pan‐tropical prediction of forest structure from the ... · bastin et al. | 3 36malaysia campus,...

18
Global Ecol Biogeogr. 2018;1–18. wileyonlinelibrary.com/journal/geb | 1 © 2018 John Wiley & Sons Ltd Received: 1 November 2017 | Revised: 31 May 2018 | Accepted: 13 June 2018 DOI: 10.1111/geb.12803 RESEARCH PAPER Pan‐tropical prediction of forest structure from the largest trees Jean‐François Bastin 1,2,3,4 | Ervan Rutishauser 4,5 | James R. Kellner 6,7 | Sassan Saatchi 8 | Raphael Pélissier 9 | Bruno Hérault 10,11 | Ferry Slik 12 | Jan Bogaert 13 | Charles De Cannière 3 | Andrew R. Marshall 14,15,16 | John Poulsen 17 | Patricia Alvarez‐Loyayza 18 | Ana Andrade 19 | Albert Angbonga‐Basia 20 | Alejandro Araujo‐Murakami 21 | Luzmila Arroyo 22 | Narayanan Ayyappan 23,24 | Celso Paulo de Azevedo 25 | Olaf Banki 26 | Nicolas Barbier 9 | Jorcely G. Barroso 26 | Hans Beeckman 27 | Robert Bitariho 28 | Pascal Boeckx 29 | Katrin Boehning‐ Gaese 30,31 | Hilandia Brandão 32 | Francis Q. Brearley 33 | Mireille Breuer Ndoundou Hockemba 34 | Roel Brienen 35 | Jose Luis C. Camargo 19 | Ahimsa Campos‐Arceiz 36 | Benoit Cassart 37,38 | Jérôme Chave 39 | Robin Chazdon 40 | Georges Chuyong 41 | David B. Clark 42 | Connie J. Clark 17 | Richard Condit 43 | Euridice N. Honorio Coronado 44 | Priya Davidar 22 | Thalès de Haulleville 13,27 | Laurent Descroix 45 | Jean‐Louis Doucet 13 | Aurelie Dourdain 46 | Vincent Droissart 9 | Thomas Duncan 47 | Javier Silva Espejo 48 | Santiago Espinosa 49 | Nina Farwig 50 | Adeline Fayolle 13 | Ted R. Feldpausch 51 | Antonio Ferraz 8 | Christine Fletcher 36 | Krisna Gajapersad 52 | Jean‐François Gillet 13 | Iêda Leão do Amaral 32 | Christelle Gonmadje 53 | James Grogan 54 | David Harris 55 | Sebastian K. Herzog 56 | Jürgen Homeier 57 | Wannes Hubau 27 | Stephen P. Hubbell 5,58 | Koen Hufkens 29 | Johanna Hurtado 59 | Narcisse G. Kamdem 60 | Elizabeth Kearsley 61 | David Kenfack 62 | Michael Kessler 63 | Nicolas Labrière 10,64 | Yves Laumonier 10,65 | Susan Laurance 66 | William F. Laurance 66 | Simon L. Lewis 35 | Moses B. Libalah 60 | Gauthier Ligot 13 | Jon Lloyd 67,68 | Thomas E. Lovejoy 68 | Yadvinder Malhi 69 | Beatriz S. Marimon 70 | Ben Hur Marimon Junior 70 | Emmanuel H. Martin 71,72 | Paulus Matius 73 | Victoria Meyer 8 | Casimero Mendoza Bautista 74 | Abel Monteagudo‐Mendoza 75 | Arafat Mtui 76 | David Neill 77 | Germaine Alexander Parada Gutierrez 21 | Guido Pardo 78 | Marc Parren 79 | N. Parthasarathy 23 | Oliver L. Phillips 35 | Nigel C. A. Pitman 79 | Pierre Ploton 9 | Quentin Ponette 37 | B. R. Ramesh 23 | Jean‐Claude Razafimahaimodison 80 | Maxime Réjou‐Méchain 9 |

Upload: duongdat

Post on 14-Aug-2019

214 views

Category:

Documents


0 download

TRANSCRIPT

Global Ecol Biogeogr. 2018;1–18. wileyonlinelibrary.com/journal/geb  | 1© 2018 John Wiley & Sons Ltd

Received:1November2017  |  Revised:31May2018  |  Accepted:13June2018DOI:10.1111/geb.12803

R E S E A R C H P A P E R

Pan‐tropical prediction of forest structure from the largest trees

Jean‐François Bastin1,2,3,4  | Ervan Rutishauser4,5  | James R. Kellner6,7 |  Sassan Saatchi8 | Raphael Pélissier9 | Bruno Hérault10,11 | Ferry Slik12 |  Jan Bogaert13 | Charles De Cannière3 | Andrew R. Marshall14,15,16 | John Poulsen17 | Patricia Alvarez‐Loyayza18 | Ana Andrade19 | Albert Angbonga‐Basia20 |  Alejandro Araujo‐Murakami21 | Luzmila Arroyo22 | Narayanan Ayyappan23,24 |  Celso Paulo de Azevedo25 | Olaf Banki26 | Nicolas Barbier9 | Jorcely G. Barroso26 |  Hans Beeckman27 | Robert Bitariho28 | Pascal Boeckx29 | Katrin Boehning‐Gaese30,31 | Hilandia Brandão32 | Francis Q. Brearley33 |  Mireille Breuer Ndoundou Hockemba34 | Roel Brienen35 | Jose Luis C. Camargo19 |  Ahimsa Campos‐Arceiz36 | Benoit Cassart37,38 | Jérôme Chave39 | Robin Chazdon40 |  Georges Chuyong41 | David B. Clark42 | Connie J. Clark17 | Richard Condit43 |  Euridice N. Honorio Coronado44 | Priya Davidar22 | Thalès de Haulleville13,27 |  Laurent Descroix45 | Jean‐Louis Doucet13 | Aurelie Dourdain46 | Vincent Droissart9 |  Thomas Duncan47 | Javier Silva Espejo48 | Santiago Espinosa49 | Nina Farwig50 |  Adeline Fayolle13 | Ted R. Feldpausch51 | Antonio Ferraz8 | Christine Fletcher36 |  Krisna Gajapersad52 | Jean‐François Gillet13 | Iêda Leão do Amaral32 |  Christelle Gonmadje53 | James Grogan54 | David Harris55 | Sebastian K. Herzog56 |  Jürgen Homeier57 | Wannes Hubau27 | Stephen P. Hubbell5,58 | Koen Hufkens29  |  Johanna Hurtado59 | Narcisse G. Kamdem60 | Elizabeth Kearsley61 | David Kenfack62 |  Michael Kessler63 | Nicolas Labrière10,64 | Yves Laumonier10,65 | Susan Laurance66 |  William F. Laurance66 | Simon L. Lewis35 | Moses B. Libalah60 | Gauthier Ligot13 |  Jon Lloyd67,68 | Thomas E. Lovejoy68 | Yadvinder Malhi69 | Beatriz S. Marimon70 |  Ben Hur Marimon Junior70 | Emmanuel H. Martin71,72 | Paulus Matius73 |  Victoria Meyer8 | Casimero Mendoza Bautista74 | Abel Monteagudo‐Mendoza75 |  Arafat Mtui76 | David Neill77 | Germaine Alexander Parada Gutierrez21 | Guido Pardo78 |  Marc Parren79 | N. Parthasarathy23 | Oliver L. Phillips35 | Nigel C. A. Pitman79 |  Pierre Ploton9 | Quentin Ponette37 | B. R. Ramesh23 |  Jean‐Claude Razafimahaimodison80 | Maxime Réjou‐Méchain9 | 

2  |     BASTIN et al.

Samir Gonçalves Rolim81 | Hugo Romero Saltos82 | Luiz Marcelo Brum Rossi81 |  Wilson Roberto Spironello32 | Francesco Rovero76 | Philippe Saner83 |  Denise Sasaki84 | Mark Schulze85 | Marcos Silveira86 | James Singh87 |  Plinio Sist10,88 | Bonaventure Sonke60 | J. Daniel Soto21 | Cintia Rodrigues de Souza24 |  Juliana Stropp89 | Martin J. P. Sullivan35 | Ben Swanepoel34 | Hans ter Steege25,90 |  John Terborgh91,92 | Nicolas Texier93 | Takeshi Toma94 | Renato Valencia95 |  Luis Valenzuela75 | Leandro Valle Ferreira96 | Fernando Cornejo Valverde97 |  Tinde R. Van Andel25 | Rodolfo Vasque77 | Hans Verbeeck61 | Pandi Vivek22 |  Jason Vleminckx98 | Vincent A. Vos78,99 | Fabien H. Wagner100 | Papi Puspa, Warsudi101 | Verginia Wortel102 | Roderick J. Zagt103 | Donatien Zebaze60

1DepartmentofEnvironmentalSystemsScience,InstituteofIntegrativeBiology,ETHZürich,Zürich,Switzerland2 CarbonCycleandEcosystemGroupVS,NASA,JetPropulsionLaboratory,CaliforniaInstituteofTechnology,Pasadena,California,USA3LandscapeEcologyandPlantProductionSystem,UniversitélibredeBruxelles.CP264‐2,Bruxelles,Belgium4Carboforexpert(carboforexpert.ch),Hermance,Switzerland5SmithsonianTropicalResearchInstitute,Balboa,Panama6DepartmentofEcologyandEvolutionaryBiology,BrownUniversity,Providence,RhodeIsland

7InstituteatBrownforEnvironmentandSociety,BrownUniversity,Providence,RhodeIsland8NASA,JetPropulsionLaboratory,CaliforniaInstituteofTechnology,Pasadena,California, USA9AMAPLab,IRD,CIRAD,CNRS,INRA,Univ.Montpellier,Montpellier,France10Cirad,URForest&Societies,Montpellier,France11INPHB(InstitutNationalPolytechniqueFélixHouphouetBoigny),Yamoussoukro,IvoryCoast12FacultyofScience,UniversitiBruneiDarusallam,Gadong,BruneiDarussalam13GemblouxAgro‐BioTech,UniversitédeLiège,Gembloux,Belgium14CIRCLE,EnvironmentDepartment,WentworthWay,UniversityofYork,York,UnitedKingdom15TropicalForestsandPeopleResearchCentre,UniversityoftheSunshineCoast,SunshineCoast,Queensland,Australia16FlamingoLandLtd,KirbyMisperton,UnitedKingdom17NicholasSchooloftheEnvironment,DukeUniversity,Durham,NorthCarolina,USA18FieldMuseumofNaturalHistory,Chicago,Illinois,USA19BiologicalDynamicsofForestFragmentProject(BDFFP–INPA/STRI),Manaus,Brazil20InstitutFacultairedesSciencesAgronomiquesdeYangambi,Yangambi,DemocraticRepublicoftheCongo21MuseodeHistoriaNaturalNoelKempffMercado,SantaCruz,Bolivia22DepartmentofEcologyandEnvironmentalSciences,PondicherryUniversity,Pondicherry,India23FrenchInstituteofPondicherry(IFP),Pondicherry,India24EmbrapaAmazôniaOcidental,Manaus,Brazil25NaturalisBiodiversityCentre,Leiden,TheNetherlands26UniversidadeFederaldoAcre,CampusFloresta,Acre,Brazil27ServiceofWoodBiology,RoyalMuseumforCentralAfrica,Tervuren,Belgium28InstituteofTropicalForestConservation,MbararaUniversityofScienceandTechnology,Mbarara,Uganda29IsotopeBioscienceLaboratory–ISOFYS,GhentUniversity,Ghent,Belgium30SenckenbergBiodiversityandClimateResearchCentre(BiK‐F),FrankfurtamMain,Germany31DepartmentofBiologicalSciences,GoetheUniversität,FrankfurtamMain,Germany32NationalInstituteforAmazonianResearch(INPA),Manaus,Brazil33SchoolofScienceandtheEnvironment,ManchesterMetropolitanUniversity,Manchester,UnitedKingdom34WildlifeConservationSociety,NewYork,NewYork, USA35SchoolofGeography,UniversityofLeeds,Leeds,UnitedKingdom

     |  3BASTIN et al.

36MalaysiaCampus,JalanBroga,Semenyih,Malaysia37UCL‐ELI,EarthandLifeInstitute,UniversitécatholiquedeLouvain,Louvain‐la‐Neuve,Belgium38EcoleRégionalePost‐universitaired’AménagementetdeGestionIntégrésdesForêtsetTerritoiresTropicaux,Kinshasa,DemocraticRepublicoftheCongo39LaboratoireEvolutionetDiversitébiologique,CNRS&UniversitéPaulSabatier,Toulouse,France40DepartmentofEcologyandEvolutionaryBiology,UniversityofConnecticut,Storrs,Connecticut,USA41DepartmentofBotanyandPlantPhysiology,UniversityofBuea,Buea,Cameroon42DepartmentofBiology,UniversityofMissouri‐StLouis,StLouis,Missouri, USA43FieldMuseumofNaturalHistoryandMortonArboretum,Chicago,Illinois, USA44Coronado,Inst.deInvestigacionesdelaAmazoniaPeruana,Iquitos,Peru45ONFpôleR&D,Cayenne,France46Cirad,UMREcoFoG(AgroParisTech,CNRS,Inra,UniversitedesAntilles,UniversitedelaGuyane),Kourou,FrenchGuiana47DepartmentofBotanyandPlantPathology,OregonStateUniversity,Corvallis,Oregon48DepartamentodeBiología,UniversidaddeLaSerena,LaSerena,Chile49UniversidadAutónomadeSanLuisPotosí,SanLuisPotosí,México50DepartmentofConservationEcology,Philipps‐UniversitätMarburg,Marburg,Germany51Geography,CollegeofLifeandEnvironmentalSciences,UniversityofExeter,Exeter,UnitedKingdom52ConservationInternationalSuriname,Paramaribo,Suriname53FacultyofScience,DepartmentofPlantBiology,UniversityofYaoundeI,Yaoundé,Cameroon54SmithCollegeBotanicGarden,Northampton,Massachusetts55RoyalBotanicGardenEdinburgh,Edinburgh,UnitedKingdom56MuseodeHistoriaNaturalAlcided’Orbigny,Cochabamba,Bolivia57PlantEcology,UniversityofGoettingen,Goettingen,Germany58DepartmentofEcologyandEvolutionaryBiology,UniversityofCalifornia,LosAngeles,California, USA59OrganizationforTropicalStudies,LaSelva,CostaRica60PlantSystematicandEcologyLaboratory,HigherTeacher’sTrainingCollege,UniversityofYaoundéI,Yaoundé,Cameroon61CAVElab–ComputationalandAppliedVegetationEcology,GhentUniversity,Ghent,Belgium62CTFS‐ForestGEO,SmithsonianTropicalResearchInstitute,Washington,USA63DepartmentofSystematicandEvolutionaryBotany,UniversityofZurich,Zurich,Switzerland64AgroParisTech,DoctoralSchoolABIES,Paris,France65CenterforInternationalForestryResearch,Bogor,Indonesia66CentreforTropicalEnvironmentalandSustainabilityScience,CollegeofScienceandEngineering,JamesCookUniversity,Cairns,Queensland,Australia67DepartmentofLifeSciences,ImperialCollegeLondon,Ascot,UnitedKingdom68DepartmentofEnvironmentalScienceandPolicy,GeorgeMasonUniversity,Fairfax,Virginia69EnvironmentalChangeInstitute,SchoolofGeographyandtheEnvironment,UniversityofOxford,Oxford,UnitedKingdom70UniversidadedoEstadodeMatoGrosso,CampusdeNovaXavantina,NovaXavantina,Brazil71UdzungwaEcologicalMonitoringCentre,UdzungwaMountainsNationalPark,Mang’ula,Tanzania72SokoineUniversityofAgriculture,Morogoro,Tanzania73FacultyofForestry,MulawarmanUniversity,Samarinda,Indonesia74EscueladeCienciasForestales,UnidadAcadémicadelTrópico,UniversidadMayordeSanSimón,Sacta,Bolivia75JardínBotánicodeMissouri,Oxapampa,Peru76MUSE‐MuseodelleScienze,Trento,Italy77UniversidadEstatalAmazónica,Puyo,Pastaza,Ecuador78UniversidadAutónomadelBeni,Riberalta,Bolivia79ScienceandEducation,TheFieldMuseum,Chicago,Illinois80CentreValBio,Ranomafana,Madagascar81EmbrapaFlorestas,Colombo,Brazil82SchoolofBiologicalSciencesandEngineering,YachayTechUniversity,Urcuquí,Ecuador83DepartmentofEvolutionaryBiologyandEnvironmentalStudies,UniversityofZurich,Zurich,Switzerland84FundaçãoEcológicaCristalinoAltaFloresta,MatoGrosso,Brazil85H.J.AndrewsExperimentalForest,BlueRiver,Oregon86MuseuUniversitário,UniversidadeFederaldoAcre,RioBranco,Brazil

4  |     BASTIN et al.

87GuyanaForestryCommission,Georgetown,Guyana88ForestsandSocieties,Univ.Montpellier,CIRAD,Montpellier,France89InstituteofBiologicalandHealthSciences,FederalUniversityofAlagoas,Maceió,Brazil90SystemsEcology,FreeUniversity,Amsterdam,TheNetherlands91FloridaMuseumofNaturalHistoryandDepartmentofBiology,UniversityofFlorida–Gainesville,Gainesville,Florida92DepartmentofBiology,JamesCookUniversity,Cairns,Queensland,Australia93Laboratoired’EvolutionBiologiqueetEcologie,FacultédesSciences,UniversitélibredeBruxelles,Bruxelles,Belgium94ForestryandForestProductsResearchInstitute,Tsukuba,Japan95EscueladeCienciasBiológicas,PontificiaUniversidadCatólicadelEcuador,Quito,Ecuador96CoordenaçãodeBotânica,MuseuParaenseEmilioGoeldi,Belém,Brazil97AndestoAmazonBiodiversityProgram,MadredeDios,Peru98DepartmentofIntegrativeBiology,UniversityofCalifornia,Berkeley,California99CentrodeInvestigaciónyPromocióndelCampesinado–NorteAmazónico,Riberalta,Bolivia100RemoteSensingDivision,NationalInstituteforSpaceResearch–INPE,SãoJosédosCampos,Brazil101TheCenterforReforestationStudiesintheTropicalRainForest(PUSREHUT),MulawarmanUniversity,EastKalimantan,Indonesia102CenterforAgriculturalResearchinSuriname(CELOS),Paramaribo,Suriname103TropenbosInternational,Wageningen,TheNetherlands

CorrespondenceJean‐FrançoisBastin,InstituteofIntegrativeBiology,DepartmentofEnvironmentalSystemsScience,ETHZürich,8092Zürich,Switzerland.Email:[email protected]

AbstractAim:Largetropical trees formthe interfacebetweengroundandairborneobserva‐tions,offeringauniqueopportunitytocaptureforestpropertiesremotelyandtoinves‐tigatetheirvariationsonbroadscales.However,despiterapiddevelopmentofmetricstocharacterizetheforestcanopyfromremotelysenseddata,agapremainsbetweenaerialandfieldinventories.Toclosethisgap,weproposeanewpan‐tropicalmodeltopredictplot‐levelforeststructurepropertiesandbiomassfromonlythelargesttrees.Location:Pan‐tropical.Time period:Early21stcentury.Major taxa studied:Woodyplants.Methods:Usingadatasetof867plotsdistributedamong118sitesacrossthetropics,wetestedthepredictionofthequadraticmeandiameter,basalarea,Lorey’sheight,communitywooddensityandabovegroundbiomass(AGB)fromtheithlargesttrees.Results:Measuringthelargesttreesintropicalforestsenablesunbiasedpredictionsofplot‐andsite‐levelforeststructure.The20largesttreesperhectarepredictedquad‐raticmeandiameter,basal area, Lorey’sheight, communitywooddensity andAGBwith12,16,4,4and17.7%ofrelativeerror,respectively.Mostoftheremainingerrorinbiomasspredictionisdrivenbydifferencesintheproportionoftotalbiomassheldinmedium‐sizedtrees(50–70cmdiameteratbreastheight),whichshowssomeconti‐nentaldependency,withAmericantropicalforestspresentingthehighestproportionoftotalbiomassintheseintermediate‐diameterclassesrelativetoothercontinents.Main conclusions:Ourapproachprovidesnewinformationontropicalforeststruc‐tureandcanbeusedtogenerateaccuratefieldestimatesoftropicalforestcarbonstocks to support the calibration and validationof current and forthcoming spacemissions.Itwillreducethecostoffieldinventoriesandcontributetoscientificunder‐standingoftropicalforestecosystemsandresponsetoclimatechange.

K E Y W O R D S

carbon,climatechange,foreststructure,largetrees,pan‐tropical,REDD+,tropicalforestecology

     |  5BASTIN et al.

1  | INTRODUC TION

The fundamental ecological function of large trees is well estab‐lished for tropical forests. They offer shelter to many organisms(Lindenmayer,Laurance,&Franklin,2012;Remm&Lõhmus,2011),regulate forest dynamics, regeneration (Harms,Wright, Calderón,Hernández,&Herre,2000;Rutishauser,Wagner,Herault,Nicolini,&Blanc,2010)andtotalbiomass(Stegenetal.,2011),andareimport‐antcontributors to theglobal carboncycle (Meakemetal.,2018).Beingmajorcomponentsofthecanopy,thelargesttreesmayalsosuffermore than sub‐canopy and understorey trees from climatechange,becausetheyaredirectlyexposedtovariationsinsolarradi‐ation,windstrength,temperatureseasonalityandrelativeairhumid‐ity(Bennett,McDowell,Allen,&Anderson‐Teixeira,2015;Laurance,Delamônica,Laurance,Vasconcelos,&Lovejoy,2000;Lindenmayeretal.,2012;Meakemetal.,2018;Nepstad,Tohver,Ray,Moutinho,&Cardinot,2007;Thomas,Kellner,Clark,&Peart,2013).Giventhattheyarevisiblefromthesky,largetreesareidealformonitoringfor‐estresponsestoclimatechangeviaremotesensing(RS;Asneretal.,2017;Bennettetal.,2015).

Largetreesencompassadisproportionatefractionoftotalabo‐vegroundbiomass (AGB) in tropical forests (Chave,Riera,Dubois,&Riéra,2001;Lutzetal.,2018),withsomevariations intheir rel‐ative contribution to the total AGB among the tropical regions(Feldpauschetal.,2012).InCentralAfrica,thelargest5%oftreesinaforestsampleplot(i.e.,the5%oftreeswiththelargestdiameterat130cm)store50%of forestplotAGBonaverage (Bastinetal.,2015).Consequently,thedensityoflargetreeslargelyexplainsvari‐ationinforestAGBatlocal(Clark&Clark,1996),regional(Malhietal.,2006;Saatchi,Houghton,DosSantosAlvalá,Soares,&Yu,2007)andcontinentalscales(Sliketal.,2013;Stegenetal.,2011).Detailingthecontributionofeach single tree to thediameter structure,weshowedpreviouslythatplot‐levelAGBcanbepredictedfromafewlargetrees(Bastinetal.,2015),withthemeasurementofthe20larg‐esttreesperhectarebeingsufficienttoestimateplot‐levelbiomasswith<15%error inreferencetogroundestimates.Thesefindingssuggested that a substantial gain of cost‐effectiveness might beachievedbyfocusingforest inventoriesonthe largesttreesratherthanallsizeclasses.Likewise,itsuggestedthatRSapproachescouldfocusonthemeasurementofthelargesttrees,insteadofpropertiesoftheentireforeststand.

SeveraleffortsareunderwaytoclosethegapbetweenRSofforestbiomassandfieldsurveys(Coomesetal.,2017;Juckeretal.,2017).However,existingRSapproachestypicallyrequiregroundmeasurement of all trees ≥10cm in diameter (D) for calibration(Asner&Mascaro,2014;Asneretal.,2012).Collectingsuchdatain the field is costly and time consuming,which therefore limitsthespatialrepresentativenessofavailableplotnetworks.Besides,extrapolationmethods of ground‐based biomass estimations onRSdatastill face important limits.For instance,usingmeancan‐opy height extracted from active sensors (HoTongMinh et al..,2016;Mascaro,Detto,Asner,&Muller‐Landau,2011)orcanopygrain derived fromoptical images (Bastin et al., 2014; Ploton et

al., 2017; Ploton, Pélissier, & Proisy, 2012; Proisy, Couteron, &Fromard,2007),thebiomassispredictedwithanerrorofonly10–20%comparedwithground‐basedestimates.However,thisgoodlevel of accuracy is limited to the extent of the RS scene used,whichdecreasesconsiderablyintheupscalingstepnecessaryfornationalorglobalmaps(Xuetal.,2017).Apromisingdevelopmenttoalleviatethisspatialrestrictionliesinthe‘universalapproach’,proposed by Asner et al. (2012) and further adapted by AsnerandMascaro (2014), inwhich plot‐level biomass is predicted bya linear combinationofground‐basedand remotely sensedmet‐rics. The ‘universal approach’ relies upon canopy heightmetricsderivedfromradarorLiDAR(topofcanopyheight,TCH),andbasalarea(BA;i.e.,thecumulatedcross‐sectionalareaofthestems)andcommunitywooddensity(i.e.,weightedbybasalarea,WDBA)de‐rivedfromfieldinventories.PlotAGBisthenpredictedasfollows(Asneretal.,2012):

Although generally performing better than approaches basedsolelyonRSoftreeheight(Coomesetal.,2017),thismodelreliesonexhaustivegroundmeasurements(i.e.,wooddensityandthebasalareaofalltrees>10cmindiameterat130cm,neitherofwhichismeasuredusinganyexistingremotelysenseddata).

Recent advances in RS allow the identification of single treesin the canopy (Ferraz, Saatchi,Mallet,&Meyer, 2016), estimationofadultmortalityratesforcanopytreespecies(Kellner&Hubbell,2017), description of the forest diameter structure (Stark et al.,2015),depictionofcrownandgapshapes(Coomesetal.,2017)andeventheidentificationofsomefunctionaltraitsofcanopyspecies(Asneretal.,2017).Giventhatroutineretrievalofsomecanopytreemetricsiswithinreach,inthepresentstudywetestthecapacityofthelargesttrees(i.e.,treesthatcanpotentiallybederivedusingRS),topredictplot‐levelbiomass.Tothisend,weadaptedEquation(1)asfollows:

wherefortheithlargesttrees,DgLTisthequadraticmeandiameter,HLTthemeanheight,andWDLTthemeanwooddensityamongtheithlargesttrees.

Using a large database of forest inventories gathered acrossthetropics(Figure1), includingsecondaryandold‐growthforestplots, we test the ability of the largest trees to predict variousmetrics estimatedat1‐haplot level, namely themeanquadraticdiameter, the basal area, the Lorey’s height (i.e., plot‐averageheightweightedbybasalarea),thecommunitywooddensity(i.e.,plot‐average wood density weighted by basal area) and meanaboveground live biomass (Supporting Information Figure 1). Bytesting different numbers of largest trees as predictors,we aimto propose a threshold for theminimal number of largest treesrequiredtopredictforestplotmetricsatapan‐tropicallevelwithnobiasandlowuncertainty(i.e.,error<20%).Althoughpreviouswork focused on estimating biomass in Central African forests(Bastin et al., 2015), the present study aims at generalizing the

(1)AGB= aTCHb1BA

b2WD

b3

BA

(2)AGB= a(DgLTiHLTiWDLTi)b1

6  |     BASTIN et al.

potentialof largetrees topredict thesedifferentplotmetricsatcontinentalandpan‐tropicalscales.Takingadvantageofauniquedatasetgatheredacrossthetropics(8671‐haplots),wealsoinves‐tigatemajordifferencesinforeststructureacrossthethreemaintropicalregions:theAmericas,AfricaandAsia.WefurtherdiscusshowthisapproachcanbeusedtoguideinnovativeRStechniquesandincreasethefrequencyandrepresentativenessofgrounddatatosupportglobalcalibrationandvalidationofcurrentandplannedspace missions. These include the NASA Global EcosystemDynamics Investigation (GEDI), NASA‐ISRO Synthetic ApertureRadar(NISAR)andESAP‐bandradar(BIOMASS)(Dubayahetal.,2014;LeToanetal.,2011).Thisstudyisastepforwardsinbring‐ingtogetherRSandfieldsamplingtechniquesforquantificationofterrestrialcarbonstocksintropicalforests.

2  | MATERIAL S AND METHODS

2.1 | Database

Forthisstudy,wecompiledstandardforestinventoriesconductedin8671‐haplotsfrom118sitesacrossthethreetropicalregions(Figure1),includingmatureandsecondaryforests.Eachsitecom‐prisesall theplots inagivengeographical location (i.e.,withina10‐km radius and collected by a principal investigator and theirteam). These consisted of 389 plots in America (69 sites), 302plotsinAfrica(35sites)and176plotsinAsia(14sites).Datawereprovided by principal investigators (see Supporting InformationTable 1) and through datasets available on the following net‐works:TEAM,CTFS(www.forestgeo.si.edu/;Conditetal.,2012)and ForestPlots (https://www.forestplots.net/) for AfriTRON(the African Tropical Rainforest Observation Network; www.af‐ritron.org)andRAINFOR(theAmazonforest inventorynetwork;networks.

We selected plots located between 23°N and 23°S, in‐cluding tropical islands,with an area of 1ha to ensure stableintra‐samplevarianceinbasalarea(Clark&Clark,2000).Plotsinwhich≥90%of thestemswere identified tospeciesand inwhich all stems with the diameter at 130cm of ≥10cm had

beenmeasuredwereincluded.Wooddensity,hererecordedasthewooddrymassdividedbyitsgreenvolume,wasassignedtoeachtreeusingthe lowestavailabletaxonomic levelofbotan‐ical identifications (i.e., speciesorgenus)and thecorrespond‐ingaveragewooddensityrecordedintheGlobalWoodDensityDatabase (GWDD; Chave et al., 2009; Zanne et al., 2009).BotanicalidentificationwasharmonizedthroughtheTaxonomicNamesResolutionService(https://tnrs.iplantcollaborative.org),forbothplot inventoriesandtheGWDD.Fortreesnot identi‐fied to species or genus (ca.5%), we used plot‐average wooddensity.Weestimatedheights of all trees usingChave et al.’s(2014)pan‐tropicaldiameter–heightmodel,whichaccountsforheterogeneity in theD–H relationship using an environmentalproxy:

whereDisthediameterat130cmandEisameasureofenvironmen‐talstress(Chaveetal.,2014).Forsiteswithtreeheightmeasurements(n=20),wedeveloped localD–Hmodels, using aMichaelis–Mentenfunction(Moltoetal.,2014).WeusedtheselocalmodelstovalidatethepredictedLorey’sheight(i.e.,plot‐averageheightweightedbyBA)fromthe largest trees,ofwhichheighthasbeenestimatedwithagenericH–Dmodel[Equation(3),Chaveetal.,2014].

Weestimatedplotbiomassasthesumofthebiomassoflivetreewithdiameterat130cmof≥10cm,usingthefollowingpan‐trop‐ical allometric model (Réjou‐Méchain, Tanguy, Piponiot, Chave, &Hérault,2017):

2.2 | Plot‐level metric estimation from the largest trees

Therelationshipbetweeneachplotmetric,namelybasalarea(BA),the quadratic mean diameter (Dg), Lorey’s height (HBA; the meanheightweightedbythebasalarea)andthecommunitywoodden‐sity(WDBA;themeanwooddensityweightedbythebasalarea),and

(3)Ln(H)=0.893−E+0.760ln(

D)

−0.0340ln(

D)2

(4)AGB=exp

{

−2.024−0.896E+0.920ln(

WD)

+2.795ln(

D)

−0.0461[

ln(

D2)]

}

F I G U R E 1  Geographicaldistributionoftheplotdatabase.Weused867plotsof1hafrom118sites.Dotsarecolouredaccordingtofloristicaffinities(Sliketal.,2015),withAmerica,AfricaandAsiainorange,greenandblue,respectively.Theyarealsosizedaccordingthetotalareasurveyedineachsite.Inthebackground,moistforestsaredisplayedindarkgreenanddryforestinlightgreen

     |  7BASTIN et al.

thosederivedfromlargesttreeswasdeterminedusinganiterativeprocedurefollowingBastinetal.(2015).Treeswerefirstrankedbydecreasingdiameterineachplot.Anincrementalprocedure(i.e.,in‐cludinganewtreeateachstep)wasusedtosumoraverageinforma‐tionoftheilargesttreesforeachplotmetric.Eachplot‐levelmetricwaspredictedbytherespectivemetricderivedfromtheithlargesttrees.Foreachincrement,theability(goodnessoffit)ofthei larg‐esttreestopredictagivenplotmetricwastestedthroughalinearregression.Toavoidoverfitting,aleave‐one‐outprocedurewasusedtodevelop independentsite‐specificmodels (n=118).Specifically,themodel tobe testedatasitewasdevelopedwithdata fromallother sites. Errorswere then estimated as the relative rootmeansquare error (rRMSE) computed between observed and predictedvalues(X):

Theformoftheregressionmodel(ie,linear,exponential)wasse‐lectedtoensureanormaldistributionoftheresiduals.

To estimate plot basal area,we used a simple power‐law con‐strainedontheorigin,aslinearmodelresultedinnon‐normalresid‐uals.Plot‐levelbasalarea(BA)wasrelatedtothebasalareaforthei largesttrees(BAi)using:

Toestimatethequadraticmeandiameter,Lorey’sheightandthewooddensityof thecommunity,weusedsimple linearmodelsre‐lating theplot‐levelmetrics and thevalueof themetrics for the i largesttrees:

Both Lorey’s height (HBA) and the average height (Hi ) of theith largest trees depend on the sameD–H allometry, which al‐wayscontainsuncertaintywhetherweusea local,acontinentalorapan‐tropicalmodel.Totestthedependenceoftheprediction of HBA on the allometric model, we used measurement fromMalebointheDemocraticRepublicoftheCongo,whereallheights were measured on the ground (see Supporting InformationFigure2).

The quality of the predictions of plot‐level metrics from thelargesttreesisquantifiedusingtherRMSEbetweenmeasuredandpredictedvaluesanddisplayedalongthecumulatednumberoflarg‐esttrees.Modelcoefficientsareestimatedforeachmetricderivedfromthelargesttrees(NLT)andaveragedacrossthe118models(seeSupportingInformationTable2).

MeanrRMSEisplottedasacontinuousvariable,anditsvariationispresentedasacontinuousareabetweenthe5thandthe95thper‐centilesofobservedrRMSE.

2.3 | The optimal number of largest trees for plot‐level biomass estimation

Theoptimalnumberoflargesttrees,NLT,wasdeterminedfromthepre‐dictionofeachplot‐levelmetricconsideredabove(i.e.,keepingasmallnumberoftreeswhileensuringalowleveloferrorforeachstructuralparameter).Wethenpredictedplot‐levelbiomassfromtheNLT model [Equation(2)].ThefinalerrorwascalculatedbypropagatingtheentiresetoferrorsrelatedtoEquation(4)(Réjou‐Méchainetal.,2017)intheNLTmodel(i.e.,errorassociatedwitheachallometricmodelused).Themodelwasthencross‐validatedacrossallplots(n=867).

2.4 | Investigating residuals: What the largest trees do not explain

TounderstandthelimitsofpredictingAGBthroughNLT,wealsoin‐vestigatedtherelationshipbetweenAGBresidualsandkeystructuralandenvironmentalvariablesusing linearmodelling.Foreststructurewas investigated through the total stem density (N), the quadraticmeandiameter(Dg),Lorey’sheight(HBA)andcommunitywoodden‐sity(WBBA).Asenvironmentaldata,weusedthemeanannualrainfallandthemeantemperaturecomputedoverthe last10yearsateachsite using the Climate Research Unit data (New et al., 1999; New,Lister,Hulme,&Makin,2002),alongwithrough informationonsoiltypes(Carré,Hiederer,Blujdea,&Koeble,2010).Majorsoiltypeswerecomputed fromthesoilclassificationof theHarmonizedWorldSoilDatabaseintoIPCC(IntergovernmentalPanelonClimateChange)soilclasses.Inaddition,consideringobserveddifferencesinforeststruc‐ture across tropical continents (Feldpausch et al., 2011, 2012) andrecentresultsonpan‐tropicalfloristicaffinities(Sliketal.,2015),wetestedforaneffectofcontinent(America,AfricaandAsia)ontheAGBresiduals.DifferencesinforeststructureandAGBamongcontinentswerealsoillustratedthroughtheanalysisoftheirdistribution.

TheimportanceofeachvariablewasevaluatedbycalculatingthetypeIIsumofsquares,whichmeasuresthedecreaseinresidualsumofsquaresowingtoanaddedvariableoncealltheothervariableshavebeenintroducedintothemodel(Langsrud,2003).Residualswerein‐vestigatedatbothplotandsitelevels,thelatteranalysedtotestforany influenceofthediameterstructure,which isusuallyunstableattheplotlevelowingtothedominanceoflargetreesonforestmetricsatsmallscales(Clark&Clark,2000).Here,weuseaprincipalcompo‐nentsanalysis(PCA)tosummarizetheinformationheldinthediameterstructurebyordinatingthesitesalongtheabundanceoftreesineachdiameterclass(from10to+100cmin10cmbins).

3 | RESULTS

3.1 | Plot‐level metrics

Plotmetricsaveragedat thesite level (867plots,118sites)pre‐sent importantvariationswithinandbetweencontinents. Inour

(5)

−150 −100 −50 0 50 100 150

−150

−100

−50

0

50

100

150

−150 −100 −50 0 50 100 150

−150

−100

−50

0

50

100

150

−150 −100 −50 0 50 100 150

−150

−100

−50

0

50

100

150

-0.6

-0.3

0.0

0.3

0 10 20 30 40

Observed residuals (Mgha-1)

Pred

icte

d re

sidu

als

(Mgh

a-1)

Pred

icte

d re

sidu

als

(Mgh

a-1)

Pred

icte

d re

sidu

als

(Mgh

a-1)

Observed residuals (Mgha-1)

Observed residuals (Mgha-1)

Obs

erve

d re

sidu

als

(Mgh

a-1)

stem density DBH 50-70

(a) (b)

(c) (d)

−150 −100 −50 0 50 100 150

−150

−100

−50

0

50

100

150

−150 −100 −50 0 50 100 150

−150

−100

−50

0

50

100

150

−150 −100 −50 0 50 100 150

−150

−100

−50

0

50

100

150

-0.6

-0.3

0.0

0.3

0 10 20 30 40

Observed residuals (Mgha-1)

Pred

icte

d re

sidu

als

(Mgh

a-1)

Pred

icte

d re

sidu

als

(Mgh

a-1)

Pred

icte

d re

sidu

als

(Mgh

a-1)

Observed residuals (Mgha-1)

Observed residuals (Mgha-1)

Obs

erve

d re

sidu

als

(Mgh

a-1)

stem density DBH 50-70

(a) (b)

(c) (d)

(6)BA= b1ΣBA�1

i

(7)Dg= a2+ b2Dgi

(8)HBA=a3+b3 Hi

(9)WDBA= a4+ b4 WDi

rRMSE=

(

Xobs−Xpred)2

n∕−X

8  |     BASTIN et al.

database,thequadraticmeandiametervariesfrom15to42cm2/ha,thebasalareafrom2to58m2/ha,Lorey’sheightfrom11to33m,andthewooddensityweightedbythebasalareafrom0.48to0.84g/cm3 (Supporting InformationFigure1).Such importantdifferences between minimal and maximal values are observedbecauseourdatabasecoverssiteswithvariousforesttypes,fromyoungforestcolonizingsavannastoold‐growthforest.However,

mostofoursitesarefoundinmatureforests,asshownbytherela‐tivelyhighaverageandmedianvalueofeachplotmetric(averageAGB=302Mg/ha;Supporting InformationFigure1). Ingeneral,highestvaluesofAGBarefound inAfrica,drivenbythehighestvaluesofbasalareaandhighestestimationsofLorey’sheight.ThehighestvaluesofwooddensityweightedbybasalareaarefoundinAmerica.

F I G U R E 2  Qualityofthepredictionofplotmetricsfromlargesttrees.Variationoftherelativerootmeansquareerror(rRMSE)ofthepredictionofplotmetricfromilargesttreesversusthecumulativenumberoflargesttreesfor:(a)basalarea,(b)quadraticmeandiameter,(c)Lorey’sheight,and(d)wooddensityweightedbythebasalarea.Resultsaredisplayedatthepan‐tropicallevel(mainplotingrey)andatthecontinentallevel(subplots;orange=America;green=Africa;blue=Asia).ThecontinuouslineandshadingshowsthemeanrRMSEandthe5thand95thpercentiles.DashedlinesrepresentthemeanrRMSEobservedforeachmodel,whenconsideringthe20largesttrees

0 20 40 60 80 100 0 20 40 60 80 100

0

0.2

0.4

0.6

0

0.2

0.4

0.6

0 20 40 60 80 100 0 20 40 60 80 100

0

0.2

0.4

0.6

0

0.2

0.4

0.6

# of cumulated largest trees # of cumulated largest trees

# of cumulated largest trees# of cumulated largest trees

relR

SE b

asal

are

a

relR

SE q

uadr

atic

mea

n di

amet

er

relR

SE L

orey

’she

ight

relR

SE c

omm

unity

woo

d de

nsity

00 50 1000 50 1000 50 100

0.50.5

00

0.5

0 50 1000 50 1000 50 100

0 50 1000 50 1000 50 100

0 50 1000 50 1000 50 100

0

0.50.5

00

0.5

0

0.50.5

00

0.5

0

0.50.5

00

0.5

(a) (b)

(d)(c)

     |  9BASTIN et al.

3.2 | Plot‐level estimation from the i largest trees

Overall,plotmetricsatthe1‐hascalewerewellpredictedbythelargest trees,withqualitative agreement amongglobal and con‐tinental models (Figure 2). When using the 20 largest trees topredict basal area (BA) and quadratic mean diameter (Dg), themeanrRMSEwas<16and12%,respectively(Figure3a,b).Lorey’sheight (HBA) and wood density weighted by basal area (WDBA)wereevenbetterpredicted(Figure3c,d),withmeanrRMSEof4%forthe20largesttrees.ThepredictionofLorey’sheightfromthelargest trees using the local diameter–heightmodel (SupportingInformation Figure 2a) yielded results similar to those obtainedusingEquation(3)ofChaveetal.(2014).Moreimportantly,italso

yieldedsimilarresultstothepredictionofLorey’sheightfromthelargesttreesusingplotswhereallthetreesweremeasuredontheground(SupportingInformationFigure2b).Thissuggeststhatourconclusionsarerobusttotheuncertaintyintroducedbyheight–di‐ameterallometricmodels.

3.3 | Aboveground biomass prediction from the largest trees

Weselected20asthenumberoflargesttreestopredictplotmet‐rics.TheresultingmodelpredictingAGB(inmegagramsperhectare)basedonthe20largesttreesis:

F I G U R E 3  Predictionofplotmetrics(yaxis)fromthe20largesttrees(xaxis).Resultsareshownfor:(a)basalarea,(b)quadraticmeandiameter,(c)Lorey’sheight,and(d)wooddensityweightedbythebasalarea.Eachdotcorrespondstoasingleplot,colouredinorange,greenandblueforAmerica,AfricaandAsia,respectively.Bothpan‐tropical(blackdashedlines)andcontinental(colouredlines)regressionmodelsaredisplayed.Theseresultsshowthatasubstantialpartoftheremainingvariance(i.e.,notexplainedbythelargesttrees)isfoundwhenpredictingthebasalareaandthequadraticmeandiameter,withslightbutsignificantdifferencesbetweencontinents

0 5 10 15 20 25 30 35

basal area top 20 (m2ha-1)

basa

l are

a (m

2 ha-1

)

0

10

20

30

40

50

60

0 50 100 150

quad

ratic

mea

n di

amet

er (m

2 ha-1

)

quadratic mean diameter top 20 (cm2ha-1)

0

10

20

30

40

50

0 10 20 30 40 50

mean Height top 20 (m)

Lore

y's h

eigh

t (m

)

0

10

20

30

40

50

0.2 0.4 0.6 0.8 1.0mean wood wensity top 20 (gcm−3)

woo

d de

nsity

wBA

(gc

m−3

)

0.2

0.4

0.6

0.8

1.0

(a) (b)

(c) (d)

10  |     BASTIN et al.

Given that theexponentwas close toone,wealsodevelopedanalternativeandmoreoperationalmodelwiththeexponentcon‐strainedtoone,givenby:

Groundmeasurements of plot AGBwere predicted by ourNLT modelwith the exponent constrained to one,with a total error of17.9%(Figure4),avaluewhichencompassestheerroroftheNLT model andtheerrorrelatedtotheallometricmodelchosen.Theleave‐one‐outcross‐validationprocedureyieldedsimilarresults(rRMSE=0.19;R2=0.81),validatingtheuseofthemodelonindependentsites.

3.4 | Determining the cause of residual variations

Theexplanatoryvariablesalltogetherexplainca.37%ofthevari‐anceinAGBatbothplotandsitelevelswhenomittingthediam‐eterstructure,andca.63%atsitelevelwhenincluded(Figure5).Ingeneral,foreststructureandparticularlythestemdensityex‐plainedmostoftheresiduals(Table1;weights:79and54%atplotandsitelevel,respectively).Thestemdensitywasfollowedbyacontinental effect (weights: 18, 28 and 1% for Africa, AmericaandAsia, respectively) andby theeffectofHBAandWDBA (re‐spectiveweights:1and0%attheplotlevel,0and11%atthesite

level,and23and0%whenaccountingforthediameterstructureatthesitelevel).Inclusionofthediameterstructureprovidedthebest explanation of residuals, with 63% of variance explained,and a weight of 69% for the first axis of the PCA (SupportingInformationFigure3). This first axis of thePCAwas related tothe general abundanceof trees at a site, and in particular,me‐dium‐sized trees (40–60cm). Among environmental variables,onlyrainfallwassignificantlyrelatedtotheresidualsatthesitelevelwhenthediameterstructurewasconsidered(2%).

3.5 | Differences among continents

Althoughdiameterstructureexplainedalargefractionofthere‐sidualvarianceofourglobalmodel,therewasamarkeddifferencein forest structure across continents (Figure 6). Consequently,we investigated differences between continents in the distri‐bution of residuals of the pan‐tropicalmodel (Figure 6a), in therelativecontributionofthe20largesttreestoplottotalbiomass(Figure6b)andinthecontributiontothetotalAGBperdiameterclass(Figure6c–f).Tothisend,weconsideredthefollowingfourclassesofdiameterat130cm:10–30,30–50,50–70and>70cm.Theresultsshowthatthepredictionofbiomassfromthe20larg‐esttreesusingthepan‐tropicalmodeltendstobeslightlyoveres‐timatedinAfrica(+3%)andunderestimatedinAmerica(−3%)andinAsia(−5%)(Figure6a).Theproportionofbiomassishigherinthehigh‐diameter class (>70cm) inAfrica, in intermediate‐diameterclasses(between30and70cm)inAmerica,andisequallydistrib‐utedamongthedifferentdiameterclassesinAsia(Figure6c,d).

(10)AGB=0.0735× (Dg20H20WD20)

1.1332(

rRMSE=0.179; R2=0.85;Akaike Criterion=−260.18)

(11)AGB=0.195× (Dg20H20WD20)

(

rRMSE=0.177; R2=0.85; AIC=−195)

F I G U R E 4  Qualityofthepredictionofabovegroundbiomass(AGB)fromlargesttreesplotmetrics.Variationoftherelativerootmeansquareerror(rRMSE)ofthepredictionofAGBfromilargesttreesversusthecumulativenumberoflargesttrees(a)anddetailedpredictionofAGBfromplotmetricsofthe20largesttrees(b).Resultsareshownforthe867plotsinthethreecontinents,colouredorange,greenandblueforAmerica,AfricaandAsia,respectively.Theregressionlineofthemodelisshownasacontinuousblackline,andthedashedblacklineshowsa1:1relationship.ThefigureshowsanunbiasedpredictionofAGBacrossthe867plots,withslightbutsignificantdifferencesbetweenthethreecontinents

0 200 400 600 800 1000

Predicted AGB from 20 largest trees (Mgha−1)

Estim

ated

AG

B fro

m th

e gr

ound

(Mgh

a−1)

0

200

400

600

800

1000

0.0

0.2

0.4

0.6

0.8

0 20 40 60 80 100

# of cumulated largest trees

rel R

SE a

bove

grou

nd b

iom

ass

0.0

0.5

0 50 100

0.0

0.5

0 50 1000.0

0.5

0 50 100

(a) (b)

     |  11BASTIN et al.

4  | DISCUSSION

4.1 | The largest trees, convergences and divergences between continents

Samplingafewlargesttreesperhectaregenerallyallowsanunbi‐asedpredictionoffourkeydescriptorsofforeststructuresacross

the tropics.There isgenerallyno improvement inpredictingbio‐mass,quadraticmeandiameter,Lorey’sheight(HBA)orcommunitywooddensitybeyondthefirst10–20largesttrees(Figures2,3and3a).Butwhenaforestplotpresentsanabundantnumberoflargetrees(Figure5d),increasingthenumberoftreessampleddoesim‐provetheaccuracyofthemodel.Thisisattributabletothefactthat

F I G U R E 5  Predictedversusobservedresidualsofabovegroundbiomasspredictedfromthe20largesttrees.Residualsareexploredatthreedifferentlevels:(a)plot,(b)site(withoutconsideringthediameterstructureasanexplanatoryvariable),(c)site(consideringthediameterstructure),and(d)alongthestemdensityofmedium‐sizedtrees.America,AfricaandAsiaarecolouredinorange,greenandblue,respectively.Thepanelsshowagoodpredictionofresidualsin(a)and(b),drivenbystemdensity,andalessbiasedpredictionin(c),drivenbythediameterstructure.Variancesofobservedresidualsarealsowellexplainedbythestemdensityofmedium‐sizedtrees(d),whichmainlydrivethefirstaxisoftheprincipalcomponentsanalysis

−150 −100 −50 0 50 100 150

−150

−100

−50

0

50

100

150

−150 −100 −50 0 50 100 150

−150

−100

−50

0

50

100

150

−150 −100 −50 0 50 100 150

−150

−100

−50

0

50

100

150

-0.6

-0.3

0.0

0.3

0 10 20 30 40

Observed residuals (Mgha-1)

Pred

icte

d re

sidu

als

(Mgh

a-1)

Pred

icte

d re

sidu

als

(Mgh

a-1)

Pred

icte

d re

sidu

als

(Mgh

a-1)

Observed residuals (Mgha-1)

Observed residuals (Mgha-1)

Obs

erve

d re

sidu

als

(Mgh

a-1)

stem density DBH 50-70

(a) (b)

(c) (d)

−150 −100 −50 0 50 100 150

−150

−100

−50

0

50

100

150

−150 −100 −50 0 50 100 150

−150

−100

−50

0

50

100

150

−150 −100 −50 0 50 100 150

−150

−100

−50

0

50

100

150

-0.6

-0.3

0.0

0.3

0 10 20 30 40

Observed residuals (Mgha-1)

Pred

icte

d re

sidu

als

(Mgh

a-1)

Pred

icte

d re

sidu

als

(Mgh

a-1)

Pred

icte

d re

sidu

als

(Mgh

a-1)

Observed residuals (Mgha-1)

Observed residuals (Mgha-1)

Obs

erve

d re

sidu

als

(Mgh

a-1)

stem density DBH 50-70

(a) (b)

(c) (d)

12  |     BASTIN et al.

thehigherthetotalAGBinaplot,thelowertheproportionoftotalAGB encompassed by the largest trees. This is particularly trueforBA,forwhichrRMSEcontinuestodecreaseupto100largesttrees(Figure2a).Incontrast,Lorey’sheightpredictionsarealteredwhen a large number of trees are included (Figure 2c; i.e.,whensmaller, often suppressed, trees draw the average down; Farrior,Bohlman,Hubbell,&Pacala,2016).Thismightexplainwhythepre‐dictionofAGBdoesnotmirrorthatofbasalarea(Figures2,3band3a) and suggests that the number of largest trees should be setindependently from each predictor considered. Interestingly, theevolutionof relativeerror inAGBpredictionasa functionof thenumberoflargesttreesconsidereddoesnotfollowthesamepathbetweencontinents.Forinstance,theerrorofpredictionsaturatesmorequicklyinAfricaandAsiathaninAmerica.Investigationofre‐sidualsshowedthatthediameterstructure(Figure5c;SupportingInformationFigure3b),and inparticular, thenumberofmedium‐sized trees (Figure5d), drives variability inAGBpredictions. It istherefore not surprising to see that in our dataset the site with

higherlevelsofunderestimationsistheonewiththehighestnum‐berofmedium‐sizedtrees,which isfoundinAsia intheWesternGhatsofIndia.

Thegoodperformanceofmodelsbasedonthe20largesttreesinpredictingLorey’sheightandcommunitywooddensityatthesitelevelwasnotsurprising.Bothmetricswereweightedbybasalarea,driven de factobythelargesttrees.Nonetheless,theirconsistencyacrosssitesandcontinentswasnotexpected,whichemphasizethegeneralityofourapproach.

Thepredictabilityofplot‐levelforeststructuremetricsfromthelargesttreesimpliesthatcharacteristicsofsmallertreesdonotvaryinacompletelyindependentmannerfromthoseofthelargertrees.Forexample,plotswherethelargesttreeshavealowbasalareatendtohavelowplot‐levelbasalarea(Figure3a),meaningthatthetotalsizeofthesmallertreesissufficientlyconstrainedthatitdoesnotcompensateforthesmallsizeofthelargesttrees.Suchconstraintscould arise through size–frequency distributions being set by al‐lometric scaling rules (Enquist,West,&Brown,2009)or couldbe

TA B L E 1  Weightofeachvariableretainedfortheexplanationofabovegroundbiomassresiduals

Level of residual Parameter Weight

Plot

Stemdensity 79

Continent 18

Lorey’sheight 1

Majorsoiltypes 1

Temperature 1

Wooddensityweightedbythebasalarea 0

Rainfall 0

Sitewithoutdiametricstructure

Stemdensity 54

Continent 28

Wooddensityweightedbythebasalarea 11

Rainfall 3

Majorsoiltypes 3

Temperature 2

Lorey’sheight 0

Sitewithdiametricstructure

PCAaxis1 69

Lorey’sheight 23

Rainfall 3

Majorsoiltypes 3

Continent 1

Temperature 1

Wooddensityweightedbythebasalarea 0

PCAaxis2 0

Note.Weightsarecalculatedasatypellsumofsquares,whichmeasuresthedecreasedresidualsumofsquaresattributabletoanaddedvariableoncealltheothervariableshavebeenintroducedintothemodel.Resultsareshownfortheexplorationofresidualsattheplotandatthesitelevel,withandwithoutconsiderationofthediameterstructure.Weightsaredominatedbystructuralvariables,andinparticular,thestemdensityandthediameterstructure.Height,wooddensityandcontinenthavealsoanon‐negligibleinfluenceonresiduals.PCA=principalcomponentsanalysis.

     |  13BASTIN et al.

attributabletothelargesttreesrespondinginthesamewayastheremainingsmallertreestoenvironmentaldrivers.

Despite the general consistency of these relationships acrosscontinents,slightdifferencesareevidentwhencomparingthepan‐tropical model residuals across continents (Figure 6; Supporting

Information Figure 4). These differences indicate biogeographicalvariation in forest structure. In America, our pan‐tropical modeltendsslightlytounderestimatebasalarea(mean:−5%)andoveresti‐mateLorey’sheight(mean:+3%)(SupportingInformationFigure4).Thissuggeststhatlargetreesmakeupasmallerproportionofbasal

F I G U R E 6  Comparisonacrosscontinentsofabovegroundbiomass(AGB)predictionpersiteandtheircontributiontodifferentsharesofthediameterstructure.Africa,AsiaandAmericaarecolouredingreen,blueandorange,respectively.(a)Thedistributionoftheresidualsofpan‐tropicalAGBpredictionfromthe20largesttreesshowsthatpredictionsareslightlyoverestimatedinAfrica(+3%),andslightlyunderestimatedinAsia(−3%)andAmerica(−5%).(b)TheproportionofAGBinthe20largesttreesishighestinAfrica(48%),followedbyAsia(40%)andAmerica(35%).(c–f)Thedecompositionacrossfourdiameterclasses[i.e.,10–30,30–50,50–70and>70cmdiameteratbreastheight(DBH)]oftheirrelativeshareofthetotalbiomassshowsthatmostofthebiomassisfoundinthelargetreesinAfricaandinthesmalltomediumtreesinAmerica.Asiapresentsamorebalanceddistributionofbiomassacrossthediameterstructure

−0.2

−0.1

0.0

0.1

0.2

America Africa Asia

resi

dual

s

0.2

0.4

0.6

America Africa Asia

Prop

ortio

n of

AG

B in

top2

0

��

0.25

0.50

0.75

1.00

America Africa Asia

Prop

ortio

n of

AG

B D

BH 1

0−30

��

0.2

0.4

0.6

America Africa Asia

Prop

ortio

n of

AG

B D

BH 3

0−50

0.1

0.2

0.3

America Africa Asia

Prop

ortio

n of

AG

B D

BH 5

0−70

��

0.1

0.2

0.3

0.4

0.5

America Africa Asia

Prop

ortio

n of

AG

B D

BH o

ver 7

0

(a) (b)

(c) (d)

(e) (f)

14  |     BASTIN et al.

areainAmericaandthatforagivendiameterwefindhighertrees(SupportingInformationFigure2),whichconfirmsthattheshapeofheight–diameterallometriesvariesbetweencontinents(Baninetal.,2012; Sullivan et al., 2018). InAfrica, large trees (i.e., diameter atbreastheight>70cm) aremoreabundant andaccount for a largefractionofplotbiomass (Figure6f).This supportspreviousobser‐vations thatAfrican forests are characterized by fewer but largerstems(Feldpauschetal.,2012;Lewisetal.,2013),whereasforestsintheAmericashavemorestemsbutgenerallyhavelowerbiomass(Sullivanetal.,2017).InAsia,thedistributionofthebiomassacrossdiameterclassesappearsmorebalanced(Figure6c–f).Suchdiffer‐encesinforeststructure,albeitlimited,suggestthattropicalforestsdifferbetweencontinentsintermsofdynamics,carboncycling,re‐sponseandfeedbacktoclimateandresiliencetoexternal forcings(e.g.,climatechange,forestdegradationanddeforestation).

Interestingly,althougharecentglobalphylogeneticclassificationoftropicalforestsgroupsAmericanwithAfricanforestsversusAsianforests (Slik et al., 2018), our study of forest structure propertiestendsmoretosingleoutAmericanforests,andinparticular,tohigh‐light thecontrastbetweenAfricanandAmerican forests.Althoughthisdeservesfurtherinvestigation,itmightrevealthelackofacloserelationshipbetweenforeststructurepropertiesandphylogenicsim‐ilarity,whichechoesrecentresultsontheabsenceofarelationshipbetweentropicalforestdiversityandbiomass(Sullivanetal.,2017).

4.2 | Largest trees: A gateway to global monitoring of tropical forests

Revealingthepredictivecapacityheldbythelargesttrees,ourresultsconstitute a major step forwards tomonitor forest structures andbiomassstocks.Thelargesttreesintropicalforestscanthereforebeusedtomakeaccuratepredictionsofvariousground‐measuredprop‐erties(i.e.,thequadraticmeandiameter,thebasalarea,Lorey’sheightandcommunitywooddensity),whereaspreviousworkhaspredictedonlybiomass‘estimates’(e.g.,Bastinetal.,2015;Sliketal.,2013).Theadvantagesofourapproachareasfollows:(a)itallowsustodescribeforeststructure independentlyofanybiomassallometricmodel; (b)itallowsustointegrateenvironmentallybasedvariationsintheD–H relationship,knowntovarylocally(Feldpauschetal.,2011;Kearsleyetal.,2013);and(c)itisalsorelativelyinsensitivetodifferencesinflo‐risticcompositionandcommunitywooddensity(Poorteretal.,2015).

Furthermore,the‘largesttrees’modelsweredevelopedforeachplot‐levelmetricandforanynumberoflargesttrees.Thus,theydonotrelyonanyarbitrarythresholdoftreediameter.Notethattheoptimalnumberoflargesttreestobemeasured(i.e.,20)wassetfordemonstrationandcanvarydependingontheneedsandcapacitiesofeachcountryorproject(seeSupportingInformationTable2).Inthe sameway, localmodels could integrate locallydevelopedbio‐massmodels,whenavailable.Consequentlyourapproach:(a)canbeusedinyoungorregeneratingunmanagedforestswithalow‘largesttree’ diameter threshold; and (b) is compatiblewith recentRS ap‐proachesabletosingleoutcanopytreesanddescribetheircrownandheightmetrics(Coomesetal.,2017;Ferrazetal.,2016).

4.3 | Aboveground biomass model from the largest trees: A multiple opportunity

Globally,theNLTmodelforthe20largesttreesallowsplotbiomasstobepredictedwith17.9%error.This result isapan‐tropicalvali‐dationof resultsobtained inCentralAfrica (Bastinetal.,2015). Itopensnewperspectivesforcost‐effectivemethodstomonitorfor‐eststructuresandcarbonstocksthroughlargesttreesmetrics(i.e.,metricsofobjectsdirectlyinterceptedbyRSproducts).

DevelopingcountrieswillingtoimplementReductionofEmissionsfromDeforestation and ForestDegradation (REDD+) activitieswillalso reporton theircarbonemissionsanddevelopanational refer‐encelevel(IPCC,2006;Maniatis&Mollicone,2010).However,mosttropical countries lack the capacity to assume multiple, exhaus‐tive and costly forest carbon inventories (Romijn,Herold,Kooistra,Murdiyarso,&Verchot,2012).Bymeasuringonlya few large treesperhectare,our results show that it ispossible toobtainunbiasedestimatesofabovegroundcarbonstocksinatime‐andcost‐efficientmanner.Assumingthat400–600treeswithD>10cmaremeasuredinatypical1‐hasampleplot,monitoringonly20treesisasignificantimprovement.Althoughfindingthe20largesttreesinaplotofseveralhundredindividualsrequiresevaluating>20trees,inpractice,acon‐servativediameterthresholdcouldbedefinedtoensurethatthe20largesttreesaresampled.Analternativecouldalsobefoundinthede‐velopmentofarelascope‐basedapproachadaptedtodetectionofthelargesttreesintropicalforests.Usingsuchapproachwouldfacilitaterapidfieldsamplinginextensiveareastoproducelarge‐scaleAGBes‐timates.ThosecouldfulfiltheneedsforcalibrationandvalidationofcurrentandforthcomingspacemissionsfocusedonAGB.

Our findings also point towards the potential effectiveness ofusingRS techniques tocharacterizecanopy trees for inferring theattributesofentireforeststands.Remotesensingdatacouldbeusedfordirectmeasurement(e.g.,tree‐levelmetrics,suchasheight,crownwidthandcrownheight)ofthelargesttreesasapotentialalterna‐tivetoindirectdevelopmentofcomplexmetrics(e.g.,meancanopyheight,texture)usedtoextrapolateforestproperties.Althoughtheuseofasingle‐treeapproachhasshownsomelimitationstoextrap‐olateplotmetrics(Coomes,Šafka,Shepherd,Dalponte,&Holdaway,2018), we have yet to investigate its potential to identify largesttrees.Somefurtherrefinementsareneeded,butmostofthetoolsrequired to develop ‘largest trees’models are readily available. Inparticular,Ferrazetal.(2016)developedanautomatedproceduretolocatesingletreesbasedonairborneLiDARdata,tomeasuretheirheight and crown area. Crown area could then be linked to basalarea,becausethelogarithmofcrownareaisconsistentlycorrelated,withaslopeof1.2–1.3,tothelogarithmoftreediameteracrossthetropics(Blanchardetal.,2016).Regardingwooddensity,hyperspec‐tralsignatureandhigh‐resolutiontopographyoffersapromisingwaytoassessfunctionaltraitsremotely(e.g.,Asneretal.,2017;Juckeretal.,2018),whichcouldpotentiallyprovideproxiesofwooddensity.Alternativeapproachescouldfocusonthedevelopmentofplot‐levelAGBpredictionbyreplacingthebasalareaofthelargesttreeswiththeircrownmetrics.Althoughthemeasurementofcrownareashas

     |  15BASTIN et al.

yettobegeneralizedwheninventoryingplots,severalbiomassallo‐metricmodelsalreadypartitiontrunkandcrownmass (Coomesetal.,2017;Juckeretal.,2017;Plotonetal.,2016).

Themainlimitationofourapproachliesinthelimitedinferencethat can be made on the understorey and sub‐canopy trees.Weshowthatmostoftheremainingvarianceisexplainedbyvariationsindiameter structures,and inparticular, in the total stemdensity.Interestingly,stemdensitywasgenerallyidentifiedasapoorpredic‐torofplotbiomassintropicalforests(Lewisetal.,2013;Sliketal.,2010).However,ourresultsshowthatstemdensityexplainsmostoftheremainingvariance(SupportingInformationTableS1).Thissug‐geststhat,inadditiontotryingtounderstandlarge‐scalevariationsinlargetreesandotherplotmetricsthatcanbequantifieddirectlyfromRS,weshouldalsoputmoreeffortintounderstandingvariationinsmallertrees,whichmainlydrivestotalstemdensityandthetotalfloristicdiversity.Smallertreesarealsoessentialtocharacterizefor‐estdynamicsandunderstandchangesincarbonstocks.Severalop‐tionsarenonethelesspossiblefromRS,consideringthevariationinLiDARpointdensitybelowthecanopylayer(D’Oliveira,Reutebuch,McGaughey,&Andersen,2012),thedistributionofleafareadensity(Starketal.,2015,2012;Tang&Dubayah,2017;Vincentetal.,2017)ortheuseofmultitemporalLiDARdatatoobtaininformationonfor‐estgapgenerationdynamicsand,consequently,onforestdiameterstructure(Farrioretal.,2016;Kellner,Clark,&Hubbell,2009).

4.4 | Large trees in degraded forests

Iflargetreesareakeyfeatureofunmanagedforests,theyarecon‐spicuouslyabsentfrommanagedordegradedforests.Indeed,largetreesaretargetedbyselectiveorillegalloggingandarethefirsttodisappearortosufferfromincidentaldamagewhentropicalforestsareexploitedfor timber (Sist,Mazzei,Blanc,&Rutishauser,2014).The loss of the largest trees drastically changes forest structuresanddiameterdistributions,andtheirlossislikelytocounteracttheconsistency in forest structures observed throughout this study.Understandinghow,orwhether,managedforestsdeviatefromourmodel predictions could help to characterize forest degradation,whichaccountsforalargefractionofcarbonlossworldwide(Baccinietal.,2017),acknowledgingthatrapidpost‐disturbancebiomassre‐covery(Rutishauseretal.,2015)willremainhardtocapture.

4.5 | Conclusion: Towards improved estimates of tropical forest biomass

Theacquisition,accessibilityandprocessingcapabilitiesofRSdatawith a very high spatial, spectral and temporal resolution has in‐creasedexponentiallyinrecentyears(Bastinetal.,2017).However,todevelopaccurateglobalmapswewillhavetoobtainalargernum‐beroffieldplotsanddevelopnewwaystouseRSdata.Ourresultsprovidea step forwards forbothby: (a) drasticallydecreasing thenumberof individual treemeasurementsrequiredtoobtainanac‐curate, yet less precise, estimateof plot biomass; and (b) opening

thewaytodirectmeasurementofplotmetricsmeasuredfromRStoestimateplotbiomass.

AshighlightedbyClarkandKellner(2012),newbiomassallo‐metricmodelsrelatingplot‐levelbiomassmeasuredfromdestruc‐tive samplingandplot‐levelmetricsmeasured fromRSproductsshouldbedeveloped,asanalternativetocurrenttree‐levelallo‐metricmodels.Suchaneffortwillgreatlyreduceoperationalcostsand uncertainties surrounding terrestrial carbon estimates and,consequently,willhelpdevelopingcountries in thedevelopmentof national forest inventories and aid the scientific communityin better understanding the effect of climate change on forestecosystems.

ACKNOWLEDG MENTS

J.‐F.B.wassupportedfordatacollectionbytheFRIA‐FNRS(FondNationalpourlaRechercheScientifique),ERAIFT(EcoleRégionalePost‐Universitaire d’Aménagement et de Gestion Intégrés desForêtsTropicales),WorldWideFundforNature(WWF)(...)WWFand by the CoForTips project (ANR‐12‐EBID‐0002); T.d.H. wassupportedbytheCOBIMFOproject(CongoBasinintegratedmon‐itoring for forest carbonmitigation and biodiversity) funded bytheBelgianSciencePolicyOffice(Belspo);C.H.G.wassupportedby the ‘Sud Expert Plantes’ project of French Foreign Affairs,CIRADandSCAC.SomeofthedatainthispaperwereprovidedbytheRAINFORNetwork,theAfriTRONnetwork,TEAMNetwork,the partnership between Conservation International, TheMissouriBotanicalGarden,TheSmithsonian InstitutionandTheWildlifeConservationSocietyandtheGordonandBettyMooreFoundation.WeacknowledgedatacontributionsfromtheTEAMnetwork not listed as co‐authors (upon a voluntary basis). WethankJean‐PhillipePuyravaud,EstaçãoCientíficaFerreiraPenna(MPEG)andtheAndrewMellonFoundationandNationalScienceFoundation (DEB 0742830). The forest plots inNova Xavantinaand Southern Amazonia, Brazil were funded by grants fromProjectPELD‐CNPq/FAPEMAT(403725/2012‐7;441244/2016‐5;164131/2013); CNPq‐PPBio (457602/2012‐0); productivitygrants(CNPq/PQ‐2)toB.H.Marimon‐JuniorandB.S.Marimon;ProjectUSA‐NAS/PEER (#PGA‐2000005316)andProjectReFlorFAPEMAT0589267/2016.Finally,wethankHelenMuller‐Landauforhercarefulrevisionandcommentsonthemanuscript.

CONFLIC T OF INTERE S T

Theauthorsdeclarethereisnoconflictofinterestassociatedwiththisstudy.

AUTHOR CONTRIBUTIONS

J.F.Bastin and E.Rutishauser conceived the study, gathered thedata,performedtheanalysisandwrotethemanuscript.Alltheco‐authors contributed by sharing data and reviewing themain text.A.R.Marshall,J.PoulsenandJ.KellnerrevisedtheEnglish.

16  |     BASTIN et al.

DATA ACCE SSIBILIT Y

DataforplotsintheCTFSnetworkareavailablethroughtheonlineportalathttps://www.forestgeo.si.edu;intheForestplotnetworkathttps://www.forestplots.net/;andintheTEAMnetworkathttps://www.teamnetwork.org/.

ORCID

Koen Hufkens https://orcid.org/0000‐0002‐5070‐8109

Jean‐François Bastin https://orcid.org/0000‐0003‐2602‐7247

Ervan Rutishauser https://orcid.org/0000‐0003‐1182‐4032

R E FE R E N C E S

Asner, G. P., Martin, R. E., Knapp, D. E., Tupayachi, R., Anderson, C.B., Sinca, F., … Llactayo,W. (2017). Airborne laser‐guided imagingspectroscopy tomap forest traitdiversityandguideconservation.Science,355,385–389.

Asner, G. P., & Mascaro, J. (2014). Mapping tropical forest carbon:CalibratingplotestimatestoasimpleLiDARmetric.Remote Sensing of Environment,140,614–624.

Asner,G.P.,Mascaro,J.,Muller‐Landau,H.H.C.,Vieilledent,G.,Vaudry,R.,Rasamoelina,M.,…vanBreugel,M.(2012).AuniversalairborneLiDARapproachfortropicalforestcarbonmapping.Oecologia,168,1147–1160.

Baccini, A.,Walker,W., Carvalho, L., Farina,M., Sulla‐Menashe, D., &Houghton, R. A. (2017). Tropical forests are a net carbon sourcebasedonabovegroundmeasurementsofgainandloss.Science,358,230–234.

Banin,L.,Feldpausch,T.R.,Phillips,O.L.,Baker,T.R.,Lloyd,J.,Affum‐Baffoe, K.,… Lewis, S. L. (2012).What controls tropical forest ar‐chitecture? Testing environmental, structural and floristic drivers.Global Ecology and Biogeography,21,1179–1190.

Bastin,J.‐F.,Barbier,N.,Couteron,P.,Adams,B.,Shapiro,A.,Bogaert,J.,…DeCannière,C.(2014).AbovegroundbiomassmappingofAfricanforestmosaicsusingcanopytextureanalysis:Towardaregionalap‐proach.Ecological Applications,24,1984–2001.

Bastin,J.F.,Barbier,N.,Réjou‐Méchain,M.,Fayolle,A.,Gourlet‐Fleury,S.,Maniatis,D.,…Bogaert,J.(2015).SeeingCentralAfricanforeststhroughtheirlargesttrees.Scientific Reports,5,13156.

Bastin, J.‐F.,Berrahmouni,N.,Grainger,A.,Maniatis,D.,Mollicone,D.,Moore,R.,…Castro,R. (2017).Theextentof forest indrylandbi‐omes.Science,356,635–638.

Bennett,A.C.,McDowell,N.G.,Allen,C.D.,&Anderson‐Teixeira,K.J.(2015).Largertreessuffermostduringdroughtinforestsworldwide.Nature Plants,1,15139.

Blanchard,E.,Birnbaum,P.,Ibanez,T.,Boutreux,T.,Antin,C.,Ploton,P.,…Couteron,P.(2016).Contrastedallometriesbetweenstemdiame‐ter,crownarea,andtreeheightinfivetropicalbiogeographicareas.Trees,30,1953–1968.

Carré,F.,Hiederer,R.,Blujdea,V.,&Koeble,R.(2010).Background guide for the calculation of land carbon stocks in the biofuels sustainability scheme drawing on the 2006 IPCC Guidelines for National Greenhouse Gas Inventories (p. 128). Luxembourg: Publications Office of theEuropeanUnion

Chave,J.,Coomes,D.,Jansen,S.,Lewis,S.L.,Swenson,N.G.,&Zanne,A.E.(2009).Towardsaworldwidewoodeconomicsspectrum.Ecology Letters,12,351–66.

Chave,J.,Réjou‐Méchain,M.,Búrquez,A.,Chidumayo,E.,Colgan,M.S.,Delitti,W.B.C.,…Vieilledent,G.(2014).Improvedallometricmodelstoestimatetheabovegroundbiomassoftropicaltrees.Global Change Biology,20,3177–3190.

Chave,J.,Riera,B.,Dubois,M.‐A.,&Riéra,B.(2001).Estimationofbio‐massinaneotropicalforestofFrenchGuiana:Spatialandtemporalvariability.Journal of Tropical Ecology,17,79–96.

Clark,D.B.,&Clark,D.A.(1996).Abundance,growthandmortalityofverylargetreesinneotropicallowlandrainforest.Forest Ecology and Management,80,235–244.

Clark,D.B.,&Clark,D.A. (2000). Landscape‐scalevariation in foreststructure and biomass in a tropical rain forest. Forest Ecology and Management,137,185–198.

Clark,D.B.,&Kellner,J.R.(2012).Tropicalforestbiomassestimationandthefallacyofmisplacedconcreteness.Journal of Vegetation Science,23,1191–1196.

Coomes,D.A.,Dalponte,M.,Jucker,T.,Asner,G.P.,Banin,L.F.,Burslem,D.F.R.P.,…Qie,L.(2017).Area‐basedvstree‐centricapproachestomappingforestcarboninSoutheastAsianforestsfromairbornelaserscanningdata.Remote Sensing of Environment,194,77–88.

Coomes, D. A., Šafka, D., Shepherd, J., Dalponte, M., & Holdaway, R.(2018).AirbornelaserscanningofnaturalforestsinNewZealandre‐vealstheinfluencesofwindonforestcarbon.Forest Ecosystems,5,10.

D’Oliveira,M. V.N., Reutebuch, S. E.,McGaughey, R. J., &Andersen,H.‐E. (2012). Estimating forest biomass and identifying low‐inten‐sity logging areas using airborne scanning lidar in Antimary StateForest, Acre State, Western Brazilian Amazon. Remote Sensing of Environment,124,479–491.

Dubayah,R.,Goetz,S.J.,Blair,J.B.,Fatoyinbo,T.E.,Hansen,M.,Healey,S.P.,…Swatantran,A.(2014).Theglobalecosystemdynamicsinvestigation.American Geophysical Union, Fall Meeting 2014, Abstract no. U14A‐07.

Enquist,B.J.,West,G.B.,&Brown,J.H.(2009).Extensionsandevalua‐tionsofageneralquantitativetheoryofforeststructureanddynam‐ics.Proceedings of the National Academy of Sciences of the USA,106,7046–7051.

Farrior, C. E., Bohlman, S. A., Hubbell, S., & Pacala, S. W. (2016).Dominanceof thesuppressed:power‐lawsizestructure in tropicalforests.Science,351,155–157.

Feldpausch, T. R., Banin, L., Phillips, O. L., Baker, T. R., Lewis, S. L.,Quesada,C.A.,…Lloyd,J.(2011).Height‐diameterallometryoftrop‐icalforesttrees.Biogeosciences,8,1081–1106.

Feldpausch, T. R., Lloyd, J., Lewis, S. L., Brienen, R. J.W., Gloor, M.,MonteagudoMendoza,A.,…Phillips,O.L.(2012).Treeheightinte‐grated intopantropical forestbiomassestimates.Biogeosciences,9,3381–3403.

Ferraz,A.,Saatchi,S.,Mallet,C.,&Meyer,V.(2016).Lidardetectionofindividualtreesizeintropicalforests.Remote Sensing of Environment,183,318–333.

Harms,K.E.,Wright,S.J.,Calderón,O.,Hernández,A.,&Herre,E.A.(2000).Pervasivedensity‐dependentrecruitmentenhancesseedlingdiversityinatropicalforest.Nature,404,493–5.

HoTongMinh,D.,LeToan,T.,Rocca,F.,Tebaldini,S.,Villard,L.,Réjou‐Méchain,M.,…Chave,J. (2016).SARtomographyfor theretrievalofforestbiomassandheight:Cross‐validationattwotropicalforestsitesinFrenchGuiana.Remote Sensing of Environment,175,138–147.

IPCC.(2006).2006 IPCC Guidelines for National Greenhouse Gas Inventories, Prepared by the National Greenhouse Gas Inventories Programme(H.S.Eggleston,L.Buendia,K.Miwa,T.Ngara,&K.Tanabe,Eds.).Hayama,Japan:IGES.

Jucker,T.,Bongalov,B.,Burslem,D.F.R.P.,Nilus,R.,Dalponte,M.,Lewis,S.L.,…Coomes,D.A.(2018).Topographyshapesthestructure,com‐positionandfunctionoftropical forest landscapes.Ecology Letters,21,989–1000.

Jucker,T.,Caspersen,J.,Chave,J.,Antin,C.,Barbier,N.,Bongers,F.,…Coomes,D.A. (2017).Allometric equations for integrating remotesensing imagery intoforestmonitoringprogrammes.Global Change Biology,23,177–190.

Kearsley, E., deHaulleville, T., Hufkens, K., Kidimbu, A., Toirambe, B.,Baert,G.,…Verbeeck,H.(2013).Conventionaltreeheight‐diameter

     |  17BASTIN et al.

relationshipssignificantlyoverestimateabovegroundcarbonstocksintheCongoBasin.Nature Communications,4,2269.

Kellner,J.R.,Clark,D.B.,&Hubbell,S.P.(2009).Pervasivecanopydy‐namics produce short‐term stability in a tropical rain forest land‐scape.Ecology Letters,12,155–164.

Kellner,J.R.,&Hubbell,S.P.(2017).Adultmortalityinalow‐densitytreepop‐ulationusinghigh‐resolutionremotesensing.Ecology,98,1700–1709.

Langsrud,Ø.(2003).ANOVAforunbalanceddata:UsetypeiiinsteadoftypeIIIsumsofsquares.Statistics and Computing,13,163–167.

Laurance,W.F.,Delamônica,P., Laurance,S.G.,Vasconcelos,H.L.,&Lovejoy, T. E. (2000). Conservation: Rainforest fragmentation killsbigtrees.Nature,404,836–836.

Le Toan, T., Quegan, S., Davidson, M. W. J., Balzter, H., Paillou, P.,Papathanassiou, K., … Ulander, L. (2011). The BIOMASS mission:Mappingglobalforestbiomasstobetterunderstandtheterrestrialcarbon cycle. Remote Sensing of Environment,115,2850–2860.

Lewis, S. L., Sonké, B., Sunderland, T., Begne, S. K., Lopez‐Gonzalez,G.,vanderHeijden,G.M.F.,…Zemagho,L. (2013).Above‐groundbiomassandstructureof260Africantropicalforests.Philosophical Transactions of the Royal Society B: Biological Sciences,368,20120295.

Lindenmayer,D.B.,Laurance,W.F.,&Franklin,J.F. (2012).Globalde‐clineinlargeoldtrees.Science,338,1305–1306.

Lutz,J.A.,Furniss,T.J., Johnson,D.J.,Davies,S. J.,Allen,D.,Alonso,A.,…Zimmerman,J.K.(2018).Globalimportanceoflarge‐diametertrees.Global Ecology and Biogeography,27,849–864.

Malhi,Y.,Wood,D.,Baker,T.R.,Wright, J.,Phillips,O. L.,Cochrane,T.,…Vinceti,B. (2006).Theregionalvariationofaboveground livebio‐mass in old‐growth Amazonian forests. Global Change Biology, 12,1107–1138.

Maniatis,D.,&Mollicone,D.(2010).Optionsforsamplingandstratifica‐tionfornationalforestinventoriestoimplementREDD+undertheUNFCCC.Carbon Balance and Management,5,9.

Mascaro, J., Detto, M., Asner, G. P., & Muller‐Landau, H. C. (2011).Evaluating uncertainty in mapping forest carbon with airborneLiDAR.Remote Sensing of Environment,115,3770–3774.

Meakem,V., Tepley, A. J., Gonzalez‐Akre, E. B.,Herrmann,V.,Muller‐Landau,H.C.,Wright,S.J.,…Anderson‐Teixeira,K.J. (2018).Roleoftreesizeinmoisttropicalforestcarboncyclingandwaterdeficitresponses.New Phytologist,219,947–958.

Molto,Q.,Hérault,B.,Boreux,J.‐J.,Daullet,M.,Rousteau,A.,&Rossi,V.(2014).Predictingtreeheightsforbiomassestimatesintropicalforests—AtestfromFrenchGuiana.Biogeosciences,11,3121–3130.

Nepstad,D.C.,Tohver,I.M.,Ray,D.,Moutinho,P.,&Cardinot,G.(2007).MortalityoflargetreesandlianasfollowingexperimentaldroughtinanAmazonforest.Ecology,88,2259–69.

New,M.,Hulme,M.,Jones,P.,New,M.,Hulme,M.,&Jones,P.(1999).Representingtwentieth‐centuryspace‐timeclimatevariability.PartI:Developmentofa1961–90meanmonthlyterrestrialclimatology.Journal of Climate,12,829–856.

New,M.,Lister,D.,Hulme,M.,&Makin,I.(2002).Ahigh‐resolutiondatasetof surface climateover global landareas.Climate Research,21,1–25.

Ploton, P., Barbier, N., Couteron, P., Antin, C. M., Ayyappan, N.,Balachandran,N., … Pélissier, R. (2017). Toward a general tropicalforest biomass predictionmodel from very high resolution opticalsatelliteimages.Remote Sensing of Environment,200,140–153.

Ploton, P., Barbier, N., Takoudjou Momo, S., Réjou‐Méchain, M.,BoyembaBosela,F.,Chuyong,G.,…Pélissier,R.(2016).Closingagapintropicalforestbiomassestimation:Takingcrownmassvari‐ation into account in pantropical allometries.Biogeosciences, 13,1571–1585.

Ploton,P.,Pélissier,R.,&Proisy,C.(2012).Assessingabovegroundtrop‐ical forest biomass using Google Earth canopy images. Ecological Applications,22,993–1003.

Poorter,L.,vanderSande,M.T.,Thompson,J.,Arets,E.J.M.M.,Alarcón,A.,Álvarez‐Sánchez,J.,…Peña‐Claros,M.(2015).Diversityenhancescarbonstorage intropical forests.Global Ecology and Biogeography,24,1314–1328.

Proisy, C., Couteron, P., & Fromard, F. (2007). Predicting and map‐ping mangrove biomass from canopy grain analysis using Fourier‐based textural ordination of IKONOS images. Remote Sensing of Environment,109,379–392.

Réjou‐Méchain, M., Tanguy, A., Piponiot, C., Chave, J., & Hérault, B.(2017). biomass: An R package for estimating above‐ground bio‐massanditsuncertaintyintropicalforests.Methods in Ecology and Evolution,8,1163–1167.

Remm,J.,&Lõhmus,A.(2011).Treecavitiesinforests—Thebroaddistri‐butionpatternofakeystonestructureforbiodiversity.Forest Ecology and Management,262,579–585.

Romijn,E.,Herold,M.,Kooistra,L.,Murdiyarso,D.,&Verchot,L.(2012).Assessing capacities of non‐Annex I countries for national forestmonitoringinthecontextofREDD+.Environmental Science & Policy,19–20,33–48.

Rutishauser,E.,Hérault,B.,Baraloto,C.,Blanc,L.,Descroix,L.,Sotta,E.D.,…Sist,P. (2015).Rapid treecarbonstock recovery inmanagedAmazonianforests.Current Biology: CB,25,R787–R788.

Rutishauser,E.,Wagner,F.,Herault,B.,Nicolini,E.‐A.,&Blanc,L.(2010).Contrastingabove‐groundbiomassbalanceinaNeotropicalrainfor‐est.Journal of Vegetation Science,672–682.

Saatchi, S. S.,Houghton, R.A.,Dos SantosAlvalá, R.C., Soares, J.V.,& Yu, Y. (2007). Distribution of aboveground live biomass in theAmazonbasin.Global Change Biology,13,816–837.

Sist,P.,Mazzei,L.,Blanc,L.,&Rutishauser,E.(2014).LargetreesaskeyelementsofcarbonstorageanddynamicsafterselectiveloggingintheEasternAmazon.Forest Ecology and Management,318,103–109.

Slik, J. W. F., Aiba, S.‐I., Brearley, F. Q., Cannon, C. H., Forshed, O.,Kitayama, K.,…VanValkenburg, J. L. C.H. (2010). Environmentalcorrelates of tree biomass, basal area, wood specific gravity andstem density gradients in Borneo’s tropical forests.Global Ecology and Biogeography,19,50–60.

Slik, J.W. F., Alvarez‐loayza, P., Alves, L. F., Ashton, P., Balvanera, P.,Bastian,M.L.,…Bernacci,L.(2015).Anestimateofthenumberoftropicaltreespecies.Proceedings of the National Academy of Sciences of the USA,112,E4628–E4629.

Slik, J. W. F., Franklin, J., Arroyo‐Rodríguez, V., Field, R., Aguilar, S.,Aguirre, N., … Zang, R. (2018). Phylogenetic classification of theworld’s tropical forests. Proceedings of the National Academy of Sciences of the USA,115,1837–1842.

Slik,J.W.F.,Paoli,G.,McGuire,K.,Amaral, I.,Barroso,J.,Bastian,M.,…Zweifel,N.(2013).Largetreesdriveforestabovegroundbiomassvariationinmoistlowlandforestsacrossthetropics.Global Ecology and Biogeography,22,1261–1271.

Stark,S.C.,Enquist,B.J.,Saleska,S.R.,Leitold,V.,Schietti,J.,Longo,M.,…Oliveira,R.C.(2015).Linkingcanopyleafareaandlightenviron‐mentswithtreesizedistributionstoexplainAmazonforestdemog‐raphy.Ecology Letters,18,636–645.

Stark,S.C.,Leitold,V.,Wu,J.L.,Hunter,M.O.,deCastilho,C.V.,Costa,F.R.C.,…Saleska,S.R.(2012).Amazonforestcarbondynamicspre‐dictedbyprofilesofcanopyleafareaandlightenvironment.Ecology Letters,15,1406–1414.

Stegen,J.C.,Swenson,N.G.,Enquist,B.J.,White,E.P.,Phillips,O.L.,Jørgensen, P. M., … Núñez Vargas, P. (2011). Variation in above‐groundforestbiomassacrossbroadclimaticgradients.Global Ecology and Biogeography,20,744–754.

Sullivan,M.J.P.,Lewis,S.L.,Hubau,W.,Qie,L.,Baker,T.R.,Banin,L.F.,…Phillips,O.L. (2018).Fieldmethods for sampling treeheight fortropicalforestbiomassestimation.Methods in Ecology and Evolution,9,1179–1189.

18  |     BASTIN et al.

Sullivan,M.J.P.,Talbot,J.,Lewis,S.L.,Phillips,O.L.,Qie,L.,Begne,S.K.,…Zemagho,L. (2017).Diversityandcarbonstorageacross thetropicalforestbiome.Scientific Reports,7,39102.

Tang, H., & Dubayah, R. (2017). Light‐driven growth in Amazon ever‐green forests explained by seasonal variations of vertical canopystructure.Proceedings of the National Academy of Sciences of the USA,114,2640–2644.

Thomas,R.Q.,Kellner,J.R.,Clark,D.B.,&Peart,D.R.(2013).Lowmor‐talityintalltropicaltrees.Ecology,94,920–929.

Vincent,G.,Antin,C.,Laurans,M.,Heurtebize,J.,Durrieu,S.,Lavalley,C.,&Dauzat,J. (2017).Mappingplantareaindexoftropicalevergreenforest by airborne laser scanning. A cross‐validation study usingLAI2200opticalsensor.Remote Sensing of Environment,198,254–266.

Xu,L.,Saatchi,S.S.,Shapiro,A.,Meyer,V.,Ferraz,A.,Yang,Y.,…Ebuta,D. (2017). Spatial distribution of carbon stored in forests of theDemocraticRepublicofCongo.Scientific Reports,7,15030.

Zanne,A.E.,Lopez‐Gonzalez,G.,Coomes,D.A.,Ilic,J.,Jansen,S.,Lewis,…Chave,J.(2009).Globalwooddensitydatabase.Dryad,235,33.

SUPPORTING INFORMATION

Additional Supporting Information may be found online in the supportinginformationtabforthisarticle.

How to cite this article:BastinJ‐F,RutishauserE,KellnerJR,etal.Pan‐tropicalpredictionofforeststructurefromthelargesttrees.Global Ecol Biogeogr. 2018;00:1–18. https://doi.org/10.1111/geb.12803

BIOSKETCHJean‐Francois Bastin is a postdoctoral Fellow of the CrowtherLaboratoryattheInstituteofIntegrativeBiology,DepartmentofEnvironmentalSystemsScience,ETH‐Zurich.Heisanecologistand a geographer using remote sensing to study the effect ofglobalchangeonterrestrialecosystems.Ervan Rutishauser is a postdoctoral Fellow at the SmithsonianTropical Research Institute, broadly interested in understand‐ing environmental resilience to natural or human‐induced dis‐turbances.Heaimsatprovidingrigorousandpracticalevidenceto trigger a shift towards better ressource management andconservation.