## 1. Introduction

Vegetation affects climate by modifying the energy, momentum, and hydrologic balance of the land surface. Boundary layer exchanges of heat and momentum, evapotranspiration, and the absorption of solar radiation, are all influenced by vegetation and have important feedbacks on the global and regional climate. Vegetation influences climate via its physiological (stomatal conductance) and structural (leaf area index, root depth and distribution, height, and albedo) properties. Biophysical effects of vegetation have been considered in a number of studies that investigate the effects of deforestation (Charney, 1975; Lean and Rowntree, 1997), the effects of changes in the structural and physiological characteristics of vegetation (Pollard and Thompson, 1995; Douville et al. 2000), and the effects of anthropogenic land cover change (Chase et al., 2000; Zhao et al., 2001). At the global scale, vegetation also affects atmospheric processes through its effects on CO_{2} exchange and the emission of other radiatively active trace gases (Shaver et al., 1992). Changes in the structural and physiological properties of vegetation, in response to increases in atmospheric CO_{2} concentration and climate warming, are expected to have important effects on the overall sensitivity of the climate system.

The interaction between vegetation and climate is represented by soil–vegetation–atmosphere transfer (SVAT) schemes and most existing SVAT schemes do not consider vegetation as a dynamic component. However, SVAT schemes with partly dynamic vegetation (e.g., Dickinson et al., 1998; Sellers et al., 1996) and dynamic vegetation models (DVMs) (e.g., Foley et al., 1996; Kucharik et al., 2000) are emerging. The most complex DVMs include a SVAT scheme coupled to a terrestrial ecosystem model that provides the SVAT scheme with time-varying structural and physiological vegetation characteristics, and dynamic geographical distribution of vegetation. Arora (Arora, 2002), for instance, illustrates how various time-varying components of vegetation, simulated in dynamic vegetation models, can be used to change the structural attributes of vegetation that are assumed constant in most SVAT schemes and hydrological models. The primary difference between coupling an atmospheric model to a standard SVAT scheme, rather than a dynamic vegetation model, is that a SVAT scheme uses prescribed surface roughness length (based on an assumed distribution of vegetation height) and a prescribed seasonal evolution of leaf area index (LAI) rather than calculating these quantities prognostically and interactively in response to local conditions as is done in a dynamic vegetation model.

Leaf, stem, and root biomass, *B*_{L}, *B*_{S}, and *B*_{R} (kg m^{–2}), respectively, are primary prognostic variables in dynamic vegetation models. Biomass in units of kilograms per square meter is closely associated to carbon content (kg C m^{–2}) since the ratio *r*_{c} of carbon to biomass is fairly stable at about 0.5 (Ajtay et al., 1979). Structural vegetation attributes such as LAI and vegetation height, *h* (m), are obtained from leaf and stem biomass using specific leaf area (SLA; m^{2} kg C^{–1}), where LAI = (1/*r*_{c}) × *B*_{L} × SLA and allometric relationships *h* = *f*(*B*_{S}). SLA is defined as the leaf area obtained per unit of dry mass of leaf biomass and is a vegetation-dependent parameter. Allometry represents the relative growth of a part in relation to the whole plant and allometric relationships relate attributes, such as weight, height, length, diameter, or area, of various plant components to one another or to the entire plant, often via power-law relationships (e.g., Enquist and Niklas, 2002). SLA has been used as a prescribed vegetation-dependent constant in the Integrated Biosphere Simulator (IBIS; Kucharik et al., 2000) and estimated as a function of leaf life span in the Ecosystem Demography (ED) and Lund–Potsdam–Jena (LPJ) models (Moorcroft et al., 2001; Sitch et al., 2002; manuscript submitted to *Global Biogeochem. Cycles*). Vegetation height (*h*) is related to stem biomass (*B*_{S}) using a simple allometric relationship (*h* ∝ *B*_{S}^{0.4}*D*) as *h* ∝ *D*^{b} (Moorcroft et al., 2001).

While LAI and vegetation height are treated as time-varying structural attributes in most dynamic vegetation models, root distribution and rooting depth are not. This is primarily because no allometric relationships exist that relate root biomass and/or vegetation age to a root distribution profile and a rooting depth. However, observational data show, not unexpectedly, that root distribution profiles and rooting depth change with plant age and root biomass. Root distribution and maximum rooting depth are also affected by soil texture and depth (Schenk and Jackson, 2002a) and by mean annual precipitation (Schenk and Jackson, 2002b).

Root distribution, among other factors, plays an important role in determining various ecosystem processes including vegetation distribution (Peters, 2002), preferential water use (Scott et al., 2000), and coexistence of different vegetation types (van Wijk and Rodriguez-Iturbe, 2002).

The use of an invariant root distribution in ecosystem models results in a constant fraction of roots in each soil layer, independent of plant age or root biomass, and this invariant root fraction is used in the calculation of transpiration and other quantities. The IBIS model (Kucharik et al., 2000), for example, assumes that the cumulative root distribution is given by (1 − *β*^{z}), where *z* is the soil depth and *β* is a vegetation-dependent coefficient (Jackson et al., 1996), which is assumed to be time invariant. Since *β* does not change with vegetation age and/or root biomass, the rooting depth (defined as the depth containing 95% of the roots) remains constant irrespective of the size of the plant. Thus a short young tree with a small root biomass, and a large old tree with a large root biomass are both assumed to have the same rooting depth, root distribution and, consequently, the same fraction of roots in each soil layer. Most vegetation and terrestrial carbon models, and almost all SVAT schemes, follow this approach (e.g., Sitch et al., 2002; Knorr, 2000; Bonan, 1996; Verseghy et al., 1993; Potter and Klooster, 1999).

The assumption that the root distribution profile and rooting depth does not change in time may be suitable for simulating energy and water balance dynamics of mature stands over a short period of time but is not expected to be optimum for dynamic vegetation models which are designed to capture the growth and death dynamics and the establishment or encroachment of vegetation in empty plots over decade to century time-scales. For example, based on stable isotope analysis in Arizona, USA, Weltzin and McPherson (Weltzin and McPherson, 1997) conclude that tree seedlings use shallow sources of soil water, while mature trees use deeper sources. The authors find that at two years of age trees were still accessing water from shallow sources and competing with grasses. Donovan and Elheringer (Donovan and Elheringer, 1994) draw similar conclusions for shrubs. They find that the seasonal development of water stress in shrubs decreased with increasing size, age, and rooting depth. Irvine et al. (Irvine et al., 2002) find that while younger stands of ponderosa pine trees (∼14 years old) experience water stress and consequently show reduced transpiration rates in the later half of the growing season, the older trees (50 and 250 years old) did not show any such behavior. They suggest these responses are a consequence of deep and shallow rooting profiles of the older and younger trees, respectively. These studies indicate that as plants grow and their root biomass increases they access water from deeper soil layers due to their increased rooting depth and deeper root profiles.

Section 2 discusses the traditional exponential root distribution used in most vegetation models and SVAT schemes and illustrates its limitations. A method for representing root distribution profile as a function of root biomass is proposed in section 3, which is intended to overcome some of the limitation in the prescription of a static root distribution profile in dynamic vegetation models. In this representation, root distribution and rooting depth are functions of time through their dependence on evolving root biomass. A comparison against observed root biomass data is presented in section 4, the application of the approach is in section 5, and the results are summarized in section 6.

## 2. The traditional exponential root distribution

^{–2}), consisting of both fine and coarse roots, provides the basic representation of root distribution, and we follow Jackson et al. (Jackson et al., 1996, hereafter J96) in adopting an exponential root density profile for vegetation type

*i*as where

*z*is the depth (m), positive downward;

*a*

_{i}= 1/

*L*

_{i}, where

*L*

_{i}is a vegetation-dependent

*e*-folding length scale; and

*A*

_{i}(

*t*) is the surface root density that is a function of time

*t*. Here

*a*

_{i}= −

*γ*ln

*β*

_{i}connects J96's representation with the exponential representation used here.

*B*

_{Ri}above depth

*z*, and omitting subscripts

*i*and

*R*for simplicity, is given by and the total root biomass is

*B*(

*t*) =

*B*

_{∞}(

*t*) =

*A*(

*t*)/

*a*. The cumulative root fraction at depth

*z*is and the rooting depth that contains the fraction

*f*of the total root mass is The “rooting depth” is typically defined as the depth that contains 95% of the root biomass, but for simplicity we use

*f*= 0.9502 and the term is the rooting depth. Rewriting the root density (for a particular vegetation type) as relates the three measures of root distribution,

*ρ*,

*f*, and

*d*[in Equations (3)–(6)], to

*a*and

*B*(

*t*). The use of a fixed

*a*in these formulas implies a constant rooting depth

*d*[Equation (5)] and a cumulative root fraction

*f*[Equation (3)] independent of the age of the plant and the total root biomass

*B*(

*t*). However, the surface root density, is directly proportional to root biomass. The result is the root density shown in Figure 1a and the constant cumulative root fraction, independent of

*B*(

*t*), shown in Figure 1b. The value

*a*= 3 characterizing trees following J96 is used in the figure. J96 provide estimates of mean root biomass for different biomes and estimate a mean value of about 4.4 kg m

^{–2}for trees for which the root profile is represented by the solid line in the figure. For other root profiles the root biomass ranges between 0.1 and 8.7 kg m

^{–2}.

Figure 1a requires that young plants attain their rooting depth early in life with long skinny roots and a small root biomass, and that older plants with large root biomass follow the same distribution so that rooting depth, and the fraction of roots in soil layers, are the same at all ages. This implies that after attaining their rooting depth with a small root biomass, roots essentially grow only horizontally. That is, at any depth the horizontal increase in root density is directly proportional to the increase in root biomass and the rooting depth does not change. This contrasts with the cumulative root distribution data in Figure 2 from four temperate coniferous sites, three tropical evergreen sites, and three tundra sites for a range of root biomasses. These data (kindly made available by R. Jackson) are a subset of the data used in J96 from the sources listed in Table 1. The data show that a constant rooting depth and root distribution profile are not very realistic. For all three biomes an increase in root biomass is associated with an increase in rooting depth and a change in the root distribution profile such that a larger fraction of roots exists in deeper soil layers as the root biomass increases. The data from 12-, 22-, and 45-yr-old Corsican Pine *(temperate coniferous)* sites in Figure 2a provide a chronological sequence, while data from the other two biomes are from sites with a range of root biomasses.

## 3. Relating root distribution profile to root biomass

*d*∝

*B*∝ 1/

*a*; that is, the parameter

*a*in Equation (6) is a function of root biomass

*B*(

*t*). A power-law relation is consistent with the assumed exponential root profile and we set which yields where

*a*,

*α*, and

*b*all depend on vegetation type. Equations (9)–(11) indicate how

*ρ*,

*f*, and

*d*vary over time in conjunction with

*B*(

*t*) in this formulation. The potential virtue of this representation is that the rooting depth is initially zero and increases with plant age and root biomass. The root profile also varies with the root biomass as expected.

*α*. For

*α*= 0,

*b*=

*a*, and the original formulation [Equations (1)–(6)] is recovered where roots grow only horizontally, without an increase in rooting depth, as

*B*(

*t*) increases (see Figure 1) and cumulative root fraction

*f*and rooting depth

*d*do not depend on time and root biomass. For

*α*= 1, by contrast, and roots grow mainly vertically downward as shown in Figures 3a and 3b. That is, the surface root density stays constant while rooting depth increases in direct proportion to root biomass. For values of

*α*between 0 and 1 roots grow both horizontally and vertically as shown in Figures 3c and 3d for the example of

*α*= 1/2. We term

*α*the “root growth direction” parameter since its value determines the rate at which roots evolve both horizontally and vertically as

*B*increases. The representation proposed here expresses the root distribution profile and rooting depth as a function of root biomass via the root growth direction parameter,

*α*.

While time-varying root distribution and rooting depth with plant age and/or root biomass is not yet a feature of most global applications of dynamic vegetation models, some agricultural and small-scale ecosystem models do incorporate such a relationship. Kage et al. (Kage et al. 2000) simulate the increasing rooting depth of cauliflower as a function of the cumulative temperature sum (the sum of daily temperatures above a certain threshold, which is used as a measure of accumulated heat) after the planting date. In their model the rooting depth increases at an exponential rate in the early days after planting and at a linear rate thereafter. Rasse et al. (Rasse et al. 2001) model the downward expansion of roots of beech and Scots pine trees as a linear function of stand age up to 100 years of age, after which the rooting depth is assumed to remain constant. These models simulate the change in rooting depth and root distribution profile as a function of the cumulative heat and plant age, respectively, rather than root biomass. By contrast, the approach proposed here takes an “allometric” view that results in a time-dependent root density distribution and rooting depth as a function of the time-dependent root biomass.

## 4. Parameter estimation and verifying the representation

*b*and

*α*and verify the appropriateness of Equations (9)–(11) for each vegetation type by fitting each to observed data. Rooting data are not widely available, however, and as a practical matter we would like to represent

*α*and

*b*in terms of more readily available information. We appeal to the mean

*e*-folding length information

*a*

*γ*ln

*β*

*B*

*a*

_{i}and

*B*

_{i}). This implies that

*b*=

^{α}and that in Equations (9)–(11). When we know

*a*

*B*

*α*that can then be used to describe the varying root distribution profile. Finally if

*α*does not vary greatly among vegetation types, then some reasonable value can be used in general. Equation (13) implies that when root biomass

*B*for a plant is equal to its mean root biomass

*B*

*a*=

*a*

*a*<

*a*

*a*>

*a*

*α*. We first estimate

*a*

*k*is the depth index and

*j*is the “biomass” or profile index labeling the individual profiles. That is, we estimate the mean exponential root distribution profile represented by

*a*

*a*

_{1}for

*a*

_{1}) and the coefficient of efficiency (

*E*) (Nash and Sutcliffe, 1970) are also shown. The coefficient of efficiency is given by

*E*= 1 −

*f*

_{jk}are the observed cumulative root fractions, and

*f*

*a*

*α*may be estimated by minimizing where

*B*

_{j}is the biomass of profile

*j*and

*B*

*α*such that the predicted profiles corresponding to the different root biomasses

*B*

_{j}are fitted to the observed data in a least squares sense. The value of

*α*thus obtained and of

*a*

*B*

*a*

_{1}, for the subset of data used here does not differ much from the

*a*

*α*equal to 0.70, 0.78, and 0.90 for temperate coniferous, tropical evergreen, and tundra biomes, respectively. The resulting values of

*E*= 1 −

_{2}) for each biome are also shown in Figure 5. The

*E*values in Figure 5 are higher, and the sum of the square of errors are lower, than that in Figure 4 because of a better fit to the cumulative root fraction data. Figure 5 indicates that Equation (16), with these values of

*α*, captures the first-order effect of the change in the root distribution profile with the root biomass. A value of

*α*larger than 0.5 implies that the rate of root growth is larger in the vertical direction than in the horizontal direction as the root biomass increases in time. The change in the root distribution profile with increasing biomass is captured especially well by Equation (16) for the three tundra sites, as seen in Figure 5c. For the temperate coniferous (Figure 5a) and tropical evergreen (Figure 5b) sites the model does somewhat less well in capturing the shallow root profile associated with the lowest root biomass.

The root growth direction parameter *α* may itself be a function of a plant's age. For example, roots may tend to grow vertically early in a plant's age and then may start growing horizontally as well. However, in the absence of a detailed analysis of chronological sequences of root biomass data from a variety of sites and biomes we cannot infer the extent to which *α* changes with plant age. A constant value of *α* appears to capture the first-order effect of change in the root distribution profile with the root biomass.

Rooting depth is also influenced by soil texture (Schenk and Jackson, 2002a) and may also be restricted by bedrock or permafrost. Rooting depth is also likely to be influenced by the nutrient status of the soils. For example, the poor nutrient status of northern soils may be a reason for the shallow rooting depths of boreal forests (J96; Schenk and Jackson, 2002a) compared to temperate and tropical forests, although this may be at least partially captured in the distribution parameters (*a**B**α*). While we suggest that the representation of root distribution and rooting depth as a function of root biomass proposed here better captures rooting behavior than the fixed profile used in most current DVMs, it is clear that other factors may also affect rooting behavior.

## 5. Application

The methodology proposed here for representing variable rooting depth and root distribution as a function of root biomass is intended to be used in the Canadian Terrestrial Ecosystem Model (CTEM; Arora, 2003) coupled to a SVAT scheme for inclusion in the Canadian Centre for Climate Modelling Analysis coupled climate model. Rooting depth will be limited to the total soil depth (which is essentially the depth to bedrock) used in the thermal and hydrological balances of the climate model's land surface scheme (Verseghy et al., 1993). In particular, the implementation uses values of *a**B**α* as a first approximation (in the absence of values of *α* for different biomes), except for the case when soil depth limits root growth.

*α*since to keep

*d*constant at the soil depth

*d** requires that hence,

*α*= ln(

*d**

*a*

*B*(

*t*)/

*B*

*α*decreases as the root biomass increases and roots grow in the horizontal direction once the rooting depth equals the soil depth. This is illustrated in Figure 6 where the special value of

*α*= 1 is used for clarity since root density in this case does not increase at the surface. A soil depth equal to 1.2 m is assumed. With

*α*= 1 roots grow vertically downward until the root biomass increases above a certain value (corresponding to

*d*=

*d**) when

*α*begins to decrease and root growth occurs in the horizontal direction such that 95% of the root biomass is always contained within 1.2 m.

The proposed representation expresses the differences in the root distribution profile of various biomes through the vegetation-dependent parameter *b* = *b* for different biomes are listed in Table 3 based on values of *a**B**α* = 0.8. The resulting value of *b* is higher for boreal forests than for temperate and tropical forests so that rooting depth *d* [Equation (11)] is shallower for boreal than that for other forests with the same mean root biomass. Similarly, when comparing temperate grasslands and tundra, which have similar average root biomass, tundra plants have shallower root profiles than temperate grassland for the same root biomass. Equation (11) provides rooting depth as a function of root biomass while taking into account the difference in the root distribution profile of different biomes. The approach proposed here is an extension of the traditional exponential root distribution used in SVAT schemes and DVMs and is intended to represent the average root profile of many plants distributed over comparatively large regions, rather than the root profile of an individual tree or stand.

## 6. Summary and conclusions

While leaf area index and vegetation height, derived from leaf and stem biomass, are considered as time-varying structural attributes of vegetation in the current generation of dynamic vegetation models, rooting depth and root distribution are usually considered to be time invariant. This implies that the fractional root distribution in soil layers, which is used to estimate transpiration, does not change no matter what the vegetation age and/or root biomass may be.

We propose a simple representation in which the root distribution profile and rooting depth are functions of the root biomass. Following J96, the root profile is assumed to be exponential (*f* = 1 − *e*^{–az}) with an associated rooting depth of *d* = 3/*a*. We connect parameter *a* of the exponential distribution to the time-dependent root biomass *B*(*t*) via a root growth direction parameter *α* as *a* = *bB*^{α}(*t*). When *α* = 0 the original distribution is recovered and this is the representation used in most current dynamic vegetation models in which the root distribution profile is independent of time and root biomass. In that case plants attain their final rooting depth very early in life, with a very small biomass, and thereafter roots grow basically horizontally. When *α* = 1, rooting depth increases in proportion to *B*(*t*) [*d* = 3*B*(*t*)/*b*] and roots grow mainly in the vertical direction. Values of *α* between 0 and 1 provide rooting profile behavior in which roots grow in depth as well as horizontally. Analysis of chronological sequences of root biomass data from a temperate coniferous site, and data from tropical evergreen and tundra sites give moderately different values of *α*. In the absence of a knowledge of *α* for each biome a value of *α* ≈ 0.8 captures the first-order effect of changes in the root distribution profile with root biomass. The primary objective here is to illustrate that this approach can be used to represent the variable root distribution profile and rooting depth. Preliminary analysis of root biomass data from temperate coniferous, tropical evergreen, and tundra sites indicates that root distribution profiles change with an increase in root biomass in a manner similar to that proposed. Implementation of variable rooting depth and profile is also consistent with increased allocation of carbon to roots in response to water stress in DVMs, which incorporates dynamic allocation.

The approach proposed in this paper offers the possibility of a somewhat more realistic treatment of the root distribution profile and rooting depth than is used in most current DVMs. However, the effect of soil texture and other factors on the root distribution profile is only implicit in the rooting profiles of various biomes via the parameter *b* =

## Acknowledgments

We gratefully acknowledge Robert Jackson's help in providing us with the root biomass data used in this study. We are also thankful to Jim Christian for providing comments on an earlier version of this paper. We would also like to thank the two anonymous reviewers whose comments helped to improved the paper.

## References

Ajtay, G. L., , P. Ketner, , and P. Duvigneaud, 1979. Terrestrial primary production and phytomass.

*The Global Carbon Cycle,*edited by B. Bolin, E. T. Degnes, S. Kempe and P. Ketner, pp. 129–181, Wiley, Chichester.Arora, V. K., 2002. Modelling vegetation as a dynamic component in soil–vegetation–atmosphere–transfer schemes and hydrological models.

*Rev. Geophys*40 2:1006. doi:10.1029/2001RG000103.Arora, V. K., . 2003. Simulating energy and carbon fluxes over winter wheat using coupled land surface and terrestrial ecosystem models.

*Agric. For. Meteorol*118 1–2:21–47.Bonan, G. B., 1996. A Land Surface Model (LSM version 1.0) for Ecological, Hydrological, and Atmospheric Studies: Technical Description and User's Guide. NCAR Tech. Note NCAR/TN-417+STR. National Center for Atmospheric Research, Boulder, CO, 150 pp.

Charney, J. G., 1975. Dynamics of deserts and drought in the Sahel.

*Q. J. R. Meteorol. Soc*101:193–202.Chase, T. N., , R. A. Pielke Sr., , T. G. F. Kittel, , R. R. Nemani, , and S. W. Running, 2000. Simulated impacts of historical land cover changes on global climate in northern winter.

*Clim. Dyn*16:93–105.Cox, P. M., 2001. Description of the TRIFFID dynamic global vegetation model.

*Hadley Centre Tech. Note 24*, Hadley Centre, Bracknell, U.K., 16 pp.Dennis, J. G., , and P. L. Johnson, 1970. Shoot and rhizome-root standing crops of tundra vegetation at Barrow, Alaska.

*Arct. Alp. Res*2:253–266.Dickinson, R. E., , M. Shaikh, , R. Bryant, , and L. Graumlich, 1998. Interactive canopies for a climate model.

*J. Clim*11:2823–2836.Donovan, L. A., , and J. R. Elheringer, 1994. Water stress and use of summer precipitation in a Great Basin shrub community.

*Funct. Ecol*8 3:289–297.Douville, H., , S. Planton, , J-F. Royer, , D. B. Stephenson, , S. Tyteca, , L. Kergoat, , S. Lafont, , and R. A. Betts, 2000. Importance of vegetation feedbacks in doubled CO2 climate experiments.

*J. Geophys. Res*105 D11:14841–14861.Enquist, B. J., , and K. J. Niklas, 2002. Global allocation rules for patterns of biomass partitioning in seed plants.

*Science*295:1517–1520.Foley, J. A., , A. Jonathan, , I. C. Prentice, , N. Ramankutty, , S. Levis, , D. Pollard, , S. Sitch, , and A. Haxeltine, 1996. An integrated biosphere model of land surface processes, terrestrial carbon balance, and vegetation dynamics (IBIS).

*Global Biogeochem. Cycles*10 4:603–628.Huttel, C., 1975. Root distribution and biomass in three Ivory Coast rain forest plots.

*Tropical Ecological Systems*, edited by F. B. Golley and E. Medina, Springer, New York, 123–130.Ignatenko, I. V., , and F. I. Khamizyanova, 1971. Soils and total phytomass reserves in dwarf birch white dryas and willow tundra of the east European northlands.

*Ekologiya*4:17–24.Irvine, J., , B. E. Law, , P. M. Anthoni, , and F. C. Meinzer, 2002. Water limitations to carbon exchange in old-growth and young ponderosa pine stands.

*Tree Physiology*22 2–3:189–196.Jackson, R. B., , J. Canadell, , J. R. Ehleringer, , H. A. Mooney, , O. E. Sala, , and E. D. Schulze, 1996. A global analysis of root distributions for terrestrial biomes.

*Oecologia*108:389–411.Kage, H., , M. Kochler, , and H. Stutzel, 2000. Root growth of cauliflower (

*Brassica oleracea L. botrytis*) under stressed conditions: Measurement and modeling.*Plant Soil*223:131–145.Khodacheck, E. A., 1969. Vegetal matter of tundra phytocoenoses in the western part of Taimyr peninsula.

*J. Bot*54:1059–1073.Knorr, W., 2000. Annual and interannual CO2 exchanges of the terrestrial biosphere: Process based simulations and uncertainties.

*Global Ecol. Biogeogr*9:225–252.Kucharik, C. J., , J. A. Foley, , C. Delire, , V. A. Fisher, , M. T. Coe, , J. D. Lenters, , C. Young-Molling, , and N. Ramankutty, 2000. Testing the performance of a dynamic global ecosystem model: Water balance, carbon balance, and vegetation structure.

*Global Biogeochem. Cycles*14 3:795–825.Lean, J., , and P. R. Rowntree, 1997. Understanding the sensitivity of a GCM simulation of Amazonian deforestation to the specification of vegetation and soil characteristics.

*J. Clim*10:1216–1235.Mensah, K. O. A., , and J. Jenik, 1968. Root system of tropical trees. 2: Features of root system of iroko.

*Preslia*40:21–27.Moorcroft, P. R., , G. C. Hurtt, , and S. W. Pacala, 2001. A method for scaling vegetation dynamics: The ecosystem demography model (ED).

*Ecol. Monogr*71 4:557–586.Nash, J. E., , and J. V. Sutcliffe, 1970. River flow forecasting through conceptual models. 1: A discussion of principles.

*J. Hydrol*10 3:282–290.Peters, D. P. C., 2002. Plant species dominance at a grassland–shrubland ecotone: An individual-based gap dynamics model of herbaceous and woody species.

*Ecol. Modell*152:5–32.Pollard, D., , and S. L. Thompson, 1995. Use of a land surface transfer scheme (LSX) in a global climate model: The response to doubling stomatal resistance.

*Global Planet. Change*10:129–161.Potter, C. S., , and S. A. Klooster, 1999. Dynamic global vegetation modelling for prediction of plant functional types and biogenic trace gas fluxes.

*Global Ecol. Biogeogr*8 6:473–488.Rasse, D. P., , B. Longdoz, , and R. Ceulemans, 2001. TRAP: A modelling approach to below-ground allocation in temperate forests.

*Plant Soil*229:281–291.Schenk, H. J., , and R. B. Jackson, 2002a. The global biogeography of roots.

*Ecol. Monogr*72 3:311–328.Schenk, H. J., , and R. B. Jackson, . 2002b. Rooting depths, lateral root spreads and below-ground/above-ground allometries of plants in water-limited ecosystems.

*J. Ecol*90:480–494.Scott, R. L., , W. J. Shuttleworth, , D. C. Goodrich, , and T. Maddock III, 2000. The water use of two dominant vegetation communities in a semiarid riparian ecosystem.

*Agric. For. Meteorol*105 1–3:241–256.Sellers, P. J., , C. J. Tucker, , G. J. Collatz, , S. O. Los, , C. O. Justice, , D. A. Dazlich, , and D. A. Randall, 1996. A revised land surface parameterization (SiB2) for atmospheric general circulation models. Part II: The generation of global fields of terrestrial biophysical parameters from satellite data.

*J. Clim*9:706–737.Shaver, G. R., , W. D. Billings, , F. S. Chapin III, , A. E. Giblin, , K. J. Nadelhoffer, , W. C. Oechel, , and E. B. Rastetter, 1992. Global change and the carbon balance of arctic ecosystems.

*BioScience*61:415–435.Sitch, S,, et al. 2002. Evaluation of ecosystem dynamics, plant geography and terrestrial carbon cycling in the LPJ dynamic global vegetation model.

*Global Change Biol*9:161–185.Vance, E. D., , and N. M. Nedkarni, 1992. Root biomass in a moist tropical montane forests.

*Plant Soil*142:31–39.van Wijk, M. T., , and I. Rodriguez-Iturbe, 2002. Tree–grass competition in space and time: Insights from a simple cellular automata model based on ecohydrological dynamics.

*Water Resour. Res*38 9:1179. doi:10.1029/2001WR000768.Verseghy, D., , N. A. McFarlane, , and M. Lazare, 1993. CLASS—A Canadian land surface scheme for GCMs. II: Vegetation model and coupled runs.

*Int. J. Climatol*13:347–370.Weltzin, J. F., , and G. R. McPherson, 1997. Spatial and temporal soil moisture resource partitioning by trees and grasses in a temperate savanna, Arizona, USA.

*Oecologia*112 2:156–164.Wright, T. W., 1955. Profile development in the sand dunes of Culbin forest, Morayshire.

*J. Soil Sci*6:270–283.Zhao, M., , A. J. Pitman, , and T. Chase, 2001. The impact of land cover change on the atmospheric circulation.

*Clim. Dyn*17 5–6:467–477.

Sources of the root biomass data used in this study

Values of biome-dependent parameters *a**B**a*

Estimated values of parameter *b* = *a**B*