Water Resources Research

Identifying the optimal spatially and temporally invariant root distribution for a semiarid environment

Authors

  • Gajan Sivandran,

    Corresponding author
    1. Department of Civil, Environmental and Geodetic Engineering, Ohio State University,Columbus, Ohio,USA
      Corresponding author: G. Sivandran, Department of Civil, Environmental and Geodetic Engineering, Ohio State University, 470 Hitchcock Hall, 2070 Neil Ave., Columbus, Ohio 43210, USA. (sivandran.1@osu.edu)
    Search for more papers by this author
  • Rafael L. Bras

    1. School of Civil and Environmental Engineering, Georgia Institute of Technology,Atlanta, Georgia,USA
    Search for more papers by this author

Corresponding author: G. Sivandran, Department of Civil, Environmental and Geodetic Engineering, Ohio State University, 470 Hitchcock Hall, 2070 Neil Ave., Columbus, Ohio 43210, USA. (sivandran.1@osu.edu)

Abstract

[1] In semiarid regions, the rooting strategies employed by vegetation can be critical to its survival. Arid regions are characterized by high variability in the arrival of rainfall, and species found in these areas have adapted mechanisms to ensure the capture of this scarce resource. Vegetation roots have strong control over this partitioning, and assuming a static root profile, predetermine the manner in which this partitioning is undertaken.A coupled, dynamic vegetation and hydrologic model, tRIBS + VEGGIE, was used to explore the role of vertical root distribution on hydrologic fluxes. Point-scale simulations were carried out using two spatially and temporally invariant rooting schemes: uniform: a one-parameter model and logistic: a two-parameter model. The simulations were forced with a stochastic climate generator calibrated to weather stations and rain gauges in the semiarid Walnut Gulch Experimental Watershed (WGEW) in Arizona. A series of simulations were undertaken exploring the parameter space of both rooting schemes and the optimal root distribution for the simulation, which was defined as the root distribution with the maximum mean transpiration over a 100-yr period, and this was identified. This optimal root profile was determined for five generic soil textures and two plant-functional types (PFTs) to illustrate the role of soil texture on the partitioning of moisture at the land surface. The simulation results illustrate the strong control soil texture has on the partitioning of rainfall and consequently the depth of the optimal rooting profile. High-conductivity soils resulted in the deepest optimal rooting profile with land surface moisture fluxes dominated by transpiration. As we move toward the lower conductivity end of the soil spectrum, a shallowing of the optimal rooting profile is observed and evaporation gradually becomes the dominate flux from the land surface. This study offers a methodology through which local plant, soil, and climate can be accounted for in the parameterization of rooting profiles in semiarid regions.

1. Introduction

[2] Noy-Meir [1973], Charney et al. [1975], Idso et al.[1975], and Eagleson [1978] are among those who pioneered the idea that the hydrologic, ecologic, and atmospheric systems are not isolated but rather part of a more complex series of multidirectional interactions. Land surface attributes, such as soil type, vegetation cover, and topography characterize the physical properties and parameters that control these interactions and govern the exchange of water and energy between the surface and atmosphere above it [Noy-Meir, 1973; Charney et al., 1975; Idso et al., 1975; Eagleson, 1978; Pielke, 2001]. The key properties that influence these interactions are the surface temperature, vegetation type and cover, surface albedo, and soil moisture. With these complex interactions in mind, several studies have shown how vegetation state [Chase et al., 2000; Pielke, 2001], root zone available moisture [Milly and Dunne, 1994; Koster and Suarez, 1996], soil moisture [Porporato et al., 2004; Yeh et al., 1984], and albedo [Charney et al., 1975] all greatly impact the modeled atmospheric system [Eltahir, 19961998].

[3] Evapotranspiration is common to both the energy and water balances for the land-atmosphere system. Evapotranspiration is a function of the spatial and temporal dynamics of soil moisture.D'Odorico et al. [2007, p. 1]describes the role of soil moisture as “the environmental variable synthesizing the effect of climate, soil and vegetation on the dynamics of water-limited systems.” Hydrologically, soil-moisture content influences infiltration rates, lateral redistribution, deep percolation below the root zone, as well as surface runoff. Atmospherically, soil moisture in the near surface influences the partitioning of incoming energy into fluxes of sensible, latent, and ground heat. Ecologically, soil-moisture content dictates the amount of moisture available to vegetation for primary productivity, which through transpiration also influences processes in the lower atmosphere [Avissar, 1998; Small and Kurc, 2003; Kurc and Small, 2004].

[4] In arid and semiarid ecosystems, vegetation can be thought of as the integrated response of the hydrologic interactions between terrestrial and atmospheric systems. It is at the interface of these systems, the land surface, where vegetation exerts its influence by impacting the energy and water balances, resulting in a complex multidirectional relationship between soils, climate, and vegetation [Eagleson, 1978; Ivanov et al., 2008a, 2008b].

[5] If we simplify the role of vegetation to that of a pump that returns moisture from the land back to the atmosphere, the efficiency of this pump is determined by the amount of water available within the soil column, the type of vegetation present at the land surface, and the atmospheric demand for moisture. The amount of water available to the pump is strongly controlled by topography, soil texture, and climate. In the absence of vegetation, these factors determine the partitioning of water into surface runoff, evaporation, and deep drainage. The presence of vegetation complicates this partitioning by adding transpiration and canopy interception as additional terms to the water balance and impacting the soil evaporation (through shading) and deep drainage (through plant water extraction).

[6] The depth and distribution of roots within the soil column controls the extent to which soil moisture can be extracted for transpiration. During interstorm periods, once the soil has begun to dry, it is this access to soil moisture deeper in the soil column via this root architecture that vegetation maintains transpiration. This transpiration flux has the ability to significantly alter the water and energy balance by tapping into water from depths out of reach of surface evaporation.

[7] The manner in which vegetation dynamics have been incorporated into modeling studies to date has been through the coupling of plant-soil water stress to transpiration fluxes and carbon assimilation [Bonan, 1996; Ivanov et al., 2008a]. In such models, the plant water uptake (i.e., transpiration) is treated as a sink of soil moisture. The manner by which this sink is extracted from the soil profile can be undertaken with various levels of sophistication. In the simplest land surface models, bucket models, the subsurface is represented as a single layer with the transpiration sink being evenly extracted throughout the soil column. In models that represent the subsurface with multiple soil layers, the rooting architecture of vegetation is described with a root-depth and/or root-shaped parameter that is dictated by the type of vegetation being modeled. These models distribute the transpiration sink based on on the fraction of roots that reside in each soil layer within the root zone.Jackson et al. [2000] details the various model treatments of root distribution, highlighting that rooting parameters are frequently determined independent of local soil texture and climatic region, thus not taking into account the strong influence that soils and variability in climatic forcings have on the partitioning of precipitation at the surface and the flow of moisture through the root zone.

[8] To get around the tedious task of the parameterization of large areas consisting of heterogeneous soil and climatic conditions, some studies have applied the evolutionary principle. This principle states that environmental (abiotic) and competitive (biotic) pressures have resulted in vegetation adapted to the local conditions by expressing traits that maximize the benefit to the plant and improve the probability of success of the individual [Kleidon and Heimann, 1998]. Kleidon and Heimann [1998]applied this philosophy to rooting depths by optimizing the rooting depths for different vegetation classes with a simple terrestrial biosphere model forced with climate data and soil texture information. The purpose of their study was to examine the change in aboveground net primary productivity as a result of using an optimized root-depth parameter rather than the model default values. The authors reported a significant increase in the mean global aboveground net primary productivity, which was accompanied by a similar increase in transpiration. The authors of the study were very cautious in drawing conclusions from these results citing the weaknesses of the ecological model, which did not include any representation of phenology, carbon allocation, stomatal control, or photosynthetic processes. In addition, this was coupled with the primary weakness of the hydrological model, which was a single layer bucket model with no representation of the vertical distribution of soil moisture.

[9] The approach taken by Kleidon and Heimann [1998]could be considered the first step within the modeling community toward incorporating rooting strategies of vegetation into large-scale modeling. Recently, several authors have explored the role of rooting depths and distributions on ecological response [Lai and Katul, 2000; Hildebrandt and Eltahir, 2007; Collins and Bras, 2007; Guswa, 2008; Schenk, 2008; Schymanski et al., 2008; Hwang et al., 2009; Schymanski et al., 2009].

[10] Collins and Bras [2007]explored the rooting strategies of plants in water-limited environments by applying the evolutionary principle that vegetation has the capability to optimize its phenological response to maximize benefit to itself.Collins and Bras [2007]varied mean annual precipitation and the potential evapotranspiration rate to simulate different climatic conditions, and documented how the optimal rooting profile was altered across different soil textures. The key assumptions of their study were that vegetation leaf-area index (LAI) was held constant (LAI equals 1) and that stochastic precipitation was applied as an instantaneous pulse. The constant LAI assumption allowed vegetation to take full advantage of available moisture as it arrives with no accounting for the growth and mortality dynamics that are evident within and between seasons. Seasonal variation in precipitation intensity, interstorm period, and duration were not explicitly taken into account, but rather artificially imposed by setting wet and dry seasons and partitioning the mean annual precipitation between the two. The instantaneous pulse model also significantly impacted the partitioning of rainfall at the surface altering the volume and depth of infiltration. This impact was clearly evident by the large fraction of surface runoff (often greater than half of the rainfall) reported by the authors.

[11] The work by Collins and Bras [2007] was a significant step forward toward improving the representation of subsurface redistribution of soil moisture and its potential impact on the water balance within a distributed hydrologic model. However, because of the simplifying assumptions made with respect to the applied rainfall and LAI, the work could not capture natural variability in climatic forcings, and hence the vegetation response to this variability was not represented.

[12] This study is an extension of the work by Collins and Bras [2007]. The tRIBS + VEGGIE modeling framework was used in order to capture the dynamic evolution of LAI and to better represent storm characteristics at an hourly timestep thereby addressing the assumptions made by Collins and Bras [2007]. The goal of this study was to examine whether the generalized results of Collins and Bras [2007] hold under a more complex representation of vegetation dynamics and climatic forcings and to offer a methodology through which biotic and abiotic conditions are utilized in the determination of rooting parameters.

2. Model Description and Experimental Setup

2.1. tRIBS + VEGGIE

[13] This study uses a fully coupled ecology-hydrology model to study the influence of root distributions. The Triangular Irregular Network (TIN)-based Real-Time Integrated Basin Simulator (tRIBS) model forms the hydrologic framework by which the interactions and feedbacks of the dynamic vegetation model can be simulated. The applications of TINs to represent of topography allows for the movement of moisture and the influence of aspect and slope on the energy balance to be simulated in a quasi three-dimensional framework [Vivoni et al., 2004]. Surface and subsurface flows from one computational element to the next are carried out along the direction of steepest descent. A finite element solution to the 1-D Richards equation for flow within the vadose zone is used to calculate the moisture transfer in the subsurface within an element [Hillel, 1980]. Through the incorporation of the explicit influence of slope and aspect on the land surface, significant spatial and temporal variability within the simulated domain can result through the representation of the lateral redistribution of moisture and the effects of aspect and slope on the energy balance [Garrote and Bras, 1995; Tucker et al., 2001; Ivanov et al., 2004].

[14] The Vegetation Generation for Interactive Evolution (VEGGIE) model incorporates plant physiology and spatial dynamics within the framework of a physically based distributed hydrology model. The coupled model simulates: biophysical energy processes: short- and long-wave radiation interactions, canopy and soil evaporation, energy flux partitioning, and transpiration; biophysical hydrologic processes: interception, stemflow, infiltration, runoff, run on, and unsaturated zone flow; and biochemical processes: photosynthesis, plant respiration, tissue turnover, vegetation phenology, and plant recruitment [Ivanov et al., 2008a, 2008b].

[15] The complexity of a vegetation model is often dictated by the spatio-temporal scale of the question being asked. The VEGGIE model can be utilized at different levels of complexity. The simplest is an annually or seasonally static vegetation cover and the more complex fully dynamic representation, which takes advantage of the strong interaction between the water and energy balances in the coupling of the tRIBS + VEGGIE ecohydrological modeling.

[16] The goal of this study is to examine the manner in which water is partitioned at the land surface in a semiarid environment. Of particular interest is the coupled influence of soil texture and vegetation on this partitioning. To explore the sensitivity of vegetation response to the vertical distribution of roots, this study aims to identify the optimal root profile, defined as the root distribution with the maximum mean transpiration over a 100-yr period, for a generic semiarid C4 grass and shrub on five different soil textures. For brevity, only changes to the model description ofIvanov et al. [2008a]will be presented along with a discussion of the key mechanisms through which the hydrology and vegetation interact within the modeling framework. For complete details on the vegetation-energy-water interaction refer toIvanov et al. [2008a].

2.1.1. Evaporation and Transpiration

[17] The calculation of evaporation and transpiration in tRIBS + VEGGIE is based on the resistivity formulations of Shuttleworth [1979]. The model first divides the computational element into vegetated and bare fractions and then applies different resistivities based on the land surface, canopy, and atmospheric properties [Ivanov et al., 2008a]. Over the bare fraction, evaporation is obtained by,

display math

where Ebare (mm h−1) is the evaporation rate over the bare soil, ρatm (kg mm−3) is the density of moist air, Cp (J Kg1 K−1) is the air heat capacity, λ (J kg−1) is the latent heat of vaporization, γ (hPa K−1) is the psychrometric constant, eatm (hPa) is the atmospheric vapor pressure, Tg (K) is the ground temperature, e* (hPa) is the saturated vapor pressure at the soil surface, raw (s m−1) is the bulk resistance to water vapor fluxes between the ground surface and the atmosphere, rsrf (s m−1) is the soil surface resistance, and hsoil (−) is the relative humidity of the soil pore space.

[18] For the vegetated fraction of the computational element the evaporative flux is divided into evaporation from below the canopy inline image (mm h−1) and evapotranspiration from the canopy itself inline image (mm h−1). These fluxes are computed by altering the formulation above to represent the two surfaces over which the flux with occur between; this impacts the vapor pressure gradient and bulk resistances,

display math
display math

where inline image is bulk resistance to flux between the ground surface and the atmosphere, inline image (s m−1) is bulk resistance to flux between the canopy surface and the atmosphere, Tc (K) is the canopy temperature, and es (hPa) is the vapor pressure of the canopy.

[19] Evaporation from the canopy can be separated into evaporation from intercepted precipitation inline image (mm h−1) and transpiration inline image (mm h−1) by,

display math
display math

where inline image (m s−1) and inline image (m s−1) are conductances. The formulations for the bulk resistances and conductances within the canopy are not presented here and are detailed in the work of Ivanov et al. [2008a]. It is important to note that the transpiration, inline image, accounts for the abiotic controls on moisture flux from vegetation, it does not include the biotic controls the plant imposes as part of its water use strategy.

2.1.2. Infiltration and Surface Runoff

[20] The amount of throughfall, qtf (mm h−1), that infiltrates into the soil is determined by,

display math

where qinf (mm h−1) is the infiltration rate, which is the minimum of the throughfall and the infiltration capacity qinf,max (mm h−1), which is given by,

display math

where, ksat|z=0 (mm h−1) is the saturated hydraulic conductivity of the surface soil layer, b(-) is the slope of the soil water retention curve, Ψb (mm) is the air entry potential, Δz (mm) is thickness over which to calculate the soil moisture gradient selected to be 100 mm, and S|z=100 (−) is the relative moisture content of the top 100 mm of the soil column [Abramopoulos et al., 1988; Entekhabi and Eagleson, 1989; Decharme et al., 2009].

[21] If qtf is greater than qinf,max, then surface runoff will be generated and can be expressed as,

display math

2.1.3. Soil Moisture

[22] This section outlines the manner in which vegetation interacts with the soil-water balance with tRIBS + VEGGIE. The model utilizes a finite element solution to the 1-D Richards equation for flow within the vadose zone.

display math

where θ (mm3 mm−3) is the soil-moisture content,D(θ) (mm2 h−1) is the unsaturated diffusivity, K(θ) (mm h−1) is the unsaturated hydraulic conductivity, inline image (radians) is the slope of the soil surface, t (h) is time, and z (mm) denotes the normal to the soil's surface coordinate.

[23] The unsaturated hydraulic conductivity can be expressed in terms of soil-moisture content [Brooks and Corey, 1964] as,

display math

where inline image (mm3 mm−3) and inline image (mm3 mm−3) are the residual and saturated soil moisture water content, respectively, and λ(−) is the pore-size distribution index.

[24] Similarly, the unsaturated diffusivity can be written as,

display math

[25] The Richards solution incorporates transpiration, infiltration, and evaporation as sources and sinks distributed within the soil moisture profile [Ivanov et al., 2004]. Infiltration is treated as a source of moisture to the top layer of the soil column, evaporation as a sink and is extracted from the topsoil layer of the soil column, and transpiration as a sink distributed by,

display math

where Ti (mm) is the transpiration from layer i, Vf (−) is the fraction of the computational element that is vegetated, which is assumed to be 1 for this study, and Bi (−) is the transpiration efficiency of layer i. ri (−) is fraction of the roots in layer i. The transpiration efficiency for each layer and the soil column are expressed as [Bonan, 1996],

display math
display math

where, BT (−) is the weighted transpiration efficiency of the root zone, θi (mm3 mm−3) is the soil-moisture content of layeri, θw (mm3 mm−3) is the wilting point of the PFT, and θ* (mm3 mm−3) is the soil-moisture content below which moisture stress begins to cause stomatal closure. 1-BT can be considered a representation of plant water stress. This formulation applies the biotic controls on the transpiration flux from vegetation calculated by equation (5). The transpiration efficiency term accounts for the distribution of the root system as well as the water available to the plant at the given time step. It is through this transpiration efficiency that different water use strategies can be applied through the opening and closing of stomata which is controlled by PFT-specific parametersθ* and θw.

[26] This formulation implicitly incorporates a “cost” to deeply rooted vegetation. For example, let us consider a rainfall event on an initially dry soil column (θ < θw). If the event infiltrates to a depth of 30 cm, thereby raising the surface layers' soil-moisture content (θ > θ*), then a plant with a uniform rooting distribution to a depth of 30 cm would transpire at the atmospheric demand (BT= 1). However, if we consider another plant with a rooting depth of 60 cm, this plant would transpire at half the rate of the shallow-rooted plant since the bottom half of the root biomass received no moisture (BT = 0.5).

2.1.4. Moisture Controls on Vegetation Dynamics

[27] The transpiration efficiency term, BT, is utilized not only in determining the flux of transpiration but also to couple the water and carbon processes within the model. Photosynthesis and foliage turnover are directly impacted by transpiration efficiency. The transpiration efficiency regulates the photosynthetic processes by reducing the area over which gas exchange can occur. In particular, the maximum catalytic capacity of photosynthetic enzyme Rubisco is scaled by BT [Collatz et al., 1991; Ivanov et al., 2008a]. The model framework also allows for stress-induced foliage loss which increases as water stress approaches the wilting point [Levis et al., 2004; Arora and Boer, 2005].

2.2. Model Setup

[28] The simulations described below were conducted on a single computational element with a slope of 5%. There is no groundwater table present in the model and a flux boundary condition determined only by the hydraulic conditions of the last soil layer is applied. No lateral subsurface flow is permitted. Only one PFT is simulated on the single element with the vegetative cover being related by the dynamic LAI which is evolved by VEGGIE. With only one PFT present within the computational element no competitive interactions between different functional types are considered.

2.2.1. Site Description

[29] The United States Department of Agriculture (USDA), Agricultural Research Service (ARS), maintains the Walnut Gulch Experimental Watershed (WGEW). The watershed is part of the San Pedro River Basin and is located near Tombstone, Arizona. Instrumentation began at the watershed in 1953, focusing on the measurement of precipitation and streamflow. The watershed has a network of over 100 rain gauges, two eddy flux towers, and ∼20 soil-moisture measurement locations. In addition to the instrumentation, extensive soil and vegetation surveys have been undertaken. Because of the volume of hydrologic data available at this location as well as the length of these data records, the watershed has been the focus of several hydrologic studies, many of which were featured in a special issue ofWater Resources Research (vol. 44, 2008), which was dedicated to the work conducted at this watershed.

2.2.2. Climatic Forcings

[30] Several studies have utilized stochastic climate forcings to drive point-scale representations of water balance and associated interactions with vegetation. These studies include the response of plants to soil moisture deficit [Porporato et al., 2001], plant suitability to climate and soil conditions [Laio et al., 2001; Porporato et al., 2003], and the coexistence of different species and functional types [van Wijk and Rodriguez-Iturbe, 2002; Fernandez-Illescas and Rodriguez-Iturbe, 2004]. These studies, albeit point scale, elucidate several of the key processes that must be considered as well as those that need an accurate spatial characterization.

[31] The abundance of grasses and shrubs in semiarid regions are strongly correlated to rainfall. For this reason it was critical that the climate forcings used to drive tRIBS + VGGIE were representative not only of macroscopic statistics such as monthly and annual rainfall volumes, but that they also captured the rainfall statistics such as interstorm period, storm duration, number of wet days, and average storm volumes. For this study, a stochastic weather generator [Ivanov et al., 2007] that had already been validated for the semiarid region of the southwest of the United States was used to force the tRIBS + VEGGIE model. To create the appropriate stochastic inputs, hourly rainfall for the period 1956–2008 from six rain gauges (rain gauge 4, 13, 42, 44, 60, and 68) at the Walnut Gulch Experimental Watershed (WGEW) in Arizona were used to derive rainfall statistics representative of a semiarid environment. The observed mean annual rainfall over this period was 331 mm, with a summer monsoon average (July, August, and September) of 188 mm. The maximum and minimum annual accumulations over this period were 528 mm and 162 mm. These rainfall statistics from the WGEW were used to parameterize the stochastic weather generator, which was then run for a 200-yr period. The generated rainfall time series returned a mean annual rainfall of 336 mm with a summer monsoon average of 198 mm. The wettest and driest years were 565 mm and 145 mm, respectively, confirming that the stochastic rainfall data matched well to the observed statistics.

[32] Figure 1 compares the statistics for the observed and stochastically generated rainfall characteristics. The generated climate statistics not only capture the mean behavior of storm duration, storm intensity, and interstorm period, but also capture the variability in these characteristics.

Figure 1.

Observed rainfall statistics derived from a 56-yr data record from Walnut Gulch Experimental Watershed, Arizona (red bars). Stochastically generated rainfall time series (blue bars). Error bars indicate the 25% and 75% quartile ranges; the squares show the mean value.

Figure 2.

Grass-dominated Kendall sub-basin (left) and shrub-dominated Lucky Hills sub-basin (right).

2.2.3. Soil Textures

[33] To explore the role that soil texture has on rainfall partitioning and vegetation growth, five generic soil textures were used to represent the expansive variability in natural soils. A sandy soil was used to represent the highly conductive, low-field capacity end of the soil texture spectrum and a clayey soil with very low conductivity and high-field capacity was used to represent the other. The parameters associated with these soil textures were obtained fromRawls et al. [1982] and are presented in Table 1. The soil at the WGEW are typically coarse-loamy, mixed, thermic Ustochreptic Calciorthids [Heilman et al., 2008]. Of the soil textures chosen for simulation under a WGEW climate, only clay loam and clay textures are likely to generate significant quantities of surface runoff.

Table 1. Soil Parameters for Five Generic Soil Typesa
ParameterSandSandy LoamLoamClay LoamClay
  • a

    Table data is from Rawls et al. [1982]. Where Ksat is the saturated hydraulic conductivity in the direction normal to the soil surface, θr and θsare the residual and saturated soil-moisture content,λ is the pore size distribution; ψb is the air entry bubbling pressure, and θfcis the soil-moisture content at field capacity.

Ksat (mm h−1)210261331
θs (mm3 mm−3)0.4370.4530.4630.4640.475
θr (mm3 mm−3)0.020.040.060.050.15
λ(-)0.5920.3220.2200.1940.131
ψb (mm)−72.6−147.0−111.5−259.0−373.0
θfc (mm3 mm−3)0.0310.1120.1750.2110.330

2.2.4. Plant Functional Types

[34] The dominant shrub species present at the site are creosotebush (Larrea tridentata), hitethorn Acacia (Acacia constricta), tarbush (Flourensia Cernua), and desert zinnia (Zinnia pumila); the dominant grasses are sideoats grama (Bouteloua curtipendula), black grama (Bouteloua eriopoda), three-awn (Aristida sp.), and cane beard grass (Bothriochloa barbinodis) [King et al., 2008]. These species are grouped within the model and represented as C3 shrubs and C4 perennial grasses (Figure 2). The vegetation parameters needed for the VEGGIE model were taken from available literature [Sellers et al., 1996; Ivanov et al., 2008a, 2008b]. See online supplementary material for parameter sets used for the VEGGIE model.

[35] The major difference in the parameterization of shrubs and grasses is related to the allocation of assimilated carbon. Shrubs maintain three carbon pools (roots, shoots, and leaves), whereas grasses maintain two (roots and leaves), which results in a greater respiration demand on the PFT. Shrubs therefore have less carbon to be allocated to leaves resulting in a slower green-up rate to early season rainfall events [Scott et al., 2000].

[36] As discussed above, transpiration is strongly controlled by θ* and θw. Grasses reach the point at which stomata start to close (θ*) and the wilting point (θw) is at higher soil-moisture levels when compared to shrubs. This difference in drought tolerance results in shrubs being capable of maintaining roots in soil layers with lower soil moisture levels without adversely impacting transpiration efficiency.

2.3. Experimental Setup

[37] A series of numerical simulations were conducted testing two forms of spatially and temporally invariant rooting schemes: uniform and logistic. For each rooting scheme, two PFTs and five soil textures were examined.

2.3.1. Uniform Root Profiles

[38] Uniform root profiles can be considered the simplest representation of root distribution within the soil column. The root distribution is controlled only by the maximum root-depth parameter, which is invariant in time and space for each simulation. Several terrestrial models still utilize this simplistic representation, such as BATS [Haxeltine and Prentice, 1996], CENTURY [Parton et al., 1993], and TEM 4 [McGuire et al., 1997], hence it was desirable to quantify the sensitivity of the water balance to this type of rooting profile. The root fraction ri within each soil layer can be represented simply as,

display math

where dzi is the thickness of the soil layer and Zr is the rooting depth parameter.

[39] Simulations were conducted varying the rooting-depth parameter from 0.1 m to 2.5 m in increments of 0.1 m, resulting in 25 simulations for each PFT soil-texture combination. Simulations were run for a total of 200 yr. The first 100 yr were used for model spin-up to ensure no model initialization influence and the second 100 yr of simulation were used for analysis. The simulation that returned the maximum value of the mean annual transpiration (as a proxy for maximum vegetation growth) was chosen as the optimal rooting profile.

2.3.2. Logistic Profiles

[40] Through a metanstudy of field studies measuring the root distribution of vegetation, Schenk and Jackson [2002]found that the shape and distribution of the root profiles could be represented best using the logistic dose-response curve (LDR),

display math

where Froot(z) () is the cumulative root fraction from the surface to a depth z (mm), D50 (mm) is the depth at which 50% of the root mass is above, and c () is the shape parameter which can be related to the D50 and D95 by [Collins and Bras, 2007],

display math

where, D95 (mm) is the depth at which 95% of the root mass is above. The root fraction, ri, of an individual layer, i, can be calculated by,

display math

where, zi is the depth of the layer of interest, and zi−1 is the depth of the layer above it. Since the subsurface discretization has a finite number of layers, D95 was considered to represent the maximum root depth for modeling purposes.

[41] Simulations were conducted to quantify the sensitivity of the water balance to different combinations of the D50 and D95root distribution parameters used in the logistic profile resulting in 80 simulations for each PFT soil-texture combination.Figure 3 illustrates the parameter state space that was explored and provides examples of logistic root profiles. Simulations were run for 200 yr, and as above only the last 100 yr were used for analysis. The simulation that returned the highest mean annual transpiration was designated as the optimal rooting distribution.

Figure 3.

(Left) D50 and D95 parameter state space. (Right) Examples of the vertical root profile corresponding to the colored circles in the left panel.

[42] Schenk and Jackson [2002]report typical rooting parameters for water-limited grass-dominated ecosystems ranging from 0.1 m to 0.3 m for theD50 parameter and 0.6 m–1.3 m for the D95parameter. For semidesert shrub-land ecosystems,Schenk and Jackson [2002] report a D50 of 0.3 m and a D95 of 1 m–1.6 m. It is important to note that even though the D50 parameter was tested to a depth of 1 m and the D95parameter to a depth of 5 m, these ranges are higher than the observed values for water-limited ecosystems and were used to provide an upper bound to the parameter space.

3. Results and Discussion

3.1. Uniform Profiles

[43] Figure 4shows the 100-yr mean values of the water balance components for each of the PFT soil-texture experiments using the uniform rooting scheme. Each panel of the figure is composed of the results obtained for each of the 25 values of the root-depth parameter space. The red lines in the figure indicate the rooting-depth parameter of the optimal rooting profile.

Figure 4.

Mean annual water balance over a 100-yr simulation for five soil textures and two plant functional types at Walnut Gulch Experimental Watershed. Evap: evaporation from soil surface; Intc: canopy interception loss; Trans: transpiration; Srf: surface runoff; Drain: deep drainage to groundwater. Red lines identify the value of the maximum rooting depth parameter corresponding to the maximum mean transpiration over the simulation period.

[44] Soil texture has a significant role in the partitioning of precipitation at the surface as well as the movement of moisture through the soil column. One of the largest controls on evaporation, drainage, and generation of surface runoff is the permeability of the soil. Well-drained, high-conductivity soils (sands) allow large volumes of water to infiltrate to depth. This results in moisture moving quickly away from the soil surface, thus reducing the soil evaporation. This high conductivity also results in the soil being able to infiltrate storms of high intensity and volume, consequently, producing little surface runoff and significant deep drainage. Poorly drained soils (clays) retain moisture in the surface layers due to a very low hydraulic conductivity decreasing the infiltration of received rainfall. This results in a significantly larger fraction of evaporation being produced and reduces deep drainage. The low conductivity also results in the surface layer quickly becoming saturated within storm events, generating significant fractions of surface runoff.

[45] From the perspective of plant water availability, differences in the partitioning of water have direct implications for the optimum rooting depth. In the near surface layers, the two sinks competing for infiltrated moisture are evaporation and transpiration. In highly conductive soils, the infiltrated water moves rapidly through the evaporative zone, quickly reaching a depth at which evaporation no longer has the ability to extract water. Therefore, from the plant's perspective, highly conductive, well-draining soils result in minimal losses to the atmosphere and thus the greatest volume of water entering the root zone. However, the high conductivity and low field capacity also acts to drain the root zone, resulting in a short residence time for moisture within the root zone. Therefore, on highly conductive soils with low field capacities, it is advantageous for plants to exhibit deep rooting profiles in order to capture as much moisture as possible before the moisture drains through the column.

[46] By contrast, low conductivity soils such as clays lose large volumes of water to surface runoff during rainfall events. Because of their slow drainage rates and high field capacities, moisture remains near the surface and available to the atmosphere to satisfy evaporative demand for a longer period than in highly conductive soils. The surface soil layers also have the highest dynamic range of soil moisture, with highs close to saturation and lows near the residual soil-moisture content. This high variability creates considerable stress to the plant. Once moisture has managed to infiltrate below the evaporative zone, moisture in the root zone has a much longer residence time and the dynamic range of soil moisture reduces. The deeper soil layers have less moisture than the surface layers but also experience less variability, creating less stress for the plant. Therefore, on low conductivity soils the plants face a tradeoff: the root is close to the surface where there is moisture but faces stress and competes with the evaporative demand, or the root is deeper in the soil column and experiences less stress but has less access to moisture.

[47] Figure 4 clearly illustrates the effects of soil texture. Across both plant types, the optimal rooting depth increases with increasing conductivity. The results also show that the magnitude of transpiration is higher for sands than for clays, such that the deeper rooting profile also correlates with more transpiration, since the high conductivity soils lead to a higher volume of available moisture. This result is consistent with the inverse texture hypothesis [Sala et al., 1988], which states that in semiarid regions, sandy soils are more productive than clayey soils due to lower losses to surface runoff and evaporation.

[48] A common trend across all soil textures and PFTs is that as maximum rooting depth increases, the fraction of evaporation also increases. This can be considered a direct result of representing the root profile with a uniform distribution. As the maximum root depth increases, a smaller proportion of the plants roots are in each soil layer. As the root density decreases, the plant's ability to extract water from that particular layer also diminishes. The increase in evaporation as roots get deeper is a consequence of a lower root density in the near surface soils layers. This reduction in density results in less moisture lost to transpiration from the surface layers, thereby increasing the volume of water available to the atmosphere for evaporation.

[49] The influence of rainfall timing can be seen when examining the simulations under an Arizona climate at the WGEW. With rainfall and the growing season both occurring during the summer months at WGEW, there is a “use it or lose it” strategy being employed by the vegetation due to the high evaporative demand [Scott et al., 2000]. This results in shallow rooting profiles across all soil textures and both PFTs at WGEW.

3.2. Logistic Profiles

[50] Simulations using a logistic profile involved more than four times the number of realizations than for the uniform profile because of the extra variable in the parameter space. Therefore, for brevity, only results for the transpiration component of the water balance are shown here since transpiration is considered a proxy for plant productivity.

[51] Figures 5 and 6show the mean annual transpiration over the 100-yr simulation period for the PFT soil-texture combinations. Each black circle in these figures represents one simulation using the associatedD50 (x-axis) andD95 (y-axis) rooting parameter values. The rooting profile that maximizes the mean annual transpiration (white circles in the figures) will be referred to as the optimum rooting profile.

Figure 5.

Mean annual transpiration (mm yr−1) for a grass on five soil textures over a 100-yr simulation for Walnut Gulch Experimental Catchment, Arizona. Solid black circles indicate parameter combinations simulated; white circles are the logistic profileD50 and D95 parameter combinations that resulted in the maximum mean transpiration, i.e., the solid white circles indicate the location of the optimal rooting profile. Note: each subplot color is scaled to match the maximum and minimum values of each panel.

Figure 6.

Mean annual transpiration (mm yr−1) for a shrub on five soil textures over a 100-yr simulation for Walnut Gulch Experimental Catchment, Arizona. Solid black circles indicate parameter combinations simulated, Solid white circles ares the logistic profileD50 and D95 parameter combinations that resulted in the maximum mean transpiration, i.e., the white circles indicate the location of the optimal rooting profile. Note: each subplot color is scaled to match the maximum and minimum values of each panel.

[52] Similar to the previous results for the uniform profiles, the influence of soil texture on the logistic rooting profiles is clearly evident in Figure 5. Moving from the sand (left panel) to the clay (right panel) soil textures, i.e., high conductivity/low field capacity to low conductivity/high field capacity, the mean annual transpiration alters significantly from 200+ mm on sand to 50 mm for clay soils. These differences can be explained as the result of the different partitioning of precipitation at the surface. In high conductivity/low field capacity soils, rainfall events are quickly infiltrated to a depth below the evaporative zone, providing a large moisture resource for the roots. There is little need to compete with evaporation at the surface due to the low residence time of soil moisture in these layers, as it is more beneficial to the plant to limit the amount of moisture lost below the root zone as a result of percolation. Consequently, we see deep optimal-rooting profiles in such soils. The low field capacity translates to a very short residence time of moisture in any given soil layer. From a plant perspective, this results in moisture levels dropping close to wilting point quickly.

[53] By contrast, the optimal D50 and D95 root parameters were shallow on lower conductivity/high field capacity soils. On low conductivity soils, a significant fraction of rainfall is lost to runoff creating less available moisture for infiltration. In the case of clayey soils, if the vegetation were to employ the same strategy as plants on sandy soils, the number of events that would reach the bottom of the root zone would be rare. The moisture stays within the evaporative zone longer, requiring the plants to compete with evaporative demand (Figure 5). Because potential evaporation rates are much higher than potential transpiration rates in semiarid regions and for these PFTs, less moisture is available for the plant activity. Consequently, vegetation found on such soil textures distributes their roots in order to capture the soil moisture from the near surface layers before evaporation can extract it. Because of the high evaporative losses of clayey soils, vegetation must employ a water-use strategy to deal with the higher variability in plant-available water.

[54] Another observation from Figure 5is that even on the sandy soils, below a certain root depth (∼2.5 m) transpiration begins to decrease. The deeper a root system, the larger the volume of soil it occupies, which translates to larger potential resource for the plant. However, an implicit cost of rooting deeper is maintaining less root fraction in the near surface. In semiarid regions evaporation is a significant flux of moisture from the soil column, and by reducing root uptake in the near surface, the plant no longer competes for near-surface moisture. Therefore, at some depth, the marginal gain in soil moisture storage by deeper roots is less than the opportunity cost of reducing the root fraction higher in the column. It is important to acknowledge that within the model the carbon cost of maintaining roots is the same irrespective of where within the soil column they are simulated. A need exists to better understand how these costs vary with depth in order to more accurately represent the tradeoffs involved with rooting very deeply.

[55] Figure 6 illustrates the mean annual transpiration for the shrub PFT at the WGEW. A similar trend across the soil texture classes can be observed for shrubs as was observed for grasses. The shrub PFT consistently transpires more efficiently with deeper root profiles than the grass PFT. This is a direct result of the different life strategies of these two PFT. Shrubs allocate the majority of their assimilated carbon to the maintenance of root and stem “infrastructure”; leaf and seed production is not prioritized as is the case with grasses. Examination of the plots in Figure 6 shows that this PFT favors a combination of a shallow D50, allowing for the extraction of moisture from the surface soil layers when available, and a deep D95 allowing for the ability to avoid severe water stress by utilizing soil moisture that manages to percolate to deeper layers [Scott et al., 2000].

4. Summary and Conclusion

[56] As this work builds on the study of Collins and Bras [2007], it is useful to compare these two papers. Both studies obtain very similar optimum profiles and the same soil texture control on these optima is observed. However, there is a significant difference in the sensitivity of the water balance fluxes between the two studies. The results of this study indicate a significant variability in optimal transpiration flux across soil textures, with the grass experiments spanning a range of 229 mm on a sandy soil compared to 54 mm on clayey soil. This is in stark contrast to the difference of only 20 mm between the transpiration flux of these two soil textures by Collins and Bras [2007]. The source of this lack of variability is attributed to a static LAI assumption by Collins and Bras [2007], which may have limited transpiration in cases where the land surface could support an LAI of greater than 1 and overestimated transpiration under conditions that would result in significant stress induced foliage turnover. The dynamic vegetation allows for the evolution of the LAI and thus produces a much larger range of annual transpiration fluxes.

[57] Another significant difference between the two results is the production of surface runoff. Collins and Bras [2007] indicate that ∼40% of all rainfall results in surface runoff. This result may be an artifact of either the instantaneous pulse representation of storm depth or the infiltration scheme applied in their study. Regardless, this flux dominates the water balance. Whereas the shallow rooting profiles by Collins and Bras [2007] may have been controlled by the surface runoff flux limiting the volume of precipitation that infiltrates the soil, in this study it was evapotranspiration that dominated the moisture balance of the soil (Table 2). It is important to note that the low contribution of the surface runoff flux in this study may be the result of the 1 h rainfall intensities used to force tRIBS + VEGGIE. At this temporal resolution, any high-intensity, short-duration events that may produce surface runoff will be smoothed.

Table 2. Annual Water Balance Fluxes Corresponding to Optimal Logistic Rooting Profile Simulationa
Water Balance TypeSandSandy LoamLoamClay LoamClay
  • a

    Profile simulation from each soil texture is for the grass plant functional type. Fluxes were rounded to whole numbers.

Evaporation (mm yr−1)82132162197215
Transpiration (mm yr−1)22918415510954
Interception (mm yr−1)15141295
Surface runoff(mm yr−1)0011347
Deep drainage (mm yr−1)4002.510

[58] The results of this study demonstrated the sensitivity of transpiration to the selection of rooting parameters within a soil texture class. Across soil textures, the magnitude of the transpiration flux of the optimal rooting profile also showed significant variability, therefore altering the partitioning of rainfall significantly. There also existed a subtle difference between PFTs with the identification of deeper shrub optimal rooting profiles than that of grasses, suggesting a difference in life strategies.

[59] The characterization of subsurface parameters, such as root depths, is at present a very difficult task to achieve over large spatial areas. This study proposes a methodology by which data of the local climate characteristics, dominant plant functional type, and the soil texture can be utilized through a brute force simulation technique to identify the optimal rooting parameters for the local biotic and abiotic conditions. The methodology relies on an evolutionary principle that plants of semiarid regions have evolved to maximize the use of resources available. This methodology would replace the current plant-dependent parameterization by also taking into account the role of abiotic conditions in the determination of root parameters.

Ancillary