What future works have the authors mentioned in the paper "Measuring β-diversity by remote sensing: a challenge for biodiversity monitoring" ?

Such measures might be used to regress species diversity against remotely sensed heterogeneity, based on new regression techniques which maximize the possibility of predicting the zones in a study area, or at larger spatial scales, of peculiar conservation value. As an example Rocchini et al. ( 2013 ) introduced the possibility of applying generalized entropy theory to satellite images with one single formula representing a continuum of diversity measures changing one parameter. As previously stated, the suggested methods for β-diversity estimation from remote sensing are mainly based on distances, but they could be effectively translated to relative abundance-based methods.

How can the authors combine remote sensing data and biodiversity data in the field?

Remote sensing data and biodiversity data in the field can be coupled by sparse canonical correlation analysis (SCCA) to produce canonical components and a community dissimilarity matrix, which are then used to build a generalized dissimilarity model (GDM) to finally derive a β-diversity map0 0SCCAGDMGDMRemote sensing data Biodiversity dataCanonical components Community dissimilarityBeta-diversity mapland cover types, one-way ANOVA tests were performed.

What is the common way to describe the occupancy of new sites by species?

A negative exponential dispersal kernel is usually adopted to mathematically describe the occupancy of new sites by species, as follows:where dik = distance between two locations i and k and a is a parameter regulating the dispersal from localized areas (low values of a) to widespread ones (high values of a; Meentemeyer, Anacker, Mark, & Rizzo, 2008).

What is the definition of a penalized canonical correlation analysis?

The SCCA is a form of penalized canonical correlation analysis based on the L1 (lasso) penalty function, and is thus designed to deal with high-dimensional data.

How is the compositional dissimilarity model used?

This is done through a linear combination of monotonic I-spline basisfunctions, under the assumption that increasing environmental dissimilarity (e.g., along a gradient) can only result in increasing compositional dissimilarity.

What are the suggested methods for -diversity estimation from remote sensing?

As previously stated, the suggested methods for β-diversity estimation from remote sensing are mainly based on distances, but they could be effectively translated to relative abundance-based methods.

How is the Rao's Q diversity index calculated?

Rao's Q is capable of discriminating among the ecological diversity of matrices (3) and (4), turning out to be 4.59 and 90.70, respectively.

What could be used to regress species diversity against remote sensing data?

Such measures might be used to regress species diversity against remotely sensed heterogeneity, based on new regression techniques which maximize the possibility of predicting the zones in a study area, or at larger spatial scales, of peculiar conservation value.

What is the example of a generalized entropy theory?

As an example Rocchini et al. (2013) introduced the possibility of applying generalized entropy theory to satellite images with one single formula representing a continuum of diversity measures changing one parameter.

(Open Access) Measuring β-diversity by remote sensing: A challenge for biodiversity monitoring (2018) | Duccio Rocchini

Q: What are the contributions in "Measuring β-diversity by remote sensing: a challenge for biodiversity monitoring" ?

In this paper, a negative exponential dispersal kernel is adopted to mathematically describe the occupancy of new sites by species, as follows: where dik = distance between two locations i and k and a is a parameter regulating the dispersal from localized areas ( low values of a ) to widespread ones ( high value of a ).

Q: What is the purpose of the self-organizing feature map?

The self-organizing feature map uses unsupervised learning to map the location of field sites within the output space on the basis of their relative similarity in species or spectral composition.

Q: What are the advantages of this approach?

The advantages of this approach are obvious: since the diversity analyses are conducted in the floristic gradient space, the resulting measures resemble field studies and are thus easier to interpret than spectral proxies and closer to the point of view of many end-users.

Q: What are the challenges of the Copernicus program?

In the perspective of global monitoring of biodiversity, and given the unprecedented remote sensing capacity allowed by the Copernicus program, including the Sentinel-2 multispectral satellites, several other challenges are foreseen and currently investigated.

Methods Ecol Evol. 2018;9:1787–1798. wileyonlinelibrary.com/journal/mee3 



1787

Received:22September2017

Accepted:11November2017

DOI: 10.1111/2041-210X.12941

IMPROVING BIODIVERSITY MONITORING

USING SATELLITE REMOTE SENSING

Measuring β-diversity by remote sensing: A challenge for

biodiversity monitoring

Duccio Rocchini

1,2,3

|

|

|

|

Daniel Doktor

|

3,8

|

|

|



|

|

|

|



|

14,15

|

|

|

Carlo Ricotta

|

|

|



|

CenterAgricultureFoodEnvironment,UniversityofTrento,S.Micheleall’Adige(TN),Italy;

CentreforIntegrativeBiology,UniversityofTrento,Povo(TN),

Italy;

DepartmentofBiodiversityandMolecularEcology,FondazioneEdmundMach,ResearchandInnovationCentre,S.Micheleall’Adige(TN),Italy;

UMR-

TETIS,IRSTEAMontpellier,MaisondelaTélédétection,MontpellierCedex5,France;

InstituteofZoology,TheZoologicalSocietyofLondon,London,UK;

SchoolofComputerScience,AstonUniversity,Birmingham,UK;

DepartmentComputationalLandscapeEcology,HelmholtzCentreforEnvironmental

Research–UFZ,Leipzig,Germany;

DepartmentofComputerScienceandEngineering,UniversityofBologna,Bologna,Italy;

InstitutfürGeographieFriedrich-

Alexander,UniversitätErlangen-Nürnberg,Erlangen,Germany;

SchoolofGeography,UniversityofNottingham,Nottingham,UK;

SchoolofBiology,Faculty

ofbiologicalScience,UniversityofLeeds,Leeds,UK;

InstituteofMediterraneanAgriculturalandEnvironmentalSciences(ICAAM),UniversidadedeEvora,

Evora,Portugal;

SchoolofBiology,UniversityofLeeds,Leeds,UK;

DepartmentLandscapeEcologyandEnvironmentalSystemAnalysis,TechnischeUniversität

Braunschweig,Braunschweig,Germany;

GeographyDepartment,Humboldt-UniversitätzuBerlin,Berlin,Germany;

DepartmentofPathology,Microbiology,and

Immunology,SchoolofVeterinaryMedicine,UniversityofCalifornia,Davis,CA,USA;

MundialisGmbH&Co.KG,Bonn,Germany;

DepartmentofEnvironmental

Biology,UniversityofRome“LaSapienza”,Rome,Italy;

KarlsruherInstitutfürTechnologie(KIT),InstitutfürGeographieundGeoökologie,Karlsruhe,Germany;

NaturalEnvironmentCentre,FinnishEnvironmentInstitute(SYKE),Helsinki,Finland;

DepartmentofRemoteSensing,RemoteSensingandBiodiversityResearch

Group,UniversityofWuerzburg,Wuerzburg,Germanyand

AzimPremjiUniversity,Bangalore,India



DuccioRocchini

Emails:ducciorocchini@gmail.com;

duccio.rocchini@fmach.it



LucyBastin,KnowledgeManagement

Unit,JointResearchCentreoftheEuropean

Commission,Ispra,Italy

HandlingEditor:FrancescaParrini

Abstract

1. Biodiversityincludesmultiscalarandmultitemporalstructuresandprocesses,with

differentlevelsoffunctionalorganization,fromgenetictoecosystemiclevels.One

ofthemostlyusedmethodstoinferbiodiversityisbasedontaxonomicapproaches

andcommunityecologytheories.However,gatheringextensivedatainthefieldis

difficultduetologisticproblems,especiallywhenaimingatmodellingbiodiversity

changesinspaceandtime,whichassumesstatisticallysoundsamplingschemes.In

thiscontext,airborneorsatelliteremotesensingallowsinformationtobegathered

overwideareasinareasonabletime.

2. Mostofthebiodiversitymapsobtainedfromremotesensinghavebeenbasedon

theinferenceofspeciesrichnessbyregressionanalysis.Onthecontrary,estimating

compositionalturnover(β-diversity)mightaddcrucialinformationrelatedtorela-

tiveabundanceofdifferentspeciesinsteadofjustrichness.Presently,fewstudies

haveaddressedthemeasurementofspeciescompositionalturnoverfromspace.

3. Extendingonpreviouswork,inthismanuscript,weproposenoveltechniquesto

measure β-diversity from airborne or satellite remote sensing, mainly based on:

(1)multivariatestatisticalanalysis,(2)thespectralspeciesconcept,(3)self-organizing

1788

Methods in Ecology and Evoluon

ROCCHINI et al.

|

Biodiversitycannotbefullyinvestigatedwithoutconsideringthespa-

tialcomponentofitsvariation.Infact,itisknownthatthedispersalof

speciesoverwideareasisdrivenbyspatialconstraintsdirectlyrelated

tothedistanceamongsites.Anegativeexponentialdispersalkernel

isusuallyadoptedtomathematicallydescribetheoccupancyofnew

sitesbyspecies,asfollows:

whered

=distancebetweentwolocationsi and k and aisaparam-

eterregulatingthedispersalfromlocalizedareas(lowvaluesofa)to

widespreadones (highvalues ofa; Meentemeyer,Anacker,Mark, &

Rizzo,2008).

Inthissense,distanceacquiresasignificantroleinecologytoesti-

matebiodiversitychange.Hence,spatiallyexplicitmethodshavebeen

acknowledgedinecologyforprovidingrobustestimatesofdiversity

atdifferenthierarchicallevels:fromindividuals(Tyre,Possingham,&

Lindenmayer,2001),topopulations(Vernesietal.,2012),tocommu-

nities(Rocchini,AndreiniButini,&Chiarucci,2005).

Whendealingwithspatialexplicitmethods,remotesensingimages

represent a powerful tool (Rocchini et al., 2017), particularly when

coupling information on compositional properties of the landscape

withitsstructure(Figure1).Remotesensinghaswidelybeenusedfor

conservationpracticesincludingverydifferenttypesofdatasuchas

nightlightsdata(Mazoretal.,2013),LandSurfaceTemperatureesti-

matedfromMODIS data(Metz,Rocchini,&Neteler,2014),spectral

indices(Gillespie,2005).

Mostoftheremotesensingapplicationsforbiodiversityestima-

tionhavereliedon the estimateoflocal diversityhotspots, consid-

eringlandusediversity(Wegmannetal.,2017)orcontinuousspatial

variabilityofthespectralsignal(Rocchinietal.,2010).Thisismainly

grounded in the assumption that a higher landscape heterogeneity

is strictly related to a higher amount of species occupying differ-

ent niches (Scmheller et al., in press). However, given two sites s

and s

,thefinal diversityisnotonlyrelatedtothespecies/spectral

richness ofs

and s

, but overallto the amount of shared species/

spectralvalues.Inotherwords,thelowerthetheirintersections

∩s

,

the higher will be the total diversity, while the lowest total diver-

sitywillbe reachedwhen s

∩s

= s

∪s

. Such intersection has been

widelystudiedinecology,afterthedevelopmentofβ-diversitytheory

(Whittaker,1960).

Tuomisto etal. (2003) demonstrated the power of substituting

distance in Equation 1 by spectral distance to directly account for

the distance between sites in an environmental space, instead ofa

merelyspatialone. However,whilespectraldistance examples exist

when measuring the β-diversity among pairs of sites (e.g.,Rocchini,

HernándezStefanoni,&He,2015),fewstudieshavetestedthepossi-

bilityofmeasuringβ-diversityoverwideareasconsideringseveralsites

atthesametime(howeverseeAlahuhtaetal.,2017;Harris,Charnock,

&Lucas,2015).Thisisespeciallytruewhenconsideringthedevelop-

mentofremotesensingtools(Rocchini&Neteler,2012)fordiversity

estimateinwhichtheconceptofβ-diversityisstillpioneering.

The aim of this paper is to present the most novel methods to

measureβ-diversityfromremotelysensedimagerybasedonthemost

recentlypublishedecologicalmodels.Inparticular,wewilldealwith:

(1)multivariatestatisticaltechniques,(2)theapplicabilityofthespec-

tralspeciesconcept,(3)multidimensionaldistancematrices,(4)met-

ricscouplingabundanceanddistance-basedmeasures.

Thismanuscriptisthefirstmethodologicalexampleencompassing

(and enhancing) most of the available methods for estimating β-di-

versityfromremotelysensedimageryandpotentiallyrelatethemto

speciesdiversityinthefield.

|





Univariatestatisticshavebeenusedtodirectlyfindrelationsbetween

spectralandspeciesdiversity.However,theamountofvariabilityex-

plainedbysinglebands/vegetationindicesversusspeciesdiversityis

generallyrelativelylow,duetothefactthatdifferentaspectsrelated

tothecomplexityofhabitatsmightactinshapingdiversity,fromdis-

turbanceandlanduseatlocalscalestoclimateandelementfluxesat

globalscales.

Ordination techniques are designed to quantitatively describe

multivariategradualtransitionsinthespeciescompositionofsampled

sites.Measuringthedistancebetweentwosamplingsitesinthemulti-

dimensionalordinationspaceisagoodproxyofthechangeinspecies

composition.Whenthismeasureisrelatedtothegeographicaldistance

(1)

∑

K=1

−d

featuremaps,(4)multidimensionaldistancematrices,andthe(5)Rao'sQdiversity.

Eachofthesemeasuresaddressesoneorseveralissuesrelatedtoturnovermeas-

urement.Thismanuscriptisthefirstmethodological exampleencompassing(and

enhancing)mostoftheavailablemethodsforestimatingβ-diversityfromremotely

sensedimageryandpotentiallyrelatingthemtospeciesdiversityinthefield.



β-diversity,Kohonenself-organizingfeaturemaps,Rao'sQdiversityindex,remotesensing,

satelliteimagery,sparsegeneralizeddissimilaritymodel,spectralspeciesconcept

1789

Methods in Ecology and Evoluon

ROCCHINI et al.

betweentheconsideredsites,thebetadiversityatthisparticularscale

canbeassessed.

Ofthevariousavailableordinationtechniques, detrendedcorre-

spondenceanalysis(DCA;Hill&Gauch,1980)isparticularlysuitable

for such analyses. The axes (i.e., gradients) of the DCA ordination

spacearescaledinSDunits,whereadistanceof4SDisrelatedtoafull

speciesturnover.Thisenablesaversatileanalysisthateasilyreveals

whethertwosampledsitesstillhavespeciesincommon.

Severalstudies havemapped the ordination space using remote

sensing data (e.g., Feilhauer & Schmidtlein, 2009; Feilhauer, Faude,

&Schmidtlein, 2011; Feilhaueretal.,2014;Gu, Singh,&Townsend,

2015; Harrisetal., 2015; Leitãoetal., 2015; Neumann etal., 2015;

Schmidtlein&Sassin,2004;Schmidtlein,Zimmermann,Schüpferling,

&Weiss,2007).Forthispurpose,theaxesscoresofthesampledsites

are regressed against the corresponding canopy reflectance values

extractedfromair-orspaceborneimagedata.Theresultingmultivar-

iateregressionmodels,oneperordinationaxisandmostoftengener-

atedwithmachine learning regressiontechniques, aresubsequently

appliedontheimagedataforaspatialpredictionofordinationscores.

Eachpixeloftheimagedataisassignedtoaspecificpositioninthe

ordinationspacethatindicatesitsspeciescomposition.Theresulting

gradientmapsareapowerfultoolforanalysesofbetadiversityacross

different spatial scales (Feilhauer & Schmidtlein, 2009; Hernandez-

Stefanonietal.,2012).

AsimpleanalysisofthevariabilityoftheDCAscoresinadefined

pixelneighbourhood(i.e.,amovingwindow)resultsinaefficientbeta

diversityassessment.Thespatialscaleofthisassessmentcanbevaried

eitherbyresamplingthegradientmaptoacoarserspatialresolution

(i.e.,pixelsize)orbychangingthekernelsizeoftheconsideredpixel

neighbourhood. Such techniques have been further developed e.g.

for spatial conservation prioritization programmes such as Zonation

(Moilanenetal.,2005;Moilanen,Kujala,&Leathwick,2009).

Figure2showsanexampleofaDCA-based assessmentofbeta

diversityonaverylocalscale(10m)followingtheapproachdescribed

inFeilhauerandSchmidtlein(2009).Theanalysedlandscapeisamo-

saicofraisedbogs,fens,transitionmiresandMoliniameadows.Fora

detaileddescriptionofthedataandsitepleaserefertoFeilhaueretal.

(2014)andFeilhauer,Doktor,Schmidtlein,andSkidmore(2016).

Analyses like this require two different datasets: (1) a sample

offielddatathatisrepresentativeforthevegetationinthestudied

areaandis used togeneratethe ordinationspace;(2) imagedata

withasufficientspectralresolutiontodiscriminatethevegetation

typeswithintheordinationspaceandwithaspatialresolutionthat

isinlinewiththesamplingdesignofthefielddata(Feilhaueretal.,

2013).

FIGURE 1 Anexampleofhowtocoupleinformationon

compositionalpropertiesofthelandscapebyopticaldatatogether

withstructural(3D)propertiesbylaserscanningLiDARdata

FIGURE 2 β-diversityassessmentwithacombinationofordinationtechniquesandremotesensing.(a)Three-dimensionaldetrended

correspondenceanalysis(DCA)ordinationspaceofn=100vegetationplotssampledinraisedbogs,fens,transitionmiresandMoliniameadows

inthealpinefoothillsofSouthernGermany.Aninter-plotdistanceof4SDcorrespondstoafullspeciesturnover.(b)Mapsoftheordinationaxes

resultingfromaspatialpredictionbasedoncanopyreflectance.Eachpixelhasapredictedpositionintheordinationspacethatisindicatedbyits

colour.Thecolourschemecorrespondsto(a).Themaphasaspatialresolutionof2×2m

,whichisinlinewiththesampledplotsize.

(c)CumulativechangeratesalongthethreeDCAaxesina5×5pixelneighbourhood.Ahighchangerateindicatesahighbetadiversity

1790

Methods in Ecology and Evoluon

ROCCHINI et al.

Using these data, the continuous spatial variability of the spec-

tralsignalintheimagepixelsistranslatedintoaspatiallycontinuous

measureofspeciescomposition.Theadvantagesofthisapproachare

obvious:sincethediversityanalysesareconductedinthefloristicgra-

dientspace,theresultingmeasuresresemblefieldstudiesandarethus

easiertointerpretthanspectralproxiesandclosertothepointofview

ofmanyend-users.Furthermore,theanalysisofordinationscoresin

definedpixelneighbourhoodsisnotrestrictedtoasinglespatialscale

butofferstheopportunitytoimplementassessmentsofbetadiversity

onmultiplescales.

|

ThespectralspeciesconcepthasbeenproposedbyFéretandAsner

(2014a) to map both α and β component of the biodiversity using

a unique framework. It is rooted in the convergence between two

otherconcepts,thespectralvariationhypothesis(SVH)proposedby

Palmer, Earls, Hoagland, White, and Wohlgemuth (2002), and the

plantopticaltypesproposedbyUstinandGamon(2010),sustained

bythetechnologicaladvances inthedomain of highspatialresolu-

tionimagingspectroscopy.TheSVHstatesthatthespatialvariability

in the remotely sensed signal, that is the spectral heterogeneity, is

relatedtoenvironmentalheterogeneityandcouldthereforebeused

asapowerfulproxyofspeciesdiversity.SVHhasbeentestedindif-

ferentsituations(Rocchinietal.,2010)andconclusionsshowthatthe

performanceofthisapproachisverydependentonseveralfactors,

includingtheinstrumentcharacteristics(spectral,spatialandtempo-

ralresolution), thetype ofvegetation investigated,and the metrics

derivedfromremotelysensedinformationtoestimatespectralheter-

ogeneity.Plantopticaltypesrefertothecapacityofsensorstomeas-

ure signals that aggregate information about vegetation structure,

phenology, biochemistry and physiology. Therefore, this concept is

alsotightlylinkedtotheperformancesofthesensorandfindspar-

ticularechowiththeincreasinguseofhighspatialresolutionimaging

spectroscopyfortheestimationandidentificationofmultiplevegeta-

tionproperties.

Thedetailsprovidedbyhighspatialresolutionimagingspectros-

copyaresufficient toperform analysesofplant optical traits at the

individualtreescaleinordertodifferentiatetreespecies,obtainin-

formationaboutleafchemicaltraitsandestimatetheαcomponentof

biodiversity(Asner&Martin,2008;Asner,Martin,Anderson,&Knapp,

2015;Chadwick&Asner,2016;Clark&Roberts,2012;Clark,Roberts,

&Clark,2005;Féret&Asner,2013;VaglioLaurinetal.,2014).These

resultsillustratethatspectralinformationcanberelatedtotaxonomic

orfunctionalinformationofthevegetation,whichsupportstheSVH

underthehypothesisthatthemetricsusedtocomputespectralhet-

erogeneityandagivencomponentofvegetationdiversityareprop-

erlydefined.Howevertheseapplicationsarecurrentlylimitedbythe

importantamountoffielddatarequiredtotrainregressionorclassi-

ficationmodels,whichisalsodirectlylinkedto theirlowgeneraliza-

tionabilityintimeandspace.Unsupervisedapproachesthenappear

asvaluablealternativesforthe analysisofecosystemheterogeneity

(Baldeck & Asner, 2013; Baldeck etal., 2014; Feilhauer, Faude, &

Schmidtlein,2011;Féret&Asner,2014b),asecologicalindicatorsofα

and βdiversityatlandscapescaleusuallyrequireoneorseverallevels

ofabstractionbeyondthecorrecttaxonomicidentification(Tuomisto

&Ruokolainen,2006).

Clustering(properlypre-processed)spectralinformationshouldre-

sultinpixelsfromthesamespeciesnaturallygroupingtogetherrather

thandistributing randomlyamong clusters,FéretandAsner(2014a)

proposedagroupingmethodaimingatassigninglabelstopixelsbased

on multiple clustering of spectroscopic data acquired at landscape

scale.Thesepixels,labelledwithasetoftheso-calledspectralspecies,

can then be used straightforwardly in orderto computevarious di-

versitymetricssuchasShannonindexforαdiversity,andBray-Curtis

dissimilarityforβ diversity.Thepre-processing stage is divided into

severalstages.Aftermaskingallnon-vegetatedpixels,anormalization

based on continuous removal is applied to each pixel and over the

fullspectraldomain,thenaprincipalcomponentanalysisisperformed

onthecontinuouslyremovedspectraldata.Thenormalizationreduces

effectsduetochangesinillumination,canopygeometryandotherfac-

torsunrelatedtovegetation,whileenhancingthesignalcorresponding

tovegetation.Thecomponentsincludingindividual-specificinforma-

tionarethe componentsofinterest.Theycan be identifiedaftervi-

sual inspection or automated routines,if initial data showsufficient

signaltonoiseratio.Oncealimitednumberofcomponentshavebeen

selected, k-means clustering is then applied to a certain numberof

subsets,andforeachofthesesubsets,centroidsarecomputedand

eachpixelintheimageislabelledbasedontheclosestcentroid.The

repetitionofclusteringbasedonvarioussubsetsoftheimagetendsto

minimizetheriskofassigningcentroidstoirrelevantgroupsofpixels.

Experimental results showed that the averaging of diversity indices

computedfrommultiplecentroidmapscanbeseenasananalogous

tosignalaveraging,whichconsistsinincreasingsignal tonoiseratio

byreplicatingmeasurements.Foreachrepetition,theclosestcentroid

correspondstothespectralspecies,andforeachspatialunitofagiven

size,thespectralspeciesdistributionisderivedinordertocompute

anydiversitymetricrequiringeitherinformationatthelocalscale,or

comparisonofinformationacrossspatiallydistantplots.

Theconceptsofspectralspeciesandspectralspeciesdistribution

havebeentestedsuccessfullyonalimitednumberofsituationsand

typesofecosystems(seeRocchinietal.,2016forareview,andLausch

etal., 2016 for an application to similar concepts). As an example,

FéretandAsner(2014a)showedabilitytoproperlyestimatelandscape

heterogeneityatmoderatespatialscale,uptofewdozensquarekilo-

metersovertropicalforests,basedonhighspatialresolutionimaging

spectroscopy (Figure 3). A generic parameterization of the method

showed robust performances for α diversity mapping across space

andtime,butmappingβdiversityacrosslargespatialscalesusingim-

agesacquiredduringdifferentairbornecampaignremainschallenging,

whichleadstoanunsolvedproblemwhenconsideringoperationalre-

gionalmapping.Intheperspectiveofglobalmonitoringofbiodiversity,

andgiventheunprecedentedremotesensingcapacityallowedbythe

Copernicusprogram,includingtheSentinel-2multispectralsatellites,

severalotherchallengesareforeseenandcurrentlyinvestigated.The

1791

Methods in Ecology and Evoluon

ROCCHINI et al.

influenceofdecreasedspatialandspectralresolutionontheabilityto

properlydifferentiateecologicallymeaningfulspectralspeciesacross

landscapesandoverregionswillneedtobeinvestigated.Theapplica-

tionofthisconceptbeyondtropicalforestsandsavannaecosystems

shouldalsobeinvestigated,asitmaynotholdwhenappliedonmoder-

atelydiverseecosystemsorsystemswithindividualswhoseindividu-

alshavedimensionswellbelowtheresolvingpoweroftheinstrument.

|

TheKohonenself-organizingfeaturemap(SOFM;Kohonen,1982)is

aneuralnetworkthatmaybeusedtoundertakeunsupervisedcluster-

ingofdata.Critically,theinputtoaSOFMcanbealargemulti-variate

datasetsuchasmaybeacquiredonspeciesfromquadrat-basedfield

surveys.TheSOFMsummarizesthedatain a low, typicallytwo,di-

mensionaloutput(Figure4).Inthisoutputspace,thedataforindivid-

ualquadratsaretopologicallyordered—withsitesthataresimilarclose

togetherwhilethoseofhighlydifferentspeciescompositionaremore

distant.Because thedata sitesin the output space arearranged by

FIGURE 3 Spectralspeciescanbeidentifiedinahyper-ormultispectralimagebyspatialclusteringmethodandtheirdistributioncanbe

mapped.Suchmapscanfurtherbeusedtoapplylocal-basedheterogeneitymeasurements(α-diversity)aswellasiterativedistance-based

methodstobuildβ-diversitymaps.ReproducedfromFéretandAsner(2014a)

FIGURE 4 Aself-organizingfeaturemapcanbebuiltstarting

fromaninputlayer,e.g.thepresenceorabsenceofatreespeciesor

ofapeculiarspectralvalue)whichisconnectedtoeveryunitinthe

outputlayerbyaweightedconnection.Theself-organizingfeature

mapusesunsupervisedlearningtomapthelocationoffieldsites

withintheoutputspaceonthebasisoftheirrelativesimilarityin

speciesorspectralcomposition.RedrawnfromFoodyandCutler

(2003)

Output layer

Input units

Measuring β-diversity by remote sensing: A challenge for biodiversity monitoring

Citations

Scaling-up biodiversity-ecosystem functioning research.

Remote sensing of terrestrial plant biodiversity

ZONATION Framework and Software for Conservation Prioritization

Radar vision in the mapping of forest biodiversity from space

Discovering floristic and geoecological gradients across Amazonia

References

Vegetation of the Siskiyou Mountains, Oregon and California

iNEXT: an R package for rarefaction and extrapolation of species diversity (Hill numbers)

Analyzing beta diversity: partitioning the spatial variation of community composition data

Using generalized dissimilarity modelling to analyse and predict patterns of beta diversity in regional biodiversity assessment

Hyperspectral discrimination of tropical rain forest tree species at leaf to crown scales

Related Papers (5)

Quantitative tools for perfecting species lists

Remote sensing of plant functional types

Airborne spectranomics: mapping canopy chemical and taxonomic diversity in tropical forests

Essential biodiversity variables

Remote sensing for biodiversity science and conservation

Frequently Asked Questions (16)

Q1. What are the contributions in "Measuring β-diversity by remote sensing: a challenge for biodiversity monitoring" ?

Q2. What future works have the authors mentioned in the paper "Measuring β-diversity by remote sensing: a challenge for biodiversity monitoring" ?

Q3. What is the purpose of the self-organizing feature map?

Q4. How can the authors combine remote sensing data and biodiversity data in the field?

Q5. What is the common way to describe the occupancy of new sites by species?

Q6. What are the advantages of this approach?

Q7. What are the challenges of the Copernicus program?

Q8. What is the definition of a penalized canonical correlation analysis?

Q9. How is the compositional dissimilarity model used?

Q10. What are the suggested methods for -diversity estimation from remote sensing?

Q11. How is the Rao's Q diversity index calculated?

Q12. What can be used to build -diversity maps?

Q13. Who proposes new techniques to measure biodiversity from airborne or satellite remote sensing?

Q14. What could be used to regress species diversity against remote sensing data?

Q15. What is the example of a generalized entropy theory?

Q16. What is the difference between averaging and signal averaging?