Multi-temporal analysis of high spatial resolution imagery for disturbance monitoring

www.elsevier.com/locate/rse

Multi-temporalanalysisofhighspatialresolutionimageryfor

disturbancemonitoring

MichaelA.Wuldera,⁎,JoanneC.Whitea,NicholasC.Coopsb,ChristopherR.Butsona,1aCanadianForestService(PacificForestryCentre),NaturalResourcesCanada,506WestBurnsideRoad,Victoria,BritishColumbia,CanadaV8Z1M5

bDepartmentofForestResourceManagement,UniversityofBritishColumbia,2424MainMall,UniversityofBritishColumbia,

Vancouver,BritishColumbia,CanadaV6T1Z4

Received20September2007;receivedinrevisedform4January2008;accepted5January2008

Abstract

Mountainpinebeetleredattackdamagehasbeensuccessfullydetectedandmappedusingsingle-datehighspatialresolution(b4m)satellitemulti-spectraldata.Forestmanagers;however,needtomonitorlocationsforchangesinbeetlepopulationsovertime.Specifically,countsofindividualtreesattackedinsuccessiveyearsprovideanindicationofbeetlepopulationgrowthanddynamics.Surveysaretypicallyusedtoestimatetheratioofgreen(current)attacktreestored(previous)attacktreesorG:R.Inthisstudy,weestimateaveragestand-levelG:Rusingatime-seriesofQuickBirdmulti-spectralandpanchromaticsatellitedata,combinedwithfielddataforthreeforestedstandsnearMerritt,BritishColumbia,Canada.UsingaratioofQuickBirdredtogreenwavelengths(Red–GreenIndexorRGI),thechangeinRGI(ΔRGI)insuccessiveimagepairsisusedtoestimateredattackdamagein2004,2005,and2006,withtruepositiveaccuraciesrangingfromto93%.Toovercomeissuesassociatedwithdifferingviewinggeometryandilluminationanglesthatimpairtrackingofindividualtreesthroughtime,segmentsaregeneratedfromtheQuickBirdmulti-spectraldatatoidentifysmallgroupsoftrees.Thesesegmentsthenserveasthevehicleformonitoringchangesinredattackdamageovertime.Alocalmaximafilterisappliedtothepanchromaticdatatoestimatestemcounts,therebyallowinganindicationofthetotalstandpopulationatriskofattack.Bycombiningtheredattackdamageestimateswiththelocalmaximastemcounts,predictionsaremadeofthenumberofattackedtreesinagivenyear.Backcastingthecurrentyear'sredattackdamagedtreesasthepreviousyear'sgreenattackfacilitatestheestimationofanaveragestandG:R.Inthisstudyarea,theseretrospectiveG:Rvaluescloselymatchthosegeneratedfromfieldsurveys.Theresultsofthisstudyindicatethatamonitoringprogramusingatimeseriesofhighspatialresolutionremotelysenseddata(multi-spectralandpanchromatic)overselectsamplelocations,couldbeusedtoestimateG:Roverlargeareas,facilitatinglandscapelevelmanagementstrategiesand/orprovidingamechanismforassessingtheefficacyofpreviouslyimplementedstrategies.

CrownCopyright©2008PublishedbyElsevierInc.Allrightsreserved.

Keywords:QuickBird;Highspatialresolution;Monitoring;Imageprocessing;Insect;Changedetection;Dynamics

1.Introduction

Thecurrentoutbreakofmountainpinebeetle(Dendroctonusponderosae)inwesternCanadaisofunprecedentedpropor-⁎Correspondingauthor.Tel.:+12503636090;fax:+12503630775.E-mailaddress:mike.wulder@pfc.cfs.nrcan.gc.ca(M.A.Wulder).1Currentaddress:ForestAnalysisandInventoryBranch,BritishColumbiaMinistryofForestsandRange,7thFloor,727FisgardStreet,Victoria,BC,CanadaV8W1R8.tions;in1999,theareaimpactedbymountainpinebeetlewasestimatedtobe1,000ha,andby2006theareaimpactedhadincreasedto9.2millionha(Westfall,2007).Itisestimatedthatby2013,80%ofthematurepineinBritishColumbiawillhavebeenkilledbythebeetle(Eng,2005).Therapidexpansionofthebeetlepopulationhasbeenfacilitatedbythelargeamountofmaturelodgepolepine,whichhastripledinthelastcenturyasaresultofintensivefiresuppressionactivities(Taylor&Carroll,2004).Severalsuccessiveyearsoffavorableclimaticconditionshaveresultedinanincreaseinclimaticallysuitableareasfor

0034-4257/$-seefrontmatter.CrownCopyright©2008PublishedbyElsevierInc.Allrightsreserved.doi:10.1016/j.rse.2008.01.010

2730M.A.Wulderetal./RemoteSensingofEnvironment112(2008)2729–2740

brooddevelopment(Logan&Powell,2001;Carrolletal.,2004,2006a)andsubsequentincreasesinmountainpinebeetlerange,northward,eastward,andtogreaterelevations.1.1.Informationneed

Countsofindividualtreesinconsecutiveyearsprovideanindicationofpopulationgrowthanddynamics.Onceahosttreehasbeenattackedandkilledbymountainpinebeetle,itsfoliagewillremaingreenforaninitialperiod,andthisisknownasthegreen(current)attackstage(Wulderetal.,2006a).Thefoliagewillgraduallyfade,andby12monthsafterattack,90%ofkilledtreeswillhaveredfoliage(Amman,1982;Henigmanetal.,1999).Thisisthevisuallydistinctredattackstage,anditisthisstageoftheinfestationthatmaybecapturedbyremotesensingmethods(Wulderetal.,2006b).By3yearsaftertheinitialattack,mostofthekilledtreeswillhavelostalloftheirneedles,andthisisknownasthegreyattackstage(BritishColumbiaMinistryofForests,1995).Thereisvariabilityintherateatwhichthefoliagewilldiscolor,dependingonspeciesandsiteconditions(Safranyik,2004).

GiventhemagnitudeandthespatialextentoftheinfestationinBritishColumbia,theinformationneedsofforestmanagersarenowfocusedonmonitoringareasontheleadingedgeoftheinfestation,particularlyareasalongtheborderbetweentheCa-nadianprovincesofBritishColumbiaandAlberta.Here,mana-gementeffortsareattemptingtoreducethespreadofthemountainpinebeetlefurthereastwardintoAlbertaandbeyondintotheborealforest.Typically,groundsurveysareconductedannuallytodeterminebeetlepopulationtrends.Thepopulationtrendofaninfestationinaparticularstandisdeterminedbyestimatingtheratioofcurrentlyattackedtrees(greenattack)toone-year-oldattackedtrees(redattack)trees(G:R).Thisratioisestimatedfromasub-sampleoftreesinthestand(i.e.,basedonrandomlylocatedtransects).Aratiogreaterthan1isindicativeofanincreasingpopulation,whilearatiolessthan1isindicativeofadecliningpopulation.TheG:RisusefulforestimatingchangesinthesizeofthebeetlepopulationandisoneofthefactorsusedtodeterminethemanagementstrategyforforestmanagementunitsinBritishColumbia(e.g.,suppression,sal-vage,monitoring)(BritishColumbiaMinistryofForests,1995).Anotherformofsurveyistheover-winteringbroodassessmentsurvey,whichisconductedinthespringbysamplingunderthebarkofinfestedtrees.Thesebroodassessmentsurveysareusedtoestimatebroodmortalityandprovideanindicationoftherateofpopulationincrease(r-value)(Safranyik&Carroll,2006).1.2.Highspatialresolutionremotelysenseddata:Single-dateredattackmapping

Highspatialresolutiondataprovidetheopportunitytotrackforestdamage,suchasthatcausedbymountainpinebeetle,atalocalscale.Thelevelofdetailandtheflexibilityfordigitalanalysesaffordedbyhighspatialresolutionsatellitedatain-creaseopportunitiesbeyondwhatwaspreviouslypossiblewithaerialphotographyandLandsatdata(Sawayaetal.,2003).Single-date,highspatialresolutionremotelysenseddatahave

beenusedtomapredattackdamageattheforeststandandindividualtreelevel.Whiteetal.(2005)usesingle-dateIKONOS4mmulti-spectraldatatodetectmountainpinebeetleredattackdamageinforeststandswithlowandmoderatelevelsofattack,andcomparedredattackdamageestimatestoestimatesgeneratedfromairphotointerpretation.Resultsindicatethatwithinaone-pixelbuffer(4m)ofidentifieddamagepixels,theaccuracyofredattackdetectionwas70.1%forareasoflowinfestation(standswithlessthan5%oftreesdamaged)and92.5%forareasofmoderateinfestation(standswithbetween5%and20%oftreesdamaged).AnalysisofredattacktreesthatweremissedintheclassificationoftheIKONOSimageryindicatethatdetectionofredattackwasmosteffectiveforlargertreecrowns(diameterN1.5m)thatwereb11mfromotherredattacktrees.

Coopsetal.(2006a)usedhelicopterGPSmeasurementsofbeetleinfestedpinetreesinnorth-centralBritishColumbiatoidentifyareasofattackandnon-attackstandsonQuickBird2.4mmulti-spectraldata(blue,green,red,andnear-infrared).Usinga50mbufferaroundeachGPSpoint,theauthorstestedtheANOVAseparabilityofeachofthespectralbandsalongwithagreennessindexandred/greenspectralratiounderfourclasses:sunlitnon-attackedcrowns,denseredattackcrowns,fadercrowns,andshadowedcrowns.BasedontheresultsoftheANOVA,spectralthresholdswereusedtogenerateabinarymapforredattackandnon-attack(combiningnon-attackedcrownswithcrownsobscuredbyshadows).TheresultsshowthattheratiooftheredtogreenQuickBirdspectralbandswasthemostsignificantbandcombinationfordetectingredattackbeetledamage,andtheredattackidentifiedwiththeQuickBirdimageryhadgoodcorrespondencetobothforesthealthsurveydataandbroaderspatialresolutionLandsatimagery.Atthelandscapescale,improvementsinmountainpinebeetleredattackdetectionandmappinghavebeenfacilitatedbytheuseofmulti-temporalLandsatimagery(Skakunetal.,2003;Coopsetal.,2006b;Wulderetal.,2006c).Theauthorsanticipatedthatasimilarimprovementindetectionaccuracywouldbepossiblewithmultipledatesofhighspatialresolutiondata.1.3.ChangedetectionwithhighspatialresolutionimageryUntilrecently,changedetectionusinghighspatialresolutionimageryhasbeenconstrainedtotheuseofaerialphotographs.Sincehighspatialresolutionsatellitesensorshaveonlybeenincommercialoperationsince1999,theacquisitionofsatellite-basedhighspatialresolutionimagesinatemporalsequencehasbeenunavailableduetolimitedimagearchiving,difficultiesinacquiringcloud-freeimagery,andprohibitivecosts.Otheris-sues,suchasviewinggeometryandilluminationconditionsalsocomplicatetheuseofhighspatialresolutionsatelliteimageryforchangedetection(dependingonthedesiredtarget).QuickBirdhasarevisitrateof3to7daysasaresultofthesensor'svariablecross-trackandin-trackviewingcapability,andwhilethisofferstemporalflexibility,theoff-nadirviewinggeometryconfoundschangedetectionapproaches.RevisitratesforIKONOSaresimilar,with3–5daysforoff-nadirand144daysfortrue-nadirimagery.

M.A.Wulderetal./RemoteSensingofEnvironment112(2008)2729–27402731

ImandJensen(2005)concludedthattraditionalapproachestochangedetectionfailtooperatesuccessfullywithhighspatialresolutiondata,primarilybecauseoftheproliferationofhighfrequency,highcontrastobjects(shadows)intheseimages,andtheimpactofoff-nadirviewanglesthatcausehorizontallayoverofverticalobjectssuchastreesandbuildings.Thisisespeciallytrueinforests,asobject-orientedcrowndelineationusingeitherhighspatialresolutionsatelliteimagery(Johansen&Phinn,2006)orsmall-formataerialphotography(Keyetal.,2001)hasbeenmetwithlimitedsuccesswiththeexceptioninveryhighspatialresolution(b30cm)images(Pouliotetal.,2002).AsnerandWarner(2003)examinedtheinfluenceofvaryingviewingandilluminationgeometriesonestimationsofshadowfraction.Theyconcludethatthevariabilityinviewingandilluminationgeometriescanhaveanimpactonstudiesofvegetationstruc-ture;significantchangesinscenereflectancecharacteristicscanresultsolelyfromdifferentobservationgeometry.Sucheffectsarenotlimitedtohighspatialresolutiondataortoforestenvi-ronments(Smith&Wise,2007).Approachestochangedetectionwithremotelysenseddataaretypicallybasedonvisualinterpretation(Clarkeetal.,2004),pixel-based(Lambin&Strahler,1994;Mas,1999;Allen&Kupfer,2001;Im&Jensen,2005),orobject-basedmethods(Descléeetal.,2006).Changedetectionmethodsoftenformanintegralpartofanymonitoringsystemthatincorporatesremotelysenseddata.Pixel-basedchangedetectionrequiresextremelyaccurateco-registrationofimages(Townshendetal.,1992),andoftenpro-ducesresultsthatareheterogeneousor“saltandpepper”inappearance(Gong&Xu,2003).Thisiscausedbybothrandomvariationinthesensor'sresponse,andintrinsicvariabilityinthetarget(e.g.,forests).Object-basedchangedetectionapproacheshaveemergedasaresultofimprovedimagesegmentationcapa-bilities(Mölleretal.,2007).Imagesegmentationistheparti-tioningofadigitalimageintoasetofjointlyexhaustive,mutuallydisjointregionsthataremoreuniformwithinthemselvesthanwhencomparedtoadjacentregions.

Liuetal.(2006)usedAirborneDataAcquisitionandRegis-tration(ADAR)data(approximately1mspatialresolution)for

Fig.1.StudyarealocatedatAngstadCreek,25kmsouthofMerritt,BritishColumbia.StandsA,B,andCaredominatedbymaturelodgepolepine—theprimaryhostofthemountainpinebeetle.Theimageontherightisahighresolutionaerialphotograph(40cmpixel)collectedin2003.

2732M.A.Wulderetal./RemoteSensingofEnvironment112(2008)2729–2740

monitoringthespreadofSuddenOakDeathoveratwo-yearperiodinaforestenvironmentofcoastalCalifornia.Theyim-plementedapixel-basedmethodofchangedetectionbycombi-ningSupportVectorMachinesandMarkovRandomFieldsmodelaccounting.Challengeswereencounteredinimage-to-imageregistrationduetothewideviewangleofthesensor(nadir±35°),variationintheterrain,lowflyingaltitudeandaircraftyaw,pitch,androll.Byincorporatingspatial–temporalcontextualinformationintotheclassificationofSuddenOakDeath,theaccuracyofthediseasedetectionimproved.

Descléeetal.(2006)implementedanobject-basedapproachtoforestlandcoverchangedetectionusing3SPOT-HRVimages(20mspatialresolution),collectedovera10-yearperiod.Theobjectiveofthisstudywastoautomatetheidentificationofchange/no-changeinamannerthatwasscene-independent.Theirmethodincorporatedimagesegmentation,imagedifferencing,andstochasticanalysisofmulti-spectraldata(OB-Reflectancemethod).Theauthorsconcludedthattheirmulti-datesegmenta-tionapproacheffectivelyconstrainedvariabilityforsubsequentstatisticalanalysis.Changewasidentifiedthroughspectraldifferencing,andtheauthorsidentifiedtheprimaryadvantagesoftheobject-basedapproachasreducedprocessingtime,lesssensitivitytopixelregistrationerrors,andnorequirementforapredefinedthresholdtodistinguishchangesegmentsfromno-changesegments.Whilethelattermaybetrueifoneisonlyinterestedinabinarychangemap,detailonaspecifictypeofchangewouldrequireathresholdandassociatedcalibrationdata.

1.4.Objectives

Theobjectiveofthisprojectwastoestablishaforesthealthmonitoringsystemthatwouldaddresstheinformationneedofforestmanagersconcernedwiththespreadofmountainpinebeetles,suchaseastwardfromBritishColumbiaintoAlbertaandbeyond.Specifically,thegoalwastousehighspatialresolutionremotelysenseddatatocapturetwocriticalpiecesofinforma-tion:first,treebasedestimatesofredattackdamage;andse-condly,anestimateofstemdensity.Bycombiningthesetwopiecesofinformation,thenumberofredattacktreescouldbeestimatedforanygivenyear,andthenbybackcastingtheseestimates,retrospectiveG:Rratioscouldbegenerated.Thismethodalsogivesanestimateofthepopulation-at-risk(totalstems),whichisinformationnottypicallyprovidedbygroundsurveys.TheapproachwouldallowG:Rtobegeneratedoverlargeareas,providingasynopticviewofspatialvariationinG:Rwhichtheninturncouldbeusedforstrategicplanningoflandscapelevelbeetlemanagementpracticesand/ortoassesstheefficacyofstrategiesandmanagementpracticesimplemented.2.Studysite

ThestudysiteislocatedatAngstadCreek,25kmsouthofMerritt,BritishColumbia,Canada,centeredatapproximately49.84°Nand120.75°W(Fig.1).Thissitewasoriginallyselec-tedforastudyonmountainpinebeetleoutbreakdevelopment,designedtoexaminethetransitionfromendemictoincipient

Table1

AsummaryofmensurationalattributesforstandsA,B,andC

StandA

StandBStandCElevation(m)116311211173Area(ha)

16.717.69.9%lodgepolepine

92.594.296.8Density(stemsperha)1342

1198

1554

DBHa(cm)22.8±0.9624.1±0.5822.7±0.96Heighta(m)21.8±0.4221.4±0.4021.2±0.58Agea(years)

110.5±1.52

116.5±1.18

107.2±1.48

DBH=diameteratbreastheight(1.3m).aValuesreportedforlodgepolepineasmeans±1standarderror(SE).

endemicpopulationlevels(Carrolletal.,2006a,b).Criteriaforsiteselectionincludedahistoricallysuitableclimateformount-ainpinebeetle(Carrolletal.,2004),andanabsenceofdetect-ablebeetleactivityatthesite,orwithin10kmofthesite,in2002.From2002to2005,regularfieldsurveyswereconductedinselectedforeststandsinthestudyareatomonitormountainpinebeetlepopulations.Threeofthesestands(denotedasA,B,andC)wereusedinouranalysis,ranginginsizefrom10to18ha,containingpineasleadingspecies,thatweregreaterthan80yearsofage,withmoderatelydensestocking(800–1500stems/ha))(Carrolletal.,2006b).Variableprismplotsineachofthestandsindicatedthatthemensurationalcharacter-isticsofthestandssurveyedweresufficientlyuniform(Table1)andhighlysusceptibletomountainpinebeetleattack(i.e.,lodgepolepine).3.Data

3.1.Remotelysenseddata

FourQuickBird-2imageswereacquiredannuallyduringthe2003to2006summergrowingseasons(Table2andFig.2).Thesunelevationatthetimeofimageacquisitionrangedfrom53to58°,andtheoff-nadirviewanglesrangedfrom7.7to14.6°(Fig.3).QuickBirdimagerycontainsfourmulti-spectralbandswitha2.5mspatialresolution:0.45–0.52μm(blue);0.52–0.60μm(green);0.63–0.69μm(red);0.76–0.90μm(nearinfra-red);andapanchromaticband(0.45–0.90μm),witha0.68mspatialresolution(Birketal.,2003).3.2.Foresthealthsurveydata

AbaselinecensusofstandsA,B,andCwascompletedin2002;asurveygridwasestablishedineachstand,andthelocationofeachtreereferencedbyabearingtothenearestgridpoint.Inthisway,eachtreebecameasurveypoint,andastemmapwasgeneratedforeachstand.Atfour-weekintervalsbe-tweenJuneandSeptemberofeachyearfrom2002to2005,thehealthstatusofalltreesinstandsAandBwereassessedandrecorded;treesinstandCweresurveyedandcategorizedforhealthstatusonlyonceperyearinmid-September(Carrolletal.,2006b).NofielddatawerecollectedinstandBafter2004,asatthatpoint,thebeetlepopulationsinthestandhadreachedepidemiclevels.Intotal,906treessurveyedbetween2002and

M.A.Wulderetal./RemoteSensingofEnvironment112(2008)2729–2740

Table2

AcquisitionparametersfortheQuickBirdimageryAcquisitiondate2003-06-042004-07-172005-07-202006-08-15

UTCtime18:5118:5919:1219:23

Localtime(PDT)11:5111:5912:1212:23

Sunazimuth(degrees)146.8148.0153.7162.1

Sunelevation(degrees)59.558.358.853.2

Satelliteazimuth(degrees)51.243.5128.556.1

Satelliteelevation(degrees)74.475.375.481.9

Intrackviewangle(degrees)11.111.6−6.25.5

Crosstrackviewangle(degrees)9.67.512.05.3

2733

Off-nadirviewangle(degrees)14.613.813.47.7

2005wereattackedbymountainpinebeetle(176instandA,663instandB,and67instandC).4.Methods

Fig.4outlinesthemethodologyusedtoestimateG:Rratiosforeachofthestands.Eachcomponentofthisschematicisdescribedindetailinthefollowingsections.4.1.Imageradiometricandgeometricpre-processingTheQuickBirdimageswerereceivedasStandardImageproducts,inat-sensorradiance,fromthedataprovider(Digital-GlobeInc.,2007).Allpre-processingonthepanchromaticandmulti-spectralbandsweredoneindependently.Multi-temporalanalysisnecessitatescalibrationofpixelvalues(Wuetal.,2005);radiancevalueswereconvertedtotop-of-atmosphere(TOA)reflectanceusingthegainsandoffsetsprovidedintheimageheaderfilesalongwithsolarexoatmosphericirradiancesestimatedfromnormalizedspectralresponsedata(DigitalGlobeInc.,2007)andastandardextraterrestrialsolarspectrumreference(AmericanSocietyforTestingandMaterials,2000).

The2003imagewasgeoreferencedtoTerrainResourceInformationManagementII(TRIMII)(BritishColumbiaMinistryofEnvironmentLandsandParks,1992)aerialphotography(1:20,000)usingathirdorderpolynomialandacubicconvolutionresamplingalgorithm,resultinginaroot-mean-square(RMS)errorof0.5pixels(1.2m).The2003imagewasthenusedasthecontrolimagetowhichthe2004,2005,and2006imagesweresubsequentlyco-registered(alsousingathirdorderpolynomialandacubicconvolutionresamplingalgo-rithm).Forallthreeimages,thefinalRMSerrorwaslessthan0.5pixels(1.2m)forthemulti-spectralbands,andlessthan1pixel(0.6m)forthepanchromaticimageband.Aminimumof15GCPswereusedforeachofthegeometriccorrections.4.2.Redattackdetection

LocationsofredattackdamagewereidentifiedbycalculatingandthresholdingtheΔRGIvaluesforeachimagepair(2003–2004;2004–2005;2005–2006).Forthisstudy,fieldsurveydatawereusedtodeterminewhichtreesexperiencedattackbymountainpinebeetle,withthecrownsgraduallyturningred(anddetectablewithremotelysenseddata)inthefollowingyear.Therefore,treesidentifiedinthesurveyashavingbeenattackedin2003,wouldhaveturnedthecharacteristicredcolorin2004,

whilecrownsoftreesattackedin2004wouldappearredinthe2005image(andsoon).

Foreachimage,aratiooftheredtogreenwavelengthswascomputed(hereafterreferredtoastheRed–GreenIndexorRGI).Thisindexemphasizesthespectralchangeinfoliagecolorfromgreentored,andhasbeenusedsuccessfullyfordetectingmountainpinebeetleredattackdamagewithsingle-dateQuick-Birdimagery(Coopsetal.,2006a).Usingamulti-dateapproachforchangedetection,theRGIvaluesfromthe2004imageweresubtractedfromthecorrespondingRGIvaluesfromthe2003image,andthechangeinRGIvalueswascalculated(hereafterreferredtoasΔRGI_2004).

A10%sub-sample(n=90)ofthefielddatawasselectedatrandomforcalibrationpurposes,withtheremainderofthefielddatausedforvalidation.Forexample,asubsetofthe2003foresthealthsurveywasusedtoiterativelydeterminetheupperandlowerlimitofΔRGI_2004valuesthatcorrespondedtoredattackdamage(asdifferencesinRGIvaluesspannedarangeofconditions,includingharvesting).Thesub-sampleofthefielddatawasusedtoassesstheomissionandcommissionerrorresultingfromthethreshold,withtheobjectivebeingtoadjustthethresholdtominimizetheseerrors.ThefinalthresholdwasthenappliedtotheΔRGI_2004valuestocreateadatalayeridentifyingredattackdamageinstandsA,B,andCin2004.Asimilarprocesswasthenusedtoidentifyredattackdamagethatmanifestedin2005(ΔRGI_2005)and2006(ΔRGI_2006).ThisprocessingprovidedlocationsofredattackdamageinstandsA,B,andC,in2004,2005,and2006.Accuracyoftheredattackdetectionwasassessedatthesegmentlevelusingfielddata.AgreementoccurredwhenasegmentcontainedredattackidentifiedbythresholdingtheΔRGI,andattackmeasuredinthefield.4.3.Imagesegmentation

Thedifferentviewinggeometryandilluminationconditions,combinedwiththepotentialforerrorinco-registeringmultipleyearsofimagerywithancillarygroundsurveys(Weberetal.,inpress),posedchallengesfortrackingthehealthstatusofindividualtreesthroughtime.Asaresult,analternativeapproachwasdevelopedthattrackedgroupsoftrees(orpixels)throughtime.Thiswasachievedbysegmentingthe2003QuickBirdmulti-spectralimage.Segmentsweregeneratedusingallfourmulti-spectralbandsineCognitionsoftware(DefiniensGmbH,München,Germany)withthefollowingparameters:equalweightsforallbands;scale=15;shape=0.9;color=0.1;

2734M.A.Wulderetal./RemoteSensingofEnvironment112(2008)2729–2740

Fig.2.QuickBird2.4multi-spectralimageryoverthestudyarea.

M.A.Wulderetal./RemoteSensingofEnvironment112(2008)2729–27402735

Fig.3.IllustrationofthedifferentviewinggeometryandilluminationconditionsunderwhichtheQuickBirdmulti-spectralimageryusedforthisstudywascollected.

Fig.4.Aschematicofhowmulti-dateQuickBirdmulti-spectralandpanchromaticimageryareusedtogeneratesegments,estimatesoftreecounts,anddetectredattackdamage;allnecessarycomponentsrequiredtoestimateagreenattack-to-redattackratio(G:R).

2736M.A.Wulderetal./RemoteSensingofEnvironment112(2008)2729–2740

Fig.5.SchematicofprocessusedtoestimateretrospectiveG:R.

compactness=1;smoothness=0).Forthethreestandsofinterest,theaveragesizeofthesegmentsrangedfrom347to405m2.StandAhad454segments;standB,596segments;andstandC,313segments.

4.4.Estimatingstemcounts

Recallthattheinformationneedassociatedwithmonitoringmountainpinebeetleattackatthetreelevelistheratioofthenumberoftreesinthegreenattackstage,tothenumberoftreesintheredattackstage.Inordertofulfillthisinformationneed,someapproximationofthenumberoftreesineachstandwasrequired.Alocalmaximafilter(Wulderetal.,2000)wasap-pliedtothepanchromaticQuickBirdimageforeachyeartoidentifyindividualtreecrowns.Pastresearchhasindicatedthatlocalmaximafilteringisbiasedtowardslargertreecrowns(Wulderetal.,2000,2004b),whichisnotaconcernforthisapplicationfortworeasons:first,theinformationneedrequiresarelativeestimateofgreenattack-to-redattacktrees(andchangestothatratioovertime)—notanexactcensusofalltreesinthestand;secondly,sincelargertreecrownsaretypi-callyassociatedwithlarger,maturetrees,anditistheselargertreesthataremostsusceptibletoattackbymountainpinebeetle(Shore&Safranyik,1992),theomissionerrorforindividualcrownsisconsideredacceptable.Theadvantageofthisap-proachoverothermethodsisthatitprovidesanindicationofthepopulation-at-riskofattackbymountainpinebeetle.AlocalmaximafilterwasappliedtoeachQBPANimageandtheimpactofdifferentviewinggeometryandilluminationcondi-tionsontreecountswereassessedusingAnalysisofVariance(ANOVA)ontreecountsbysegment.4.5.EstimatingG:R

Fig.4summarizeshowtheremotelysensedoutputsofredattackdamage,imagesegments,andestimatesofstemcountwerecombinedtoestimatetheG:Rratio.First,theremotelysensedoutput(calculatedfromΔRGI)isusedtodeterminetheareaofredattackdamagefoundwithinthesegment.Thepro-portionofredattackarearelativetototalsegmentareaisthen

calculated,andthatproportionisthenappliedtothesegmentstemcount(calculatedfromapplyingthelocalmaximafilter),resultinginanestimateofthenumberoftreesattackedinthesegmentforagivenyear.Forexample,ifΔRGIindicatesthat25%ofasegmentisredattackin2004andthesegmenthas100trees,then25treesareassumedtoberedattackinthesegmentin2004.Theredattacktreesinanygivenyearareassumedtohavebeengreenattackinthepreviousyear;hence,redattacktreesin2005areassumedtohavebeengreenattackin2004.Usingthisapproach,weareabletoretrospectivelyconstructaG:Rratioforagivenyear,aswellasestimatearateofpopu-lationincreaseformultipleyears(Fig.5).G:Rratioswerenotcalculatedforallsegmentsinthestand,rather,segmentswerestratifiedbasedontheirhealthstatusasdeterminedusingΔRGI(onlysegmentswithattackwereused).Fromthesearandomsampleofattackedsegmentsrepresentingapproximately15%ofthestandareawasusedincalculatingtheG:Rratios;thestandG:RratiosaretheaverageoftheG:Rineachoftherandomlyselectedsegmentswithineachstand.5.Results

5.1.Redattackdetection

RedattackdamageidentifiedfromtheQuickBirdimagerywascomparedtolocationsofredattackdamagerecordedbyfieldsurvey.Agreementbetweenthefieldsurveyandtheremotelysensedoutputsoccurredwhenasegmentcontainedagroundsurveypoint(tree)identifiedasredattack(2002–2005),andalsocontainedredattackintheremotesensingoutputs(2004–2006).TheresultsoftheaccuracyassessmentaresummarizedinTable3andindicateastrongcorrespondencebetweenthefieldsurveyandtheremotelysensedchangemapping,withtruepositiveaccu-raciesrangingfrom(standsAandC)to93%(standB).5.2.Estimationofstemcounts

Asdiscussedearlier,varyingviewinggeometryandillu-minationconditions(Table2),combinedwithcomplexforeststructure,resultinvariabilityinthenumberandpositionoflocalmaximafromoneimagetothenext.Thisshiftinatree'spositionovertimeprecludesmonitoringofindividualtreesthroughtime(usingdataofthisspatialresolution,withthesecollectionparameters).However,iftheimageissegmentedbasedonitsspectralproperties,theresultingsegments,repre-sentingsmallgroupsoftrees,couldbemonitoredthroughtimeandaccountforthisvariabilityinpositionalaccuracy.For

Table3

AccuracyassessmentofredattackdetectionStandTotal#offield

#ofsegmentsTruepositive95%

#ofsurveycontainingfieldrate(redconfidencesegmentspointssurveypointsattack)(%)intervalA4541768481%–94%B5966632079388%–95%C

313

67

37

75%–96%

M.A.Wulderetal./RemoteSensingofEnvironment112(2008)2729–27402737

monitoringpurposes,itisdesirabletoidentifythepopulation-at-riskinthefirstyearofthestudy(e.g.,throughabaselinecensus).Formonitoringredattackdamage,alocalmaximafilterwouldideallybeappliedtothefirstyearofimagerytoprovideanestimateofstemcountpersegment,withtheas-sumptionthatthestemcountwouldremainconstantovertime(dependingonthetimehorizonformonitoring;along-termstudy,N5–10yearswouldlikelyhavetoaccountfortreeregenerationandmortality).Toverifythisassumptionthatstemcountspersegmentwouldnotvarysignificantlythroughtime,weappliedalocalmaximafiltertoeachyearofimageryandgeneratedastemcountfromeachyear.AnANOVAonthetreecounts,bysegment,foundthattreecountsdidnotvarysignificantlyfromoneyeartothenext(Table4).Asaresult,weusedthelocalmaximaoutputfromthe2003imagetorepresentindividualstemcountswithineachsegment.TheaveragestemcountpersegmentinstandsA,B,andCin2003was22,18,and20,respectively.5.3.EstimationofG:R

Table5includestheestimatedG:Rforeachstandin2004and2005.ThestandG:RistheaverageoftheG:Rforeachofthesampledsegmentswithineachstand(withsamplesizesforeachstandindicatedinTable5).ForstandsAandB,theG:RisverysimilartotheG:Rderivedfromthefieldsurvey.InstandB,theremotelysensedestimateofG:RisgreaterthanthefieldsurveyedG:Rin2004,andremainsconstantin2005.Unfortu-nately,nofielddatawerecollectedinstandBin2004andsonocomparablefieldbasedG:Risavailablefor2005.Basedon2003fieldsurveyhowever,weknowthattheG:Rwas4in2003—indicatingthatthepopulationinthisstandhasbeensteadilyincreasinginpastyears(Carrolletal.,2006b).

Fig.6providesanexampleofsegment42414instandB.Thissegmenthadanareaof0.14haandcontainsapproximately73trees(asestimatedbytheLMfilterapproach).ΔRGI_2004indicatedthat1%ofthesegment,or1treehadredattackdamagein2004.By2005,themountainpinebeetlepopulationinstandBhadincreaseddramaticallyandΔRGI_2005indi-catedthat33%ofthesegmentarea,orapproximately24newtreeshadredattackdamage.TheG:Rfor2004wastherefore24:1.Thislevelofincreaseisextremeandtheoreticallyimpro-bable—emphasizingtheneedtoestimatetheaverageG:Roveralargerspatialunit(BritishColumbiaMinistryofForests,1995).In2006,therateofredattackexpansiondecreasedfrom

Table4

Totalnumberoftreesperstand,asestimatedbylocalmaximafilteringandfieldsurveyStand

LocalmaximafilterField-2003

200420052006ANOVAp-valuebased

(α=0.05,df=3)A10,12610,33610,74310,2320.86022,405B10,82010,95911,32610,9660.85821,092C

6379

04

6822

6381

0.805

15,385

TheANOVAwascalculatedusingeachyear'sstemcount/segment,forallsegmentsineachstand.

Table5

EstimatedG:RfromfieldsurveyversusremotesensingmethodsStand

TotalSamplesizeTotalFieldRemotesensingarea(ha)(%)

#of

G:RaG:R(ha)

segments(sample2004

2005

2004

2005

size)A16.72.4(14.3%)454(32)2:13:12.04:12.76:1B17.62.75(15.6%)596(39)4:1N/Ab5.08:14.80:1C

9.91.45(14.6%)313(18)

353.21:15.44:1

aFromdatapublishedin:Carroll,A.L.,Aukema,B.H.,Raffa,K.F.,Linton,D.A.,Smith,G.D.,Lindgren,B.S.2006b.Mountainpinebeetleoutbreakdevelopment:Theendemic–incipientepidemictransition.MPBIProject#1.03FinalReport.PacificForestryCentre,Victoria,BritishColumbia.22p.bNodatacollected.

2005,with11%ofthestandor8additionaltreeshavingredattackdamage.6.Discussion

Thedifferentviewinggeometryandilluminationconditionsforeachoftheimagesusedinthisstudy,asillustratedinFig.3and6,confoundedeffortstotrackindividualtreesthroughtime;imagesin2003and2004wereacquiredwithfairlysimilarparameters,whereasthe2005imagehasaverydifferentsolarazimuthandin-trackviewangle(Table2).Fortheestablish-mentofamonitoringsystemintendedtocoverlargeareas,differencesinviewingandilluminationgeometryisanissue,notonlytemporallybutalsospatially.Furthermore,thecostofacquiringhighspatialresolutiondataoverlargeareascanbeprohibitive,andincreasesoverheadwiththeprocessingrequire-mentsofmultiplescenesoveralargearea.Theauthorsthereforerecommendthattheapproachpresentedinthispaperbeimple-mentedthroughasamplingprotocol,wherebyanareaofinterestisstratifiedusingdataappropriateforidentifyingareaswithahighlikelihoodofbeetleattack(e.g.,leadingspecies,age,elevation,Landsatredattackmaps,etcetera).Satellitesampleplotscouldthenbeestablished(8kmby8km),andQuickBirdorsomeotherhighspatialresolutiondatasourceacquiredovertheseareas.Further,itshouldbestressed,asisevidentinFig.5,thatthegeometricmatchbetweenscenesisgood,itistheinteractionofthesun/surface/sensorgeometrywiththeforeststructuralconditionsthatresultsintheuniquemanifestationoftheforestandthesubsequentimageprocessingandtree-to-treematchingdifficulties(Wulderetal.,2004a).

Forlocalscalecharacterizationwithhighspatialresolutiondata,highlevelsofprecisionarerequiredtoaccuratelyco-registerimages(Stow,1999)andsimilarly,toco-registerimageoutputswithfieldsurveys(Weber,2006;Weberetal.,inpress).Othershavesuggestedtakingimageryintothefieldtoensurethatsurveydataandremotelysenseddataareproperlyalignedasthedataarecollected(Sawayaetal.,2003).Unfortunately,fieldandimagedataarenotoftencollectedinthismanner,withfielddatasometimesoriginatingfromadifferentproject(andthereforecollectedatadifferenttime,withdifferentobjectives).Thereinliestheappealofobject-basedapproaches;general-izationtoasmaller(andoftenmeaningful)spatialunitallows

2738M.A.Wulderetal./RemoteSensingofEnvironment112(2008)2729–2740

Fig.6.StandB,segment42414(yellow);time-seriesofQuickBirdmulti-spectralimagery.Segmentsweregeneratedusingthe2003QuickBirdmulti-spectralimage.

positionalambiguitytobeaccountedfor.Inthecaseofhighspatialresolutiondata,thesesegmentsoftenrepresentindivi-dualtreesorsmallgroupsoftrees.Inthecaseofmountainpinebeetle,thetimingofdatacollectionisparticularlyproblematic,duetothespecificbio-windowwithinwhichsurveydatacanbecollected(Wulderetal.,2006b).

ItmustalsobereiteratedthattherearedatasourcesandproceduresthatareusedoperationallytoaddresstheG:Rinfor-mationneed.Coarselevelaerialoverviewsurveysareconductedprovince-wideonanannualbasis.Thelocationsofdamageidentifiedfromthesesurveysarethenusedtodirectmoredetailedaerialsurveys,whichinturnareusedtodirectthelocationofgroundsurveys,suchasthoseusedtogenerateestimatesofG:R.Furthermore,theseestimatescanoftenbegeneratedrelativelyquickly,whereastheapproachpresentedhereinisprimarilyretrospective,sinceitrequiresatleastthreedatesofimagerytoestimateG:R.Thisapproachdoeshoweverprovidesuniquebe-nefitsforareaswhere,forwhateverreason,therewasatemporalgapindatacollection,fieldcrewsareunabletoaccessthesite,orthereisaneedtoassesstheefficacyofpreviouslyimplementedmanagementpractices.Also,thisapproachisaconsistentandrepeatablemethodforgeneratingG:Roverlargerareas,whereeconomiesofscalewouldmakethisapproachmoreaffordable

thanthedeploymentoffieldcrews.Theauthorsarenotproposingthatthisdatawouldsupplantothersources,butrathercouldsupplementandaugmenttheseotherpiecesofinformation.Whileitwaspossibletosuccessfullyacquiremulti-yearhighspatialresolutiondataforthisstudyarea,therearerisksasso-ciatedwiththeassumptionthatqualityimagerywillbeavailableforthespecificbio-windowdesired.Commerciallyavailablehighresolutionsatellitedataarenotroutinelyarchived,sosensorsmustbetaskedtoacquiredataoverareaofinterest.Asdemandforthisdataincreases,itbecomesincreasinglydifficulttogetdatawhenandwhereitisrequired.Typically,animageordermustbeputinmonthsinadvance.Particularrequestsmaybemadetolimitviewingangles,solarconditions(collectiondaterange),cloudcover,mosaickingofmultipleover-passes,etcetera.Whilethesetypesofparametersmaybecustomized,anylimitationstotheconditionsunderwhichanimagemaybecollecteddecreasesthelikelihoodthatimagerywillbesuc-cessfullyobtained.7.Conclusion

Inthispaperwepresentedanapproachforestimatingratesofmountainpinebeetlepopulationchange,viaG:R,withhigh

M.A.Wulderetal./RemoteSensingofEnvironment112(2008)2729–27402739

spatialresolutionremotelysenseddata.TheaverageG:Risanimportantsourceofinformationforforestmanagerswhoneedtomonitorchangesinbeetlepopulationsovertime.Usingboththemulti-spectralandpanchromaticbandsfrommultipledatesofQuickBirddata,wewereabletocharacterizeredattackdamageandestimatestemcounts.Bycombiningthesetwopiecesofinformation,thenumberofredattacktreescouldbeestimatedforanygivenyear,andthenbybackcastingtheseestimates,G:Rratioscouldbegeneratedretrospectively.Thismethodalsogivesanestimateofthepopulation-at-risk(totalstems),informationthatistypicallynotcollectedwhenconduc-tinggroundsurveys.Thechallengesandrisksassociatedwithusinghighspatialresolutionremotelysenseddatatomonitorforestdamageovertimehavealsobeenenumerated.Theap-proachpresentedwouldallowG:Rtobegeneratedoverlargeareas,providingasynopticviewofspatialvariationinG:Rwhichtheninturncouldbeusedforstrategicplanningofland-scapelevelbeetlemanagementpracticesand/ortoassesstheefficacyofstrategiesandmanagementpracticesimplemented.Acknowledgements

WethankDr.AllanCarroll,oftheCanadianForestService,forprovidingthefielddatausedforthisproject.Dr.GeoffHay,oftheUniversityofCalgary,isthankedforvaluablediscussionsthatfortifiedourprojectdesign.Theinsightfulandconstructivecom-mentsofthereviewersareverymuchappreciated.ElementsofthisprojectwerefundedbytheGovernmentofCanadathroughtheMountainPineBeetleInitiative,a6-year,$40millionprogramadministeredbytheNaturalResourcesCanada,CanadianForestService(http://mpb.cfs.nrcan.gc.ca).References

Allen,T.R.,&Kupfer,J.A.(2001).Spectralresponseandspatialpatternof

Fraserfirmortalityandregeneration,GreatSmokyMountains,USA.PlantEcology,156,59−74.

Asner,G.P.,&Warner,A.S.(2003).CanopyshadowsinIKONOSsatellite

observationsoftropicalforestsandsavannas.RemoteSensingofEnviron-ment,87,521−533.

AmericanSocietyforTesting&Materials(ASTM).(2000).Standard

extraterrestrialSolarspectrumreference,E-490-00.(AccessedonlineAugust14,2007).http://rredc.nrel.gov/solar/spectra/am0/ASTM2000.htmlAmman,G.D.(1982).Themountainpinebeetle—Identification,biology,

causesofoutbreaks,andentomologicalresearchneeds.InD.M.Shrimpton(Ed.),ProceedingsoftheJointCanada/USAWorkshoponmountainpinebeetlerelatedproblemsinWesternNorthAmerica,EnvironmentCanada,CanadianForestryService,PacificForestResearchCentre,Victoria,InformationReportBC-X-230(pp.7−12).87pp.

Birk,R.J.,Stanley,T.,Snyder,G.I.,Hennig,T.A.,Fladeland,M.M.,&

Policelli,F.(2003).Governmentprogramsforresearchandoperationalusesofcommercialremotesensingdata.RemoteSensingofEnvironment,88,3−16.

BritishColumbiaMinistryofEnvironment,Lands,andParks.(1992)British

ColumbiaSpecificationsandGuidelinesforGeomatics.ContentSeriesVolume3:Digitalbaselinemappingat1:20000.GeographicDataBC,Victoria,BritishColumbia.(AccessedonlineAugust14,2007).http://ilmbwww.gov.bc.ca/bmgs/pba/trim/specs/specs20.pdf

BritishColumbiaMinistryofForests.(1995).Barkbeetlemanagement

guidebook(ForestPracticesCode).Victoria,BritishColumbia:ForestPracticesBranch.45pp.Carroll,A.L.,Taylor,S.W.,Régnière,J.,&Safranyik,L.(2004).Effectsof

climateandclimatechangeonthemountainpinebeetle.InT.L.Shore,J.E.Brooks,&J.E.Stone(Eds.),Proceedingsofthemountainpinebeetlesymposium:Challengesandsolution.October30–31,2003,Kelowna,BritishColumbia,Canada.CanadianForestService,PacificForestryCentre,InformationReportBC-X-399(pp.223−232).298pp.

Carroll,A.L.,Régnière,J.,Logan,J.A.,Taylor,S.W.,Bentz,B.J.,&Powell,J.A.

(2006a).Impactsofclimatechangeonrangeexpansionbythemountainpinebeetle.MountainPineBeetleInitiativeWorkingPaper2006-14.20p.

Carroll,A.L.,Aukema,B.H.,Raffa,K.F.,Linton,D.A.,Smith,G.D.,&

Lindgren,B.S.(2006b).Mountainpinebeetleoutbreakdevelopment:Theendemic–incipientepidemictransition.MountainPineBeetleInitiativeProject#1.03FinalReport.22pp.

Clarke,D.B.,Castro,C.S.,Alvarado,L.D.A.,&Read,J.M.(2004).

Quantifyingmortalityoftropicalrainforesttreesusinghigh-spatialresolutionsatellitedata.EcologyLetters,7,52−59.

Coops,N.C.,Johnson,M.,Wulder,M.A.,&White,J.C.(2006a).Assessment

ofQuickBirdhighspatialresolutionimagerytodetectredattackdamageduetomountainpinebeetleinfestation.RemoteSensingofEnvironment,103,67−80.

Coops,N.C.,Wulder,M.A.,&White,J.C.(2006b).Integratingremotely

sensedandancillarydatasourcestocharacterizeamountainpinebeetleinfestation.RemoteSensingofEnvironment,105,83−97.

Desclée,B.,Bogaert,P.,&Defourny,P.(2006).Forestchangedetectionby

statisticalobject-basedmethod.RemoteSensingofEnvironment,102,1−11.DigitalGlobeInc.(2007).QuickBirdimageryproducts:Productguide.Revision

4.7.3.(AccessedonlineAugust14,2007).http://www.digitalglobe.com/downloads/QuickBird%20Imagery%20Products%20-%20Product%20Guide.pdf

Eng,M.(2005).Mountainpinebeetleprojections:Answeringquestionsor

makingmoonshine.Forum,12,19.

Gong,P.,&Xu,B.(2003).Remotesensingofforestsovertime:Changetypes,

methodsandopportunities.(Chapter11).InM.A.Wulder&S.E.Franklin(Eds.),Remotesensingofforestenvironments:Conceptsandcasestudies(pp.301−333).Boston:Kluwer.519pp.

Henigman,J.,Ebata,T.,Allen,E.,Holt,J.,&Pollard,A.(Editors).(1999).Field

guidetoforestdamageinBritishColumbia,2ndEdition.CanadianForestServiceandBritishColumbiaMinistryofForests,Victoria,BritishColumbia.FRDAIIRep.No.17.348p.

Im,J.,&Jensen,J.R.(2005).Achangedetectionmodelbasedonneighborhood

correlationimageanalysisanddecisiontreeclassification.RemoteSensingofEnvironment,99,326−340.

Johansen,K.,&Phinn,S.(2006).Mappingstructuralparametersandspecies

compositionofriparianvegetationusingIKONOSandLandsatETM+datainAustralianTropicalSavannahs.PhotogrammetricEngineeringandRemoteSensing,72,71−80.

Key,T.,Warner,T.A.,McGraw,J.B.,&Fajvan,M.A.(2001).Acomparisonof

multispectralandmultitemporalinformationinhighspatialresolutionimageryforclassificationofindividualtreespeciesinatemperatehardwoodforest.RemoteSensingofEnvironment,75,100−112.

Lambin,E.F.,&Strahler,A.H.(1994).Changevectoranalysisinmultitemporal

space:Atooltodetectandcategorizeland-coverchangeprocessesusinghightemporalresolutionsatellitedata.RemoteSensingofEnvironment,48,231−244.

Liu,D.,Kelly,M.,&Gong,P.(2006).Aspatial–temporalapproachto

monitoringforestdiseasespreadusingmulti-temporalhighspatialresolutionimagery.RemoteSensingofEnvironment,101,167−180.

Logan,J.A.,&Powell,J.A.(2001).Ghostforests,globalwarming,andthe

mountainpinebeetle(Coleoptera:Scolytidae).AmericanEntomologist,47,160−173.

Mas,J.F.(1999).Monitoringland-coverchanges:Acomparisonofchange

detectiontechniques.InternationalJournalofRemoteSensing,20,139−152.

Möller,M.,Lymburner,L.,&Volk,M.(2007).Thecomparisonindex:Atool

forassessingtheaccuracyofimagesegmentation.InternationalJournalofAppliedEarthObservationandGeoinformation,9,311−321.

Pouliot,D.A.,King,D.J.,Bell,F.W.,&Pitt,D.G.(2002).Automatedtree

crowndetectionanddelineationinhigh-resolutiondigitalcameraimageryof

2740M.A.Wulderetal./RemoteSensingofEnvironment112(2008)2729–2740

coniferousforestregeneration.RemoteSensingofEnvironment,82,322−334.

Safranyik,L.2004.Mountainpinebeetleepidemiologyinlodgepolepine.Pages

33–40inMountainpinebeetlesymposium:Challengesandsolutions,October30–31,2003.Shore,T.L.,Brooks,J.E.,Stone,J.E.(editors).Kelowna,BritishColumbia,Canada.NaturalResourcesCanada,CanadianForestService,PacificForestryCentre,Victoria,BritishColumbia,Infor-mationReportBC-X-399.298p.

Safranyik,L.,&Carroll,A.L.(2006).Thebiologyandepidemiologyofthe

mountainpinebeetleinlodgepolepineforests.Pages3–66(Chapter1)inSafranyik,L.,&Wilson,W.R.(Editors).Themountainpinebeetle:Asynthesisofbiology,management,andimpactsonlodgepolepine.NaturalResourcesCanada,CanadianForestService,PacificForestryCentre,Victoria,BritishColumbia,304p.

Sawaya,K.E.,Olmanson,L.G.,Heinert,N.J.,Brezonik,P.L.,&Bauer,M.E.

(2003).Extendingsatelliteremotesensingtolocalscales:Landandwaterresourcemonitoringusinghigh-resolutionimagery.RemoteSensingofEnvironment,88,144−156.

Shore,T.L.,&Safranyik,L.(1992).Susceptibilityandriskratingsystemsfor

themountainpinebeetleinlodgepolepinestands.ForestryCanada,PacificandYukonRegion,PacificForestryCentre,Victoria,BritishColumbia.InformationReportBC-X-336.17p.

Skakun,R.S.,Wulder,M.A.,&Franklin,S.E.(2003).Sensitivityofthe

thematicmapperenhancedwetnessdifferenceindextodetectmountainpinebeetleredattackdamage.RemoteSensingofEnvironment,86,433−443.Smith,M.J.,&Wise,S.M.(2007).Problemsofbiasinmappinglinear

landformsformsatelliteimagery.InternationalJournalofAppliedEarthObservationandGeoinformation,9,65−78.

Stow,D.A.(1999).Reducingtheeffectsofmisregistrationonpixel-level

changedetection.InternationalJournalofRemoteSensing,20,2477−2483.Taylor,S.W.,&Carroll,A.L.(2004).Disturbance,forestage,andmountain

pinebeetleoutbreakdynamicsinBC:Ahistoricalperspective.Pages41–51inShore,T.L.,Brooks,J.E.,&Stone,J.E.(Editors).Proceedingsofthemountainpinebeetlesymposium:Challengesandsolution.October30–31,2003,Kelowna,BritishColumbia,Canada.CanadianForestService,PacificForestryCentre,InformationReportBC-X-399.298p.

Townshend,J.R.G.,Justice,C.O.,Gurney,C.,&McManus,J.(1992).The

impactofmisregistrationonchangedetection.IEEETransactionsonGeo-scienceandRemoteSensing,30,1056−1060.

Weber,K.T.(2006).Challengesofintegratinggeospatialtechnologiesintorangelandresearchandmanagement.RangelandEcologyandManagement,59,38−43.

Weber,K.T.,Théau,J.T.,&Serr,K.(inpress).Effectofcoregistrationerroronpatchytargetdetectionusinghigh-resolutionimagery.RemoteSensingofEnvironment.doi:10.1016/j.rse.2007.06.016

Westfall,J.(2007).2006summaryofforesthealthconditionsinBritishColumbia.ForestPracticesBranch,BritishColumbiaMinistryofForestsandRange,73pp.

White,J.C.,Wulder,M.A.,Brooks,D.,Reich,R.,&Wheate,R.D.(2005).Detectionofredattackstagemountainpinebeetleinfestationwithhighspatialresolutionsatelliteimagery.RemoteSensingofEnvironment,96,340−351.

Wu,J.,Wang,D.,&Bauer,M.E.(2005).Image-basedatmosphericcorrectionofQuickBirdimageryofMinnesotacropland.RemoteSensingofEnvironment,99,315−325.

Wulder,M.,Niemann,K.O.,&Goodenough,D.G.(2000).Localmaximafilteringfortheextractionoftreelocationsandbasalareafromhighspatialresolutionimagery.RemoteSensingofEnvironment,73,103−114.

Wulder,M.A.,Hall,R.J.,Coops,N.C.,&Franklin,N.C.(2004a).Highspatialresolutionremotelysenseddataforecosystemcharacterization.BioScience,54,511−521.

Wulder,M.A.,White,J.C.,Niemann,K.O.,&Nelson,T.(2004b).Comparisonofairborneandsatellitehighspatialresolutiondatafortheidentificationofindividualtreeswithlocalmaximafiltering.InternationalJournalofRemoteSensing,25,2225−2232.

Wulder,M.A.,Dymond,C.C.,White,J.C.,&Erickson,B.(2006a).Detection,mapping,andmonitoringofthemountainpinebeetle.Pages123–154(Chapter5)inSafranyik,L.,&Wilson,W.R.(Editors).Themountainpinebeetle:Asynthesisofbiology,management,andimpactsonlodgepolepine.NaturalResourcesCanada,CanadianForestService,PacificForestryCentre,Victoria,BritishColumbia,304p.

Wulder,M.A.,Dymond,C.C.,White,J.C.,Leckie,D.G.,&Carroll,A.L.(2006b).Surveyingmountainpinebeetledamageofforests:Areviewofremotesensingopportunities.ForestEcologyandManagement,221,27−41.

Wulder,M.A.,White,J.C.,Bentz,B.,Alvarez,M.F.,&Coops,N.C.(2006c).Estimatingtheprobabilityofmountainpinebeetleredattackdamage.Re-moteSensingofEnvironment,101,150−166.