Climate Dynamics

, Volume 25, Issue 7, pp 755–776

Coupled atmosphere-ocean-vegetation simulations for modern and mid-Holocene climates: role of extratropical vegetation cover feedbacks


    • Center for Climatic ResearchUniversity of Wisconsin, Madison
  • Robert Jacob
    • Argonne National Laboratory
  • John Kutzbach
    • Center for Climatic ResearchUniversity of Wisconsin, Madison

DOI: 10.1007/s00382-005-0054-z

Cite this article as:
Gallimore, R., Jacob, R. & Kutzbach, J. Clim Dyn (2005) 25: 755. doi:10.1007/s00382-005-0054-z


A full global atmosphere-ocean-land vegetation model is used to examine the coupled climate/vegetation changes in the extratropics between modern and mid-Holocene (6,000 year BP) times and to assess the feedback of vegetation cover changes on the climate response. The model produces a relatively realistic natural vegetation cover and a climate sensitivity comparable to that realized in previous studies. The simulated mid-Holocene climate led to an expansion of boreal forest cover into polar tundra areas (mainly due to increased summer/fall warmth) and an expansion of middle latitude grass cover (due to a combination of enhanced temperature seasonality with cold winters and interior drying of the continents). The simulated poleward expansion of boreal forest and middle latitude expansion of grass cover are consistent with previous modeling studies. The feedback effect of expanding boreal forest in polar latitudes induced a significant spring warming and reduced snow cover that partially countered the response produced by the orbitally induced changes in radiative forcing. The expansion of grass cover in middle latitudes worked to reinforce the orbital forcing by contributing a spring cooling, enhanced snow cover, and a delayed soil water input by snow melt. Locally, summer rains tended to increase (decrease) in areas with greatest tree cover increases (decreases); however, for the broad-scale polar and middle latitude domains the climate responses produced by the changes in vegetation are relatively much smaller in summer/fall than found in previous studies. This study highlights the need to develop a more comprehensive strategy for investigating vegetation feedbacks.

1 Introduction

The past two decades have seen major advances in our knowledge and ability to simulate past climate changes. The mid-Holocene period around 6,000 year BP, a period when civilization started to rapidly evolve into modern form, was a time with measurable differences in environmental conditions (climatic and terrestrially biotic) from today. The evidence has been gleaned from an ever increasing volume of paleo-environmental data that has been gathered (Prentice et al. 2000; Webb et al. 1997; Tarasov et al. 1998; Harrison et al. 2003; Bigelow et al. 2003; MacDonald et al. 2000) and from simulations of climate with ever improving models (Kutzbach and Otto-Bliesner 1982; Joussaume et al. 1999; Hewitt and Mitchell 1998; Kutzbach et al. 1998; Texier et al. 1997; Ganopolski et al. 1988; Crucifix et al. 2002). This dual effort to understand past times, such as for 6,000 year BP, has led to collaborative model/data comparison projects (COHMAP 1988; Kutzbach et al. 1998; Webb et al. 1997; Harrison et al. 1998; Prentice et al. 1997); the modeling community benefits from validation of models tested under a climate change scenario and paleo-environmental scientists benefit from the modelers by getting a potential global picture of the climate change that can aid them in the interpretation of their data.

The basic picture of mid-Holocene climate (derived principally from models) is one of the enhanced seasonality brought about by an orbitally driven increase in the seasonal cycle of solar radiation (more in northern summer, less in winter). Summer monsoons strengthen over North Africa to Asia and over the Americas partly in response to warmer continental interiors. Reduced sea ice cover (and thickness) in high latitudes extends the polar warming in summer to fall/winter. In addition to summer warming, many models simulate dryer conditions over the extratropical continents. Consistent with this picture from models, paleo-environmental evidence indicates that the northern margin of the boreal forest pushed poleward in many areas, temperate forest encroached on present day boreal forest, xerophytic vegetation expanded in middle latitudes at the expense of forest cover, and vegetation expanded into desert areas (North Africa) in response to enhanced monsoons.

Climate models have steadily undergone improvements in their parameterizations of the complex atmosphere/ocean/biosphere interactions that are important to realistic determination of present day climate and climate change (past or potentially future). Most recently, a considerable effort is being made to incorporate dynamic vegetation components of the terrestrial biosphere into climate models (Texier et al. 1997; Ganopolski et al. 1998; Crucifix et al. 2002; Levis and Bonan 2004; Wohlfahrt et al. 2004). These models allow at various levels of complexity an examination of crucial land/vegetation/climate feedbacks, especially as these may be critical to realistic assessment of future climate change. Concerns for future climate change under scenarios of plausible rises in human-induced greenhouse gases has put a pressing need on the scientific community to not only include many key interactive components such as vegetation, but also to assess the accuracy of the climate model simulations (IPCC Report: Prentice et al. 2001; CCSP/SGCR 2003). An examination of the mid-Holocene climate represents a good test for models as there is a reasonable amount of data available in terms of vegetation change for verifying model results. Furthermore, as the models evolve in complexity so too will it be necessary to understand the potential feedback interactions of the various ocean/atmosphere/biosphere processes. In the light of the potential importance of vegetation feedbacks on future climate change a number of studies have started to address the role of vegetation feedbacks on a known past climate (Texier et al. 1997; Ganopolski et al. 1998; Levis and Bonan 2004; Doherty et al. 2000; Braconnot et al. 1999; Wohlfahrt et al. 2004).

In this study we examine the extratropical mid-Holocene climate and vegetation structure (relative to modern) from experiments performed using a near state-of-the-art global atmosphere/ocean climate model coupled synchronously to the Lund-Potsdam-Jena (LPJ) dynamic vegetation model (Sitch et al. 2003; Haxeltine and Prentice 1996). The LPJ component part of the model determines competitively the year-to-year vegetation structure (e.g. height and seasonal leaf cover) and the fractional coverage of a grid cell occupied by a set of Plant Function Types or PFTs (e.g. temperate deciduous tree, C3 grass etc.). The experimental results are derived from sufficiently long runs (400 years) to allow for adequate successional development of model vegetation. The focus of this study is twofold: present in some detail the simulated extratropical vegetation/climate results for modern and mid-Holocene times and assess the nature of vegetation feedbacks on the extratropical climate response. The first objective is undertaken to document the performance of the coupled model, which we feel represents an advancement over the two classes of previously published atmosphere/ocean/vegetation models: those that fall into the category of atmosphere/ocean models of intermediate complexity coupled synchronously to a vegetation component (Ganopolski et al. 1998; Crucifix et al. 2002; Brovkin et al. 2002) and those that are near state-of-the-art climate models coupled asynchronously to a vegetation/biome model (Texier et al. 1997; Wohlfahrt et al. 2004). In our study, we adopt a synchronous vegetation/climate coupling more like that used in the first class of models, but with a climate modeling approach more like that used in the second class of models.

A number of previous studies provide a foundation for our second objective of examining extratropical vegetation feedbacks. These have typically been done using prescribed vegetation changes either based on observational evidence (Foley et al. 1994; Douville and Royer 1996; Gallimore and Kutzbach 1996) or from model results derived by the aforementioned two classes of coupled climate/vegetation models (Ganopolski et al. 1998; Texier et al. 1997; Wohlfahrt et al. 2004). For this study we assess the feedback by placing the long-term structure of simulated vegetation obtained from the synchronously coupled runs into parallel uncoupled experiments for modern and mid-Holocene times. An important element of the extratropical vegetation feedback that is examined involves the role of vegetation/snow albedo feedbacks on the climate response associated with grass expansion in middle latitudes (at the expense of forest) and boreal expansion into tundra at higher latitudes. The recent history of investigation of the vegetation/snow albedo impact on high latitude climate includes a number of studies (Bonan et al. 1992; Foley et al. 1994; Gallimore and Kutzbach 1996; Wohlfahrt et al. 2004, to name a few). It should be noted that for this study, extratropical vegetation feedbacks are defined to be those produced in the presence of full ocean/atmosphere interaction; it is not a goal of this study to further decompose this feedback definition into the vegetation feedback under prescribed ocean conditions and the synergy between vegetation and ocean feedbacks such as it was been done in some previous work (Ganopolski et al. 1998; Wohlfahrt et al. 2004).

We first describe the models and some important modifications that were made to the models for proper coupling (Sect. 2). The specific full coupled and fixed vegetation experiments and their spin-up procedures are described in Sect. 3. The results section (Sect. 4) is divided into a discussion of the extratropical climate and vegetation for 0 ka (present day pre-industrial) and 6 ka (mid-Holocene at 6,000 year BP), and the climate response to vegetation feedbacks. Conclusions and further discussion are given in Sect. 5.

2 Model and methods

2.1 FOAM

The basic coupled climate model is the Fast Ocean Atmosphere Model or FOAM (Wu et al. 2003; Jacob 1997). The atmospheric component of FOAM (PCCM3) comprises a parallelized version of the NCAR CCM2 (Drake et al. 1995) that has been upgraded with CCM3.6 physics (Kiehl et al. 1998) and has a horizontal resolution at R15 (48 longitudes, 40 latitudes) and 18 vertical levels. The ocean component (OM3) is conceptually like the GFDL MOM (Pacanowski 1996). OM3 utilizes a z-coordinate system with a horizontal resolution of 1.4 latitude×2.8 longitude, 24 vertical levels, and an explicit free surface. Freshwater input into the oceans from the land is determined via a river routing/discharge model (Coe 1998). A simple three-layer thermodynamic sea model, without ice dynamics, is incorporated into FOAM. FOAM is highly efficient for long century-scale integrations on parallel computing platforms (Jacob 1997; Wu et al. 2003; Harrison et al. 2003) and exhibits minimal drift in ocean temperature without the use of flux corrections (Wu et al. 2003). FOAM captures most features of the observed climate (Jacob 1997; Wu et al. 2003; Harrison et al. 2003). This includes a reasonably realistic simulation of tropical monsoons, ENSO behavior, subtropical arid belts, and the seasonal cycle of climate in the extratropics. Notable biases include a colder than observed high northern latitudes, especially in winter.

The basic CCM2 land model (Hack et al. 1993), used in the standard version of FOAM (which implicitly assumes prescribed vegetation structure), has been replaced with a land component incorporating the Lund-Potsdam-Jena Global Dynamic Vegetation Model (LPJ-DVGM) in which vegetation structure and cover are predicted interactively with the evolving FOAM climate. The land grid is the same as for the ocean (1.4 latitude×2.8 longitude) and the atmospheric inputs needed to drive the LPJ-based land component are interpolated from the coarse R15 atmospheric grid to the higher resolution surface grid via a grid meshing coupler scheme analogous to that used for the NCAR CCSM. In substituting the original FOAM land component with the LPJ-based one, we retain the original FOAM CCM2-based surface/soil diffusive temperature calculation scheme that assumes four soil layers, but replaces the simple, single layer (bucket) soil water component (Harrison et al. 2003) with the two-layer soil water scheme of LPJ (Sitch et al. 2003; Haxeltine and Prentice 1996). Local runoff is calculated when the soil water of a layer exceeds water holding capacity and is used as input into the river routing scheme for freshwater discharge into the oceans.

2.2 LPJ

LPJ is designed for efficient computation of the interactions between terrestrial ecosystem structure (vegetation height, biomass, composition) and function (energy absorption, coupled carbon and water exchanges). LPJ determines the plant functional type (PFT), its average properties (height, crown area, leaf-stem-root biomass), and the number of individuals per unit area. Biogeochemical processes and representations of PFTs in LPJ are based on the BIOME 3 model of Haxeltine and Prentice (1996). Vegetation dynamics are represented by a formulation which captures the essential processes simulated by individual-based models (self-thinning, succession, gap-phase dynamics) in a parameterized way for large-scale application in a global climate model. Natural disturbance by fire is expressed as a probability, determined by minimum fuel availability and surface moisture content.

LPJ incorporates nine PFT types in the version we coupled to FOAM (seven tree and two grass types; Table 1). Using climatic input from the FOAM part of the simulation, the LPJ vegetation modules (Fig. 1) sort out which PFTs can establish, survive, and dominate from an equal opportunity to compete for resources. The LPJ version used for this study calculates only natural vegetation cover (associated with these nine PFTs); anthropogenic land use, such as including agricultural crop PFTs, is not considered. We also note that the current LPJ does not explicitly determine shrubs, but, for example, shrub-like tundra vegetation is produced in LPJ when arid/cold conditions limits the growth of trees. LPJ, when forced with CRU (Climate Research Unit) climate data, produces a realistic structure of natural vegetation cover, including most areas of observed boreal, temperate, and tropical forests (Sitch et al. 2003). Grasslands are also reasonably well simulated, although LPJ has a bias toward over estimating woody PFTs relative to grass in some areas (African Sahel and US Great Plains).
Table 1

List of plant function types (PFT) used in LPJ (Sitch et al. 2003)




Tropical broadleaved evergreen tree


Tropical broadleaved raingreen tree


Temperate neeedleleaved evergreen tree


Temperate broadleaved evergreen tree


Temperate broadleaved summergreen tree


Boreal needleleaved evergreen tree


Boreal summergreen tree


C3 grass


C4 grass
Fig. 1

Schematic of time-stepping and coupling for the FOAMLPJ land model with the LPJ dynamic vegetation components. The model in fully coupled mode provides two-way interaction between the atmosphere and land model via a coupler interface. The time-stepping involves three distinct update calculations; sub-daily (20 min), daily, and annual, with this latter calculation used to update the annual state of vegetation structure. For the fixed vegetation cover runs, the calls to the yearly time-stepping calculations are bypassed in favor of the initial prescribed state of vegetation cover in the restart input

The coupling of LPJ to FOAM necessitated strapping of the component models on three different time scales (FOAM’s basic diurnal time step of 20 minutes with LPJ’s intrinsic daily and annual time steps). The basic connection of the time stepping links is illustrated in Fig. 1. LPJ supplies FOAM with a daily value of canopy conductance, soil water, and runoff (from the LPJ soil water model), leaf-on fraction (phenology)/daily leaf area index (LAI), and an annual value of the distribution of PFT coverage for each land grid cell. The annual net primary productivity (ANPP), used in computing the vegetation cover and structure at the end of each year (Fig. 1), is determined by summing up over the year the daily gross primary productivity (derived from the LPJ photosynthesis model) less the daily respiration cost. From these LPJ-derived vegetation inputs to FOAM, the biophysical effects of the vegetative/soil water influence on the surface fluxes are calculated and passed to the atmospheric component of FOAM via the coupler. In turn LPJ requires daily means of the FOAM calculations of the downward solar flux at the surface, net surface radiation, surface temperature, rainfall, snowfall, and snowmelt.

2.3 Modifications to FOAM and LPJ needed for coupling

The coupling of FOAM and LPJ necessitated developing a common meshing parameterization for some key physical processes (evaporation and surface albedo) that are represented by separately defined formulations in each model component. In FOAM, surface evaporation is given by an aerodynamic-based representation (Hack et al. 1993) that assumes the latent heat exchange between the surface and atmosphere is proportional to wind speed, a roughness length and stability dependent on drag coefficient, and the surface to air gradient of specific humidity. In LPJ surface, evaporation is given empirically by an evapotranspiration process within plants (Sitch et al. 2003; Haxeltine and Prentice 1996). Specifically, it comprises taking the minimum of two functions; one defining the availability of soil water for root uptake (supply) and the other the availability of light (day time net radiation) for opening leaf stomata (demand). For FOAMLPJ, the LPJ evaporation process is used for space occupied by leaf covered vegetation and the FOAM formulation is used for space covered by leafless vegetation or bare ground. In coupling LPJ to FOAM, we modified the original LPJ daily evapotranspiration process to allow for the diurnal calculation of soil temperature (Fig. 1).

We modified the FOAM surface albedo to more realistically account for snow in forest canopies with a resulting better fit to observations (Douville et al. 1995a, b; Dorman and Sellers 1989; Roesch et al. 2001). Originally FOAM utilized the CCM2 scheme whereby the fraction of snow area in a grid cell depends on the snowdepth and surface roughness. This formulation led to higher than desired snow albedos and fractions for forest covered regions. The scheme adopted in FOAMLPJ takes into account the fraction of a forest canopy area covered with snow and the fraction of snow area on the ground that is not masked by leaves and stems; for a deciduous forest, only stems mask the snow covered ground. For a grass/bareground surface the snow cover fraction is determined as in the original CCM2-based FOAM prescription. The choice of fixed parameters in the formulation (fraction of stem area, fraction of canopy covered by snow, etc.) was made to produce calculated albedos of about 0.25 and 0.33, respectively, for boreal evergreen and deciduous forests with deep snowcover (25 cm).

A number of changes (mostly small) were made to LPJ in order to improve the vegetation simulation during the spin-up cycle (see below). The most notable of these involved the LPJ tree survival mechanism. In the original LPJ, all trees in a grid cell would be abruptly killed if the 20-year mean temperature of the coldest month falls below (even ever so slightly) a survival threshold (e.g. temperate deciduous trees do not survive if the coldest monthly mean temperature of the previous 20 years falls below 256 K). In FOAMLPJ, this catastrophic tree kill is ameliorated by providing a 3°C transition zone whereby within this thermal interval the fraction of trees killed increases linearly as the temperature decreases below the survival threshold with total kill occurring when the 20-year mean temperature of the coldest month reaches 3°C below the survival threshold (e.g. 253 K for temperate deciduous trees).

3 Experiments and spin-up procedure to equilibrium

3.1 Full coupled pre-industrial and mid-Holocene runs

Following full coupling of FOAM and LPJ, the model (FOAMLPJ) was run in testing mode starting with bare soil, but with equal opportunity for establishment and growth of all PFTs. The testing mode for the buildup of global vegetation was completed in 185 years of integration in which a number of coupling issues were found that necessitated modifications in parameterization to achieve reasonably realistic results (see Sect. 2.3). We then commenced with two fully coupled experiments starting from year 410 of a FOAM present day run without LPJ coupling and the vegetation structure that was obtained at the end of the testing runs (Year 185). The first coupled experiment (denoted 0 ka) consisted of a pre-industrial control run with modern orbital forcing and an atmospheric CO2 concentration set at 280 ppm. The second full-coupled experiment (denoted 6 ka) was a mid-Holocene run with 6,000 year BP orbital forcing and also a pre-industrial CO2 setting of 280 ppm. The same CO2 level was used in forcing both the climate and the plant physiology. The 0 and 6 ka experiments were each run 635 years from the same starting point with the last 400 years of integration used for comparative analysis in this study (Table 2). The monthly and seasonal means for 0 and 6 ka were obtained using a modern calendar definition (see e.g. Joussaume and Braconnot 1997).
Table 2

Interactive/fixed vegetation experiments


CO2 (ppm)

Orbital forcing (ka)

Annual vegetation cover

Seasonal leaf cover

Years run

0 ka






6 ka









Fixed at 400 year

average of 0 ka coverage






Fixed at 400 year

average of 0 ka coverage






Fixed at 400 year

average of 6 ka coverage



The simulated climate shows no major drift or trend in ocean temperature or vegetation cover (not shown). The runs were performed without flux correction in the ocean part of the simulations. Furthermore, no effort was made to adjust simulated land rainfall/temperature toward observations in order to remove model biases that might degrade the quality of the vegetation simulation. By not adjusting the model climate forcing, we can better assess how well the seasonal climate/vegetation interactions are working in coupled mode and to pinpoint potential problems in specifying parameterizations for vegetation response to simulated climatic conditions. In this study, we examined the vegetation response in the context of both modern and perturbed (6,000 year BP) climate runs.

3.2 Fixed vegetation experiments: assessing vegetation/climate feedbacks

A key aspect of this study is to examine not only the change in extratropical vegetation structure that results from the orbitally induced changes in radiation between mid-Holocene times and today, but also to assess the possible feedback role of the vegetation change on the climate response. The former is examined by comparing the full-coupled 0 and 6 ka experiments, while the latter objective necessitates carefully designed fixed vegetation experiments. In this study we focus on the impact of the long-term change in fractional coverage of vegetation in a set of three experiments described as follows (see also Table 2):

R0V0, a 400-year run of FOAMLPJ with modern orbital forcing (R0) but with each year of integration using the same 400-year mean PFT fractional coverage deduced from the full-coupled 0 ka run (V0).

R6V0, a 100-year run of FOAMLPJ with 6,000 year BP orbital forcing (R6) but with each year of integration using the same 400-year mean PFT fractional coverage deduced from the full-coupled 0 ka run (V0).

R6V6, a 100-year run of FOAMLPJ with 6,000 year BP orbital forcing (R6) but with each year of integration using the same 400-year mean PFT fractional coverage deduced from the full coupled 6 ka run (V6).

These experiments, somewhat analogous to those performed by Texier et al. (1997), are designed to explore the climatic response to extratropical vegetation changes, such as expansion of boreal forest into tundra and expansion of grassland in the mid-latitudes. In examining the various fixed vegetation experiments, the orbital forcing effect is isolated by comparison of R6V0 with R0V0, whereas the effect of mean vegetation cover changes alone is isolated by comparing R6V6 with R6V0. In addition a comparison of 6 ka minus 0 ka differences with R6V6 minus R0V0 differences provides insight into the potential effects of variability of vegetation cover on the climate response, since in the former set of differences year-to-year vegetation changes can occur, whereas in the latter case the vegetation cover is set perpetually as the long-term mean. Table 3 lists the various difference combinations and the associated forcing processes that are isolated that will be discussed in results section below.
Table 3

Isolation of orbital forcing/vegetation feedbacks

Experiment difference

Orbital forcing

Vegetation feedbacks

Mean of annual cover

Variability of cover

Seasonal leaf cover

6 − 0 ka





R6V6 − R0V0





R6V0 − R0V0





R6V6 − R6V0





For the fixed vegetation experiments, FOAMLPJ was integrated with the daily part of LPJ coupling turned on and the annual part (the part that determines PFT fractional coverage) turned off (Fig. 1). The fixed vegetation runs were performed by replacing the PFT coverage in the initializing restart file by that of the 400-year mean coverage and then holding that coverage invariant over the time integration of the model. Since the daily part of LPJ coupling was not skipped during the integration, soil water and summergreen leaf cover (deciduous trees and grass) were not held seasonally fixed in the experiments. Thus changes in seasonal leaf cover (LAI) brought about by, for example, the changes in seasonal rainfall/temperature that affect the growing season length, are allowed to occur in the experiments designed to isolate the orbital forcing effects (e.g. R6V0 vs. R0V0). Enabling the phenology in the fixed vegetation cover experiments could produce, in some cases, unrealistic seasonal leaf cover in the presence of the assigned vegetation cover, however, the most extreme consequences of this (e.g. leafless trees) occur rarely with the mid-Holocene summer warming of the extratropics. For this present study, we did not examine the feedback role of daily LPJ coupled processes (e.g. soil water, LAI) on the climate response.

4 Results

4.1 0 ka pre-industrial climate and vegetation

FOAMLPJ possesses a notable cold (warm) bias in the Northern (Southern) Hemisphere during boreal winter (December–February, Fig. 2). The most severe winter bias is the much colder than observed conditions by over 20°C in the Atlantic sector poleward of 60N and northwestern Europe. A Southern Ocean/Antarctic warm bias reaches up to 10°C and is also seen to be comparably large in June–August. In boreal summer (June–August) there are local temperature errors as large as 5°C, but these are less spatially systematic compared to those in December–February; the cold bias present over the North Atlantic/European region in winter is also evident in summer, but is much less severe (around 5°C).
Fig. 2

The global distribution of surface temperature difference (°C) for northern summer (JJA, top) and winter (DJF, bottom) between the 400-year full-coupled FOAMLPJ 0 ka simulation and the NCEP reanalysis data (e.g. Kalnay et al. 1996)

Generally FOAMLPJ captures the major features of the observed distribution of rainfall, which include the summer monsoons and subtropical arid regions. Notable biases (Fig. 3) include a wet central and a dry eastern US (June–August), a more poleward summer monsoon in North Africa, a too dry eastern Indonesia (New Guinea) resulting from a penetration of the equatorial cold tongue too far into the western Pacific, and a wetter Australia (December–February) and a dryer Amazonia (in both summer and winter).
Fig. 3

As in Fig. 2, but for the precipitation difference (mm/day)

FOAMLPJ calculates a substantial natural forest cover, both in explicitly simulated PFT coverage and in offline-computed biomes based on model vegetation structure (Figs. 4, 5). The model generally produces forest cover where satellite observations (DeFries et al. 1999, 2000) also show natural forest (Fig. 4), taking into account that much of the discrepancy between the satellite observations and model forest cover is due to human land use which is not incorporated in the simulations (i.e. FOAMLPJ simulates a potential natural vegetation cover in the absence of human influence). The model produces too much total forest cover over eastern Asia, Australia, western half of the US, tropical Africa, and the North African Sahel. As it was found in the one-way forced runs of LPJ (Sitch et al. 2003), FOAMLPJ produces too much boreal deciduous tree cover compared to evergreen—a result that is correctable by increasing the mortality rate for boreal deciduous trees relative to that for the evergreens (Bonan et al. 2003). A comparison of the simulated tree cover poleward of 30N with that of a recent coupled simulation by Crucifix et al. (2002) using the VECODE dynamic vegetation model (Brovkin et al. 1997) reveals similar results except for the latitude belt 35–50N where FOAMLPJ produced more forest cover (not shown). Our results for modern global tree cover also compare favorably with the simulation of Brovkin et al. (2002)
Fig. 4

The global distribution of deciduous, evergreen, and total tree cover (%) from the 400-year full-coupled FOAMLPJ 0 ka simulation (left) and from satellite observations (right). The observed maps were prepared by Notaro et al. (2005) using the 1992–1993 global continuous fields of vegetation data taken from DeFries et al. (1999, 2000)
Fig. 5

The global distribution of the full-coupled FOAMLPJ 0 ka (top) and 6 ka (middle) biomes and the potential natural vegetation biomes (bottom). The model biomes are derived from output of PFT cover and height and growing degree days. The potential natural vegetation map is taken from Notaro et al. (2005) and was produced by collapsing a larger array of biomes (Ramankutty and Foley 1999) into a comparable set of biomes for comparison with the model-derived biomes

In an attempt to remove the conflicting issue of land use in measuring the model results, we computed a global map of seven biome categories for comparison with a similar set of potential natural vegetation biomes (Fig. 5). The FOAMLPJ biomes were computed from an algorithm using the simulated (400-year mean) PFT fractional coverage and height, and total annual growing degree days (Joos et al. 2004), whereas the potential natural vegetation biomes were subjectively aggregated into seven analogous biome categories (Notaro et al. 2005) from a larger independently derived biome data set that was obtained using observed climate forcing (Ramankutty and Foley 1999). The comparison of these two sets of biomes should be viewed as somewhat qualitative since the rules for each biome computation are not identical.

In broad terms, the model-produced biomes and the potential natural vegetation biomes (Fig. 5) correspond reasonably well in location of desert, tundra, and forest (tropical, temperate, and boreal) eco-systems, but not so well for grassland; the observed grass areas in the western US, central-eastern Asia, and Australia are not adequately captured and the model overproduces grass in the eastern US. The largest discrepancy in biome results is the greater (lesser) area of boreal (temperate) forest in the FOAMLPJ simulation, in part, caused by the model cold bias (Fig. 2).

While LPJ has a known bias for overproducing woody PFTs relative to grass that may contribute to the FOAMLPJ tree bias in the US Great Plains area (Sitch et al. 2003; our reconstruction of LPJ vegetation (not shown) driven offline using the monthly NCEP reanalysis rainfall and temperature data that was interpolated to daily forcing (Kalnay et al. 1996)), by far the greatest cause for model vegetation error lies with the biases in simulated climatology. Based on our comparison of the coupled run with the offline reconstruction, some of these climate bias driven vegetation errors are found to include:
  1. 1.

    An overly extensive tundra/polar desert in northern Europe/Siberia (Fig. 5) resulting from a combination of severe winter cold that precluded all but the hardiest of boreal trees and colder than observed summer conditions (Fig. 2) that restricted the growing season.

  2. 2.

    An overextensive and too far equatorward push of boreal forest in eastern Asia (Figs. 5) resulting from colder winters and wetter conditions than observed (Fig. 2, 3).

  3. 3.

    An overly extensive temperate forest in the US Great Plains and Australia (Fig. 5) resulting from too much simulated rainfall (Fig. 3).

  4. 4.

    An overextensive grass cover in the US Great Lakes region, eastern Europe/western Asia around 50N and extreme eastern Asia (30–50N) (Fig. 5) resulting from a combination of near observed summer warmth and notably colder than observed winters (Fig. 2), which together inhibited the LPJ survival and development of boreal and temperate tree PFTs leaving grass as the only vegetative alternative.

  5. 5.

    A too far poleward push of vegetation into the African Sahel (Figs. 4, 5) resulting from a too far northward shift of summer monsoon rainfall (Fig. 3).

  6. 6.

    A reduced Amazonian rainforest (Figs. 4, 5) resulting from an underestimate of rainfall (Fig. 3).


4.2 6 ka extratropical climate and vegetation

The simulated FOAMLPJ climate for the mid-Holocene shows many of the differing features from modern (Fig. 6) that have been reported in previous studies (Joussaume et al. 1999; Hewitt and Mitchell 1998; Kutzbach et al. 1998; Texier et al. 1997; Gallimore and Kutzbach 1989; Kutzbach and Gallimore 1988). These include a broad summer warming of the northern continents, reaching more than 2.5°C over areas of the large Euro-Asian land mass and up to 1.5°C over the smaller North American continent. While the most significant rainfall feature is the increased northern summer monsoons at 6 ka, there is also a notable tendency for increased (decreased) rains over extratropical areas where tree cover increased (decreased) between 6 ka and present (Figs. 6, 7). Over most of the US and southern Canada, the FOAMLPJ rainfall anomalies are similar to those associated with the simulated intensification of the American monsoon at 6 ka as described in Harrison et al. (2003). The dipole rainfall anomaly in far eastern Asia around 30–60N (negative just inland from the Asian coast and positive further west) is linked to a westward push of maximum summer rains to more inland locations from the Asian coast at 6 ka in response to the enhanced warming of the continental interior; while a similar response is also noted in Liu et al. (2003, 2004), the coarseness of FOAMLPJ orography could be unrealistically impacting the monsoon sensitivity around the Tibetan plateau (Figs. 3, 6). FOAMLPJ simulates a generally increased aridity in association with a decrease in soil water over much of the interior of the extratropical continents (especially Eurasia). This mainly occurs in response to increased summer solar radiation and warmth at 6 ka (Fig. 6), which together drive enhanced surface latent heat (evapo-transpiration) and sensible heat fluxes. In some areas the general aridity is augmented by decreased rainfall, whereas in some other areas (notably in central North America near the Great Lakes region) it is alleviated by enhanced rainfall that is of sufficient magnitude to produce increased levels of soil water in summer.
Fig. 6

The distribution of anomalous (6 − 0 ka) summer (JJA) surface temperature (top, °C), precipitation (middle, mm/day) and top-layer soil water (bottom, fraction of saturation) from the full- coupled 400-year FOAMLPJ experiments

The simulated change in vegetation between 6 and 0 ka involves an increase in tree cover in the region 60–90N and a decrease in tree cover in the middle latitude belt (35–60N) (Figs. 7, 8). The increased tree cover poleward of 60N reflects the simulated expansion of boreal forest into tundra/polar desert, especially in Siberia and northwestern North America. The expansion of boreal forest cover is aided primarily by the increased summer warmth (Fig. 6), which fuels a more active growing season at 6 ka. A positive feedback of increased spring warmth associated with the boreal tree expansion further enhanced the 6 ka growing season (see below). Most of the expanded vegetation cover derives from the hardiest model PFTs (grasses and boreal summergreens) as the simulated winter conditions at 6 ka are comparably harsh to 0 ka conditions. For regions such as northern Europe, the severity of winter cold precludes an accurate simulation of the vegetation sensitivity.
Fig. 7

The distribution of anomalous (6 − 0 ka) tree cover (%) from the full-coupled 400-year FOAMLPJ experiments
Fig. 8

Four hundred year time series of fractional tree cover from the full- coupled FOAMLPJ 6 ka (dashed) and 0 ka (solid) simulations for the zonal land areas 60–90N minus Greenland (top) and 35–60N (bottom). Also shown for reference is the long term 400-year mean tree cover for each time plot (heavy horizontal line)

The decreased mid-latitude tree cover at 6 ka (35–60N), reflected by the simulation of expanded grasslands at the expense of forest cover, is most evident over central North America, far eastern Asia, and eastern Europe extending to central Asia. The increase in 6 ka grassland is primarily a consequence of two factors, enhanced mid-Holocene seasonality and summer drying. The seasonality factor is a direct consequence of the orbital-enhanced insolation difference between summer and winter; the sensitivity of the mid-latitude tree cover to this enhanced radiation is increased by a greater than observed present day seasonality of temperature that becomes further exaggerated at 6,000 year BP to the point where in some areas excessive warm summers preclude boreal trees and excessive cold winters inhibit temperate trees. The importance of these two components of the model’s climatic change (enhanced seasonality and summer drying) in contributing to the expansion of grass cover varies regionally. Areas of increased grass cover in south-central Canada, eastern and central Asia, and parts of Europe coincide with decreased soil water levels and precipitation (Figs. 6, 7). In contrast, tree cover declines in the US Great Lakes region occur in response to enhanced seasonality of temperature; for this region, water availability is not an issue as soil water and precipitation at 6 ka are at or greater than for 0 ka. We note further that the expansion of grassland produced a positive feedback of colder spring conditions that limited the growing season support for tree cover (see below).

The simulation of a poleward expansion of boreal forest and middle latitude expansion of grass cover is consistent with many previously published model results, e.g. simulated vegetation changes derived from coupled climate/vegetation models (Crucifix et al. 2002; Ganopolski et al. 1998; Texier et al. 1997; Brovkin et al. 2002; Wohlfahrt et al. 2004) and vegetation changes inferred using anomalous climate forcing (mid-Holocene minus modern) from uncoupled climate models (J. Wohlfahrt et al., submitted; Harrison et al. 1998; Kutzbach et al. 1998). Some of the causes for the simulated vegetation changes noted in this study (summer/fall warmth and interior continental drying) are amongst those also noted in previous studies and are nicely summarized in some detail in Harrison et al. (1998) and J. Wohlfahrt et al. (submitted).

The model results for expanded forest cover into the tundra/polar desert areas of northern Canada (Mackenzie delta), western Europe, and western and central Siberia is supported by paleo-observations (Texier et al. 1997; Harrison et al. 1998; Tarasov et al. 1998; MacDonald 1995; MacDonald et al. 2000; Prentice et al. 2000; Bigelow et al. 2003), although the amount of simulated expansion into these areas is greater than observed. The simulated northward expansion of boreal forest in far eastern Siberia, Alaska, and parts of northeastern Canada is not supported by the paleo-evidence (Prentice et al. 2000). The mismatch in northeastern Canada (where the model shows expanded boreal tree cover, while the observations indicate an even more southerly position of the tree line in Labrador and Keewatin at 6 ka than present) may in part be due to the neglect of the small remnant ice sheet around 6,000 year BP in the simulations (Ruddiman 2001; J. Wohlfahrt et al., submitted). There is some evidence supporting an expansion of grassland/shrubland in middle latitude areas that occurred in our FOAMLPJ 6 ka simulation, e.g. an expansion of prairie north and east of its present range in North America (Prentice et al. 2000; Harrison and Prentice 2003); however, the paleo-data, although somewhat sparse, suggests little change over Eurasia (Harrison et al. 2003, 1998; Tarasov et al. 1998; Prentice et al. 2000; Wohlfahrt et al. 2004). Paleo-evidence also indicates a decided northward shift of temperate forest into present-day boreal forest area in eastern North America, Europe, and western Asia (Harrison and Prentice 2003; Wohlfahrt et al. 2004). This result for expansion of temperate into boreal forest was generally not simulated in our FOAMLPJ experiments (Fig. 5) because of the aforementioned effects of model enhanced seasonality (especially colder winters).

4.3 6 ka climate response to vegetation feedback

FOAMLPJ produced large-scale changes in the fractional tree coverage between 6 and 0 ka. These changes include a significant 6 ka reduction of tree cover between 35–60N (Fig. 8), concentrated in eastern Asia and central North America (Fig. 7), and an enhanced tree cover poleward of 60N (Fig. 8), concentrated over northwestern Canada and most of Eurasia, especially in far eastern Siberia (Fig. 7). For the zonal belt 60–90N the fractional cover of trees increased from 0.31 to 0.49 between 0 and 6 ka; in contrast, the fractional tree cover for 35–60N decreased from 0.69 to 0.57.

There is a moderate inter-decadal variability in tree cover in high latitudes (Fig. 8). This is mainly contributed by multi-decadal scale fluctuations in tree cover at the northern fringe of the boreal forest. It will be the aim of future study to examine the nature of the variability in detail, but suffice it to say that the maximum range of the tree cover variability in polar latitudes (about 0.1 in Fig. 8, top) is significantly less than the mean state difference between 6 and 0 ka (about 0.18).

The magnitude of the time-averaged tree cover changes suggests a potentially large vegetation/snow albedo feedback in the model. In FOAMLPJ a forest with deep snow cover has an albedo ranging from 0.25 (evergreen) to 0.33 (deciduous), whereas grassland/bareground with deep snow has an albedo exceeding 0.7. To assess the feedback effects of this vegetation change, we ran three fixed vegetation cover experiments to compare to the full-coupled 6 and 0 ka cases; 0 ka orbital forcing with fixed 0 ka vegetation cover (R0V0), 6 ka orbital forcing with 0 ka fixed vegetation cover (R6V0), and 6 ka orbital forcing with 6 ka fixed vegetation cover (R6V6) (see Tables 2, 3; Sect.3.2 for more details). We then formed three sets of differences between the fixed vegetation cover experiments (Figs. 9, 10, 11, 12) in an effort to isolate the separate effects of the orbital forcing changes from the vegetation changes (Table 3). Specifically, R6V6 − R0V0 differences reveal the combined effects of orbital forcing and mean state vegetation cover changes, R6V0 − R0V0 differences reveal the effects of orbital forcing changes and R6V6 − R6V0 differences reveal the feedback effects of the changes in mean state vegetation cover.
Fig. 9

The spring (MAM—top four panels) and summer (JJA—bottom four panels) distribution of anomalous surface temperature (°C) in the extratropics (35–90N) from the FOAMLPJ full coupled and fixed vegetation cover experiments. Each set of four panels shows the full coupled 6 − 0 ka differences or total change (top left) and three combinations of fixed vegetation cover differences: R6V6 − R0V0 or Orb+Veg change (top right) to isolate the combined effects of the change in orbital forcing plus vegetation cover, R6V0 − R0V0 or Orb change (bottom left) to isolate the effects of the change in orbital forcing, and R6V6 − R6V0 or Veg change (bottom right) to isolate the effects of the change in vegetation cover
Fig. 10

The summer (JJA) distribution of anomalous precipitation (mm/day) in the extratropics (35–90N) from the FOAMLPJ full coupled and fixed vegetation cover experiments. The set of four panels are arranged as in Fig. 9
Fig. 11

Annual cycle of anomalous a surface temperature (°C), b snow depth (cm H2O water equivalent), c precipitation (mm/day), d top layer soil water (fraction of saturation), e latent heat of total evaporation (W/m2), f latent heat of evapotranspiration (W/m2), and g latent heat of ground evaporation (W/m2) averaged over the land area 35–60N from the FOAMLPJ full coupled and fixed vegetation cover experiments. Plotted are the full coupled 6 − 0 ka difference or total change (thick solid-dot) and three combinations of fixed vegetation cover differences: R6V6 − R0V0 or Orb+Veg change (thick dashed-dot) to isolate the combined effects of the change in orbital forcing plus vegetation cover, R6V0 − R0V0 or Orb change (thin solid) to isolate the effects of the change in orbital forcing, and R6V6 − R6V0 or Veg change (thin dashed) to isolate the effects of the change in vegetation cover
Fig. 12

As in Fig. 11 except averaged for the land area 60–90N minus Greenland

The similarity in the spatial structure of the anomalous northern spring/summer temperatures (6 − 0 ka vs. R6V6 − R0V0) indicates that the combined effects of orbital and mean state vegetative cover changes dominate as the cause for the total difference response (6 − 0 ka) in the full-coupled experiments (Fig. 9). The total spring (MAM) temperature change consists primarily of cooling with maximum temperature decreases exceeding 2°C over central North America and eastern Asia. There is also a modest warming in northwestern Canada and extreme northern Eurasia, especially far eastern Siberia. While the change in orbital forcing (R6V0 − R0V0) induces much of the total response, there is a notable contribution induced by the vegetation cover changes (R6V6 − R6V0). In particular, spring cooling of 1–2°C occurs in central North America and central and eastern Asia where tree cover decreased (Figs. 7, 9) and warming occurs in northwestern Canada and across northern Eurasia where tree cover increased. As will be discussed in more detail below, this response is linked to the vegetation/snow/albedo feedback whereby increased (decreased) tree cover led to a lowered (raised) surface albedo and a coincident decrease (increase) in snow cover and increase (decrease) in temperature.

In northern summer (JJA), by far the main cause for the total temperature change (mainly warming) is the changed orbital forcing (compare 6 − 0 ka, R6V6 − R0V0, R6V0 − R0V0 differences in Fig. 9). Compared to spring, the contribution to the total summer temperature change from the changed vegetation cover is quite small (R6V6 − R6V0 difference, Fig. 9). The spatial connection between the change in summer temperature and vegetation cover is more complicated than in spring. In some areas, the temperature change is of the same sign as in spring, but of smaller magnitude. For example, a modest summer warming of about 0.25–0.5°C occurs in northwest Canada/Alaska and much of eastern Siberia where tree cover increased, and a modest cooling of about the same magnitude occurs over central North America and mid-latitude east Asia where tree cover decreased. For these regions there appears to be a carryover of the larger temperature signal from spring. In other areas the temperature change reverses from spring, such as in Europe into western Asia, where at 50–60N we have tree cover decreases and warming (up to 0.5°C) and further north in polar latitudes we have tree cover increases and cooling.

The anomalous summer rainfall distribution (Fig. 10) provides a clue into the reasons behind the more complicated summer thermal response in association with the vegetation change. As with the temperature response, the orbitally forced precipitation anomalies (R6V0 − R0V0) account for much of the total precipitation change seen in the full experiment (6 − 0 ka); the association of this rainfall response to the model aridity and vegetation changes was discussed in Sect. 4.2 and will not be further elaborated here. In the experiment that isolates the role of vegetation cover change (R6V6 − R6V0), there is a tendency for increased rainfall to occur over the polar areas of greatest tree cover increase (Figs. 7, 10) and decreased rainfall to occur over the principle centers of mid-latitude tree cover decrease (e.g. the regions of North America north of the US Great Lakes, Eastern Europe (40E, 55N), central Asia (80E, 55N), and extreme Eastern Asia (120E, 40–30N)—see Figs. 7, 10). Part of the structure of temperature response in the vegetation cover experiments is connected to the rainfall anomalies; where rainfall increases (decreases) cooling (warming) occurs. This is particularly the case in the European/western Asian region. Physically, the increased (decreased) rainfall and associated lower (higher) temperatures for these regions are further linked to increased (decreased) cloud cover, a generally reduced (enhanced) solar flux to the surface, higher (lower) surface evaporation, and wetter (dryer) soils (figures not shown).

In the following sections, we examine in more detail the seasonal structure of the anomalous climate response to the changes in orbital forcing and vegetation cover in the experiments. We do this for the latitude belts 35–60N and 60–90N, which encompasses the main areas of extratropical tree cover decreases and increases, respectively (Figs. 7, 8). In general, the differences in anomalous seasonal results between the 6 − 0 ka and R6V6 − R0V0 cases are quite small, essentially verifying that the variability of vegetation cover has little influence on the temporal mean state differences in the climate. Of the combined effects, the orbital forcing change (R6V0 − R0V0) contributes the most to the full response, but there are significant mitigating influences coming from the changes in the vegetation cover (R6V6 − R6V0).

4.3.1 Anomalous response to changed orbital forcing and vegetation for 35–60N

The changed orbital forcing alone (R6V0 − R0V0) induces winter/spring cooling (summer/fall warming) (Fig. 11). Associated with the spring cooling is a greater snow depth (maximum in April) and subsequent increase in soil water (maximum in May) when the excess snow melts (Fig. 11). The enhanced 6 ka summer radiation combined with the added soil water drives increased local rainfall and evaporation in early summer, with the total evaporation rate increase (maximum about 4.8 W/m2 or 0.16 mm/day) exceeding the precipitation rate increase (0.12 mm/day). Because evaporation exceeds precipitation, the positive spring pulse of soil water gives way to a drying (negative soil water fraction anomaly) in summer/fall. The positive total evaporation anomaly from late spring into summer is first comprised of a positive contribution from ground evaporation in late spring (following snow melt, but preceding full leaf emergence) and then of a net positive contribution from positive evapo-transpiration over the negative ground evaporation anomaly in mid- summer under full leaf cover.

In the changed vegetation cover experiments (R6V6 − R6V0), the shift from tree to grass cover produced a comparable increase in snow depth (April) and soil water (May) and decrease in spring surface temperature as was found in the orbital only case (Fig. 11). The associated mechanisms behind this response are the greater fractional coverage of snow, higher surface albedo, and reduced surface absorption of solar energy that accompanies the decreased tree cover in middle latitudes. Compared to the orbital change experiment (R6V0 − R0V0), the anomalous response in the changed vegetation cover case (R6V6-0R6V0) is small over the subsequent summer/fall seasons; the enhanced surface radiation at 6 ka is clearly the major driver for the total summer/fall response in the full-coupled experiments (6 − 0 ka). Although the changes in total evaporation in response to decreased tree cover are small, there are significant compensating changes in the component fluxes of evapo-transpiration and ground evaporation. Over much of the late spring to summer season (especially May/June) a negative evapotranspiration anomaly is balanced by a positive anomaly for ground evaporation. This structure of counterbalancing component evaporation fluxes is a result of increased bare soil exposure (more bare ground and less leaf cover) with the shift to more grass and bareground cover between 35–60N at the 6 ka compared to 0 ka. It should be reiterated that while the anomalous summer/fall response to the decreased tree cover is relatively small, when averaged over the 35–60N latitude band (Fig. 11), there are modest regional responses to local vegetation cover changes as was discussed in Sect. 4.3 (Figs. 7, 9, 10).

4.3.2 Anomalous response to changed orbital forcing and vegetation for 60–90N

Although a bit more complicated in nature, the responses in the full cases (6 − 0 ka, R6V6 − R0V0) and the changed orbital case (R6V0 − R0V0) for the area 60–90N (Fig. 12) carry elements of the same sensitivity features as for the area 35–60N (Fig. 11); the principle anomalous cooling (warming) for both areas is in spring (summer/fall). In contrast to 35–60N, fall warming extends well into the polar night winter, when there is no contributive cooling impact from a negative radiative change. As also found in previous studies (Kutzbach and Gallimore 1988; Ganopolski et al. 1998), reduced sea ice cover and thickness (not shown) contributed to a longer anomalous warming season at 6 ka over polar land areas. The long warm period from fall to winter is associated with reduced snow accumulation in the 6 ka cases. This leads to a decrease in snow melt for spring and less spring soil water than for modern (Fig. 12), a result in contrast to that for 35–60N. Interestingly, there is a saw tooth pattern in the negative anomalous snow depths in spring that in part reflects the impact of more radiation cooling (warming) in May (June) at 6 ka leading to less (more) relative snow melt compared to 0 ka. As was found for 35–60N, the warmer summer period shows a 6 ka drying of the soil resulting from an increase in evaporation exceeding the increase in rainfall. In fall the situation reverses and soil water increases in response to more precipitation over evaporation. As was also found for 35–60N, the orbitally forced, positive total evaporation anomaly derives from an early contribution from a positive ground evaporation anomaly and then later from a positive anomalous evapo-transpiration.

The expansion of boreal forest into the polar area produces an opposite response (R6V6 − R6V0 differences) to that of the 35–60N area where tree cover decreased (Fig. 11, 12). This is most evident in spring (May) where slightly higher temperatures and much reduced snow depths are produced as a consequence of the lowered albedo accompanying increased boreal tree cover under snow covered conditions. Coupled with these changes are small anomalous increases in precipitation, soil water (from earlier snow melt), and evaporation. As was found for 35–60N, the changes in summer temperature, soil water, and precipitation are much smaller than the orbitally produced responses. The generally positive anomalous total evaporation from spring to fall derives from a positive ground evaporation from exposed soil before leaf emergence and then from increased plant transpiration in summer (when the bare ground evaporation component decreases relative to modern, i.e. displays a negative anomaly). For reasons not entirely clear, there is a positive snow depth anomaly in the cold season for the vegetation change case (i.e. greater snow depths but with increased tree cover at 60–90N).

5 Conclusions and discussion

In this study, a full global atmosphere/ocean model coupled to a dynamic vegetation model is used to examine the extratropical modern versus mid-Holocene vegetation structure in the context of the climatic processes driving the vegetation response and potential climate/vegetation feedbacks. The results are extracted from a 400-year period of integration during which the vegetation structure has sufficiently equilibrated following a multi-century interval of competitive successional development. The present-day natural vegetation produced by the model is relatively realistic: most notable discrepancies are a model cold bias that pushes colder biome types (polar desert, tundra, and boreal forest) further south of their observed range and regional wet/dry biases that lead to local errors in vegetation cover (e.g. too much tree cover in a wet US Great Plains and Australia, reduced Amazonia rainforest because of sporadic seasonal dryness).

The mid-Holocene simulation, in response to the changed orbital forcing between 6 ka and present, produces many of the climatic features found in other models (interior northern continents that are warmer and somewhat dryer in summer and colder in winter, reduced sea ice leading to an extension of polar warming into fall/winter, expanded and/or intensified monsoons). The simulated mid-Holocene climate changes (from present) produced a northward expansion of boreal forest into modern tundra areas (because of increased summer/fall warmth) and an expansion of mid-latitude grassland at the expense of forest (because of summer dryness and/or an enhanced seasonal range of temperature); these changes in vegetation have been noted in the observational record but to a lesser extent.

Experiments were designed to examine the impact of the extratropical vegetation cover changes on the simulated climate; the dominant component of the feedback comprises the vegetation/snow/albedo mechanism, whereby a forest cover with deep snow has an appreciably lower surface albedo than a grass cover. Except during spring the orbital forcing change from 0 to 6 ka produced the preponderance of climate change seen in the coupled experiments. These include a colder spring and an increase in evaporation exceeding an increase in rainfall with a resultant drying and increased warmth in summer. The shift from tundra to boreal forest in polar latitudes produced a warmer spring with reduced snow cover, which partially counteracts the orbitally forced response; in the middle latitudes the shift from forest to grass cover led to a colder spring with greater snow cover and a delayed pulse of soil water input from snow melt, thus, essentially reinforcing the orbitally forced response. In summer/fall the broad-scale climate changes associated with the vegetation cover changes are relatively small, mainly a result of the balancing effects between the changes in plant transpiration and the direct evaporation from exposed soil. Locally, however, there is a tendency for summer rains to increase (decrease) in the areas of greatest tree cover increases (decreases).

The vegetation feedback with full ocean/atmosphere coupling, determined in this study from the R6V6 − R6V0 differences, can be compared with results from analogous experiments performed in some previous studies (e.g., OAV 6 ka − OA 6 ka differences in Wohlfahrt et al. 2004; HOL5 − HOL0 differences in Texier et al. 1997; AOV − AO differences in Ganopolski et al. 1998). The feedback of spring warming associated with the expansion of boreal forest cover in polar latitudes (Fig. 12) is comparable to the 0.32°C warming found by Texier et al. (1997), but appreciably less than the warming (1°C or more) obtained by Wohlfahrt et al. (2004). For mid-latitudes and in contrast to our study (which shows spring cooling—Fig. 11), Wohlfahrt et al. (2004) report a significant spring warming of 2°C for 50–60N and 0.89°C for 40–50N (their OAV − OA differences). For the summer season, the vegetation feedback on surface temperature in our experiments is weak, which contrasts with the significant warming found in the previous studies (Ganopolski et al. 1998; Texier et al. 1997; Wohlfahrt et al. 2004).

The reasons for these differences in results are difficult to trace as the models are notably different with regard to both individual components (e.g. the atmospheric model used in the Ganopolski et al. 1998 study is a very course resolution dynamical-statistical model, whereas in our study it is a R15 version of CCM3) and the coupling strategy (e.g. Texier et al. 1997 and Wohlfahrt et al. 2004) obtained their vegetation distribution via an asynchronous iteration of BIOME1 driven by averaged climate anomalies from their atmospheric model; whereas in our study the atmospheric and dynamic vegetation models are in continuous interaction on a variety of time scales). One factor potentially influencing the differing results is that in both the Texier et al. (1997) and Ganopolski et al. (1998) studies the total extratropical forest cover increased whereas it decreased in our experiments—the result of the mid-latitude expansion of grass cover at the expense of forest outweighing the increase in boreal forest cover in polar latitudes. However, the conflicting differences between our study and the Wohlfahrt et al. (2004) study regarding the surface temperature response in spring to mid-latitude vegetation change (cooling in our case and warming in their case) is not obvious, as both models produced grass expansions at the expense of forest in the mid-Holocene simulations.

The continued examination of potential vegetation/climate feedbacks with fully coupled atmosphere/ocean/biosphere models will require a careful strategy of experimental design that is only partially realized in this study. Our experiments only considered the change in PFT coverage. Our present effort to isolate the separate roles of changed orbital forcing (6 ka vs. 0 ka) from the changed vegetation cover thus only partially achieves its goal, since within the orbital only experiments the leaf phenology part of the vegetation change is not held fixed. Future work will need to look at the role of changes in seasonal leaf phenology (timing of leaf emergence, maximum coverage, and senescence) as well as soil properties (texture, color etc.). New work by Levis and Bonan (2004), who have coupled a hybrid version of LPJ to the community land model (CLM) of the community climate system model (CCSM), is examining some of these additional issues of vegetation/soil/climate feedbacks for mid-Holocene North Africa. We are planning a similar course of strategy for a more comprehensive examination of the feedbacks operating in the extratropics. For this purpose, it would be instructive to run a FOAMLPJ mid-Holocene experiment with prescribed modern ocean conditions (SST and sea ice) and vegetation cover. With this experiment the R6V6 − R6V0 difference case (defined here, as previously noted, to be the vegetation feedback but with allowance for the interactive ocean response) can be decomposed further into the separate vegetation feedback under fixed ocean conditions and the synergy between vegetation and ocean feedbacks. Performing this fixed SST experiment would also allow a decomposition of the orbital only case (R6V0 − R0V0 differences) into its separate atmosphere and ocean feedbacks (see Wohlfahrt et al. 2004 for more details).

Future plans call for doing a more quantitative model/data comparative analysis; an examination of the role of ocean feedbacks in altering both the mean state and variability of climate; and extending the analysis of vegetation feedback to the subtropics/tropics. Finally, work is planned on improving the model climate as the present biases in simulated temperature extremes restrict the realistic development of tree cover in some regions of the extratropical latitudes.


We thank Dr. Colin Prentice for his assistance and guidance on issues of vegetation/climate coupling, Mr. Mark Marohl for providing technical assistance in manuscript preparation, and Ms. Pat Behling for final preparation of the figures. We also would like to thank the two anonymous reviewers whose input helped to improve the paper. This work was sponsored by the National Science Foundation and the experiments were performed with computer resources at the National Center for Atmospheric Research.

Copyright information

© Springer-Verlag 2005