The regional effects of CO
2
and landscape change using
a coupled plant and meteorological model
JOSEPH L. EASTMAN,*,³ MICHAEL B. COUGHENOUR² and
ROGER A. PIELKE, SR.³
*Department of Biology, University of New Mexico, Albuquerque, NM 87131, ²Natural Resources Ecology Laboratory, Colorado
State University, Fort Collins, CO 80523, ³Department of Atmospheric Sciences, Colorado State University, Fort Collins,
CO 80523, USA
Abstract
A meteorological model, the Regional Atmospheric Modelling System (RAMS),
and a plant model, the General Energy and Mass Transfer Model (GEMTM), are
coupled in this study. The integrated modelling system was used to investigate
regional weather conditions in the central grasslands of the USA for three experi-
mental scenarios:
d land cover is changed from current to potential vegetation;
d radiative forcing is changed from 1 3 CO
2
to 2 3 CO
2
;and
d biological CO
2
partial pressures are doubled.
Results indicate that the biological effect of enriched CO
2
, and of land-use change
exhibit dominant effects on regional meteorological and biological ®elds, which
were observed for daily to seasonal time scales and grid to regional spatial scales.
Simulated radiation impacts of 2 3 CO
2
were minimal, with interactive effects
between the three experimental scenarios as large as the radiational impact alone.
Model results highlight the importance of including 2 3 CO
2
biological effects
when simulating possible future changes in regional weather.
Keywords: CO
2
, feedbacks, land-use, mesoscale, radiation, regional, vegetation
Received 22 December 1999; revised version received and accepted 25 September 2000
Introduction
The role of CO
2
and/or land-use change on climate has
been the focus of several research efforts (e.g. Bonan et al.
1992; Lyons et al. 1993; Claussen 1994, 1998; Dirmeyer &
Shukla 1994; Foley 1994; Avissar 1995; Cutrim et al. 1995;
Sivillo et al. 1997; Texier et al. 1997; De Ridder & Gallee
1998; Kanae et al. 1999; Kiang & Eltahir 1999; Li et al. 2000;
O'Brien 2000; Wang & Eltahir 2000a; see also the review
in Pielke 2001). Foley et al. (1998) suggest, however, that
linked global and vegetation models have been limited
by the consideration of only equilibrium responses of
vegetation to shifting climate and the utilization of
separate models that might not be physically consistent.
It has also been suggested that the wide timescales at
work make this an initial value problem with numerous
possible outcomes (Pielke 1998; Pielke et al. 1999).
Recent work simulating palaeoclimate changes on the
global scale (De Noblet et al. 1996; Claussen & Gayler
1997; Crowley & Baum 1997; Texier et al. 1997; Brostrom
et al. 1998; Ganopolski et al. 1998) and on the regional
scale in the west Africa region (Kiang & Eltahir 1999;
Wang & Eltahir 2000a,b,c,d) indicate a crucial role for
biospheric feedbacks on climate. Several biophysical,
biogeochemical, and biogeographical parameters have
been shown to have an impact on the atmosphere. These
include, for example, albedo, stomatal conductance,
rooting pro®le, fractional coverage, roughness length
and displacement height, the time of planting and
harvesting, vegetation phenology, and plant growth.
Their effects on the atmosphere will in¯uence, amongst
other things, the diurnal temperature evolution, wind
pro®le, turbulent transfer, Bowen ratio, precipitation, soil
*Correspondence: Joseph Eastman, Department of Atmospheric
Science, Colorado State University, Fort Collins, CO 80523, USA,
tel/fax +1/970-491-8293, e-mail dallas@cobra.atmos.colostate.
edu
Global Change Biology (2001) 7, 797±815
ã 2001 Blackwell Science Ltd
797
wetness and soil temperature, and evaporation.
Although the relative role of radiative and stomatal
conductance in a global model has been compared
(Sellers et al. 1997), other biophysical effects, such as
the effect of enriched CO
2
concentration on plant growth,
have not yet been incorporated. In the Betts et al. (1997)
evaluation of the physiological and structural feedbacks
on a doubled CO
2
climate, it was found that structural
changes in vegetation partially offset the physiological
vegetation±climate feedbacks over the long term. It was
also suggested that overall vegetation feedbacks could
cause signi®cant regional-scale effects. Previous vegeta-
tion±climate feedback simulations did not incorporate
iteratively coupled model components, and therefore
may not have accounted accurately for important
physiological features of vegetation.
Stomatal conductance is sensitive to changes in atmos-
pheric CO
2
, with elevated CO
2
resulting in decreased
stomatal conductance, and either constant or increasing
photosynthesis rates. The net result is a general increase
in water-use ef®ciency (WUE): the ratio of photosynthetic
carbon gain to transpiration water loss. Clearly, this will
also alter the ratio of the sensible and latent heat ¯ux,
providing another feedback on local weather conditions.
In addition, changing free atmosphere CO
2
will affect
growth and respiration rates (Idso 1992; Ryan et al. 1997).
Changing the proportions of live and dead matter will
produce changes in the nutrient cycling. There are other
implications resulting from increasing CO
2
as outlined in
Bazzaz (1990). The authors point out the following:
d C3 plants are more responsive than C4 plants to
elevated CO
2
.
d Photosynthesis is enhanced by CO
2
but this enhance-
ment may decline with time.
d The response to CO
2
is more pronounced under high
levels of other resources, especially water, nutrients,
and light.
d Adjustment of photosynthesis during growth occurs in
some species but not in others, and this adjustment
may be in¯uenced by resource availability.
d Species, even of the same community, may differ in
their response to CO
2
.
These points have been demonstrated in ®eld studies
(Ryan 1991; Owensby 1993). In terms of modelling
efforts, these points have been generally ignored to
varying degrees.
In order to investigate the effects of changing CO
2
and
land-use on a regional scale, a model capable of repre-
senting the aforementioned effects is required. A mod-
elling system that is highly mechanistic in its formulation
needs to be used to ascertain the effects resulting from
landscape change, radiative CO
2
forcing, and changing
plant growth patterns as a consequence of enhanced CO
2
concentrations. The model must be able to simulate the
nonlinear interactions between these effects at regional
and local spatial scales. This mechanistic formulation
was pointed out in Reynolds & Acock (1985), where they
argued that plant growth models used to investigate
elevated CO
2
must be mechanistic or semimechanistic in
the areas that directly affect plant behaviour. This is
reinforced in Chen et al. (1994) where much of the basis
for the plant growth model used here was developed. In
their work describing C4 photosynthesis, they suggest
that this mechanistic approach will address the short-
term effects of increasing CO
2
, as well as the long-term
effects resulting from plant acclimation to rising CO
2
.
Photosynthetic acclimation to elevated CO
2
, as pointed
out by Bazzaz (1990), is species- or ecosystem-dependent.
In nitrogen-poor environments, acclimation is a bene®t to
the plant because it needs to conserve the scarce
resource. Earlier studies generally employed potted
plants in their experimental design, but Eamus (1996)
points out that acclimation may not be as prevalent when
the roots are unconstrained. In Bryant et al. (1998) down-
regulation was observed, yet the net rates of photosyn-
thesis were still enhanced. In this same study, it was
found that two forbs had actually reversed their decrease
in rubisco activity and did not show evidence of down-
regulation. In the present paper, the problem of tackling
acclimation and subsequent down-regulation is not
addressed by this modelling system owing to the
complexity and existing lack of understanding of this
issue.
Modelling tools
The meteorological model used in this investigation is
the Regional Atmospheric Modelling System (RAMS).
RAMS is a 3D model that employs fundamental equa-
tions of ¯uid dynamics. It was further modi®ed (Liston &
Pielke 2000) for longer integrations and is named
ClimRAMS. ClimRAMS, a mesoscale model, has several
additions to RAMS that make it desirable for assessing
regional climate effects. Some of the main features of the
3D atmospheric model include:
d variable initialization and 3D nudging; 3D nudging
adds a time-scale dependent prognostic term to force
model-simulated quantities like wind, temperature,
and moisture, towards observations at the outer
boundaries (within 5 gridpoints of the lateral bound-
ary for these simulations);
d dump bucket precipitation (Cotton et al. 1995)Ðthis
scheme evaluates the saturation levels in the atmos-
phere, and assuming an ef®ciency for precipitation,
removes the excess moisture as precipitation;
d Kuo cumulus convection parameterization (Kuo &
Raymond 1980); this is a simple parameterization that
simulates cumulus convection and is triggered by
798 J . L . E A S T M A N et al.
ã 2001 Blackwell Science Ltd, Global Change Biology, 7, 797±815
exceeding a threshold of vertical motion at the model-
simulated cloud base;
d longwave and shortwave radiative ¯uxes (Mahrer &
Pielke 1977)Ðthis scheme is rather simple, and does
not account for attenuation of shortwave radiation by
clouds;
d multilevel soil model to simulate variable soil tem-
perature and moisture (Tremback & Kessler 1985);
d additional subroutines to simulate grid-scale snow
accumulation, snow melt, and their effects on surface
hydrology and surface energy budgets.
ClimRAMS advects and diffuses the atmospheric CO
2
and simulates diurnal, as well as seasonal changes in
CO
2
that are representative of observational studies, such
as Bazzaz & Williams (1991). The CO
2
concentrations
resulting from these calculations can be used in both
biological and radiative calculations, or for each one
separately, leaving the other constant.
ClimRAMS has been successfully linked to the General
Energy and Mass Transfer Model (GEMTM; Chen et al.
1994). The ClimRAMS and GEMTM models are physic-
ally consistent in their formulations, as both are based on
similar theory for prognosing surface energy budgets
(Eastman 1999). The version of GEMTM used in this
study contains the following:
d explicit C3 (Farquhar et al. 1980) and C4 (Collatz et al.
1992; Chen & Coughenour 1994) photosynthesis;
d allocation, respiration, plant growth and mortality
rates based on temperature and moisture relation-
ships;
d spatially explicit root model for ef¯uence and uptake,
including branching and lengthening algorithms
(Chen & Lieth 1992, 1993);
d multilevel canopy radiation model for prediction of
diffuse and direct radiation (Goudriaan 1977; Chen
1983);
d variable biomass initialization;
d unlimited species types.
Both models are interfaced on a common timestep of
90 s. The meteorological model calculates several soil
and surface related quantities, such as soil moisture and
temperature, as well as water and heat ¯uxes. In the
calculation of soil moisture in the root zone, a dynam-
ically changing root resistance model is employed. The
model allows water to be absorbed and released from the
roots. It is assumed that the transpiration rate is equal to
the soil water uptake for calculations of the plant water
potential at the base of the xylem. These calculations
involve several surface-vegetation related quantities
including albedo, fractional coverage, Leaf Area Index
(LAI, m
2
m
±2
), and roughness length, with the latter three
updated daily. Surface radiation, soil temperature, and
soil moisture parameters are in turn employed for
algorithms used to calculate direct and diffuse radiation,
with a separate photosynthesis calculation performed for
direct and diffuse photosynthetically active radiation
(PAR). The model calculates respiration, mortality,
growth rates, and allocation to roots, shoots, leaves,
and seeds, using temperature and moisture relations. At
the end of the modelling day, 0 GMT here, the plant
model updates root characteristics. This involves alloca-
tion to branching and lengthening and subsequent root
weighted density according to the available net photo-
synthate. The allocation is based on water and tempera-
ture stress at a given root level.
The updated stomatal conductance, calculated accord-
ing to Ball et al. (1987), LAI, roughness length, and
fractional coverage are then returned to the meteoro-
logical model. These quantities are then used in similar-
ity theory calculations in ClimRAMS to modify the
surface heat and moisture ¯ux. The fractional coverage is
used to weight the net heat and moisture ¯ux due to bare
and shaded soil, water, and vegetation.
Initialization
The different components of the modelling system
require several datasets in order to initialize simulations.
The domain used is shown with the natural and current
vegetation distributions (Fig. 1a,b). Horizontal grid spa-
cing is 50 km. For the vertical coordinates, an initial
increment of 100 m is used at the surface, which is then
slowly increased to 1.5 km near the top of the model.
There are 30 points in the E±W direction, 26 points in the
N±S direction, and 20 points in the vertical. Topography
was initialized using the USGS 30 s dataset.
All simulations were integrated for 210 days over the
growing season, beginning on 1 April 1989. In order to
simulate the 1989 growing season, the model uses a
Newtonian relaxation method at the outer 3 grid points
of the domain. Newtonian relaxation adds a tendency
term to the prognosed quantity that drives it towards the
observations from the NCEP reanalysis product (Kalnay
et al. 1996). The reanalysis product was used to create the
initial ®elds, as well as the lateral boundary conditions.
The data are processed through an isentropic analysis
package, which is interpolated to the ClimRAMS grid.
The Vegetation Mapping and Analysis Project
(VEMAP, Kittel et al. 1995) provided the vegetation
distribution for the current and potential landscapes
(Ku
È
chler 1964; Ku
È
chler 1975). In order to set the above-
and belowground biomass ®elds, the Advanced Very
High Resolution Radiometer (AVHRR) derived
Normalized Difference Vegetation Index (NDVI) product
was used to calculate spatially explicit LAI using a
separate algorithm for woody and herbaceous vegetation
(Asrar et al. 1984; Nemani & Running 1989). Leaf biomass
was in turn determined from a speci®c leaf area consist-
C O U P L E D P L A N T A N D M E T E O R O L O G I C A L M O D E L 799
ã 2001 Blackwell Science Ltd, Global Change Biology, 7, 797±815
ent with the biome type. Shoot biomass was determined
through a literature survey of typical leaf and shoot
biomass observed for different vegetation classes.
Finally, a plant-speci®c root-to-shoot ratio was used to
determine the belowground biomass for herbaceous
vegetation. In the case of woody vegetation, root biomass
was determined using the shoot biomass corresponding
to the yearly maximum LAI determined from the
AVHRR±NDVI data.
A spatially variable objective analysis was employed to
initialize the soil model. The soil class data are directly
read from the STATSGO soils dataset (U.S. Department
of Agriculture 1994). Soil textural class is assumed
uniform within the vertical column. Data used to
variably initialize the soil moisture and temperature
pro®les are a product of the ECMWF/TOGA Advanced
Operational Analysis Data. The soil wetness and tem-
perature are based on this product following the work of
Mintz & Sera®ni (1981). The surface soil climatological
moisture corresponds to a layer from the surface to a
depth of 7.2 cm. The deep-layer climatological soil
moisture and temperature values are derived for a
layer from 7.2 cm to a depth of 42 cm. Soil data are
de®ned spatially on a 1° 3 1° grid, while the temporal
resolution is a one-month averaged ®eld. The initial soil
temperature and moisture ®elds are de®ned by linear
interpolation between the 42-cm-deep climatology soil
temperature and moisture, and the RAMS objectively
analysed surface temperature ®elds. Soil grid points
below the 42 cm depth are set to the 42 cm value. The
bottom soil layer, at 2 m below ground, then serves as a
nudging boundary condition and is temporally interpol-
ated between monthly values of the climatological
dataset for moisture only.
Hypothesis and experimental design
In the introduction, the various interactions for CO
2
and
vegetation were discussed brie¯y. Using a mechanistic
model, the various perturbation experiments (which we
will call factors), can be quanti®ed individually through
a separation technique designed by Stein & Alpert (1993).
This separation also enables the examination of the
nonlinear interactions between the factors. This study
examines the contributions of vegetation change (poten-
tial and current), 13 and 2 3 CO
2
radiative forcing, and
13 and 2 3 CO
2
concentrations felt only by the biota
(called CO
2
biology from here on). It also examines the
Fig. 1 (a) Modelling domain and natural vegetation distribu-
tion. (b) modelling domain and current vegetation distribution.
Classes represent: 1, tundra; 2, subalpine; 3, temperate conifer;
4, temperate deciduous; 5, temperate xeromorphic; 6, temper-
ate coniferous xeromorphic; 7, savanna and deciduous; 8, C3
shortgrass; 9, C4 tall grass; 10, temperate arid shrub; 11, spring
wheat/small grass; 12, small grains; 13, winter wheat; 14, corn;
15, irrigated crop; 16, deciduous forest crop; 17, subtropical
mixed forest; and 18, grassland and grain.
Table 1 Summary of the 8 simulations performed and the
symbols used to refer to them
Simulation
symbol Vegetation
CO
2
radiation
CO
2
biology
f
0
Current 13 13
f
1
Natural 13 13
f
2
Current 23 13
f
3
Current 13 23
f
12
Natural 23 13
f
13
Natural 13 23
f
23
Current 23 23
f
123
Natural 23 23
800 J . L . E A S T M A N et al.
ã 2001 Blackwell Science Ltd, Global Change Biology, 7, 797±815
various interactive combinations of these factors. The
eight simulations needed to perform the analysis are
outlined in Table 1, and the difference ®elds of each
factor contribution are shown in Table 2.
Signi®cant differences in the domain-averaged prog-
nosed meteorological and biological ®elds between the
control simulation and perturbation simulations are
examined when:
d land cover is changed from current to potential
vegetation;
d radiative forcing is changed from 1 3 CO
2
to 2 3 CO
2
;
and
d biological CO
2
concentrations are doubled.
The results are examined through temporal domain
averages, domain- and temporal-averaged means and
biases, and spatial time series at individual gridpoints.
The analysis is performed for daily updated variables
and 2-h averaged data. The Kolmogorov±Smirnoff (KS)
test (Anderson & Darling 1952) is used to determine
whether signi®cant differences are present in the cumu-
lative distribution function.
Results
Validation
Meteorology. The control simulation, f
0
, was initially
compared against observed data from standard
meteorological stations, to quantify the ability of the
simulation in capturing major trends in regional
weather patterns. The meteorologically observed data
were produced by objectively analysing surface station
data for the ClimRAMS model grid. The meteorological
variables tested were maximum daily temperature,
minimum daily temperature, and daily precipitation.
Spatially explicit plots were created from a comparison
of the temporal series at each gridpoint for the
modelled and observed variable.
A spatially explicit correlation coef®cient, r, and bias
were calculated for the temperature ®elds and are shown
in Figs 2(a)±(b) and 3(a)±(b), respectively. Modelled
temperature ®elds exhibit an acceptable degree of
correlation with the model for both maximum and
minimum daily temperature, with an r-value ranging
from 0.72 to 0.92. There is one isolated area in the south-
east portion of the domain that exhibits a relative
minimum in correlation, although the values are still
higher than 0.7. This is likely a consequence of its
distance from the in¯ow boundaries, because the ¯ow is
from west or north-west to the east and south-east
portions of the domain. This implies that this area would
receive the least amount of information from the lateral
boundaries.
Fig. 3(a)±(b) presents the spatially explicit plots for the
bias in the indicated variable. The warmest maximum
temperature bias is indicated in the area where the
lowest correlation was found. The magnitudes are
around 2±3 °C throughout the area. There also appears
to be a signi®cant cool bias in the western portions of the
domain. These likely result from the in¯uence of topog-
raphy, which is not well resolved with the 50-km grid.
Domain-averaged correlation, bias, and KS statistic
were calculated for temperature ®elds. These are
summarized in Table 3. The KS statistic indicates that
neither the maximum nor minimum daily temperature
distributions are signi®cantly different between the
model and observations at the a = 0.05 signi®cance
level. A value of 1.0 would indicate a perfect match
between the datasets. The correlation coef®cients are also
relatively high on a domain-averaged scale. Finally, the
biases indicate that the domain-averaged maximum
daily temperature is undersimulated by nearly a degree,
while the domain-averaged minimum daily temperature
is undersimulated by just over a tenth of a degree. These
numbers indicate that the model has captured the
regional weather patterns observed during the 1989
growing season. Such validation experiments show that
the precipitation was considerably more dif®cult than
temperature to predict. Simulated precipitation was
found to be about 60% of the observed values, except
during large-scale synoptic events in the spring and fall,
where the tendency was to overpredict.
A plot of the domain-averaged precipitation is dis-
played in Fig. 4. Validation experiments indicate that the
present model does a reasonable job of capturing the
temporal progression of the domain-averaged rainfall.
Summer period precipitation predictions appear to be
consistently undersimulated.
Biological. In order to help validate plant model
predictions, AVHRR±NDVI-derived LAI was compared
to the simulated ®elds using a 10-d averaged temporal
Table 2 Description of the difference ®elds and their meaning
Difference Contribution resulting from
f
1 ±
f
0
natural vegetation
f
2
± f
0
2 3 CO
2
radiation
f
3
± f
0
2 3 CO
2
biology
f
12
± (f
1
+ f
2)
+ f
0
the interaction of natural vegetation and
2 3 CO
2
radiation
f
13
± (f
1
+ f
3)
+ f
0
the interaction of natural vegetation and
2 3 CO
2
biology
f
23
± (f
2
+ f
3
) + f
0
the interaction of 2 3 CO
2
radiation and
2 3 CO
2
biology
f
123
+ (f
1
+ f
2
+ f
3
)
± (f
12
+ f
13
f
23)
± f
0
the interaction of natural vegetation,
2 3 CO
2
radiation, and 2 3 CO
2
biology
C O U P L E D P L A N T A N D M E T E O R O L O G I C A L M O D E L 801
ã 2001 Blackwell Science Ltd, Global Change Biology, 7, 797±815
