Comparing Impacts across Climate Models

Abstract

In this paper we combine a climate-forecasting model, COSMIC, with a global impact model, GIM, to compare the market impacts of climate change projected by 14 general circulation models. Given a specific date (2100), carbon dioxide concentration (612 ppmv), and global temperature sensitivity (2.5C), predicted impacts to economies are calculated using climate-response functions from Experimental and Cross-sectional evidence. The Cross-sectional impact model predicts small global benefits across all climate models, whereas the Experimental impact model predicts a range from small benefits to small damages. High-latitude countries are less sensitive to temperature increases than low-latitude countries because they are currently cool. Uniform global temperature changes overestimate global damages because they underestimate the benefits in polar regions and overestimate the damages in tropical regions compared to the GCM predictions.

Full text

Integrated Assessment 1 (2000) 37–48 37
Comparing impacts across climate models
Robert Mendelsohn
a
, Michael Schlesinger
b
and Larry Williams
c
a
Yale University, 360 Prospect Street, New Haven, CT 06511, USA
b
University of Illinois, 105 South Gregory, Urbana, IL 61801, USA
c
Electric Power Research Institute, PO Box 10412, Palo Alto, CA 94303, USA
Received 8 September 1999; revised 19 November 1999
In this paper we combine a climate-forecasting model, COSMIC, with a global impact model, GIM, to compare the market impacts of
climate change projected by 14 general circulation models. Given a specific date (2100), carbon dioxide concentration (612 ppmv), and
global temperature sensitivity (2.5
◦
C), predicted impacts to economies are calculated using climate-response functions from Experimental
and Cross-sectional evidence. The Cross-sectional impact model predicts small global benefits across all climate models, whereas
the Experimental impact model predicts a range from small benefits to small damages. High-latitude countries are less sensitive to
temperature increases than low-latitude countries because they are currently cool. Uniform global temperature changes overestimate
global damages because they underestimate the benefits in polar regions and overestimate the damages in tropical regions compared to
the GCM predictions.
1. Introduction
Two tools have recently been developed to project an-
thropogenically induced climate changes and their impacts.
COSMIC [25] uses the global-mean surface temperature
calculated by a simple climate/ocean model to scale in time
the geographical patterns of changes in surface temperature
and precipitation simulated by any of 14 General Circula-
tion models (GCMs) to generate country-specific climate
projections. GIM [19] combines country-specific climate
projections, market-sector data, and climate-response func-
tions to predict market impacts from warming by sector and
nation. In this paper we combinethese two tools to compare
the projections of the 14 GCMs whose results are included
in COSMIC. This large number of GCMs was selected to
illustrate the variation across models. We thought it im-
portant to show more than one climate-response function
again to illustrate differences. However, it is important that
policy analysts understand that the scenarios do not demon-
strate all possible sources of uncertainty. For example, we
do not explore the effect of alternative baseline emission
scenarios, variations in temperature sensitivity, alternative
time lags due to the ocean, and alternative climate-response
functions. Instead, the scenarios emphasize the new in-
sights from making country-level associations between cli-
mate forecasts and impacts. The integration of GCMs and
country-level impact models provides a new mechanism for
comparing GCM forecasts and for understanding the distri-
bution of impacts across the earth.
There are many links in the chain from carbon dioxide
emissions to climate projections [13]. We know a lot about
each step and yet each step remains a source of uncertainty.
Alternative assumptions about the path of the economy al-
ter emissions over space and time. The carbon cycle de-
termines how these emissions affect concentrations in the
atmosphere. Changes in carbon dioxide and other man-
made greenhouse gases alter the net radiative flux at the
top of the atmosphere, which constitutes a radiative forcing
of the climate system. Changes in temperature at different
heights in the atmosphere will in turn affect clouds and ice,
either compounding or reducing this radiative effect, gen-
erating global temperature sensitivity. Finally, the ocean
will gradually respond to temperature change, warming at
different levels over time, creating a dynamic response that
may take decades to centuries to play out. COSMIC com-
bines a simple climate/ocean model with the geographical
distributions of the changes in surface temperature and pre-
cipitation simulated by 14 GCMs. Assuming that the geo-
graphic climate distributions from each model can be scaled
by the global-mean surface temperature change, COSMIC
predicts country-specific climate projections for each GCM.
GIM ties the country-specific projections of changes in
surface temperature and precipitation to predictions of mar-
ket impacts by sector and country. First, GIM projects what
the future economy will look like at the time of the impact.
Models of each sector must reference baseline projections
of economic activity for that sector in order to calculate fu-
ture climate impacts. For example, agriculture is predicted
to grow more slowly than the rest of the economy, cooling
will grow relative to warming in space conditioning, and the
world economy will be about ten times larger by 2100. Sec-
ond, the climate-response function of each sector must be
predicted. Unfortunately, this is one of the weakest links in
global-warming research. It is very uncertain how the entire
earth will respond to any given climate change. Climate-
response functions have been estimated only for market
effects, that is, impacts to economic sectors [18]. Although
these projections are presented as point estimates, they are
clearly uncertain. We present two projections that come
 Baltzer Science Publishers BV

38 R. Mendelsohn et al. / Comparing impacts across climate models
from entirely different empirical approaches to illustrate the
range of uncertainty. Even this is an understatement of the
true level of uncertainty since we have very little empirical
information about the impacts of climate change on devel-
oping countries. Finally, the climate-response functions in
this paper have been calibrated only for the United States.
The climate-response functions for the quality of life such
as ecological changes, human health, and aesthetic impacts,
are still under development; hence, they are not yet included
in GIM. Thus, the market impacts discussed in this paper
cannot be compared directly against estimates of all impacts
such as in [4,22,32].
Since climate itself is changing very slowly, it is ex-
tremely difficult to measure across time the sensitivity of
market sectors to climate change. The literature has conse-
quently resorted to two alternative approaches, Experimen-
tal and Cross-sectional. The Experimental approach con-
structs process-based simulation models from carefully con-
ducted scientific experiments. Using laboratory-controlled
settings, experiments are run on crops, trees, and other sub-
jects to determine their sensitivity to temperature, precip-
itation, and carbon dioxide. Simulation models are con-
structed from the experimental evidence to predict what
will happen in the aggregate. The Cross-sectional ap-
proach uses evidence from alternative locations to make
predictions. By comparing one area with a warmer site,
the Cross-sectional approach is able to discern what would
happen to that place in the long run if it warmed. Both
approaches have strengths and weaknesses. The Experi-
mental approach is able to isolate the impact of each ele-
ment through laboratory controls, while the Cross-sectional
approach is subject to unwanted variation from factors it
fails to control. This advantage of the Experimental ap-
proach leads to a weakness. The controls imposed by the
Experimental approach may eliminate important responses
by subjects, that is, adaptations that limit damages and en-
hance benefits. Since the Cross-sectional approach includes
responses people have already made to where they live,
the Cross-sectional approach captures adaptation. Because
the two methods have different strengths and weaknesses,
and both are highly uncertain, it is prudent to include both
methods in impact analysis.
In this paper we explore a set of future conditions to
compare the impacts calculated for the changes in sur-
face temperature and precipitation simulated by 14 GCMs.
We utilize both Experimental and Cross-sectional climate-
response functions to evaluate the climate projections based
on each of the GCMs. By examining the impacts, we can
gain new insights into which projected climate changes are
important, which are consistent across the GCMs, and how
best to describe these projections for the globe. We focus
on impacts in 2100 to obtain a long-run, but still relevant,
perspective. We assume that humankind commits itself to
a maximum equivalent CO
2
concentration of 750 ppmv,
which implies a carbon dioxide concentration of 612 ppmv
in 2100 [13]. We assume that global population has dou-
bled to 10 billion and that the global gross domestic product
(GDP) is 217 trillion, a ten-fold increase. Finally, we as-
sume that global temperature sensitivity is 2.5
◦
C, that is,
the equilibrium global-mean surface temperature increase
for a CO
2
doubling is 2.5
◦
C.
In the next section we describe various features of
COSMIC and GIM in detail. In section 3 we describe the
aggregate results for each GCM. In section 4 we analyze
these alternative results to explain which features of the
climate projections generate large impacts, which projec-
tions are consistent, and how to aggregate climate forecasts
across the earth from an impacts perspective.
2. COSMIC and GIM
2.1. COSMIC
Analyses of the impacts of anthropogenically induced
climate changes require time-dependent scenarios of the
geographical distributions of these climate changes. If the
anthropogenic emissions of greenhouse gases (GHGs) and
sulfur dioxide (SO
2
) and the sensitivity of the climate
system were known, the best-possible method of con-
structing geographical scenarios of climate change would
be to perform climate-change simulations with coupled
atmosphere–ocean general circulation models (CGCMs),
from pre-industrial time into the future. However, the fu-
ture anthropogenic emissions of GHGs and CO
2
are highly
uncertain (e.g., [30]), as is the radiative forcing due to
sulfate (SO
4
) aerosol created in the atmosphere from the
emitted SO
2
[13]. Thus it is computationally impossible
to perform with a CGCM the multitude of climate-change
simulations required to span the ranges of possible emis-
sion scenarios, climate sensitivities and sulfate radiative
forcing (∆F
SO
4
(1990)). Accordingly a simpler, computa-
tionally practicable method of constructing these numerous
geographical scenarios of climate change is needed.
The method of scenario construction we employ in the
Country Specific Model for Intertemporal Climate (COS-
MIC) was developed by [23] and further refined at the
Climatic Research Unit, University of East Anglia, Nor-
wich, UK [14], and used in the scenario generation code,
SCENGEN [15]. The method has also been used else-
where [24,25,29,30]. An atmospheric GCM with a mixed-
layer ocean (AGC/MLO) model is used to simulate a con-
trol (con) equilibrium climate and the equilibrium experi-
ment (exp) climate for an enhanced CO
2
concentration. The
geographical distribution of CO
2
-induced equilibrium cli-
mate change for any climatic quantity, Q(λ, ϕ,m), where λ
and ϕ are the longitude and latitude of the AGC/MLO
model’s grid cells and m is the calendar month, is then
calculated and normalized by the corresponding change in
annual global-mean surface-air temperature, T
exp
− T
con
,
∆Q
N
(λ,ϕ,m) =
Q
exp
(λ,ϕ, m) − Q
con
(λ,ϕ, m)
T
exp
− T
con
. (1)

R. Mendelsohn et al. / Comparing impacts across climate models 39
The change in annual global-mean surface-air temperature,
∆T (y) = T (y) − T (y
0
), (2)
from some pre-industrial time, y
0
(1765), to the present
is calculated with our energy-balance–climate/upwelling–
diffusion–ocean model [28] using the historical radia-
tive forcing of the Intergovernmental Panel on Climate
Change [13] and then into the future for a prescribed
climate-change scenario and a prescribed climate sensitiv-
ity, G. The changes in annual global-mean surface-air tem-
perature relative to a reference year, y
ref
(= 1990), are then
calculated from
δT (y) = T (y) − T (y
ref
) = ∆T (y) − ∆T (y
ref
). (3)
The time-dependent geographical distributions of monthly
climate change in year y relative to y
ref
for each climate-
change scenario of equivalent CO
2
concentration
1
and pre-
scribed climate sensitivity are then determined from
δQ(λ,ϕ, y,m) = δT (y)∆Q
N
(λ,ϕ, m), (4)
the normalized pattern of greenhouse-gas-induced climate
change being taken to be the same as the pattern of
CO
2
-induced climate change, ∆Q
N
(λ,ϕ, m). Finally, the
monthly climatic quantities in year y are obtained from
Q(λ,ϕ, y,m) = Q
obs
(λ,ϕ, y
ref
,m) + δQ(λ,ϕ,y,m), (5)
where Q
obs
(λ,ϕ, y
ref
,m) are the observed climatic quanti-
ties for month m of the reference year or period containing
the reference year.
In COSMIC the geographical distributions of the nor-
malized changes in monthly mean surface-air temperature
and precipitation, ∆Q
N
(λ,ϕ, m), can be chosen for any of
the 14 GCMs listed in table 1. The changes in annual
global-mean surface-air temperature, δT (y), can be cal-
culated by COSMIC for any of 7 main scenarios of fu-
ture concentrations of greenhouse gases [39], each for low,
medium and high sulfate aerosol emission rates
2
[39], as
well as for two ways (proposed by [13] and [37]) of stabi-
lizing the CO
2
concentration at either 350, 450, 550, 650
or 750 ppmv. Each calculation can be performed for a
wide range of values of G and ∆F
SO
4
(1990). COSMIC
calculates the country-specific annual cycles of surface-air
temperature and precipitation rate for 177 countries, as well
as the global-mean sea-level rise, and is available gratis on
compact disk.
3
1
The amount of CO
2
required to give the same radiative forcing as all
the greenhouse gases together.
2
In Version 1 of COSMIC, the sulfate aerosol burden influences only
δT (y), that is, no account is taken of the geographical distribution of
climate change due to the sulfate aerosol. The latter will be included in
Version 2 of COSMIC.
3
To request a no-cost license contact: Larry J. Williams; Electric Power
Research Institute; 3412 Hillview Avenue; E-mail: ljwillia@epri.com;
fax: (650) 855-2950.
2.2. GIM
GIM is a spreadsheet model that begins with a country-
specific set of climate changes and then predicts market
impacts. A separate model is designed for each sensi-
tive market sector: agriculture, forestry, energy, water, and
coastal structures. A separate calculation is made for each
sector and country that combines the change in climate,
sector data, and a climate-response function. This leads
to calculations of damages or benefits by sector and coun-
try. Quality-of-life effects such as changes in ecosystems,
health, and aesthetic losses are not included in this version
of the model as climate-response functions for these effects
are not yet available.
The current version of GIM responds to annual temper-
ature and precipitation. Future versions of the model will
move to seasonal climate variables to gain more detailed
insight into climate impacts. Annual climate by country
is one of the inputs to the model. These projections are
obtained from COSMIC for each GCM.
For each country, key parameters of each sector are col-
lected. For example, area of cropland, area of forestland,
and length of coastline provide important insights into agri-
culture, forestry, and coastal structures, respectively. Gross
Domestic Product (GDP) by country is also a key in several
sectors. As GIM becomes more sophisticated, additional
parameters will be collected for each country.
The heart of GIM is its climate-response functions. Ear-
lier impact research predicted impacts from a limited set
of climate scenarios. Examining individual scenarios be-
comes cumbersome when it is important to evaluate a large
number of scenarios and when one evaluates a path of cli-
mate change. Consequently, the literature had begun to
develop climate-response functions, descriptions of how
impacts change within a sector as climate changes [18].
Many integrated assessment models have climate-response
functions to measure damages, given a path of climate
change [12,16,21,22]. Unfortunately, many of these re-
sponse functions were invented by the authors or were fit
to very limited observations (for example, current condi-
tions and doubling of greenhouse gases). In this paper, we
rely upon climate-response functions based on empirical
research [20]. In that study, over a dozen of the lead-
ing impact researchers did empirical studies of each of the
climate-sensitive sectors of the US economy. There were
four key elements in this new research: inclusion of effi-
cient adaptation, broad sectoral estimates, dynamic analysis
when appropriate, and use of future economic conditions.
The research relied upon the two major alternative meth-
ods of measuring the response to climate. Several studies
relied upon the Experimental method, which begins with
carefully controlled laboratory studies and uses these to
construct simulation models. The remainder of the studies
relied on Cross-sectional evidence. By comparing farms
and households in cool versus warm locations, one can es-
timate how people have adapted to their resident climates
and how they may react as these climates change in the

40 R. Mendelsohn et al. / Comparing impacts across climate models
Table 1
GCM model simulations: global annual averages for doubling the pre-industrial carbon dioxide concentration.
Acronym Institution ∆T ∆P Reference
(
◦
C) (%)
BMRC Bureau of Meteorology Research Center 2.11 2.38 [10]
CCC Canadian Climate Centre 3.50 4.00 [2,3,17]
GF30 Geophysical Fluid Dynamics Laboratory
(R30 run)
4.00 8.3 [35,36]
GFDL Geophysical Fluid Dynamics Laboratory
(first run)
4.00 8.3 [35,36]
GFQF Geophysical Fluid Dynamics Laboratory
(Q-flux run)
4.00 8.30 [35,36]
GISS Goddard Institute for Space Studies 4.20 11.00 [7–9]
HEND Henderson-Sellers using CCM1 at NCAR 2.50 5.60 [11]
OSU Schlesinger and Zhao at Oregon State
University
2.40 7.80 [26]
POLD Pollard and Thompson-GENESIS with
dynamic sea-ice
2.27 3.13 [31]
POLS Pollard and Thompson-GENESIS with
static sea-ice
2.27 3.13 [31]
UIUC Schlesinger at University of Illinois at
Urbana-Champaign
3.37 5.53 [28]
UKMO United Kingdom Meteorological Office 5.20 15.00 [38]
WANG Wang et al. at State University of New
York at Albany and NCAR
a
3.90 6.90 [33]
WASH Washington and Meehl using CCM1
at NCAR
4.82 4.75 [34]
a
NCAR is the National Center for Atmospheric Research.
Table 2
Aggregate impacts in 2100 by GCM model experimental responses (billions of 1990 $/year).
GCM Continent
a
Total Africa Asia LatAm WEur Comm NAm Ocean
BMRC 54 −112 −31 −67 1 224 48 −9
CCC 28 −139 −52 −87 9 250 62 −13
GF30 210 −79 −9 −30 14 245 83 −16
GFDL 203 −100 −8 −37 12 267 81 −12
GFQF 134 −113 −1 −49 13 224 78 −17
GISS 45 −103 −86 −59 15 217 73 −13
HEND −69 −163 −103 −73 9 216 66 −21
OSU −33 −111 −157 −45 13 209 68 −10
POLS 147 −134 40 −92 10 230 114 −22
POLD 163 −103 −77 −44 20 270 112 −14
UIUC −139 −186 −161 −97 10 223 85 −12
UKMO 27 −139 −97 −62 14 245 82 −17
WANG −29 −143 −145 −72 18 239 90 −17
WASH 25 −123 −90 −55 12 219 79 −18
AVERAGE 55 −125 −70 −62 12 234 80 −15
a
The continents above are Africa, Asia, Latin America, Western Europe, the former Soviet Union
and Eastern bloc, North America, and Oceania.
long run. The strength of the Experimental method is that
it can isolate climate effects from other factors in the en-
vironment. Further, it can explore the effect of factors that
are not yet evident in the environment, such as higher levels
of carbon dioxide. The weakness of the approach is that
experiments are designed to control responses, both en-
vironmental and human. Adaptations that ecological sys-
tems and people make to climate change are suppressed,
thereby exaggerating the damages and reducing the bene-
fits from warming. The Cross-sectional approach is able
to capture efficient adaptations because the method com-
pares systems currently adapted to different climates. For
example, the farm in a cool place is compared to a farm in
a warm place, given all the adaptations that farmers have
made to where they live. This advantage of Cross-sectional
evidence comes at a cost. Cross-sectional studies are vul-
nerable to unmeasured factors that may be correlated with
climate. If these factors are not taken into account, they

R. Mendelsohn et al. / Comparing impacts across climate models 41
Table 3
Aggregate impacts in 2100 by GCM model Cross-sectional responses (billions of 1990 $/year).
GCM Continent
a
Total Africa Asia LatAm WEur Comm NAm Ocean
BMRC 150 −10 32 −3 2 100 29 −1
CCC 152 −18 31 −6 5 108 33 −2
GF30 185 −5 35 3 6 106 41 −2
GFDL 184 −9 31 2 5 114 42 −1
GFQF 165 −12 35 0 6 98 41 −3
GISS 131 −15 17 −7 7 94 38 −2
HEND 97 −28 8 −10 5 95 32 −4
OSU 116 −15 0 −3 6 93 37 −1
POLS 173 −16 39 −7 6 101 53 −4
POLD 175 −10 21 −2 8 112 48 −2
UIUC 98 −31 −1 −14 5 99 42 −2
UKMO 136 −21 16 −5 6 104 39 −3
WANG 119 −22 1 −9 7 102 43 −3
WASH 143 −13 22 −2 5 96 38 −3
AVERAGE 145 −16 21 −5 6 102 40 −2
a
The continents above are Africa, Asia, Latin America, Western Europe, the former Soviet Union
and Eastern bloc, North America, and Oceania.
can be confused with climate effects, thereby leading to
misleading results. This is not a problem for the carefully
controlled experimental studies. Consequently, the Experi-
mental and Cross-sectional methods complement each other
well, and we rely upon both of them in this study.
The climate-response functions in these studies were
quadratic in temperature. That is, the response function
indicated a hill-shaped relationship between impacts and
temperature. This is an essential feature of the model and
explains many of the results in this paper. Countries that
are currently cooler than optimal are predicted to benefit
from warming. Countries that happen to be warmer than
optimal are predicted to be harmed by warming. Although
the quantitative measures shown in the paper remain highly
uncertain, these qualitative insights are likely to be robust.
3. Results
Combining the projections of COSMIC and GIM, one
can examine the impacts from a wide set of climate mod-
els. COSMIC provides a consistent set of conditions so that
the scenario, global temperature sensitivity, and ocean dy-
namics are the same. Given these identical conditions, one
can then study the alternative distributional patterns pre-
dicted by each GCM and examine their effect on impacts.
Although there have been a number of GCM comparisons
conducted by atmospheric scientists, these studies focused
on climate projections, not the resulting impacts [6]. Previ-
ous comparisons have consequently not been able to iden-
tify which aspects of these projections are important, what
impacts do these GCMs consistently agree upon, which as-
pects lead to a wide range of impacts, and how best to
aggregate climate projections across the earth.
To compare the 14 GCMs using a consistent set of start-
ing conditions, we make a number of assumptions. First,
we assume that carbon emissions are on a global path con-
sistent with reaching a maximum of 750 ppmv [13]. Sec-
ond, we examine the impacts in 2100. Given the IPCC path
specified above, carbon dioxide will reach 612 ppmv
by 2100. Third, we specify a global temperature sensitivity
of 2.5
◦
C. The model predicts a global-average temperature
of 2.21
◦
C by 2100. Fourth, we assume that the economy
grows according to medium projections so that global GDP
is $217 trillion by 2100. Given these assumptions, we cal-
culate the country-specific climate outcomes according to
each of the 14 GCMs in COSMIC. These climate changes
are then used to predict impacts by market sector for each
country.
In tables 2 and 3 we present continental estimates of
aggregate market impacts for each of the 14 GCMs us-
ing the Experimental and Cross-sectional climate-response
functions, respectively. Compared to the size of the econ-
omy in 2100 ($217 trillion), the market effects are small.
Global net impacts have a broad range across GCMs us-
ing the Experimental climate-response functions: from
$139 billion of damages to $210 billion of benefits, with
an average of $55 billion of benefits. The Cross-sectional
climate-response functions imply a narrower range of im-
pacts across GCMs: from $97 to $185 billion of benefits,
with an average of $145 billion of benefits a year. The
Experimental climate-response functions are more steeply
hill-shaped and thus they respond more sharply to tem-
perature increases in the polar and tropical regions. This
explains why the Experimental results are more sensitive
to the variety of GCM projections.
The results are also quite different across countries.
First, the GCMs generally agree that temperature change
increases with latitude. The GCMs also agree that pre-

42 R. Mendelsohn et al. / Comparing impacts across climate models
cipitation changes will not be uniform, although the pre-
cipitation projections are not consistent across the GCMs.
However, the impact models suggest that the magnitude of
impacts will depend not only on the changes in temperature
and precipitation, but also on the base conditions in each
country. Countries that are already hot or dry will be more
vulnerable to warming. Countries that are cold, in contrast,
are likely to benefit from warming. These initial conditions
lead to different outcomes across countries.
The results indicate that there will be large benefits from
warming in the Former Communist bloc (the former So-
viet Union and Eastern Bloc countries). The benefits in
this region almost offset losses throughout the tropics in
the Experimental results. The Soviet benefits account for
two-thirds of the net global benefits in the Cross-sectional
results. The results also suggest that there will be large ben-
efits in North America and small benefits in Western Eu-
rope. The critical factor that these benefiting countries have
in common is that they are currently cool so that warming
is helpful. The Experimental model predicts sizeable dam-
ages from warming in Africa, Latin America, Oceania, and
often Asia because these areas are currently already hot.
In contrast, the Cross-sectional model predicts that Africa,
Oceania, and Latin America will suffer only modest dam-
ages because of the compensating effects of carbon fertil-
ization and adaptation, and that Asia will likely benefit.
It is interesting to compare this geographic pattern
against the predictions of other authors. Tol predicts ben-
efits for more polar countries and damages for low lat-
itude countries as in this analysis [32]. However, both
Fankhauser and Nordhaus predict more uniform damages
across the entire world [4,22]. It appears that both of these
latter studies fail to fully account for initial climate condi-
tions in predicting warming impacts in a country. Because
it is important, future integrated assessment models will
need to do a better job of integrating geographically spe-
cific climate predictions and impacts.
The most important market impact from warming is agri-
cultural. According to the Experimental results, agriculture
was responsible for average global benefits of $88 billion,
compared to total net benefits of only $55 billion. The
Cross-sectional results were similar; agriculture would pro-
vide average benefits of $163 billion compared to total net
market benefits of $145 billion. Forestry was also per-
ceived as being beneficial, contributing an additional $20
and $29 billion in the Experimental and Cross-sectional re-
sults, respectively. The remaining sectors were expected
to generate net damages. Water damages were expected to
average $32 billion, energy damages were expected to be
about $9 billion, and coastal impacts were anticipated to be
$6 billion.
In addition to generating the largest expected effect, agri-
culture also explains most of the variation both across coun-
tries and across the GCMs. The standard error of aggre-
gate market impacts across the 14 GCMs is $101 billion
in the Experimental results and $54 billion in the Cross-
sectional results. The standard error for agriculturalimpacts
is $94 billion in the Experimental results and $52 billion
in the Cross-sectional results. Agriculture is the source of
most of the variation across models. In comparison, wa-
ter has a standard error of only $9 billion and energy only
$4 billion.
The three GCMs that predict the largest benefits in ta-
bles 2 and 3; GF30, GFDL, and POLD, all predict large in-
creases in temperature at high latitudes and small increases
at low latitudes. The benefits in the Communist bloc coun-
tries and North America are consequently higher and the
damages in Latin America and Africa are lower. In con-
trast, the three GCMs that predict the greatest damages or
smallest benefits in tables 2 and 3; HEND, OSU, and UIUC,
predict more uniform temperature changes; relatively high
values for low latitudes and relatively modest increases at
high latitudes. The benefits to the polar countries are con-
sequently smaller and the damages in the tropical countries
are higher.
The variability of estimated global impacts that result
from differences in the GCM climate forecasts can be seen
in more detail with the maps shown in figures 1 and 2. Fig-
ure 1 shows annual percentage changes in GDP as calcu-
lated by the Experimentalclimate-response functions driven
by different climate-change projections. The GCMs used
to prepare the three maps were chosen to represent the
maximum, average, and minimum impacts. There are sev-
eral ways in which the maximum and minimum impact
maps could be chosen. Total global market welfare losses
(shown in tables 2 and 3) could be used. This would result
in a measure mainly dependent on the countries with the
largest GDPs. An alternative would be to compare area- or
population-weighted percent GDP changes. In some sense
this would result in maps with the most/least red, and would
place the most weight on countries with the largest areas
or populations. Instead, we chose a method that weights
each country equally, independent of GDP, area, or popula-
tion. This method ranks 14 possible maps according to the
sum of percentage changes in GDP across all 177 countries
included in the GIM model.
The maximum-impact map in both figures 1 and 2 re-
sulted from using the UIUC GCM climate simulation. This
GCM generated larger impacts because it predicts relatively
more warming in the tropics than the other models. The
minimum-impactmap (again for both figures) was produced
with the POLD simulation. This GCM predicts more ben-
efits because it predicts relatively more warming for more
polar countries and less warming in the tropics. The top
map in figure 1 and 2 is the minimum impact map and the
maximum map is at the bottom of the figures. The middle
map shows the average impact, calculated by averaging the
impacts estimated by each of the 14 climate models used
in this analysis.
The most striking feature of figures 1 and 2 is the simi-
larity between maps going from top (minimum impact) to
bottom (maximum impact). Of course, the choice of “bins”
shown in the legend strongly affects the main features of
the maps. Nevertheless, these maps support the points

R. Mendelsohn et al. / Comparing impacts across climate models 43
Figure 1. Range of impacts calculated using Experimental climate-response functions. The POLD model produced smaller impacts than most other Gcms.
The UIUC model led the high impact end of the group.

44 R. Mendelsohn et al. / Comparing impacts across climate models
Figure 2. Range of impacts calculated using Cross-sectional climate-response functions. The POLD model produced smaller impacts than most
other Gcms. The UIUC model led the high impact end of the group.

R. Mendelsohn et al. / Comparing impacts across climate models 45
Table 4
Regressions of aggregate impacts on average global climate
change.
a
Impact
Experimental
= 720 − 341 T
Pop
+ 1711 P
Pop
186 77 687
Impact
Experimental
= 809 − 317 T
Area
+ 639 P
Area
423 163 917
Impact
Cross-sectional
= 375 − 109 T
Pop
+ 206 P
Pop
62 26 228
Impact
Cross-sectional
= 381 − 94 T
Area
− 29 P
Area
126 49 274
a
Dependent variable is net global market effects for 2100 in
billions of 1990$. Climate variables measure the aggregate
change in temperature and precipitation weighted by either
population or area. The standard errors are below the regres-
sion equation.
evident in tables 2 and 3. Developing countries in the trop-
ics are likely to be harmed by expected climate change,
while the developed countries, and transition economies, in
temperate and northern climates will see a net improvement
in the market sectors that are most responsive to climate
changes.
Although it has become customary to average tem-
perature increases across the entire globe when reporting
global changes, global impacts are more sensitive to the
population-weighted average change. We compare two
alternative aggregations of temperature and precipitation
across nations. The area measure weights climate in coun-
tries by total area. The population measure weights all
climate changes by the number of people in each coun-
try. Countries with more people get more weight. Ta-
ble 4 reports regressions of global net impacts across the
14 models on aggregate temperature and precipitation us-
ing the two alternative weights. The population-weighted
measure of both temperature and precipitation is statisti-
cally more significant and can explain a greater fraction of
the variance of global impacts across the climate models.
Population-weighted temperature and precipitation changes
are better predictors of impacts than land-weighted aver-
ages. This insight is likely to apply to national averages
as well. Weighting grids by population can give a better
estimate of the average temperature change than weighting
grids by area.
Table 4 provides another key insight. The coefficient
on temperature change is negative and significant in all the
models. Although the net impacts of climate change are
beneficial relative to an unchanged state, the models imply
that higher temperatures are harmful. There are two expla-
nations of this result. First, the climate-response function
for temperature is hill-shaped, not linear. Starting from
a cool climate, warming is beneficial at first. However,
as warming continues, more countries exceed the optimum
and warming becomes increasingly harmful. By 2100, all
the GCMs predict that the unweighted global temperature
change will exceed 2
◦
C at which point further warming
is harmful. Second, changes in precipitation and carbon
dioxide are beneficial. Thus the overall net impact of all
Table 5
Regressions of regional impacts on regional climate
change (Experimental).
a
Impact
Africa
= 44 − 86 T
Pop
+ 287 P
Pop
63 14
Impact
Asia
= 214 − 157 T
Pop
+ 994 P
Pop
103 42 332
Impact
LatAmer
= 42 − 59 T
Pop
+ 357 P
Pop
15 8 34
Impact
WEur
= 9 − 1T
Pop
+ 93 P
Pop
52 11
Impact
Soviet
= 99 + 39 T
Pop
+ 235 P
Pop
41 13 76
Impact
NAmer
= 84 − 7T
Pop
+ 216 P
Pop
28 7 42
Impact
Oceania
= 6 − 11 T
Pop
+ 28 P
Pop
11 6 17
a
Dependent variable is regional impacts in 2100 in bil-
lions of 1990$. Regional climate change is average
change in region.
Table 6
Regressions of regional impacts on regional climate
change (Cross-sectional).
a
Impact
Africa
= 37 − 26 T
Pop
+ 38 P
Pop
11 3
Impact
Asia
= 110 − 44 T
Pop
+ 103 P
Pop
27 11 85
Impact
LatAmer
= 34 − 21 T
Pop
+ 44 P
Pop
32 7
Impact
WEur
= 8 − 1T
Pop
+ 25 P
Pop
21 4
Impact
Soviet
= 56 + 14 T
Pop
+ 60 P
Pop
16 5 29
Impact
NAmer
= 51 − 5T
Pop
+ 54 P
Pop
10 3 20
Impact
Oceania
= 5 − 4T
Pop
+ 7P
Pop
31 4
a
Dependent variable is regional impacts in 2100 meas-
ured in billions of 1990$. Regional climate change
is average change in region.
the changes is beneficial, even though the marginal effect
of additional temperature is harmful by 2100.
The sensitivity of each sector to climate is not uni-
form across all regions. Tables 5 and 6 display the re-
gional sensitivity. These sensitivities were calculated by
regressing the Experimental and Cross-sectional impacts
on the population-weighted climate measures for each re-
gion. North America, Western Europe, and the Soviet bloc
all have positive or small negative temperature coefficients
because they are currently cool. In contrast, Africa, Asia,
and Latin America have large negative temperature coeffi-
cients because they are currently hot. The resources each
continent possesses determine the size of the coefficients.
Asia has large coefficients because it has the most people,
whereas Oceania has few people and thus small coefficients
throughout. Tables 5 and 6 also reveal that the temperature
sensitivity of the Experimental results is greater in magni-

46 R. Mendelsohn et al. / Comparing impacts across climate models
Table 7
Market impacts from uniform climate change.
a
Measure Continent
b
Total Africa Asia LatAm WEur Comm NAm Ocean
Experimental
Average GCM 59 −125 −69 −60 14 234 80 −15
Area −130 −145 −154 −82 15 187 67 −19
Population −72 −123 −120 −67 15 172 65 −15
Cross-sectional
Average GCM 146 −16 21 −4 6 102 40 −2
Area 95 −21 2 −12 7 87 36 −3
Population 114 −15 12 −7 7 81 36 −2
a
Impacts are measured in billions of 1990 $/year. The area-weighted uniform temperature change is 2.49
◦
C with
a precipitation increase of 5.5% and the population-weighted uniform temperature change is 2.21
◦
C, with a
precipitation increase of 5.2%.
b
The continents are Africa, Asia, Latin America, Western Europe, the former Soviet Union and Eastern bloc, North
America, and Oceania.
tude than that of the Cross-sectional results. As mentioned
earlier, this heightened sensitivity is due to the more steeply
shaped Experimental climate-response functions.
In order to shed more light on these results, we compare
the GCM results to the impacts from a uniform climate
change. We examine the impacts predicted by GIM using
two uniform climate predictions: the area and population-
weighted average temperature and precipitation change.
The results are displayed in table 7. Even though uni-
formity implies the same change in temperature and pre-
cipitation in every country, the impacts vary widely. Coun-
tries that begin cool benefit whereas countries that begin
warm are harmed. Comparing the uniform results to the
impacts generated by the GCMs reveals significant differ-
ences. The average uniform scenarios predict large dam-
ages in low-latitude countries in Africa, Asia, and Latin
America, and smaller benefits in high-latitude countries.
The uniform climate changes overestimate global damages.
The population-weighted results are better than the area-
weighted estimates, but they suffer from the same prob-
lems. The uniform scenarios miss the important variation
in temperature across latitudes.
This paper aggregates market effects across continents
without making any adjustments for the incomes of the im-
pacted countries. Some authors have argued that such ad-
justments should be made [5]. Unfortunately, the world has
yet to agree on a social-welfare function that could generate
such weights. Further, one would also have to weigh costs.
As costs become more evenly distributed across countries,
the importance of such weights diminishes [1]. Figures 1
and 2 both present impacts as a fraction of GDP. The results
indicate that impacts will be small relative to GDP. Because
the low-income countries tend to be clustered in the low lat-
itudes, the maps also indicate that any weighting scheme
that placed higher weights on low-income countries would
tend to emphasize the damages in the more tropical coun-
tries relative to the benefits in the more temperate countries.
Thus, the more weight one gave to low-income countries,
the more the resulting index would lean towards damages.
4. Conclusion
This paper combines COSMIC, a climate-projection
tool, and GIM, an impact-projection tool, to examine the
country-specific market impacts predicted by 14 GCMs
for 2100. Although there is considerable uncertainty about
the exact magnitude of country-specific impacts, there are
a number of insights from this research. First, the mod-
est climate-change scenarios expected by 2100 are likely to
have only a small effect on the world economy. The mar-
ket impacts predicted in this analysis do not exceed 0.1%
of global GDP and are likely to be smaller. Second, the
market impacts will vary from country to country across the
globe. High-latitude countries are expected to gain and low-
latitude countries are expected to be harmed by warming.
Third, although the overall effects of warming and carbon
fertilization on the globe in 2100 are near zero, the mar-
ginal effect of higher temperature is expected to be harmful.
Temperature changes beyond 2
◦
C are expected to reduce
benefits and increase damages. Fourth, the GCMs pre-
dict greater warming near the poles and less warming near
the equator relative to a uniform climate change scenario.
These consistent deviations reduce damages (increase ben-
efits) relative to a uniform climate change and should be
taken into account. This research is intended to illustrate
the power of COSMIC and GIM as forecasting tools. The
research is also intended to reveal weaknesses or problems
with these forecasts. For example, the current use of coun-
trywide estimates of climate change is problematic for large
countries because climates vary sufficiently within national
borders that more localized estimates would be preferable.
Another weakness in these forecasts is the reliance on an-
nual temperature changes. Future models should attempt
to model seasonal changes. A third prominent weakness
involves the reliance on United States evidence to cali-
brate the responses to climate change. Clearly it would
be preferable to have estimates of regional responses to
climate change from around the world. Finally, in many
sectors, it would be attractive to get more detailed infor-
mation about each country. For example, there is no soil

R. Mendelsohn et al. / Comparing impacts across climate models 47
data in the current agriculture or forestry models, little in-
formation about space heating and cooling in the energy
model, and little data about runoff in the water models.
Prudent policy-makers should understand that the country-
specific estimates of impacts are consequently preliminary
and are likely to change as the models become more so-
phisticated.
Finally, we have measured only market effects from pre-
dicted climate changes. Preliminary research indicates that
climate change is also likely to impact the quality of life.
Effects on ecosystems, health, and aesthetics have not been
taken into account in this analysis. Impacts from changes
in extreme events or catastrophes should also be measured.
As research in these areas develops, the model can be
revised to include these more complete measures of im-
pacts.
Acknowledgements
This research was funded by the Electric Power Research
Institute and the US National Science Foundation (Grant
No ATM-9522681).
References
[1] C. Azar, Weight factors in cost-benefit analysis of climate change,
Env. and Res. Econ. 13 (1999) 249–268.
[2] G.J. Boer, N. McFarlane and M. Lazare, Greenhouse gas induced
climate change simulated with the Canadian Climate Centre second
generation general circulation model, J. Clim. 5 (1992) 1045–1077.
[3] S.J. Cohen, G.J. Boer, N. McFarlane, J.P. Blanchet, M. Lazare, N.E.
Sargent, F.G. Majaess, D.P. Phillips, M. Webb and T. Cutler, Ap-
plication of the Canadian Climate Center general circulation model
output for regional climate impact studies, Canadian Climate Centre,
Downsview, Ontario, Canada (1990).
[4] S. Fankhauser, Valuing Climate Change, The Economics of the
Greenhouse (Earthscan, London, 1995).
[5] S. Fankhauser, R. Tol and D. Pearce, The aggregation of climate
change damages: A welfare theoretic approach, Env. and Res. Econ.
10 (1997) 249–266.
[6] W.L. Gates, A. Henderson-Sellers, C. Boer, A. Folland, B. Kitoh, F.
McAvaney, N. Semazzi, A. Smith, Q. Weaver and C. Zeng, Climate
models – evaluation, in: Climate Change 1995: The Science of
Climate Change, eds. J.T. Houghton et al. (Cambridge University
Press, Cambridge, 1996).
[7] J. Hansen, G. Russell, D. Rind, P. Stone, A. Lacis, S. Lebedeff,
R. Ruedy and L. Travis, Efficient three-dimensional global models
for climate studies: Models I and II, Mon. Wea. Rev. 111 (1983)
609–662.
[8] J. Hansen, A. Lacis, D. Rind, L. Russell, P. Stone, I. Fung, R. Ruedy
and J. Lerner, Climate sensitivity: Analysis of feedback mechanisms,
in: Climate Processes and Climate Sensitivity, Geophys. Monogr.
Vol. 29, eds. J. Hansen and T. Takahashi (American Geophysical
Union, Washington, DC, 1984) pp. 130–163.
[9] J. Hansen, I. Fung, A. Lacis, D. Rind, S. Lebedeff, R. Ruedy, L.
Russell and P. Stone, Global climate changes as forecast by Goddard
Institute for Space Studies three-dimensional model, J. Geophys.
Res. 93 (1988) 9341–9364.
[10] T.L. Hart, W. Bourke, B.J. McAvaney, B.W. Forgan and J.L.
McGregor, Atmospheric general circulation simulations with the
BMRC global spectral model: The impact of revised physical para-
meterizations, J. Clim. 3 (1990) 436–459.
[11] A. Henderson-Sellers, R.E. Dickinson, T.B. Durbidge, P.J. Kennedy,
K. McGuffie and A.J. Pitman, Tropical deforestation: Modelling
local to regional-scale climate change, J. Geophys. Res. 98 (1993)
7289–7315.
[12] C. Hope, J. Anderson and P. Wenman, Policy analysis of the green-
house effect: An application of the PAGE model, Energy Policy 21
(1993) 327–338.
[13] J.T. Houghton, L.G. Meira Filho, B.A. Callander, N. Harris, A. Kat-
tenberg and K. Maskell, eds., Climate Change 1995: The Science of
Climate Change (Cambridge University Press, Cambridge, 1996).
[14] M. Hulme, S.C. Raper and T.M. Wigley, An integrated framework
to address climate change (ESCAPE) and further developments of
the global and regional climate modules (MAGICC), Energy Policy
23 (1995a) 347–355.
[15] M. Hulme, T. Jiang and T.M. Wigley, SCENGEN, a climate change
scenario generator, a user manual, Climatic Research Unit, Univer-
sity of East Anglia, Norwich, UK (1995b).
[16] A. Manne, R. Mendelsohn and R. Richels, MERGE: A model for
evaluating regional and global effects of GHG reduction policies,
Energy Policy 23 (1993) 17–34.
[17] N.A. McFarlane, G.J. Boer, J.P. Blanchet and M. Lazare, The Cana-
dian Climate Centre second generation general circulation model and
its equilibrium climate, J. Clim. 5 (1992) 1013–1044.
[18] R. Mendelsohn and M.E. Schlesinger, Climate-response functions,
Ambio 28 (1999) 362–366.
[19] R. Mendelsohn, W. Morrison, M.E. Schlesinger and N.G.
Andronova, Country-specific market impacts of climate change, Cli-
matic Change (1999), in press.
[20] R. Mendelsohn and J. Neumann, eds., The Economic Impact of Cli-
mate Change on the United States Economy (Cambridge University
Press, Cambridge, 1998).
[21] W.D. Nordhaus, To slow or not to slow, Econ. J. 5 (1991) 920–937.
[22] W.D. Nordhaus, Managing the Global Commons: The Economics of
Climate Change (The MIT Press, Cambridge, 1994).
[23] B.D. Santer, T.M. Wigley, M.E. Schlesinger and J.F. Mitchell, De-
veloping climate scenarios from equilibrium GCM results, Report
No. 47, Max-Planck-Institut f
¨
ur Meteorologie, Hamburg, Germany,
1990.
[24] M.E. Schlesinger and N. Andronova, Regional climate-change sce-
narios based on 2 × CO
2
and 4 × CO
2
equilibrium climate-changes
simulated by the UIUC atmospheric general circulation/mixed-layer
ocean model, Report to EPRI, Urbana, IL, 1994.
[25] M.E. Schlesinger and L.J. Williams, COSMIC – Country Spe-
cific Model for Intertemporal Climate, Computer Software, Electric
Power Research Institute, Palo Alto, 1997.
[26] M.E. Schlesinger and Z.C. Zhao, Seasonal climate changes induced
by doubled CO
2
as simulated by the OSU atmospheric GCM/mixed-
layer ocean model, J. Clim. 2 (1989) 459–495.
[27] M.E. Schlesinger, N. Andronova, A. Ghanem, S. Malyshev, T.
Reichler, E. Rozanov, W. Wang and F. Yang, Geographical scenar-
ios of greenhouse gas and anthropogenic-sulfate aerosol induced cli-
mate changes, Climate Research Group, Department of Atmospheric
Sciences, University of Illinois at Urbana-Champaign, Urbana, IL,
1997a.
[28] M.E. Schlesinger, N. Andronova, B. Entwistle, A. Ghanem, N.
Ramankutty, W. Wang and F. Yang, Modeling and simulation of
climate and climate change, in: Past and Present Variability of the
Solar-Terrestrial System: Measurement, Data Analysis and Theo-
retical Models. Proceedings of the International School of Physics
“Enrico Fermi” CXXXIII, eds. G. Castagnoli and A. Provenzale (IOS
Press, Amsterdam, 1997b).
[29] M.E. Schlesinger, N. Andronova, A. Ghanem, S. Malyshev, E.
Rozanov, W. Wang and F. Yang, Geographical scenarios of green-
house gas and anthropogenic-sulfate aerosol induced climate change,
Do We Understand Global Climate Change? (Norwegian Academy
of Technological Sciences, Oslo, 1998).
[30] M.E. Schlesinger, S. Malyshev, E.V. Rozanov, F. Yang and N.G.
Andronova, Geographical Distributions of Temperature Change for

48 R. Mendelsohn et al. / Comparing impacts across climate models
the SRES Scenarios of Greenhouse Gas and Sulfur Dioxide Emis-
sions. Technological Forecasting and Social Change, 1999, submit-
ted.
[31] S.L. Thompson and D. Pollard, A global climate model (GENESIS)
with a land surface transfer scheme (LSX) Part 1. Present climate
simulation, J. Clim. 8 (1995) 732–761.
[32] R. Tol, The damage costs of climate change: Towards more com-
prehensive calculations, Env. and Res. Econ. 5 (1995) 353–374.
[33] W.C. Wang, M.P. Dudek and X. Liang, Inadequacy of effective CO
2
as a proxy to assess the greenhouse effect of other radiative gases,
Geophys. Res. Lett. 19 (1992) 1375–1378.
[34] W. Washington and G. Meehl, Greenhouse sensitivity experiments
with penetrative cumulus convection and tropical cirrus albedo ef-
fects, Clim. Dyn. 8 (1992) 211–233.
[35] R.T. Wetherald and S. Manabe, An investigation of cloud cover
change in response to thermal forcing, Climatic Change 8 (1986)
5–23.
[36] R.T. Wetherald and S. Manabe, Cloud feedback processes in a gen-
eral circulation model, J. Atmos. Sci. 45 (1988) 1397–1415.
[37] T.M. Wigley, R. Richels and J.A. Edmonds, Alternative emissions
pathways for stabilizing CO
2
concentrations, Nature 379 (1996) 240–
243.
[38] C.A. Wilson and J.F. Mitchell, A doubled CO
2
climate sensitivity
experiment with a global climate model including a simple ocean,
J. Geophys. Res. 92 (1987) 13,315–13,343.
[39] G.W. Yohe and M.E. Schlesinger, Sea level change: The expected
economic cost of protection or abandonment in the United States,
Climatic Change 38 (1998) 447–472.