Modeling the Exchanges of Energy, Water, andCarbon Between Continents and the Atmosphere

P. J. Sellers,* R. E. Dickinson, D. A. Randall, A. K. Betts, F. G. Hall, J. A. Berry, G. J. Collatz,A. S. Denning, H. A. Mooney, C. A. Nobre, N. Sato, C. B. Field, A. Henderson-Sellers

Atmospheric general circulation models used for climate simulation and weather fore-casting require the fluxes of radiation, heat, water vapor, and momentum across theland-atmosphere interface to be specified. These fluxes are calculated by submodelscalled land surface parameterizations. Over the last 20 years, these parameterizationshave evolved from simple, unrealistic schemes into credible representations of the globalsoil-vegetation-atmosphere transfer system as advances in plant physiological andhydrological research, advances in satellite data interpretation, and the results of large-scale field experiments have been exploited. Some modern schemes incorporate bio-geochemical and ecological knowledge and, when coupled with advanced climate andocean models, will be capable of modeling the biological and physical responses of theEarth system to global change, for example, increasing atmospheric carbon dioxide.

Until the early 1980s, global atmosphericgeneral circulation models (AGCMs) in-corporated very simple land surface param-eterizations (LSPs) to estimate the ex-changes of energy, heat, and momentumbetween the land surface and the atmo-sphere. These have since evolved into afamily of schemes that can realistically de-scribe a comprehensive range of land-atmo-sphere interactions. These advancedschemes will be needed to understand theresponse of the biosphere and the climatesystem to global change, for example, in-creasing atmospheric CO2 (1–3).

Three generations of models have tak-en us from the early LSPs to where westand now. The first, developed in the late1960s and 1970s, was based on simpleaerodynamic bulk transfer formulas andoften uniform prescriptions of surface pa-rameters (albedo, aerodynamic roughness,and soil moisture availability) over the

continents (4). In the early 1980s, a sec-ond generation of models explicitly recog-nized the effects of vegetation in the cal-culation of the surface energy balance (5,6). At the same time, global, spatiallyvarying data of land surface propertieswere assembled from ecological and geo-graphical surveys published in the scien-tific literature (7). The latest (third gen-eration) models use modern theories relat-ing photosynthesis and plant water rela-tions to provide a consistent description ofenergy exchange, evapotranspiration, andcarbon exchange by plants (8–10). Someare beginning to incorporate treatments ofnutrient dynamics and biogeography, sothat vegetation systems can move in re-sponse to climate shifts. A series of large-scale field experiments have been execut-ed to validate the process models and scal-ing assumptions involved in land-atmo-sphere schemes (3). These experimentshave also accelerated the development ofmethods for translating satellite data intoglobal surface parameter sets for the models.

Theoretical Background and theFirst-Generation Models

It has been understood for nearly 200 yearsthat the continents and the atmosphereexchange energy, water, and carbon witheach other. However, it was not until thelate 1960s with the construction of the firstAGCMs, in the guise of numerical weatherprediction (NWP) models, that scientificinterest became focused on how to describethese exchanges with useful accuracy.

In AGCMs, the motion of the atmo-sphere is defined by fluid dynamics equa-tions that incorporate the mechanicalforces of gravity, the Earth’s rotation, pres-

sure and temperature gradients, and fric-tion (2, 11). Energy transfer processes in-clude radiative heating and cooling; heattransport by means of convection, con-densation, and evaporation; and the trans-fer of energy, water, and momentum acrossthe lower boundary of the atmosphere,that is, between the land or ocean surfacesand the atmosphere. The AGCMs usethree-dimensional grid systems to repre-sent the vertical and horizontal structureand state of the atmosphere and integratefinite difference versions of the governingequations to predict successive states ofthe atmosphere. The early models typical-ly used horizontal resolutions of around10° to 20° in longitude and latitude, twoto seven layers in the vertical, and timesteps of about 30 min. As computer per-formance improved, spatial resolutions forglobal NWP models were refined toaround 1° of horizontal resolution and 20or more layers in the vertical. Time stepsstill range from a few minutes to a half-hour or so.

The coarse grids did not lend them-selves to exacting validation. However, asthe models improved along with their spa-tial resolutions, it became obvious thatshortcomings in the description of manyphysical processes—such as radiative ex-change, convection, and condensation—were at least as important as the fluiddynamics problems that had absorbedmost of the early scientific effort (2). Withregard to land-atmosphere interactions,the important processes from the AGCMpoint of view were (i) the exchanges ofradiation, (ii) the fluxes of sensible andlatent heat (evapotranspiration) betweenthe surface and atmosphere, and (iii) thefrictional deceleration of the lower atmo-sphere resulting from drag forces as thewind passes over the land. All three ofthese processes are closely related througha simple set of equations.

The net amount of radiant energy ab-sorbed by the land surface and available fortransfer into other energy forms is definedas the sum of the absorbed solar energy andthe absorbed downwelling long-wave radia-tion emitted by the overlying atmosphereminus the long-wave radiation emitted bythe surface (Fig. 1A)

Rn 5 S(1 2 a) 1 Lw 2 εsTs4 (1)

P. J. Sellers is at the NASA Johnson Space Center, MailCode CB, Houston, TX 77058, USA. R. E. Dickinson is inthe Department of Atmospheric Sciences, University ofArizona, Tucson, AZ 85721, USA. D. A. Randall is in theAtmospheric Sciences Department, Colorado State Uni-versity, Fort Collins, CO 80523, USA. A. K. Betts is atAtmospheric Research, Rural Route 3, Box 3125, Pitts-ford, VT 05763, USA. F. G. Hall and G. J. Collatz are atNASA Goddard Space Flight Center, Code 923, Green-belt, MD 20771, USA. J. A. Berry and C. B. Field are in theDepartment of Plant Biology, Carnegie Institution, Stan-ford, CA 94305, USA. A. S. Denning is in the School ofEnvironmental Science and Management, University ofCalifornia, Santa Barbara, CA 93106–5131, USA. H. A.Mooney is in the Department of Biological Sciences,Stanford University, Stanford, CA 94305, USA. C. A. No-bre is at INPE/CPTEC, Caixa Postal 01, cep 12630-000,Cachoeira Paulista, SP, Brazil. N. Sato is in the NumericalPrediction Division, Japan Meteorology Agency, 1-3-4,Ootemachi, Chiyoda-ku, Tokyo, Japan 100. A. Hender-son-Sellers is at the Royal Melbourne Institute of Tech-nology, Plenty Road, Post Office Box 71, Bundoora, VIC3083, Australia.

*To whom correspondence should be addressed.

ARTICLE

SCIENCE z VOL. 275 z 24 JANUARY 1997502

where Rn is the net radiation, S is theinsolation, a is the surface albedo, Lw is thedownward long-wave flux, ε is the surfaceemissivity (>1.0), s is the Stefan-Boltz-mann constant, and Ts is the land surfacetemperature. Insolation S is calculated as afunction of latitude, longitude, time of day,and the cloudiness simulated or prescribedwithin the AGCM; Lw is a function of thetemperature, humidity profile, and cloudi-ness of the atmosphere; and Ts is usuallypredicted (time stepped) within the model.From Eq. 1 we see that the land surfaceparameter that has the most influence onthe surface radiation budget is the albedo a,which is defined as the integrated reflec-tance of the surface over the solar spectrum(0.0 to 4.0 mm).

Radiation Rn is partitioned into three(nonradiation) heat flux terms

Rn 5 G 1 H 1 lE (2)

where G is the ground heat flux, H is thesensible heat flux, E is the evapotranspira-tion rate, and l is the latent heat of vapor-ization. The calculation of G has generallybeen approached as a simple heat diffusionproblem. The early models used two-layerforce-restore schemes (12), which crudelysimulated the diurnal and seasonal compo-nents of the ground heat flux. Many LSPs

have moved on to incorporate multilayerdescriptions of coupled heat and moistureflow, together with descriptions of theeffects of snowmelt and water phasechanges within the soil profile (13). Gen-erally, G is a small proportion, 10% or less,of Rn when averaged over a diurnal cyclebut can be an important term in the sea-sonal energy budget.

The fluxes of H and lE have profoundeffects on weather and climate (Fig. 1B).The sensible heat released from the landsurface raises the temperature of the over-lying air column and warms the planetaryboundary layer. The latent heat flux lEis the energy equivalent of the water evap-orated from the surface or transpiredthrough vegetation (the term evapotrans-piration covers both sources of moisture).The evaporated water vapor is often trans-ported to great heights through convec-tion, releases heat to the atmosphere dur-ing condensation, forms clouds (whichhave strong effects on the atmosphericradiation budget), and produces pre-cipitation P. Thus, unlike sensible heat,the release of latent heat from the sur-face usually has a nonlocal impact on theatmosphere.

Given the different and important rolesof sensible and latent heat within the cli-mate system, Eq. 2 should be solved asaccurately as possible over all surface grid-points within an AGCM. In the first mod-els, transports of H and lE were commonlytreated as quasi-diffusive processes, whichcan be written in the potential differenceresistance form (Fig. 2A)

H 5Ts 2 Tr

rarcp (3)

lE 5 bFe*~Ts! 2 er

rsG rcp

g(4)

where Tr is the air temperature within thelowest layer of the atmospheric model, ra isthe aerodynamic resistance between thesurface and the lowest layer of the atmo-sphere, r and cp are the density and specificheat of air, b is the moisture availabilityfunction (0 # b # 1), e*(Ts) is the satu-rated vapor pressure at temperature Ts, er isthe vapor pressure within the lowest layer ofthe atmospheric model, and g is the psy-chrometric constant.

Thus, H is proportional to the differ-ence between surface and atmospherictemperatures and inversely proportional toan aerodynamic resistance ra, which canbe thought of as a turbulent diffusion-related term impeding the transfer of heator mass from the surface to the air. Almostall of the first AGCMs made use of Eq. 4to describe evapotranspiration; it is a di-rect analog of Eq. 3 except that water

vapor pressure replaces temperature andthe land surface is assumed to be a satu-rated source of water [e*(Ts)]. A moistureavailability term, b, was included to re-duce evaporation rates in dry areas (4, 14),and an accounting procedure for soil mois-ture, W, was applied to each grid square,whereby the soil column was depleted byevapotranspiration and replenished byprecipitation up to a maximum capacityWmax, after which excess precipitation wasassumed to run off as streamflow. Thevalue of b was usually set to unity whenthe so-called “bucket” model was “full”(W 5 Wmax) and to zero when empty. Avariety of functions was proposed to de-scribe b for intermediate cases (Fig. 2A).

The aerodynamic resistance ra is com-monly derived from simple turbulencemodels. It is inversely dependent on windspeed ur and the logarithm of the surfaceroughness length z0; z0 is a function of thedrag properties of the land surface and isabout 10% of the vegetation height. Sta-bility corrections are applied to accountfor the effects of convection on ra becauselarge fluxes of H can significantly augmentturbulent transfer and reduce ra (15). Thefrictional stress t exerted by the land sur-face on the atmosphere is proportional tour/ra; many formulations also account fordifferences between the transports of heat,water vapor, and momentum in ra.

Inspection of Eqs. 1 through 4 showshow the atmospheric forcing variables (S,Lw, Tr, er, ur, and P) are used to calculatethe energy available to the surface (Rn) andthe fluxes (H, lE, G, and t) given thesurface parameters (a, z0, and Wmax): Tobegin with, a and z0 were simply prescribedas global, often uniform, fields, and Wmaxwas set to a single value, typically 150 mm,everywhere.

A computer code incorporating theseequations or their analogs is referred to asan LSP. The first-generation LSPs wereused to explore the roles of albedo, surfaceroughness, and moisture availability inAGCM climatologies (16).

The land surface albedo varies fromabout 9 to 12% (boreal and tropical for-ests) to around 35% (Sahara desert) overthe snow-free land (7, 17, 18). SomeAGCM sensitivity experiments that fo-cused on albedo changes in the Sahel-Sahara region indicated that an increasein the regional albedo would lead to de-creases in the surface evaporation and pre-cipitation rates in the same area (19, 20).In another study, the roughness length z0of the world’s deserts was reduced from 45cm to 0.2 mm, which improved the simu-lated fields of horizontal water vapor con-vergence and convective precipitation(21). Many sensitivity studies have fo-

H

H

Atmosphericboundary layer

G

λE

λE

λP

εσT4sS

αS

S

A

B

Fig. 1. Interactions between the land surfaceand the atmosphere that have direct impacts onthe physical climate system. (A) Surface radia-tion budget. (B) Effect of heat fluxes on theatmosphere.

SCIENCE z VOL. 275 z 24 JANUARY 1997 503

cused on the role of soil moisture in con-trolling sensible and latent heat fluxesbetween the surface and atmosphere (20,22). Experiments ranged from global pre-scriptions of totally wet versus totally dryland surfaces to alterations in regional soilmoisture capacities. In almost all of thesimulations, reduced land surface evapora-tion rates led to reduced precipitationrates in the continental interiors.

Biophysics and theSecond-Generation Models

The early AGCM sensitivity experimentsdemonstrated that the specification of albe-do, roughness, and, perhaps most impor-tantly, the “surface wetness” could haveimportant impacts on atmospheric fields.Although illustrative, the experiments wereunrealistic because the fluxes of radiation,

sensible and latent heat, and momentumacross the lower boundary of the atmo-sphere were treated as independent process-es (5, 23).

In nature, plants are not the passive,spongelike structures implied by the first-generation bucket models, in which thevegetation was viewed as a pervious sheetseparating the soil from the atmosphere(Fig. 2A). Consequently, the strategyadopted in formulating the second-genera-tion biophysical LSPs (5, 6) was to modelthe vegetation-soil system itself and let thisdetermine the ways in which the land sur-face interacts with the atmosphere (Fig.2B). These interactions can be summarizedas follows.

1) Radiation absorption. The spectralproperties of leaves and multiple reflectionsbetween them make vegetative canopieshighly absorbent in the visible [photosyn-thetically active radiation (PAR)] wave-length interval (0.4 to 0.72 mm) and mod-erately reflective in the near-infrared region(0.72 to 4.0 mm). In contrast, bare groundgenerally exhibits a gradual increase in re-flectivity with wavelength between 0.4 and4.0 mm.

2) Momentum transfer. Vegetative cano-pies usually present a rough, porous surfaceto the planetary boundary layer airflow.The resultant turbulence enhances thetransport of sensible and latent heat awayfrom the surface while exerting a drag forcethat may be significantly larger than thatproduced by bare ground.

3) Biophysical control of evapotranspira-tion. Plant metabolism is based on the pho-tosynthetic reaction, in which shortwaveradiation energy is used to combine waterand atmospheric CO2 into sugars and otherorganic compounds. To do this, plants mustallow for the transfer of CO2 from the at-mosphere to the cellular sites of photosyn-thesis located inside the leaves. This flowrequires an open pathway between the at-mosphere and the water-saturated tissuesinside the leaf, which leads to an inevitableloss of water vapor over the same route(Figs. 2B and 3). Higher plants regulate theamount of gas exchange (and hence waterloss) by means of adjustable valvelike struc-tures on the leaf surface called stomates.When the first biophysical LSPs were beingformulated in the 1980s, working hypothe-ses for quantitatively describing the linkedphotosynthesis-stomatal conductance sys-tem were in the early stages of development(24) and were not yet recognized by theclimate community. However, empiricalwork (25) had correlated stomatal conduc-tance gs (the conductance of the leaf sto-mates to the passage of water vapor fromthe saturated leaf interior to the free air justoutside the leaf) to the environmental con-

PAR (mEin m-2 s-1)

0.50.40.3

0.20.10 1000 2000

g s (c

m s

-1)

g s (c

m s

-1)

g s (c

m s

-1)

g s (c

m s

-1)

CL (MPa)

Cl (MPa)

0

0.25

0.5 0.5δe (kPa)

1.0

1.5

-2.0 -1.0 0

0.3

0.2

0.1

0–2.0 –1.0

Tr

Tse*(Ts)

Wmax = 150 mm

W

e r

ra ra/β

λE

H

W

b

Wmax0.00.0

1.0

Reference height z r

Surface properties: α, z 0

Aerodynamic pathwayin lower atmosphere

Bucket modelfor hydrology

Reference height z r

Canopy air space

Atmosphere

Surface soil layer

Root zone layer

T (°C)

0

0.25

0.5

-10 10 30

A

B

λEc +

λE g

λEg

H c +

Hg

Tr

Ta

ra

h soile*(Tg)

e*(Tc) e a

rd rdrc 2rb

rsoil

ra

e r

Hc

Hg

Tg

λEc

Canopy

Fig. 2. Development of LSPs in AGCMs. (A) First-gen-eration “bucket” model (4). (Left) The sensible and latentfluxes flow from the surface to the atmosphere throughan aerodynamic resistance ra. (Right) The latent heatflux is regulated by the level of moisture W in the bucket through the moisture availability function b. (B)Second-generation model (5, 6) with separated vegetation canopy and soil. Moisture fluxes from the(dry) vegetation canopy escape to the canopy air space and free atmosphere through a bulk canopysurface resistance [rc 5 1/gc (29)]. Sensible and latent heat fluxes from the ground (subscript “g”) andcanopy (subscript “c”) combine in the canopy air space to give the total fluxes passed to the atmo-sphere. (Right) Terms in Eq. 6 that relate leaf conductance (gs) to light intensity (PAR), vapor pressuredeficit (de), temperature (T ), and leaf water potential ( Cl). [Redrawn with permission from (25)] The pointsdenote observations; the lines denote a hypothetical limit to the f (Cl) relation. The canopy resistance rccan be calculated by integrating gs over the depth of the canopy and inverting it (27, 28). (C) Third-generation model (8, 10). A carbon flux pathway is added to the moisture and heat flux pathways shownin (B). (Left) The canopy A-gs model controls canopy carbon and water fluxes simultaneously andconsistently. (Upper right) Dependence of leaf assimilation A on light (PAR) intensity, CO2 concentration,and water stress. (Lower right) Relation between gs and A for a number of species, with a line of best fit(see also Eq. 7).

c r

csc a

R soi

lA C

- R

D -

R soi

l

AC - RDci

1.6rc 2.8r bCanopy

Atmospheric CO2, etc Tracer - transport model

Canopy photosynthesis - conductance model

Soil respiration model

C

g s (m

ol m

-2 s

-1)

A (hs/cs)

0.80.70.60.50.40.30.20.10.00 .01 .02 .03 .04 .05 .06 .07 .08 .09

A-R d

(µm

ol m

-2 s

-1)

2 × CO2 (700 ppm)

1 × CO2 (350 ppm)

1 × CO2 (350 ppm),water stress

PAR (W m-2)

40

30

20

10

00 250 500 750 1000

SCIENCE z VOL. 275 z 24 JANUARY 1997504

ditions that control photosynthesis (Fig.2B).

gs 5 gs~PAR!@f(de)f(T)f(Cl)] (5)

where gs(PAR) is the PAR-regulated (un-stressed) value of leaf conductance andf(de), f(T), and f(Cl) are the environmentalstress factors that account for the effects ofvapor pressure deficit de, temperature T,and leaf water potential Cl, respectively.

Equation 5 captures the important re-sponses of leaf stomates to the environ-ment. As PAR (sunlight) increases, the un-stressed leaf prepares for photosynthesis byopening its stomates, and so gs(PAR)monotonically increases from near zero atzero PAR to an asymptote at high lightlevels. The stress factors are all scaled be-tween 1, for optimal conditions, to 0, whenenvironmental conditions are severeenough to shut down photosynthesis andclose the stomates (Fig. 2B, right). Thef(de) term is particularly interesting: almostall plants maintain open stomates in humidair, when CO2 can be taken up freely witha relatively small loss of leaf water vapor. Asthe external air dries, the stomates progres-sively close, and f(de) decreases, presumablyto protect the leaf from desiccation and toconserve water. The f(T) function reaches amaximum around the mean environmentalgrowing season temperature and tapers offto zero for warmer or cooler temperatures;this action is related to the enzyme kineticsof photosynthesis and conductance, whichhave been “tuned” through evolution towork efficiently at particular temperatures(in fact, this optimal temperature for aplant’s enzymatic function can vary to ac-commodate seasonal and interannual tem-perature variations). The leaf water poten-tial Cl represents the chemical energy ofthe liquid water in the leaf cells; its valuehas to be negative to ensure a continuoussuction pathway of moisture from the rootzone to the leaves (26). The stomates closewhen the soil moisture is limiting or whenthe transpiration rate is excessive, causingCl to drop. The leaf-scale model of Eq. 5 hasbeen modified by a number of techniques toprovide a canopy-scale estimate of con-ductance gc (27). One scheme (28) assumesthat the stress factors are near uniform andthat PAR is attenuated exponentially downthrough the vegetation canopy, which per-mits analytical integration of Eq. 5 to yieldan estimate of gc for a canopy of known leafarea index (LAI, the one-sided area of leavesper unit of ground area; dense vegetation hasan LAI of around 5). The inverse of gc givescanopy resistance rc 5 1/gc, which can beused to calculate evapotranspiration (29)(Fig. 2B). In the absence of significant soilevaporation, this takes the form

lE 5 bFe*~Ts! 2 er

ra 1 rcG rcp

g(6)

Equations 6 and 4 should be compared.Equation 4 has a moisture limitation termb, which is applied externally to an esti-mate of the maximum evaporation rate; bycontrast, the use of rc in series with ra in Eq.6 realistically separates aerodynamic andsurface resistance terms (Fig. 2B). Undernormal unstressed conditions (for example,in dense green forests), ra ' 10 s m21 and rc' 100 s m21, so that evapotranspirationrates calculated by Eq. 6 are almost alwaysmuch lower than those calculated by Eq. 4.This effect becomes even more markedwhen soil moisture is limiting and calculat-ed values of Ts are high.

4) Precipitation interception and intercep-tion loss. Vegetation canopies also interceptprecipitation, and some can store the equiv-alent of about a millimeter of water on leafsurfaces. The evaporation of this intercept-ed water reduces the precipitation inputinto the soil, reduces the sensible heat flux,and can substantially increase the totalevaporation rate. For example, one-third toone-half of the rainfall falling on Amazoniais estimated to be re-evaporated to the at-mosphere through interception loss (30).

5) Soil moisture availability. The depthand density of root systems determine theamount of soil moisture available for evapo-transpiration. Empirical models were usedto relate f(Cl) to soil water content in theroot zone, the root density, and the transpi-ration rate (6).

6) Insulation. The soil surface under adense vegetation canopy intercepts lessradiation and may also be aerodynamicallysheltered. For these reasons, the energyavailable to the covered soil is small, andthe component terms of the soil energybudget (evaporation, sensible heat flux,and ground heat flux) are correspondinglyreduced.

Global parameter sets for these modelswere assembled from reports on ground-based ecological surveys (7). Estimates of

seasonally varying fields of LAI and green-leaf fraction in these data were used todefine global monthly fields of albedo,roughness, and unstressed canopy con-ductance with the use of a series of simplemodels. Thus, the important surface param-eters were made mutually consistent inthese second-generation models; vegetationstructure, density, and optical propertieswere used to determine a, z0, and rc (5, 6).The models were then used to investigatecontinental hydrometeorology and to con-duct some “land cover change” AGCM ex-periments (16).

The use of second-generation models ledto improved simulation of continental hy-drometeorology. The results of a run pro-duced by a biophysical model linked to anAGCM were compared with those pro-duced by a conventional bucket hydrologymodel linked to the same AGCM (31). Thecontinental evaporation rates calculated bythe biophysical simulation were consistent-ly lower and in closer agreement with avail-able observations compared with the resultsfrom the control run, mainly because of theinclusion of the surface resistance term (rc)in the biophysical model. These reducedevapotranspiration rates caused reducedand more realistic continental precipitationfields. More recently, a four-layer prognos-tic soil model coupled to a canopy resis-tance model (32) and an improved bound-ary layer model (carefully validated againstdata from land surface experiments) wasintroduced into a global forecast model,leading to improvements in the forecast ofprecipitation over the continents. This im-proved NWP model realistically simulatedthe precipitation anomaly that led to themidwestern floods in the United States inthe summer of 1993 (33).

A series of “land cover change” simula-tion experiments directly benefited fromthe second-generation LSPs. Biophysicallybased models were used to study the impactof large-scale Amazonian deforestation onthe regional and global climate (34). Theresults from some of these studies show

H

λE

A-Rd

Sensibleheat

Lightreactions

Lightreactions

Calvincycle

PARPAR

Stomate(1/gs)

Leaf boundarylayer (1/g b)

Photosynthate(C6H12O6, etc)

Leafinterior

Stem androots

Watervapor

CO2

H2Ofrom Soil

CO2

O2

ca

eaese*(Ts)

Ts Ta

csci

ATP, NADPH

Fig. 3. Schematic of carbonand water exchange in a leafas conceptualized in a com-bined photosynthesis-con-ductance model (39, 40, 42,43). The stomatal conduc-tance (gs) is a direct functionof photosynthesis, CO2 con-centration at the leaf surface(cs), and relative humidity atthe leaf surface [hs 5 es/e*(Ts); es 5 vapor pressureat leaf surface] (see Eq. 7).

ARTICLE

SCIENCE z VOL. 275 z 24 JANUARY 1997 505

decreases in regional evapotranspirationand precipitation linked to increases in sur-face temperature of around 3 to 5 K associ-ated with large-scale deforestation. The in-fluence of vegetation on precipitation pat-terns in the Sahelian region of Africa hasalso been investigated (35); replacement ofseasonal forest and grassland by desert led toa simulated reduction in evapotranspirationand rainfall over the same areas, resulting ina net displacement of the seasonal rainfallpatterns to the south.

The effects of spatial or temporal varia-tions in land surface properties have alsobeen studied. The contributions of “mosa-ics” of different vegetation types within asingle AGCM grid square were investigatedto explore different averaging schemes; itwas found that some straightforward aver-aging of key parameters could produce re-sults comparable to fully discretized treat-ments (36). Biophysical LSPs have alsobeen used within mesoscale models to showthat sharp variations in land surface prop-erties—for example, forest-agricultureboundaries—may initiate mesoscale circu-lations under low-wind, high-radiation con-ditions (37).

The incorporation of biophysics into thesecond-generation LSPs made them inter-nally consistent, realistic, and capable ofcalculating surface-atmosphere fluxes moreaccurately than their first-generation coun-terparts. However, they were still focusedon calculating energy and water budgets

because these fluxes have immediate andlarge effects on the physical climate systemas represented in AGCMs. A number ofcircumstances provided the motivation forthe development of more advanced models.

The Carbon Cycle andThird-Generation Models

By the late 1980s, scientific interest hadbecome focused on global change, particu-larly on the “greenhouse effect” (globalwarming) and associated impacts (38). Theneed for more complete models of the cli-mate system—including biological andchemical processes in the ocean, land, andatmosphere—became apparent.

Plant physiological research made signif-icant advances during the 1980s and early1990s. A series of biochemical models ofleaf photosynthesis were developed that de-scribe CO2 assimilation by chloroplasts orleaves as rate-limited by (i) enzyme kinet-ics, specifically the amount and cycle timeof the carboxylating enzyme Rubisco, (ii)electron transport, which is a function ofincident PAR, and (iii) the efficiency of theleaf’s light-intercepting apparatus (chloro-phyll) (39). These models describe the leafassimilation (or gross photosynthetic) rateas approaching the minimum of three lim-iting rates: wc, we, and ws, which describethe assimilation rate as limited by the effi-ciency of the photosynthetic enzyme system(Rubisco-limited), the amount of PAR cap-

tured by the leaf chlorophyll, and the ca-pacity of the leaf to export or utilize theproducts of photosynthesis, respectively.

The physiological limit on assimilation,wc, is primarily a function of the leaf’senzyme reserves, which can be thought of asthe biochemical processing capacity of theleaf. A model parameter Vmax represents themaximum catalytic capacity of the leaf’sphotosynthetic machinery and determinesthe upper bound of wc; it is directly relatedto leaf nitrogen content. Significantly, wc isalso a direct function of leaf internal CO2concentration ci, which is itself linked tothe CO2 concentration of the external air ca(Fig. 3). The light-limited rate of assimila-tion we is a linear function of incident PARand is also dependent on ci; ws may bedefined as a function of the biochemicalcapacity of the leaf and made proportionalto Vmax (40).

Field and laboratory studies have docu-mented the tight linkage between leaf pho-tosynthesis and conductance, and theoreti-cal work suggests that stomates function soas to maximize the efficiency of plant wateruse (24, 41). A semi-empirical model of leafconductance gs has been proposed (42) (Fig.2C)

gs 5 mAn

cshsp 1 b (7)

where m is an empirical coefficient fromobservations ('9 for most C3 vegetationand '4 for C4 vegetation), An is the net

NP

60

30

EQ

-30

-60

SP

NP

60

30

EQ

-30

-60

SP

1800.1 0.3 0.4 0.5 0.6 0.7 0.8 0.90.2

120 W 60 W 0

FPAR

60 E 120 E 180 18010 16 22 28 34

Transpiration (W m-2)

40 46 52 58120 W 60 W 0 60 E 120 E 180

180150 300 450 600 750 900 1050 1200 1350

120 W 60 W 0 60 E 120 E 180 180360 362 364 366 368 372 374 376370

120 W 60 W 0 60 E 120 E 180

CO2 concentration (ppm)NPP (g C m-2)

A B

C D

Fig. 4. Global fields used inor generated by a third-gen-eration LSP. (A) Global fieldof FPAR calculated fromAVHRR SVI data (Eq. 9). (B)Canopy transpiration and(C) canopy net photosyn-thetic productivity (NPP, ingrams of carbon per squaremeter) calculated by a third-generation LSP from withinan AGCM, using the FPARfield shown in (A) (8, 48). (D)Annual mean CO2 concen-tration in the planetaryboundary layer (55).

SCIENCE z VOL. 275 z 24 JANUARY 1997506

CO2 assimilation, cs is the CO2 concen-tration at the leaf surface, hs is the relativehumidity at the leaf surface, p is the atmo-spheric pressure, and b is the minimumvalue of gs ('0.01 for C3 vegetation and'0.04 for C4 vegetation). The principalvegetation-dependent parameter definingA is Vmax, so a total of only three param-eters (Vmax, m, and b) are used in thecalculation of A and gs, which is a consid-erable simplification over the more empir-ical models of leaf conductance used inthe second-generation models. In thesenew photosynthesis-conductance (A-gs)models, the partial pressures of CO2 (ciand cs) and the leaf surface relative hu-midity hs are linked to conditions in thecanopy air space through gs, the leafboundary-layer conductance gb, the netflux of CO2 (A 2 Rd, where Rd is the leafrespiration rate), and leaf transpiration(Fig. 3). The complete equation set can besolved to yield mutually consistent valuesof leaf photosynthesis and transpiration.

This leaf-level model can be integratedover the depth of a vegetation canopy anddriven by satellite data given a couple ofsimplifying assumptions (43). The value ofVmax has been observed to decrease withcanopy depth in parallel with the attenua-tion of PAR; this arrangement seems tomake optimal use of plant nitrogen, whichis usually a scarce and valuable resource innature (41). This relation between thewithin-canopy profiles of PAR and leaf ni-trogen (Vmax) can be exploited to drastical-ly simplify the integration of A and gs. Thephotosynthetic rate and conductance of anentire canopy can be estimated by multiply-ing a calculation of the performance of theuppermost leaves in the canopy, exposed to

the maximum (incident) PAR flux (PAR0)and hence having the highest photosyn-thetic capacities (Vmax0), by a canopy PARuse parameter, P 5 FPAR/k, where FPARis the fraction of incident PAR absorbed bythe green leaves in the canopy and k is thecanopy extinction coefficient for PAR.Both FPAR and k are time-mean, radiation-weighted quantities. The complete set ofleaf-scale and canopy-integrated equationsgoverning photosynthesis and conductancecan then be summarized as

Ac,gc 5 [Vmax0,PAR0][B1. . .B6][P] (8)

or, reexpressed in words, the canopy scaleAc is the leaf physiology or radiation limittimes the parameters B1 through B6 [whichdescribe the effects of temperature, humid-ity, CO2 concentration, soil moisture stress,and so forth on Ac and gc (43)] times P(which varies from zero, for no vegetationcover, to between 1 and 1.5, for dense greenvegetation); gc is calculated in a similarmanner. The PAR use parameter P is avegetation property that is amenable to re-mote sensing (43, 44). FPAR and LAI havebeen related to the so-called spectral vege-tation indices (SVI), which are combina-tions of the radiances (or calculated reflec-tances) in the visible and near-infrared re-gions as observed over vegetated land sur-faces by satellite sensors (45). Theoreticalwork (28, 43, 46) has explored these rela-tions and has demonstrated that the sensorwavebands on the Advanced Very HighResolution Radiometers (AVHRR) mount-ed on polar-orbiting satellites are well suit-ed for providing SVI values, which shouldbe near linearly related to FPAR over awide range of conditions. The FPAR termin P should be near linearly related to the

simple ratio (SR) SVI when the soil back-ground is dark (43, 46)

P 5 FPAR/k } SR 5 aN/aV (9)

where aN and aV are the near-infrared andvisible reflectances (or instrument counts),respectively (sensor-dependent). The ap-proximate scale-invariance of Eq. 9 is key toits application on large spatial scales, suchas in an AGCM. Comparing Eqs. 8 and 9,we see that there is a chain of (near-) linearrelations between Ac, gc, P, FPAR, and SR.Because the area integral of a linear func-tion is proportional to the area of thatfunction over the domain, the mean valueof a suitable SVI over a large area (as sup-plied by a coarse-resolution satellite sensor)should provide good estimates of non-stressed Ac and gc over the same areathrough P in Eq. 8. Global sets of SVI datahave been assembled from satellite observa-tions (47) and have been further processedand transformed to provide global 1° 3 1°monthly fields of FPAR (Fig. 4A) and LAI,and hence albedo, roughness length, and P,which can be used directly by third-gener-ation LSPs (8, 9, 48).

The third-generation models (8, 10)have some advantages over their predeces-sors (Fig. 2). First, they are more realisticbiologically in that linked A-gs models areused to calculate the coupled fluxes of en-ergy, water, and carbon. These new modelsalso require fewer parameters. Second, animportant parameter governing the surfacefluxes, P, can be obtained continuously,globally, and consistently from satellitedata. Third, physiological models can bedirectly responsive to changes in atmo-spheric CO2 in a realistic way.

A model that incorporates all of these

Table 1. Completed and planned large-scale land-atmosphere field experiments.

Experiment Date of fieldphase Location area Area of study

(km 3 km) Primary foci

HAPEX-MOBILHY (51) 1986 Southwest France 100 3 100 Energy-water exchange, mesoscale modelingFIFE (52) 1987, 1989 Kansas,

United States15 3 15 Energy-water-carbon exchange process studies,

scaling, remote sensing scienceKUREX (57 ) 1991 Kursk, Russia – Energy-water-carbon exchange, remote sensing

scienceEFEDA (58) 1991 Central Spain 100 3 100 Energy-water-carbon exchange, process studies,

scaling, remote sensing scienceHAPEX-SAHEL (59) 1992 Niger, Africa 100 3 100 Energy-water-carbon exchange, process studies,

scaling, remote sensing science mescoscalemodeling

BOREAS (17 ) 1993–1996 Saskatchewan andManitoba, Canada

1000 3 1000 Energy-water-carbon exchange, carbon cyle andbiogeochemistry, terrestrial ecology processstudies, scaling, remote sensing science, mesoscalemodeling

GCIP (60) 1995–2000 Mississippi basin,United States

2000 3 2000 Energy-water exchange, scaling studies, mesoscalemodels, application of remote sensing

LBA (61) 1998–2000 Amazon basin, Brazil 2000 3 2000 Energy-water-carbon exchange, carbon cycle andbiogeochemistry, terrestrial-ecology processstudies, scaling, remote sensing science, mesoscalemodeling

ARTICLE

SCIENCE z VOL. 275 z 24 JANUARY 1997 507

features has been used to compare the radi-ative and physiological effects of doubledCO2 on climate (49). A series of 30-yearGCM runs were conducted in which theeffects of increased CO2 on stomatal func-tion were simulated in addition to the con-ventional CO2 “greenhouse effect” (radia-tive warming of the atmosphere). The im-pacts were studied separately and in combi-nation. The results indicated that fordoubled CO2 conditions, evapotranspira-tion would drop over the continents andthat air temperatures would increase signif-icantly over the tropical land masses, am-plifying the changes resulting from atmo-spheric radiative effects.

Process Studies and theDevelopment of Remote

Sensing Techniques

The process models embedded in LSPs areprimarily based on local-scale observationsand paradigms. For example, the leaf-scaleA-gs models and associated canopy radiativetransfer schemes were originally based onstudies conducted on spatial scales of a fewmicrometers (chloroplasts) up to a fewmeters (a typical soil-vegetation experi-ment plot). Such submodels were usuallyapplied unchanged and largely untestedwithin AGCMs to describe area-averagedprocesses over thousands of square kilome-ters. Some landscape-scale validation ofLSPs was seen as being essential (50). It wasalso recognized that satellite remote sensingoffered the most feasible, consistent, andaccurate means of providing global fieldsof land surface parameters, provided thattechniques could be developed to rigor-ously interpret the radiances and convertthem into useful biophysical quantities. Aseries of large-scale field experiments wasset up to solve both problems (Table 1).The experiments have been conducted atdifferent times and places and have haddiffering scientific emphases. However,they all use a nested framework that per-mits a progressive comparison of measure-ments made by surface instrumentation(scale: 1 to 10 m), surface flux equipment(10 m to 1 km), airborne remote sensingequipment (a few hundred meters to sev-eral kilometers), airborne eddy correlation(several kilometers), and satellite remotesensing (30 m to global scale). Coverage ofa range of biomes (grassland, agriculture,arid zone, boreal forest, and tropical for-est) will be achieved over a period of morethan a decade (Table 1). The rate ofprogress has been limited by resourceavailability and the need for the sciencecommunity to assimilate the results ofthese complex experiments as theyprogress. A number of summary texts have

been published for each experiment (Ta-ble 1), so only the salient results are men-tioned here.

In general, the local-scale representa-tions of radiative transfer, turbulent ex-change, and carbon and water fluxes scaledremarkably well over a wide range of spatialscales. Simple “aggregation” rules can oftenbe used to define effective parameter valuesover large areas, at least for energy, water,and carbon fluxes over grasslands and tem-perate and boreal forests (51, 52). Satelliteremote sensing techniques have beenshown to provide useful estimates of surfaceradiation budget components (S, Rn, εsTs

4,and sometimes Lw), surface reflectances,surface soil moisture, FPAR, and LAI (52).For surface soil moisture, FPAR, and LAI, aseries of numerical experiments made use ofthe fine-resolution data from field observa-tions to compare “aggregate calculations,”that is, the sum of many pixel-by-pixel cal-culations, with “bulk” calculations per-formed for the same area using average val-ues of the surface and atmospheric forcingparameters. The results agreed to within afew percent and to well within the uncer-tainties of the estimates provided by aver-aging surface and airborne eddy correlationmeasurements ('10 to 20%), except forvery heterogeneous soil moisture conditions(53). Remote sensing techniques have beenimproved to the point where global satellitedata can be transformed into fields ofFPAR, LAI, albedo, roughness length, andother surface parameters and released to thescientific community (54).

Future Directions

A third-generation LSP-GCM has beenused to calculate credible time-series fieldsof global atmospheric CO2 concentrationand realistic surface-atmosphere carbon andwater fluxes for a number of instrumentedsites around the world (48, 55). This kindof simulation, in which physical climateand carbon-cycle processes are directly cou-pled, will be useful in advancing our under-standing of the dynamics of the global car-bon cycle.

Diagnostic analyses have shown that theglobal carbon cycle is intricately linked tothe physical climate system. There are in-dications that the terrestrial biosphere act-ed as a large sink for atmospheric CO2during the late 1980s, apparently as a resultof anomalies in land surface temperatures,and possibly precipitation anomalies, whichled to imbalances in photosynthesis andrespiration (56). Large-scale physiologicaland physical climate system effects willboth be important in determining the rateof increase in atmospheric CO2 and thephysical and biological responses of the

Earth system to it over the next few decades(49). The need for realistic and accuratemodels becomes more urgent.

The third-generation LSPs point theway to future land models that can be cou-pled with comprehensive atmospheric andocean models to explore different globalchange scenarios. We expect future modelsto incorporate better descriptions of hydrol-ogy, soil respiration, and ecological respons-es to climate change than we have today.They will also be required to calculate iso-topic fractionation processes as the ob-served patterns of atmospheric carbon andoxygen isotopes, as well as CO2 and O2concentrations, can provide insights intocarbon cycle dynamics. Running thesemodels over the 1980 to 1995 period, usinghistorical satellite data as boundary condi-tions, should greatly increase our under-standing of the linked energy-water-carboncycles. In particular, the analysis of modelresults for anomalous years—for example,years with large air temperature and CO2concentration excursions—will help us toassess the likely future behavior of the Earthsystem in response to increasing atmospher-ic CO2.

REFERENCES AND NOTES___________________________

1. J. T. Houghton et al., Eds., Climate Change 1995,Science of Climate Change, Technical Summary(Cambridge Univ. Press, Cambridge, 1996), pp.9–97.

2. K. Trenberth, Ed., Climate Systems Modeling (Cam-bridge Univ. Press, Cambridge, 1992).

3. P. J. Sellers et al., Remote Sensing Environ. 51, 3(1995).

4. S. Manabe, Mon. Weather Rev. 97, 739 (1969);iiii, J. Holloway, J. Leigh Jr., J. Geophys. Res.80, 1617 (1975); S. Manabe and R. T. Wetherald, J.Atmos. Sci. 44, 1211 (1987); G. A. Meehl and W. M.Washington, ibid. 45, 1476 (1988); J. F. B. Mitchell,in Variations in the Global Water Budget, A. Street-Perrott, M. Beran, R. Ratcliffe, Eds. (Reidel, Hing-ham, MA, 1983), pp. 429–446; S. H. Schneider andR. E. Dickinson, Rev. Geophys. Space Phys. 12, 447(1974).

5. R. E. Dickinson, Geophys. Monogr. Am. Geophys.Union 29, 58 (1984).

6. P. J. Sellers, Y. Mintz, Y. C. Sud, A. Dalcher, J.Atmos. Sci. 43, 305 (1986).

7. A. W. Kuchler, Goode’s World Atlas (Rand McNally,New York, ed. 16, 1983), pp. 16–17; J. L. Dormanand P. J. Sellers, J. Appl. Meteorol. 28, 833 (1989);E. Matthews, J. Clim. Appl. Meteorol. 22, 474 (1984);J. S. Olson, J. A. Watts, L. J. Allison, Carbon in LiveVegetation of Major World Ecosystems (U.S. Depart-ment of Energy, Washington, DC, 1983); C. J. Will-mott and K. Klink, in ESA SP-248 (European SpaceAgency, Paris, 1986), pp. 109–112; M. F. Wilsonand A. Henderson-Sellers, J. Clim. 5, 119 (1985).

8. P. J. Sellers et al., J. Clim. 9, 676 (1996).9. P. J. Sellers et al., ibid., p. 706.

10. G. B. Bonan, J. Geophys. Res. 100, 2817 (1995); R.E. Dickinson, R. Bryant, L. Graumlich, in preparation.

11. W. M. Washington and C. L. Parkinson, An Introduc-tion to Three Dimensional Climate Modeling (Univer-sity Science Books and Oxford Univ. Press, Oxford,1986).

12. J. W. Deardorff, J. Geophys. Res. 83, 1889 (1977).13. R. E. Dickinson, J. Clim. 1, 1086 (1988); D. A. War-

rilow, in ESA SP-248 (European Space Agency, Par-is, 1986), pp. 143–150.

14. M. I. Budyko, Climate and Life (Academic Press,

SCIENCE z VOL. 275 z 24 JANUARY 1997508

New York, 1974).15. C. A. Paulson, J. Appl. Meteorol. 9, 857 (1970); M. R.

Raupach and A. S. Thom, Annu. Rev. Fluid Mech.13, 97 (1981).

16. J. R. Garratt, J. Clim. 6, 419 (1993).17. P. J. Sellers et al., Bull. Am. Meteorol. Soc. 76, 1549

(1995).18. W. J. Shuttleworth et al., Q. J. R. Meteorol. Soc. 110,

1163 (1984); W. J. Shuttleworth et al., ibid., p. 1143.19. J. G. Charney, W. J. Quirk, S. H. Chow, J. Kornfield,

J. Atmos. Sci. 34, 1366 (1977); Y. C. Sud and M. J.Fennessy, J. Clim. 2, 105 (1982).

20. D. J. Carson and A. B. Sangster, in Numerical Ex-perimentation Programme Report (UK MeteorologyOffice, Bracknell, UK, 1981), no. 2, pp. 5.14–5.21.

21. Y. C. Sud and W. E. Smith, Boundary Layer Meteo-rol. 33, 15 (1985).

22. Y. Mintz, in The Global Climate, J. Houghton, Ed.(Cambridge Univ. Press, Cambridge, 1984), pp. 79–105; J. Shukla and Y. Mintz, Science 215, 1498(1982); J. M. Walker and P. R. Rowntree, Q. J. R.Meteorol. Soc. 103, 29 (1977); D. L. Warrilow, A. B.Sangster, A. Slingo, Met 0 20, Tech. Note 38 (UKMeteorology Office, Bracknell, UK, 1986); D. L. War-rilow and E. Buckley, Ann. Geophys. 7, 439 (1989).

23. D. J. Carson, in Proceedings of the JSC Study Con-ference on Land-Surface Processes in AtmosphereGeneral Circulation Models, P. S. Eagleton, Ed.(Cambridge Univ. Press, Cambridge, 1981), pp. 67–108.

24. I. R. Cowan and G. D. Farquhar, Symp. Soc. Exp.Biol. 31, 471 (1977).

25. P. G. Jarvis, Philos. Trans. R. Soc. London Ser. B273, 593 (1976).

26. P. S. Nobel, Introduction to Biophysical Plant Phys-iology (Freeman, San Francisco, 1974).

27. R. E. Dickinson, A. Henderson-Sellers, C. Rosenz-weig, P. J. Sellers, Agric. For. Meteorol. 54, 373(1991).

28. P. J. Sellers, Int. J. Remote Sensing 8, 1335 (1985).29. J. L. Monteith, Principles of Environmental Physics

(Arnold, London, 1973).30. R. E. Dickinson, Rev. Geophys. 33 (suppl.), 217

(1995); K. L. Brubaker, D. Entekhabi, P. S. Eagleson,J. Clim. 6, 1077 (1993).

31. N. Sato et al., J. Atmos. Sci. 46, 2757 (1989).32. P. Viterbo and A. C. M. Beljaars, J. Clim. 8, 2716

(1995).33. A. C. M. Beljaars, P. Viterbo, M. J. Miller, A. K. Betts,

Mon. Weather Rev. 124, 362 (1996); A. K. Betts, J.

H. Ball, A. C. M. Beljaars, M. J. Miller, P. Viterbo, J.Geophys. Res. 101, 7209 (1996).

34. R. E. Dickinson and A. Henderson-Sellers, Q. J. R.Meteorol. Soc. 114, 439 (1988); A. Henderson-Sell-ers and V. Gornitz, Clim. Change 6, 231 (1984); C. A.Nobre, P. J. Sellers, J. Shukla, J. Clim. 4, 957 (1991);A. Henderson-Sellers et al., J. Geophys. Res. 98,7289 (1993); J. Lean and D. A. Warrilow, Nature 342,411 (1989); J. Lean and P. R. Rowntree, Q. J. R.Meteorol. Soc. 119, 509 (1993).

35. Y. Xue and J. Shukla, J. Clim. 6, 2232 (1991).36. R. Koster and M. Suarez, J. Geophys. Res. 97, 2697

(1992).37. J. Noilhan, P. Lacarrere, P. Bougeault, Mon. Weath-

er Rev. 119, 2393 (1991); R. A. Pielke and R. Avissar,Landscape Ecol. 4, 133 (1990).

38. J. T. Houghton, G. J. Jenkins, J. J. Ephraums, inClimate Change: The IPCC Scientific Assessment,World Meteorological Organization–United NationsEnvironment Programme, Eds. (Cambridge Univ.Press, Cambridge, 1990).

39. G. D. Farquhar, S. von Caemmerer, J. A. Berry,Planta 149, 78 (1980); G. J. Collatz, J. T. Ball, C.Grivet, J. A. Berry, Agric. For. Meteorol. 54, 107(1991); S. von Caemmerer and G. D. Farquhar, inProceedings of the 1983 Conference at Tallinn, J.Vill, G. Girishina, A. Laisk, Eds. (Estonian Academy ofSciences, Tallinn, 1985), pp. 46–58.

40. For some tropical heat–adapted (C4) species, theterms wc and we still refer to Rubisco and light limi-tations, respectively, but ws refers to another bio-chemical (phosphoenolpyruvate carboxylase) limita-tion [G. J. Collatz, M. Ribas-Carbo, J. A. Berry, Aust.J. Plant Physiol. 19, 519 (1992)].

41. O. Bjorkman, in Plant Physiological Ecology I, vol.12A of Encyclopedia of Plant Physiology, O. L.Lange et al., Ed. (Springer-Verlag, Berlin, 1981), pp.57–107; T. Hirose and M. J. A. Werger, Oecologia72, 520 (1987); C. B. Field and H. A. Mooney, in Onthe Economy of Plant Form and Function, T. J.Givnish, Ed. (Cambridge Univ. Press, Cambridge,1986), pp. 22–55.

42. J. T. Ball, thesis, Stanford University (1988).43. P. J. Sellers, J. A. Berry, G. J. Collatz, C. B. Field, F.

G. Hall, Remote Sensing Environ. 42, 187 (1992).44. F. G. Hall, K. F. Huemmrich, S. J. Goetz, P. J. Sellers,

J. E. Nickeson, J. Geophys. Res. 97, 19061 (1992).45. G. Asrar, M. Fuchs, E. T. Kanemasu, J. L. Hatfield,

Agron. J. 76, 300 (1984); C. J. Tucker, B. N. Holben,J. H. Elgin, E. McMurtrey, Remote Sensing Environ.

11, 171 (1981).46. P. J. Sellers, Remote Sensing Environ. 21, 143

(1987); F. G. Hall, K. F. Huemmrich, S. N. Goward,ibid. 32, 47 (1990); R. B. Myneni, G. Asrar, D. Tanre,B. J. Choudhury, IEEE Trans. Geosci. Remote Sens-ing 30, 302 (1992).

47. S. O. Los, C. O. Justice, C. J. Tucker, Int. J. RemoteSensing 15, 3493 (1994); C. J. Tucker, I. Fung, C. D.Keeling, R. H. Gammon, Nature 319, 195 (1986).

48. D. A. Randall et al., J. Clim. 9, 738 (1996).49. P. J. Sellers et al., Science 271, 1402 (1996).50. S. I. Rasool and H.-J. Bolle, Int. Satell. Land Surf.

Climatol. Proj. Rep. 1 (National Center for Atmo-spheric Research, Boulder, CO, 1983).

51. J.-C. Andre et al., Ann. Geophys. 6, 477 (1988).52. F. G. Hall and P. J. Sellers, J. Geophys. Res. 100,

25383 (1995); P. J. Sellers and F. G. Hall, Eds., ibid.97 (no. 17) (1992).

53. P. J. Sellers, M. D. Heiser, F. G. Hall, ibid. 97, 19033(1992); P. J. Sellers et al., ibid. 100, 25607 (1995).

54. P. J. Sellers et al., Bull. Am. Meteorol. Soc. 77, 1987(1996); B. W. Meeson et al., ISLSCP Initiative I: GlobalData Sets for Land-Atmosphere Models, 1987–1988[CD-ROM] (GSFC-DAAC, NASA Goddard SpaceFlight Center, Greenbelt, MD, 1995), vols. 1–5 (USA_NASA_GDAC_ ISLSCP_001_ USA_NASA_GDAAC_ISLSCP_005).

55. A. S. Denning et al., Tellus B 48, 521 (1996); A. S.Denning, D. A. Randall, J. G. Collatz, P. J. Sellers,ibid., p. 543.

56. P. Ciais, P. P. Tans, M. Trolier, J. W. C. White, R. J.Francey, Science 269, 1098 (1995); A. S. Denning, I.Y. Fung, D. A. Randall, Nature 376, 240 (1995); C. D.Keeling, T. P. Whorff, M. Wahlen, J. van der Plicht,ibid. 375, 666 (1995); P. P. Tans, I. Y. Fung, T.Takahashi, Science 247, 1431 (1990).

57. D. Deering and V. Kozoderov, in preparation.58. H.-J. Bolle et al., Ann. Geophys. 11, 173 (1993).59. J. P. Goutorbe et al., ibid. 12, 53 (1994); S. D. Prince

et al., Remote Sensing Environ. 51, 215 (1995).60. GEWEX Continental-Scale International Project

(GCIP) J. Geophys. Res. 101 (no. 3) (1996).61. J. C. Gash et al., “A Concise Plan for the Large-Scale

Biosphere-Atmosphere Experiment in Amazonia,”report available from C. A. Nobre.

62. Supported by NASA Earth Observing System funds(Sellers-Mooney Interdisciplinary Science Project).We gratefully acknowledge encouragement andsupport from G. Asrar and D. L. Williams and thankL. East and V. Corey for typing the manuscript.

ARTICLE

SCIENCE z VOL. 275 z 24 JANUARY 1997 509