gomez,P.Gong,andG.S.Biging.2008.Treedensityestimationinatropical woodlandecosystemwithmultiangularMISRandMODISdata.RemoteSensingof Environment112(5):2,5232,537. Sesnie,S.E.,P.E.Gessler,B.Finegan,andS.Thessler.2008.IntegratingLandsatTMand SRTMDEMderivedvariableswithdecisiontreesforhabitatclassificationand changedetectionincomplexneotropicalenvironments.RemoteSensingof Environment112(5):2,1452,159. Sexton,J.O.,T.Bax,P.Siqueira,J.J.Swenson,andS.Hensley.2009.Acomparisonoflidar, radar,andfieldmeasurementsofcanopyheightinpineandhardwoodforestsof southeasternNorthAmerica.ForestEcologyandManagement257(3):1,1361,147.

FAGAN AND DEFRIES

114

Siegert,F.,andG.Ruecker.2000.UseofmultitemporalERS2SARimagesforidentification ofburnedscarsinsoutheastAsiantropicalrainforest.InternationalJournalof RemoteSensing21(4):831837. Short,NicholasM.2009.TheRemoteSensingTutorial.http://rst.gsfc.nasa.gov Slatton,K.C.,M.M.Crawford,andB.L.Evans.2001.Fusinginterferometricradarandlaser altimeterdatatoestimatesurfacetopographyandvegetationheights.IEEE TransactionsonGeoscienceandRemoteSensing39(11):2,4702,482. Smith,M.L.,S.V.Ollinger,M.E.Martin,J.D.Aber,R.A.Hallett,andC.L.Goodale.2002.Direct estimationofabovegroundforestproductivitythroughhyperspectralremote sensingofcanopynitrogen.EcologicalApplications12(5):1,2861,302. Song,C.2007.Estimatingtreecrownsizewithspatialinformationofhighresolutionoptical remotelysensedimagery.InternationalJournalofRemoteSensing28(15):3,305 3,322. Song,C.,T.A.Schroeder,andW.B.Cohen.2007.Predictingtemperateconiferforest successionalstagedistributionswithmultitemporalLandsatThematicMapper imagery.RemoteSensingofEnvironment106(2):228237. Soudani,K.,G.leMaire,E.Dufrene,C.Francois,N.Delpierre,E.Ulrich,andS.Cecchini.2008. Evaluationoftheonsetofgreenupintemperatedeciduousbroadleafforests derivedfromModerateResolutionImagingSpectroradiometer(MODIS)data. RemoteSensingofEnvironment112(5):2,6432,655. Souza,C.M.,andD.Roberts.2005.MappingforestdegradationintheAmazonregionwith Ikonosimages.InternationalJournalofRemoteSensing26(3):425429. Souza,C.M.,D.A.Roberts,andM.A.Cochrane.2005.Combiningspectralandspatial informationtomapcanopydamagefromselectiveloggingandforestfires.Remote SensingofEnvironment98(23):329343. Steininger,M.K.1996.TropicalsecondaryforestregrowthintheAmazon:Age,areaand changeestimationwithThematicMapperdata.InternationalJournalofRemote Sensing17(1):927. Steininger,M.K.2000.Satelliteestimationoftropicalsecondaryforestaboveground biomass:datafromBrazilandBolivia.InternationalJournalofRemoteSensing21(6 7):1,1391,157. StOnge,B.,Y.Hu,andC.Vega.2008.Mappingtheheightandabovegroundbiomassofa mixedforestusinglidarandstereoIkonosimages.InternationalJournalofRemote Sensing29(5):1,2771,294. Sun,G.,K.Ranson,J.Masek,A.Fu,andD.Wang.2007.PredictingTreeHeightandBiomass fromGLASData.Paperpresentedatthe10thInternationalSymposiumonPhysical MeasurementsandSignaturesinRemoteSensing.March2007,Davos,Switzerland.

FAGAN AND DEFRIES

115

Swanson,A.,L.Abelson,T.Manning,andL.Newhart.2009.RemoteSensingAssessmentof ForestCarbonContent,NorthropGrummanAerospaceSystems,Northrop GrummondCorporation,Washington,DC. TerMikaelian,M.T.,andM.D.Korzukhin.1997.BiomassequationsforsixtyfiveNorth Americantreespecies.ForestEcologyandManagement97(1):124. Thenkabail,P.S.,E.A.Enclona,M.S.Ashton,C.Legg,andM.J.DeDieu.2004.Hyperion, IKONOS,ALI,andETMplussensorsinthestudyofAfricanrainforests.Remote SensingofEnvironment90(1):2343. Thiel,C.,P.Drezet,C.Weise,S.Quegan,andC.Schmullius.2006.Radarremotesensingfor thedelineationofforestcovermapsandthedetectionofdeforestation.Forestry 79(5):589597. Thompson,S.D.,S.E.Gergel,andN.C.Coops.2008.Classificationoflateseralcoastal temperaterainforestswithhighspatialresolutionQuickBirdimagery.Canadian JournalofRemoteSensing34:S460S470. Tian,H.,J.M.Melillo,D.W.Kicklighter,A.D.McGuire,J.V.K.Helfrich,B.Moore,andC.J. Vorosmarty.1998.Effectofinterannualclimatevariabilityoncarbonstoragein Amazonianecosystems.Nature396(6712):664667. Ticehurst,C.,A.Held,andS.Phinn.2004.IntegratingJERS1imagingradarandelevation modelsformappingtropicalvegetationcommunitiesinFarNorthQueensland, Australia.PhotogrammetricEngineeringandRemoteSensing70(11):1,2591,266. Townsend,P.A.,T.R.Lookingbill,C.C.Kingdon,andR.H.Gardner.2009.Spatialpattern analysisformonitoringprotectedareas.RemoteSensingofEnvironment113(7): 1,4101,420. Townshend,J.R.,C.O.Justice,andV.T.Kalb.1987.Characterizationandclassificationof SouthAmericanlandcovertypesusingsatellitedata.InternationalJournalofRemote Sensing8:1,1891,207. Treuhaft,R.N.,G.P.Asner,andB.E.Law.2003.Structurebasedforestbiomassfromfusion ofradarandhyperspectralobservations.GeophysicalResearchLetters30(9). Treuhaft,R.N.,andP.R.Siqueira,2000.Verticalstructureofvegetatedlandsurfacesfrom interferometricandpolarimetricradar,RadioScience,35(1),141177. Tucker,C.J.,J.R.Townshend,andT.E.Goff.1985.Africanlandcoverclassificationusing satellitedata.Science227:369375. Tucker,C.J.,andJ.R.G.Townshend.2000.Strategiesformonitoringtropicaldeforestation usingsatellitedata.InternationalJournalofRemoteSensing21(67):1,4611,471. UCS(UnionofConcernedScientists).2009.UCSSatelliteDatabase.UnionofConcerned Scientists. http://www.ucsusa.org/nuclear_weapons_and_global_security/space_weapons/tec hnical_issues/ucssatellitedatabase.html(accessedJuly23,2009).

FAGAN AND DEFRIES

116

Uhl,C.,R.Buschbacher,andE.A.S.Serrao.1988.AbandonedpasturesineasternAmazonia. 1.Patternsofplantsuccession.JournalofEcology76(3):663681. UNFCCC(UnitedNationsFrameworkConventiononClimateChange).2001.COP7:The Marrakechaccords.(ed.byUNFCCCSecretariat.)Bonn,Germany. http://www.unfccc.int UNOOSA(UnitedNationsOfficeforOuterSpaceAffairs).2009.GeneralAssembly Resolution41/65:PrinciplesRelatingtoRemoteSensingoftheEarthfromOuter Space.UnitedNationsOfficeforOuterSpaceAffairs. http://www.oosa.unvienna.org/oosa/en/SpaceLaw/gares/html/gares_41_0065.ht ml(accessedJuly23,2009). Uriarte,M.,L.W.Rivera,J.K.Zimmerman,T.M.Aide,A.G.Power,andA.S.Flecker.2004. Effectsoflandusehistoryonhurricanedamageandrecoveryinaneotropicalforest. PlantEcology174(1):4958. Ustin,S.L.,D.A.Roberts,J.A.Gamon,G.P.Asner,andR.O.Green.2004.Usingimaging spectroscopytostudyecosystemprocessesandproperties.Bioscience54(6):523 534. Ustin,S.L.,andQ.F.Xiao.2001.Mappingsuccessionalborealforestsininteriorcentral Alaska.InternationalJournalofRemoteSensing22(9):1,7791,797. vanAardt,J.A.N.,R.H.Wynne,andJ.A.Scrivani.2008.Lidarbasedmappingofforestvolume andbiomassbytaxonomicgroupusingstructurallyhomogenoussegments. PhotogrammetricEngineeringandRemoteSensing74(8):1,0331,044. vanderSanden,J.J.,andD.H.Hoekman.1999.Potentialofairborneradartosupportthe assessmentoflandcoverinatropicalrainforestenvironmentRemoteSensingof Environment68(1):2640. vanderWerf,G.R.,J.Dempewolf,S.N.Trigg,J.T.Randerson,P.S.Kasibhatla,L.Gigliof,D. Murdiyarso,etal.2008.Climateregulationoffireemissionsanddeforestationin equatorialAsia.ProceedingsoftheNationalAcademyofSciencesoftheUnitedStates ofAmerica105(51):20,35020,355. vanderWerf,G.R.,J.T.Randerson,L.Giglio,G.J.Collatz,P.S.Kasibhatla,andA.F.Arellano. 2006.Interannualvariabilityinglobalbiomassburningemissionsfrom1997to 2004.AtmosphericChemistryandPhysics6:3,4233,441. Waggoner,P.E.(ed.).2009.ForestInventories:DiscrepanciesandUncertainties"The WorldsForests:DesignandImplementationofEffectiveMeasurementand Monitoring"Discussionpaper0929.Washington,DC:ResourcesfortheFuture. Walker,W.S.,J.M.Kellndorfer,E.LaPoint,M.Hoppus,andJ.Westfall.2007.Anempirical InSARopticalfusionapproachtomappingvegetationcanopyheight.RemoteSensing ofEnvironment109(4):482499. Wallerman,J.,andJ.Holmgren.2007.Estimatingfieldplotdataofforeststandsusing airbornelaserscanningandSPOTHRGdata.RemoteSensingofEnvironment110(4): 501508.

FAGAN AND DEFRIES

117

Wang,C.Z.,J.G.Qi,andM.Cochrane.2005.Assessmentoftropicalforestdegradationwith canopyfractionalcoverfromlandsatETMplusandIKONOSimagery.Earth Interactions9. Wang,X.Q.,Z.Y.Li,X.E.Liu,G.Deng,andZ.H.Jiang.2007.Estimatingstemvolumeusing QuickBirdimageryandallometricrelationshipsforopenPopulusxiaohei plantations.JournalofIntegrativePlantBiology49(9):1,3041,312. West,G.B.,B.J.Enquist,andJ.H.Brown.2009.Ageneralquantitativetheoryofforest structureanddynamics.ProceedingsoftheNationalAcademyofSciencesofthe UnitedStatesofAmerica106(17):7,0407,045. Williams,M.2008.Anewlookatglobalforesthistoriesoflandclearing.AnnualReviewof EnvironmentalResources33:345367. Wilson,E.H.,andS.A.Sader.2002.Detectionofforestharvesttypeusingmultipledatesof LandsatTMimagery.RemoteSensingofEnvironment80(3):385396. Wolter,P.,P.TownsendandB.Sturtevant.inpress,2009.Estimationofforeststructural parametersusing5and10meterSPOT5satellitedata.RemoteSensingof Environmentforthcoming. Woodcock,C.E.,S.A.Macomber,M.PaxLenney,andW.B.Cohen.2001.Monitoringlarge areasforforestchangeusingLandsat:Generalizationacrossspace,timeandLandsat sensors.RemoteSensingofEnvironment78(12):194203. Wulder,M.A.,R.J.Hall,N.C.Coops,andS.E.Franklin.2004.Highspatialresolutionremotely senseddataforecosystemcharacterization.Bioscience54(6):511521. Wulder,M.A.,J.C.White,N.C.Coops,andC.R.Butson.2008.Multitemporalanalysisofhigh spatialresolutionimageryfordisturbancemonitoring.RemoteSensingof Environment112(6):2,7292,740. Xian,G.,C.Homer,andJ.Fry.2009.Updatingthe2001NationalLandCoverDatabaseland coverclassificationto2006byusingLandsatimagerychangedetectionmethods. RemoteSensingofEnvironment113(6):1,1331,147. Xiao,J.F.,Q.L.Zhuang,D.D.Baldocchi,B.E.Law,A.D.Richardson,J.Q.Chen,R.Oren,etal. 2008.EstimationofnetecosystemcarbonexchangefortheconterminousUnited StatesbycombiningMODISandAmeriFluxdata.AgriculturalandForestMeteorology 148(11):1,8271,847. Zhao,K.G.,S.Popescu,andR.Nelson.2009.Lidarremotesensingofforestbiomass:Ascale invariantestimationapproachusingairbornelasers.RemoteSensingofEnvironment 113(1):182196. Zheng,D.L.,J.Rademacher,J.Q.Chen,T.Crow,M.Bresee,J.leMoine,andS.R.Ryu.2004. EstimatingabovegroundbiomassusingLandsat7ETM+dataacrossamanaged landscapeinnorthernWisconsin,USA.RemoteSensingofEnvironment93(3):402 411.

FAGAN AND DEFRIES

118

Zink,M.,G.Krieger,H.Fiedler,andA.Moreira.2007.TheTanDEMXMission:Overviewand Status.PaperpresentedattheIEEEInternationalGeoscienceandRemoteSensing Symposium.July2007,Barcelona,Spain.

FAGAN AND DEFRIES

119

Vous aimerez peut-être aussi