Journal of Geophysical Research: Atmospheres

Effects of paleoclimate and time-varying canopy structures on paleowater fluxes

Authors


Abstract

[1] We combined a long-term (18,000 years) climatological record and time-varying vegetation conditions to evaluate the role that climate change and vegetation might play in paleowater fluxes in arid settings. The HYDRUS-1D model, which solves Richards Equation for variably saturated flow, the convection-dispersion equation for chloride transport and the heat flow equation, was used to simulate water flux and chloride (Cl) transport. Six distinct case studies were compared for different boundary conditions and root distributions. A Mojave Desert–type canopy including evergreens, drought deciduous shrubs, annuals, grasses, and succulents was used as representative vegetation to transpire soil water and was modeled using ground cover percentage and leaf area index (LAI) as the bases for partitioning evapotranspiration (ET). The results showed that under water limited conditions, realistic root zone distributions and climate sequences (including extreme events) were both needed to simulate the accumulation of Cl in Mojave Desert soils. Results also showed that increasing precipitation intensity affected paleowater fluxes. However, contrary to the results of other researchers, we found that simulated chloride bulges were located at depths of around 20–30 cm, rather than at the base of the root zone, if current and normal climate conditions were applied. Moreover, the climatic shift beginning in the late Pleistocene was not the major reason for the chloride accumulation.

1. Introduction

[2] In arid regions where evapotranspiration (ET) rates are high and precipitation rates are low, water resources are often limited. How water is partitioned between surface runoff, recharge, evaporation, and transpiration depends on linkages between climate, vegetation, soil properties, and water balances. Among these, climate is the key driving force affecting the desert ecosystem. If the linkages between past climatic and ecohydrological responses can be understood, it might be possible to predict how desert ecosystems will respond to potential future climate changes.

[3] Studies of global climate change [Bartlein et al., 1998; Kutzbach et al., 1993, 1998], regional climate change [Smith et al., 1997], and the associated changes in soil properties [McDonald et al., 1996, 2003; Young et al., 2004] and vegetation structure [van Devender and Spaulding, 1979; Spaulding, 1985, 1990] have shown that surface conditions vary through time and are effected by changes in environmental conditions. These sometimes subtle relationships can be included in long-term simulations, where past atmospheric changes are used as boundary conditions to evaluate variations in soil water status and water partitioning.

[4] However, long-term (decadal) soil water data are mostly unavailable. So, investigators have used different solute accumulation rates, specifically chloride, to derive long-term averaged water fluxes [Allison and Hughes, 1978; Allison et al., 1994; Murphy et al., 1996; Tyler et al., 1996] in thick vadose zones. The chloride mass balance (CMB) technique [Allison and Hughes, 1978; Scanlon, 1991; Cook et al., 1992; Ginn and Murphy, 1997; Scanlon et al., 2003] is one method that can be used to evaluate average paleorecharge rates (e.g., water that percolates below the root zone). As indicated by Cook et al. [1992, p. 2728], “variations in solute and isotopic concentration within a soil profile can arise from a variety of causes, other than recharge or climatic variability.” An increase in chloride concentration could be caused either by decreased precipitation or by increased root growth. For this reason, merely detecting changes in chloride concentration cannot provide sufficient information to explain differences in water fluxes.

[5] Forward simulations of water and solute movement can be rather complex depending on the simulation time and variability of boundary conditions. In some cases [e.g., Ginn and Murphy, 1997; Walvoord et al., 2002b; Scanlon et al., 2003], variable boundary conditions and the complexity of climate change were omitted to simplify the problem. However, errors in estimated fluxes could occur if time-invariant boundary conditions and climate are assumed, because dynamically changing boundary conditions could significantly affect water balance conditions. In addition, long-term averaged precipitation neglects possible episodic flushing, which is very important to the redistribution of soil moisture in arid regions [Sandvig and Phillips, 2006]. These different processes may result in different soil water profiles, even given similar boundary conditions. Though uniform boundary conditions may be valid for shorter time frames (i.e., decades), they could have larger impacts on soil water fluxes as time periods lengthen, especially when those time periods span major climatic and plant community structural switches.

[6] The purpose of this study is to conduct forward numerical modeling to evaluate the effects of past climate changes on soil water and solute (chloride) concentration, while considering environmental factors including the evolution of canopy structure and plant rooting patterns in a Mojave Desert ecosystem. The modeling period in this study is from 18,000 years ago (18 ka) to present. This period was chosen because of the large environmental changes that occurred after the last glacial period (i.e., late Pleistocene and early Holocene Periods). The environmental factors listed above were reconstructed on the basis of the knowledge of palaeoclimatology. The paper will discuss the roles played by these factors in a climate-plant-soil water system, and will include possible explanations for the existence of chloride profiles in arid soil profiles.

2. Methods

2.1. Physical Model Description

[7] The soil material used in this research is located at the Amargosa Desert Research Site (ADRS), located in the Mojave Desert near Beatty, Nevada, about 20 km east of Death Valley National Park, California, USA. Five significant soil layers within the upper 5 m have been visually identified using texture, cohesiveness, and color. Physical properties of these five soil layers were described by Andraski [1996]. Following the assumptions of Scanlon et al. [2003], the hydraulic properties of the top 100-cm soil for this study were considered to be representative of the soil material (described below) in this study (Table 1). Groundwater at ADRS is present at depths ranging from 85 to 110 m below land surface [Fischer, 1992]. The bottom of the model domain was set as 15 m, significantly above the known groundwater level. This shallower model domain was also chosen because most of the Cl information is preserved in this depth range. For example, Stonestrom et al. [2003] examined Cl concentration profiles at nine locations with different upper boundary conditions in the Amargosa Desert and showed that most of the variations in concentration were restricted to the top 15 m of the soil.

Table 1. Hydraulic Properties of Representative Soil [Andraski, 1996; Scanlon et al., 2003]
USDA TextureSand, %Silt, %Clay, %Ks, cm d−1θs, cm3 cm−3θr, cm3 cm−3α, cm−1n
Loamy sand8014643.00.290.0260.0261.42

2.2. Numerical Model Description

[8] This study requires a numerical approach that considers liquid water flow, thermal and isothermal vapor transport, chloride transport, root water uptake by plants, and evaporation from the soil surface. These processes account for some of the complexities of the near-surface water balance of arid and semiarid landscapes. The numerical package (HYDRUS-1D [Simunek et al., 2005]) used to simulate these processes accounts for these processes in one-dimensional variably saturated media. HYDRUS-1D approximates the solution to the governing equation for water flow:

equation image

where θ is the volumetric water content, t is time, h is the water pressure head, x is a spatial coordinate, and K is the unsaturated hydraulic conductivity, which is a function of the saturated hydraulic conductivity (Ks) and water content. The parameter, S, in equation (1) represents a sink term, which accounts for the uptake of soil water by vegetation [Feddes et al., 1978]:

equation image

where S(h) is the water uptake rate, α is a water stress response function, and Sp is the potential water uptake rate. The value S(h) is partitioned into each layer according to the depth-specific root density. The van Genuchten [1980] and Mualem [1976] representations for unsaturated hydraulic properties used in this study are given by:

equation image
equation image
equation image

where θr is residual soil water content, θs is saturated soil water content, and α and n are parameters for the water retention curve. To aggregate the vegetation effects of each plant type, a weighted linear combination was used to calculate a “lumped root density,” as shown by:

equation image

where N is an index for the growth form (i.e., plant type), r is the root density of each growth form, and c is the percent ground cover for each growth form. We discretized the 15-m soil column into 175 adjoining elements. Element thickness ranges from 2.5 cm (at ground surface) to 50 cm (at bottom of domain). Equation (6) is applied to each element, allowing for transpiration losses in each element as well.

[9] The governing equation in HYDRUS-1D for solute transport is given by:

equation image

where Dw is the dispersion coefficient, c is the solute concentration, and q is the volumetric flux density. The dispersion coefficient can be represented by:

equation image

where DL is the longitudinal dispersivity, Dw is the diffusion coefficient in free water, and τw is a tortuosity factor in the liquid phase. According to Millington and Quirk [1968], the tortuosity factor τw can be described as:

equation image

We note that a few assumptions are intrinsic when applying this model: (1) chloride in the soil is in steady state with no “sources” (i.e., no mineral dissolution or formation), or “sinks” (i.e., no interaction with soil or uptake by vegetation) and (2) the effect of salt concentration gradient on the driving force of water movement can be disregarded because the modeled site does not have clay-rich soil [Hillel, 1998]. Table 1 shows that the soil profile contains only 6% clay.

[10] The governing equation in HYDRUS-1D for thermal and isothermal vapor flow was provided by Scanlon et al. [2003]:

equation image

where qvh is isothermal vapor flux, qvT is thermal vapor flux, Kvh is isothermal vapor conductivity, KvT is thermal vapor conductivity, D is vapor diffusivity in soil, ρw is water density, ρvs is saturated vapor density, M is molecular weight of water, g is gravitational acceleration, R is gas constant, Hr is relatively humidity, η is an enhancement factor, and T is temperature. Given the tortuosity, volumetric air content, saturated water content, and mass fraction of clay in soil, parameters D and η can be solved [Scanlon et al., 2003].

[11] The governing equation for heat transport can be represented by a convection-dispersion-type equation:

equation image

where λ(θ) is the apparent thermal conductivity of the soil, and Cp(θ) and Cw are the volumetric heat capacities of the porous medium and the liquid phase, respectively. De Vries [1963] expressed the volumetric heat capacity as the sum of individual components of the bulk soil. The apparent thermal conductivity λ(θ) can be expressed as:

equation image

where λ0(θ) is the thermal conductivity of the porous medium, βt is the thermal dispersivity and q is the velocity. The Chung and Horton [1987] model was used to calculate λ0(θ). Parameters in equations (11), (12) and in Chung and Horton's [1987] model are given in HYDRUS-1D. In this study, we used the default parameters for loam.

2.3. Boundary Conditions

2.3.1. Paleoclimate Reconstruction

[12] Climate change can significantly impact the percent coverage and types of vegetation that are likely to flourish through long time periods. For example, Walvoord et al. [2002a] indicated that, when paleoclimate shifted to a warmer and drier climate, the mesic vegetation was replaced by xeric vegetation. Recently, various paleoclimate reconstructions have been studied in the Sierra Nevada area and in the deserts of the southwestern United States [Spaulding, 1985; Kutzbach et al., 1996, 1998; Smith et al., 1997; Bartlein et al., 1998; Sharpe, 2004]. These results support the understanding that a significant climate switch occurred in most regions in the southwestern United States after the Last Glacial Maximum (LGM), when decreases in effective environmental moisture were observed in paleoclimate proxies. Some studies [e.g., Tyler et al., 1996] concluded that this shift to drier climate conditions allowed chloride to accumulate in the soil profile. In the southwestern United States, a “wetter” LGM could be explained by either higher precipitation rates [McDonald et al., 1996], or a combination of either similar or lower precipitation rates and lower evapotranspiration rates due to the lower mean temperatures [Brakenridge, 1978; Galloway, 1983]. More recent evidence [e.g., Menking et al., 2004] supports the former hypothesis. Spaulding [1985] suggested that the precipitation in the LGM in the current Mojave Desert was about 1.4 times larger than the current value. McDonald et al. [1996] indicated that the annual precipitation in the LGM would have to increase by as much as 100% to simulate the observed soil carbonate. A study conducted at the Nevada Test Site, Nevada [Thompson et al., 1999] derived precipitation values of 2.5X and 2.6X (X: modern precipitation multiplied by this number) in the LGM. We referenced this study because its location is in our area of interest.

2.3.1.1. Construction of Precipitation Record

[13] A temporal resolution on the order of thousands of years is common in paleoclimate proxies; however, numerical models typically require higher temporal resolution, especially given the potential impact of short-term, episodic rainfall on recharge [e.g., Sandvig and Phillips, 2006]. If annual precipitation rates were evenly distributed in the model over an entire year, given the time step of 1 d, soil water recharge would be zero, because the extremely low (averaged) daily precipitation rates will be much smaller than daily evapotranspiration rates. Therefore, we require time steps that can account for those specific time periods when the precipitation rate is higher than the evapotranspiration rate. This allows water to (perhaps temporarily) percolate to depths below the root zone, and potentially recharge deep soil material.

[14] To account for the extreme temporal heterogeneity in precipitation patterns found in the southwestern U.S., the Community Climate Model (Version 1, CCM1) [Kutzbach et al., 1996; Bartlein et al., 1998] was used to generate paleoclimate data from 18 ka to present in 3,000-year sequences. CCM1 quantitatively interprets past climatic changes and takes into account intra-annual variations. These time periods were designated as: 18–15 ka, full-glacial environment; 15–12 ka, late-glacial environment; 12–9 ka, early-Holocene environment; 9–6 ka, mid-Holocene environment; and 6 ka to the present, late Holocene environment. The paleoclimate proxies in the CCM1 model include the stratigraphic pollen records, plant macrofossil assemblages from packrat middens, and lake level records. The climate generated from 18 to 15 ka shows an annual temperature 5.7°C lower than the present value and an annual precipitation 13% higher than the present value.

[15] Using paleoclimate reconstruction from Thompson et al. [1999], we replaced CCM1 precipitation values from 18–15 ka and 15–12 ka to 2.5X and 2.6X, respectively. The simulation time interval is daily in this reconstruction. However, the simulated precipitation is the same every day within each month for successive years (i.e., no within-month variation from year to year). Thus, to make the CCM1 simulations more representative of climate in the northern Mojave Desert, a few steps were taken.

[16] (1) All climate data were averaged over four CCM1 model grids (bounded between 120 and 112.5°W longitude and 37.77 and 33.77°N latitude) surrounding the northern Mojave Desert. (2) The number of days in each month was multiplied with the daily precipitation, producing the monthly precipitation. (3) The monthly precipitation record for each 3,000-year period was derived from recorded (i.e., modern) precipitation [Stonestrom et al., 2003] when using the CCM1 precipitation transition trends. (4) The monthly precipitation data were randomly distributed to each raining day within the month using the recorded average (1972–2005) monthly precipitation days from the meteorological station Beatty 8N in Beatty, Nevada, and the appropriate multipliers as suggested by Thompson et al. [1999].

[17] Using steps 2 to 4 for each month, the CCM1 daily precipitation data was used to create a precipitation series with variations within each month.

[18] To account for the impact of different precipitation intensities, a second precipitation series was created for case 4 (described below) by forcing daily precipitation into a 6-h period: 0:00–6:00, 6:00–12:00, 12:00–18:00, or 18:00–24:00. Potential ET (PET), however, was not evenly distributed in this 6-h series. Rather, 10% PET was assigned to occur from each 0:00–6:00 and 18:00–24:00, and 80% PET was assigned to the rest of the day. So the time step in HYDRUS-1D becomes hourly rather than daily to account for variations in daily precipitation. Figure 1 shows the reconstructed monthly precipitation and PET patterns for the simulation period.

Figure 1.

Reconstructed PET and P.

[19] To examine the potential impacts of a “wet year” on water flow and Cl balance, one wet year was randomly embedded into each 100-year period in each 3,000-year climate sequence for case 5 (described below), thus simulating a 100-year return period. Each wet year, with a total of 48.63 cm precipitation, was based on data recorded at the Beatty 8N station during an 87-year (1918–2005) yearly precipitation record. The frequency analysis in this study was described by Chow et al. [1998]. Precipitation was then randomly distributed across the 52 raining days within the wet year. These raining days are the same as those recorded in 1998, which is the wettest year among the 87-year record and has a yearly precipitation value of 32.05 cm.

2.3.1.2. Construction of Evapotranspiration Record

[20] To acquire Sp as listed in equation (2), the potential evapotranspiration (PET) was first determined. We chose the approach given by Hargreaves and Allen [2003] and Wu [1997]:

equation image

where ETo is the reference ET (equivalent to PET), T is the daily mean temperature (Celsius) and Rs is the incident solar radiation in units of MJ/m2/d. This simpler approach was necessary because the parameters needed in the commonly used (and more complicated) Penman equation are not available, even in General Circulation Models. Output from the CCM1 model includes the daily T and Rs values, and considers the controls of paleoclimate variations across North America [Bartlein et al., 1998]. The parameters T and Rs do not vary within each month.

[21] Potential ET rates were partitioned into potential soil evaporation (PE) and potential plant transpiration (PT) using an approach described by Kemp et al. [1997]. This approach is as follows:

equation image
equation image

where k is radiation extinction by the canopy and is related to the value of LAI (leaf area index). The parameter k ranges from 0.5 to 0.75 when LAI is between 0.2 and 2.0 [Nichols, 1992], and was linearly interpolated on the basis of the value of LAI. Table 2 shows the calculation of LAI. By using equations (14) and (15) and Table 2, partitioned PT is positively correlated to the LAI and the vegetation ground coverage. Similar to equation (6), PE is a weighted linear combination of PE for each growth form, based on reconstructed ground cover in each simulation period.

Table 2. Calculation of LAI for Different Growth Formsa
Vegetation TypeLAI
Evergreen (L. tridentada)0.65 × cover
Deciduous (subshrub)5.70 × cover
Annual1.17 × cover
Grass3.60 × cover

[22] To calculate the vapor transport driven by the thermal gradient, temperature at 15-m depth was set at 22.7°C for 3–0 ka, as derived on the basis of the temperature of groundwater 26.5°C at 110-m depth [Walvoord et al., 2004] and a geothermal gradient of 40°C/km [Scanlon et al., 2003]. Hence, the transient trends of yearly averaged atmospheric temperatures were used to derive the bottom temperature for each 3,000-year period on the basis of the temperature at 15-m depth for 3–0 ka period.

2.3.2. Paleovegetation Reconstruction

[23] On the basis of the existing plant community in the northern Mojave Desert, vegetation growth forms can be classified into five different categories: evergreen, deciduous, grasses, annuals and succulents. A total of five guilds were used in the model depending on the time period, as shown in Figure 2 (note that succulents were not present in modern times in the Mojave Desert [Spaulding, 1990], and thus are not included in the last guild). Rooting depth ranges from 0 cm to 100 cm. Though some research indicates that xeric shrubs may extend their roots to 5-m depth [Schenk and Jackson, 2002; Seyfried et al., 2005], root density generally decreases exponentially with depth because of limited water resources, such that more than 90% of the root mass can be found in soils at 0–100 cm depth in desert regions [Jackson et al., 1996]. Moreover, desert plant species usually have most of their fine roots, roots that are involved in transpiration, shallower than 1-m depth [Wilcox et al., 2004; Peek et al., 2005]. Sandvig and Phillips [2006] also reported that most of roots in their study were found in the upper 40 cm of excavation pits.

Figure 2.

Reconstructed vegetation coverage.

[24] Observed root distributions of similar categories were published by Kemp et al. [1997] when classifying creosotebush (Larreatridentada) as an evergreen and subshrubs as deciduous (Table 3). Jackson et al. [1996] described root distributions of 11 different biomes using a model based on an asymptotic equation [Gale and Grigal, 1987]. Jackson et al.'s [1996] general model was used to describe the root distribution of succulent in this study (Table 3). Furthermore, to account for phenology of root water uptake and vegetation coverage, root distribution was allowed to vary according to season, with three seasons accounting for winter, late spring, and summer/fall periods. We assumed the same annual transitions of vegetation coverage and root growth throughout the 18-ka modeling period, and a constant root distribution as long as the growth forms were the same. We also assumed that annuals and grasses would be classified as separate growth forms with identical ground coverage but distinct root distributions in the numerical model. Figure 2 shows the reconstructed percent vegetation coverage.

Table 3. Root Density Distributions Used in Model
Depth, cmGrowth Form
EvergreenaDeciduousaGrassaAnnualaSucculentb
  • a

    After Kemp et al. [1997].

  • b

    After Jackson et al. [1996], when using equation Y = 1 − βd, where Y is the cumulative root fraction, d is depth in the soil (cm), and β is the fitted parameter. Here we use β = 0.975 as suggested value for desert biome.

0–100.00.10.20.30.22
10–200.10.20.30.50.18
20–300.20.20.20.20.13
30–400.20.20.200.11
40–600.30.20.100.14
60–800.10.05000.09
80–1000.10.05000.05

2.4. Case Analysis

[25] Scanlon [1991] and Tyler et al. [1996] reported one or more nonuniform Cl “bulges” at their sites. They (and others) attributed the unevenly distributed Cl concentration to the changing ratio of P to PET with time due to the climate change. Major climate shifts during the late Pleistocene–early Holocene periods are likely one of the important factors that initiated Cl accumulation in the soil profile as the near-surface water balance changed [Scanlon, 1991; Phillips, 1994; Tyler et al., 1996; Scanlon et al., 2003]. Because the climate shift and vegetation responses occurred contemporaneously, it is difficult to identify whether one factor dominates another, or whether each factor has some influence on water and solute fluxes. We attempted to examine the role of changing climate and vegetation through a series of case studies in which some processes were held constant and others were allowed to vary according to known amounts. Table 4 is a breakdown of characteristics of each case. Because these cases were conducted sequentially, several boundary conditions or model parameters were determined through the completion of earlier runs. This allowed for direct comparisons between simulated soil water potential, volumetric water content, thermal regime, and Cl profiles in different cases.

Table 4. Characteristics of Simulation Casesa
CaseBase CasePaleoflux RecordRoot Water UptakeCompressed Daily Precipitation100-Year Storm EventsDeeper Active Root Zone
  • a

    Solid dots represent selections of boundary conditions that were used in each test case.

1     
2     
3    
4   
5   
6   

[26] The cases vary as follows: (1) Case 1 is a base case in which model boundary conditions are similar to those measured currently at ADRS in the northern Mojave Desert. (2) In case 2 we use a reconstructed paleoclimate sequence. (3) In case 3, we use root water uptake processes, including reconstructed vegetation parameters and paleoclimate. (4) In case 4 we compress daily precipitation to occur within a 6-h period, rather than a 24-h period (described in section 2.3.1.1). (5) In case 5 we embed design-basis storm events (i.e., 100-year return period) into each century in each 3,000-year climate sequence (described in section 2.3.1). (6) In case 6 we modify the root characteristics to include highest root density at 200-cm depth and an active root zone at 300-cm depth, rather than 100-cm in previous cases.

[27] Note that the depth of the plant roots (Figure 3) is restricted to the upper 100 cm of soil for cases 3–5, and extended to 300-cm depth for case 6 to better understand the influence of the maximum root density on the Cl profile.

Figure 3.

Comparison of root distributions. The distribution represents the root conditions in spring at 0 ka simulation.

[28] We also assumed the following:

[29] 1. Average Cl concentration in the precipitation (0.0016 mg/ml) was assumed to be constant over the entire simulation period [Scanlon et al., 2003].

[30] 2. The initial Cl concentration in the soil was set to zero, because of the high water fluxes before 18 ka. This wettest glacial climate is supported by most paleoclimate proxies [Tyler et al., 1996].

[31] 3. Initial soil matric potential was arbitrarily set to −0.098 MPa (equivalent to −10 m water pressure head) to represent a relatively wet environment during the LGM.

[32] 4. For solute transport, longitudinal dispersivity was set to 100 cm on the basis of the soil profile scale [Domenico and Schwartz, 1998], and the diffusion coefficient of Cl in water was set to 1.3 cm2 d−1 [Cook et al., 1992].

[33] 5. Value of the soil matric potential, below which root water uptake ceases, was set at −4.9 MPa (equivalent to −500 m water pressure head), which is the average of the measured low water potential under an undisturbed, vegetated site in the Amargosa Desert [Andraski, 1997].

[34] 6. In cases 1–2, a free drainage lower boundary condition was used to preclude affecting root suction processes. In cases 3–6, the lower boundary condition was set at a constant matric potential of −2.45 MPa (equivalent to −250 m water pressure head), roughly equaling the measured matric potential at the same depth in the Amargosa Desert [Scanlon et al., 2003].

3. Model Results

3.1. Matric Potential and Water Content Profiles

[35] Simulated matric potential and water content profiles are shown in Figure 4. Note that the test cases are split into two groups (group 1 includes cases 1 and 2 and group 2 includes cases 3 to 6) because the largest discrepancies are observed between these two groups. In Figures 4a and 4b (group 1), soil water potential and water content showed very little variation in depth because of a lack of root water uptake. In Figure 4a (case 1), matric potential and water content were nearly constant over 18,000 years, reflecting the steady meteorological input. Both water potential and water content profiles show very steep gradients in the top 30 cm of soil and are then constant in deeper soil. Once the simulation reaches steady state, yearly averaged net water flux (liquid + vapor) at the lower boundary is approximately 1.2 mm a−1. In the uppermost soil, the direction of liquid flux became upward because of the effects of evaporation, even though summer thermal gradients induce downward thermal vapor transport. Figure 4b shows two distinct matric potential and water content profiles due to the large differences in precipitation between pluvial and dry periods. For example, matric potential in 15 and 12 ka (top graph in Figure 4b) shows an upward potential gradient in soil shallower than 20 cm and a downward potential gradient in soil from 40 to 60 cm. The divergence of the gradient at about 30-cm depth is a direct result of a water bulge that existed in the profile from a precipitation event that occurred near the end of the 3,000 year simulation period. This water bulge dissipated because of evaporation or percolation to deeper soil, or both, after the precipitation event. During the final 6,000 years of simulation, water flow rates and directions in the majority of the soil profile were unchanged, and the downward potential gradient approached zero, because the supply of water at the surface was not sufficient to sustain a potential gradient. In these two cases, response time of the water content to the surface climate change is short relative to the simulation period.

Figure 4.

(a and b) Water potential and volumetric water content (group 1). The figure does not show the simulated water potential and water content at the ending points of 9 ka and 3 ka. Matric potential on ground surface in both cases is −10 MPa, which is the lowest allowed water potential in the numerical model. (c–f) Water potential and volumetric water content (group 2). The figure does not show the simulated water potential at the ending points of 9 ka and 3 ka. Note that the horizontal scales in matric potential are different between group 1 and group 2.

Figure 4.

(continued)

[36] After the processes of root water uptake were accounted for in group 2 (Figures 4c4f), the soil water potential throughout the profile decreased substantially, especially in the upper 100 cm in cases 3–5, and in the upper 300 cm in case 6. Soil water content profiles also showed the same trends. Large differences in water content were observed below the root zone between cases 3, 4 and 5, especially during the periods of 15 and 12 ka. This observation is supported by the reduced net water flux at the bottom of the domain which previously was unaffected by variations in surface climatic conditions. For example, at 15 ka, bottom water fluxes were −0.13 mm a−1 (negative represents upward water flux), 0.07 mm a−1 and 1.11 mm a−1 in cases 3, 4 and 5, respectively. However, when the climate was much drier, for example at 0 ka, water fluxes were −0.16 mm a−1, 0.06 mm a−1 and −0.16 mm a−1, respectively. Therefore, although 100-year return period events were introduced in case 5, they only increased the recharge when total water supply was relatively large (e.g., 15 ka and 12 ka). During drier periods, however, the effect of 100-year storm events was buffered by a long drying process, so the upward bottom flux was the same as observed in case 3 (e.g., −0.16 mm a−1 at 0 ka), which had no extreme events. On the basis of the simulation results, only case 4 produced recharge during dry periods of 6 ka to 0 ka, reflecting the impact of increased precipitation intensity imposed on the upper boundary (i.e., precipitation occurred during 6-h period instead of 24-h period). Moreover, even though the climate changed between 15 ka and 0 ka, the difference in flux was 0.01 mm a−1 (i.e., 0.07 versus 0.06 mm a−1).

[37] Bottom net fluxes in case 6 at 15 ka and 0 ka were −0.16 mm a−1 and −0.23 mm a−1, respectively, which were the lowest fluxes for all cases that included root water uptake. The results indicate that the increased root mass essentially stopped liquid water from percolating below the root zone, and then induced a net upward water flux as has been observed by others [Walvoord et al., 2002a; Scanlon et al., 2003]. The water content profile (Figure 4f) also supports this observation. Also, after 6,000 years of simulation (time = 12 ka), soils in the uppermost 30 cm became significantly drier, showing the effect of higher PET rates (Figure 1) toward the end of the last glacial period. In case 6, where the root zone distribution was extended to 300 cm below ground surface, reduced water contents begin at 55 cm rather than 30 cm in other cases, illustrating the importance of where plant roots are placed in the simulation.

[38] An important phenomenon is that bottom net fluxes in group 2 (Figures 4c4f) during dry climates are actually upward, except in case 4. Considering that a large downward matric potential gradient was artificially created by the constant soil water matric potential of −2.45 MPa (equivalent to −250 m water pressure head) at the bottom of the domain, root water extraction and upward thermally driven vapor transport exceeded the downward liquid fluxes induced by the downward gradient, causing water to move vertically upward toward ground surface. Therefore, under such conditions, water from soil deeper than 15 m could be extracted given sufficient time.

3.2. Chloride Distributions

[39] Figure 5 shows Cl distributions for cases 1–6, divided into groups similar to those described for the water potential and water content profiles. Cases 1 and 2 (Figures 5a and 5b) show nearly uniform Cl profiles at the end of the simulation period without discernible peaks. Some differences are noted between the two cases, however. For example, the Cl concentration profile in case 1 is uniform with depth, reaching steady state relatively quickly. In case 2 (Figure 5b), two distinct concentration profiles exist, illustrating the different Cl accumulation rates caused by the climatic switch. In both cases, even though root water uptake is not occurring, soil evaporation does remove enough soil water to eventually create a profile with steadily increasing Cl concentrations very close to ground surface. Figure 5b also shows that during the period when climate switched (12 ka), Cl accumulated in the entire soil domain rather than only close to the soil surface. Also, one may notice that the stored Cl concentration at the end of the simulation record (i.e., time = 0 ka) is lower than observed at the end of 12,000 years of simulation (or time = 6 ka) (Figure 5b). This implies that Cl output rates during this time exceeded input rates. The net liquid water fluxes were equal to 0.96 cm a−1 and 1.19 cm a−1, respectively. When taking into account the upward vapor transport, however, the downward liquid fluxes were actually larger.

Figure 5.

(a and b) Chloride concentrations (group 1). The figure does not show the simulated chloride concentration at the ending points of 9 ka and 3 ka. (c–f) Chloride concentrations (group 2). The figure does not show the simulated chloride concentration at the ending points of 9 ka and 3 ka. Note that the scales are different between group 1 and group 2.

Figure 5.

(continued)

[40] The importance of vegetation in this environmental system is revealed in cases 3–6 (Figures 5c5f). Here, the Cl concentrations at depths approaching 15 m are essentially zero, again illustrating the small liquid flux occurring below the root zone during the 18,000 years of simulation. In case 3, the root water uptake clearly led to significant increases in Cl concentration near ground surface, with values almost 100 times higher than in case 2, which did not include transpiration. Case 4 shows a very similar Cl distribution to those presented when daily precipitation was applied during a 24-h period (case 3). However, the increased precipitation intensity in case 4 flushed Cl deeper into the profile and through the base of the domain. We noted that the average percent of Cl mass stored in the profile was only 59.2% of the total input, which is less than in all other cases (Table 5).

Table 5. Percentage of Chloride Remaining in the Model Domain at Representative Time Perioda
Period EndingCase 1Case 2Case 3Case 4Case 5Case 6
  • a

    Unit is %.

15 ka0.11.099.886.856.299.9
0 ka0.0–5.971.359.268.491.9

[41] Figure 5e (case 5) reflects the addition of a single “wet year” for each 100 years of simulation. Significantly lower Cl concentrations are noted near ground surface after simulation periods 15 ka and 12 ka, versus those from cases 3, 4 and 6. Figure 5e also shows lower Cl concentrations in other simulation periods, though we note that these lower concentrations resulted from low initial Cl mass in the profile after 12 ka. Adding 100-year storm events during the LGM caused larger downward Cl fluxes; however, during modern times with lower annual precipitation, Cl was flushed deeper into the soil profile but through the bottom of the domain.

[42] The most significant finding in case 5 is the formation of a distinct Cl bulge at approximately 90–100 cm depth at 0 ka (Figure 5e). The depth of maximum Cl concentration, which corresponds to the base of the root zone (Figure 3), is also much deeper than simulated in cases 3, 4 and 6. Apparently, Cl is transported further downward during the simulated wetter (100-year storm) periods and then became concentrated deeper within the root zone. When compared to cases 3 and 4 (Figures 5c and 5d), the results show the importance of the precipitation record on the Cl profile. For example, using the “normal” precipitation record (case 3), the Cl bulge remains close to 10–20 cm depth with concentrations of about 40 mg L−1. Compressing the precipitation record into 6-h (case 4) led to increased Cl fluxes from the bottom of the domain during all climate periods and a lower overall Cl mass remaining in the profile at the end of the simulation (Table 5). Adding periodic wet years to the record (case 5) led to a deeper Cl bulge (Figure 5e) with lower concentrations than seen in cases 3 and 4. Moreover, we note in case 5 that significant bottom Cl fluxes occurred only during wet climates (i.e., 15 ka and 12 ka). When the climate became drier, almost no downward bottom Cl fluxes were observed.

[43] When roots were extended deeper into soil (case 6), the depth of the highest Cl concentration increased to 100-cm depth in 15 ka and 12 ka (Figure 5f). When the climate became drier (after 12 ka), the Cl peak moved upward in the profile to about 20–40 cm depth. We contrast these results to those of case 3, where maximum root zone density and rooting depth were 20–30 cm and 100 cm, respectively, and the Cl bulge was simulated at about 20-cm depth. Apparently, the distinct Cl bulges observed for the wetter 15-ka and 12-ka conditions in case 6 were deep enough into the soil profile to buffer the bulge against diffusion. Moving the maximum root zone density to 200-cm depth allowed water and Cl to percolate and migrate deeper in the soil profile, away from effects of soil evaporation.

4. Discussion

[44] The measured Cl peak at ADRS is 230 cm below ground surface and the peak concentration is 9 mg/ml [Scanlon et al., 2003]. Some differences exist between our model predictions and field data for Cl concentration reported by Scanlon et al. [2003], but, given the long simulation time and spatial variability in root zone distribution in desert systems, the results are generally close and represent spatially averaged hydrologic processes in the northern Mojave Desert, and not at a particular site. In another study, Sandvig and Phillips [2006] showed matric potential and Cl profiles under different species and climatic conditions. Most of their experiment sites were located in arid/semiarid regions in central New Mexico, and consisted of sandy-textured soils. Their results also showed sharp decreases in average matric potential (from three sites) from the soil surface to a depth of about 100 cm, and an average depth to the Cl peak of 250 cm. The field conditions for Sandvig and Phillips [2006] at locations 1, 2 and 3 are closest to our simulation conditions described for case 5 (Figures 4e and 5e), and the results are conceptually similar. Differences in water potential and Cl are attributed to several potential factors, including soil conditions, climate and the structure of the numerical model (in our study) where root water uptake is strictly associated with the root locations.

[45] The results of the numerical experiments by Walvoord et al. [2002a] and Scanlon et al. [2003] illustrated that Cl will concentrate at the depth of a fixed subroot zone, which in their cases were implemented using a single sink node kept constant at −4.9 MPa at a depth of 230 cm. However, according to Sandvig and Phillips [2006], who examined conditions in the upper 10 m of soil, large negative matric potentials exist through the whole root zone as well as large part of the subroot zone. In our study, a root zone distribution was calculated for a mixed canopy with different rooting characteristics for each plant type, and with different maximum root zone densities. We showed that in cases 3 and 4, the depth of elevated Cl concentrations was associated with the depth of the maximum root density. But, in case 6, the association no longer holds during dry periods. Given that the peak did move deeper in 15 and 12 ka, we propose that the location of the Cl peak is secondarily associated with the distribution of the root zone, and primarily associated with climatic conditions. The results indicate that, in arid regions, root water uptake rapidly removes water that infiltrates the soil, and continues throughout the entire root zone.

[46] Recharge below the root zone, potentially resulting in groundwater recharge, apparently will become more significant during episodes of plant mortality or widespread plant loss (e.g., from brush fires or disease), as was demonstrated in case 2. However, at other times when vegetation is present and active, on average, almost all water infiltrating through the soil surface eventually will be transpired (case 3). Although other models suggest that transpiration responds to the average soil water potential [Fahey and Young, 1984], rather than to the water potential in layers where roots exist (as simulated in HYDRUS-1D), the simulated extremely dry conditions throughout the root zone are still likely to sustain the constant root sink. Moreover, during dry time periods, thermally driven vapor water transport could sustain an upward net flux. Our results are consistent with Walvoord et al. [2002a] and Scanlon et al. [2003], who demonstrated that vapor water transport in arid regions plays important roles on the distribution of the matric potential in the soil, but has a relatively unimportant role in the absolute values of soil water content (especially within the active root zone). Thus, inasmuch as the effective diffusion coefficient of Cl is a function of soil water content, and given that Cl is nonvolatile, the redistribution of Cl affected by the enhanced vapor flux could be ignored. In addition, Scanlon [1991] and Walvoord et al. [2002a] showed that osmotic potential has a much smaller effect than matric potential on water flow in arid vadose zones.

[47] Present-day root zone distributions for mixed canopies in arid climates generally have coexisting shallow and deep roots which are active for different seasons and lengths of time. The paradox, then, is how Cl bulges deeper than 100 cm could form and persist given current root zone distributions and plant phenologies. For example, only in case 5 of our study, which included the most realistic environmental setting, did the depth of penetrating moisture correspond to the rooting depth for the entire guild. In cases 3, 4 and 6, the depth of the Cl bulge was significantly above the base of the roots. These results illustrate the importance of including realistic root zone distributions and climate sequences when simulating paleowater fluxes.

[48] Other physical and/or biological processes may explain the deep presence of Cl. For example, soil aggradation processes that create desert pavement surfaces [McFadden et al., 1998] could lead to buried soil horizons containing higher chloride levels, though soil aggradation is typically measured on the order of cm, not 10s of cm. Similarly, pedologic development (including the creation of macropores) may cause significant changes of soil hydraulic properties given the timescale of thousands of years [Young et al., 2004], and a stronger coupling of soil properties and canopy characteristics as shown by Shafer et al. [2007]. Another explanation could be an inaccurate dry fallout chloride rate in the simulation. Although this dry deposition is variable [Zhu et al., 2003], Dettinger [1989] suggested a 33% uncertainty in the Great Basin, which would explain the discrepancy between our simulated results and the Cl concentrations reported by others.

5. Conclusion

[49] The results showed that the total amount of water entering the soil through precipitation, the precipitation intensity, and the vegetation root distribution play important roles in the water flux mechanisms in arid regions. Rather than assuming that the Cl profile is controlled only by climate and the climate switch, our results show that the depth of root water uptake has a substantial influence on the depth of the Cl bulge commonly preserved in thick vadose zones of the southwestern United States. Moreover, under the conditions of typical desert vegetative cover and normal precipitation patterns (e.g., case 3), recharge rates below the root zone became almost zero, even in the more pluvial 15–12 ka climate. Any recharge into the deep vadose zone occurring during this period is most probably focused in topographic lows like channels and washes. Large intensity precipitation events may contribute to recharge or perhaps to groundwater recharge. Large storms such as 100-year precipitation events (case 5) may also contribute to recharge, but our results showed that this occurrence was restricted to the pluvial period (i.e., 15 and 12 ka) when the initial soil water content was higher. The yearly averaged net upward water flux found in case 5 during the present time (−0.16 mm a−1) is larger than −0.01 mm a−1 that was simulated by Walvoord et al. [2004], and is smaller (and opposite in direction) than the net downward flux calculated by Andraski [1996] at 5-m depth in a vegetated soil plot at the end of September 1990 (0.09 cm a−1).

[50] The results of the simulations provide a better understanding of soil water flux and Cl concentration variations under the influence of long-term climate change. The results also suggest that further research on water fluxes in arid regions should focus on the development of desert plant roots and the interactions between roots and soil moisture under the conditions of climate change. These are the most significant factors that govern water fluxes in vadose zone.

Acknowledgments

[51] Funding for J.Y. was made available by the DRI-UNLV PhD fellowship program. Funding is also acknowledged for M.Y. and Z.Y. from NSF EPSCoR under the grant EPS-0447146. We thank the anonymous reviewers' valuable suggestions that improved this manuscript.

Ancillary