SEARCH

SEARCH BY CITATION

Keywords:

  • shrub;
  • dynamic global vegetation model;
  • arid and semiarid regions

Abstract

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. Observational Characteristics of Temperate Shrubs
  5. 3. Brief Description of CLM-DGVM
  6. 4. Shrub Submodel for the CLM-DGVM
  7. 5. Numerical Simulations
  8. 6. Conclusions and Further Discussions
  9. Acknowledgments
  10. References
  11. Supporting Information

[1] Arid and semiarid regions represent a large fraction of global land, but most of the existing dynamic global vegetation models (DGVMs) do not include shrubs or do not effectively distinguish shrubs from grasses, and hence cannot realistically reproduce the ecosystem formation and variability there. A shrub submodel is developed here for the Community Land Model–DGVM (CLM-DGVM), and the major revisions include (1) explicit consideration of shrubs' drought tolerance in the photosynthesis computation; (2) use of appropriate phenology type and morphology parameters for shrubs; (3) consistent treatment of fractional vegetation coverage; (4) development of tree/grass/shrub hierarchy for light competition; and (5) improvement of the allocation scheme to avoid unrealistic behaviors. Preliminary global offline CLM-DGVM simulations for 400 years show that, with the shrub submodel, the simulated global distribution of temperate shrubs agrees with Moderate Resolution Imaging Spectroradiometer (MODIS) data. The simulated shrub coverage reaches its peak around annual precipitation (Pann) of 300 mm, the grass coverage reaches its peak over a broad range of Pann (from 400 to 1100 mm), and the tree coverage reaches its peak for Pann = 1500 mm or higher, all in good agreement with MODIS data.

1. Introduction

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. Observational Characteristics of Temperate Shrubs
  5. 3. Brief Description of CLM-DGVM
  6. 4. Shrub Submodel for the CLM-DGVM
  7. 5. Numerical Simulations
  8. 6. Conclusions and Further Discussions
  9. Acknowledgments
  10. References
  11. Supporting Information

[2] Vegetation covers most of the global land surface. It strongly affects the land-atmosphere exchanges of energy, momentum and materials, and thus influences climatic variables such as temperature and precipitation. For these reasons, understanding and modeling the formation and variability of terrestrial ecosystems have been emphasized in recent years by various national and international programs [e.g., iLEAPS, 2005].

[3] These studies can be divided into different categories, each with some strengths and weaknesses. Conceptual and analytically tractable ecological models are useful in capturing the essential physical and ecological mechanisms of specific events such as ecosystem transition [e.g., Brovkin et al., 1998; Claussen et al., 1999; Zeng et al., 2005] and species competition [Pacala and Tilman, 1994; Tilman, 2004] in a large area over a long period. However, they are essentially equilibrium models and hence neglect the transient behaviors and details of vegetation-soil-atmosphere interactions. Ecological models for individual plants; e.g., the forest gap model [Acevedo et al., 1995; Moorcroft et al., 2001], can simulate the dynamics of an individual plant (e.g., tree) with high spatial and temporal resolutions, but they are difficult to apply to the global simulations over a long period. The third approach, dynamic global vegetation models (DGVMs), a group of intermediate complexity models, are designed for coupling with a land surface model to represent both the fast-scale vegetation-soil-atmosphere interactions and the slow-scale ecosystem dynamics. Therefore, DGVMs are more suitable for the study of global ecosystem-climate interactions. For instance, they can reproduce the regimes of current major plant functional types and simulate the evolution of ecosystems in the near future under the changing climate [Kucharik et al., 2000; Cramer et al., 2001; Bonan et al., 2003; Sitch et al., 2003], including the possible collapse of the Amazon forest in response to the climate change associated with the doubling of CO2 concentration in the next 50 years [Cox et al., 2000; Huntingford et al., 2000].

[4] This paper focuses on the ecosystem formation and its change in the arid and semiarid regions that cover a large fraction of global land (Figure 1). The typical ecosystems over these regions are desert, steppe, grassland, shrubland, woodland, and savanna [Shmida, 1985], where shrubs, dry grasses and other drought-tolerant species are the dominant vegetation types. Ecosystems over these regions are vulnerable and sensitive to the changes in climate and environment because precipitation is sparse and irregular. For example, shrub expansions in the North American semiarid grassland have been observed on decadal to millennial timescales [Archer et al., 1995; Betancourt, 1996; van Auken, 2000], especially in the last 150 years [Bahre, 1995]. In Inner Mongolia of China, large areas of the former steppe have been degraded into sandy dune and desert. These increasing degradations of the terrestrial ecosystems and desertification have resulted in significant economical losses as well. The causes of such dramatic ecosystem changes are still under debate, and they may include overgrazing, cultivation, wildfire, or changes of precipitation (either annual amount or seasonal pattern) and temperature. Revegetation in the desert grassland, e.g., reversing the shrubland to grassland over large areas, is very difficult (if not impossible), and requires a long-term effort [Roundy and Biedenbender, 1995].

image

Figure 1. Global land cover type distribution based on the 0.05° Moderate Resolution Imaging Spectroradiometer (MODIS) data. Black and blue solid lines indicate the contours for annual precipitation of 300 mm and 800 mm, respectively, based on the Intergovernmental Panel on Climate Change 1981–1990 averaged 0.5° precipitation data.

Download figure to PowerPoint

[5] Most of the existing DGVMs do not include shrubs [Brovkin et al., 1997; Foley et al., 1996; Haxeltine and Prentice, 1996; Sitch, 2000; Levis et al., 2004], or do not effectively distinguish shrubs from grasses or trees [Woodward et al., 1998; Hickler et al., 2006]. Only a few models [Kucharik et al., 2000; Cox, 2001] have specific shrub types.

[6] The purpose of this paper is to develop a shrub submodel for the Community Land Model–Dynamic Global Vegetation Model (CLM-DGVM) so that shrubs can grow properly over global arid and semiarid regions. The overall approach is also expected to be relevant to other DGVMs without shrubs.

[7] Section 2 provides observational characteristics of temperate shrubs which are the basis for our shrub submodel development. Section 3 briefly discusses the CLM-DGVM, while section 4 discusses the new shrub submodel. Section 5 demonstrates the impact of the shrub submodel on the CLM-DGVM simulations. Section 6 provides summary and further discussions.

2. Observational Characteristics of Temperate Shrubs

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. Observational Characteristics of Temperate Shrubs
  5. 3. Brief Description of CLM-DGVM
  6. 4. Shrub Submodel for the CLM-DGVM
  7. 5. Numerical Simulations
  8. 6. Conclusions and Further Discussions
  9. Acknowledgments
  10. References
  11. Supporting Information

[8] Shrub is not strictly a botanical category of plant. In most definitions [e.g., Allaby, 1994], shrub is a woody plant that has several stems branching below or above the ground level with none of them dominant. Shrub is usually less than 3 m tall. On the contrary, trees are generally defined as woody plants more than 6 m tall, having a dominant stem, or trunk, and a definite crown shape. The intermediate plants (3–6 m tall) between shrubs and trees are called arborescences (treelike shrubs). In practice, a large number of plants can be either shrubs or trees, depending on the growing conditions they experience.

[9] Shrubs have a large family with many subtypes and are distributed widely over the world, especially in arid and semiarid regions. For example, there are more than 5000 species of shrubs or subshrubs (i.e., short shrubs) within the United States, with California and Texas each having more than 1000 species [Francis, 2000]. In this study, we will focus on the general characteristics of temperate shrubs over global arid and semiarid regions where shrubs and grasses are the dominant plants and trees are rare. Figure 1 shows the map of global land cover types based on the Moderate Resolution Imaging Spectroradiometer (MODIS) satellite data [Friedl et al., 2002] overlapped with contours of annual precipitation of 300 mm and 800 mm based on the Intergovernmental Panel on Climate Change precipitation data (available at http://www.ipcc-data.org/download_data/obs/cpre6190.zip). It is clear from Figure 1 that most of the arid regions (with annual precipitation less than 300 mm) are either barren or dominated by shrubs.

[10] In arid and semiarid regions, summer is usually very hot, and annual precipitation is limited (less than 300 mm) and is mainly concentrated in specific seasons. In order to adapt to the environment, shrubs have developed two traits (i.e., high tolerance to drought and high temperature) through several strategies. First, leaves respond quickly to rain (e.g., more quickly than tree leaves). Shrubs in arid habitats usually have relatively small leaves which are less costly to grow compared with large leaves [Smith et al., 1997]. After major rainfall events, shrubs grow new leaves [Mooney, 1981], or the existing leaves quickly turn to fully open status and expand, so that they can efficiently use the rare and irregular water. In the hot and dry season, leaves are withered or dropped, or the stomata are less open to reduce transpiration [Smith et al., 1997]. Therefore shrubs can be either evergreen or raingreen (i.e., drought-deciduous). Second, shrubs have a relatively high efficiency in taking soil water by roots. Shrubs usually have roots penetrating in soil much deeper and wider than grasses, and hence can use soil water unavailable to grasses [Burgess, 1995]. Shrubs can also maintain a relatively high level of photosynthesis even under drought condition [Wilson, 1998]. Third, shrubs expand more horizontally than vertically, so that a larger amount of the net primary production (NPP) is distributed to the productive structures (leaves and roots) rather than the woody supporting structures (stems). Finally, shrubs have a relatively low leaf area index (LAI) (usually around or even less than 1), and their roots can extract soil water from a large area (comparable to or even much larger than the crown area) to support the productive processes [Kummerow, 1981]. However, such strategies of adaptation are less competitive in the wetter areas where species other than shrubs become dominant (see Figure 1).

[11] Because of the lower LAI, the annual productivity averaged over the shrub-covered ground is small. The maximum photosynthesis levels (i.e., under the optimal temperature and soil moisture) of shrubs are also smaller than grasses and trees (e.g., as recognized by Oleson et al. [2004]). Furthermore, compared with trees, shrubs have a higher percentage of leaves and roots over its total dry mass, but the portion of dry mass distributed to leaves is mostly lost every year. All these factors result in the relatively slow growth rate of shrubs.

[12] In the following we introduce these features, as summarized in Table 1, to a dynamic global vegetation model, CLM-DGVM, in order to properly represent shrubs. While some of the details discussed below are relevant to the CLM-DGVM only, the overall approach is expected to be relevant to other DGVMs that do not explicitly have a shrub component.

Table 1. Major Features of Shrubs, Grasses, and Treesa
 TreesShrubsGrasses
  • a

    LAI: leaf area index.

Growth ratemediumlowhigh
Heighttallmediumlow
LAIhighlowmedium
Drought tolerancelowhighmedium
Leaves' response to rain eventslow/mediumquickquick

3. Brief Description of CLM-DGVM

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. Observational Characteristics of Temperate Shrubs
  5. 3. Brief Description of CLM-DGVM
  6. 4. Shrub Submodel for the CLM-DGVM
  7. 5. Numerical Simulations
  8. 6. Conclusions and Further Discussions
  9. Acknowledgments
  10. References
  11. Supporting Information

[13] The CLM-DGVM represents the Community Land Model 3 (CLM 3) coupled with a dynamic global vegetation model (DGVM). The CLM and DGVM have been well documented by Oleson et al. [2004] and Levis et al. [2004], respectively. Here we give a brief description of the model with a main focus on the vegetation dynamics part that is related to our revisions and improvements. While the CLM can be run without DGVM, DGVM has to be run with a land surface model (CLM).

[14] In CLM, the fractional vegetation coverage is prescribed and hence does not have interannual variability. In CLM-DGVM, however, the fractional coverage of each plant functional type (PFT) in a grid cell is determined by equations of vegetation dynamics, as driven by climatic conditions and the annual carbon balance of the plant type. If a PFT is present in a grid cell, the related biogeochemical, biophysical, and biogeographic processes are calculated with different time steps. The fastest processes, such as canopy photosynthesis, growth and maintenance respiration, are calculated at each time step of the model (which is defined by a user and usually varies from 20 min to 2 h). The intermediate-scale processes, e.g., plant phenology, are calculated daily. The slowest processes are calculated once a year, and the major processes are turnover of living tissues (leaf, root, and sapwood), allocation of annual NPP to living tissues, competition among different PFTs for fractional coverage, and the establishment and survival of PFTs according to the climatic conditions. After these slowest processes are computed, the presence of PFTs in a grid cell, their fractional coverage, as well as their morphological values such as LAI, stem area index, and vegetation height, are updated and used in the fast and intermediate-scale processes in the next year.

[15] The standard simulation of CLM-DGVM starts with a completely bare soil surface. The establishment of a specific PFT in a grid cell begins at the end of the first year. An established PFT can survive only if it has a positive NPP (photosynthesis minus respiration) and can grow and expand if its NPP is larger than its loss of biomass due to turnover and other processes. PFTs in a grid cell usually reach an equilibrium state in a few hundred years.

[16] The CLM includes 16 PFTs (eight types of trees, three types of shrubs, three types of grasses, and two types of crops) as well as the additional type of bare ground. Each soil column may have up to four PFTs with different fractional coverages [Oleson et al., 2004]. In CLM-DGVM, the list of PFTs is shortened to 10 by not including shrubs and crops and by merging two types of boreal deciduous trees into one, but each column can have up to 10 PFTs [Levis et al., 2004].

[17] The CLM-DGVM is able to simulate the global biogeography, but it underestimates global tree coverages and overestimates grass coverages [Levis et al., 2004; Bonan and Levis, 2006]. Furthermore, because CLM-DGVM does not include shrubs, it is not appropriate for arid and semiarid regions. In the next section, we will modify the existing CLM-DGVM processes so that shrubs can grow properly in these regions.

4. Shrub Submodel for the CLM-DGVM

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. Observational Characteristics of Temperate Shrubs
  5. 3. Brief Description of CLM-DGVM
  6. 4. Shrub Submodel for the CLM-DGVM
  7. 5. Numerical Simulations
  8. 6. Conclusions and Further Discussions
  9. Acknowledgments
  10. References
  11. Supporting Information

[18] In order to grow temperate shrubs, most of the CLM-DGVM components need to be revised because of their complicated internal connections and mutual interactions. Our overall strategy is to make as few revisions as possible.

[19] Recognizing the importance of shrubs and for the convenience of future development of a shrub submodel, the original DGVM developers [Levis et al., 2004] have provided DGVM-related parameters for shrubs as place holders in the code. These shrub parameters are simply copied from the tree PFTs with similar physiological and morphological traits. For instance, the parameters for broadleaf temperate deciduous shrubs are set to the same as broadleaf temperate deciduous trees. There is a switch to turn off shrubs in DGVM so that shrubs do not appear in the default DGVM simulations.

[20] As a sensitivity test, we have turned on the switch, and run the CLM-DGVM simulation (see section 5 for model setup) using the default shrub parameters. It is found that shrubs may establish for a period of time in some grid cells but they cannot survive (figures not shown).

4.1. Photosynthesis and Respiration

[21] The photosynthesis scheme in CLM3 comes from Bonan [1996]. In contrast to the consideration of solar absorption by both sunlit and shaded leaves given by Bonan [1996], however, only the solar absorption by sunlit leaves is considered in CLM3 [Oleson et al., 2004]. Our sensitivity tests show that, using the photosynthesis scheme in CLM3 or Bonan [1996], the annual photosynthesis level of the temperate shrub in semiarid regions is smaller than the maintenance respiration of leaves and roots in the default CLM-DGVM (e.g., Figure 2). Therefore, shrubs cannot survive, let alone grow. In order to grow shrubs, either the photosynthesis level should be increased, or the respiration level decreased. Because the photosynthesis from both sunlit and shaded leaves is considered by Bonan [1996], we have implemented the Bonan scheme as the starting point. Further discussion on this issue is provided by Thornton and Zimmermann [2007].

image

Figure 2. Monthly averaged photosynthesis (FPSN) and leaf/root maintenance respiration (FRMF/FRMR) of temperate shrubs simulated by the default Community Land Model–DGVM (CLM-DGVM) with solar absorption by sunlit and shaded leaves at the grid cell centered at 114.4°W, 31.4°N.

Download figure to PowerPoint

[22] Because plant photosynthesis is calculated within CLM, independent of the DGVM modules, it is possible to further evaluate it in CLM with prescribed vegetation coverages. We have done 20-year CLM3 simulations using 7 × 6 T42 (2.8 deg by 2.8 deg horizontal resolution) grid cells in the southwestern United States in which trees, grasses, and shrubs exist. Figure 3a shows the monthly averaged photosynthesis in the last year for different PFTs over the whole area. The photosynthesis level of shrubs is smaller than that of grasses and two types of trees by at least an order of magnitude. In contrast, our analysis of the MODIS land cover data [Friedl et al., 2002] and photosynthesis and net primary production data [Zhao et al., 2005] shows that, over roughly the same area, the photosynthesis level of shrubs is comparable to that of grasses, and is about 1/3 to 1/2 of the level of trees (Figure 3b). Evidently, CLM substantially underestimates the photosynthesis for shrubs in arid and semiarid regions using the default shrub parameters.

image

Figure 3. Area averaged monthly photosynthesis of C3 nonarctic grass, temperate shrub, broadleaf deciduous temperate tree (BDT), and needleleaf evergreen temperate tree (NET) (a) calculated from offline CLM simulation over the southwestern United States (109∼120°W, 33°∼43°N, covered by 7 × 6 grid cells) with prescribed vegetation coverage for each plant functional type (PFT); and (b) based on the MODIS data over roughly the same area.

Download figure to PowerPoint

[23] In CLM, leaf photosynthesis A is calculated as the minimum value of three terms; i.e., the RuBP carboxylase (Rubisco) limited rate of carboxylation Wc, the maximum rate of carboxylation allowed by the capacity to regenerate RuBP Wj, and export limited rate of carboxylation We [Oleson et al., 2004]. In arid and semiarid regions where sunlight is sufficient and soil water supply is limited, Wj is almost always the largest among these three terms, hence A is mostly determined by Wc and We. Both Wc and We are related to the “maximum rate of carboxylation” Vmax that depends on temperature and soil water [see Oleson et al., 2004, equation 8.8]:

  • equation image

where αvmax = 2.4, Vmax25 is the prescribed value at 25°C, Tv is the leaf temperature, and βt is the soil moisture limitation function that is independent of PFT.

[24] On the basis of the discussion of the shrubs' drought tolerance in section 2, shrubs should have a relatively high efficiency of taking soil water by roots, and should be able to maintain a relatively high level of photosynthesis even under drought condition. This implies that the impact of βt on Vmax should be PFT dependent. Therefore, we rewrite equation (1) as

  • equation image

that is, we replace the original term βt by a PFT-dependent function g(βt). Because Vmax25shrub is much smaller than Vmax25grass (in CLM Vmax25shrub = 17 and Vmax25C3grass = 43), the competition between shrubs and grasses in arid and semiarid regions implies that g (βt)∣shrub should be much larger than g (βt)∣grass when βt is small.

[25] There are many methods to construct g(βt). The original scheme in CLM corresponds to the special case of g(βt) = βt (Figure 4, curve 1). Generally, the curve of g(βt) of a PFT can be higher (e.g., curves 2 and 3) or lower (e.g., curve 4) than curve 1. For simplicity, g(βt) for grasses and trees keeps unchanged (curve 1), but curve 2 is used for shrubs:

  • equation image

Photosynthesis of shrubs based on (2) and (3) can still maintain at a relatively high level when βt is small, and becomes saturated as βt > βt0. Thus, shrubs become drought-tolerant in the revised model.

image

Figure 4. Example of possible functional curves of g(βt) as used in equation (2). Curve 1 is used for grasses and trees (as in the default CLM), while curve 2 (i.e., equation (3)) is applied to shrubs.

Download figure to PowerPoint

[26] As mentioned earlier, the two-leaf canopy scheme [Bonan, 1996] that considers the contribution to photosynthesis by both sunlit and shaded leaves is more realistic. However, our sensitivity tests demonstrate that its implementation in CLM-DGVM would result in unreasonably large LAI in the tropical forest regions (e.g., the Amazon), because another important process; that is, the control of the vertical distribution of Vmax within canopy by nitrogen concentration [Thornton and Zimmermann, 2007] is not considered in CLM. Therefore, Vmax in equation (2) has been replaced by the vertical integral of the Vmax profile in the canopy from Dickinson et al. [2002].

4.2. Phenology and Establishment

[27] Phenology type affects the survival and growth of vegetation in the model, because it determines the strategy for plants to grow and drop leaves. In CLM-DGVM, woody vegetation has three phenology types: evergreen, summergreen, and raingreen. The evergreen plant does not drop leaves and keeps constant LAI within a year. The summergreen plant grows leaves when the accumulated growing degree days are higher than a critical value and drops leaves in the cold season. The raingreen plant grows and drops leaves depending on its average level of photosynthesis and leaf maintenance respiration. In arid and semiarid regions, the growth of temperate shrub is mainly limited by soil water availability rather than air temperature. Moreover, spring rainfall (in the southwestern United States) and winter-spring rainfall (in the Mediterranean climate region) are very important water resources for the shrub growth there. This cannot be reasonably taken into account by the summergreen type of phenology. Therefore, for simplicity, we set temperate shrub to the existing raingreen type in the CLM-DGVM so that shrubs can maintain a relatively high LAI in the wet season to obtain high photosynthesis, and keep a relatively low LAI in the dry and hot season to avoid high respiration. Furthermore, in order to reflect the heat tolerance of shrubs, the warmest minimum monthly air temperature for establishment is set to “no limit”, so that shrub can grow in the very hot regions such as the Middle East and the southwestern Australia.

4.3. Plant Morphology and the Allocation of Annual NPP

[28] Plant morphology determines the allocation of yearly NPP to leaves, roots, and stems. In CLM-DGVM all woody PFTs share the same morphological parameters. Although this might be acceptable among tree PFTs as the first-order approximation, the morphological differences between shrubs and trees are large and cannot be ignored. Compared with trees, shrubs are shorter with a lower LAI and possess a larger ratio of crown area over height. For example, in Arizona, the typical shrub is usually 1∼2 m tall with LAI usually less than 1, and has numerous thin branches with the crown area about 1∼3 m2. Therefore, a new set of parameters for shrub morphology, as provided in Table 2, is used in the revised CLM-DGVM.

Table 2. Parameters for Shrub and Tree Morphologya
Parameterkla:sa (m2 m−2)kallom1kallom2CAmax (m2)
  • a

    The proportionality coefficient between sapwood area and leaf area is denoted by kla:sa, CAmax denotes the maximum crown area, and kallom1 and kallom2 are used to link canopy height and crown area, respectively, to stem diameter, by Levis et al. [2004].

Shrub400025085
Tree80001004015

[29] Furthermore, our sensitivity tests show that the default iteration scheme needed in the allocation computation of CLM-DGVM could sometimes result in an unrealistic solution (e.g., negative heartwood biomass), especially for the newly introduced shrub PFT (figures not shown). Therefore, we have also developed a new iteration scheme and applied it to the allocation computation for all PFTs. For all the cases we have tested, the new scheme does not generate any unrealistic behaviors.

4.4. Definition of Fractional Coverages of PFTS

[30] In both CLM and CLM-DGVM, several PFTs can coexist at a model grid cell, with each PFT covering a fraction of ground surface. The total PFT coverage (including bare soil) is always equal to 100%, implying that the model does not allow the overlapping between different PFTs. In CLM, the fractional coverage of each PFT is prescribed, while in CLM-DGVM it is computed as the fractional foliar projective cover (FPC) [Levis et al., 2004, equation 28] and varies from year to year. However, FPC is always smaller than the fraction of area covered by a PFT (e.g., the crown area) as used in CLM, with the ratio being FPCind [Levis et al., 2004, equations 27–28]. While photosynthesis is calculated using LAI, which is calculated as the total leaf area divided by the plant crown area (CA) [Levis et al., 2004, equation 26], the plant maintenance respiration are averaged over FPC [Levis et al., 2004, equations 4–6]. Therefore, the difference between FPC and CA leads to the inconsistent computation of NPP [Levis et al., 2004, equations 8–9] as the difference between photosynthesis (over CA) and maintenance respiration (over FPC) in CLM-DGVM.

[31] Such an inconsistency is especially significant for PFTs with a small LAI, such as shrubs, and hence is removed in our revision by directly computing FPC as the average crown area of individual plants multiplied by the PFT population density. In other words, the FPC of individual plants (i.e., the term FPCind) for both trees and shrubs is dropped from equation 28 of Levis et al. [2004]. Note that this correction is not needed for grasses because the term FPCind is not computed. The importance of this revision will be further discussed in section 6.

4.5. Competition for Light

[32] Because trees are taller than grasses and hence have the advantage in capturing incoming solar radiation, the competition of trees and grasses for available space roughly follows the hierarchy of tree-grass in CLM-DGVM, in that the summation of FPCtree is limited to 95% of the naturally vegetated area, and grasses can only grow over the remaining area that is not occupied by trees. Although shrubs are a kind of woody plant, they exist primarily over arid and semiarid regions. Because of their slow growth rate, shrubs cannot compete effectively with trees or grasses when water supply is sufficient. Therefore, we expand the above competition strategy into the hierarchy of tree-grass-shrub. The rules for trees and grasses remain unchanged, but shrubs can grow only over the area that is not occupied by trees or grasses. This strategy can explicitly capture the nature of the competition and coexistence of these three vegetation categories. A more realistic and physical rule to capture the ecological and biophysical mechanisms of shrubs competing with trees and grasses is left for future research.

5. Numerical Simulations

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. Observational Characteristics of Temperate Shrubs
  5. 3. Brief Description of CLM-DGVM
  6. 4. Shrub Submodel for the CLM-DGVM
  7. 5. Numerical Simulations
  8. 6. Conclusions and Further Discussions
  9. Acknowledgments
  10. References
  11. Supporting Information

[33] As discussed in section 4.1, the scheme of solar absorption by sunlit leaves only in the default CLM should be replaced by the absorption by both sunlit and shaded leaves. Furthermore, the impact of nitrogen concentration distribution in the canopy on Vmax in equation (2) should be considered for physical reasons and to avoid unnecessarily large LAI over the tropical forest regions. Because these issues have been addressed by Thornton and Zimmermann [2007], hereafter for brevity, CLM-DGVM refers to the default CLM-DGVM with these two modifications, and its simulation is referred to as the “control simulation”. The default CLM-DGVM with all the revisions discussed in section 4 is referred to as “revised model” and its simulation is called “new simulation”. While the differences between the new and control simulations are caused by the introduction of shrubs in the new simulation over arid and semiarid regions, these two simulations may yield different results over other regions because of the revised formulation of PFT fractional coverage in section 4.4 and the revised iteration scheme in the allocation computation in section 4.3 for all PFTs.

[34] As a preliminary evaluation of the impact of the shrub submodel on the CLM-DGVM simulations, we have done global offline simulations for 400 years at T-42 resolution (128 × 64 grid cells) using the CLM-DGVM (“control simulation”) and the revised model (“new simulation”). The near-surface atmospheric forcing data (of temperature, humidity, wind, precipitation, and downward solar and longwave radiation) are generated by cycling the National Centers for Environmental Prediction (NCEP) reanalysis data of the year 1998. In other words, there is only seasonal variation but no interannual variability in the forcing data.

5.1. Evolution and Competition of PFTs Over Selected Grid Cells

[35] We first analyze the 400-year simulations at selected locations with different annual precipitation (Pann) to demonstrate the evolution and competition of shrubs, grasses, and trees in the model. The four locations include Case 1: New Mexico grid cell centered at (105.5°W, 32.3°N) with Pann = 196 mm; Case 2: Arizona grid cell centered at (111.1°W, 35.2°N) with Pann = 355 mm; Case 3: Oklahoma grid cell (99.8°W, 35.2°N) with Pann = 480 mm; and Case 4: Kentucky grid cell (83.0°W, 38.0°N) with Pann = 1157 mm, respectively.

[36] Figures 5 and 6show the PFT percent coverages over the 400-year simulation for both control and new simulations for these four cases. Shrubs survive alone in the driest case (Case 1; Figure 5a), and coexist with grasses in the wetter case (Case 2; Figure 5c). As precipitation increases, shrubs cannot compete with grasses (Case 3; Figure 6a) and finally cannot exist as the land surface is fully covered by trees and grasses (Case 4; Figure 6b). In the first two cases, the occurrence of shrubs reduces the bare soil fraction, slightly increases transpiration and total evapotranspiration over the whole grid cell, and hence results in slightly lower total soil moisture in the new simulations compared with the control simulations (see Figures 5b, 5d, and Table 3). In Case 2, this decreases the grass coverage to 52% in the new simulation compared with 62% in the control simulation. The grid cell averaged GPP and NPP (Table 3) are higher in the new simulation, but they may still be too low in Case 1 (e.g., compared with Reynolds et al. [1980]) because of only 20% shrub coverage (Figure 5a) and very small LAI (Table 4). In the wetter Cases 3 and 4, grasses (Figure 6a) and trees (Figure 6b) are dominant, and their coverages are not affected by shrubs.

image

Figure 5. 400-year CLM-DGVM control (dotted lines) and new (solid lines) simulations over two grid cells with different annual precipitation. (a) PFT percent coverages and (b) soil water (in mm) in the top seven layers (down to 0.83 m depth) over a New Mexico grid cell centered at (105.5°W, 32.3°N) with Pann = 196 mm; (c) PFT percent coverages and (d) soil water (in mm) over an Arizona grid cell centered at 111.1°W, 35.2°N with Pann = 355 mm.

Download figure to PowerPoint

image

Figure 6. 400-year CLM-DGVM control (dotted lines) and new (solid lines) simulations over two grid cells with higher annual precipitation than in Figure 5. (a) PFT percent coverages over an Oklahoma grid cell (99.8°W, 35.2°N) with Pann = 480 mm and (b) PFT percent coverages over a Kentucky grid cell (83.0°W, 38.0°N) with Pann = 1157 mm.

Download figure to PowerPoint

Table 3. Annual Evapotranspiration, Transpiration, Soil Moisture in the Top Seven Layers (Down to 0.83 m Depth), Gross Primary Production, and Net Primary Productiona
 ET (mm a−1)Transpiration (mm a−1)Soil Moisture (mm)GPP (gC m−2 a−1)NPP (gC m−2 a−1)
  • a

    Averaged over all plant functional types (PFTs) over the two grid cells (Case 1: New Mexico; Case 2: Arizona) in Figure 5. The corresponding values for individual PFTs are also shown. Note that all PFTs share the same soil moisture in the grid cell. ET is evapotranspiration; GPP is gross primary production; and NPP is net primary production.

Case 1 New
Grid cell average184.61.9126.96.83.4
Shrub273.39.0-32.516.2
 
Case 1 Ctrl
Grid cell average183.80.4129.21.00.2
 
Case 2 New
Grid cell average321.172.3147.6293.7129.5
Shrub382.642.7-184.478.7
C3 grass265.499.2-392.8175.6
 
Case 2 Ctrl
Grid cell Average320.170.9148.5275.2120.4
C3 grass317.8114.0-442.7193.7
Table 4. Morphologies of Shrubs, Grasses, and Trees at the End of the 400-Year New Simulation at the Four Locations Discussed in Section 5.1a
P (mm a−1)196 (Figure 5a)355 (Figure 5c)480 (Figure 6a)1157 (Figure 6b)
  • a

    C3 denotes C3 nonarctic grass; tree1 denotes needleleaf evergreen temperate tree; tree2 denotes broadleaf deciduous temperate tree; P denotes annual precipitation; PFT denotes plant functional type; FC denotes fractional coverage; LAI denotes annual averaged leaf area index (which for shrubs is about half of the maximum LAI); and Htop denotes canopy top height.

PFTshrubshrubC3shrubC3C3tree1tree2
FC (%)19.747.552.49.890.05.028.866.2
LAI0.110.652.391.023.812.025.874.96
Htop (m)0.511.090.610.931.260.4515.119.4

[37] Table 4 summarizes the fractional coverage, LAI, and canopy height (Htop) of each PFT at the above locations at the end of the 400-year new simulation. The LAI of shrubs varies from 0.1 to 1.0, and Htop from 0.5 m to 1.1 m, all satisfying the criteria of shrubs defined in sections 2 and 4. In contrast, trees have the largest LAI and are the tallest (Htop > 15 m).

5.2. Global Simulations

[38] Figure 7 shows the global distribution of bare soil, shrubs, trees (sum of seven PFTs) and grasses (sum of three PFTs) from the new simulation. Shrubs can grow over arid and semiarid regions in the southwest of North America, Australia, Middle East, central Asia, northern China, southern South America, and South Africa, consistent with the satellite observations in Figure 1. There is a narrow band of shrubs near the southern edge of the Sahara Desert from the MODIS data in Figure 1, which is largely missing in Figure 7, because of the coarse resolution (2.8 deg) of the simulation, as confirmed using even slightly higher resolution data. The observed boreal shrubs over high latitudes in Figure 1 are absent in the model simulation (Figure 7), because they are a different PFT and cannot be represented by the temperate shrub submodel developed here. The treatment of boreal shrubs is a task for our future research.

image

Figure 7. Global distribution of the percent coverages of (a) trees, (b) grasses, (c) bare soil, and (d) shrubs at the end of the 400-year new simulation.

Download figure to PowerPoint

[39] Figure 8 shows the fractional coverage differences in trees, grasses, and bare soil between the new and control simulations. Comparison of Figure 7d with Figure 8c demonstrates that the new simulation grows shrubs primarily at the expense of reduced bare soil coverage. Comparison of Figure 7d with Figure 8b also suggests that the existence of shrubs reduces the grass coverage over some areas in agreement with Figure 5c, because of the reduction of soil moisture over these areas (Figure 8d).

image

Figure 8. Global distribution of (a) tree, (b) grass, and (c) bare soil percent coverage differences between the new and control simulations. (d) Soil water difference (in mm) in the top seven layers (down to 0.83 m depth).

Download figure to PowerPoint

[40] Figures 8a and 8b show that there are also differences in grass or tree coverage between the new and control simulations over areas with minimal or no shrub existence, particularly over the Siberia. There are also changes in soil moisture (Figure 8d) associated with these changes in PFT coverages. As mentioned at the beginning of section 5, these differences are caused by the revised formulation for the computation of PFT fractional coverage and the iteration scheme for the allocation computation for all PFTs. This issue will be further discussed in section 6.

[41] The fractional shrub coverage in Figure 7d cannot be quantitatively evaluated against the MODIS land cover data in Figure 1 because of the binary nature (i.e., 100% cover or no cover) of land cover data. In our earlier work [Zeng et al., 2000, 2003; Miller et al., 2006], we have developed global satellite-based fractional vegetation data. Combination of these data with the land cover data in Figure 1 provides the MODIS-based shrub fractional coverage in Figure 9b. Comparison of Figure 9a (same as Figure 7d except between 60°N and 60°S) with Figure 9b indicates that the overall geographic distribution of simulated shrubs is similar to that based on the MODIS data. The exact values of simulated and satellite-based shrub coverages in each grid cell, however, are different, which might be partially caused by the use of atmospheric forcing data for 1 year only (i.e., the NCEP reanalysis for the year 1998), the relatively coarse resolution of T42 in the simulation, and the inhomogeneity of natural (but not the model's) geographic environment in each grid cell. For instance, when the DGVM with the shrub submodel is coupled to the newer version of CLM and forced by cycling the atmospheric data from 1948 to 2004, the overall geographic distribution of shrubs remains the same, but some of the regional details are altered (e.g., more shrubs than shown in Figure 9a over Australia; Sam Levis, personal communication, 2008).

image

Figure 9. (a) Global distribution of shrub percent coverages from the new simulation (same as Figure 7d except between 60°N and 60°S) and (b) MODIS-based shrub coverages aggregated to model grid cells.

Download figure to PowerPoint

[42] Because the annual precipitation amount (Pann) is a crucial parameter in the competition of PFTs, Figure 10 shows the dependence of the average coverage of PFTs between 60°N and 60°S as a function of Pann. In the new simulation, shrub coverage reaches its peak around Pann = 300 mm (Figure 10a) in excellent agreement with the MODIS data (Figure 10c). The simulated shrub coverage drops below 5% for Pann greater than 500 mm, while the MODIS-based shrub coverage still exists above 5% level even around Pann = 900 mm, because of the inclusion of some boreal shrubs (e.g., Figure 9b). In contrast to the narrow peak of shrub coverage around Pann = 300 mm, grass coverage reaches about 40% for a large range of Pann (between 400 mm and 1100 mm) in the new simulation. Direct comparison of model grass coverage with the MODIS data is difficult because savanna, as a land cover type in the MODIS data, is a mixture of grasses and trees. As Pann is further increased above 1200 mm, the simulated tree coverage becomes dominant and reaches its peak at Pann = 1500 mm or higher, consistent with the MODIS data (Figure 10c). The simulated peak tree coverage of around 90% is higher than the MODIS data (around 70%), partly because of the separate consideration of savanna in the MODIS data.

image

Figure 10. The dependence of the average percent coverages of PFTs as a function of annual precipitation using results between 60°N and 60°S (a) from new simulation, (b) from control simulation, (c) based on MODIS land cover and fractional vegetation cover data, and (d) using MODIS land cover data alone. The MODIS data have been aggregated to model grid cells in Figures 10c and 10d.

Download figure to PowerPoint

[43] Figure 10 also quantifies the earlier statement that land cover data alone are not appropriate to evaluate the model results. In fact, using the MODIS land cover data alone (Figure 10d) would more than double the peak value of the shrub coverage based on both land cover and fractional vegetation cover data (Figure 10c). The grass percent coverages using the land cover data alone (Figure 10d) are also much larger than those in Figure 10c.

[44] The tree and savanna coverages in Figure 10d are only slightly larger than those in Figure 10c, because the fractional vegetation coverages of these PFTs are relatively close to 100%. The bare soil fraction in Figure 10d is smaller than in Figure 10c, because the percent coverages of shrubs, grasses, savannas, and trees are all larger in Figure 10d.

[45] With our focus on dry regions, we have also done the student t test using data in Figure 10 for Pann less than 500 mm. The mean bare soil fraction difference of 7.8% (or −2.2%) between the control (or new) simulation and MODIS (land cover and fractional vegetation cover) data is statistically significant (or insignificant) at the 5% level, demonstrating that the shrub submodel enables the DGVM to simulate shrubs and improves the DGVM simulation of bare soil fraction as well. The corresponding difference of −25.8% between the land cover data alone and MODIS (land cover and fractional vegetation cover) data is also statistically significant at the 5% level and is much larger in magnitude than the above differences between models and MODIS data. Similarly, the mean shrub fraction difference of 15.5% between the land cover data alone and MODIS (land cover and fractional vegetation cover) data is statistically significant at the 5% level, in contrast to the insignificant and much smaller difference of 0.5% between the new simulation and MODIS data. This further demonstrates the importance of using both MODIS land cover and fractional vegetation cover data for DGVM evaluations.

[46] Comparison of the new (Figure 10a) and control (Figure 10b) simulations shows that shrubs grow in the new simulation primarily by decreasing the bare soil fraction and, to a lesser degree, by reducing the grass coverage. For Pann = 1200 mm or higher, results from the two simulations are very similar. For Pann = 600–1200 mm at which shrub coverage is small, the percent cover of trees is slightly larger than that of grasses in the new simulation, which is opposite to the control simulation. This issue will be further addressed in section 6.

6. Conclusions and Further Discussions

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. Observational Characteristics of Temperate Shrubs
  5. 3. Brief Description of CLM-DGVM
  6. 4. Shrub Submodel for the CLM-DGVM
  7. 5. Numerical Simulations
  8. 6. Conclusions and Further Discussions
  9. Acknowledgments
  10. References
  11. Supporting Information

[47] Temperate shrubs cover a large fraction of global land, but they are not considered by most DGVMs (including CLM-DGVM). To improve the simulation of these DGVMs over arid and semiarid regions, we have developed a shrub submodel based on observational characteristics of shrubs and previous ecological studies. In particular, we have coupled it with CLM-DGVM. The major elements of this submodel include (1) explicit consideration of shrubs' drought tolerance in the photosynthesis computation; (2) use of appropriate phenology type and morphology parameters for shrubs; (3) consistent treatment of fractional vegetation coverage; (4) development of tree/grass/shrub hierarchy for the light competition; and (5) improvement of the allocation scheme to avoid unrealistic behaviors.

[48] Preliminary global offline CLM-DGVM simulations for 400 years show that, with the shrub submodel, the simulated global distribution of temperate shrubs agrees with MODIS data. Compared with the control simulation, shrubs grow primarily by reducing the bare soil coverage and, to a lesser degree, by decreasing the grass coverage. The simulated shrub coverage reaches its peak around annual precipitation (Pann) of 300 mm, the grass coverage reaches its peak over a broad range of Pann (from 400 to 1100 mm), and the tree coverage reaches its peak for Pann = 1500 mm or higher, all in good agreement with MODIS data. While our shrub submodel has been tested in CLM-DGVM only, most of the above elements are relevant to other DGVMs and their impact on other DGVMs will be addressed in the future.

[49] The MODIS data analyses in this study also reveal that using the land cover data alone is not appropriate for the evaluation of simulated fractional coverages of PFTs. In particular, compared with the combined land cover and fractional vegetation cover data, using the land cover data alone would more than double the peak shrub coverage and significantly overestimate the peak grass coverage.

[50] In addition to the above differences between the new and control simulations where shrubs are present, differences have also been found where shrub coverage is zero or minimal. To further address this issue, we pick up a model grid cell centered at (90.2°E, 69.8°N) in Siberia. In the control simulation, C3 arctic grass is dominant with 98.1% coverage, while needleleaf evergreen trees are dominant with 89.5% coverage in the new simulation. In both simulations, shrub coverage is zero. These substantial changes are entirely caused by the use of revised FPC formulation as well as a new iteration scheme for the allocation computation to avoid unrealistic behaviors at some model grid cells. If the original FPC formulation (see section 4.4) is used in the new simulation, results are similar to the control simulation: C3 arctic grass is dominant with 91.0% coverage. This suggests that the change of the FPC formulation is the primary reason for the substantial change between the control and new simulations. Further analysis indicates that, in the new simulation, the needleleaf evergreen tree at this grid cell grows very slowly at the beginning (e.g., with a coverage of only 2.2% after 50 years of simulation). This implies that the annual carbon production (through photosynthesis) and loss (through respiration and turnover, etc.) are nearly in balance for trees at this location, with a slightly higher production over loss in the new simulation. As discussed in section 4.4, respiration is inconsistently averaged over FPC in the default CLM-DGVM, in contrast to the averaging over the crown area (CA) in the new simulation. Because the ratio of FPC over CA is less than 0.6 in the control simulation over this grid cell, the carbon loss in the control simulation becomes slightly larger than carbon production, preventing the presence of trees.

[51] As mentioned earlier, the default CLM-DGVM considers the solar absorption by sunlit leaves only and does not consider the control of nitrogen concentration on Vmax in equation (2). These two deficiencies have been removed here by considering the solar absorption by both sunlit and shaded leaves and by including the control of nitrogen concentration in the control simulation. The 400-year simulation using the default CLM-DGVM would yield C3 arctic grass coverage of 45.9% with bare soil fraction of 54.1% (and zero tree coverage) over the above grid cell in Siberia. These results are different from those in the control or new simulation. All these results are also different from the MODIS data with 53.4% shrub coverage and 46.6% bare soil coverage over this grid cell. However, the bare soil fraction from the default CLM-DGVM simulation is most similar to the MODIS data.

[52] The above analyses over this model grid cell demonstrate the sensitivity of the CLM-DGVM over high latitudes to minor changes in the model formulations. Because of the significantly different snow albedos over areas dominated by grasses versus trees [e.g., Barlage et al., 2005], changes in the dominant PFT are expected to significantly affect the land-atmosphere coupled simulations through the widely recognized snow albedo feedback mechanism [e.g., Hall and Qu, 2006]. With a focus on arid and semiarid regions in this study, it will be our future task to carefully addressing this issue over high latitudes.

Acknowledgments

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. Observational Characteristics of Temperate Shrubs
  5. 3. Brief Description of CLM-DGVM
  6. 4. Shrub Submodel for the CLM-DGVM
  7. 5. Numerical Simulations
  8. 6. Conclusions and Further Discussions
  9. Acknowledgments
  10. References
  11. Supporting Information

[53] This work was supported by NASA (NNG06GA24G, NNG04G061G) and CAS (KZCX2-YW-219). The authors are grateful to the IAP/CAS group on Earth System Dynamic Model for valuable discussions. Two anonymous reviewers are appreciated for helpful comments.

References

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. Observational Characteristics of Temperate Shrubs
  5. 3. Brief Description of CLM-DGVM
  6. 4. Shrub Submodel for the CLM-DGVM
  7. 5. Numerical Simulations
  8. 6. Conclusions and Further Discussions
  9. Acknowledgments
  10. References
  11. Supporting Information

Supporting Information

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. Observational Characteristics of Temperate Shrubs
  5. 3. Brief Description of CLM-DGVM
  6. 4. Shrub Submodel for the CLM-DGVM
  7. 5. Numerical Simulations
  8. 6. Conclusions and Further Discussions
  9. Acknowledgments
  10. References
  11. Supporting Information
FilenameFormatSizeDescription
gbc1493-sup-0001-t01.txtplain text document0KTab-delimited Table 1.
gbc1493-sup-0002-t02.txtplain text document0KTab-delimited Table 2.
gbc1493-sup-0003-t03.txtplain text document1KTab-delimited Table 3.
gbc1493-sup-0004-t04.txtplain text document1KTab-delimited Table 4.

Please note: Wiley Blackwell is not responsible for the content or functionality of any supporting information supplied by the authors. Any queries (other than missing content) should be directed to the corresponding author for the article.