• Non ci sono risultati.

environments a before–after with control-impact statisticaldesign Remote monitoring sensing of land restoration interventions insemi-arid International Journal of Applied Earth Observation andGeoinformation

N/A
N/A
Protected

Academic year: 2021

Condividi "environments a before–after with control-impact statisticaldesign Remote monitoring sensing of land restoration interventions insemi-arid International Journal of Applied Earth Observation andGeoinformation"

Copied!
11
0
0

Testo completo

(1)

ContentslistsavailableatScienceDirect

International Journal of Applied Earth Observation and Geoinformation

j ou rn a l h o m epa 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

Remote sensing monitoring of land restoration interventions in

semi-arid environments with a before–after control-impact statistical design

Michele Meroni

a,∗

, Anne Schucknecht

a

, Dominique Fasbender

a

, Felix Rembold

a

, Francesco Fava

b

, Margaux Mauclaire

c

, Deborah Goffner

d

, Luisa M. Di Lucchio

e

, Ugo Leonardi

f

aEuropeanCommission,JointResearchCentre,DirectorateDSustainableResources,FoodSecurityUnit,ViaFermi2749,21027Ispra,VA,Italy

bInternationalLivestockResearchInstitute,P.O.Box30709,00100Nairobi,Kenya

cUniversityofBordeaux3,LabexDRIIHMandLesAfriquesdanslemonde(LAM),IEPdeBordeaux,alléeAusone11,Domaineuniversitaire,33607Pessac Cedex,France

dFrenchNationalCentreforScientificResearch,CNRS/UMIn3189Environment,HealthandSocieties,BdPierreDramard51,13344MarseilleCedex15, France

eUniversityofCopenhagen,DepartmentofGeosciencesandNaturalResourceManagement,Rolighedsvej23,1958Frederiksberg,Denmark

fFoodandAgricultureOrganizationoftheUnitedNations,SomaliaWaterandLandInformationManagementProject,P.O.Box30470-00100,Nairobi, Kenya

a r t i c l e i n f o

Articlehistory:

Received11January2017

Receivedinrevisedform17February2017 Accepted20February2017

Availableonline16March2017

Keywords:

Restorationinterventions Biophysicalimpact Landsat

MODIS

BACIsamplingdesign

a b s t r a c t

Restorationinterventionstocombatlanddegradationarecarriedoutinaridandsemi-aridareasto improvevegetationcoverandlandproductivity.Evaluatingthesuccessofaninterventionovertimeis challengingduetovariousconstraints(e.g.difficult-to-accessareas,lackoflong-termrecords)andthe lackofstandardisedandaffordablemethodologies.Weproposeasemi-automaticmethodologythatuses remotesensingdatatoprovidearapid,standardisedandobjectiveassessmentofthebiophysicalimpact, intermsofvegetationcover,ofrestorationinterventions.TheNormalisedDifferenceVegetationIndex (NDVI)isusedasaproxyforvegetationcover.Recognisingthatchangesinvegetationcoverarenatu- rallyduetoenvironmentalfactorssuchasseasonalityandinter-annualclimatevariability,conclusions aboutthesuccessoftheinterventioncannotbedrawnbyfocussingontheinterventionareaonly.We thereforeuseacomparativemethodthatanalysesthetemporalvariations(beforeandaftertheinter- vention)oftheNDVIoftheinterventionareawithrespecttomultiplecontrolsitesthatareautomatically andrandomlyselectedfromasetofcandidatesthataresimilartotheinterventionarea.Similarityis definedintermsofclasscompositionasderivedfromanISODATAclassificationoftheimagerybefore theintervention.Themethodprovidesanestimateofthemagnitudeandsignificanceofthedifferencein greennesschangebetweentheinterventionareaandcontrolareas.Asacasestudy,themethodologyis appliedto15restorationinterventionscarriedoutinSenegal.Theimpactoftheinterventionsisanalysed using250-mMODISand30-mLandsatdata.Resultsshowthatasignificantimprovementinvegetation coverwasdetectableonlyinonethirdoftheanalysedinterventions,whichisconsistentwithindepen- dentqualitativeassessmentsbasedonfieldobservationsandvisualanalysisofhighresolutionimagery.

Ruraldevelopmentagenciesmaypotentiallyusetheproposedmethodforafirstscreeningofrestoration interventions.

©2017TheAuthor(s).PublishedbyElsevierB.V.ThisisanopenaccessarticleundertheCCBYlicense (http://creativecommons.org/licenses/by/4.0/).

∗ Correspondingauthor.Tel.:+390332786429;fax:+390332785162/9029.

E-mailaddress:michele.meroni@ec.europa.eu(M.Meroni).

1. Introduction

Desertification,definedaslanddegradationinarid,semi-arid anddrysub-humidareasresultingfromvariousfactors,including climate variation and human activities (UNCCD, 1994), repre- sentsamajorthreattopopulationsandecosystems(Low,2013;

Reynoldsetal.,2007).Besidesphysicallyaffectingecosystems,land http://dx.doi.org/10.1016/j.jag.2017.02.016

0303-2434/©2017TheAuthor(s).PublishedbyElsevierB.V.ThisisanopenaccessarticleundertheCCBYlicense(http://creativecommons.org/licenses/by/4.0/).

(2)

degradationcausesvarioussocio-economicproblems,suchasfood insecurityandconflicts(Mbowetal.,2015).Restorationinterven- tionsareamongthestrategiesthatcanbeputinplacetocombat landdegradation.Restorationactionsofteninvolvetheimprove- mentofvegetationcover(Zuccaetal.,2015),throughtheplanting ofappropriatespecies(e.g.Niangetal.,2014)orthroughimproved soil,waterandlandmanagement.

Thedefinition of “effectiveness”of a restoration action may coverdifferentaspectsoftheintervention,rangingfromthepurely biophysicaltotheecologicalandsocio-economicones(Shackelford etal.,2013).Withrespecttothebiophysicalimpact,guidelines for the ecological evaluation of restoration interventions focus onthecomparison betweentherestoration and referencesites for a number ofattributes measuredin thefield, rangingfrom speciescomposition,toecosystemfunctionandstability,andto landscapecontext(SocietyforEcologicalRestorationInternational Science&PolicyWorkingGroup,2004).Althoughcomprehensive, thisapproachisexpensiveandrequiresextensivefieldoperations.

Independentassessmentofthesuccessofrestorationprojects isoftenchallengingbecauseinterventionsmaybelocatedinareas thataredifficulttoaccessandhavepoorinfrastructure.Additional challengesrefertothelackofaffordableandstandardisedmethod- ologies/criteriaandthedifficultyofobtaininglong-termdatato monitortheeffectofaninterventionoutsidetheproject’stimes- pan. Verification performed by theimplementing agent is also frequentlynotavailable.Forexample,inarecentsurveyofrestora- tionprojectsintheMediterraneanBasinconductedbyNunesetal.

(2016)among restoration professionalpractitioners, restoration successwasnotevaluatedin22%oftheprojectsandevaluatedonly inthefirstyearaftertheplantationin19%oftheprojects.When conducted,theevaluation wasbasedonplantcover and diver- sity(69%oftheprojects)andplantvitality(48%).Lackoffunds, togetherwithcapacity constraintsandlackofknowledge,were identifiedasobstaclestoprojectmonitoringbyrestorationpracti- tionersinSouthAfrica(Ntshotshoetal.,2015)andcanbeassumed torepresentcommonlimitationsinotherruralareasacrossthe continent.

The lack of evaluation and dissemination of the results of restorationstillrepresentaconstrainontheapplicationofthebest technologiesandapproachesavailable(Bautistaetal.,2010).Asa results,thereiswidespreadconsensusontheneedforinnovative approachesfor thesystematic evaluationoftheeffectivenessof restorationactions(Bautistaetal.,2010;Benayasetal.,2009;Birch etal.,2010;Papanastasisetal.,2015).

Remotesensing(RS)canhelpcopewiththewidespreadlack oftimely,long-term,reliable,andhomogeneousgroundinforma- tion,especiallyinAfricandrylands.FewexamplesoftheuseofRS datatoassessrestorationinterventionsareavailable.TheFoodand AgricultureOrganisation–Somalia WaterandLandInformation Managementproject(FAO-SWALIM)uses commercialvery high resolution(VHR)imagerytovisuallyappraisetheimplementation ofsurfacerun-offcontrolinfrastructuresinSomalia(e.g.rockdams, gabions,watercatchments)operatedbyvariouscontactors(FAO, 2015).Inthisway,however,itistheimplementationoftheinfras- tructurethatisscrutinised,notitsimpactorsuccesswithrespect tovegetationdynamics.Photointerpretationoftimeseriesofaerial photographywasusedbyRangoetal.(2002)toqualitativelyeval- uatethe long-termeffectiveness of restoration interventionsin NewMexicointermsofpersistencyintimeofrecognisablestruc- turessuchasterraces,grubbingpatterns,revegetatedareas,etc.

Recently,theOpenforisinitiativeoftheFAOprovidedafreeand open-sourcetool,namedCollectEarth,whichfacilitatesthevisual interpretationof VHR time series imageryof GoogleEarth and MicrosoftBingforpointsamplingandlandusechangedetection (Beyetal.,2016).Despitetheirusefulness,theresultsoftheanal- ysisare pronetointerpretationerrors asall ofthese examples

make useofphotointerpretation.A quantitativeevaluation ofa restorationinterventionusingAtriplexnummulariaplantationsin MoroccowasinsteadperformedbyZuccaetal.(2015) utilising SPOT5imageryandground-basedbiomassmeasurementstoderive thedrybiomassyieldoftheplantationsinMoroccoascompared toknownreferences.Landcoverclassificationandspatialpattern metricshavebeenanalysedbyFavaetal.(2015)tostudytheimpact ofrestorationactionsinMediterraneanrangelands.

VegetationindicessuchastheNormalisedDifferenceVegeta- tionIndex (NDVI;Rouseet al.,1974)canbeusedasproxies to monitorthefractionofvegetationcover,i.e.thefractionofground coveredbygreenvegetation(CarlsonandRipley,1997).However, evaluatingthe“greening”of arestoration intervention presents a challenge, becausethe direct comparison of the NDVI of the areabeforeandaftertheinterventionwouldnotbeinformative.

Infact,vegetationcoverwillchangeovertimeindependentlyof therestorationproject.Twomainsourcesdrivethetemporalvari- abilityofvegetationstatus:theannualseasonaldevelopmentcycle (oneormore)andtheinter-annualclimatevariability.Bothfluc- tuations hamperthepossibilityof makinga direct comparison.

In fact,even in theabsenceof disturbances(e.g.fires,pests), a differencein NDVIbetweentwo observations takenbeforeand aftertheinterventioncouldbeduetotheinterventionitself,the stageofdevelopmentofthevegetationatthoseparticulartimesof observation,andtheweatherconditionsexperiencedbythevege- tationintheweeks/monthsprecedingtheobservations.Assuming thatclimaticconditionsareratherhomogeneousintheneighbour- hoodoftherestorationproject,theproblemcanbeapproachedby comparingtheconditionsoftherestorationareabeforeandafter theinterventionwiththoseofsimilarareasnearby,asin Zucca etal.(2015).Therationaleisthattheanthropogenicintervention willcauseadifferentpatternofchangefrombeforetoafterthe interventioncomparedwithnaturalchangesinundisturbedand similarareas.Thisconceptformsthebasisofthebefore/aftercon- trol/impact(BACI)samplingdesign(Underwood,1992),originally developed inecology toassesstheimpactof astress (typically inducedbyindustrialactivities)ontheenvironment.BACIhasbeen successfullyappliedtostatisticallyevaluatepotentialenvironmen- talandecologicalimpacts(Smith,2002),buthasnotbeenusedby theRScommunitysofar.

InthisstudywemakeuseoftheBACIdesigntodevelopamethod toassesstheimpactofa restorationinterventiononvegetation fractionalcover solely basedonRS information(i.e.NDVI).The methodisintendedtoperformacost-effectiveverificationofthe effectivenessoftherestorationinterventionthatmaybeusedas afirstscreening,usabletoplanadditionalfieldverificationcam- paigns, andas amedium- tolong-termimpactmonitoringtool whenappliedrepeatedlyovertime.It isacknowledgedthatthe proposedmethodissuitedtorestorationinterventionsthatinvolve anincreaseinvegetationcover,whichisnotthecaseforanum- berofinterventiontypes(e.g.agreenlandscapeofinvasivespecies wheretherestorationwouldaimtochangetheplantcommunity composition;soilconservationmeasuressuchasrockdamstostop gullyerosion).

ToillustratetheapproachweapplyittoacasestudyinSene- gal,whereanumberofrestorationinterventionswereperformed inthecontextoftheGreatGreenWallfortheSaharaandtheSahel Initiative(GGWSSI), apan-Africaninitiativetocombatdesertifi- cation (African Union&Pan-AfricanAgency oftheGreat Green Wall,2012).ThebiophysicalimpactwasassessedusingRSdata attwodifferentspatialresolutions,namelytheModerateResolu- tionImagingSpectroradiometer(MODIS)at250mandLandsatat 30m,andcomparedwithqualitativeinformationfromfieldobser- vationsandphotointerpretationofVHRimagery.Theprosandcons ofusingMODISandLandsatdataarediscussed.

(3)

2. Studyarea

The test case-study encompasses severalinterventions con- ducted in theLinguère departmentof the Lougaregion and in theRaneroudepartmentoftheMatamregionofnorthernSene- gal(Fig.1).TherelativelyflatstudyareabelongstotheSahelian acacia savannah ecoregion (Olson et al., 2001), and is charac- terisedbyahotariddesertclimate(BWh)accordingtotheupdated Köppen–Geiger climate classification (Peel et al., 2007). Mean annualtemperatureandprecipitationinthestudyarearangefrom 27to28C(ECMWFERA-Interimovertheperiod1990–2014;Dee etal.,2011)andfrom270to390mm(CHIRPSrainfallestimates overthesameperiod;Funketal.,2015).Themajorityofprecipita- tionfallsduringtherainyseason,whichoccursbetweenJulyand September,andisrelatedtotheWestAfricanMonsoon(Nicholson, 2013).In thearea, severalrestoration projects, includingrefor- estationandimprovedforageproduction,havebeenimplemented between2007 and 2011 in the context of the GGWSSI by the GreatGreenWallagencyundertheresponsibilityoftheSenegalese MinistryofEnvironment.However,thetechnicalrationaleforthe selectionofprojectsandthecompletedescriptionoftheprojects (where,what,how,successrate,etc.)is,toourknowledge,not available.

3. Data

3.1. Remotesensing

Theanalysiswasperformedonfreelyavailablesatelliteimagery attwodifferentspatialscales:250-mMODISNDVIproductand30- msurfacereflectancesfromtheLandsatmissions.Forthemoderate resolution,weusedtheeMODISproductprovidedbytheUnited StatesGeologicalSurvey(USGS)andbasedonMODISdataacquired bytheTerra satellite. Theproductis a 10-day maximumvalue NDVIcomposite(Jenkersonetal.,2010)temporallysmoothedwith theSwetsalgorithm(Swetsetal.,1999).Compositesareproduced everyfivedays,resultinginsixtemporallyoverlappingcompos- itespermonth.Hereweonlyusedthecompositesfordays1–10, 11–20,and21-lastdayofeachmonth.Boththetimeseriesof10-day observationsandthemaximumannualNDVIvalue,representing vegetationpeakdevelopment,wereusedintheanalysis.

Inspectionof MODIS multi-annual temporal profiles for the interventionareaspermittedustodeterminetheperiodofvegeta- tiongrowth,whichroughlyrangesfromJunetoSeptember,with maximumdevelopmentreachedinlateAugust.Cloud-freeLand- satimagerywasselectedduringthisperiod.Landsat8Operational LandImages(OLI)dataareavailablesince2013;beforethenwe hadtorely onLandsat 5Thematic Mapper(TM)andLandsat 7 EnhancedThematicMapperPlus(ETM+)data.However,Landsat7 ETM+imagerycollectedafter31/05/2002hasdatagapsduetothe ScanLineCorrectorfailure(SLC-off;Andrefouetetal.,2003).The issuedoesnotpreventtheanalysisbuthastobeproperlytreated, asexplainedinthemethodssection.

Althoughnotstrictlyrequired,theBACIdesign benefitsfrom havingmultipletime observations beforeandafter thetime of intervention.WhereasgatheringmultipleMODISobservationsis straightforward,itisverychallengingforLandsat5and7inthese geographicalsettingswheretheavailabilityofcloud-freeimages duringthegrowingseasonisverylimited.Forinstance,alargedata gapexistsbetween2003and2007andbetween2007and2012, whennotasinglecloud-freeimageisavailableintheperiodof maximumvegetationdevelopment.ThelistoftheLandsatimages usedintheanalysisispresentedinTable1.

Landsat-based NDVI was computed using the red and near infraredbandsofsurfacereflectanceproducts(USGS,2016a,2016b)

Table1

AcquisitiondateandsensorofLandsatdataused(path204androw49).

Sensor Date

Landsat7ETM+ 19/07/2003

Landsat7ETM+ 16/09/2007

Landsat7ETM+ 13/09/2012

Landsat8OLI 24/09/2013

Landsat8OLI 11/09/2014

retrievedfromtheUnitedStatesGeologicalSurvey.Largelycloud- freeimagerywasselected,and theCFmaskbandofthesurface reflectanceproduct wasused tomask sparseclouds and cloud shadows.

Insummary,MODISandLandsat-basedanalysesdifferinthree aspects:i)thespatialresolution(250mvs.30m),ii)theRSvari- ableused(maximumseasonalNDVIvs.NDVIataspecific,anddata availability-driven,dateduringtheseason),andiii),thetemporal periodcoveredbeforeandaftertheintervention(uptofiveyears ofacquisitionsvs.asingleacquisition).

Finally,tochecktheconsistencyofBACIresults,VHRimagery fromGoogleEarth(GE)wasusedforthequalitativeandvisualeval- uationoftherestorationinterventions.ThevisualanalysisofVHR imagerybeforeandaftertheinterventiondateaimedtospotsigns ofinterventions,rangingfromsignsoftractorploughingtovisible patternsofregularplantationsandthegrowthofnewtrees.When imagerybeforetheinterventionwasnotavailableinGE(8cases outof15),theassessmentoftheinterventionwasperformedon theimageryonlyaftertheintervention,andwasbasedonacom- parisonofthevegetationcoverinsidetheinterventionareawith thatoftheareaoutside,withobviouslimitationsonthepossible interpretation.

3.2. Fieldmissionsandanalysedinterventions

Theoutlineoftheprojectpolygonsandthemainprojectinfor- mation(typeandyearofintervention)wereobtainedduringthree fieldvisits(2014–2015)performedbytheFrenchNationalCentre forScientificResearch(CNRS)andsupportedbytheSenegalState ServiceofWaterandForests.Asacentralisedandpublicrecord ofrestorationprojectsdoesnotexist,thelocationoftheinterven- tionprojectstobevisitedwasdefinedwiththestaffoftheSenegal StateServiceof WaterandForestsand theSenegaleseNational GreatGreenWallAgency.Thispreliminaryinformationwascom- plementedby visualinterpretationusing VHRsatelliteimagery fromGEbeforethefieldcampaignsandinterviewswithlocalcom- munitiesduringthecampaigns.Projectareaswerethendelineated inthefieldusingGPS.

Restoration interventions mainly involved tree plantations (Acacianilotica,Acaciasenegal,AcaciaseyalandBalanitesaegypti- aca),thefencingofplotstoenhancethenaturalregenerationof woodyspeciesandrestorerangeland grasses,andthecombina- tionofthetwo.TreeplantingusuallyoccurredinAugust,during therainyseason.Activitiesweredesignedtoimprovelandproduc- tivityoverthelongrununderthehypothesisthattheincreasein vegetationcoverduetotheinterventionwouldrestoresoilfertility andatthesametimeproviderelevantecosystemservicesforlocal communities(e.g.gumarabicproductionfromAcaciaSenegal,fruits fromBalanitesaegyptiaca,andgrassstrawtobeharvestedattheend oftheseasonandeitherusedorsold).Restorationinterventions wereimplementedbytheSenegaleseNationalGreatGreenWall Agencywithintheframeworkofacash-for-workprogramme.Itis notedthatoneoftheinterventionsconsideredinthetestcase(i.e.

projectno.81ofFig.1)doesnotbelongtotheGGWSSI,butrefers toanAcaciaSenegalplantationimplementedbyaprivatecompany fortheproductionofgumarabic.

(4)

Fig.1. Locationoftheinterventionsconsideredinthecasestudy(greenpolygonsandidentificationnumber,detailsinSection3.2).Areaswithinsufficientdocumentation aboutthetimingoftheintervention,interventionssmallerthan0.25km2,andareassubjectedtootherinterventiontypes(i.e.conservation)areingrey.Theredboxdelineates theboundariesoftheLandsatimageryused.Backgroundimageryisatruecolourcomposite(source:Esri).(Forinterpretationofthereferencestocolourinthisfigurelegend, thereaderisreferredtothewebversionofthisarticle.)

Totest theproposedmethodology,fromthelistofidentified interventionsitesweselectedthosewiththefollowingcharacter- istics:i) havingdocumentationoftheperiodofintervention,ii) coveredbyLandsatpath204androw49,iii)implementedafteryear 2002,andiv)withanareagreaterthan0.25km2(i.e.aminimum offourMODISpixels).Thisresultedinatotalof15interventions (greenpolygonsinFig.1).Interventionswithoutsuchcharacteris- ticsandanareasubjectedtonaturalconservation(greypolygons inFig.A)wereexcludedfromthecontrolsitesearch algorithm describedinthemethodssection.Abriefdescriptionofthevar- iousprojects,includingthetime and type ofintervention, field missionandVHRanalysisevaluation,ispresentedintheresults section(Table3).

Aqualitative evaluationof thesuccessof theintervention is availableforfivesitesthatwerevisitedinOctober2015andAugust 2016.Variouselementsweretakenintoaccountinthis evalua- tion:presenceandhealthstatusofnewlyplantedtrees,treeand herbaceouscoverdifferencewithrespecttosurroundings,informal interviewswithlocals.Thisinformation,togetherwiththevisual interpretationofVHRimagery,wasusedtocarryoutaconsistency checkwiththeresultsoftheproposedmethodology.

4. Methods

InBACIdesign,toaccountfornaturalchanges,theNDVIofthe restorationinterventionarea(i.e.the“impact”site)iscompared toanothersite,whichisreferredtoasthe“control”site(Smith, 2002).Theuseofmultiplecontrolsites(i.e.BACIwithmultiplesites) extendsthisideaandavoidsthecriticismthattheresultsofthe BACIexperimentaresolelyduetoapoorchoiceofthecontrolsite.

Thelocationofcontrolsisselectedrandomlyamongsitesthatare similartotheimpactsite(detailsinSection4.1).

4.1. Spatialsampling

Withrespecttotheimpactsite,acontrolareashouldhavethe followingcharacteristics:

i)similarlandcoverbeforetheintervention;

ii)relativelycloseinspaceinordertoexperiencethesameweather variability;

iii)not subjected to anthropogenic changes during the whole before–afterperiodbeinganalysed;

iv)randomlyselected.

Inaddition,evenifnotstrictlyrequiredbytheBACIdesign,we optedforselectingcontrolareaswithasizesimilartothatofthe impactareatoensureamorebalancedsamplingsize.Similarityin soilcharacteristics,knowntobeimportantdeterminantsofveg- etationinaridsystems,is expectedtobeimplicitlyensuredby conditioni.

Inordertofulfiltheserequirements,weproceedasfollowsfor eachoftheimpactsites.Whendifferentsettingsareusedforthe MODISandLandsatanalysis,thisisexplicitlymentionedinthetext andreportedinTable2.Someoftheintermediateproductsofthe analysisfortheLandsatdataandimpactsitenumber9areshown inFig.2asanexample.

First,werestricttheareafromwhichcontrolsareselectedto acircularareacentredonthecentroidoftheimpactsite.Pixels affectedbycloudcontaminationandSLC-offineitherthebefore

(5)

Fig.2. Exampleofintermediateresultsofthedescribedprocessingforprojectno.9(yellowpolygon,otherprojectsinred).Landsatimagesarefromthefollowingdates:

19/07/2003(before),13/09/2012(after).(A)nearinfraredfalsecolourcompositionLandsatimagebefore;maskedpixels(i.e.outsidesearcharea,SLC-off,detectedasclouds orcloudshadowsinthebeforeorafterimagery)areinblack, ¨stripes”areoriginatedbySLC-offaffectedpixels;(B)fiveclassesISODATAclassificationofthevalidpixels;

(C)classcompositionRMSEwithrespecttotheinterventionarea;RMSEofthewindowassignedtothecentralpixel;pixelswhosewindowwouldoverlapotherprojects aremaskedout(black);(D)greensquarepolygonsaretheselectedcontrols;NDVIdifference(valueafter–before)inthebackground.(Forinterpretationofthereferencesto colourinthisfigurelegend,thereaderisreferredtothewebversionofthisarticle.)

Table2

ListofMODIS-andLandsat-specificparametersusedintheanalysis.

MODIS Landsat

Ratiorbetweensearchandimpactarea 600

NumbernofISODATAclasses 5

Similaritythresholds 0.9

Numberofcontrolsrandomly extracted(nc)

20

Targetvariable Maximumannual

valueofsmoothed NDVI

NDVIof within-season

imagery Additionalcloudscreening None Visualanalysis Temporalsampling(beforeandafter) 5+5samples 1+1sample

orafterimageryaremaskedout(Fig.2A).Therestrictiontosucha circularareahastheobjectiveoffulfillingconditionii,i.e.defining

“aneighbourhood”whereclimaticconditionsshouldnotsignifi- cantlychange.Theextentofthisareaisdefinedasamultipleof theimpactareasize(searcharea/impactarea=r).Wemadethe searchareaproportionaltothesizeoftheimpactareatoensure thatitcontainsaroughlyconstantnumberofpotentialcontrols, independentoftheimpactareasize.Theratiorwassetforthis studyto600.Iftheimpactareahadacircularshape,thiswouldcor- respondtoaratiobetweentheareasearchedandtheimpactradius of24.5.Inthecasestudy,thisresultedinanaveragesearchradius

of25km(range=9–61km),wheresimilarclimaticconditionscan bereasonablyexpected.

Second,weusetheimagesacquiredintheperiodbeforethe intervention(Fig.2A)toperformaniterativeself-organisedunsu- pervisedclusteringalgorithm(ISODATA)withnclassesspatially restrictedtothesearcharea(Fig.2B).Withatrialanderrorprocess basedonthequalitativecomparisonoftheISODATAclassification mapandVHRimagery,wesetn=5inthisstudy.Alargernumber ofclassescanbeselectedifthelandscapeismoreheterogeneous.

TofullycomparetheresultsgatheredwiththeLandsatandthe MODISanalysis,weperformtheclassificationusingeitherLandsat orMODISdataforthetwotypesofanalysis,implyingthatdifferent controlsitesareselected.FortheLandsatanalysis,allthebandsin thereflecteddomainofasingleimageareusedfortheclassifica- tion,whereasforMODISwefollowtheapproachproposedbydeBie etal.(2011),usingthemulti-temporalNDVItrajectoryinsteadof themultispectralinformation.Theclassificationisthusperformed onafive-yearmulti-temporaldatasetof10-daycompositesend- ingtheyearbeforetheimplementationoftherestorationaction.

Afterthisclassificationstage,thefractionalclasscompositionofthe impactareaiscomputed.

Third,wedefineagenericcontrolasasquarespatialwindow withthesameareaoftheimpactsite.Thepopulationofpotential controlsisthusformedbyallthepossibleandoverlappingwin-

(6)

dowscentredoneachofthepixelsbelongingtothesearchareas.

Potentialcontrolsthatoverlapotherimpactsitesorexcludedareas (theareasubjecttoenvironmentalconservationinourcasestudy) areexcluded.Potentialcontrolshavingmorethan50%ofinvalid pixels(astheyarecoveredbythecloudandshadowmask)arealso excluded.Then,thefractionalclasscompositioniscomputedfor eachpotentialcontrol.

Fourth,thelandcoversimilaritybetweeneachpotentialcon- trolandtheimpactisdefinedasthecomplementoftherootmean squareerrorbetweenthefractionalcompositionsandone,i.e.sim- ilaritys=1–RMSE(Fig.2C).Valuesclosetoonethusindicatenearly identicaloverallclasscompositionofapotentialcontrolandthe impact.NotethatthesimilarityofNDVIvaluesbeforetheinter- ventionisnotconsideredhereastheBACIdesigndoesnotrequire similarlevelsofthevariableofinterest.

Fifth,wesubsamplethepopulationofpotentialcontrolsbydis- cardingthosewithasimilaritysmallerthans(0.9).Atthispoint wehaveasampleofpotentialcontrolsthatfulfilconditionsiand ii.Fromthissamplewerandomlyextractnccontrolsites(nc=20in thisstudy,Fig.2D).Randomextractionisexecutedusingprobability proportionaltosizesampling(Lohr,2010),inwhichtheselection probabilityforeachelementisproportionaltoitssimilaritytothe impactsite.Inthisway,themostsimilarcontrolshaveahigher probabilityofbeingselected.Oncea controlisextracted,allits overlappingpotentialcontrolsareexcludedforfurtherselection andtherandomextractionisrepeateduntilalltherequiredcon- trolsareselected.Itisnotedthatthisproceduredoesnotguarantee thatalldesirednccontrolsareactuallyavailable.Ifthenumberof selectedcontrolsisconsideredtobeinsufficient,onemayincrease thesearchareaorreducetherequiredsimilaritystoincreasethe populationofcandidatecontrolsandthusthenumberofselected controls.

Oncethelocation ofthecontrolsis established,theNDVI is extractedforallvalidpixelsbelongingtotheimpactandcontrol areasfortheperiodbeforeandaftertheintervention.Theselection processdescribedsofarwasimplementedinIDL(HarrisGeospatial Solution,Inc.)andfullyautomatised.

Finally,conditioniiiwastestedbyvisuallyinspectingtheavail- abletimeseriesofLandsatimageryoftheselectedcontrolsites.It isnotedthatonlyclearlandusechanges,forinstancefromnat- uralvegetationtocroplandor tosettlements,are detectablein suchaway.Thepossibleoccurrenceoflessvisiblechanges,such asunreportedrangelandmanagementpractices,canthereforenot beexcluded.Theimpactofthepotentialselectionofsuchunsuit- ablecontrolsisexpectedtobemitigatedbygatheringarelatively largenumberofcontrolsites.

4.2. TreatmentofLandsat7SLCgaps

Anestimated22%oftheLandsat7scenesislostbecauseofthe SLC failure (http://landsat.usgs.gov/productsslcoffbackground.

php).TheSLC-offeffectsaremostpronouncedalongtheedgesof thescene andgraduallydiminishtowarditscentre.Theprecise location of the missing scan lines varies from scene to scene.

Therefore,itisdifficulttoanticipatethefractionofmissingdata forindividual impactareas.Withourtest cases,thefractionof missingdatavariedbetween0%and40%andoperatedasarandom subsampling with no expected consequences on the following BACItest.InthepresenceofthisSLCproblem,theaffectedpixels wereconsideredasbelongingtoanadditionallandcover class, thuscontributing tothesimilaritymeasure described above.In thiswaywefavouredtheselectionofcontrolsshowingasimilar fractionofSLC-affectedpixels.

4.3. Temporalsampling

Multipletemporalsamplingbeforeandaftertheputativeimpact ispreferableasitensuresthatcoincidentaltemporalfluctuations in either locationdo not confound thedetection ofthe impact (Underwood, 1992). Due to the limited frequency of temporal acquisition,wecouldnotretrievemultipleobservationimagery beforeand after theintervention from Landsat, and we conse- quently applied BACI based ona single couple of before–after observations and multiple control sites. The closest cloud-free imagesbeforeandafterthetimeofinterventionwerethusselected foreachrestorationsite.ThemorerobustBACIdesign,withobser- vationsfrommultipledatesandsites,wasinsteadusedwiththe hightemporalfrequencyMODISdata.Thatis,uptofiveannualval- uesofmaximumannualNDVIwereextractedfromtheMODIStime series.

4.4. Statisticalanalysis

Alinearmixed-effectsmodelonNDVIsite-levelaverageswas usedtotesttheimpactoftherestorationinterventionasinSchwarz (2015). In this context, the period (before/after), the site class (impact/control)and theinteractionofsite classandperiodare fixed effects while the site and the samplingtime, beingnon- exhaustivesamplesofthepotentialsitesandsamplingtimes,are consideredtoberandomeffects.Linearmixed-effectsmodelsuse maximumlikelihoodtoestimatetheparametersofthelinearfunc- tioncontainingboth fixedandrandomeffects. Outputisin the formofapproximatez-ratiosornormaldeviates,whichallowssta- tistical testsonanylinearcombination ofthefixedparameters (PinheiroandBates,2000).Toevaluatetheimpactoftheinter- ventionwewereinterestedintheinteractionoftheperiodand thesiteclass(the so-calledBACIeffect)representingthediffer- entialchangebetweenimpactandcontrolsitescomparedbefore andaftertheintervention.The(null)hypothesisofnochangewas rejectedattheconventional5%significancelevel.

TheBACIanalysisprovidestwoimportantstatistics(amongoth- ers):thesignificancelevel(i.e.P-value)oftheBACIeffecttest(i.e.

nochangenullhypothesis)andtheBACIcontrast.TheBACIcon- trastiscalculatedasthedifference(controlsvs.impact)between themeandifferences(aftervs.before):

BACIcontrast=(␮CA−␮CB)−(␮IA−␮IB) (1)

Where␮isthesite-specificspatialmeanofthevariableselected torepresenttheimpact(hereNDVI);CA,IAstandforControland ImpactAfter,respectively;CBandIBforControlandImpactBefore, respectively.Byconvention,anegativecontrastindicatesthatthe variablehasincreasedmore(ordecreasedless)intheimpacted sitewithrespecttocontrolsinthetimeperiodrangingfrombefore toaftertheimplementationoftherestorationproject. TheBACI contrastisexpressedinthesameunitsofthevariableofinterest, hereNDVI.Inordertohighlightthemagnitudeofthecontrastwith respecttotheinitialconditions,wenormaliseitbythemeanofthe NDVIoftheimpactareabeforetheinterventiontookplace(␮IB) andexpressitasapercentage.Thisderivedvariableisreferredto as“relativecontrast”inthefollowing.

Itisnotedthat,despitethefactthatNDVIcomputedfromLand- sat7and8(ETM+andOLIsensors)maybeslightlydifferentbecause ofthedifferentspectralresponsesofthebands(Royetal.,2015)and differentatmosphericcorrectionalgorithm,thisimpactsboththe projectsiteandthecontrolsandhencedoesnothaveaneffecton theBACIanalysis,whichworksonthedifferencebetweenthetwo typesofarea.

(7)

Table3

Maininformationofanalysedinterventions,fieldmissionevaluation,visualinterpretationofGoogleEarthVHRimageryandBACIresultsonMODISandLandsatdata.n.a.

standsfornotavailable.ThemeanoftheRSvariableiscomputedastheoverallmeanextractedbeforetheintervention(allsites,allsamplingdates).Green(likelysuccess), lightgreen(moderateorambiguoussuccess)andgreybackground(likelyfailure)isusedtoranktheintervention’ssuccessbasedonthefieldmissionandVHRqualitative evaluation.GreenbackgroundisusedintheBACIsectiontohighlightnegativeBACIcontrasts(inbold)thataresignificantatthe0.05P-value.Greybackgroundindicatesa non-significantBACIeffect.

Theopensourcestatistical softwareR(RDevelopment Core Team,2016)wasusedtodevelopascripttoautomatisethesta- tisticaltestfollowingSchwarz(2015).

5. Resultsanddiscussion

ResultsoftheBACIanalysis,alongwithprojectinformation,VHR photointerpretationandfieldmissionqualitativeevaluation,are reportedinTable3.Thenumberofcontrolsitesexcludedfromthe analysisaftervisualinspectionrangedfromzerotoamaximum ofsix.Theanthropogenicchangesdetectedintheperiodafterthe interventionmainlyrefertoappearanceofagriculturalfieldsand settlements.

5.1. BACIanalysis

AsignificantlynegativeBACIcontrast(i.e.improvementinNDVI withrespecttocontrolsaftertheintervention)wasdetectedinfive andfouroutof15sitesusingMODISandLandsatdata,respectively.

Forthemajorityofsites,the(null)hypothesisofnochangecould notberejected.Forthreesites,thecontrastwasindeedpositive, i.e.therewasarelativedecreaseinNDVIintherestorationarea.

FocussingonthesitesforwhichasignificantBACIeffectwas detected, the average relative contrast is −20% and −27% for MODISandLandsatdata,respectively.ConsideringNDVIasarough approximationofthefractionalvegetationcover,thesenumbers translateintoasignificantimprovementinthevegetationcover withrespecttothecontrols.

AsanexampleofthedatausedfortheBACIanalysis,impact andcontrolaveragesareshowninFig.3forfourrepresentative interventions:no.9,whereasignificantlynegativeBACIeffectis foundusingbothLandsatandMODIS;no.81,wherethenegative contrastissignificantatthe0.05levelforMODISonly(P<0.1for Landsat);no.17withapositivebutnon-significantcontrast;and

finallyno.4withapositiveandnon-significantcontrast(P<0.1for Landsat).

InordertogaininsightsintothedifferencebetweentheMODIS andLandsatanalyseswefocusontheagreementbetweentworel- evantBACIstatistics(i.e.contrastandP-value).First,Table3shows aperfectmatchintheBACIcontrastsign.Thatis,bothtypesofanal- ysisagreeintheevaluationofthesignoftheintervention,either re-greening(negativecontrast)ordegradation(positivecontrast) oftheimpactsitecomparedtothecontrols.Themagnitudeofthe contrastandthemeanoftheRSvariablecanbedifferentbetween thetwotypesofanalysisbecausetheRSvariableisdifferent:the maximumseasonalNDVIforMODIS,andtheNDVIvalueduring thegrowingseasonataspecificsamplingdatedictatedbyimage availabilityforLandsat.

Second,largeagreementinthedetectionof a significantre- greening of the intervention (i.e. negative BACI contrast with P-value<0.05)existsbetweenthetwotypesofanalysis.Onlyone caseofminordisagreementisfoundforsiteno.81(Table3and Fig.3),forwhichthevisualanalysisofGoogleEarthVHRimagery indicatesthat theplantationwas actuallyimplemented. In this casebothtypesofanalysiscomputeanearlyidenticalandnega- tiveBACIcontrast(−0.12)whereastheydifferinthesignificance levelattributed.However,theP-valueofLandsat(0.058)isnotfar fromthethreshold(0.05)usedtorejectthe(null)hypothesisofno change.Asaresult,thechangedetectedusingLandsatdatahasa lowerconfidencelevel(P<0.10).

Otherminordifferences(i.e.notleadingtodifferenttestout- come)betweentheresultsofthetwotypesofanalysisreferto differentmagnitudesoftheP-value.TheP-valueofMODISisgen- erallylowerthanthatofLandsat.Fig.4showstheP-valueofthe twotypesofanalysisvs.theabsolutevalueoftherelativeBACI contrast.

Bothtypesof analysisshowareduction in theP-valuewith increasingabsolutevalueoftherelativecontrast,asthetestessen- tiallybuilds(also)onthemagnitudeoftheBACIcontrast.However,

(8)

Fig.3.TemporalprofilesofmeanNDVIvaluesforselectedimpact(bluelines)andcorrespondingcontrolsites(redlines)forLandsat(left)andMODIS(right)data.Sampling datesbeforeandaftertheinterventionareseparatedbytheverticalblackline.TheP-value(P)andthepercentrelativecontrast(RC)arereported.(Forinterpretationofthe referencestocolourinthisfigurelegend,thereaderisreferredtothewebversionofthisarticle.)

theMODISP-valuesaremostlylowerthanthoseofLandsatforsim- ilarrelativecontrasts.Therefore,themultipletemporalsampling thatcanbeachievedusingMODISdataappearstobeinstrumental inincreasingthesignificancelevelofthetestwithrespecttothe singletimeanalysisofLandsat.Thisislikelyduetothreereasons:

i)increasedsamplesizeforMODISanalysis,ii)betterrepresen- tationof theoverall vegetationcover offeredby themaximum NDVIwithrespecttothesingledateNDVI,andiii)reduceddepen- dencyofMODISonaspecificyearandtime.Concerningthelatter, withtheMODISset-upweanalysethecontrol/impactdifferential behaviourinamulti-yeartime span,makingit lesssensitiveto possibleyear-specificpeculiaritiesthatmayaffectthesingle-year high-resolutionLandsatset-up(seeFig.3).Itisnotedthatthesame multiple-period design canbeapplied tohigh-resolutionfreely availabledataingeographical settingswithahigheravailability

of cloud-free Landsat imagery, or when analysing more recent projectsthatcanexploitthemorefrequentavailabilityofLandsat8 imageryandotherrecentlyavailableinstruments(e.g.Sentinel2).

Theuncertaintiesconnectedtotheuseofasingleimagebeforeand aftertheinterventionarewellexemplifiedbythetemporalevolu- tionofNDVIforprojectno.81inFig.3.Anon-significantBACIeffect is detectedusing thesingle-imageset-up ofLandsat (P=0.058) despiteaquitelargenegativerelativecontrast(−27.95%).Forthe samesite,theMODISmulti-yearprofileshowsalargeinter-annual variability.TheLandsatsingle-imageset-uppickedup2007asthe year“before”,whenthecontrolhadthethird-highestMODISNDVI.

Ifadifferentyearwereavailable,forexample2006whenthecon- trolhadthesecond-lowestMODISNDVI,thismayhaveresultedin adifferent(lower)P-value.

(9)

0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1

0 10 20 30 40

P-v al u e

|% relave contrast|

Landsat MODIS P=0.05 P=0.1

Fig.4.ScatterplotoftheabsolutevalueoftherelativeBACIcontrast(equalto100*

|contrast|/meanofRSvariablebeforetheintervention)vs.theP-valueoftheBACI test.NullhypothesisrejectionoftypicalP-valuethresholdsof0.05and0.1areshown asdashedlines.

Projectno.4 (showingdegradationwithrespecttocontrols) showsanoppositebehaviour:lowerP-value(i.e.higherconfidence) fortheLandsatanalysis.Here,thesmallsizeoftherestorationarea playsarole,resultinginapoorMODISspatialsampling,ontheone handbyreducingthesamplesizeandthepowerofthetest,andon theotherbymakingthefewMODISsampleslessreliable.Infact, theactualareasensedbytheinstrumentisgreaterthanthenomi- nalspatialresolution,andhasanellipticalshapecontrolledbythe sensorcharacteristicsandobservationgeometry(Duveilleretal., 2011;DuveillerandDefourny,2010).Thus,afractionofthesignal inpixelslocatedattheborderoftheprojectareamayoriginate fromanareaoutside.Thiseffectmaybenon-negligiblewhenthe projectareaiscomposedofonlyafewMODISpixels,asforproject no.4.

Besidesthestatisticaltestresult(i.e.rejectionofthenullhypoth- esisofnochange),therelativeBACIcontrastcanprovideadditional insights into the extent of the successof a given intervention project.Forinstance,withMODISanalysis,thisrangesfrom+6.3%

(degradationforprojectno.4)to−27.7%(improvementforproject no.14),indicatingadifferentmagnitudeoftheeffectofthedifferent restorationinterventions.

5.2. BACIresultsvs.qualitativeinformation

A general agreement between the qualitative information extracted from Google Earth VHR imagery and BACI results is observed.Inallsiteswherenosignsofinterventionsornodiffer- encewiththesurroundingareaswasobservedinVHRimagery,the BACIeffectisnotsignificant.Inallsiteswhereapatternofregularly plantedandestablishedtreeswereobserved,theBACIcontrastis negativeandtheBACIeffectissignificant,withtheexceptionof siteno.81whichisnotsignificantwhenLandsatisused.Thetest oftheBACIeffectalsoagreeswiththefieldqualitativeevaluation availableforfivesites.Amongthefivesites,twowereevaluatedas beingrelativelysuccessfulandarematchedbyasignificantBACI effect(sitesno.14and15),andthreewerenegativelyevaluated andarematchedbyanon-significantBACIeffect(sitesno.9,16 and44).Siteno.5,wherethepresenceofreforestationintervention wasnotvisible,wasinsteadfoundtohaveasignificantandnegative BACI.However,thefieldevaluationdidnotprovideanyinforma- tionaboutthegrasslandcoverthatmayhaveimprovedafterthe

fencingintervention,thustriggeringthestatisticaldetectionofa greeningeffect.

5.3. Applicabilityofthemethodtodifferentinterventiontypes

Albeitrestorationinterventionsthatdonotinvolvea“green- ing”cannotbescrutinisedusingNDVI,therangeofapplicability maybeexpandedusingthesamestatisticalframeworkwithother RS-based quantitative indicators, when considered relevant for assessingthesuccessofaspecifictypeofinterventionatagiven scale of analysis. For instance, soil erosion processes couldbe assessed by detecting erosionfeatures and eroded areas or by estimatingerosion-controllingfactors,suchassoilmoistureand surfaceroughness(AndersonandCroft,2009;Vrieling,2006).Spa- tialpatternmetricscouldsupporttheassessmentofrestoration interventionsthatimpacthabitatcomposition,fragmentation,and connectivityatlandscapelevel,alsoinrelationtolanddegrada- tionprocesses(Favaetal.,2015;Kéfietal.,2007).Asanadditional example,fine-scalequantitativemappingofspecificplantspecies (e.g.invasive)couldbecriticaltomonitortheeffectivenessofplant removalorcontrolefforts(PysekandRichardson,2010).

5.4. Applicabilityofthemethodtodifferentlandscapesettings

Topographicvariationsarenotexplicitlyaccountedforinthe describedmethod.Althoughnotanissueintheflatcase-studyland- scape,twoeffectsoftopographycanbeenvisagedinregionswith significantrelief.First,differentvegetationtypesgrowinlocations withdifferentelevation,slopeand aspect.Thus,controlsshould beselectedwithsimilartopographiccharacteristicswithrespect totherestorationsite.Asweexpectdifferentvegetationtypesto bespottedbytheclassificationofRSimagery,thisfirsteffectdoes nothampertheproposedmethod.Inaddition,inthecaseswhere suchtopographiccharacteristicsareexpectedtobeimportant,they couldbeaddedtotheinputlayersoftheclassification.Thesecond effectoftopographyisonthegeometryofthesun-target-sensor system,andthusonthereflectance.Moderatereliefvariationsare expectedtohaveaminorimpactonthemethodastheuseofaband ratiosuchastheNDVIwillreducethetopographiceffect(Leeand Kaufman,1986).Inaddition,differentilluminationconditions(at leastthoserelatedtothedirectlightcomponent)canbenormalised using,forinstance,slope-aspectcorrections(e.g.Teilletetal.,1982).

Thus,topographycanbetreatedanddoesnotlimittheapplicability ofthemethod.

TheBACIanalysisisacomparativemethodinwhichthetem- poral variability due to natural environmental conditions (i.e.

weather)isaccountedforusingcontrols.Therandomselectionof multiplecontrolsitesandthevisualinspectionoftheirstability overtheanalysisperiodminimisestheimpactoftheselectionof unsuitablecontrols(i.e.affectedbynonweather-drivenchangesin greennessafterthetimeofintervention).However,ifthelandscape aroundtherestoration areais subjectedto widespreadanthro- pogenicchanges(e.g.agriculturalintensification,urbanisation),the possibilityofselectingmultiplesuitablecontrolswillbeseverely limited,affectingthediscriminationpowerofthetest.Onthecon- trary,possiblenaturaldisturbancessuchasfiresorpestscanbe accountedforbythetest.Infact,adecreaseingreennesswouldbe detectedifthedisturbanceaffectedonlytherestorationsitewhile arelativeincreasewouldbemorelikelytobedetectedifthedis- turbanceaffectedseveralcontrols.Thechangeingreennessmaybe theninterpretedasdecreased(orincreased)vulnerabilitytosuch disturbancesduetotheintervention.

(10)

6. Conclusions

Forthefirsttime,abefore/aftercontrol/impact(BACI)design wasappliedtoRSdatatoevaluatethebiophysicalimpactofrestora- tionprojects. Large agreementwasfoundin thestatistical test outcomesusingeitherMODISorLandsatdata.Theavailabilityof frequentMODISobservationsmakesthedataofthisinstrument wellsuitedtothemostrobustBACIdesign,exploitingmultiplecon- trolsandmultipleobservationsbeforeandaftertheintervention.

TheuseofLandsatdatainourtestcasestudywaslimitedbythe pooravailabilityofcloud-freeimagery,compellingtheapplication ofasingle-timeBACIdesignandresultingingenerallylowercon- fidence(i.e.highersignificancelevel,P-value)ofthetestresults.

Theanalysisofmorerecentinterventionprojectswillbenefitof theavailabilityofmorefrequentsatelliteobservationsfromLand- sat8andSentinel2satellites.Thecombinationofhighspatialand temporalresolutionofferedbysensorssuchastheSentinels2may considerablyincrease thepotential ofthe proposedmethod.In addition,forearlierproject,theuseofcommercialsatellite(e.g.

SPOT4and5,RapidEye)maybeconsideredtocomplementthe freeimageryandincreasedataavailability.

Resultsofthestatisticalanalysiswereinagreementwiththe qualitativeinformationprovidedbyfieldobservationsandvisual interpretation of the VHR imagery in Google Earth. The pro- posedapproachcanbeconsideredafirstscreeningofrestoration interventionsthatmaydrivefurtherandcomplementaryinsitu analyses,thusincreasingthecost-efficiencyandfeasibilityofthe evaluationofrestorationinterventions.Inaddition,themethod- ology can be usedfor thelong-term monitoring of restoration interventions,thusallowingthebenefitsoftheinitialinvestment anditssustainabilitytobeevaluated.

WhenNDVIisused,theapplicabilityoftheproposedmethod islimitedtotheverificationofabiophysicalimpactintermsof variationinvegetationcover.Thisisnotlimitedtoreforestation andrangelandimprovementbuttoarangeofinterventions(e.g.

soilconservation,surfacewaterrun-offcontrol,infrastructuresfor irrigation,improvedlandgovernanceandmanagement,etc.)that alsocausere-greening.Theuseofotherremote-sensing-derived variables(e.g.soilmoisture,surfaceroughness,fragmentation,VHR plantspeciesmapping)mayfurtherextendtheapplicabilityofthe statisticalframeworktootheraspectsofrestorationinterventions.

Insitu analysesremainof fundamentalimportance,notonlyto provideamoredetailedsetofbiophysicalindicatorstargetedat thespecificrestoration,butalsotoconsiderotherkeyaspectsof restorationrelatedtosocialperceptionandeconomicimpacts.

Acknowledgements

WewouldliketothankMoustaphaBassimbéSagnaforthesup- portduringthefieldmissionsandtheCNRSHumanEnvironment Observatory(Tessekeré,Senegal)forthefinancialsupportforfield missions.Theworkwasfunded bytheAdministrativeArrange- mentbetweentheEuropeanCommissionDGDEVCOandtheJRC for ¨Technicalandscientific supporttoagricultureand foodand nutritionsecuritysectors”(TS4FNS2,ref.33272).

References

AfricanUnion,PanafricanAgencyoftheGreatGreenWall,2012.Harmonised regionalstrategyforimplementationofthe“GreatGreenWallInitiativeofthe SaharaandtheSahel”.,pp.1–33(http://www.greatgreenwallinitiative.org/

sites/default/files/publications/harmonizedstrategyGGWSSI-EN.pdf).

Anderson,K.,Croft,H.,2009.Remotesensingofsoilsurfaceproperties.Prog.Phys.

Geogr.33,457–473,http://dx.doi.org/10.1177/0309133309346644.

Andrefouet,S.,Bindschadler,R.,BrownDeColstoun,E.C.,Choate,M., Chomentowski,W.,Christopherson,J.,Doorn,B.,Hall,D.K.,Holifield,C., Howard,S.,Kranenburg,C.,Lee,S.,Masek,J.B.,Moran,M.,Mueller-Karger,F., Ohlen,D.,Palandro,D.,Price,J.,Qi,J.,Reed,B.,Samek,J.,Scaramuzza,P.,Skole,

D.,Schott,J.,Storey,J.,Thome,K.,Torres-Pulliza,D.,Vogelmann,J.,Williams, D.L.,Woodcock,C.,Wylie,B.,2003.PreliminaryAssessmentofthevalueof Landsat7ETM+datafollowingscanlinecorrectormalfunction.In:USGS TechnicalReport.,pp.1–88,avaliableat:https://landsat.usgs.gov/sites/default/

files/documents/SLCoffScientificUsability.pdf.

Bautista,S.,Aronson,J.,Vallejo,V.,2010.LandRestorationtoCombat

Desertification:InnovativeApproaches,QualityControlandProjectEvaluation.

FundaciónCEAM,Valencia,Valencia.

Benayas,R.J.M.,Newton,A.C.,Diaz,A.,Bullock,J.M.,2009.Enhancementof biodiversityandecosystemservicesbyecologicalrestoration:ameta-analysis.

Science325,1121–1124,http://dx.doi.org/10.1126/science.1172460.

Bey,A.,Sanchez-PausDiaz,A.,Maniatis,D.,Marchi,G.,Mollicone,D.,Ricci,S., Bastin,J.-F.,Moore,R.,Federici,S.,Rezende,M.,Patriarca,C.,Turia,R.,Gamoga, G.,Abe,H.,2016.CollectEarth:landuseandlandcoverassessmentthrough augmentedvisualinterpretation.RemoteSens.8(807),1–24,http://dx.doi.

org/10.3390/rs8100807.

Birch,J.C.,Newton,a.C.,Aquino,C.a.,Cantarello,E.,Echeverria,C.,Kitzberger,T., Schiappacasse,I.,Garavito,N.T.,2010.Cost-effectivenessofdrylandforest restorationevaluatedbyspatialanalysisofecosystemservices.Proc.Natl.

Acad.Sci.USA107,21925–21930,http://dx.doi.org/10.1073/pnas.1003369107.

Carlson,T.N.,Ripley,D.A.,1997.OntherelationbetweenNDVI,fractional vegetationcover,andleafareaindex.RemoteSens.Environ.62,241–252.

deBie,C.a.J.M.,Khan,M.R.,Smakhtin,V.U.,Venus,V.,Weir,M.J.C.,Smaling,E.M.a., 2011.Analysisofmulti-temporalSPOTNDVIimagesforsmall-scaleland-use mapping.Int.J.RemoteSens.32,6673–6693,http://dx.doi.org/10.1080/

01431161.2010.512939.

Dee,D.P.,Uppala,S.M.,Simmons,A.J.,Berrisford,P.,Poli,P.,Kobayashi,S.,Andrae, U.,Balmaseda,M.A.,Balsamo,G.,Bauer,P.,Bechtold,P.,Beljaars,A.C.M.,vande Berg,L.,Bidlot,J.,Bormann,N.,Delsol,C.,Dragani,R.,Fuentes,M.,Geer,A.J., Haimberger,L.,Healy,S.B.,Hersbach,H.,H??lm,E.V.,Isaksen,L.,K??llberg,P., K??hler,M.,Matricardi,M.,Mcnally,A.P.,Monge-Sanz,B.M.,Morcrette,J.J., Park,B.K.,Peubey,C.,deRosnay,P.,Tavolato,C.,Th??paut,J.N.,Vitart,F.,2011.

TheERA-interimreanalysis:configurationandperformanceofthedata assimilationsystem.Q.J.R.Meteorol.Soc.137,553–597,http://dx.doi.org/10.

1002/qj.828.

Duveiller,G.,Baret,F.,Defourny,P.,2011.Cropspecificgreenareaindexretrieval fromMODISdataatregionalscalebycontrollingpixel-targetadequacy.

RemoteSens.Environ.115,2686–2701,http://dx.doi.org/10.1016/j.rse.2011.

05.026.

Duveiller,G.,Defourny,P.,2010.Aconceptualframeworktodefinethespatial resolutionrequirementsforagriculturalmonitoringusingremotesensing.

RemoteSens.Environ.114,2637–2650,http://dx.doi.org/10.1016/j.rse.2010.

06.001.

FAO,FoodandAgricultureOrganizationoftheUnitedNations,2015.SWALIM SupportsGulleyErosionRehabilitationMonitoringSysteminPuntland.

SWALIMUpdate,September-December2015,pp.1–10.Availableonlineat http://www.faoswalim.org/resources/sitefiles/

SWALIM%20Update%20Issue%20%2010.pdf.

Fava,F.,Pulighe,G.,Monteiro,A.T.,2015.Mappingchangesinlandcover compositionandpatternforcomparingmediterraneanrangelandrestoration alternatives.LandDegrad.Dev.27(3),671–681,http://dx.doi.org/10.1002/ldr.

2456.

Funk,C.,Peterson,P.,Landsfeld,M.,Pedreros,D.,Verdin,J.,Shukla,S.,Husak,G., Rowland,J.,Harrison,L.,Hoell,A.,Michaelsen,J.,2015.Theclimatehazards infraredprecipitationwithstations—anewenvironmentalrecordfor monitoringextremes.Sci.Data2,150066,http://dx.doi.org/10.1038/sdata.

2015.66.

Jenkerson,C.B.,Maiersperger,T.,Schmidt,G.,2010.eMODIS:AUser-FriendlyData Source:U.S.GeologicalSurveyOpen-FileReport2010-1055.

Kéfi,S.,Rietkerk,M.,Alados,C.L.,Pueyo,Y.,Papanastasis,V.P.,ElAich,A.,deRuiter, P.C.,Kefi,S.,Rietkerk,M.,Alados,C.L.,Pueyo,Y.,Papanastasis,V.P.,ElAich,A.,de Ruiter,P.C.,2007.Spatialvegetationpatternsandimminentdesertificationin Mediterraneanaridecosystems.Nature449,213–217,http://dx.doi.org/10.

1038/nature06111.

Lee,T.Y.,Kaufman,Y.J.,1986.Non-Lambertianeffectsonremotesensingofsurface reflectanceandvegetationindex.IEEETrans.Geosci.RemoteSens.,699–708.

Lohr,S.L.,2010.Sampling:DesignandAnalysis,secondedition.,http://dx.doi.org/

10.1017/CBO9781107415324.004.

Low,P.S.(Ed.),2013.EconomicandSocialImpactsofDesertification,land degradationanddrought.WhitePaperI.UNCCD2ndScientificConference, preparedwiththecontributionsofaninternationalgroupofscientists.Bonn, Germany.

Mbow,C.,Brandt,M.,Ouedraogo,I.,deLeeuw,J.,Marshall,M.,2015.Whatfour decadesofearthobservationtellusaboutlanddegradationintheSahel?

RemoteSens.7,4048–4067,http://dx.doi.org/10.3390/rs70404048.

Niang,K.,Sagna,M.B.,Ndiaye,O.,Thiaw,A.,Diallo,A.,Akpo,L.E.,Saleh,M.M., Diome,N.,Diatta,S.,Faye,M.N.,Gueye,M.,Guissé,A.,Goffner,D.,2014.

RevisitingtreespeciesavailabilityandusageintheFerloregionofSenegal:a rationaleforindigenoustreeplantingstrategiesinthecontextofthegreat greenwallfortheSaharaandtheSahelinitiative.J.Exp.Biol.Agric.Sci.2, 529–537.

Nicholson,S.E.,2013.TheWestAfricanSahel:areviewofrecentstudiesonthe rainfallregimeanditsinterannualvariability.ISRNMeteorol.2013,32.

Ntshotsho,P.,Esler,K.J.,Reyers,B.,2015.Identifyingchallengestobuildingan evidencebaseforrestorationpractice.Sustainability7,15871–15881,http://

dx.doi.org/10.3390/su71215788.

Riferimenti

Documenti correlati

2 Pension provision in the recent past Public pensions and replacement rates The pay-as-you-go social security system is by far the most important pension system in Germany.. It

Due to the different research new complete dentures do not change unstimulated whole salivary flow rate significantly however another researcher found that unstimulated whole

Figure 6-31 (right) Bulk temperature and pin average temperature for the SS70 and OF70 cases as function of the height from the heated region inlet plane; (left) Pressure drop

Filopanti, Degli usi idraulici della tela :memoria letta all’Accademia delle scienze 26/11/186, Estratto Nuovi Annali delle scienze naturali di Bologna Fasc. Filopanti, Al

The results are very similar when all the single sources considered in the analysis (SNRs and PWNe) contribute to the anisotropy, which is dominated by Vela YZ. Considering the

Absent some convergence in growth models, the correction of future imbalances will impose a burden only on countries with a demand-driven growth model because they will have no

As mentioned in the previous section, the tool consists of a client and a server. The client is executed on MONROE nodes, whereas the server is executed on a dedicated machine.

To test whether this trend changes with redshift, we separate the sample into three redshift inter- vals with equal number of sources (17 per bin). A trend of higher AGN