Biospheric carbon stocks reconstructed at the Last Glacial Maximum: comparison between general...

22
Biospheric carbon stocks reconstructed at the Last Glacial Maximum: comparison between general circulation models using prescribed and computed sea surface temperatures Dominique Otto * , Daniel Rasse, Jed Kaplan, Pierre Warnant, Louis Franc ßois Laboratoire de Physique Atmosphe ´rique et Plane ´taire, Institut d’Astrophysique et de Ge ´ophysique, Universite ´ de Lie `ge, alle ´e du Six Aou ˆt B5c, B-4000 Lie `ge, Belgium Received 20 March 2001; received in revised form 27 August 2001; accepted 7 September 2001 Abstract The terrestrial biosphere model Carbon Assimilation in the Biosphere (CARAIB) was improved by introducing two vegetation storeys and implementing a new module which simulates the equilibrium distribution of the vegetation inferred from physiological processes and climatic constraints. In this fourth version of CARAIB, we differentiate ground-level grasses from tree canopies, which allows us to determine the light available to grasses as a direct function of the leaf area index (LAI) of the forest canopy. Both of these storeys are potentially composed of several plant functional types (PFT). The cover fraction of each PFT within each storey is estimated according to its respective net primary productivity (NPP). A biome is assigned to each grid cell on the basis of three physiological criteria: (1) the cover fraction, (2) the NPP, and (3) the LAI; and two climatic constraints: (1) the growing degree-days (GDD) and (2) the lowest temperature reached during the cold season (T min ), which are well-known indices of vegetation expansion boundaries. Total biospheric carbon stocks (vegetation + soil) are reconstructed by forcing the model with eight climatic scenarios of the Last Glacial Maximum (LGM, 21 ka BP), which were obtained from the Paleo-Modelling Intercomparison Project (PMIP) from four general circulation models (MRI2, UGAMP, LMD4, and GEN2) using prescribed and computed sea surface temperatures (SSTs). The model was also forced with a current climate together with a preindustrial atmospheric CO 2 level of 280 ppm as reference simulation. To validate the model, current biome distribution is reconstructed and compared, for the modern climate, with two distributions of potential vegetation and, for the LGM, with pollen data. The model simulations are in good agreement with broad-scale patterns of vegetation distribution. The results indicate an increase in the total biospheric carbon stock of 827.8– 1106.1 Gt C since the LGM. Sensitivity analyses were performed to discriminate the relative effects of the atmospheric CO 2 level (‘‘fertilization effect’’), the climate (present or LGM), and the sea level. Our results suggest that the CO 2 fertilization effect is mostly responsible for the total increase in vegetation and soil carbon stocks. The four GCMs diverged in their predicted responses of continental climate to calculated SSTs. Only one of them, i.e., MRI2, predicted a marked decline of the continental temperatures in response to lower calculated SSTs. For this GCM, the effect of reduced SSTs on continental biospheric carbon stocks was a decrease of 544.1 Gt for the soil carbon stock and of 283.7 0921-8181/02/$ - see front matter D 2002 Elsevier Science B.V. All rights reserved. PII:S0921-8181(02)00066-8 * Corresponding author. Tel.: +32-4-366-9780; fax: +32-4-366-9729. E-mail address: [email protected] (D. Otto). www.elsevier.com/locate/gloplacha Global and Planetary Change 33 (2002) 117 – 138

Transcript of Biospheric carbon stocks reconstructed at the Last Glacial Maximum: comparison between general...

Biospheric carbon stocks reconstructed at the Last

Glacial Maximum: comparison between general

circulation models using prescribed and

computed sea surface temperatures

Dominique Otto*, Daniel Rasse, Jed Kaplan, Pierre Warnant, Louis Franc�oisLaboratoire de Physique Atmospherique et Planetaire, Institut d’Astrophysique et de Geophysique, Universite de Liege,

allee du Six Aout B5c, B-4000 Liege, Belgium

Received 20 March 2001; received in revised form 27 August 2001; accepted 7 September 2001

Abstract

The terrestrial biosphere model Carbon Assimilation in the Biosphere (CARAIB) was improved by introducing two

vegetation storeys and implementing a new module which simulates the equilibrium distribution of the vegetation inferred

from physiological processes and climatic constraints. In this fourth version of CARAIB, we differentiate ground-level

grasses from tree canopies, which allows us to determine the light available to grasses as a direct function of the leaf area

index (LAI) of the forest canopy. Both of these storeys are potentially composed of several plant functional types (PFT).

The cover fraction of each PFT within each storey is estimated according to its respective net primary productivity (NPP).

A biome is assigned to each grid cell on the basis of three physiological criteria: (1) the cover fraction, (2) the NPP, and

(3) the LAI; and two climatic constraints: (1) the growing degree-days (GDD) and (2) the lowest temperature reached

during the cold season (Tmin), which are well-known indices of vegetation expansion boundaries. Total biospheric carbon

stocks (vegetation + soil) are reconstructed by forcing the model with eight climatic scenarios of the Last Glacial Maximum

(LGM, 21 ka BP), which were obtained from the Paleo-Modelling Intercomparison Project (PMIP) from four general

circulation models (MRI2, UGAMP, LMD4, and GEN2) using prescribed and computed sea surface temperatures (SSTs).

The model was also forced with a current climate together with a preindustrial atmospheric CO2 level of 280 ppm as

reference simulation. To validate the model, current biome distribution is reconstructed and compared, for the modern

climate, with two distributions of potential vegetation and, for the LGM, with pollen data. The model simulations are in

good agreement with broad-scale patterns of vegetation distribution. The results indicate an increase in the total biospheric

carbon stock of 827.8–1106.1 Gt C since the LGM. Sensitivity analyses were performed to discriminate the relative effects

of the atmospheric CO2 level (‘‘fertilization effect’’), the climate (present or LGM), and the sea level. Our results suggest

that the CO2 fertilization effect is mostly responsible for the total increase in vegetation and soil carbon stocks. The four

GCMs diverged in their predicted responses of continental climate to calculated SSTs. Only one of them, i.e., MRI2,

predicted a marked decline of the continental temperatures in response to lower calculated SSTs. For this GCM, the effect

of reduced SSTs on continental biospheric carbon stocks was a decrease of 544.1 Gt for the soil carbon stock and of 283.7

0921-8181/02/$ - see front matter D 2002 Elsevier Science B.V. All rights reserved.

PII: S0921 -8181 (02 )00066 -8

* Corresponding author. Tel.: +32-4-366-9780; fax: +32-4-366-9729.

E-mail address: [email protected] (D. Otto).

www.elsevier.com/locate/gloplacha

Global and Planetary Change 33 (2002) 117–138

for the vegetation carbon stock, which means a decrease in the total biopsheric carbon stock of 827.8 Gt. D 2002 Elsevier

Science B.V. All rights reserved.

Keywords: prescribed and computed sea surface temperatures; Last Glacial Maximum; vegetation distribution; continental carbon stocks;

climate and vegetation models; asynchronous coupling

1. Introduction

The relationship between biospheric carbon stocks

and atmospheric CO2 concentrations during the gla-

cial– interglacial cycles, i.e., the Pleistocene epoch

during the past 2 million years, may help us under-

stand the mechanisms which will drive climate

change during the 21st century and beyond. Measure-

ments in ice and deep-sea sediment cores revealed

that the atmospheric CO2 concentration has fluctuated

widely in concert with temperature variations during

this period. From the Last Glacial Maximum (LGM,

21,000 years BP) to preindustrial time, atmospheric

CO2 concentration increased from 200 to 280 ppmv

(Petit et al., 1999). Earlier research hypothesised that

the reservoir size and turnover time of the ocean were

the sole responsible factors for this increase and

invoked a redistribution of carbon in the ocean–

atmosphere system attributable to changes in the

oceanic circulation (Broecker and Takahashi, 1984)

and nutrient cycles (Broecker and Peng, 1987). How-

ever, foraminifera data suggest that the d13C value of

oceanic carbon at the LGM was from 0.3x to

0.7xlower than at present (Shackleton, 1977;

Duplessy et al., 1988; Curry et al., 1988). This change

of the oceanic carbon isotopic composition implies

that a transfer of 470–1100 Gt of carbon from the

ocean to the biosphere occurred during deglaciation.

This calculation rests on the hypothesis that the

terrestrial biosphere is the only reservoir having

exchanged carbon with the ocean during the deglaci-

ation, at least at an isotopic signature different from

that of the ocean. It also assumes that the average

d13C fractionation of photosynthesis at the LGM was

the same as today. However, taking into account

possible variations of this photosynthetic fractionation

would only slightly modify this range probably

towards lower values, since C4 species (exhibiting a

lower fractionation factor) are thought to have been

more widespread at the LGM than today (Bird et al.,

1994; Franc�ois et al., 1998). Two additional methods

support the concept that massive amounts of carbon

were removed from the ocean to the land biosphere

during deglaciation. First, reconstructions of paleove-

getation from palynological and sedimentological

proxy data suggest that the increase in the biospheric

carbon stock from the LGM to the present ranges

from 700 to 1600 Gt C (Adams et al., 1990; Van

Campo et al., 1993; Crowley, 1995; Adams and

Faure, 1998). Second, biospheric models forced with

outputs of general circulation models (GCM) estimate

this change to range from 0 to 700 Gt C (Prentice and

Fung, 1990; Friedlingstein et al., 1992, 1995; Prentice

et al., 1993; Esser and Lautenschlager, 1994; Franc�oiset al., 1998, 1999). In addition to increased carbon

stocks in the land biosphere, the dissolution of ter-

restrial calcrete, i.e., soil carbonate, during deglacia-

tion appears as another potential sink of atmospheric

CO2 (Adams and Post, 1999). Both estimates of

change in biospheric carbon stock and calcrete imply

an increased efficiency of the oceanic mechanisms

increasing the atmospheric CO2 level during deglaci-

ation for the net effect to explain the observed 80

ppmv shift between the LGM and the pre-industrial

period.

Sea surface temperatures (SSTs) are pivotal to the

simulations of atmospheric global circulation models

(AGCM), which drive the biospheric models used for

estimating the evolution of LGM carbon stocks.

Tropical SST values during the LGM remain uncer-

tain, which propagates uncertainties to biospheric

model estimates of LGM carbon stocks. The first

global SST reconstructions for the LGM were gen-

erated by CLIMAP Project Members (1981). They

were prescribed based on the abundances of plank-

tonic fossils in deep-sea sediment cores (‘‘fixed

SST’’). CLIMAP reconstructions of tropical SSTs

are on average 1 jC cooler than present-day tropical

SSTs. A considerable controversy has arisen about the

validity of CLIMAP SST estimates. Numerous geo-

D. Otto et al. / Global and Planetary Change 33 (2002) 117–138118

chemical studies indicate that the tropics were 4–6 jCcooler at the LGM than under present conditions and

provide some support for the hypothesis that glacia-

tions were global. These studies include: (1) lowered

tropical snowlines (Rind and Peteet, 1985); (2) fora-

miniferal records (Curry and Oppo, 1997); (3) Sr/Ca

ratios from Barbados corals (Guilderson et al., 1994);

(4) tropical ice cores (Thompson et al., 1995); (5)

noble gases in Brazil aquifer (Stute et al., 1995). This

new evidence of sensitivity of the tropics to climate

change could have dramatic implications for forecasts

of the future global warming. The CLIMAP results

implie that the tropics, which represents 40% of

Earth’s surface, might not be affected by the future

warming. Opposite to these results, geochemical stud-

ies described above suggest that the tropics might

experience similar warming to that of higher latitude

regions. On the other hand, analysis of the temper-

ature-dependent production of alkenone molecules by

marine organisms, which responds to changes in

water temperature by altering the molecular compo-

sition of their cell membranes (Eglinton et al., 1992;

Herbert and Schuffert, 1998), yields small temperature

changes of about 2 jC, closer to the CLIMAP

estimates. Recently, slab ocean models, i.e., atmos-

phere-mixed layer ocean models, (Broccoli, in press),

have simulated air–sea interactions and computed the

SST distribution at the LGM (‘‘calculated SST’’). The

computed tropical cooling is comparable to recon-

structions based on alkenones, but smaller than the

cooling inferred from other geochemical studies.

The objective of this research is to analyse the

sensitivity of LGM biospheric carbon stocks and

vegetation distribution to SSTs. For this purpose, we

used the biospheric model CARAIB (Warnant et al.,

1994; Warnant, 1999) forced with four sets of two

AGCM scenarios, which differed only on their SST

inputs: (1) SST are prescribed based on the CLIMAP

(CLIMAP Project Members, 1981) reconstructions,

and (2) SST are computed with a thermodynamic slab

ocean model integrated as a submodule in the AGCM.

Since most recent geochemical studies indicate that

tropical SST during LGM were cooler than SST

prescribed by CLIMAP, we aim at providing a more

realistic estimate of the change in biospheric carbon

stocks during the last deglaciation. For this study, we

hypothesized that colder SSTs resulted in decreased

continental temperatures.

2. The model

2.1. General structure

CARAIB is a global model of the carbon cycle in

the continental biosphere (Warnant et al., 1994;

Nemry et al., 1996). It calculates carbon fluxes

between the atmosphere and the terrestrial biosphere,

and estimates the evolution of carbon pools resulting

from these fluxes. Five pools are considered: (1) the

leaves (GC, for ‘‘green carbon’’); (2) the rest of the

plant, i.e., branches, stems and roots (Ludeke et al.,

1994) (RC, for ‘‘remaining carbon’’); (3) the litter

from GC (GL, for ‘‘green litter’’); (4) the litter from

RC (RL, for ‘‘remaining litter’’); (5) the humus, i.e.,

the product of litter decomposition (SC, for ‘‘soil

carbon’’). Eight plant functional types (PFTs) are

considered: (1) C3 grasses; (2) C4 grasses; (3) needle-

leaved evergreen trees; (4) needleleaved deciduous

trees; (5) temperate broadleaved evergreen trees; (6)

tropical broadleaved evergreen trees; (7) temperate

broadleaved deciduous trees; (8) tropical broadleaved

deciduous trees. Carbon contents and fluxes in and

out of each pool are estimated daily for each grid cell

and each PFT. The model contains no nutrient cycle. It

should be included in future. The different carbon and

water fluxes are described in the following subsec-

tions.

2.1.1. Soil water budget

The Improved Bucket Model (IBM) developed by

Hubert et al. (1998) is used to calculate the soil water

budget of a soil layer of a given thickness with a time

step of 1 day. This budget is described by the follow-

ing equation

dw

dt¼ P � E � D� SR ð1Þ

where w is the water content in mm of the soil layer

and the water fluxes P, E, D, and SR are respectively

precipitation, actual evapotranspiration, deep drain-

age, and surface runoff in units of mm year � 1.

Precipitation is an input of the model. Actual evapo-

transpiration is estimated as a fraction of the potential

evapotranspiration rate. This fraction depends on soil

wetness, while the potential evapotranspiration rate is

calculated from Penman’s equation (Mintz and

Walker, 1993). Deep drainage, i.e., the downward

D. Otto et al. / Global and Planetary Change 33 (2002) 117–138 119

water flux at the bottom of the soil layer, is estimated

from the soil hydraulic conductivity parameterized as

a function of soil texture, i.e., % sand, % silt, and %

clay, and soil wetness, according to Saxton et al.

(1986). Surface runoff occurs when precipitation is

too high and exceeds maximum infiltration into the

soil. The model requires as inputs daily mean values

of air temperature, precipitation, cloudiness, relative

air humidity, and wind speed. The last variable is used

to calculate the aerodynamic resistance needed in

Penman’s equation. The balance between the two

water fluxes, i.e., precipitation, which is imposed by

the GCM data, and evapotranspiration, which is

driven by wind data, is critical for determining the

vegetation distribution and the extent of desertic

areas.

2.1.2. Photosynthesis

Photosynthesis of C3 and C4 plants is computed

according to the methodologies of Farquhar et al.

(1980) and Collatz et al. (1992), respectively. The

canopy is divided in several layers for computing the

absorption of the photosyntheticaly active radiation

(PAR). PAR is estimated for each layer following the

method of Goudriaan and van Laar (1994), which

separates the effects of direct and diffuse light. Photo-

synthetic fluxes are estimated on a 2-h time step, to

take into account the diurnal cycle of the solar

insolation. The stomatal conductance is from Leu-

ning’s (1995) model modified by Wijk et al. (2000) as

follows: the empirical coefficients of the original

stomatal conductance model from Leuning were opti-

mized by comparing modeled and measured transpi-

ration fluxes, and the stomatal response to soil water

content was incorporated in this formulation by multi-

plying the stomatal conductance by the standard

response function proposed by Jarvis (1976) and

Stewart (1988).

2.1.3. Autotrophic respiration

The autotrophic respiration is divided between two

fluxes: maintenance and growth respiration. Mainte-

nance respiration was parametrized as an exponential

function of temperature and a linear function of

carbon content of the GC or RC pools (Warnant,

1999). The growth respiration is assumed proportional

to the biomass increase (Raich et al., 1991; Parton et

al., 1993; Ruimy et al., 1996). For tree woody tissues,

only sapwood is respiring. The sapwood fraction was

estimated from data reported by Duvigneaud et al.

(1971), Dubroca (1983), Ryan and Waring (1992) and

collected by Ruimy et al. (1996).

2.1.4. Allocation of photosynthates and reserve use

The carbon assimilated during photosynthesis is

partitioned between the GC and the RC pools accord-

ing to the simulated environmental conditions. When

temperature falls between a minimum and a maximum

value and soil water content is above a critical value

(Table 1), which have been calibrated to reproduce a

reasonable vegetation distribution, one half of the

assimilated carbon is allocated to each pool. When

these conditions are not met, the assimilated carbon is

entirely allocated to the RC pool. The GC pool is

limited to a maximum value, defined by an allometric

relationship with the RC content (Ludeke et al., 1994).

A new leaf layer is created only if its productivity is

higher than its mortality rate, which prevents CAR-

AIB from generating leaf layers with a negative car-

bon budget. Budburst is simulated by a transfer of

carbon from the RC pool to the GC pool. The carbon

Table 1

Tresholds used for determination of stress conditions

PFTs Tmin1 Tmin2 Tmax1 Tmax2 SWmin1 SWmin2

(1) C3 grass � 55 � 55 40 40 0.1 0.0

(2) C4 grass 0 0 50 50 0.1 0.0

(3) Needle-leaved

evergreen

boreal/temperate

� 40 � 40 30 30 0.3 0.3

(4) Needle-leaved

summergreen

boreal/temperate

� 50 � 55 30 35 0.3 0.0

(5) Broad-leaved

evergreen

temperate

� 10 � 10 35 35 0.3 0.3

(6) Broad-leaved

evergreen tropical

0 0 45 45 0.2 0.2

(7) Broad-leaved

summergreen

boreal/temperate

0 � 25 40 40 0.3 0.0

(8) Broad-leaved

raingreen tropical

0 0 40 45 0.2 0.0

Tmin is the absolute minimum temperature (jC), Tmax is the absolute

maximum temperature (jC), SWmin is the minimum soil water

content, expressed as a fraction of the field capacity and limited to

the wilting point. Index 1 refers to leaves and index 2 refers to the

rest of the plant.

D. Otto et al. / Global and Planetary Change 33 (2002) 117–138120

available for buds is computed as a fraction of the

maximum value of the GC pool of the previous year.

2.1.5. Litter production and mortality

Litter is producted by the GC and RC mortality,

resulting from three contributions: (1) seasonal leaf fall

for deciduous species controlled by three external

parameters, i.e., temperature, soil water content, and

PAR; (2) plant death due to the natural regeneration of

the canopy, with a mortality characteristic time depend-

ing on the type of plant, i.e., grass or tree, evergreen or

deciduous, and the reservoir, i.e., GC or RC; and (3)

plant death due to unfavourable climatic conditions

(see Table 1) with a characteristic time of 1 week.

2.1.6. Heterotrophic respiration

Heterotrophic respiration is due to organic matter

decomposition by soil bacteria. Litter decomposition

is computed as a function of temperature, soil water

content, and litter carbon content (GL or RL). The

adopted temperature dependence has been fitted by

Nemry et al. (1996) on the data reported by Raich and

Schlesinger (1992) for all major world’s ecosystems.

A fixed fraction (20%) of this flux contributes to the

SC formation (Johnson et al., 1987) and the rest is

returned to the atmosphere as CO2. The SC mineral-

ization is computed using an equation similar to that

for the litter decomposition, but with a proportionality

coefficient reduced by a factor of 100 (Esser, 1984).

2.2. The new biome prediction module

A biome prediction module was developed, using

the NPP and LAI outputs of the CARAIB model,

which allows us to determine the cover fractions of

the different PFTs and to assign a biome to each grid

cell. This assignment is made in three steps: (1) the

application of climatic constraints; (2) the simulation

of competition between PFTs; and (3) the assignment

of a biome.

2.2.1. Application of climatic constraints

Two climatic constraints, which are indices of the

geographical extent of vegetation, have been chosen

with the aim of selecting the species potentially

present in each grid cell (Table 2): the absolute

minimum temperature (Tmin), i.e., the average temper-

ature of the coldest day of the year, and the growing

degree-days based on a threshold temperature of 5 jC(GDD5), defined by

GDD5 ¼X365

i¼1ðbTi>5ÞðTi � 5Þ ð2Þ

where Ti is the mean temperature of day i (jC).According to Woodward (1987), there is a strong

correlation between the distribution of dominant spe-

cies of trees and their winter freezing resistance.

Tropical species die as soon as temperature falls

below 0 jC, while boreal and temperate species must

endure freezing temperatures. Broadleaved evergreen

species resist to � 10 jC. All plants require a period

with temperatures warm enough for growth. At high

latitudes, it is a severe limitation to the growth of

leaves, and therefore establishment of LAI. Two

values of GDD5 have been adjusted: (1) 50 simulates

Table 2

Climatic constraints applied to PFTs

PFTs Tmin GDD5 new PFTs

(1) C3 grass / / (1) C3 grass

(2) C4 grass / / (2) C4 grass

(3) Needle-leaved

evergreen

boreal/temp

< 0 [50, 1350

z 1350

(3) Needle-leaved

evergreen boreal

(4) Needle-leaved

evergreen temperate

(4) Needle-leaved

summergreen

boreal/temp

< 0 [50, 1350]

z 1350

(5) needle-leaved

summergreen boreal

(6) needle-leaved

summergreen

temperate

(5) Broad-leaved

evergreen

temperate

[� 10,0] > 50 (7) broad-leaved

evergreen temperate

(6) Broad-leaved

evergreen

tropical

> 0 > 50 (8) broad-leaved

evergreen tropical

(7) Broad-leaved

summergreen

boreal/temp

< 0 [50, 1350]

z 1350

(9) broad-leaved

summergreen boreal

(10) broad-leaved

summergreen

temperate

(8) Broad-leaved

raingreen

tropical

> 0 > 50 (11) broad-leaved

raingreen tropical

Tmin is the absolute minimum temperature and GDD5 is the growing

degree-days based on a threshold temperature of 5 jC.

D. Otto et al. / Global and Planetary Change 33 (2002) 117–138 121

the limit of the ice sheet extent, and (2) 1350 simu-

lates the limit between temperate and boreal climates,

which allows us to generate three tree PFTs represent-

ing stages between temperate and boreal ecosystems,

and to raise the number of PFTs from 8 to 11 (see the

second column of Table 2). Further in this work, we

will adopt this new set of 11 PFTs, which contains two

grass PFTs and nine tree PFTs.

2.2.2. Simulation of competition between PFTs

Simulation of competition between PFTs results

from the introduction of two vegetation storeys. In

the previous version of the model, there was only one

vegetation storey, which means that all PFTs present in

any given grid cell received the solar irradiance avail-

able at the top of the canopy, and grew independently

from one another without being affected by the poten-

tial shading from surrounding plants. In this new

version, we differentiate ground-level vegetation from

tree canopies, which allows us to determine the light

available to grasses as a direct function of the LAI of

the forest canopy. The canopy is composed of a

maximum of 16 layers with a LAI of 0.5 each. The

bottom layer may be only partly filled with leaves if the

cumulated LAI is not a multiple of 0.5. The carbon

content is evaluated daily for each pool (GC and RC)

and each of the six tree PFTs. Using for each PFT a

constant value of the specific leaf area (SLA), it is then

possible to compute the LAI of all tree PFTs and,

hence, the number of layers occupied by their leaves

(ntree), from the respective GC contents. The net

primary productivity (NPP) is then estimated for each

tree PFT and each day. On the last day of each month,

the cover fractions (fractree,i) are estimated according to

fractree,i ¼NPPi

X8

j¼3

NPPj

ð3Þ

where the index i represents each of the six tree PFT

(see the first column of Table 2) and NPP refers to the

monthly NPP. The average number of layers occupied

by trees is obtained as

ntree ¼X8

i¼3

ntree,i*fractree,i ð4Þ

where ntree,i and ntree are not integers, since they are

derived from the LAI itself proportional to the GC

pool. Then the photosynthesis of grasses is calculated

with the PAR available under the ntree layers of tree

leaves. Finally, the NPP of the two grass PFTs is

estimated by limiting their canopy development to

16� ntree layers, i.e., a LAI of 8� ntree*0.5. Their

cover fraction is estimated such as for trees, i.e.

fracgrass,i ¼NPPi

X2

j¼1

NPPj

ð5Þ

The NPP of each PFT simulated by the model

reflects differences in metabolism between plants, and

their aptitude to use the available water and light

resources, and represents a physiological criteria of

competitivity among species, which is used for the

biomes determination described below.

2.2.3. Assignment of a biome

Biomes are determined according to the classifica-

tion scheme described in Table 3. The new set of 11

PFTs is used. The polar ice sheet and the tundra

biomes are assigned on a climatic criteria, i.e.,

GDD5, as done in other biospheric models (IBIS:

Foley et al., 1996; BIOME3: Haxeltine and Prentice,

1996). The value of 350 jC-day for the tundra biome

is the same as in the BIOME3 and IBIS models, and

the value 50 jC-day for the limit of the ice sheet has

been adjusted to reproduce as well as possible the two

available present potential natural vegetation maps

(Melillo et al., 1993; Matthews, 1983). The other

biomes are assigned on the basis of two outputs of

the CARAIB model, i.e., the LAI and the NPP of each

PFT, or of parameters directly derived from these two

variables. Let us define the LAI of grasses and the

LAI of trees as the weighted sum of the LAI of the

PFTs present in their respective storeys

LAIgrasses ¼X2

i¼1

LAIi*fracgrass,i

LAItrees ¼X11

i¼3

LAIi*fractree,i ð6Þ

These quantities may vary with the season, espe-

cially if deciduous species are present. Thus, in the

framework of the biome assignment criteria, the

D. Otto et al. / Global and Planetary Change 33 (2002) 117–138122

maximum LAI reached by the PFT over the course of

the year is used for LAIi and the cover fractions are

defined on an annual basis. The total LAI of the pixel

is then defined as the sum of the LAI over both

vegetation storeys, i.e.

LAItot ¼ LAIgrasses þ LAItrees ð7Þ

In a similar way, for the NPP, we define

NPPgrasses ¼X2

i¼1

NPPi*fracgrass,i ð8Þ

NPPtrees ¼X11

i¼3

NPPi*fractree,i

where NPPi here is the annual NPP. The NPP of all

PFTs is given by

NPPtot ¼ NPPgrasses þ NPPtrees ð9Þ

The assignment of non-tundra biomes is then

performed as follows. Desert is assigned whenever

NPPtot is equal to 0 and semi-desert is assigned where

LAItot falls below 0.4. Grasslands and grassy savan-

nas are separated from forests and woody savannas

when NPPgrasses is higher than 3/2*NPPtrees, and

distinguished from each other on a LAItrees criteria

fixed at 0.5. The limit between temperate or tropical

forests and woody savannas is fixed at a LAItrees of

1.5. To reduce the number of biomes, the distinction

between grassy and woody savanna is not specified.

The temperate mixed forest is predicted whenever

the cover fraction of any of the temperate tree PFTs

present in the pixel does not reach 0.6. These

thresholds have been adjusted by comparing the

simulated present vegetation distribution with the

two present potential natural vegetation maps used

in this work, i.e., Matthews (1983) and Melillo et al.

(1993).

3. Input data

The climatic fields used to force the CARAIB

model for modern climate, i.e., air temperature, pre-

cipitation, and cloud cover, are monthly mean

observed fields from Cramer et al. (unpublished data,

1995), an updated version of the dataset of Leemans

and Cramer (1991). These climatic data correspond to

a long-term mean over the period 1931–1960. They

are widely used within the biospheric community, as

for instance in the Potsdam ’95 intercomparison of

Table 3

Biome assignment scheme

Biome GDD5 NPPtot LAItot R LAItrees D F

(1) Ice < 50

(4) Tundra [50, 350]

(2) Desert >350 = 0

(3) Semi-desert >350 p 0 < 0.4

(5) Grassland >350 p 0 z 0.4 >1.5 < 0.5

(6) Temperate/tropical savanna >350 p 0 z 0.4 >1.5 z 0.5

(13) Boreal evergreen forest/woodland >350 p 0 z 0.4 V 1.5 3

(14) Boreal deciduous forest/woodland >350 p 0 z 0.4 V 1.5 5 or 9

(6) Temperate/tropical savanna >350 p 0 z 0.4 V 1.5 < 1.5 all � {3,5,9}

(11) Temperate mixed forest >350 p 0 z 0.4 V 1.5 z 1.5 4,6,7 or 10 < 0.6

(12) Temperate conifer forest >350 p 0 z 0.4 V 1.5 z 1.5 4 or 6 z 0.6

(9) Temp broad-leaved evergreen forest >350 p 0 z 0.4 V 1.5 z 1.5 7 z 0.6

(10) Temp broad-leaved deciduous forest >350 p 0 z 0.4 V 1.5 z 1.5 10 z 0.6

(7) Tropical rainforest >350 p 0 z 0.4 V 1.5 z 1.5 8

(8) Tropical seasonal forest >350 p 0 z 0.4 V 1.5 z 1.5 11

GDD5 is the growing degree-days based on a threshold temperature of 5 jC.NPPtot and LAItot are the sum of the NPP and LAI of each PFT weighted by their respective cover fraction. R is the ratio between the total NPP

of the two grass PFTs (NPPgrasses) and the total NPP of the nine tree PFTs (NPPtrees). D is the dominant tree PFT. F is the cover fraction of the

dominant tree PFT. PFTs are numbered from 1 to 11 (see the new set of PFTs in Table 2).

D. Otto et al. / Global and Planetary Change 33 (2002) 117–138 123

biospheric models (Cramer et al., 1999). Absolute

anomalies of these fields between the LGM and the

present are calculated from four different AGCM

models participating in the Paleo-Modelling Inter-

comparison Project (Joussaume and Taylor, 1995):

MRI2 (Meteorological Research Institute-Japan),

UGAMP (Universities Global Atmospheric Model-

ling Programme-UK), LMD4 (Laboratoire de Mete-

orologie Dynamique-France) and GEN2 (GENESIS

model, National Center for Atmospheric Research-

USA). Model descriptions can be found on the web

at http://www-pcmdi.llnl.gov/pmip. These anomalies

are then added to Cramer et al.’s (1999) present

dataset to obtain LGM fields of air temperature,

precipitation and cloud cover. Two sets of SST fields

were used as boundary conditions for each GCM: (1)

SST prescribed from estimates given by CLIMAP

Project Members (1981) and (2) SST computed

using coupled atmosphere-mixed layer ocean models

(Dong and Valdes, 1998). The CARAIB model also

requires surface horizontal wind speed and air rela-

tive humidity fields to calculate evapotranspiration

fluxes. For present-day conditions, these fields were

obtained from monthly average ECMWF data for

1991. The relative humidity data were missing from

the GCM outputs. The present-day field was used for

the LGM simulation after extrapolation to the LGM

grid, which is larger than the present-day one due to

a lower sea level, using a distance weighted average

procedure. The relative anomalies of wind speed

between the LGM and the present simulations of

the GCM were calculated and multiplied by the

observed field to obtain the corrected LGM one.

The relative anomaly Dqrel is calculated by

Dqrel ¼qLGM

qpresð10Þ

and the relative corrected field qrel,corr is given by

qrel,corr ¼ qobs*Dqrel ð11Þ

where qLGM and qpres are the values of the variable

in the GCM simulations for LGM and present

conditions, respectively, while qobs is the observed

present-day value from Cramer et al. (1999). In total,

eight LGM climatic fields were used to force the

CARAIB model, and three modern simulations were

performed. Their characteristics are summarized in

Table 4. The CARAIB model contains a stochastic

generator of meteorological variables (Hubert et al.,

1998), which transforms the monthly mean data into

diurnal values. In the procedure used, a normaliza-

tion is performed after the stochastic generation to

ensure that the monthly mean values of the variables

are not altered. GCM outputs were interpolated down

to a 0.5� 0.5j latitude–longitude resolution.

The glacial atmospheric CO2 level was set to 200

ppmv (Petit et al., 1999). For modern simulations, it

was set to its pre-industrial value, i.e., 280 ppmv,

while we consider that vegetation is clearly out of

equilibrium with the atmospheric CO2 level since its

rapid increase at the beginning of industrial times. To

be fully consistent, we should have used a pre-

industrial climate instead of a present-day one, but

such a dataset is not available. Anyway, the difference

between the pre-industrial and the modern climates

remains small compared to the change between the

LGM and the present.

The orbital parameters used to calculate the solar

radiation fluxes are the same as in the PMIP experi-

ments (listed in Franc�ois et al., 1999). The present

continental area is 132� 106 km2. At the LGM, the

sea level is lowered by 105 m (Tushingham and

Peltier, 1991) and the continental area is increased

by 22� 106 km2.

Soil texture data from Zobler (1986) were used for

the present. In the absence of any information on past

soil texture, the same dataset was used for the LGM

and extrapolated to the emerged shelf by assuming a

medium soil texture on these pixels.

Table 4

Characteristics and codes of the 11 simulations performed

Code Climate [CO2]atm(ppmv)

Sea level SST

PD280 present-day 280 present-day

PD200 + present-day 200 present-day

PD200� present-day 200 LGM

MRI2C LGM 200 LGM calculated

MRI2F LGM 200 LGM fixed

UGAMPC LGM 200 LGM calculated

UGAMPF LGM 200 LGM fixed

LMD4C LGM 200 LGM calculated

LMD4F LGM 200 LGM fixed

GEN2C LGM 200 LGM calculated

GEN2F LGM 200 LGM fixed

D. Otto et al. / Global and Planetary Change 33 (2002) 117–138124

4. Validation of the pre-industrial simulation

The reference simulation for the present is the

PD280 one (Table 4), i.e., a simulation with present-

day climate and continent size, and a pre-industrial

atmospheric CO2 level of 280 ppmv.

4.1. Vegetation distribution

The reference simulation PD280 has been com-

pared with two maps of present potential natural

vegetation distribution (Fig. 2): (1) the map simulated

by the TEM (Melillo et al., 1993) adapted from the

maps of Matthews (1983), Olson et al. (1983), Isa-

chencko (1990) and Kuchler (1964) (Haxeltine and

Prentice, 1996), which we will call further the

‘‘Melillo adapted’’ map and (2) the map of Matthews

(1983) (Fig. 1). A comparison at the pixel level has

been performed. We assessed the agreement among

the three distributions at the pixel level, by comparing

each pixel to pixels included within a given radius r of

the original pixel. A radius of 0 km is equivalent to

treat grid cells independently. At the global scale, the

appropriate scale for comparison may be larger than a

0.5j cell, due to the relatively coarse effective reso-

lution of the interpolated climatic fields, and to the

limited accuracy of the potential natural vegetation

distribution. Since the mean size of a pixel of

0.5� 0.5j is approximately 50 km, we considered a

radius of 50 or 100 km, which corresponds to a circle

of 1 or 2 pixels of radius.

The CARAIB model correctly predicts the ice and

tundra distributions, since these biomes are assigned

on a GDD5 criteria, which has been adjusted. Deserts

are too extended at the expense of semi-deserts. Two

parameters are principally responsible for this deserti-

fication: wind and precipitation. Indeed, all the areas

where desert appears in the PD280 map and not in the

potential natural vegetation maps are characterised by

a strong wind and weak precipitations: Western North

America, South Argentina, Somalia, Southern Sahara,

Southern Arabia, Western Caspian Sea, Northern

China, and Mongolia. The tropical rainforest biome

is too extended. Seasonal forest has approximately the

same extension as in the natural maps, but is slightly

shifted because it is invaded by tropical rainforest and

it extends on the savanna area. For the North-Western

of India, tropical seasonal forest is replaced by grass-

land and semi-desert. This region is characterised by a

strong wind and weak precipitations. Savanna is

reduced and replaced by seasonal forest. In northeast-

ern Argentina and Uruguay, the model predicts trop-

ical forest instead of grasslands because precipitations

are sufficiently abundant. Haxeltine and Prentice

(1996) put forward the hypothesis that the cause of

this discrepancy may be connected with the occur-

rence of specific soil conditions in this region. For the

temperate region, the model tends to predict a uniform

mixed forest because all trees have approximately the

same productivity. The transition between grasslands

and closed forests is not as gradual as in the potential

natural maps because trees reach too rapidely a high

LAI. Boreal forest areas are correctly predicted,

except in some regions where forest is replaced by

grassland and semi-desert: North-Eastern of Russia

and North-Western of North America. These regions

are characterised by weak precipitations, and hence

small soil water contents.

It appears clearly that the vegetation simulated by

CARAIB is very sensitive to wind speed, which

affects evapotranspiration, and thus the soil water

content, by the way of the aerodynamic resistance.

The expression of this resistance may need to be

refined in the future.

The CARAIB model shows a better agreement

with the Melillo adapted map than with the Matthews

one. However, for the tropical regions, the agreement

is better with the Matthews map. A global comparison

on all biomes was performed (G in Fig. 2). The

overall levels of agreement with the Melillo adapted

and Matthews maps are both 61% for a radius of 100

km. A comparison between the two potential natural

vegetation maps has also been performed and shows a

global agreement of 74%.

4.2. Global carbon stocks values

Previous reconstructions of pre-industrial bio-

spheric carbon stocks (Adams et al., 1990; Prentice

and Fung, 1990; Van Campo et al., 1993; Prentice et

al., 1993; Crowley, 1995; Franc�ois et al., 1998, 1999)give carbon stocks that range from 1115 to 1379 Gt

for the soil carbon, with a mean of 1247 Gt. For the

vegetation carbon, these reconstructions range from

622 to 924 Gt, with a mean of 773 Gt, and for the total

biospheric carbon, the range is from 1891 to 2422 Gt,

D. Otto et al. / Global and Planetary Change 33 (2002) 117–138 125

Fig. 1. Present-day vegetation distribution simulated with the CARAIB model (PD280) and two maps of present potential natural vegetation

distribution: (1) the map simulated by the TEM (Melillo et al., 1993) adapted from the maps of Matthews (1983), Olson et al. (1983),

Isachencko (1990) and Kuchler (1964) (Haxeltine and Prentice, 1996) and (2) the map of Matthews (1983).

D. Otto et al. / Global and Planetary Change 33 (2002) 117–138126

with a mean of 2157 Gt (Table 5). The values

predicted by the PD280 simulation are slightly lower:

1150 Gt for the soil carbon, 761 Gt for the vegetation

carbon, and 1911 Gt for the total biospheric carbon.

These values are directly calculated from the model,

i.e., by multiplying the estimated carbon stocks (the

sum of all vegetation, litter, and soil carbon pools) for

each PFT by their cover fraction in each grid cell. This

estimation is thus made at the PFT level, and hence, is

totally independent of the biome assignment scheme.

Note that the IPCC values of Table 5 correspond to a

present-day climate, with an atmospheric CO2 level of

358 ppmv and the vegetation distribution strongly

disturbed by human activities. The difference between

the PD280 stocks and these values is probably mainly

due to the rise in the atmospheric CO2 level since pre-

industrial times.

This pre-industrial validation gives us confidence

in our biogeography model, since the precision of the

predicted vegetation distribution is comparable to that

of existing present potential natural vegetation maps.

Moreover, the predicted carbon stocks fall within the

uncertainty limits of previous reconstructions.

5. LGM results

5.1. Climatic data

The global continental mean of the climatic data

changes from present to LGM for the eight LGM

simulations are presented in Fig. 3. The general trends

which emerge from these data are that LGM climatic

conditions were colder, less rainy, and more windy.

Fig. 2. Comparison between present vegetation distribution simulated with the CARAIB model (PD280) and two maps of present potential

natural vegetation distribution: Melillo adapted and Matthews. Each pixel of the CARAIB map is compared with the corresponding pixel of the

natural vegetation map, center of a circular area of radius r. The two natural maps are also compared. The comparison is made for each biome;

‘‘global’’ refers to the comparison for all biomes together, i.e., for all pixels of the continents.

D. Otto et al. / Global and Planetary Change 33 (2002) 117–138 127

The cloudiness data do not allow us to conclude on

any trend. These changes are parallel to the develop-

ment of extensive ice caps at high to mid-latitudes of

the Northern Hemisphere in response to a smaller

contrast between summer and winter insolations in

these regions. At lower latitudes, changes in precip-

itation and winds fields are linked to the strengthening

of the Hadley circulation and the reduced seasonal

shifts of the Inter-tropical Convergence Zone (Nich-

olson and Flohn, 1980).

The difference between fixed and calculated SST

vary from one GCM to another. Therefore, the effect

of SSTs on continental climatic data computed by

GCMs remains uncertain. The MRI2 and UGAMP

models present a colder continental temperature in the

calculated version with a smaller difference for

UGAMP. The LMD4 model shows the inverse trend

with a small difference and the GEN2 model presents

no difference. Following the UGAMP, GEN2, and

LMD4 models, SSTs have little to no influence on the

continental temperatures and have thus not affected

the variations of the biospheric carbon stocks since the

LGM. By contrast, MRI2 predicted a marked decline

in continental temperatures in response to lower SSTs

as obtained in the calculated scheme. These reduced

continental temperatures could have a substantial

effect on the estimation of continental carbon storage

at the LGM. Although it is not proven at this point

that MRI2 is the GCM that most accurately predicts

continental temperature response to SSTs, we decided

to further investigate the effects on LGM carbon

stocks of SST-induced changes in continental temper-

atures, as predicted by the MRI2 model.

5.2. Comparison with pollen data

The LGM simulation has been compared with 245

pollen data from Crowley (1995) and Elenga et al.

(2000). The comparison is made at the data point

level. The interpolated map of Crowley, the validity of

which is highly speculative, has not been used. The

comparison technique is the same as for the present

(Fig. 2), i.e., for each of the 245 pollen data, we verify

whether the biome type inferred from the pollen

record is present or not in the model simulation within

a circle of 0 (same 0.5� 0.5j pixel), 50, and 100 km.

The extension of the comparison to a circle of 100 km

around the pollen data is particularly justified here,

since (1) pollens can be transported relatively far from

their production site and (2) GCMs actually provide

climatic fields at much coarser spatial resolutions than

the 0.5� 0.5j one to which all inputs of the biospheremodel have been interpolated.

The overall agreement obtained with the 245

pollen data points (i.e., without distinction of biome

type) is presented for all models in Table 6. The pixel-

by-pixel agreement (0 km) varies from 17% to 28%

depending on the GCM simulation considered. This

agreement is strongly improved if the comparison is

extended to a circle of 100 km radius, since the scores

exhibited by the models then reach 36–53% of

agreement with the data. For the pixel-by-pixel com-

parison, the best agreement is obtained for the MRI2C

simulation, while in the 100 km case, LMD4F obtains

the highest score. For the MRI2 GCM, the calculated

SST version exhibits a better agreement with the

pollen data than the fixed SST one, while it is the

reverse for the other three GCMs. The agreement is

on average inferior by 15–20% with respect to the

comparison between PD280 and the Melillo adapted

map.

The percentages of agreement per biome for the

MRI2 model are presented in Fig. 4. The fixed and

calculated SST versions exhibit a lack of agreement

for the savanna biome. From a general point of view,

the agreement is good for desertic and semi-desertic

regions, the tundra and the broadleaved mixed forest

Table 5

Global values of continental biospheric carbon stocks (soil and

vegetation) for the pre-industrial simulation PD280

C soil

(Gt C)

C veg

(Gt C)

C tot

(Gt C)

PD280 1150 761 1911

Franc�ois et al. (1999) 1334–1379 646–650 1984–2025

Franc�ois et al. (1998) 1322 622 1943

Crowley (1995) 2167–2422

Prentice et al. (1993) 2122

Van Campo et al. (1993) 2116

Prentice and Fung (1990) 1143–1313 748–834 1891–2147

Adams et al. (1990) 1115 924 2039

IPCC values

(Climate Change, 1995)

1580 610 2190

Some references of pre-industrial estimations are given for com-

parison. The IPCC values are present-day values (atmospheric CO2

level of 358 ppmv and vegetation distribution disturbed by human

activities).

D. Otto et al. / Global and Planetary Change 33 (2002) 117–138128

biomes, and poor for the conifers and the grasslands.

The mismatch concerning conifers could be due to the

fact that the pollen of the conifers are among those

who spread far from the originating trees and that they

are produced with high amounts which could bias the

estimates for the conifer biome. The agreement for the

tropical seasonal forest, which is based on only two

data points, is not very relevant.

The fact that the scores reached by the LGM

simulations are relatively high (especially for the

100 km radius comparison) and only slightly lower

than those obtained for the present-day biome com-

parison is quite encouraging. However, refinements

are still needed in the comparison methodology: (1)

more pollen data are needed to get a better coverage of

the LGM continents; (2) the biome classification and

the biomisation scheme of the model should be

identical to those used for the pollen data; and (3)

possible transport of pollens should be taken into

account.

Fig. 3. Change from present to LGM of the continental mean of climatic variables for the eight sets of GCM results used as inputs to the

CARAIB biospheric model. Note that wind speed was not available for the LMD4C simulation and was taken from LMD4F simulation.

Table 6

Overall agreement of the eight LGM vegetation distributions with

245 pollen data from Crowley (1995) and Elenga et al. (2000)

Simulations Overall agreement for all biomes (%)

r = 0 km r = 50 km r= 100 km

MRI2F 21.6 26.5 42.8

MRI2C 28.2 37.1 51.0

UGAMPF 17.2 23.7 40.8

UGAMPC 16.8 20.8 36.3

LMD4F 25.3 35.1 53.5

LMD4C 22.9 31.8 47.3

GEN2F 25.3 33.1 51.4

GEN2C 22.9 32.2 48.2

The comparison is made all biomes together.

D. Otto et al. / Global and Planetary Change 33 (2002) 117–138 129

5.3. Simulation of the biome distribution and

vegetation area changes from the present

The vegetation distributions reconstructed from the

MRI2C and MRI2F simulations are displayed in Fig.

5. Globally, the fixed SST simulation produces a more

desertic landscape than the calculated SST one,

despite the higher precipitation rate in the former.

The reason is that the fixed SST version is also

globally warmer by f 2 jC, and hence, evapotrans-

piration can be expected to be higher. This effect is

reinforced by the substantially stronger winds of the

fixed SST version. Stronger winds indeed tend to

decrease the aerodynamic resistance, and hence, stim-

ulate evapotranspiration. Thus, wind speed appears to

be a critical factor for vegetation distribution. This

factor has been disregarded in all earlier reconstruc-

tions of LGM vegetation distribution. Locally, Central

America, Northern South America, Somalia, Pakistan,

and Northwestern India are more desertic in MRI2C

than in MRI2F, again at least partly in response to

stronger winds in these regions. For the tropical rain-

forest of South America, two opposite effects are

observed: for MRI2F, tropical rainforest area de-

creases with respect to the present, while for MRI2C,

it increases, which is contrary to all previous LGM

vegetation distribution reconstructions (Van Campo et

al., 1993; Crowley, 1995; Franc�ois et al., 1999). Thisis presumably associated with the colder temperatures

prevailing over South America in the MRI2C simu-

lation, since colder temperatures tend to reduce evap-

otranspiration. It would be interesting to have a more

complete set of pollen data for this region.

The global trends of vegetation changes in MRI2

simulations are summarized in Fig. 6 and compared to

those of other GCMs. This figure has been constructed

on the basis of CARAIB results at the PFT level, and

hence, it does not rest on the biome assignment

scheme. All models predict essentially the same gen-

eral trend: an increase of the proportion of ice sheets

Fig. 4. Comparison between the MRI2C and MRI2F simulations and the pollen data from Crowley (1995) and Elenga et al. (2000). Each of the

245 pollen data is compared with the corresponding pixel of the CARAIB map, center of a circle of radius r. The biome classification is from

Crowley (1995). The numbers between brackets represent the numbers of pollen data for each biome.

D. Otto et al. / Global and Planetary Change 33 (2002) 117–138130

and deserts at the LGM compared to the present,

mostly at the expense of trees. Grasses show only

small to moderate reductions. These general trends are

largely consistent with those obtained from previous

reconstructions based on palynological and sedimen-

tological data (Adams et al., 1990; Van Campo et al.,

1993; Crowley, 1995; Adams and Faure, 1998), except

that these studies tend to predict an increase of the

proportion of grass ecosystems. In general, our recon-

structions appear more forested in the tropics and

slightly more desertic in the extra-tropics. These pro-

portions of ice sheets, deserts, grasses and trees are

changed by less than 2–3% in the fixed SST version of

a given GCM compared to the calculated SST results.

As seen before for the MRI2 model, the fixed SST

version generally tends to be more desertic, except in

the case of the UGAMP model.

PFT and biome areas reconstructed with the MRI2

model for present-day and LGM times are presented

in Fig. 7. As mentioned, the area of grasses was

reduced at the LGM compared to the present in these

simulations. C3 grasses are slightly more affected than

Fig. 5. LGM vegetation distribution simulated by the CARAIB model with the two versions of the MRI2 GCM: calculated (MRI2C) and fixed

(MRI2F) SSTs.

D. Otto et al. / Global and Planetary Change 33 (2002) 117–138 131

C4, since the CO2 reduction from 280 to 200 ppmv

tends to favor C4 species. Tree areas also decrease,

except that of broad-leaved evergreen tropical trees

which increases. Needle-leaved trees are more

affected in MRI2F and broad-leaved trees are more

affected in MRI2C. At the biome level, the areas of

ice sheets, deserts, semi-deserts and tundras are

strongly increased at the LGM compared to the

present. Tropical rainforests and temperate broad-

leaved evergreen forests show a moderate increase.

Temperate conifer forests and savannas exhibit essen-

tially no change. A moderate decrease is observed for

boreal deciduous and temperate broad-leaved decid-

uous forests. Boreal evergreen forests, temperate

mixed forests, tropical seasonal forests, and grass-

lands show strong decreases.

5.4. Carbon stocks and fluxes

Global values of the model biospheric carbon

stocks and the corresponding changes from present

are presented in Table 7. These simulated changes are

larger than those obtained from previous vegetation

reconstructions by the way of biospheric models,

which range from 0 to � 700 Gt C (Prentice and

Fung, 1990; Friedlingstein et al., 1992, 1995; Prentice

et al., 1993; Esser and Lautenschlager, 1994; Franc�oiset al., 1998, 1999), and than those suggested by

foraminifera isotopic data, which range from � 300

to � 1000 Gt C (Shackleton, 1977; Duplessy et al.,

1988; Curry et al., 1988; Bird et al., 1994). They are

in better agreement with interpretations from palyno-

logical and sedimentological proxy data, which range

Fig. 6. Proportion of the total continental area of ice sheet, desert, grasses, and trees for the eight LGM simulations, changes of these proportions

from the present to the LGM, and differences of these proportions between fixed and calculated SSTs for each of the four GCMs. Ice sheets are

regions with GDD5 < 50 jC day, deserts are pixels devoid of vegetation, while grass and tree proportions are obtained as the global cover

fraction of the corresponding PFTs predicted by CARAIB.

D. Otto et al. / Global and Planetary Change 33 (2002) 117–138132

Fig. 7. PFT and biome areas of 5 simulations: MRI2C, MRI2F, PD200� , PD200+ and PD280.

D. Otto et al. / Global and Planetary Change 33 (2002) 117–138 133

from � 700 to � 1600 Gt C (Adams et al., 1990; Van

Campo et al., 1993; Crowley, 1995; Adams and Faure,

1998). For all GCMs, the total biospheric carbon

stock is larger in the calculated SST version than in

the fixed one. MRI2 shows the largest difference with

209 Gt C, while UGAMP exhibits almost no differ-

ence with 10.7 Gt C. The reason why carbon storage

is larger in the calculated SST experiments is possibly

linked to the systematically smaller wind speed which

reduces evapotranspiration and hence leads to higher

soil water contents and vegetation productivity.

Nevertheless, these changes in carbon storage remain

very difficult to interpret, in view of the complex

dependence of vegetation net primary productivity

and microbial respiration on surface air temperature

and soil moisture.

5.5. Sensitivity experiments

The individual effects of atmospheric CO2 level

increase, sea level drop, and climate variation on the

total biospheric carbon stock since the LGM have

been analysed by comparing the PD280, PD200+ and

PD200� simulations (defined in Table 4) with

MRI2C. The atmospheric CO2 level effect has been

separated in two parts, i.e., the vegetation shift due to

the increase of this level and the pure fertilization

effect without vegetation migration, by comparing the

PD280 simulation with the PD200+ interm, in which

carbon stocks are calculated with the PFT distribution

of PD280 and the carbon distribution of PD200+

(Table 8). The CO2 fertilization effect is clearly

dominant and almost solely responsible for the total

biospheric carbon stock increase since the LGM. The

sea level drop, which decreases the continental area,

tends to decrease the carbon stocks, while the combi-

nation of the modification of the climatic variables,

i.e., the increase of temperature, precipitation and the

decrease of wind speed (Fig. 3), tends to increase the

carbon stocks and fluxes. The vegetation shift asso-

ciated with the atmospheric CO2 level increase has

rather weak impact on the carbon stocks.

The effect of the CO2 level on vegetation is

obtained by comparing the PD280 and PD200+ sim-

ulations (Fig. 7). The lower CO2 level appears as a

major contributor to the desertification during the

LGM. Our simulations indicate that this lower CO2

level is sufficient to explain the transformation of

present-day arid regions into desert during the LGM,

i.e., western North America, southern South America,

Australia, South Africa, and Somalia. It must be

pointed out that these regions received approximately

the same precipitation rate at the LGM than at present

according to the MRI2C simulations. On the other

hand, desertification of Europe and most of Asia

cannot be explained only by the CO2 level decrease,

but is associated with a substantial reduction of

precipitations. The C3 grass area decreases by

2.1*106 km2 while that of C4 grass increases by

2.2*106 km2. The ratio of C3/C4 grass areas is 0.66

for PD200+ and 0.80 for PD280, corresponding to an

increase of 0.14 (Fig. 7). As expected, C4 grasses are

thus dominant with respect to C3 in low atmospheric

CO2 level conditions. Globally, the productivity of all

Table 8

Effect of climate change, sea level increase, and atmospheric CO2

level increase (fertilization + PFT shift effects) on the increase of the

total biospheric carbon stock since the LGM to present-day

Simulations C total

(Gt)

Effect Change from

LGM (MRI2C)

(Gt)

MRI2C 1083.1 0

PD200� 1505.4 climate:

LGM! present

+ 422.3

PD200 + 1161.4 sea level rise � 344.0

PD200 + interm 1071.4 PFT shift

due to CO2

level increase

� 90.0

PD280 1910.9 CO2 level

increase without

PFT shift

+ 839.5

The total carbon stock of the PD200+ interm simulation is cal-

culated with the distribution of vegetation of PD280 and the carbon

distribution of PD200+.

Table 7

LGM global values of continental biospheric carbon stocks (soil and

vegetation), with the change from present in parentheses

Simulation C soil (Gt C) C veg (Gt C) C tot (Gt C)

MRI2C 605.9 (� 544.1) 477.2 (� 283.7) 1083.1 (� 827.8)

MRI2F 461.8 (� 688.2) 412.4 (� 348.5) 874.1 (� 1036.8)

UGAMPC 416.8 (� 733.2) 398.7 (� 362.2) 815.5 (� 1095.4)

UGAMPF 422.9 (� 727.1) 382.0 (� 378.9) 804.8 (� 1106.1)

LMD4C 532.9 (� 617.1) 450.8 (� 310.1) 983.7 (� 927.2)

LMD4F 502.8 (� 647.2) 419.0 (� 341.9) 921.8 (� 989.1)

GEN2C 575.1 (� 574.9) 493.8 (� 267.1) 1068.9 (� 842.0)

GEN2F 495.8 (� 654.2) 426.2 (� 334.7) 922.0 (� 988.9)

D. Otto et al. / Global and Planetary Change 33 (2002) 117–138134

PFTs decreases in response to the CO2 drop. Biomes

like grasslands or savanna become desertic and closed

forests become woodlands. Tropical rainforest is

maintained because it contains PFTs of a sufficiently

high productivity to remain dominant, even if their

productivity decreases.

The effect of the continental area change associated

with the lower sea level at the LGM can be observed

by comparing the PD200+ and PD200� simulations.

This sensitivity experiment corresponds to a sea level

drop under current climatic conditions. The biomes

which tend to colonize the emerged shelf in this

experiment are ice sheets, tundra, tropical rainforest,

temperate mixed forest, and boreal evergreen forest/

woodland. In the case of the tropical rainforest, the

corresponding area increase is larger than the decrease

induced by the climate change. Consequently, the

expansion of the tropical rainforest biome at the

LGM in these simulations is associated with the

colonization of the emerged shelf (mostly in Indonesia

as obvious from Fig. 5), while on the present-day

continents, the area of the tropical rainforest actually

decreases which is more consistent with the trend

indicated by the data.

In view of the apparently important role of wind

speed in the LGM simulations, an additional sensi-

tivity test has been performed with the MRI2 model,

in which all inputs were set to the standard input fields

of the MRI2C simulation, except wind speed which

was set to the present-day value (eventually extrapo-

lated to the emerged shelf). In view of the globally

smaller wind speed (see Fig. 3), this new simulation

resulted in a reduction of the desert area by 0.7*106

km2 compared to MRI2C, at the benefit of grass and

tree areas which respectively increased by 0.4*106

km2 and 0.3*106 km2. These changes in the areas of

the PFTs remain relatively limited. However, the

global biospheric carbon stock increased by 50.5 Gt

C with respect to MRI2C, which is not negligible in

the global carbon budget. Wind speed thus occurs as a

significant factor which should taken into account in

future reconstructions of the LGM carbon stocks.

6. Conclusion

CARAIB simulates the vegetation distribution in

equilibrium with a given climate and atmospheric

CO2 concentration. This distribution is determined

by simulated NPP, resulting from a direct response

of photosynthesis and stomatal conductance to climate

and CO2. The model successfully reproduces the

broad-scale patterns in potential natural vegetation

for the current climate. It also produces estimates of

the biospheric carbon stocks, which fall in the range

of previous estimates for pre-industrial times. The

competition for light between grass and woody plant

types was modeled by introducing two vegetation

storeys, i.e., ground-level grasses and tree canopies.

This competition for light tends to favor trees at the

expense of grasses. An important application of CAR-

AIB is the simulation of the equilibrium response of

vegetation to changes in climate and atmospheric CO2

concentration.

A validation of simulated vegetation responses to

climate changes is only possible with past data.

CARAIB was forced with LGM climatic data from

four GCMs using fixed and calculated SST fields, i.e.,

eight scenarios. The simulated carbon stocks are in

reasonable agreement with reconstructions based on

palynological and sedimentological proxy data. These

values are larger than those obtained from previous

vegetation reconstructions by the way of biospheric

models, and from foraminifera data. The main char-

acteristic common to all vegetation reconstructions

performed with CARAIB for the LGM is the global

expansion of deserts and ice and a global reduction of

forests consistent with previous reconstructions (Van

Campo et al., 1993; Crowley, 1995; Franc�ois et al.,

1999). However, the model predicts a decrease of

grasslands and a slight expansion of the tropical

rainforest, in conflict with these previous reconstruc-

tions. Note, nevertheless, that the expansion of the

tropical rainforest is linked to the colonization of the

emerged shelf, while on the present-day continents the

tropical rainforest tends to be reduced, except region-

ally such as in South America for the MRI2C simu-

lation.

The four GCMs are not consistent with one another

concerning the effect of SSTs on the continental

climate, and therefore on the vegetation distribution

in equilibrium with this climate. However, it appears

that the use of calculated SSTs in the GCMs system-

atically leads to higher biospheric carbon stock pre-

dicted by the model. A more detailed study was

performed with the MRI2 model. The choice of this

D. Otto et al. / Global and Planetary Change 33 (2002) 117–138 135

model was motivated by the fact that the predicted

continental temperature is substantially colder than the

one calculated by its fixed SST version, which is

consistent with the recent studies indicating that the

tropics were colder than suggested by the CLIMAP

database. Moreover, the calculated SST version of this

model was among those which led to the best scores

in the comparison of the predicted biomes with pollen

data. There are two main differences between the

results obtained with the fixed and the calculated

SST versions of this GCM. Firstly, the fixed SST

version predicts a reduction of the tropical rainforest

of South America from the present to the LGM, while

the calculated one predicts its expansion, which is in

total disagreement with previous reconstructions. This

is probably associated with the colder tropical temper-

atures in the calculated SST version, which decreases

the continental surface evaporation and therefore

increases the soil water content in this region. Sec-

ondly, from a more global perspective, the fixed SST

version is more desertic than the calculated one. This

is probably due to increased wind speed. Therefore,

the calculated SST version of MRI2 leads to larger

biospheric carbon stocks than its fixed SST version.

Sensitivity tests suggest that CO2 fertilization

effect is among the most important factors responsible

for the changes in carbon stocks and vegetation

patterns between LGM and present times. C4 grasses

are less affected by the decrease in the atmospheric

CO2 concentration during glaciation than C3.

A series of model refinements may improve the

accuracy of the vegetation reconstructions. Firstly, the

vegetation distribution seems to be very sensitive to

wind speed data. All the vegetation distributions

reconstructed with CARAIB, for current or LGM

climate, are too desertic and present weaker carbon

stocks. The expression of the aerodynamic resistance,

i.e., the only way by which wind speed influences

vegetation, should be improved. Secondly, relative air

humidity data used for the LGM simulations are

present-day data. This approximation may alter the

simulated NPP, and therefore vegetation response.

Thirdly, a more accurate determination of the dynamic

equilibrium between grass and tree types would

require the implementation into the model of natural

disturbances, such as fires, which is known as an

important factor in determining this equilibrium (Hop-

kins, 1992) and which tends to favor grasses with

respect to trees. This would lead to a better represen-

tation of the transition between grasslands and closed

forests, which is too abrupt in the current vegetation

reconstructions with the CARAIB model.

Acknowledgements

The authors wish to thank S. Jousseaume and Gilles

Ramstein who made the PMIP database available for

this study. Funding for this research was provided by

the Communaute Franc�aise de Belgique-Direction de laRecherche Scientifique-Actions de Recherches Con-

certees (Contract No. ARC 98/03-219) and by ‘‘The

global carbon cycle and future atmospheric CO2

levels’’ (First Multiannual Scientific Support Plan for

a Sustainable Development Policy (SPSD I) of the

OSTC, Contract CG/DD/11A, 12/1996–11/2000). L.

Franc�ois is supported by the Belgian National

Foundation for Scientific Research (FNRS).

References

Adams, J., Faure, H., 1998. A new estimate of changing carbon

storage on land since the last glacial maximum, based on global

land ecosystem reconstruction. Global Planet. Change 16, 3–24.

Adams, J.M., Post, W.M., 1999. A preliminary estimate of changing

calcrete carbon storage on land since the last glacial maximum.

Global Planet. Change 20 (4), 243–256.

Adams, J.M., Faure, H., Faure-Denard, L., McGlade, J.M., Wood-

ward, F.I., 1990. Increases in the terrestrial carbon storage from

the last glacial maximum to the present. Nature 348, 711–714.

Bird, M.I., Lloyd, J., Farquhar, G.D., 1994. Terrestrial carbon stor-

age at the LGM. Nature 371, 566.

Broccoli, A.J., 2000. Tropical cooling at the last glacial maximum:

an atmosphere-mixed layer ocean simulation. J. Clim. In press.

Broecker, W.S., Peng, T.-H., 1987. The role of CaCO3 compensa-

tion in the glacial to interglacial atmosphere CO2 change. Global

Biogeochem. Cycles 1 (1), 15–29.

Broecker, W.S., Takahashi, T., 1984. Is there a tie between atmos-

pheric CO2 content and ocean circulation? In: Hansen, J.E.,

Takahashi, T. (Eds.), Clim. Process. Clim. Sens. AGU, Wash-

ington, DC, pp. 314–326.

CLIMAP Project Members, 1981. Seasonal reconstruction of the

earth’s surface at the last glacial maximum. Geol. Soc. America

Map Chart Ser. MC-36.

Climate Change, 1995. The science of climate change. Contribution

of Working Group I to the Second Assessment Report of the

Intergovernmental Panel on Climate Change.

Collatz, G., Ribas-Carbo, M., Berry, J., 1992. Coupled photosyn-

thesis-stomatal conductance model for leaves of c4 plants. Aust.

J. Plant Physiol. 19, 519–538.

D. Otto et al. / Global and Planetary Change 33 (2002) 117–138136

Cramer, W., Kicklighter, D., Bondeau, A., Moore III, B., Churkina,

G., Nemry, B., Ruimy, A., Schloss, A., The participants of the

Potsdam NPP Model Intercomparison, 1999. Comparing global

models of terrestrial net primary productivity (npp): overview

and key results. Global Change Biol. 5 (Suppl. 1), 1–15.

Crowley, T.J., 1995. Ice age terrestrial carbon changes revisited.

Global Biogeochem. Cycles 9 (3), 377–389.

Curry, W., Oppo, D., 1997. Synchronous, high-frequency oscilla-

tions in tropical sea surface temperatures and north atlantic deep

water production during the last glacial cycle. Paleoceanography

12, 1–14.

Curry, W.B., Duplessy, J.-C., Labeyrie, L.D., Shackleton, N.J., 1988.

Changes in the distribution of d13C of deep-water ACO2 between

the last glaciation and the Holocene. Paleoceanography 3 (3),

317–341.

Dong, B., Valdes, P., 1998. Simulations of the last glacial maximum

climates using a general circulation model: prescribed versus

computed sea surface temperatures. Clim. Dyn. 14, 571–591.

Dubroca, E., 1983. Evolution saisonniere des reserves dans un taillis

de chataigniers Castanea sativa Mill., avant et apres coupe. PhD

thesis, Universite Paris XI.

Duplessy, J.-C., Shackleton, N.J., Fairbanks, R.G., Labeyrie, L.,

Oppo, D., Kallel, N., 1988. Deepwater source variations during

the last climatic cycle and their impact on the global deepwater

circulation. Paleoceanography 3 (3), 343–360.

Duvigneaud, P., Kestemont, P., Ambroes, P., 1971. Productivite

primaire des forets temperees d’essences feuillues caducifoliees

en Europe occidentale. Proc. of the Conference: ‘‘Productivite

des ecosystemes forestiers’’. Bruxelles, Oct. 27–31 1969, 259–

270.

Eglinton, G., Bradshaw, S.A., Rosell, A., Sarnthein, M., Pflaumann,

U., Tiedemann, R., 1992. Molecular record of secular sea sur-

face temperature changes on 100-year timescales for glacial

terminations I, II and IV. Nature 356, 423–426.

Elenga, H., Peyron, O., Bonnefille, R., Jolly, D., Cheddadi, R.,

Guiot, J., Andrieu, V., Bottema, S., Buchet, G., de Beaulieu,

J.-L., Hamilton, A., Maley, J., Marchant, R., Perez-Obiol, R.,

Reille, M., Riollet, G., Scott, L., Straka, H., Taylor, D., Campo,

E.V., Vincens, A., Laarif, F., Jonson, H., 2000. Pollen-based

biome reconstruction for southern europe and africa 18,000 yr

bp. J. Biogeogr. 27, 621–634.

Esser, G., 1984. The significance of biospheric carbon pools and

fluxes for the atmospheric CO2: a proposed model structure.

Por. Biometeorol. 3, 253–294.

Esser, G., Lautenschlager, M., 1994. Estimating the change of car-

bon in the terrestrial biosphere from 18,000 bp to present using a

carbon cycle model. Environ. Pollut. 83, 45–53.

Farquhar, G., von Caemmerer, S., Berry, J., 1980. A biochemical

model of photosynthetic CO2 assimilation in leaves of C3 spe-

cies. Planta 149, 78–90.

Foley, J., Prentice, I., Ramankutty, N., Levis, S., Pollard, D., Sitch,

S., Haxeltine, A., 1996. An integrated biosphere model of land

surface processes, terrestrial carbon balance, and vegetation dy-

namics. Global Biogeochem. Cycles 10 (4), 603–628.

Franc�ois, L.M., Delire, C., Warnant, P., Munhoven, G., 1998. Mod-

elling the glacial – interglacial changes in the continental bio-

sphere. Global Planet. Change 16–17 (1–4), 37–52.

Franc�ois, L.M., Godderis, Y., Warnant, P., Ramstein, G., de Noblet,

N., Lorenz, S., 1999. Carbon stocks and isotopic budgets of the

terrestrial biosphere at mid-Holocene and last glacial maximum

times. Chem. Geol. 159 (1–4), 163–189.

Friedlingstein, P., Delire, C., Muller, J.-F., Gerard, J.-C., 1992. The

climate-induced variation of the continental biosphere: a model

simulation of the last glacial maximum. Geophys. Res. Lett. 19

(9), 897–900.

Friedlingstein, P., Prentice, K.C., Fung, I., John, J., Brasseur, G.,

1995. Carbon–biosphere–climate interactions in the last glacial

maximum climate. J. Geophys. Res. 100 (D4), 7203–7221.

Goudriaan, J., van Laar, H.H., 1994. Modelling potential crop

growth processes, textbook with exercices. Current Issues in

Production Ecology, vol. 2. Kluwer Academic Publishing, Dor-

drecht, The Netherlands, 256 pp.

Guilderson, T.P., Fairbanks, R.G., Rubenstone, J.L., 1994. Tropical

temperature variations since 20,000 years ago: modulating in-

terhemispheric climate change. Science 263, 663–665.

Haxeltine, A., Prentice, I., 1996. Biome3: an equilibrium terrestrial

biosphere model based on ecophysiological constraints, re-

source availability, and competition among plant functional

types. Global Biogeochem. Cycles 10 (4), 693–709.

Herbert, T., Schuffert, J., 1998. Alkenone unsaturation estimates of

late miocene through late pliocene sea-surface temperatures at

site 958. Proceedings of the Ocean Drilling Program, Scientific

Results 159T.

Hopkins, B., 1992. Ecological processes at the forest – savanna

boundary. In: Furley, P.A., Proctor, J., Ratter, J.A. (Eds.), Nature

and Dynamics of Forest –Savanna Boundaries. Chapman &

Hall, New York, pp. 21–33.

Hubert, B., Franc�ois, L., Warnant, P., Strivay, D., 1998. Stochastic

generation of meteorological variables and effects on global

models of water and carbon cycles in vegetation and soils. J.

Hydrol. 212–213 (1–4), 318–334.

Isachencko, T., 1990. Map of vegetation in the USSR. Inst. Geog. of

the Siberian Dep. of U.S.S.R. Acad. of Sci., Inst. Bot. of

U.S.S.R. Acad. of Sci. and Moscow State Univ. Geog. Dep.,

Minsk, Russia.

Jarvis, P., 1976. The interpretation of the variations in leaf water

potential and stomatal conductance found in canopies in the field.

Philos. Trans. R. Soc. London, Ser. B: Biol. Sci. 273, 593–610.

Johnson, H., Bergstrom, L., Jansson, P., 1987. Simulated nitrogen

dynamics and losses in a layered agricultural soil. Agric., Eco-

syst. Environ. 18, 333–356.

Joussaume, S., Taylor, K., 1995. Status of the paleoclimate modeling

intercomparison project (PMIP). In: Gates, W.L. (Ed.), Proceed-

ings of the First International Atmospheric Model Intercompar-

ison Project (AMIP) Scientific Conference. No. 732 inWMOTD,

JPS (WCRP/WMO), Geneva, 425–430 (Available online at

URL http://www-pcmdi.llnl/pmip/overview.html.

Kuchler, A., 1964. Potential Natural Vegetation of the Contermi-

nous United States Am. Geogr. Soc., New York.

Leemans, R., Cramer, W., 1991. The IIASA database for mean

monthly values of temperature, precipitation and cloudiness

on a global terrestrial grid. IIASA-Report RR-91-18. Interna-

tional Institute for Applied Systems Analysis, Laxenburg, Aus-

tria.

D. Otto et al. / Global and Planetary Change 33 (2002) 117–138 137

Leuning, R., 1995. A critical appraisal of a combined stomatal-

photosynthesis model for c3 plants. Plant Cell Environ. 18,

339–356.

Ludeke, M., Badeck, F.-W., Otto, R., Hager, C., Donges, S., Kinder-

mann, J., Wurth, G., Lang, T., Jakel, U., Klaudius, A., Ramge,

P., Habermehl, S., Kohlmaier, G.H., 1994. The frankfurt bio-

sphere model. A global oriented model for the seasonal and

longterm CO2 exchange between terrestrial ecosystems and

the atmosphere. Clim. Res. 4, 143–166.

Matthews, E., 1983. Global vegetation and land use: new high

resolution data bases for climate studies. J. Clim. Appl. Mete-

orol. 22, 474–487.

Melillo, J., McGuire, A., Kicklighter, D., Moore III, B., Vorosmarty,

C., Schloss, A., 1993. Global climate change and terrestrial net

primary production. Nature 363, 234–240.

Mintz, Y., Walker, G., 1993. Global fields of soil moisture and

surface evapotranspiration derived from observed precipitation

and surface air temperature. J. Appl. Meteorol. 32, 185–188.

Nemry, B., Franc�ois, L.M., Warnant, P., Robinet, F., Gerard, J.-C.,

1996. The seasonality of the CO2 exchange between the atmos-

phere and the land biosphere : a study with a global mechanistic

vegetation model. J. Geophys. Res. 101 (D3), 7111–7125.

Nicholson, S., Flohn, H., 1980. African environmental and climatic

changes and the general circulation in late pleistocene and hol-

ocene. Clim. Change 2, 313–348.

Olson, J., Watts, J., Allison, L., 1983. Carbon in Live Vegetation of

Major World Ecosystems ORNL-5862, Oak Ridge Nat. Lab.,

Oak Ridge, TN.

Parton, W.J., Scurlock, J., Ojima, D., 1993. Observations and mod-

elling of biomass and soil organic matter dynamics for the grass-

land biome worldwide. Global Biogeochem. Cycles 7, 785–809.

Petit, J.-R., Jouzel, J., Raynaud, D., Barkov, N.I., Barnola, J.-M.,

Basile, I., Bender, M., Chappellaz, J., Davis, M., Delaygue, G.,

Delmotte, M., Kotlyakov, V.M., Legrand, M., Lipenkov, V.Y.,

Lorius, C., Pepin, L., Ritz, C., Saltzman, E., Stievenard, M.,

1999. Climate and atmospheric history of the past 420,000 years

from the Vostok ice core, Antarctica. Nature 399, 429–436.

Prentice, K.C., Fung, I.Y., 1990. The sensitivity of terrestrial carbon

storage to climate change. Nature 346, 48–51.

Prentice, I.C., Sykes, M.T., Lautenschlager, M., Harrison, S.P., De-

nissenko, O., Bartlein, P., 1993. Modelling global vegetation

patterns and terretsrial carbon storage at the last glacial maxi-

mum. Global Ecol. Biogeogr. Lett. 3, 67–76.

Raich, J.W., Schlesinger, W.H., 1992. The global carbon dioxide

flux in soil respiration and its relation to vegetation and climate.

Tellus 44B, 81–99.

Raich, J., Rastetter, E., Melillo, J., Kicklighter, D., Steudler, P.,

Peterson, B., Grace, A., Moore III, B., Vorosmarty, C., 1991.

Potential net primary productivity in South America: application

of a global model. Ecol. Appl. 1, 299–429.

Rind, D., Peteet, D., 1985. Terrestrial conditions at the last glacial

maximum and climap sea surface temperature estimates: are

they consistent? Quat. Res. 24, 1–22.

Ruimy, A., Dedieu, G., Saugier, B., 1996. Turc: a diagnostic model

of continental gross productivity and net primary productivity.

Global Biogeochem. Cycles 10, 269–286.

Ryan, M., Waring, R., 1992. Maintenance respiration and stand

development in a subalpine lodgepole pine forest. Ecology 73

(6), 2100–2108.

Saxton, K., Rawls, W., Romberger, J., Papendick, R., 1986. Esti-

mating generalized soil –water characteristics from texture. Soil

Sci. Soc. Am. J. 50, 1031–1036.

Shackleton, N.J., 1977. Carbon-13 in Uvigerina: tropical rainforest

history and the equatorial pacific carbonate dissolution cycles.

In: Andersen, N.R., Malahoff, A. (Eds.), The Fate of Fossil Fuel

CO2 in the Oceans. Plenum, New York, pp. 401–427.

Stewart, J., 1988. Modelling surface conductance of pine forest.

Agric. For. Meteorol. 43, 19–35.

Stute, M., Forster, M., Frischkorn, H., Serejo, A., Clark, J.F.,

Schlosser, P., Broecker, W.S., Bonani, G., 1995. Cooling of

tropical Brazil (5 jC) during the last glacial maximum. Science

269, 379–383.

Thompson, L.G., Mosley-Thompson, E., Davis, M.E., Lin, P.-N.,

Henderson, K.A., Cole-Dai, J., Bolzan, J.F., Liu, K.-B., 1995.

Late glacial stage and Holocene tropical ice core records from

Huascaran, Peru. Science 269, 46–50.

Tushingham, H., Peltier, W., 1991. A global model of late pleisto-

cene deglaciation based upon geophysical predictions of post-

glacial relative sea level change. J. Geophys. Res. 96, 4497–

4523.

Van Campo, E., Guiot, J., Peng, C., 1993. A data-based re-appraisal

of the terrestrial carbon budget at the last glacial maximum.

Global Planet. Change 8, 189–201.

Warnant, P., 1999. Modelisation du cycle du carbone dans la bio-

sphere continentale a l’echelle globale. PhD thesis, Universite

de Liege, Liege. (In French).

Warnant, P., Franc�ois, L.M., Strivay, D., Gerard, J.-C., 1994. CAR-

AIB: a global model of terrestrial biological productivity. Global

Biogeochem. Cycles 8 (3), 255–270.

Wijk, M.V., Dekker, S., Bouten, W., Bosveld, F., Kohsiek, W.,

Kramer, K., Mohren, G., 2000. Modeling daily gas exchange

of a douglas-fir forest: comparison of three stomatal conduc-

tance models with and without a soil water stress function. Tree

Physiol. 20, 115–122.

Woodward, F., 1987. Climate and plant distribution. Cambridge

Studies in Ecology. Cambridge Univ. Press, Cambridge,

174 pp.

Zobler, L., 1986. A World Soil File for Global Climate Modelling.

Tech. Rep NASA Goddard Institute for Space Studies (GISS),

New York.

D. Otto et al. / Global and Planetary Change 33 (2002) 117–138138