climate change effects on vegetation distribution and ... · pdf fileclimate change effects on...

F O R E S T B I O D I V E R S I T Y U N D E R G L O B A L C H A N G E

Climate Change Effects onVegetation Distribution and Carbon

Budget in the United StatesDominique Bachelet,1* Ronald P. Neilson,2 James M. Lenihan,3 and

Raymond J. Drapek4

1Department of Bioresource Engineering, Oregon State University, Corvallis, Oregon 97331, USA; 2USDA Forest Service, ForestrySciences Laboratory, Corvallis, Oregon 97331, USA; 3Department of Botany and Plant Pathology, Oregon State University,Corvallis, Oregon 97331, USA; and 4Department of Forest Science, Oregon State University, Corvallis, Oregon 97331, USA

ABSTRACTThe Kyoto protocol has focused the attention of thepublic and policymarkers on the earth’s carbon (C)budget. Previous estimates of the impacts of vegeta-tion change have been limited to equilibrium “snap-shots” that could not capture nonlinear or thresholdeffects along the trajectory of change. New modelshave been designed to complement equilibrium mod-els and simulate vegetation succession through timewhile estimating variability in the C budget and re-sponses to episodic events such as drought and fire. Inaddition, a plethora of future climate scenarios hasbeen used to produce a bewildering variety of simu-lated ecological responses. Our objectives were to usean equilibrium model (Mapped Atmosphere–Plant–Soil system, or MAPSS) and a dynamic model (MC1)to (a) simulate changes in potential equilibrium veg-etation distribution under historical conditions andacross a wide gradient of future temperature changesto look for consistencies and trends among the manyfuture scenarios, (b) simulate time-dependentchanges in vegetation distribution and its associated Cpools to illustrate the possible trajectories of vegeta-tion change near the high and low ends of the tem-

perature gradient, and (c) analyze the extent of theUS area supporting a negative C balance. Both modelsagree that a moderate increase in temperature pro-duces an increase in vegetation density and carbonsequestration across most of the US with smallchanges in vegetation types. Large increases in tem-perature cause losses of C with large shifts in vegeta-tion types. In the western states, particularly southernCalifornia, precipitation and thus vegetation densityincrease and forests expand under all but the hottestscenarios. In the eastern US, particularly the South-east, forests expand under the more moderate scenar-ios but decline under more severe climate scenarios,with catastrophic fires potentially causing rapid vege-tation conversions from forest to savanna. Both mod-els show that there is a potential for either positive ornegative feedbacks to the atmosphere depending onthe level of warming in the climate change scenarios.

Key words: global climate change; simulationmodel; biogeography; carbon budget; MAPSS;MC1.

INTRODUCTION

The Kyoto protocol agreement of December 1997has focused the attention of the public and policy-

makers on the earth’s carbon (C) budget. It hasfostered a continuing search for a more accuratequantification of global terrestrial C sources andsinks to mitigate global climate change by conserv-ing or increasing C sequestration. To estimate thesize of C pools and fluxes, scientists have used bio-geochemical models that simulate steady-state con-ditions using long-term average climatic records, or

Received 12 May 2000; accepted 22 November 2000.*Current address: Forestry Sciences Laboratory, 3625 93rd Avenue, Olym-pia, Washington 98512-9193, USA; e-mail: [email protected]

Ecosystems (2001) 4: 164–185DOI: 10.1007/s10021–001–0002-7 ECOSYSTEMS

© 2001 Springer-Verlag

164

temporal dynamics and interannual variability us-ing annual climatic records. However, the climatecan influence the C budget not only directly byaffecting the flux rates but also indirectly by affect-ing both the disturbance regime and the vegetationtype. Earlier research on climate change focused onchanges in vegetation types and their C pools underdoubled atmospheric carbon dioxide (CO2) concen-tration and associated climatic changes (VEMAP1995). MAPSS (Mapped Atmosphere–Plant–SoilSystem), a biogeography model, has been used inthis context to simulate vegetation distribution un-der several equilibrium climate change scenarios(VEMAP 1995; Neilson and others 1998; Neilsonand Drapek 1998). The resulting vegetation mapshave then been used by biogeochemistry models todetermine the carbon budget estimates.

The equilibrium models require only long-termaverage climate and use biogeographic rules basedon that climate to determine the vegetation typeand its density. So the questions that remainedwere: How did this vegetation type come to be? Hasit been a smooth transition from one vegetationtype to another, or have there been abrupt changesand many transition states? Has there been a lineardecline or an increase in biomass? To complementthe “equilibrium snapshots” provided by the staticmodels and answer these questions, new modelshave been designed to use transient climate andsimulate dynamic changes in the vegetation typeand its associated biogeochemical cycle. MC1 (Dalyand others 2000) is one example of the new dy-namic global vegetation models (DGVM) that cannow illustrate year-to-year variability in the C bud-get due to climatic variations such as drought anddisturbances such as fire under more realistic cli-mate change scenarios that include the influence ofdynamic oceans and aerosol forcing.

The existence of a large number of future climatechange scenarios suggests that there is considerableuncertainty about possible future ecological im-pacts, particularly since some scenarios produce op-posite sign ecological responses. However, Neilsonand Drapek (1998) have discerned some patternsamong six equilibrium scenarios and developed thehypothesis that moderate warming could produceincreased vegetation growth over broad areas in theconterminous United States but greater warmingcould also produce large areas of drought stress andC losses.

In this paper, we further tested this hypothesis byproviding a bridge between the sensitivity analysisof a static model (MAPSS) under many equilibriumscenarios and a time series analysis of a dynamicmodel (MC1) using only two of these scenarios but

treating them as transient scenarios—one with onlya small amount of warming and one with a largeamount of warming across the United States.MAPSS was run under seven equilibrium climatechange scenarios at 10-km resolution. MC1 was rununder two transient scenarios at 0.5° latitude/lon-gitude resolution (approximately 50-km). The rea-son for the discrepancy between the number andthe spatial resolution of the two types of scenarios isthat (a) there are fewer transient climate changescenarios than equilibrium scenarios and (b) thereare no transient climate change scenarios at 10-kmresolution for the conterminous United States. Tocomplete the bridge between equilibrium and tran-sient analyses, we had to first show that the twomodels were reasonably consistent with each otherin their responses under the various scenarios. TheMAPSS and MC1 models are quite different in theirstructure and conception, so we did not expectperfect consistency. However, the broad patterns offuture changes, particularly with respect to the signand the location of the changes, had to be generallyconsistent. Therefore, MAPSS was also run at 0.5°resolution for the same two transient scenarios afterthey were averaged and transformed into equilib-rium scenarios, and the results were compared tothose of MC1.

Our objectives were thus to use both the staticmodel MAPSS and the dynamic model MC1 tosimulate (a) vegetation distribution (both models),(b) associated LAI (leaf area index—that is, area ofleaves per unit area of ground, MAPSS) or biomass(MC1) and the interannual variations of the Cfluxes (MC1), and (c) the resulting change in theextent and location of the stress areas (defined asareas of vegetation density reduction) in the con-terminous US under historical conditions and un-der various climate change scenarios. Both modelssimulate vegetation changes using biogeographicrules based on climatic indices. Both simulate veg-etation density. The equilibrium model worksmuch more simply, using just LAI, which is a sur-rogate for vegetation density (high LAI for closedcanopy, low LAI for low vegetation density),whereas the dynamic model simulates vegetationand soil C pools and fluxes. We focused our analysison the conterminous United States, in part, becauseit may constitute a large C sink (Fan and others1998) and also because we relied on the climate andsoils datasets of the Vegetation/Ecosystem Model-ing and Analysis Project (VEMAP), which are cur-rently the highest quality datasets available (Kitteland others 1995).

The summaries of C gains or losses for the entireUS can mask dramatic regional impacts. In our

Climate Change Effect 165

study, simulated vegetation density in virtually alllocations across the conterminous United States ei-ther increased or decreased under future climatescenarios. Even when the national average indi-cates an overall C gain under conservative climatechange scenarios, regional droughts and fires canstill cause significant distress to local ecological andeconomic systems. The rate of increase in forestvegetation density is constrained by its growth rate,which can lag behind climate change. On the otherhand, the rate of decline in vegetation density isphysically constrained by the rate of climate changeand by episodic disturbance, and it can occur rap-idly, even in forests. Simulated changes in eitherLAI or biomass were calculated to provide someindication of regional stress due to either drought orcold temperatures and some indication of regionalgains due to moderate warming and increased pre-cipitation. The location and extent of these regionalstress areas is reported here because they shouldbecome the focus of the attention of land managersinterested in sustainability issues.

METHODS

Models

Although there are many future climate changescenarios for equilibrium conditions that span awide temperature increase gradient, there are stillfew transient scenarios available. Therefore, weused the MAPSS equilibrium model to illustratepotential vegetation responses to the variety ofequilibrium climate change scenarios, and we alsoused MC1 under two transient climate scenarioscorresponding to the extremes of the temperaturerange of the equilibrium scenarios to illustrate thepossible trajectories of vegetation types and C pools.

The MAPSS model (Neilson 1995) is an equilib-rium biogeography model that includes a mecha-nistic hydrology module that calculates plant avail-able water and a set of biogeography rules thatdetermine climatic zone, life form, and plant type asa function of temperature thresholds and wateravailability (http://www.fs.fed.us/pnw/corvallis/mdr/mapss). MAPSS determines the maximum po-tential LAI a site can support, based on 1 year oflong-term monthly average climate data, assumingthe vegetation can use all available soil water. Itsimulates a CO2-induced increase in water use ef-ficiency by reducing stomatal conductance by 35%at double the present CO2 concentration (Eamus1991). MAPSS uses an aerodynamic evapotranspi-ration approach that is sensitive to canopy charac-teristics. Grasses and trees have different rooting

depths in a three-layer soil and compete for avail-able soil water, while shading by trees limits grassgrowth. Vegetation classification in MAPSS is basedon climatic thresholds and the presence/absenceand LAI values of three life forms—trees, shrubs,and grasses—with differing leaf characteristics,thermal affinities, and seasonal phenology. Eithertrees or shrubs are assumed to be dominant andmutually exclusive. There are 52 possible vegeta-tion classes, but we aggregated them into 11 tosimplify the visualization of the results in this paper(Table 1). MAPSS also includes a fire module thatmaintains prairie grasslands and transition zonessuch as the prairie peninsula. MAPSS cannot di-rectly convert areas of decline in vegetation densityto specific predictions of the amount of C lost be-cause the simulations are based on leaf area, not Cpool size.

MCI (http://www.fsl.orst.edu/dgvm) is the firstversion of a dynamic global vegetation model basedon the linkage of MAPSS (Neilson 1995) and CEN-TURY (Parton and others 1987). The model (Dalyand others 2000) reads monthly climate time seriesand calls first a set of biogeography rules that de-termines the vegetation type. Secondly, the modelcalls a modified version of the CENTURY model,where parameters vary as a function of the life formcombination defined by the biogeography rules forthat year. Once the C budget has been establishedand the soil moisture estimated, a fire module iscalled to determine if fuel load and fuel moistureare conducive to fire. Given adequate climatic con-ditions, a fire is simulated and C pools modifiedconsequently. If the fire is extreme, the vegetationtype may then be modified; for example, a burntforest can become a grassland or a savanna.

MC1 simulates mixtures of deciduous/evergreenand needleleaf/broadleaf trees and C3/C4 grass lifeforms using climatic thresholds modified from theoriginal MAPSS biogeography rules. MC1 classifieswoody and herbaceous life forms into 24 differentvegetation classes based on their leaf biomass sim-ulated by a biogeochemistry module. The 24 classeswere aggregated to 11 classes to simplify the visu-alization of the results in this paper (Table 1). Thebiogeochemistry module, a modified version of theCENTURY model, simulates plant production, soilorganic matter decomposition, and water and nu-trient cycling. It includes competition between treesand grasses for light, nutrients, and water. Produc-tion increases and transpiration decreases by 25%as atmospheric CO2 concentration increases to 700ppm. The hydrology module is a simple bucketmodel that, unlike MAPSS, does not include unsat-urated flow. Potential evapotranspiration is calcu-

166 D. Bachelet and others

Table 1. Vegetation Classes: Comparison between MAPSS, VEMAP, and the Simplified Classification Usedin this Paper

Simplified Classes (this Paper) VEMAP Classesa MAPSS Classes

1. Tundra 1. Tundra 601. Tundra2. Taiga–Tundra 22. Taiga 600. Taiga–Tundra3. Conifer Forest 2. Boreal Coniferous Forest 107. Forest EN Taiga

3. Maritime Temperate Coniferous Forest 108. Forest Mixed warm EN4. Continental Temperate Coniferous Forest 112. Forest EN Maritime23. Boreal Larch Foresta 113. Forest EN Continental

4. NE Mixed Forest 5. Cool Temperate Mixed Forest 102. Forest Mixed Cool5. Temperate Deciduous Forest 7. Temperate Deciduous Forest 100. Forest Deciduous Broadleaf

111. Forest Hardwood cool6. SE Mixed Forest 6. Warm Temperate—Subtropical Mixed Forest 101. Forest Mixed Warm DEB7. Tropical Broadleaf Forest 8. Tropical Deciduous Forest 105. Forest EB Tropical

9. Tropical Evergreen Forest8. Savannas and Woodlands 11. Temperate Coniferous Xeromorphic Woodland 109. Forest Seasonal Tropical ED

13. Temperate Subtropical Savanna 110. Forest Savanna Dry Tropical ED15. Temperate Coniferous Savanna 200. Tree Savanna DB16. Tropical Deciduous Savanna 201. Tree Savanna Mixed Warm DEB

205. Tree Savanna Mixed Cool EN206. Tree Savanna Mixed Warm EN207. Tree Savanna EN Maritime208. Tree Savanna EN Continental209. Tree Savanna PJ Continental210. Tree Savanna PJ Maritime211. Tree Savanna PJ Xeric Continental

9. Shrubs and Woodlands 10. Temperate Mixed Xeromorphic Woodland 301. Open Shrubland—No Grass12. Tropical Thorn Woodland 302. Shrub Savanna DB14. Warm Temperate Subtropical Mixed Savanna 303. Shrub Savanna Mixed Warm DEB19. Mediterranean Shrubland 307. Shrub Savanna Mixed cool EN20. Temperate Arid Shrubland 308. Shrub Savanna EN

310. Shrub Savanna Subtropical Mixed311. Shrubland Subtropical Xeromorphic312. Shrubland Subtropical Mediterranean313. Shrubland Temperate Conifer314. Shrubland Temperate XeromorphicConifer423. Grass Semi Desert C3424. Grass Semi Desert C3-C4

10. Grasslands 17. C3 Grasslands 414. Grass Tall C318. C4 Grasslands 415. Grass Mid C3

416. Grass Short C3417. Grass Tall C3 C4418. Grass Mid C3 C4419. Grass Short C3 C4420. Grass Tall C4421. Grass Mid C4422. Grass Short C4

11. Arid Lands 21. Subtropical Arid Shrubland 305. Shrub Savanna Tropical EB309. Shrub Savanna Mixed Warm EN404. Grass Semi Desert425. Grass Semi Desert C4500. Desert Boreal501. Desert Temperate502. Desert Subtropical503. Desert Tropical504. Desert Extreme

E, evergreen; N, needleleaf; D, deciduous; B, broadleafaMC1 simulates two categories beyond the 21 VEMAP classes: 22. Taiga and 23. Boreal Larch Forests.


lated using Linacre’s (1977) equations. Nitrogendemand is always met in the current version of themodel. Parameterization of the biogeochemical pro-cesses is based on the life form composition, whichis updated annually by the biogeography module.The model also simulates the occurrence, behavior,and effects of severe wild fires (Lenihan and others1998). Allometric equations are used to convertaboveground live and dead biomass to fuel classes.Fuel loading and fuel moisture thresholds are usedto determine fire occurrence. Plant mortality, fireemissions, and live and dead biomass consumptionare estimated as functions of fire spread, fire lineintensity, and vegetation structure. Fire effects feedback to the biogeochemistry module to adjust thelevels of the C and nutrient pools.

Because MC1 is a carbon accounting model, veg-etation and soil carbon pools need to be initialized.Soil C pools require a long time to build up. Con-sequently, the model is run on long-term meanclimate (1 year of average monthly climate data)until the slow soil C pool equilibrates. This mayrequire up to 3000 simulation years, depending onthe ecosystem being simulated (Daly and others2000). Because dynamic fire events cannot be sim-ulated meaningfully using a mean climate, fireevents are scheduled at regular intervals that varywith vegetation type. Once the soil C has equili-brated, the model is run on variable climate (spin-up period) allowing for fuel buildup and variablefuel moisture conditions until the aboveground Cpools are at equilibrium with dynamic fire events.The spin-up climate timeseries is based on historicalclimate. Since the historical records contain knowntrends in both temperature and precipitation, theywere detrended using a 30-year filter (VEMAPMembers unpublished). After detrending, the long-term temperature and precipitation monthly aver-ages were set equal to those of the first 15 years ofthe observed record (1895–1909) to provide asmooth transition into the observed records in1895.

Climate Scenarios

Seven future climate scenarios generated by generalcirculation models (GCM) were used by MAPSS at10-km resolution (Kittel and others 1995; Neilsonand Drapek 1998), and two of them were used byMC1 and MAPSS at a 0.5° latitude/longitude reso-lution. Fine-scale features of the climate, related totopographic effects, are better represented in thehigher-resolution (10-km) data set, and the largenumber of equilibrium scenarios provides a greatercontext to assess possible future changes. However,

there are currently no 10-km transient climate datasets.

The scenarios span a range of about 2.8–6.6°C inprojected average annual temperature increaseover the conterminous United States near the endof the 21st century (Figure 1). Four are equilibriumscenarios (GFDL-R30, GISS, UKMO, OSU) thatwere included in the First Assessment Report of theIntergovernmental Panel on Climate Change(IPCC) (Cubasch and Cess 1990). They include asingle-layered ocean and assume an instantaneousdoubling of CO2. Three scenarios are transient and

Figure 1. GCM-simulated changes in (a) temperatureand (b) precipitation between historical and future con-ditions (at or near doubled CO2 values), aggregated overthe conterminous United States. Transient model datawere averaged for 1961–90 and 2061–99 and the differ-ence are reported here. Transient GCMs includeHADCM2SUL, HADCM2GHG (Johns and others 1997),and CGCM1 (Boer and others 1999). Equilibrium (23CO2) GCMs include OSU (Schlesinger and Zhao 1989),(Johns and others 1997), GFDL-R30 (Manabe and Weth-erland 1990), GISS (Hansen and others 1988), andUKMO (Wilson and Mitchell 1987).


were included in the Second Assessment Report ofthe IPCC (Gates and others 1996). Two transientscenarios come from the Hadley Climate Center(HADCM2GHG and HADCM2SUL, the latter ofwhich includes effects of sulfate aerosols), and onecomes from the Canadian Climate Centre (CGCM1,also including aerosols). Transient GCM include afully dynamic 3-D ocean and are run from the1800s to the present using observed CO2 increasesand into the future using IPCC projections of futuregreenhouse gas concentrations (IS92a) (Kattenbergand others 1995). The last 30 years of the threetransient scenarios were averaged so they could betreated as equilibrium scenarios by the biogeogra-phy model, MAPSS. However, the transient scenar-ios were clearly not at equilibrium, having attainedonly about half to two-thirds of their eventual tem-perature change, due to thermal inertia of theoceans (Gates and others 1996). OnlyHADCM2SUL and CGCM1 were used to run MC1.The 0.5° latitude/longitude transient climate dataset was generated in the context of the VEMAP,aggregating some of the 10-km data, such as pre-cipitation. The baseline historical climate corre-sponds to VEMAP Phase 1 baseline climate.

Calculation of Stress Area

Using both models (MAPSS and MC1), we simu-lated the fractional area of US land that underwenta decline in vegetation density (where vegetationdensity was less than its long-term mean) under allscenarios (seven for MAPSS, two for MC1) andtermed this the “stress area index,” or SAI. UsingMC1, we followed the temporal trajectory of theSAI through the past 100 years and into the futureto the end of the 21st century. Because MAPSSsimulates LAI rather than biomass, we usedchanges in LAI as an approximation of changes invegetation density and compared them withchanges in biomass calculated by MC1. Eventhough LAI and biomass are curvilinearly related,changes in both should follow the same directionand occur in the same geographic regions.

In most cases, the cause of stress is drought thatfollows either a reduction in precipitation, an in-crease in evaporative demand (higher tempera-tures), or both. Our analysis is similar to the spatialmapping of the Palmer drought severity index(PDSI) (Palmer 1965). The objective is to relateregional patterns of drought or vegetation stress tolarge-scale atmospheric circulation changes on bothshort and long time scales (see for example, Nigamand others 1999), because the SAI essentially tracksthe regional wet and dry zones partitioned by per-sistent jet stream positions.

MAPSS simulations at 10-km resolution. For eachgrid cell, we calculated the difference between theaverage LAI under a future climate change scenario(between 2070 to 2099 for the three originally tran-sient scenarios) and the average LAI for the histor-ical period from 1961 to 1990. The fraction of theUS lands and forested areas where that differencewas negative (LAI2070–2099, LAI 1961–1990) was de-fined as the SAI.

MCI simulations at 0.5° latitude/longitude. We firstcalculated the long-term average live vegetationcarbon simulated for the spin-up climate time series(detrended historical climate) described earlier. Wethen calculated, for each year of the simulation andfor each grid cell, the difference between the cur-rent year’s simulated live vegetation C and thatfrom the spin-up period. The fraction of the USlands and forested areas where that difference wasnegative (live vegetation carbonyear , live vegeta-tion carbon spin-up average) was defined as the SAI.

RESULTS AND DISCUSSION

Potential Vegetation DistributionMAPSS Results under seven GCM scenarios (10-km

resolution). Over 40% of the coniferous forests arereplaced by savannas under the UKMO scenario(Figures 2 and 3). Under all other scenarios (exceptGISS), coniferous forests expand slightly (Figure 2).The temperate deciduous forest shifts to morenorthern locations (Figure 3) and is replaced byeither the southeast mixed forest or savannas undermost scenarios except for the mildest one(HADCM2SUL). Northeast mixed forests are re-placed by either savannas (UKMO, GFDL) or thenorthward-shifting temperate deciduous forest.Southeast mixed forest are replaced mostly by sa-vannas under three scenarios: the UKMO scenarioand CGCM1, which project the largest increases intemperature (both above 5°C), and the GFDL sce-nario. Southeast mixed forest are even replacedpartially by grasslands under the two warmer sce-narios (Figure 3). Tropical forests appear as a newvegetation type mostly in Louisiana (Figure 3),where they replace the original Southeast mixedforest. The area covered by all forest types tends todecrease by the end of the 21st century under thewarmest scenarios (over 40% under the UKMOscenario, about 20% for CGCM1 and the GFDLscenario) (Figure 4a). In contrast, moderately warmscenarios (HADCM2SUL and HADCM2GHG) pro-duce increases in forest area of about 20%.

MAPSS simulates decreases in shrubland areaunder all scenarios except the UKMO scenario andHADCM2SUL, where shrubs replace grasses in ar-


eas of the Great Plains (Figure 3). Most shrublandlosses occur in the Great Basin, where increasedprecipitation drives the replacement of shrubs bysavannas. Savannas are simulated to increase bymore than 50% under three warm scenarios(UKMO, CGCM1, GFDL) (Figure 2). The area oc-cupied by grasslands is relatively stable (Figure 2),except for a 20% increase in area simulated underthe UKMO scenario, where they expand into Min-nesota and Wisconsin, and a 20% decrease underHADCM2SUL, where they are replaced mostly bysavannas (Figure 3). Simulations for the westernUnited States show 60% or greater reductions inthe area of deserts under HADCM2SUL,HADCM2GHG, and CGCM1 (Figure 2). MAPSSsimulates about a 50% decrease in the area of aridland under the GFDL scenario, but with only smallchanges to the extent of the Sonoran Desert (Figure3). MAPSS simulates a 20% increase in desert areaonly under the warmest scenario, UKMO (Figure2).

MAPSS simulates some of the largest percentagechanges at high elevations where taiga–tundra and

tundra decrease by more than 80% by the end ofthe 21st century (Figure 2).

MAPSS and MC1 results under two transient scenarios(0.5° latitude/longitude resolution). Under historicalclimate, MC1 simulates that over 40% of the US iscovered by grasslands, whereas MAPSS simulatesthat only about 25% of the US is covered by grass-lands (Figure 5a). MAPSS simulates about 15% ofthe US being covered by shrubs and woodlands,whereas MC1 simulates only about 1% of the UScovered by shrubs and woodlands.

In general, the two models agree with each otheron both the sign and magnitude of the simulatedfuture changes. Both models agree that the largestpercentage change is the disappearance of thetaiga–tundra and tundra vegetation types from thecontinental US due to increases in temperature(Figure 5). Both models simulate decreases in thearea of arid lands, due to increased precipitationand simulated increased water-use efficiency underelevated atmospheric CO2 (Figures 5 and 6). Bothmodels agree that grassland area decreases and thatshrub and woodland areas increase under both fu-

Figure 2. Percentage changein area (10-km resolution) of11 simplified vegetationtypes for the conterminousUS between current and fu-ture conditions under theseven GCM scenarios listedin Figure 1.


ture climate change scenarios (Figure 5). However,MAPSS simulates a 75% increase in savanna areaunder CGCM1, whereas MC1 simulates a 30% de-crease. Under HADCM2SUL, both models simulatelittle change in savannas and woodlands (small de-creases with MCI, small increases with MAPSS).Both models predict an encroachment of tropicalforests along the Gulf Coast in the Southeast underboth scenarios.

Both models predict small decreases in the area oftemperate deciduous forests under CGCM1 andsmall increases under HADCM2SUL (Figure 5). Thetwo models show losses of northeast mixed forestsunder CGCM1, large for MAPSS (approximately90%), moderate for MC1 (approximately 20%).However, the two models disagree on the fate ofnortheast mixed forests under HADCM2SUL: sig-nificant losses for MAPSS (approximately 65%) andmoderate gains for MC1 (about approximately30%), especially in Minnesota and Wisconsin,where they replace savannas and grasslands. Thetwo models also disagree on the sign of the responseof coniferous forests.

Comparison between the two models. At coarse res-olution, MAPSS simulates a smaller extent of his-torical grasslands and shrublands and a larger ex-tent of shrublands and woodlands than MC1 ingreater agreement with Kuchler’s (1975) map ofpotential vegetation (Figure 5a). This large differ-ence between the two models is due in part to the

dynamic fire model in MC1. The grassland areasthat are classified as shrub and woodlands byMAPSS are areas of high fire-return intervals. MC1biogeography rules may underestimate the contri-bution of the shrubs to the landscape using tempo-ral averages of total leaf biomass because of therecurring loss of shrub leaf biomass in fires. On theother hand, MAPSS does not include a dynamic firemodule; thus, it may miss the secondary effect ofdroughts, fueling fire events, and it may overesti-mate vegetation density. This difference betweenthe two models is carried into simulations of thefuture. In these simulations, MAPSS always simu-lates more shrubs and woodlands than MC1, andMC1 simulates more grasslands than MAPSS (Fig-ure 6).

MAPSS and MC1 simulate similar changes in fu-ture vegetation distribution. However, they dis-agree about the extent of savanna area primarilybecause of the simulated response of the southeastmixed forests. With MAPSS, savannas replacemuch of the southeast forests; with MC1, theSoutheast remains dominated by forests eventhough their density is lower than that of the orig-inal southeast mixed forests (Figure 6). The twomodels also disagree on the sign of the response ofconiferous forests. Coniferous forests include bothnortheastern boreal coniferous forests and north-western temperate coniferous forests. The borealforests in the Northeast shift north and decrease in

Figure 3. MAPSS-simulatedvegetation distribution forthe conterminous US(10-km resolution) underthe seven GCM climate sce-narious listed in Figure 1.


area for both models and under both scenarios.Thus, the disagreement between the models is re-ally about the fate of western coniferous forests.Western coniferous forests increase in area underboth scenarios with MAPSS at the 10-km resolutionand with MC1 at the 0.5° resolution, but they showlittle change under either scenario for MAPSS at thecoarser resolution. Because the higher-resolutionclimate data set captures more accurately the com-plex topography of the western states, MAPSS re-sults at the 10-km resolution should be more reli-able.

It is more difficult to explain why MC1 simula-tions with coarse resolution climate do not agreewith MAPSS results at the same resolution. It maybe due to scaling of climate or different sensitivitiesof the models to subtle differences in climate.MAPSS is very sensitive to variations in vapor pres-

sure deficit (VPD), a variable not utilized by MC1.VPD was calculated and scaled differently betweenthe two different resolutions of climate data (Neil-son 1995; VEMAP 1995) and may be causing thedifferent responses. The large sensitivity of MAPSSto VPD arises from the aerodynamic evapotranspi-ration algorithm used in the model (Marks andothers 1998). It has been shown that different al-gorithms produce different sensitivities to climatechange (Mckenney and Rosenberg 1993). How-ever, a full analysis of the MAPSS VPD sensitivitywould require a complete replacement of theevapotranspiration algorithm and a recalibration ofthe model; it is therefore beyond the scope of thispaper.

The different responses of the Northeast mixedforests between MAPSS and MC1 result from un-certainties regarding the relative importance of wa-ter and/or temperature in controlling the competi-tive relationship between temperate deciduous andNortheast mixed forests. Under both scenarios,there are subregions in the northeastern region ofthe US that become wetter and other subregionsthat become drier. Under both scenarios, tempera-tures get warmer. Thus, the limiting factors of tem-perature and drought play complex roles in bothmodels in this region and different nuances in im-plementing these two factors produce different re-sponses.

Comparison of MAPSS results across scales. Dis-agreement in the sign of change between fine andcoarse resolution MAPSS results (Figure 5c, d) oc-curs for shrubs and woodlands and for westernconiferous forests. Because these vegetation typesare typically found in western states, we believe thehigher resolution climate, which more accuratelysimulates small climatic variations due to the com-plex topography, enables the model to better sim-ulate the vegetation response to climate changes.We have thus relied primarily on the 10 km resultsfor past assessment work. In the case of savannasunder the HADCM2SUL scenario, the magnitude ofthe simulated change is small under both scenarios,and the disagreement in the sign of change betweenthe two resolutions is probably due to a simpleaccounting of the pixels involved, —a resolutionerror rather than a disagreement in the prediction.

There are differences in the magnitude of changebetween fine- and coarse-resolution MAPSS resultsespecially for the northeast mixed forests, which arereplaced by temperate deciduous forests at thecoarser resolution under HADCM2SUL (Figure 6d).As with the conifer response in the West, the dif-ference between resolutions may be a result of dif-ferences in the VPD calculations.

Figure 4. Percentage changes in (a) forest area and (b)forest LAI for the conterminous US between current andfuture conditions for the seven GCM climate scenarioslisted in Figure 1.


LAI and Vegetation BiomassMAPSS-simulated LAI under seven CGM scenarios (10

km resolution). MAPSS simulates large increases inLAI both in the Southwest (California, Arizona, andNevada) and in the East (Figure 7) under bothHadley Climate Center scenarios, which simulatethe greatest increases in precipitation (more than22%) and also produce only modest warming. Un-der CGCM1, MAPSS also simulates large increasesin LAI in the West but large decreases in LAI in theEast. Under the OSU, GFDL, GISS, and UKMO sce-narios, MAPSS simulates large decreases in LAI inthe Great Lakes area, New England, and along thesouthern coast (Figure 7).

MAPSS consistently predicts increases in LAI forsavannas, shrublands, and arid lands under all sce-narios (Figure 8) with LAI increasing by nearly500% in desert areas under CGCM1. The LAI ofboth southeast mixed forests and temperate decid-uous forests increase by up to 70% under the wet-test climate scenarios with modest warming (suchas HADCM2SUL), but they decrease under thewarmer climate scenarios. Large decreases in LAIare simulated for the Northeast mixed forest exceptunder the more modest warming scenario,HADCM2SUL. Increases in LAI are simulated forthe taiga–tundra area (up to 1241%) except underthe warmest scenario (UKMO), where the decrease

in LAI reaches almost 50%, and under the GFDL-R30 scenario (Figure 8). The overall trend for all theforest types combined is similar to that observedwhen looking at the change in area, with forest LAIincreasing under wetter scenarios with modestwarming and declining under the warmer scenarios(Figure 4).

Change in LAI (MAPSS) and live vegetation carbon(MC1) under two transient scenarios (0.5° latitude/lon-gitude resolution). MAPSS and MC1 simulate lowervegetation density for coniferous forests under bothtransient climate change scenarios (Figure 9a). Thisdecrease pertains only to those areas currently popu-lated by conifers—that is, the Northeast and theNorthwest. Drought-induced vegetation density re-ductions occur in parts of both regions, even thoughincreased precipitation allows conifers to expand intonorthern California and the Great Basin. Both tran-sient climate change scenarios produce precipitationincreases in the southern half but some decreases inthe northern half of the western US, thus causingsome vegetation declines in those areas.

MC1 simulates an increase in vegetation densityin the northeast mixed forest under HADCM2SUL,whereas MAPSS simulates no change. However,under CGCM1, both models simulate a decrease invegetation density (Figure 9). Both models predictan increase in the LAI of temperate deciduous for-

Figure 5. (a) Percentageland cover of simplified po-tential vegetation types sim-ulated by MC1 (1990) andMAPSS (baseline historicalclimate) over the contermi-nous US compared toKuchler’s (1975) potentialvegetation map. (b) Percent-age change in area of thesimplified vegetation types assimulated by MC1 under twotransient scenarios for theyear 2095. (c) Percentagechange in area of the simpli-fied vegetation types as sim-ulated by MAPSS at 10-kmresolution under an averageclimate (2070–99) from thesame two scenarios (sincethere were no tropical broad-leaf forests under historicalconditions, percentagechange has been arbitrarilyset to 150%). (d) Same as c,but for 0.5° latitude/longi-tude resolution.


Figure 6. Potential vegeta-tion distribution simulatedby MAPSS and MC1 for cur-rent conditions (baseline his-torical climate for MAPSS,1990 for MC1) and for fu-ture conditions (2070–99 forMAPSS and 2095 for MC1)under two scenarios:HADCM2SUL and CGCM1at 0.5° latitude/longituderesolution. The color legendis identical to that of Figure3.

Figure 7. MAPSS-simulat-ed LAI at 10-km resolutionfor current climate condi-tions and percentagechange in LAI under theseven GCM climate scenar-ios listed in Figure 1.


ests under both scenarios. For southeast mixed for-ests, MC1 and MAPSS simulate an increase in veg-etation density under HADCM2SUL and a decreaseunder CGCM1.

MC1 simulates a decrease in vegetation C in trop-ical broadleaf forests under both scenarios. MAPSSdoes not include tropical forests under historicalclimate conditions. Both models simulate increasesin the vegetation density of savannas, shrublands,grasslands, and arid lands.

Comparison of MAPSS results across scales. Simula-tions by MAPSS of coniferous forests LAI (Figure 8)showed an increase under most climate change sce-narios at 10 km resolution except under the moreextreme UKMO scenario, whereas they showed adecrease at 0.5° latitude/longitude resolution (Fig-ure 9). Coniferous forests correspond in our simpleclassification to a combination of northeastern bo-real and western conifer forests. We showed earlierthat they respond differently to climate change andthat, because of the complex topography of thewestern US, western conifer forest simulationswere sensitive to scaling. We have more confidencein the validity of high-resolution results because ofthe greater accuracy of the climate simulation atthat scale.

MAPSS simulates an increase in the LAI of tem-perate deciduous forests under both transient sce-narios at coarse resolution in contrast to the 10 km

resolution results under CGCM1 (Figures 8 and 9).The differences between the results from MAPSS atthe two resolutions may be due to the VPD differ-ences in the baseline climates, but in any case theyhighlight the sensitivity of western vegetation tosubtle variations in climate.

Temporal dynamics of net biological production (NBP)and carbon pools (MC1 simulation). Total C storageremains stable (135 Pg C) in the early part of the20th century until about 1940, when it begins toincrease (Figure 10a). The upward trend beginningaround 1940 corresponds to an increase in precip-itation over North America and the beginning ofthree decades of Northern Hemisphere cooling(Karl 1998). This trend is also reflected in the netbiological production (NBP) trace (Figure 10d).NBP, calculated as the net change in C storage fromone year to the next and equivalent to net primaryproduction NPP minus heterotrophic respirationand fire emissions, averages –0.6 Pg C y21 from1895 to 1940, when a warmer climate with fre-quent droughts favors C fluxes from decomposi-tion, respiration, and fire; by contrast, from 1940 to1971, cool climate favors C sequestration and NBPaverages 10.12 Pg C y21.

Impacts of large fire events and droughts (partic-ularly in the 1910s and 1930s) in the early part ofthe 20th century are clearly visible in the evolutionof NBP. The same impacts are also probably respon-

Figure 8. MAPSS-simulatedpercentage change in LAI atthe 10-km resolution foreach simplified vegetationtype over the conterminousUS between current and fu-ture conditions under theseven GCM scenarios listedin Figure 1.


sible for the small decline in soil C during the sameperiod (Figure 10c). Nicholls and others (1996) re-port that the Northern Hemisphere shifted from 3decades of cooling to a persistent warming trend in1972 or 1976, depending on the region, and wasaccompanied by a large increase in precipitationover North America (Karl 1998). Again, this regimeswitch is captured in the simulation of NBP, whichdecreases slightly to an average of 10.09 Pg C y21

between 1971 and 1993, when we can assumeplant and soil respiration are again enhanced bywarmer temperatures.

To further verify MC1 simulations, we comparedNBP results with recently published values. MC1simulated NBP averages 10.06 Pg C y21 between1961 and 1990. Even though these NBP valuesinclude no history of land use, such as agriculturalconversion and forest harvest, they agree well withsimulations by other VEMAP models (0.08 Pg Cy21) that included agricultural ecosystems (Schimeland others 2000), but they are lower than the ob-served record for forests of about 0.3 Pg C y21

(Birdsey and Heath 1995; Houghton and others1999). Regrowth of eastern US forests followingharvest is thought to be responsible for the higherobserved NBP (Schimel and others 2000) and thediscrepancy between observations and simulationsthat do not include land-use history.

Figure 9. MAPSS-simulatedLAI (a) and percentagechange in LAI (b). MC1-sim-ulated live vegetation carbon(c) and percentage change invegetation carbon (d) for his-torical and future conditionsunder HADCM2SUL andCGCM1 at 0.5° latitude/lon-gitude resolution for 11 sim-plified vegetation types.MAPSS results are for theaverage climate of 2070–99.MC1 results are averagedover the decade of the2090s.

Figure 10. Temporal dynamics (simulated by MC1) of(a) total vegetation and soil C (b) live vegetation C (c)litter and soil C, and (d) net ecosystem productivity (NBP,defined as gain or loss in total C from one year to thenext) under HADCM2SUL and CGCM1.


The role of fire in the carbon budget (MC1 simulation).MC1 simulates an increase in the frequency of firesduring the 1930’s drought and a large fire event in1988, a major fire year in the western United States(Sampson 1997), when the model simulated 2 Pg Cy21 consumed (under potential natural condi-tions— that is, without agriculture) (Figure 11).MC1 simulates that fires consumed about 720 Tg Cy21 between 1895 and 1993. This is consistent withLeenhouts’s (1998) estimate of 530–1228 Tg C y21

during the same period for potential natural vege-tation in the conterminous United States.

Simulations of C pools by MC1 differ dramaticallybetween the two transient scenarios. MC1 simu-lates a continuous increase in total biomass, litter,and soil organic C under HADCM2SUL (Figure10a). However, under CGCM1, it simulates a 15%decline in total live biomass (Figure 10b) and a 5%decrease in litter and soil organic matter (Figure10c) over the course of the 21st century. Biomassconsumed by fire increases in the future underHADCM2SUL (Figure 11). Around 2044, fires con-sume about 1 Pg C, causing a sharp decline inbiomass (Figure 10a, b). No single fire event seemscorrelated with the sharp decline in live vegetationfrom 2057 to 2067, although the large negativeNBP during that period indicates a strong droughtresponse (Figures 10c, d). Under CGCM1, total livebiomass decreases while biomass consumed by fireincreases, especially around 2030, when fires con-sumed almost 2.5 Pg C, largely in eastern forests(Figures 11 and 12). Around 2085, also underCGCM1, large fires correspond to large declines intotal biomass (Figures 10a and 11). Fire increases inthe West under both scenarios (Figures 12e, f), butespecially under CGCM1, because fuel loads in-crease with increased precipitation coupled withseveral wet–dry cycles (El Nino/La Nina).

Stress AreaLinear relationship between stress area and tempera-

ture increase. Across the seven climate change sce-narios (10-km resolution), there is a significant re-lationship between the projected increase intemperature and the stress area simulated byMAPSS over the conterminous United States (Fig-ure 13). MAPSS simulates an 11% increase in thisarea per degree of temperature increase (Figure13a). Considering only the forest lands, the stressarea simulated by MAPSS increases at a rate of 17%of the total forest area per degree of increased tem-perature (Figure 13b). The stress area simulated byMC1 (0.5° latitude/longitude resolution) at the endof the 21st century compares well with that simu-lated by MAPSS (Figure 13) for both conterminousUS lands and forested areas under both scenarios.Under HADCM2SUL, at the low end of the temper-ature gradient represented by the seven scenarios,MC1 simulates an area of vegetation stress of about20% of the conterminous US, whereas MAPSS sim-ulates 10%. Under CGCM1, MC1 simulates almost60% and MAPSS almost 40% of the total areaundergoing some stress (Figure 13a). However,both models show greater sensitivity of forests un-der the warmer scenario with MC1 simulatingnearly 80% and MAPSS about 55% of current for-est area losing carbon under drought stress by theend of the 21st century (Figure 13b). The MC1results under the two scenarios lie within the rangeof variation of the MAPSS results under all sevenscenarios. Both models also show that forest areasare more sensitive than non-forest areas to poten-tial future temperature increases. The overall re-sponse of both models across a range of scenariossuggests that an average annual temperature in-crease of 4.5°C could produce a reduction in vege-tation density over about 50% of conterminous USforest lands, while the remaining lands would ex-perience increased growth. This temperature in-crease corresponds to about the middle of the pro-jected temperature change over the conterminousUS by the end of the 21st century.

Time Series of Historical Stress Area Compared to theDrought-Area Index (MC1 Simulation)The stress area simulated by MC1 varies consider-ably from year to year. Large droughts clearly standout. The drought of the 1930s is simulated as themost severe of the century, affecting about 49% ofthe simulated forest area and about 60% of the USland surface between 1933 and 1940 (Figure 14).These results agree with a time series of a drought-area index (DAI) based on the PDSI (Palmer 1965).The DAI is the area of land with a PDSI value

Figure 11. Simulated (MC1) total annual C consumed bywildfire for the conterminous US under HADCM2SULand CGCM1. These results are potential dynamic vegeta-tion only and do not include land-management activities,such as fire suppression, forest harvest, or conversion toagriculture.


showing moderate to extreme drought (Diaz 1983).The DAI indicates that the three major droughtepisodes during the 1930s affected between 50%and 80% of the conterminous US (Diaz 1983). Thedrought of the 1950s (1951–57), as simulated byMC1, affects about 49% of the forest area and about53% of the entire conterminous United States.Again, this result is confirmed by the DAI, whichindicates that an area of 50% to more than 60% ofthe US was involved in that drought (Diaz 1983).Finally, as simulated by MC1, between 1988 and1992, about 39% of the US was losing C underdrought stress. This result agrees with Changnon’s(1989) estimate (based on the PDSI) that about40% of the US was under severe to extremedrought in 1988 alone. MC1 simulated that thebrunt of the 1988 drought was in nonforested zoneswith only about 25% of the simulated forest areabeing affected by it.

Evidence of the 1972 climatic shift. The stress areasimulated by MC1 averages 49% of the contermi-nous US during the historical period until about1972, when it drops to 34% and remains at thatlevel until 1993. The decline in the stress area in1972 is particularly large across forest lands (Figure14b), dropping from a pre-1972 average of 43% toa post-1972 average of about 27%. This result com-pares well with climatic observations. The rapiddecline in stress area in 1972 can be linked to a

dramatic increase in the precipitation regime overNorth America and the shift from 3 decades ofcooling to persistent warming (Karl 1998).

Projection of the future extent of the stress area. Thestress area simulated by MC1 in the future is quitedifferent under the two scenarios (Figure 14). Un-der HADCM2SUL, there is an initial increase fol-lowed by a continuous decline in the stress area,which implies that increased precipitation associ-ated with moderate warming and coupled with CO2

effects is favoring vegetation growth (Figure 14a,b). Under CGCM1, the stress area increases in theearly decades of the 21st century, returning to thelevel of the 1930s drought, and remains at that leveluntil the end of the century (Figure 14a). Theamount of forested stress area under CGCM1 showsthe same pattern as that of the overall stress area,but it is far greater than during any of the droughtepisodes of the past century (Figure 14b), reachingnearly twice the average level of the past century(from a pre-1972 value of 43% to about 80% ofcurrent forest area).

SYNTHESIS

Changes in Vegetation DistributionConsistent patterns across all climate change scenarios.

Our first objective was to see if there were consis-tent patterns of vegetation change across a large

Figure 12. (a), (b) Simu-lated percentage change inLAI (MAPSS model at 0.5°latitude/longitude resolu-tion) vs (c, d) change in livevegetation carbon (MC1)under HADCM2SUL andCGCM1 scenarios. MAPSSresults are the change be-tween the 2070–99 periodand the baseline historicalperiod. MC1 results com-pare average vegetation car-bon of the last decade of thecentury to that of the1961–90 period. (e, f) Abso-lute change in average an-nual biomass consumed assimulated by MC1, compar-ing the long-term averagefor the entire historical pe-riod (1895–1993) to that ofthe entire future climatechange period (1994–2099)under HADCM2SUL andCGCM1.


gradient of temperatures simulated by GCM usingthe equilibrium vegetation model, MAPSS. Manyresults are indeed consistent across all scenarios.For example, under most scenarios, MAPSS simu-lates increases in vegetation density in the south-western states, where large increases in precipita-tion contribute to the reduction of arid land areas(Figure 3). MAPSS also simulates some vegetationdecline in the Great Lakes region, particularly in thearea of northeast mixed forests, subject to warmerand drier climatic conditions in the 21st century(Figure 3). Under most scenarios, MAPSS simulatesdecreases in the area of tundra and taiga–tundrathat disappear as their cooler temperature optima

disappear (Figure 2). On the other hand, MAPSSsimulates increases in the area of southeast mixedforests and savannas under most scenarios (Figure2).

There are also trends across the scenarios thatseem related to the magnitude of the potential in-crease in future temperatures. MAPSS simulates anoverall increase in vegetation density with moder-ate warming and a decrease with greater warming.Biomes that appear most sensitive to elevated tem-peratures include the temperate deciduous forest,which decreases in area under the warmer scenar-ios but increases under the more moderate scenar-ios, and the southeast mixed forest, which shiftsnorthward under the warmer scenarios and is re-placed by savannas and grasslands in its currentlocation (Figure 2). Conifer forests show some in-dication of increased sensitivity with increasingtemperature, but their overall response is less clearbecause they increase slightly in area under mostbut not all scenarios (Figure 2).

“Early green-up, later browning” hypothesis.MAPSS results across the seven climate change sce-narios imply a monotonic change in climate: nei-ther interannual nor interdecadal variability affectthese equilibrium simulations. Were the climate tochange in such a smooth manner and temperaturesto increase to the level projected by the warmestscenario, simulation results suggest the possibility

Figure 13. Fractional area of LAI decline (stress area) asa function of the average change in temperature over theconterminous US from the seven GCM scenarios listed inFigure 1, as simulated by MAPSS (F) for (a) the conter-minous US land area and (b) the forested lands onlywithin the conterminous US at 10-km resolution, includ-ing the least-squares regression line. For comparison, thestress areas simulated by MC1 (Œ) are also shown for twotransient scenarios (HADCM2SUL and CGCM1), repre-senting the regions of live biomass decline, comparing thelast decade of the 21st century to the historical period(1961–90) at 0.5° latitude/longitude resolution.

Figure 14. MC1-simulated stress area under historicalclimate and future climate scenarios for (a) the conter-minous US and (b) forested lands only. The stress area isthe area of the US or of the US forest lands where livevegetation C density is less than that of the long-termaverage carbon density calculated for the 100-yearspin-up period.


for an early green-up in response to a moderatewarming, followed later by vegetation density de-clines due to temperature-induced droughts (Neil-son and Drapek 1998). We had hoped to test thishypothesis with a dynamic global vegetation model,such as MC1, using a climate change scenario fromthe warmer end of the temperature gradient, suchas CGCM1. However, even though temperaturestend to increase somewhat monotonically in thesescenarios, precipitation exhibits considerable inter-decadal variability, which can override the simpli-fied trajectory implied by the hypothesis. Althoughprecipitation under CGCM1 increases significantlyby the end of the 21st century, it first decreases by4% by mid-century before increasing rapidly to theend of the century. Because temperatures increasein the first few decades of the 21st century as pre-cipitation slightly decreases, the hypothesized earlygreen-up does not materialize under CGCM1. In-stead, there is a rapid loss in vegetation density untilmid-century, after which increasing precipitationand water-use efficiency roughly balance the in-creasing evaporative demand (Figures 10 and 14). Itis not clear what would have prevailed had CGCM1continued with increased warming beyond the 21stcentury. A more thorough test of the early-green-ing, later-browning hypothesis would require theuse of several transient climate change scenarios.

Moderate warming benefits vegetation but additionalwarming could cause droughts. In addition to testingthe early-greening, later-browning hypothesis, wewere interested in comparing the results from thetwo types of models and in integrating these resultsin an overall synthesis that would enhance ourconfidence in the conclusions. Although there aredifferences between MAPSS and MC1 results, thesimilarities are striking. In general, both modelsagree that, under moderate warming, vegetationdensity increases due to the projected increase inprecipitation and CO2-induced increases in water-use efficiency, resulting in forest expansion and Csequestration across the United States. They alsoagree that greater warming can lead to lower veg-etation density and conversions of forests to savan-nas and grasslands with possible overall losses of Cto the atmosphere in regions where neither precip-itation increases nor direct CO2 effects can compen-sate for the exponential increases in evaporativedemand.

Regional projections illustrate this pattern. Bothmodels agree that warming produces a northwardshift of the various eastern forest types and analtitudinal shift of the colder taiga–tundra and tun-dra vegetation types, which may have disappearedby the end of the 21st century from the contermi-

nous United States (Figure 5). However, steepmountain slopes with unstable or poorly developedsoils may limit the upslope migration of forests. Thetwo models also agree on the contraction of theSouthwest arid land area (Figure 5) associated withlarge increases in LAI or biomass in several south-western states (California, Nevada, and Arizona),due to the predicted increases in precipitation (andCO2-induced water-use efficiency) accompanyingthe rise in temperature (Figure 12).

Both models simulate increases in the area ofsoutheast forests (Figure 5) under HADCM2SULand decreases in their vegetation density underCGCM1 (Figure 9). The C losses suggested byMAPSS are sufficient to convert much of the South-east forests to savannas and grasslands, whereaslosses simulated by MC1 are considerable but notsufficient to cause a conversion to savanna. Simu-lated fires by MC1 convert large sections of theSoutheast to savannas and grasslands by the middleof the 21st century. However, these savannas re-cover to forests by the end of the 21st century, eventhough their biomass remains approximately 30%lower than before the fires,—not far in characterfrom the savannas simulated by MAPSS. Areas inthe Southeast that were near the fire zones, but notburned, also experience drought-induced vegeta-tion density declines of about 30%.

The fate of the western coniferous forests underwarmer climates is less clear. MC1 simulates a largeexpansion of the coniferous forests across the west-ern states under CGCM1, even though it simulatesa decrease in their C density over the area of theircurrent distribution. MAPSS, on the other hand,simulates little change in their area but a decreaseof their LAI under the two transient scenarios atlow resolution (0.5° latitude/longitude). However,MAPSS simulates an increase in the western conif-erous forest area and in their LAI at higher resolu-tion (10-km) under most scenarios (except UKMO).Because the higher-resolution climate takes intoaccount complex terrain, it includes less scaling er-ror and thus should produce more reliable simula-tions. It is important to note that both models as-sume that there are no barriers to species migration,such as seed dispersal limitations or habitat loss dueto urban or agricultural expansion.

Under CGCM1, both models simulate decreasesin biomass or LAI in the Great Plains states (Col-orado, New Mexico, Texas, Oklahoma, Kansas).MC1 also simulates large decreases in biomass inArkansas, Missouri, Oklahoma, and Kansas andin the Carolinas, West Virginia, and Ohio with


conversions to savannas and grasslands underCGCM1 (Figure 12). In these eastern forest areas,MC1 simulates increases in fire frequency. Be-cause MAPSS does not include a dynamic firemodule, it may miss the secondary effect ofdroughts, fueling fire events, and it may overes-timate vegetation density.

Two regions where the two models disagree andwhere the scale of the input data affects simulationresults stand out for their apparent sensitivity toclimatic changes and should be the object of furtherresearch: (a) the Southeast, where the more severeclimate change scenarios induce large declines in Cstocks, and, (b) the northwestern forests, wherecomplex topography and uncertainties about pre-cipitation change and potential CO2-induced water-use efficiency render predictions more tentative.

Carbon Budget under Future ClimateConditions

A warming threshold separates the future greenworld from the brown world. Under HADCM2SUL,MC1 simulates a C gain of approximately about 15Pg in the US terrestrial biosphere from now to theend of the 21st century, whereas under CGCM1, itsimulates a loss of around 7 Pg C over the conter-minous US (Figure 10). Even though neither posi-tive nor negative feedbacks from the biospherehave been included in the climate change scenarios,feedbacks are implied in the simulation results be-cause the conterminous US can become either a Csource or a C sink depending on the scenario. Theseresults suggest that there is a warming thresholdtransition point. Below this temperature threshold(for example, HADCM2SUL), plants could thriveand their C uptake could slow global warming (neg-ative feedback). However, above this threshold(CGCM1), regional drought stress could occur andcause net C emissions (for instance, from droughtand fires) that could accelerate global warming,producing additional vegetation stress (positivefeedback). Past simulations suggest that these con-cepts apply globally, rather than simply to the US(Neilson and others 1998; Neilson and Marks1994).

Stress Area TrajectoriesHistorical records confirm model projections of location

and extent of stress areas. The stress area trajectories(Figure 14) are analogous to the widely used ap-proach of the Palmer drought severity index totrack national changes in overall drought stress andto map the area over which drought impacts are

important. MC1 tracks the drought area index(based on PDSI) quite well over the historical timeframe, capturing the well-known droughts of the1930s, 1950s, and late 1980s, among others. Thus,these results provide some measure of validation ofthe model. The stress area trajectories also hint of aclimate regime shift about 1972, which correspondsto a large shift in both the temperature and rainfallregimes over North America at that time (Karl1998). PDSI maps are often used to indicate large-scale atmospheric circulation regimes, their rela-tionship to interdecadal oceanic circulation re-gimes, and their shifts to alternative states (Nigamand others 1999). The regime shift identified byNigam and others (1999) occurred in 1976. Bothyears, 1972 and 1976, appear to be important at-mospheric change points. A climate regime shiftalso occured in 1940 (Neilson 1986), correspondingto a change from Northern Hemisphere warming to3 decades of cooling and also to an increase inprecipitation over North America (Nicholls and oth-ers 1996). All of these atmospheric circulation re-gime shifts are apparent in the evolution of NBP(Figure 10), shifting from a long-term negative an-nual average of –0.06 Pg C y21 to a positive averageof 10.1 Pg C y21 after 1940. The climate shift in1972 also produced a minor reduction in NBP from10.12 (1940–71) to 10.09 Pg C y21 (1972–1993).

Under the canadian future climate change scenario, thelargest stress area is in the southeast. Future stressarea trajectories are of interest when looking at thepotential consequences of mild vs more extremewarming. The position of storm tracks or jet streampatterns influences the location of regions of in-creased moisture and of drought-stressed areas. Un-der HADCM2SUL, MC1 simulates a continuous re-duction in the drought-stressed area; underCGCM1, it simulates an overall increase in that area(Figure 14). The stress area under CGCM1 increasesrapidly over the 3–4 decades, then stabilizes for therest of the 21st century, suggesting a large regimeshift in the atmospheric circulation. As tempera-tures increase during these first decades, GCGM1precipitation decreases slightly during the same pe-riod. However, both temperature and precipitationincrease during the latter period, but the area overwhich stress impacts occur appears to stabilize bymid-century. Most of the drought region is concen-trated in the eastern to southeastern US, with apredominant impact on forests. These temporal andregional changes suggest a significant reorganiza-tion of the upper atmospheric circulation patterns.Which of these two scenarios, if either, is more


likely to occur cannot be stated at this time. Moretransient scenarios must be analyzed to see if thereare any overall tendencies, or if there are uniquelydifferent atmospheric circulation regimes that tendto occur under the different scenarios.

Validation Efforts

Cramer and others (1999) and Scurlock and others(1999) have emphasized the need for improvedvalidation methods to more accurately evaluate thesensitivity and validity of model responses to cli-matic signals. Cramer and others (2000) comparedsix DGVM under HADCM2SUL as they simulatevegetation changes and C fluxes for the globe. LikeMC1, two of these DGVM are derived from equilib-rium biogeography models—LPJ from BIOME3(Sitch and others 2000), and SDGVM (Woodwardand others 1998) from DOLY—and include bothbiogeography rules and mechanistic physiologicaland biogeochemical processes, but they also includea detailed calculation of photosynthesis. Three oth-ers—IBIS (Foley and others 1996), TRIFFID (P. Coxpersonal communication) and VECODE (Brovkinand others 1997)—also include feedback fluxes tothe atmosphere because they were originally de-signed to be included in coupled atmosphere–bio-sphere models. MC1 does not include thesebiofeedbacks, which can greatly modify atmo-spheric responses to changes in albedo and evapo-ration fluxes, for example. Cramer and others(2000) compared their vegetation distribution re-sults to a satellite-derived map of the world andtheir simulated fluxes with published estimates (forexample, see Ciais and others 1995; Fung and oth-ers 1997). Because of the coarse spatial resolutionof their results, the vegetation distribution maps aredifficult to compare with those presented here.However, in terms of C fluxes, when MC1 was runwith the same climatic data, it showed a muchlower sensitivity to CO2, probably due to its lack ofa detailed representation of photosynthesis (includ-ing a direct relationship to atmospheric CO2 con-centration), and projected a future global C sourcerather than a C sink for the 21st century. Future netecosystem production (NEP) simulated by MC1 didnot show the CO2 fertilization effect compensatingfor the impacts of global warming. One of theDGVM—LPJ (Sitch and others 2000)—used in Cra-mer and others (2000), has also been run with the0.5° latitude/longitude resolution climate data forVEMAP and has displayed a much stronger re-sponse to enhanced CO2 than MC1, confirming theimportance of adequate representation of CO2 ef-fects on plant processes, the lack of data about CO2

impacts at the regional scale during the 20th cen-tury, and the resulting inability of modelers to re-fute either projection. Clearly, more work needs tobe done to create databases that can be used bymodelers to verify their projections.

The simulation by MC1 of biomass consumed byfire over the historical period bears some relation-ship to the stress area calculations (Figures 11 and14). Fires tend to occur in the heart of the droughtzones, given sufficient fuel (Figure 12). However,the stress-area and biomass-consumed curves arenot exact overlays, because most biomass con-sumed comes from forests and the historicaldroughts affected forested and nonforested regionsdifferently. Even so, the fact that MC1 accuratelycaptured known high-fire years, such as 1910 and1988, provides another test of the validity of themodel. Several other validation efforts are underway, comparing model results with extensive soiland vegetation databases available for the states ofOregon and Washington (M. Harmon, B. Law, S.Remillard personal communication), as well asother data sets gathered from the literature. Be-cause the models do not simulate historical or fu-ture land use, it is difficult to directly compare theresults of the simulations with field data. Land-useimpacts, forest harvest patterns, and disturbances(for example, fire, pests, disease) need to be takeninto account during the validation exercises. Effortsare currently under way to begin including land-use changes in simulations of historical vegetationdynamics and these changes also need to be con-sidered in future scenarios. In addition, models canalways be improved through reexamination andimprovement of existing model processes.

Conclusions and Options for the Future

We have attempted to bring some clarity to theconfusion surrounding the multitude of possiblefuture climates and the associated ecological re-sponses. Rather than focusing on a single scenario,which may seem less confusing but is inherentlydeceptive, we have chosen to examine as manyscenarios as possible to see if there were any con-sistent patterns. The use of the MAPSS equilibriumbiogeography model allowed us to examine sevenscenarios, but only as “snapshot” comparisons ofcurrent vs future conditions. That is, it provides noindication of how the biosphere might evolve dy-namically between the current conditions and theend of the 21st century.

Consistent patterns have emerged from our com-parison of the seven scenarios. In some instances,


all seven scenarios produced the same sign ofchange—for example, spatial shifts of cold-limitedecosystems. In other instances, certain trends fol-lowed the increase in temperature across all sevenscenarios, where regional differences in precipita-tion produced the “noise” around the regressionline; for example, forest area might increase undermild warming but decrease under greater warming.Similarly, the area of the US subjected to droughtstress appeared to increase linearly with respect tothe projected temperature change. A 4.5°C rise intemperature could cause drought stress in about50% of US forest area (while the other 50% showsincreased growth), suggesting that a temperatureincrease near 4.5°C could be a threshold belowwhich US ecosystems would sequester C but abovewhich they would lose C.

The MC1 results provided some sense of howthe terrestrial biosphere could change along twotrajectories chosen among the seven scenarios—one near the mild end of the temperature changegradient and one near the warm end. The overallresults from MC1 were quite consistent withthose from MAPSS, even though there were somedifferences in the details. As hypothesized fromthe MAPSS results, the moderately warm Hadleyscenario (HADCM2SUL) produced increased veg-etation growth and reduced drought stressthroughout the 21st century. Also as anticipated,the warmer Canadian scenario, which exceedsthe 4.5°C threshold, produced large areas ofdrought stress, resulting in net C losses by the endof the century. However, the Canadian scenariodeviated from the “linear” logic of the MAPSS-based hypothesis of early green-up followed bylater browning. The hypothesis presumed thatprecipitation increased linearly with temperature,which did not happen in the Canadian scenariobecause drought stress began almost immedi-ately. These results should not be taken too liter-ally, since a different Canadian simulation withdifferent initial conditions might produce a dif-ferent trajectory. Nevertheless, these results un-derscore the importance of interannual and in-terdecadal climate variability, the potentiallylarge impact of climate variations on ecosystems,and the need for further use and development ofdynamic vegetation models using various ensem-bles of climate change scenarios.

Finally, both transient scenarios included largechanges in regional weather patterns. Each sce-nario, even the milder HADCM2SUL scenario,produced regional impacts of drought and fire

that could cause significant distress to regionalecological and economic systems, while warmerand wetter climates could benefit other regions.

Given the uncertainty surrounding future sce-narios, managers would be well advised to de-velop contingency plans for alternative futures,increased vegetation growth, or increased vege-tation stress, with specific regional patterns andtiming to both. Monitoring could be configured toidentify these alternative conditions as they oc-cur. One of the greatest uncertainties in theseresults is the importance of the CO2-induced wa-ter-use efficiency, which is incorporated in allresults presented here. If the effect is less thanthat simulated, then the early greening would beless marked than presented here and may notoccur in all ecosystems.

ACKNOWLEDGMENTS

We thank Tim Kittel and the VEMAP data group atNCAR (Boulder, Colorado) for providing us withthe climate scenarios, and David Yates from theResearch Application Program, NCAR (Boulder,Colorado), for providing us with PDSI data. We alsothank Steve Wondzell, Andy Hansen, Virginia Dale,and two anonymous reviewers for helpful com-ments on the manuscript. This work was funded inpart by the US Department of Energy, NationalInstitute for Global Environmental Change, GreatPlains Region (LWT 62-123-06509); the US Geolog-ical Survey, Biological Resources Division, GlobalChange Program (CA-1268-1-9014-10); and theUSDA Forest Service, PNW, NE, SE Stations (PNW95-0730).

REFERENCES

Birdsey RA, Heath LS. 1995. Carbon changes in U.S. forests. In:Joyce A, editor. Productivity of America’s forest and climatechange. General technical report RM-271 Fort Collins, (CO):USDA Forest Service, Rocky Mountain Forest and Range Ex-periment Station, p 56–70.

Brovkin V, Ganopolski A, Svirezhev Y. 1997. A continuousclimate–vegetation classification for use in climate–biospherestudies. Ecol Model 101:251–61.

Changnon SA. 1989. The 1988 drought, barges, and diversion.Bull Am Meteorol Soc 70:1092–104.

Ciais P, Tans PP, Trolier M, White JWC, Francey RJ. 1995. A largenorthern hemisphere terrestrial CO2 sink indicated by the 13C/12C ratio of atmospheric CO2. Science 269:1098–102.

Cramer W, Bondeau A, Woodward FI, Prentice IC, Betts R,Brovkin V, Cox PM, Fischer V, Foley JA, Friend AD, andothers. 2001. Global responses of terrestrial ecosystem struc-ture and function to CO2 and climate change: results from sixdynamic global vegetation models. Global Change Biol. Forth-coming.


Cramer W, Kicklighter DW, Bondeau A, Moore IIIB, ChurkinaG, Nemry B, Ruimy A, Schloss AL, and the participants ofthe Potsdam NPP model intercomparison. 1999. Comparingglobal models of terrestrial net primary productivity (NPP):overview and key results. Global Change Biol 5 Suppl 1:1–15.

Cubasch U, Cess RD. 1990. Processes and modeling. In: Hough-ton HT, Jenkins GJ, Ephraums JJ editors. Climate change: theIPCC scientific assessment. Cambridge (UK): Cambridge Uni-versity Press, p 69–91.

Daly C, Bachelet D, Lenihan JM, Neilson RP, Parton W, Ojima D.2000. Dynamic simulation of tree–grass interactions for globalchange studies. Ecol Appl 10(2):449–69.

Diaz HF. 1983. Some aspects of major dry and wet periods in thecontiguous United States, 1895–1981. J Climate Appl Meteo-rol 22:3–16.

Eamus D. 1991. The interaction of rising CO2 and tempera-tures with water use efficiency. Plant Cell Environ 14:843–52.

Fan S, Gloor M, Mahlman J, Pacala S, Sarmiento JL, TakahashiT, Tans P. 1998. Atmospheric and oceanic CO2 data and modelsimply a large terrestrial carbon sink in North America. Sci-ence 282:442–46.

Foley JA, Prentice IC, Ramankutty N, Levis S, Pollard D, SitchS, Haxeltine A. 1996. An integrated biosphere model of landsurface processes, terrestrial carbon balance, and vegetationdynamics. Global Biogeochem Cycles 10:603–28.

Fung I, Field CB, Berry JA, Thompson MV, Randerson JT,Malmstrom CM, Vitousek PM, Collatz GJ, Sellers PJ, Ran-dall DA, and others. 1997. Carbon 13 exchanges betweenthe atmosphere and biosphere. Global Biogeochem Cycles11:507–33.

Gates WL, Henderson-Sellers A, Boer GJ, Folland CK, KitohA, McAvaney BJ, Semazzi F, Smith N, Weaver AJ, ZengQ-C. 1996. Climate models— evaluation. In: Houghton JT,Meira Filho LG, Callander BA, Harris N, Kattenberg A,Maskell K, editors. Climate change 1995: the science ofclimate change. Contribution of Working Group 1 to theSecond Assessment Report of the Intergovernmental Panelof Climate Change. Cambridge (UK): Cambridge UniversityPress. p 235– 84.

Houghton RA, Hackler JL, Lawrence KT. 1999. The U.S. carbonbudget: contributions from land-use change. Science285:574–78.

Karl TR. 1998. Regional trends and variations of temperatureand precipitation. In: Watson RT, Zinyowera MC, Moss RH,Dokken DJ, editors. The regional impacts of climate change:an assessment of vulnerability. Cambridge (UK): CambridgeUniversity Press. p 412–25.

Kattenberg A, Giorgi F, Grassl H, Meehl GA, Mitchell JFB,Stouffer RJ, Tokioka T, Weaver AJ, Wigley TML. 1996. Cli-mate models—projections of future climate. In: Houghton JT,Meira Filho LG, Callander BA, Harris N, Kattenberg A, MaskellK, editors. Climate change 1995: the science of climatechange. Contribution of Working Group 1 to the Second As-sessment Report of the Intergovernmental Panel on ClimateChange. Cambridge (UK): Cambridge University Press. p 285–357.

Kittel TGF, Rosenbloom NA, Painter TH, Schimel DS, VEMAPModeling Participants. 1995. The VEMAP integrated databasefor modeling United States ecosystem/vegetation sensitivity toclimate change. J Biogeog 22:857–62.

Kuchler AW. 1975. Potential natural vegetation of the UnitedStates. 2nd ed. Map 1:3,168,000. New York: American Geo-graphic Society.

Leenhouts B. 1998. Assessment of biomass burning in the con-terminous United States. Conserv Ecol 2(1):1. [online]

Lenihan JM, Daly C, Bachelet D, Neilson, RP. 1998. Simulatingbroad-scale fire severity in a dynamic global vegetation model.Northwest Sci 72:91–103.

Linacre ET. 1977. A simple formula for estimating evaporationrates in various climates, using temperature data alone. AgricMeteorol 18:409–24.

Marks D, Kimball J, Tingey D, Link T. 1998. The sensitivity ofsnowmelt processes to climate conditions and forest coverduring rain-on-snow: a case study of the 1996 Pacific North-west flood. Hydrol Proc 12:1569–87.

Mckenney MS, Rosenberg NJ. 1993. Sensitivity of some poten-tial evapotranspiration estimation methods to climate change.Agric For Meteorol 64:81–110.

Neilson RP. 1986. High-resolution climatic analysis and south-west biogeography. Science 232:27–34.

Neilson RP. 1995. A model for predicting continental scale veg-etation distribution and water balance. Ecol Appl 5(2):362–85.

Neilson RP, Drapek RJ. 1998. Potentially complex biosphereresponses to transient global warming. Global Change Biol4:505–21.

Neilson RP, Marks D. 1994. A global perspective of regionalvegetation and hydrologic sensitivities from climatic change. JVeget Sci 5:715–30.

Neilson RP, Prentice IC, Smith B, Kittel TGF, Viner D. 1998.Simulated changes in vegetation distribution under globalwarming. In: Watson RT, Zinyowera MC, Moss RH, DokkenDJ, editors. The regional impacts of climate change: an assess-ment of vulnerability. Cambridge (UK): Cambridge UniversityPress. p 439–56.

Nicholls N, Gruza GV, Jouzel J, Karl TR, Ogallo LA, Parker DE.1996. In: Houghton JT, Meira Filho LG, Callander BA, HarrisN, Kattenberg A, Maskell K, editors. Climate change 1995: thescience of climate change. Contribution of Working Group 1 tothe Second Assessment Report of the Intergovernmental Panelof Climate Change. Cambridge (UK): Cambridge UniversityPress. p 133–192.

Nigam S, Barlow M, Berbery EH. 1999. Analysis links Pacificdecadal variability to drought and streamflow in the UnitedStates. EOS 80:621–5.

Palmer WC. 1965. Meteorological drought. Research paper no.45. Washington (DC): US Weather Bureau, 58 p.

Parton WJ, Schimel DS, Cole CV, Ojima D. 1987. Analysis offactors controlling soil organic levels of grasslands in the GreatPlains. Soil Sci Soc Am 51:1173–9.

Sampson NR. 1997. Forest management, wildfire and climatechange policy issues in the 11 western states. CR-820-797-01-0. p 1–44 (http://www.amfor.org)

Schimel DS. 1995. Terrestrial ecosystems and the global carboncycle. Global Change Biol 1:77–91.

Schimel DS, Melillo J, Tian H, McGuire AD, Kicklighter D, KittelTGF, Rosenbloom N, Running S, Thornton P, Ojima D, andothers. 2000. The contribution of increasing CO2 and climateto carbon storage by natural and agricultural ecosystems of theUS 1980–1993. Science 287:2004–6.

Scurlock JMO, Cramer W, Olson RJ, Parton WJ, Prince SD.1999. Terrestrial NPP: toward a consistent data set for globalmodel evaluation. Ecol Appl 9:913–9.


Sitch S. 2000. The role of vegetation dynamics in the control ofatmospheric CO2 content [dissertation]. Lund University(Sweden): Institute of Ecology.

[VEMAP] Members. Vegetation/Ecosystem Modeling and Anal-ysis Project. 1995. Vegetation/ecosystem modeling and anal-ysis project: comparing biogeography and biogeochemistry

models in a continental-scale study of terrestrial ecosystemresponses to climate change and CO2 doubling. Global Biogeo-chem Cycles 9:407–37

Woodward FI, Lomas MR, Betts RA. 1998. Vegetation–climatefeedbacks in a greenhouse world. Philo Trans R Soc London B353:29–38.


climate change effects on vegetation distribution and ... · pdf fileclimate change effects on...

Documents