Mapping burn severity in a disease-impacted forest landscape using Landsat and MASTER imagery

(1)

ContentslistsavailableatScienceDirect

International

Journal

of

Applied

Earth

Observation

and

Geoinformation

j ou rn a l h o m ep a g e :w w w . e l s e v i e r . c o m / l o c a t e / j a g

Mapping

burn

severity

in

a

disease-impacted

forest

landscape

using

Landsat

and

MASTER

imagery

Gang

Chen

a,∗

_,

_Margaret

_R.

_Metz

b,c

_,

_David

_M.

_Rizzo

c

_,

_Ross

_K.

_Meentemeyer

d,e

a_Department_of_Geography_and_Earth_Sciences,_University_of_North_Carolina_at_Charlotte,₉₂₀₁_University_City_Blvd,_Charlotte,_NC_28223,_USA b_Department_of_Biology,_Lewis_&_Clark_College,₀₆₁₅_S.W._Palatine_Hill_Road,_MSC_53,_Portland,_OR_97219,_USA

c_Department_of_Plant_Pathology,_University_of_California,₁_Shields_Ave,_Davis,_CA_95616,_USA

d_Center_for_Geospatial_Analytics,_North_Carolina_State_University,₃₁₂₀_Jordan_Hall,_Raleigh,_NC_27695,_USA

e_Department_of_Forestry_and_{Environmental}_Resources,_North_Carolina_State_University,₃₁₂₀_Jordan_Hall,_Raleigh,_NC_27695,_USA

a

r

t

i

c

l

e

i

n

f

o

Articlehistory:

Received20December2014 Accepted7April2015

Keywords: Burnseverity Forestﬁre Diseaseinvasion MASTER Landsat Suddenoakdeath Interactingdisturbances

a

b

s

t

r

a

c

t

Globalenvironmentalchangehasincreasedforestvulnerabilitytotheoccurrenceofinteracting distur-bances,includingwildfiresandinvasivediseases.Mappingpost-fireburnseverityinadisease-affected forestoftenfaceschallengesbecauseburnedandinfestedtreesmayexhibitahighsimilarityinspectral reflectance.Inthisstudy,wecombined(pre-andpost-fire)Landsatimageryand(post-fire)high-spectral resolutionairborne MASTERdata [MODIS (moderateresolution imagingspectroradiometer)/ASTER (advancedspacebornethermalemissionandreflectionradiometer)]tomapburnseverityinaCalifornia coastalforestenvironment,whereanon-nativeforestdiseasesuddenoakdeath(SOD)wascausing sub-stantialtreemortality.ResultsshowedthattheuseofLandsatplusMASTERbundleperformedbetterthan usingtheindividualsensorsinmostoftheevaluatedforeststratafromgroundtocanopylayers(i.e., sub-strate,shrubs,intermediate-sizedtrees,dominanttreesandaverage),withthebestmodelperformance achievedatthedominanttreelayer.Themidtothermalinfraredspectralbands(3.0–12.5␮m)from MAS-TERwerefoundtoaugmentLandsat’svisibletoshortwaveinfraredbandsinburnseverityassessment. Wealsofoundthatinfestedanduninfestedforestssimilarlyexperiencedmoderatetohighdegreesof burnswhereCBI(compositeburnindex)valueswerehigherthan1.However,differencesoccurredinthe regionswithlowburnseverity(CBIvalueslowerthan1),whereuninfestedstandsrevealedamuchlower burneffectthanthatininfestedstands,possiblyduetotheirhigherresiliencetosmallfiredisturbances asaresultofhigherleafwatercontent.

PublishedbyElsevierB.V.

1. Introduction

Forestsplayakeyroleasglobalterrestrialcarbonsinks respon-sibleforanannualuptakeofoveronequarteroftheanthropogenic carbon dioxide (CO2) from the atmosphere (Pan et al., 2012). Comparedtothenetcarbonsequestrationthroughtreegrowth, carbon loss may beoccurring at a faster pace through natural disturbancesincludingwildfire(Kurzetal.,2008a;Asner,2013). Recently,increasedfireactivitieshavebeenreportedinmajor for-estbiomes,suchasBrazilianAmazon,Canadianborealandwestern USA(Carcailletetal.,2001;Westerlingetal.,2006;Mortonetal., 2013).Intheshortterm,forestfiresdirectlytransformlivingand deadorganicmatteratorabovethesoilsurfacetocharredand blackenedresidues(Kokalyetal.,2007), whichinthelongterm

∗Correspondingauthor.Tel.:+17046875947;fax:+17046875966. E-mailaddress:[email protected](G.Chen).

mayaffectthestructureandfunctionoftheecosystem,and influ-encethespatialpatternsofecologicalsuccession (Turneretal., 1998;Metzetal.,2012,2013).Hence,itiscriticaltounderstand thecombustiondegreeoforganicmattersandtheimpactofheat onsoilchemicalpropertiesinforestsimmediatelyfollowingfire events,wherethepracticeistypicallyreferredtoasburnseverity assessment(Nearyetal.,2005;KeyandBenson,2006;Lentileetal., 2006;Kokalyetal.,2007;Keeley,2009).Overthelastthreedecades, remotesensinghasproveneffectiveinassessingtheimpactsoffire onforestecosystemsatlocal,regionalandcontinentalscales(Hall etal.,1980;Milne,1986;Jakubauskasetal.,1990;Whiteetal.,1996; HudakandBrockett,2004;Lentileetal.,2006;Veraverbekeetal., 2012).Whilethedetectionoffireextents(“where”)usingremote sensingisrelativelywellestablished,recentresearcheffortshave advanced themeasurement of “howseverely” wildfires change post-disturbanceforestlandscapes(i.e.,burnseverity)by capitaliz-ingonthevariationinspectralreflectancefromburnedvegetation andsoilshortlyafterthefireevents(MillerandYool,2002;Epting

(2)

etal.,2005;Royetal.,2006;Lentileetal.,2006;MillerandThode, 2007;Harrisetal.,2011;Quintanoetal.,2013).Here,weadopted thedefinitionofburnseverityas“thedegreetowhichan ecosys-temhaschangedowingtothefire”(Lentileetal.,2006).Statistical regressionandimageclassificationaretypicalapproachesusedto linkspectralreflectancefromaburnedforestwithafield-measured burnseverityindex,suchascompositeburnindex(CBI)ranging fromzerotothreeorlowtohigh(KeyandBenson,2006;Hultquist etal.,2014).

Althoughtherehasbeena growingbodyofliteratureonthe application of remote sensing to assess burn severity across a diversity of environments [see reviewsby Lentile et al. (2006) andKeeley(2009)],fewstudieshavediscussedtheuncertaintyin mappingburnseverityinaforestthatisalsosubjecttosevere dis-turbancescausedbyinvasivepestsorpathogens.Whileincreased fire events have been observed mainly due to the impact of warmingfostereddrought,theintensityandfrequencyofinsect invasionsanddiseaseepidemicsarelikelytobeaccelerated result-ingfromglobal trade(Bergotet al.,2004; Wulder etal., 2006; Kurzetal.,2008b;Lamsaletal.,2011;Olssonetal.,2012). Stud-ieshave foundthat it is becoming morefrequent that wildfire affectsaforestinvadedbyanexoticinsectordisease(Jenkinsetal., 2008;Metzetal.,2013).Sincetheoutbreaksofinsectordisease coulddramaticallyincreasehostfuelabundanceandexhibit dif-ferentpatterns(duetopatchydispersalbythepest)(Meentemeyer etal.,2012),accurateassessmentofforestburnseverityrequires acomprehensiveinvestigationofhownon-nativespecies, includ-ing infectious disease, affect fire behavior. However, relatively littleconsiderationhasbeengiventotherolethatinvasivepests or pathogens play in shaping spatial heterogeneities of burn severity.

Todate, remotely senseddata used in themajority of burn severitystudieshavebeenfromtraditionalsensors,suchas Land-satand SPOT(SatellitePourl’ObservationdelaTerre)(e.g.,Roy etal., 2006;Fox et al.,2008).Althoughpromising resultswere reportedwiththesebroadbandsensorsusingtheVNIR-SWIR (vis-ibletoshortwaveinfrared,0.4–3.0␮m)spectralregion,itremains unclearwhethersuchdatasets canprovidesatisfactory perfor-mancemappingburnseverityindiseasedforests,becausepost-fire spectral reflectance is complicated by both the degree of burn severityanddisease-causedtreemortality.Thedisturbancesoffire andforestdiseasemaychangespectralreflectanceinsimilarways. Recently,researchersfoundthatsensorswithanabilitytoacquire imageryofhigherspectralresolutionandbroaderspectralrange mayprovidegreaterinsightsintotheimpactof firesonforests (Wagtendonket al.,2004Wagtendonk etal.,2004Kokaly etal., 2007;Veraverbekeetal.,2012).MASTER[MODIS(moderate res-olutionimagingspectroradiometer)/ASTER(advancedspaceborne thermalemissionandreflectionradiometer)]airbornesimulatoris anexampleofasensorcapableofcollecting50-bandimagery cov-eringboththeVNIR SWIRandMIR TIR(midtothermalinfrared, 3.0–12.5␮m)spectralranges.Hence,MASTERmayhavethe poten-tialtorecognizedisturbancerelatedsubtlevariationsinspectral reflectance,andaugmentburnseverityassessmentsfrom broad-bandLandsatdataindiseasedforests.

Capitalizing onthe emerging infectious disease sudden oak death(SOD)ascasestudyofforesthealthdeclinepriortofire,we asktwoquestionsthataddressthepotentialforcombingMASTER airborneandLandsatTMdatatoassesstheimpactofapre-fire bioticdisturbanceonforestburnseverity:(i)whatisthe perfor-manceofcombiningMASTERandLandsatTMdatainburnseverity assessmentscomparedtousingdatafromindividualsensors?(ii) DoesSODtreemortalityinfluenceburnseveritymeasuredusing CBIinvariousforeststratafromgroundtocanopylayers(i.e., sub-strate,herbs,shrubs,intermediatetrees,dominanttreesandthe average)?

Fig.1.StudyarealocatedintheBigSurecoregiononthewesternﬂankoftheSanta LuciaMountainsinCalifornia.TheMASTERimageisfromacolorcompositeusing bands5(red),3(green)and1(blue).The2007and2008LandsatTMimagesarefrom acolorcompositeusingbands3(red),2(green)and1(blue).(Forinterpretationof thereferencestocolorinthisﬁgurelegend,thereaderisreferredtothewebversion ofthisarticle.)

2. Methods

2.1. Studyarea

(3)

Table1

Summarystatistics(i.e.,minimum,maximum,mean,medianandstandarddeviation)ofallsixtypesofCBI(compositeburnindex)scores(i.e.,substratelayer,herblayer, shrublayerintermediate-sizedtreelayer,dominanttreelayer,andaverage).

TypeofCBIscore Minimum Maximum Mean Median Standarddeviation

Substratelayer 0.15 2.66 1.79 1.82 0.48

Herblayer 0.50 3.00 2.71 3.00 0.53

Shrublayer 0.00 3.00 2.40 2.50 0.45

Intermediate-sizedtreelayer 0.38 3.00 1.53 1.50 0.59

Dominanttreelayer 0.38 3.00 1.31 1.25 0.67

Average 0.58 2.93 1.96 1.96 0.40

Fig.2. Anareaaffectedbypre-ﬁreSOD,withmanylargedeadtanoaksandpiles oftanoaklogs(topimage),andseverelydamagedbythe2008BasinComplexFire (bottomimage).

andrainfallconditionsforpathogensporulationfollowedby

devel-opmentofstemcankersinsusceptiblehoststanoakandsomeoaks,

whichmaygirdleatreeoverseveralyearsormakethetreemore

susceptibletoattackbyotherpathogensorinsects.In2008,major

wildﬁresoccurredfortheﬁrsttimeinforestsaffectedbySOD.The

largestﬁre,theBasinComplexFire,ignitedfromadrylightning

storminlateJuneandburnedover95,000ha(USDAForestService,

2008).Ourstudysystempresentsauniqueopportunitytostudy theimpactsofadestructiveforestdiseaseonalargewildﬁre.Fig.2 showsanareaaffectedbypre-ﬁreSOD,andseverelydamagedby the2008BasinComplexFire.

2.2. Fielddata

Anetworkoflong-termSOD monitoringplots(500m2 _each) inBigSurwasestablishedin2006and2007tounderstandthe responsesofforestcommunities(e.g.,hostmortality)tothe inva-sionofSOD(Meentemeyeretal.,2008).InSeptemberandOctober 2008,immediatelyfollowingthecontainmentoftheBasinfire,we revisitedatotalof61plotstoassessburnseveritiesinbothinfested anduninfestedforests.Amongtheserandomlysampledplots,42 wereinfectedbySOD,while19werenot(Metzetal.,2011). Plot-levelfireseveritywasratedusingthecompositeburnindex(CBI),a ratingfromzerotothreerepresentingthedamagetomultipleforest strataasmeasuredacrosstheentireplot.Specifically,wefollowed KeyandBenson(2006)byusingastandardCBIdataformto mea-surethedepthofashdeposition,charredandintactduff,charred soil,soilwithdestroyedsoilstructure,andsoilthathadbeenfiredto

aredcolor,withdetailsrevealedinMetzetal.(2011).Thefive for-eststrataevaluatedforburnseverityincluded:(1)substratelayer, measuredaschangestocoarsewoodydebris,soil,duff,andleaf litter;(2)herblayer,changesorresponsesofvegetationlessthan 1m;(3)shrublayer,changesofvegetationhigherthan1mbutless than5m;(4)intermediate-sizedtreelayer,anytreeshigherthan 5mbutstandingunderthedominanttrees;and(5)dominanttree layer(Metzetal.,2011).WeassessedCBIforthefiveforeststrata andcalculatedameanscorerepresentinganoverallassessmentof fireseverity(i.e.,averageCBI).Table1showstherange(minimum andmaximum),mean,medianandstandarddeviationofallthesix typesofCBI.

2.3. MASTERimagery

TheMASTER airbornesimulator wasoriginallydeveloped to collectASTER-likeandMODIS-likedatasetsforsupportingthe vali-dationofASTERandMODISalgorithms(Hooketal.,2001).MASTER isalsosimilartooneofthetwoinstrumentsonboardthenext generation satelliteHyperspectral Infrared Imager (HyspIRI)that offersmultitemporalnarrowbandwidthVSWIR MTIRdatawith globalcoverage(NASA,2015).Clearly,studiesthatemployMASTER airbornesimulatorofferpotentialtofacilitatefutureapplications ofthespaceborne HyspIRI acrossregionaltocontinental scales. TheimagerycollectedbyMASTERincluded50bandscoveringthe visibletoshortwaveinfrared(VNIR SWIR:bands1–25),themid infrared(MIR:bands26–40),andthethermalinfrared(TIR:bands 41–50)regions,withmoredetailsfoundinHooketal.(2001).

Inthisstudy,12transectsofMASTERimagescoveringtheBasin ComplexFireareawerecollectedbyNASA-JPLfrom12:14to14:00 pmonAugust26,2008(approximatelyonemonthaftertheﬁre perimeterwascontained).Thesensorwaskeptinasmallrange ofaltitudes,andhadanaveragegroundresolutionof4.0m.We obtainedtheMASTERdata asa Level1Bproductstandard that wasradiometricallycalibratedandgeo-locatedandencapsulated inhierarchicaldataformat.Thedimensionofeachimagescene was716(thenumber ofpixels)bythenumberofscanlinesby 50(numberofMASTERbands).Thescanlinesofthese12images rangedfrom2639to7539,andthetotalnumberofscanlineswas 71,033.Alltheimageswereprocessedtomitigatetheview-angle brightnessgradienteffect,whichwasfollowedbyradiometricand topographiccorrections(Chenetal.,2015).Themosaicked MAS-TERimagewasresampledto30m(facilitatingthecomparisonwith LandsatTMimagery),andco-registeredwiththeLandsatTM2007 image(seemoredetailsaboutLandsatimagecollectionand pro-cessinginSection2.4)yieldingaRMSE(rootmeansquarederror) of0.46pixels.

2.4. Landsatimagery

(4)

Table2

Independentvariablesusedinburnseveritymodeling(NIR:nearinfrared,SSWIR:shortershortwaveinfrared,lSWIR:longershortwaveinfrared,MIR:midinfrared,E: emissivity).

Variabletype Variablename Description

Spectralresponse–MASTER SRiM SpectralresponseoftheMASTER’sithband(i=1,...50)

Spectralindex–MASTER NBRM MASTERnormalizedburnratio=(NIR−lSWIR)/(NIR+lSWIR)(Keyetal.,2006)

NDVIM MASTERnormalizeddifferencevegetationindex=(NIR−Red)/(NIR+Red)(Tucker,1979) NSEv1M NIR−SWIR-emissivityversion1=(NIR−lSWIR)/(NIR+lSWIR)×E(Veraverbekeetal.,2011)

NSEv2M NIR−SWIR-emissivityversion2=(NIR−(lSWIR+E))/(NIR+lSWIR+E)(Veraverbekeetal.,2011)

VI3M Vegetationindex3=(NIR−MIR)/(NIR+MIR)(KaufmanandRehmer,1994)

SMIM SWIR−MIRindex=(SSWIR−MIR)/(SSWIR+MIR)(Veraverbekeetal.,2012)

Spectralresponse–Landsat SRiL Spectralresponseofthe2008Landsat’sithband(i=1,...5,7)

Spectralindex–Landsat NBRL Landsat2008NBR

dNBRL DifferencedNBR=Landsat2007NBR–Landsat2008NBR(Wagtendonketal.,2004)

NDVIL Landsat2008NDVI

dNDVIL DifferencedNDVI=Landsat2007NDVI–Landsat2008NDVI(Xiaoetal.,2002)

Red:MASTERband5,Landsatband4;NIR:MASTERband9,Landsatband4;SSWIR:MASTERband12,Landsatband7;lSWIR:MASTERband24,Landsatband7;MIR:MASTER band29;E:MASTERband43.

and135.9◦,respectively.Forbothscenes,bands1–5and7were convertedtoat-sensorradiances,whichwerethenconvertedto top-of-atmosphere(TOA)reﬂectance(Chanderet al.,2009).The darkobjectsubtraction algorithm wasthenusedtoderive sur-facereﬂectance(Chavez,1996),whichwasfollowedbytopographic correctionusingtheC-correctionmethodandtheASTER-derived DEMdata(Teilletetal.,1982;ASTERGDEMValidationTeam,2009). Finally,image-to-imageco-registrationwasconductedusingthe 2007Landsatsceneasthebase.Asecond-orderpolynomial warp-ingmethodandbilinearneighborresamplingwereappliedbased on40tiepoints,yieldingaRMSEof0.16pixels.

2.5. Burnseveritymodeling

Oneofthemajorobjectivesof thisstudywastounderstand theimpactoftheforestdiseaseSODonburnseverity.Todothis, wedevelopedtwo groupsofmodelstoassessburnseverityfor infestedanduninfestedforests,respectively.Withineachgroup,we furtheranalyzedtherelationshipsbetweenﬁeld-assessedCBIand remotesensing-derivedvariablesfortheﬁveforeststrataincluded intheCBI(i.e.,substrate,herb,shrub,intermediate-sizedtree,and dominanttreelayers),aswellastheaverageCBI.Wealsocompared threetypesofremotesensingmodelsusingMASTERplusLandsat bundle,MASTERonly,andLandsatonly.Consequently,atotalof36 modelsweredevelopedinourburnseverityanalysis.

Whilethedependentvariableofburnseverity(i.e.,CBI)was measuredinthefield,theindependentvariableswereextracted fromtheremotelysenseddata:(i)spectralresponsesand(ii) spec-tralindices(Table2).Specifically,(i)spectral responsesinclude allthe50bandsfromtheMASTERdata,andsixbands(i.e.,bands 1–5,7)fromthe2008post-fireLandsatimage.(ii)Spectralindices extractedfromthetwosensorswereslightlydifferent.For Land-sat,we extracted normalized burn ratio (NBR)and normalized differencevegetation(NDVI)fromthe2008post-firedata.Since the 2007pre-fire Landsat data were available, we further cal-culateddifferenced NBR (dNBR)and differenced NDVI(dNDVI). Thesefourindiceshavebeenwidelyusedinpreviousburn sever-ityassessmentsgeneratingpromisingresults(Tucker,1979;Xiao etal., 2002).For MASTER, wefirst extracted NBRand NDVI by followingHarrisetal.(2011)toselectappropriatered,NIR,and lSWIR(longershortwaveinfrared)bands.Itwasnotpossibleto calculatedNBRanddNDVIduetothelackofpre-firedata; how-ever,wecapitalizedontheMASTER’suniqueMIRandTIRbands andcalculatedextrafourindices:NIR SWIR emissivityversion1 and2(NSEv1andNSEv2),Vegetationindex3(VI3),andSWIRMIR index(SMI)(KaufmanandRehmer,1994;Veraverbekeetal.,2011, 2012).Specifically,VI3andSMIweredevelopedtoemploytheMIR

spectralregionandrevealedhighrobustnessagainstthepresence of smoke,while NSEv1and NSEv2werecreatedto incorporate emissivityfromthethermalbandsandshowedrelativelystrong potential for discriminating immediate post-ﬁre burned areas. PleaserefertoTable2fordetailedequationsandbandsused.

Multiplelinearregressionmodelsweredevelopedtoestimate burnseverityfromtheremotesensingpredictors.Therationaleof choosingmultipleregressionwasduetoitswideapplicationin burnseveritymappinggeneratingsatisfactoryresults(e.g.,Miller andThode,2007;Harrisetal.,2011).However,wealsonotethat physicalmodelsandmachinelearningapproacheshavebeenused forsimilarmappingpurposes(e.g.,DeSantisetal.,2007;Hultquist etal.,2014Hultquistetal.,2014).Toavoidtheoverfittingproblem commontoregressionmodelsandreducemulticollinearity, cor-relationcoefficientswerefirstcalculatedbetweenallindependent (i.e.,spectralresponsesandindices)anddependent(i.e.,CBI) vari-ables.Eachindependentvariablewasevaluatedandretainedunder tworules:(1)itscorrelationvaluewithanyotherindependent vari-ableshouldbelowerthan0.7;or(2)ifitscorrelationwithanother oneorseveralindependentvariablesishigherthan0.7,thevariable shouldhavethehighestcorrelationvaluewiththefield-measured CBIofallthesevariables(Chenetal.,2011).Afterdiscarding redun-dantindependentvariables,aforwardstepwisemethodwasused intheregressionanalysistodevelopmodelsata0.05significance level.Tocalibratemodels,werandomlyselected50%ofthefield data(i.e.,50%oftheinfestedanduninfestedplots,respectively), whiletheremainderwasusedinmodelvalidation.

Once the burn severity models were developed from the ground-truthedobservationsin61plots,theywereapplied sep-aratelyininfestedanduninfestedforestsbasedonatreemortality mapproducedbyMeentemeyeretal.(2008)whoestimatedthe number of trees killed bySOD measuredusing high-resolution aerialimagescalibratedwithﬁelddata.Apparently,theforest infes-tationstatuswasacquiredpriortotheﬁreevent.

3. Resultsanddiscussion

3.1. ComparisonbetweenLandsatandMASTER

(5)

Fig.3. Comparisonofcoefﬁcientofdetermination(R2₎_using_Landsat,_MASTER,_and_the_Landsat_and_MASTER_bundle_for_modeling_burn_severity_in_infested_and_uninfested

forests.

progressively decreased from the relatively high intermediate-sizedtreelayer,theshrublayertothelowsubstratelayer.This wasexpectedasthesignalsreceivedbyremotelysensedoptical sensorsaretypicallybiasedtotheupperlevelofforestcanopies (Chenetal.,2012).Fortheherblayer(vegetationlessthan1.0min height),bothLandsatandMASTERperformedpoorly,withR2_close to0.Besidesthefactoroflowsignalpenetrationrate,oneparticular reasonwasthatburnseverityatthislayerwasveryhigh(e.g.,max CBIof3.0)in80%theﬁeldplots,withmeanandmedianCBIvalues closeorequalto3.0(Table1).BecauseofsuchhighCBIvaluesfor allplots,notrendswerepresentinthedataforthislayer.Inthe followingsections,herblayerwillberemovedfromourdiscussion. WefurtherfoundthattheestimationofaverageCBI,representing anoverallassessmentofburnseverityofalllayers,showedaverage performanceamongallthemodelsdeveloped(Fig.3).

Inaddition,Fig.3showsthatthecombinationofLandsatand MASTERconsistentlygeneratedbetterresultsthanusingLandsator MASTERonlyforalmostalltheevaluatedforeststrata.Forinfested forests, on average, the data combination improvedthe model performanceby14–15%inR2 _compared_to_the_use_of_Landsat_or MASTER,whilethemodelRMSEdecreasedby0.04–0.06.Similarly, foruninfestedforests,theaveragemodelimprovementswere13% and17%inR2_,_with_decreases_of_0.06_and_0.10_in_RMSE._Clearly,_the combinationofLandsatandMASTERdemonstratedhighpotential toimproveburnseverityassessmentinbothtypesofforestswith andwithoutdiseaseinfection.However,thecomparisonbetween LandsatandMASTERshowedslightlydifferenttrends(Fig.3).For uninfestedtrees,theapplicationofLandsathadbettermodel per-formanceinalmostalltheforeststrata(exceptthesubstratelayer), withanaverageincrease of5%in R2 _and_a _decrease _of_0.06 _in RMSE.Forinfestedtrees,whileLandsatoutperformedMASTERat theshrub,dominantandaveragetreelayers,MASTERwasbetter for estimatingburn severityatthesubstrate and intermediate-sizedtreelayers(Fig.3).Thecomparisonsconﬁrmedthehighvalue ofapplyingthetypicalpre-andpost-ﬁreLandsatdatabundlein burnseveritymapping,whileMASTERdatawithahigherspectral resolutionandextramid-thermalbandsmayoutperformLandsat undercertaincircumstances(e.g.,theintermediate-sizedlayerin infestedforests).However,wenotedthattheadditionofMASTER toLandsatprovidedthebestsolution(i.e.,highermodelR2 _and lowerRMSE)comparedtotheuseofMASTERorLandsatdataonly (Fig.3).

ScatterplotsoftheburnseveritymodeledbyLandsatand/or MASTERdataversusﬁeldvalidationmeasurementsarepresented in Figs. 4 and 5, representing infested and uninfested forests,

respectively. Similar to our findings during model calibration (Fig.3),burnseverityassessmentsusingthecombinationof Land-sat and MASTERshowed betterresults (i.e.,higher agreements withfieldmeasurements)thanthoseusingdatafromindividual sensorsformostoftheevaluatedforeststrata,andburn sever-ityatthedominanttreelayerwasfoundtobemoreaccurateto estimate(Figs.4and5).However,exceptionsremained.For exam-ple,data combination didnot improve themodel performance atthesubstrateandaveragelayers,withestimationresultseven poorerthanusingLandsatonlyfortheinfestedforests(Fig.4).This canbeexplainedbythelowvarianceofburnseverityCBIvalues acquiredfromtheinfestedforests,wheretheadditionor subtrac-tionofindividualfieldmeasuresmayhavesignificantlyaffected model performance, resulting in less robust models.Especially fortheunderstorylayers,SOD increasedfuelloads (Metzetal., 2011),whichwasexpectedtobethemainreasoncausing simi-larburneffectswithlowvariance.Comparedtouninfestedforests, infestedstandsclearlyintroducedhighervariationtoforestspectral reflectance,makingitchallengingtoremotelysenseburnseverity indiseasedtrees.

Toexplore thepotentialreasonswhy themodelscombining bothLandsatandMASTERdatasetsledtoimprovedburn sever-ity assessmentsin theevaluated forestlandscape, weanalyzed theindependentvariablesthathavesignificantlycontributed to theintegratedmodels.Table3showsthatLandsat-derived vari-ables (e.g.,NDVI, NBR, dNBR,red and NIRbands)still played a criticalroleinburnseverityestimationinbothinfestedand unin-festedforests. Similar toseveralpreviousfindings,this may be explained by thefact that fire can cause remarkabledecreases inchlorophyllcontentandvegetationmoistureleadingtolower visible andNIRreflectance andhigher SWIRreflectance(White et al.,1996; De Santis and Chuvieco, 2007).Among these vari-ables,themostcommononewasNDVIthatappearedinfiveofthe total10modelsandwasnegativelycorrelatedwithburnseverity (Table3).Surprisingly,thisfindingisslightlydifferentfromsome ofthoseintheliterature,whereNBRordNBR,ratherthanNDVI, extractedfromtheNIR SWIRbandsshowedstrongercorrelations withCBI(DeSantisandChuvieco,2007).Becauseourstudyarea waslocatedinamountainouscoastalregion,itwaspossiblethat microclimatesledtohighvariationinvegetationwatercontent, thusincreasingvariationintheSWIRspectralbandsandNBRand dNBRvalues.

(6)

Fig.4. ScatterplotsoftheburnseveritymodeledbyLandsatand/orMASTERdataversusﬁeldvalidationmeasurementsforinfestedforests,with1:1linesusedforacomparison purpose.

oftheindependentvariablesusedinthefinalintegratedmodels werebasedonthe2008data(Table3).ComparedtoLandsat,the uniquefeaturesofMASTERdatalaidintwoaspects:(i)higher spec-tralresolution (i.e., 25 MASTERbands versussix Landsat bands in VNIR SWIR),and (ii) extra25 MIR TIR bands. Thus, it was meaningfultoinvestigatewhetherthesefeaturescanexert posi-tiveimpactsonburnseveritymapping.InTable3,wefoundthat 12variablesfromtheMASTERsensor(versusfivevariablesfrom Landsat)contributedtotheburnseverityassessments.Specifically, fourvariableswerewithinVNIR SWIR,withfivevariablesfrom MIR TIR,andthreespectralindices(i.e.,NSEv1M,SMIMandNDVIM) derivedfromboththeVNIR SWIR,MIR TIRspectralranges.The reasonthatalargernumberofMASTER-derivedvariablesshowed

significanceinthefinalmodelscouldbeexplainedbythefactthat theMASTERnarrowspectralbandsweresensitivetothechanges inforestconditions.Evenifsuchchangesweresubtle,thevariables thathadahigherabilitytoexplainvarianceinburnseveritymay vary,probablyshiftingtoothervariableswhenmodelingdifferent foreststrata.Resultsalsorevealedthesignificantcontributionsof severaluniqueindices(e.g.,NSEv1MandSMIM)thatwerederived fromtheMASTER’sMIR TIRbands.Theseextendedspectralranges werepossiblyrobustagainstfiresmoke-causedscatteringenabling a‘clear’viewofgroundburnseverity,similartothefindings dis-coveredbyVeraverbekeetal.(2012).

Hence,wemayconcludethat(i)highspectralresolution MAS-TER data (using extended MIR TIR spectral bands) have the

Table3

BurnseveritymodelsusingsynergizedLandsatandMASTERdataformultipleforeststrata(excepttheherblayer).

Forestlayer Model

Substrate YInfested=1.14+21.65×NBRL+0.94×SMIM

YUninfested=−39.1−17.73×DN3L−6.28×DN4L+15.46×DN34M+32.56×DN48M+2.49×NSEv1M

Shrub YInfested=0.91+10.54×DN11M−10.57×DN23M−9.06×NDVIM YUninfested=11.50+5.33×DN3L−3.90×NDVIL−16.76×DN18M

Intermediate-sizedtree YInfested=−15.62−2.81×NDVIL−3.91×DN18M+21.72×DN49M YUninfested=−9.39−3.40×NDVIL+4.93×DN17M+11.43×DN36M

Dominanttree YInfested=1.44+3.32×DN5L−1.32×NBRL−2.25×NSEv1M YUninfested=−12.82−3.10×NDVIL+21.68×DN34M−3.98×DN42M

Average YInfested=2.72−1.62×NDVIL+0.22×DN18M

YUninfested=4.66+1.61×dNBRL+0.04×DN11M+14.62×DN34M−18.11×DN42M

dNBRL=Landsatdifferencednormalizedburnratio;

SRiL=Spectralresponseofthe2008Landsat’sithband(i=1,...5,7); NBRL=Landsat2008normalizedburnratio;

NDVIL=Landsat2008normalizeddifferencevegetationindex; SRiM=SpectralresponseoftheMASTER’sithband(i=1,...50); SMIM=MASTERSWIR–MIRindex;

NDVIM=MASTERnormalizeddifferencevegetationindex; NSEv1M=MASTERNIR–SWIR-emissivityversion1; SWIR=Shortwaveinfrared;

(7)

Fig.5. ScatterplotsoftheburnseveritymodeledbyLandsatand/orMASTERdataversusﬁeldvalidationmeasurementsforuninfestedforests,with1:1linesusedfora comparisonpurpose.

potentialtoaugmenttheperformanceofapplyingLandsatimagery

toassessforestburnseverityinbothhealthyanddiseasedtrees;

and(ii)thecombinationofMASTERandLandsatdatacollectedafter

ﬁredisturbancesmayretainorevenimprovetheperformanceof

burnseverityestimationusingthetypicalpre-andpost-ﬁre

Land-satdatabundle,facilitatinglandscape-scaleforestﬁremanagement

ifpre-ﬁrehigh-quality remote sensingdata arenot possibleto

obtain.

3.2. Comparisonbetweeninfestedanduninfestedforests

Asdiscussedintheprevioussection,datacombinationusing

Landsat and MASTER demonstrated high potential to improve

theaccuracyofassessingburnseveritythanusingindividual

sen-sorsin bothinfested anduninfestedforests. Andamong allthe

foreststrataevaluated,thedominanttreelayergeneratedthemost

accurateresults.Hence,thefollowingcomparisonsofburneffects

(8)

Table4

Summarystatistics(i.e.,area,percentageoftotalarea,minimum,maximum,mean,medianandstandarddeviation)oftheburnseverityinforestsaffectedbythedisease.

Burnseverity Area Percentageoftotalarea Minimum Maximum Mean Median Standarddeviation

(ha) (%)

Low(0–1) 3528.5 61.2 0.00 0.99 0.41 0.42 0.33

Moderate(1–2) 1579.8 27.4 1.00 1.99 1.41 1.37 0.28

High(2–3) 654.4 11.4 2.00 3.00 2.51 2.47 0.33

Table5

Summarystatistics(i.e.,area,percentage,minimum,maximum,mean,medianandstandarddeviation)oftheburnseverityinforestswithoutdiseaseinvasion.

Burnseverity Area Percentage Minimum Maximum Mean Median Standarddeviation

(ha) (%)

Low(0–1) 5511.9 38.7 0.00 0.99 0.16 0.00 0.32

Moderate(1–2) 5017.1 35.2 1.00 1.99 1.52 1.53 0.28

High(2–3) 3722.0 26.1 2.00 3.00 2.44 2.40 0.30

onlyfocusonapplyingtheintegratedmodels(Table3)tothe

dom-inanttreelayer.Fig.6showsthreeburnseveritymapsforinfested, uninfestedandtheentire(combined)forestcommunity.A rain-bowlegendwasused,withgreencolorindicatinglowseverityof burns,redcolorshowingsevereburns,andwhitecolor represent-ingnon-forestedlands.Inset(c)isasampleburnedforeststand comprisedoftrees(a)affectedbydiseaseand(b)thosethatwere not,withblacklinesin(c)representingtheboundariesofthesetwo communities.Surprisingly,burnseverityunveiledasmooth tran-sitionontheboundariesofinfestedanduninfestedstands(Fig.6), withsimilarphenomenonobservedinmanyotherlocationsofthe studyarea.Whileitwasexpectedthatburnseveritymayincrease ininfestedforestsasaresultofreducedfoliarmoisturecontentand increasedforestfuels,suchphenomenonwasnotclearlyexposed intheburnseveritymapcreatedinourstudy.Besidesinfestation, wefoundthatcomplexlocalmicroclimates(e.g.,causedby topog-raphy)mayalsoplayaroleindeterminingburnseverity(Leeetal., 2009).

Toquantifythedifferencesinburnseveritybetweeninfested anduninfestedforests,wesummarizedtheirarea,percentage, min-imum,maximum,mean,medianandstandarddeviationforthree burnseveritycategoriesoflow(CBI

[0,1]),moderate(CBI∈[1,2]) andhigh(CBI∈[2,3]),respectively(Tables4and5).Inthisresearch, wefoundthatahigherareapercentageoflow-severityburnsanda smallerareapercentageofhigh-severityburnsoccurredininfested thanuninfestedforests(i.e.,low:61.2%versus38.7%;high:11.4% versus26.1%).Thiscouldbeexplained by twofacts.First, most oftheinfestedtreeswerewithinthenarrowstripinthewestof thestudyarea(Fig.6), asa consequenceof SODinvasion start-ingfromthecoastofPaciﬁcOceaninthewest(Rizzoetal.,2005). However,theuninfestedtreesstilloccupiedalargeportionofthe area,i.e.,6.3timeslargerthanthecoverageoftheinfestedforests (Fig.6).Second,mostofthesevereburnsoccurredinthecentral partofourstudyarea,outofthedistributionsofthemajorityof infestedtrees(Fig.6).Intermsofthecomparisonofthesamelevel ofburnseverity,wefoundsimilarstatisticalresults(i.e.,similar mean,medianandstandarddeviation)withinthetwocategories ofmoderateandhighlevelsofburnsforbothinfestedand unin-festedforests(Tables4and5).However,theresultsalsoshowed thatinthecategoryoflowburnseverity,uninfestedforeststended tohaveamuchlowerburneffectthanthatininfestedforests(e.g., mean:0.16versus0.41;median:0.00versus0.42).Thiswas possi-blyduetothehigherwatercontentinhealthytreeleavesthatwere moreresilienttosmallﬁres.

4. Conclusions

Today’sforestecosystemsareincreasinglyaffectedbyavariety ofdisturbancesincludingﬁresandemerginginfectiousdiseases,

which makes it challenging toaccurately assess post-fire burn severity.In this study, we evaluated the performance of using (pre-andpost-fire)Landsatand(post-fire)MASTERdatatoassess burnseverityinaCaliforniacoastalforestecosystem,whichhas beenaffected bya non-native infectiousdisease SOD sincethe mid-1990s.ResultsshowedthatthecombinationofLandsatand MASTER consistently generated better resultsthan those using LandsatorMASTERonlyinalmostalltheevaluatedforeststrata (e.g.,substrate,shrubs, intermediate-sizedtrees,dominanttrees andtheaverage),amongwhich thebestresult wasachievedat thedominanttreelayer.TheapplicationofMASTERMIRandTIR spectralbands(3.0–12.5␮m)wasfoundtoaugmentburnseverity assessmentsfromLandsatforbothuninfestedanddiseasedforests. Wefurtherfoundimprovedmodelperformanceusingintegrated MASTERandLandsatdatacollectedafterthefireevent,ratherthan thetypicalpre-andpost-fireLandsatbundle,suggestingthat post-firehighspectralresolutionMASTERdatamayhavethepotentialto retainorevenimprovetheburnseverityestimationaccuracyif pre-firehigh-qualityremotesensingdataarenotpossibletoobtain(e.g., duetohighpercentcloudcover).Itwasalsodiscoveredthatthe burnseveritymodelsdevelopedfortheuninfestedforests outper-formedtheonesfortheinfestedtrees,indicatingthattheinvasion ofSODhasintroducedhighspectralvariationtolocalforests,where singlesensorsandsimplemodels(e.g.,linearregression)maynot besufficienttofullydescribethecomplexrelationshipbetween remotelysensedspectralreflectanceandburnseverity.In addi-tion,statisticalresultsshowedthatboththeinfestedanduninfested forestssimilarly experiencedmoderateandhighlevelsofburns (CBI>1).Thedifferenceoccurredintheforestswithlowburn sever-ity(CBI<1),whereuninfestedtreeshadamuchlowerburneffect thanthatininfestedtrees,possiblyduetothefactthathealthytree withhigherwater contentandless fuelsweremoreresilientto smallfiredisturbances.Becauseseveralfactors(e.g.,the character-isticsofdataandstudyarea)mayaffectourfindings,futureresearch isneededtoveritytherobustnessofcombininghigh-spectral MAS-TERandmulti-dateLandsatinburnseverityassessmentoverother diseasedforests.

Acknowledgements

(9)

Mehl,A.Oguchi,E.Paddock,K.Pietrzak,M.Vaclavikova,J.Vieregge, L.WaksandA.Wickland.

References

ASTERGDEMValidationTeam,2009.ASTERGlobalDEMValidation:Summary Report.June2009.Availableonline:https://lpdaac.usgs.gov/sites/default/ﬁles/ public/aster/docs/ASTERGDEMValidationSummaryReport.pdf

Asner,G.P.,2013.Geographyofforestdisturbance.Proc.Natl.Acad.Sci.U.S.A.110, 3711–3712.

Bergot,M.,Cloppet,E.,Pérarnaud,V.,Déqué,M.,Marcais,B.,Desprez-Loustau, M.-L.,2004.Simulationofpotentialrangeexpansionofoakdiseasecausedby Phytophthoracinnamomiunderclimatechange.GlobalChangeBiol.10, 1552–1939.

Carcaillet,C.,Bergeron,Y.,Richard,P.J.H.,Fréchette,B.,Gauthier,S.,Prairie,Y.T., 2001.ChangeofﬁrefrequencyintheeasternCanadianborealforestsduring theHolocene:doesvegetationcompositionorclimatetriggertheﬁreregime? J.Ecol.89,930–946.

Chander,G.,Markham,B.L.,Helder,D.L.,2009.Summaryofcurrentradiometric calibrationcoefﬁcientsforLandsatMSS,TM,ETM+,andEO-1ALIsensors. RemoteSens.Environ.113,893–903.

Chavez,P.,1996.Image-basedatmosphericcorrections–Revisitedandimproved. Photogramm.Eng.RemoteSens.62,1025–1036.

Chen,G.,Hay,G.J.,Castilla,G.,St-Onge,B.,Powers,R.,2011.Amultiscale geographicobject-basedimageanalysistoestimatelidar-measuredforest canopyheightusingQuickbirdimagery.Int.J.Geogr.Inf.Sci.25,37–41. Chen,G.,Wulder,M.A.,White,J.C.,Hilker,T.,Coops,N.C.,2012.Lidarcalibration

andvalidationforgeometric-opticalmodelingwithLandsatimagery.Remote Sens.Environ.124,384–393.

Chen,G.,Metz,M.R.,Rizzo,D.M.,Dillon,W.W.,Meentemeyer,R.K.,2015. Object-basedassessmentofburnseverityindiseasedforestsusing high-spatialandhigh-spectralresolutionMASTERairborneimagery.ISPRSJ. Photogramm.RemoteSens.102,38–47.

Davis,F.W.,Borchert,M.I.,Flint,A.,Meentemeyer,R.K.,Rizzo,D.M.,2010. Pre-impactforestcompositionandongoingtreemortalityassociatedwith suddenoakdeathdiseaseintheBigSurregion,California.For.Ecol.Manage. 259,2342–2354.

DeSantis,A.,Chuvieco,E.,2007.Burnseverityestimationfromremotelysensed data:performanceofsimulationversusempiricalmodels.RemoteSens. Environ.108,422–435.

Epting,J.,Verbyla,D.,Sorbel,B.,2005.Evaluationofremotelysensedindicesfor assessingburnseverityininteriorAlaskausingLandsatTMandETM+.Remote Sens.Environ.96,328–339.

Fox,D.M.,Maselli,F.,Carrega,P.,2008.UsingSPOTimagesandﬁeldsamplingto mapburnseverityandvegetationfactorsaffectingpostforestﬁreerosionrisk. CATENA75,326–335.

Hall,D.K.,Ormsby,J.P.,Johnson,L.,Brown,J.,1980.Landsatdigitalanalysisofthe initialrecoveryofburnedtundraatKokolikRiver,Alaska.RemoteSens. Environ.10,263–272.

Harris,S.,Veraverbeke,S.,Hook,S.,2011.Evaluatingspectralindicesforassessing ﬁreseverityinChaparralecosystems(SouthernCalifornia)using

MODIS/ASTER(MASTER)airbornesimulatordata.RemoteSens.3, 2403–2419.

Hook,S.J.,Myers,J.J.,Thome,K.J.,Fitzgerald,M.,Kahle,A.B.,2001.The MODIS/ASTERairbornesimulator(MASTER)–anewinstrumentforearth sciencestudies.RemoteSens.Environ.76,93–102.

Hudak,A.T.,Brockett,B.H.,2004.MappingﬁrescarsinasouthernAfricansavanna usingLandsatimagery.Int.J.RemoteSens.25,3231–3243.

Hultquist,C.,Chen,G.,Zhao,K.,2014.AcomparisonofGaussianprocess regression:randomforestsandsupportvectorregressionforburnseverity assessmentindiseasedforests.RemoteSens.Lett.5,723–732.

Jakubauskas,M.E.,Lulla,K.P.,Mausel,P.W.,1990.Assessmentofvegetationchange inaﬁre-alteredforestlandscape.Photogramm.Eng.RemoteSens.56, 371–377.

Jenkins,M.J.,Hebertson,E.,Page,W.,Jorgensen,C.A.,2008.Barkbeetlesfuels,ﬁres andimplicationsforforestmanagementintheIntermountainWest.For.Ecol. Manage.254,16–34.

Kaufman,Y.,Rehmer,L.,1994.Detectionofforestsusingmid-IRreﬂectance:an applicationforaerosolstudies.IEEETrans.Geosci.RemoteSens.32,672–683. Keeley,J.E.,2009.Fireintensity,ﬁreseverityandburnseverity:abriefreviewand

suggestedusage.Int.J.WildlandFire18,116–126.

Key,C.,Benson,N.,2006.Landscapeassessment:groundmeasureofseverity;the compositeburnindex,andremotesensingofseverity,thenormalizedburn index.In:Lutes,D.,Keane,R.,Caratti,J.,KeyBenson,C.,Sutherland,N.,Gangi, S.L.(Eds.),FIREMON:FireEffectsMonitoringandInventorySystem.Rocky MountainsResearchStationUSDAForestService,FortCollins,CO,USA,pp. 1–51.

Kokaly,R.F.,Rockwell,B.W.,Haire,S.L.,King,T.V.V.,2007.Characterizationof post-ﬁresurfacecover,soils,andburnseverityattheCerroGrandeFire,New Mexico,usinghyperspectralandmultispectralremotesensing.RemoteSens. Environ.106,305–325.

Kurz,W.A.,Stinson,G.,Rampley,G.,2008a.Couldincreasedborealforest ecosystemproductivityoffsetcarbonlossesfromincreaseddisturbances? Philos.Trans.R.Soc.B363,2259–2268.

Kurz,W.A.,Dymond,C.C.,Stinson,G.,Rampley,G.J.,Neilson,E.T.,Carroll,A.L., Ebata,T.,Safranyik,L.,2008b.Mountainpinebeetleandforestcarbonfeedback toclimatechange.Nature452,987–990.

Lamsal,S.,Cobb,R.C.,HallCushman,J.,Meng,Q.,Rizzo,D.M.,Meentemeyer,R.K., 2011.Spatialestimationofthedensityandcarboncontentofhostpopulations forPhytophthoraramoruminCaliforniaandOregon.For.Ecol.Manage.262, 989–998.

Lee,S.-W.,Lee,M.-B.,Lee,Y.-G.,Won,M.-S.,Kim,J.-J.,Hong,S.,2009.Relationship betweenlandscapestructureandburnseverityatthelandscapeandclass levelsinSamchuck,SouthKorea.For.Ecol.Manage.258,1594–1604. Lentile,L.B.,Holden,Z.A.,Smith,A.M.S.,Falkowski,M.J.,Hudak,A.T.,Morgan,P.,

Lewis,A.A.,Gessler,P.E.,Benson,N.C.,2006.Remotesensingtechniquesto assessactiveﬁrecharacteristicsandpost-ﬁreeffects.Int.J.WildlandFire15, 319–345.

Meentemeyer,R.K.,Rank,N.E.,Shoemaker,D.,Oneal,C.,Wickland,A.C.,Frangioso, K.M.,Rizzo,D.M.,2008.Impactsofsuddenoakdeathontreemortalityinthe BigSurecoregionofCalifornia.Biol.Invasions10,1243–1255.

Meentemeyer,R.K.,Haas,S.E.,Vaclavik,T.,2012.Landscapeepidemiologyof emerginginfectiousdiseasesinnaturalandhuman-alteredecosystems.Annu. Rev.Phytopathol.50,379–402.

Metz,M.R.,Frangioso,K.M.,Meentemeyer,R.K.,Rizzo,D.M.,2011.Interacting disturbances:wildﬁreseverityaffectedbystageofforestdiseaseinvasion. Ecol.Appl.21,313–320.

Metz,M.R.,Frangioso,K.,Meentemeyer,R.K.,Rizzo,D.M.,2012.Anemergent diseasecausesdirectionalchangesinforestspeciescompositionincoastal California.Ecosphere3,art86.

Metz,M.R.,Varner,J.M.,Frangioso,K.M.,Meentemeyer,R.K.,Rizzo,D.M.,2013. Unexpectedredwoodmortalityfromsynergiesbetweenwildﬁreandan emerginginfectiousdisease.Ecology94,2152–2159.

Miller,J.D.,Thode,A.E.,2007.Quantifyingburnseverityinaheterogeneous landscapewitharelativeversionofthedeltanormalizedburnratio(dNBR). RemoteSens.Environ.109,66–80.

Miller,J.D.,Yool,S.R.,2002.Mappingforestpost-ﬁrecanopyconsumptionin severaloverstorytypesusingmulti-temporalLandsatTMandETMdata. RemoteSens.Environ82,481–496.

Milne,A.K.,1986.Theuseofremotesensinginmappingandmonitoring vegetationalchangeassociatedwithbushﬁreeventsinEasternAustralia. GeocartoInt.1,25–32.

Morton,D.C.,LePage,Y.,DeFries,R.,Collatz,G.J.,Hurtt,G.C.,2013.Undestoreyﬁre frequencyandthefateofburnedforestsinsouthernAmazonia.Philos.Trans. R.Soc.B368,20120163.

NASA,2015.HyspIRIMissionStudy.Availableonline:http://hyspiri.jpl.nasa.gov Wildlandﬁreinecosystems:effectsofﬁreonsoilsandwaterNeary,D.G.,Ryan, K.C.,Debano,L.F.(Eds.),2005.Gen.Tech.Rep.RMRS-GTR-42-,vol.4.UT:U.S. DepartmentofAgriculture,ForestService,RockyMountainResearchStation, Ogden,pp.5–17.

Olsson,P.-O.,Jönsson,A.M.,Eklundh,L.,2012.AnewinvasiveinsectinSweden– Physokermesinopinatus:tracingforestdamagewithsatellitebasedremote sensing.ForestEcol.Manage.285,29–37.

Pan,Y.,Birdsey,R.A.,Fang,J.,etal.,2011.Alargeandpersistentcarbonsinkinthe world’sforests.Science333,988–993.

Quintano,C.,Fernández-Manso,A.,Roberts,D.A.,2013.Multipleendmember spectralmixtureanalysis(MESMA)tomapburnseveritylevelsfromLandsat imagesinMediterraneancountries.RemoteSens.Environ.136,76–88. Rizzo,D.M.,Garbelotto,M.,Hansen,E.M.,2005.Phytophthoraramorum:integrative

researchandmanagementofanemergingpathogeninCaliforniaandOregon forests.Annu.Rev.Phytopathol.43,309–335.

Roy,D.P.,Boschetti,L.,Trigg,S.,2006.Remotesensingofﬁreseverity:assessingthe performanceofthenormalizedburnratio.IEEETrans.Geosci.RemoteSens. Lett.3,112–116.

Teillet,P.M.,Guindon,B.,Goodenough,D.G.,1982.Ontheslope-aspectcorrection ofmultispectralscannerdata.Can.J.RemoteSens.8,84–106.

Tucker,C.,1979.Redandphotographicinfraredlinearcombinationsfor monitoringvegetation.RemoteSens.Environ.8,127–150.

Turner,M.G.,Baker,W.L.,Peterson,C.J.,Peet,R.K.,1998.Factorsinﬂuencing succession:lessonsfromlarge,infrequentnaturaldisturbances.Ecosystems1, 511–523.

USDAForestService,2008.ExecutiveSummary:BasinComplexFire/IndiansFire BAERInitialAssessment.USDAForestService,pp.16.

Veraverbeke,S.,Harris,S.,Hook,S.,2011.Evaluatingspectralindicesforburned areadiscriminationusingMODIS/ASTER(MASTER)airbornesimulatordata. RemoteSens.Environ.115,2702–2709.

Veraverbeke,S.,Hook,S.,Hulley,G.,2012.Analternativespectralindexforrapid ﬁreseverityassessments.RemoteSens.Environ.123,72–80.

Westerling,A.L.,Hidalgo,H.G.,Cayan,D.R.,Swetnam,T.W.,2006.Warmingand earlierspringincreasewesternUSforestwildﬁreactivity.Science313, 940–943.

White,J.D.,Ryan,K.C.,Key,C.C.,Running,S.W.,1996.Remotesensingofforestﬁre severityandvegetationrecovery.Int.J.WildlandFire6,125–136.

Wulder,M.A.,Dymond,C.C.,White,J.C.,Leckie,D.G.,Carroll,A.L.,2006.Surveying mountainpinebeetledamageofforests:Areviewofremotesensing opportunities.For.Ecol.Manage.221,27–41.