Impacts of trait variation through observed trait–climate relationships

Biogeosciences 10 5497ndash5515 2013wwwbiogeosciencesnet1054972013doi105194bg-10-5497-2013copy Author(s) 2013 CC Attribution 30 License

EGU Journal Logos (RGB)

Advances in Geosciences

Open A

ccess

Natural Hazards and Earth System

Sciences

Open A

ccess

Annales Geophysicae

Open A

ccess

Nonlinear Processes in Geophysics

Open A

ccess

Atmospheric Chemistry

and Physics

Open A

ccess

Atmospheric Chemistry

and Physics

Open A

ccess

Discussions

Atmospheric Measurement

Techniques

Open A

ccess

Atmospheric Measurement

Techniques

Open A

ccess

Discussions

Biogeosciences

Open A

ccess

Open A

ccess

BiogeosciencesDiscussions

Climate of the Past

Open A

ccess

Open A

ccess

Climate of the Past

Discussions

Earth System Dynamics

Open A

ccess

Open A

ccess

Earth System Dynamics

Discussions

GeoscientificInstrumentation

Methods andData Systems

Open A

ccess

GeoscientificInstrumentation

Methods andData Systems

Open A

ccess

Discussions

GeoscientificModel Development

Open A

ccess

Open A

ccess

GeoscientificModel Development

Discussions

Hydrology and Earth System

Sciences

Open A

ccess

Hydrology and Earth System

Sciences

Open A

ccess

Discussions

Ocean Science

Open A

ccess

Open A

ccess

Ocean ScienceDiscussions

Solid Earth

Open A

ccess

Open A

ccess

Solid EarthDiscussions

The Cryosphere

Open A

ccess

Open A

ccess

The CryosphereDiscussions

Natural Hazards and Earth System

SciencesO

pen Access

Discussions

Impacts of trait variation through observed traitndashclimaterelationships on performance of an Earth system model aconceptual analysis

L M Verheijen 1 V Brovkin 2 R Aerts1 G Bonisch3 J H C Cornelissen1 J Kattge3 P B Reich45 I J Wright 6and P M van Bodegom1

1VU University Amsterdam Systems Ecology Department of Ecological Science De Boelelaan 1085 1081 HV Amsterdamthe Netherlands2Max Planck Institute for Meteorology Bundesstrasse 55 20146 Hamburg Germany3Max Planck Institute for Biogeochemistry Hans Knoell Strasse 10 07745 Jena Germany4University of Minnesota Department of Forest Resources 1530 Cleveland Avenue North St Paul MN 55108 USA5University of Western Sydney Hawkesbury Institute for the Environment Penrith NSW 2753 Australia6Macquarie University Department of Biological Sciences Sydney NSW 2109 Australia

Correspondence toL M Verheijen (lmverheijenvunl)

Received 12 October 2012 ndash Published in Biogeosciences Discuss 19 December 2012Revised 12 June 2013 ndash Accepted 26 June 2013 ndash Published 15 August 2013

Abstract In many current dynamic global vegetation mod-els (DGVMs) including those incorporated into Earth sys-tem models (ESMs) terrestrial vegetation is represented bya small number of plant functional types (PFTs) each withfixed properties irrespective of their predicted occurrenceThis contrasts with natural vegetation in which many planttraits vary systematically along geographic and environmen-tal gradients In the JSBACH DGVM which is part of theMPI-ESM we allowed three traits (specific leaf area (SLA)maximum carboxylation rate at 25C (Vcmax25) and maxi-mum electron transport rate at 25C (Jmax25)) to vary withinPFTs via traitndashclimate relationships based on a large traitdatabase TheR2

adjusted of these relationships were up to083 and 071 for Vcmax25 and Jmax25 respectively ForSLA more variance remained unexplained with a maximumR2

adjustedof 040 Compared to the default simulation allow-ing trait variation within PFTs resulted in gross primary pro-ductivity differences of up to 50 in the tropics ingt 35 different dominant vegetation cover and a closer match witha natural vegetation map The discrepancy between defaulttrait values and natural trait variation combined with thesubstantial changes in simulated vegetation properties to-gether emphasize that incorporating climate-driven trait vari-ation calibrated on observational data and based on ecolog-

ical concepts allows more variation in vegetation responsesin DGVMs and as such is likely to enable more reliable pro-jections in unknown climates

1 Introduction

Terrestrial vegetation plays a pivotal role in landndashatmosphereinteractions modifying carbon water and heat fluxes via bio-chemical processes such as photosynthesis and respiration orvia biophysical vegetation properties such as stomatal con-ductance and albedo Therefore a correct representation ofterrestrial vegetation and its dynamics in Earth system mod-els (ESMs) is essential especially for future climate projec-tions In these models vegetation can be simulated by dy-namic global vegetation models (DGVMs) which integratevegetation dynamics with land surface models They are de-veloped to predict vegetation distribution and fluxes of en-ergy water and carbon under past current and future climateregimes As such they allow analysis of transient vegetationresponses and feedbacks to climate (Foley et al 1998 Pren-tice et al 2007)

Compared to earlier models vegetation dynamics and in-teractions between the biosphere and atmosphere have been

Published by Copernicus Publications on behalf of the European Geosciences Union

5498 L M Verheijen et al Trait variation impacts on global vegetation model performance

much improved over the past decade although many issuesstill need to be resolved (Quillet et al 2010) One of theseissues is the way in which plant functional types (PFTs)are used to represent vegetation PFTs are classes of plantspecies with presumably similar roles in an ecosystem re-sponding in a comparable manner to environmental con-ditions like water and nutrient availability (Lavorel et al1997) They are defined by a combination of attributes suchas plant growth form (eg trees shrubs grasses herbs) phe-nology (evergreen raingreen summergreen) and bioclimatictolerances (eg minimum temperature requirements) No-tably most current PFT classifications use constant param-eter values for some key plant traits ie plant properties thatreflect the way plants cope with their environment (Violleet al 2007) Using constant plant traits in PFTs has seri-ous limitations (Ordonez et al 2009 Van Bodegom et al2012) as such constant values contrast with observed traitvariation (Ackerly and Cornwell 2007 Freschet et al 2011Westoby et al 2002) therefore not accounting for local en-vironmental constraints Furthermore global trait databaseanalyses have shown that the variation in plant traits is largewithin PFTs and often even greater than the difference inmeans among PFTs (Laughlin et al 2010 Wright et al2005a) Even though PFTs may capture a significant part ofthe global plant trait variation a large part (for some traits upto 75 ) may still be unexplained (Kattge et al 2011) and isthus not represented in current DGVMs

Given that plants can adjust to the environment viachanges in traits and that such changes influence ecosystemfunctioning (Diaz et al 2004 Lavorel and Garnier 2002)implementing these trait-driven interactions within PFTs ofDGVMs is highly relevant By simulating variation in planttrait responses it is better possible to quantify plant adap-tation to climate and to account for plantndashatmosphere feed-backs in DGVMs Indeed the limitations of inflexible PFTsis increasingly acknowledged by modelers and several at-tempts to allow for more (trait) variation within PFTs canbe found in for example JeDi-DGVM (Pavlick et al 2012)or aDGVM (Scheiter et al 2013) However none of theapproaches so far tried to maximally include trait varia-tion based on observational trait data and capture multiplesources of this variation by relating trait data to environmen-tal variables This inclusion may be possible by making useof the relationship between traits and multiple environmen-tal drivers as occurring at both regional and global scalesfor instance in relation to temperature water and nutrientavailability or disturbances (Ordonez et al 2009 Van Om-men Kloeke et al 2012 Wright et al 2005b) Such rela-tionships between environmental conditions and traits canpotentially be understood via ecological assembly theorywhich describes the processes that determine species assem-blages (Cornwell and Ackerly 2009 Cornwell et al 2006Gotzenberger et al 2012) An important abiotic assemblyprocess is habitat filtering (Keddy 1992) which describeshow local environmental drivers (eg soil fertility or pre-

cipitation) constrain the range of potential species and re-lated trait values in a given habitat For many traits such asspecific leaf area leaf nitrogen and wood density this con-tributes to trait convergence within communities prevailingover trait divergence (Freschet et al 2011 Swenson and En-quist 2007) and resulting in global relationships betweencommunity trait means and climatic drivers Community traitmeans match closely to the model resolution of DGVMsgiven that the majority of the models provide a mean valueof a trait for a PFT (in a given grid cell) By identifying theenvironmental drivers of variation in community trait meansmultiple causes of variation are determined and the observednatural trait variation can be modeled with a high level ofaccuracy

The aim of this study is thus to model climate-driventrait variation within PFTs determined as comprehensivelyas possible from observations and to identify the impactsthrough vegetation responses and vegetationndashatmospherefeedbacks on DGVM model behavior So far observationaltrait data have been used to derive mean parameter values anduncertainties for PFTs in the context of different vegetationmodels (Kattge et al 2009 Ziehn et al 2011) but this studyprovides a proof of concept to predict mean parameter valuesper PFT for each individual grid cell based on a systematicanalysis of observed traitndashclimate relationships using a largeobservational trait database

The traitndashclimate relationships were implemented in theJSBACH DGVM which is part of the Max Planck InstituteEarth System Model (MPI-ESM) (Brovkin et al 2009 Rad-datz et al 2007 Roeckner et al 2003) and is used in thecontext of ESMs and carbon cycle model intercomparisons(Friedlingstein et al 2006) We simulate variation in threeoriginally constant and PFT-specific key leaf traits in JS-BACH for which sufficient georeferenced observational datacould be selected to calibrate the traits These traits were spe-cific leaf area (SLA) maximum carboxylation rate at a refer-ence temperature of 25C (Vcmax25) and maximum electrontransport rate at 25C (Jmax25) To determine relationshipsbetween these three traits and climatic drivers for each PFTcommunity means of trait values from the TRY global planttrait database (Kattge et al 2011) were related to global cli-matic data Based on these relationships between traits andclimate SLA Vcmax25 and Jmax25 were reparameterizedfor each grid cell on a yearly basis depending on the localclimatic conditions in a grid cell Compared to the defaultmodel this enabled enhanced feedbacks between plants andenvironment as traits within natural PFTs could vary dynam-ically in space and time A simulation with variable traits iscompared for trait distribution productivity and vegetationdistribution to a default simulation with the original trait val-ues of the model and to an additional simulation with con-stant but observation-based trait values

As DGVMs are already parameterized to produce ap-proximate realistic results our variable traits simulationwill not necessarily approach reality better than the default

Biogeosciences 10 5497ndash5515 2013 wwwbiogeosciencesnet1054972013

L M Verheijen et al Trait variation impacts on global vegetation model performance 5499

simulation Therefore in this paper the focus lies on sim-ulation intercomparisons to evaluate the impact of incorpo-rating climate-driven trait variation The simulations are notmeant to represent current vegetation or climate states butwere run into equilibrium states with their own internal cli-mate and vegetation composition in order to study the im-pacts of trait variation within PFTs However we includecomparisons with global biomass and gross primary produc-tivity (GPP) estimates and potential vegetation distribution toevaluate the realism of the various simulation results

2 Materials and methods

21 Model description

Simulations were performed with the ESM of the MaxPlanck Institute (MPI-ESM) The model setup consisted ofthe JSBACH DGVM a land surface model (Raddatz et al2007) with a vegetation dynamics module (Brovkin et al2009) coupled to the atmosphere model ECHAM5 (Roeck-ner et al 2003) to allow for full inclusion of vegetationndashatmosphere feedbacks The model version on spatial resolu-tion of T63 (approx 1875 dividing the world into 18 432grid cells) with 47 atmospheric layers was used with atmo-spheric CO2 concentration kept constant at 3539 ppm Sea-sonal sea surface temperatures and sea ice were prescribedfrom a default simulation with the MPI ocean model (MPI-OM) (Marsland et al 2003) Vegetation dynamics wereinteractive to allow for vegetation shifts and vegetationndashatmosphere feedbacks through trait dynamics Terrestrialgrid cells contained multiple PFTs each occupying a cer-tain fraction depending on their competitive ability (basedon net primary productivity ndash NPP) This study does not in-clude crops and pastures but focuses on responses of naturalvegetation represented by eight PFTs These were tropicalbroadleaved evergreen trees (TrET) tropical broadleaved de-ciduous trees (TrDT) extratropical (both temperate and bo-real) evergreen trees (ExTrET) extratropical deciduous trees(ExTrDT) raingreen shrubs (RgSh) colddeciduous shrubs(DSh) C3 grasses (C3G) and C4 grasses (C4G) No anthro-pogenic land use or land cover change was simulated

22 Selected trait and climate data

All PFT-dependent trait parameters in the model for whichsufficient georeferenced observational trait data could becollected were selected This resulted in selection of 3traits SLA (m2 kgminus1 carbon) Vcmax25 (micromol mminus2 sminus1) andJmax25 (micromol mminus2 sminus1) SLA is used to determine theamount of carbon that can be stored in the green and reservecarbon pools In contrast to most DGVMs SLA is decou-pled from the phenology routine in JSBACH and thereforedoes not play a role in determining productivity Vcmax25(micromol mminus2 sminus1) and Jmax25 (micromol mminus2 sminus1) are the refer-ence values at 25C used for calculation of the Vcmax and

Jmax values at the actual temperature These traits are partof the photosynthesis routine in JSBACH which is modeledfollowing Farquhar et al (1980) for C3 plants and Collatz etal (1992) for C4 plants (see Supplement S1 for a more de-tailed description with equations of the functioning of traitsin JSBACH) Vcmax25 also determines the reference darkrespiration at 25C (Rd25) Leaf carbon assimilation is thelowest of carboxylation and electron transport rate (calcu-lated from actual Vcmax and Jmax) minus dark respirationand scaled to canopy level with the use of the leaf area index(LAI) Supplement S2 provides a description and a flowchartwith the role of the selected traits in JSBACH processes andthe way variation in these traits might propagate in the model

Only georeferenced trait observations from field samplingand field experiments were used and selected from the TRYdatabase (Kattge et al 2011) with additional data for SLAfrom the database by Van Bodegom et al (2012) and forVcmax25 and Jmax25 from Domingues et al (2010) (seeTable 1 for all references) From these databases only Vc-max and Jmax data that could be standardized to a refer-ence temperature of 25C were used Most Vcmax and Jmaxvalues had already been standardized to this temperature viathe formulation of the photosynthesis model by Farquhar etal (1980) used in the context of JSBACH (Kattge and Knorr2007) For other records for which the temperature duringmeasurement was recorded standardization was done ac-cording to this formulation For C4 grasses variation in PEP-case CO2 specificity (PEP mmol mminus2 sminus1) instead of Jmax25was determined following Collatz et al (1992) Since forC4 grasses insufficient observational PEP and Vcmax25 datawere available these traits were estimated more indirectlyapplying equations from Simioni et al (2004) who deter-mined PEP and Vcmax25 based on leaf nitrogen (N) content(g mminus2) (regressions based on two C4 species) Thereforefor C4 grasses additional information on leaf N was obtainedfrom the TRY database

Each trait observation was linked to a PFT based on in-formation on growth form (shrub grass tree) leaf habit (de-ciduousevergreen) and photosynthetic pathway (C3C4) forthe species involved The climatic domain of a species (tropi-cal boreal etc) was determined based on the KoppenndashGeigerclimate classification (Kottek et al 2006) and applied to thegeoreferenced observations

We used PFT-specific community means of traits to matchthe scale of modeled fluxes most closely as JSBACH pro-vides mean values per PFT (in a given grid cell) Commu-nity means were weighted by the square root of the numberof observations (data points) per location This resulted in12 394 observations for SLA distributed over 2869 speciesand 1052 (PFT-specific) communities 761 observations over129 species and 70 communities for Vcmax25 and 402 ob-servations over 108 species and 56 communities for Jmax25(see Supplement S3 for a global map with locations of traitdata)

wwwbiogeosciencesnet1054972013 Biogeosciences 10 5497ndash5515 2013


Table 1References for trait data used in this study

Reference SLA Vcmax25 Jmax25

Ackerly and Cornwell (2007) xBahn et al (1999) xCavender-Bares et al (2006) xCornelissen et al (2003) xCornelissen et al (2004) xCornwell et al (2006) xDomingues et al (2010) x xFyllas et al (2009) xGarnier et al (2007) xKattge et al (2009) x x xKattge et al (2011) xKleyer et al (2008) xKurokawa and Nakashizuka (2008) xKurokawa (unpublished data) xLaughlin et al (2010) xLeishman (unpublished data) xLouault et al (2005) xMedlyn et al (1999) xNiinemets (1999) x xNiinemets (2001) x xOgaya and Penuelas (2007) xOgaya and Penuelas (2008) xOrdonez et al (2010) xPatino et al (2012) xPyankov et al (1999) xReich et al (2008) xReich et al (2009) xShipley (1995) xShipley and Vu (2002) xSoudzilovskaia (unpublished data) xSwaine (2007) xvan Bodegom et al (2012) xvan Bodegom (unpublished data) xVile et al (2006) xWeiher (unpublished data) xWohlfahrt et al (1999) xWright et al (2004) xWright et al (2006) x

23 Traitndashclimate relationships

The observed natural variation of SLA Vcmax25 and Jmax25was described for each PFT by the simplest models pos-sible using linear regressions between individual traits andmultiple observational climatic variables The derived traitndashclimate relationships likely reflect habitat filtering whichcommonly leads to trait convergence at community scalesThe relationships were implemented in JSBACH to predicttraits based on the internally simulated climate As a con-sequence in our choice of climate variables we were con-strained by the environmental predictors that are modeled inJSBACH

Global climate data on mean annual precipitation (MAPmm yrminus1) mean annual relative humidity (Reh ) mean an-nual temperature (MATC) and mean temperature of cold-est and warmest month (Tmin andTmax C) were collected

from a global 10 min gridded dataset of mean monthly cli-mate data based on weather stations from the Climatic Re-search Unit (CRU) (New et al 2002) Mean annual net short-wave radiation (NSWR W mminus2) was calculated based ondistance to sun and percentage sunshine (from CRU) accord-ing to Allen et al (1998) on a global 30 min spatial resolu-tion Mean annual soil moisture was taken from GLEAM amethodology that estimates soil moisture and evapotranspi-ration based on remotely sensed data at a global 15 min spa-tial resolution (SoilMoist m3 mminus3 for 1 m depth Miralles etal 2011) Given the available modeled climatic drivers inJSBACH soil moisture was the best available estimate of adrought index Soil moisture was the only edaphic control oftrait variation since soil N was not modeled in this versionof JSBACH

To determine traitndashclimate relationships we tested all pos-sible climate combinations in relation to each trait Regres-sions with the highest explained variance (highestR2

adjusted)were selected after checking for significance of climaticdrivers and distribution of residuals Table 2 shows for eachPFT the selected climatic drivers of the best model as wellas their directional relationship with the traits (third column)and the correspondingR2

adjusted In case of co-linearity be-tween environmental drivers (Pearsonrsquos correlation higherthan 07 or lower thanminus07) the driver with lowest signif-icance was omitted Due to the low number of entries forVcmax25 and Jmax25 data for the two tropical tree PFTswere combined as well as data for the two shrub PFTs Eventhough we acknowledge that mean trait values for evergreenand deciduous vegetation types will differ in reality combin-ing these will give better estimates than modeling these traitson their few individual data points alone Due to differencesin other vegetation characteristics these PFTs will still dif-fer in their emergent properties as the results will show Intotal there were eight relationships for SLA and six for bothVcmax25 and Jmax25 These relationships were subsequentlyimplemented in JSBACH

In addition to allow trait calculations to go beyond cli-matic regions used in the regressions but to still maintaintraits within ranges as observed in nature we applied twoconstraints on simulated trait ranges First of all predictedtraits were constrained to the 25ndash975 quantiles of all indi-vidual observations within a PFT (instead of the communitymeans) Secondly for Jmax25 and Vcmax25 an additionalconstraint was applied to maintain the strong physiologicalcorrelation between them (Medlyn et al 2002 Wullschleger1993) keeping Jmax25 and Vcmax25 values within the 95 confidence interval of the linear regression among these twotraits (as determined per PFT) Predicted values outside thisrange were adjusted to the confidence interval border basedon the shortest distance needed to reach the border Sincethis version of JSBACH did not simulate N-cycling soil Ncould not be used to parameterize Vcmax25 even though N-availability is a strong determinant (see references in Kattgeet al 2009) However observational data was used to limit



Table 2Properties of the selected traitndashclimate relationships used to parameterize traits

No of entries Environmental drivers R2adjustedof Lower and upper

(observations)1 of traits2 relationships boundaries3

SLA (m2 kgminus1 carbon)

Tropical evergreen trees 69 (933) minusMAP minusNSWR (+int) 0092 741 4829Tropical deciduous trees 38 (129) minusTmin minusMAP +Reh 030 1207 4954Extratropical evergreen trees4 363 (308) +MAT +MAPminusNSWR 029 458 4246Extratropical deciduous trees4 177 (282) minusTmax minusReh (+int) 0056 624 9050Raingreen shrubs 178 (696) minusMAP minusNSWR 026 366 7269Colddeciduous shrubs 39 (111) +MAT 013 1782 6153C3 grasses 153 (309) +Tmax +Reh 012 308 9658C4 grasses 35 (101) minusNSWR 040 541 7219

Vcmax25 (micromol mminus2 sminus1)

Tropical evergreen trees amp 9 (85) +MAT +MAP (minusint) 083 180 677Tropical deciduous trees5

Extratropical evergreen trees 15 (284) +MATminusMAP +Reh 028 184 1799Extratropical deciduous trees 19 (273) minusRehminusSoilMoist (+int) 034 182 1300Raingreen shrubs amp 15 (75) minusMAP minusReh (+int) 045 251 1291Colddeciduous shrubs5

C3 grasses 4 (15) minusMAT 037 244 1184C4 grasses 8 (29) +NSWR 051 217 463

Jmax25 (micromol mminus2 sminus1) ndash for C4 grasses PEP (mmol mminus2 sminus1)


Extratropical evergreen trees 12 (113) +Tmin minusMAP 028 575 3296Extratropical deciduous trees6 10 (135) +Tmin +NSWR (minusint) 071044 411 2069

minusMAP minusSoilMoist (+int)Raingreen shrubs amp 13 (59) +MAPminusSoilMoist 064 654 2235Colddeciduous shrubs5


1 Number of entries available for the regression analysis in parentheses the number of observations for these entries2 Environmental drivers are mean annual temperature (MAT) mean temperature of coldest and warmest month (Tmin andTmax) mean annualprecipitation (MAP) mean annual relative humidity (Reh) soil moisture (SoilMoist) and net shortwave radiation (NWSR) + = positive relationshipminus = negative relationship and (int) = interaction effect between two drivers3 Lower and upper boundaries of the predicted traits in the variable traits simulation based on the 25 and 975 quantiles of all individual observationswithin PFTs4 Extratropical trees include both temperate and boreal trees5 For Vcmax25 and Jmax25 tropical trees have the same relationships with climatic drivers as they were combined due to low number of entries Thesame holds for shrubs6 As the best model covered a relatively small climatic range an additional regression model was applied to these areas that fell outside the climaticrange of the first model

the above two constraints on Vcmax25 in this way indirectlyincorporating N-limitations via observed Vcmax25 values

24 Simulation setups

Three different scenarios were performed (1) a simulationcontaining the default parameterization of JSBACH withconstant parameter values per PFT based on Raddatz etal (2007) and Kattge et al (2009) and adapted to approx-imate realistic vegetation functioning within the vegetationdynamics module (Brovkin et al 2009) hereafter called ldquode-

fault simulationrdquo (2) an ldquoobserved traits simulationrdquo againwith constant values for SLA Vcmax25 and Jmax25 per PFTbut based on observational data only (the weighted commu-nity means for each PFT from here on called ldquoobservedglobal meansrdquo) and (3) a ldquovariable traits simulationrdquo inwhich traits were allowed to vary depending on local climaticconditions In the latter simulation at the beginning of everyyear each of the three key traits was reparameterized for eachPFT in every terrestrial grid cell of the world depending onthe local simulated climatic conditions in each grid cell Theobserved traits simulation was performed in order to separate



the effects of replacing default trait constants by observation-based constants from the effects of adding trait variation inthe model As this simulation mainly serves as a control wewill focus our results and discussion on the variable traitssimulation

To get vegetation into quasi-equilibrium with the simu-lated climate (but with CO2 concentration fixed) the coupledmodel (JSBACHECHAM5) was run for 150 years with veg-etation dynamics in an accelerated mode (ie vegetation wassimulated 3 times for each year of simulated climate) Nextto get the slow soil carbon pools into equilibrium the uncou-pled carbon model of JSBACH (ldquoCBALANCErdquo) was run for1500 yr The coupled model was continued for 160 yr withupdated carbon pools (with the vegetation module in an ac-celerated mode for the first 50 yr) until carbon pools and veg-etation cover had reached equilibrium The last 10 yr wereaveraged and used for further analysis which in this setup(with prescribed seasonal sea surface temperatures) was suf-ficient to account for the interannual climate variation

25 Data comparison

Model performance was investigated by comparing vegeta-tion distribution GPP and biomass of the simulations withobservations Vegetation distribution of the dominant PFT(Fig 5) was compared to the potential (natural) vegetationmap of Ramankutty and Foley (1999) which in part is usedto initialize simulations in JSBACH The mismatch in spa-tial resolution (the potential vegetation maps is at a global30 min spatial resolution) was solved by counting the num-ber of PFTs according to Ramankutty and Foley (1999) thatwere present in each JSBACH grid cell The PFT with thehighest occurrence was compared with the simulated dom-inant PFT of JSBACH To match the PFT classificationsaggregated PFTs were constructed (Supplement Table S61and Fig S61) as the PFTs of Ramankutty and Foley (1999)could not be directly related to those of JSBACH Temper-ate and boreal evergreen trees were merged to match extrat-ropical evergreen trees in JSBACH The same was done fordeciduous trees Shrubs (marginal in JSBACH) were mergedinto a single PFT (ldquoshrubsrdquo) in both JSBACH and the veg-etation map As savannas consist of grasslands and wood-lands matches between savannas and C4 grasses or tropicalbroadleaved deciduous trees were both classified as correctTundra and mixed forests were omitted from the comparisonas no equivalent PFT was available in JSBACH Model per-formance was determined with Cohenrsquos kappa (κ) (Cohen1960) which is the proportion of agreement between veg-etation maps while accounting for chance agreement Gridcells with sea or ice as the dominant cover were not takeninto account and neither were cases in which more than1 PFT shared the highest occurrence (ie equal number ofgrid cells) This resulted in a comparison of 2819 grid cells

Latitudinal patterns of median GPP were compared withdata taken from Beer et al (2010) who combined observa-

tional data (eddy covariance fluxes) with diagnostic modelsto approximate GPP

Simulated biomass was compared to estimates fromRobinson (2007) which include adjusted estimates of be-lowground (root) carbon in plant biomass Estimates basedon rootshoot quotients from Mokany et al (2006) and Graceet al (2006) were used As we simulated only natural veg-etation and the estimates of Robinson (2007) are currentestimates including land use change and crops comparingglobal total biomass would result in an overestimation byour simulations Therefore comparisons were made per m2

per (aggregated) PFT JSBACH does not have a separatesavanna-like PFT Therefore tropical forests and C4 grass-lands were averaged and compared with averages of tropi-cal forests and savanna Furthermore extratropical trees wereaveraged and compared with averaged temperate and borealtrees of Robinson (2007) Mediterranean shrubs and tundrawere omitted as there were no comparable PFTs in JSBACHFor the tropics an additional comparison of biomass wasmade to Saatchi et al (2011)

3 Results

31 Mismatch between observed trait data and defaulttrait settings

Fig 1 shows the observed trait ranges as collected fromthe databases Grey diamonds reflect the PFT-specific de-fault trait values in JSBACH and black circles indicate thePFT-specific global means based on the observed commu-nity means These observed global means were used to re-place the default trait values in the observed traits simulationFor all three traits the default trait values deviated stronglyfrom the observed global means Differences were on av-erage 323 for Vcmax25 268 for Jmax25 and 173 for SLA (but for specific PFTs going up to 734 576 and 352 respectively) Such mismatch between observa-tional trait data and default trait values has also been reportedby Kattge et al (2011) who showed how PFT-specific con-stant SLA values in DGVMs often differ from the meansof observed data and in some cases are at the low end ofthe observed data distribution In JSBACH SLA default val-ues were always lower than the observed global means ex-cept for the tropical broadleaved deciduous trees but theyalmost all fell within the 25 quartiles of the observedrange of SLA except for deciduous shrubs However forboth Vmax25 and Jmax25 default values fell outside the 25 quartiles for more than half of the PFTs (five out of eightPFTs for both) and in some cases even outside the minimumand maximum values (tropical deciduous trees C4 grassesand C3 grasses for Vcmax25 and Jmax25 and additionallyextratropical evergreen trees for Jmax25) In contrast to SLAthere was no clear direction of the trait differences for eitherVcmax25 or Jmax25 half of the default values were lower



than the observed global means (tropical deciduous trees ex-tratropical evergreen trees raingreen shrubs and C4 grasses)the other half higher These differences point to a strong mis-match between default trait parameters in JSBACH with ob-served natural trait means

Figure 1 also shows that within each PFT large traitvariation was apparent as well as large overlap in traitranges among PFTs which corresponds to other trait anal-ysis (Laughlin et al 2010 Wright et al 2005a) This wasmost manifested for SLA where all PFTs overlapped witheach other For Vcmax25 the only PFTs of which the 95 confidence interval did not overlap were both tropical ever-green trees and C4 grasses with C3 grasses and for Jmax25both tropical trees and deciduous shrubs with C3 grasses

32 Simulated trait variation based on climatic drivers

The observed variation in PFT-specific community means (asgiven by Table 1) was related to environmental drivers by ap-plying multiple regression Table 2 (third column) presentsfor each trait and each PFT the selected climatic drivers usedin the regressions and their directional relationship with thetraits All regressions were significant except for C3 grassesfor Vcmax25 and for Jmax25 Although the selected environ-mental drivers differed among PFTs NSWR and MAP weremost frequently selected as drivers for SLA as were MAPMAT and RH for Vcmax25 and MAP for Jmax25 R2

adjustedwas up to 083 for Vcmax25 and 071 for Jmax25 For SLAmore variation remained unexplained with a maximum valueof 040 forR2

adjustedFig 2 presents the predicted trait ranges in JSBACH as

calculated by applying the regressions on simulated climatein JSBACH As in Fig 1 grey diamonds and black cir-cles again represent default trait values and observed globalmeans For each grid cell only simulated trait values ofthe dominant PFTs are selected to prevent trait values ofmarginal PFTs with almost no ecosystem impact to obscuretrait ranges most important for model output Raingreenshrubs were almost never dominant and predicted variationwas low while deciduous shrubs were never dominant

For all three traits a strong mismatch of default traits withthe simulated trait variation was apparent for SLA defaultvalues (grey diamonds) fell (at least) outside the 25ndash75 in-terquartile range in four PFTs (vs one in the observed data)and for Vcmax25 and Jmax25 for every PFT

In contrast to the default and observed trait simulationwhere PFTs had one constant value per trait a large climate-driven trait range was predicted for the variable traits simu-lation The higher SLA of deciduous versus evergreen treesobserved in natural vegetation was reflected in the simulatedSLA ranges As for the observational trait data (Fig 1) therewas large overlap in predicted trait ranges among PFTs aswell (Fig 2) particularly for Vcmax25 and Jmax25 Simu-lated SLA ranges were narrower and overlapped less amongPFTs compared to observed ranges For some PFTs (eg ex-

Fig 1Observed trait ranges of PFT-specific community means(a)SLA (m2 kgminus1 carbon)(b) Vcmax25 (micromol mminus2 sminus1) (c) Jmax25(micromol mminus2 sminus1) For C4 grasses PEPcase CO2 specificity (PEP)in mmol mminus2 sminus1 is modeled Box plots show the median (mid-dle line) the 25 and 75 quartiles (hinges) the outer valuewithin the 15lowast interquartile range (whiskers) and outliers (opencircles) Grey diamonds are trait values used in the default simu-lation black circles are observed global means used in the observedtraits simulation and therefore do not match the medians in the boxplots PFTs are tropical broadleaved evergreen trees (TrET) tropicalbroadleaved deciduous trees (TrDT) extratropical (both temperateand boreal) evergreen trees (ExTrET) extratropical deciduous trees(ExTrDT) raingreen shrubs (RgSh) colddeciduous shrubs (DSh)C3 grasses (C3G) and C4 grasses (C4G)

tratropical deciduous trees and C3 grasses for Vcmax25 andJmax25) simulated trait variation for Vcmax25 and Jmax25(Fig 2) was larger than observed variation (Fig 1) Predictedtraits may have fallen outside the ranges of observed varia-tion when climatic ranges as simulated in JSBACH exceededobservational climate ranges used to derive the traitndashclimaterelationships However this did not result in unrealistic trait



Fig 2 Simulated trait ranges of the dominant PFTs of each gridcell in the variable traits simulation(a) SLA (m2 kgminus1 carbon)(b) Vcmax25 (micromol mminus2 sminus1) (c) Jmax25 (micromol mminus2 sminus1) For C4grasses PEPcase CO2 specificity (PEP in mmol mminus2 sminus1) insteadof Jmax25 is modeled Box plots and symbols as in Fig 1 PFTsas in Fig 1 but without colddeciduous shrubs (DSh) because thisPFT was never dominant Observed global means are added againfor illustrative purposes Note that direct comparison with simulateddata is not appropriate here as the range of climatic conditions donot overlap for observed and simulated trait data

values as limits were set on maximum trait values basedon individual observations Furthermore we implemented atrade-off between Vcmax25 and Jmax25 keeping them withinrealistic ranges (even outside climatic ranges used for deter-mination of the regressions) For SLA such trade-off was nottaken into account but unrealistic SLA values were not verylikely as SLA was based on a much larger dataset coveringa very large range of climatic conditions

33 Inclusion of trait variation alters predicted globalpatterns of productivity

The observed traits and variable traits simulation showed asimilar pattern of median GPP along the latitudinal gradi-ent compared to the default simulation (Fig 3) but in ab-solute values GPP was much higher in the variable traitssimulation in the geographic zone between about 30 S and30 N (sim 50 around the equator and twice as much be-tween 25 and 30 N) GPP of the observed traits simulationalso reached higher values (up to twice as much) as the de-fault between 25 and 30 N but had lower values than thevariable traits simulation between 20 S and 15 N and high-est values (64ndash67 ) at higher latitudes

NPP differences between 30 S and 30 N were less pro-found than for GPP showing that trait variation affects globalpatterns of GPP and NPP in different ways Again at higherlatitudes the observed traits simulation predicted highestNPP For the variable traits simulation the reduction fromgross to net primary productivity around the tropics is muchstronger than for either other simulation resulting in smallerNPP differences among simulations compared to GPP dif-ferences (even though differences may still go up to 75 be-tween 25 and 30 N for both observed and variable traits sim-ulations) The large reduction in differences between NPPand GPP among simulations cannot solely be attributed toa higher respiration of PFTs due to higher Vcmax25 (seeFig 4) and is mainly a consequence of net productivity ex-ceeding the maximum sizes of the different carbon pools(eg leaves wood reserves) as defined in JSBACH result-ing in respiration of excess carbon This loss of excess carbonis visible in the tropical regions for all simulations but forthe variable traits simulation this resulted in a proportionallylarger amount of productivity removed This was enhancedby the higher SLA in these regions which partly determinesthe amount of carbon stored in the living parts of the plantsconsisting of leaves fine roots and sapwood

These results show that while reproducing global patternsof productivity incorporating global trait variation in themodel leads to strong changes in predicted productivity Eventhough GPP is not only determined by the photosynthetic pa-rameters (Vcmax25 and Jmax25) as it is affected by for ex-ample water availability as well on average shifts in GPPwere generally accompanied by shifts in similar directionsby mean Vcmax25 and Jmax25 (weighted by fractional cov-erage of PFTs) compared to the default simulation (Fig 4)Exceptions occurred as for example the drop in Vcmax25and Jmax25 above 40 N did not lead to a coinciding drop inGPP As expected as SLA does not play a role in productiv-ity changes in GPP could not be related to similar shifts inSLA



Fig 3 Latitudinal gradient of median GPP and NPP (gram car-bon mminus2 yrminus1) for default simulation (DEFAULT) observed traitssimulation (OBS TR) and variable traits simulation (VAR TR) andGPP taken from Beer et al (2010) (BEER)

Table 3Dominant PFT coverage () of vegetated grid cells for thethree different simulations

coverage of dominant PFT

Observed VariablePFT Default traits traits

simulation simulation simulation

Tropical evergreen trees 2306 0 1255Tropical deciduous trees 0024 2138 789Extratropical evergreen trees 1364 3631 2145Extratropical deciduous trees 1472 0 866Raingreen shrubs 366 130 012Colddeciduous shrubs 0048 0 0C3 grasses 2458 2450 2147C4 grasses 483 519 837

Bare surface 1544 1132 1950

34 Major shifts in vegetation distribution

Figure 5 and Table 3 show how the global distribution ofdominant vegetation types as predicted by the simulationsstrongly changes when incorporating observed global traitvalues or including climate-driven trait variation A PFT wasconsidered dominant if it had the highest fractional cover-age in a grid cell this ranged from coverage of almost 100 (mostly in tropical regions) to only 30 in some areas athigher latitudes (see Supplement S4 for fractional coverageof the dominant PFTs) Predicted dominant PFTs differedfrom the default simulation in 354 of the terrestrial gridcells for the variable traits simulation and in 505 of thegrid cells for the observed traits simulation

In the observed traits simulation tropical evergreen treesdominant in the default simulation were taken over by de-ciduous trees in Africa South America and Australia andextratropical deciduous trees were replaced by extratropi-

Fig 4 Latitudinal gradient of mean trait values weighted by frac-tional cover of the PFTs(a) SLA (m2 kgminus1 carbon)(b) Vcmax25(micromol mminus2 sminus1) and(c) Jmax25 (micromol mminus2 sminus1) For Jmax25 C4grasses are left out as PEP instead of Jmax25 is modeled for thisPFT Lines and abbreviations as in Fig 3

cal evergreen trees as the dominant PFT resulting in lessspatially heterogeneous dominant vegetation In the variabletraits simulation these shifts occurred as well (see Fig 5 andTable 3) However in contrast to the observed traits simu-lation both changes in dominant tree cover only occurredin limited areas which resulted in more spatial variation invegetation in the areas where trees were dominant The shiftsfrom tropical evergreen to tropical deciduous trees cannot beexplained by Vcmax25 and Jmax25 since these tropical PFTswere parameterized with the same values for these traits Themost profound difference between these tropical PFTs seemsto be their leaf turnover rate which is higher for the decidu-ous than for the evergreen trees As a consequence tropicaldeciduous trees had somewhat lower leaf area index whichmeant lower productivity in favorable periods but also lesscarbon loss in more stressful circumstances (eg drier peri-ods) In some areas this could have resulted in a higher totalyearly NPP for tropical deciduous trees thereby outcompet-ing evergreen trees



Fig 5Global distribution of dominant PFTs(a) Default simulation(b) observed traits simulation and(c) variable traits simulation



Another shift in predicted dominant vegetation in the vari-able traits simulation was an increase in C4 grasses (from48 in the default simulation to 84 ) This occurredmostly in Africa and Australia at the expense of tropical treesand raingreen shrubs This expansion of C4 grasses below theSahara coincided with higher fractions of burned area whichpromoted the expansion of grasses at the cost of trees

In the variable traits simulation bare ground increased asthe dominant cover type (from 154 in the default simu-lation to 195 ) in the southwest of the United States (andMexico) northern Canada and northeast of Siberia at the ex-pense of C3 grasses and deciduous trees resulting in a shiftof the boreal treeline toward lower latitudes These shiftsoften coincided with a decrease in Vcmax25 (see Supple-ment S5) suggesting lower productivity and consequentlyless expansion of these PFTs

35 Modulation of climate by traits

Predicted climate differed among simulations In the vari-able traits simulation enhanced climatendashvegetation feed-backs were possible both through changes in traits (mod-ifying PFT properties) as well as through shifts in vege-tation distribution (see Supplement S2 describing such in-teractions) However differences in climate cannot be re-lated unambiguously to either traits or vegetation shifts Asdifferences in temperature in unvegetated areas like Green-land and Antarctica show there does not have to be a directspatial relation between climate and vegetation Moreoversome vegetation shifts (compared to the default) do occur inplaces where climates differ among simulations (eg north-ern Canada) whereas in other areas no such effect occurred(eg in tropical zones)

Precipitation differences were local and showed no clearspatial pattern (Fig 6c and d) but in the variable traits andobserved traits simulations it was drier in Canada Asia andAustralia as well as in large parts of the Amazon rainforestscompared to the default simulation

Compared to the default simulation in the observed traitssimulation mean annual surface-air temperatures were pro-foundly higher (over 1C) in eastern Siberia Alaska the USand Australia and lower (up to 1C) in large parts of Eu-rope and Russia South Africa and South America (Fig 6a)meaning that (at least) temperature is very sensitive to pa-rameterization of traits Temperature differences were lessprofound between default and the variable traits simulation(Fig 6b) but still went up to around 1C Changes in temper-ature did not correlate with clear changes in traits or vegeta-tion shifts (eg treendashgrass shifts) but in the Southern Hemi-sphere (Australia Africa South America) corresponded todifferences in transpiration where cooler areas coincidedwith higher transpiration This could be related to the higherVcmax25 in these areas resulting in higher GPP and conse-quently an increase in transpiration

The differences between the observed traits simulation andvariable traits simulation indicate that by allowing traits tovary and respond to environmental conditions (as in the vari-able traits simulation) feedbacks between climate and traitsresult in more moderate temperature shifts showing the sig-nificant effect of adaptive traits on climate

36 Comparison of model output with observationaldata

Cohenrsquosκ indicating the correspondence of the global mapof potential (natural) vegetation of Ramankutty and Foley(1999) with simulated vegetation distribution was 02890282 and 0334 for the default observed traits and vari-able traits simulations respectively These values were some-what lower than theκ rsquos of other DGVMs egκ = 040 forthe Lund-Potsdam-Jena DGVM (LPJ) orκ = 042 for LPJwith implementation of plant hydraulic architecture bothwith 18 PFTs (Hickler et al 2006) orκ = 042 for a con-sensus map of multiple DGVMs with 6 PFTs (Cramer et al2001) However these models were compared to differentvegetation maps and had a different number of classes mak-ing comparisons difficult Our simulations with 7 vegetationclasses performed less well but this might partly depend onthe chosen vegetation map (Ramankutty and Foley 1999)A low κ means that for either simulation a large proportionof the grid cells did not match the potential vegetation map(Supplement Fig S62) However there is a substantial in-crease in similarity to observed vegetation from the defaultand observed traits simulation toward the variable traits sim-ulation

Mismatches occurred in large parts of the US and Canadawhere the simulations predicted mostly C3 grasslands whileaccording to the potential vegetation map also forests shouldbe present The same holds for large parts of Europe Thepotential vegetation map shows less bare ground than anysimulation resulting in mismatches in the US but also otherparts of the world Furthermore almost the whole continentof Australia did not correspond to this map shrubs and sa-vanna are dominant according to the map and even thoughthe models did predict C4 grasses there it was in differentareas Where the default and observed traits simulation hadlow correspondence with the vegetation map in Africa andSouth America the variable traits simulation performed bet-ter mainly with respect to the tropical trees Even thoughdifferences in performance are small the variable traits sim-ulation matched the potential vegetation map most closely

Comparing latitudinal patterns of median GPP with es-timates from Beer et al (2010) (thin blue line in Fig 3)each simulation produces substantial higher GPP at most lat-itudes (on average leading to 2 24 and 26 times higherGPP for the default observed traits and variable traits sim-ulations respectively) with the default simulation in gen-eral having the smallest differences Through the imposedobservation-based variation in Vcmax25 and Jmax25 in the



Fig 6Difference in mean annual temperature (C) and annual precipitation (mm yrminus1) Observed traits simulation minus default simulationfor (a) temperature and(c) precipitation and variable traits simulation minus default simulation for(b) temperature and(d) precipitation

variable traits simulation GPP particularly in the tropics wasoverestimated As nothing has been changed in the photosyn-thesis schemes this probably points to a parameter mismatchin the (default) model as in the variable traits simulation thetraits are supposed to be more close to realistic values than inthe default model

Chen et al (2012) suggest that global GPP estimates basedon remotely sensed LAI are underestimated by 9 when leafclumping is not taken into account as this would result in anunderestimation of the contribution of shaded leaves to GPPwith the strongest underestimation occurring in the tropicsThis implies that the estimates of GPP by Beer et al (2010)might be too low as remotely sensed LAI (and fAPAR) datawere used to extrapolate GPP estimates from flux-towers toglobal maps and thus the actual differences with our simu-lations may be less Also many other DGVMs show higherGPP in the tropical areas than the observed median GPP byBeer et al (2010) and the GPP estimates of the variable traitssimulation still do fall within the upper range of GPP predic-tions by other DGVMs

Comparisons of simulated biomass per m2 per (aggre-gated) PFT with current biomass estimates by Robinson(2007) show that for the combined tropical trees and sa-vannas the variable traits simulation (1312 kg C mminus2) mostclosely matched biomass estimates (1314 kg C mminus2) (Ta-

ble 4) For extratropical (temperate and boreal) forestand temperate grasslands the default simulation under-estimates carbon in vegetation (708 kg C mminus2) Both theobserved traits and variable traits simulation are closer(964 and 965 kg C mminus2) to the global estimates forforests (883 kg C mminus2) although they overestimate biomassFor temperate grasslands either simulation underestimatesbiomass with the default simulation deviating the least fromglobal estimates (016 vs 067 kg C mminus2) even though dif-ferences among simulations are small Overall this impliesthat of the three simulations the variable traits simulation isclosest to global biomass estimates per PFT per m2

The largest differences in GPP were found in the trop-ical areas Comparing simulated biomass of the tropicalzone with estimates of Saatchi et al (2011) either sim-ulation overestimates biomass in most areas (see Supple-ment S7) especially in the south of Brazil and large partsof southern Africa (areas with dry forests and savannas) Ingeneral for either simulation the areas with high biomass(gt 200 Mg C haminus1) are much more extensive than accordingto Saatchi et al (2011) However it has to be noted that forthese areas with high biomass the uncertainties in biomassestimates of Saatchi et al (2011) are large ranging from25 to over 50 meaning differences might be smaller inreality



Table 4Biomass estimates from Robinson (2007) compared to simulated biomass of the three simulations

Biomass (kg C mminus2)

Observed VariableRobinson PFT Biomass JSBACH PFT Default traits traits

(kg C mminus2) simulation simulation simulation

Tropical forest 2023 Tropical evergreen forest 1615 1463 2207Savannah amp grassland 605 Tropical deciduous forest 1084 1755 1696

C4 grasses 031 029 032Average 1314 910 1082 1312

Temperate forest 1365 Extratropical evergreen forest 579 1096 904Boreal forest 401 Extratropical deciduous forest 838 833 1025Average 883 708 964 965

Temperate grassland 067 C3 grasses 016 013 012

4 Discussion

The aim of this study is to identify the impacts of climate-driven trait variation within PFTs through vegetation re-sponses and vegetationndashatmosphere feedbacks on DGVMmodel behavior We determined observed trait variationwithin PFTs as comprehensively as possible based on rela-tionships between measured traits and climate and soil mois-ture representing major assembly processes by the abioticenvironment As model intercomparisons have shown thatlarge uncertainties exist in projections of land carbon up-take by DGVMs (Cramer et al 2001 Friedlingstein et al2006 Sitch et al 2008) incorporation of variation in veg-etation responses is important to allow more feedbacks be-tween vegetation and climate and to increase plausibility ofmodel predictions especially under strong climate changeHere we incorporated variation in plant responses based onrelationships between observed trait and climate data into theJSBACH DGVM which revealed profound effects on carbonfluxes and vegetation distribution

41 Challenges of modeling trait variation based ontraitndashclimate relationships

In our approach we used traitndashclimate relationships to de-scribe the observed natural trait variation and implementedthese in JSBACH These relationships identify and capturemultiple (approximate) drivers of natural trait variation andare thought to reflect abiotic assembly processes They inte-grate multiple vegetation responses at different temporal andspatial scales including acclimation adaptation of speciesand species replacement into a spatially and temporally vary-ing trait mean

One of the issues that might affect the reliability of modelperformance is the fact that we derived climate-driven traitvariation for each trait independently not explicitly account-ing for trait trade-offs The advantage of a changed trait oftenimplies a cost in another trait (eg the leaf economics spec-

trum of Wright et al 2004) possibly dampening the positiveeffect of the other trait and modifying plant performanceImplementing trait trade-offs helps to restrict possible traitcombinations or plant responses especially when simulatedtraits are modeled outside the observational climatic rangesused to derive the traitndashclimate relationships However trade-offs can only be included if the traits involved are explic-itly represented in the model and if they are PFT specificThis complicates including and evaluating the importanceof trade-offs in DGVMs In our situation we took care ofthe trade-off among Vcmax25 and Jmax25 by constrainingthe values of one trait by the other Moreover in JSBACHchanges in these traits are coupled to respiration and tran-spiration processes as well likely leading to consistent re-sponses Trade-offs between SLA and other traits were notaccounted for given the specific way SLA in JSBACH isused in relation to productivity independent of LAI and phe-nology This decoupling of SLA expresses that early phenol-ogy is strongly driven by remobilization of carbon from pre-vious year reserves and not from current year productivityInstead in JSBACH phenology is determined by the envi-ronmental drivers temperature and soil moisture Thereforethe commonly observed strong trade-off between SLA andleaf life span (LLS) (Wright et al 2004) is not expressedin JSBACH in contrast to many other DGVMs In fact byincorporating environment-driven variation in SLA the de-coupling between SLA and LLS has been partly diminishedcompared to the default model more closely representingnatural plant strategies Trait trade-offs thus need to be eval-uated depending on the model formulation

We updated trait values once per year This time step isa balance between computational efficiency and ecologicalrealism By this approach we avoid evaluating ontogenic im-pacts on trait values and whether environmental impacts dif-fer for different parts of the growing season for which cur-rently insufficient information is available Despite presumeddifferences in plasticity among species and PFTs (although



we are not aware of studies analyzing this) current analyseson within-species variation of leaf economics traits (Kattgeet al 2011 Messier et al 2010) suggest that all PFTs willbe sufficiently plastic to adjust their leaf economic traits atyearly timescales to the extents forced by year-to-year dif-ferences in climatic drivers The selected leaf traits in thisstudy are rather plastic and can vary within a year (Dubey etal 2011 Misson et al 2006) Other traits like wood traitswill have longer adaptation times than a year as changeswould involve individual or species turnover SLA may besomewhat less plastic than the two photosynthetic traits buta yearly reparameterization is reasonable as leaves of decid-uous species are shed every year and vary in relation to en-vironmental changes (Ma et al 2011) The maximum lifes-pan for leaves of evergreen PFTs is three years in JSBACHmeaning yearly SLA shifts occur in (on average) a third ofthe leaves resulting in a slight overestimation of SLA varia-tion in evergreens However the yearly shift in leaf trait val-ues may not only reflect acclimation but genetic adaptationand species replacements may also contribute In contrast to amechanistic approach in our approach the impacts and (un-known) timescales of those processes leading to trait shiftsdo not have to be differentiated

42 Implementing variation in PFTs comparingapproaches

Determining traitndashclimate relationships is broadly acceptedand applied in ecology (Ordonez et al 2010 Wright et al2005b) but using these relationships in combination with ob-servational trait data to calibrate trait variation and imple-ment these relationships in DGVMs is novel Recently thereare various attempts to implement more (trait) variation inPFTs or calibrate parameters on observational data For ex-ample Alton (2011) allowed for more variation in a num-ber of traits in the land surface model JULES-SF via tun-ing of traits to ranges set by observational data (eg eddy-covariance fluxes satellite data) but this trait variation wasmodeled as stochastic processes Some trait variation basedon mechanistic principles has been incorporated in O-CN(Zaehle and Friend 2010) there trait variation was not theprimary goal but a means to enable feedbacks in the mod-eling of nutrient cycles Some DGVMs also implement theconcept of environmental filtering like the JeDi-DGVM(Pavlick et al 2012) This DGVM models functional diver-sity by creating a continuum of PFTs or plant growth strate-gies (PGSs) which are determined by trait trade-offs andhabitat filtering However this model does not include mea-surable traits (Reu et al 2011) and as such does not allow forlinking traits to observational trait data or evaluation of traitcombinations Also aDGVM (Scheiter et al 2013) generatestrait variation and viable trait combinations are selected andinherited via species performance This means that environ-mental filtering only acts on trait values through the next gen-

eration aDGVM has not been validated with observationaldata nor does it include trait trade-offs

In contrast to these approaches we aim to comprehen-sively model trait variation in PFTs by identifying multipledrivers of trait variation and calibrating these relationships onobservational trait data Our proposed method is correlationaland does not explain mechanistically the adaptation acclima-tion or turnover processes behind trait variation (Pavlick et al2012) Nor does it fully account for constraints by biotic in-teractions trait trade-offs or dispersal limitation on trait val-ues (similar to most DGVMs) Still in our opinion it is animportant and necessary step as it reflects the observed corre-lation between traits and climatic drivers (Niinemets 2001StPaul et al 2012 Wright et al 2005b) as changing withchanging climate conditions Importantly it has the advan-tage that it does identify and quantify multiple abiotic driversof trait means and in this way captures a large part of ob-served trait variation as shown by the substantialR2

adjustedofmost regressions

43 Implications of incorporating observation-basedtrait variation

The observed global mean trait values of natural vegetationas used in the observed traits simulation strongly deviatedfrom trait values in the default simulation indicating a mis-match between default PFT trait means and means obtainedfrom natural vegetation Moreover either set of constant val-ues contrasts strongly with the large trait variation observedin natural vegetation (Fig 1) While we applied the mostcomprehensive database available today we are aware thatestimates of observed trait variation are still uncertain (Ta-ble 2) and need to be improved in future applications Nev-ertheless the wide range of observed trait values illustrateshow simulations with constant traits do not reflect naturaltrait variation In contrast this variation was reflected by thevariable traits simulation where trait variation representedabiotic assembly processes (Fig 2)

To investigate the effects of trait variation on vegetationndashclimate feedbacks it was essential to incorporate vegetationdynamics to allow trait shifts to alter vegetation distribu-tion and in this way modulate productivity and climate Incontrast to the simulations with constant traits (both defaultand observed traits simulations) the variable traits simula-tion enabled enhanced interactions between vegetation andclimate to occur via shifting traits Such changes in traits didnot have a direct effect on climate but propagated indirectlyto climate by modifying different plant properties and fluxes(see Supplement S2 for these pathways) These traitndashclimateinteractions ultimately resulted in more spatial variation indominant vegetation compared to the other two simulationsAs such predicted vegetation distribution is more a result oftemporal dynamics in vegetation properties than is the case inthe other simulations where these vegetation properties wereprescribed



In the current model setup direct and indirect effects (viaclimate changes in vegetation properties or distribution) oftraits are not easily disentangled which makes it difficult topinpoint how and to what degree traits directly modulatedmodel output Our main aim however is to identify the in-tegrated impact of trait variation on model behavior via veg-etation responses and feedbacks with climate So we weremore interested in whether and to what extent trait varia-tion alters model performance than to know the exact path-ways as we acknowledged in advance this is difficult with acoupled model setup Offline simulations with prescribed cli-mate could help to reduce the number of possible variablescausing the changes and may clarify some patterns Howeverfor our model such offline simulations were for technicalreasons not yet feasible

Provided the strong effect exerted by climate on traits ma-jor differences among the simulations in predicted vegeta-tion distribution and productivity were expected the latterespecially when parameters that affect assimilation rate areconcerned as sensitivity analyses of DGVMs have shown(White et al 2000 Zaehle et al 2005) Indeed for the ob-served traits simulation and variable traits simulation this re-sulted in large differences in the new equilibrium state bothcompared to the default and to each other Not only werevegetation properties affected but also climate changed andmean temperatures were altered by up to more than 1CThis confirms that JSBACH (and other DGVMs) are quitesensitive to changes in a few key parameters and this maycast some doubts on the reliability and predictive power ofDGVMs in general

As the simulations in this study provide equilibrium statesand as such do not necessarily correspond to current climateor vegetation composition model comparisons with obser-vational data must be interpreted with care although it doesprovide insights in the realism of the simulations DGVMsare parameterized to produce approximately realistic resultsand therefore our simulations were not expected to approachobservations better than the default simulation Even thoughthe variable traits simulations produced high GPP for trop-ical areas its biomass estimates and vegetation distributionmore closely resembled observational data than the defaultsimulation

The large differences in model output and the mismatchbetween default trait values in the model and observed traitvariation in nature demonstrate that ndash besides a correct rep-resentation of plant physiology (eg photosynthesis transpi-ration) ndash integration of ecological theory will be a step for-ward to help improve vegetation representation in DGVMsand ESMs Allowing for variable vegetation responses to cli-mate and soil will have important consequences for predic-tions by vegetation models for vegetation distribution andproductivity as well as for global current and future climateA model intercomparison with this approach under elevatedCO2 projections for which large uncertainties in predictions

by current DGVMs exist (Friedlingstein et al 2006 Sitch etal 2008) should therefore be one of the next steps

5 Conclusions

In this study we identified the impacts of modeling climate-driven trait variation in PFTs calibrated on observationaldata on JSBACH model behavior The current mismatch ofconstant trait values in JSBACH with observed natural traitvariation and the impact of incorporation of trait variation onmodel behavior with respect to vegetation distribution pro-ductivity and global climate together emphasize the need forimplementation of more observation-based trait variation andconcomitant ecological concepts The suggested approachbased on such data and concepts reflects vegetation accli-mation and adaptation to the environment and will help en-able more reliable modeling of vegetation behavior under un-known climates

Supplementary material related to this article isavailable online athttpwwwbiogeosciencesnet1054972013bg-10-5497-2013-supplementpdf

AcknowledgementsThis study has been financed by the Nether-lands Organisation for Scientific Research (NWO) ThemeSustainable Earth Research (project number TKS09-03) Theauthors are grateful to TERRABITES COST Action ES0805 forfinancial support and stimulation of discussions on modeling traitsin the terrestrial biosphere Furthermore this study was supportedby the TRY initiative on plant traits (httpwwwtry-dborg) TRYis hosted developed and maintained by J Kattge and G Bonisch(Max Planck Institute for Biogeochemistry Jena Germany) andis supported by DIVERSITAS IGBP the Global Land ProjectQUEST and GIS ldquoClimat Environnement et Societerdquo FranceFinally the authors thank Veronika Gayler (Max Planck Institutefor Meteorology Hamburg Germany) for the technical andmethodological support on JSBACHECHAM5 and Bob Doumafor inspiring discussions and help with R programming

Edited by K Thonicke

References

Ackerly D D and Cornwell W K A trait-based approach to com-munity assembly partitioning of species trait values into within-and among-community components Ecol Lett 10 135ndash145doi101111j1461-0248200601006x 2007

Allen R G Pereira L S Raes D and Smith M Crop evap-otranspiration ndash Guidelines for computing crop water require-ments FAO Irrigation and drainage paper 56 Food and Agricul-ture Organization of the United Nations Rome 1998

Alton P B How useful are plant functional types in global sim-ulations of the carbon water and energy cycles J GeophysRes-Biogeo 116 G01030 doi1010292010JG001430 2011



Bahn M Wohlfahrt G and Haubner E Leaf photosynthesis ni-trogen contents and specific leaf area of 30 grassland speciesin differently managed mountain ecosystems in the EasternAlps in Land-Use Changes in European Mountain EcosystemsECOMONT ndash Concepts and Results edited by Cernusca ATappeiner U and Bayfield N Blackwell Wissenschaft Berlin247ndash255 1999

Beer C Reichstein M Tomelleri E Ciais P Jung M Carval-hais N Roedenbeck C Arain M A Baldocchi D Bonan GB Bondeau A Cescatti A Lasslop G Lindroth A LomasM Luyssaert S Margolis H Oleson K W Roupsard OVeenendaal E Viovy N Williams C Woodward F I and Pa-pale D Terrestrial Gross Carbon Dioxide Uptake Global Dis-tribution and Covariation with Climate Science 329 834ndash838doi101126science1184984 2010

Brovkin V Raddatz T Reick C H Claussen M and Gayler VGlobal biogeophysical interactions between forest and climateGeophys Res Lett 36 L07405 doi1010292009GL0375432009

Cavender-Bares J Keen A and Miles B Phylogenetic structureof floridian plant communities depends on taxonomic and spatialscale Ecology 87 S109ndashS122 2006Chen J M Mo G Pisek J Liu J Deng F Ishizawa M andChan D Effects of foliage clumping on the estimation of globalterrestrial gross primary productivity Global Biogeochem Cy26 18 GB1019 doi1010292010gb003996 2012

Cohen J A coefficient of agreement for nominal scales EducPsychol Meas 20 37ndash46 doi1011770013164460020001041960

Collatz G J Ribas-Carbo M and Berry J A Coupledphotosynthesis-stomatal conductance model for leaves of C4plants Aust J Plant Physiol 19 519ndash538 1992

Cornelissen J H C Cerabolini B Castro-Diez P Villar-Salvador P Montserrat-Marti G Puyravaud J P MaestroM Werger M J A and Aerts R Functional traits of woodyplants correspondence of species rankings between field adultsand laboratory-grown seedlings J Veg Sci 14 311ndash322doi101111j1654-11032003tb02157x 2003

Cornelissen J H C Quested H M Gwynn-Jones D VanLogtestijn R S P De Beus M A H Kondratchuk ACallaghan T V and Aerts R Leaf digestibility and lit-ter decomposability are related in a wide range of sub-arctic plant species and types Funct Ecol 18 779ndash786doi101111j0269-8463200400900x 2004

Cornwell W K and Ackerly D D Community assemblyand shifts in plant trait distributions across an environmen-tal gradient in coastal California Ecol Monogr 79 109ndash126doi10189007-11341 2009

Cornwell W K Schwilk D W and Ackerly D DA trait-based test for habitat filtering convex hullvolume Ecology 87 1465ndash1471 doi1018900012-9658(2006)87[1465ATTFHF]20CO2 2006

Cramer W Bondeau A Woodward F I Prentice I C BettsR A Brovkin V Cox P M Fisher V Foley J A FriendA D Kucharik C Lomas M R Ramankutty N SitchS Smith B White A and Young-Molling C Global re-sponse of terrestrial ecosystem structure and function to CO2and climate change results from six dynamic global vegetationmodels Glob Change Biol 7 357ndash373 doi101046j1365-

2486200100383x 2001Diaz S Hodgson J G Thompson K Cabido M Cornelissen

J H C Jalili A Montserrat-Marti G Grime J P Zarrinka-mar F Asri Y Band S R Basconcelo S Castro-Diez PFunes G Hamzehee B Khoshnevi M Perez-HarguindeguyN Perez-Rontome M C Shirvany F A Vendramini F Yaz-dani S Abbas-Azimi R Bogaard A Boustani S CharlesM Dehghan M de Torres-Espuny L Falczuk V Guerrero-Campo J Hynd A Jones G Kowsary E Kazemi-Saeed FMaestro-Martinez M Romo-Diez A Shaw S Siavash BVillar-Salvador P and Zak M R The plant traits that driveecosystems Evidence from three continents J Veg Sci 15295ndash304 doi101111j1654-11032004tb02266x 2004

Domingues T F Meir P Feldpausch T R Saiz G Vee-nendaal E M Schrodt F Bird M Djagbletey G HienF Compaore H Diallo A Grace J and Lloyd J Co-limitation of photosynthetic capacity by nitrogen and phospho-rus in West Africa woodlands Plant Cell Environ 33 959ndash980doi101111j1365-3040201002119x 2010

Dubey P Raghubanshi A S and Singh J S Intra-seasonal vari-ation and relationship among leaf traits of different forest herbsin a dry tropical environment Curr Sci 100 69ndash76 2011

Farquhar G D Caemmerer S V and Berry J A A biochem-ical model of photosynthetic CO2 assimilation in leaves of C3Planta 149 78ndash90 doi101007BF00386231 1980

Foley J A Levis S Prentice I C Pollard D and ThompsonS L Coupling dynamic models of climate and vegetation GlobChange Biol 4 561ndash579 doi101046j1365-24861998t01-1-00168x 1998

Fyllas N M Patino S Baker T R Bielefeld Nardoto G Mar-tinelli L A Quesada C A Paiva R Schwarz M Horna VMercado L M Santos A Arroyo L Jimenez E M LuizaoF J Neill D A Silva N Prieto A Rudas A Silviera MVieira I C G Lopez-Gonzalez G Malhi Y Phillips O Land Lloyd J Basin-wide variations in foliar properties of Ama-zonian forest phylogeny soils and climate Biogeosciences 62677ndash2708 doi105194bg-6-2677-2009 2009

Freschet G T Dias A T C Ackerly D D Aerts R van Bode-gom P M Cornwell W K Dong M Kurokawa H LiuG Onipchenko V G Ordonez J C Peltzer D A Richard-son S J Shidakov I I Soudzilovskaia N A Tao J andCornelissen J H C Global to community scale differences inthe prevalence of convergent over divergent leaf trait distribu-tions in plant assemblages Global Ecol Biogeogr 20 755ndash765doi101111j1466-8238201100651x 2011

Friedlingstein P Cox P Betts R Bopp L Von Bloh WBrovkin V Cadule P Doney S Eby M Fung I Bala GJohn J Jones C Joos F Kato T Kawamiya M Knorr WLindsay K Matthews H D Raddatz T Rayner P ReickC Roeckner E Schnitzler K G Schnur R Strassmann KWeaver A J Yoshikawa C and Zeng N Climate-carbon cy-cle feedback analysis Results from the C4MIP model intercom-parison J Climate 19 3337ndash3353 doi101175JCLI380012006

Garnier E Lavorel S Ansquer P Castro H Cruz P DolezalJ Eriksson O Fortunel C Freitas H Golodets C GrigulisK Jouany C Kazakou E Kigel J Kleyer M Lehsten VLeps J Meier T Pakeman R Papadimitriou M Papanas-tasis V P Quested H Quetier F Robson M Roumet C



Rusch G Skarpe C Sternberg M Theau J-P Thebault AVile D and Zarovali M P Assessing the effects of land-usechange on plant traits communities and ecosystem functioningin grasslands A standardized methodology and lessons from anapplication to 11 European sites Ann Bot-London 99 967ndash985doi101093aobmcm215 2007

Gotzenberger L de Bello F Brathen K A Davison J DubuisA Guisan A Leps J Lindborg R Moora M Partel MPellissier L Pottier J Vittoz P Zobel K and Zobel MEcological assembly rules in plant communities-approaches pat-terns and prospects Biol Rev 87 111ndash127 doi101111j1469-185X201100187x 2012

Grace J San Jose J Meir P Miranda H S and Montes R AProductivity and carbon fluxes of tropical savannas J Biogeogr33 387ndash400 doi101111j1365-2699200501448x 2006

Hickler T Prentice I C Smith B Sykes M T and ZaehleS Implementing plant hydraulic architecture within the LPJDynamic Global Vegetation Model Global Ecol Biogeogr 15567ndash577 doi101111j1466-8238200600254x 2006

Kattge J and Knorr W Temperature acclimation in a biochem-ical model of photosynthesis a reanalysis of data from 36species Plant Cell Environ 30 1176ndash1190 doi101111j1365-3040200701690x 2007

Kattge J Knorr W Raddatz T and Wirth C Quantifying pho-tosynthetic capacity and its relationship to leaf nitrogen contentfor global-scale terrestrial biosphere models Glob Change Biol15 976ndash991 doi101111j1365-2486200801744x 2009

Kattge J Diaz S Lavorel S Prentice C Leadley P BoenischG Garnier E Westoby M Reich P B Wright I J Cornelis-sen J H C Violle C Harrison S P van Bodegom P M Re-ichstein M Enquist B J Soudzilovskaia N A Ackerly DD Anand M Atkin O Bahn M Baker T R Baldocchi DBekker R Blanco C C Blonder B Bond W J BradstockR Bunker D E Casanoves F Cavender-Bares J ChambersJ Q Chapin F S III Chave J Coomes D Cornwell WK Craine J M Dobrin B H Duarte L Durka W ElserJ Esser G Estiarte M Fagan W F Fang J Fernandez-Mendez F Fidelis A Finegan B Flores O Ford H FrankD Freschet G T Fyllas N M Gallagher R V Green WA Gutierrez A G Hickler T Higgins S I Hodgson J GJalili A Jansen S Joly C A Kerkhoff A J Kirkup D Ki-tajima K Kleyer M Klotz S Knops J M H Kramer KKuehn I Kurokawa H Laughlin D Lee T D LeishmanM Lens F Lenz T Lewis S L Lloyd J Llusia J LouaultF Ma S Mahecha M D Manning P Massad T MedlynB E Messier J Moles A T Mueller S C Nadrowski KNaeem S Niinemets U Noellert S Nueske A Ogaya ROleksyn J Onipchenko V G Onoda Y Ordonez J Over-beck G Ozinga W A Patino S Paula S Pausas J GPenuelas J Phillips O L Pillar V Poorter H Poorter LPoschlod P Prinzing A Proulx R Rammig A Reinsch SReu B Sack L Salgado-Negre B Sardans J Shiodera SShipley B Siefert A Sosinski E Soussana J F Swaine ESwenson N Thompson K Thornton P Waldram M Wei-her E White M White S Wright S J Yguel B ZaehleS Zanne A E and Wirth C TRY ndash a global database of planttraits Glob Change Biol 17 2905ndash2935 doi101111j1365-2486201102451x 2011

Keddy P A Assembly and response rules two goals forpredictive community ecology J Veg Sci 3 157ndash164doi1023073235676 1992

Kleyer M Bekker R M Knevel I C Bakker J P Thomp-son K Sonnenschein M Poschlod P van Groenendael J MKlimes L Klimesova J Klotz S Rusch G M Hermy MAdriaens D Boedeltje G Bossuyt B Dannemann A En-dels P Goetzenberger L Hodgson J G Jackel A K KuehnI Kunzmann D Ozinga W A Roemermann C Stadler MSchlegelmilch J Steendam H J Tackenberg O Wilmann BCornelissen J H C Eriksson O Garnier E and Peco B TheLEDA Traitbase a database of life-history traits of the NorthwestEuropean flora J Ecol 96 1266ndash1274 doi101111j1365-2745200801430x 2008

Kottek M Grieser J Beck C Rudolf B and Rubel F Worldmap of the Koppen-Geiger climate classification updated Mete-orol Z 15 259ndash263 doi1011270941-294820060130 2006

Kurokawa H and Nakashizuka T Leaf herbivory and decompos-ability in a Malaysian tropical rain forest Ecology 89 2645ndash2656 doi10189007-13521 2008

Laughlin D C Leppert JJ Moore MM and Sieg CH Amulti-trait test of the leaf-height-seed plant strategy scheme with133 species from a pine forest flora Funct Ecol 24 493ndash501doi101111j1365-2435200901672x 2010

Lavorel S and Garnier E Predicting changes in community com-position and ecosystem functioning from plant traits revisitingthe Holy Grail Funct Ecol 16 545ndash556 doi101046j1365-2435200200664x 2002

Lavorel S McIntyre S Landsberg J and Forbes T D A Plantfunctional classifications from general groups to specific groupsbased on response to disturbance Trends Ecol Evol 12 474ndash478 doi101016S0169-5347(97)01219-6 1997

Louault F Pillar V D Aufrere J Garnier E and Soussana JF Plant traits and functional types in response to reduced dis-turbance in a semi-natural grassland J Veg Sci 16 151ndash160doi101111j1654-11032005tb02350x 2005

Ma S Baldocchi D D Mambelli S and Dawson T E Are tem-poral variations of leaf traits responsible for seasonal and inter-annual variability in ecosystem CO2 exchange Funct Ecol 25258ndash270 doi101111j1365-2435201001779x 2011

Marsland S J Haak H Jungclaus J H Latif M and RoskeF The Max-Planck-Institute global oceansea ice model withorthogonal curvilinear coordinates Ocean Model 5 91ndash127doi101016S1463-5003(02)00015-X 2003

Medlyn B E Badeck F W De Pury D G G Barton C V MBroadmeadow M Ceulemans R De Angelis P ForstreuterM Jach M E Kellomaki S Laitat E Marek M PhilippotS Rey A Strassemeyer J Laitinen K Liozon R PortierB Roberntz P Wang K and Jarvis P G Effects of elevated[CO2] on photosynthesis in European forest species a meta-analysis of model parameters Plant Cell Environ 22 1475ndash1495 doi101046j1365-3040199900523x 1999

Medlyn B E Dreyer E Ellsworth D Forstreuter M Harley PC Kirschbaum M U F Le Roux X Montpied P Strasse-meyer J Walcroft A Wang K and Loustau D Temperatureresponse of parameters of a biochemically based model of photo-synthesis II A review of experimental data Plant Cell Environ25 1167ndash1179 doi101046j1365-3040200200891x 2002



Messier J McGill B J and Lechowicz M J Howdo traits vary across ecological scales A case for trait-based ecology Ecol Lett 13 838ndash848 doi101111j1461-0248201001476x 2010

Miralles D G Holmes T R H De Jeu R A M Gash J HMeesters A G C A and Dolman A J Global land-surfaceevaporation estimated from satellite-based observations HydrolEarth Syst Sci 15 453ndash469 doi105194hess-15-453-20112011

Misson L Tu K P Boniello R A and Goldstein A H Season-ality of photosynthetic parameters in a multi-specific and verti-cally complex forest ecosystem in the Sierra Nevada of Califor-nia Tree Physiol 26 729ndash741 2006

Mokany K Raison R J and Prokushkin A S Critical analysisof root shoot ratios in terrestrial biomes Glob Change Biol 1284ndash96 doi101111j1365-24862005001043x 2006

New M Lister D Hulme M and Makin I A high-resolutiondata set of surface climate over global land areas Clim Res 211ndash25 doi103354cr021001 2002

NiinemetsU Components of leaf dry mass per area - thickness anddensity ndash alter leaf photosynthetic capacity in reverse directionsin woody plants New Phytol 144 35ndash47 doi101046j1469-8137199900466x 1999

Niinemets U Global-scale climatic controls of leafdry mass per area density and thickness in treesand shrubs Ecology 82 453ndash469 doi1018900012-9658(2001)082[0453GSCCOL]20CO2 2001

Ogaya R and Penuelas J Tree growth mortality and above-ground biomass accumulation in a holm oak forest under afive-year experimental field drought Plant Ecol 189 291ndash299doi101007s11258-006-9184-6 2007

Ogaya R and Penuelas J Changes in leaf delta(13)C anddelta(15)N for three Mediterranean tree species in rela-tion to soil water availability Acta Oecol 34 331ndash338doi101016jactao200806005 2008

Ordonez J C van Bodegom P M Witte J P M Wright IJ Reich P B and Aerts R A global study of relationshipsbetween leaf traits climate and soil measures of nutrient fer-tility Global Ecol Biogeogr 18 137ndash149 doi101111j1466-8238200800441x 2009

Ordonez J C van Bodegom P M Witte J P M BartholomeusR P van Hal J R and Aerts R Plant Strategies in Relationto Resource Supply in Mesic to Wet Environments Does TheoryMirror Nature Am Nat 175 225ndash239 doi1010866495822010

Patino S Fyllas N M Baker T R Paiva R Quesada CA Santos A J B Schwarz M ter Steege H PhillipsO L and Lloyd J Coordination of physiological and struc-tural traits in Amazon forest trees Biogeosciences 9 775ndash801doi105194bg-9-775-2012 2012

Pavlick R Drewry D T Bohn K Reu B and Kleidon AThe Jena Diversity-Dynamic Global Vegetation Model (JeDi-DGVM) a diverse approach to representing terrestrial biogeog-raphy and biogeochemistry based on plant functional trade-offs Biogeosciences Discuss 9 4627ndash4726 doi105194bgd-9-4627-2012 2012

Prentice I C Bondeau A Cramer W Harrison SP HicklerT Lucht W Sitch S Smith B and Sykes MT Dynamicglobal vegetation modelling quantifying terrestrial ecosystem

responses to large-scale environmental change in TerrestrialEcosystems in a Changing World edited by Canadell J GPataki D E and Pitelka L F Springer Heidelberg Germany175ndash192 2007

Pyankov V I Kondratchuk A V and Shipley B Leafstructure and specific leaf mass the alpine desert plants ofthe Eastern Pamirs Tadjikistan New Phytol 143 131ndash142doi101046j1469-8137199900435x 1999

Quillet A Peng C and Garneau M Toward dynamic globalvegetation models for simulating vegetation-climate interactionsand feedbacks recent developments limitations and future chal-lenges Environ Rev 18 333ndash353 doi101139A10-016 2010

Raddatz T J Reick C H Knorr W Kattge J Roeckner ESchnur R Schnitzler K G Wetzel P and Jungclaus J Willthe tropical land biosphere dominate the climate-carbon cyclefeedback during the twenty-first century Clim Dynam 29565ndash574 doi101007s00382-007-0247-8 2007

Ramankutty N and Foley J A Estimating historical changes inglobal land cover Croplands from 1700 to 1992 Global Bio-geochem Cy 13 997ndash1027 doi1010291999gb900046 1999

Reich P B Tjoelker M G Pregitzer K S Wright I J OleksynJ and Machado J-L Scaling Scaling of respiration to nitrogenin leaves stems and roots of higher land plants Ecol Lett 11793ndash801 doi101111j1461-0248200801185x 2008

Reich P B Oleksyn J and Wright I J Leaf phospho-rus influences the photosynthesis-nitrogen relation a cross-biome analysis of 314 species Oecologia 160 207ndash212doi101007s00442-009-1291-3 2009

Reu B Proulx R Bohn K Dyke J G Kleidon A PavlickR and Schmidtlein S The role of climate and plant func-tional trade-offs in shaping global biome and biodiversity pat-terns Global Ecol Biogeogr 20 570ndash581 doi101111j1466-8238201000621x 2011

Robinson D Implications of a large global root biomass for carbonsink estimates and for soil carbon dynamics P Roy Soc B-BiolSci 274 2753ndash2759 doi101098rspb20071012 2007

Roeckner E Baeuml G Bonaventura L Brokopf R EschM Giorgetta M Hagemann S Kirchner I Kornblueh LManzini E Rhodin A Schlese U Schulzweida U andTompkins A The general circulation model ECHAM5 Part IModel description report 349 Max-Planck-Institute for Meteo-rology Hamburg 2003

Saatchi S S Harris N L Brown S Lefsky M MitchardE T A Salas W Zutta B R Buermann W Lewis SL Hagen S Petrova S White L Silman M and MorelA Benchmark map of forest carbon stocks in tropical regionsacross three continents P Natl Acad Sci USA 108 9899ndash9904 doi101073pnas1019576108 2011

Scheiter S Langan L and Higgins S I Next-generation dy-namic global vegetation models learning from community ecol-ogy New Phytol 198 957ndash969 doi101111nph12210 2013

Shipley B Structured interspecific determinants of specific leafarea in 34 species of herbaceous angiosperms Funct Ecol 9312ndash319 doi1023072390579 1995

Shipley B and Vu T T Dry matter content as a measure of drymatter concentration in plants and their parts New Phytol 153359ndash364 doi101046j0028-646X200100320x 2002

Simioni G Le Roux X Gignoux J and Walcroft AS Leaf gas exchange characteristics and water- and



nitrogen-use efficiencies of dominant grass and tree speciesin a West African savanna Plant Ecol 173 233ndash246doi101023BVEGE00000293237452380 2004

StPaul N K M Limousin J M Rodriguez-Calcerrada J Ruf-fault J Rambal S Letts M G and Misson L Photosyn-thetic sensitivity to drought varies among populations of Quer-cus ilex along a rainfall gradient Funct Plant Biol 39 25ndash37doi101071FP11090 2012

Swaine E K Ecological and evolutionary drivers of plant commu-nity assembly in a Bornean rain forest PhD thesis Universityof Aberdeen Aberdeen 2007

Swenson N G and Enquist B J Ecological and evolutionary de-terminants of a key plant functional trait Wood density and itscommunity-wide variation across latitude and elevation Am JBot 94 451ndash459 doi103732ajb943451 2007

van Bodegom P M Douma J C Witte J P M OrdonezJ C Bartholomeus R P and Aerts R Going beyondlimitations of plant functional types when predicting globalecosystemndashatmosphere fluxes exploring the merits of traits-based approaches Global Ecol Biogeogr 21 625ndash636doi101111j1466-8238201100717x 2012

van Ommen Kloeke A E E Douma J C Ordonez J C ReichP B and van Bodegom P M Global quantification of contrast-ing leaf life span strategies for deciduous and evergreen speciesin response to environmental conditions Global Ecol Biogeogr21 224ndash235 doi101111j1466-8238201100667x 2012

Vile D Shipley B and Garnier E Ecosystem productivity canbe predicted from potential relative growth rate and speciesabundance Ecol Lett 9 1061ndash1067 doi101111j1461-0248200600958x 2006

Violle C Navas M-L Vile D Kazakou E Fortunel C Hum-mel I and Garnier E Let the concept of trait be functionalOikos 116 882ndash892 doi101111j20070030-129915559x2007

Westoby M Falster D S Moles A T Vesk P A and WrightI J Plant ecological strategies Some leading dimensions ofvariation between species Annu Rev Ecol Syst 33 125ndash159doi101146annurevecolsys33010802150452 2002

White M A Thornton P E Running S W and Nemani R RParameterization and sensitivity analysis of the BIOME-BGCterrestrial ecosystem model net primary production controlsEarth Interact 4 1ndash85 2000

Wohlfahrt G Bahn M Haubner E Horak I Michaeler WRottmar K Tappeiner U and Cernusca A Inter-specific vari-ation of the biochemical limitation to photosynthesis and relatedleaf traits of 30 species from mountain grassland ecosystemsunder different land use Plant Cell Environ 22 1281ndash1296doi101046j1365-3040199900479x 1999

Wright I J Reich P B Westoby M Ackerly D D BaruchZ Bongers F Cavender-Bares J Chapin T Cornelissen JH C Diemer M Flexas J Garnier E Groom P K Gu-lias J Hikosaka K Lamont B B Lee T Lee W Lusk CMidgley J J Navas M L Niinemets U Oleksyn J OsadaN Poorter H Poot P Prior L Pyankov V I Roumet CThomas S C Tjoelker M G Veneklaas E J and Villar RThe worldwide leaf economics spectrum Nature 428 821ndash827doi101038nature02403 2004

Wright I J Reich P B Cornelissen J H C Falster D S Gar-nier E Hikosaka K Lamont B B Lee W Oleksyn J Os-ada N Poorter H Villar R Warton D I and Westoby MAssessing the generality of global leaf trait relationships NewPhytol 166 485ndash496 doi101111j1469-8137200501349x2005a

Wright I J Reich P B Cornelissen J H C Falster D SGroom P K Hikosaka K Lee W Lusk C H NiinemetsU Oleksyn J Osada N Poorter H Warton D I andWestoby M Modulation of leaf economic traits and trait re-lationships by climate Global Ecol Biogeogr 14 411ndash421doi101111j1466-822x200500172x 2005b

Wright I J Reich P B Atkin O K Lusk C H Tjoelker M Gand Westoby M Irradiance temperature and rainfall influenceleaf dark respiration in woody plants evidence from comparisonsacross 20 sites New Phytol 169 309ndash319 doi101111j1469-8137200501590x 2006

Wullschleger S D Biochemical limitations to carbon assimilationin C3 plants ndash a retrospective analysis of ACi curves from 109species J Exp Bot 44 907ndash920 doi101093jxb4459071993

Zaehle S and Friend A D Carbon and nitrogen cycle dynamicsin the O-CN land surface model 1 Model description site-scaleevaluation and sensitivity to parameter estimates Global Bio-geochem Cy 24 GB1005 doi1010292009GB003521 2010

Zaehle S Sitch S Smith B and Hatterman F Effectsof parameter uncertainties on the modeling of terrestrialbiosphere dynamics Global Biogeochem Cy 19 GB3020doi1010292004GB002395 2005

Ziehn T Kattge J Knorr W and Scholze M Im-proving the predictability of global CO2 assimilation ratesunder climate change Geophys Res Lett 38 L10404doi1010292011GL047182 2011



much improved over the past decade although many issuesstill need to be resolved (Quillet et al 2010) One of theseissues is the way in which plant functional types (PFTs)are used to represent vegetation PFTs are classes of plantspecies with presumably similar roles in an ecosystem re-sponding in a comparable manner to environmental con-ditions like water and nutrient availability (Lavorel et al1997) They are defined by a combination of attributes suchas plant growth form (eg trees shrubs grasses herbs) phe-nology (evergreen raingreen summergreen) and bioclimatictolerances (eg minimum temperature requirements) No-tably most current PFT classifications use constant param-eter values for some key plant traits ie plant properties thatreflect the way plants cope with their environment (Violleet al 2007) Using constant plant traits in PFTs has seri-ous limitations (Ordonez et al 2009 Van Bodegom et al2012) as such constant values contrast with observed traitvariation (Ackerly and Cornwell 2007 Freschet et al 2011Westoby et al 2002) therefore not accounting for local en-vironmental constraints Furthermore global trait databaseanalyses have shown that the variation in plant traits is largewithin PFTs and often even greater than the difference inmeans among PFTs (Laughlin et al 2010 Wright et al2005a) Even though PFTs may capture a significant part ofthe global plant trait variation a large part (for some traits upto 75 ) may still be unexplained (Kattge et al 2011) and isthus not represented in current DGVMs

Given that plants can adjust to the environment viachanges in traits and that such changes influence ecosystemfunctioning (Diaz et al 2004 Lavorel and Garnier 2002)implementing these trait-driven interactions within PFTs ofDGVMs is highly relevant By simulating variation in planttrait responses it is better possible to quantify plant adap-tation to climate and to account for plantndashatmosphere feed-backs in DGVMs Indeed the limitations of inflexible PFTsis increasingly acknowledged by modelers and several at-tempts to allow for more (trait) variation within PFTs canbe found in for example JeDi-DGVM (Pavlick et al 2012)or aDGVM (Scheiter et al 2013) However none of theapproaches so far tried to maximally include trait varia-tion based on observational trait data and capture multiplesources of this variation by relating trait data to environmen-tal variables This inclusion may be possible by making useof the relationship between traits and multiple environmen-tal drivers as occurring at both regional and global scalesfor instance in relation to temperature water and nutrientavailability or disturbances (Ordonez et al 2009 Van Om-men Kloeke et al 2012 Wright et al 2005b) Such rela-tionships between environmental conditions and traits canpotentially be understood via ecological assembly theorywhich describes the processes that determine species assem-blages (Cornwell and Ackerly 2009 Cornwell et al 2006Gotzenberger et al 2012) An important abiotic assemblyprocess is habitat filtering (Keddy 1992) which describeshow local environmental drivers (eg soil fertility or pre-

cipitation) constrain the range of potential species and re-lated trait values in a given habitat For many traits such asspecific leaf area leaf nitrogen and wood density this con-tributes to trait convergence within communities prevailingover trait divergence (Freschet et al 2011 Swenson and En-quist 2007) and resulting in global relationships betweencommunity trait means and climatic drivers Community traitmeans match closely to the model resolution of DGVMsgiven that the majority of the models provide a mean valueof a trait for a PFT (in a given grid cell) By identifying theenvironmental drivers of variation in community trait meansmultiple causes of variation are determined and the observednatural trait variation can be modeled with a high level ofaccuracy

The aim of this study is thus to model climate-driventrait variation within PFTs determined as comprehensivelyas possible from observations and to identify the impactsthrough vegetation responses and vegetationndashatmospherefeedbacks on DGVM model behavior So far observationaltrait data have been used to derive mean parameter values anduncertainties for PFTs in the context of different vegetationmodels (Kattge et al 2009 Ziehn et al 2011) but this studyprovides a proof of concept to predict mean parameter valuesper PFT for each individual grid cell based on a systematicanalysis of observed traitndashclimate relationships using a largeobservational trait database

The traitndashclimate relationships were implemented in theJSBACH DGVM which is part of the Max Planck InstituteEarth System Model (MPI-ESM) (Brovkin et al 2009 Rad-datz et al 2007 Roeckner et al 2003) and is used in thecontext of ESMs and carbon cycle model intercomparisons(Friedlingstein et al 2006) We simulate variation in threeoriginally constant and PFT-specific key leaf traits in JS-BACH for which sufficient georeferenced observational datacould be selected to calibrate the traits These traits were spe-cific leaf area (SLA) maximum carboxylation rate at a refer-ence temperature of 25C (Vcmax25) and maximum electrontransport rate at 25C (Jmax25) To determine relationshipsbetween these three traits and climatic drivers for each PFTcommunity means of trait values from the TRY global planttrait database (Kattge et al 2011) were related to global cli-matic data Based on these relationships between traits andclimate SLA Vcmax25 and Jmax25 were reparameterizedfor each grid cell on a yearly basis depending on the localclimatic conditions in a grid cell Compared to the defaultmodel this enabled enhanced feedbacks between plants andenvironment as traits within natural PFTs could vary dynam-ically in space and time A simulation with variable traits iscompared for trait distribution productivity and vegetationdistribution to a default simulation with the original trait val-ues of the model and to an additional simulation with con-stant but observation-based trait values

As DGVMs are already parameterized to produce ap-proximate realistic results our variable traits simulationwill not necessarily approach reality better than the default



















































25 Data comparison






3 Results








































































4 Discussion

























5 Conclusions





References






















































































































































25 Data comparison






3 Results








































































4 Discussion

























5 Conclusions





References










































































































































25 Data comparison






3 Results








































































4 Discussion

























5 Conclusions





References





























































































































25 Data comparison






3 Results








































































4 Discussion

























5 Conclusions





References








































































































25 Data comparison






3 Results








































































4 Discussion

























5 Conclusions





References









































































































































































4 Discussion

























5 Conclusions





References




























































































































































4 Discussion

























5 Conclusions





References




















































































































































4 Discussion

























5 Conclusions





References






































































































































4 Discussion

























5 Conclusions





References



































































































































4 Discussion

























5 Conclusions





References






















































































































4 Discussion

























5 Conclusions





References














































































































4 Discussion

























5 Conclusions





References






















































































































5 Conclusions





References











































































































5 Conclusions





References























































































































































































































































































































































Documents

Impacts of trait variation through observed trait–climate relationships