SEARCH

SEARCH BY CITATION

Keywords:

  • biosphere;
  • climate change;
  • ecosystem;
  • potential natural vegetation

Abstract

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. Model and Methodology
  5. 3. Results
  6. 4. Discussion and Conclusions
  7. Acknowledgments
  8. References

[1] A number of previous modeling studies have assessed the implications of projected CO2-induced climate change for future terrestrial ecosystems. However, although current understanding of possible long-term response of vegetation to elevated CO2 and CO2-induced climate change in some geographical areas (e.g., the high-latitude regions) has been strengthened by dint of accumulating evidence from these past studies, it is still weak in others. This study examines the responses of global potential natural vegetation distribution, net primary production (NPP), and fire emissions to future changes in atmospheric CO2 concentration and climate using the National Center for Atmospheric Research Community Land Model's dynamic global vegetation model. The model is run to vegetative equilibrium (i.e., with respect to leaf area index (LAI) and vegetation coverage) driven with preindustrial climate and future climate near 2100, respectively, simulated by eight general circulation models (GCMs). The simulated potential vegetation under the preindustrial control mean climate (CO2 concentration held at 275 ppm) is compared with that under the SRESA1B 2100 mean climate (CO2 concentration stabilizes at 720 ppm beyond 2100). Simulated vegetation response ranges from mild changes of the fractional coverage of different plant functional types to the rather dramatic changes of dominant plant functional types. Although such response differs significantly across different GCM climate projections, a quite consistent spatial pattern emerges, characterized by a considerable poleward spread or shift of temperate and boreal forests in the Northern Hemisphere high latitudes, and a substantial degradation of vegetation type in the tropics (e.g., increase of drought deciduous trees coverage at the expense of evergreen trees) especially in portions of West and southern Africa and South America. Despite the widespread degradation of vegetation type in the tropics, NPP, and growing season LAI are predicted to increase under most GCM scenarios over most of the globe. Carbon fluxes to the atmosphere due to fire generally increase too across the globe. Such responses of NPP and fire occurrence result from the synergistic effects of CO2 concentration changes, climate changes, and vegetation changes. In the HadCM-driven simulation, however, extreme responses are shown in some regions: Deciduous forest is replaced by grasses in large areas in the middle latitudes, and substantial areas in northern South America and southern Africa predominantly covered by evergreen forest are replaced with grasses while NPP and fire emissions reduce drastically (by more than 100%). A future paper will examine how the biosphere response documented here influences the impact of climate change on surface hydrological conditions.

1. Introduction

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. Model and Methodology
  5. 3. Results
  6. 4. Discussion and Conclusions
  7. Acknowledgments
  8. References

[2] Increasing concentrations of atmospheric greenhouse gases, particularly carbon dioxide (CO2), and the accompanying global warming, are all too apparent. Changes in regional and global hydrological conditions are among the major consequences of greenhouse gas warming [Labat et al., 2004; Thomson et al., 2005]. The synergistic effects of elevated atmospheric CO2, and associated temperature increases and water availability changes, may impose substantial control on terrestrial vegetation [Smith et al., 2005]. In fact, considerable evidence exists that global warming is already driving changes in vegetation productivity [Zhou et al., 2001; Nemani et al., 2003] and distribution [Root et al., 2003]. The mechanistic framework underpinning such impacts of greenhouse gas-induced climate changes on vegetation structure and function resides mainly in the interactive physiological and biochemical influences of atmospheric CO2 concentration, temperature and water availability on plant growth. Generally, elevated atmospheric CO2 tend to increase photosynthesis, water use efficiency, and hence vegetation productivity [Gerber et al., 2004; Faisal and Parveen, 2004] especially for C3 plants. In warmer regions, higher temperatures (therefore faster evapotranspiration) and/or reduced rainfall may lead to water stress and forest dieback, whereas higher temperature can lengthen the growing season and increase vegetation productivity in colder regions such as the middle to high latitudes where vegetation growth is limited by temperature instead of water availability [Wang, 2005]. Therefore, as atmospheric CO2 concentration continues to rise into the future, how vegetation will change (as a result of the combined atmospheric CO2 changes and climate changes) is expected to vary considerably according to regions and localities, and is a function of the initial environmental conditions, properties of species and competitive advantages among species [Shaver et al., 2000].

[3] Growing scientific and public concerns about the potential impacts of greenhouse gas-induced climate change incited a coordination of climate model experiments and data analysis by the international climate modeling community for the 4th Assessment Report of the Intergovernmental Panel on Climate Change (IPCC AR4). The response of the terrestrial biosphere to climate change is of obvious concern owing to its implications for the provision of food, timber and other life essentials [Churkina and Running, 1998; Norby and Luo, 2004], as well as for biodiversity, carbon sequestration and feedbacks to the climate system. General circulation models (GCMs) are the major workhorses for climate change assessments. The GCMs that participated in IPCC AR4 did not incorporate vegetation dynamics. The potential responses of vegetation to CO2 and climate changes predicted by these GCMs and subsequent feedbacks are therefore not known and not accounted for. The different GCMs vary substantially in their predictions of future climatic changes, particularly in the magnitude of temperature increase, and in both the sign and magnitude of precipitation changes [Wang, 2005]. This therefore compounds the uncertainty in how the terrestrial biosphere may change in the future.

[4] Dynamic global vegetation models (DGVMs) have emerged in the last decade and have been used as tools in several modeling studies for simulating transient and long-term changes in vegetation characteristics in response to changing climate and atmospheric CO2 concentration [e.g., Joos et al., 2001; Gerber et al., 2004; Lucht et al., 2006; Schaphoff et al., 2006]. A common finding from these studies is the simulated northward expansion of boreal forests, thus concretizing our understanding of the broad pattern of potential response of high-latitude ecosystems to CO2 and its induced climate changes. For other geographical regions such as the tropics and subtropics, however, substantial dissent exists in predicted responses, limiting current understanding of potential ecosystem response to climate change in these regions. Schaphoff et al. [2006] recently used the Lund-Potsdam-Jena dynamic global vegetation model (LPJ-DGVM) to study the response of the land biosphere to climate projections by five different GCMs under the same emission scenario, focusing on carbon storage. The GCMs did not concur on the sign of response they produced in the tropics and subtropics due to large differences in their precipitation projections.

[5] Ideally, a DGVM can be synchronously coupled to a GCM in predicting future climate, which would allow the simulation of not only the response of the biosphere to CO2 and climate changes, but also the feedback of such biosphere dynamics to climate. However, with the lack of vegetation dynamics in the IPCC AR4 simulations, a pragmatic way to examine future vegetation changes is to use a DGVM coupled to a land surface model driven with climate change scenarios produced by the different GCMs. Although this approach cannot account for the feedback from vegetation to climate change, such a multi-GCM-based analysis offers information on robustness across models regarding the direction of future vegetation changes, as well as on a range of possible spatial patterns of changes. This will permit the identification of especially vulnerable regions and allow anticipation of possible ecological disruptions. However, few studies (especially in the context of IPCC AR4 climate predictions) have addressed the responses of vegetation to future climate changes produced by different GCMs. Scholze et al. [2006], in their climate change risk analysis for the terrestrial biosphere based on multiple scenarios from sixteen GCMs, ascertained that larger areas of the world's terrestrial ecosystems were affected for GCMs and scenarios that predicted higher global warming levels than for those that produced lower warming levels. Salazar et al. [2007] used an equilibrium vegetation model to study vegetation changes in South America in response to IPCC AR4 climate change scenarios, and identified a decrease of tropical forest area as a robust feature of vegetation response across the climate scenarios applied.

[6] The present study contributes to current understanding of potential biosphere response to CO2-induced climate change by examining the responses of simulated natural potential vegetation to a combination of elevated CO2 and climate changes projected by eight GCMs that participated in IPCC AR4. Here, the National Center for Atmospheric Research (NCAR) Community Land Model (CLM3.0) dynamic global vegetation model (CLM-DGVM) is used to simulate the potential vegetation. Our aim is to determine whether consistent patterns of response of the world's ecosystems emerge from simulations using the different GCM scenarios. The CLM-DGVM and LPJ-DGVM [Sitch et al., 2003] (the model used by Scholze et al. [2006]) share some common components but differ in their land surface model and plant phenology model. This difference however allows a qualitative intercomparison between our study and Scholze et al. [2006] that also used IPCC AR4 scenarios.

2. Model and Methodology

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. Model and Methodology
  5. 3. Results
  6. 4. Discussion and Conclusions
  7. Acknowledgments
  8. References

2.1. CLM-DGVM

[7] CLM-DGVM [Levis et al., 2004] simulates natural vegetation structure and distribution and their transient changes. The model is composed of a land surface scheme CLM3.0 [Oleson et al., 2004], a phenology module [Kucharik et al., 2000], and biogeochemistry and vegetation dynamics modules (based on LPJ-DGVM [Sitch et al., 2003]). Biogeophysical and biogeochemical processes are simulated with a 20-min time step while plant phenology is evaluated daily. Vegetation structure and distribution are updated yearly based on knowledge of the processes integrated at faster time steps.

[8] CLM3.0 is forced by the atmospheric forcing data at any given model time step to simulate biogeophysical, physiological and biogeochemical processes. The results are state of the land surface (e.g., soil moisture and temperature, and vegetation temperature) and fluxes of water, heat, and carbon (e.g., gross primary production (GPP), autotrophic, and heterotrophic respiration). Net primary production (NPP) is then calculated as GPP minus autotrophic respiration (the sum of growth and maintenance respiration), and summed annually to update plant carbon stores [Sitch et al., 2003]. The phenology module updates leaf area index (LAI) daily based on annual growing degree days, temperature, soil moisture and NPP [Foley et al., 1996]. The vegetation dynamics component of CLM-DGVM allocates the annual NPP (less reproductive cost) to plant leaves, sapwood, heartwood and roots according to allometric relationships [Sitch et al., 2003]. The model uses 10 plant functional types (PFTs), namely, needleleaf evergreen temperate trees, needleleaf evergreen boreal trees, broadleaf evergreen tropical trees, broadleaf evergreen temperate trees, broadleaf deciduous tropical trees, Broadleaf deciduous temperate trees, broadleaf deciduous boreal trees, C3 arctic grasses, C3 nonarctic grasses and C4 grasses. Up to 10 PFTs may coexist in each grid cell. Bioclimatic rules in terms of temperature, growing degree days and precipitation determine the survival and establishment of PFTs; competition among PFTs for water and light under the prevailing soil and climatic conditions is governed by the PFT's foliage projective cover, LAI, height, and rooting profiles [Bonan et al., 2003; Levis et al., 2004]. Leaf and root turnover, as well as carbon from PFTs upon mortality, are converted to litter, while sapwood turnover is transferred to heartwood. Fire is simulated over every grid cell annually, and its occurrence and effect depend on the above ground litter, topsoil moisture, and surface air temperature [Levis et al., 2004].

[9] The DGVM of CLM3.0 has been evaluated and found to simulate global vegetation distribution and NPP broadly consistent with observations [Bonan et al., 2003; Sitch et al., 2003]. However, CLM-DGVM, like many other DGVMs [Cramer et al., 2001], is not without biases. Dickinson et al. [2006] identified biases in land climatologies of simulations by the Community Climate System Model (CCSM3), a global climate model which uses the same land surface scheme (CLM3.0) as in CLM-DGVM. They suggested that better treatments of canopy interception, soil water storage, runoff and transpiration could improve a prominent dry bias in the Amazon. In evaluating CLM-DGVM against observations, Bonan and Levis [2006] noted that reducing canopy-intercepted precipitation improved the offline (i.e., not coupled to a climate model) simulated global vegetation distribution and NPP. Importantly, the version of CLM-DGVM used in this study includes an improved canopy hydrology scheme that reduces canopy interception by accounting for impacts of precipitation subgrid variability. A detailed description of the scheme is given by Wang and Wang [2007]. In this study, CLM-DGVM was configured for a ∼2.81 longitude by ∼2.81 latitude (T42 resolution) global land grid, excluding Greenland and Antarctica, and utilizes a model time step of 30 min. CLM-DGVM can run at any spatial resolution. Here, the T42 resolution was chosen to accommodate planned future studies using the simulated vegetation in a climate model running at T42 resolution.

2.2. Climate Data

[10] The driving forcing for CLM-DGVM were derived from the monthly atmospheric forcing (temperature, precipitation, specific humidity, solar radiation, and wind) from the preindustrial control (PICNTRL) and SRESA1B stabilization experiments of eight GCMs that participated in IPCC AR4. In the PICNTRL, atmospheric CO2 concentration is held at a constant value of 275 ppm, whereas in the SRESA1B experiments, it follows the SRESA1B scenario [Nakicenovic and Swart, 2000] and stabilizes at 720,ppm after 2100. The eight GCMs are (1) U.S. National Center for Atmospheric Research CCSM3.0 (hereafter referred to as CCSM), (2) U.S. Geophysical Fluid Dynamics Laboratory GFDL-2.1 (hereafter referred to as GFDL), (3) U.S. Goddard Institute for Space Studies GISS-ER (hereafter referred to as GISS), (4) UK Hadley Center for Climate Prediction and Research UKMO-HadCM3 (hereafter referred to as HadCM), (5) Germany Max Planck Institute for Meteorology ECHAM5/MPI-OM (hereafter referred to as ECHAM), (6) Japan Center for Climate System Research's MIROC3.2 (hires) (hereafter referred to as MIROC), (7) Canadian Centre for Climate Modeling and Analysis CCCma-CGCM3.1/T47 (hereafter referred to as CGCM), and (8) China Institute of Atmospheric Physics FGOALS-g1.0 (hereafter referred to as FGOALS).

[11] For each of these GCMs, PICNTRL climatology was derived based on a 30-year period of model integration from the preindustrial control experiment, and SRESA1B after-stabilization climatology based on the period 2101–2130. For CCSM and MIROC, monthly output data beyond 2100 were not available from the SRESA1B experiments. The climatologies for these two GCMs were based on the period 2071–2100. For the same GCM, the differences in climate between 2101–2130 and 2071–2100 are much smaller than the difference between PICNTRL and either of the two periods. This therefore has little qualitative impact on our results, especially on the direction of future vegetation changes. Spatial resolutions differ among the GCMs. In this study, the climatological data for each GCM were spatially interpolated to our model's T42 resolution. Daily data were obtained by linear interpolation between monthly values. We then derived diurnal cycles from the daily data using a stochastic weather generator [Foley et al., 1996].

[12] Of particular importance to vegetation growth are changes in temperature and precipitation in the future. The projected changes in mean annual precipitation and in annual average temperatures between the PICNTRL and SRESA1B climatologies derived from the various GCMs are shown in Figure 1. The differences in the predicted changes highlight the considerable uncertainty stemming from different GCM climate predictions. In the Northern Hemisphere (NH) high latitudes, all the GCMs generally predict an increase in precipitation. A strong decrease in precipitation over a large portion of northern Amazonia is seen in the HadCM projection whereas a strong increase is seen in the GISS projection for a large area in Amazonia. GISS also predicts a strong increase in precipitation in the central portions of Amazonia and Africa. While all other models predict drier conditions of varying magnitude and spatial extent in Africa and Amazonia, CCSM predicts wetter conditions in almost the whole of these regions. One commonality to all the GCM predictions is the decrease in precipitation in a portion of southwestern United States. All the GCMs predict warmer conditions over the entire globe for the future, with HadCM and MIROC predicting the strongest warming in northern Amazonia and North America, respectively. GISS projects the weakest warming in the NH middle and high latitudes, while CCSM projects the weakest over a large portion of Africa.

image

Figure 1. Projected changes (SRESA1B minus PICNTRL) (left) in annual precipitation (in mm) and (right) in annual mean temperature (in °C) by the eight GCMs.

Download figure to PowerPoint

2.3. Experimental Design

[13] Pertaining to each GCM, two simulations were carried out using CLM-DGVM initialized with bare ground: one for the PICNTRL potential vegetation (simulation PIC) and the other for the future potential vegetation under the SRESA1B emission scenario (simulation A1B). In each PIC simulation, CLM-DGVM was forced by the preindustrial climatology and run for 200 years (to a near-equilibrium state with respect to LAI). Atmospheric CO2 concentration was specified at 275 ppm. In each A1B simulation, CLM-DGVM was driven by the SRESA1B climatology and run for 200 years with atmospheric CO2 concentration specified at a constant value of 720 ppm. The differences in vegetation structure and distribution between simulations A1B and PIC depicts the impact of both CO2 changes and CO2-induced climate changes. We compare changes in the distribution of dominant vegetation type (i.e., vegetation type with the highest fractional coverage on each model grid cell), fractional coverage of various vegetation types, NPP, and carbon flux due to fire disturbance for the different GCM scenarios. Since the carbon fluxes may vary substantially from year to year, we use averages of the last 10 years of simulations for our analysis.

3. Results

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. Model and Methodology
  5. 3. Results
  6. 4. Discussion and Conclusions
  7. Acknowledgments
  8. References

3.1. Changes in Vegetation Distribution

3.1.1. Dominant Vegetation Types

[14] We aggregated the simulated PFTs into five vegetation types as follows: desert (no vegetation), sparse (fractional coverage <40%) and dense (fractional coverage > 40%) grasses (C3 arctic, C3 nonarctic and C4), deciduous trees (broadleaf tropical, broadleaf temperate and broadleaf boreal), and evergreen trees (needleleaf temperate, needleleaf boreal, broadleaf tropical, and broadleaf temperate). Figure 2 shows the simulated distributions of dominant vegetation types under the PICNTRL and SRESAIB climates, and the changes in the distributions in response to CO2 and climate changes for all the GCM scenarios applied.

image

Figure 2. Distribution of dominant vegetation types for the eight GCMs in (left) simulation PIC and (middle) simulation A1B and (right) the changes in dominant vegetation type (A1B-PIC) for the eight GCMs. In Figures 2 (left) and 2 (middle), type 0 is desert (red), type 1 is grass with fractional coverage of less than 40% (orange), type 2 is grass with fractional coverage of less than 40% (light yellow), type 3 is deciduous trees (green), and type 4 is evergreen trees (blue). In Figure 2 (right), red (−2) represents 2-grade vegetation degradation, e.g., 2 in PIC becoming 0 in A1B; orange (−1) is 1-grade degradation, light yellow (0) is no change; green (1) is 1-grade enhancement, e.g., 1 in PIC becoming 2 in A1B; blue (2) is 2-grade enhancement.

Download figure to PowerPoint

[15] In the PIC simulations (Figure 2, left), it is noticeable that for all the GCM scenarios almost no evergreen trees are identified as the dominant vegetation type in the northern middle and high latitudes (e.g., Canada and Russia) where one would expect the dominance of needleleaf evergreen trees [Bonan et al., 2002]. Instead, dominance of deciduous trees is seen in extensive areas of North America, west Europe, and Russia under the GFDL, HadCM, and ECHAM scenarios. The dominance is substantially less extensive for the other GCM scenarios. Note that the lack of dominance by needleleaf evergreen trees does not mean a complete absence. In fact, for most GCMs, some coverage of needleleaf evergreen trees is simulated. In most of the tropics, evergreen trees dominate the PIC potential vegetation for all the GCM scenarios, although of varying spatial extents. GISS produces the least extensive dominance of tropical evergreen trees (in central Africa, South America and the maritime subcontinents) whereas GFDL and HadCM produce the most extensive dominance. CLM-DGVM has been found in a previous study by Bonan and Levis [2006] to underestimate global forest cover in favor of grasses, and evergreen trees in favor of deciduous ones.

[16] In response to elevated atmospheric CO2 and warming, a poleward expansion or shift of boreal forests into northern Russia and northern Canada is evident for all the GCMs except for GISS, though of less spatial extent for HadCM. In terms of species dominance, large areas of grasslands/tundra transition to deciduous forests in these regions under the future climate. More extreme vegetation shifts (i.e., from grasses to evergreen trees) are seen in Alaska, northern Russia and northern Canada under the CGCM climate change scenario due to the fact that the CGCM-simulated temperature for the PICNTRL temperature is not warm enough to support the growth of tree life forms. In the SRESA1B climate, the temperature increase in such regions predicted by CGCM is large enough to support the growth of evergreen trees in CLM-DGVM. Distinctly, under the HadCM predicted future climate, deciduous forest is replaced by grasses in large areas of Russia and United States.

[17] Under the CCSM scenario, the dominant vegetation type in portions of West Africa changes from grasses in PIC to deciduous trees in A1B due to the strong increase in precipitation in the CCSM-predicted future climate. The other GCMs show a degradation in vegetation in portions of West Africa, characterized by the transition of grass cover dominance to desert (GISS, GFDL and FGOALS), and transition of evergreen tree dominance to deciduous ones (HadCM, MIROC, ECHAM and CGCM). Some spatially fragmented areas in southern Africa also experience shifts in the distribution of dominant vegetation types especially under the GFDL, HadCM, GISS, ECHAM, and FGOALS scenarios. These shifts mainly include transitions of the dominant vegetation type from evergreen trees to deciduous ones in some scenarios, and from deciduous trees to grasses in others, both as a degradation of vegetation type. Under the HadCM scenario, however, the most dramatic shifts are seen in South America and southern Africa, with substantial areas of evergreen forest dominance replaced with grass cover. Climate change forcing from most of the GCMs favor the degradation of dominant vegetation types in substantial parts of northern South America, even though of relatively small spatial extents under the CCSM and GISS scenarios. For most of the GCMs, portions of the dominantly evergreen Amazon forest turn deciduous under the future climate, with the HadCM scenario producing the most extreme response (with the replacement of evergreen forest by grassland). A full-scale Amazonian forest dieback has previously been reported in climate–carbon cycle projections carried out with HadCM3 coupled to an ocean carbon cycle model and dynamic global vegetation model [Cox et al., 2000]. Generally, increased water stress due to the GCM-simulated higher temperatures and/or reduced precipitation promotes the vegetation degradation in the tropics observed in our CLM-DGVM simulations.

[18] In many areas across the globe (Figure 2 right), no shifts in dominant vegetation types are shown. That is, changes of vegetation are not large enough to cause shifts of dominant vegetation types in these areas. However, such landscapes may be undergoing gradual changes in terms of the fractional coverage of the different vegetation types, as shown in the next subsection.

3.1.2. Fractional Coverage of Vegetation Types

[19] Figures 3a and 3b depict the geographic distribution of changes in fractional coverage of four categories of plant functional types (needleleaf evergreen trees (NET), broadleaf evergreen trees (BET), deciduous trees (DT), and grasses (GR)) for the different GCM scenarios. Relative to the geographic variations of shifts in dominant vegetation types (i.e., Figure 2), more widespread changes in fractional coverage of the different vegetation types across the globe are shown for all the GCM scenarios (Figures 3a and 3b). The simulations driven by GFDL, HadCM, MIROC, and CGCM predictions all show large increases in fractional coverage of NET and concomitant decreases in grass cover in portions of northern Russia, consistent with the shift from dominance by grasses to dominance by evergreen trees in these areas. Total fractional coverage of forest increases strongly in these regions (Figure 4). In the HadCM-driven simulation, increases of GR coverage at the expense of DT in large areas of southern Russia and United States are consistent with shifts from DT to GR dominance, and total forest coverage shrinks drastically (by 80% or more in some parts) (Figures 3a and 4). DT coverage decreases in portions of eastern United States and eastern China in almost all GCM scenarios, but the magnitude is not large enough to cause changes in dominant vegetation types. In parts of southwestern Europe, BET coverage expands at the expense of DT under most GCM scenarios (CCSM, GFDL, HadCM, MIROC and ECHAM), consistent with the observed shifts from DT to ET dominance in these areas. Grass cover generally shrinks in middle and high latitudes in future climate as tree cover expands (Figures 3a, 3b, and 4). However, for some GCM scenarios (e.g., GISS and FGOALS), substantial increases of grass cover are simulated, but not at the expense of tree growth. Rather, these increases are future enhancements of growth over currently existing open grasslands.

image

Figure 3a. Geographic distribution of changes in fractional coverage (A1B-PIC), as percent of vegetated portion of grid cell, of four categories of plant functional types: needleleaf evergreen trees (NET), broadleaf evergreen trees (BET), deciduous trees (DT), and grasses (GR) for four of the eight GCMs.

Download figure to PowerPoint

image

Figure 3b. Same as Figure 3a but for the other four GCMs.

Download figure to PowerPoint

image

Figure 4. Geographic distribution of changes in fractional coverage (A1B-PIC), as percent of vegetated portion of grid cell, of forest (all woody plant functional types) for the eight GCMs.

Download figure to PowerPoint

[20] The eight GCMs generally show declines in coverage of ET in the tropics especially in South America, and West and southern Africa, most markedly under the HadCM scenario, owing to increased water stress as a result of warming and/or reduced precipitation. Such vegetation degradation is much more extensive in space than the changes in dominant vegetation type. That is, DT has not overtaken ET yet as the dominant vegetation type in many areas, thus revealing a trend toward evergreen tropical forests turning deciduous in portions of the tropics. Decreases in total fractional coverage of forest are seen in these portions, especially under the HadCM, GFDL, MIROC, and CGCM scenarios. However, under the CCSM scenario, in some parts of West Africa, DT coverage increases at the expense of grasses leading to an increase in total forest coverage, while the desert area in North Africa is reduced in the future climate as grass cover expands into some portions that are initially desert. This results from the strong increase in precipitation and less warming in that region in the CCSM predictions.

[21] Overall, the spatial extents of fractional coverage changes are much larger than those of the dominant vegetation type changes, as many of these changes in fractional coverage are not enough to change the dominant vegetation type on the landscape. Not surprisingly, large magnitudes of fractional coverage changes tend to coincide with the dominant vegetation type changes.

[22] We compare our results, in terms of broad patterns of changes from nonforest to forest and vice versa, with results from previous predictions by Scholze et al. [2006] using LPJ-DGVM driven with the IPCC AR4 scenarios. Under the CCSM, ECHAM, GFDL, GISS, and HadCM scenarios, our results show a transition of grasses to trees in portions of the high latitudes (Figures 2 and 4), which is broadly consistent with the results of Scholze et al. [2006]. However, under the GFDL scenario, we do not observe the widespread change of forest to nonforest in the midlatitudes shown in the Scholze et al. [2006] simulation. Also, our results do not support the conversion of forest to nonforest in eastern Amazonia under the ECHAM scenario found by Scholze et al. [2006], although we observe degradation from evergreen to deciduous trees. There is consistency in the simulated replacement of forest with grasses in Amazonia, but not in southern Africa, under the HadCM scenario.

3.2. Changes in NPP

[23] Figure 5 shows spatial variations of annual NPP changes under different GCM climate change scenarios. All the GCM scenarios, except for HadCM, broadly produce enhancements in NPP in most continental areas across the globe. However, in large areas of northern China, Australia and the Middle East, NPP changes are negligible. NPP in portions of southwest United States decreases substantially in the HadCM and FGOALS scenarios. Declines in NPP are also seen in portions of West and southern Africa and northern South America for all the GCMs except for CCSM. Enhanced photosynthesis under the higher CO2 concentration, and increases in temperature and the resulting lengthening of the growing season all promote the NPP increases in the northern middle and high latitudes. In the water-limiting ecosystems, reduced precipitation and/or warming-induced acceleration of evapotranspiration lead to decreases in NPP as particularly seen with the HadCM scenario. The HadCM scenario leads to a strong reduction in NPP in Amazonia and southern Africa. Except for the HadCM, all other GCM scenarios result in strong increase in NPP in central Africa and most of Amazonia despite the degradation of vegetation in some portions of these areas. Decreases of NPP in these models can be found in southwestern United States, North Africa, and some areas in South America.

image

Figure 5. Simulated changes in NPP (A1B-PIC) (in kg C m−2 of vegetated portion of grid cell) for the eight GCMs. Typical values of simulated preindustrial (PIC) annual NPP for typical grid cells in various regions are southern Africa, 1.0 kg m−2; eastern United States, 0.6 kg m−2; western Europe, 0.6 kg m−2; northeastern Middle East, 0.1 kg m−2; eastern China, 0.7 kg m−2; and northern South America, 1.1 kg m−2.

Download figure to PowerPoint

[24] We explored the roles of vegetation dynamics and climate changes (together with CO2 enrichment) in causing the NPP changes by performing one additional experiment A1B_PICVEG for each GCM scenario. In A1B_PICVEG, CLM-DGVM was forced with SRESA1B climate with vegetation prescribed based on the potential vegetation from PIC. Thus A1B_PICVEG and A1B share the same climate forcing as well as CO2 concentration, but differ in vegetation distribution. The difference in NPP between simulations A1B_PICVEG and A1B therefore reveals the impact of vegetation dynamics on the greenhouse-induced NPP changes, whereas A1B_PICVEG minus PIC isolates the impact of climate plus CO2 (not shown). Vegetation dynamics (including shifts in PFTs and changes in fractional coverage) act to contribute positively to NPP in portions of the middle and high latitudes owing to the substantial spread of forests at the expense of the less productive grasses with the CO2-induced climate warming.

[25] Apart from HadCM, all the GCMs generally agree on vegetation dynamics tending to reduce NPP in large areas in the tropics. This is attributable to the widespread transition from broadleaf evergreen trees to the less productive deciduous trees and grasses in the future climate. In the HadCM scenario, however, vegetation dynamics (in the form of replacement of evergreen trees with deciduous trees and grasses), act to contribute positively toward NPP in future climate since the original evergreen trees are unable to survive (therefore have negative NPP) in future climate. NPP of the new vegetation types (deciduous trees and grasses) are higher than the evergreen trees in future climate, although they are less productive than the evergreen trees in the preindustrial climate.

[26] On the other hand, climate change and CO2 enrichment together, enhances NPP all over the globe in all scenarios except HadCM. This at its first glance may seem to contradict the simulated vegetation degradation in the tropics shown in Figures 2, 3a and 3b, but it does not. Vegetation degradation occurs because the increase of NPP for the original vegetation is smaller than the NPP increase for the new vegetation type, leading to changes in vegetation competition. In the HadCM scenario, the negative impacts of the strong warming and drying in the Amazon and Africa dominates over the potential positive impacts of warming and CO2 enrichment on NPP.

[27] Consistent with the widespread enhancement of NPP, vegetation density (as reflected by leaf area index in the peak growing season) will be higher in the future (Figure 6). This increase of LAI, similar to the NPP increase, occurs despite the predicted degradation of vegetation type in the tropics. In other words, with the tropics, the ecosystem is expected to produce more, but the product will be of a lower quality (if one ranks the ecosystem service from evergreen forest higher than deciduous forest and deciduous forest higher than grassland).

image

Figure 6. Simulated changes in LAI of potential natural vegetation (A1B-PIC) (in m2 m−2) for the eight GCMs.

Download figure to PowerPoint

3.3. Changes in Disturbance Due to Fire

[28] Simulated carbon flux to the atmosphere due to fire generally increases across the globe but more prominently in the tropics (Figure 7). The HadCM scenario, however, produces a decrease of fire emission in eastern Amazonia, while GFDL and CGCM produce decreases in a part of West Africa. The increase in carbon release to the atmosphere through natural fires across the globe is associated with increased fire frequency due to the lower moisture level in future climate and increased fuel load due to enhanced NPP. The strongest increases are in the tropics. However, the degradation of vegetation is so severe that it reduces the fuel loading, leading to decrease of fire carbon release in eastern Amazonia, and part of West Africa under the HadCM scenario, and GFDL and CGCM scenarios, respectively.

image

Figure 7. Simulated changes in carbon flux to the atmosphere due to fire (A1B-PIC) (in kg C m−2 of vegetated portion of grid cell) for the eight GCMs. Typical values of simulated preindustrial (PIC) fire emission for typical grid cells in various regions are southern Africa, 0.15 kg m−2; eastern United States, 0.03 kg m−2; western Europe, 0.04 kg m−2; northeastern Middle East, 0.01 kg m−2; eastern China, 0.01 kg m−2; and northern South America, 0.12 kg m−2.

Download figure to PowerPoint

4. Discussion and Conclusions

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. Model and Methodology
  5. 3. Results
  6. 4. Discussion and Conclusions
  7. Acknowledgments
  8. References

[29] The ultimate objective was to determine if simulations based on different GCMs' climate predictions could support any general statements on the sign and magnitude of future vegetation changes in various regions across the globe. This was achieved using a land surface model coupled to a dynamic vegetation model (CLM-DGVM).

[30] Despite the considerable differences in the simulated responses of the biosphere as already highlighted, the results of this study support a generalization that the terrestrial biosphere may respond to greenhouse gas increase and its induced climate changes with a considerable poleward shift or spread of temperate and boreal forests in the Northern Hemisphere high latitudes, and a substantial degradation of vegetation cover in the tropics, especially in portions of West and southern Africa and South America. A surprising consensus is found across different GCM scenarios for an enhancement of NPP and carbon fluxes to the atmosphere due to fire across the globe in future climate.

[31] The baseline (i.e., PICNTRL) temperature and moisture conditions, elevated atmospheric CO2 concentration, as well as the spatial patterns of precipitation and temperature changes in each GCM scenario, are the main factors that influence the trends in simulated vegetation function, structure and distribution. The warmer temperatures shown in all the GCM scenarios, coupled with increased atmospheric CO2, had a strong positive impact on vegetation growth especially in regions initially too cold under the PICNTRL climate. As a result, the poleward shift or expansion of temperate and boreal forests stands out as a robust feature across all the GCM scenarios, consistent with findings from other previous studies [e.g., Cramer et al., 2001; Joos et al., 2001; Gerber et al., 2004; Lucht et al., 2006]. In the tropics, moisture condition changes (as a result of evapotranspiration rate increase or precipitation decrease or both) reduce the competitiveness of evergreen trees relative to drought deciduous trees and grasses, leading to widespread vegetation degradation despite the NPP enhancement. Note that the term “degradation” is used in this paper to specifically refer to change of vegetation type to a lower grade, which may or may not be accompanied by vegetation density changes toward the same direction.

[32] As a step toward examining the robustness of our results, we compared our results with previous predictions using LPJ-DGVM driven with IPCC AR4 climate scenarios [Scholze et al., 2006]. Although our results are consistent with theirs in some regions and for some GCM scenarios, they conflict in others. While the differences may be, in part, due to methodological differences in the two studies, it seems that simulated biosphere responses may be considerably model-dependent. This provides strong motivation for future intercomparison studies using different DGVMs.

[33] The NPP enhancement found in this study, however, may seem to contradict some findings based on a different DGVM [Higgins and Vellinga, 2004]. The differences may be related to the fact that the Higgins and Vellinga [2004] study did not (while ours does) include the direct physiological CO2 effect on plant growth that tends to cause strong increase in NPP and enhancement of vegetation growth. The CO2 fertilization response of vegetation, however, is still disputable. The physiological effect of CO2 on plant function in CLM-DGVM is parameterized in a manner consistent with observations that have been carried out in controlled environmental settings. For example, guided by observations, photosynthesis in C4 plants is set to saturate at an ambient CO2 of about 400 ppm [Oleson et al., 2004]. The response of plants to CO2 enrichment in natural ecosystems may differ considerably from that observed in controlled settings [Neilson and Drapek, 1998], and this may set a different limitation on the growth enhancement response of vegetation to elevated CO2 than that currently modeled by CLM-DGVM. The CO2 fertilization of NPP in our model thus represents an important uncertainty in our results, but without the physiological effects of elevated CO2, the simulated degradation of vegetation in the tropics is likely to be more since higher CO2 tends to reduce water stress by enhancing water use efficiency. The differences between our simulated NPP patterns and those of Higgins and Vellinga may also be partly due to differences in parameterization of the vegetation and land surface processes.

[34] CLM-DGVM, as in many other DGVMs, assumes that plant species can move to other locations without barriers in order to survive if the climate in their current locations becomes unsuitable. That is, seeds are assumed to be always available. However, dispersal and human land use change may limit the migrational capabilities of species, thereby invalidating modeled ecosystem responses [Higgins and Harte, 2006].

[35] The results presented in this paper are not predictions of the future state of the terrestrial biosphere per se, but shed light on a range of possible changes in natural vegetation structure and function in response to atmospheric CO2 enrichment and the resulting climate change. For instance, our results suggest that for vegetation cover in the tropics, especially in portions of West and southern Africa and South America, the response ranges from forest dieback as with the HadCM scenario to less or no degradation or even enhancements with the CCSM scenario. Such uncertainty needs to be considered in decision making on mitigating potential climate impacts and reducing vulnerability.

[36] It is important to recognize that while the simulations for different GCM scenarios generally agree on NPP increases under future climate for most parts of the globe, substantial shifts in dominant vegetation types revealed in our simulations will have important hydrological, ecological and socioeconomic implications. For example, changes in vegetation structure will influence runoff [Cramer et al., 2001], and therefore the amount and seasonal pattern of streamflow; vegetation changes will influence wildlife inhabitants, triggering widespread ecosystem response; transition of a grass-dominated land cover to deciduous or evergreen tree dominance in the middle or high latitudes may affect the aesthetic value the inhabitants placed on their original environment; the replacement of current forests by grasslands will influence the supply of wood and pharmaceutical products; the increased fire frequency across the globe in future climate imply higher risks of destruction of lives and property through natural fires [Scholze et al., 2006]. Last, but not the least, changes in vegetation distribution and growth can cause feedbacks to climate through both biogeophysical and biogeochemical pathways [Bonan et al., 1992; Levis et al., 1999; Betts, 2000; Cox et al., 2000; Wang and Eltahir, 2002].

[37] Given that the primary objective of this study was to illustrate the direction of future vegetation changes, it sufficed to run CLM-DGVM offline on average preindustrial and SRESA1B climates until its vegetative equilibrium. A major advancement of this research will be to couple the same dynamic land-vegetation model (e.g., CLM-DGVM) to the different GCMs, perform long–term transient PICNTRL and SRESA1B climate simulations, and then evaluate vegetation changes based on the coupled model simulations. This will enable the exploration of biosphere responses that include the feedbacks between vegetation and climate [Foley et al., 2000]. A further step will involve incorporating agricultural and urban land cover changes into the simulations, for example based on the Integrated Model to Assess the Global Environment (IMAGE) 2.2 projections of global cropland cover changes under the IPCC SRES scenarios [Alcamo et al., 1996; Wang et al., 2006]. Coupled biosphere-atmosphere simulations that use detailed trajectories of atmospheric CO2 concentration changes and human land changes into the future would provide more realistic predictions of the future state of the biosphere.

Acknowledgments

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. Model and Methodology
  5. 3. Results
  6. 4. Discussion and Conclusions
  7. Acknowledgments
  8. References

[38] The authors would like to thank Paul Higgins and an anonymous reviewer for their helpful comments on an earlier version of this manuscript. We acknowledge the international modeling groups for providing their data for analysis, the Program for Climate Model Diagnosis and Intercomparison (PCMDI) for collecting and archiving the model data, the JSC/CLIVER Working group on Coupled Modeling (WGCM) and their Coupled Model Intercomparison Project (CMIP) and Climate Simulation Panel for organizing the model data analysis activity, and the IPCC WG1 TSU for technical support. The IPCC Data Archive at Lawrence Livermore National Laboratory is supported by the Office of Science, U.S. Department of Energy.

References

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. Model and Methodology
  5. 3. Results
  6. 4. Discussion and Conclusions
  7. Acknowledgments
  8. References