It is based on the coupling of the three following submodels; a leaf assimilation model including estimates of stomatal conductance and leaf respiration, a canopy model describing prin
Trang 1GLOBAL BIOGEOCHEMICAL CYCLES, VOL 8, NO 3, PAGES 255-270, SEPTEMBER 1994
CARAIB: A global model of terrestrial biological productivity
P Warnant, L Francois, D Strivay, and J.-C G•rard
Laboratoire de Physique Atmosph•rique et Plan•t•ire, Institut d'Astrophysique, Universite de Liege, Liege, Belgium
Abstract CARAIB, a mechanistic model of carbon assimilation in the biosphere
estimates the net primary productivity (NPP) of the continental vegetation on
a grid of 1 ø x 1 ø in latitude and longitude The model considers the annual
and diurnal cycles It is based on the coupling of the three following submodels;
a leaf assimilation model including estimates of stomatal conductance and leaf
respiration, a canopy model describing principally the radiative transfer through
the foliage, and a wood respiration model Present-day climate and vegetation
characteristics allow the discrimination between ecotypes In particular, specific
information on vegetation distribution and properties is successfully used at four
levels; the leaf physiological level, the plant level, the ecosystem level, and the global
level The productivity determined by the C ARAIB model is compared with local
measurements and empirical estimates showing a good agreement with a global
value of 65 Gt C yr -1 The sensitivity of the model to the diurnal cycle and to the
abundance of C4 species is also tested The productivity slightly decreases (10%)
when the diurnal cycle of the temperature is neglected By contrast, neglecting the
diurnal cycle of solar irradiance produces unrealistically high values of NPP Even
if the importance of this increase would presumably be reduced by the coupling
of CARAIB with a nutrient cycle model, this test emphasizes the key role of the
diurnal cycle in a mechanistic model of the NPP Uncertainties on the abundance
and spatial distribution of Ca plants may cause errors in the NPP estimates,
however, as demonstrated by two sensitivity tests, these errors are certainly lower
than 10% at the global scale as shown by two tests
Introduction
Continental vegetation plays an important role in the
climatic system Indeed, the hydrological cycle is mo-
dified by the extraction, transpiration, and storage of
soil water in plants Moreover, photosynthesis of green
plants is an important sink of carbon Atmospheric
CO2 is, after water vapor, the most important green-
house gas, and its assimilation by the biosphere modi-
fies the atmospheric reservoir and therefore the heating
or cooling of Earth's atmosphere In particular, globatL
climate changes induced by human activities may be in-
fiuenced by the coupled atmosphere-biosphere system
In this scope the study of the global carbon uptake by
the biosphere is of primary importance
During the last years, global models of carbon as
sirnilation by the biosphere based on empirical para-
meterizations have been developed [e.g., Esser, 1991]
These models estimate biospheric carbon pool sizes and fluxes However, the parameterizations they use are calibrated with present-day CO2 levels and climatic conditions but may be less appropriate for future con- ditions On the other hand, mechanistic models have often been applied only at the leaf, plant, or, eventually, the canopy level The study described in this paper ten- tatively uses mechanistic models [Farquhar et al., 1980; Collatz et al., 1992] to predict the net primary produc-
tivity (NPP) at a global scale Even if this scaling-up
method rests upon many simplifications, it is a first step toward the modeling of CO2 assimilation by continen- tal vegetation, and it gives results as realistic as those
of previous estimations Furthermore, improvements of this model are expected as physiological knowledge and scaling-up methodology progress The sensitivity of the model to the diurnal cycle is tested It is shown that ignoring this cycle may introduce important errors in NPP estimates
Copyright 1994 by the American Geophysical Union
Paper number 94GB00850
0886-6236/94/94GB-00850510.00
Model Description
The C ARAIB model has been built to estimate the net primary productivity of continental vegetation at
Trang 2WARNANT ET AL.: GLOBAL MODEL OF TERRESTRIAL BIOLOGICAL PRODUCTIVITY
a global scale using vegetation information and clima-
tic data A spatial resolution of 1 ø x 1 ø in latitude
and longitude has been chosen because it permits the
description of relatively fine spatial variations of the
NPP, while staying computationally manageable At
this resolution the continents cover 15,347 grid points
The NPP is calculated independently at each of these
grid points The functioning of the model is outlined
in Figure 1, which the diagram illustrates the various
submodels coupled in C ARAIB with the processes they
consider, the input data they require, and the timescale
(day or season) to which they apply In the lower right
of Figure I is a summary of the symbols used in the
text (sections 2.3 and 3.2) for the area and carbon al-
location fractions of the different vegetation covers of a
grid point CARAIB considers the main two solar cy-
cles, the diurnal and annual cycles Because of the large
variation of the photosynthetic rate during the day, the
CO2 uptake is calculated on an hourly basis The hourly
values are subsequently summed up to provide the daily
assimilation However, since a monthly mean climatic
data set is used, random day-to-day variations of wea-
ther conditions cannot be taken into account The daily
NPP is thus estimated for a midmonth day and multi-
plied by the month length to obtain the monthly value
CARAIB intends to be as mechanistic as possible
and is based on the coupling of the three following
submodels: a leaf assimilation model including esti- mates of stomatal conductance and leaf respiration, a canopy model describing principally the radiative trans-
fer through the foliage, and a wood respiration model
Photosynthesis is the major carbon flux determining
plant growth In this study the emphasis is thus put on
the estimation of gross primary productivity (GPP) By
contrast, respiration of leaves and wood is not known accurately Only a rough estimate of respiration rates will thus be performed here, mainly to provide the value
of net primary productivity and to enable the compari- son of model results with in situ measurements (existing
for NPP but not for GPP)
Leaf Assimilation Submodel
The leaf gross assimilation rate A (/zmol m -2 s -1)
is described by two quadratic equations [Collatz et al., 1991]'
OAp • - Ap(A1 + A•) + A1A• = 0 (1)
/SA • - A(Ap + Aa) + A;,Aa = 0 (2)
LEAF LEVEL
SOIL HYDROLOGY
Air humicity
• water
Bucket
CANOPY LEVEL
Trendat
PLANT LEVEL
Respiration , Construction
Wood Biota are
VEGETATION AREA FRACTION ALLOCATION
GROUND C, f o • ( I - f •,)
,
CARbon Assimilation In the Biosphere ( CARAIB )
Figure 1 A schematic diagram showing the structure of the carbon assimilation in the biosphere
(CARAIB) model The parameters defining the area and carbon allocation fractions (see text)
of the different vegetation covers of a grid point are summarized at lower right
Trang 3WARNANT ET AL.: GLOBAL MODEL OF TERRESTRIAL BIOLOGICAL PRODUCTIVITY 257
where 0 and /• are parameters; A1, A2 and As are
functions describing limitations of the assimilation rate
(•umol m -2 s-i); and Ap is the assimilation rate resul-
ting from the coupling of the first two limitations (/zmol
m -2 s-l)
These two equations indicate that the assimilation
rate is limited by three processes, with a coupling be-
tween them represented by the parameters 0 and
For Ca species, A1 is the ribulos-biphosphate car-
boxylase oxygenase (Rubisco) limited rate, mud A2 is
the electron transport limited or light-limited rate [Far-
quhar et al., 1980] They are given by
Pi - F
+ +
where Vcma• is the maximum catalytic capacity of Ru-
bisco (/zmol m -2 s -1); Pi is the intercellular CO2 pres-
sure (10 -6 Pa); O2 is the intercellular O2 pressure (Pa);
F is the CO2 compensation point in the absence of
dark respiration (10 -6 Pa); Kc is the Michaelis-Menten
constant for CO2 (10 -6 Pa); and Ko is the Michaelis-
Menten constant for O2 (Pa)
pi - F
where J is the potential rate of electron transport (/zEq
m -2 s -1) V•m•, F., Kc, and Ko are functions of tem-
perature, while J is a function of temperature and of
the absorbed irradiance J saturates at a level Jmax at
high irradiance
As is the rate limited by the capacity for the export or
the utilization of the products of photosynthesis [Collatz
et al., 1991]'
Ycmax
&= 2 (s)
Similar equations describe the C4 species behavior
[Collatz et al., 1992] In this case, A1 and A2 are inde-
pendent of p•, and As is a CO2-1imited rate proportional
to Pi
where V•m• is a "Q10 function" (Q10 is the factor by
which the rate, i.e Vcmax, is multiplied for each 10øC
increase in + •empera•ure• of temp + erasure •orrec•ed to
limit the assimilation rate at low or high temperature;
a is the slope of the photosynthetic response to light
(molco•_/mO]photon); I is the irradiance absorbed by the
2 1
leaf (•molphoton m- s- ; k is the initial slope of pho-
tosynthetic C02 response (mo] m -2 s -1) and is a
function of temperature; and P is the atmospheric pres-
sure (Pa)
Finally, the leaf net assimilation rate An (/zmol m -2
s -1) is given by
where Rd, the dark respiration rate, is assumed to be proportional to Vcmax The proportionality constant
used is 0.015 for Ca plants [Sellers et al.,1992] and 0.020 for C4 plants (estimated from Gollatz et al 's [1992] va- lue of the dark respiration rate at 25øC)
The CO2 pressure in interee!!ular spaces is related to
the atmospheric CO• pressure, p•tm (Pa), by a diffusion
equation
Pi Patm
where g, the total conductance to
is given by
I I 1
g gst
where gst is the stomatal conductance; and gbl is the leaf boundary layer conductmace
The stomatal conductance to CO• is estimated fol-
lowing Ball et al., [1987]'
where h8 is the air relative humidity at the leaf surface; and F is the compensation point (10 -6 Pa)
The factor 1.6 accounts for the ratio of the diffusi-
vities of CO2 and H20 vapor in the stomates [Collatz
et al., 1992]; go = 0.01 (mol•_o m -2 s -1) and gl= 9.2
for Cs species [Leuning, 1990]; mud go = 0.08 (moln2o
m -2 s -1) and gl= 3.0 for C4 species [Collatz et al., 1992] For low air relative humidity (hs _• 0.46), gl
is decreased linearly with the available soil water frac- tion (wn•_o - wp)/(fc - wp) = (hs - 0.1)/0.9 (see sec-
tion 3.1) This linear decrease tends to simulate the
observed behavior of the stomatal conductance at low water availability [Mc Mutttie et al., 1992] This for- mulation of the stomatal conductance involves the hy- pothesis that the stomates open and close to optimize the uptake of CO2, while limiting the H20 losses In the present version of the model a constant value g• =
0.0714 molco•_ m -2 s -1 is used as a first approximation
for the leaf boundary layer conductmace
Equations (1), (2), and (9)-(12) have to be solved si-
multaneously to calculate the leaf net assimilation rate
An Since A1, A2, and As are not identical functions
of pi for Cs and C4 specie, the mathematical •!ution
of these equations will also differ For Cs, species equa-
tions (10)-(12) are combined to give the intercellular
pressure of CO2, pi, as a function of the assimilation
rate, An This function is then introduced in (3) (re- spectively (4)), assuming that A1 (respectively A2) is
the only limiting rate This procedure leads to two cu- bic equations, the solutions of which yield A1 and A2
Trang 4258 WARNANT ET AL.: GLOBAL MODEL OF TERRESTRIAL BIOLOGICAL PRODUCTIVITY
Finally, (1) and (2) are solved to provide Ap, and A,
and thus AN by making use of (9) Since A x and A2 are
not functions of pi for Ca species, (1), (2), (6)-(8), and
(10)-(12) can be combined into a cubic equation direc-
tly providing the leaf gross assimilation rate, A (Collatz
et al., 1992) A• is then calculated from (9)
Canopy Submodel
The CO2 uptake by the green vegetation is directly
related to the net assimilation of a single leaf, assuming
that physiological parameters are constant throughout
the canopy The canopy is divided into layers of equal
thickness in leaf area index (LAI) The assimilation rate
of each layer is determined as described in section 2.1.,
and layer values are added to provide the canopy assi-
milation In order to perform this integration easily, the
LAI is reduced to the nearest multiple of the layer thick-
ness, taken as 0.2 in this study The temperature, rela-
tive humidity, and CO2 pressure of the air are supposed
to be constant throughout the canopy, while light is ab-
sorbed within the canopy An exponential attenuation
of the solar flux Ir within the foliage is implemented
[Sellers, 1985]
where Io is the irradiance at the top of the canopy; L is
the cumulative LAI; and kr is the extinction coefficient
As leaves are assumed to be spherically distributed,
kLis given by
(1 - •)o.•
2p where p is the cosine of the zenith angle of the solar
beam; and •v is the scattering coefficient (0.175)
Net Primary Productivity of a Grid Point
Using the leaf and canopy submodels presented above,
the total leaf net assimilation (LNA) rate per unit area
of ground surface at a model grid point is calculated
from the relationship
[f• Lc
L=I
]
+ (1- fc,)
L=I
where the summation extends over canopy layers, A, r (C4
and A•(C3) are the net assimilation of layer L for C4
and C3 species calculated from (9) with solar irradiance
derived from (13); fc4 is the fraction of vegetation us-
ing the C4 photosynthetic pathway at the grid point
considered; fo is the fraction of the soil surface cove-
red by vegetation; and Lc, the number of layers in the
canopy, is determined from the leaf area index of the
canopy, LAIc (assuming, as mentioned above, that the layer thickness in LAI is 0.2)
To obtain the net primary productivity of the grid
point (per unit area of ground surface), the respiration
rate R•, of the woody parts of the vegetation must be substracted from the total leaf net assimilation rate
NPP - LNA- R• (16)
Consequently, an estimate of the woody respiration rate is needed before the net primary productivity of the vegetation can be calculated Woody respiration is very poorly known quantitatively, and a very crude es- timate is made here only for completeness of the model
The approach adopted here is similar to that used by Raich et al., [1991] and McGuire et al., [1992] in their calculation of total plant respiration Maintenance and construction respiration rates are calculated separately, since the former is proportional to the biomass and the latter to the net carbon assimilation Thus R,, can be
written as
where R• is proportional to woody biomass, and R•,
is proportional to the part of net assimilation allocated
to wood growth
Following McGuire et al., [1992], the increase of/•
with temperature is represented by a "Qm relation- ship," in which Q•o is, itself, temperature dependent and is calculated from a third-order polynomial fit to observations So we have
[j•o T ln(Q•o) 5 T] (18)
R• m = Kr Bw exp 10
where B•ois the woody phytomass in standing vegeta- tion; Kr is the respiration rate per unit mass of woody material at 0øC; and T is temperature in øC Since on-
ly the living part of the woody phytomass respires, K•
must be substantially lower than the values listed by
McGuire et al., [1992] for average vegetation (leaves plus wood) in different ecotypes In the absence of pre-
cise measurements, the K• value has been chosen here
in such a way that the global and annual average of the respiration rate is comparable to the value reported by Harvey [1989] The woody phytomass B•ois not cal- culated explicitly in the model Rather, it is estimated from a simple parameterization linking the annual mean NPP and the phytomass [Esser, 1984, 1991]
B•o = 0.59181 NPP•, •%o.7v2•6 (19) where •"o is the mean stand age of woody material,
and NPP,, is the annual mean NPP allocated to wood growth This annual mean value of B•o is used to
calculate R•, assuming that B•o is roughly constant
throughout the year The value of •% depends on the
ecotype Note that since the annual mean net primary
productivity of the wood NPP,, is not known until the
Trang 5WARNANT ET AL.: GLOBAL MODEL OF TERI•STRIAL BIOLOGICAL PRODUCTIVITY 259
calculation is performed over the whole year, it will be
necessary to use an initial guess of Bw while starting
an iterative procedure (see below) The NPP allocated
to wood growth is related to total NPP by the simple
relationship
NPP• = (1 - H) Nee (20)
where the herbaceous factor H is also a characteristic of
the ecotype and is calculated here from the fraction •
of the vegetated surface covered by ground vegetation
(as opposed to trees) as follows:
where h0 is the Ëaction of tree NPP allocated to leaf
growth
Using a similar approach to that of Raich et a/.,[1991],
we assume that the rate of construction respiration/•
is given by
where
NA• - LNA - H x NPP - R• TM (23)
is the net assimilation (NA) left for wood growth and
wood construction respiration when the amounts allo-
cated to growth (H x NPP) of ground vegetation and
tree leaves and to wood maintenance respiration
have been substracted from leaf net assimilation (LNA)
By substituting NPP with its value derived from (16)
and (17) and rearanging, (23) can be rewritten in the
equivalent form
NA• (1 - H) (LNA- R•) + H R•, (24)
Introducing this value of NA• into (22) and solving for
/•w, it becomes
(25)
When fo, fc,, LAIc, •, and h0 (characteristics of the
ecotype) are known,(15)-(18) together with (21) and
(25) can be solved simultaneously to yield the net pri-
mary productivity (NPP) on an hourly, diurnal, month-
ly, or annual basis (depending on which time interval
A• and /• are calculated), provided that an initial
value of B•o is known As mentioned earlier, in this
first version of the model the woody phytomass B•ois
assumed constant over the year and is estimated from
(19) and (20), that is, from the annual mean NPP As
a result, the calculations of NPP and B•omust be re-
peated iteratively until convergence In this way the
model is capable of estimating the NPP from hourly to
annual timescales The method used to calculate the
wood respiration rate is very preliminary and has been
adopted because it avoids an explicit calculation of the
biomass from mass conservation equations This sim-
plified method neglects the seasonality of carbon alloca-
tion (parameters H and h0) associated with the pheno- logical changes, but it, nevertheless, allows a correction
of the net assimilation for wood respiration, so that the model NPP can be compared with average measure- ments in the major ecotypes of the world
Input Data
The model requires two kinds of inputs, climatic data used to estimate the photosynthetic rate and vegetation data used to differentiate the ecosystems
Climatic Data All climatic inputs are monthly mean data Mean temperature T,• and precipitation P,• are given by the International Institute for Applied Systems Analysis database [Leeroans and Cromer, 1991] on a regular grid
of 0.5 ø x 0.5 ø and are averaged over the 1 ø x 1 ø grid elements of the model The monthly mean maximum and minimum temperatures, Truax and Train are given
by spatial interpolation of observations at about 3000 stations situated all over the world These observa-
tion files are those used by May et al., [1992], and a
linear interpolation is carried out, ecosystem by ecosys- tem The surface irradiances are the monthly mean va-
lues for 1989 and come from the International Satellite
Cloud Climatology Project (ISCCP) database available
at Goddard Institute for Space Studies in New York [Bishop and Rossow, 1991] These inputs are treated to
provide the mean diurnal air relative humidity and the
hourly values of temperature and irradiance
Air Relative Humidity A simple bucket model
of the surface hydrological cycle considering one single
soil pool is used to estimate the soil water content W•ao
In this model the precipitating water which is not eva-
porated fills in the soil pool until the field capacity fc is
reached Excess precipitation leaves the site as runoff The potential evapotranspiration rate is calculated with
a parameterization, depending on temperature and so-
lax irradiance (Turc formula) The thickness of the soil layer (rooting depth) depends on the vegetation type and the soil texture It varies between 1.0 and 2.5 m,
except for lithosols, for which a value of 0.1 m is as- sumed The soil water content is limited to a minimum
equal to the wilting point, wp The field capacity and the wilting point are fianctions of soil texture The air
relative humidity h8 is estimated with a formula adap-
ted from Sellers [1983]
In this formula, wH,O is limited to top to be consistent
with the hydrological model
Trang 6260 WARNANT ET AL.: GLOBAL MODEL OF TERRESTRIAL BIOLOGICAL PRODUCTIVITY
Temperature The diurnal variation of tempera-
ture is introduced in the model using a rough estimate
A T cos(2•r
= + (h- 14) 24 )
where h is the local solar hour, and A T = Tm.x- Train
Surface Irradiance The hourly irradiance /surf
(W m -2) is deduced from daily mean values and from
the irradiance calculated at the top of the atmosphere,
I,o, (W m -2)
Is,rf(h) = Itoa(h)exp[-kd(h)] (28)
where d(h), the length of the path of the solar beam
through the atmosphere (km), is calculated from the
solar zenith angle at hour h, assuming a spherical at-
mosphere The monthly mean atmospheric extinction
coefficient k (km -1) is determined by comparing the
ISCCP monthly means of daily irradiances at the sur-
face, Isurf (W m-2), with the daily mean irradiances at
the top of the atmosphere calculated for the middle of
the same month Its average daily value, used in (28),
is obtained from linear interpolation of the monthly va-
lueso
The direct and diffuse parts of the irradiance/air and
Iaig are estimated, as suggested, by Grunt et al [1989]
Iaig = 1.2 {1-exp[-kd(h)]) I•urf (29)
= - Finally, similarly to the formulation of Raich et al
[1991], it is supposed that 45% of the direct and 65% of
the diffuse radiation are photosynthetically active radia-
tion (PAR), while 0.825 (= 1-•) of the PAR is absorbed
by the vegetation [Sellers, 1985] The average energy of
one absorbed PAR photon is 3.6 x 10 -x9 J
Vegetation Characteristics
The ecotype classification by Wilson and Henderson-
Sellers [1985] is used It gives the distribution of con-
tinental ecotypes on a 1 ø x 1 ø grid For each eco-
type, several pieces of informations about the vegetation
are required including physiological parameters ( V•max,
Jm,x), canopy or plant characteristics (LAI, stand age
of the woody material, the fraction h0 of tree NPP al-
located to leaf growth), and spatial distributions (the
fraction f0 of the soil surface covered by vegetation,
the fraction fc• of vegetation using C4 photosynthe-
tic pathway, and the relative areas covered by ground
vegetation (•) and by trees (1-•))
Physiological parameters Measurements of the
maximum rate of carboxylation and of the maximum
rate of electron transport have been listed and grouped
into broad categories by Wullschleger [1993] These da-
ta are used to estimate the value of Vcm•x and Jm•x
appropriate to each ecotype To do this, values repre-
sentative of ground vegetation and trees are averaged
with the weight factor • V•m•x, Jm•x, and • are listed
in Table 1 For C4 species the values of the physiological
parameters are those of Collatz et al [1992]
Canopy or plant characteristics At this stage
the leaf area index of the canopy LAIc is fixed by a pa-
rameterization depending on the monthly temperature
Tm [Pitman et al., 1991]
LAIc = LAIm,x - ALAI[1 - f(T•,)] (31)
= o <_ ooc f(Tm) = 1-0.0016(25-Tin) 0øC <Tm < 25øC;
f(Tm)= I Tm _> 25øC
where LAIm,x is the maximum leaf area index, and ALAI is the amplitude of the seasonal variation of the leaf area index LAIc is minimum for temperatures lo- wer than 0øC, increases quadratically with temperature, and reaches its maximum at 25øC The maximum va- lue LAImax and the amplitude of the seasonal variation ALAI of the leaf area index differ between ecotypes de- pending on their seasonal behavior and their growth po- tentiality Owing to this rough parameterization, some canopy layers may have negative annual NPP These layers are not considered, and the canopy LAI is lowe- red accordingly The values used here for LAIm•x and
ALAI are taken from Pitman et al [1991] and are given
in Table 1
The stand age of woody material rw is listed in Ta- ble I as a function of the ecotype; these values are de- rived from Esser [1991] The Ëaction h0 of tree NPP allocated to leaf growth has been estimated from da-
ta for some species of tropical, temperate deciduous and coniferous forests reported by Bray and Gorham [1964], Dajoz [1982], Duvignaud [1971], and Schlesinger [1991] The average values obtained are 0.30 for tropical species, 0.27 for temperate deciduous trees, and 0.20 for coniferous trees These values are assumed to be con- stant over the year This hypothesis of constancy is made for the sake of simplicity, although it is obviously not correct For instance, cold deciduous leaves grow essentially during spring However, the consequence of this hypothesis on the model results discussed in this paper are minor Indeed, h0 is used only to calculate
the herbaceous factor H used in (19) and (25) to es- timate B•, and R•, The first of these two variables,
B•,, is calculated on a yearly averaged basis and thus the mean annual value of h0 must be used, while the
second variable, R•,, is a flux of secondary importance
for the estimate of the NPP
Spatial Distribution The seasonal variation of the Ëaction f0 of the surface covered by vegetation is described with a parameterization similar to that used for LAIc [Pitman et al., 1991]
fo f0,max Afo (1 - f(T.,)) (32)
where fO,m•x is the maximum fo value over the year
The amplitude A fo of the seasonal variation of fo is
Trang 8generally small, except for cultivated ecotypes, where
it simulates the cycle of sowing and harvesting As for
LAImax and ALAI, the values used here for fo and Afo
are taken from Pitman et al [1991] and are given in
Table 1 Values of • listed in Table I have been de-
fined so that the herbaceous factors H calculated from
(21) are comparable to those given by Esser [1991] for
similar ecotypes The fraction • is assumed constant
throughout the year
For agricultural vegetation the fraction of C4 species
is fc4 = 0.5 when the ecotype is maize or cane sugar
(assuming only 50% of the 1 ø x 1 ø grid element is cove-
red with corn or cane sugar, the rest of the cell being
covered with natural grassland) and is fc4 = 0.0 for the
other agricultural ecosystems
In the absence of global data for natural vegetation
the C4 species are very roughly distributed in natural
ecosystems as follows: fca = 0.1 • between 15øS and
15øN latitude; fc, 0.1 • to 0, decreasing linearly be-
tween 15 ø and 30ø; and fca 0.0 at latitudes higher
than 30 ø
Results
The spatial distribution of the annual NPP calcu-
lated by the CARAIB model is presented in Plate 1
The general features appear realistic; the productivi-
ty is high in tropical regions (Amazionian and tropi-
cal forests) and decreases at higher latitudes; desertic
areas (Sahara, Australian low-productivity regions, Hi- malaya) are clearly visible; and even smaller geographi-
cal entities such as the Alps or the Ural Mountains can
be seen Very high NPP values in the United States
and in central Europe are due to the presence of maize crops Since maize uses the C4 photosynthetic pathway,
its productivity may be much higher than that of other cultures and even than that of natural species when in-
tensively cultivated The desertic areas appear to be too
extensive, as, for example, in South Africa This fea- ture can be attributed to the hydrological model which underestimates the soil water content of semidesertic ar-
eas Presumably, this low soil water content is, at least partly, linked to the assumption made in the hydrologi- cal model that the monthly total precipitation of a site
is distributed uniformly over time within the month This assumption, obviously not correct in semidesertic
areas, leads to too low daily precipitation, implying that
precipitated water can be readily reevaporated without filling the soil pool
The model global net primary productivity is 65 Gt C yr-•, slightly higher than other estimates For example,
Tinker and Incson [1990] give NPP ranging between 45 and 62 Gt C yr -• with a most probable value of 60
Gt C yr -• This relatively high global NPP value is certainly due, in large part, to the lack of coupling with
a model of nutrient cycle in the vegetation and soils This coupling would certainly limit the productivity of certain areas and therefore decrease the global annual
estimate
0 - 1 O0
100 - 200
œ'L'• - 300
40O - 50O
500 - 7O0
700 - 900
900-
IlOO - !300
t500 - ] 50O
Annual Net Primary Productivity (gO m-2 y-l)
Plate 1 Global map of the annual mean net primary productivity (NPP) calculated with the
C ARAIB model Note that the color scale is nonlinear
Trang 9In Figure 2 the annual values of the NPP calculated
by our model are compared with the measured local
NPP used to calibrate the terrestrial ecosystem model
[Raich et al., 1991; McGuire et al., 1992] covering a
wide range of natural ecotypes The climatic data used
in this test are those of the grid point covering the mea-
surement site The representative ecotypes are chosen
to be as close as possible to the vegetation types of the
measurements Two different behaviors appear clear-
ly in Figure 2; in some cases (twelve grid points) the
estimated NPP is equal to or higher than the measure-
ments, in the other cases (seven grid points) the NPP
is underestimated by the model In the first category
we find the boreal and temperate ecosystems Only the
less productive vegetation (arid shrubland or tundra)
shows a large discrepancy If we do not consider the
three grid points that belong to these latter ecotypes,
then the mean error is 33% Regarding the second cate-
gory for which carbon assimilation is underestimated, a
mean error of 36% is obtained Local conditions, in par-
ticular, regional climatic ones, can explain, in part, the
differences between estimates and measurements The
test of the model results at a global scale will, for exam-
ple, show that the calculated NPP of tropical savanna
is relatively high, while it is underestimated in this first
test However, even at a global scale the tundra seems
to be less productive than expected from the model,
and equatorial forests may have a higher productivity
than the calculated ones
A further test of our model is performed by com-
paring our results with empirical estimates gathered in
the Intergovernmental Panel on Climate Control (IPCC)
15oo
1 ooo
5O0
meosured NPP (gC m-' yr-') Figure 2 Comparison of calculated annual mean net
primary productivities with local measurements which
were used to calibrate the terrestrial ecosystem model
[Raich et al., 1991; McGuire et al., 1992]
[Houghton et al., 1990] scientific assessment of climatic
change (Table 2) To make such a comparison possible,
we grouped the ecosystems of Wilson and Henderson- Sellers [1985] into 13 broad vegetation types, defined as
follows (the numbers in parentheses refer to Wilson and Henderson-Sellers ecotypes): desert (70,71,73), tundra (61,62), grassland (30,31,34,36), savanna (32,33,37),
shrub land (16,24,27,28,35,39), needle leaf forest (10,18),
boreal and temperate woodland (11,13,14,17,21), tem- perate broadleaf and mixed forest (12,15,19,20), tropi-
cal woodland (23,26), tropical deciduous forest (25,52), equatorial rain forest (50,51), wetland (2,5), and culti-
vation (4,40-49,80) Three other ecotypes for which we
do not calculate the NPP (inland water, semidesert, hu- man area) are added to provide a list compatible with
the IPCC published one Unfortunately, even the broad
vegetation types differ between authors This is ob-
viously emphasized by the differences in global areas
covered by each ecotype (Table 3) In particular, it
is clear from Table 3 that the ecotypes named here as woodlands are considered forests by other authors This discrepancy may induce some differences between the various estimates Moreover, it is not always possible to find a correspondence between Wilson and Henderson- Sellers and IPCC's ecotypes, so that we compare our calculated NPP for needle leaf forest with the boreal forest estimates As already pointed out, the global NPP estimated by CARAIB is rather high Consistent-
ly, for almost all ecotypes the calculated NPP is in the
higher part of the spectrum of empirical values How- ever, the relative NPP variations between ecotypes are similar to empirical ones Moreover, in almost all ca-
ses, the CARAIB NPP is within the range of the IPCC
values and even very close to the values of Atjay et al [1979] In fact, our estimates differ from Atjay et al.'s
values by less than 20% for seven ecotypes (grassland,
savanna, shrub land, needle leaf forest, temperate for-
est, tropical deciduous and equatorial forest), this dif-
ference being less than 10% for five of them For deserts the difference is 3 gC m -2 yr -1 For four ecotypes (in- land water, semidesert, boreal temperate, and tropical
woodland) the comparison is impossible, and for wood-
lands our estimates are relatively close to the values of corresponding forests Only two ecotypes show impor- tant discrepancies, tundra and wetland The reasons for the differences in the NPP of tundra may be multi- ple First of all Atjay et al [1979] (as do Whittaker and Likens [1975]) obviously use a definition of tundra more restrictive than ours, as can be seen by comparing the
global area covered by tundras (Table 3) Some of our
tundras are more southern ones and are thus more pro- ductive However, this is certainly not the only reason for the discrepancy, and uncertainties in the hydrologi- cal cycle, the absence of nutrient cycles, or the inaccu- racy of physiological parameters may be responsible for some of the difference The problem caused by wetlands
is not surprising, since C ARAIB does not take into ac- count the specificities of such an ecotype Nevertheless,
Trang 10264 WARNANT ET AL.: GLOBAL MODEL OF TERRESTRIAL BIOLOGICAL PRODUCTIVITY
Table 2 Net Primary Productivity (NPP)
NPP
NPP values are in gC m -2 yr -1, and global values are in GtC yr -1 The results of the CARAIB model (column 0) are compared with estimates cited in Houghton et al., [1990]; [Whittaker and Likens, 1975] (column 1); [Atjay et 199] (column 2); and [Olson et al., 1985] (column 3) It is assumed that the needle leaf forests are representative of the boreal forests
Table 3 Surface Area as a Function of the Ecosystem
Surface Area, 106 km 2
[Olson et al., 1985] (column 3) It is assumed that the needle leaf forests are representative of the boreal forests Note that desert and semidesert categories of columns 1, 2, and 3 include Antarctica (-• 14 x 106 km2), a continent not considered by CARAIB