Climate Model Simulations of CO2-Induced Climatic Change

CLIMATE MODEL SIMULATIONS OF COZ-INDUCED CLIMATIC CHANGE MICHAEL E. SCHLESINGER Department of Atmospheric Sciences and Climatic Research Institute Oregon State University Corvallis. Oregon 1. Introduction. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2. Mathematical Climate Models. . . . . . . . . . . . . . . . . . . . . . . . . . . 2.1. Energy Balance Models (EBMs) . . . . . . . . . . . . . . . . . . . . . . . . 2.2. Radiative-Convective Models (RCMs) . . . . . . . . . . . . . . . . . . . . . 2.3. General Circulation Models (GCMs) . . . . . . . . . . . . . . . . . . . . . . 3. ComparisonofModelSimulationsofCO,-InducedClimaticChange . . . . . . . . . . 3.1. Description of GCMs and C02 Simulations . . . . . . . . . . . . . . . . . . 3.2. Simulated Temperature Changes. . . . . . . . . . . . . . . . . . . . . . . . 3.3. Simulated Precipitation Changes. . . . . . . . . . . . . . . . . . . . . . . . 3.4. Simulated Soil Moisture Changes . . . . . . . . . . . . . . . . . . . . . . . 4. Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4.1. Model-Dependent Results . . . . . . . . . . . . . . . . . . . . . . . . . . 4.2. Time Required to Reach Equilibrium. . . . . . . . . . . . . . . . . . . . . . 4.3. Statistical Significance . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5 . Conclusions and Recommendations . . . . . . . . . . . . . . . . . . . . . . . . References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

. .

141 143 144 144 145 152 158 165 19 1

206 216 2 17 2 19 222 228 230

1 , INTRODUCTION

Measurements taken since 1958 at Mauna Loa, Hawaii, and at other locations show that although carbon dioxide (CO,) makes up only about 0.03% by volume of the earth's atmosphere, its concentration has been increasing (Bolin and BischoE, 1970; Keeling et al., 1976a,b; Keeling and Bacastow, 1977). A study by Rotty (1982) indicated that the C02concentration increased from 1860to 1973due to a nearly constant 4.6%/yr growth in the consumption of fossil fuels (gas, oil, coal), and is continuing to increase due to the diminished 2.3%/yr growth in fossil fuel consumption since 1973. Projections of the future usage of fossil fuels predict that the CO, concentration may reach double the 1860 value of about 295 parts per million by volume (ppmv) sometime in the next century (Baes et al., 1976; Keeling and Bacastow, 1977; Rotty and Marland, 1980), and could eventually peak at 8 to 10 times the preindustrial level early in the twenty-second century (Council on Environmental Quality, 1981). As the CO, level increases, less of the temperature-dependent infrared 141 ADVANCES IN GEOPHYSICS, VOLUME 26

Copyright 0 1984 by Academic Press, Inc. All rights of reproduction in any form reserved. ,C*?.T

n

I,

n,oo?r

A

142

MICHAEL E. SCHLESINGER

radiation emitted by the earth can escape through the atmosphere to space, but the amount of solar radiation absorbed by the earth remains almost unchanged. It is to be expected that this so-called greenhouse effect will result in a warming of the average temperature of the earth such that a balance between the solar heating and infrared cooling will be restored. However, it is also to be expected that the temperature change at any location may be higher, lower, or even of opposite sign from the average temperature change, and that there may be concomitant changes in other climatic elements such as precipitation and soil moisture. The question of the possible climatic effects of increased atmospheric CO, has in recent years received more attention than any other global anthropogenic effect (National Academy of Sciences, 1976, 1977, 1979, 1982;Council on Environmental Quality, 1981). Much of this interest stems from the potential impacts that a significant climate change could have on agricultural production and energy-use practice, and hence on the patterns of global economics. One approach to estimate what the climate of a future warmer earth might be like is the analog merhod. In the analog method the seasonaland regional patterns of past warm climates are used to construct scenarios for a future warm climate (Kelloggand Schware, 1981). An advantage of this method is that the scenarios represent “surprise-free”projections (Kahn et al., 1976)in the sense that they are based on warm climatesthat have actually existed. A disadvantage of the method is that the quality of the reconstructionsof past climates, which are based on proxy data such as tree rings and ice cores, decreases with age before the present. Also, it is not possible to reconstruct all of the elements of climate that may be of interest. More importantly, the causes of most of the earth’s past warm climates are not known, and it is likely that not all were the result of elevated levels of atmospheric CO,. Consequently, a future C0,-induced warmer climate may differ substantially from the “surprise-free” scenarios based on past warm climates. Another approach to estimate what a C0,-induced climate change might be like is the physical method. In the physical method the behavior of the components of the earth’s climate system, namely, the atmosphere, oceans, snow and ice, vegetation (biomass),and land surface (Fig. I), is determined on a physical basis from the fundamental laws of nature such as the conservation of energy. These physical laws are expressed mathematically to form a mathematical climate model. The advantage of such a model is that it can be used to simulate, in a physically consistent manner, not only the present climate, but also how that climate would change in response to a change in some “external” forcing such as in the energy received from the sun or in the composition of the atmosphere. A disadvantage of the physical method is the inherent inability to construct a model that has perfect similitude to the actual climate system.

143

MODELS OF C02-INDUCED CLIMATIC CHANGE Changes of Solar Radiation

+ I

6

SPACE

Terrestrial Radiation

ATMOSPHERE

a Clouds

He01 N , S O , , C O , , O , ,

O~C.

Aerosols A i r h e Coupling

Precipitation. Evaporation

Air/biomass/ land Coupling

AtmospherdOcean Couph

FIG. 1. Schematic illustration of the atmosphere/ocean/ice/land/biomassclimatic system, with some examplesof physical processes responsible for climate and climatic change. [From Gates (1979).]

The object of this article is to formulate and describe the current issues attendingthe physical method in the study of possible C0,-induced climatic change. To this end an elaboration of mathematical climate models is given in the next section. This is followed in Section 3 by a comparison of the results of these models for the climatic change resulting from increased levels of atmospheric CO, . In Section 4 the issues that are raised by that comparison are described, and Section 5 contains recommendations for resolving those issues. 2. MATHEMATICAL CLIMATE MODELS Several types of mathematical climate model have been developed that differ in the comprehensiveness of their treatment of the climate system components. Individual models of the climate model hierarchy can be classifiedbroadly as either thermodynamic or hydrodynamic models. Thermodynamic climate models explicitly predict temperature but either ignore the motion field and its influence on the temperature, or incorporate that influence in a highly simplified and approximate way. Hydrodynamic cli-

144

MICHAEL E. SCHLESINGER

mate models explicitly predict both the temperature and the motion fields and their mutual interactions. Hydrodynamic climate models therefore allow conversion between the two forms of energy- the total potential energy, which is proportional to the temperature, and the kinetic energy, which is proportional to the square of the velocity-while thermodynamic climate models do not. In the following discussions we describe the characteristics of two thermodynamic climate models, the energy balance and radiative- convective models (EBMs and RCMs), and one hydrodynamic model, the general circulation model (GCM).

2.1. Energy Balance Models (EBMs) In their simplest "zero-dimensional" formulation EBMs determine the effective radiating temperature of the planet, Tp,from the radiative equilibrium condition that the infrared radiation emitted by the earth to space, ep~T;47ra2,equals the solar radiation absorbed by the earth, S( 1 - ol,)na2. Here aT: is the infrared radiation per unit area that would be emitted by the planet if it were a blackbody radiator, with Q the Stefan-Boltzmann constant, 4naZ the surface area of the planet with radius a, eP the effective emissivity of the planet, S the solar radiation per unit area at the top of the atmosphere (insolation), .a2 the cross-sectional area of the planet, and (1 or,) the planetary absorptivity, with crp the albedo (reflectivity). If the earth emitted as a blackbody, ePwould be unity and Tp= - 18.6"C, which is considerably colder than the observed global-mean surface air temperature of 14.2"C. This warmer temperature is the result of the effects of the infrared-absorbing gases of the atmosphere (and clouds), i.e., the greenhouse effect, which yields eP 0.6. The value of ePthus depends on the composition of the atmosphere and must be determined by EBMs. The planetary albedo is frequently made a function of the size of the polar icecap through a prescribed dependence of a, on Tp. In one-dimensional EBMs the north south (meridional)distribution of temperature is determined by including a simplified, semiempiricalformulation for the effective meridional transport of heat by the motion fields of the atmosphere and ocean; however, as noted earlier, neither of these motion fields is determined by EBMs. [For further information, see the review article by North et al. ( 198I).] 5

2.2. Radiative- Convective Models (RCMs) Radiative- convective models determine the equilibrium vertical temperature distribution for an atmospheric column and its underlying surface for given insolation and prescribed atmospheric composition and surface al-

MODELS OF C02-INDUCED CLIMATIC CHANGE

145

bedo. An RCM includes submodels for the transfer of solar and terrestrial (infrared)radiation, the turbulent heat transfer between the earth's surface and atmosphere,the vertical redistributionof heat within the atmosphereby dry or moist convection, and the atmospheric water vapor content and clouds. The radiative transfer models used in RCMs are frequently identical to those used in GCMs. The surface heat exchange is treated either as an equivalent radiative exchange, or is parameterized] as a Newtonian exchange with a prescribed transfer coefficient. The vertical heat redistribution by convective atmospheric motions is modeled as an adjustment whereby the temperature lapse rate of the atmosphere is prevented from exceeding some given value. The amount of water vapor is determined in RCMs either by prescribing the absolute humidity or the relative humidity; in the latter case the amount of water vapor increases (decreases) with increasing (decreasing) temperature. Finally, the fractional cloudiness and the temperature or altitude of the cloud tops are prescribed [see the review article by Ramanathan and Coakley ( 1 978)] or predicted (Wang et al., 198 1 ; Hummel and Kuhn, 1981; Charlock, 1982).

2.3. General Circulation Models (GCMs) The principal prognostic variables2of an atmospheric GCM are the temperature, horizontal velocity, and surface pressure, which are governed, respectively, by the thermodynamic energy equation, the horizontal momentum equation, and the surface pressure tendency equation. With the mass continuity equation and the hydrostatic approximation and appropriate boundary conditions, these equations form a closed system for an adiabatic and frictionlessatmosphere. But the general circulation of the atmosphere is the large-scale, thermally driven field of motion in which there are interactions between the heating and motion fields. Therefore, several additional prognostic variables, with corresponding governing equations and appropriate boundary conditions, must be added to simulate the heating. Of these, the most important is the water vapor, which is governed by the water vapor continuity equation. Because the atmosphere is largely heated by the underlying surface through the exchange of sensible and latent heat, and because snow lying on the ground can have a large influence on the surface albedo, the ground temperature, soil moisture, and mass of snow on the ground are prognostic variables, governed by energy, water, and snow The treatment of physical processeswhose characteristicsize is smallerthan the smallest size resolved by a model. Variables whose time rates of change of magnitude are determined by the governing equations.

146

MICHAEL E. SCHLESINGER

TABLE I. THEPRINCIPAL PROGNOSTIC AND DIAGNOSTIC VARIABLES IN ATMOSPHERIC GENERAL CIRCULATION MODELS Prognostic variables

Diagnostic variables

Surface pressure Temperature Horizontal velocity Water vapor concentration Soil temperature Soil moisture Snow mass

Vertical velocity Geopotential height Density Cloudsb Surface albedob

a In spectral models (see text) the vertical component of vorticity and the horizontal divergence replace the horizontal velocity as the prognostic variables, and the latter are determined diagnostically from the former. These quantities are prescribed in some models.

budget equations for the ground. In addition to the prognostic variables, GCMs have many diagnostic variablesY3 among which clouds may be one of the most important. A summary of the prognostic and diagnosticvariables in GCMs is given in Table I. The governing equations of GCMs are nonlinear, partial differential equations whose solution cannot be obtained except by numerical mathematical methods on the fastest computers. These numerical methods subdivide the atmosphere vertically into discretelayerswherein the variables are “carried” and computed (Fig. 2). For each layer the horizontal variations of the predicted quantitiesare determined either at discretegrid points over the earth, as in the grid point (’nite diference) models (Fig. 3), or by a finite number of prescribed mathematical functions, as in the spectral models. The values of the predicted variables for each layer (including the surface) and grid point (or mathematical function) are determined from the governing equations by “marching” (integrating) forward in time in discrete steps starting from some given initial conditions(Fig. 4). To prevent the solution from becoming numerically unstable the time step must be made smaller than a value that depends on the speed of the fastest moving disturbance (wave), the size of the grid (or smallest scale mathematical function),and the integration method. The spatial resolution of GCMs is constrained for practical reasons by the speed and memory capacity of the computer used to perform the numerical Variables whose magnitudes are determined by the governing equations.

MODELS OF C02-INDUCED CLIMATIC CHANGE

147

FIG.2. The vertical structure and principal variables of the Oregon State University (OSU) two-level atmospheric general circulation model. Here u = (p - pT)/(p,- pT), where p is pressure, pTthe (constant) pressure at the model top, andp, the (variable) pressure at the earth’s surface; u and v are the eastward and northward velocity components, Tthe temperature, CD the geopotential, q the water vapor mixing ratio, Sand R the solar and terrestrial radiation at the top of the model atmosphere (subscript 0)or at the earth’s surface (subscript s), apand cu, the planetary and surface albedos, Q and Fthe diabatic heating and friction, H, the surface sensible heat flux,Pthe precipitation rate, E, the surface evaporation rate, and GWthe ground wetness. CL,-CL, denote the model’s cloud types. [From Schlesinger and Gates (1979).]

integrations. Increasing the resolution not only increases the memory required (linearly for vertical resolution and quadratically for horizontal resolution), but also generally requires a reduction in the integration time step. Consequently, the computer time required increases rapidly (nonlinearly) with increasing resolution. Contemporary GCMs have from two to about nine vertical layers, a horizontal resolution of a few hundred kilometers, and a time step ranging from 10 to 40 min. These models require from one-half

148

MICHAEL E. SCHLESINGER

to a few minutes to simulate 1 day on a fifth-generation computer such as the Cray 1 and CYBER 205. Due to their limited spatial resolution GCMs do not resolve several physical processes of importance to climate. However, the effects of these subgrid-scale processes on the scales resolved by the GCM are incorporated in MATSUNO

1

L E A P F R O G ___)

FIG. 4. Sequence of time steps in the time integration of equations of the form ay/at = D ( y ) S(y)in the OSU atmospheric GCM. Here y is any prognostic variable; dy/at the time rate of change of' y at a fixed location; S(y) the source term, e.g., the heating term in the thermodynamic energy equation; and D(y) all other terms, including the transport by the motion field. The circumflex (-) refers to the predictor estimate of the Matsuno integration scheme, the single prime (') refers to the corrector estimate of the Matsuno scheme and to the leapfrog scheme estimate, and the tilde (-) refers to the estimate before the source terms are added. The time step At,, = 6A2,and At = 10 min. [After Ghan el al. (1982).]

+

MODELS OF CO2-INDUCED CLIMATIC CHANGE

149

TABLE 11. SUBGRID-SCALE PROCESSES THATARE PARAMETERIZED IN ATMOSPHERIC GENERAL CIRCULATION MODELS Turbulent transfer of heat, moisture, and momentum between the earth’s surface and the atmosphere Turbulent transfer of heat, moisture, and momentum within the atmosphere by dry and moist (cumulus) convection Condensation of water vapor Transfer of solar and terrestrial radiation Formation of clouds and their radiative interaction Formation and dissipation of snow Soil heat and moisture physics

the model by relating them to the resolved-scale variables themselves. Such a relation is called a parameterization, and is based on both observational and theoretical studies. The subgrid-scaleprocesses that are parameterized are shown in Table 11. To simulate climate and climate change with an atmospheric GCM it is necessary to prescribe certain parameters and boundary conditions as shown in Table 111. For the earth it is also necessary to include the ocean and ice components of the climate system (Fig. 1). How this is done depends on whether the purpose of the simulation is to test (validate) the atmospheric GCM or to simulate a climate change. To validate an atmospheric GCM it is possible to treat the sea surface temperature (SST) and sea ice thickness as given boundary conditions rather than as prognostic variables of the climate system. Then, since it is the ability of the GCMs to simulate climate change that is of interest, and since the seasons are the best-documented climate changes, the seasonal performance of the models can be evaluated from a simulation in which the SST and sea ice distributions are taken as equal to their observed values. This has been done most frequentlyby simulatingsingle winter and summer months, TABLE 111. THEPRESCRIBED PARAMETERS AND BOUNDARYCONDITIONS IN ATMOSPHERIC GENERAL CIRCULATION MODELS Radius, surface gravity, and rotation speed of the planet Solar constant and orbital parameters of the planet Total atmospheric mass and composition Thermodynamic and radiation constants of the atmospheric gases and clouds Surface albedo Surface elevation

150

MICHAEL E. SCHLESINGER

usually January and July, and comparing the simulated atmospheric variables with their observed counterparts. [See, for example, Global Atmospheric Research Programme (1979), Schlesinger and Gates (1979), Shukla et al. (1 98 I), and Pitcher et al. (1983).] The seasonal performance of several models has also been determined from extendedintegrations over more than one annual cycle wherein the SST and sea ice distributions are prescribed to repeat their observed annual cycles (Manabe and Hahn, 1981; Schlesinger and Gates, 1981b; Hansen et al., 1983). The ability ofan atmospheric GCM to simulate a climate different from that of the present has also been evaluated by prescribing the SST and sea ice distributions equal to those that have been reconstructed for the most recent ice age (the Wisconsin glacial period 18,000yr ago), along with the different geography of land, ocean, and continental ice sheets, and comparing the simulated surface air temperature with the temperature field reconstructed from fossil pollen and other periglacial evidence (Gates, 1976a,b). These validation studies show that atmospheric GCMs are capable of simulating many of the characteristic differences between the present summer and winter climates, and between the present interglacial climate and its glacial antecedent. To simulate a climate change such as that which may be induced by elevated CO, levels, however, it is not desirable to treat the sea surface temperature and sea ice distributions as given boundary conditions. To do so would, as is illustrated later, proscribe the response of the climate system by preventing the interaction and feedback among the atmosphere, ocean, and sea ice components. Consequently, atmospheric GCMs have been coupled with different ocean and sea ice models, and these models also form a model hierarchy wherein the individual models may be classified as either thermodynamic or hydrodynamic models. The simplest thermodynamic ocean/sea ice model is one in which both the heat storage and the heat transport ofthe ocean are ignored and the SST is diagnostically determined such that the net energy exchange at the air - sea interface is zero (Manabe, 1969a; Manabe and Wetherald, 1975). This model is called a swamp model because of its similarity to perpetually wet land. In a swamp model the existence of sea ice is predicted whenever the SST is below the temperature at which sea water normally freezes. It is not desirable to include either the diurnal or the annual solar cycles in a simulation with a coupled atmospheric GCM/swamp ocean model because the absence of oceanic heat storage would result in the freezing of the ocean in the nighttime hemisphere and in the region of the polar night. To permit simulations of climate and climate change with the annual and diurnal cycles included, atmospheric GCMs have been coupled to slab models of the uppermost layer of the ocean, the oceanic mixed layer, wherein the temperature is relatively uniform with depth. In slab models a fixed

MODELS OF COyINDUCED CLIMATIC CHANGE

151

depth of the mixed layer is prescribed such that the simulatedannual cycle of SST, and therefore of the heat storage, is in close agreement with the observed annual cycle (Manabe and Stouffer, 1980). In this type of thermodynamic model the sea ice thickness is predicted based on a thermodynamic energy budget that includes the accumulation of snowfall, sea water freezing, and ice melting and sublimation. Because the fixed depth of the slab model is chosen to reproduce a periodic climate change (the annual cycle), while it is likely that a climate change induced by rising COz (and other forcing mechanisms) will be secular, it is desirable to have models of the mixed layer in which the depth is a prognostic variable. Such a variable-depth mixed-layer model has been coupled to an atmospheric GCM and tested in a 16-month simulation (Pollard, 1982a). In this model the horizontal heat transport as well as the heat storage is included through the hydrodynamic prediction of the mixed-layer currents, while the sea ice thickness is determined thermodynamically as described above. An extension of this hydrodynamic ocean model has been made in which the depth, temperature, and currents of the seasonal thermocline4are predicted, as well as the corresponding quantities for the mixed layer (Pollard, 1982a). Although the hydrodynamical models of the upper ocean include the storage and horizontal transport of heat, they do not include the vertical transport of heat associated with the large-scale upwelling and downwelling of water. However, this vertical heat transport is of particular importance in the heat budgets of the equatorial and polar seas. For this reason models of the general circulation of the ocean have been developed that are the dynamical counterparts to the atmospheric GCMs. In oceanic GCMs the prognostic variables are the temperature, horizontal currents, and salinity, and the diagnostic variables include density, pressure, and the vertical velocity. There are subgrid-scale processes that must be parameterized in ocean GCMs, including the turbulent transfers of heat, momentum, and salt in both vertical and horizontal directions, and parameters and boundary conditions similar to those in Table I1 that must be prescribed. The solution of the governing equations is obtained numerically in a manner similar to that used for atmosphericGCMs. The ocean is subdividedvertically into layers and horizontally into grid boxes, and the predicted quantities are determined as a function of time by numerical integration (Fig. 4). Oceanic GCMs have been tested by prescribing either the air - sea fluxes of heat, momentum, and water, or the state of the atmosphere (World Meteorological Organization, 1977). Oceanic GCMs have also been coupled with atmospheric GCMs, The layer immediately beneath the mixed layer wherein the temperature decreases with depth.

152

MICHAEL E. SCHLESINGER

and simulations have been performed to validate these coupled (or joint) GCMs (Manabe, 1969b; Bryan, 1969;Wetherald and Manabe, 1972; Manabe et al., 1975; Bryan et al., 1975; Manabe et al., 1979; Washington et al., 1980). In all but one of these joint GCMs (Washington et al., 1980) the transport of sea ice by the upper ocean currents was included in the thermodynamic sea ice model, but the life cycle of ice leads5and their effects were uniformly ignored. These have been predicted in a dynamic sea ice model (Hibler, 1979), but not yet in association with a coupled GCM. To validate a coupled atmosphere/ocean/sea ice GCM it is possible to treat the remaining components of the climate system- the biomass, land ice, and land (Fig. 1)-as given boundary conditions in a manner similar to that used in the validation of atmospheric GCMs alone. Whether any one of these climate system components can also be regarded as known in the simulation of climate change with the coupled GCMs depends on its characteristic time scale relative to that for the climate change. Only in the case when the time scale for the climate change is much less than that for the climate system component can that component be regarded as constant. For example, it is reasonable to assumethat the geography of the land, ocean, and land ice will not change sufficiently in the next 50 yr due to increased CO, to warrant its inclusion in coupled GCMs as a predicted quantity. That is, it is not likely that the West Antarctic ice sheet will surge in the next 50 yr, which would result in a rise in sea level and flooding of low-lying coastal regions. However, a similar assumption regarding the biomass component seems unjustified. Ironically, several models have been developed for land ice (for example, Sergin, 1980; Pollard, 1982b), albeit none has yet been included in a coupled GCM, but very little attention has been devoted toward the development of a completelyinteractive biomass model wherein, for example, old forests perish and new ones grow in response to a changing climate.

3. COMPARISON OF MODELSIMULATIONS OF C 0 2 - I ~ ~ ~ ~ ~ CLIMATIC CHANGE The climate change induced by increased atmospheric carbon dioxide has been simulated by models of each of the types described earlier. Reviews of these simulationshave been given by Schneider (1973, the National Academy of Sciences (1979, 1982), Watts (1980), Gates (1980a,b), and Kellogg and Schware ( 1981). The principal simulationsare summarized in Table IV in terms of the models’ basic characteristics. As described below, two types Open water lanes within the sea ice pack.

TABLE IV. (Continued) Type of simulation

-

u, P

Equilibrium quadrupling (4x1

Energy balance models (EBMs) Idso (1980) [23] (S,N,N), (I,N,N), (A,T,A), <0.3"C Kande1(1981)[27] (S,N,N), (N,N,N), (G,E,A), 0.7 to 8.8'C SchneiderandThompson (1981) (P,G,N), O,N,D), (RE,A); lag time varies with latitude with minimum in tropics Raswl andSchneider(l97I) [4] (P,N,N),(N,N,N),(B,H,A), 1.1 to 1.3'C Weare and SneU (1974) [7] (P,N,N), (N,N,N), (R,P,A), 1.1 "C Tempkin and Snell(1976) [I I] (PB,N), (I,N,N), (R,P.A), 2.6"C

Equilibrium decupling

Rasool and Schneider(l971) [4] (P,N,N), (N,N,N), (R,H,A), 2.5-C

Nonequilibrium

Robock (1978) (P,GA". (I,N,F), (R,E,S) Thompson and Schneider ( I 979) (P,G,N), (I,N,D), (E,E,A); lag time of 5-20 yr depends on depth of ocean's response and surface-deep water mixing rate

(lox)

Radiative-convective models (RCMs)

General circulation models (GCMs)

Augustrson and Ramanathan (1977) 1121 W,NM, ("9(R,H.A), 4.4'C

Bryan ef al. (1982) W,G,S), (I,N,O), (P,H,A); 10- to 25-yr lag time, a function of latitude for firsI 10 yr with more rapid adjustment in tropics

Hoffert d 01. ( 1980) (P,N,N), (I,N,D), (R,H,A); lag time of 10-20 yr depends on prescribed ratio of temperature changes in polar sea to surface mixed layer Cess and Goldenberg(1 98 1) (P.N,N),(I,N,D),(E,E,A), lO-to2O-vlastime Schneider and Thompson (198 1) (P,G,N). (I,N,D), (E,E,A); lag time depends on latitude Michael ef a/. (1981) (P,N,N), (I,N,D),(R,H,A); lag time about 4Oyr

The models’ characteristics are shown by three sets of three symbolseach. The I hset shows the domain (vertical,latitude, longitude). The second set shows (land/ocean distribution, topopphy. ocean temperature model). The third set show (humidity model, cloud model, insolation). The symbols for each of these nine characteristicsare defined below. The simulated surface air

2

temperatw change averaged over each model’s horizontal domain (unless otherwise noted) and/or the Simulated lag time (time raluired for temperature change to reach the equilibrium temperature change) are alsn given. The numbers [ I ] through 1.341 following the citation are shown in Fa 20 for reference. Vertical: S, surface energy balance; P,planetary energy balance; T,troposphere; U, troposphere and upper atmosphere. latitude: N, none; H, equator to pole; G, pole to pole. Longitude: N, none; Z, zonally averaged; S, 120” sector, G, 360”. Land/disuibution: N, none; I, idealized, R, realistic (for given resolution). Topography: N, none; 1, idealized, R, realistic (for given resolution). Ocean tempeXaNre model: N, none; G,prescribed sea surface temperature; S, swamp; SO, swamp with 6xed sea ice; F,mixed layer with @kd depth; D,F with deeper owan; P,mixed layer with predicted depth; 0,oceanic general circulation model. Humidity model:A, hxed absolute humidity; R, 6xed relative humidity; B,both 6xed absolute and relative humidiv, E,empiricallybuilt into longwave parameterization;G, humidity change prescribed; P, predicted. Cloud model: N, no cloud; F, fixed cloud; 8, both 6xed and no cloud; H, fixed cloud “height”; T, 6xed cloud temperature; C, both fmed cloud “height” and temperature; E,empiricallybuilt into longwave radiation parametetization; P, predicted. YHeight” is b e d only with altitude as the vertical coordinate. With pressure ( p ) or normalized pressure (a)as the vertical c ~ o ~ d i ~the t eactual , altitude for 6xed p or u varies with temperature followingthe hydrostaticequation.] Insolation: A, annual mean; S, seasonally varying;J, January.

156

MICHAEL E. SCHLESINGER

of study have been performed with these models, an equilibriumstudy and a nonequilibrium study. In an equilibrium study a control simulation is made with a fixed CO, concentration (typicallyabout 300 ppmv), and an experiment simulation is made with another fixed CO, concentration. Both the control and the experiment are run sufficiently long to achieve their respective equilibrium climates as illustrated in Fig. 5. The object of most equilibrium studiesis to determine what the change in the climate would be if the CO, concentration increased to some constant value and the climate system reached a new equilibrium for that higher CO, level; however, the time required to attain that new equilibrium has not been ascertained from the equilibrium simulations. As shown in Table IV, equilibrium simulationshave been performed for elevated experiment COzconcentrationsthat are double (2X), quadruple (4X), and 10 times (IOX) their control values. The 2X simulations have been made because, as noted in Section 1, it is projected that the CO, level will reach twice the preindustrial value sometime in the next century. The 4X and 1OX simulationswith GCMs have been made not because such large increases are foreseen, but rather to increase the statistical significanceof the results. In GCM simulations,as in nature, there is an inherent variability for any I

I

I

I

I

1

I

-

a:

; W

-

0.-

I-

-5

-

-10

-

--------.--------MASS-AVERAGED TEMPERATURE

21c02 IXCOp

Averaging Period

-

MODELS OF CO2-INDUCED CLIMATIC CHANGE

157

time-averaged quantity, even though the climate is unchanged. This natural variability constitutes noise against which the signal of the difference between the experiment and control must be contrasted. If the ratio of the signal to the noise is sufficientlylarge, then it can be said that the experiment climate is not just another realization (sample) of the control climate, but is in fact a climate different from that of the control. To increase the signal-tonoise ratio the noise can be reduced by averaging over longer time periodsthis requires extending the length of the simulations. Alternatively, the magnitude of the signal call be increased by increasing the magnitude of the forcing, e.g., by quadrupling the C02 level. This approach, which may be called the superanomaly method, has been used previously to study the influence of sea surface temperature anomalies on climate (e.g., Chervin, 1979). Its validity depends on whether there is a known transferfunction from the superanomaly response to the anomaly response. In the case of increased levels of CO,, Augustsson and Ramanathan ( 1977)have shown by an RCM calculation that the contribution to the surface air temperature warming by the CO, 15-pm absorptanceband increaseslogarithmicallywith increasing CO, , while that due to the weaker C02bands increases linearly. Their results show that the combined effect of the 15-pm and weak bands is such that the warming due to quadruplingis about 2.4 times greater than the warming due to doubling. Based on this it has been assumed that the thermal response for a CO, doubling could be taken as approximately half that obtained from a superanomaly CO, quadruplingexperiment. We shall reconsider this point subsequently. In a nonequilibrium study the C02concentration in the experiment either is changed at the initial time and held constant or is made to increase with time. (See Fig. 6 for an example of the latter.) The response of the climate system in the experiment at any time is compared with the equilibrium response for the C02 concentration at the same time, the equilibrium response having been determined from a control equilibrium simulation. The object of the nonequilibrium study is to determine the lag in the response of the climate system, that is, the difference between the times when the nonequilibrium and equilibrium responses reach any specified value. Knowledge of the lag time is required to estimate when a C0,-induced climate change may become detectable. As shown in Fig. 6, the lag time depends critically on the rate of heat exchange between the upper ocean (mixed layer) and the intermediate ocean (thermocline), with the lag time increasing with the rate of heat exchange. Because the value of the heat exchange rate is poorly known from observations, the time of first detectability is uncertain by about 10 yr, as shown in Fig. 6 (Schlesinger, 1983a). In this article attention is focused on comparing the simulations from the three-dimensional GCMs because it is the geographical distribution of a

*

158

MICHAEL E. SCHLESINGER

Ea!

c

e a2

E

a2 +

1.5 -

c .0)

c CI, .0 c V

E

.-0 e A

c-

gg c z

801

0 V

0.5

-

3zzzzzlo I860

1900

I950

2000

2050

250 I860

I900

1950

2000

2050

2100

Year FIG.6. The change in surface air temperature (top) induced by the change in C02concentration (bottom). A,, is the coefficient of heat exchange between the mixed layer and deep ocean (W m-* K-I). [From Schlesinger (1983a).]

possible C0,-induced climatic change that is of importance to humanity. Moreover, since only one nonequilibrium study has been performed with a GCM (see Table IV), the comparison can be made only for the equilibrium GCM studies. However, the global-mean surface temperature changes from these GCM studies will be compared with those from the EBMs and RCMs. Before presenting the comparisons, a brief overview of the characteristics of the GCMs and CO, simulations is given.

3.1, Description of GCMs and CO,Simulations Simulations of the climate changes induced by elevated CO, levels have been carried out with GCMs at the Geophysical Fluid Dynamics Laboratory/NOAA, Princeton University, New Jersey (hereafter abbreviated as GFDL);the NASA Goddard Space Flight Center Institute for Space Studies, New York, New York (GISS);the Lawrence Livermore National Laboratory, Livermore, California (LLNL); the National Center for Atmospheric Research, Boulder, Colorado (NCAR); the Climatic Research Institute, Or-

MODELS OF C0,-INDUCED CLIMATIC CHANGE

159

egon State University, Corvallis, Oregon (OSU); the United Kingdom Meteorological Office, Bracknell, Berkshire, United Kingdom (UKMO); and the Computing Center of the USSR Academy of Sciences, Moscow (USSR). These simulations are referenced in Table IV by author(s) and are listed in chronological order. Not all of the simulations listed in Table IV (and shown in Fig. 2 1) are included in the present comparison: the LLNL simulations (Potter, 1978, 1980) are not included because the model is only two dimensional (latitude and height); the GISS simulations [J. Hansen, 1978, 1979, as reported by the National Academy of Sciences (1979)l and the USSR simulations (Aleksandrov et al., 1983) because a description of the model and/or results were not available; and the UKMO simulations (Mitchell, 1983) either because of the similarity of their prescribed-SST results to those of the OSU simulation (Gates et al., 198 1) or because of the ad hoc 2°C increase in SST that was prescribed in the second UKMO doubled-CO, simulation. The simulationsthat are included in the present comparison are listed in Table V along with some characteristicsthat will be discussed subsequently. Some of the characteristicsof the GCMs used to perform the CO, simulations are presented in Table VI. The principal differences among these GCMs are in their domain, land/ocean geography, orography, ocean/sea ice model, clouds, and the albedo of snow and sea ice. 3.I . 1. Domain, LandlOcean Geography, Orography. The first GCM simulation of C0,-induced climate change was performed at GFDL by Manabe and Wetherald (1975). In that simulation the horizontal domain extended over only 120" of longitude and from the equator to 81.7"N,as shown in Fig. 7. In this sector model cyclical boundary conditions were imposed at 0 and 120" longitude, and an idealized land/ocean geography was prescribed along with no topography, that is, the surface elevation was uniform everywhere. As shown in Fig. 7, slightly different sector models were employed by Manabe and Wetherald (1980) and Wetherald and Manabe ( 1 98 1 ). In all other simulationslisted in Table VI the horizontal domain was global,that is, 360" of longitude and extendingfrom pole to pole, and the landlocean geography and orography were realistic within the models' horizontal resolution. (SeeFigs. 10, 1 1, and 18 for the OSU, NCAR, and GFDL models, respectively.) All the simulationsexcept those carried out at OSU were performed with nine-layer models that included both the troposphere and the stratosphere. The OSU simulationsemployed a two-layer model of only the troposphere (see Fig. 2).

3.1.2. OceanlSeaIceModel. In the simulationsperformed with the OSU model by Gates et al. (1981) both the sea surface temperature and sea ice

160

MICHAEL E. SCHLESINGER

TABLEV. EQUILIBRIUM GCM SIMULATIONS OF CLIMATIC CHANGEINDUCED BY C02 DOUBLING AND QUADRUPLING Model GFDL Manabe and Wetherald (1975) Manabe and Wetherald (1980) Manabe and Stouffer ( 1980); Manabe et al. (1981) Wetherald and Manabe (1981) NCAR Washington and Meehl (1983b)

osu

Gates et al. ( 198 1) Schlesinger (1983b)

2 X C02

4 X C02

Annual cycle

Length of simulation (yr)

Averaging period of results (yr)

Yes

No

No

2.2

0.3

Yes

Yes

No

3.3

1.4

No

Yes

Yes

No

Yes

Yes

4

No

7 (1.1) 5’ 8 ( 1.6) 6’ 8(1.2)1lU 9(1.7) 1 1 26(1.1) 1 . 1 1.6, 1.8b

1

Yes

Yes

No

Yes Yes

Yes No

Yes No

1.3 2.0

3

0.5

1 0.5‘

~~

In these simulations the annual cycle of the atmospheric GCM is first accelerated with respect to the annual cycle of the ocean model. The ocean model is integrated through the number of annual cycles shown by the first number in the row. The atmospheric model is integrated through the same number of annual cycles; however, since these annual cycles are accelerated, the amount of time simulated is less, as shown by the second number in the row (in parentheses). The accelerationofthe atmospheric annual cycle is gradually reduced until there is no acceleration. The models are then synchronouslyrun for the number of years shown by the third number in the row. The first row is for the control, the second row for the experiment when different from the control. For the simulations with fixed clouds and computed clouds, respectively. Unless otherwise noted.

distributionswere prescribed to vary according to their observed present-day annual cycles. In these simulations, then, neither the SST nor the sea ice could respond and feed back on the changed atmospheric climate. The only other simulations that included the annual cycle of solar insolation were those reported by Manabe and Stouffer (1 980) and Wetherald and Manabe ( 1981) [see Table V under heading “Annual cycle”; note that the papers by Manabe and Stouffer (1980) and Manabe et al. (198 1) report different aspects of the same simulation]. These simulations were performed with the GFDL atmospheric GCM coupled with a 68-m slab model ofthe oceanic mixed layer and with the thermodynamic sea ice model noted in Section 2.3. All the remaining simulations with the models listed in Table VI employed the swamp ocean/sea ice model and, therefore, were made with only annually averaged solar insolation (Table V).

MODELS OF CO2-INDUCED CLIMATIC CHANGE

161

-00

90-N

FIG.7. Horizontal domain and land/ocean distribution of the sector models of Manabe and Wetherald (1975) (top), Manabe and Wetherald (1980) (middle), and Wetherald and Manabe (1981) (bottom).

3.1.3. Clouds. As already noted in Table I, clouds are diagnostic variables in some atmospheric GCMs and are prescribed in others. In three of the four GFDL models shown in Table VI clouds are prescribed and, therefore, cannot respond and feed back on the C0,-induced climate change. Clouds are predicted variables in the simulations with both the OSU and NCAR

TABLE VI.

CHARACTERISTICS OF GENERAL CrRCULATlON

MODELS USED TO SIMULATE C02-INDUCED

CLIMATIC CHANCES

~

~

Domam Model

vatlcal (z)

GFDL

e

N

Manabe and Wetherald (1975) Manah and Wetherald (1980) Manabe and stouffcr (1980); Manabe er al. (1981) Wetherald and Manabe (1981) NCAR Washington and Meehl (1983b)

osu

Gates er a/. (1981)

Schlesinger (1983b)

Surface t o p = 0 (9 layen)

Longitude(A)

Landlocean distribution

F.quator to 81.7' (grid point, A@ = variable)

120" sector (grid point, M = 6')

Idealized land/azan

Equator to pole (grid point,

120' sector (grid point, M = 5 ' )

Latitude (CP)

A@

Topography None

Pole to pole @ P e c w 15

Idealired land/-

None

Realistic

120' sector (spccrml, waves 3,6.9, 12, 15)

Idealized Land/azan

None

360" (spectral, I5 waves)

Realistic

Realistic

360' (grid point,

Realistic

Reatistic

Pole to pole (grid point,

Surface to p = 200 mbar (2 layen)

Pole to pole (gnd point,

A@

A@

M=5')

Realistic

= 4")

occumrra

Predicted

68-m mixed layer (no heat -sport)

ThiChesS

Rediacd

69-m mixed layer (no heat -sport)

ThiChCSS

Swamp (no heat opacity or transport)

Occurrence predicted when To = 1.8"C

Predicted

Toannual cycle

Sea ice annual

Predicted (Ghan el al.. 1982)

Swamp (no heat cawcity or

PmiM

= 4')

360' (grid point, M =5')

Realistic

Realistic

Sail moisture and snow

Predicted (Manah, 1969a)

transport)

360' (spectral, 15 waves and 2 I waves)

sea Ice.

Oaxurence predicted when T,---Z'C

Swamp (no heat -port)

waves) Surface t o p = 200 mbar (2 layen)

ocean'

capacity or

= 4.5')

Pole to pole (spectral, 15 waves and 2 1 Wave)

Surface to p = 0 (9 layen)

Surface treatment

Swamp (no heat capacity or transport)

predicted when To= -2°C

predicted (Bryan.1969); TD--2'C when sea ice exists

predid (Bryan, 1969); T,=-2'C when sea ice exists

-

cycle pmribed Occurrence predicted when To= - 1.6-C

(Manabc, 19696

(Manah, 1969a)

Predicted ( M a ~ hI969a) ,

Washington and Williamson, 1977)

Predicted (Ghan eral.. 1982)

Radiation Solar radiation GFDL Manabe and Wetherald (1975)

Annual mean; H,O,CO, , 0,. clouds, Rayleigh

scatming(Manaheand Manabe and Wetherald (1980) Manate and Stouffer (1980) Manate ef a/. (1981) Wetherald and Manabe (1981)

(1983h)

osu

Gatesetal. (1981)

Schlesinger( I 983b)

.

Wetherald, 1967) Annual mean; H,O, CO,, 0,. clouds,Raylei& scattering (Manah? and Wetherald,1967) Annual cycle, no diurnal cycle; H,O, CQ,, 03, clouds, Rayleigh scattering (Lacis and Hansen, 1974) Annual mean and seasonal Variation runs,H,O, CQ, 0 3 , clouds, Rayleigtl scattering (Lacis and Hansen, 1974)

Terrestrial radiation

H,O,CO,, O,, clouds

(Manabe and Wetherald, 1967)

H,O,

m,

O,, clouds (StoneandManah?, 1968)

HIO,CO,,O,, clouds (StoneandManah?, 1968)

surface albedocd

cloud@

RMibed annual mean

fnct (z,0)

Land and ocean premibed(Manate,1969a);snow and sea ice = 0.70 (T, <-25"C), -0.45 and 0.35 (T,>-25'C)

Predicted when condensation

Land and ocean prescribed (Manabe, 1969a); snow and sea ice = 0.70 (T, < - IO'C), -0.45 and 0.35 (T,2 - 10°C)

Rescribedannual mean fnct (z, 0 )

Land and ocean prescribed (Poseyand Clapp, 1964); snow s 0.80, hct (a,snow mass, underlyingsurface); sea ice S 0.70, fnct (0,ice thickness, melting)

-

Land and ocean &bed (Manate,1969a);snow and sea ice 0.70 (T, < IO"C),-0.60 (T, 2 IO'C), -0.45 for melting sea ice

-

-

Annual mean; H20, CO,, O,, 01,clouds, Rayleigh scattering (Ramanathan ef al. 1983)

H,O, CQ, 0,.clouds

Annual and diurnal cycle; H,O,O,, Rayleigh scattering (Ghan efa/., 1982) Annual mean; H,O,O,, Ftayleigb scattering (Ghan ef a/.. 1982)

H,O, CO,, clouds (Ghan ef al.. 1982)

Predictedwhen cumulus convection or

Land and ocean premibed (Posey and Clapp, 1964); snow s 0.80, fnct (snow mass, underlying surface); sea ice 0.45

HIO,CO,, clouds (Ghanef

Predictedwhen cumulus

Land and ocean &ted (Posey and Clapp, 1964); snow s 0.70, fnct (snow mass, underlying surface); sea ice = 0.35

(Ramanathana a/. 1983)

a/., 1982)

Rediaed when

RH > 80%

(convective when re< 0, nonconvective when r, a 0);also run with prnrribed clouds

RHa 100%

convection or

),0.55 (0.9-4pm); Land = 0.13; desert = 0.25; snow = 0.80 (0-0.9~ sea ice = 0.70

-

RHa 100%

T,, Ocean temperature. RH, Relative humidity; re,vertical gradient of equivalent potential temperature. Tv Temperature of earth'ssurface. The surface albedos for sea ice are for the condition of zero snow cover. The surface albedos for snowcovered sea ice are as shown for snow.

164

MICHAEL E. SCHLESINGER

models; however, doubled- and quadrupled-CO, experiments with fixed clouds as well as computed clouds were conducted with the NCAR model to assess the importance of clouds in the simulated climate change. 3.1.4. Albedo of Snow and Sea Ice. The albedo of snow and sea ice in GCMs may be of particular importance in the simulation of C0,-induced warming in high latitudes. As shown in Table VI, the albedo of snow is a discontinuous function of the surface temperature T, in the simulations reported by Manabe and Wetherald (1975), Manabe and Wetherald (1 980), and Wetherald and Manabe ( 198l), with a value of 0.70 for T, < -25, - 10, and - 10°C,respectively, and 0.45,0.45, and 0.60 for T, 2 -25, - 10, and - 10°C,respectively. On the other hand, in the simulationsby Manabe and Stouffer (1 980), Gates et al. (1 98 l), Schlesinger (1983b), and Washington and Meehl(1983b) the albedo of snow is independent of T,. In the latter simulation the albedo is simply taken as equal to a constant value of 0.80, while in the former three simulations it is a function of the nature of the underlying surface, for example, forest or tundra, and depends on the snow depth (water equivalent) up to some value. The albedo of snow-free sea ice is also a discontinuous function of Tgin the simulations of Manabe and Wetherald (1979, Manabe and Wetherald (1980), and Wetherald and Manabe (1981), with a value of 0.70 for T, < -25, - 10, and - 10°C, respectively, and 0.35, 0.35, and 0.60 for T, 2 -25, - 10, and - 10°C, respectively; in the simulation of Wetherald and Manabe (1 98 1) the albedo was further reduced to 0.45 during the melting of sea ice. In contrast, the albedo of snow-freesea ice was independent of T, in the simulationsby Washington and Meehl (1983b), Gates et al. (198l), and Schlesinger (1983b), and was taken as equal to the constant value of 0.70, 0.45, and 0.35, respectively. 3.1.5. Length of Simulation and Averaging Period. The length of the different COz simulations is shown in Table V along with the length of the averaging period for the results that follow. Of the simulationswithout the annual cycle of solar insolation, all but the results reported by Washington and Meehl(1983b) are for at least 2 yr; the averaging period for the results varies from 0.3 yr (100 days) to 1.4 yr (500 days). Of the simulations with the annual solarcycle, that by Gates eta). (198 1)with prescribed SST and sea ice was for only 1.3 yr ( 16 months), while those performed at GFDL with the slab mixed-layer model were for at least 12 yr (given by the sum of the first and third numbers in the appropriate row); the simulationby Wetherald and Manabe (198 1) with the slab model but without the annual cycle extended over 27 yr. The number of realizations (samples) in the averaging of the results in the simulationswith the annual cycle ranges from 1 (for example, 1 January, 1 winter, or 1 yr), as reported by Gates et al. (1981), to 4 (for

MODELS OF C02-INDUCED CLIMATIC CHANGE

165

example, the averageof 4 Januaries, 4winters, or 4 yr), as given by Wetherald and Manabe (1 98 1). However, Wetherald and Manabe (1 98 1) averaged the results from the two hemispheres shown in Fig. 7 to effectively double the averaging period to 8 yr. In the followingsectionswe present and compare results for the changes in temperature, precipitationrate, and soil moisture simulated as the result of a CO, doubling and a CO, quadrupling.

3.2. Simulated Temperature Changes 3.2.1. CO,Doubling. 3.2.1.I . Latitude- height cross sections. Latitude - height cross sections of the change in zonal-mean6air temperature simulated for a CO, doubling are presented in Fig. 8. [In this and other figures for CO, doubling, hatching (or dense shading) indicates warming greater than 4"C, sparse stipple a warming between 2 and 4 "C,and dense stipple a cooling. A similar scheme is used for the CO, quadrupling but with these temperatures doubled.] The two panels on the left side of Fig. 8 are from the earliest simulationsthat were performed at GFDL with sector models by Manabe and Wetherald (1 975, 1980) (see Tables V and VI); the two panels on the right are from the recent simulations performed at NCAR with a global model by Washington and Meehl(l983b). In the top panels, for both the GFDL and NCAR results, the clouds were fixed, whereasthe clouds were computed for the results shown in the bottom panels for each model. Each of the four simulations shows that stratospheric temperatures decrease and tropospheric temperatures increase in response to the doubled CO, . The magnitudes of the stratosphericcooling simulated by the GFDL and NCAR models are nearly equal, particularly for the simulations with fixedclouds. However, the tropospheric warmings simulatedby the models are quite different,with the warming ofthe NCAR model being considerably less than that of the GFDL model. The stratosphericcooling increases with increasing altitude in the GFDL and NCAR simulations. In the Manabe and Wetherald (1975) simulation with fixed clouds, the cooling at any altitude in the stratosphere is a maximum in the tropics and approachesa smaller constant value in the poleward direction. In the Manabe and Wetherald (1980) simulation with computed clouds, the C0,-induced stratospheric temperature decrease also becomes smaller from the tropics toward the poles above about 26 km. Below this altitude, however, the cooling decreases with latitude only to the subtropics The average over the longitudinal domain of a model.

m

'd/d

-

3

0

w

MODELS OF COZ-INDUCED CLIMATIC CHANGE

167

and then increases toward the pole. This results in a minimum cooling in the subtropical stratosphere. In both of the NCAR simulationsthe cooling is relatively independent of latitude near 18 km except near the poles, where the cooling extends downwardto lower altitudes. In the upper model stratosphere in both simulations the cooling is minimum in the tropics and increases toward the middle latitudes in both hemispheres. In the simulation with fixed clouds the cooling continues to increase to the South Pole, while there is a cooling minimum near 70"s latitude in the simulation with computed clouds. The tropospheric warming increases from the surface upward to a maximum value at about 10 km in tropical and subtropical latitudes in the Manabe and Wetherald (1975) simulation. This upward amplification of the tropospheric warming was attributed to the maintenance of the moist adiabatic temperature lapse rate by moist (cumulus) convection, a parameterized subgrid-scale process (see Table 11),and to the fact that this lapse rate decreases with increasing temperature. The upward amplification is also found in the Manabe and Wetherald (1980) simulation, although it is somewhat more confined to low latitudes. A similar upward amplification was obtained at almost all latitudesin the global model simulationby Schlesinger (1983b) (not shown in Fig. 8). In contrast, an upward amplification of the tropospheric temperature increase is not evident in either simulation with the NCAR model. The GFDL simulations also display a poleward amplification of the warming in the lower half of the troposphere, with maximum temperature increases at the surface near 80" latitude that are five to six times the minimum increases in the tropics. This poleward amplificationwas attributed to the ice albedo feedback mechanism, whereby an initial warming in high latitudes is amplified by melting snow and/or sea ice that results in a large decrease in surface albedo and an increase in the absorbed solar radiation, and to the vertical confinement of this surface warming by the low-level temperature inversion (Manabe and Wetherald, 1975). Both of the NCAR simulations show a weak poleward amplification in the Southern Hemisphere but not in the Northern Hemisphere. Although the GFDL models used by Manabe and Wetherald in 1975 and 1980 differ in more than their treatment of clouds (in particular, there is a difference in the parameterization of longwave radiative transfer; see Table FIG.8. The zonal-mean temperature differences ("C) for doubled C 0 2 simulated by the GFDL models of Manabe and Wetherald (1975, 1980) (a and b) and the NCAR model of Washington and Meehl (1983b) (c and d; fixed and computed clouds, respectively). Dense stipple indicates a temperature decrease, light stipple an increase between 2 and 4 ° C and hatching an increase larger than 4°C. P is pressure in millibars, P* is surface pressure.

168

MICHAEL E. SCHLESINGER

VI), a comparisonof the GFDL simulationswith fixed and computed clouds suggests that the influence of clouds on C0,-induced tropospheric temperature change is of secondary importance. A similar comparison of the NCAR simulationswith fixed and computed clouds also suggeststhat clouds may not be of primary importance, at least insofar as the zonal-mean temperatures are concerned. 3.2.1.2. Geographical distribution of surface air temperature change. The geographical distributions of the surface air temperature change simulated by the sector models of Manabe and Wetherald (1975, 1980) are presented in Fig. 9. The averageswith respect to longitudeof the temperature changes shown in this figure are those shown at the 990-mbar level in the left-hand panels of Fig. 8. Thus the surface air temperature in these and the other GFDL models shown in Tables V and VI represents the temperature at about 200 m above the earth's surface. Figure 9 clearly shows the poleward amplification of the warming of the surface air, with a maximum of 12°C at 82" latitude in the Manabe and Wetherald (1975) simulation and 8°C at the same latitude in the Manabe and Wetherald (1980) simulation. It may be that the 4" smaller maximum warming in the 1980 simulation is the result ofevaporative cooling over the high-latitude ocean that did not exist poleward of 66.5' latitude in the 1975 simulation. In both simulations land/ocean contrasts in the warming are found equatorward of about 45 latitude. In particular, in both simulations there is a secondary warming maximum near 40" latitude over about the western half of the continent without an oceanic counterpart. As can be seen from Figs. 22 and 3 1, these continental warming maxima occur where there was a decrease in the precipitation rate and a large decrease in the soil moisture. Another region of maximum continental warming is found over the southeastern coast in the 1975 simulation, but this region exhibits a minimum warming in the 1980 simulation. The latter result is co-located with an increase in precipitation rate (Fig. 22) and soil moisture (Fig. 3 1), while the former result occurs in associationwith a decrease in the precipitation rate (Fig. 22) and soil moisture (Fig. 3 1). The negative correlation that is found between the changes in surface air temperature and soil moisture occurs presumably through the change in surface evaporation, with increased (decreased)soil moisture resulting in increased (decreased)evaporative cooling and smaller (larger)warming of the earth's surface and surface air. The geographical distribution of the surface air temperature change simulated by the OSU global model of Schlesinger (1983b) is presented in Fig. 10. In this model, as in the model employed by Gates et al. (1981), the surface air temperature is determined by a constant-flux layer approximation (see Ghan et al., 1982)and thereforerepresentsthe temperatureat about O

169

MODELS OF COyINDUCED CLIMATIC CHANGE 90"

FIG.

co2.

than 4

I

I

I

X 3 9 N I S T I H 3 S ‘3 73VH3IM

011 N0 6 NOL

NO9 N OE

NO1 SOL

s OE s 09 SOL SO6

10 m above the surface. Figure 10 shows a warming of the surface air temperature over most of the earth. The warming increasesfrom the tropics to the subtropics of both hemispheres, as was also found from the 1980 GFDL sector model (Fig. 9), but then decreases in the middle latitudes before again increasing toward both poles. As a result of the decrease in warming in middle latitudes, the poleward amplification is not monotonic and the polar warming is only three times the tropical warming. This is in contrast to the amplificationof six and five obtained with the 1975and 1980 GFDL sector models, respectively. Figure 10also shows a significant longitudinal variation of the surface air temperature change at virtually all latitudes, some of which is clearly due to contrast between land and ocean. (Recall that the OSU model has realistic orography as well as land/sea contrastswhereas the GFDL sector models do not.) In the Northern Hemisphere, regions of warming in excess of 4°C are located over western and northern Greenland, the Arctic Ocean north of Spitsbergen, the Kara Sea, northeast of the Caspian Sea, and the Sahara, Arabian, and Gobi deserts. Similar but smaller warming regions are found in the Southern Hemisphere in South Africa, southwesternAustralia, the Ross ice shelf, and over Antarctica near 30"E. A region of large cooling is located in central east Africa. Figure 32 shows a large increase in soil moisture in this region, and Fig. 23 reveals a large increase in the precipitation rate there. Further analysis indicates that in general there is a negative correlation over land between the

MODELS OF C02-INDUCED CLIMATIC CHANGE

171

changes in surface air temperature and soil moisture, as was evident in the GFDL sector models' results. The changes in the surface air temperature simulated by the NCAR global model with computed and fixed clouds are presented in Fig. 1 I. This figure shows that although most of the earth's surface is warmed in response to the

FIG.1 1. NCAR model simulationsof the change in surface airtemperature ("C)for doubled C02with computedclouds(top) and fixed clouds(bottom). Heavy stipple indicatesa temperature decrease, light stipple an increase between 2 and 4"C,and hatching an increase larger than 4°C. [From Washington and Meehl(1983b); unpublished results.]

172

MICHAEL E. SCHLESINGER

COz doubling, there are regions of cooling simulated by the NCAR model, particularly with computed clouds. Ignoring these cooling regions for the moment, it can be seen that the warming tends to be small in the tropics and middle latitudes and large in the subtropics and polar latitudes. In the simulation with computed clouds, warming in excess of 4°C occurs over Antarctica inland from the Ross ice shelf. Regions of warming in excess of 3 "C are found extending from Greenland over Europe toward the Arabian Sea, inland from the Gulf of Mexico over the United States, and inland west of the Sea of Okhotsk. Regions of smaller warming are seen over the Sahara Desert, eastern Brazil, South Africa, and Australia. The NCAR simulation with fixed clouds shows some features that are nearly identical to those with computed clouds, such as the regions of maximum warming located between the Black and Caspian seas. However, there are fewer similarities than dissimilarities between these model simulations. This suggests that clouds may be important in the geographical distribution of C0,-induced temperature changes. 3.2.I .3. Zonal-mean surface air temperature change. A comparison of the zonal-mean surface air temperature changes simulated by the models of the preceding section is shown in Fig. 12 along with the change simulated by the OSU model with prescribed SST and sea ice (Gates et al., 1981). The zonal means for the GFDL sector models of Manabe and Wetherald (1975, 1980)are the longitudinally averaged data of Fig. 9, which are plotted symmetrically about the equator in Fig. 12. These curves both show a minimum warming of about 1.5"Cin the tropics followed by an increase toward the subtropics. The rise is more rapid in the 1975 simulation and reaches a maximum value of2.5 "C near 15 " latitude. The warming then decreases to about 2°C between 20 and 30" latitude and increases virtually monotonically toward a maximum value of nearly 11"C at the highest model latitude. On the other hand, the 1980 simulation does not display a maximum value in the tropics, but rather increases monotonically to a maximum value of about 7.5 "Cat 83 latitude and then decreasesto about 5.5 "C near the pole. The zonal means of the OSU global simulation by Schlesinger (1983b), that is, the longitudinally averaged data of Fig. 10, are quite similar to those of the 1980 GFDL simulation between 30"s and 25"N, with a minimum warming of about 1.25"C in the tropics and an increase to about 3 "C in the subtropics. However, in marked contrast to the almost 1 "C increase in warming that occurs between 34 and 38 latitude in the 1980GFDL simulation, there is a decrease in warming in the OSU simulation to about 1.5 "Cat 5 5 " s and 40"N. This is similar to the result ofthe 1975 GFDL simulation, albeit there is about a 10" latitude difference in positions of the resultant low-latitude warming maxima. Poleward of these latitudes the warming in the OSU simulation again increases with latitude to a value that is larger in O

173

MODELS OF C02-INDUCED CLIMATIC CHANGE

-I

90N

70

50

30

ION

IOS

30

50

70

9 S

Latitude

FIG.12. The change in zonal-mean surface air temperature (AT,) simulated by seven GCMs for doubled CO,. The data of Manabe and Wetherald (1975, 1980) (curves a and b, respectively, for 1975and 1980) are plotted symmetricallyabout the equator. Other curves: c, Gates et al. (1981); d, Schlesinger (1983b); Washington and Meehl(1983b); e, predicted clouds; f, prescribed clouds. (All data are unpublished results.)

the Arctic than in the Antarctic, namely, 3.9"C at 86"N and 3.0"C at 86"s. If the OSU curve is shifted upward such that the midlatitude minimum warming in each hemisphere intersects the 1980GFDL curve, it is seen that the poleward amplification in the OSU simulation is approximately half that of the 1980 GFDL simulation. The zonal means of the NCAR simulation with computed clouds, that is, the longitudinally averaged data of the top panel in Fig. 1 1, show a minimum warming of about 0.8"C in the tropics and a relatively uniform warming of about 1.3"Cin the midlatitudes of the Southern Hemisphere, with a poleward amplification there of about two. In the Northern Hemisphere the

32 0

- 30

-20

-E1

A!

k

f

-10

-n.

E

.

180

; Y

I (3 w 121 I

a

I

(3

g

83 5.5

N 90" 70° 50° 30" 10" -1 0" -30°-50"-70" -90" S LATITUDE

32 0

-E

15

20

E

15 5

I-

f 10 (3 w

I

7 074

e P

180 y

I-

189

12.1

336

83

5

S LATITUDE

N 90"70" 50" 30' 10" -1 0" -30" -50' -70' -90'' S LATITUDE

I (3

I;!

MODELS OF COyINDUCED CLIMATIC CHANGE

175

warming is a maximum of 2°C near 40"Nlatitude and decreasestoward the pole. Similar results are obtained in the fixed-cloud simulation, again indicating that the influence of clouds may not be important for the zonal-mean temperature changes. The zonal-mean warming in both NCAR simulationsis smaller than that obtained by the GFDL and OSU simulationsalmost everywhere. Figure 12 also shows that the zonal-mean surface air temperature change simulated by the OSU model with prescribed, noninteractiveSST and sea ice (Gateset al., 1981) is about an order of magnitude smaller than the warming obtained by the OSU model with an interactive swamp ocean (Schlesinger, 1983b).

3.2.2. CO, Quadrupling. 3.2.2.I. Latitude- height cross sections. Figure 13 showslatitude- height cross sections of the change in zonal-mean air temperature simulated for a CO, quadrupling. The top left panel is from the sector model of Manabe and Wetherald (1980)and is the counterpart to the cross section for the CO, doubling shown in the lower left panel of Fig. 8. The bottom left panel of Fig. 13 is the annually averaged temperature change from the global model of Manabe and Stouffer (1980),in which the ocean was treated as a 68-m slab mixed layer, the solar insolation was varied over its annual cycle, and the clouds were fixed. The two panels on the right are from the global model of Washington and Meehl(1983b) with the swamp ocean model and annually averaged insolation. The simulations with fixed and computed clouds are shown in the top and bottom panels, respectively, as in Fig. 8 for the CO, doubling. Each of the cross sections in Fig. 13 shows a warming of the troposphere and a cooling of the stratosphere in response to the CO, quadrupling, with the stratosphericcooling increasingwith altitude. This is the same response as that obtained for the CO, doubling (Fig. 8). Comparingthe top left panel of Fig. 13with the lower left panel of Fig. 8, and the right-hand panels of Fig. 13 with the right-hand panels of Fig. 8, it is seen that the response of the GFDL model for CO, quadrupling is everywhere very nearly double the response for CO, doubling, while a similar linear behavior is clearly evident only in the stratosphere in the NCAR simulations. Although there are tropospheric regions in the NCAR simulations where the warming for CO, quadrupling is twice as large as the warming for doubling, it is difficult to FIG.13. The zonal-meantemperaturedifferences("C)for quadrupledC 0 2simulated by the GFDL models of Manabe and Wetherald (1980) (a) and Manabe and Stouffer (1980)(b), and the NCAR model of Washington and Meehl (1983b) with prescribed (c) and predicted (d) clouds. Heavy stipple indicatesa temperature decrease, light stipple an increase between 4 and 8"C, and hatching an increase larger than 8°C.

176

MICHAEL E. SCHLESINGER

discern such a relation elsewhere owing to the coarse contour interval shown for the doubling results (Fig. 8). Comparing the left-hand panels in Fig. 13 shows that the annually averaged response of the tropospherein the Manabe and Stouffer (1980) simulation with the annual insolation cycle is smaller than the tropospheric response of the Manabe and Wetherald (1 980) simulation with annually

t!

z

a 205

10 350 515

5

660 830 940 990 90

0

80

70

60

50

40

25

6

n

10

0

95 15

-5 c

205

c

lo

t!

a

20

20

Y

$

30

25

E

?! a

p

CF

v)

?! a

.-0,

I'

350 515

5

660

8% 990

0

Latitude

FIG.14. The zonal-mean temperature differences ("C)for quadrupled CO, simulated by the GFDL model of Wetherald and Manabe (1981, unpublished results). (a) Annual model; (b) seasonal model. Heavy stipple indicates a temperature decrease, light stipple an increase between 4 and 8"C, and hatching an increase larger than 8°C.

MODELS OF C02-INDUCED CLIMATIC CHANGE

177

averaged insolation. In particular, the maximum high-latitude surface warming decreases from 15 to 9 "C and occurs in the polar cap instead of at 83' latitude. Furthermore, the inclusion ofthe Southern Hemisphere in the Manabe and Stouffer (1980) model reveals hemispheric differences that could not be shown by the sector models of Manabe and Wetherald (1975, 1980). In particular, the surface warming in the Arctic is larger than that in the Antarctic. A similar hemispheric asymmetry was found in the doubling simulation with the OSU model by Schlesinger(1983b)(Fig. 12). However, the NCAR quadrupling simulations display nearly equal warmings in the Arctic and Antarctic. Finally, it should be noted that the stratospheric cooling of the Manabe and Stouffer (1 980) simulation with the annual solar cycle is considerably larger than that obtained by Manabe and Wetherald (1980) without the annual solar cycles, particularly in high latitudes. A more comprehensive and rigorous assessment of the effects of the annual insolation cycle on C02-induced climatic change was conducted by Wetherald and Manabe (1 98 1) by performing two simulationswith the same (sector)model, one with the annual solar cycle (hereafter called the seasonal model) and the other with annual-mean insolation (the annual model). In this model the ocean was treated as a 68-m slab mixed layer (see Tables V and VI). The zonal-mean temperature differences of the annual and seasonal models in response to quadrupled C02are presented in Fig. 14. Comparing these annual-mean temperature differencesshowsthat the principal effect of the annual insolation cycle is to reduce the warming of the zonal-mean surface air temperature at all latitudes, with the reduction increasing with latitude from about 0.5"C in the tropics to 4°C poleward of 70" latitude, 3.2.2.2. Geographical distribution of surface air temperature change. The geographical distribution of the surface air temperature change simulated by the sector model of Manabe and Wetherald (1980) is shown in Fig. 15. This figure for quadrupled CO, is the counterpart of the lower panel of Fig. 9 for doubled COz. To aid in discerning whether the quadrupling-induced warming is equal to twice the doubling-induced warming, these two figures have dense shading for warming greater than 8 "C for the quadrupling (dense stipple, Fig. 15) and 4°C for the doubling (hatching, Fig. 9) and light shading for warming between 4 and 8' C for the quadrupling (light stipple, Fig. 15) and between 2 and 4°C for the doubling (stipple, Fig. 9). Comparing Fig. 15 with Fig. 9 shows that the response for the CO, quadrupling is qualitatively quite similar to that for the doubling, with a poleward amplification of the warming, maximum warming near 83" latitude over land and ocean, land/ocean contrasts only equatorward of about 45 latitude, a secondary warming maximum near 40" latitude over the western part of the continent, and minimum warming over the southeasterncoast. A comparison of the magnitude of the two C0,-induced temperature increases shows O

178

MICHAEL E. SCHLESINGER

LONGITUDE

FIG.15. GFDL model simulation of the change in surface air temperature ("C)for quadrupled C02. Light stipple indicates a temperature increase between 4 and 8"C,dense stipple an increase larger than 8°C. [From Manabe and Wetherald (1980).]

that the warming for the quadrupled CO, level is slightly less than twice the warming for doubled CO, over most of the sector domain. Comparing Fig. 15 with Fig. 35 shows again the negative correlation between changes in the surface air temperature and soil moisture; that is, the warming is large where the soil is desiccated and small where there is moistening. Figure 16 shows the geographical distribution of the surface air temperature change simulated by the NCAR model of Washington and Meehl (1983b). Comparing this figure with its counterpart for the C02doubling, Fig. 1 1, shows that the warming patterns have several similarities. In particular, for the simulations with computed clouds there are regions of maximum warming extending from Greenland over Europe toward the Arabian Sea; inland from the Gulf ofMexico over the United States;over the Atlantic Ocean, the Sahara Desert, the regions east of the Caspian Sea and west of the Sea of Okhotsk; and over eastern Brazil, South Africa, and Australia. Fur-

179

MODELS OF CO1-INDUCED CLIMATIC CHANGE

thermore, for both simulations with computed and fixed clouds most of the regions of cooling for the doubling (Fig. 1 I ) are eliminated for the C02 quadrupling and are replaced by regions of minimum warming. This result is what would be expected if the cooling regions in Fig. 1 1 for the doubling are manifestations of noise. The superposition of “large” noise on a “small”

90N 70 N 50N 30 N

ION 10s

30s 50s

70s 90s 180

150W

120W

90W

60W

30W

0

30E

60E

90E

120E

150E

FIG.16. NCAR model simulationsof the change in surface air iemperature (“C)for quadrupled C02with computed clouds (top)and fixed clouds (bottom). Stipple indicates a temperature increasebetween 4 and 8°C. [From Washingtonand Meehl(1983b);unpublished results.]

180

180

MICHAEL E. SCHLESINGER

0

60

I20

FIG.17. GFDL model simulationsof the change in surface air temperature ("C)fox.quadm pled C02. (a) Annual model; (b) seasonal model. Stipple indicates a temperature increase between 4 and 8°C and hatching an increase larger than 8°C. [From Wetherald and Manabe (1 98 1); unpublished results.]

MODELS OF CO2-INDUCED CLIMATIC CHANGE

181

doubling signal could produce negative temperature changes, but the superposition of such noise on the enhanced quadrupling signal would result in both fewer and weaker cooling regions. It is in fact the purpose of a superanomaly experiment such as a CO, quadrupling to increase the signal-to-noiseratio by increasingthe signal. As noted earlier, the validity of this approach lies in the assumed known relationship between the responses for quadrupled and doubled CO, . A comparison of the regions of maximum warming in Figs. 1 1 and 16 indicates that the warming for quadrupling is somewhat less than twice that for the doubling, as is also found for the GFDL sector model results. However, it is not clear whether such quasi-linearity exists in those regions where the cooling for the CO, doubling has changed to warming for the CO, quadrupling. The only way to verify this would be to extend the doubling simulation sufficiently long that the time-averaged signal significantly exceeds the timeaveraged noise so that the cooling regions disappear. If this were done, however, the need to perform the superanomaly quadrupling experiment would de facto be eliminated. Nevertheless, the occurrence of extensive cooling regions for the doubling lowers its global-mean warming and tends to make the ratio of the global-meanwarmings for quadruplingand doubling larger than the linear value of two (see Table VII). The geographical distributions of the change in surface air temperature simulated by the annual and seasonal models of Wetherald and Manabe ( 1 98 1) are shown in Fig. 17. A comparisonofthe annual-mean temperature differences reveals that the longitudinal variation of the response of the seasonal model is smaller than that of the annual model, particularly equatorward of 45 ' latitude. Furthermore, the seasonalmodel is less sensitiveto the CO, quadrupling than is the annual model, especially in high latitudes where the maximum warming is 4°Csmaller.' This reduced sensitivity of the seasonal model is attributed to the absence of the ice albedo feedback mechanism in summer when there is no snow cover or sea ice in both the seasonal model experiment and the control, whereas ice albedo feedback apparently exists perpetually in the annual model (Wetherald and Manabe, 198 1). This comparison indicates that the influence of the seasons on the annual-mean climate change is not negligible. The geographical distribution of the change in the annual-mean surface air temperature simulated by the global GFDL seasonal model of Manabe and Stouffer (1980) is shown in Fig. 18 along with the corresponding maps for December, January, and February and June, July, and August. The Preliminaryresults from a version of the NCAR model with a 50-m slab mixed-layerocean and the annual solar cycle indicate a greater sensitivity of this seasonal model compared to the annual model of Washington and Meehl(1983b) (Washington and Meehl, 1983a).

MODELS OF C02-INDUCED CLIMATIC CHANGE

183

annual-mean temperature change shows warming everywhere, with minimum values in the tropics and maximum values in the polar latitudesin both hemispheres. These features are in general agreement with the changes simulated by the NCAR model for quadrupled-CO, concentration (Fig. 16). The GFDL model simulation shows that the warming in the Arctic is about 4°C larger than that in the Antarctic, in contrast with the more nearly equal warming obtained by the NCAR model. Furthermore, less longitudinal variation in the warming is simulated by the seasonal GFDL model than is simulated by the annual NCAR model. This is what was also found for the seasonal and annual GFDL model simulations of Wetherald and Manabe (198 1). Figure 18 also shows that there is a large seasonal variation of the C02-induced warming in the middle and high latitudes of both hemispheres, with large warming in winter and small warming in summer. There is even a small region of cooling located over the Ellsworth Highland in austral summer, as well as over northern Australia and central South America. This nonuniformity of the C0,-induced temperature change throughout the year is even more strikingly revealed by Fig. 19, in which the annual cycle of the change in the zonal-mean surface air temperature is shown. It can be seen here that the zonal-mean warming is maximum in winter and minimum in summer over both the oceans and the continents in the high latitudes of the Northern Hemisphere. This indicates a large reduction of the amplitude of the annual temperature cycle in these latitudes and is attributed to the change of the thermal insulation effect of sea ice (Manabe and Stouffer, 1980). A similar seasonal asymmetry in the warming is found over the high-latitude oceans of the Southern Hemisphere, but not over Antarctica. Furthermore, there is little variation of the warming throughout the year in the tropics. 3.2.2.3. Zonal-mean surface air temperature change. A comparison of the zonal-mean surface air temperature change simulated by the models of the preceding section is shown in Fig. 20 along with the change simulated by the OSU model with prescribed SST and sea ice (Gates et al., 1981). The results for the GFDL sector models of Manabe and Wetherald (1 980) and Wetherald and Manabe (I 98 1) are shown plotted symmetrically about the equator, and the curves for the seasonal models of Manabe and Stouffer (1980), Wetherald and Manabe (1 98 I), and Gates et al. (198 1) represent the

FIG. 18. GFDL model simulation of the change in surface air temperature for quadrupled CO,. (a) Annual mean; (b) December/January/February; (c) June/July/August. Sparse shading indicates a temperature increase between 5 and 7.5"C, dense shading an increase greater than 7.5"C. [From Manabe and Stouffer (1980).]

184

MICHAEL E. SCHLESINGER a

C

30

60

905

J

F

M

A

M

J

J

A

S

O

N

D

Month

b 90h

60

30

Month

FIG.19. GFDL model simulation of the annual cycle of the change in zonal-mean surface air temperature ("C)for quadrupled C02, (a) Oceans and continents; (b) oceans; (c) continents. Sparse shading indicates an increase between 4 and 8"C,dense shading an increase larger than 8°C. [From Manabe and Stouffer (1980).]

changes in the annual-mean surface air temperature. The simulations by the GFDL models all show a minimum warming in the tropics and a poleward amplification. The latter is smaller in the seasonal models of Manabe and Stouffer (1980) and Wetherald and Manabe (1 98 1) than in the annual models of Manabe and Wetherald ( 1980) and Wetherald and Manabe (198 1). There is also an asymmetry in the poleward amplification in the global model of Manabe and Stouffer (1980), with greater warming in the Arctic than in the Antarctic. In contrast, the NCAR model simulations of

J

185

MODELS OF C02-INDUCED CLIMATIC CHANGE

Washington and Meehl(1983b) show a smaller poleward amplification and little difference in the Arctic and Antarctic warnings. The zonal-mean warming in the NCAR simulationsis almost everywheresmaller than that in any of the GFDL simulations. However, the warming obtained by the NCAR model is everywhere considerably larger than that obtained with the OSU model when the sea surface temperature and seaice were not allowed to respond and feed back on the C0,-induced climatic change (Gates et al. 1981).

3.2.3. Comparison of Global-Mean Warmings Simulated for Doubled and Quadrupled CO,. 3.2.3.1. GCMsimulations. A summary of the temperature changes simulated by eight GCMs for CO, doubling and quadrupling is shown in Table VII in terms of the area-averaged surface air temperature. For the four

models with the annual solar cycle, that is, the seasonal models of Manabe and Stouffer (1980), Wetherald and Manabe (1981), Mitchell (1983), and Gates et al. ( 1 98 I), the annual-mean change is presented in Table VII. For I

0

I

I

I

I

I

I

I

....... c .......... . ........ . . ........... .... ......... I ....' 1 ..........(. I y ......... I. .........i

90N

70

50

30

ION

IOS

Latitude

30

50

70

91 S

FIG.20. The change in zonal-mean surface air temperature (AT,) simulated by seven GCMs for quadrupled COz. The data of Manabe and Wetherald (1980) (a) and Wetherald and Manabe ( 1981;curves c and d: seasonal and annual insolation, respectively) are plotted symmetrically about the equator. Other curves: b, Manabe and Stouffer (1980); c, Gates et al. (1981); Washington and Meehl(1983b): f, predicted clouds; g, prescribed clouds. (All data are unpublished results.)

186

MICHAEL E. SCHLESINGER

TABLEVII. AREA-MEAN ANNUAL-MEAN SURFACE AIRTEMPERATURE AND COz-INDUCED TEMPERATURE CHANGES ("C) Mode1

1 X C02

I X C02-OBS'

2 X COZ- 1 X CO2

Manabe and Wetherald (1975) [81b(300, 600, --y Manabe and Wetherald (1 980) [24] (300,600,1200) Manabe and Stouffer (1980) I201 (300, -, 1200) Wetherald and Manabe (1981)d (301 (300, -, 1200) Washington and Meehl(1983b3)'[33] (330,660, 1320) Mitchell (1983) [28] (320.7, 641.4, -) Gates et 01. (1 98 1) [25](326,644, 1289) Schlesinger (1983b) (321 (326,644, -)

20.9

6.7

2.9

21.3

6.9

3.0

14.8

0.6

4.1

16.0 16.8

1.8 2.6

6.0 4.8

11.6 11.6

- 2.6 - 2.6

1.3 1.3

3.4 2.7

12.3

-0.9

0.2

0.4

14.8

0.6

0.2

17.9

3.7

2.0

4 X COZ- 1 X C02

5.9

Observed value of 14.2"C based on data of Crutcher and Meserve (1970) and Taljaard et al. (1 969) as given by Jenne (1975). These numbers correspond to those in Tables IV and IX and Figs. 21,29, and 30. C 0 2concentrationsin ppm for 1 X CO,, 2 X C02,and 4 X CO,. First row is for annual-mean insolation, second row is for seasonal insolation. First row is for computed clouds, second row is for prescribed clouds.

both the doubling and the quadrupling each simulation shows an increase in the global-mean surface air temperature. The smallest warming for each increased C02 level is obtained by the UKMO and OSU models with prescribed SST and sea ice (Mitchell, 1983; Gates et af.,198 1). However, the OSU model with an interactive (swamp) ocean and sea ice (Schlesinger, 1983b)gives a 10-fold increase in the warming, at least for the C02doubling. A similar amplification was first obtained by the author from a pair of RCM sensitivity calculations, one in which the surface temperature in the experiment was fixed and the other in which it was predicted. In both the RCM and GCM calculations with prescribed surface temperatures (the latter only for the ocean), the surface (ocean) acts as an infinite heat sink. That is, the increased downward infrared radiation

MODELS OF C02-INDUCED CLIMATIC CHANGE

187

from the atmosphere to the earth’s surface, which is a direct result of the increased opacity of the C0,-enriched atmosphere, simply passes into the surface (ocean) and is lost. As pointed out by Watts (1982), in this case the models’ climate system is not energetically closed. When the surface (ocean)temperature is free to respond, however, it warms and returns energy to the atmosphere through the fluxes of latent, sensible, and radiative (infrared) heat. This in turn warms the atmosphere and increases the downward infrared (IR) radiation to the surface and the upward IR radiation to space. This feedback process continues to warm the surface and the atmosphere until the increasein the IR radiation to space due to the warmer temperature compensates the decrease that initially occurred from the increased C02 opacity. When this compensation occurs, the balance between the solar heating and IR cooling of the climate system is restored. It is not restored when the surface (ocean) temperature is fixed and energy is lost through the lower surface. The response of the NCAR model with both fixed and computed clouds (Washington and Meehl, 1983b) is smaller than that of the OSU model (Schlesinger, 1983b)that has computed clouds, while the warming obtained by the GFDL sector model with computed clouds (Manabe and Wetherald, 1980) is larger than that of the OSU model. For the quadrupled-CO, concentration, the global-mean warming of the seasonal model of Wetherald and Manabe (1 98 1) is smaller than that of their annual model, and a similar reduced sensitivity is found for the seasonal model of Manabe and Stouffer ( 1980)in comparison with the annual model of Manabe and Wetherald (1980), albeit these two models also differ in several other ways (see Table VI). The warming ofthe annual NCAR model of Washington and Meehl(1983b) is not only smaller than the warming of the annual GFDL models, it is also smaller than that of the seasonal GFDL models. In fact, the global-mean warming of the NCAR model simulations for quadrupled COzis comparable to the warming of the GFDL models for doubled CO, (see Figs. 2 1 and 29). As Table VII shows, four models have been used to simulate both a CO, doubling and a quadrupling. The OSU model (Gates et al., 1981), the GFDL model (Manabeand Weatherald, 1980),and the NCAR model simulation with fixed clouds (Washington and Meehl, 1983b) show that the warming for quadrupling is virtually equal to twice the warming for doubling, while the NCAR model simulation with computed clouds gives a ratio closer to three. Table VII also shows the global-mean surface air temperature of the control (1 X CO,) simulation for each model, the observed value, and their difference. This reveals that of all the models’ simulations of the present climate, only those of the UKMO and NCAR models are colder than the

188

MICHAEL E. SCHLESINGER

TABLEVIII. RCM STUDYOF THE DEPENDENCE OF THE SURFACE AIR TEMPERATURE WARMING INDUCED BY DOUBLED CO, ON THE TEMPERATURE OF THE CONTROL SIMULATION

10.6

5.3

3.3 1 3.27

13.9 8.5

3.10 3.23

observed temperature. It is interesting therefore that, excluding the simulations by Mitchell (1983) and Gates et al. (198 1) with prescribed SST and sea ice, the warming simulated for both doubled and quadrupled C 0 2 concentrations isa minimum for the NCAR model. Moreover, as shown by the top panel of Fig. 29, there is an increasing relation between the C0,-induced surface air temperature warming and the surface air temperature of the control. That is, in the sense of intermodel comparison, the warmer the control the larger the C0,-induced warming.8 This relation could be due to the combined effect of the nonlinear increase of evaporation with surface temperature as suggested by Newel1 and Dopplick (1979), and the resultant increase in the atmosphere's water vapor content and its contribution to the greenhouse effect. I have made a preliminary test of this hypothesis with an RCM in which the relative humidity was fixed; there were no clouds; the surface heat exchange was parameterized as a latent heat flux C,(q,*- q,), with C, a prescribed transfer coefficient in units of langleys per day, @ the saturation mixing ratio at the temperature ofthe ground and surface pressure, and qsthe mixing ratio of the surface air that changes with surface air temperature T, due to the fixed relative humidity. In this test the temperature of the control was made to increase by decreasing the prescribed surface albedo from 0.1 to 0.05. Table VIII shows that increased warming does occur with increased temperature of the control for the smaller value of C,,but the opposite occurs for the larger value. This suggests that the sensitivity of the warming to the control temperature may depend on the dynamics. In any event, the changes shown in Table VIII are small in comparison with the differences between the models' warnings shown in Table VII and Fig. 29. This indicates that some explanation other than the nonlinear increase ofevaporation with surface temperature may have to be sought to explain these differences.

* However, in the sense of inlramodelcornpurrson. as revealed by the solar constant sensitivity studiesofwetheraidand Manabe( 1975,1980), the warmerthecontrol thesmallerthe temperature sensitivity.

MODELS OF C02-INDUCED CLIMATIC CHANGE

189

3.2.3.2. EBM and RCM simulations. Figure 21 shows the change in surface temperature simulated by EBMs, RCMs, and GCMs for COz concentrations that are halved, doubled, quadrupled, and decupled. The characteristics of the individual models used to perform these simulations are summarized in Table IV. Focusing attention on the C02 doubling we see that both the maximum and minimum values of 9.6 and 0.1 "C have been obtained by the EBMs, namely, by Moller (1963) (model) [ 13 and Sellers (1973) [5], respectively. Similar large temperature changes have been obtained by Kandel (1981) [27], and small changes by Rasool and Schneider (1971) [4], Newell and Dopplick (1979) [22], Idso (1980) [23], and Kandel(l981) [27]. Manabe and Wetherald ( 1967), Schneider ( 1979, and Manabe ( I 983) have shown that the wide range of warming obtained by Moller (1963) [ 11 was due to his

Ratio of C02 concentration to preindustrial value 1. Moller 11963) 2. Manabe and Wetherald 119671 3. Manabe 11971) 4. Rosool and Schneider 11971) 5. Sellers (19731 6. Sellers 119741 7. Weore and Snell 119741 8. Manabe and Wetherald (19751

9. Romonothon 119751 10. Schneider (19751 I I. Rmkin and SnelI 119761 12. Augustssan and Romonothan 119771 13. Patter 11978, 19801 14. Hansen 119781 15. Rowntreeand Walker 11978) 16. HOnSen (19791 17. Humrnel and Reek 119791

18. Hunt and Wells 119791 19. MacDonold et Of. (1979) 20. Monabe and Stouffer (1979, 19801 21. RamonathanefaL 11979) 22. Newell and Oopplick 119791 23. Idso (19801 24. Monabe and Wetherald 119801 25. Gat%se/of. 119811

26. 27. 28. 29. 30. 31. 32. 33. 34.

Honsen ef a/. 11981I Kandel (1981 I Mitchell (1983) Ramanathan 11981I Wetherald and Monabe 119811 Hall ct a/. 11982) Schlesinger 11983b) Washington and Meehl I1983 bl Alrkrondrov ctol. 119831

FIG.2 1. The change in surface temperature induced by halving (fx), doubling(2X), quadrupling (4X), and decupling (1OX) the preindustrial C02concentration as simulated by energy balance models (EBMs), radiative-convective models (RCMs), and general circulation models (GCMs). [After Schlesinger (1983a).]

190

MICHAEL E. SCHLESINGER

use of a model that required an energy balance for the earth's surface rather than for the entire earth - atmosphere climate system as described in Section 2.1. Kandel ( 1981) found a similar explanation for the small values obtained by Newell and Dopplick (1 979) [22] and Idso (1980) [23], but argued that the surface energy balance model can be reconciled with the planetary energy balance model. However, uncertainty in the response of the atmospheric water vapor gives the wide range shown for Kandel's model [27], in agreement with that obtained by Moller (1963) [l]. In a study by Watts (1 982) it was shown that the temperature changes predicted by surface energy balance models are highly sensitiveto the values of their parameters, so much so that Watts concluded that such models are not useful and should be replaced with tropospheric heat balance models. Watts (1982) also showed that the small warming obtained by Newell and Dopplick (1979) [22] is the result of not having iterated their surface energy balance model to its equilibrium solution. Interestingly, Sellers ( 1974)also discounted his earlier small warming (Sellers, 1973 [S]) as due to not having run his model sufficiently long to reach equilibrium. Schneider (1975) has discussed the reasons for the small warming of his earlier study (Rasool and Schneider, 1971 [4])and for that of Weare and Snell(l974) [7]. If we exclude all of the EBM simulations discussed above from consideration, the range of the warming simulated by EBMs for a C02 doubling is 1.3-3.3"C. The range ofwarming simulatedby RCMs for doubled C02is 1.3 - 3.2 "C, in remarkable agreement with that given by EBMs. Augustsson and Ramanathan (1977) have shown that this range of RCM warmings can be explained by the models' different treatments of the water vapor and clouds. It was first shown by Manabe and Wetherald (1 967) [2] that prescribing the relative humidity instead of the specific humidity approximately doubles the RCM's surface warming, from 1.3 to 2.4"C with prescribed average cloudiness and from 1.4 to 2.9"C for a cloud-free atmosphere. This enhancement, or positive feedback, occurs because the atmospheric water vapor must increase with increasing temperature to maintain constant relative humidity, and because of the strong greenhouse effect of water vapor. Augustsson and Ramanathan (1977) showed that when the relative humidity itself was made an increasing function of the surface temperature, the C02-induced warming increased (for the same reasons as described above). They also showed an enhancement of the warming when the temperature of the cloud tops was fixed rather than when the cloud top altitude was fixed. This finding was confirmed by the study of Hansen et al. (1981). The range of surface warming simulated by the GCMs when the prescribed-SST/sea ice simulations of Gates et al. (1981) and Mitchell (1983) are excluded from consideration is somewhat larger than that of the purely thermodynamical models, namely, 1.3- 3.9"C.

MODELS OF C02-INDUCED CLIMATIC CHANGE

191

3.3. Simulated Precipitation Changes

3.3.1. CO, Doubling. The geographical distributions of the change in precipitation rate simulated by the sector models of Manabe and Wetherald (1975, 1980) for a CO, doubling are presented in Fig. 22. (In this and subsequent precipitation figures for both the doubling and quadrupling, dense shadingindicates a decrease, sparse shadingan increase between 2 and 5 mm/day, and hatching an increase greater than 5 mm/day.) This figure shows that the precipitation rate in both simulationsincreased almost everywhere poleward of about 45" latitude, albeit by less than about 1 mm/day. Equatorward of this latitude there are large regions where the precipitation rates decreased as well as increased, and the magnitude of the changes is larger than that in the high latitudes. Accompanying this in both simulations there is a much larger longitudinal variation of the changes in low and middle latitudes than in high latitudes. These features are reflected in the changes in the zonal-mean precipitation rates that are shown in Fig. 24 (upper left panel). The longitudinal variation equatorward of about 45 latitude causes cancellation in the zonal means there such that the increased precipitation rate of about 0.3 mm/day in high latitudes is comparable to the maximum decrease in the zonal mean near 40" latitude and the maximum increase near 25 latitude. As shown by Fig. 22 the decrease in the midlatitude zonal-mean precipitation rate is the result of the extensive region of decreased precipitation that extends from the center of the ocean to the center of the continent in both simulations. [Recall that cyclical boundary conditionsare imposed in the sector models. Thus there is ocean to the left (west in the Northern Hemisphere) of the continent equatorward of 66.5" latitude in the 1975 simulation and at all latitudes in the 1980 simulation.] This latter region is the location of a maximum warming in both simulations, as already shown in Fig. 9. Figure 22 shows that the increased zonalmean precipitation rate near 25"latitude is the result of the increased precipitation rate over the continent and ocean at this latitude in the 1980 simulation,and of a similar continentalincreasecombined with a minimum maritime decrease in the 1975 simulation. Figure 24 reveals that the largest differencebetween the two simulations occursin the equatorial region where the 1975 simulation gives an increase of 0.8 mm/day and the 1980 simulation gives a decrease of 0.35 mm/day. Figure 22 shows that these equatorial differences occur predominantly over the ocean where increased precipitation occurs in the 1975 simulation and decreased precipitation occurs in the 1980 simulation. While clouds were predicted in the 1980 simulation and were fixed in the 1975simulation, it is premature to ascribe the differences in equatorial maritime precipitation to cloud feedback (R. T. Wetherald, personal communication, 1982). O

192

MICHAEL E. SCHLESINGER

80

0

60

I20

Longitude

FIG.22. GFDL model simulations of the change in precipitation rate (millimeters/day) for doubled CO,. Top, data from 1975; bottom, 1980. Dense shading indicates a rate decrease, sparse shading an increase larger than 2 mm/day. [From Manabe and Wetherald (1975, 1980); unpublished results.]

193

MODELS OF CO2-INDUCED CLIMATIC CHANGE 9 0N

70N 50N 3 0N 10N 10s

30 S 50s

70s 90s 180

150W

120W

90W

60W

30W

0

30E

60E

90E

120E

150E

FIG.23. OSU model simulation of the change in precipitation rate (millimeters/day) for doubledCOz. Dense shading indicates a rate decrease, no shading an increase between 0 and 2 mm/day, sparse shading an increase between 2 and 4 mmlday, and hatching an increase in excess of 4 mm/day. [From Schlesinger (1983b).]

Figure 23 shows the geographical distribution of the change in precipitation rate simulated by the OSU model (Schlesinger, 1983b) for C02 doubling. It can be seen that unlike the surface air temperature (Fig. lo), the precipitation rate decreased as well as increased over a large fraction of the earth's surface. In general the largest changes in precipitation occurred between 30"s and 30"N latitudes. Increases in the precipitation rate by more than 2 mm/day are located predominantly on or near the equator over Indonesia, northeast of New Guinea, the central Pacific Ocean, Equador, and central east Africa. The latter region coincides with the decrease in surface air temperature shown in Fig. 10. Decreases in the precipitation rate by more than 2 mm/day are located over the Arabian Sea- Indian Ocean, the Coral Sea, north of the Gilbert' Islands, the Atlantic Ocean, and north Africa. The regions of increased precipitation dominate, however, so that there is an increase in the zonal-mean precipitation rate around the equator, as shown by Fig. 24 (upper right panel). This is in agreement with the increase simulated by the GFDL sector model of Manabe and Wetherald (1975). However, there is disagreement between the changes in these models' zonal means in the Northern Hemisphere, where the OSU model simulates a decreased precipitation rate in the subtropics and an increased rate in midlatitudesand both the 1975and 1980GFDL models simulate the reverse. Interestingly, there is better agreement in the Southern Hemi-

180

194

MICHAEL E. SCHLESINGER

sphere, not only in the subtropicsand midlatitudes,but also in high latitudes, where all three models simulate an increase. (However, recall that there is only one hemisphere in the sector models.) In the high latitudes of the Northern Hemisphere all three models simulate an increase in the precipitation rate, with that of the OSU model being smaller than that of the GFDL models. This occurs, at least in part, because there are regions of decreased precipitation rate in the high latitudes of both hemispheres in the OSU simulation (Fig. 23), while such negative regions are virtually absent from high latitudes in the GFDL simulations (Fig. 22).

-

v

r. -0.4

0 0

90N

I

70

l

50

I

30

l

ION

I

10s

1' U-&*t 50

0 j 3

70 90s-0.490N

30 1

50 1

70 1

ION I

90N

30 I

50 I

70 1

1

O 0.6' I

-0.41

IOS 1

I

70

I

50

I

30

I

ION

8

IOS

I

30

I

50

I

I

70 90s

Latitude

FIG. 24. The change in zonal-mean precipitation rate (AP) simulated by six GCMs for doubled CO,. The data of (A) Manabe and Wetherald (1975, curve a; 1980, curve b) are plotted symmetrically about the equator. (B) Data from Gates et af. (1981, curve a) and Schlesinger (1983b, curve b). (C) Data from Washington and Meehl (198313): curve a, predicted clouds; curve b, prescribed clouds. (All data are unpublished results.)

1

90s

MODELS OF C02-INDUCED CLIMATIC CHANGE

195

The changes in the zonal-mean precipitation rate for the doubled-CO, simulations by the NCAR model (Washington and Meehl, 1983b) with computed and fixed clouds are shown in the lower panel of Fig. 24. (The corresponding geographical distributions were not available.) Both of the NCAR simulations show an increase in the precipitation rate in the higher latitudes of both hemispheres, in agreement with the GFDL and OSU (Schlesinger, 1983b) simulations. Both NCAR simulations also show a band of decreased precipitation near 30"N latitude, as do the GFDL simulations and, somewhat more equatorward, the OSU simulation. The NCAR results show disagreement between each other in the tropics, where the model with computed clouds simulatesa decrease (increase) in the Northern (Southern) Hemisphere and the model with fixed clouds simulates the reverse. This result is similar to that found for the GFDL models with fixed clouds (Manabe and Wetherald, 1975) and computed clouds (Manabe and Wetherald, 1980). Figure 24 (upper right panel) also showsthe changes simulatedby the OSU model with prescribed sea surface temperature and sea ice. It is evident that these changes are not only generally considerably smaller than those of the other models, but that they are negative almost everywhere.

3.3.2. CO,Quadrupling. The geographical distribution of the change in precipitation rate simulated by the sector model of Manabe and Wetherald (1980) for a C02quadruplingis presented in Fig. 25. A comparison of this figure with the lower panel of Fig. 22 for the CO, doubling with this model showsthat there are many qualitative similarities. In particular, the precipitation rate increases almost everywhere poleward of about 45 latitude,there are decreases as well as increases equatorward of this latitude, the magnitude of the changes in low latitudes is much larger than that in high latitudes, and the longitudinal variations are larger in low latitudes than in middle and high latitudes. These similaritiesin the geographicaldistributions of the changes in precipitation rate for the quadrupling and doubling simulations of Manabe and Wetherald (1 980) lead to the similarity of the respective changes in the zonal-mean precipitation rates that are evident from a comparison of Figs. 28 and 24 (upper left panels). Both figures show a decrease near the equator that is the result of the decreased precipitation rate over the ocean shown by Figs. 25 and 22. The zonal means for the quadrupling and doubling show increased precipitation rates in the subtropics and high latitudes and a small region of decreased precipitation in the middle latitudes. This latter feature is the result of the decreasein precipitation rate that occursover the western half of the continent and eastern half of the ocean (if the sector is taken to be the Northern Hemisphere) at 40" latitude for both the quadrupling and the doubling. There is also agreement in the locations of the O

196

MICHAEL E. SCHLESINGER

80 70 60 50 3 c_

t

5

40

30 20 10

0

0

60

I20

Longitude

FIG. 25. GFDL model simulation of the change in precipitation rate (millimeters/day) for quadrupled C 0 2 . Dense shading indicatesa rate decrease, sparse shading an increase between 2 and 5 mm/day, and hatching an increase larger than 5 mm/day. [From Manabe and Wetherald ( 1 980); unpublishedresults.]

regions of increased and decreased precipitation over the southeastern section of the continent. However, the increased precipitation that occurred over the southwestern coast in the doubling is replaced with a weak decrease in the quadrupling. It is evident that there are other differences between the two simulations, and that the changes shown for the quadrupling are not simply twice those forthe doubling. There is more similarity for the changes in the zonal means for the quadrupling and doubling, but, again, the former is not simply twice the latter. Figure 26 shows the geographical distributions of the change in the annual-mean precipitation rates simulated by the annual and seasonal models of Wetherald and Manabe ( 1981) for the COz quadrupling. A comparison of these results reveals several interesting similarities and differences. First, there are only increases in the precipitation rate poleward of about 45" latitude in both simulations, and there are both decreases and increases equatorward of this latitude. Second, both the annual and seasonal simulations display reduced precipitation rates in midlatitudes over the eastern ocean and western continent, as well as over the western part of the ocean in

197

MODELS OF C02-INDUCED CLIMATIC CHANGE

0

60

I20

Longitude

FIG.26. GFDL model simulations of the change in precipitation rate (millimeters/day)for quadrupled C02. Top, annual model;bottom, seasonal model. Shadingas in Fig. 25. [From Wetherald and Manabe (1 98 I); unpublished results.]

198

MICHAEL E. SCHLESINGER

low latitudes. However, the seasonal model simulates increased precipitation rates everywhere over the continent in low latitudes, as well as over the eastern ocean, while the annual model does not. These similarities and differences are reflected in the changes in the zonal-mean precipitation rate shown in Fig. 28 (upper right panel). In the equatorial region there is an increase in the zonal mean in the seasonal model simulation, in qualitative agreement with the larger increase in the simulation with the annual model. Also, both model simulationsshow increased zonal-mean precipitation rates in the subtropicsand high latitudes,while only the annual model simulation has a decrease in midlatitudes. Undoubtedly, some of the differencesin the zonal-mean precipitation rate reflect the considerably smaller changes seen in the simulation with the seasonal model compared with those for the annual model simulation (Fig. 26). This does not mean that the changes in the seasonal simulation are small throughout the year; rather, it indicates that there is cancellation due to the seasonal variation of the changes. (This is evidenced by the latitude-time distributions shown by Wetherald and Manabe, 1981 .) A similar effect was seen in the seasonal model simulation of surface air temperature in comparisonwith that for the annual model (Fig. 17). The results for the precipitation rate also indicate that there is a nonnegligible influence of the seasons on the annual-mean climate. The geographicaldistribution of the change in the annual-mean precipitation rate simulated by the seasonal model of Manabe and Stouffer ( 1980)is shown in Fig. 27. [The hydrological aspects of this simulation are also referred to in the following discussion as those of Manabe et al. (1981).]

60E

120E

I80

120w

60W

0

FIG. 27. GFDL model simulation of the change in the annual-mean precipitation rate (miIlimeters/day) for quadrupled C02. Shading indicates a rate decrease. [From Manabe and Stouffer ( 1 980); Manabe et a/. (1981); unpublished results.]

MODELS OF CO2-INDUCED CLIMATIC CHANGE

199

This figureshows the predominance of precipitation rate increases poleward of about 45 latitude in both hemispheres. This agrees with the results of the GFDL sector models (Figs. 22, 25, and 26), but not with that of the OSU model (Fig. 23), and is also apparent in the increased zonal-mean precipitation rates in high northern and southern latitudes shown in Fig. 28 (lower right panel). Figure 27 also shows regions of both decreased and increased precipitation rates between about 45"N and 45"S, with the largest changes between 30"N and 30"slatitudes, in agreement with the OSU global model and the GFDL sector models. The magnitude of the largest changes in Fig. 27 for the CO, quadrupling are smaller than those of the OSU model for doubling (Fig. 23). As suggested by the discussion in the preceding paragraph, this probably occurs because the GFDL simulation is seasonal, while the OSU simulation is not. However, it is interesting that the location and sign of several of the extreme precipitation rate changes agree in the two simulations. For example, both models simulate increased precipitation over central east Africa and decreased precipitation extending southeastward from the equator at 60 'E longitude. There are also many differences between the two simulations;for example, over most of the United Statesthe OSU model simulates an increase, and the GFDL model a decrease, in the precipitation rate. Figure 28 (lower right panel) shows that the changes in the zonal-mean precipitation rate simulated by the global GFDL model (Manabe et al., 1981) are positive almost everywhere and, in particular, there is again no large decrease near the equator as found by Manabe and Wetherald (1980). Similar changes are seen for the NCAR model simulations with both computed and fixed clouds (lowerleft panel). (The geographicaldistribution of the change in precipitation rate was not available for the NCAR model.) In contrast, the change in zonal-mean precipitation rate simulated by the OSU model with prescribed SST and sea ice (Gates et al., 1981;Fig. 28, upper right panel) is again negative almost everywhere. O

3.3.3. Comparison of Global-Mean Precipitation Rate Changesfor Doubled and Quadrupled CO,. A summary of the precipitation rate changes simulated by eight GCMs for C02 doubling and quadrupling is shown in Table IX in terms of the area-averaged precipitation rate. As in Table VII, the annual-mean change is presented in Table IX for the seasonal models of Manabe and Stouffer (1980), Wetherald and Manabe (198 l), Mitchell (1983), and Gates et al. (198 1). The changesin precipitation rate simulated by the models are also shown in Table IX as a percentage of the corresponding control precipitation rate. The latter is also shown in the table, along with its difference from the observed global-mean precipitation rate.

TABLEIX. AREA-MEAN ANNUAL-MEAN PRECIPITATION RATEAND C 0 2 - I PRECIPITATION ~ ~ ~ ~ RATE ~ ~ CHANGE

1

8

N

xco*

1XC0,-0BS"

2 x c 0 , - 1 XCO, (mm/day)

100

2 x c 0 , - I XCO, 1 x CO,

4 X C 0 2 - 1 XCO2

4 x c 0 , - 1 XCO, 1 XCO,

Model

(mm/day)

(mm/day)

Manabe and Wetherald

2.55

-0.10

0.20

7.8

-

2.58

-0.07

0.18

7.0

0.30

11.6

0.18

6.7

0.30 0.24

12.8 10.0

(%)

(mm/day)

I00

(%)

(1975) [8Jb (300, 600,

-Y

Manabe and Wetherald (1 980) [24] (300, 600,

1200) Manabe and Stouffer

2.69

0.04

( 1980) [20]

(300, -, 1200)

Wetherald and Manabe (1981y 1301 (300. -, 1200)

2.35 2.40

-0.30 -0.25

N

2

Washington and Meehl (1983b3)’ [331(330, 660, 1320) Mitchell (1983) [28] (320.7, 64 1.4, -) Gates et al. (1981) [25] (326, 644, 1289) Schlesinger (1983b) [321(326, 644,-)

0.12 0.10

6.0 6.5

1.01 1.02

2.83

0.18

-0.07

-2.5

-

-

2.69

0.04

-0.04

- 1.5

-0.09

- 3.3

2.13

0.08

5. I

-

0.14

Observed value of 2.65 mm/day from Jaeger (1976). These numbers correspond to those in Tables IV and VII and Figs. 29 and 30. CO, concentrations in ppm for 1 X CO,, 2 X C 0 2 ,and 4 X CO, . First row is for annual-mean insolation, second row is for seasonal insolation. First row is for predicted clouds, second row is for prescribed clouds.

3.3 2.7

0.22 0.24

3.66 3.67

-

202

MICHAEL E. SCHLESINGER

IA

t 0.8

f$L

0.6 0.4

0.2 0

.... .. .

-0.2

.

. .....,....:

-0.4 -

0” 9

-06-0.8-

v

P

a

-1.0-

-ISl 90N

’

70

I

50

’

30

I

ION

1

IOS

30

’

50

I

)S

70 D

-

-

1.0

0.8-

-

0.8-

1.0

0

\I ..’

-

-0.2

q!7

”

’

l

i

i

l

’

-

0

- -0.2-

‘,’

I

ii’

L’

-0.4

FIG.28. The change in zonal-rnean precipitation rate (AP)simulated by seven GCMs for quadrupled C 0 2 . The data of (A) Manabe and Wetherald ( I 980, curve a) and (B) Wetherald and Manabe ( 1981;curves a and b are seasonal and annual, respectively) are plotted symmetrically about the equator. Other curves: (A) Gates ef a/. (1981), curve b; (C) Washington and Meehl(1983b), predicted (a) and prescribed (b) clouds; (D) Manabe et al. (1981).

203

MODELS OF CO2-INDUCED CLIMATIC CHANGE

The precipitation rate Pfor each control simulation is shown plotted in the lower panel of Fig. 29 versus its corresponding surface air temperature T,. The observed values of precipitation rate and surface air temperature are denoted by "+OBS." It is evident that the seasonal simulations with the global GFDL model of Manabe and Stouffer( 1980)[20] and the OSU model of Gates et al. (1981) [25] are identical and closest to the observed climate, albeit the OSU model used prescribed SST and sea ice whereas the GFDL model predicted the SST from a 68-m mixed-layer wean model and the sea

4.0

x

0

I

I

3.6 -

I

I

I

t

I

I

I

I

I

I

1

-

B

3.2 - -2 2.8 -

E

8

+*20,25

-

2.0

-

028

-

*32 8..24

OBS

a: 2.4 -

-

-

*33

030 30

'

"

~

~

'

'

'

'

~

'

-

'

'

'

FiG. 29. Top: the change in global-mean surface air temperatureAT, induced by doubled (e) and quadrupled (+) C02 versus the global-mean surface air temperature Ts of the control simulation for eight GCMs. Bottom: the global-mean precipitation rate P of the control simulation versus the global-mean surface air temperature of the control simulation for eight GCMs. +OBS, Observed value.

204

MICHAEL E. SCHLESINGER

ice from a thermodynamic model. The seasonal and annual simulationsof Wetherald and Manabe (198 1) [30] employed the same ocean and sea ice models as did the simulations of Manabe and Stouffer [20], but Wetherald and Manabe’s simulated T,and Pare warmer and smaller, respectively,than the observed values. This increased discrepancy of the model [30] simulation compared with that of 1201 is likely, in part, the result of [30] using a sector model in contrast to the global model of [20]. When the mixed-layer Ocean and thermodynamic sea ice models are replaced with a swamp ocean in the sector model, as in [24] (Manabe and Wetherald, 1980),the temperature error increases considerably while the precipitation error decreases. A more accurate portrayal of the present climate with a swamp model, at least insofar as temperature, is given by the OSU model [32] (Schlesinger, 1983b). However, it is seen that the NCAR simulations with a swamp model for both computed and fixed clouds [33] have relatively larger temperature and precipitation errors. The results shown in the bottom panel of Fig. 29 suggest a decreasing relation between the precipitation and surfaceair temperature in the control simulations; that is, in the sense of intermodel comparison, the warmer the control, the smaller the control precipitation rate. Table IX shows that, with the exceptions of the UKMO and OSU models with prescribed SST and sea ice (Mitchell, 1983;Gates et al., 198l), all of the models simulate an increase in the global-mean precipitation rate for both doubled and quadrupled C 0 2 . The decreased precipitation rates simulated by the models in which the SST is prevented from warmingin response to the enhanced downward IR radiation from the increased CO, may occur because the surface evaporation is thereby inhibited from increasing while the atmosphere warms slightly (see Table VII). The joint effect would be to reduce the relative humidity of the atmosphere and thereby reduce at least the nonconvective precipitation. Table IX shows for doubled CO, that the NCAR model simulation with prescribed clouds (Washington and Meehl, 1983b) gives the smallest increase in the global mean precipitation rate. The increase is slightly larger when clouds are predicted in the NCAR model. The largest increase in the global-mean precipitation rate is simulated by the GFDL model of Manabe and Wetherald (1975), while the OSU model of Schlesinger (1983b) simulates an increase somewhat larger than that of the NCAR model with predicted clouds. If the simulated precipitation rate increases A P for the doubling and quadrupling are plotted versus their corresponding surface air temperature increases ATs, as shown in the top panel of Fig. 30, it can be seen that there is an increasing relation. That is, in the sense of intermodel comparison, the larger the simulated warming, the largerthe increase in the simulated precip-

205

MODELS OF CO2-INDUCED CLIMATIC CHANGE I

I

I

'

324 %

30 +20

+33 24.

08

+33

.32

-

5:

:8 1.25

4.0

h 0

33.

I

l-

-

*33

I

I

I

I

I

I

I

I

I

I

I

1

3.6 -

33.

033

0 , 3.2 -

-

Y

2

E 0

-

2.8

-+25

OZ8

+20

32.

.25

-

u

249

a- 2.4-

-

2.o

I

I

I

I

+24

08

I

-

-

33+ +33

+3O

.30

I

I

-

-

in global-mean precipitation rate induced by doubled (0)and quadrupled (+) C02 for eight GCMs.

itation rate. Because the global-mean surface evaporation rate must equal the global-mean precipitation rate for an equilibrium simulation,the above relation impIies an identical relatioc between the surface evaporation increase and the surface air temperature warming. Both relations are likely the result of the increase of the surface saturation mixing ratio with temperature (due to the Clausius-Clapeyron relation between saturation vapor pressure and temperature) and, presumably, a smaller increase in the water

206

MICHAEL E. SCHLESINGER

vapor mixing ratio of the surface air. The latter needs to be verified by further analysis of the models' results. If the preceding relation between A P and AT, for any particular model were linear, then the ratio of A P for quadrupling to A P for doubling would equal the ratio of AT, for quadrupling to AT, for doubling. (This assumes that A P = 0 for AT, = 0.) The results for the Manabe and Wetherald (1980) doubling and quadrupling simulations, aswell as those for the NCAR model, show that this is not the case. This indicates, therefore, that the APand AT, relation is nonlinear, at least for these models. Since A P increases with AT,, and AT, increaseswith T, of the control, AP must increase with T,. In other words, again in the sense of intermodel comparison, the warmer the control, the larger the C02-induced precipitation rate increase. For this reason the precipitation rate increase simulated by the seasonal model of Manabe and Stouffer (1980) [20] is smaller than that of the annual model of Manabe and Wetherald (1980) [24], and similarly for the seasonaland annual models of Wetherald and Manabe ( 1981) [301. As already discussed, the bottom panel of Fig. 29 shows that P of the control simulation tends to decreasewith increasing T, of the control. Since A P increaseswith T, ,there is a tendency for A P to decrease with increasingP of the control. This is substantiatedby the bottom panel of Fig. 30. (There must be separate relations for the doubling and quadrupling.) As a result, in the sense of intermodel cornparison, there is a tendency toward the relation that the warmer the control, the larger the percentage increase in the precipitation rate. 3.4. Simulated Soil Moisture Changes

3.4.1. CO,Doubling. The geographical distributions of the soil moisture change simulated by the sector models of Manabe and Wetherald (1975, 1980) are presented in Fig. 3 1. This figure shows that in both simulations the soil moisture decreased almost everywhere poleward of 35 latitude and increased over most of the continent equatorward of this l a t i t ~ d e . The ~ maximum drying of the soil occurs in a band that stretches from coast to coast centered near 35" and 40" latitude in the 1975 and 1980 simulations, respectively. This drying band in each simulation is the most prominent feature of the zonal-mean soil moisture change shown in Fig. 33 (from data over the continent only). This latter figure also reveals the weak drying in high latitudes and the larger moistening in the subtropics and tropics. The changes of soil moisturein high latitudesin the annual-meansectormodelsmay not be a meaningful indicator of hydrologic change there because this area is covered by snow (R. T. Wetherald, personal communication, 1982).

MODELS OF CO2-INDUCED CLIMATIC CHANGE

207

90" SO"

70°

70"

60"

60'

2

50'

c I-

5 40°

3 0"

30'

zoo

20"

10"

10"

0"

0" 0"

LONGITUDE

60'

0"

LONGITUDE

FiG. 31. GFDL model simulation of the change in soil moisture (centimeters)for doubled CO,. Sparse shading indicates a decrease smaller than 0.5 cm, dense shading a decrease larger than 0.5 em. Left, data from 1975; right, 1980. (From Manabe and Wetherald [ 1975 (unpublished results), 19801.)

Figure 3 1 shows a soil moisture decrease over the southeastern coast in the 1975 simulation, but a soil moisture increase in the 1980 simulation. A comparison of Fig. 31 with Fig. 9 shows that there is a tendency for the tropical and midlatitude regions of minimum and maximum warming to occur where the soil is moistened and dried, respectively. Similarly,there is a tendency for the tropical and midlatitude regions of increased and decreased soil moisture to occur where the precipitation rate increased and decreased, respectively, as can be seen by comparing Figs. 3 1 and 22. Although the above relation between the changes in precipitationrate and soil moisture appears to be intuitively correct, that it must hold it is not entirely evident from the governing soil moisture balance equation

awiat = P,+ s, - E, - R

(3.1) Here W is the soil moisture, which changes in time (awlat)due to (1) the and by the melting addition of water by the precipitation that fallsas rain (P,)

208

MICHAEL E. SCHLESINGER

of snow (S,) and (2) the subtraction of water by the surface evaporation (E,) and runoff (R).In equilibrium, the time-averaged (denoted by the overbar) soil moisture (@) is constant @%/at = 0), and the time-averaged soil moisture sources and sinks must balance, that is

F, +s,= E, + R

When the CO, level is changed there is a change (denoted by A) in the equilibrium climate. The difference between the equilibrium climates, insofar as the soil moisture balance is concerned, is given by Eq. (3.2) as or

AF, + A3m = AE, + A R

(3.3) In other words, the changes in the soil moisture sources and sinks must balance. However, Eq. (3.3) does not give any information about the change in the equilibrium soil moisture A E That information can be obtained only as a result of the integrated effect of the soil moisture sources and sinks over time as the climate changes from equilibrium state 1 to equilibrium state 2 as a result of increased CO,. That is, from Eq. (3.1)

Consequently,it is interesting that there appears to be a positive correlation between A@and AF,. What may be occumng is the following relation. and runoff El are both zero Suppose in equilibrium state 1 the snowmelt so that by Eq. (3.2) Es,l= and @ is constant. Consider now that the CO, concentration is changed and in response P, changes to PI, A& but Es remains essentially equal to Esl. Then by Eq. (3.1) aW/dt = Pr,! AP, - gA, = AP,, and the soil moisture must change in the same direction as AP,. As this occurs, E, must change in the same direction toward F,,I AP, so that a new equilibrium can be reached in which Es,2= and g2is constant. Thus, if AF, S 0, then A @ 2 0 and AE, S 0. Since evaporation acts to cool the surface, these changes should result in ATg S 0, where Tgis the ground temperature. Figure 32 shows the geographical distribution of the change in soil moisture simulated by the OSU model of Schlesinger (1983b). This figure shows a small moistening of the soil over most of the contiguous United States, western Europe, central Asia, and northern Australia, and a larger increase in the soil moisture over the Plateau of Tibet and in central east Africa. A large drying of the soil is simulated in Canada, north Africa, and to the east of the Himalaya Mountains, and smaller drying in Mexico, central Europe to

sm,

+

+

+

209

MODELS OF C02-INDUCED CLIMATIC CHANGE 90 N 70 N 50 N 30 N ION

10s 30 S

50 S

70 s 90s 180

l50W 12OW

9OW

60W

30W

0

30E

60E

90E

120E

150E

180E

FIG. 32. OSU model simulation of the change in soil moisture (centimeters)for doubled CO, . Dense shading indicates a decrease, sparse shading an increase. The unshaded regions indicate either ocean or glacier ice. [From Schlesinger (1983b).]

western Asia, and southern Australia. By comparing Fig. 32 with Fig. 10 it can be seen that the changes in surface air temperature tend to be negatively correlated with the changes in soil moisture; that is, regions of maximum warming often occur where the soil moisture is decreased, particularly over the deserts, while regions of reduced warming or even cooling occur where the soil is moistened. The latter is particularly evident in central east Africa, where the surface air temperature actually decreased as a consequenceofthe enhanced evaporation (not shown) over the moistened soil. A comparison ofFig. 32 with Fig. 23 also showsthe tendency for the change in soil moisture to be positively correlated with the change in precipitation rate. For example, both the precipitation rate and soil moisture decreased over north Africa, and both increased substantially over central east Africa. These correlations between ATs and A Wand between A Wand AFr are the same as evidenced by the sector model of Manabe and Wetherald (1980). The changes in the zonal-mean soil moisture over land simulated by the GFDL and OSU models are presented in Fig. 33. A comparison of the results of the 1975and 1980sector model simulationsof Manabe and Wetherald shows qualitative agreement in the tropics and subtropics, where the soil moisture increased in response to the doubled CO,, and quantitative agreement in the position and intensity of the midlatitude desiccation region. However, differences in intensity and sign are evident in high latitudes. These may reflect the different land/ocean geometries poleward of 66.5" latitude in these sector models. For the OSU model the soil moisture increasein east Africa and, to a lesser extent, in northern Australialeadsto an

2 10

MICHAEL E. SCHLESINGER 2.4

-2.4 1 90N

I

I

70

I

I

I

50

I

30

I

I ION

I

I

IOS

I

I

I

30

I

50

I

I

70

s

Loti tude

FIG.33. The change in zonal-meansoil moisture (over land only) simulatedby three GCMs for doubledCO, . The data of Manabe and Wetherald ( 1975, curve a; 1980,curve b) are plotted symmetricallyabout the equator. Curve c, Schlesinger (1 983b). (All data are unpublished results.)

increase in the zonal-mean soil moisture in the tropics. This generally agrees with the result of the GFDL models. The zonal-mean soil moisture in the OSU simulation decreased near 30"Ndue to the desiccation in the Sahara and decreased near 40"s due to the drying in South Africa and southern Australia. There is an increase in soil moisture at 50"s that is solely due to the change in the single grid box in South America at this latitude, and an increase near 35"Npredominantelydue to the increase over the United States and Himalayas. This increase is in marked contrast to the midlatitudedrying simulated by the GFDL model. . However, both the OSU and the GFDL model simulate drying over most of the high latitudes of the Northern Hemisphere.

FIG.34. Simulation of the change in soil moisture for doubled C02given by the NCAR model with fixed clouds. Crosshatchingindicates regions where there was a decrease in soil moisture in the experiment compared to the control during all seven 30day segments of a 2 loday period near the end of the integration. Hatching indicates regions where six of seven segments were drier, and stippling, five of seven segments. Dashed-line contours indicate regions of increased soil moisture during at least five of the seven 30-day segments. [From Washington and Meehl(1983b).]

212

MICHAEL E. SCHLESINGER

The change in soil moisture simulated by the NCAR model with fixed clouds (Washington and Meehl, 1983b) is shown in Fig. 34. This figure shows the regions of persistent drying and moistening rather than the actual values of the soil moisture changes, as in Figs. 3 1 and 32. Figure 34 shows that persistent soil moisture increases (areas within dashed contours) occurred over north central Siberia and Canada, northwestern Mexico, and southwestern United States, as well as the Gulf states. Persistent soil moisture decreases (crosshatching,hatching, and stipple)are located over Australia, the Amazon River Basin, the central eastern United States, central Europe, most of Asia, and Ethiopia. A comparison of Fig. 34 with Fig. 32 indicates that the changes in soil moisture simulated by the NCAR model (with fixed clouds) and the OSU model (with computed clouds) are of the same sign in some regions (southern Australia, the Amazon River Basin, central eastern United States, eastern Canada, Sahara, central Europe,

LONGITUDE

FIG.35. GFDL model simulation ofthe change in soil moisture(centimeters)for quadrupled CO,. Sparse shading indicatesa decrease smaller than 0.5 cm, dense shading a decrease larger than 0.5 cm. [From Manabe and Wetherald (1980).]

MODELS OF CO1-INDUCED CLIMATIC CHANGE

213

northeast of the Caspian Sea, and Southeast Asia), and of opposite sign in others (northern Australia, Peru, and southwestern United States).

3.4.2. CO, Quadrupling. Figure 35 shows the change in soil moisture simulated by the sector model of Manabe and Wetherald (1980)for quadrupled CO, . Comparing this figure with Fig. 3 1 for the doubling shows that there are many similarities: in particular, the general decrease of soil moisture poleward of about 35 latitude, the band of strong drying centered near 40"latitude stretching across the continent, and the general increase in soil moisture equatorward of 35 latitude, particularly along the east coast. The principal differences between the soil moisture changes induced by quadrupled and doubled CO, are found in the polar region, where there is an O

O

90'

Longitude

60'

0'

0'

Longitude

FIG.36. GFDL model simulations of the change in soil moisture (centimeters)for quadrupled C02. (a) Annual model; (b) seasonal model. Stipple indicates a decrease. [From Wetherald and Manabe (198 1); unpublished results.]

60'

214

MICHAEL E. SCHLESINGER

increase for the quadrupling but not for the doubling (see footnote on p. 206) and along the west coast in the tropics and subtropics, where there is a drying for the quadrupling but generally not for the doubling. The latter differenceis also seen for the precipitation rate, particularly near the equator where there was an increase for the doubling (Fig. 22) but a decrease for the quadrupling(Fig. 25). Finally, it is evident that the changes in soil moisture for the quadruplingare in magnitude quite similar to those for doubling, and are certainly not twice as large. The changes in the annual-mean soil moisture simulated by the annual and seasonal models of Wetherald and Manabe (198 I) are shown in Fig. 36. Comparison of these results reveals several notable differences. First, the seasonal model simulatesa very large moistening in high latitudesin contrast to the small drying simulated by the annual model. Second, the intense midlatitude drying belt in the annual model simulation is broader and much stronger than that in the seasonal model simulation. Finally, the extremes of drying and moistening simulated by the seasonal model in the tropics, particularly over the east coast, are weaker than those simulated by the annual model. The smoothing and weakening of the changes in soil moisture in low and middle latitudes that are simulatedby the seasonal model in

60

S

30

0

J

F

Y

A

M

J

J

A

S

O

N

D

J

MONTHS

FIG.37. GFDL model simulation of the seasonal variation of the change in (a) the zonalmean P - E(centimeters/day)and @) soil moisture (centimeters) for quadrupled CO,. [From Wetherald and Manabe ( 198 1 ).]

-

6,,

POS

-

Higher resolution model (621)

-

FIG.38. GFDL model simulation of the change in soil moisture (centimeters) for quadrupled CO,. The upper and lower panels (a and b) are simulations with 15 and 2 1 waves, respectively (for both longitude and latitude; see Table VI), and the left and right panels are for northern hemisphere spring (March, April, May) and summer (June,July, August). [From Manabe et ul. (198 I).]

216

MICHAEL E. SCHLESINGER

comparison with the annual model were also found for the surface air temperature (Fig. 17)and the precipitation rate (Fig. 26). What was not shown in those temperature and precipitation comparisons, however, is the reversal of sign and amplification exhibited by the soil moisture changes in high latitudes. As shown in the bottom panel of Fig. 37, this is the result of the large asymmetry in the seasonal variation of the soil moisture changes in high latitudes, with the large increases from October through April dominating the small decreasesfrom May through September. This summer drying is also simulated by the global seasonal model of Manabe and Stouffer (1 980) over both middle and high latitudes of Asia and North America, as shown in Fig. 38.

4. DISCUSSION

In the preceding section we compared the changesin temperature, precipitation rate, and soil moisture induced by doubled and quadrupled CO, concentrations as simulated by atmospheric GCMs coupled to two different models of the ocean. For the first, or swamp ocean model, the simulations were performed only for annual-mean insolation; for the second, or slab mixed-layer ocean model, the simulations employed the annual cycle of insolation. The comparisons that have been made between these seasonal and annual models clearly show that the annual-mean climate change simulated by a model is dependent upon whether or not the annual cycle of solar forcing is included. For the temperature and precipitation rates there is a general smoothing of the change in the annual mean due to the geographical shift of the changes with season, while this leads to a sign reversal and an intensification for the annual-mean soil moisture change in high latitudes in the sector model simulations of Wetherald and Manabe ( 1981). Moreover, it is the seasonal cycle of the C0,-induced climatic change that is likely to be of importance with respect to the impact on humanity, notjust the change in annual-mean climate. Consequently, future simulations of possible C0,induced climatic changes should be performed with seasonal models. This will require an ocean model other than the swamp model and, therefore, extended integration periods to reach equilibrium. Furthermore, such simulations should be performed with hydrodynamic ocean models so that the oceanic heat transport is included. This may be of particular importance in the calculation of the change in sea ice resulting from increased CO, and, in turn, the poleward amplification of the C0,-induced warming. The above notwithstanding, the simulations of CO,-induced climate change that have been made with atmospheric GCMs coupled with swamp

MODELS OF CO2’INDUCED CLIMATIC CHANGE

217

ocean models have been useful to identify whether the projected increasing level of CO, could cause a climatic change of nontrivial proportion. The utility of these models is that they allow the ocean to participate and feed back on the climate change of the atmosphere, unlike the case when the sea surface temperature and sea ice are prescribed, and they reach their equilibrium climatic state in a relatively short time due to the absence of oceanic heat storage (see later). Furthermore, because such simulations have been carried out with different atmospheric GCMs, it is now possible to identify the model dependenceof the simulated C0,-induced climate changes. This will also be possible to do for atmospheric GCMs coupled with more realistic models of the ocean, but, as noted previously, this will require considerably longer numerical integrations. Consequently, it behooves us to make the maximum possible use of the extant simulationswith swamp ocean models, and to continue to use these models as necessary to understand further the causes for the model dependencies. In the following section we summarize the model-dependent results that are apparent from the preceding comparison, and discuss two possible causes of such model dependence. 4.1, Model-Dependent Results

The comparisons of Section 3 have shown that the simulated changes in the global-mean surface air temperature and precipitation rate induced by doubled and quadrupled CO, are model-dependent results (Table VII, Fig. 2 1, and Table IX). However, when these results are plotted as in Figs. 29 and 30, we see that in the sense of intermodelcomparison (1) the warmer the control, the larger the C0,-induced warming; (2) the warmer the control, the larger the C0,-induced precipitation rate increase; (3) the warmer the control, the greater tendency toward a larger percentage increase in the precipitation rate; and (4) the warmer the control, the smaller the control precipitation rate. These findings suggest that the differences among the models’ sensitivitiesto increased CO, levels, at least insofar as the changes in globalmean temperature and precipitation are concerned, are linked to the differences among the models’ simulations of the control climate. If this is correct, then the models’ sensitivitiesshould convergeas their simulated control climates converge, and it appears that this convergence should happen simultaneously for precipitation and temperature. Figure 29 shows that the controls for models [20] and [25]are quite close to the observed present-day climate. What is remarkable about this is not that the simulated temperatures are identical and closeto the observed, because the sea surfacetemperature was prescribed in model [25] and some tuning was performed in model

218

MICHAEL E. SCHLESINGER

[20] (as evidenced by the choice of mixed-layer depth), but rather that the simulated precipitation rates for both models are also identical and close to the observed value. This suggeststhat if the control temperature of (annual) model [33], for example, were somehow made closer to that of (seasonal) model [20], the control precipitation rate of model [331 would become closer to that of model [20]. Then not only should ATs of model [33] approach that of model [20], but A P of model [33] should also approach that of model [20]. If this conjecture is valid, then the differencesin the models’ sensitivities lies in the differences in the models’ control simulations, not only for temperature, but for precipitation and perhaps other quantities as well. In this case it is of paramount importance to understand the reasons for the differences in the models’ control simulations. Modeldependent results are also exhibited by the zonal-mean climatic changes resulting from CO, concentrations that are doubled and quadrupled. Of particular prominence is the difference in the poleward amplification of the surface temperature warming: the GFDL models give polar warmings about five times greater than those simulated in the tropics, the OSU model about three times, and the NCAR models about three times in the Southern Hemisphere but only about one time in the Northern Hemisphere for doubled CO, . While the lower value of poleward amplification simulated by the OSU model compared with that of the GFDL model could be the result of a weaker or missing inversion due to the OSU model’s coarse vertical resolution, this cannot be the cause of the small poleward amplification of the NCAR model because it has the same vertical resolution as the GFDL model. Alternatively, the comparatively large warming of at least the GFDL sector models may be due to the temperature dependence of the surface albedo for snow and ice, which was not in the NCAR or OSU models. As shown in Table VI and discussed in Section 3.1.4, the albedo in the GFDL sector models decreased discontinuously at some critical temperature below freezing. Consequently, there should be an ice albedo feedback (temperature increase causes a reduction in ice/snow and thus a decrease in the surface albedo, which causes an increase in absorbed solar radiation and thereby a further temperature rise) at the subfreezingcritical temperature in addition to the “normal” ice albedo feedback at 0°C. It should be possible to use an RCM to test whether this gives increased warming compared with the case in which there is no temperature dependence of the ice/snow albedo. However, it is likely that this will not explain all of the poleward amplification since such amplification was also obtained by the global GFDL model (Manabe and Stouffer, 1980), which did not have a temperature-dependent ice/snow albedo (see Table VI). Although comparisons have been made for the global and zonal means without regard to the differences between the geography and orography of

MODELS OF COyINDUCED CLIMATIC CHANGE

219

the sector and global models (see Table VI), it is not really possible to comparethe geographicaldistributions of the C02-inducedclimatic changes simulated by such widely different models. Moreover, the global model simulationperformed at GFDL includes the annual solar cycle, and we have seen that there are appreciable differences between the simulations of the seasonal and annual models with the same or similar geography and orography. Consequently,it is not desirable to compare the geographicaldistributions of the OSU and NCAR annual models with the annual-mean distribution of the GFDL seasonal model. Therefore, the only intermodel comparisons that can be made for the geographical distributions of C02-induced climate change are between the OSU and NCAR models for temperature and soil moisture; the latter can be compared only in a qualitative way because of the different presentations for these models. In the light of the differences between the global- and zonal-mean climate changes simulated by the models, we should not expect good agreement between the models’ geographical distributions of climate change. However, before endeavoring to understand the physical significance of these differences,it is essential to establish that they represent differences between the models’ equilibrium climates and that they are statistically significant. Since these issues are of fundamental importance to the interpretation of all GCM results, they are discussed further in the following section.

4.2. Time Required to Reach Equilibrium Most of the simulationspresented in the preceding discussions were performed with an atmospheric GCM coupled to a swamp ocean model for which there is zero heat capacity. Therefore,the Ocean (and land) surface is in thermal equilibrium at each time step of the numerical integration, and the time required to reach equilibrium depends only on the atmosphere. The conventional wisdom is that this time is not more than about 60 -90 days. To show that this is not the case, Fig. 39 presentsthe time evolution of the differencebetween the experiment and control simulations of the globalmean surface air and mass-averaged temperatures for the OSU model (Schlesinger, 1983b). It is the latter quantity that is of predominant importance in the assessment of whether a simulation with an atmospheric GCM/ swamp ocean model has reached equilibrium. Figure 39 clearly shows that the OSU model simulations did not reach equilibrium by day 100, and are just reaching equilibrium by day 200. The figure suggests that a conservative estimate of the time required for the global-mean temperatures to reach equilibrium is about 300 days. While 300 days appears to be sufficiently long for the global-mean temper-

220

MICHAEL E. SCHLESINGER 3.0I

t-

I

I

I

I

I

I

I

1

I

I

I

TIME, days

FIG.39. Time evolution of the difference between experiment (2 X C 0 2 )and control ( 1 X CO,) simulationsof the global-meansurface air and mass-averagedtemperatures for the OSU atmospheric GCM/swamp Ocean model (Schlesinger,1983b). The average temperatures over the last 180 days are shown by horizontal lines.

atures in an atmosphericGCM/swamp ocean model to reach equilibrium, is this sufficiently long for the geographical distribution to reach equilibrium? To address this question, Fig. 40 shows the time evolution over days 320720 of the difference between the experiment and control surface air temperatures at four geographical locations from the OSU doubling study. These locations correspond to several of the surface air temperature extrema as can be seen by reference to Fig. 10. Although the first 320 days are not shown in Fig. 40,the last 400 days of this simulation pair (experiment minus control) do not indicate any long-term trend at these four pointsas was evident forthe global-mean temperatures during the first 200 days (Fig. 39). This suggests that 300 days is sufficient for the geographical distribution of the C0,-induced temperature changes to reach equilibrium. However, it is clear from Fig. 40, as it was from Fig. 39, that there are both low-frequency and high-frequency oscillationsin some of these temperature records. Consequently, in order to obtain a statisticallysignificant measure of the C0,-induced temperature changes once equilibrium has been reached requires averaging over long periods to reduce the noise. We will focus on this issue in the following section. So far we have considered the time required to reach equilibrium only for the atmosphericGCM/swamp ocean models. When an atmospheric GCM is instead coupled to a slab mixed-layer model, as was done, for example, by Manabe and Stouffer (1980), it is to be expected that the heat capacity of the

22 1

MODELS OF C02-INDUCED CLIMATIC CHANGE

mixed layer will require many years of integration for the ocean to achieve equilibrium. This is of course the reason for the long integrations that have been made with this type of ocean model, as shown in Table V. While it appearsthat these simulationshave been of sufficientlength to reach equilibrium [in particular, see Figs. 4 - 6 of Manabe and Stouffer ( 1980)], the length of their averaging period may not have been lohg enough to establish statistically significant changes for some variables, as is discussed below.

20.

1 0

d o

'.bo'

-do'

'

520

'

'do-bbo'

. .640'

'

'6b'

.

'7-0

0

Time (days)

FIG.40. Time evolution of the difference between experiment (2 X COz)and control (1 X COz)simulations ofthe surface air temperature AT, at four geographicallocations for the OSU model (Schlesinger, 1983b). The averaged temperaturesfor the last 180 days and last 400 days are shown by the horizontal dashed lines. (a) Greenland,60°W, 82"; (b) Sahara, NE Africa, O"E,26"N, (c) Tanzania, 35"E, 10"s;(d) Antarctica, 170"E, 78"s.

222

MICHAEL E. SCHLESINGER

4.3. Statistical Significance

The statistical significance of the changes in surface air temperature, precipitation rate, and soil moisture simulated by the OSU model for doubled CO, has been determined for the global and zonal means and for the geographical distributions. For the surface air temperature and soil moisture changes, the statistical significance has been determined by the parametric time-series modeling approach developed by Katz ( 1980, 1982b) for nonindependent samples of climatic quantities that are continuous in time. In this method the variance V(1XCO,) of the population meanp( 1XCO,) of the control and the variance V(2XC0,) of the population mean p(2XCO2) of the experiment are estimated from the sample variance for each time series by fitting an autoregressive process (AR) of order p to each time series, with 0 I.p 5 5 chosen to minimize a certain statistic (the Bayesian information criterion). The estimated variances for the lXC0, and 2XC02 simulations are then used to generate the signal-to-noiseratio statistic

Z E Z Bm . =

X(2XC0,) - F(lXC0,) {(l/N)[V(2XCO2) V2( lXC0,)])”2

+

(4.1)

where $1 XCO, ) and x(2XC0,) are the Nday sample means of the 1XC02 and 2XC0, simulations, respectively, and the subscript sim is a reminder that the value of Z is calculated from the data of the two simulations. The signal-to-noise ratio statistic 2,under the null hypothesis that the difference p(2XC0,) - p( IXCO,) is zero, has a Gaussian distribution with zero mean and unit variance in the limit as the number of points Nin the time series is indefinitely increased. The significance level Pis defined as the probability that I ZI can be larger than any particular value IZ-1 of the simulationspurely by chance, and can be computed from the Gaussian distribution. For example, for IZ,, I = 1 and 3,l ZI > IZ-1 can occur by chance with a probability P = 3 1.74 and 0.26%, respectively. The confidence interval for the change in the population means can also be calculated. A [loo( 1 - a)]% confidence interval for p(2XC02) - p( 1XCO, ) is (4.2) where Z,/, is determined such that the probability that Z > Z,,, = a/2. For example, a 95% confidence interval gives a = 0.05 and, using the Gaussian distribution, Z,: = 1.96. For the precipitation rate the statistical significance has been determined by the parametric time-seriesmodeling approach developed by Katz ( 1982a, 1983) for climatic quantities whose occurrence is discontinuous in time. The mean and variance for both the control and the experiment are esti-

223

MODELS OF C02-INDUCED CLIMATIC CHANGE

mated from their individual time seriesby estimatingthe mean and variance of the amount of precipitation only on those days with precipitation (wet days), and the mean and variance of the total number of wet days. Because there are no days with negative precipitation, the probability distribution of precipitation amounts on wet days is not Gaussian. However, the probability distribution of the logarithm of precipitation amounts is approximately Gaussian. Therefore the above estimates are obtained for logarithmically transformed precipitation data rather than for the precipitation data themselves. Then the statistical significance test proceeds as given by Eq. (4.1) and the preceding description. However, the corresponding confidence interval for the transformed data that could be obtained from Eq. (4.2) would correspond to the difference of the medians, not the means, of the control and experiment precipitation rates. For this reason it is not dealt with in the following discussion. The statistical significance parameters obtained by the preceding methods for the changes in global-mean surface air temperature, precipitation rate, and soil moisture are presented in Table X. The 2.00"C change in the global-mean surface air temperature induced by doubled COz (the signal) is 32 times larger than the noise (see Figs. 5 and 39), hence the probability that this difference is due to chance is virtually zero. The corresponding 95% confidence interval is 0.12"C, that is, the probability that the simulated global-mean warming is smaller than 1.88"C or larger than 2.12"C is only 5%. Similarly, the change in the logarithmically transformed global-mean precipitation rate is significant at virtually the 0% level. However, the 0.02-cm decrease in the global-mean soil moisture is less than the noise, consequently this change could occur by chance with a probability of 72%. The 95% confidence interval of 0.09 means that the probability that the TABLEX. STATISTICAL SIGNIFICANCE PARAMETERS OF THE CHANGE IN GLOBAL MEAN QUANTITIES SIMULATED BY THE osu MODELFOR DOUBLED COz4

Surface air temperature ("C)

Experiment Control Difference Signal/noise Significance 95% confidence interval

precipitation [lo&(mm/da~)l

Soil moisture (cm)

19.87 (2.5 X 10-2)b 1.05 (3.5 X 10-3)b 3.41 (3.0 X 10-2)b 17.88 (5.7 X 10-2)b 1.00 (3.6 X 3.43 (3.5 X 10-2)b 2.00 0.05 -0.02 32.1 9.6 -0.4 0.0% 0.0% 7 I .9% 0.12 No estimate 0.09

a From Schlesinger (1983b). The results are for the last 180 days of the 72Oday simulation. The first and second numbers are the mean and estimated standard deviationof the mean.

224

MICHAEL E. SCHLESINGER 20 I6

-

c- I2

Q

L

Z B

N

4

c

0 I00

0) c

-g Y

80

J

W

> 60 W

-I

w

U

z

40

a

u

k

z 2 v)

20 0 90N

70

50

30

ION

IOS

30

50

70

90s

Latitude

FIG.41. Top: the absolute value of the signal-to-noise ratio for the zonal-mean surface air temperaturechanges simulated for doubled C02by the OSU atmosphericGCM/swamp ocean model (Schlesinger, I983b). Bottom: the significance level ofchanges in zonal-mean precipitation rate (transformed}and soil moisture. Results are for averagesover the last 180 days of the 720-day simulation.

change is smaller than -0.1 1 or larger than +0.07 is 5%. Because this interval includes zero, that is, no change, the simulated change in globalmean soil moisture is not statistically significant. We now consider the statistical significance of the changes in the zonalmean surface air temperature, precipitation rate, and soil moisture. The top panel of Fig. 4 1 shows the absolute value of the signal-to-noise ratio for the zonal-mean surface air temperature. Because the values are everywhere greater than three except at the South Pole, the corresponding changes are significant at better than the 0.1% level. The bottom panel of Fig. 4 1 shows directly the significance levels for the changes in zonal-mean precipitation rate and soil moisture. This shows that the precipitation rate changes are not everywhere significant at the 10%level or less, and the soil moisture changes are significant at the 10% level almost nowhere. In other words,

MODELS OF C02-INDUCED CLIMATIC CHANGE

225

there is a degradation in the statistical significance of the changesin the zonal means compared to that of the global means, at least for precipitation rate. The statisticalsignificanceof the simulated changes is further degraded for the geographical distributions, as is evident from Fig. 42. In this figure the regions where the significance level P I 1% are unshaded, the regions where 1Yo IP I10%are lightly shaded, and the regions where P > 10%are heavily shaded. The top panel is the significancelevel of the surface air temperature change shown in Fig. 10. It can be seen that most of the simulated temperature changes are significant at below the 1% level over most of the ocean, while many of the simulated changesover land are above the 109'0 level. It is particularly noteworthy that the cooling over central east Africa is not statistically significant. The data for Tanzania in Fig. 40c show that this is the case because of the large noise in the 180-dayaverage caused by the low-frequency variations in the time series. Returning to Fig. 42, the middle panel shows the significancelevel of the (transformed) precipitation rate change shown in Fig. 23. Here it can be seen that the significancelevel is higher than 10%almost everywhere, indicating that the changes are not statistically significant. It is interesting, however, that the large increase in precipitation rate simulated over central east Africa is significant at below the lYo level. This is also true for the increased soil moisture in this region (Fig. 32), as well as for the drying in north Africa. The changes in soil moisture almost everywhereelse over the continents are not statistically significant. The point that must be taken from Fig. 42 is that it is senselessto compare the climate changes simulated by different models without first establishing the statistical significance of those simulated climate changes. Otherwise, we may simply be comparing the models' noise levels, a not very fruitful exercise. But, having performed the analysis of statistical significance as illustrated in Fig. 42, what should be done about the nonsignificant changes such as for surface air temperature in central east Africa? It may be that although the simulated change is not statistically significant, it is too small based on some other criterion to be of interest even if its statistical significance were established somehow. In this case, we can simply not continue the analysis. On the other hand, if the simulated change is of such a magnitude as to be of interest, then its statistical significance can be established by reducing the noise through extending the averaging period. How long an averaging period is required to reduce the noise, say, by a factor of two? The conventional wisdom would suggest a fourfold increase in the averagingperiod, at least for a reasonably well-behaved quantity such as temperature. To illustrate this noise reduction by increased averaging period, Fig. 43 shows the significancelevel of a 400-day period in compari-

081

30EI

3021

309

306

309

306

30E

I

I

M06

M09

MOE

0

M06

M09

MOE

0

MOZI MOEL

I

081 SO6

SOL

SOE

s OE .sot NO1

.NO€ N OE a N OL N 06

081

I

1

I

3021

30SI

1

30E

I

I

I

MOZl MOE!

I

I

081

I NO6

221

MODELS OF C02-INDUCED CLIMATIC CHANGE I

..

I

I

I

I

I

I

9dW

6dW

3dW

b

36E

I

I

I

1

I

1

I

1

6i)E

96E

12bE

l5bE

1

I

1

I

I

1

70N 50 N 3 0N ION

10s 30 S 50s

70 S 90 s 180 9aNI

156W l26W I

1

180 4

I

4?

50 7 0NN k ,

.:'io..> 30Ni ION

10s

0

30 S

50 S

70s 90s

180

150W 120W

90W

60W

30W

0

30E

60E

90E

120E

150E

180

FIG.43. The significance level (percentage) of the change in surface air temperature simulated by the OSU atmospheric GCM/swamp ocean model for doubled COz. The averaging period is the last 400 days (top) and last 180days (bottom) of the 720-day simulation. Shading as in Fig. 42.

son with that of a 180-dayperiod for surface air temperature. As is evident, there is a reduction in the significancelevel over the oceans and most of the continents. But it is noteworthy that more than doubling the averaging period does little to improve the statistical significanceover several regions, FIG.42. The significance level (percentage) of the change in surface air temperature (top), precipitation rate (transformed) (middle), and soil moisture (bottom) simulated by the OSU atmospheric GCM/swamp Ocean model for doubled C02. The averagingperiod is the last 180 days ofthe 72O-day simulation. Dense stipple indicates a level greater than 10%; sparsestipple, between 1 and 10%;and no stipple, less than 1%.

228

MICHAEL E. SCHLESINGER

both where the signal is small and large, as, for example, over the United States and central east Africa, respectively. And, while it might be argued that this occurs for surface air temperature because of the absence of thermal inertia for the earth’s surface by the assignment of zero heat capacity everywhere, there is also little improvement for the soil moisture that has hydrologic inertia and for the precipitation rate (not shown). In summary, it may require very long integrationsto establish the statistical significance of the climatic change simulated for increased CO, levels. Even the 8-yr analysis period of the Wetherald and Manabe (1981) simulation may not be sufficientlylong to establish the statistical significanceof the geographical distributionsof many ofthe climaticquantities of interest. For example, if the statistical significanceof the simulated changes for a particular month is desired, then there are at most 248 data points in an 8-yr record. This is less than the 400 data points used to create the top panel of Fig. 43. The number of data points can of course be increased by extending the period from a month to a season or longer. But then the seasonal cycle must be removed lest it increase the variance. The analysis of the statistical significance of multiyear simulations deserves increased attention. 5 . CONCLUSIONS AND RECOMMENDATIONS

As stated in Section 1, the object of this article is to formulate and describe the current issues concerning the study of possible C0,-induced climatic change by the physical method, that is, by the use of mathematical climate models. In this article we have focused on the general circulation models and their simulations of C0,-induced climatic change because it is the geographical distribution of that change that is of importance to humanity, and because only the GCMs simulate that geographical distribution. The equilibrium simulations of eight GCMs for both doubled and quadrupled concentrations of CO, have been considered and the geographical distributions, zonal means, and global means of the C0,-induced changes in surface air temperature, precipitation rate, and soil moisture have been compared. While these comparisons reveal similarities and differences among the models’ simulations, it may be premature to draw firm conclusions at this time. The reasons for this are as follows. First, the differences between the models’ geography/orography (sector with idealized land/sea geography at zero elevation versus realistic land/sea geography and realistic orography), ocean treatment (swamp ocean versus slab mixed-layer ocean), solar forcing (annually averaged insolation versus the annual cycle), and vertical resolution (number of levels in the vertical) reduce the comparisons that can be rigorously made between the models’ simulations. Second, it may be that

MODELS OF CO2-INDUCED CLIMATIC CHANGE

229

some of the simulations were not run sufficiently long for equilibrium to have been reached by the time at which the averaging of the results was begun. Finally, it is likely that not all of the simulated climatic changes are statistically significant. The following recommendations are made to reduce these problems in comparing the GCM simulations of C0,-induced climate change: 1. The existing simulations should be extended as required to ensure that they reach equilibrium. 2. The statistical significanceof the C0,-induced climate changes should be determined for the existing or extended simulations. 3. The simulations should be extended further as required to obtain statistically significant results or to decide that the nonstatistically significant changes are too small to be of interest, even if they were subsequentlyshown to be statistically significant. 4. The comparison of this article should be expanded to include other climatic quantities, in particular the cryospheric quantities of sea ice and snow. 5. EBMs and RCMs should be used where possible, and GCMs where necessary, to perform studies to understand the causes for the differences among the existing (or extended) simulations. 6. Seasonal model simulationsof C0,-induced climatic change should be performed with models other than the GFDL model, and the comparison of this article and recommendations (1) through ( 5 ) should be repeated as necessary. 7. Seasonal model simulationsof C0,-induced climatic change should be performed with coupled atmosphere/ocean GCMs to incorporate the oceanic horizontal and vertical heat transports, and such simulationsshould be compared with each other and with the simulations made with the simpler ocean models.

ACKNOWLEDGMENTS I would like to thank Syukuro Manabe and Richard T. Wetherald of the Geophysical Fluid Dynamics Laboratory and Warren M. Washington and Gerald A. Meehl of the National Center for Atmospheric Research for making their results available to me, and for their discussions of those results. I especiallywant to thank R. T. Wetherald and G. A. Meehl for providingme with the unpublished results that appear in this article. I also want to thank Michael R. Riches ofthe Carbon Dioxide Research Division, Office of Energy Research, Department of Energy, for inviting me to prepare this article for presentation at the DOE C02 Research Conference “Carbon Dioxide, Science and Consensus,”which was held at the Coolfont Conference Center, Berkeley Springs, West Virginia, September 19-23, 1982.

230

MICHAEL E. SCHLESlNGER

I express my gratitude to W. L.Gates for reviewing a preliminary version of the manuscript, to R. L. Mobley, D. S. Christopherson, and C. S. Mitchell for assistingwith the computations and graphics, to C. Beck, L. Riley, and N. Zielinski for typing the manuscript, and to J. Stark for drafting the figures. This research was supported by the National Science Foundation and the U.S. Department of Energy under Grants ATM 80-01702 and ATM 82-05992. REFERENCES

Aleksandrov, V. V., Arkhipov, P. L., Parkhomenko, V. P., and Stenchikov, G. L. (1983). The global joint ocean-atmosphere model and its sensitivity to the CO, variation. Zm.Akad. Nauk SSSR, Fiz. Atmos. Okenna 19,45 1-458. Augustsson, T., and Ramanathan, V. (1977). A radiative-convectivemodel study ofthe CO, climate problem. J. Amos. Sci. 34,448 -45 1. Baes, C. F., Goeller, H. E., Olson, J. S., and Rotty, R. M. (1976). The global carbon dioxide problem. Oak Ridge Nail. Lab. [Rep.] OWL. ORNLS194, 1-72. Bolin, B., and Bischof, W. (1970). Variations in the carbon dioxide content ofthe atmosphere of the Northern Hemisphere. Tellus 22,43 1-442. Bryan, K. (1969). Climate and the Ocean circulation. 111. The ocean model. Mon. Weather Rev. 97,806-827. Bryan, K., Manabe, S., and Pacanowski, R. C. (1975). A global ocean-atmosphere climate model. Part 11. The oceanic circulation. J. Phys. Oceunogr. 5, 30-46. Bryan, K., Komro, F. G., Manabe, S., and Spelman, M. J. (1582). Transient climate response to increasing atmospheric w b o n dioxide. Science 215, 56-58. Cess, R. D., and Goldenberg, S. D. (1981). The effect of ocean heat capacity upon the global warming due to increasing atmospheric carbon dioxide. JGR, J. Geophys. Res. 86,498 502. Charlock, T. P. (1 982). Cloud optical feedback and climate stability in a radiative-convective model. Tellus 34,245-254. Chervin, R. M. (1979). Response of the NCAR general circulation model to changed ocean surface temperature. CARP Pub[. Ser. 22,46 1 -48 I. Council on Environmental Quality (1 98 1). “Global Energy Futures and the Carbon Dioxide Problem.” U. S. Govt. Printing Office, Washington, D.C. Crutcher, H. L., and Meserve, J. M. (1970). “Selected Level Heights, Temperatures and Dew Points for the Northern Hemisphere,” NAVAIR 50-1C-52. Naval Weather ServiceCommand, Washington, D.C. Gates, W. L. (1976a). Modeling the ice-age climate. Science 191, 1138- 1144. Gates, W. L. (1976b). The numerical simulation of ice-age climate with a global general circulation model. J. Afmos.Sci. 33, 1844- 1873. Gates, W. L. (1979). “An Essay on the Physical Basis of Climate,” Rep. No. 7. Climatic Research Institute, Oregon State University, Corvallis. Gates, W. L. ( 1980a). “A Review of Modeled Surface Temperature Changes due to Increased Atmospheric CO,,” Rep. No. 17. Climatic Research Institute, Oregon State University, Corvallis. Gates, W. L. ( I980b). Modeling the surface temperature changes due to increased atmospheric C02. In “Interactions of Energy and Climate” (W. Bach, J. Pankrath, and J. Williams, eds.), pp. 169- 190. Reidel Publ., Dordrecht, Netherlands. Gates, W. L., Cook, K. H., and Schlesinger, M. E. (1981). Preliminary analysis ofexperiments on the climatic effectsof increased C02with an atmospheric general circulation model and a climatological Ocean. JGR, J. Geophys. Res. 86,6385-6393.

MODELS OF COz-INDUCED CLIMATIC CHANGE

23 1

Ghan, S. J., Lingaas, J. W., Schlesinger, M. E., Mobley, R. L., and Gates, W. L. (1982). “A Documentation of the OSU Two-level Atmospheric General Circulation Model,” Rep. No. 35. Climatic Research Institute, Oregon State University, Corvallis. Global Atmospheric Research Programme (1979). “Report of the JOC Study Conference on Climate Models: Performance, Intercomparison and Sensitivity Studies,” Vols. I and 11, GARP Publ. Ser. No. 22. World Meteorological Organization-International Council of Scientific Unions, Geneva. Hall, M. C. G., Cacuci, D. G., and Schlesinger,M. E.(1982). Sensitivity analysis of a radiativeconvective model by the adjoint method. J. Atmos. Sci. 39,2038-2050. Hansen, J., Johnson, D., Lacis, A., Lebedeff, S., Lee, P., Rind, D., and Russell, G. (1981). Climate impact of increasing atmospheric carbon dioxide. Science 213,957 -966. Hansen, J., Russell, G., Rind, D., Stone, P., Lacis, A., Lebedeff, S., Ruedy, R., and Travis, L. (1983). Efficient three-dimensional global models for climate studies: Models I and 11. Mon. Weather Rev. 111,609-662. Hibler, W. D., I11 (1979). A dynamic thermodynamic sea ice model. J. Phys. Oceanogr. 9, 8 15-846. Hoffert, M. I., Callegari,A. J., and Hsieh, C.-T. (1980). The role of deep sea heat storagein the secular response to climate forcing. JGR, J. Geophys. Res. 85,6667-6679. Hummel, J. R., and Kuhn, W. R. (198 1). An atmospheric radiative-convdve model with interactive water vapor transport and cloud development. Tellus 33,372-38 1. Hummel, J. R., and Reck, R. A. (1979). Carbon dioxide and climate: The effects of cloud specification in radiative-convective models. Res. Publ. Gen. Mot. Corp., Res. Lab. GMR-3151, 1-6. Hunt, B. G. (198 1). An examination of some feedback mechanisms in the carbon dioxide climate problem. Tellus 33, 78-88. Hunt, B. G., and Wells, N. C. ( I 979). An assessment of the possible future climatic impact of carbon dioxide increases based on a coupled one-dimensional atmosphericaeanic model. JGR, J. Geophys. Res. 84,787-791. Idso, S. B. (1980). The climatological significance of a doubling of earth’s atmospheric carbon dioxide concentration. Science 207, 1462- 1463. Jaeger, L. (1976). Monatskarten des Niederschlags fur die ganze Erde. Ber. Dtsch. Wetterdienstes 139, 1- 38. Jenne, R. (1975). “Data Sets for Meteorological Research,” NCAR Tech. Note TN/lA-1 11. National Center for Atmospheric Research, Boulder, Colorado. Kahn, H., Brown, W., and Martel, L. (1976). “The Next 200 Years, A Scenario for America and the World.” Morrow, New York. Kandel, R. S. (198 1). Surface temperature sensitivity to increased atmospheric CO,. Nature (London) 293,634-636. Katz, R. W. (1980). “Statistical Evaluation of Climate Experiments with General Circulation Models: Inferences about Means,” Rep. No. 15. Climatic Research Institute, Oregon State University, Corvallis. Katz, R. W. ( 1982a). “Procedures for Determining the Statistical Significanceof Precipitation Changes Simulated by an Atmospheric General Circulation Model,” Rep. No. 33. Climatic Research Institute, Oregon State University, CorvaHis. Katz, R. W. (1982b). Statistical evaluation of climate experiments with general circulation models: A parametric time series modeling approach. J. Atmos. Sci. 39, 1446- 1455. Katz, R. W. (1983). Statistical procedures for making inferences about precipitation changes simulated by an atmospheric general circulation model. J. Atmos. Sci. 40,2 193-220 1. Keeling, C. D., and Bacastow, R. B. (1977). Impact of industrial gases on climate. In “Energy and Climate,” Stud. Geophys., pp. 72-95. Natl. Acad. Sci., Washington, D.C.

232

MICHAEL E. SCHLESINGER

Keeling, C. D., Bacastow, R. B., Bainbridge, A. E., Ekdahl, C. A., Guenther, P. R., Waterman, L. S., and Chin, J. F. S. (1 976a). Atmospheric carbon dioxide variations at Mauna Loa Observatory, Hawaii. Tellus 28, 538-551. Keeling, C. D., Adams, J. A., Ekdahl, C. A,, andGuenther, P. R. (1976b). Atmosphericcarbon dioxide variations at the South Pole. Tellus 28, 552-564. Kellogg, W. W., and Schware, R. (1981). “Climate Change and Society.” Westview Press, Boulder, Colorado. Lacis, A. A.: and Hansen, J. E. (1974). A parameterization for the absorption of solar radiation in the earth’s atmosphere. J. Atmos. Sci. 31, 118- 133. MacDonald, G. F., Abarbanel, H., Carruthers, P., Chamberlain, J., Foley, H., Munk, W., Nierenberg, W., Rothaus, O., Ruderman, M., Vesecky, J., and Zachariasin, F. (1979). “The Long Term Impact of Atmospheric Carbon Dioxide on Climate,” JASON Tech. Rep. JRS-78-07. SRI International, Arlington, Virginia. Manabe, S. (1969a). Climate and the Ocean circulation. I. The atmospheric circulation and the hydrology of the earth’s surface. Mon. Weather Rev. 97,739-774. Manabe, S. (1969b). Climate and the ocean circulation. 11. The atmospheric circulation and the effect of heat transfer by ocean currents. Mon. Weather Rev. 97, 775-805. Manabe, S. (1971). Estimate of future changes in climate due to increase of carbon dioxide concentration in the air. In ”Man’s Impact on the Climate” (W. H. Mathews, W. W. Kellogg, and G. D. Robinson, eds.), pp. 249-264. MIT Press, Cambridge, Massachusetts. Manabe, S.(1983). Carbon dioxide and climatic change. Adv. Geophys. 25, 39-82. Manabe, S., and Hahn, D. G. (1981). Simulation of atmospheric variability. Mon. Weather Rev. 109,2260-2286. Manabe, S., and Stouffer, R. J. ( 1 979). A CO,-climate sensitivity study with a mathematical model of the global climate. Nature (London) 282,491 -493, Manabe, S., and Stouffer, R. J. (1980). Sensitivity of a global climate model to an increase of CO, concentration in the atmosphere. JGR, J. Geophys. Res. 85,5529- 5554. Manabe, S., and Wetherald, R. T. (1967). Thermal equilibrium ofthe atmosphere with a given distribution of relative humidity. J. Atmos. Sci. 24, 241 -259. Manabe, S., and Wetherald, R. T. (1975). The effects ofdoubling the C 0 2concentration on the climate of a general circulation model. J. Atmos. Sci.32, 3- 15. Manabe, S., and Wetherald, R. T. ( 1 980). On the distribution ofclimate change resulting from an increase in C0,-content of the atmosphere. J. Atmos. Sci. 37, 99- I 18. Manabe, S.. Bryan. K., and Spelman, M. J. (1975). A global ocean-atmosphere climate model. Part I. The atmospheric circulation. J. Phys. Oceanogr. 5, 3-29. Manabe, S., Bryan. K., and Spelman, M. J. ( 1979). A global ocean-atmosphere climate model with seasonal variation for future studies of climate sensitivity. Dyn. Atmos. Oceans 3, 393-426. Manabe, S., Wetherald, R. T., and Stouffer, R. J. ( 1981). Summer drynessdue to an increase of atmospheric CO, concentration. Clim. Change 3, 347 - 386. Michael, P., Hoffert, M., Tobias, M., and Tichler, J. (1981). Transient climate response to changing carbon dioxide concentration. Clim. Change 3, 137- 153. Mitchell, J. F. B. (1983). The seasonal response of a general circulation model to changes in CO, and sea temperatures. Q. J. R. Meteorol. SOC.109, 1 13- 152. Moller. F. (1963). On the influence of changes in CO, concentration in air on the radiative balance of the earth’s surface and on the climate. J. Geophys. Res. 68, 3877-3886. National Academy of Sciences (1976). “Climate and Food,” Report of the Committee on Climate and Weather Fluctuations and Agricultural Production. Board on Agriculture and Renewable Resources, Natl. Acad. Sci., Washington, D.C.

MODELS OF C02-INDUCED CLIMATIC CHANGE

233

National Academy of Sciences(1977). “Energy and Climate,” Report ofthe Panel on Energy and Climate. Geophys. Res. Board, Natl. Acad. Sci., Washington, D.C. National Academy of Sciences (1979). “Carbon Dioxide and Climate: A Scientific Assessment,” Report of an Ad Hoc Study Group on Carbon Dioxide and Climate. Climate Res. Board, Natl. Acad. Sci., Washington, D.C. National Academy of Sciences(1982). “Carbon Dioxide and Climate: A Second Assessment,” Report of the CO,/Climate Review Panel. Climate Board, Natl. Acad. Sci., Washington, D.C. Newell, R. E., and Dopplick, T. G. (1979). Questions concerning the possible influence of anthropogenic CO, on atmospheric temperature. J. Appl. Meteorol. 18,822 -825. North, G. R., Cahalan, R. F., and Coakley, J. A. (198 1). Energy balance climate models. Rev. Geophys. Space Phys. 19,9 1 - 121. Pitcher, E. J., Malone, R. C., Ramanathan, V., Blackmon, M. L., Puri, K., and Bourke, W. ( 1983). January and July simulationswith aspectral generalcirculation model. J. Atmos. Sci. 40,580 - 604. Pollard, D. (1982a). “The Performance of an Upper-ocean Model Coupled to an Atmospheric GCM: Preliminary Results,” Rep. No. 3 1. Climatic Research Institute, Oregon State University, Corvallis. Pollard, D. (1982b). A simple ice sheet model yields realistic 100 kyr glacial cycles. Nature (London) 296,334-338. Posey, J. W., and Clapp, P. F. (1964). Global distribution of normal surface albedo. Geojis. Int. 4, 33-48. Potter, G. L. (1978). Preliminary zonal model analysis of the climatic change resulting from increased atmospheric carbon dioxide. Lawrence Livermore Lab. [Rep.] NTIS UCRL-81575,1-30. Potter, G. L. (1980). Zonal model calculation of the climatic effect of increased CO,. In “Environmental and Climatic Impact of Coal Utilization” (J. J. Singh and A. Deepak, eds.), pp. 433-454. Academic Press, New York. Ramanathan, V. (1975). A study ofthe sensitivity ofradiativaconvective models. Proc. Con$ Atmos. Radiat. 2nd, 1975, pp. 124-125. Ramanathan, V. (198 1). The role of ocean-atmosphere interactions in the CO, climate problem. J. Atmos. Sci. 38, 918-930. Ramanathan, V., and Coakley, J. A. (1978). Climate modeling through radiative-convective models. Rev. Geophys. Space Phys. 16,465-489. Ramanathan, V., Lian, M. S., and Cess, R. D. (1979). Increased atmospheric COz: Zonal and seasonal estimates of the effects on the radiation energy balance and surface temperature. JGR, J. Geophys. Res. 84,4949-4958. Ramanathan, V., Pitcher, E. J., Malone, R. C., and Blackmon, M. L. (1983). The response of a spectral general circulation model to refinements in radiative processes. J.Atmos. Sci. 40, 605-630. Rasool, S. I., and Schneider, S. H. (1971). Atmospheric carbon dioxide and aerosols:Effects of large increases on global climate. Science 173, 138- 141. Robock, A. (1978). Internally and externally caused climate change. J. Atmos. Sci. 35, 11 11-1 122. Rotty, R. M. (1982). Distribution of and changes in industrial carbon dioxide production. Memo. Proc. OR4 U/IEA- Oak Ridge Assoc. Univ., Inst. Energy Anal. ORAUjIEA-822(M), 1-23. Rotty, R. M., and Marland, G. (1980). Constraints on fossil fuel use. In “Interactions of Energy and Climate” (W. Bach, J. Pankrath, and J. Williams, eds.), pp. 191-212. Reidel Publ., Dordrecht, Netherlands.

234

MICHAEL E. SCHLESINGER

Rowntree, P. R., and Walker, J. (1978). The effects of doubling the CO, concentration on radiative-convective equilibrium. In “Carbon Dioxide, Climate and Society” (J. Williams, ed.), pp. 18l - 19l. Pergamon, Oxford. Schlesinger, M. E. ( 1983a). A review of climate models and their simulation of C0,-induced warming. In?.J. Environ. Stud. 20, 103- 114. Schlesinger, M. E. (1983b). “A Review of Climate Model Simulations of C0,-Induced Climatic Change.” Rep. No. 41, Climatic Research Institute, Oregon State University, Corvallis. Schlesinger, M. E., and Gates, W. L. (1979). “Numerical Simulation of the January and July Global Climate with the OSU Two-level Atmospheric General Circulation Model,’’ Rep. No. 9. Climatic Research Institute, Oregon State University, Corvallis. Schlesinger,M. E., and Gates, W. L. ( 198la). “Preliminary Analysis ofthe Mean Annual Cycle and Interannual Variability Simulated by the OSU Two-level Atmospheric General Circulation Model,” Rep. No. 23. Climatic Research Institute, Oregon State University, Corvallis. Schlesinger,M. E., and Gates, W. L. (198 1b). “Preliminary Analysis of Four General Circulation Model Experiments on the Role of the Ocean in Climate,” Rep. No. 25. Climatic Research Institute, Oregon State University, Corvallis. Schneider, S.H. (1975). On the carbon-dioxide-climateconfusion. J. Atmos. Sci. 32,20602066. Schneider, S. H., and Thompson, S.L. (198 1). Atmospheric CO, and climate: Importance of the transient response. JGR. J. Geophys. Res. 86,3135-3147. Sellers, W. D. (1973). A new global climatic model. J. Appl. Meteorol. 12,241 -254. Sellers,W. D. ( 1974). A reassessment of the effect of CO, variations on a simple global climate model. J. Appl. Meteorol. 13, 831 -833. Sergin,V. Ya. ( 1980). Origin and mechanism of large-scaleclimatic oscillations. Science 209, 1477- 1482. Shukla, J., Straus, D., Randall, D., Sud, Y.,and Marx, L. (1981). “Winter and Summer Simulations with the GLAS Climate Model,” Tech.Memo. 83866. Laboratory for Atmospheric SciencesModeling and Simulation Facility, NASA Goddard Space Right Center, Greenbelt, Maryland. Stone, H. M., and Manabe, S.(1968). Comparison among various numerical models designed for computing infrared cooling. Mon. Weather Rev. 96,735-741. Taljaard, J. J., van Loon, H., Crutcher, H. L., and Jenne, R. L. (1969). “Climate of the Upper Air: Southern Hemisphere,” Vol. 1, NAVAIR 50-1C-55. Naval Weather Service Command, Washington, D.C. Temkin, R. L., and Snell, F. M. (1 976). An annual zonally-averaged hemispherical climate model with diffuse cloudiness feedback. J. Atmos. Sci. 33, 1671- 1685. Thompson, S.L., and Schneider, S.H. (1979). A seasonal zonal energy balance climate model with an interactive lower layer. JGR, J. Geophys. Res. 84,2401 -2414. Wang, W.-C., Rossow, W. B., Yao, M.-S., and Wolfson, M. (1981). Climate sensitivity of a onedimensional radiative-convective model with cloud feedback. J. Atmos. Sci. 38, 1167-1178. Washington, W. M., and Meehl, G. A. (1983a). A summary of recent NCAR general eirculation experiments on climatic effects of doubled and quadrupled amounts of C 0 2 , In “Proceedings of the U.S. Department of Energy CO, Research Conference: Carbon Dioxide, Science and Consensus,” Coolfont Conference Center, Berkley Springs,West Virginia. Washington, W. M., and Meehl, G. A. (1983b). Generalcirculation model experimentson the climatic effects due to a doubling and quadrupling of carbdn dioxide concentration. JGR, J. Geophys. Res. 88,6600-6610.

MODELS OF C02-INDUCED CLIMATIC CHANGE

235

Washington, W. M., and Williamson, D. L. (1977). A description of the NCARglobal circulation models. Methods Comput. Phys., 17, 1 1 1 - 172. Washington, W. M., Semtner, A. J., Jr., Meehl, G. A., Knight, D. J., and Mayer, T. A. (1980). A general circulation experiment with a coupled atmosphere, ocean and sea ice model. J. Phys. Oceanogr. 10, 1887- 1908. Watts, R. G. (1980). Climate models and C0,-induced climatic changes. Clim. Change 2, 387 - 408. Watts, R. G. (1982). Further discussion of “Questions concerning the possible influence of anthropogenic CO, on atmospheric temperature.” J. Appf. Meteorol. 21,243- 247. Weare, B. C., and Snell, F. M. (1 974). A diffuse thin cloud structure as a feedback mechanism in global climatic modeling. J. Atmos. Sci. 31, 1725- 1734. Wetherald, R. T., and Manabe, S. (1972). Response of the joint ocean-atmosphere model to the seasonal variation of the solar radiation. Mon. Weather Rev.100,42-59. Wetherald, R. T., and Manabe, S. (1975). The effects of changing the solar constant on the climate of a general circulation model. J. Atmos. Sci. 32,2044-2059. Wetherald, R. T., and Manabe, S. (1980). Cloud cover and climate sensitivity. J. Atmos. Sci. 37, 1485-1510.

Wetherald, R. T., and Manabe, S. (198 1). Influence of seasonal variation upon the sensitivity of a model climate. JGR, J. Geophys. Res. 86, 1194 - 1204. World Meteorological Organization (1977). “Report of the JOC/SCOR Joint Study Conference on General Circulation Models of the Ocean and their Relation to Climate (Helsinki, 23-27 May 1977),” Vols. I and 11. World Meteorological Organization, Geneva.