ebook img

Seasonality and drought effects of Amazonian forests observed from multi-angle satellite data PDF

14 Pages·2015·3.47 MB·English
by  HilkerThomas
Save to my drive
Quick download
Download
Most books are stored in the elastic cloud where traffic is expensive. For this reason, we have a limit on daily download.

Preview Seasonality and drought effects of Amazonian forests observed from multi-angle satellite data

Seasonality and drought effects of Amazonian forests observed from multi-angle satellite data de Moura, Y. M., Hilker, T., Lyapustin, A. I., Galvão, L. S., dos Santos, J. R., Anderson, L. O., ... & Arai, E. (2015). Seasonality and drought effects of Amazonian forests observed from multi-angle satellite data. Remote Sensing of Environment, 171, 278-290. doi:10.1016/j.rse.2015.10.015 10.1016/j.rse.2015.10.015 Elsevier Version of Record http://cdss.library.oregonstate.edu/sa-termsofuse RemoteSensingofEnvironment171(2015)278–290 ContentslistsavailableatScienceDirect Remote Sensing of Environment journal homepage: www.elsevier.com/locate/rse Seasonality and drought effects of Amazonian forests observed from multi-angle satellite data YhasminMendesdeMouraa,⁎,ThomasHilkerb,AlexeiI.Lyapustinc,LênioSoaresGalvãoa, JoãoRobertodosSantosa,LianaO.Andersond,e,CélioHelderResendedeSousab,EgidioAraia aNationalInstituteforSpaceResearch(INPE),AvenidadosAstronautas,1758SãoJosédosCampos,SP,Brazil bOregonStateUniversity,CollegeofForestry,Corvallis,OR97331,USA cNASAGoddardSpaceFlightCenter,Greenbelt,MD20771,USA dNationalCentreforMonitoringandEarlyWarningofNaturalDisasters(CEMADEN),EstradaDoutorAltinoBondensan,500,SãoJosédosCampos,SP,Brazil eEnvironmentalChangeInstitute,UniversityofOxford,OxfordOX13QY,UK a r t i c l e i n f o a b s t r a c t Articlehistory: SeasonalityanddroughtinAmazonrainforestshavebeencontroversiallydiscussedintheliterature,partiallydue Received25March2015 toalimitedabilityofcurrentremotesensingtechniquestodetectitsimpactsontropicalvegetation.Weusea Receivedinrevisedform9October2015 multi-angleremotesensingapproachtodeterminechangesinvegetationstructurefromdifferencesindirection- Accepted22October2015 alscattering(anisotropy)observedbytheModerateResolutionImagingSpectroradiometer(MODIS)withdata Availableonline5November2015 atmosphericallycorrectedbytheMulti-AngleImplementationAtmosphericCorrectionAlgorithm(MAIAC). Ourresultsshowastronglinearrelationshipbetweenanisotropyandfield(r2=0.70)andLiDAR(r2=0.88) Keywords: basedestimatesofLAIevenindensecanopies(LAI≤7m2m−2).Thisallowedustoobtainimprovedestimates Amazon Drought ofvegetationstructurefromopticalremotesensing.WeusedanisotropytoanalyzeAmazonseasonalitybased Anisotropy onspatiallyexplicitestimatesofonsetandlengthofdryseasonobtainedfromtheTropicalRainfallMeasurement Greening Mission(TRMM).Anincreaseinvegetationgreeningwasobservedduringthebeginningofdryseason(across Browning ~7%ofthebasin),whichwasfollowedbyadecline(browning)laterduringthedryseason(across~5%ofthe MAIAC basin).Anomaliesinvegetationbrowningwereparticularlystrongduringthe2005and2010droughtyears MODIS (~10%ofthebasin).Weshowthatthemagnitudeofseasonalchangescanbesignificantlyaffectedbyregional differencesinonsetanddurationofthedryseason.Seasonalchangesweremuchlesspronouncedwhen assumingafixeddryseasonfromJunethroughSeptemberacrosstheAmazonBasin.Ourfindingsreconcile remotesensingstudieswithfieldbasedobservationsandmodelresultsastheyprovideasounderbasisforthe argumentthattropicalvegetationgrowthincreasesduringthebeginningofthedryseason,butdeclinesafter extendeddroughtperiods.Themulti-angleapproachusedinthisworkmayhelpquantifydroughttolerance andseasonalityintheAmazonianforests. ©2015ElsevierInc.Allrightsreserved. 1.Introduction couldalterspeciescomposition,biodiversity(Asner&Alencar,2010; Asner,Nepstad,Cardinot,&Ray,2004,April20;Phillipsetal.,2009) Vulnerabilityoftropicalforeststoclimatechangehasreceivedbroad andplantproductivity(Aragaoetal.,2007;Gattietal.,2014;Meir, attentionbythescientificcommunityasincreaseinequatorialseasur- Metcalfe,Costa,&Fisher,2008). facetemperature(SST)canleadtolongerdryseasons(Fuetal.,2013; Overthelastdecade,theAmazonregionhasexperiencedtwosevere Marengo,Tomasella,Alves,Soares,&Rodriguez,2011)andmorefre- droughts,onein2005andanotherin2010(Marengoetal.,2011).How- quent,severedroughtevents(Lewis,Brando,Phillips,vanderHeijden, ever,thebroadscaleresponseofvegetationtotheseeventsremains &Nepstad,2011;Malhi,Aragão,Galbraith,etal.,2009;Malhi,Aragão, controversial.Saleska,Didan,Huete,anddaRocha(2007)reportedan Metcalfe,etal.,2009;Marengoetal.,2008).Thefeedbacksofsuch increaseingreenness(higherEVI)forthe2005drought,aresultthat drying on global climate change could be substantial; the Amazon wassubsequentlychallenged(Atkinson,Dash,&Jeganathan,2011; rainforestaloneaccountsforabout15%ofglobalphotosynthesisand Samantaetal.,2010).Xuetal.(2011)observedawidespreaddecline hostsperhapsaquarteroftheworld'sterrestrialspecies(Malhietal., ingreeningforthe2010drought.Similarly,theprevailingviewofsea- 2008).Fieldstudieshaveindicatedthatsuchextremedroughtevents sonalityofvegetationhasrecentlybeendiscussed.Severalfindings (Brandoet al., 2010;Graham,Mulkey,Kitajima, Phillips,&Wright, ⁎ Correspondingauthor. 2003; Huete et al., 2006; Hutyra et al., 2007; Myneni et al., 2007; E-mailaddress:[email protected](Y.M.deMoura). Samantaetal.,2012;Wagner,Rossi,Stahl,Bonal,&Hérault,2013) http://dx.doi.org/10.1016/j.rse.2015.10.015 0034-4257/©2015ElsevierInc.Allrightsreserved. Y.M.deMouraetal./RemoteSensingofEnvironment171(2015)278–290 279 supporttheviewthatphotosyntheticactivityincreasesinitiallyduring 2.Materialandmethods thedryseasoninresponsetoanincreaseinincidentphotosynthetically active radiation (PAR). However, a recent study based on NASA's 2.1.Quantificationofmulti-anglescattering ModerateResolutionImagingSpectroradiometer(MODIS)(Morton etal.,2014)arguedthatseasonalchangesaredrivenbyartifactsofthe MAIACdatawereobtainedfrom12MODIStiles(h10v08toh13v10, sun-sensorgeometry. spanning10°Nto20°Sinlatitudeand80°Wto42°Winlongitude) Agrowingbodyofliteraturesuggestsuncertaintiesinremote from Terra and Aqua satellites between 2000 and 2012. MAIAC is sensing of atmospheric aerosol loadings (Samanta et al., 2010, basedonMODISLevel1B(calibratedandgeometricallycorrected)ob- 2012)anddeficienciesinclouddetection(Hilkeretal.,2012)tobe servations,whichremovemajorsensorcalibrationdegradationeffects partiallyresponsibleforthesecontradictingresults.Whileprogress present in earlier collections (Lyapustin et al., 2014). MAIAC grids hasbeenmadeaddressingsomeofthesechallengesbyusingalterna- MODISL1Bdatatoa1kmresolution,andaccumulatesmeasurements tivedatasets(Hilkeretal.,2014)orhigherspatialresolutionimagery ofthesamesurfaceareafromdifferentorbits(viewgeometries)for (Zelazowski,Sayer,Thomas,&Grainger,2011),observationsbased upto16daysusingamovingwindowapproach.Thesedataareused onremotelysensedvegetationindicesarelimitedintheirabilityto toderivespectralregressioncoefficientsrelatingsurfacereflectancein detectchangesinvegetationcoverduetoawell-documentedsatu- theblue(0.466μm)andshortwaveinfrared(2.13μm)foraerosolre- rationeffectinareaswithhighbiomassandleafarea(Carlson& trievals,andtoobtainparametersofsurfacebi-directionalreflectance Ripley,1997). distributionfunction(BRDF)(Lyapustinetal.,2011;Lyapustin,Wang, Asanalternativetoobservationsfromonlyoneviewangle,the Laszlo,Hilker,etal.,2012).Assumingthatvegetationisrelativelystable combinationofmultipleviewanglesmayprovidenewopportunities duringthisperiod,thesurfacedirectionalscatteringcanbecharacterized to mitigate these saturation effects, and allow better insights into usingtheRoss-ThickLi-Sparse(RTLS)BRDFmodel(Roujean,Leroy,& seasonal and inter-annual changes of tropical forests. Biophysical Deschamps,1992).Duringperiodsofrapidsurfacechange(e.g.,green- changesin thecanopy structure affect thedirectional scattering of uporsenescence)MAIACfollowsanapproachoftheMODISBRDF/ lightandtheseeffectsareobservablefrommulti-angulardata(Chen, albedo algorithm (MOD43, Schaaf et al., 2002) to scale the BRDF Menges,&Leblanc,2005).Withtheadvanceofmulti-angularsensors model with the latest measurement to adjust the magnitude of such as theMulti-angleImaging SpectroRadiometer (MISR) (Diner reflectancewhileassumingthattheshapeofBRDFdoesnotchange etal.,1998),progresshasbeenmadeindescribingthedependenceof significantly. This approach preserves spectral contrasts of actual reflectance on observation angles (Barnsley, Settle, Cutter, Lobb, & surfacecharacteristics. Teston,2004;Dineretal.,1998).Forinstance,theangularcomponent OneadvantageofusingtheRTLSmodelratherthanreflectancedi- ofsurfacereflectance(anisotropy)hasbeenlinkedtoopticalproperties rectlyisthepossibilitytomaintainconstantsun-observergeometry and geometric structure of the target (Widlowski et al., 2004; and extrapolate measurements to the principal plane to describe Widlowski,Pinty,Lavergne,Verstraete,&Gobron,2005)suchascanopy backscatterandforwardscatterdirections.Inthisstudy,weselecteda roughness(Strahler,2009),leafangledistribution(Roujean,2002),leaf viewzenithangle(VZA)of35°ratherthantheabsolutehotspotlocation areaindex(LAI)(Walthall,1997)andfoliageclumping(Chenetal., atVZA=45°inordertokeepthemodeledreflectanceclosertotheac- 2005). tualrangeofanglesobservedbyMODIS,therebyminimizingpotential The theoretical basis for the influence of canopy structure on errorsresultingfromextrapolationoftheBRDF.Forlandvegetatedsur- multi-anglereflectancehasbeendeveloped(Bicheron,1999;Chen, faces,directionalscatteringdominatesintheNIRregionduetothehigh Liu,Leblanc,Lacaze,&Roujean,2003;Gao,2003;Leblancetal.,2005; absorptionofvisiblelight.RatherthanobtaininganisotropyoftheNIR Mynenietal.,2002).However,multi-anglereflectanceisnoteasily bandalone,wecalculatedforwardandbackscatterforblue,redand obtainedfromtraditionalsurfacereflectancealgorithms,evenwhen NIR reflectance and then obtained the Enhanced Vegetation Index data is acquired from multiple viewangles.Pixel basedalgorithms (three-bandversionEVI)forbothdirections.Theobjectiveofusing often assume a Lambertian reflectance model, which reduces the EVIratherthansurfacereflectanceofagivenbandwastominimize anisotropyofthederivedsurfacereflectance(Lyapustin&Muldashev, theeffectofnon-photosyntheticallyactiveelementswhileoptimizing 1999;Wangetal.,2010)thusdecreasingtheabilitytodetectdirectional thesensitivitytogreencanopystructure.Itcan,however,beshown scattering(Hilkeretal.,2009). thatdifferencesbetweenforwardandbackwardscatterEVIislargely Newmethodsforprocessingremotesensingdata,suchastheMulti- the result of differences in scattering in the near infrared region AngleImplementationofAtmosphericCorrection(MAIAC),canhelp (Moura,Galvão,dosSantos,Roberts,&Breunig,2012). overcomethislimitationbyusingaradiativetransfermodelthatdoes ThespectralerrorofMAIACsurfacereflectancewasevaluatedasthe notmakeaLambertianassumption(Lyapustin&Knyazikhin,2001). standarddeviationbetweenobservedsurfacereflectance(BRF)and MAIACisacloudscreeningandatmosphericcorrectionalgorithmthat BRDFmodelprediction1)overatime(usinganareaof100×100km usesanadaptivetimeseriesanalysisandprocessingofgroupsofpixels toobtainsufficientstatisticsgivenhighcloudcoverinAmazonia)and toderiveatmosphericaerosolconcentrationandsurfacereflectance.A 2)inspace(pixelbypixel)fortheexampleoftwo30dayperiods,in detaileddescriptionofthealgorithmcanbefoundinLyapustinetal. JuneandSeptember. (2011),Lyapustin,Wang,Laszlo,Hilker,etal.,(2012).Inthispaper, wetakeadvantageofMAIACtostudychangesinanisotropyacrossthe 2.2.LiDARbasedestimatesofleafareaindex AmazonBasinusingthirteenyearsofmulti-angleMODISobservations. Wedefineanisotropyasdifferenceinreflectancebetweenthebackscat- Estimates of anisotropy were validated against existing and tering(relativeazimuthangle(RAA)=180°)andtheforwardscatter- independentfieldobservationsofLAI(n=16)obtainedfromthelitera- ing (RAA = 0°) directions for a fixed view and sun zenith angle. ture(Andreae,2002;Figueraetal.,2011;Domingues,Berry,Martinelli, Estimatesofsuchdefinedanisotropywerethenrelatedtofieldand Ometto,&Ehleringer,2005;Doughty&Goulden,2000a,b;Galvaoetal., LiDAR-basedestimatesofLAIinordertovalidateitsrelationtovegeta- 2011;Malhi,Aragão,Galbraith,etal.,2009;Malhi,Aragão,Metcalfe, tionstructure.Ourobjectivesweretodemonstratespatialandtemporal et al., 2009; Negrón Juárez, da Rocha, Figueira, Goulden, & Miller, changesinanisotropy,particularlyduringtheonsetofthedryseasonas 2009;Restrepo-Coupe et al., 2013; Scurlock, Asner, & Gower, 2001; ameasureofchangesinvegetation.Were-visitedthetwolastmajor Zanchietal.,2009),andAirborneLaserScanning(ALS)asanexample droughtsintheAmazonBasin(2005and2010)toevaluateanomalies ofameasureofcanopystructure.LiDARdatawereacquiredbytheSus- inanisotropyandinvestigatevegetationresponsetothesedrought tainableLandscapesBrazilprojectsupportedbytheBrazilianAgricultural eventsonamonthlybasis. ResearchCorporation(EMBRAPA),theUSForestService,USAID,andthe 280 Y.M.deMouraetal./RemoteSensingofEnvironment171(2015)278–290 USDepartmentofState.AdetaileddescriptionoftheLiDARdatacanbe averageofMODIStileh12v09(south-centralAmazon)asanexample. foundathttp://mapas.cnpm.embrapa.br/paisagenssustentaveis.Inorder Wevariedthesolarzenithanglebetween45°(whichisthedefault toallowacomparisonbetweenLiDARbasedLAIandanisotropy,the angleforBRDFnormalizationinMAIAC)and25°degrees(whichis areaofairborneLiDARacquisitionwasfirstsubdividedinto1×1km more commonly found in tropical latitudes). Results shows strong tilesmatchingtheMODISpixels.Theprobabilityofcanopygapwithin seasonalvariationsforredandNIRreflectance(variationsintheblue eachtilewasthendeterminedasthesumofthetotalnumberofhits bandarenotshown),andresultingEVI,irrespectiveofthemodeled downtoaheightz,relativetothetotalnumberofindependentLiDAR SZA.WhilereflectanceandEVIanisotropyincreasedwithincreasing shots(N)(Lovell,Jupp,Culvenor,&Coops,2003;Reading,Bedunah,& SZA,theseasonaldifferenceremainedsimilar.Theseasonalrobustness Amgalanbaatar,2006): withrespecttoagivenSZAmaybeexplainedbythefactthattheregion aroundhotspotanddarkspotareaisrelativelysmooth(Fig.1a,cande), 1−z¼Xzmax#z makingestimatesofseasonalanisotropyrelativelyinsensitivetothe j particularsun-sensorconfigurationselected,aslongasthisconfigura- PgapðzÞ¼ zN¼j ð1Þ tionremainsconstant. Figs.A.2andA.3provideaquantitativeanalysisofthestandard where#zisthenumberofhitsdowntoaheightzabovetheground(or deviation(σ)betweenobservedsurfacereflectanceandBRDFmodel therangetowhichthegapprobabilityistaken).Finally,theleafarea prediction.Fig.A.2showsthebehaviorofthestandarddeviationover profile L(z) was modeled as a logarithmic function of P (Lovell atime,usinganareaof100×100kmareaasanexample(65°0′0″W, gap etal.,2003)assuminganexponentialextinctionoflightwithinthe 5°0′0″S).Themeanstandarddeviationswere0.005and0.019forthe canopy: redandNIRreflectance,respectively,whichisabout10–15%ofthe (cid:2) (cid:3) seasonalchangesillustratedinFig.1aandb.Slightseasonalvariability LðzÞ¼−log PgapðzÞ : ð2Þ inσwasfound,likelyasaresultofincreasedcloudcoverduringthe wetseason.Fig.A.3illustratesthespatialvariabilityofthestandard Adetaileddescriptionofthemethodappliedcanbefoundin(Coops deviationinJune(Fig.A.3a)andSeptember(Fig.A.3b).Forreasonsof etal.,2007). brevityonlythevariabilityinEVIispresented.SimilartoFig.A.2,the standarddeviationbetweenobservedEVIandmodelpredictionpre- 2.3.EstimatingonsetanddurationofAmazondryseasons sentedinFig.A.3wasonaverageabout5–10%oftheobservedseasonal changes(compareFig.1). Themostcommonperiodusedintheliteraturefordescribingdry Fig. 2 illustrates the spatio-temporal variation in anisotropy by seasonsacrossAmazoniaisJunethroughSeptember(Saleskaetal., meansofitsfirstprincipalcomponent(PC1)fortheperiodbetween 2007; Samanta et al.,2010; Xu et al., 2011). It is, however, widely 2000 and 2012. The droughts years 2005 and 2010 were excluded acknowledgedthattheactualonsetanddurationofthedryseason fromthisanalysistoonlyrepresentnormalyearvariations.PC1ex- variesgreatlyacrosstheAmazonBasin(Silvaetal.,2013).Inordertoin- plainedabout89%ofthetotalvarianceinanisotropy;consequently, vestigatetheeffectsofregionalvariabilityinprecipitation,onsetand wefocusonthisfirstcomponenttoillustrateregionalvariabilityin length of dry season were calculated for each year using monthly thedataset.Redandyellowrepresentareaswithrelativelyhigheran- estimatesofwaterdeficitfromprecipitationobtainedfromTropical isotropy,whilegreentowhiteshowareaswithrelativelyloweranisot- RainfallMeasuringMission(TRMM)(3B43v7and7A,at0.25°spatial ropy.NotabledifferenceswerefoundnotonlybetweentheAmazonian resolution).TRMMdatahasbeenextensivelyusedtocharacterizethe rainforestandnon-forestedsavannahregions,butalsowithinthefor- seasonal and inter-annual variability in rainfall across the Amazon estedareaitself(comparetraditionalvegetationindicesforinstancein region (Aragao et al., 2007). Dry season months were determined Hilkeretal.,2014,Fig.S2).Highanisotropywasfoundpredominantly byusingtheassumptionthatmoisttropicalforeststranspireabout inthemoredenselyforestedareasinnorthernandeasternAmazonia 100mm·month−1(Anderson,2012;Aragaoetal.,2007):Whenrainfall whereastheopenforesttypesinthesouthernregionsyielded,onaver- dropsbelow100mm·month−1,evapotranspirationexceedsprecipita- age,lowervaluesofanisotropy.ThepointsymbolsinFig.2illustratethe tion,andsoilwateravailabilitydeclines(Borchert,1998;A.Strahler& plotlocationsforfieldbasedobservationsofLAIandLiDARderived D.Jupp,1990;Williamsetal.,1998). measurements.WhilethetotalnumberofindependentLAIestimates islimitedacrossremoteforestedareassuchastheAmazonBasin,the 3.Results fieldlocationsindicatedinFig.2representedareasonablerangeof foresttypeswithintheAmazonrainforest. TheleftcolumninFig.1showsanexampleofaBRDFsurfacefittedto Measuresofanisotropywerestronglyrelatedtoindependentfield retrieveforwardandbackscatterforredreflectance(a)andNIRreflec- observationsofLAI(n=16)obtainedfromtheliterature(Fig.3a,r2= tance(c).WealsoillustratedaBRDFsurfacecalculatedforEVI(e)to 0.70pb0.05)andLAIestimatesderivedfromairborneLiDAR(Fig.3b, demonstratetheanisotropyofthisindex.TheRTLSsurfacesareshown r2=0.88pb0.05).Importantly,bothrelationshipswerefoundtobe fora1×1kmareaofAmazonforest(65°0′0″W,5°0′0″S)usingall linear,atleastwithintheobservedrangeofLAI≤7m2m−2andyielded observations acquired between January 1 and 14, 2006. The polar animproveddescriptionofstructureindenselyvegetatedareascom- coordinatesrepresenttheviewzenithandazimuthangles,thez-axis paredtoestimatesobtainablefromnadirEVIimagesalone(Fig.3c). showsthecorrespondingreflectance(ρ)intheredandNIRbands, TherelationshipbetweenbothfieldmeasuredandLiDARLAIestimates andEVI,respectively.TheblackdotsrepresenttheMODISobservations withanisotropyfollowedalmosttheidenticallinearfunctionalform, thatwereusedtoparameterizetheBRDFsurface.Theredandbluedots whichallowedustodescribeleafareaacrossarangeofvegetation showthemodeledforwardandbackscatterreflectance(Fig.1,leftcol- typeswithintheAmazonianrainforestfrombothdatasources. umn)withafixedsunobservergeometry(SZA=45°,VZA=35°, Inadditiontoalargeheterogeneityinvegetationstructure(Fig.2), RAA=180°inthebackscatterdirectionandSZA=45°,VZA=35°, ouranalysisconfirmedalsoalargevariabilityinprecipitationacross RAA=0°intheforwardscatterdirection).Wefittedonesuchsurface thebasin(Villaretal.,2009).Fig.4representsestimatesofmonthly foreachpixeland14-dayperiodtoderivebi-weeklyanisotropyacross waterdeficit.Areaswithlowwaterdeficitareshowninblue,whereas theAmazonBasin. red indicates high water deficits. Areas with no water deficit are Fig.1b,dandfshowtemporalvariationsinanisotropy(forred,NIR presentedwithoutcolor.Highlevelsofwaterdeficitwerefoundinthe andEVI,respectively)fordifferentsunobservergeometriestoverify northernAmazonregionmainlybetweenJanuaryandMarch,corre- therobustnessofthemethodapplied.Thetimeseriesshowsaspatial spondingtothedryseasoninthenorthernhemisphere,whereasMay Y.M.deMouraetal./RemoteSensingofEnvironment171(2015)278–290 281 Fig.1.ModeledBRDFsurfacefora1×1kmareaofAmazonforest(65°0′0″W,5°0′0″S)forredreflectance(a),NIRreflectance(c)andEVI(e).TheblackdotsrepresenttheactualMODIS observationsaccumulatedovera14dayperiod,thebluedotrepresentsthemodeledforwardscatterdirection(darkspot),thereddotrepresentsthemodeledbackscatterdirection (hotspot).Fig.1b,dandfshowatimeseriesofanisotropy(red,NIRandEVI,respectively)usingthemeantimeseriesofMODIStileh12v09.Sun.zenithangles(SZA)variedbetween 45°and25°degreestoinvestigatethesensitivitywithrespecttothesunsensorconfiguration. toAugustmarkedthedryseasonmonthsacrosslargepartsofthesouth- changes are labelled “greening”, negative changes are labeled ernhemisphere.Overall,thelargestwaterdeficitwasfoundduringJune “browning”;allchangesarenormalizedwithrespecttotheirstandard andJuly(focusingonthesouth-easternborderoftheAmazon),whereas deviations).Absolutechanges(non-normalized)inanisotropyatbegin- thelowestlevelsofwaterdeficitwereobservedduringMarch,with ningandendofdryseasonarepresentedinA.1.Onlythosechangesthat precipitationexceeding100mmmonth−1almostacrosstheentire exceededtheRMSEofthefieldvalidation(Fig.3aandb)arepresented. basin.Thebeginningandlengthofdryseason(Fig.5aandb,respective- Non-forestedareaswereexcludedfromallanalysisusingtheMODIS ly)variedaccordinglyandfollowedasouth-westnorth-eastgradient land cover product (collection 5, Friedl et al., 2010). Fig. 6a uses a withupto5monthsofwaterdeficitinthesouth-west.Bycontrast, fixeddryseasonassumptionfromJunethroughSeptembertoderive largeareasofAmazonasstate,centralAmazon,showed,onaverage, conventionalmeasuresofgreening/browning.Fig.6bshowschanges no water deficit during the observed years (gray area in the map, ofgreening/browningbasedononsetandlengthofdryseasonderived comparealsoSteege&Pitman,2003). fromwaterdeficit.Inbothcases,thedroughtyearsof2005and2010 Anisotropy changes in response to these seasonal variations in wereexcludedfromtheanalysestoreflect“normalyear”situations.In precipitationareshowninFig.6.Fig.6aandbshowtotaldifferences caseofthefixeddryseasonassumption(Fig.6a),negativenetchanges inanisotropybetweenbeginningandendofthedryseason(positive inanisotropywerefoundlargelyinthewestandnorth-westernregion 282 Y.M.deMouraetal./RemoteSensingofEnvironment171(2015)278–290 Fig.2.Thefirstprincipalcomponentofanisotropybetween2000and2012.Thedroughtsyears2005and2010wereexcluded.ThelocationsofthefieldandLiDARestimatesofLAIare shown.LiDARestimateswereobtainedfromSustainableLandscapeProjectinthreelocations:AdolphoDuckeForestReserve,Amazonasstate,Brazil(⊗);RioBrancomunicipality,Acre State,Brazil( )andTapajósNationalForest,ParáState,Brazil(⦾).TheotherfieldestimatesofLAIwerecollectedfromtheliterature:Malhi,Aragão,Galbraith,etal.(2009),Malhi,Aragão, Metcalfe,etal.(2009)(•),Dominguesetal.(2005)(○),DoughtyandGoulden(2008c)(*),NegrónJuárezetal.(2009)(x),Andreaeetal.(2002)(□),Zanchietal.(2009)(◊),Restrepo- Coupeetal.(2013)(Δ),Figueraetal.(2011)(b),Scurlocketal.(2001)(˃),Galvaoetal.(2011)(+). oftheAmazonBasin,whilesmallgreeningeffectswereobservedinthe moremonthswhenaccountingforactualdryseasononset.Monthly Amapástateregionandsouth-centralAmazonia.Whenconsideringthe changes in LAI (as deviations from annual means) were calculated specificlengthofdryseason(Fig.6b),itbecomesapparentthatmostof usingthelinearrelationshipstofieldandLiDARmeasurementspresent- theareashowingnetbrowningeffectsdidactuallynotexperiencea edinFig.3.ThedashedlineinFig.6canddrepresentestimatesbasedon seasonalwaterdeficit(Fig.6a),atleastonaveragewithinthetime therelationshipwithfieldobservations(Fig.3a),whilethesolidline periodobserved.NetgreeningeffectsshowninFig.6bweresimilarto representsestimatesbasedontherelationshipwithLiDARobservations thosepresentedinFig.6a.However,regionally,considerabledifferences (Fig.3b).Consistentwiththenetchangesinareaofgreeningandbrow- werefound,particularlyinthesouthwesternpartofthestudyarea. ning,ourresultssuggestedthattotalleafareaincreasesduringthe Comparedtothenormalizedresults,non-normalizeddifferencesinan- beginningofthedryseasonbyonaverage0.2m2m−2acrossthebasin, isotropybetweenthebeginningandendofdryseasonwereprominent whileLAIdroppedbelowtheannualmeanafterabout2monthsofdry acrossmostoftheAmazonBasin(A.1),whichcanbeexplainedby season (0.1 m2 m−2). However, these results should be interpreted therelativelysmallRMSEobtainedfromthevalidationdataset(Fig.3a carefullyaschangesinLAIvariedgreatlyacrossspaceandmayalsobe andb). theresultofchangesinotherstructuralparameters(seediscussion). Fig.6canddshownetgreeningandbrowningeffectsinpercentage Thetwodroughtyearsresultedinstrongbrowningeffects;spatial oftotalareapermonthofdryseason.Greeningandbrowningeffects andtemporalpatternsofanomaliesinanisotropyforthe2005and weredefinedaspercentageofpixelswithsignificantincrease/decrease 2010droughtsarepresentedinFig.7.Forbothyears,onlythespecific inanisotropy(≥2σ)comparedtotheannualmean.Fig.6cusesafixed beginningandendofthedryseason(basedonthewaterdeficit)are dryseasonassumption(JunethroughSeptember),whileFig.6dshows shown.Fig.7showsanomaliesi.e.deviationsfromthenormalyearpat- changesbasedononsetofdryseasonderivedfromwaterdeficit.In ternspresentedinFig.6;allanomalieswerenormalizedtothestandard caseofFig.6d,wealsoshowthelastmonthbeforeawaterdeficitwas deviation(≥2σ)oftheyears2000–2012,excluding2005and2010. observed,inordertoillustratechangesinphotosyntheticactivitywith Whilepositiveandnegativeanomalieswereapproximatelybalanced thereduction of rainfall towardstheendof therainy season.Both atthebeginningofthedryseason(Fig.7candd),negativeanomalies estimatesshowedincreasedanisotropyduringthebeginningofthe inanisotropyoutweighedpositiveeffectsafteraboutthreemonths, dryseasonwithabout2%ofareaexperiencing“greening”whenusing especially during 2010, with negative anomalies being three times a fixed dry season assumption and over 5% of total area greening largerthanpositiveones.EstimatesofderivedLAIshowedsmallposi- whenexplicitlyconsideringdryseasononsetforeachpixel.Inboth tiveanomaliesduringthebeginningofthe2005droughtperiod,but analyses,greeningeffectsturnedintonetbrowningeffectsafteran confirmed large negative effects with extended dry season length. extendedlengthofdryseason,reachingabout7%oftheareaafter3or In 2010, change was similar to normal years (anomalies were Y.M.deMouraetal./RemoteSensingofEnvironment171(2015)278–290 283 Fig.3.RelationshipbetweenanisotropyandLAI;a)fromfieldvaluescollectedintheliterature(seeFig.2),andb)fromLiDARestimates.c)Relationshipbetweendirectionallynormalized (nadir)EVIandLAI.ThecorrelationswereperformedusingthedatesdescribedinthefielddatawiththeclosestMODISacquisitionsavailable.ThelocationoftheplotsareprovidedinFig.2. RMSEforFig.3aandbwere0.08and0.02(unitsofanisotropy),respectively. small)duringthefirstfewmonthsofthedryseason.However,areas sourcesfromundetectedcloudstogriddinguncertainties.Itshouldfur- with6monthsofdryseasonshowednegativedeviationfromnormal theraccountforlimitationsoftheRTLSmodeltodescribetheBRDF yeardeclineinstructure(Fig.6d)ofanadditional−0.2m2m−2across shapeandanisotropyoftheMAIACdata. thebasin(Fig.7d). Ourresultssuggestthatstructuralinformationofvegetationmaybe obtainedfrequentlyoverlargeareasfromMODIS.Furtherresearch, 4.Discussion however,willbeneededtoinvestigatepotentialsforotherecosystems andregions.TheBRDFmodelselectedinthisstudyallowedustoderive Thisstudyusedmulti-angleobservationsfromtheMODISinstru- seasonalanisotropyindependentofthesunobservergeometry,which menttoinvestigatespatialandtemporalvariabilityinvegetationstruc- isanimportantconsiderationforseparatingvegetationseasonality tureacrosstheAmazonBasin.Whiletherangeofviewanglesacquired fromartifactsduetoseasonalchangesinthesun/sensorconstellation fromMODISisrelativelysmall(Fig.1),astheinstrumentwasnotspecif- (Mortonetal.,2014). icallydesignedformulti-angleacquisitions,anisotropystillprovidedan While it is acknowledged that vegetation may change over a effectivemeanstocharacterizevegetationstructureacrosstheAmazon 14dayperiod,asusedinourBRDFapproach,thistechniqueshould forest.Changesinthesun-sensorconfigurationovertheyeardonotal- stillallowustoobservemostseasonalityofvegetationandhasprov- waysallowtomodelforwardandbackscatteringobservationswithin en useful in other composite products (Huete et al., 2002; Schaaf thesamplingrangeoftheMODISinstruments.However,theanalysis et al., 2002). Data scarcity may prevent frequent updates of BRDF presentedinFig.1hasdemonstratedarelativerobustnesswithrespect shapes,whichinsomecaseslimittheabilitytodetermineanisotropy totheselectedsun-sensorconfiguration.Thestandarddeviationbe- and may lead to misinterpretation of changes in canopy structure. tweenobservedandmodeledMAIACreflectance(Figs.A.2andA.3) However, analysis of observation frequency across the Amazon wasabout10%oftheobservedvariationinanisotropy(Figs.6and7), Basin(Hilkeretal.,2015)hasshownthatMAIACprovidesonaverage confirmingtheabilityofourapproachtodetectseasonalandinter- between10and60observationsinanygivenmonthfromTerraand annual changes. The results were further within the range of the Aqua, respectively, which should allow stable BRDF inversions for RMSEreportedinFig.3,therebyconfirmingthesignificanceoftherela- most pixels. Other methods to derive multidirectional reflectance tionshiptocanopystructure.Theapproachshouldaccountforerror (Franch,Vermote,Sobrino,&Fédèle,2013;Schaafetal.,2002)have 284 Y.M.deMouraetal./RemoteSensingofEnvironment171(2015)278–290 Fig.4.Monthlyestimatesofwaterdeficit(inmmmonth−1),basedonTRMMobservationsfrom1998to2012.Areaswithlowwaterdeficitareshowninblue,whereastheredcolorin- dicateshighwaterdeficits;areaswithnowaterdeficitarepresentedwithoutcolor. been published. Their usefulness to derive MODIS anisotropy will D.L.B.Jupp,1990).Ourstronglinearrelationshipfoundbetweenanisot- havetobeaddressedseparately. ropyandLAIestimates(Fig.3aandb,r2=0.70,r2=0.88pb0.05, Theabilityofmulti-angleobservationtoderivevegetationstructural respectively)confirmsthesefindings. attributesiswellsupportedbypreviousstudiesoftemperateecosys- TheRMSEfortherelationshipsbetweenanisotropyandfieldobser- tems(Chen etal.,2003;Gao,2003).Multi-angledata decreasethe vationshasallowedustolinkseasonalchangesinanisotropywith dispersionandsaturationingeometricallycomplexcanopies(Zhang, changesinvegetationstructure.However,wedoacknowledgethat Tian,Myneni,Knyazikhin,&Woodcock,2002)andarethereforebetter otherstructuralvariablescaninfluenceseasonalpatternsofanisotropy suitedtodescribethethreedimensionalstructureofforestscompared indifferentways.Forinstance,anisotropyisalsoaffectedbycanopy to mono-angle acquisitions (Chen & Leblanc,1997;A.H.Strahler& roughness(Strahler,2009),leafangledistribution(Roujean,2002) Fig.5.Beginning(a)andlength(b)ofdryseasonacrosstheAmazoncalculatedonperpixelbasisusingmonthlywaterdeficits.Thisapproachwasperformedforeachyearseparatelyin ordertoconsiderinter-annualvariability.Thefigureshowsmeanonsetandlengthofdryseasonforallyears. Y.M.deMouraetal./RemoteSensingofEnvironment171(2015)278–290 285 Fig.6.(a)SpatialdistributionofchangesinanisotropynormalizedbythestandarddeviationusingadryseasonperiodfromJunetoSeptember(forallyearsbetween2000and2012,ex- cept2005and2010).Thegrayregionsrepresentsnodryseasonornon-forestedareas.(b)Spatialdistributionofchangesinanisotropynormalizedbythestandarddeviationusingspecifc beginandendofdryseasonbasedonthewaterdeficitmaps.Figurescanddshowthecorrespondingchangesingreening(bluebars)andbrowning(redbars)bymonthsofdryseason (p=0.05).ThedashedlinesinFigurescanddrepresentthenetchangesinLAI(averagedacrossthebasin)modeledbythelinearrelationshipbetweenanisotropyandLAI(Fig.2a).The solidlineshowsthecorrespondingestimatesbasedonthemodelderivedfromLiDAR(Fig.2b). andfoliageclumping(Chenetal.,2005).Furthermore,theselected of Amazon forests as it increases the sensitivity of optical observa- approachofmodelingLAIfromLiDARdepends,tosomeextent,onfoot- tionstochangesacrossdensevegetationtypes.Thisshouldconsider- printsizeandpointdensity.Whilethetechniqueutilizedherehasbeen ably improve our understanding of Amazon forest seasonality and validatedelsewhere(Coopsetal.,2007,Lovelletal.,2003),differences droughttolerance. inecosystemtypesandLiDARconfigurationmayaffectLAIestimates. ThefindingspresentedinFigs.4and5suggestthatrainfallpatterns Asaresult,thefindingspresentedwithrespecttochangesinleafarea inAmazonianforestsvariedgreatly,causingdifferencesinseasonality shouldbeinterpretedwithcareandshouldbeunderstoodmoreasan acrosstheregion.Estimatesofwaterdeficit(Fig.5aandb),followeda exampleofhowanisotropymaybelinkedtostructuralvariables.Further south-westnorth-eastgradientwithupto5monthsofwaterdeficit analysiswillberequiredwithrespecttochangesinanisotropyasaresult inthesouth-westwhereaslargeareasofAmazonasstateshowed,on ofchangesincanopyroughnessorotherparameters.Nonetheless,the average,nowaterdeficitduringtheobservedyears(compareSteege linearfunctionalformsuggeststhatmulti-angleobservationsmaypro- &Pitman,2003).Previousresearchhassuggestedthatvegetationsea- videanopportunitytoaddresscurrentlimitationscausedbysaturation sonalitymayfollowthisgradientclosely,ashigherprecipitationlevels ofconventional(nadir)vegetationindices(Hueteetal.,2006;Knyazikhin supporthigherleafareasinthewetterregionswhilevegetationin etal.,1998)atleastwithintherangeofobservedLAI(≤7m2m−2). drier areas is limited by available soil water(Myneni et al., 2007). While mono-angle observations have been shown to indicate Wateravailabilitymayfurthercontributetochangesinthespatialdis- levelsof vegetation greenness, theyare less well suitedtodescribe tributionofleaves(Guanetal.,2015;Malhado&Costa,2009;Schurr thethreedimensionalstructureofforestcanopies(Chen& Leblanc, etal.,2012;terSteegeetal.,2006;Wagneretal.,2013).Also,thelength 1997; A.H. Strahler & D.L.B. Jupp, 1990). The selected approach ofdryseasonhasbeenshowntocorrelatewithabovegroundbiomass usinganisotropymayprovidenewinsightsintostructuralvariability (Saatchi,Houghton,DosSantosAlvala,Soares,&Yu,2007)andtree 286 Y.M.deMouraetal./RemoteSensingofEnvironment171(2015)278–290 Fig.7.Spatialdistributionofthestandardizedanomaliesinanisotropyfor2005(a)and2010(b),consideringspecificallybeginandendofdryseason(basedonthewaterdeficitmaps). Thegrayregionsrepresentsnodryseasonornon-forestedareas.Figurescanddshowthecorrespondinganomaliesingreening(bluebars)andbrowning(redbars)bymonthsintodry season(p=0.05).CirclesrepresentsanapproximationoftheepicentersofthedroughtsdescribedbyLewisetal.(2011).ThedashedlinesinFigurescanddrepresenttheanomaliesinLAI (averagedacrossthebasin)modeledbythelinearrelationshipbetweenanisotropyandLAI(Fig.2a).Thesolidlineshowsthecorrespondingestimatesbasedonthemodelderivedfrom LiDAR(Fig.2b). speciescomposition(terSteegeetal.,2006;Wagneretal.,2014).Esti- observationstodeterminewhetherachangeissignificantornot,but matingvegetationseasonalityfromwaterdeficitprovidedasimple didnotnormalizebythestandarddeviation.OurresultsinFig.A.1 buteffectiveapproachtocapturethisregionalvariabilityinprecipita- showedmuchincreaseseasonalitywhenusingtheRMSEoffieldand tion.Otherapproaches,forexample,basedonavailablephotosyntheti- LiDAR data (Fig. 3a and b) to determinesignificance. Also, Myneni callyactiveradiation(PAR)arepossibleandmayresultindifferent etal.(2007)calculatedseasonalityasthedifferencebetweenthemax- definitionofdryandwetseason. imum4-monthaverageLAIinthedryseasonminustheminimum Whileourfindingsareingoodagreementwithpreviousreports 4-month average LAI in the wet season for those regions with dry (Steege&Pitman,2003),itshouldbenotedthatinthenorthernpart seasonslongerthan3months.Forallotherregions,theycalculated oftheAmazon,cloudcoverisconsiderablyhigherthaninthesouth, seasonalityasthedifferencebetweenthedry-seasonaverageLAIand whichmayincreasemeasurementnoiseintheTRMMdataandcontrib- theminimum4-monthaverageLAIinthewetseason. utetolargerspatialvariationinonsetandlengthofdryseasonobserved. Theseasonalityinanisotropy(Figs.1,6)cannotbeexplainedbydi- ThefindingsprovidedinFigs.2,A.1and6suggestlargespatialand rectionaleffects,asallobservationshavebeennormalizedtoafixedfor- seasonalvariabilityofAmazonianforests.Thisresultrelateswellto wardandbackscattergeometry(Lyapustin,Wang,Laszlo,&Hilker, previousstudiesonvegetationstructureandseasonalityoftheAmazon 2012).OppositefindingsbasedonconventionalMODISdata(Morton (terSteegeetal.,2006;Villaretal.,2009)aswellasestimatesofabove- etal.,2014)willrequirefurtheranalysistobeaddressedseparately. groundcarbon(Saatchietal.,2011).Considerationofthisvariability Onepossibleexplanationmightbenoiseinthedataset(Hilkeretal., willbecriticalforinterpretingthebiophysicalresponsesofvegetation 2012)renderingresidualchangesbelowastatisticalsignificancelevel. tochangesinclimate. Changes in anisotropy (greening/browning) during the dry season TheresultspresentedinFigs.6and7confirmseasonalswingsinthe (Figs.A.1,6and7)coincidedwellwithpreviousreportsonAmazonsea- Amazon(Mynenietal.,2007).Whilethetotalareawithsignificant sonality.Theresultssupporttheviewthatphotosyntheticactivityini- change(Fig.6)isrelativelysmallcomparedtoMynenietal.(2007), tiallyincreasesduringthedryseasoninresponsetoanincreasein thesevariationscanbeexplainedbythedifferenceinmethodsapplied. incidentPAR(Brandoetal.,2010;Grahametal.,2003;Hueteetal., First,Mynenietal.(2007)usedtheRMSEoftherelationshiptofield 2006;Hutyraetal.,2007;Malhi,Aragão,Galbraith,etal.,2009,Malhi,

Description:
dx.doi.org/10.1016/j.rse.2009.08.016. Fu, R., Yin, L., Li, W., Arias, P. A., Dickinson, R. E., Huang, L., & Chakraborty, S. (2013). Increased dry-season
See more

The list of books you might like

Most books are stored in the elastic cloud where traffic is expensive. For this reason, we have a limit on daily download.