GLOBAL BIOGEOCHEMICAL
CYCLES, VOL. 12, NO. 1, PAGES 1-24, MARCH 1998
Evaluation of terrestrial carbon cyclemodelsthrough simulations
of the seasonal
cycleof atmosphericCO2:
First resultsof a modelintercomparisonstudy
M.Heimann,
1G.Esser,
2A.Haxeltine,
3J.Kaduk,
1'8D.W.Kicklighter,
4W.Knorr,
1
G.H. Kohlmaier,
5A.D. McGuire,
6J.Melillo,
4B.MooreIII,7R.D. Ottofi
I. C.Prentice,
3W.Sauf,
1A.Schloss,
7S.Sitch,
3'9U.Wittenberg,
2andG.Wtirth
5
Abstract. Resultsof anintercomparison
amongterrestrialbiogeochemical
models(TBMs) arereported,in whichonediagnostic
andfive prognostic
modelshavebeenrunwiththesamelong-term
climateforcing.Monthlyfieldsof netecosystem
production
(NEP),whichis thedifferencebetween
netprimaryproduction
(NPP)andheterotrophic
respiration
RH, at0.5øresolution
havebeengeneratedfortheterrestrial
biosphere.
Themonthlyestimates
of NEPin conjunction
withseasonal
CO2
flux fieldsgenerated
by theseasonal
HamburgModelof theOceanicCarbonCycle(HAMOCC3)
andfossilfuelsource
fieldsweresubsequently
coupled
to thethree-dimensional
atmospheric
tracer
transport
modelTM2 forcedby observed
winds.Theresultingsimulated
seasonal
signalof theatmospheric
CO2 concentration
extractedat thegridcellscorresponding
to thelocationsof 27 backgroundmonitoring
stations
of theNationalOceanicandAtmospheric
Administration/Climate
MonitoringandDiagnostics
Laboratorynetworkis compared
with measurements
fromthesesites.
TheSimpleDiagnostic
Biosphere
Model(SDBM1),whichistunedtotheatmospheric
CO2concentrationatfive monitoring
stations
in thenorthernhemisphere,
successfully
reproduced
theseasonal
signalof CO2 attheothermonitoringstations.
TheSDBM1 simulations
confirmthatthenorth-south
gradient
intheamplitude
of theatmospheric
CO2 signalresults
fromthegreaternorthern
hemisphere
landareaandthemorepronounced
seasonality
of radiationandtemperature
in higherlatitudes.
In
southern
latitudes,
ocean-atmosphere
gasexchange
playsanimportant
rolein determining
theseasonalsignalof CO2. Mostof thefive prognostic
models(i.e.,modelsdrivenby climaticinputs)includedin theintercomparison
predictin thenorthernhemisphere
a reasonably
accurate
seasonal
cyclein termsof amplitude
and,to someextent,alsowithrespectto phase.In thetropics,however,
theprognostic
modelsgenerally
tendtooverpredict
thenetseasonal
exchanges
andstronger
seasonal
cyclesthanindicatedby the diagnosticmodelandby observations.
The differencesfrom the observedseasonal
signalof CO2 maybecausedby shortcomings
in thephenology
algorithms
of the
prognostic
models
orbynotproperly
considering
theeffects
oflanduseandvegetation
firesonCO2
fluxesbetweentheatmosphere
andterrestrialbiosphere.
1Max-Planck-Institut
fiirMeteorologie,
Hamburg,
Germany.
2Institut
fiirPflanzen6kologie,
Justus-Liebig-Universitfit,
Giessen,
Ger-
1. Introduction
The terrestrialbiosphere
takesup an amountof CO2 equivalent
to about one sixth of the total atmosphericinventoryevery year
Lund, Sweden.
[see,e.g., $chimel,1995]. This carbonflux into the biosphererep4TheEcosystems
Center,
Marine
Biological
Laboratory,
Woods
Hole, resentsgrossprimary production(GPP), that is, carbonfixed in
Massachusetts.
(and not rapidly releasedin photorespiration)by
5Institut
farPhysikalische
undTheoretische
Chemie,
J.W.Goethe- photosynthesis
terrestrialplants.Abouthalf of GPP is returnedto the atmosphere
Universitfit,Frankfurt,Germany.
6Alaska
Cooperative
FishandWildlife
Research
Unit,National
Biolog- asplant(autotrophic)respiration.The remainderrepresents
netpriical Service,Universityof Alaska,Fairbanks.
mary production(NPP), that is, carbonthat accruesto the growth
7Center
forComplex
Systems,
Institute
fortheStudy
ofEarth,
Oceans, of plants.
andSpace,Universityof New Hampshire,Durham.
In the absenceof disturbanceand major climatic fluctuations,
8NowatDepartment
ofPlant
Biology,
Carnegie
Institution
ofWashmany.
3Global
Systems
Group,
Department
ofEcology,
University
ofLund,
ington,Stanford,California.
9NowatPotsdam
Institute
forClimate
Impact
Research,
Potsdam,
Germany
Copyright1998bytheAmericanGeophysical
Union.
Papernumber97GB01936.
0886-6236/98/97GB-01936512.00
annual
netprimary
production
(NPPyear)
isbelieved
tobealmost
inbalance
withannual
heterotrophic
respiration
(RHyear)
bysoil
microorganisms,
sothatannual
netecosystem
prod•iction
(NEPear),which
represents
thenetexchange
between
theterrestrial
iosphereand the atmosphere(NEP = NPP - RH), is relatively
small. However, the ratesof photosynthesis
and detritus/soilorganicmatterdecompositionare underthe controlof differentenvi-
2
HEIMANN ET AL.: EVALUATION OF TERRESTRIAL CARBON CYCLE MODELS
ronmentalfactorsthat are not synchronizedduring the seasonal
cycle. On a weekly or monthlybasis,NEP can thereforefluctuate
overa largerangeof positiveor negativevalues.PositiveNEP (i.e.,
NPP > RH) indicatesa terrestrialsinkfor atmospheric
CO2, whereas negativeNEP (i.e., NPP < RH) indicatesa terrestrialsourcefor
atmospheric
CO2. Thisseasonal
fluctuationof NEP is primarilyre-
time, theyreflectstronglythe local weatherandsoil characteristics
at thetimeof measurement.
Thustheydonotdirectlytestthelargescaleaggregated
flux computations
thataremoreimportantfor the
model'sapplications
to the globalcarboncycle.
An alternativeapproachis providedby the availabletime series
have the greatestamplitude[Fung et al., 1983, 1987, Heimann et
al., 1989]. The role playedby land-atmosphere
fluxesin causing
theseseasonalvariationsis graphicallydemonstrated
by "flying
carpet"diagramsthat showthe contrastin magnitudeand a 180ø
phasedifferencebetweenthe cyclesin the northernand southern
hemispheres
andphaseplotsthat showannuallyrepeatedhysteresis patternsin the relation betweensatellite-derived"greenness"
(asnormalized-difference
vegetationindex(NDVI)) anddetrended
atmospheric
CO2 concentrations
in the northernand tropicallatitude bands[Fung et al., 1987].
Recognitionof the terrestrialbiosphere's
key role in the global
carboncyclehaspromptedthe developmentof severalmodelsto
quantifythemetabolismof terrestrialecosystems
at a globalscale.
Suchmodelsare of two maintypes:diagnosticmodels,whichuse
weekly to monthlyremotesensingdatasuchas composites
of the
normalized-differencevegetationindex (NDVI) as input, and
prognosticmodels, which use only environmental(climate and
soil)dataasinput.The firstdiagnostic
terrestrialbiosphere
(or biogeochemical)models(TBMs) expressed
NPP asa simplefunction
of incidentphotosynthetically
activeradiation(PAR) and NDVI
[Funget al., 1987;HeimannandKeeling, 1989]. More recentmodelstakeinto accountmoremechanistic
aspectsof primaryproduction, including the constraintson the light use efficiency of
evergreenvegetationimposedby low temperaturesand drought
[Potteret al., 1993;Ruimyet al., 1994]. The first prognosticmodels expressed
NPP as an empiricalfunctionof annualcharacteristics of climate (e.g., temperature, precipitation, and
evapotranspiration
[Lieth, 1975]); again, more recentmodelsincludemoremechanistic
representations
of processes
includingexplicit simulationof vegetationfoliage cover and phenologyas
responses
to the environment.Thus the two typesof modelsare
converging,andsomecurrentmodelscanbe run in eitherdiagnostic or prognosticmodeaccordingto whetherthe seasonalvariations
in foliagecoverarepredictedor prescribed.
GlobalTBMs aremultiplyingrapidly,as shownby recentpublications[e.g.,Raichet al., 1991;Melillo et al., 1993;Potteret al.,
clear pictureof the seasonalcycles,latitudinalgradients,interannual variability,and (increasingly)stableisotopecomposition
of
atmosphericCO2 [Keeling et al., 1989, 1995; Conwayet al.,
1994a],all of whicharein principlepredictableby TBMs. However, suchcomparisons
arecomplicatedby theneedto specifyin addition fossil fuel emissionsand to model not only the terrestrial
biosphere-atmosphere
carbonexchangesbut alsothe corresponding ocean-atmosphere
exchangesand the three-dimensional
transportof CO2 by windsfromtheatmospheric
andoceanicexchange
sitesto the remotemeasurementsites.A rigorousevaluationthus
requiresthat the TBMs are linked to accuraterepresentations
of
fossilfuelCO2 emissions,
ocean-atmosphere
seasonal
CO2 fluxes,
and atmospheric
tracertransport[Heimannet al., 1989].Despite
thefactthatneitherof theseadditionalfactorsis knownat present
to sufficientaccuracy,the approach,nevertheless,
providesa consistentframeworkwithin which to assesssomeaspectsof largescalecarbonfluxesas simulatedby TBMs.
Thispaperfocuses
onmeanseasonal
cyclesof atmospheric
CO2
concentration,
asrecordedat 27 backgroundmonitoringstationsof
the National Oceanic and AtmosphericAdministration/Climate
Monitoring and Diagnostic Laboratory (NOAA/CMDL) flask
samplingnetwork[Conwayet al., 1994a] coveringall the major
of CO2 measurements
in the remoteatmosphere,
whichoffer an
sponsible
for theobserved
seasonal
variationsin atmospheric
CO2 appropriatelylargespatialscalefor evaluatingthe aggregated
carcontent,especiallyat highnorthernlatitudeswherethesevariations bonfluxessimulated
by TBMs. The CO2 measurement
datagivea
climatic zones of the earth. It describes the results of a TBM inter-
comparison
in whichsixTBMs (onediagnostic
andfive prognostic
models)have been run under similar protocolson a 0.5ø grid
(55.6x
55.6km2attheequator).
Each
TBMwas
runtoequilibrium
for eachgridelement;thatis, theannualtotalsof NPP andRH are
equal
(NPPyea
r =RH,
year).
Asaconsequence,
theNEPyea
r esti-
matesof the TBMs are not appropriatefor examiningme spatial
patternof the anthropogenically
inducedterrestrialsinkfor atmosphericCO2 believedto be operatingtoday[Schimelet al., 1995].
However,thetemporalNEP estimatesareusefulfor simulatingthe
seasonal
dynamicsof CO2 throughout
thebiosphere.
The spatially
explicitmonthlyresultsof the TBMs (NPP andRH), combined
with equivalentseasonalCO2 flux fieldsgeneratedby the Hamburg Model of the OceanicCarbonCycle (HAMOCC3) [Maier1993;Foley, 1994;Ladekeet al., 1994;Ruimyet al., 1994;War- Reimer, 1993, Six and Maier-Reimer, 1996] and fossil fuel source
nant et al., 1994; Friend, 1995; Woodwardet al., 1995; VEMAP fieldsderivedfrom standardsources[Marland et al., 1989],have
Members,1995;Kaduk and Heimann, 1996]. The modelsare still beenusedto providethe lowerboundaryconditionto a three-diquitediversein theirstructure
(for example,in whethernitrogen mensional
tracertransport
model(TM2) forcedby observed
winds
cyclingis treatedasanactivepathwaycontrolling
NPPor aspassiveconsequence
of a system
drivenby lightandwater);quantitative functions(thereis a wide rangeof valuesusedfor key
parameters,
suchas the temperature
dependence
of respiration);
andresults(modelsdifferconsiderably
in simulated
spatialpat-
(Figure1). The seasonal
cyclesat locations
corresponding
to the
sampling
stations
havebeenextracted
andcompared
withthe(detrended)measurements
at eachstation.The objectivewasto determine the extent to which different models can simulate the
observed
seasonal
cyclesandlatitudinalpatterns.
ternsandseasonality
of NPP aswell asin their simulatedvaluesfor
totalglobalNPP,evenwhenforcedby identicalinputs(Potsdam
1994/1995workshops,
[Lurinet al., 1994])).The existence
of such
2. Model Descriptions
differences
indicates
a needfor evaluation
to discriminate
among
differentformulations.
However,sofar, relativelylittleefforthas
2.1. Ocean Model
beenput into evaluatingthe performance
of TBMs. Field-based
Monthlyfluxesof CO2 betweentheatmosphere
andthesurface
NPP measurements
representa minimal benchmark.However, layer of the oceanwere obtainedfrom a standardrun of the Hamtheyareimprecise,
andbeingrelatedto specificpointsin spaceand burgModel of theOceanCarbonCycle(HAMOCC3) [Maier-Re-
HEIMANN ET AL.: EVALUATION
MeteorologicalAnalyses
OF TERRESTRIAL CARBON CYCLE MODELS
3
..•[Simulated
CO2
Atmospheric
Transport
•'-[Concentration
Model (TM2)
Intercompariso
Observations
from Global
Monitoring
Network
Fossil
OceanicCO2
Exchange
co2
Terrestrial
Bio-
culatedby
sphericCO2
ExchangeFluxes
(calculatedby
HAMOCC3+
TBM)
Fluxes (Cal-
Figure 1. Flowchartof thedataflow andandthemodellinks.
imer, 1993] enhancedwith a recently developedmodel of the
marinebiosphere[Kurz, 1993; Six and Maier-Reimer, 1996]. The
oceancarboncycle modelis embeddedin the globalthree-dimensional, seasonalflow field of the Hamburg Large-ScaleGeostrophic ocean circulation model (LSG) [Maier-Reimer et al.,
1993]. HAMOCC3 describesthe cycling of oceaniccarbonin its
threeinorganicformson a three-dimensional
grid with a horizontal
resolutionof approximately3.5ø and 12 layersin the verticaldimension.The marinebiosphereis describedin the modelby two
organiccarbonpools which representphytoplanktonand zooplankton.Gasexchangeat the surfaceis computedin the modelby
theair-seadifferencein partialpressure
of CO2 (pCO2) multiplied
by a wind speedand temperaturedependentgasexchangeformulation accordingto Liss and Merlivat [1986], scaledto satisfythe
constraintimposed by the global bomb radiocarbonbalance
[Heirnannand Monfray, 1989].
In thisstudywe employedmonthlynet sea-to-airfluxesof CO2
as calculatedby a steady state run of HAMOCC3 with a prescribed,constantatmospheric
CO2 concentration.
The modelcomputedpCO2 fields, which are the major driving force for the
seasonal
air-seaexchange
flux of CO2, showa fair agreement
with
observationaltime seriesand regionalsurveydata [Kurz, 1993].
Furthermore,an alternativevalidationbasedon the seasonalcycle
of oxygenin the atmosphere
alsoconfirmsthatthe model-calculated effectsof the marinebiosphereon the oceancarbonsystemare
reasonablywell simulated,at least in the southernhemisphere
[Kurz, 1993; Six and Maier-Reimer, 1996].
The standardcoarsegrid resolutionemployedin this studyis approximately7.83ø latitudeby 10ø longitudeandninelayersin the
verticaldimension.
Tracertransportis describedin the modelby
meansof specifiedthree-dimensional
time-varyingwindfieldsand
by a verticaltransport
parametrization
representing
verticalmixing
by cumuluscloudsandby turbulentdiffusion.The meteorological
wind fields usedin the simulationexperimentsare basedon 12hourlyanalyses
of theEuropean
Centrefor Medium-Range
Weather Forecastsfor the year 1987.
Thereexistquite a few approaches
to modelatmospherictransport,anda recentintercomparison
amongglobaltransportmodels
[Law et al., 1996] revealed substantialdifferencesin the concentration fields computedfrom the sameprescribedsurfacesourcesby
differenttransportmodels.The TM2 modelin manyrespectsperformedcloseto the "typical" coarse-gridtransportmodel [Law et
al., 1996]. Clearly, an assessment
of the sensitivityof the present
resultswith respectto atmospherictransport,for example,by making useof differenttransportmodels,wouldbe worthwhilebut is
beyondthe scopeof the presentpaper.
For the simulationexperimentsdescribedbelow, each source
wasprescribedas a seasonallyvaryingsurfaceflux and the transportmodelswasrunfor 4 yearsuntilthemodeledatmospheric
CO2
concentration
reachedan approximatlycyclo-stationarystate.The
resultsof the fourth year were usedin the subsequent
analysis.
2.3. Terrestrial Biosphere Models
Six globalterrestrialbiogeochemical
modelswereexaminedin
thisintercomparison
study:the BIOME2 model,theFrankfurtBio2.2. AtmosphericTransport Model
sphere Model (FBM), the High-ResolutionBiosphereModel
The three-dimensional
atmospheric
transportmodelusedin this
(HRBM), the Simple DiagnosticBiosphereModel (SDBM1), the
studyis the TM2 model [Heimann,1995], representinga further SimulatingLand Vegetationand NPP Model (SILVAN), and the
development
of the TM1 modelusedin numerouspreviousatmo- TerrestrialEcosystemModel (TEM). Although all thesemodels
sphericCO2 simulationstudies[e.g.Heimannand Keeling, 1989; simulatetheexchangeof carbonbetweenatmosphere
andterrestriHeimann et al., 1989] and which originatedfrom the GoddardInal biosphere,they use different approaches
for the calculationof
stitutefor SpaceStudiestracermodel[Russelland Lerner, 1981].
NPP. The diagnosticmodel includedin this study(SDBM1) uses
TM2 solvesthe continuityequationof an atmosphericconstituent monthlyNDVI data to estimateabsorbedradiationwhich is then
on a three-dimensional
Eulerian grid spanningthe whole globe. translatedinto NPP throughan efficiencycoefficient.In contrast,
HEIMANN
ET AL.: EVALUATION
OF TERRESTRIAL
the five prognosticmodelseithercalculateNPP directlyfrom statistical relationsbetweentemperatureand precipitationwhich is
thenmodifiedas functionsof soil characteristics
(HRBM), or they
estimateNPP as the differencebetweenGPP and autotrophicrespiration (BIOME2, FBM, SILVAN, and TEM). In HRBM,
SDBM1, and TEM, climate data enter into the flux calculations as
monthlymeans,while BIOME2, FBM, andSILVAN applysimple
algorithmsto interpolateto daily valuesfrom the monthlyclimate
dataanddeterminefluxeson a daily time step.All modelsconsider
carbonandwater cycling,but the TEM alsoincludesnitrogendynamics.Among the TBMs, phytomassis representedby different
numbers of carbon pools (BIOME2, 3; FBM, 2; HRBM, 4;
SDBM1,0; SILVAN, 3; andTEM, 1). Grossphotosynthesis
(GPP)
is eithercalculatedwith a multiple limitation approach(FBM and
TEM) or by modifiedversionsof theFarquharmodel[Farquharet
al., 1980](BIOME2 andSILVAN). AlthoughRH in all modelsis
calculatedby relationshipsthat dependon soil carbon,temperature, and soil moisture,either the formulationsof the relationships
or theparameters
of similarformulationsdifferamongthemodels.
The brief modeldescriptionsgiven in the followingsubsections
2.3.1. to 2.3.6. will concentrateon thehistoricdevelopmentandthe
main featuresand assumptionsof each model that determinethe
computation
of theCO2 exchangefluxesontheseasonal
timescale.
A shortmodelcomparisonis givenin Table 1; moredetailedinformationcanbe foundin thereferencescitedat thebeginningof each
subsection.
2.3.1. BIOME2.
CARBON
CYCLE
MODELS
dependson temperature(T), ambientcarbondioxideconcentration
(CO2) andthe day length(d). Foliar projectedcover(FPC) is the
variable used to define leaf area in B IOME2.
It is measured as the
fractionof groundcoveredby foliageverticallyaboveit (i.e., valuesrangingfrom 0-1). FPC is calculatedfrom one-sidedleaf area
index(LAI) usingBeer'slaw [Monsiand Saeki,1953]:
-LAI?2
FPC = 1 - e
(2)
An optimalvalueof FPC is predictedfor eachmodelgrid square,
andthisoptimalvalue of FPC is usedin the final NPP calculation.
The potentialGPP is thenreducedby a scalarin orderto take
intoaccount
theeffectsof drought
stress
(•D) andtemperature
extremes
( •r ):
GPP - GPPma
x •a •T
(3)
Finally,netprimaryproduction
is calculated
by subtracting
whole
plantrespiration
costs,whicharecurrentlyestimated
asbeing50%
of GPPma
x.
Heterotrophic
respiration(RH)is calculatedfrom a relationship
derivedby Lloyd and Taylor [1994] and Howard and Howard
[1993],whereCO2 evolutionis relatedto bothmonthlysoiltem-
peratureandaveragemonthlysoilmoisturecontent.Monthlysoil
temperature
wasderivedfromair temperature
assuming
an attenuationfactoranda phaselagcomputed
by thepropagation
of anannual heatwave throughsoil [Campbell1977].RH is calculated
usingthefollowingrelationship:
BIOME2 [Haxeltine et al., 1996, Haxeltine
andPrentice,1996] is a coupledcarbonandwater-fluxsimulation
modelthathasbeendesignedfor usein predictingtheresponseof
global natural vegetationto changingclimates.Thus the model
doesnotusea mapof globalvegetationasa forcingdatasetbut insteadpredictsglobal vegetationpatternsas a model output.The
BIOME2 hasa simpletwo-layer hydrologymodelwhich allows a
realisticsimulation
of thedroughtstress
experienced
by vegetation
anda mechanistic
carbonbalancemodel.Modeloutputconsists
of
predictions
of NPP, leaf areaindex,andvegetation
type,whichis
thecombination
of planttypeswhichmaximizes
wholeecosystem
netprimaryproduction
(NPP)at a particularsite.Woodyandgrass
planttypesaredifferentiated
according
to theirphysiological
and
phenological
attributesandrootingstrategies.
The modelingapproach
hasbeento producea mechanistic
model of naturalecosystems
which,nevertheless,
hasa level of com-
plexityappropriate
for useat theglobalscalewithmonthlyclimate
data.Themonthlyclimatefieldsarelinearlyinterpolated
yielding
quasi-dailyvaluesfor usein the water-fluxmodel[Haxeltineet al.,
1996].
Rn -- Otf2(rsoil)f3(Wsoil)
(4)
whereWsoil
is thesoilmoisture.
Thevaluea canberegarded
asthe
productof thesoildecayrate,whichis dependent
onthesoiltype,
and the mass of soil carbon. An initial estimate of a was subse-
quently scaledsuch that the steadystate constraintcondition
(NPPyea
r --RH,year)
wassatisfied.
Thus,
monthly
heterotrophic
respirationestimateswere obtained.This methodavoidsthe needto
explicitlydefinethevarioussoildecayratesandalsotheneedfor
estimatingthe soil carboncontent.
2.3.2. Frankfurt BiosphereModel (FBM). The FBM [Kindermannet al., 1993; Liideke et al., 1994; Kohlmaier et al., 1997]
consistsof a mechanisticcarbonmodel and a simpleone-layer
bucketmodelfor soil moisture.Litter and soil organiccarbonare
combinedin one soil compartment.
Vegetationis represented
by
two compartments
to which assimilatedcarbonis allocatedby an
allometricrelation.Model outputincludesdaily carbonexchange
fluxes,leaf area,phytomassandsoil carbon.
The two vegetationcompartmentsare greencarboncompartment(GC) comprisingleavesandfeederrootswith a lifetimeof up
to about4 yearsanda residualcarboncompartment
(RC) whichincludesthecarbonmassstoredin branches,
stem,androots.The dynamicsof all threecarboncompartments
resultfromthedailyinput
and outputfluxes as follows:
1. Carbonassimilation(GPP) is calculatedusinga factorialapproachin which a vegetationtype specificmaximumassimilation
Photosynthesis
is calculated
usinga semimechanistic
lightuse
efficiency(LUE) model.TheLUE modelisbasically
anoptimized
versionof theFarquharphotosynthesis
model[Collatzet al., 1991,
1992].Instead
of prescribing
valuesfortheparameter
Vm(themaximumcatalyticcapacityof theenzyme"ribulosebiphosphate
carboxylase
oxygenase"
perunitleafarea),an optimalvaluefor Vm
givena particularvegetation
type(C3 or C4 plants)andsetof enrate(GPPmax)
ismultiplied
bydifferent
factors
forthedependence
vironmental
conditions
is calculated.
The resultis an equationfor
on leaf areaindex (LAI) (calculatedfrom leaf massanda vegetaphotosynthesis
whichis linearlydependent
onabsorbed
photosyn- tiontypedependentspecificleaf area),incidentphotosynthetic
actheticallyactiveradiation(APAR) suchthat
GPPma
x = FPCfl(CO 2,T, d) APAR
(1)
where GPPma
x is the potentialor "maximumpossible"GPP,
fl(CO2,T,d)represents
the semimechanistic
LUE equation
which
tive radiation (PAR), air temperature(T), the ratio of actual to
potentialevapotranspiration
(AET/PET), and atmosphericCO2
concentration
AET
GPP= GPPmaxf4(PAR,
LAI)fs(r) p--•-•
f6(CO2) (5)
HEIMANN ET AL.' EVALUATION
OF TERRESTRIAL CARBON CYCLE MODELS
<
<
<
<
E
Z
Z
Z
Z
•o
<C
<C
<C
<C
'•
Z
Z
•
Z
Z
Z
Z
Z
Z
0
o
.,•
o
o
.,•
o
o
<C
.,•
-
'{::1
o
Z
(J
.•
ß
=
o
•
•.
• • '-=
•
o
o
Z
Z
Z
Z
Z
"'
O'
•
•
Z
•
o
5
o
o.
o
o
ø,
o
6
HEIMANN
ET AL.: EVALUATION
OF TERRESTRIAL
CARBON CYCLE MODELS
2.3.3. High-ResolutionBiosphereModel (HRBM). The
2. Autotrophic
respiration
fromGCandRCwhichdepends
on
compartment
sizeandanexponential
function
of temperature
with HRBM is thesuccessor
of theOsnab•ck Biosphere
Model [Esser,
a constant
Qlovaluespecified
foreachvegetation
type.
1987, 1991]. It represents
thebiosphere
by ninecarbonpoolsand
3. Litterproduction
(LP) is proportional
to thecompartmentin its version3.0 includesmodulesdealingwith vegetationfires
sizewitha constant
coefficient
except
fortheGCcompartment
of and humaninfluencesthroughland use changesand emissions
deciduousvegetationwhich will shedthe leaveswithin about
fromfossilfuel andfromindustry.A detaileddescription
is given
1 month.
by Esseret al. [ 1994].
4. Heterotrophic
respiration
(RH)usesthetemperature
depenThe nine biosphericcarbonpoolsare herbaceous
and woody
dence
of Funget al. [1987],lineardependence
oncompartmentlive phytomass,litter from herbaceousandwoodymaterial(dead
size,anda soilmoisture
factoranalogous
to themoisture
depen- phytomass),
andsoilorganiccarbon,withthelive anddeadphytodenceof photosynthesis.
masspoolssubdivided
into aboveground
andbelowground
pools.
Thephenological
behavior
of themodelisgoverned
bya setof The biosphericfluxesare net primaryproductivity(NPP), litter
rulesforcarbon
partitioning
ofevergreen
anddeciduous
vegetation production
(LP), soilorganiccarbonproduction
andheterotrophic
types.Theleafshooting
phasestartswhenthecarbongainfrom respiration
(RH).NPP iscalculated
individuallyfor eachof thefour
photosynthesis
isgreater
thanthecarbon
losses.
Thesystem
allo- compartments
of live phytomass.
catesmostof theassimilates
to theGC compartment
untilGC corresponds
toRC asdefinedby anallometric
relation[Janecek
etal.,
1989]
b
RC -- a GC'
(6)
In the HRBM, empiricalmodelsareusedto calculateannualnet
primary
production
(NPPyear)
[Esser,
1987,1991
] andlitterproduction
(LPyear)
[Esser,
1991
], which
areseasonally
distributed
afterward.
The distribution
of NPPyea
r followsmonthlyactual
evapotranspiration
(AET), whichis derivedfrom a simplebucket
(with a andb dependent
on vegetationtype), which statesthat a
minimumamountof RC is requiredto supporta given massof
leavesand feederroots.During the secondary
growthphasethe
systemis forcedto allocatesimultaneously
into the GC and RC
compartments
in sucha way that(6) remainsvalid. The endof the
vegetationperiod is defined by unfavorableweatherconditions
which result in biomass decrease.
For deciduousvegetationtypesa leaf abscission
phasefollows
in which the GC compartmentis reducedto a residualamount
which is definedby the function
model
[Prentice
etal.,1992].
Thedistribution
ofLPyea
r fromherbaceousphytomassfollows the relativedecreaseof AET between
2 months.In biomeswith seasonallitter fall ("temperatedeciduous," "cool mixed," "cold mixed," "cold deciduous," and "tun-
dra"),litterproduction
startswhenmonthlyair temperature
drops
to half of themeanmonthlytemperature
of thewarmestmonth(on
thecentigrade
scale)andreduces
theherbaceous
phytomass
byhalf
within 14 days.In contrast,the litter productioncoefficientfrom
woodyphytomassis assumedto be 1/12 of the annualvalueeach
month.
The litter decomposition
flux is proportionalto the sizeof the
litterpool.Thevalueof thecoefficientdepends
onthecomposition
(with c andd dependent
on vegetationtype).In thefollowingdor- of thelittermaterialandonclimate.While decomposition
increasmancyphasethe carbonlosses(litter productionand autotrophic esexponentially
with air temperature,
thedependence
on precipirespiration)aredistributedamongthe compartments
sothat (7) retation is a skewedmaximumfunctionwith an initial steep
mains valid.
decomposition
increase
with increasing
precipitation
followedby
For evergreenvegetationtypes the vegetationperiod is fola slowerdecrease
in decomposition
at highprecipitation
rates.The
lowed by a standbyphasein which the lossesof GC andRC are maximumdecomposition
rate is alsodependent
on temperature:
characterized
by constant
proportions.
All flux equations
contain highertemperatures
shiftthemaximumdecomposition
to higher
free parameters
which are determinedthroughcalibrationfor a precipitationrates.
steadystatesuchthatthemeanannualgrossandnetprimaryproThe primaryfactorsthatcontrolseasonal
CO2 exchanges
beare the seasonal
air temductivityandheterotrophic
respiration
of a vegetation
typeequal tweenthe biosphereandthe atmosphere
averageecologicalestimates.
The modelrequireshourlyvaluesof peratureandprecipitation
signalwhichaffectmonthlyNPP,LP,
temperature
andradiationaswell asdailyprecipitation
anda vari- anddecomposition.
In thisintercomparison
experiment
themodety of staticdataaboutvegetation
andsoil.Daily precipitation
is ulesthatreferto landuseandvegetation
fireshavenotbeenconobtainedfromthemonthlyvaluesby meansof a smoothredistribu- sidered.
tion.In a similarway the monthlymeantemperature
andmean
2.3.4. SimpleDiagnosticBiosphereModel (SDBM1). The
cloudiness
are interpolated
to daily values.The hourlyvaluesof SDBM1[KnorrandHeimann,1995]is a simple,
diagnostic,
glotemperaturearegeneratedwith thehelpof a sinusoidalfunctionus- ballyuniformmodelof monthlyCO2 exchange
between
theatmo-
RC= c GCd
(7)
ingindependent
estimates
of dailytemperature
rangeforeachvegetationtype.The percentage
cloudiness
is constant
overtheday,
andthediurnal
cycle
ofradiation
iscalculated
using
the•ngstr6m
relationship.
It isthenpossible
to calculate
theseasonal
courses
of LAI, phytomass,GPP,NPP, andRH in a steadystate(NPPyear -- LPyear --
sphere
andtheterrestrial
biosphere.
ModeledNPPandRH are
computedfrom remotesensingandclimatedata.The modelversionemployedin this studyis the formulationI describedin the
KnorrandHeimann[1995]work,in whichbothNPPandRH dependalsoonwaterstress
computed
usinga simplebucketmodel.
Carbonuptakeby vegetation
(NPP) is calculated
fromestimates
RH,¾ear)
aswellasthelong-term
development
ofnonclimax
vegetation.
of photosynthetically
activeradiation
(PAR)absorbed
by vegetation,whichis determined
froma combination
of bi-weekly
maxi-
An applicationof the FBM to differentclimatescenariosfor the
northernforestbiomesis described
by Liidekeet al. [ 1995]andthe
ma of the normalized-difference
vegetationindex(NDVI) of the
NOAA 9 satellite(1985-1989)[Gallo, 1992]andcloudiness
data,
validationof themodel-predicted
phenology
usingNDVI datais
regridded
fromtheoriginalmercator
projection
grid(approximate-
given in Liideke et al. [ 1996].
ly 10' resolutionat theequator)to the0.5øx0.5ø standard
latitude-
HElMANN ET AL.: EVALUATION OF TERRESTRIAL CARBON CYCLE MODELS
7
longitudemodel grid. A one-layerbucketmodel calculateswater mean daily temperaturesabove 5øC since December 1 for the
stressas the ratio betweenactualand potentialevapotranspiration northernand June 1 for the southernhemispheres.)Relationsof
(AET/PET) from monthlymeansof precipitation,temperature,and thistypehavebeenshownto describethedateof budburstin temcloudiness.
peratebiomesratherwell [Murreyet el., 1989].The biomedependentparametervaluesin theformulaehavebeendeterminedby an
Specifically,NPP is calculatedfrom
evaluationof satellitedata. In contrastto tropical vegetationthis
NPP= a 1.222
(NDVI_
0.1566
PAR (8) modelingapproachimpliesa dormancyperiodfor the temperate
•,0.559
P•
vegetation,sincegrowthonly startsafter the temperaturesumexwherePAR is assumedto be half of meanmonthlyglobalradiation
AET
atthesurface
[W m'2].Carbon
release
fromsoilsisassumed
todependon relativeevapotranspiration
andexponentiallyon air temperature,T, with no cutoffat low freezingtemperatures
ceeds a threshold.
The allocationstrategyemployedin SILVAN aims at maximization of NPP. Allocation of carbon is directed from the assimilate
pool to the leavesuntil an optimalLAI is reached.This optimal
LAI is definedby resultingin maximalNPP giventhe currentenvironmentalconditionsandcurrentplant state.Furtherassimilates
rzandQlo aregloballyuniformparameters
determined
fromfitting are allocatedto wood accordingto an allometricrelationof sapthemodel-derived
seasonal
cycleto atmospheric
CO2 observations woodandLAI aslongas thisleadsto an increasein NPP thereby
from five key stationsin the northernhemisphere,selectedto cover increasingthe potentialLAI that the vegetationcan supportwith
approximatelyuniformly the major regionsin temperatelatitudes waterby increasingsapwoodcross-sectional
area.Excessassimilates are directed to dead structural wood.
with stronglyseasonal
vegetation(Qlo = 1.5 andct= 0.68 gC (MJ
PAR)-I).Thevalueoftheparameter/l
isnotglobally
uniform,
but
No a priori limits are imposedon the simulatedLAI. Typical
is determined
fromthecondition
of equilibrium
(NPPyea
r = meanannualLAI for a grid cell (biomeaverage)rangesfrom 0.2
to 3.8 dependingon the biome.Intrabiomevariationis typically
RH,year).
2.3.5. Simulating Land Vegetation and NPP Model (SILlow becauseof similar climate and constantbiological parameter
r/10
AET
RH= [lQlO ( p• )
(9)
VAN). SILVAN [Kaduk and Helmann, 1996, Kaduk, 1996] con-
values.
sistsof a mechanisticcarbonmodel and a simpleone-layerbucket
modelfor theestimationof the waterbalance.The biosphereis representedby threepoolsof living biomass(assimilates,herbaceous,
and woody phytomass)and three pools of dead biomass(herbaceousand woody litter and soil organiccarbon).Model outputincludes daily exchange fluxes of carbon, LAI (single sided),
In all biomesa low continuousleaf mortality is in effect at all
times,mimickingherbivoreconsumption
and decayof photosyntheticpotentialthroughoutthe growingseason.Beyondthis base
mortalityrate,leaf andfine root sheddingin tropicalbiomesfollowsthe productionconditions;thatis, if AET/PET > 0.4 andthe
currentLAI is higherthanthe optimalLAI estimatedfor the time
step,a fractionof theexcessherbaceous
biomassis shed.If AET/
biomass, and soil carbon.
The assimilatepool representscarbohydratesserving as reservesfor theplants.Herbaceousbiomassincludesleavesandfine
roots,woodybiomasssapwood,andheartwoodof stemsandroots.
Fixed fractionsof the herbaceousandwoody carbonpoolsare consideredas belowgroundcarbon.Leaf areaindex is linearlyrelated
to abovegroundherbaceouscarbon and constrainedby growing
conditionsand the diagnosticallycomputedsapwoodcross-sec-
PET<0.4, then all herbaceousbiomassis shed within 2 weeks. In
tional area.
ousversionsof the HRBM [Esser,1991]; heterotrophicrespiration
is the productof theratesandcurrentpool sizes.
In summary,SILVAN featuresa fully climate-drivenphenology using simpleempiricalfunctionswhich are basedon results
from field experiments.
Note thatin SILVAN phenologyandallocationarenotonlycrucialfor productionbut alsofor heterotrophic
respiration,astheydeterminetimingandamountof litter produc-
Assimilationat the leaf level is simulatedby applyinga simplified versionof thephotosynthesis
modelof Farquhar et al. [ 1980].
Leaf assimilationis thenscaledto the canopywith the currentleaf
areaindexand a fixed light extinctioncoefficientof k -- 0.5. Integratingthediurnalcycleof temperature
andPAR resultsin nonwater-stresseddaily assimilation.This rate is then adjustedby the
ratio of actualto potentialevapotranspiration.
Assimilatedcarbonis storedin the assimilatepool andusedfor
respirationand growth. Maintenanceand growth respirationare
simulatedfor leaf, fine root, and sapwoodcompartments.
Maintenancerespirationratesare tied to the maximal assimilationrate.
The temperaturedependence
of maintenancerespirationis simulated by an Arrheniusrelationshipadjustedto be equivalentto a
Q•o of 2 at the optimaltemperature
for net assimilation.
Growth,startingwith budburst,is triggeredin tropicalbiomes
whenproductionconditionsare good,that is, whendroughtstress
is low (AET/PET>0.4). Thus the model doesnot imposeany dormancyfor the tropicalvegetation,hencewhich may respondinstantaneouslyto improved weather conditions. In temperate
biomes,growthbeginswhen the growingdegreeday temperature
sumexceedsa certainthresholddependingon chillingdaysandthe
biome.(The growingdegreeday sumis definedby the sumof the
temperature
biomes,sheddingof herbaceous
biomassis initiated
when daily minimumtemperaturedropsbelow 0øC and it is assumedto last 1 month.Woody litter productionproceedswith a
fixed rate throughoutthe whole year.
Finally, litter decomposition
rates are modeledby empirical
functionsdependingon temperatureandprecipitationas in previ-
tion.
2.3.6. Terrestrial EcosystemModel (TEM). The Terrestrial
EcosystemModel (TEM) is describedin detail by Reich et el.
[ 1991], McGuire et el. [ 1992, 1993, 1995, 1997], andMelillo et el.
[1993]. To date, TEM has been usedto examinepatternsof NPP
for potentialvegetationin SouthAmerica[Reichet el., 1991] and
North America [McGuire et el., 1992] and to examinethe potential
response
of NPP andcarbonstorageto climatechange[McGuireet
el., 1993, 1995, 1996, 1997;Melillo et el., 1993, 1995;Joyceet el.,
1995]. In this studywe useversion4.0 of TEM [McGuire et el.,
1995].
The TEM is a process-based
ecosystemsimulationmodelthat
usesspatiallyexplicitdataon climate,soiltexture,vegetation,and
wateravailabilityto make monthlyestimatesof importantcarbon
andnitrogenfluxesandpoolsizes.The TEM is a highlyaggregated
modelin whichterrestrialecosystems
are representedby oneveg-
10.0
5.0
0.0
-5.0
-10.0
'
'
'
'
i
J,
,
,
,
i
,
,
!
i
'
'
I
I
i, •
.....
,
I
'
'
I
I
I •
CHF1
2.0
I
I
SMO
T•:
.1."1'
-2.0
-4.0
....
'
' , '
.......
' .....
' ',
, ,
1.0
0.5
.....
¾•..........
]I -0.5
0.0
--- o•n
--
SDBMN
-•
SDBM
S
-1.0
•
SDBM T
•--- ½om_p.o_._sit•_]
-1.5
Jan
Jul
Jan
Jan
Jul
Jan
Jan
Jul
Jan
Plate 1. Comparison
of theobserved
seasonal
cycleof CO2 at nineselected
monitoring
stations
withthesimulated
seasonal
cycleproduced
by couplingthemonthlyestimates
of netecosystem
production
by theSimpleDiagnostic
Biosphere
Model(SDBM1)andfossilfuel emissions
withtheHamburgoceanandatmospheric
transport
models.
SDBM1 N, SDBM S, and SDBM T denotethe contributionsfrom NEP north of 30øN, southof 30øS and from the
tropics(30øS-30øN),
respectively.
Meanandstandard
deviation
areshownfor theseasonal
cyclederivedfromthe
NOAA dataset[Conwayet al., 1994]asdescribed
in thetext.Thefirst6 monthsof eachcyclearedisplayed
twiceto
revealtheannualvariationmoreclearly.
x M BC (Obs.)
x ALT (Ohs.)
10.0
x KTL(Obs.)
_
0.0
-10.0
CBA (Ohs.)
x STM (Obs.)
10.0
.
--
0.0
-lO.O
10.0
:----BIOME21
.
. xSHM
(Ohs.)[ %CMO
(Obs.)
.
0.0
----FBM
.----HRBM
---TEM
I---- SILVAN
0.0
.
L---SDBM___j
Jan
Jul
Jan
Jan
Jul
Jan
Jan
Jul
Jan
Plate2a. Comparison
of theobserved
seasonal
cycleof CO2 withthesimulated
seasonal
cycleproduced
bycoupling
the monthlyestimates
of net ecosystem
production
by the six terrestrialbiosphere
models(TBMs) andfossilfuel
emissions
with the Hamburgoceanandatmospheric
transport
modelsfor northernmonitoringstations.
The first 6
monthsof eachcyclearedisplayedtwiceto revealtheannualvariationmoreclearly.Meanandstandard
deviations
are shown for the observed data.
HElMANN
ETAL.:EVALUATION
OFTERRESTRIAL
CARBON
CYCLE
MODELS
9
8.0
xNWR
(Obs.)
4.0
xMID(Ohs.)
KEY
(Obs.)
-•Ar•.
.
[%,
-4.0
--,-
-8.0
4.0
2.0
.•.
o
••LO
(Ob•.) •
'
•KU•
(Ob•.)
'
o.o
-2.0
O
Z•..
-4.0
2.0
-i-BIOME2
----FBM
•O•l
(0•.)
'
'•
' I //••,RPB
(Obs.) xCHR
(Obs)
--
i
0.0
----HRBM
i•TEM
,,----SILVAN
--iSDBM
-2.0
':..
-4.0
Jan
i
Jul
I I I ]
Jan
ß ß
Jan
i
Jul
• ,,
i
ß ::,,_
Jan
,•
,, , I ,
Jan
,i , i I
Jul
i
,,
Jan
Plate 2b. As in Plate2a but for tropicalmonitoringstations.
2.0
xSEY(Obs.) "•-"•'
1.o,•
.•
-1.0
-2.0
.
•Asc
(Ohs.)
'•i,
x SMO (Obs.)
-
ß
ß d • , ß I
2.0
d ß d • ß I
ß ,
1.0
0.0
-1.0
-2.0
1.5
r
[IiBIOME2-I 0.5
/ "'-'FBM J
i"'' HRBM
/
j•-• TEM /
-0,5
xHBA
(Ohs.)
1
xSPO
(Obs)
"]•..• "'"'"'
ß
!I SILVAN
i -1.5
L_
_I.__S_D_.B•M_
!
Jan
dul
Jan
dan
dul
dan
dan
Plate2c. As in Plate2abutfor southern
monitoring
stations.
Jul
dan
i d
10
HEIMANNET AL.:EVALUATIONOFTERRESTRIAL
CARBONCYCLEMODELS
etationcarbonpool, two vegetationnitrogenpools(structuraland
labile), one soil organiccarbonpool, and two soil nitrogenpools
(organicandinorganic).Hydrologyfor TEM is determinedby the
water balancemodel of Vtrtsmarty et al., [ 1989].
Parametersin TEM are vegetation-specific,soil-specific,or
constant.The vegetationusedby TEM in this studyis an updated
versionof the vegetationgiven by Melillo et al. [ 1993]. Although
mostof the vegetation-specific
parametersaredefinedeitherfrom
the literatureor from statisticalanalysesof data,the rate-limiting
parametersin mostflux equationsare determinedby calibration.
The calibrationof theseparameters
is onetechniquefor addressing
issuesof temporaland spatialscalein the model [seeRastetteret
al., 1992]. In TEM the rate-limitingparametersare determinedby
calibratingthemodelsothatit estimatestheequilibriumfluxesand
poolsof anintensivelystudiedfield site,thatis, thecalibrationsite
for thevegetationtype.Data usedto calibratetherate-limitingparameters of version 4.0 are documented in Table 1 of McGuire
et
al. [1995].
The seasonality
of carbonexchangebetweentheterrestrialbiosphereandthe atmosphere
is influencedby differencesin the seasonalityof NPP and RH. In TEM, NPP is calculatedas the
differencebetweenGPP andplantrespiration.A numberof factors
influencethe seasonality
of GPP, whichis calculatedmonthlyas
tween the minimum and optimum temperatureof photosynthesis
[seeRaich et al., 1991] and remains at 1 until it reachesthe maximum temperature
of photosynthesis
whereit decreases
parabolicallyto 0 [seeMcGuireet al., 1995].The minimumandmaximum
temperatures
of photosynthesis
are vegetation-specific
parameters
thataredefinedby thegrowingseasonlimitsof thevegetation
type
[seeMcGuire et al., 1995].The optimumtemperatureof photosynthesisis grid-cell-specific
andis definedasthemonthof maximum
relativeleaf phenologyfor thegrid cell [seeMcGuire et al., 1995].
Thusf8 tendsto increaseGPP mostduringmonthswhenthetemperatureexceedsthetemperature
of themonthwith maximumleaf
area.
Thefunctionf9(CO2,H20)
describes
thesimultaneous
effectsof
atmospheric
CO2 and wateravailabilityon C assimilation.
This
functioncalculatesa scalarthat increaseshyperbolicallyfrom 0 to
1 astheintercellularconcentration
of CO2 increases[seeRaichet
al., 1991]. IntercellularCO2 is theproductof atmospheric
CO2 and
relativecanopyconductance,
whichis approximately
a linearfunction of theratioof actualestimatedevapotranspiration
to potential
evapotranspiration
(EET/PET) [seeRaich et al., 1991;McGuire et
al., 1992].Becausetheseasonal
variationof atmospheric
CO2 is a
smallfractionof theatmospheric
CO2 concentration,
theseasonality of f9 is primarilydetermined
by the seasonality
of EET/PET,
follows
whichis highestin wet monthsandlowestin dry months.
The functionflo(NA) describeshow carbon-nitrogen
statusof
GPP-- GPPma
x LEAF f7(PAR)f8(T)f9(CO2,H20) flo(NA) (10) the vegetation,i.e., nitrogenavailabilityor supply,influencescarwhereGPPma
x is themaximumrateof C assimilation,
PARispho- bon assimilation.Nitrogensupplyis the sum of nitrogenuptake
tc•synthetically
activeradiation,LEAF is leaf arearelativeto max- plusnitrogenmobilizedfrom the vegetationlabile nitrogenpool
[McGuireet al., 1993].Nitrogensupplyis generallygreatestearly
imum leaf area, T is monthlymean air temperature,CO2 is the
in
thegrowingseasonwhenvegetationis ableto mobilizenitrogen
atmospheric
concentration
of carbondioxide,H20 is wateravailfrom
storage.Nitrogenuptakedependson temperature,
soilmoisability,andNA is nitrogenavailability.The parameterGPPma
x is
therate-limitingparameterin the GPP equationandis determined ture, and inorganicnitrogenin the soil solution[seeRaich et al.,
1991] and increasesfor higherair temperature,highersoil moisby calibration.Of the severalfunctionsin theGPPequation,LEAF
ture,
and higherinorganicnitrogen.Net nitrogenmineralization,
playsa majorrole in the controllingthe seasonality
of C assimilawhich
dependsin parton decomposition,
replenishes
inorganicnition.This functionsimulatesrelativechangesin thephotosynthetic
andhighcapacityof maturevegetationfrom estimatedactualevapotranspi- trogen,andtendsto be greaterfor higherair temperature
er soil moisture.Thus, the seasonalityoffl o is influencedby the
rationandthe previousmonth'sphotosynthetic
capacity[Raichet
dynamicsof vegetationnitrogenstorage,nitrogenuptake,andnial., 1991]:
trogenmineralization.
In TEM, the calculationof plant respirationconsidersboth
LEAFj= a(EETj/EETmax)
+ b(LEAFj_l) + c (11)
maintenance
andconstruction
respiration.Maintenancerespiration
where
LEAFjistherelative
leafphenology
inthecurrent
month
j, (Rm)is calculatedasfollows:
EETjis theestimated
actualevapotranspiration
in thecurrent
month,EETmax is the maximummonthlyevapotranspiration,
and
LEAFj.1isrelative
leafphenology
intheprevious
month.
Notethat
LEAF is constrained to be between 1.0 and a minimum relative leaf
phenology.The parametersa, b, and c are vegetation-specific
parameters[seeMcGuire et al., 1992] thataredeterminedby regressing the relative leaf phenologyof the currentmonth with EET/
EETmax and the previousmonth'sleaf phenologyfor a site with
seasonaldataon leaf area.Relativeleaf phenologyis calculatedas
rTT
Rm -- Kr Cv e
(12)
whereKr is theper-gramrespiration
rateof vegetation
biomassat
0øC, Cv is the massof carbonin the vegetation,T is the mean
monthlyairtemperature,
andrTis theinstantaneous
rateof change
in respiration
with air temperature.
The parameter
Kr is theratelimitingparameter
in theRmequation
andisdetermined
bycalibra-
the ratio of leaf area index in the current month to the maximum
tion.Theparameter
rTisequalto ln(Q1o)/10whereQ•oistherate
of changeof respiration
dueto a 10øCincrease
in temperature.
In
monthlyleaf areaindex.
TEM, Qlo is modeledasfollows[seeMcGuire et al., 1992]:
The functions
f7, fs, f9, andflo alsoinfluencetheseasonality
of
C assimilation.
The functionf7(PAR)describes
theeffectof photosynthetically
activeradiationonC assimilation.
Becausef7
is a scalar function that increaseshyperbolicallyfrom 0 to 1 as PAR
Q•o-- 2.35665
+ 0.05308
T + 0.00238
T2- 0.00004
T4
(13)
whereT is meanmonthlyair temperature.
This functionapproximatesa linearincreasein Q•o from 1.5 to 2.0 between40ø and
increases [see Raich et al., 1991], it tends to increase GPP more
20øC,a constantQ•o of 2 between20ø and5øC,anda linearinduringmonthswith greatersolarradiationat thetopof thecanopy. creasein Qlo from 2.0 to 2.5 between5ø and 0øC [seeLarcher,
1980].The seasonality
of Rmis determinedprimarilyby the seaThe functionf8(T) describes
the effectsof air temperature
on C
in T andsecondarily
by theseasonality
in Cv. Theseasonassimilation.This functionincreasesparabolicallyfrom 0 to 1 be- sonality
HElMANN ET AL.: EVALUATION OF TERRESTRIAL CARBON CYCLE MODELS
11
alityof construction
respiration,
whichis calculated
as20% of the
For eachstationrecordthefollowingprocedurewasusedto exdifference
betweenGPPandRm,depends
ontherelativeseasonal- tract the seasonalcycle of atmosphericca 2 from the station
ity of GPP andRm.
A numberof factorsinfluencethe seasonality
of RH, whichis
records.First,a smoothcontinuous
representation
of theinterannual ca 2 concentration
trendwasobtainedby interpolating
theannucalculated as follows:
al meanconcentrations
with Hermitecubics.Therebythe needed
firstderivativesof thetrendfunctionatthecenterof eachyearwere
(14)
RH= Kd sac fll(T) f12(Srn)
estimatedby centereddifferencesfrom the annualmeanconcentrationsof the adjacentyears.Second,trend-corrected
monthlyconwhereKd is the per-gramrespirationrate ofsoil organiccarbon
centrationvalueswere calculatedby subtractingthe interannual
(sac) at 0øC, T is meanmonthlyair temperature,
andSmis voluca 2 trendfunctionfromthemonthlyobservations.
Third,monthly
metricsoilmoisture.The parameterKd is therate-limitingparamestimatesof the seasonal
cycleandof its variabilitywereobtained
eter in the RH equationand is determinedby calibration.The
by computingaverageand 1 standarddeviationof all trend-correctseasonality
of RH is primarilydetermined
by the seasonality
ofJ)l
ed January,February,etc., valueswithin the time window 1983andJ)2. The functionJ)l(T) describes
the effectof meanmonthly
1992.The standarddeviationsreflectthe interannualvariabilityof
air temperatureon heterotrophicrespirationand is modeledas an
the seasonalcycle within the selectedtime window. A total of 27
exponential
functionof T with a Q•o of 2.0. Thustheseasonality
of
stationsfrom the NaAA network containedsufficientlylong
J)• dependson the seasonality
of air temperature.
The function
recordsto performthisprocedure(Figure2). A list of the selected
)•2(Srn)describesthe effect of soil moistureon decomposition,
stationsis given in Table 2.
which is a parabolicrelationshipthat dependson volumetricsoil
moistureandhasan optimumat approximatelyfield capacity.Thus
3.3. CO 2 SourceFrom FossilFuel Burning
theseasonality
off• 2 depends
on the seasonality
of soilmoisture.
The component
in theatmospheric
ca 2 concentration
resulting
fromthereleaseof ca 2 fromfossilfuel burningandcementman3. Data
ufacturewascomputedbasedon a global1øx1ø mapcompiledby
Marland
et al. [1989].Fossilfuel ca 2 emissions
wereassumed
to
3.1. Input Data for the TBMs
beconstantin timeduringeachyear.In severalregionsof theglobe
An equilibriumrun wasperformedfor eachof the six terrestrial this is probablynot the case:indeed,Rotty [1987] determineda
biosphere
modelsundera prescribedatmospheric
ca 2 concentra- seasonalityin the northernhemisphereof 18%, while Levin et al.,
tion of 340 partsper million by volume(ppmv).All of the models [1989],
based
on14Cmeasurements,
inferred
anevenlarger
sea-
used
thesame
0.5øx0.5
øhorizontal
grid.
Theclimate
used
todrive sonality
innorthern
Europe.
It isexpected
thatfossil
fuelusefor
themodels
isbased
ontheprecipitation,
temperature,
and
cloudi' heating
inwinter
(inhigher
latitudes)
and
forcooling
insummer
(in
ness
data
sets
ofCramer
and
Leemans
[W.Cramer,
personal
corn- lower
latitudes)
induces
aseasonal
release
pattern.
However,
these
munication,1994], which is an updatedversionof the Leeroans
and Cramer [1991] database.These data are long-termaverage
monthlymean valuesand are given on a 0.5ø grid for the entire
land surface(withoutAntarctica).If necessary,the monthlyvalues
wereinterpolated
to dailyandhourlytimesteps
usingsimpleinterpolation schemes and periodic functions as described in
sourcestypicallycontributeonly aboutonethirdto thetotalindustrial ca 2 emissions,which themselvesare substantiallysmaller
thanthe seasonalfluxesof interestin this study.On the basisof
theseconsiderations
we do not expectthe assumption
of constant
industrialca 2 emissionsto significantlyaffect the modeledresults.
section 2.3.
All fluxesare calculatedfor potentialvegetation(disregarding
4. Results
human influences)exceptfor the SDBM1 which implicitly includesland use throughits applicationof satellitedata (NDVI).
To evaluatehowwell a simulationreproduced
theobservedseaThe descriptions
of the patternof vegetationtypes(biomes)and
sonalsignalof atmospheric
ca 2 at a monitoringstation,we calcusoil characteristics(soil type and texture) are different in each lateda normalizedmean-squared
deviation(NMSD) definedas
model,thoughall soil datalayersusedhere are basedon the FAO
soil map of the world [FoodandAgricultureOrganization,19711979]. Vegetationdistributionis either taken from observation(15)
basedmapsor derivedfrom climate(BIaME2, HRBM, and SILrn-!
Orn
VAN, seesections2.3.1, 2.3.3, 2.3.5.).
NMSD
=•2•12
rn
+CF
'rn+Co
'rn_COBS,
m
(,CT,
)2
whereCT,m,CF,m,
andCo,marethemonthly
ca 2 concentrations
3.2. Observationsof the AtmosphericCO 2 Concentration
resultingfromthecorresponding
biospheric,
fossilfuel,andocean
Atmospheric
observations
of theca 2 concentration
usedin the
presentstudywere obtainedfrom the comprehensive
monitoring
programof the NationalOceanographic
andAtmosphericAdministration.This networkconsistsof approximately30 stationscovering latitudesfrom 82øN to 90øS, with the highestdensityof
flux, respectively,
COBS,
rnis the 10-year-mean
observed
value
stationsin the Americas, in Antarctica, and on Pacific islands and
thebiggestgapsovertheEurasianandAfrican continents[Conway
et al., 1994a].We usedthemonthlyaveragedstationdataprovided
by the CarbonDioxide InformationandAnalysisCenter[Conway
et al., 1994b].
(1983-1992),and o mis the corresponding
standarddeviationof
theobserved
valuefor eachmonth(m=1.....12) of theyear.TheresultingNMSDsof theTBMsforeachof 27 monitoring
stations
are
documented in Table 3.
4.1. Comparisonsto CO2 MonitoringStations:
The Simple DiagnosticBiosphereModel
The SimpleDiagnostic
Biosphere
Model(SDBM1)is a diagnostictoolthatwasdesigned
to helpexaminetemporal
patterns
of
12
HEIMANNET AL.:EVALUATIONOFTERRESTRIAL
CARBONCYCLEMODELS
NEP
90N
....
90N
60N
60N
30N
30N
EO
•'
•'
EQ
,.lOS
30S
60S-
60S
BIOME2
90S
,,,,,
JAN
,,
,,
APR
,
JUL
.... ,
OCT
, ,.... ,
JAN
FBM
90S
APR
J/•N AISRJOL OCT
' JAN
' AISR JUL
JUL
90N
90N
60N
60N
30N
30N
EQ
EQ
30S
30S
60S
60S
TEM ]
90S
JAN
APR
JUL
OCT
JAN
APR
!
ill i IlllI
90N
90N•
6ON-
60N
3ON'
30N
I
I
Illl
EQ
30S-
30S
60S-
60S
SILVAN
90S
I
j,•N AiSRJUL OCT JAN APR JUL
JUL
EQ-
iiiii
,'
,
...... , ...... ,
JAN
APR
-60
JUL
OCT
-45
.... ,
'-•'
JAN
-30
SDBM
'
APR
90S
JUL
-15
0
,
,, ....
, ......
,,
,
,
JAN APR JOL OCT JAN APR JUL
15
30
45
60
gCmonth
-• m-2
Plate3a. Zonally
averaged
netecosystem
production
(Nl•P)foreach
oftheterrestrial
biosphere
models
shown
versus
latitude
andtimeoftheyear.
Units
areingCmonth
-] m-Zof
landarea
intheterrestrial
biosphere.
HElMANN
ET AL.' EVALUATION
OF TERRESTRIAL
CARBON CYCLE MODELS
13
NPP
90N
90N
....
60N
60N
30N
30N
EQ
EQ
30S
30S
60S-
60S
BIOME2
90S
, '
dAN
90N
!
!
APR
JUL
I
ii
OCT
I
i
JAN
i i I iii
i ii
APR
90S
i
JUL
..............
FBM
•.............•'
JAN APR JOL O•T
JAN
APR
JUL
90N .....
6ON-
60N
:
30N
30N
,.
..•,.
EO
EQ
30S
30S
60S
60S
TEM
90S
,
,
90S
JAN A!•R JOE OCT J•N A•R JUL
90N
90N
60N-
60N
3ON-
AN
JUL
OCT
JAN
APR
JUL
-.
30N
ß
EQ-
EQ
30S-
30S
60S-
60S
....
SDBM
SILVAN
90S
JAN
A•R JO
L O•T J•N A•R
15
30
45
90S
60
JOL O•T J•N APR JUL
JAN
JUL
75
90
105
120
gCmonth
-] m-2
Plate 3b. As in Plate3a butfor net primaryproduction(NPP).
135
14
HEIMANNET AL.:EVALUATIONOFTERRESTRIAL
CARBONCYCLEMODELS
90øN
i
i
i
i
i
i
i
.•l•W
i
i
.,=bo
60 ø
30 ø
i
,
.......
....
i
i
i
i
i
',
i
i
r
i
',
i
i
:
.... i .......
•B--•
-
i
i
•
I
I
I
I
•'
I
I
I
'
:
............................
i : ' ' ' ' •
-; ....... : .......
•--f ....... • ..............
.... j .......%
............
; .... %-4................
• ....... t ........
_:___
........
.....
30 ø
60 ø
2....
90'%
180øW
T....... ..............
150 ø
120 ø
90 ø
C02
..... ........ ].....
60 ø
30 ø
Monitoring
0ø
.....
30 ø
Stotion
....... F....... 'T
.......
60 ø
90 ø
120 ø
150 ø
180øE
Network
Figure2. Locationsof the27 CO2 monitoringstationsconsidered
in thisstudy[Conwayet al., 1994b].The two solid
linesdelineatethe middleandhigh-latitudenorthern,tropical(30øS- 30øN)andmidlatitudesouthernsourceregions
from which the contributionsto the seasonalstationsignalswere computedwith the SimpleDiagnosticBiosphere
Model (SDBM1) (seePlate 1).
sources
andsinksof CO2 throughout
theterrestrialbiosphere.
The
globalparameters
in the NPP (Ix) andRH (Q]o) relationships
of
SDBM1 havebeencalibratedsuchthatthesimulated
seasonal
signalof atmospheric
CO2 represents
a goodfit totheseasonal
signals
measured
at 5 of the27 monitoringstations.
As described
by Knorr
and Heimann [1995], the five calibration stationswere selectedto
coverapproximately
uniformlythe majorregionsin the northern
hemisphere
temperate
latitudeswith stronglyseasonal
vegetation
(seeTable2). Not surprisingly,
theagreement
betweentheSDBM1
simulations
andtheobservedseasonal
signalsis goodat moststations:themedianNMSD amongthe27 stations
is 1.7andtherange
of NMSD is between0.3 and8.4 (Table3). Exceptions
areStation
"M" (STM), Cold Bay (CBA), ShemyaIsland(SHM), andCape
Grim (CGa), where the normalizedmean-squared
deviation
(NMSD) is greaterthan5 (Table3). At STM thereis closeagree-
of CO2 at themonitoring
stations
[KnorrandHeimann,1995,Figure 1].
The SDBM1 simulationis usefulin identifyingtherelativecontributionsof fossilfuel emissions,
oceanicCO2 exchange,
andter-
restrialCO2 exchangein controllingthe seasonalsignalof
atmospheric
ca 2 at themonitoring
stations.
To evaluatetheserelative contributions,we ran the model with each of the different
sourcecomponents
separately.
In orderto separate
effectsof the
middleandhighlatitudesversusthetropics,we alsoperformedan
atmospheric
transportmodelsimulationwith the SDBM1 fluxes
from the latitudes north of 30øN, 30øS-30øN, and south of 30øS
only.
Plate 1 showstherelativecontributions
to the seasonalsignalof
CO2 at ninemonitoringstations
thatoccurat differentlatitudes:
Alert (ALT, 83øN);ShemyaIsland(SHM, 53øN);Midway (MID,
ment betweenthe model simulationsand the observations,but be28øN); Cape Kumukahi (KUM, 20øN); ChristmasIsland (CHR,
cause of the small standard deviation of the observations at this
2øN); Samoa (SMO, 14øS); AmsterdamIsland (AMS, 38øS);
station(approximately
0.5 ppmv),theNMSD is large.At CBA and PalmerStation(PSA, 65øS);and southpole (sPa, 90øS).At the
SHM thesimulationdiffersmoderately
fromthemeasured
signal, three northernmoststations(ALT, SHM, and MID) the seasonal
especiallyin JuneandJuly. At CapeGrim (CGa) the simulation signalof atmospheric
CO2 iscontrolled
almostentirelyby CO2 exdifferssubstantially
fromtheobservations,
whichmaybe attribut- changeof the terrestrialbiospherenorthof 30øN;fossilfuel emised in partto thefact thatin thetransportmodeltheCapeGrim losions,oceanicexchange,and terrestrialexchangesouthof 30øN
cationlies in a grid box with a substantialland surfacearea,which havelittle effecton the seasonal
dynamicsof atmospheric
CO2.
impliesthatthesimulatedsignalat thestationis largelydominated Theexchangeof CO2 betweentheatmosphere
andthenorthernterby thelocalsources.
However,theobservations,
in ordertoberep- restrialbiosphereplays a large role in the seasonaldynamicsof
resentative
of southern
hemisphere
background
air, areselectedacca 2 at northernstationsbecause
muchof thegloballandmasslies
cordingto wind directionand wind speed.This inconsistency
is northof 30øN(43%, excludingAntarctica)andNEP is highlysearesponsible
for a largefractionof the discrepancy
at this station sonalin thisregion.The effectof theexchangeis sostrongthatthe
[Ramonetand Monfray, 1996].
peak-to-peakamplitudemeasuredat northernstationsvaries beFor the terrestrialbiospherethe SDBM1 estimatesbothan antweenapproximately8 ppmvand 12 ppmv,with greateramplitude
nual
NPPand
anannual
RHof60.0PgC(1015
g C).Inthecontextat more
of simulatingtheseasonalsignalof atmospheric
CO2, thetemporal
patternof NEP at any spatialscaleis morerelevantthantheabsolutevaluesof NPP andRH;NPP andRH estimates
couldbebiased
eitherhighor low andstill producea goodfit to theseasonal
signal
northern stations.
The controlof theterrestrialbiosphereon the seasonal
signalof
stationsin the tropicsdiminishesfrom north to south(Plate 1). At
CapeKumukahi,whichis locatedin thenortherntropics,the seasonalsignalof CO2 is controlledalmostentirelyby exchangebe-
HElMANN ET AL.' EVALUATION
OF TERRESTRIAL CARBON CYCLE MODELS
15
Table 2. StationsFrom the NOAA StationNetwork [Conwayet al., 1994b] Selectedfor the PresentStudy
StationCode
Station
Country
Northern
Latitude
Longitude
Elevation,[m]
210
Stations
ALT
Alert, North West Territories
Canada
82ø27'N
62ø3l'W
MBC
MouldBay,NorthWestTerritorie. Canada
76ø14'N
119ø20'W
KTL
KotelnyIsland,Siberia
Russia
76ø06'N
137ø36'E
BRW*
Point Barrow, Alaska
United States
71ø19'N
156ø36'W
STM*
OceanStationM
Norway
66ø00'N
2ø00'E
15
5
11
6
CBA*
ColdBay, Alaska
UnitedStates
55ø12'N
162ø43'W
25
SHM
ShemyaIsland
UnitedStates
52ø43'N
174ø06'E
40
CMO
CapeMeares,Oregon
UnitedStates
45ø29'N
124ø00'W
30
NWR
Niwot Ridge,Colorado
UnitedStates
40ø03'N
105ø38'W
3749
AZR*
Azores(TerceiraIsland)
Portugal
38ø45'N
27ø05'W
MID
SandIsland,Midway
UnitedStates
28ø13'N
177ø22'W
4
24ø40'N
80ø12'W
3
KEY
Key Biscayne,Florida
UnitedStates
30
Tropical Stations
3397
MLO
Mauna Loa, Hawaii
United States
19ø32'N
155ø35'W
KUM*
CapeKumukahi,Hawaii
UnitedStates
19ø31'N
154ø49'W
3
AVI
St. Croix,Virgin Islands
UnitedStates
17ø45'N
64ø45W
3
GMI
Guam
UnitedStatesTerritory
13ø26'N
144ø47'E
2
RPB
RaggedPoint
Barbados
13ø10'N
59ø26'W
3
CHR
Christmas Island
Kiribati
2ø00'N
157 ø19'W
3
SEY
Seychelles(Mahe Island)
Seychelles
4ø40'S
55ø10'E
3
ASC
AscensionIsland
UnitedKingdom
7ø55'S
14ø25'W
54
SMO
AmericanSamoa
UnitedStatesTerritory
14ø15'S
170ø34'W
30
AMS
Amsterdam Island
France
37ø57'S
77ø32'E
CGO
CapeGrim,Tasmania
Australia
40ø41'S
144ø41'E
94
PSA
PalmerStation(AnversIsland)
Antarctica
64ø55'S
64ø00'W
10
SYO
SyowaStation
Antarctica
69ø00'S
39ø35'E
11
HBA
HalleyBay
Antarctica
75ø40'S
25ø30'W
10
SPO
AmundsenScott(SouthPole)
Antarctica
89ø59'S
24ø48'W
Southern Stations
150
2810
*Stations
used
tocalibrate
thediagnostic
model
SDBM1.
tweenthe atmosphereandthe northernterrestrialbiosphere.Close
to the equator,the seasonalsignalof CO2 at ChristmasIslandis
still dominatedby terrestrialexchanges,althoughseasonalvariationsin atmospherictransportinducea substantialseasonalsignal
from the fossil fuel source [Heimann et al., 1989]. In contrast, the
seasonalsignalat Samoa,whichis locatedin the southerntropics,
is controlledapproximatelyequally by fossil fuel emissionsand
oceanicexchangeandnorthernterrestrialexchange,the latternow
showsa 6-monthphaseshift leadingto a relativemaximumin August;tropicaland southernterrestrialexchangehaslittle effect on
seasonaldynamicsof atmosphericCO2 at this station.Although
much of the global land massoccursbetweenlatitudes30øN and
30øS(45%), the SDBM1 simulationindicatesthat NEP is highly
aseasonal
in the tropics,so that the northernterrestrialbiosphere
exchangestill playsa role in thenortherntropicsbut haslittle effect in the southerntropics.The diminishingrole, from north to
south,of the atmospheric
exchangewith northernterrestrialbiospherein controllingthe seasonalsignalof CO2 in the tropicsis
also reflectedin the peak-to-peakamplitude,which is approximately8 ppmvat CapeKumukahi,4 ppmvat ChristmasIsland,
and 2 ppmv at Samoa.
At the three southernmoststations(AMS, PSA, and SPO) the
seasonal
signalof CO2 is largelycontrolledby thedifferencebetween oceanicexchangeand the combinationof terrestrialexchangeand fossilfuel emissions
(Plate 1). Oceanicexchangeof
CO2 withtheatmosphere
is importantin thesouthern
latitudesbe-
16
HEIMANN ET AL.: EVALUATION OF TERRESTRIAL CARBON CYCLE MODELS
RH
90N '
90N
6ON'
60N
30N
30N
EQ
......
EO
30S
30S
60S '
60S
BIOME2
90S
.......
i
'l
............
OCT
• ......
•
90S
....
JUL
I I, ....
' ......... •.....
• .....
JAN APR JUL OCT JAN
90N
90N
60N
60N
30N
3ON-
EQ
i
i
u . i i i .
L.
L
....
ii
iiiiii iiiiiii.
i
i
JUL
] i
EO-
50S
30S'
60S
60S
HRB•
90S
APR
,:
JAN
,..............
APR JUL
,•
OCT
'
JAN
TEM
......
•..............
APR JUL
'
'
......
90SJ/•NAPRJO
L
i"
OCT
,
'
!
......
JANAPRJUL
9ON'
90N
.....
:..
......
60N
6ON'
30N
3ON'
EO
EQ
30S
30S
60S
60S
SILVAN
90S
,
JAN
,
APR
15
JUL
OCT
,:30
JAN
45
A
SDBM
JUL
60
JAN
75
90
APR
105
JUL
OCT
120
gCmoQth
-! m-2
Plate3c. As in Plate3a butfor heterotrophic
respiration
(RH).
JAN
135
APR
JUL
HEIMANN ET AL.: EVALUATION OF TERRESTRIAL CARBON CYCLE MODELS
17
Table 3. Mean SquaredDeviationBetweenObservedand SimulatedSeasonalCycle, Normalizedby Estimated
StandardDeviationsof Observations(Equation(15))
BIOME2
FBM
HRBM
SILVAN
TEM
SDBM 1
ALT
44.5
15.5
55.9
7.4
18.6
2.7
MBC
24.0
10.9
31.7
3.9
12.8
2.3
KTL
5.4
3.7
9.6
0.5
2.0
1.7
BRW
20.5
7.5
23.3
4.5
10.0
3.7
STM
25.7
11.2
27.7
14.8
12.2
5.2
CBA
13.0
9.8
15.6
10.2
6.7
4.9
SHM
13.0
9.3
16.0
12.0
5.2
5.6
CMO
15.2
6.9
62.7
1.1
6.9
2.8
NWR
7.1
4.7
5.5
9.6
2.5
1.6
AZR
4.5
2.7
4.6
6.8
1.6
0.5
MID
12.9
7.6
11.7
9.7
2.2
1.2
KEY
11.0
5.5
4.2
23.3
4.4
2.3
MLO
19.0
6.6
11.3
22.8
2.7
2.2
KUM
20.5
10.5
15.0
23.5
4.2
3.6
AVI
5.3
2.2
1.6
19.9
2.0
1.7
0.7
GMI
4.6
1.3
3.1
4.1
2.0
RPB
6.7
6.6
2.0
19.9
26.4
2.0
CHR
4.0
7.5
1.1
4.3
4.0
0.6
SEY
0.6
0.8
0.5
2.1
1.0
0.6
ASC
4.5
8.5
4.7
22.5
13.0
1.3
SMO
1.5
6.4
2.4
5.1
10.4
0.8
AMS
0.3
2.2
0.8
2.9
2.1
0.3
CGO
14.7
7.5
4.9
35.0
24.7
8.4
PSA
12.5
10.9
2.6
41.6
19.1
0.9
SYO
6.9
7.6
3.1
19.0
10.8
0.9
HBA
3.1
6.1
1.5
10.0
9.6
0.3
SPO
23.9
27.3
8.1
73.6
40.6
1.4
Median
11.0
7.5
4.9
10.0
6.7
1.7
Minimum
0.3
0.8
0.5
0.5
1.0
0.3
Maximum
44.5
27.3
62.7
73.6
40.6
8.4
Median
13.0
7.6
15.8
8.5
6.0
2.5
Minimum
0.3
0.8
0.5
0.5
1.0
0.3
Maximum
44.5
15.5
62.7
23.3
18.6
5.6
Median
4.6
6.6
2.4
19.9
4.0
1.3
Minimum
0.6
0.8
0.5
2.1
1.0
0.6
Maximum
20.5
10.5
15.0
23.5
26.4
3.6
Median
9.7
7.6
2.8
27.0
14.9
0.9
Minimum
0.3
2.2
0.8
2.9
2.1
0.3
Maximum
23.9
27.3
8.1
73.6
40.6
8.4
All Stations
Twelve Northern Stations (ALT- KEY)
Nine Tropical Stations(MLO - SMO)
Six Southern Stations (AMS- SPO)
18
HEIMANNETAL.:EVALUATION
OFTERRESTRIAL
CARBON
CYCLEMODELS
causemostof thesouthern
hemisphere
is coveredby waterandlit-
tleof thegloballandmassoccurs
southof 30øS(12%).Theeffects
of terrestrial
exchange
andfossilfuel emissions
on the seasonal
signalof CO2, although
smallin southern
latitudes,
aregenerally
in phasewitheachotherandoutof phasewiththeeffectsof oceanicexchange.
Evenat thesouthpolethebiospheric
signals
from
thenorthern,
tropical,andsouthern
regionsareof similarmagnitudeandarein phase,suchthatthetotalbiospheric
signalis of the
samesizeastheoceanicsignal.Because
theeffectsof oceanicex-
(Figure3a;Kruskal-Wallis
test,H = 13.8,P -- 0.008,andd.f. (degreesof freedom)-- 4). Pairwisemultiplecomparisons
(StudentNewman-Keuls,P < 0.05) indicatethatthefits of theFBM (median
is 7.6), SILVAN (medianis 8.5), andTEM (medianis 6.0) simula-
tionsarenotstatistically
differentfromeachotherbutarestatistically differentfrom the fits of the HRBM (medianis15.8)and
BIOME2 (medianis 13.0)simulations;
the fits of the HRBM and
BIOME2 simulationsare statisticallyindistinguishable.
4.2.2. Tropical stations. Nine monitoringstationsoccurin
change
tendto canceltheeffectsof terrestrial
exchange
andfossil the tropicsbetweenlatitudes20øN and 14øS(Plate 2b; MLO,
KUM, GMI, AVI, RPB, CHR, SEY, ASC, andSMO). Threeof the
stationsoccurin thenortherntropicalPacificOcean(MLO, KUM,
tropicalAtlanticOcean(AVI
The factthatin thesouthern
hemisphere
southof 30øStheoce- andGMI), two occurin thenortheast
aniccontribution
to the seasonal
signalis of similarmagnitudeas andRPB), andfouroccurnearor southof theequator(CHR, SEY,
the terrestrial contributions makes these locations less suited as
ASC, andSMO). At the threePacificstationsin the northerntropvalidationsitesfor themodeledseasonal
terrestrialCO2 exchang- ics the HRBM and BIOME2 simulationstend to estimateearly
es. Pendingaccuratetechniquesto separatethe terrestrialfrom the drawdownandrecoveryof atmospheric
CO2. In contrast,
theSILoceanic
signals
in theobservations
(e.g.,concurrent
13C/12C
or VAN simulationestimateslate drawdown and recovery. The
O2/N2 measurements),
any discrepancybetweensimulationand HRBM, BIOME2, and SILVAN simulations tend to underestimate
observationcouldequallywell be ascribedto errorsin terrestrialor the degreeof drawdownat the threesites.Recoveryalsois underoceanic(or a combinationof both) exchangeformulations.The estimatedby the threesimulations,exceptfor the SILVAN simumain constraintprovidedby the stationssouthof 30øSis a bound lation at Guam. For both the FBM and TEM simulations both the
on the seasonalamplitude which must be smaller than about timingandthe degreeof drawdownandrecoveryare similarto the
2 ppmv (peakto peak) [Knorr and Heimann, 1995].
observedsignal.
At thetwo Atlanticstationsin thenortherntropicsthe SILVAN
4.2. Comparisonsto CO2 Monitoring Stations:
simulationestimateslate drawdownandrecoveryof atmospheric
The Five PrognosticTBMs
CO2; the othersimulationsare moreor lesssynchronous
with the
Estimatesof annualNPPandof annualRH for theterrestrialbio- observedsignal.Both the BIOME2 andSILVAN simulationstend
to underestimatethe degreeof drawdownat both stations.The
sphereby the five prognosticTBMs in this studyrangefrom 46.4
BIOME2
simulationalsounderestimates
thedegreeof recovery.In
Pg C to 61.0 Pg C (FBM, 50.3; HRBM, 46.4; BIOME2 56.6; SILthe degreeof
VAN, 61.0; TEM, 49.0; seeTable 1). Becausecontrolsof the sea- contrast,the TEM simulationtendsto overestimate
the degreeof drawsonalsignalof CO2 varylatitudinally,we analyzetheperformance recoveryat both stationsand overestimates
tendsto overestimate
recovery
of the TBMs in simulatingthe seasonalsignalof CO2 separately downat RPB. The FBM simulation
at
RPB.
Both
the
timing
and
degree
of
drawdown
and
recovery
of
for northern,tropical,and southernmonitoringstations.
4.2.1. Northern stations. At the twelve monitoring stations the HRBM simulationare similarto the observedsignalat thetwo
stations.
that occur between latitudes83øN and 26øN (ALT, MBC, KTL,
At eachof the tropicalstationsnearor southof the equator,the
BRW, STM, CBA, SHM, CMO, AZR, NWR, MID, and KEY), the
simulationsof the TBMs generallycapturethe seasonalplacement patternof simulationsare unique.At ChristmasIsland all of the
of wintermaximaandsummerminimain theatmospheric
CO2 sig- simulations,except for HRBM, tend to estimateearly drawdown
nature (Plate 2a), but the phasingof the seasonalsignal differs and early recovery. The SILVAN and TEM simulationstend to
overestimate
the degreeof recovery.At Seychellesall of the simuamongmodels.In comparisonto the observedseasonalsignalof
drawCO2 at eachstation,the HRBM andBIOME2 simulations
consis- lations,exceptfor SILVAN, tend to estimatesynchronous
tently estimateearly drawdownandearly recoveryof atmospheric down and recoveryin comparisonwith the observedsignal' the
CO2. The TEM simulationtendsto estimateearly drawdownand SILVAN simulationindicateslate drawdownand recovery.The
degreeof drawdownandrecoveryis similarto thatof the observed
early recoveryat thefour northernmost
stations(ALT, MBC, KTL,
and BRW) but estimatesapproximatelysynchronousdrawdown signalfor all simulations,mostlybecauseat this sitethe seasonal
a largeseasonal
signalfromthefosandrecoveryfor theothereightstations.
Thedrawdownof CO2 es- monsooncirculationgenerates
sil fuel source [Heimann et al., 1989]. At AscensionIsland all of
timatedby the FBM simulationis approximatelysynchronous
for
the simulationsestimatelate drawdownand late recoveryin comall the stations,but similarto the TEM simulation,the recoveryis
early for the four northernmoststationsand synchronous
for the parisonto theobservedsignal.The HRBM simulationoverpredicts
othereight stations.The SILVAN simulationtendsto estimatelate
recoveryand the FBM, TEM, and SILVAN simulationsoverestidrawdownfor all stationsandlate recoveryat the eight southern- matethedegreeof bothdrawdownandrecovery.All TBMs tendto
moststations;the recoveryat the four northernmost
stationsis ap- overpredict
theobserved
peak-to-peak
variationby a factorof 2. At
proximately synchronous.All of the simulations tend to
Samoathe FBM, TEM, and HRBM simulationsestimateearly
underestimate
themagnitudeof thedrawdownfor the northernsta- drawdownandrecovery,while the SILVAN simulationestimates
tions,with the HRBM andBIOME2 simulationsgenerallyindicat- late drawdownandrecovery.The FBM, SILVAN, and TEM siming the greatest degree of underestimationand the SILVAN
ulationsalsooverestimate
thedegreeof drawdownandrecoveryin
simulationgenerallyindicatingthe least amountof underestima- comparisonto the observedsignal.
tion.
For the tropicalstationsthe medianNMSD indicesfor the simOn the basis of the NMSD index the simulations differ in their
ulationsrangefrom 2.4 to 19.9 (FBM, 6.6; HRBM, 2.4; BIOME2,
fits to the seasonalsignalof CO2 amongthe 12 northernstations 4.6; SILVAN, 19.9; and TEM, 4.0). On the basisof the NMSD in-
fuelemissions,
thepeak-to-peak
amplitude
of CO2 at southern
stationsis lessthan 2 ppmv.
HElMANN ET AL.: EVALUATION
OF TERRESTRIAL CARBON CYCLE MODELS
(a)
NMSD
'
30-
simulations
areconsistently
approximately180ø out of phasewith
theobservedseasonalsignalat all stations.The otherthreesimulationsall estimatelatedrawdownandrecoveryin comparisonto the
observedseasonalsignal;the FBM andTEM simulationsare approximatelyin phasewith eachotherandare later than the HRBM
simulation.The degreeof drawdownandrecoveryareconsistently
overestimatedby the FBM, SILVAN, and TEM simulations.The
degreeof drawdownand recoveryestimatedby the HRBM and
BIOME2 simulationsaresimilarto theobservedseasonalsignalof
CO2 at the six stations.
On the basis of the NMSD
(b)
401
35
index the simulations differ in their
fits to the seasonalsignalof CO2 amongthe six southernstations
(Figure3c; Kruskal-Wallistest,H -- 9.51, P -- 0.0495, and d.f. -- 4).
Pairwisemultiplecomparisons
(Student-Newman-Keuls,
P < 0.05)
indicatethat fits of the FBM (medianis 7.6), BIOME2 (median is
9.7), SILVAN (median is 27.0), and TEM (median is14.9) simulationsare not statisticallydifferentfrom eachotherbut are statistically different from the fits of the HRBM simulation(median is
30-
25-
NMSD'
19
20-
2.8).
15-
4.3. Comparisons of Latitudinal-Seasonal Patterns of NEP:
PrognosticTBMs Versus the SDBM1
10-
5-
On the basis of the NMSD
0-
80-
(c)
70-
6O
index the simulations
of the five
prognosticTBMs andthe SDBM1 differ in their fits to the seasonal
signal of CO2 among the 27 monitoringstations(Figure 4;
Kruskal-Wallis test, H -- 37.7, P < 0.0001, and d.f. = 5). Pairwise
multiple comparisons(Student-Newman-Keuls,
P < 0.05) indicate
that the fits of the FBM (median is 7.5), HRBM (median is 4.9),
NMSD
5O
8ONMSD'
,•o
30-
70-
20-
10-
60-
O-
50FBM
HRBM
BIOME2
SILVAN
TEM
40-
Figure3. Comparison
of thedistribution
of thenormalized
mean-
squared
deviation
between
theobserved
seasonal
cycleandthe
simulated
seasonal
cycleproduced
bycoupling
themonthly
esti-
30-
mates
ofnetecosystem
production
bytheprognostic
terrestrial
bio-
sphere
models
andfossil
fuelemissions
withtheHamburg
ocean 20andatmospheric
transport
models
for (a) northern
stations,
(b)
tropical
stations,
and(c)southern
stations.
Theboxplotsindicate 10themedian,
25th,and75thpercentile
and10thand90thpercentile
-7
intervals.
_
dexthesimulations
donotdifferin theirfitsto theseasonal
signal
of CO2 amongtheninetropicalstations
(Figure3b;Kruskal-Wallis
FBM HRBM BIOME2 SILVAN
TEM
SDBM1
test,H = 6.21,P = 0.1841,andd.f.-- 4). A poweranalysis
for a oneway analysisof varianceindicatesthatthepowerto detectdiffer- Figure 4. Comparisonof thedistributionof thenormalizedmeansquareddeviationbetweenthe observedseasonalcycle and the
encesis low (0.36 versusdesired0.80).
simulatedseasonal
cycleproducedby couplingthe monthlyesti4.2.3. Southernstations. At the sixmonitoringstations
that matesof netecosystem
production
by terrestrialbiosphere
models
occurbetweenlatitudes38øSand90øS(AMS, CGO, SYO, PSA, andfossilfuel emissions
withtheHamburgoceanandatmospheric
HBA, andSPO),thesimulations
of theTBMsdonotgenerally
cap- transportmodelsfor all of the 27 monitoringstations.The box
turethe seasonalplacementof winterand summermaximain the plotsindicatethe median,25th, and75th percentileand 10thand
atmospheric
CO2 signature
(Plate2c). The BIOME2 andSILVAN
90th percentileintervals.
20
HEIMANN ET AL.:EVALUATIONOF TERRESTRIALCARBONCYCLEMODELS
BIOME2 (medianis 11.0), SILVAN (medianis 10.0), and TEM HRBM estimates;
thetimingandmagnitudeof CO2 releasesesti(medianis 6.7) simulations
arenotstatistically
differentfromeach matedby the otherthreemodelsare similar to the SDBM1 in the
other but are statisticallydifferentfrom the SDBM1 simulation southernportionsof the terrestrialbiosphere.
(median is 1.7). The SDBM1 simulation indicatesthat control of
theterrestrial
biosphere
overtheseasonal
dynamics
of CO2 in the 5.
atmosphere
diminishes
fromnorthto southandsuggests
thatNEP
ishighlyseasonal
in northern
regions
andhighlyaseasonal
in tropicalregions.Becauseof thesingularlyclosefit of theSDBM1 simulationto theseasonal
signature
of CO2 at themonitoring
stations,
thelatitudinal-seasonal
patternof NEP estimated
by theSDBM1is
a usefulstandardfor comparison
to theotherTBMs. However,it is
Discussion
and Conclusions
The simulationswith the diagnosticmodel SDBM1, which is
tunedto the atmospheric
CO2 seasonalsignalat five monitoring
stationsin the northernhemisphere,successfully
reproducedthe
seasonalsignalof CO2 at the othermonitoringstations.The additionalsimulationexperiments
with the SDBM1 sourcesplitamong
importantto keepin mind thatthe simulationswith the SDBM1 do tropicsand extratropicallatitudesdemonstratethat the seasonal
notincludeCO2 exchanges
caused
by landuse(e.g.,biomass
burn- signalboth north and within the tropicsis dominatedby the exing).
changeswith the terrestrialbiosphere.In southernlatitudes,howFornorthernregionsof theterrestrial
biosphere,
theSDBM1 in- ever, ocean-atmosphere
gasexchangecontributessignificantlyto
dicatesthatNEP is highlyseasonal
(Plate3a).All of thefive prog- the seasonalsignalof CO2.
Amongthe27 monitoringstationstheNMSD (equation(15)) of
nosticmodelsalso indicatethat northernregionsare highly
seasonal,
butthemagnitude
andtimingof CO2 uptakeandrelease the SDBM1 simulationwas greaterthan 5 at only four of the staby the terrestrialbiospherediffersamongthe models.The maxi- tions(StationM, Cold Bay, Shemya,and CapeGrim). Severalasmumuptakeof CO2 in thenorthernterrestrial
biosphere
estimated pectsof the methodologyof the presentstudymay be responsible
by FBM, HRBM, andTEM is similarto themaximumuptakees- for the differences.First,thereis uncertaintyasto how well the attimatedby theSDBM1,butthetimingof maximumuptaketends mospherictracertransportmodel(TM2) performsat highnorthern
to be earlier,andthegrowingseason
appearsto be shorterin com- latitudes.Poorperformanceof TM2 at northernlatitudesmightexparisonto theSDBM1.Thereleaseof CO2 to theatmosphere
esti- plainthe poorerfits at StationM, Cold Bay, andShemya.Second,
mated by the three modelsduring the winter is less than that ourassumption
of time invariantfossilfuel emissionsmay be inacestimated
by theSDBM1.Thetimingof maximum
uptakeof CO2 curate.For instance,Levinet al. [1989] foundthatthe seasonality
estimatedby the SILVAN modelis laterthanthat of the SDBM1, of the fossilfuel flux fromEuropeis about3 timeshigherthanthe
but the magnitudeof uptakeappearsto be similar.The releaseof seasonalityestimatedfrom emissionstatistics.This mightcontribCO2 estimatedby the SILVAN modelis similarto that estimated uteto the seasonalsignalobservedat StationM. Third, errorsin the
by theSDBM1. In comparison
to theSDBM1 theBIOME2 model simulatedoceanexchangewith the atmospheremay becomeimestimates
lessreleaseof CO2 to theatmosphere
duringwinter;the portantin the southernhemisphere.It shouldalsobe recognized
timingof uptakeis earlier,andthemagnitude
of uptakeis smaller. thatwe'havenot includedin our analysesbiomassburningnor the
Thus, althoughthe estimatesof the prognostic
modelsand the globalterrestrialsinkwhichcanbe inferredfromtheglobalbudget
CO2 [Schimeletal., 1995].Somenoticeable
effects
SDBM1 agreethat NEP in the northernterrestrialbiosphereis of atmospheric
highlyseasonal,
theprognostic
modelsdifferfrom the SDBM1 in from tropicalvegetationfires may be expectedat AVI, RPB, and
themagnitude
and/ortimingof CO2 exchange
withtheatmosphere ASC accordingto a studyby Iacobelliset al. [1994]. Vegetation
in thisregion.
firesin the tropicsandsubtropics
areexpectedto occurprimarily
For tropicalregionsof theterrestrialbiosphere
the SDBM1 in- duringthe dry seasonwhenNPP is reducedbecauseof waterstress.
dicatesthatthe seasonality
of NEP is muchlesspronounced
than Henceif vegetationfiresmakeup for a significantatmospheric
sigin northernregions(Plate3a). In contrast,the NEP estimatesof all nalin atmospheric
CO2, thenouranalysis,whichneglected
thisadfiveprognostic
modelsarehighlyseasonal.
Themaximumuptake ditionalflux, shouldresultin an underestimation
of the amplitude
of CO2 in thetropicalterrestrial
biosphere
estimated
by theFBM, of the seasonalsignal. These additionaleffects are difficult to
HRBM, SILVAN, and TEM is greaterthan the maximumuptake quantify,however,because
of anuncertain
database
ontheCO2 reuptakeby regrowing
estimatedby the SDBM1. The maximumuptakeestimatedby the leasefrom vegetationfiresandthesubsequent
BIOME2
model is similar to that of the SDBM1 in the northern
vegetation.Anotherexplanationfor differencesmay be the useof
tropicsbut is greaterin the southerntropics.All five of the prog- 1987wind fieldsin TM2 whichmay not be representative
for the
nosticTBMs estimatesubstantial
releaseof CO2 to theatmosphere entireperiodof CO2 measurements
(1983-1992);in a sensitivity
duringthe tropicaldry season;the estimatedreleasesaremostpronouncedfor the FBM, HRBM, andTEM. In general,seasonality
of
NPP in the tropicsdiffers more betweenthe prognosticand diag-
study, Knorr and Heimann [1995, Table 2] documenteda 15%
changein theoverallagreement
whenusingwinddatafrom 1986.
Finally, becausewe comparemodel outputof TM2 at approximately7.83ø by 10ø resolutionwith point stationmeasurements,
nosticmodelsthandoestheseasonality
of RH (Plates3b and3c).
we potentiallyneglectthe influenceof local sources.
For southernregionsof theterrestrialbiosphere,the SDBM1 indicatesthatNEP is highly seasonal(Plate3a). All of theprognostic
Thereareseveralaspects
of theSDBM1 extrapolations
for NPP
TBMs estimatethattheuptakeof CO2 fromtheatmosphere
occurs andRH thatmay influencethefit betweenthe SDBM1 simulations
betweenOctoberand April, which is generallyin agreementwith andtheobservations.
First,theuseof constant
globallightuseefthe SDBM1 estimates. The FBM and HRBM estimates indicate
ficiencyandQlO valuesis clearlyan oversimplification
[Ruimyet
highermaximumuptakeof CO2 thanthe SDBM1 estimates,
and al.,1994].Also,theuseof monthlymaximumNDVI froma 5-year
the BIOME2, SILVAN, and TEM estimatesindicate lower maxiperiodin the SDBM simulations
maynotbe representative
for the
mum uptakeof CO2. In comparison
to the SDBM1 the estimated entireperiodof CO2 measurements.
Finally,becausethe SDBM1
releaseof CO2 to the atmosphereis extendedin the FBM and calculatesfluxes at monthlytemporalresolution,there may be
HEIMANN ET AL.: EVALUATION OF TERRESTRIAL CARBON CYCLE MODELS
phasedifferencesof up to two 2 betweensimulatedandobserved
signals.
Eventhoughtherearea numberof concernsthatcouldinfluence
thefit of theSDBM1simulations
withtheobserved
seasonal
signal
21
from north to south.For example,at AscensionIsland (ASC,
7ø55'S),locatedin thetropicalAtlantic,all theprognostic
models
consistently
overpredict
theobserved
seasonality,
mostprominently seenin the largepeakappearingduringNovemberandDecem-
of atmospheric
CO2, thesimulation
withthediagnostic
modelsig- ber.The seasonal
cycleat thisstationis primarilyinfluenced
by
nificantlyfitstheseasonal
signalbetterthananyof thesimulations seasonal
terrestrialCO2 exchanges
in tropicalsouthern
Africa and
withtheprognostic
TBMs.OnlyatCapeGrimdid simulations
with SouthAmerica [Kaminskiet al., 1996].
prognostic
TBMs producesmallerNMSD valuesthanthe SDBM1
Onehypothesis
to explaintheoverestimation
of CO2 releaseof
simulation(seeTable 3), which,however,may be a coincidence the terrestrialbiosphere
duringthe dry periodin the tropicalreconsideringthe problemof data selectionat this site as discussed gionsis thatnoneof theTBMs allowtropicaltreesaccess
to deep
in section4.1. The mostlikely explanation
for thedifferentfits to soil water.In someregionsof the tropics,evergreenforestshave
theobserved
datais theuseof phenology
algorithms
in theprog- beenableto maintainevapotranspiration
during5-monthdryperinosticmodelsversusthe useof NDVI whichessentially
defines odsby absorbing
waterfromsoildepthsof morethan8 m [Nepstad
phenology
in thediagnostic
model.The variabilityin phaseof the et al., 1994].TheTBMs in thisstudyhaverootingzonesfor tropiseasonal
cycleamongsimulations
with prognostic
modelsis pre- calforeststhatrangefrom1 m to 3 m, depending
onthemodel.The
sumablycausedby variationin the phenologyalgorithmsof the useof NDVI by theSDBM1to someextentimplicitlyaccounts
for
models.An additional,
although
minorsourceof variabilityamong theeffectsof deeprootingonphenology.
Furthermore,
sincein the
theprognostic
TBMs maybedueto differentvegetation
mapsand SDBM1 thesamewaterstressfactoris appliedontheformulations
differenttechniques
to interpolate
themonthlyclimatevariablesto of NPPandRH, itseffectsontheresulting
netseasonal
CO2 fluxis
thetimeresolution
of themodel.It mustalsoberecognized
thatthe minimized.
useof NDVI in thediagnostic
modelimplicitlymaypartiallyacAnotherhypothesis
to explaindifferencesin thelatitudinal-seacountfor the effectsof landuseon terrestrialCO2 exchange.
In sonalpatternof NEP betweenthe estimatesof the prognostic
contrast,the prognostic
modelsusepotentialvegetationhencedo TBMs andthe SDBM1 in the tropicsis thatplantrespiration
is
not considerland use.However,preliminarysensitivitysimula- overestimated
by theprognostic
TBMs. Uncertaintyin respiration
tionsin whichthe prognostic
modelsmaskedout agriculturally budgetsoccursbecausethereis poor informationon the relative
usedlandpixelsshowedonlyverysmalldifferencesin theseasonal amountsof activelyrespiringplant tissue(leaves,fine roots,and
cycleat themonitoringstations
ascompared
to thebasesimulation. sapwood)andinactiveplanttissue(heartwoodandcoarseroots).
The fits of theprognostic
simulations
dependon theregionof
In SILVAN thedelayof theseasonal
cycleby approximately
2
the globe.At thenorthernstationsall modelstendto underestimate monthsin tropicallatitudes
appears
to becausedby therelatively
theseasonal
amplitudeand,in general,predictearlydrawdownand earlyreductionof heterotrophic
respiration
(seePlate3, compare
recovery.However, the generalphasingof the northernhemi- SILVAN with SDBM) asa resultof thereducedprecipitation
dursphereCO2 concentration
with a minimumin late summeranda ingthedry season.
Hencetheaccumulated
litterfromtheprevious
maximumin springis capturedby all models.Onehypothesis
to growingseasontendsto be preserveduntil the onsetof thefollowexplaintheunderestimation
of theseasonal
amplitudeis thatall the ing rainyseason,
whenit is respired,yieldinga largepulsewhich
modelspredictin high latitudesa too strongseasonality
of het- tendsto delaytheconcurrent
drawdownof theCO2 concentration
erotrophic
respiration,
which,in theseregions,iscontrolled
prima- by new production.
rily by theannualcycleof temperature.
Thisdefectmightbedueto
It is difficultto attributethelargedifferences
betweenthefitsof
the fact thatthe TBMs implementbasemetabolicratesof soilde- theprognostic
anddiagnostic
modelsin southern
regionsto particcomposition
thathaveno temperature
dependence.
Emergingdata ular deficiencies
in modelformulations
or particularproblemarindicatethatthebasemetabolicrateof decomposition
above10øC eas. However, a thoroughinvestigationis not warrantedin this
is muchhigherthanbelow10øC(J. M. Melillo, unpublished
data region because,as discussedin section4.1., the northernhemi1996;K. J. Nadelhoffer,
unpublished
data,1996),thatis, thereap- spheremiddle and high latitudes,the tropicsand the southern
pearsto be a thresholdtemperature
for basemetabolicrates.
hemisphere
midlatitudes,
andalsotheoceancomponent
contribute
The consistently
underestimated
seasonal
amplitudegenerated aboutequallytotheseasonal
signal.Furthermore,
thesignalsin the
by BIOME2 throughout
theentirenorthernhemisphere
stationsis southernhemispheremidlatitudesare small and show substantial
attributed
in partto a toostrongseasonality
in heterotrophic
respi- interannual
variations.
In thefuture,high-precision
stableisotope
rationwhichis generated
fromtheHowardandHoward[1993] data
(13C/12C)
might
provide
anadditional
tooltodiscriminate
beformulation
andwhichhence
cancels
a toolargefraction
ofthesea- tween the different contributions.
sonaldrawdown
of CO2 by themodeledNPP.
The relativelylate drawdownof SILVAN in the northernlati-
In a similarstudy,Hunt et al. [1996]haveuseda description
of
actualvegetation
to comparethesimulated
seasonal
signalof CO2
tudesresults
fromthemaximalNPPbeinglatecompared
to the
othermodels
(Plate3).In thetemperate
andboreal
biomes
thephenologyof leaves
asa function
of thedateof theyearin SILVANis
determined
onthebasis
ofheatsums
andnotonthebasis
ofpotentialNPPcomputed
fromconcurrent
climateandvegetation
state.
with the BIOME-BGC model,drivenby 1987 weatherdata[Piper
and Stewart,1996],to themeasured
CO2 signalfor theyear 1987.
Also, they derivedannualmaximumLAI from 1987 NDVI data
This appearsto leadto a laterbudburstandthusto laterNPP sim-
ulatedby SILVAN thanassimulated
by theothermodelsin the
and used this maximum LAI to initialize the carbon and nitrogen
pools.In ourstudywe haveusedpotential
vegetation
to makethe
comparison
of thesimulatedseasonal
signalof eachof five prog-
nosticTBMs, drivenby a long-termaverageclimate,to theaverage
seasonal
signalof theperiod1983-1992.Althoughthenatureof the
At thetropicalandsouthernstationsthedifferencesin thefits of comparisons
are somewhat
different,the analyses
presented
here
the prognostic
anddiagnostic
modelsgenerally
becomelarger andbyHuntetal. [1996]areimportantfor evaluatingthedynamics
northern latitudes.
22
HElMANN
ET AL.: EVALUATION
OF TERRESTRIAL
CARBON CYCLE MODELS
of globalTBMs andfor gainingfurtherinsightintothebiosphere's FoodandAgricultureOrganization,SoilMap of the WorM, vols. I-X, scale
1:5,000,000,Food and AgriculturalOrganizationand U. N. Educ., Sci.
rolein controlling
theseasonal
signalof atmospheric
CO2 concenandCult. Organ.,Paris,France,1971-1979.
tration.The differencesin thepredictions
by the five modelswill
Friend,A.D., PGEN: An integratedmodelof leaf photosynthesis,
transpirabe furtheranalyzedby developing
a modularized
CommunityTertion, and conductance,Ecol. Model., 77, 233-255, 1995.
restrial BiosphereModel which allows us to combinemodules
Fung,I. Y, K. Prentice,E. Matthews,J. Lerner,andG. Russell,Three-dimensionaltracermodelstudyof atmospheric
CO2:Response
to seasonal
exchangeswith the terrestrialbiosphere,J. Geophys.Res., 88, 1281-1294,
from the different TBMs.
The studyformspart of the ElectricPower ResearchInstitute
1983.
CarbonCycleModelLinkageProject,whichis usingcomprehensivemodelsincludingterrestrial,oceanic,andatmospheric
componentsto analyzethedynamicsof theglobalcarboncycleduringthe
historical(fossilfuel) and contemporary
periodsand to examine
the responseof the carboncycle to differentscenariosof future
emissions.
In thepresentanalysis,interannual
variabilityandsecular trendswere factoredout so as to test the ability of current
TBMs to simulateonespecificaspectof the contemporary
carbon
cycle,thatis,theseasonal
cycleof CO2 concentrations
in theatmosphere.Simulations
with theprognostic
TBMs successfully
reproducedaspects
of the seasonal
signalof atmospheric
CO2 which
Fung,I. Y., C. J. Tucker,andK. C. Prentice,Applicationof advancedvery
high resolutionradiometervegetationindex to studyatmosphere-biosphereexchange
of CO2,J. Geophys.
Res.,92, 2999-3015,1987.
Gallo, K. P., Experimentalglobalvegetationindex from AVHRR utilizing
pre-launchcalibration,cloudand Sun-anglescreening,digital data,Nat.
Ocean. and Atmos. Admin., Nat. Geophys. Data Cent., Boulder, Co.
1992.
Haxeltine,A., andI. C. Prentice,A generalmodelfor thelight-useefficiency
of primaryproduction,
Funct.Ecol., 10, 551-561, 1996.
Haxeltine,A., I. C. Prentice,andI.D. Cresswell,A coupledcarbonandwater
flux modelto predictvegetationstructure,
J. Veg.Sci.,7, 651-666, 1996.
Heimann M., The TM2 tracer model, model descriptionand user manual,
DKRZ Rep. 10, 47 pp., Ger. Clim. Comput.Cent.,Hamburg,1995.
HeimannM. andC. D. Keeling,A threedimensional
modelof atmospheric
CO2 transport
basedon observed
winds,2, Modeldescription
andsimulatedtracerexperiments,in Aspectsof ClimateVariability in the Pacific
and the WesternAmericas,Geophys.Monogr.Ser.,vol. 55, editedby D.
H. Peterson,pp. 237-275, AGU, Washington,D.C., 1989.
Heimann,M. andP. Monfray, Spatialandtemporalvariationof the gasexchangecoefficientfor CO2, 1, Dataanalysisandglobalvalidation,MaxPlanck-Inst.for Meteorol.Rep. 31, 29 pp., Max-Planck-Inst.for Meteorol., Hamburg, 1989.
Heimann,M., C. D. Keeling,andC. Tucker,A threedimensionalmodelof
atmospheric
CO2 transportbasedon observed
winds,3, Seasonal
cycle
andsynoptictime scalevariations,in Aspectsof ClimateVariabilityin the
PacificandtheWesternAmericas,Geophys.Monogr.Ser.,vol. 55, edited
by D. H. Peterson,pp. 277-303, AGU, Washington,D.C., 1989.
Howard,D. M., andP. J. A. Howard,Relationships
betweenCO2 evolution,
moisturecontent,andtemperature
for a rangeof soiltypes,in TheGlobal
Carbon Cycle and Its Pertubation by Man and Climate, Rep. EPOCCT90-0017(MNLA), EuropeanUnion,Brussels,Belgium,1993.
couldnot be anticipatedbecausethesemodelswere not calibrated
to theatmospheric
CO2 signal.We considerthisdemonstration
essentialto establish
theusefulness
of TBMs for subsequent
analyses
in whichwe will relaxtheassumption
of equilibriumin theterres-
trialcarbonbalanceandtakeintoaccountthephysiological
effects
of CO2 increase,
climatevariability,andclimatechange.
Acknowledgments.We cordiallythankLou Pitelkafrom the Electric
PowerResearchInstitute(EPRI, PaloAlto, California,U.S.A.) for hishelp
andstimulating
discussions.
We gratefullyacknowledge
a constructive
review by StevePiper.This studywasconducted
in partwithinthe Carbon
Cycle Model LinkageProjectwith supportby EPRI (RP 3416-01), IGBPGAIM, and the GermanMinistry for Educationand Research(BMBF).
Computingsupportwasprovidedby theGermanClimateComputing
Center (DKRZ) in Hamburg.
References
Campbell,
G.S.,AnIntroduction
toEnvironmental
Biophysics,
Springer,
New York, 1997.
Hunt,E. R., S.C. Piper,R. Nemani,C. D. Keeling,R. D. Otto,andS.W.
Running,
Global
netcarbon
exchange
andintra-annual
atmospheric
CO2
concentrations
predicted
by an ecosystem
process
modelandthree-di-
Collatz,G. J.,J.T. Ball,C. Grivet,andJ.A. Berry,Physiological
andenvimensional
atmospheric
transport
model,
GlobalBiogeochem.
Cycles,
10,
ronmental
regulation
of stomatal
conductance,
photosynthesis
andtran431-456, 1996.
spiration:
A modelthatincludes
a laminarboundary
layer,Agric.For. Iacobellis,
S.F.,R. Frouin,H. Razafimpanilo,
R. C. J.Somerville,
andS.C.
Meteorol., 54, 107-136, 1991.
Piper,NorthAfricansavanna
firesandatmospheric
carbon
dioxide,
J.
Collatz,G. J.,M. Ribas-Carbo,
andJ.A. Berry,Coupled
photosynthesis-sto-Geophys.Res.,99, 8321-8334, 1994.
matalconductance
modelforleavesof C4plants,
Aust.J. PlantPhysiol., Janecek,
A., G. Benderoth,
M. K. B. Ladeke,J.Kindermann,
andG. H. Kohl19, 519-538, 1992.
maier,Modelof theseasonal
andperennial
carbon
dynamics
in deciduConway,
T. J.,P.Tans,L. S.Waterman,
K. W. Thoning,
D. R. Buanerkitzis, ous-type
forests
controlled
byclimatic
variables,
Ecol.Modell.,49, 101K. A. Masarie,
andN. Zhang,Evidence
forinterannual
variability
ofthe
124, 1989.
carboncyclefromtheNOAA/CMDLglobalair sampling
network,
J. Jensen,
M. E.,andH. R. Haise,Estimating
evapotranspiration
fromsolarraGeophys.Res.,99, 22831-22855, 1994a.
diation,
J. Irrig.Drain.Div.,Am.Soc.Civ.Eng.,89(IR4),15-41,1963.
Conway,
T. J.,P.P.Tans,andL. S.Waterman,
Atmospheric
CO2fromsites Joyce,
L. A., J. Mills,L. Heath,A.D. McGuire,
R. W. Haynes,
andR. A.
in theNOAA/CMDLairsampling
network,
in Trends
'93:A Compendi- Birdsey,
Forest
sector
impacts
fromchanges
inforest
productivity
under
umofDataonGlobalChange,
editedbyT. A. Bodenetal.,Rep.ORNL/
climatechange,J. Biogeogr.,22,703-714, 1995.
CDIAC-65,
pp.41-119,Carbon
DioxideInf.Anal.Cent.,OakRidgeNat. Kaduk,
J.,Simulation
derKohlenstoffdynamik
derglobalen
Landbiosph'are
Lab., Oak Ridge,Tenn.,1994b.
mit SILVAN - Modellbeschreibung
undErgebnisse,
Ph. D. thesis,
Esser,
G.,Sensitivity
ofglobalcarbon
poolsandfluxestohuman
andpoten157pp.,Univ.ofHamburg,
Hamburg,
Germany,
July,1996.
tial climateimpacts,Tellus,Ser.B, 39, 24-60, 1987.
Esser,G., Osnabriick
Biosphere
Model:Structure,
construction,
results,in
ModernEcology
- BasicandApplied
Aspects,
editedbyG. EsserandD.
Overdieck,pp. 679-709,Elsevier,New York, 1991.
Esser,G., J. Hoffstadt,
F. Mack,andU. Wittenberg,
HighResolution
Bio-
Kaduk,
J.andM. Heimann,
A prognostic
phenology
scheme
forglobal
modelsof theterrestrial
biosphere,
Clim.Res.,6, 1-19,1996.
Kaminski,
T.,R.Giering,
andM.Heimann,
Sensitivity
oftheseasonal
cycle
ofCO2atremote
monitoring
stations
withrespect
toseasonal
surface
ex-
change
fluxes
determined
withtheadjoint
of anatmospheric
transport
sphereModel - Documentation,
Mitteil. 2, Inst.far Pflanzen6kol.derJusmodel,Phys.Chem.Earth,21,457-462, 1996.
ms-Liebig-Univ.
Giessen,
Germany,1994.
Keeling,
C. D., R. B. Bacastow,
A. F. Carter,S.C. Piper,T. P. Whorf,M.
Farquhar,
G. D., S. vonCaemmerer,
andJ. A. Berry,A biochemical
model
Heimann,W. G. Mook,andH. Roeloffzen,
A threedimensional
modelof
ofphotosynthetic
CO2 in leavesof C3species,
Planta,149,78-90,1980.
Federer,C. A., Transpirational
supplyanddemand:Plant,soilandatmospheric
effectsevaluated
by simulation,
WaterResour.
Res.,18(2),355362, 1982.
atmospheric
CO2 transport
based
onobserved
winds,1,Analysis
of observational
data.In Aspects
of Climate
Variability
in thePacificandthe
Western
Americas,
Geophys.
Monogr.
Ser.,vol.55,edited
byD. H. Peterson,pp. 165-236,AGU, Washington,
D.C., 1989.
Foley,J.A., Netprimary
productivity
in theterrestrial
biosphere:
Theappli- Keeling,C. D., T. P. Whorf,M. Wahlen,andJ. vanderPlicht,Interannual
cationof a globalmodel,J. Geophys.
Res.,99, 20773-20783,1994.
extremes
inthegrowthof atmospheric
CO2,Nature,375,666-670,1995.
HEIMANN
ET AL.: EVALUATION
OF TERRESTRIAL
CARBON CYCLE MODELS
23
and carboncycling in grasslandsand conifer forests,in Global Change:
Effectson ConiferousForestsand Grasslands,SCOPE 56, editedby A.
Soil Pollut., 70, 675-684, 1993.
I. Breymeyer,D. O. Hall, J. M. Melillo, and G.I. Agren, pp. 389-411,
JohnWiley, New York, 1996.
Knorr, W., and M. Heimann,Impact of droughtstressand other factorson
seasonalland biosphereCO2 exchangestudiedthroughan atmospheric McGuire, A.D., J. M. Melillo, D. W. Kicklighter, Y. Pan, X. Xiao, J. Heltracertransportmodel, TellusSer. B, 47, 4, 471-489, 1995.
frich, B. Moore III, C. J. Vorosmarty,andA. L. Schloss,EquilibriumreKohlmaier, G. H., et al., The Frankfurt BiosphereModel: A global process
sponses
of globalnet primaryproductionand carbonstorageto doubled
atmospheric
carbondioxide:Sensitivityto changesin vegetationnitrogen
orientedmodelfor theseasonal
andlongtermCO2 exchange
betweenterconcentration,Global Biogeochem.Cycles,11, 173-189, 1997.
restrial ecosystemsand the atmosphere,2, Global resultsfor potential
McNaughton,K. G., andP. G. Jarvis,Predictingeffectsof vegetationchangvegetationin an assumedequilibriumstate,Clim. Res., 8, 61-87, 1997.
es on transpirationand evaporation.In Water Deficit and Plant Growth,
Kurz, K. D., Zur saisonalenVariabilitfitdesozeanischenKohlendioxidpareditedby T. T. Kozlowski,pp. 1-47, AcademicPress,New York, 1983.
tialdrucks,Ph.D. thesis,107pp., Univ. of Hamburg,Hamburg,Germany,
November 1993.
Melillo, J. M., A.D. McGuire, D. W. Kicklighter,B. Moore III, C. J. V6r6smarty, and A. L. Schloss,Global changeand terrestrialnet primary proLarcher, W., PhysiologicalPlant Ecology, 303 pp., Springer-Verlag,New
Kindermann,J., et al., Structureof a globalcarbonexchangemodel for the
terrestrialbiosphere:The FrankfurtBiosphereModel (FBM), Water Air
York, 1980.
duction, Nature, 363, 234-240, 1993.
Law, R. M., et al., Variationsin modeledatmospherictransportof carbondi- Melillo, J. M., D. W. Kicklighter, A.D. McGuire, W. T. Peterjohn,and K.
M. Newkirk, Global changeandits effectson soil organiccarbonstocks,
oxideandthe consequences
for CO:•inversions,
GlobalBiogeochem.
Cycles, 10, 783-796, 1996.
in Role of Nonliving Organic Matter in the Earth's Carbon Cycle, edited
Leemans,R., and W. P. Crmner,The IIASA databasefor meanmonthlyvalbyR. G. ZeppandC. So.n.
tag,pp. 175-189,JohnWiley,NewYork, 1995.
ues of temperature,precipitation,and cloudinesson a global terrestrial Monsi, M. and T. Saeki, Uber den Lichtfaktor und den Pflanzengesellschaftenund seineBedeutungftir die Stoffproduktion,
Jpn.J. of Bot., 14,
grid,Rep. IIASA RR-91-18, Laxenburg,Int. Inst. Appl. Syst.Anal., Austria, 1991.
Levin, I., J. Schuchard,B. Kromer, and K. O. Mtinnich, The continental Eu-
22-52, 1953.
Murray, M. B., M. G. R. CannellandR. I. Smith,Date of budburstof fifteen
treespeciesin Britain following climatewarming,J. Appl. Ecol., 26, 693-
ropeanSuess-effect,Radiocarbon32(1), 431-440, 1989.
700, 1989.
Levin, I., R. Graul, and N. B. A. Trivett, Long-termobservationsof atmosphericCO:•and carbonisotopesat continentalsitesin Germany,Tellus, Nepstad,D.C., C. R. de Carvalho,E. A. Davidson,P. H. Jipp,P. A. Lefebvre,
Ser. B, 47, 23-24, 1995.
G. H. Negreiros,E. D. da Silva, T. A. Stone,S. E. Trumbore,and S. Vieira, The role of deeprootsin the hydrologicalandcarboncyclesof AmaLieth, H., Modeling the primaryproductivityof the world, in Primary Prozonianforestsandpastures,Nature, 372,666-669, 1994.
ductivityof the Biosphere,editedby H. Lieth and R.H. Whittaker, pp.
237-263, Springer-Verlag,New York, 1975.
Nevison,C. D., G. Esser,E. A. Holland,A globalmodelof changingN20
emissionsfrom naturaland perturbedsoils,Clim. Change,32, 327-378,
Liss,P.S., andL. Merlivat, Air-seagasexchangerates:Introductonandsyn1996.
thesis,in The Role of Air-Sea Exchangein GeochemicalCycling,edited
Piper,S.C., andE. F. Stewart,A griddedglobaldatasetof daily temperature
by P. Buat-M6nard,pp. 113-127,D. Reidel, Norwell, Mass., 1986.
andprecipitationfor terrestrialbiosphericmodeling,Global Biogeochem.
Lloyd, J., andJ. A. Taylor, On the temperaturedependence
of soil respiration, Funct. Ecol., 8, 315-323, 1994.
Cycles,10, 757-782, 1996.
Ltideke, M. K. B., et al., The FrankfurtBiosphereModel: A globalprocess Potter, C. S., J. T. Randerson, C. B. Field, P. A. Matson, P.M. Vitousek, H.
A. Mooney,and S. A. Klooster,Terrestrialecosystemproduction:A proorientedmodelfor theseasonal
andlongtermCO2 exchangebetweentercess model based on global satellite and surface data, Global Biorestrialecosystemsand the atmosphere,1, Model descriptionand illusgeochem.Cycles,7, 811-842, 1993.
tratingresultsfor the vegetationtypescold deciduousand borealforests,
Clim. Res., 4, 143-166, 1994.
Prentice,I. C., W. Cramer, S. P. Harrison,R. Leemans,R. A. Monserud,and
Ladeke,M. K. B., S. Dfnges,R. D. Otto,J. Kindermann,F.-W. Badeck,P.
A.M. Solomon,A globalbiomemodelbasedon plantphysiologyand
Ramge,U. Jfikel,and G. H. Kohlmaier,Responses
in NPP and carbon
dominance,soil propertiesand climate, J. of Biogeogr., 19, 117-134,
1992.
storesof thenorthernbiomesto a CO2-induced
climaticchange,asevaluatedby theFrankfurtBiosphereModel (FBM), Tellus,Ser.B, 47, 191- Raich,J. W., E. B. Rastetter,J. M. Melillo, D. W. Kicklighter,P. A. Steudler,
205, 1995.
B. J. Peterson,A. L. Grace,B. Moore III, andC. J. V6r6smarty,Potential
Ladeke,M. K. B., P. H. Ramge,andG. H. Kohlmaier,The useof satellitenetprimaryproductivityin SouthAmerica:Applicationof a globalmoddetectedNDVI data for the validation of global vegetationphenology
el, Ecol. Appl., 1,399-429, 1991.
modelsandapplication
to theFrankfurtBiosphere
Model,Ecol.Modell., Ramonet,M., and P. Monfray, P., Selectionof baselineconditionsin a 3D
91,255-270, 1996.
atmospherictransportmodel: applicationto the seasonaland synoptic
Lurin, B., W. Cramer, B. Moore III, and S. I. Rasool,Global terrestrialnet
variationsof CO2, Tellus,Ser.B, 48, 502-520, 1996.
primaryproductivity,
GlobalChangeNewsLett.(IGBP), 19, 6-8, 1994. Rastetter,E. B., A. W. King, B. J. Cosby,G. M. Hornberger,R. V. O'Neill,
Maier-Reimer,E., Geochemical
cyclesin an oceangeneralcirculationmodandJ. E. Hoppie,Aggregatingfine-scaleecologicalknowledgeto model
el: Preindustrialtracerdistributions,Global Biogeochem.Cycles,7, 645coarser-scale
attributesof ecosystems,
Ecol.Appl., 2, 55-70, 1992.
677, 1993.
Rotty,R., Estimates
of the seasonal
variationin fossilfuel CO2 emissions,
Tellus, $er. B, 39, 203-208, 1987.
Maier-ReimerE., U. Mikolajewicz,andK. Hasselmann,
Meancirculation
of
theHamburgLSG OGCM andits sensitivity
to thethermohaline
surface Ruimy,A., B. Saugier,andG. Dedieu,Methodology
fortheestimation
of terforcing,J. Phys.Oceanogr.,23, 731-757, 1993.
restrialnet primaryproductionfrom remotelysenseddata,J. Geophys.
Marland,G., T. A. Boden,R. C. Griffin, S. F. Huang,P. Kancirukand T.R.
Res.,99, 5263-5283, 1994.
Nelson,Estimates
of CO2emissions
fromfossilfuelburningandcement Russell,G., andJ. Lerner,A new finite-differencing
schemefor the tracer
manufacturing,
basedon the U.S. Bureauof Minescementmanufactur- transportequation,J. Appl. Meteorol., 20, 1483-1498, 1981.
ing data,Rep. ORNL/CIAC-25,NDP-030, CarbonDioxide Inf. Anal. Schimel,D. S., Terrestrialecosystems
andthe carboncycle,Global Change
Biol., 1, 77-91, 1995.
Cent.,Oak Ridge Nat. Lab., Oak Ridge,Tenn., 1989.
McGuire,A.D., J. M. Melillo, L. A. Joyce,D. W. Kicklighter,A. L. Grace, Six, K. D., andE. Maier-Reimer,Effectsof planktondynamicson seasonal
B. MooreIII, andC. J. V6r6smarty,Interactions
betweencarbonandnicarbonfluxesin an oceangeneralcirculationmodel,GlobalBiogeochem.
Cycles,10, 559-583, 1996.
trogendynamicsin estimating
netprimaryproductivity
for potentialvegetationin North America,GlobalBiogeochem.
Cycles,6, 101-124, 1992. Thornthwaite,C. W., andJ. R. Mather,Instructionsand Tablesfor ComputMcGuire,A.D., L. A. Joyce,D. W. Kicklighter,J. M. Melillo, G. Esser,and
ingPotentialEvapotranspirationand theWaterBalance,Publ. Climatol.,
vol. 10, No. 3, Lab. of Climatol., Drexel Inst. of Technol., Centerton,New
C. J. V6r6smarty,Productivity
response
of climaxtemperate
foreststo elevatedtemperature
andcarbondioxide:A NorthAmericancomparison Jersey,1957.
betweentwo globalmodels,Clim. Change,24, 287-310, 1993.
VEMAP Members,Vegetation/ecosystem
modelingand analysisproject:
McGuire,A.D., J. M. Melillo, D. W. Kicklighter,andL. A. Joyce,EquilibComparingbiogeography
andbiogeochemistry
modelsin a studyof terriumresponses
of soilcarbonto climatechange:Empiricalandprocess- restrialecosystem
responses
to climatechange
andCO2doubling,
Global
Biogeochem.Cycles,9, 407-437, 1995.
basedestimates,J. Biogeogr.,22, 785-796, 1995.
McGuire,A.D., D. W. Kicklighter,andJ. M. Melillo, Globalclimatechange V6r6smarty,C. J., B. Moore III, A. L. Grace,M.P. Gildea,J. M. Melillo, B.
24
HEIMANN ET AL.: EVALUATION OF TERRESTRIALCARBON CYCLE MODELS
J. Peterson,
E. B. Rastetter,andP. A. Steudler,Continental
scalemodels ton, 260 Panama Street, Stanford, CA 94305-1297. (e-mail: joerg@jasof waterbalanceandfluvialtransport:
An application
to SouthAmerica, per.stanford.edu)
D. W. KicklighterandJ. Melillo, The Ecosystems
Center,MarineBiologGlobalBiogeochem.
Cycles,3,241-265, 1989.
Warnant,
P., L. Franqois,
D. Strivay,andJ. C. G6rard,CARAIB:A global ical Laboratory,WoodsHole, MA 02543. (e-mail: [email protected];
modelof terrestrial
biologicalproductivity,
GlobalBiogeochem.
Cycles, jmelillo•lupine.mbl.edu)
G. H. Kohlmaier,R. D. Otto, andG. Wtirth, Institutftir Physikalische
und
8, 255-270, 1994.
Woodward,F. I., T. M. Smith,andW. R. Emanuel,A globalprimaryproduc- TheoretischeChemie, J. W. Goethe-Universitfit, Marie-Curie-Strasse 11, Dtivityandphytogeography
model,GlobalBiogeochem.
Cycles,9, 471- 60439 Frankfurt am Main, Germany. (e-mail: g.kohlmaier•chemie.unifrankfurt.de;[email protected])
A.D. McGuire,AlaskaCooperativeFish andWildlife ResearchUnit, Na-
490, 1995.
G. EsserandU. Wittenberg,
Institutftir Pflanzen6kologie,
Justus-Liebig-tionalBiologicalService,Universityof Alaska,FairbanksAK 99775.
(e-mail: [email protected])
Universitfit,
Heinrich-Buff-Ring
38, D-35392Giessen,
Germany.(e-mail:
B. Moore III and A. Schloss,Centerfor ComplexSystems,Institutefor
[email protected];
[email protected] Study of Earth, Oceans,and Space, University of New Hampshire,
A. Haxeltine,andI.C. Prentice,GlobalSystemsGroup,Departmentof Durham,NH 03824. (e-mail: [email protected];[email protected])
S. Sitch,PotsdamInstitutefor ClimateImpactResearch,P.O. Box 60 12
Ecology
University
of Lund,S61vegatan
37,S-22362Lund,Sweden.
03, D-14412 Potsdam,Germany.(e-mail: [email protected])
(e-mail:[email protected];
[email protected])
M. Helmann,W. Knorr,andW. Sauf,Max-Planck-Institut
ftir Meteorolgiessen.de)
ogie,Bundesstrasse
55,D-20146Hamburg,
Germany.
(e-mail:
[email protected];
knorr•z.de;
[email protected])
(ReceivedJuly 8, 1966; revisedJune16, 1997;
J.Kaduk,Department
of PlantBiology,Carnegie
Institution
ofWashing- acceptedJuly 2, 1997.)