ground biomass in Amazonian forests

18 Pages·2016·1.04 MB·English

GlobalChangeBiology(2016),doi:10.1111/gcb.13315
Variation in stem mortality rates determines patterns of
above-ground biomass in Amazonian forests:
implications for dynamic global vegetation models
MICHELLE O. JOHNSON1, DAVID GALBRAITH1, MANUEL GLOOR1,
HANNES DE DEURWAERDER2, MATTHIEU GUIMBERTEAU3,4, ANJA RAMMIG5,6,
KIRSTEN THONICKE6, HANS VERBEECK2, CELSO VON RANDOW7,
ABEL MONTEAGUDO8, OLIVER L. PHILLIPS1, ROEL J. W. BRIENEN1,
TED R. FELDPAUSCH9, GABRIELA LOPEZ GONZALEZ1, SOPHIE FAUSET1,
CARLOS A. QUESADA10, BRADLEY CHRISTOFFERSEN11,12, PHILIPPE CIAIS3,
GILVAN SAMPAIO7, BART KRUIJT13, PATRICK MEIR11,14, PAUL MOORCROFT15,
KE ZHANG16, ESTEBAN ALVAREZ-DAVILA17, ATILA ALVES DE OLIVEIRA10, IEDA
AMARAL10, ANA ANDRADE10, LUIZ E. O. C. ARAGAO8, ALEJANDRO ARAUJO-
MURAKAMI18, ERIC J. M. M. ARETS13, LUZMILA ARROYO18, GERARDO A. AYMARD19,
CHRISTOPHER BARALOTO20, JOCELY BARROSO21, DAMIEN BONAL22, RENE BOOT23,
JOSE CAMARGO10, JEROME CHAVE24, ALVARO COGOLLO25, FERNANDO CORNEJO
VALVERDE26, ANTONIO C. LOLA DA COSTA27, ANTHONY DI FIORE28, LEANDRO
FERREIRA29, NIRO HIGUCHI10, EURIDICE N. HONORIO30, TIM J. KILLEEN31, SUSAN G.
LAURANCE32, WILLIAM F. LAURANCE32, JUAN LICONA33, THOMAS LOVEJOY34,
YADVINDER MALHI35, BIA MARIMON36, BEN HUR MARIMON JUNIOR36, DARLEY C. L.
MATOS29, CASIMIRO MENDOZA37, DAVID A. NEILL38, GUIDO PARDO39, MARIELOS
PEN˜ A-CLAROS33,40, NIGEL C. A. PITMAN41, LOURENS POORTER40, ADRIANA PRIETO42,
HIRMA RAMIREZ-ANGULO43, ANAND ROOPSIND44, AGUSTIN RUDAS42,
RAFAEL P. SALOMAO29, MARCOS SILVEIRA45, JULIANA STROPP46, HANS TER STEEGE47,
JOHN TERBORGH41, RAQUEL THOMAS44, MARISOL TOLEDO33,
ARMANDO TORRES-LEZAMA43, GEERTJE M. F. VAN DER HEIJDEN48,
RODOLFO VASQUEZ9, IMA CE(cid:1)LIA GUIMARA~ES VIEIRA29, EMILIO VILANOVA43,
VINCENT A. VOS49,50 andTIMOTHY R. BAKER1
1SchoolofGeography,UniversityofLeeds,LeedsLS62QT,UK,2CAVElabComputational&AppliedVegetationEcology,Faculty
ofBioscienceEngineering,GhentUniversity,CoupureLinks653,B-9000Gent,Belgium,3LaboratoiredesSciencesduClimatetde
l’Environnement,LSCE/IPSL,CEA-CNRS-UVSQ,Universite´Paris-Saclay,F-91191Gif-sur-Yvette,France,4UMR7619
METIS,IPSL,SorbonneUniversite´s,UPMC,CNRS,EPHE,75252Paris,France,5TUMSchoolofLifeSciencesWeihenstephan,
TechnicalUniversityMunich,Hans-Carl-von-Carlowitz-Platz2,85354Freising,Germany,6PotsdamInstituteforClimateImpact
Research(PIK),TelegrafenbergA62,POBox601203,D-14412Potsdam,Germany,7INPE,Av.DosAstronautas,1.758,Jd.
Granja,CEP:12227-010SaoJosedosCampos,SP,Brazil,8Jard´ınBota´nicodeMissouri,ProlongacionBolognesiMz.e,Lote6,
Oxapampa,Pasco,Peru,9Geography,CollegeofLifeandEnvironmentalSciences,UniversityofExeter,RennesDrive,ExeterEX4
4RJ,UK,10INPA,Av.Andre´Arau´jo,2.936,CEP69067-375Petro´polis,Manaus,AM,Brazil,11SchoolofGeosciences,University
ofEdinburgh,EdinburghEH93FF,UK,12EarthandEnvironmentalSciencesDivision,LosAlamosNationalLaboratory,POBox
1663,LosAlamos,NM 87545,USA,13ALTERRA,Wageningen-UR,POBox47,6700AAWageningen,TheNetherlands,
14ResearchSchoolofBiology,AustralianNationalUniversity,Canberra,ACT0200,Australia,15DepartmentofOrganismicand
EvolutionaryBiology,HarvardUniversity,26OxfordStreet,Cambridge,MA 02138,USA,16CooperativeInstituteforMesoscale
MeteorologicalStudies,UniversityofOklahoma, NationalWeatherCenter, Suite2100,120DavidL.BorenBlvd,Norman,OK
73072,USA,17Fundacio´nCon-Vida,Cr68A46A-77Medell´ın,Medell´ın,Colombia,18MuseodeHistoriaNaturalNoelKempff
Mercado,UniversidadAutonomaGabrielReneMoreno,Casilla2489,Av.Irala565,SantaCruz,Bolivia,19UNELLEZ-Guanare,
ProgramadeCienciasdelAgroyelMar,HerbarioUniversitario(PORT),MesadeCavacas,EstadoPortuguesa3350,Venezuela,
20DepartmentofBiologicalSciences,InternationalCenterforTropicalBotany(ICTB),FloridaInternationalUniversity,112200
SW8thStreet,OE167,Miami,FL33199,USA,21UniversidadeFederaldoAcre,CampusdeCruzeirodoSul,RioBranco,Brazil,
22INRA,UMR1137“EcologieetEcophysiologieForestiere”,54280Champenoux,France,23TropenbosInternational,POBox232,
6700AEWageningen,TheNetherlands,24Universite´PaulSabatierCNRS,UMR5174EvolutionetDiversite´Biologique,
Correspondence:TimothyR.Baker,tel.[phone],fax
[phone],e-mail:[email protected]
©2016TheAuthors.GlobalChangeBiologyPublishedbyJohnWiley&SonsLtd. 1
ThisisanopenaccessarticleunderthetermsoftheCreativeCommonsAttributionLicense,whichpermitsuse,
distributionandreproductioninanymedium,providedtheoriginalworkisproperlycited.
2 M. O. JOHNSON etal.
baˆtiment4R1,31062Toulouse,France,25Jard´ınBota´nicodeMedell´ınJoaqu´ınAntonioUribe, Calle73#51D14Medell´ın,
Colombia,26AndestoAmazonBiodiversityProgram,PuertoMaldonado,MadredeDios,Peru´,27CentrodeGeociencias,
UniversidadeFederaldoPara,CEP66017-970Belem,Para,Brazil,28DepartmentofAnthropology,UniversityofTexasatAustin,
SACRoom5.150,2201SpeedwayStopC3200,Austin,TX78712,USA,29MuseuParaenseEmilioGoeldi,Av.Magalha˜esBarata,
376-Sa˜oBraz,CEP:66040-170Bele´m,PA,Brazil,30InstitutodeInvestigacionesdelaAmazon´ıaPeruana,Av.Jose´Quin˜oneskm
2.5,Iquitos,Peru´,31WorldWildlifeFund,125024thStNW,Washington,DC20037,USA,32CentreforTropicalEnvironmental
andSustainabilityScience(TESS)andCollegeofMarineandEnvironmentalSciences,JamesCookUniversity,Cairns,Qld4878,
Australia,33InstitutoBolivianodeInvestigacio´nForestal,C.P.6201SantaCruzdelaSierra,Bolivia,34EnvironmentalScienceand
PolicyDepartmentandtheDepartmentofPublicandInternationalAffairsatGeorgeMasonUniversity(GMU),3351Fairfax
Drive,Arlington, VA22201,USA,35EnvironmentalChangeInstitute,SchoolofGeographyandtheEnvironment,Universityof
Oxford,SouthParksRoad,OxfordOX13QY,UK,36UniversidadedoEstadodeMatoGrosso,CampusdeNovaXavantina,Caixa
Postal08,CEP78.690-000NovaXavantina,MT,Brazil,37EscueladeCienciasForestales(ESFOR),Av.FinalAtahuallpas/n,
Casilla447,Cochabamba,Bolivia,38FacultaddeIngenier´ıaAmbiental,UniversidadEstatalAmazo´nica,Pasolateralkm21/2via
Napo,Puyo,Pastaza,Ecuador,39UniversidadAutonomadelBeni,CampusUniversitario,Av.Eje´rcitoNacional,final,Riberalta,
Beni,Bolivia,40ForestEcologyandForestManagementGroup,WageningenUniversity,POBox47,Wageningen6700AA,The
Netherlands,41CenterforTropicalConservation,DukeUniversity,Box90381,Durham,NC27708,USA,42DoctoradoInstituto
deCienciasNaturales,UniversidadNacionaldeColombia,Bogota´,Colombia,43InstitutodeInvestigacionesparaelDesarrollo
Forestal,UniversidaddeLosAndes,AvenidaPrincipalChorrosdeMilla,CampusUniversitarioForestal,EdificioPrincipal,
Me´rida,Venezuela,44IwokramaInternationalCentreforRainforestConservationandDevelopment,77HighStreetKingston,
Georgetown,Guyana,45MuseuUniversita´rio,UniversidadeFederaldoAcre,RioBranco,AC69910-900,Brazil,46Instituteof
BiologicalandHealthSciences,FederalUniversityofAlagoas, Av.LourivalMeloMota,s/n,TabuleirodoMartins,Macei(cid:1)o, AL
57072-900,Brazil,47NaturalisBiodiversityCenter,POBox 9517,2300RALeiden,TheNetherlands,48SchoolofGeography,
UniversityofNottingham,NottinghamNG72RD,UK,49CentrodeInvestigacio´nyPromocio´ndelCampesinado,regionalNorte
Amazo´nico,C/NicanorGonzaloSalvatierraN˚362,Casilla16,Riberalta,Bolivia,50UniversidadAuto´nomadelBeni,Avenida6de
AgostoN(cid:1)64,Riberalta,Bolivia
Abstract
Understandingtheprocessesthatdetermineabove-groundbiomass(AGB)inAmazonianforestsisimportantforpre-
dictingthesensitivityoftheseecosystemstoenvironmentalchangeandfordesigningandevaluatingdynamicglobal
vegetationmodels(DGVMs).AGBisdeterminedbyinputsfromwoodyproductivity[woodynetprimaryproductiv-
ity(NPP)]andtherateatwhichcarbonislostthroughtreemortality.Here,wetestwhethertwodirectmetricsoftree
mortality(theabsoluterateofwoodybiomasslossandtherateofstemmortality)and/orwoodyNPP,controlvaria-
tioninAGBamong167plotsinintactforestacrossAmazonia.Wethencomparetheserelationshipsandtheobserved
variation in AGB and woodyNPP with the predictions of four DGVMs. The observations show that stem mortality
rates,ratherthanabsoluteratesofwoodybiomassloss,arethemostimportantpredictorofAGB,whichisconsistent
withtheimportanceofstandsizestructurefordeterminingspatialvariationinAGB.Therelationshipbetweenstem
mortalityratesandAGBvariesamongdifferentregionsofAmazonia,indicatingthatvariationinwooddensityand
height/diameterrelationshipsalsoinfluencesAGB.Incontrasttopreviousfindings,wefindthatwoodyNPPisnot
correlated with stem mortality rates and is weakly positively correlated with AGB. Across the four models, basin-
wideaverageAGBissimilartothemeanoftheobservations.However,themodelsconsistentlyoverestimatewoody
NPP and poorly represent the spatial patterns of both AGB and woody NPP estimated using plot data. In marked
contrast to the observations, DGVMs typically show strong positive relationships between woody NPP and AGB.
Resolvingthesedifferenceswillrequireincorporatingforestsizestructure,mechanisticmodelsofstemmortalityand
variationinfunctionalcompositioninDGVMs.
Keywords: allometry,carbon,dynamicglobalvegetationmodel,forestplots,productivity,tropicalforest
Received3October2015;revisedversionreceived5February2016andaccepted1March2016
basininparticularcomprisesapproximately50%ofthe
Introduction
world’s tropical forests, and therefore, any perturba-
Tropical forests are the most carbon-rich and produc- tions to this ecosystem will have important feedbacks
tive of all forest biomes (Pan et al., 2011). The Amazon onbothcarboncyclingandclimateworldwide(Zhao&
©2016TheAuthors.GlobalChangeBiologyPublishedbyJohnWiley&SonsLtd.,doi:10.1111/gcb.13315
STEM MORTALITY CONTROLS FOREST BIOMASS 3
Running,2010;Wanget al.,2014).Itisthereforeimpor- processes that drive variation in above-ground carbon
tant that we understand the processes that determine stocks, which can also be used to evaluate and cali-
current patterns of carbon storage and cycling to pre- brate DGVMs. For example, the paradigm to emerge
dict how the productivity and carbon stores of these from previous analysis of plot data in Amazonia is
forests will respond to changing environmental condi- that there is a positive association between woody
tions. NPP and stem mortality rates, linked to a reduction
Ourknowledgeofthesensitivityofrainforestecosys- in AGB (Baker et al., 2004; Malhi et al., 2004; Quesada
tems to environmental change is based on three et al., 2012). This finding has been used to evaluate
sources. Firstly, observational data from networks of the architecture and outputs of DVGMs (Negro´n-
permanent plots, flux towers, remote sensing and air- Jua´rez et al., 2015) and has stimulated attempts to
craft measurements of greenhouse gas concentrations make direct links between mortality and woody NPP
have demonstrated the sensitivity of these ecosystems in these models (Delbart et al., 2010; Castanho et al.,
to environmental change, particularly in response to 2013).
drought(e.g.Phillipset al.,2009;Restrepo-Coupeet al., More generally, observational data are valuable for
2013;Gattiet al.,2014).Secondly,experimentalmanipu- informing how the fundamental processes that influ-
lations of water stress have probed the mechanisms enceAGBshouldbeincludedinvegetationmodels.For
behind these responses (e.g. Nepstad et al., 2007; da example, the residence time of woody biomass, s
w
Costaet al.,2010;Meiret al.,2015;Rowlandet al.,2015). (years), is often used as a measure of mortality in
Thirdly, process-based ecosystem models, especially DGVMsandisdefinedforaforestatsteadystateas:
dynamicglobalvegetationmodels(DGVMs),havebeen
AGB
used to explore the future sensitivity of Amazon vege- sw ¼ W : ð1Þ
P
tation to increasing temperatures, carbon dioxide con-
centrationsandwaterstress(e.g.Galbraithet al.,2010). This parameter varies almost sixfold among tropical
Coupled with climate models, DGVMs have high- forest plots (Galbraith et al., 2013). However, surpris-
lighted the sensitivity (Cox et al., 2004), and more ingly, in several commonly used vegetation models,
recently, the resilience (Rammig et al., 2010; Hunting- thisparameterisconstant;Galbraithet al.(2013)found
fordet al.,2013)ofAmazonianforeststoenvironmental that 21 of the 27 vegetation models they compared use
change. However, observations of above-ground bio- single, fixed values for this parameter. In addition,
mass (AGB, Mg C ha(cid:1)1) and woody productivity (the observational data suggest that the ultimate cause of
amount of net primary productivity (NPP) allocated to variationintreemortality,WPandhenceAGBisvaria-
above-groundwoodygrowth:W ,MgCha(cid:1)1 yr(cid:1)1)are tion in edaphic properties (Quesada et al., 2012). Que-
P
still little used to parameterize and evaluate DGVMs sada et al. (2012) found that spatial differences in WP
(e.g. Delbart et al., 2010; Castanho et al., 2013), despite correlated most strongly with total soil phosphorus,
substantial progress increasing the spatial distribution whereas stem mortality rates correlated with a soil
ofsuchinsituobservations(e.g.Feldpauschet al.,2011; physical structure index which combined soil depth,
Quesada et al., 2012; Mitchard et al., 2014). Integrating texture, topography and anoxia. Most DGVMs, how-
the insights from such observational studies into the ever,onlyincludeverylimitedfeedbacksbetweenveg-
design, calibration and validation of DGVMs would etation and edaphic properties. Soil properties such as
enhance our ability to make convincing predictions of texturearemainlyimplementedintoDGVMstoparam-
thefutureoftropicalcarbon. eterize hydraulic processes (e.g. Marthews et al., 2014)
Observational data can either be used to evaluate and soil structure and nutrient content are rarely con-
the outputs of models, or more fundamentally, cali- sideredforotherprocessessuchasstemmortality.
brate and inform the processes that models should Overall,theaimofthisstudyistocomparehowvari-
aim to include. For example, networks of inventory ation in WP and mortality control variation in AGB in
plots have revealed strong differences in AGB among Amazonia using both field observations and four
terra firme forests in north-east and south-western DGVMs, to inform the future development of vegeta-
Amazonia (Baker et al., 2004; Malhi et al., 2006; Bar- tion models. In terms of the analysis of observations,
aloto et al., 2011; Quesada et al., 2012; Mitchard et al., we build on previous work (e.g. Baker et al., 2004;
2014). Such observations have been used to evaluate Malhi et al., 2004, 2015) in two ways. Firstly, we com-
the predictions of Amazonian forest biomass from parepatternsofAGBwithvariationintwodirectmea-
both remote sensing (e.g. Mitchard et al., 2014) and surements of mortality from each plot: the absolute,
DGVM studies (e.g. Castanho et al., 2013). These field stand-level rate of woody biomass loss (WL; Mg C
observations also yield information about the ha(cid:1)1 yr(cid:1)1) and the rate of stem mortality (l; % yr(cid:1)1).
©2016TheAuthors.GlobalChangeBiologyPublishedbyJohnWiley&SonsLtd.,doi:10.1111/gcb.13315
4 M. O. JOHNSON etal.
Previousstudieshaveuseds toexaminehowmortal- climate, forest structure and species composition (e.g.
w
ity influences AGB (e.g. Malhi et al., 2004, 2015; Gal- terSteegeet al.,2006;Feldpauschet al.,2011).Here,we
braith et al., 2013). However, although s is a useful use data from these regions to test whether the para-
w
parameterinthecontextofvegetationmodellingandto digmofapositiveassociationbetweenwoodyNPPand
partition ecosystem carbon fluxes, its dependency on stem mortality rates, linked to a reduction in AGB, is
AGB (see Eqn 1) means that this term is not an inde- foundacrossthefullrangeofSouthAmericanlowland
pendentcontrolofAGBstocks:itisinevitablethatAGB moisttropicalforests.
is inversely related to s . In addition, as s is defined In terms of the analysis of the DGVMs, we aim
w w
foraforestatsteadystate,itcannotbeeasilyrelatedto firstly to establish the reliability of land vegetation
specific short-term processes, such as droughts, which simulation for the Amazon basin by comparison of
ultimately cause tree mortality. Here, we therefore test modelling results with kriged maps of field observa-
the sensitivity of AGB to direct independent measures tions of W , mortality and AGB that illustrate the
P
ofbothstand-levelandstem-levelvariationinmortality major patterns of variation in these variables. We then
rates,asthesemeasuresmayultimatelyprovideamore test how well the four DGVMs capture these spatial
appropriatebasisformodellingmortalityinDGVMs. patterns and the overall magnitude of AGB and W .
P
Secondly, we greatly extend the spatial coverage of Finally, we explore the relationships between simu-
observations. The first large-scale studies of Amazon lated AGB, W and s . By comparing our findings
P w
forest dynamics (Baker et al., 2004; Malhi et al., 2004; from the analysis of the observations and simulation
Phillipset al.,2004)focusedonthewestern,andcentral results, we conclude by making recommendations for
andeasternportionsofthebasin,butincludedfewdata model developments and data collection that will
from forests on the Guiana and Brazilian Shields improve our ability to model Amazonian vegetation
(Fig. 1). These areas, however, have distinctive soils, carbon stocks.
(a) (b)
(c)
Fig.1 Locationofplotsusedtocalculate(a)above-groundwoodybiomass,(b)above-groundwoodyproductivityandstemandbio-
mass-basedmortalityand(c)thepositionofthekriged1°mapgridcells.TheAmazonbasinincludingforestsontheGuianaShieldis
splitintoregions(shownbydifferentcolours)thataredefinedinFeldpauschetal.(2011).Plotlocationsarenotgeographicallyexact
butareoffsetslightlytoimprovethevisualizationofplotswhichareinverycloseproximitytoeachother.
©2016TheAuthors.GlobalChangeBiologyPublishedbyJohnWiley&SonsLtd.,doi:10.1111/gcb.13315
STEM MORTALITY CONTROLS FOREST BIOMASS 5
where D is stem diameter (cm), q is stem wood density
Materialsandmethods (gcm(cid:1)3),Hisstemheight(m)andnisthenumberoftreesin
the stand. We retained the use of this biomass equation for
Plotobservations
this study, instead of using the recent biomass equation of
Chaveetal.(2014),toprovideestimatesofW thatareconsis-
We used tree inventory data from permanent sample plots P
tent with Mitchard et al. (2014). Estimates of AGB for moist
locatedthroughoutAmazoniacompiledaspartoftheRAIN-
tropicalforestsareinfactsimilarusingeitherequation(Chave
FORandTEAMnetworkstoestimatestocks(AGB)andfluxes
etal., 2014). The height of each tree was estimated from tree
of carbon (woody NPP, stem and biomass mortality) within
diameter using a height-diameter Weibull equation with dif-
Amazonian forest stands (Fig. 1). For analysis of AGB, we
ferent coefficients for each region, based on field-measured,
used the data for the 413 plots analysed by Mitchard et al.
height-diameter relationships (Feldpausch etal., 2011). We
(2014) (Fig. 1a). For properties which can only be calculated
usedthismethodtoestimatetreeheight,ratherthanpredict-
by observing change over time and thus require more than
onecensus,plotsinintact,moist,lowland(<1000masl)forest ing height on the basis of climate as in Chave et al. (2014),
becauseamongmoistforestsinAmazonia,theprincipalvaria-
werechosenwhichhadaminimumtotalmonitoringperiodof
tion in height/diameter allometry is due to the contrast
2 years between 1995 and 2009 inclusive. Data for 167 plots
between the particularly tall-statured forests on the Guiana
thatmetthesecriteriaforanalysisofdynamicpropertieswere
Shield and shorter-statured forest in other regions (Feld-
downloadedfromForestPlots.net(Lopez-Gonzalezetal.,2011,
pausch etal., 2011). This difference is related to the unique
2012;Johnsonet al.,2016;Fig. 1bandTableS1)andtheTEAM
species composition of forests on the Guiana Shield rather
website (http://www.teamnetwork.org/data/query; data set
than variation in climate (Feldpausch etal., 2011). The wood
identifier codes 20130415013221_3991 and 20130405063033_
density of each tree wasassigned on a taxonomicbasis from
1587). For this data set, mean plot size is 1.09ha, the mean
the pan-tropical database of Zanne etal. (2009) and Chave
dateofthefirstcensusis2000.2andthemeandateofthefinal
etal. (2009), following Baker etal. (2004). Mean plot wood
censusis2008.5.Meancensusintervallengthis3.70yearsand
density values were used when taxonomic information was
plot mean total monitoring period is 8.3years. Most of the
missingforindividualtrees.
plotsweremonitoredformostofthetimeperiod:onaverage,
To estimate total above-ground woody biomass, we
76% of plots were being monitored in any given year from
2000–2008(Fig.S1).Alltreeswithadiameteratbreastheight assumed that carbon is 50% of total dry biomass (Penman
etal., 2003) and to account for the unmeasured, small trees
(dbh)greaterthan10cmwereincludedintheanalyses.
(<10cm), we added an additional 6.2% of carbon to each of
Plotswereclassifiedintofourregionsoflowlandmoistforest
theplots,followingMalhietal.(2006).Wedonotincludethe
definedbythenatureandgeologicalageofthesoilsubstrate
unknown contributions from lianas, epiphytes, necromass,
(Fig.1;Feldpauschetal.,2011).ThesoilsandforestsoftheGui-
shrubsandherbs.
anaandBrazilianShieldshavedevelopedonold,Cretaceous,
crystallinesubstrates,whereastheforestsofWesternAmazonia
are underlain by younger Andean substrates and Miocene Mortalityandproductivity
deposits(Irion,1978;Quesadaetal.,2010;Higginsetal.,2011).
East-central Amazonia contains reworked sediments derived Stem mortality rates were calculated as the exponential mor-
fromtheotherthreeregionsthathaveundergonealmostcontin- talitycoefficientl[%yr(cid:1)1;Sheil&May(1996)]:
uousweatheringformorethan20millionyears,leadingtovery
nutrientpoorsoils(Irion,1978;Quesadaetal.,2010).Previous l¼lnðn0Þ(cid:1)lnðn0(cid:1)ndÞ(cid:3)100; ð3Þ
comparativestudieshavenotedsubstantialdifferencesinforest t
dynamicsbetweenWesternandEast-centralAmazonia(Baker
where n is the number of stems at the start of the census
etal.,2004,2014;Quesadaetal.,2012),butlargelyexcludedfor- 0
interval,n isthenumberofstemsthatdieintheintervaland
ests on the Guiana and Brazilian Shields. This classification d
tisthecensusintervallength.Asestimatesofmortalityrates
thereforeallowsustotesttheimpactofincludingthesedistinc-
in heterogeneous populations are influenced by the census
tiveforestsonAmazon-widepatternsofforestdynamics.
interval, we standardized our estimates of l to comparable
censusintervalsusingtheequationofLewisetal.(2004).We
Above-groundbiomass calculated corrected values of l for each census interval for
each plot in the data set, and calculated average values of l
ForAGBvalues,weusedthedatasetpresentedbyMitchard perplot,weightedbythecensusintervallength.
etal.(2014)andLopez-Gonzalezet al.(2014).Inbrief,forthis TotalNPPcannotbecalculatedfromtreeinventoriesasthis
dataset,theAGB(MgDWha(cid:1)1)ofeachplotwascalculated includes both the growth of the stem as well as litterfall and
using the Chave etal. (2005) moist forest allometric equa- rootproductionwhichhasonlybeenmeasuredatarelatively
tionwhichincludesmeasurementsofdiameter,wooddensity small number of Amazonian sites (Malhi et al., 2015). There-
andheight: fore,we arerestricted tocalculatingW ,whichcanbecalcu-
P
P
nð0:0509qD2HÞ lated from repeated censuses of tree diameters within
AGB¼ 1 ; ð2Þ inventoryplots.Comparableoutputcanbeobtainedfromveg-
1000
©2016TheAuthors.GlobalChangeBiologyPublishedbyJohnWiley&SonsLtd.,doi:10.1111/gcb.13315
6 M. O. JOHNSON etal.
etation models as DGVMs typically partition total above- from four DGVMs. The DGVMs included in this study are
groundNPPintodifferentcarbonpoolsusingvariouscarbon the joint uk land environment simulator (jules), v. 2.1. (Best
allocation algorithms, ranging from fixed coefficients (e.g. etal.,2011;Clarket al.,2011),theLund-Potsdam-JenaDGVM
INLAND) to approaches based on resource limitation (e.g. for managed Land (LPJmL; Sitch etal., 2003; Gerten et al.,
ORCHIDEE). For comparison with measurement data, we 2004; Bondeau et al., 2007), the INtegrated model of LAND
used the fraction of simulated above-ground NPP that the surface processes (INLAND) model (a development of the
modelsallocatetowoodygrowth.Boththeobservedmeasure- IBISmodel,Kuchariketal.,2000)andtheOrganisingCarbon
mentsandmodelsexcludethecontributiontoW thatismade andHydrology InDynamicEcosystEms(ORCHIDEE) model
P
bythelossandregrowthoflargewoodybranches.Thiscom- (Krinner etal., 2005). A brief description of each of the four
ponentisapproximately1MgCha(cid:1)1 a(cid:1)1inAmazonianfor- models and how output data are derived is included in the
estsor10%ofabove-groundNPP(Malhiet al.,2009).W was supplementary information (Appendix S2). The models each
L
calculated as the sum of the biomass of all trees that died followed the standardized Moore Foundation Andes-Ama-
withinagivencensusinterval. zon Initiative (AAI) modelling protocol (Zhang et al., 2015).
EstimatesofW andW areinfluencedbythecensusinter- The simulated region spanned 88°W to 34°W and 13°N to
P L
val over which they are calculated, because more trees will 25°S. Simulations from each model included a spin-up per-
recruit anddie without being recorded during longer census iod from bare ground of up to 500years with pre-industrial
intervals(Talbotetal.,2014).WefollowedthemethodsofTal- atmospheric CO (278ppm). The models were then forced
2
bot et al. (2014) for calculating W with forest inventory data by recycling 39year, 1° spatial resolution, bias-corrected
P
tocorrectforthisbias(Supportinginformation,AppendixS1). NCEP meteorological data (Sheffield etal., 2006) for 1715–
Thus, we calculated W as the sumof (i) the growth of trees 2008 with increasing CO concentrations, as in Zhang et al.
P 2
that survive the census period, and the estimated growth of (2015). Figure S3 shows the spatial distribution of mean
(ii) trees that died during the census interval, prior to their meteorological variables for 2000–2008 across the Amazon
death, (iii) trees which recruited within the interval, and (iv) basin. As well as precipitation, temperature and short-wave
treesthatbothrecruitedanddiedduringthecensusinterval. radiation we also show maximum cumulative water deficit
Similarly, to calculate W we summed the biomass of trees (MWD), calculated from monthly precipitation values to
L,
thatdiewithinacensusintervalwithcomponents(ii)and(iv) indicatedroughtseverityacrossthebasin,asinAragaoet al.
above.WecalculatedcorrectedvaluesofW andW foreach (2007).Thetimeperiodofmodeloutputis2000–2008.
P L
census interval for each plot in the data set, and calculated TocomparesimulatedwoodyNPPwithobservedW ,cor-
P
averagevaluesperplot,weightedbycensusintervallength. rections were applied to the simulated total woody NPP to
calculate above-ground woody NPP only, by assuming a
Analysisofobservationaldata below-groundtoabove-groundallocationratioof0.21(Malhi
etal.,2009).InthecaseofJULES,onlyafractionoftheNPPis
ThecurrentparadigmforAmazonianforestssuggeststhatW
P allocated tobiomass growth, astheremainder is allocatedto
and l are positively correlated and that both correlate nega-
‘spreading’ of vegetated area – an increase in the fraction of
tively withAGB(Malhiet al.,2002;Quesadaetal., 2012).We
grid cell cover (Cox, 2001). To facilitate comparison with
tested whether these relationships are supported by the data
observationsandothermodels,wethereforerescaledW from
from across South American tropical lowland moist forest, P
JULES,retainingtherelativeallocationtowoodbutassuming
includingplotsfromtheGuianaandBrazilianShield.Firstly,
thatalloftheNPPwasusedforgrowth.
weexploredwhetherdifferentregionshavedistinctivepatterns
We compared model outputs to kriged maps of AGB, W
ofcarboncyclingbycomparingW ,W ,landAGBamongthe P
P L and mortality to understand how well the DGVMs captured
fourregionsusingANOVA.Secondly,weexploredtherelation-
the major differences in AGB, W and mortality across the
ships between these terms using generalized least squares P
basin. The forest properties were mapped onto a region
regression.WetestedwhetherW andeitherW orlweresig-
P L defined as Amazonia sensu stricto (Eva et al., 2005) which is
nificantlyrelatedtoAGBandwhethertheserelationshipsdif-
divided into 1° by 1° longitude–latitude grid cells (Fig. 1c).
fered among the four regions. We accounted for spatial
Model output was provided for the same grid. The kriged
autocorrelation by specifying a Gaussian spatial correlation
mapswerecreatedusingordinarykrigingwiththegstatpack-
structure,whichisconsistentwiththeshapeofthesemivari-
ageinR(Pebesma,2004).Toassessthepredictiveabilityofthe
ograms for these forest properties across the plot network
kriging method, we performed a leave-one-out cross-valida-
(Fig.S2).Stemmortalityratesandabsoluteratesofwoodybio-
tiontechnique.Thisinvolvesleavingonesiteoutinturnand
masslosswerelog-transformedpriortoanalysistoensurethe
performingthekrigingusingtherestoftheobservations.The
residuals were normally distributed. Model evaluation was
kriging prediction for this location was then compared with
performedonthebasisofAkaikeinformationcriterion(AIC)
theobservation.Resultsfromthecross-validationdemonstrate
values.AnalyseswerecarriedoutusingthenlmepackageinR
thattherewasnospatialbiasinthekrigingmethod(Fig.S4).
(RDevelopmentCoreTeam,2012;Pinheiroetal.,2015).
Therewasalsonotendencyforthekrigingtooverestimateor
underestimatevaluesforthewholebasin.However,thekrig-
Modelsimulationsandcomparisonwithobservations
ing method was not able to capture the few locations with
We tested how well a range of DGVMs perform for Amazo- veryhighmortalityvalues(Fig.S5).Thisproblemiscommon
nia by comparing observed AGB, W and s to the output to any interpolation method which is effectively averaging
P w
©2016TheAuthors.GlobalChangeBiologyPublishedbyJohnWiley&SonsLtd.,doi:10.1111/gcb.13315
STEM MORTALITY CONTROLS FOREST BIOMASS 7
observed values. The median percentage bias between the regionsisassociatedwithdifferentpatternsinW .Inthe
P
leave-one-out cross-validation and the measured plot values western Amazon, the lower biomass values are associ-
was13.6%,12.7%and23.0%forAGB,WPandstemmortality ated with high WP (Fig. 2a–c). In contrast, the particu-
raterespectively. larly low biomass forests of the Brazilian Shield have
Wedonotintendthekrigedmapstobeadetailed,accurate highratesofstemmortalityandlowW (Fig. 2a–c).
P
description of Amazon forest properties: ecological patterns
Analysis of the relationships using generalized
are a mix of smooth gradients (e.g. related to climate) and
least squares allows the relative importance of W
more abrupt boundaries (e.g. related to edaphic properties) P
and l for determining AGB to be explored in more
thatcannotbeshownusingthesemethods.Rather,weintend
detail. Stem mortality rate is the key parameter that
thesemapsasbroadscaletoolstoprovideameansofevaluat-
ingtheperformanceofthevegetationmodels. controls variation in AGB (Table 2, Fig. 4c). This rela-
Finally, we compared how well the DGVMs captured the tionship between AGB and stem mortality rates is
mean and variability in AGB, W and s (calculated using not because there is a correlation between AGB and
P w
averagevaluesforWPandAGBacrossallgridcellsfor2000– stem number, as these two variables are unrelated
2008 from model outputs using Eqn 1) for grid cells where (Fig. S6). In contrast, the alternative measure of mor-
thereisobservationaldata,andcontrastthecontrolsonAGB tality, W , is not related to AGB (Fig. 4b): all models
L
betweenobservationsandmodelsintermsofW andmortal-
P including stem mortality rates, rather than W , show
L
ity. We acknowledge that the models will predict a small
substantially better fit and lower AIC values
increaseinW overthetimeperiodofstudyduetoCO fertil-
ization(~0.35PMgCha(cid:1)1 a(cid:1)1;Lewiset al.,2009).Howev2er,the (Table 2).
effectofthisprocessonestimatesofs issmall. The effect of stem mortality rate on AGB also differs
w
amongregions(Fig.4c).Forexample,forastemmortal-
ityrateof1.5%yr(cid:1)1,forestsontheGuianaShieldstore
Results
approximately 75% more carbon as (above-ground)
wood than forests on the Brazilian Shield (Fig. 4c). In
Observedlinksbetweenwoodybiomass,mortalityand
addition,thestrengthoftherelationshipbetweenAGB
productivity
and stem mortality rates varies among regions: the
There is a strong variation in AGB (F = 72.1, slope of this relationship is comparatively shallow
3,163
P < 0.001), l (F = 23.6, P < 0.001) and W amongtheplotsinwesternAmazonia(Fig.4c).Finally,
3,163 P
(F = 22.7,P < 0.001)amongthefourregions,butnot W is significantly positively correlated with variation
3,163 P
W (F = 1.49,ns;Table 1,Fig. 2).ForestsontheGui- in AGB, although the relationship is weak (Table 2,
L 3,163
ana Shield are characterized by the highest AGB of all Fig.4a).
Amazonianforests,associatedwithlowstemmortality
ratesandhighW (Fig. 2a–c).East-centralAmazonfor-
P Modelprojectionsandcomparisonwithobservations
estsalsohavecomparativelyhighAGBandsimilar,very
lowstemmortalityrates.However,W islowerinthese The comparisons of simulated AGB and above-ground
P
sites (Fig. 2b). Compared with these regions, forests in W reveal considerable differences both between the
P
the western Amazon and on the Brazilian Shield have individual models and between the models and obser-
lower AGB. However, the lower biomass in these two vations(Table 3,Figs5,6,S7andS8).Forthewholeof
Table1 Observedforestproperties(mean (cid:4) SE)calculatedfromplotdataforeachregionofAmazonia
East-central Western
Basin GuianaShield Amazon Amazon BrazilianShield
Meanabove-groundbiomass(MgCha(cid:1)1) 153.48(cid:4) 2.82 211.91(cid:4)5.03 167.64(cid:4) 4.95 126.26(cid:4) 2.38 107.73(cid:4) 4.48
n= 413 n= 110 n = 78 n =149 n= 76
Meanabove-groundwoodyproductivity 2.97(cid:4) 0.06 3.51(cid:4)0.13 2.41(cid:4) 0.07 3.06(cid:4) 0.07 2.40(cid:4) 0.15
(MgCha(cid:1)1yr(cid:1)1) n= 167 n= 41 n = 37 n =76 n= 13
Stem-basedmortalityrate(%yr(cid:1)1) 1.96(cid:4) 0.08 1.66(cid:4)0.16 1.38(cid:4) 0.08 2.62(cid:4) 0.12 3.19(cid:4) 0.38
n= 167 n= 41 n = 37 n =76 n= 13
Meanabove-groundbiomasslosses 2.46(cid:4) 0.13 3.06(cid:4)0.44 2.12(cid:4) 0.16 2.43(cid:4) 0.15 1.57(cid:4) 0.12
(MgCha(cid:1)1yr(cid:1)1) n= 167 n= 41 n = 37 n =76 n= 13
Meanwooddensity(gcm(cid:1)3) 0.63(cid:4) 0.00 0.69(cid:4)0.00 0.67(cid:4) 0.01 0.58(cid:4) 0.00 0.61(cid:4) 0.01
n= 413 n= 110 n = 78 n =149 n= 76
Basalarea(m2 ha(cid:1)1) 26.64(cid:4) 5.53 29.10(cid:4)0.49 28.24(cid:4) 0.51 25.98(cid:4) 0.41 22.73(cid:4) 0.66
n= 413 n= 110 n = 78 n =149 n= 76
©2016TheAuthors.GlobalChangeBiologyPublishedbyJohnWiley&SonsLtd.,doi:10.1111/gcb.13315
8 M. O. JOHNSON etal.
Fig.2 Boxplotsofplotmeasurementsof(a)above-groundbiomass,(b)above-groundwoodyproductivity,(c)stemmortalityratesand
(d)absoluteratesofwoodybiomasslossinfourregionsofAmazonia.GuShld=GuianaShield,ECAmaz=EastCentralAmazon,
WAmaz=WesternAmazon,BShld=BrazilianShield.
Table2 GeneralizedleastsquaresmodelsrelatingAGBtovariationin(A)above-groundwoodyproductivity(W ),stemmortality
P
rates(l)orratesofwoodybiomassloss(W );(B)landW ;(C)W andW among167plotsacrossfourregionsofAmazonia.Mod-
L P L P
els incorporated region as an additional factor and interactions as appropriate. Terms for mortality were log-transformed before
analysis.Allmodelsincorporated aGaussianspatialerrorcorrelation structuretoaccountforspatial autocorrelation. Themodel
withthestrongestsupportishighlightedinbold;thismodelwasusedtoquantifytherelationshipsinFig. 3
Model Terms Interactions Loglikelihood AIC Pseudorsquared
A.Includingeithermortalityorgrowth
1 l,Region (cid:1)813.7 1643.3 0.65
2 W ,Region (cid:1)830.1 1676.3 0.57
L
3 W ,Region (cid:1)829.3 1674.5 0.58
P
B.IncludingW andlasmortalityterm
P
4 W ,l,Region (cid:1)810.8 1639.6 0.66
P
5 W ,l,Region l9Region (cid:1)805.0 1634.0 0.68
P
6 W ,l,Region W 9 Region (cid:1)808.8 1641.6 0.67
P P
C.IncludingW andW asmortalityterm
P L
7 W ,W ,Region (cid:1)829.0 1676.1 0.58
P L
8 W ,W ,Region W 9Region (cid:1)826.7 1677.4 0.59
P L L
9 W ,W ,Region W 9 Region (cid:1)826.6 1677.2 0.59
P L P
AGB,above-groundbiomass.
the Amazon basin, mean AGB is highest for ORCHI- underestimate mean AGB (Table 3). However, the
DEE,andlowestforINLAND;incontrast,woodyNPP modelensemblemeanAGBvalue(163.87MgCha(cid:1)1)is
is highest for LPJmL and lowest for JULES (Table 3). closetotheobservedmean(153.48MgCha(cid:1)1).Incon-
Compared with the plots, different models over- and trast, all models overestimate above-ground W
P
©2016TheAuthors.GlobalChangeBiologyPublishedbyJohnWiley&SonsLtd.,doi:10.1111/gcb.13315
STEM MORTALITY CONTROLS FOREST BIOMASS 9
Table3 Basin mean values, standard errors and root mean square error (RMSE) for above-ground wood biomass (AGB; Mg C
ha(cid:1)1)andabove-groundwoodynetprimaryproductivity(woodyNPP;MgCha(cid:1)1yr(cid:1)1)fromtheplotobservationsandmeanval-
uesfromfourDGVMsfortheplotlocations.Abelow-groundtoabove-groundallocationratioof0.21isappliedtotheDGVMval-
uestoconvertfromtotalNPPwoodtoabove-groundwoodyNPP
AGB(Obsmean = 153.48) W (Obsmean = 2.97)
P
AGBwood AGNPPwood
Model ORCHIDEE JULES INLAND LPJmL ORCHIDEE JULES INLAND LPJmL
Modelmean 218.00(cid:4) 3.16 137.93(cid:4) 2.09 125.43(cid:4) 1.35 174.10(cid:4) 2.89 7.80(cid:4) 0.10 4.05(cid:4) 0.09 7.46(cid:4) 0.11 9.92(cid:4) 0.10
RMSE 91.84 76.98 61.36 73.65 5.00 1.89 4.73 7.06
NPP,netprimaryproductivity;DGVMs,dynamicglobalvegetationmodels.
comparedwiththemeanfortheplots,bybetween36% betweenwoodyNPPandAGBfromtheplotdatacom-
(JULES)and234%(LPJmL;Table 3,Fig.5).Variationin parativelywell(Fig.7).
s inevitably reflects the variation in mean AGB and Simulated AGB and W from all four models show
w P
woody NPP with average values for ORCHIDEE and strong relationships with climatological drivers. Corre-
JULES (27.9 and 33.2 years) approximately twice the lations between W and precipitation are particularly
P
valuesforINLANDandLPJmL(16.7and17.5 years). strong for INLAND and LPJmL and all models apart
Thereareconsiderabledifferencesbetweentheobser- fromJULESexhibitstrongcorrelationsbetweenrainfall
vations and the predictions across the four models in and AGB (Fig. S9). Weaker correlations are observed
thespatialvariabilityofAGBandW (Figs5,6andS7). between temperature and short-wave radiation and
P
JULES and INLAND both simulate very little spatial simulatedW andAGB(Fig.S10).
P
heterogeneityinAGBintheAmazonbasin,incontrast
to the strong pattern in the observations: compared
Discussion
with the observations, they simulate a very narrow
range of AGB values and underestimate both the AGB
UnderstandingspatialvariationintheAGBofAmazon
oftheGuianaShieldandthebasinasawhole(Table 3,
forests
Fig.5c, e).LPJmLandORCHIDEEdisplaygreatervari-
abilityintheirpredictionsofAGB(Fig.5g, i).However, Overall,ourresultsextendandenrichtheoriginalpara-
LPJmL predicts highest AGB in the north-west of the digm concerning the controls on forest dynamics in
basin in contrast to the observations (Fig. 5i). ORCHI-
DEE is the only model that provides a reasonable
matchwiththespatialpatternsintheobservations,but
thismodelstilloverestimatesAGBformostofthebasin
comparedwiththeplotobservations(Table 3,Fig.5g).
IntermsofW ,LPJmL(Fig.5j)istheonlymodelthat
P
captures the higher observed values in the Guiana
ShieldandWesternAmazoncomparedwiththeBrazil-
ian Shield and East-central Amazon (Fig. 5b). In con-
trast, INLAND, ORCHIDEE and JULES simulate very
little variability in W across the majority of basin
P
(Fig.5d, f, h).
Forallmodels,thespatialvariationins issimilarto
w
that of AGB (Fig. 6). LPJmL demonstrates the greatest
spatialvariationinresidencetimeswiththehighestval-
uesfoundinthenorth-westofthebasin(Fig.6).JULES
and INLAND display little variation in s across the
w
basin. Overall, JULES, LPJmL and INLAND display a
much stronger positive relationship between woody
NPP and AGB (Fig. 7) than seen in the observations
(Fig. 4a), although the form of this relationship varies. Fig.3 Relationship between woody net primary productivity
In contrast, the relationship predicted by ORCHIDEE (NPP) and stem mortality rates for 167 forest plots in four
matches the variability and form of the relationship regionsofAmazonia.
©2016TheAuthors.GlobalChangeBiologyPublishedbyJohnWiley&SonsLtd.,doi:10.1111/gcb.13315
10 M. O. JOHNSON etal.
Amazonia. The previous paradigm described corre- Incontrasttothesepatternsforabsoluteratesofloss
latedwesttoeastgradientsinW ,stemmortalityrates ofbiomass,therearestrongrelationshipsbetweenstem
P
and AGB across the Amazon basin, maintained by a mortality rates and AGB (Fig. 4c). This result suggests
soil-mediated, positive feedback mechanism (Malhi that variation in the numbers and diameters of trees
et al., 2004; Quesada et al., 2012). Our findings agree that die in different locations is a key control on AGB:
thatvariationinmortalityisthekeydriverofvariation high rates of stand-level biomass loss and W can be
P
in AGB across Amazonian forests (Table 2, Fig. 4). associated with high AGB if stem mortality rates are
However, our results modify the current paradigm low, and biomass loss is concentrated in a few large
aboutvariationinforestdynamicsinAmazoniainfour trees, but can also be associated with comparatively
importantways. lowAGBifstemmortalityratesarehigh,andmortality
Firstly,theplotdatademonstratethatthereisnocor- is concentrated in a larger number of smaller trees
relation between W (above-ground woody productiv- (Fig. 4). Stem mortality rates may influence AGB
P
ity) and stem mortality rates with the new, broader because they affect the size structure of forests: demo-
data set: they vary independently (Fig. 3). Previous graphic theory demonstrates how higher stem mortal-
studies have strongly focused on western Amazonia ityratesareassociatedwithasteeperslopeoftreesize/
and some East-central Amazon sites. However, the frequency distributions and therefore fewer large trees
inclusion of data from the Guiana Shield in particular (Coomes et al., 2003; Muller-Landau et al., 2006). In
demonstrates that low stem mortality rates can also be turn,variationinthenumberoflargetreesisakeypre-
associatedwithhighW (Fig. 3). dictorofspatialvariationinbiomassamongforestplots
P
Secondly, our results demonstrate that variation in (e.g. Baker et al., 2004; Baraloto et al., 2011). Impor-
stem mortality rates, rather than absolute rates of car- tantly, this result indicates that incorporating stem
bon loss, is the key aspect of mortality that determines diameter distributions within modelling frameworks
variationinAGB.ThelackofcorrelationbetweenAGB will be important for obtaining accurate predictions of
andabsoluteratesofbiomassloss(Fig. 4b)issomewhat AGB.
surprising: for a forest stand at approximately steady Thirdly, our results resolvea paradox in the original
state, we might expect this relationship to at least mir- paradigm–thatW showedanegativecorrelationwith
P
ror the weak correlation between AGB and stand W AGB(Malhi,2012).Here,withabroaderrangeofsites,
P
(Fig. 4a).Thisresultmaybebecauseestimatesofabso- theexpectedpositivecorrelationisfound,althoughthe
luteAGBlossaresubjecttogreatersamplingerrorthan strengthoftherelationshipremainsweak(Fig. 4a).Pos-
W duetostochasticvariationintreemortality(e.g.see itivecorrelationsbetweenAGBandW areafeatureof
P P
wide variationin values on the x axis ofFig. 4b).Sam- theoutputofDGVMs(e.g.Fig.7).Thisanalysis,atleast
pling over longer time intervals may reveal stronger to an extent, demonstrates consistency between one
correlationsbetweenabsoluteratesofbiomasslossand aspect of the models and the data, although the
AGB.
Fig.4 RelationshipsbetweenAGBand(a)woodyNPP,(b)absoluteratesofwoodybiomasslossand(c)stemmortalityratesfor167
forestplotsinfourregionsofAmazonia.LinesrelatetosignificantrelationshipsasgivenbyfinalstatisticalmodelinTable3.NPP,net
primaryproductivity;AGB,above-groundbiomass.
©2016TheAuthors.GlobalChangeBiologyPublishedbyJohnWiley&SonsLtd.,doi:10.1111/gcb.13315