Classifying Vineyards from Satellite Images: a Case Study on Burgundy’S Côte D’Or
Total Page:16
File Type:pdf, Size:1020Kb
04-ducati_05b-tomazic 08/01/15 21:42 Page247 CLASSIFYING VINEYARDS FROM SATELLITE IMAGES: A CASE STUDY ON BURGUNDY’S CÔTE D’OR JorgeR.DUCATI1,2,4,*,MagnoG.BOMBASSARO1 andJandyraM.G.FACHEL3 1:CentroEstadualdePesquisasemSensoriamentoRemotoeMeteorologia,UniversidadeFederaldoRioGrande doSul,Av.BentoGoncalves9500,CEP91501-970,PortoAlegre,Brazil 2:DepartamentodeAstronomia,InstitutodeFísica,UniversidadeFederaldoRioGrandedoSul, Av.BentoGoncalves9500,CEP91501-970,PortoAlegre,Brazil 3:DepartamentodeEstatística,InstitutodeMatemática,UniversidadeFederaldoRioGrandedoSul, Av.BentoGoncalves9500,CEP91501-970,PortoAlegre,Brazil 4:Visitingprofessor(2011),ÉcoleSupérieured‘Agricultured’Angers,GroupeESA,55rueRabelais, 49007Angers,France Abstract Résumé Aim:TouseRemoteSensingimageryandtechniquesto Objectif :Différencierlescatégoriesdeparcellesdu differentiatecategoriesofBurgundianvineyards. vignoblebourguignonparl’utilisationd’imagessatellites. Methods and results :Asampleof201vineplotsor Méthodes et résultats:Unéchantillonde201parcellesou “climats”fromtheCôted’OrregioninBurgundywas “climats”delaCôted’OrenBourgogneaétésélectionné, selected,consistingofthreevineyardcategories(28Grand formépartroiscatégoriesdevignobles(28GrandCru, Cru,74PremierCru,and99Communale)andtwogrape 74 PremierCruet99Communale)etdeuxcépages(Pinot varieties(PinotnoirandChardonnay).Amaskformedby noiretChardonnay).Unmasquecomposéparles thepolygonsofthesevineplotswasmadeandprojectedon polygonesdecesparcellesaétéconstruitetensuiteprojeté foursatelliteimagesacquiredbytheASTERsensor, surquatreimagessatellitesdelarégion,collectéesparle coveringtheCôted’Orregioninyears2002,2003(winter senseurASTERen2002,2003(enhiver),2004et2006. image),2004and2006.Meanreflectanceswereextracted Lesréflectancesmoyennespourlespixelsàl’intérieurde frompixelswithineachpolygonforeachofthenine chaquepolygoneontétécalculéespourchaqueannéeet spectralbands(visibleandinfrared)coveredbyASTER. pourchacunedesneufbandesspectrales(visibleet Thedatabasehadatotalof797reflectancespectra infrarouge).Labasededonnéesestforméepar797spectres assembledoverthefourimages.Statisticaldiscriminant deréflectance.Desanalysesdiscriminantesdupourcentage analysisofpercentageclassificationaccuracywasmade d'exactitudeduclassementontétéfaitesséparémentpour separatelyforCôtedeNuitsandCôtedeBeaune,andfor CôtedeNuitsetCôtedeBeauneetpourchaqueannée.Les eachyear.Resultsshowedthatforindividualyearsand résultatsmontrentquelaclassificationauniveaudes Côtes,classificationaccuracyforvineyardcategorywasas catégoriesdevignobleaétépréciseentre66,7 %(Beaune highas73.7 %(Beaune2002)andaslowas66.7 % 2003)et73,7 %(Beaune2002).Aucunedifférence (Beaune2003).Therewerenosignificantdifferencesin significativedanslaprécisionn'aététrouvéeentreles accuracybetweenspring,summerandwinterimages. saisonsdel’annéedesimages(printemps,étéethiver). ClassificationaccuracyforgrapevarietyinCôtedeBeaune PourlaséparationdescépagessurlaCôtedeBeaune,la overthefourstudyyearswasbetween73.5 %forPinot précisiondel’analysediscriminanteaétéentre73,5 %pour noirclimatsin2004and91.9 %forChardonnayclimatsin lesparcellesdePinotnoiren2004et91,9 %pourles 2006,includingthewinterimage.Concerningthe parcellesdeChardonnayen2006,comprisel’image vegetationindexNDVI,therewerenosignificant d’hiver.Pourl’indicedevégétationNDVI,aucunes differencesbetweenvineyardcategories. différencessignificativesentrelescatégoriesn’ontété Conclusions:Satellitedataisshowntobefunctionalto trouvées. revealvineyardquality.Spectraldifferencesbetween Conclusion:L’analysespectraledesdonnéessatellitaires categoriesofBurgundianvineyardsareatleastpartially peutdonneruneindicationdelaqualitéd’unvignoble duetoterroircharacteristics,whicharetransmittedtovine bourguignon.Lesdifférencesspectralesentrelescatégories andvinecanopy. devignoblessontduesaumoinsenpartieauxpropriétésdu Significance and impact of the study :Thiswork terroirtransmisesàlavigne. indicatesthatRemoteSensingtechniquescanbeusedasan Signification et impact de l’étude:Cetravailmontreque auxiliarytoolforthemonitoringofvineyardqualityin latélédétectionpeutêtreunoutilsupplémentairepour establishedviticulturalregionsandforthestudyofquality l’observationderégionsviticolesétablies,etaussipour potentialinnewregions. l'étudedelaqualitépotentielleattenduedansdenouvelles Key words :Burgundianclimats,RemoteSensing, régions. vineyardsspectra,leafreflectance,satelliteimages Mots clés:climatbourguignon,télédétection,spectredes vignobles,réflectancedesfeuilles,imagessatellites manuscript received 9th September 2014 - revised manuscript received 19th July 2014 J. Int. Sci. Vigne Vin , 2014, 48, 247-260 *Correspondingauthor:[email protected] - 247 - ©Vigne et Vin Publications Internationales (Bordeaux, France) 04-ducati_05b-tomazic 08/01/15 21:42 Page248 JorgeR.DUCATI et al. INTRODUCTION techniquesallowsnotonlytheseparationof vineyardsfromothervegetation,butalso,toacertain TheobservationofEarthfromremoteplatformslike degree,theidentificationofgrapevarieties(Cemin airplanesorsatelliteshasprovedtobeapowerful andDucati,2011).Thesepossibilitieswerealready resourceforlandstudies,withapplicationsto perceivedfromlaboratorymeasurements(Lacaret geology,agriculture,environmentalsciences,urban al.,2001),butnowitbecomesclearthatsatellite andmarinemonitoring,andmanyotherfields. imageshavetheirownpotentialinviticultural Presently, most of these Remote Sensing studies.Afterusingsatelliteimagestostudy investigationsareperformedusingdigitalimages vineyardsinFrance(Bordeaux,Champagne,Loire), collectedfromsatellites,whichprovidelow-cost ChileandBrazil,wenowfocusourstudieson data,withtheadvantageofbeingre-acquiredatnew Burgundy´sCôted’Or.Thischoiceisjustifiedby over-flights.ThetypicalRemoteSensingprocess threebasicfactors: involvescamerasandsensorsaboardthesatellite, whichcollectsunlightreflectedfromtheEarth’s a)ThehierarchicaldivisionoftheBurgundian surface;duringreflectionbytypesorclassesof vineyardishistoricalandemblematical,havingbeen surfacecover,likesoilorvegetation,thesolar theobjectofcountlessstudies,butuptothepresent spectrumundergoesmodifications.Theresulting dayfewpapers,ifany,haveusedobservationsfrom reflectancespectracarrycharacteristicfeaturesofthe space; classes present in the imaged surface and identificationoftheseclassesispossible.For b)ThetypicalsizeofvineparcelsinBurgundyisof example,reflectancespectrafromplantsare theorderoffewhectares,beingadequatelyresolved characterizedbylowreflectanceinvisiblelight,with bymultispectralimageslikethosefromASTER apeakat550nmduetochlorophyll,whichisthe sensor; reasonforthegreencolorofvegetation;atnear- infrared(NIR)wavelengthsthereisanabrupt c)TheCôted’Orregionisgenerallyorientedfacing transitiontowardsstrongerreflectances(theRed east(PitiotandServant2010;Atkinson2011),andso Edge);andatlongerwavelengths(theShort-Wave mostvineyardsreceivethemorningsunlightinfairly Infrared,SWIR)reflectancefalls,carryingthetypical equalinclinationsofsolarrays.Thisfactisrelevant featuresofabsorptionbywaterat1,400and sincetheASTERimagergetsdatainthemorning 1,900 nm.Thespectralsignaturesofmineralsor (around10h30AM).Theilluminationofparcels, waterarequitedifferent,andthisallowsthe whichingeneralareongentleslopes,tendstobe identificationofclassesofsoilandlandcoverin homogeneous;thisperceptionwasgainedduring RemoteSensingimageswhichhavetheadequate severalfieldtripstotheregioninthelastyearsbythe spectralsensitivity;forcomprehensivereviewson firstauthor. theapplicationsofRemoteSensingimagerytoland InBurgundy,thehierarchyofGrandCru,Premier monitoring,seeJensen(2007)orCampbelland Cruandmoregenericappellations(Côtes,Villages, Wynne(2011). Communales,etc.)seemstobelinkedtosoil Thisworkdealswithreflectancespectraofvineyards characteristics,whichareattheveryrootofthe asaparticularclassofvegetation.Applicationsof terroirconcept(VanLeeuwenandSeguin,2006). RemoteSensingtechniquestovineyardstudiesare Fromaprincipalcomponentanalysis,Wittendal stillintheirinfancy.Upuntilnow,themajorityof (2004)gaveweighttoawidespreadperception, studieshavefocusedonprecisionviticulture indicatingthatmostGrandCrusoilshaveaparticular managementinprivatepropertiesoflimitedsurface structurethatissignificantlydifferentfromthesoils and,forthisreason,arebasedonairbornesensors, ofothercategories.Therefore,theobjectiveofthis eithermultispectralorhyperspectral(Bramleyand investigationwastoverifyifthesequalitycategories, Proffitt,1999;Zarco-Tejadaet al.,2005).Remote whicharetransmittedfromsoiltowine,arealso Sensingimageryfromsatellitescoversmuchlarger transmittedfromsoiltovineleavesandiftheycanbe areasandissuitableforregionalsurveysand detectedinthespectralinformationcontainedinthe monitoring.Thisfieldofresearchisnewandmuch images.Thisisbecausetheobservationparameterin groundbreakingworkhastobedone.Inthisaspect, digitalimages,thereflectance,originatesmainlyfrom wereportedinaseriesofpapersstudiesperformed vineleavesreflectingsunlight,ifweareusingnon- overseveralviticulturalareasinEuropeandSouth winterdata.Atthehighplantdensityusedin America(DaSilvaandDucati,2009;Blauthand Burgundy(upto10,000vines/hectare),thesoilis Ducati, 2010 ; Ducati et al., 2014). It was almostentirelycoveredbytheplantcanopy;besides, demonstrated that Remote Sensing data and atthemomentofimageacquisition(10h30AM), J. Int. Sci. Vigne Vin , 2014, 48, 247-260 ©Vigne et Vin Publications Internationales (Bordeaux, France) - 248 - 04-ducati_05b-tomazic 08/01/15 21:42 Page249 thereisanimportantprojectionofshadowbetween than40pixelsof225 m2 each),withadequate vinerowsandlittlesunlightisreflectedfromthe geometry(themoresquare,thebetter),andevenly