the interannual variability of africa’s ecosystem ... · the lpj-guess has the most advanced...

Biogeosciences 6 285ndash295 2009wwwbiogeosciencesnet62852009copy Author(s) 2009 This work is distributed underthe Creative Commons Attribution 30 License

Biogeosciences

The interannual variability of Africarsquos ecosystem productivity amulti-model analysis

U Weber12 M Jung2 M Reichstein2 C Beer2 M C Braakhekke2 V Lehsten3 D Ghent4 J Kaduk4 N Viovy5P Ciais5 N Gobron6 and C Rodenbeck2

1Department of Forest Science and EnvironmentUniversity of Tuscia Viterbo Italy2Max Planck Institute for Biogeochemistry Jena Germany3GeoBiosphere Science Centre Lund University Sweden4Department of Geography University of Leicester UK5Laboratoire des Sciences du Climate et de lrsquo Environnement Gif-sur-Yvette France6European Commission ndash DG Joint Research Centre Institute for Environment and Sustainability Global EnvironmentMonitoring Unit Ispra (VA) Italy

Received 4 August 2008 ndash Published in Biogeosciences Discuss 10 October 2008Revised 2 February 2009 ndash Accepted 17 February 2009 ndash Published 25 February 2009

Abstract We are comparing spatially explicit process-model based estimates of the terrestrial carbon balance andits components over Africa and confront them with remotesensing based proxies of vegetation productivity and atmo-spheric inversions of land-atmosphere net carbon exchangeParticular emphasis is on characterizing the patterns of inter-annual variability of carbon fluxes and analyzing the factorsand processes responsible for it For this purpose simula-tions with the terrestrial biosphere models ORCHIDEE LPJ-DGVM LPJ-Guess and JULES have been performed usinga standardized modeling protocol and a uniform set of cor-rected climate forcing data

While the models differ concerning the absolute magni-tude of carbon fluxes we find several robust patterns of inter-annual variability among the models Models exhibit largestinterannual variability in southern and eastern Africa regionswhich are primarily covered by herbaceous vegetation In-terannual variability of the net carbon balance appears to bemore strongly influenced by gross primary production thanby ecosystem respiration A principal component analysis in-dicates that moisture is the main driving factor of interannualgross primary production variability for those regions Onthe contrary in a large part of the inner tropics radiation ap-pears to be limiting in two models These patterns are partlycorroborated by remotely sensed vegetation properties fromthe SeaWiFS satellite sensor Inverse atmospheric modeling

Correspondence toU Weber(uweberbgc-jenampgde)

estimates of surface carbon fluxes are less conclusive at thispoint implying the need for a denser network of observationstations over Africa

1 Introduction

Understanding terrestrial sources and sinks of CO2 and itsvariability is important for understanding the carbon cycle-climate feedback Extensive research in this field has con-centrated on the highly developed parts of the world in par-ticular North America and Europe with a strong research in-frastructure In a recent review Williams et al (2007) identi-fied Africa as ldquoone of the weakest links in our understandingof the global carbon cyclerdquo Africa is the second largest con-tinent of the world occupying about 20 of global land massand inhabits a large variety of ecosystems ranging from per-humid tropical forest to semi-arid and arid grass and shrubcommunities Although Africarsquos decadal scale mean carbonbalance appears to be neutral the continent contributes abouthalf of the interannual variability of the carbon balance onglobal scale (Williams et al 2007) This large interannualvariability results primarily from climatic perturbations re-lated to the El Nino phenomenon that directly affects theecosystemsrsquo productivity and due to concomitant biomassburning (eg Le Page et al 2007 Anyamba et al 20022003 Myneni et al 1996 Kogan 2000)

Given the scarcity of observation sites for atmosphericCO2 concentrations and land ndash atmosphere CO2 exchangethe uncertainties of African carbon cycle research remain

Published by Copernicus Publications on behalf of the European Geosciences Union

286 U Weber et al Interannual variability of Africas ecosystem productivity

high in particular regarding the spatial localization of hotspotregions of variability and the underlying driving forces Sim-ulations of terrestrial ecosystem models can provide insightshere however these models are also associated with largeuncertainties (eg McGuire et al 2001 Friedlingstein et al2006) in particular for water limited conditions (eg Moraleset al 2005 Jung et al 2007) and in addition have gener-ally not been tested and parameterized specifically for AfricaThus confidence of a single model analysis is limited and amulti-model study is warranted to identify coherent and dis-similar behavior between different biosphere models

In this study we assess the interannual variability ofAfricarsquos carbon cycle using four different terrestrial carboncycle models in conjunction with remotely sensed indicatorsfor the state of the vegetation as well as carbon balance esti-mates from global atmospheric inversions We aim to iden-tify (1) regions of largest carbon balance interannual variabil-ity (2) the primary process (photosynthesis or respiration)dominating the carbon balance variability and (3) which cli-mate variables are driving the ecosystemrsquos carbon cycle inthe models

2 Materials and methods

21 Model descriptions

The four ecosystem models ORCHIDEE (Krinner et al2005) LPJ-DGVM (Sitch et al 2003) LPJ-GUESS (Smithet al 2001) and JULESTRIFFID (Cox et al 2001 Es-sery et al 2001 Hughes et al 2006) applied in thisstudy are coupled biogeography-biogeochemistry modelsie they combine representations of both vegetation dynam-ics and land-atmosphere carbon and water exchanges (dy-namic global vegetation models ndash DGVMs) The concept ofplant functional types (PFT) (Smith et al 1997) is used todiscretize differences in physiology and allometry of speciesincluding adaptations to climatic conditions and disturbanceregime PFTs compete for resources like light and waterRepresentations of vegetation structure like allometry andfunction like phenology allocation mortality and establish-ment are essential for this Gross primary production (GPP)is calculated based on a coupled photosynthesis-water bal-ance scheme after (Farquhar et al 1980 Collatz et al 19911992) where simplifications have been incorporated in thedifferent models as described in detail in the above refer-ences Heterotrophic respiration is assumed to be representedby first-order decay of organic material with decay rates fora few pools (Foley 1995) which depend on temperature andmoisture following (Lloyd and Talor 1994) or a Q10 formulaFire the main disturbance in the region is simulated follow-ing (Thonicke et al 2001) within LPJ-DGVM ORCHIDEEand LPJ-GUESS

The applied models differ in the temporal resolution and inthe resolution of vegetation structure representations While

LPJ-DGVM and LPJ-GUESS are designed as stand-alonemodels running on a daily time step ORCHIDEE andJULES can be coupled to General Circulation Models func-tioning as land-surface schemes Therefore latter models re-solve the diurnal cycle with time steps of 30 to 60 min Usu-ally the LPJ family of models is driven by monthly climaticdata which are interpolated to pseudo-daily using a whethergenerator for precipitation In this study the LPJ-DGVM isdriven by daily climatic inputs In doing so net radiationis approximated from global radiation after Linacre (1969)The LPJ-GUESS has the most advanced representation ofvegetation structure including age with features of a forest-gap model (Shugart 1984) resolving forest succession LPJ-DGVM and ORCHIDEE have intermediate complexity ap-plying the concept of ldquoaverage individualsrdquo for a whole gridcell (Sitch et al 2003) while JULESTRIFFID employs aheuristic approach to determine the vegetation coverage andcarbon allocation to each PFT

22 Model drivers

All models are driven by the same climate data atmosphericCO2 concentration and soil texture type on a 1 grid from1982 to 2006 which has been derived as follows Meteo-rological forcing (near surface air temperature specific hu-midity wind speed radiation and precipitation) originatesfrom 6-hourly NCEP-DOE Reanalysis-2 (Kanamitsu et al2002) that were spatially interpolated to 1 from the origi-nal T62 Gaussian grid Despite substantial improvements ofNCEP R2 over R1 considerable precipitation biases remainin comparison to various independent data sets (Fekete et al2004) To account for these limitations precipitation wascorrected using more reliable data sets based on observationsfrom the satellite based Tropical Rainfall Measuring Mission(TRMM 3B43) available from 1998ndash2006 (Kummerow etal 1998) and from interpolated station data provided by theClimate Research Unit (CRU) from 1961ndash2003 (CRUTR21Mitchell and Jones 2005) The general calibration methodfollows Williams et al (2008) Ngo-Duc et al (2005) andSheffield et al (2006) where the daily values of the origi-nal NCEP data for each grid cell are scaled to match in theirmonthly totals those of the corresponding CRU and TRMMdata For the CRU period (1979ndash1997) precipitation was cor-rected following

CAl NCEPymdh =NCEPymdh times CRUym

NCEPymtimes (1)sum2006

1998TRMMmsum20061998CRUm

For the TRMM period (1998ndash2006) the calibration used

CAl NCEPymdh = NCEPymdh timesTRMMym

NCEPym(2)

Biogeosciences 6 285ndash295 2009 wwwbiogeosciencesnet62852009

U Weber et al Interannual variability of Africas ecosystem productivity 287

Fig 1 African ecoregions

The adjustment yielded an overall reduction of 152 pre-cipitation

Soil texture is given by the IGBP-DIS map at 1 (Tempel etal 1996) Data on the annual CO2 concentration was takenfrom measurements at Mauna Loa (wwwesrlnoaagovgmdccggtrends)

23 Experimental setup

The simulation of the terrestrial carbon and water budgets arecarried out at a spatial resolution of 1

times1 for entire conti-nental Africa for the target period 1982ndash2006 Spin-up cal-culations were performed by repeating the years 1982ndash1992using the meteorological data sets of the appropriate yearsand a fixed CO2 concentration of 34113 ppm (Mauna Loavalue in 1982) until carbon pools reach equilibrium Afterthe spin-up model simulations start in 1982 with rising CO2concentration per annum Potential vegetation distributionwas dynamically simulated by the models

24 Data analysis

Definition of regions

We define six major regions of sub-saharan Africa basedon the broad distribution of ecosystem types (available fromglobal land cover maps) and thus implicitly accordingto bioclimatic conditions Northern Savannah Belt Cen-tral African tropical forest Horn of Africa Southern rain-green woodlands South African grasslands and Madagascar(Fig 1) Given that Madagascar is small in comparison to theother regions and very heterogeneous we do not discuss thisregion extensively

Quantification of interannual variability

Interannual variability (IAV) is calculated for each pixeland model as the standard deviation of the respective quantity(eg GPP) for a yearly time step across all years resulting ina grid of IAV for each model and variable For identifyingspatial patterns of relatively high and low interannual vari-ability the IAV grid of each model was z-transformed

z(IAV i) =

(IAV i minus IAV

)σIAV

Where z(IAVI) is the standardized IAV of grid cell iIAVandσIAV are the spatial mean and standard deviation of theIAV respectively Hence z(IAVI ) measures the degree ofvariability for each pixel in units of standard deviations ievalues larger than 0 refer to above average interannual vari-ability and for example a value of two indicates variabilityof 2 standard deviations above the mean variability of thecontinent Consistent spatial patterns between all modelsare defined as model agreement and are implemented as thesum of models per grid cell which show variability largerthan one standard deviation above the mean variability ofthe continent

Principal component analysis

To analyze the relationship between meteorologicalconditions and simulated carbon fluxes on annual scale wefollow the approach of Jung et al (2007) First we reducethe array of several meteorological driver variables to theirprincipal components and then calculate the correlationbetween meteorological principal components and relativeflux variations for each grid cell The principal componentanalysis (PCA) of the meteorological data reduces thedimensionality of the data set and often extracts majorweather patterns or gradients We used four annual datafields as input to the PCA precipitation air temperatureradiation specific air humidity

Corroboration against satellite data

We use the fraction of absorbed photosynthetic active radi-ation (FAPAR) product of Gobron et al (2006) based on theSeaWiFS satellite sensor as a proxy for vegetation productiv-ity (available atwwwfaparjrcit) It has been designed as anoptimized indicator for the state and health of the vegetationand overcomes several limitations of the classically used nor-malized difference vegetation index (NDVI) (Gobron et al2000) Jung et al (2008) have shown that the annual sumof FAPAR growing season values correlates strongly withannual gross primary production from eddy covariance mea-surement sites in Europe At first glance it appears to be moreconsistent to compare simulated FAPAR by the models withthe FAPAR satellite retrievals instead of using the FAPAR as

wwwbiogeosciencesnet62852009 Biogeosciences 6 285ndash295 2009


a proxy for GPP However there are several reasons why itmakes more sense to interpret the remotely sensed FAPAR asan indicator for productivity (1) ecosystem models tend tocapture interannual variability of GPP primarily via interan-nual variations of radiation use efficiency and not via changesof leaf area (Jung et al 2007) while the interannual anomalypatterns of the SeaWiFS-FAPAR provide a realistic picture ofthe GPP anomalies (Jung et al 2008 Gobron et al 2005)and (2) there are conceptual mismatches between the FAPARfrom the satellite and the FAPAR simulated by models Forexample the response of herbaceous understorey which isvery sensitive to eg water stress plays likely an importantrole in the anomaly patterns of the satellite FAPAR by indi-cating the direction of change of the ecosystem (see in Junget al 2008 for more discussion) In addition changes ofthe remotely sensed FAPAR may originate from changes ofleaf colour (eg leaf darkening or yellowing) which indi-cates changes of chlorophyll content and thus radiation useefficiency In contrast the leaves in the models do not changetheir reflective properties effects of leaf aging on photosyn-thesis are in some cases captured as Vcmax being a functionof leaf age as in ORCHIDEE

We calculate mean annual FAPAR as a proxy for GPPafter filling gaps of the FAPAR time series as describedin Jung et al (2008) In addition we compare the meanseasonal cycles of simulated GPP and remotely sensedFAPAR for the defined regions

Comparison with atmospheric inversions

We compare the simulated variations of the African car-bon balance in terms of Net Ecosystem Productivity (NEP)and Net Biome Production (NBP) with results from globalatmospheric inversions from Rodenbeck (2005) (versions96v31) covering the period of 1997 to 2006 based on glob-ally 51 stations of atmospheric CO2 records This way wealso determine the role of fire on the interannual variabilityof the carbon balance Simulated NEP is calculated as GPPminus Terrestrial Ecosystem Respiration (TER) ModeledNBP is based on the difference of NEP and model specificfire emissions except JULES which does not include fireThe inversions detect carbon emissions from fire To facili-tate comparability between NEP simulations and inversionswe used the Global Fire Emission Database (GFED version21 available atftpdaacornlgovdataglobalvegetationfire emissionsv21) from Randerson et al (2007) and Vander Werf et al (2006) to correct for carbon fire emissions inthe atmospheric inversions

Table 1 Continental annual average GPP NPP per model (temporalstandard deviation) all numbers are in Pg C y-1

LPJ DGVM LPJ GUESS JULES ORCHIDEE

GPP 3968 [173] 1658 [104] 3150 [091] 2980 [120]NPP 1728 [112] 916 [067] 1201 [048] 1538 [077]

3 Results and discussion

31 Mean annual carbon fluxes

We use annual sums of GPP and Net Primary Production(NPP) over the entire study period to investigate the meanannual carbon fluxes at the continental as well as on the re-gional scale and compare them to global numbers based onthe IPCC Fourth Assessment Report (IPCC 2007) Mod-elled absolute NEP estimates strongly depend on factors thatcannot be taken into account (eg change in land-use his-tory) and thus are relatively meaningless Hence we dis-cuss NEP only in terms of interannual variability as previ-ously done in other contexts (eg Ciais et al 2005 Vet-ter et al 2008) (see below) At continental scale annualGPP estimates range from 1658 to 3968 Pg C y-1 (14ndash33of global GPP (120 Pg C y-1)) and 916 to 1728 Pg C y-1for NPP (14ndash27 of global NPP (65 Pg C y-1)) respectively(Table 1) Previous modeling studies indicated mean annualNPP between 7 and 13 Pg C y-1 (Cramer et al 1999 Cao etal 2001 Potter 2003) Possibly the too extensive forestcover simulated by LPJ-DGVM (1728 Pg C y-1) and OR-CHIDEE (1538 Pg C y-1) may have caused an overestima-tion of continental scale NPP A realistic representation ofsavannah ecosystems is still a challenge for dynamic vegeta-tion models

The comparison of simulated interannual variations of theAfrican carbon balance with atmospheric inversions revealsa consistent pattern but some discrepancies remain (Fig 2)There is agreement of above average net uptake for 19972000 2001 and 2006 and a below average carbon balancefor 1998 2002 2003 and 2005 The general pattern of conti-nental scale IAV is consistent between NEP and NBP whichindicates a small contribution of fire emissions on the vari-ability of the African carbon balance in the considered pe-riod Relatively low interannual variability of fire emissionsin Africa is consistent with van der Werf et al (2006) Someof the deviations between the models and the inversions iscertainly also related to uncertainties of the latter given thatthe density of atmospheric CO2 measurement stations is low

Differences between models are even more pronounced atregional level (Fig 3) GPP numbers are similar betweenJULES ORCHIDEE and LPJ-DGVM for the Northern Sa-vannah Belt the Central Tropical Forest and the SouthernRaingreen Woodland where LPJ-Guess represents only 50percent of other modeled GPP numbers GPP estimates for



Fig 2 Comparison of Atmospheric Inversion and Models based on standardized NEP(a) and NBP(b) (1997ndash2006) Due to the standardiza-tion method 0 represents the mean of each individual distribution but no neutral balance NEP for models was calculated as the differenceof GPP and TER NEP from atmospheric inversions was calculated by removing fire emissions using GFED (van der Werf et al 2006) NBPrefers to the originally inversion results and NEP minus model specific fire emissions (except JULES which does not simulate fire)

the Horn of Africa and the South African Grassland are con-sistent between JULES and LPJ-DGVM but significantlylower for ORCHIDEE but in the same range as LPJ-GuessAnnual NPP estimates are more consistent between all mod-els Exceptions are lower LPJ-Guess NPP at the NorthernSavannah Belt twice as much higher ORCHIDEE NPP thaninter-model average at the Central Tropical Forest as wellas a high and low plateau situation in the South AfricanGrassland region as represented by JULESLPJ-DGVM andORCHIDEELPJ-GUESS

Despite the discrepancies among models regarding the ab-solute flux magnitudes consistent patterns emerge betweenmodels and with satellite observations regarding seasonalchanges of photosynthesis in the different regions (Fig 4)Both remotely sensed FAPAR and simulated GPP show thatthe seasonality of photosynthesis varies in concert with rain-fall in all regions except for the inner tropical forest wherelow intra-annual variation of rainfall creates climatic con-ditions without water limitation reducing the effect of pre-cipitation on photosynthesis The heterogeneous vegetationof Madagascar together with known extensive anthropogenictransformations (Green and Sussman 1990) are likely a ma-jor reason for the disagreement between the models and be-tween models and remotely sensed FAPAR

32 Regions of large interannual variability

Similar spatial patterns of interannual variability for GPP andNEP are estimated by all models (Fig 5) Despite some dif-ferences of the patterns among the models we can identifyareas of large interannual variability of GPP and NEP in eastand south Africa predicted by all models (Fig 6) Both re-gions are dominated by herbaceous vegetation including ex-tensive agricultural land and are known to be strongly influ-enced by El Nino conditions (Plisnier et al 2003 Kogan2000) The independent remote sensing based FAPAR dataconfirms these two variability hotspot regions (Fig 7) How-ever ORCHIDEE and JULES further simulate considerablevariability in the northern savannahs and partly in the tropicalforest which is not evident in the satellite observations

33 What drives NEP interannual variability ndash Photosyn-thesis or respiration

The previous section indicated that regions of large inter-annual NEP variability are associated with also large GPPvariability suggesting that variations of photosynthesis aredriving the variations of the net carbon balance Table 2 fur-ther shows that NEP anomalies in the five defined regions



Fig 3 Annual average GPP NPP and NEP per region (g Cm2yr)

are strongly correlated with GPP anomalies in all models Inmany cases anomalies of ecosystem respiration are also pos-itively correlated with the NEP anomalies which may appearcounterintuitive at first glance If respiration would drivethe carbon balance we would expect negative correlationsthe positive correlations instead originate from the tight cou-pling of GPP and TER in the models which is also evidentin eddy-covariance based estimates of GPP and TER (egReichstein et al 2007 Baldocchi 2008) In such case a pos-itive anomaly of TER results in a positive anomaly of NEPbecause the GPP increase is even higher than the (GPP in-duced) TER stimulation

Although the overall pattern of GPP controlled NEP in-terannual variability is robust and consistent among modelsand regions we find differences in the strength of this re-lationship depending on the model and region JULES andLPJ-GUESS show strongest correlations between GPP andNEP anomalies (RGPPNEPgt092) in all regions while LPJ-DGVM and ORCHIDEE simulate larger inter-regional vari-ability and slightly weaker correlations

4 The response of simulated gross primary productionto meteorological conditions

Having identified GPP as a crucial driver of the carbon bal-ance and the hotspot regions of largest interannual variabilityin east and South Africa we now infer which meteorologicalconditions drive the variability of GPP We use a principal

Table 2 Correlation coefficient (R) between GPP and NEP anoma-lies (left) and TER and NEP anomalies 1 (right) per region andmodel

Northern Savannah Belt R (GPPNEP) R (TERNEP)

LPJ DGVM 071 019LPJ GUESS 097 082ORCHIDEE 083 067JULES 099 099Central African tropical forest R (GPP NEP) R (TER NEP)LPJ DGVM 083 057LPJ GUESS 092 067ORCHIDEE 079 063JULES 094 091Horn of Africa R (GPPNEP) R (TERNEP)LPJ DGVM 094 084LPJ GUESS 099 092ORCHIDEE 082 060JULES 099 099Southern raingreen woodland R (GPPNEP) R (TERNEP)LPJ DGVM 089 064LPJ GUESS 099 091ORCHIDEE 047 012JULES 096 093South African Grassland R (GPPNEP) R (TERNEP)LPJ DGVM 094 078LPJ GUESS 100 096ORCHIDEE 081 060JULES 100 099

Table 3 Eigenvectors and percent variance per Principal Compo-nent Modes

Eigenvectors precipitation temp2m iswrad shumid variance

Mode 1 041 020 minus026 042 5509Mode 2 000 minus079 minus061 minus001 2519Mode 3 minus051 067 minus087 minus035 1557Mode 4 167 014 minus015 minus179 414

component analysis to reduce the dimensionality of the me-teorological dataset and to extract typical weather gradients(see Sect 24) The first two modes of the PCA of the annualmeteorological data explain 80 of its variability (Table 3)and are used to infer the primary driving factor of GPP in-terannual variability in the models The first mode is moststrongly associated with precipitation and specific humidityand thus represents a gradient of moisture availability Thefactor loadings of the second mode show that it representsa gradient of increasing temperature and radiation with de-creasing values of mode 2

The correlation maps between the first principal compo-nent and GPP clearly show that moisture availability is themain limiting factor for photosynthesis (Fig 8) This find-ing is in conjunction with Williams et al (2008) who iden-tified water stress as the primary governing factor of IAV



Fig 4 Standardized mean seasonal cycle GPP from SEAWIFS-FAPAR and participating models

Fig 5 Standardized interannual variability of modeled GPP (top)NEP (bottom) from left LPJ-DGVM ORCHIDEE JULES LPJ-GUESS

Fig 6 Model agreement of interannual variability defined as countof models per grid cell where standardized interannual variability isgreater than one standard deviation (GPP (left) NEP (right) 1982ndash2006)

Fig 7 GPP model agreement of interannual variability defined ascount of models per grid cell where standardized interannual vari-ability is greater than one standard deviation 1998ndash2005 (left) andstandardized SeaWiFs-FAPAR interannual variability greater onestandard deviation 1998ndash2005 (right)

of photosynthesis Correlation maps between the meteoro-logical PCAs and the mean annual FAPAR from SeaW-iFS (Fig 9) confirm that primary productivity responds tomoisture in south and east Africa which are the regions oflargest interannual variability (see Sect 32) The correla-tions between moisture and FAPAR on interannual scale inthe Northern Savannah Belt are not as extensive as indicatedby the models This is possibly an artifact related to the rel-atively short time series (8 years) and coarse meteorologicaldata Camberlin et al (2006) found more widespread strongcorrelations between integrated NDVI and rainfall for a 20year time series in the northern savannahs We also findno correlations between moisture and FAPAR in the Horn



Fig 8 Correlation between modeled GPP and Meteorology PCA1 not significant correlations in grey (confidence interval=090)

Fig 9 Top Correlation between FAPAR and Meteorology PCA 1(left) and PCA 2 (right) not significant correlations in grey (con-fidence interval=090) bottom Model Agreement of significant(confidence interval=090) positive correlations between PCA1 andGPP (left) and of significant negative correlations between PCA2and GPP (right)

of Africa region while the models suggest such relationshipFor this region Camberlin et al (2006) confirm no correla-tion between precipitation and vegetation productivity for thenorthern corner of the African Horn but indicate strong cor-relations with rainfall in the southern part of the region


The partly different spatial correlation patterns betweenPCA1 and GPP among the models is likely related to dif-ferent parameterizations for water stress effects such as rootprofiles soil depth and the coupling between photosynthesisand transpiration via canopy conductance which largely de-termines soil water depletion The simulation of water stresseffects on photosynthesis has been previously identified as amajor source for differences among models regarding inter-annual variability of GPP in Europe (Jung et al 2007 Vetteret al 2008)

The general water limitation of African vegetationrsquos pri-mary production is consistent with Churkina and Run-ning (1998) based on Biome-BGC simulations and Jollyet al (2005) based on the analysis of remote sensing datafrom MODIS Using a production efficiency model Nemaniet al (2003) find that tropical Africa is primarily radiationlimited while both Churkina and Running (1998) as wellas Jolly et al (2005) masked the inner tropical regions andconcluded that no climatic constrain limits productivity hereInterestingly LPJ-DGVM and JULES also suggest that notrainfall but radiation limits GPP in some regions especiallyin the inner tropics This is represented by the negativecorrelations between PCA 2 and GPP in the tropical for-est where simulated GPP increases with increasing radiationand temperature (Fig 10) Given that temperature limitationdoes likely not play a role here we can interpret the neg-ative correlations between PCA2 and GPP as an indicationfor light limitation The latter also makes sense since fre-quent cloud cover is present in the tropics which controlsincoming radiation with otherwise favorable climatic condi-tions for productivity Light limitation for parts of the Ama-zon tropical forests has been inferred from studying wet to



dry season transitions using in-situ measurements of CO2gas exchange (Saleska et al 2003) manipulative experi-ments (Graham et al 2002) and remote sensing based stud-ies (Huete et al 2006 Xiao et al 2006 Nemani et al2003) Light limitation for parts of the Amazon result fromaccess to groundwater during dry seasons when the vegeta-tion capitalizes from the increases incoming radiation Fromthe correlation maps between the remote sensing based FA-PAR and PCA2 we also find patches in the inner tropics ofAfrica where productivity increases with increasing radiationbut to a much smaller extent than suggested by some of themodels Thus the role of light availability in the Africantropics remains controversial There are several possible rea-sons why we find only small areas with significant corre-lation between radiation and productivity from the remotesensing data (1) the short time series (8 years) in combina-tion with small variances make it difficult to achieve largecorrelations (2) the uncertainty of the meteorological datathat originate from global reanalysis (3) possible errors inthe satellite FAPAR retrievals due to subpixel cloud contam-ination (4) a strong role of non-climatic limitations of pro-ductivity such as nutrients If nutrients are not most limitingecosystems are more sensitive to climatic variations Possi-bly light limitation in the African tropics occurs where lessphosphorous depleted soils are present since phosphorous isbelieved to be generally the main limiting factor for produc-tivity in the tropics (Vitousek 1984) The overriding effectof nutrient availability may explain why we find only smallareas with light limitation from the analysis using the satel-lite data in contrast to the extensive areas of light limitationindicated by LPJ-DGVM JULES and Nemani et al (2003)because none of the latter considers explicitly nutrient cycles

5 Conclusions

Using four terrestrial ecosystem models in combination withremote sensing based information of vegetation productivitywe were able to identify that (1) the largest interannual vari-ability of gross primary production and net ecosystem pro-ductivity are concentrated in east and south Africa (2) inter-annual variations of gross primary production is driving netecosystem production in the models and (3) the availabilityof moisture is the primary determinant of interannual varia-tions of gross primary production and consequently the netcarbon balance

Nevertheless our current simulations reveal substantialdiscrepancies among models regarding the actual flux mag-nitudes Future simulations should be performed with im-proved forcings by prescribing the actual distribution of veg-etation types provided by a remote sensing based land covermap and improved meteorological reanalysis In addition itshould be investigated to what extent the models are able toreproduce the ecological patterns found in the synthesis of in-situ measurements of carbon and water fluxes across Africa

(Merbold et al 2009) Given the significance of moisturecontrol on the carbon fluxes in most parts of Africa particu-lar attention should be dedicated to the modelsrsquo ability to ac-curately simulate soil hydrological conditions and the sensi-tivity of photosynthesis and soil respiration to soil moisture

AcknowledgementsThe work was supported by the EuropeanCommission via the FP6 project CarboAfrica (EU Contract No037132) We thank Neil Hanan and Christopher Williams for theirsupport on the experimental setup and implementation

Edited by F Joos

This Open Access Publication isfinanced by the Max Planck Society

References

Anyamba A Tucker C J and Mahoney R From El Nino toLa Nina Vegetation Response Patterns over East and SouthernAfrica during the 1997ndash2000 Period J Clim 15(21) 3096ndash3103 2002

Anyamba A Justice C O Tucker C J and Mahoney R Sea-sonal to interannual variability of vegetation and fires at SA-FARI 2000 sites inferred from advanced very high resolutionradiometer time series data J Geophys Res 108(D13) 8507doi1010292002JD002464 2003

Baldocchi D rsquoBreathingrsquo of the terrestrial biosphere lessonslearned from a global network of carbon dioxide flux measure-ment systems Australian J Botany 56 1ndash26 2008

Camberlin P Martiny N Philippon N and Richar Y Deter-minants of the interannual relationships between remote sensedphotosynthetic activity and rainfall in tropical Africa RemoteSens Environ 106(2) 199ndash216 2006

Cao M K Zhang Q F and Shugart H H Dynamic responsesof African ecosystem carbon cycling to climate change ClimRes 17 183ndash193 2001

Churkina G and Running S Contrasting climatic controls on theestimated productivity of global terrestrial biomes Ecosystems1 206ndash215 1998

Ciais P Reichstein M Viovy N Granier A Ogersquoe J Al-lard V Aubinet M Buchmann N Bernhofer Chr CarraraA Chevallier F De Noblet N Friend A D FriedlingsteinP Gruenwald T Heinesch B Keronen P Knohl A Krin-ner G Loustau D Manca G Matteucci G Miglietta FOurcival J M Papale D Pilegaard K Rambal S SeufertG Soussana J F Sanz M J Schulze E D Vesala Tand Valentini R Europe-wide reduction in primary produc-tivity caused by the heat and drought in 2003 Nature 437doi101038nature03972 2005

Collatz G J Ball J T Grivet C and Berry J A Physio-logical and environmental-regulation of stomatal conductancephotosynthesis and transpiration a model that includes a lami-nar boundary-layer Agric Forest Meteorol 54(2ndash4) 107ndash1361991



Collatz G J Ribas-Carbo M and Berry J A Coupledphotosynthesis-stomatal conductance model for leaves of C4plants Aust J Plant Physiol 19(5) 519ndash538 1992

Cox P M Description of the TRIFFID Dynamic Global Vegeta-tion Model Technical Note 24 Hadley Centre Met Office UK16 2001

Cramer W Kicklighter D W Bondeau A Moore B ChurkinaG Nemry B Ruimy A and Schloss A L Comparing globalmodels of terrestrial net primary productivity (NPP) overviewand key results Glob Change Biol 5 1ndash15 1999

Cramer W Bondeau A Woodward F I Prentice I C BettsR A Brovkin V Cox P M Fisher V Foley J FriendA D Kucharik C Lomas M R Ramankutty N Sitch SSmith B White A and Young-Molling C Global responseof terrestrial ecosystem structure and function to CO2 and cli-mate change results from six dynamic global vegetation modelsGlob Change Biol 7(4) 357ndash373 2001

Essery R L H Best M J and Cox P M MOSES 22 TechnicalDocumentation Technical Note 30 Hadley Centre Met OfficeUK 30 2001

Farquhar G D von Caemmerer S and Berry J A A biogeo-chemical model of photosynthesis in leaves of C3 species Planta149 78ndash90 1980

Fekete B M Vorosmaty C J Roads JO and Willmott C JUncertainties in Precipitation and Their Impacts o Runoff Esti-mates J Climate 17 294ndash304 2003

Friedlingstein P Cox P Betts R Bopp L Von Bloh WBrovkin V Cadule P Doney S Eby M Fung I Bala GJohn J Jones C Joos F Kato T Kawamiya M KnorrW Lindsay K Matthews H D Raddatz T Rayner P Re-ick C Roeckner E Schnitzler K G Schnur R StrassmannK Weaver A J Yoshikawa C and Zeng N Climate-carboncycle feedback analysis Results from the (CMIP)-M-4 modelintercomparison J Clim 19 3337ndash3353 2006

Foley J A Prentice I C Ramankutty N Levis S Pollard DSitch S and Haxeltine A An integrated biosphere model ofland surface processes terrestrial carbon balance and vegetationdynamics Global Biogeochem Cy 10( 4) 603ndash628 1996

Gobron N Pinty B Verstraete M M and Widlowski J-LAdvanced vegetation indices optimized for up-coming sensorsDesign performance and applications Geosci Remote Sens38(6) 2489ndash2505 2000

Gobron N Pinty B Melin F Taberner M Verstraete M MBelward A Lavergne T and Widlowski J-L The state ofvegetation in Europe following the 2003 drought Int J RemoteSens Lett 26 2013ndash2020 2005

Graham F A Mulkey S S Kitajima K Phillips N G andWright S S Cloud cover limits net CO2 uptake and growthof a rainforest tree during tropical rainy seasons PNAS 100(2)572ndash576 2003

Green G M and Sussman R W Deforestation history of theeastern rain forests of Madagascar from satellite images Science248 212ndash215 1990

Huete A R Didan K Shimabukuro Y E Ratana P SaleskaS R Hutyra L R Yang W Nemani R R and Myneni RAmazon rainforests green-up with sunlight in dry season Geo-phys Res Lett 33 L06405 doi1010292005GL025583 2006

Hughes J K Valdes P J and Betts R Dynamics of a global-scale vegetation model Ecol Model 198(3ndash4) 452ndash462 2006

IPCC 2007 Climate Change 2007 The Physical Science BasisContribution of Working Group I to the Fourth Assessment Re-port of the Intergovernmental Panel on Climate Changem editedby Solomon S Qin D Manning M Chen Z Marquis MAveryt K B Tignor M and Miller H L Cambridge Univer-sity Press Cambridge UK and New York NY USA

Jung M Vetter M Herold M Churkina G Reichstein M Za-ehle S Cias P Viovy N Bondeau A Chen Y TrusilovaK Feser F and Heimann M Uncertainties of modelling GPPover Europe A systematic study on the effects of using differ-ent drivers and terrestrial biosphere models Global BiogeochemCy 21 GB4021 doi1010292006GB002915 2007

Jung M Verstraete M Gobron N Reichstein M Papale DBondeau A Robustelli M and Pinty B Diagnostic assess-ment of European gross primary production Global Change Bi-ology 14 1ndash16 doi101111j1365-2486200801647x 2008

Jolly W Nemani R and Running S W A generalized bio-climatic index to predict foliar phenology in response to cli-mate Global Change Biol 11 619ndash632 doi101111j1365-2486200500930x 2005

Kalnay E Kanamitsu M Kistler R Collins W Deaven DGandin L Iredell M Saha S White G Woollen JZhuy Chelliah M Ebisuzaki W Higgins W Janowiak J MoKC Ropelewski C Wang J Leetmaa A and Reynolds RThe NCEPNCAR 40-year reanalysis project B Amer MeteorSoc 77 437ndash470 1996

Kanamitsu M Ebisuzaki W Woollen J Yang S-K HniloJ J Fiorino M and Potter G L NCEP-DEO AMIP-II Reanalysis (R-2) B Atmos Meterol Soc 1631ndash1643doi101175BAMS-83-11-1631 November 2002

KistlerR Kalnay E CollinsW SahaS WhiteG WoollenJChelliah M Ebisuzaki W Kanamitsu M Kousky V vanden Dool H Jenne R and Fiorino M The NCEPndashNCAR 50-Year Reanalysis Monthly Means CD-ROM and DocumentationB Am Meteorol Soc 82 247ndash267 2001

Kogan F N Satellite ndash Observed Sensitivity of World LandEcosystems to El NinoLa Nina Remote Sens Environ 74445ndash462 2000

Krinner G Viovy N de Noblet-Ducoudre N Ogee J PolcherJ Friedlingstein P Ciais P Stitch S and Prentice C A dy-namic global vegetation model for studies of the coupled atmo-sphere biosphere system Global Biogeochem Cy 19 GB1015doi1010292003GB002199 2005

Kummerow C Barnes W Kozu T Shiue J and Simpson JThe Tropical Rainfall Measuring Mission (TRMM) sensor pack-age J Atmos Ocean Tech 15 809ndash817 1998

Le Page Y Pereira J M C Trigo R Da Camara C OomD and Mota B Global fire activity patters (1996ndash2006)and climatic influence an analysis using the World Fire AtlasAtmos Chem Phys 8 1911ndash1924 2008httpwwwatmos-chem-physnet819112008

Linacre E T Net Radiation to Various Surfaces J Appl Ecol6(1) 61ndash75 1969

Lloyd J and Taylor J A On the temperature dependence of soilrespiration Funct Ecol 8 315ndash323 1994

McGuire A D Sitch S Clein J S Dargaville R Esser G Fo-ley J Heimann M Joos F Kaplan J Kicklighter D WMeier R A Melillo J M Moore B Prentice I C Ra-mankutty N Reichenau T Schloss A Tian H Williams



L J and Wittenberg U Carbon balance of the terrestrialbiosphere in the twentieth century Analyses of CO2 climateand land use effects with four process-based ecosystem modelsGlobal Biogeochem Cy 15 183ndash206 2001

Mitchell T D and Jones P D An improved method of con-structing a database of monthly climate observations and as-sociated high-resolution grids Int J Climatol 25 693ndash712doi101002joc1181 2005

Morales P Sykes M T Prentice I C Smith P Smith B Bug-mann H Zierl B Friedlingstein P Viovy N Sabate SSanchez A Pla E Gracia C A Sitch SArneth A andOgee J Comparing and evaluating process-based ecosystemmodel predictions of carbon and water fluxes in major Europeanforest biomes Global Change Biol 11 1ndash23 2005

Myneni R B Los S O and Tucker C J Satellite based identi-fication of linked vegetation index and sea surface anomaly areasfrom 1982ndash1990 for Africa Australia and South America Geo-phys Res Lett 23(7) 729ndash732 1996

Nemani R R Keeling C D Hashimoto H Jolly W MPiper S C Tucker C T Myneni R B and Running SW Climate-Driven Increases in Global Terrestrial Net Pri-mary Production from 1982 to 1999 Science 300 5625doi101126science1082750 2003

Ngo-Duc T Polcher J and Laval K A 53-year dataset for land surface models J GeophyRes 110 D06116doi1010292004JD005434 2005

Plisnier P D Serneels S and Lambin E F Impact of ENSOon East African ecosystems a multivariate analysis based on cli-mate and remote sensing data Global Ecol Biogeogr 9 481ndash497 2000

Potter C S Klooster S and Brooks V Interannual variability interrestrial net primary production Exploration of trends and con-trols on regional to global scales Ecosystems 2 36ndash48 1999

Potter C Klooster S Myneni R Genovese V Tan P-Nand Kumar V Continental-scale comparisons of terrestrial car-bon sinks estimated from satellite data and ecosystem modeling1982ndash1998 Glob Planet Change 39 201ndash213 2003

Randerson J T van der Werf G R Giglio L Collatz G Jand Kasibhatla P S Global Fire Emissions Database Version2 (GFEDv21) Data set available onlinehttpdaacornlgovfrom Oak Ridge National Laboratory Distributed Active ArchiveCenter Oak Ridge Tennessee USA 2007

Reichstein M Papale D Valentini R Aubinet M BernhoferC Knohl A Laurila T Lindroth A Moors E PilegaardK and Seufert G Determinants of terrestrial ecosystem carbonbalance inferred from European eddy covariance flux sites Geo-phys Res Lett 34 L01402262 doi1010292006GL0278802007

Rodenbeck C Estimating CO2 sources and sinks from atmo-spheric mixing ratio measurements using a global inversion ofatmospheric transport Technical Report 6 Max Planck Institutefor Biogeochemistry Jena 61 2005

Saleska S R Miller S D Matross D M Goulden M LWofsy S C Da Rocha H R De Camargo P B Crill PDaube B C De Freitas H C Hutyra L Keller M Kirch-hoff V Menton M Munger J W Hammond Pyle E Rice AH and Silva H Carbon in Amazon Forest Unexpected Sea-sonal Fluxes and Disturbance ndash Induced Losses Science 3021554ndash1557 2003

Sheffield J Goteti G and Wood E F Development of a 50-yearHigh Resolution Global Dataset of Meteorological Forcings forLand Surface Modeling J Clim 19 3088ndash3110 2006

Sitch S Smith B Prentice I C Arneth A Bondeau ACramer W Kaplan J O Levis S Lucht W Sykes M TThonicke K and Venevsky S Evaluation of ecosystem dy-namics plant geography and terrestrial carbon cycling in the LPJdynamic global vegetation model Glob Change Biol 9 161ndash185 2003

Smith T M Shugart H H and Woodward F I (Eds) PlantFunctional Types Cambridge University Press Cambridge1997

Smith B Prentice I C and Sykes M T Representation ofvegetation dynamics in the modelling of terrestrial ecosystemscomparing two contrasting approaches within European climatespace Global Ecol Biogeogr 10 621ndash637 2001

Shugart H H A Theory of Forest Dynamics The Ecological Im-plications of Forest Succession Models Springer-Verlag NewYork USA 1984

Tempel P Batjes N H Collaty G J and van Engelen V W PIGBP-DIS soil data set for pedotransfer function developmentWorking paper and Reprint 9605 International Soil Referenceand Information Centre (ISRIC) Wageningen 1996

TRMM 3B43- Tropical Rainfall Measuring Mission Science Dataand Information System (TSDIS) Interface Control Specifica-tion ftpdisc2nascomnasaGovdataTRMMGridded3B43V6 last access 31 May 2007

Thonicke K Venevsky S Sitch S and Cramer W The roleof fire disturbance for global vegetation dynamics coupling fireinto a Dynamic Global Vegetation Model Global Ecol Bio-geogr 10 661ndash677 2001

Van der Werf G R Randerson J T Giglio L Collatz G Jand Kasibhatla P S Interannual variability in global biomassburning emission from 1997 to 2004 Atmos Chem Phys 63423ndash3441 2006

Vetter M Churkina G Jung M Reichstein M Zaehle SBondeau A Chen Y H Ciais P Feser F Freibauer AGeyer R Jones C Papale D Tenhunen J Tomelleri ETrusilova K Viovy N and Heimann M Analyzing thecauses and spatial pattern of the European 2003 carbon fluxanomaly using seven models Biogeosciences 5 561ndash583 2008httpwwwbiogeosciencesnet55612008

Vitousek M Litterfall Nutrient Cycling and Nutrient Limitationin Tropical Forests Ecology 65 285ndash298 1984

Williams C W Hanan N P Neff J C Scoles R J Berry J ADenning A S and Baker D F Africa and the global carboncycle Carbon Balance and Management 23 doi1011861750-0680-2-3 2007

Williams C A Hanan N P Baker I Collatz G J Berry Jand Denning A S Interannual variability of photosynthesisacross Africa and its attribution J Geophys Res 113 G04015doi1010292008JG000718 2008

Xiao X Hagen S Zhang Q Keller M and Moore III BDetecting leaf phenology of seasonally moist tropical forestsin South America with multi-temporal MODIS images RemoteSens Environ 103 465ndash473 2006

Trends in Atmospheric Carbon Dioxide ndash Mauna Loawwwesrlnoaagovgmdccggtrends access 12 July 2007



high in particular regarding the spatial localization of hotspotregions of variability and the underlying driving forces Sim-ulations of terrestrial ecosystem models can provide insightshere however these models are also associated with largeuncertainties (eg McGuire et al 2001 Friedlingstein et al2006) in particular for water limited conditions (eg Moraleset al 2005 Jung et al 2007) and in addition have gener-ally not been tested and parameterized specifically for AfricaThus confidence of a single model analysis is limited and amulti-model study is warranted to identify coherent and dis-similar behavior between different biosphere models

In this study we assess the interannual variability ofAfricarsquos carbon cycle using four different terrestrial carboncycle models in conjunction with remotely sensed indicatorsfor the state of the vegetation as well as carbon balance esti-mates from global atmospheric inversions We aim to iden-tify (1) regions of largest carbon balance interannual variabil-ity (2) the primary process (photosynthesis or respiration)dominating the carbon balance variability and (3) which cli-mate variables are driving the ecosystemrsquos carbon cycle inthe models

2 Materials and methods

21 Model descriptions

The four ecosystem models ORCHIDEE (Krinner et al2005) LPJ-DGVM (Sitch et al 2003) LPJ-GUESS (Smithet al 2001) and JULESTRIFFID (Cox et al 2001 Es-sery et al 2001 Hughes et al 2006) applied in thisstudy are coupled biogeography-biogeochemistry modelsie they combine representations of both vegetation dynam-ics and land-atmosphere carbon and water exchanges (dy-namic global vegetation models ndash DGVMs) The concept ofplant functional types (PFT) (Smith et al 1997) is used todiscretize differences in physiology and allometry of speciesincluding adaptations to climatic conditions and disturbanceregime PFTs compete for resources like light and waterRepresentations of vegetation structure like allometry andfunction like phenology allocation mortality and establish-ment are essential for this Gross primary production (GPP)is calculated based on a coupled photosynthesis-water bal-ance scheme after (Farquhar et al 1980 Collatz et al 19911992) where simplifications have been incorporated in thedifferent models as described in detail in the above refer-ences Heterotrophic respiration is assumed to be representedby first-order decay of organic material with decay rates fora few pools (Foley 1995) which depend on temperature andmoisture following (Lloyd and Talor 1994) or a Q10 formulaFire the main disturbance in the region is simulated follow-ing (Thonicke et al 2001) within LPJ-DGVM ORCHIDEEand LPJ-GUESS

The applied models differ in the temporal resolution and inthe resolution of vegetation structure representations While

LPJ-DGVM and LPJ-GUESS are designed as stand-alonemodels running on a daily time step ORCHIDEE andJULES can be coupled to General Circulation Models func-tioning as land-surface schemes Therefore latter models re-solve the diurnal cycle with time steps of 30 to 60 min Usu-ally the LPJ family of models is driven by monthly climaticdata which are interpolated to pseudo-daily using a whethergenerator for precipitation In this study the LPJ-DGVM isdriven by daily climatic inputs In doing so net radiationis approximated from global radiation after Linacre (1969)The LPJ-GUESS has the most advanced representation ofvegetation structure including age with features of a forest-gap model (Shugart 1984) resolving forest succession LPJ-DGVM and ORCHIDEE have intermediate complexity ap-plying the concept of ldquoaverage individualsrdquo for a whole gridcell (Sitch et al 2003) while JULESTRIFFID employs aheuristic approach to determine the vegetation coverage andcarbon allocation to each PFT

22 Model drivers

All models are driven by the same climate data atmosphericCO2 concentration and soil texture type on a 1 grid from1982 to 2006 which has been derived as follows Meteo-rological forcing (near surface air temperature specific hu-midity wind speed radiation and precipitation) originatesfrom 6-hourly NCEP-DOE Reanalysis-2 (Kanamitsu et al2002) that were spatially interpolated to 1 from the origi-nal T62 Gaussian grid Despite substantial improvements ofNCEP R2 over R1 considerable precipitation biases remainin comparison to various independent data sets (Fekete et al2004) To account for these limitations precipitation wascorrected using more reliable data sets based on observationsfrom the satellite based Tropical Rainfall Measuring Mission(TRMM 3B43) available from 1998ndash2006 (Kummerow etal 1998) and from interpolated station data provided by theClimate Research Unit (CRU) from 1961ndash2003 (CRUTR21Mitchell and Jones 2005) The general calibration methodfollows Williams et al (2008) Ngo-Duc et al (2005) andSheffield et al (2006) where the daily values of the origi-nal NCEP data for each grid cell are scaled to match in theirmonthly totals those of the corresponding CRU and TRMMdata For the CRU period (1979ndash1997) precipitation was cor-rected following

CAl NCEPymdh =NCEPymdh times CRUym

NCEPymtimes (1)sum2006

1998TRMMmsum20061998CRUm

For the TRMM period (1998ndash2006) the calibration used

CAl NCEPymdh = NCEPymdh timesTRMMym

NCEPym(2)









24 Data analysis





z(IAV i) =

(IAV i minus IAV

)σIAV














GPP 3968 [173] 1658 [104] 3150 [091] 2980 [120]NPP 1728 [112] 916 [067] 1201 [048] 1538 [077]
















































5 Conclusions





Edited by F Joos


References














































































24 Data analysis





z(IAV i) =

(IAV i minus IAV

)σIAV














GPP 3968 [173] 1658 [104] 3150 [091] 2980 [120]NPP 1728 [112] 916 [067] 1201 [048] 1538 [077]
















































5 Conclusions





Edited by F Joos


References














































































GPP 3968 [173] 1658 [104] 3150 [091] 2980 [120]NPP 1728 [112] 916 [067] 1201 [048] 1538 [077]
















































5 Conclusions





Edited by F Joos


References
















































































































5 Conclusions





Edited by F Joos


References







































































the interannual variability of africa’s ecosystem ... · the lpj-guess has the most advanced...

Documents