Simulating plant water availability in dry lands under climate change: A generic model of two soil layers

Abstract

[1] Dry lands are exposed to a highly variable environment and face a high risk of degradation. The effects of climate change are likely to increase this risk; thus a profound knowledge of the system dynamics is crucial for evaluating management options. This applies particularly for the interactions between water and vegetation, which exhibit strong feedbacks. To evaluate these feedbacks and the effects of climate change on soil moisture dynamics, we developed a generic, process-based, spatially explicit soil moisture model of two soil layers, which can be coupled with vegetation models. A time scale relevant for ecological processes can be simulated without difficulty, and the model avoids complex parameterization with data that are unavailable for most regions of the world. We applied the model to four sites in Israel along a precipitation and soil type gradient and assessed the effects of climate change by comparing possible climatic changes with present climate conditions. The results show that in addition to temperature, the total amount of precipitation and its intra-annual variability are an important driver of soil moisture patterns. This indicates that particularly with regard to climate change, the approach of many ecological models that simulate water dynamics on an annual base is far too simple to make reliable predictions. Thus, the introduced model can serve as a valuable tool to improve present ecological models of dry lands because of its focus on the applicability and transferability.

1. Introduction

[2] Water is a key factor in determining plant growth in arid and semiarid ecosystems [Walker et al., 1981; Ehleringer et al., 1991; Kemp et al., 1997; Rodriguez-Iturbe et al., 1999]. Active biological processes, namely growth and reproduction only take place if the soil is moist enough [Noy-Meir, 1973]. Soil moisture is driven by a variety of factors. The most obvious of these is precipitation, which in dry lands varies greatly both within and between years [Noy-Meir, 1973; Schlesinger et al., 1990] and which is predicted to vary even more in the course of climate change [Easterling et al., 2000]. The effect of single and multiple precipitation events on different ecological responses is explored by precipitation pulse theory [e.g., Huxman et al., 2004; Loik et al., 2004; Schwinning and Sala, 2004]. It finds that only events of a sufficient size or multiple events trigger essential ecosystem responses such as seed production, germination, or establishment by exceeding thresholds of infiltration depth and/or of soil moisture [Beatley, 1974; Ehleringer et al., 1991; Bowers, 1996; Marone et al., 2000].

[3] However, such large precipitation events also generate runoff, when infiltration capacity is exceeded (Hortonian runoff) or when the soil is saturated with water (saturation runoff). Hortonian runoff in particular has been observed in dry lands world wide [see Stomph et al., 2002]. During extreme precipitation events, both runoff types can markedly reduce infiltration intensity.

[4] Both, infiltration and runoff, are strongly affected by the prevailing vegetation cover [Bartley et al., 2006]. It promotes infiltration through stemflow and the creation of macropores [see, e.g., Walker et al., 1981; HilleRisLambers et al., 2001]. Vegetation cover also increases the surface roughness, which slows down runoff flows [Howes and Abrahams, 2003]. Many studies in dry lands have found that vegetation facilitates the establishment of other plants by increasing the local water availability [Callaway and Walker, 1997; Shachak et al., 1998; Maestre et al., 2001]. This can lead to mosaics of bare patches and vegetated patches, in which vegetated patches profit from the high runoff from bare patches [see Thiery et al., 1995; Ludwig et al., 2005]. Furthermore, the vegetation cover influences water losses of the soil due to evaporation and transpiration [Huxman et al., 2005].

[5] Climate change will clearly have an impact on future soil moisture patterns in arid and semiarid systems. Regional climate change predictions remain difficult [Weltzin et al., 2003], but all SRES scenarios of the Intergovernmental Panel on Climate Change (IPCC) [2007] predict that arid and semiarid areas will experience a warming of at least 1.5°C by the end of this century. Such a temperature increase will likely lead to lower soil moisture due to enhanced evapotranspiration. Projections for mean annual precipitation are ambiguous: the total annual water availability is predicted to increase for some dry lands but decrease for others [IPCC, 2007]. Independent of the mean annual precipitation, the variance of interannual as well as of intra-annual precipitation likely will increase [Easterling et al., 2000; Dore, 2005]. In other words, in many regions there will be less, but more extreme precipitation events. The effects of this increased variability on vegetation can even surpass those resulting from changes in average values [Puigdefábregas, 1998]. Changed vegetation patterns, caused by altered soil moisture conditions, will in turn feedback on the soil moisture dynamics.

[6] A sound understanding of the dynamics of the coupled water-vegetation system is essential to assess the impacts of climate change on vegetation cover in dry lands. Thus, when studying climate change, it is necessary to look beyond the ecological responses to precipitation events and instead to include all relevant drivers and feedbacks of soil moisture dynamics [Tietjen and Jeltsch, 2007]. The emerging discipline of ecohydrology addresses this issue by explicitly establishing the link between ecology and hydrology [e.g., Rodriguez-Iturbe, 2000; Porporato et al., 2002; Rodriguez-Iturbe and Porporato, 2004; Wilcox and Newman, 2005]. In this field, several models have been developed, for example to explain the formation of vegetation patterns [Okayasu and Aizawa, 2001; von Hardenberg et al., 2001; Rietkerk et al., 2004] or to analyze water availability to plants [Loik et al., 2004; Porporato et al., 2004].

[7] Many of these models focus on annual dynamics or steady state situations and preclude the effects of intra-annual precipitation patterns. However, pulse theory has revealed that different precipitation pulses lead to different ecological responses. That is, ecological responses can be affected strongly if climate change alters intra-annual precipitation patterns [Fravolini et al., 2005]. Additionally, timing of effective precipitation events can be of great importance, e.g., for germination or for the establishment of plants [Noy-Meir, 1973; Schwinning and Sala, 2004]. Here, different plant species can react differently to changing soil moisture patterns, which can lead to a changing competitive situation in the ecosystem [Ehleringer et al., 1991]. Similarly, positive interactions like facilitation by increasing the local water availability can be affected by changing spatiotemporal rainfall patterns. If the soil moisture regime changes, niches for new species can be created or existing niches can vanish. This also leads to modified conditions for invading plant species [Huxman et al., 2004].

[8] These findings suggest clearly that accounting for changes in mean responses to climatic variability is not sufficient when it comes to climate change [Porporato et al., 2004]. There is a great need to expand the existing set of ecohydrological models by approaches that address in detail the consequences of single precipitation events or series of events. Changed temporal precipitation patterns and primarily extreme events also strongly influence the amount of water that does not infiltrate directly but is redistributed by runoff. Although a variety of ecohydrological models consider runoff [e.g., Rodriguez-Iturbe et al., 1999; Laio et al., 2001], it is to our knowledge mostly modeled as saturation runoff and not as Hortonian runoff (but see Rigby and Porporato [2006], Collins and Bras [2007], and Ivanov et al. [2008]). This is a clear shortcoming in most of the present models used in ecology, since Hortonian runoff dominates lateral water redistribution in semiarid and arid regions [Saco et al., 2006; Wilcox and Newman, 2005].

[9] We are aware that observing and modeling patterns of hydrological processes in microscale and mesoscale catchments is one of the most complicated tasks in hydrology. The main reason for this is the enormous heterogeneity of natural soils and ecosystems [Western et al., 2004; Zehe et al., 2007] that, together with the nonlinear nature of soil hydrological processes, leads to an incompatibility between observation scales and process scales [Blöschl and Sivapalan, 1995]. Additionally, the interactions between soil moisture, infiltration and runoff generation get even more complex when preferential pathways such as earthworm burrows and root channels are present [Bronstert and Plate, 1997; Villholth et al., 1998; Zehe and Fluhler, 2001]. Sound modeling of the interactions of soil moisture dynamics, runoff production and preferential flow in heterogeneous soils requires numerical models based on the Richards equation often combined with a double continuum approach [Simunek et al., 2003; Zehe and Blöschl, 2004] as well as detailed information on the spatial pattern of subsurface heterogeneity. However, such information is typically unavailable for most places in the world. Moreover, because such models work on a time scale in the order of seconds to minutes in contrast to ecological models, which simulate vegetation dynamics on daily, weekly, monthly or even longer time steps, a simple coupling of a detailed hydrological process model with an ecological model is not feasible.

[10] The objective of this study is therefore to introduce a generic, spatially explicit hydrological model to simulate abiotic boundary conditions for plant growth in dry lands as a parsimonious hydrological tool for ecologists. The major difficulty is to make the model necessarily complex to ensure acceptable hydrological performance, mainly with respect to soil moisture dynamics and lateral redistribution, while on the other hand keeping parameterization of the model feasible thereby not limiting its application to intensively monitored research catchments. We developed a spatially distributed model (HydroVeg) to simulate hydrological surface processes and soil water dynamics within two soil layers. The model conceptualizes infiltration and preferential flow into the deep soil via roots, water losses by evapotranspiration, and lateral and vertical water redistribution by surface runoff and fluxes between soil layers. These processes are tightly coupled to cover of two plant types, which is assumed to be static in the present study. By coupling HydroVeg with a dynamic vegetation model in future studies, bidirectional feedbacks between water and vegetation can be accounted for.

[11] In the following we will first introduce the model structure and the process concepts with special emphasis on how vegetation cover affects key processes. The model is then applied to four sites along the climate gradient in Israel and simulated soil water regime is compared to observations for different soil types. A sensitivity analysis will then shed light on the importance of lateral water redistribution and on how changes in plant cover affect key hydrological processes. Finally, we investigate the impact of changes in precipitation pattern and of enhanced temperature on soil water as primary source for plants.

2. HydroVeg: Structure and Process Descriptions

[12] The model (HydroVeg) is designed to simulate the spatially and temporally varying moisture conditions for plant growth in semiarid areas rather than to simulate the water balance of a catchment. We divided the landscape into grid cells of 5 by 5 m at a total spatial extent of 25 ha. Figure 1 provides a scheme of the spatial structure and processes simulated in each grid cell. Target state variables are soil moisture in the upper (W_L1) and the lower (W_L2) soil layer as well the overland flow/surface water depth (W_SF). This two-layer approach allows simulating the differing controls of vegetation on hydrological processes such as on evapotranspiration from the upper soil, water retention by shading, as well as root extraction from the deeper soil. A higher resolution would be of course more precise, however, parameterization would be less feasible and it would mean additional computational cost.

Figure 1

Model structure and processes simulated in each grid cell.

[13] Hourly precipitation (PI) accumulates as surface water and infiltrates into both soil layers (F_L1, F_L2). Surface water that does not infiltrate is laterally redistributed to lower areas by means of surface runoff (Q_in, Q_out). Losses from both soil layers and surface water are allowed by means of either evapotranspiration (ET_L1, ET_L2, ET_SF) or by drainage to deeper soil layers (F_drainL1, F_drainL2). Additionally, soil water flow between the two layers (F_L1L2) is accounted for. As will be shown in the following all processes are strongly dependent on grass and shrub cover (c_g, c_s), which are given as fractions ranging from zero to one. Zero shrub cover and zero grass cover indicate a bare soil grid cell. Combined cover of grass and shrubs can exceed 1.0, since grass may grow underneath shrubs. The total mass balance for surface water and soil moisture in both layers of a grid cell is thus:

These model equations can be deemed as a trade-off between the necessary complexity required to capture all essential drivers of the system and the need to have a parsimonious model, based on parameters assessable from the literature, simple field measurements and careful estimates of parameters that are not directly observable. As the model will frequently used in areas were no gauging data are available, calibration of model parameters to observed discharge is not feasible.

2.1. Infiltration Into the Soil Surface and Into the Deep Soil

[14] The model distinguishes between infiltration at the soil surface and through preferential pathways (roots) into the deep soil, which we regard as necessary to account for feedbacks between shrub cover and soil water dynamics. Fast, preferential infiltration F_L2 into the lower soil layer occurs via macropores due to cracks or roots:

Infiltration of surface water W_SF over time step Δt [mm/h] (as substitute for precipitation rate PI) is given by infiltration rate F_L2,frac [dimensionless] that is in a range of [0, 1], and a saturation function g₁ of shrub cover c_s [dimensionless] that is also in a range of [0, 1]:

Thus, fast infiltration through macropores is essentially linked to shrub cover. The latter (1) can be predicted by a vegetation model and (2) is an observable quantity and therefore allows for evaluation of model parameterization within simulations with dynamic vegetation. In absence of shrubs, the saturation function equals bare soil infiltration F_L2,bare [dimensionless]. A low value F_L2,bare accounts for possible water repellency of the soil at bare sites, while a high value of F_L2,bare indicates preferential flow phenomena that are not dependent on vegetation cover. Infiltration increases with shrub cover, thus, equation (2) also allows for an enhanced infiltration at vegetated patches, which is a first-order control of vegetation patterns forming semiarid ecosystems [Wilcox and Newman, 2005]. Total infiltration F_L2 is additionally limited by a maximal hourly infiltration rate F_L2,max [mm/h] that depends on soil properties, similar to the maximum water flow in macropores in the kinematic wave model of Germann and Beven [1981]. As the parameters F_L2,frac and F_L2,bare are difficult to observe in the field, they have to be estimated with great care as discussed below.

[15] Infiltration of the remaining surface water (W_SF,rem, mm) into the upper soil layer is calculated according to the Green and Ampt [1911] approach. The idea is that a wetting front proceeds into the soil, above which the soil is saturated and below which the water has not infiltrated yet. A wetting front forms only, if rainfall intensity is higher than the hydraulic conductivity K. If the available surface water W_SF,rem divided by time step Δt is smaller than K, the remaining surface water infiltrates into the upper soil layer. Otherwise, ponding occurs and the saturation time t_s is calculated according to Rawls et al. [1992], until the upper few millimeters of the soil are saturated:

Here, n_a describes the fillable porosity [m³/m³] and is determined by the difference between field capacity fc [m³/m³] and the current volumetric soil moisture W_L1,V. S_f [mm] is the effective suction at the wetting front. The hydraulic conductivity K [mm/h] is assumed to equal the geometric mean of saturated and unsaturated hydraulic conductivity (K_s and K_u, respectively) to account for the interplay of the saturated zone above and the unsaturated zone below the wetting front [see also, e.g., Germann and Beven, 1981; Zehe et al., 2001]. K_u is calculated out of K_s as described by Kemp et al. [1997]: K_u = . A more complex approach to parameterize unsaturated soil conductivity such as the Van Genuchten [1980] and Mualem [1976] approach would likely be more realistic. However, often in the target areas neither direct measurements nor reliable data for pedotransfer functions to estimate these parameters are at hand. We modified equation (4) by a function g₂(c_g, c_s) of grass and shrub cover to include the influence of vegetation cover on infiltration into the soil surface:

Function g₂ approaches a limit of 1.0. Parameters a₁ and a₂ are chosen such that g₂(0, 0) = 0.2 to account for water repellency of bare soils. At a grass and shrub cover of 50% each, g₂(0.5, 0.5) = 0.9 is close to the limit of 1.0. Grass has a higher effect on infiltration into the upper soil layer, due to the presence of fine roots. If saturation does not occur within one time step (t_s > Δt), the entire surface water can infiltrate, i.e., F_s = W_SF,rem. Otherwise, infiltration F_s until saturation of the surface is calculated by:

Integrating the subsequent infiltration intensity as suggested by Dyck and Peschke [1989] and adjusting the result by the vegetation function g₂ leads to the following total infiltration rate after saturation:

with

[16] Thus, the total infiltration rate into the upper layer adds up to

2.2. Vertical Soil Water Movement After Infiltration

[17] If soil moisture in one of the layers exceeds field capacity, water drains to deeper layers. F_drainL1 and F_drainL2 [mm/h] equal the excess water from the upper and lower layer, respectively. Excess water from the upper layer is added to the lower layer, while water from the lower layer is assumed to recharge the groundwater stock.

[18] Additionally to drainage, diffusive soil water flow between both layers can occur. Here, similar to Quintana Seguí et al. [2008] we adopt Darcy's law (e.g., described by Rawls et al. [1992]), but in moisture based form where the flow F_L1L2 from the upper to the lower layer is expressed as:

W_L1,V and W_L2,V are the volumetric water contents of both layers with depth_L1 and depth_L2 [mm]. Water flux is thus equal to the moisture gradient times the water diffusion coefficient. The latter is estimated on the basis of the hydraulic conductivity K [mm/h] times a calibration constant d [dimensionless] times the depth of the respective layer depth_Lx. The flux F_L1L2 is positive when there is flow from the upper to the lower layer. This approach is a simplification because the term in brackets is an estimate of the real water diffusion coefficient. However, to account for this, we would have had to introduce at least four additional parameters to describe soil water characteristics [Van Genuchten, 1980]. This information is not at hand for most sites of interest. The suggested approach is a reasonable approximation as long as the two soil layers have similar water retention curves and the parameter d is estimated with care as discussed below.

2.3. Evapotranspiration

[19] The model first evaluates potential evapotranspiration according to the temperature based approach suggested by Hargreaves [1974]. Actual evapotranspiration is calculated dependent on the current surface and soil water and the vegetation cover. Droogers and Allen [2002] showed that this approach leads to sound predictions of evapotranspiration in semiarid and arid areas that are comparable to the results of the much more parameter intensive Penman Monteith method.

[20] Hargreaves' approach to calculate daily potential evapotranspiration uses daily mean, minimal and maximal temperature [°C] (, T_min, and T_max, respectively), as well as extraterrestrial radiation R_a [mm/d]. Additionally, we account for spatial heterogeneity in potential evapotranspiration due to slope sl [rad] and aspect effects af [dimensionless] on solar radiation.

Extraterrestrial radiation is calculated on the basis of latitude and Julian day as suggested by Stull [1988]:

with d_r relative distance between Earth and Sun [dimensionless] d_r = 1 + 0.033 · cos

The aspect factor af allows for differences in radiation between south and north facing slopes and slightly corrects the calculated values for inclination values above 5°. For reasons of simplification, the aspect and inclination refer to the lowest neighboring cell, values are taken from Shevenell [1999] and are given in Table 1.

Table 1. Correction Factor af to Account for the Aspect of the Grid Cell

Direction	Correction Factor
NW	0.97
N	0.9
NE	0.95
W	1.02
E	0.98
SW	1.05
S	1.1
SE	1.03

[21] Actual evaporation from the surface water ET_SF is linearly reduced by grass and shrub cover, to account for shading:

To calculate actual evapotranspiration from both soil layers (ET_L1 and ET_L2), we adopted an approach of the conceptual HBV model [Hundecha and Bárdossy, 2004]. This results in a similar approach as many other ecohydrological models [e.g., Laio et al., 2001; Porporato et al., 2001]. If the volumetric soil moisture W_Lx,V in layer x is above a level wsc [m³/m³] at which plants begin to close their stomata, ET_L1 is maximal (Figure 2) and equals the potential evapotranspiration ET_pot times function g₃ of vegetation cover. The same applies for ET_L2 except that here, potential evapotranspiration is simply multiplied by shrub cover, since mainly shrub roots extract water from lower soil layer. Below this level, actual evapotranspiration decreases and stops at a soil type specific residual water content value rw [m³/m³]:

Constant a_ET accounts for evaporation reduction e.g., caused by soil crusts. Normally, a_ET should not influence evaporation and should be set to 1. However, if soil crusts play a major role at a respective site, a_ET should be adopted, e.g., by comparing measured evaporation values of crusted patches with bare patches without a crust. Since mainly shrub roots extract water from the lower soil layer, transpiration from this layer is assumed to be driven only by shrub cover. In contrast, evapotranspiration from the upper layer is driven by both plant types and is described by a linear decreasing function g₃:

Because of shading, vegetation has an inhibiting effect on evapotranspiration from the upper soil layer, which is to a bigger extent driven by evaporation than by transpiration. The contrary applies to the lower soil layer, where water losses are driven by transpiration and increase therefore with vegetation cover. A high vegetation cover can even result in a total evapotranspiration that is higher than the calculated potential evapotranspiration, since the latter refers to a reference area that can be exceeded by a high leaf area index (LAI). Evaporation and transpiration stop when soil moisture drops to a critical residual water content.

Figure 2

Schematic overview of the actual evapotranspiration dependent on soil moisture, with rw, residual water content; wsc, water content of beginning stomatal closure; and ET_pot, potential evapotranspiration.

2.4. Surface Runoff and Lateral Distribution

[22] The lateral redistribution of water is calculated at an hourly time step for each grid cell. We assume that runoff from a cell only contributes to the lowest neighboring cell and that overland velocity is fast enough to redistribute water in the simulated area within one time step.

[23] First, the runoff of the most upslope cells is evaluated that do not receive any runoff from surrounding cells. Following the approach of Manning-Strickler [Dingman, 1994] their runoff Q_out is calculated dependent on the surface water depth, the vegetation cover as an estimate for surface roughness, and the downward slope sl:

Afterward, cells receiving water from upslope cells are handled. The total water Q_in a cell x_i,j receives from the eight directly surrounding cells is calculated as:

The surface water height is increased by Q_in · Δt and the resulting runoff of the actual cell is evaluated according to equation (15). Afterward, the runoff of cells at a position more downslope is evaluated.

[24] Cells at the borders of the grid that have no lower neighboring cell within the grid are assumed to lose water from the system. Here, we set the slope sl to the same value as the slope to the directly neighboring cell that is not at the border.

3. Generic Model Test, Model Sensitivity, and the Role of Lateral Redistribution

3.1. Study Area and Model Setup

[25] We applied the model to four sites along a precipitation gradient in Israel. The sites range from mesic Mediterranean in the north to arid in the south (Ein Ya'aqov: mean annual precipitation 780 mm, mean temperature 18.1°C; Matta: 537 mm, 17°C; Lahav: 300 mm, 18.4°C; Sde Boqer: 90 mm, 19.1°C) [Holzapfel et al. 2006]. All sites share the same calcareous bedrock (hard limestone), with the soil type Terra Rossa at Ein Ya'aqov and Matta, Brown Rendzina at Lahav and desert Lithosol at Sde Boqer [Holzapfel et al., 2006]. Thus, the sites cover not only a broad range of climatic conditions, but also have different soil conditions ranging from clay in the north to sandy loam in the south. The general soil type was determined in the field [Zwikel et al., 2007]; afterward most soil-specific parameters were assessed from literature data (see Table 2). Wilting point and field capacity were determined from similar sites under field conditions (S. Pariente, personal communication, 2007) and correspond to a soil matric potential of −1500 kPa and −6 kPa, respectively. Parameter values that conceptualize fast preferential infiltration (F_L2,frac, F_L2,max, F_L2,bare) are not directly observable. These parameters were estimate within trial and error calculations. Also the constant d, which scales the water diffusion coefficient, was estimated in such a way that the overall diffusion coefficients were of similar order as observed by Bai et al. [2007] and allowed at the same time a sufficient fast exchange between the layers to reproduce observed soil moisture time series. All soil parameters were assumed to be homogeneous within one simulation site of 100 by 100 grid cells, i.e., 25 ha.

Table 2. Model Parameters for Four Sites in Israel

	Abbreviation	Unit	Ein Ya'aqov	Matta	Lahav	Sde Boqer	Source
Latitude			33.01	31.71	31.39	30.87
Longitude			35.24	35.07	34.9	34.77
Climate			mesic-Mediterranean	Mediterranean	semiarid	arid	Holzapfel et al. [2006]
Soil type			clay	clay	clayey-loam	sandy loam	Zwikel et al. [2007]
Effective suction	Sf	mm	313.6	313.6	208.8	110.1	Rawls et al. [1992]^aa Literature data, based on soil type.
Saturated hydraulic conductivity	Ks	mm/h	0.6	0.6	2	21.8	Rawls et al. [1992]^aa Literature data, based on soil type.
Residual water content	rw	m^3/m^3	0.09	0.05^bb Value reduced because of field observations.	0.027	0.041	Rawls et al. [1992]^aa Literature data, based on soil type.
Water content of beginning stomatal closure	wsc	m^3/m^3	28.5	23.5	16.5	8	estimated as matric potential of –1500 kPa
Field capacity	fc	m^3/m^3	41.5	36.5	29.5	24	S. Pariente (personal communication, 2007)
Depth of layer 1	depthL1	mm	100	100	100	200	estimated
Depth of layer 2	depthL2	mm	900	900	900	800	estimated
Constant for diffusion coefficient	d	–	0.01	0.01	0.01	0.01	estimated according to Bai et al. [2007]
Mean grass cover	cg,mean	–	0.2	0.4	0.25	0.02	M. Köchy (personal communication, 2007)
Mean shrub cover	cs,mean	–	0.7	0.5	0.25	0.05	M. Köchy (personal communication, 2007)
Evapotranspiration factor	aET	–	1	1	1	0.4	calibrated^cc Calibrated value for Sde Boqer, factor does not influence evapotranspiration at other sites.
Bare soil infiltration rate into layer 2	FL2,bare	–	0.1	0.1	0.5	0.8	estimated dep. on K
Infiltration rate into layer 2	FL2,frac	–	0.1	0.1	0.1	0.1	estimated
Maximal total infiltration into layer 2	FL2,max	mm/h	0.15	1.5	1.5	1.5	estimated

• a Literature data, based on soil type.
• b Value reduced because of field observations.
• c Calibrated value for Sde Boqer, factor does not influence evapotranspiration at other sites.

[26] To assess the topography we interpolated data of available digital elevation models with a resolution of 30 m by 30 m to a resolution of 5 m by 5 m. Mean slopes are 28% at Ein Ya'aqov, 34% at Matta, 17% at Lahav, and 12% at Sde Boqer. Each site is characterized by a long history of domestic grazing, with mainly goats and sheep in the south and cattle in the north.

[27] Hourly precipitation and temperature data available at each site for the years 2002–2004 were used as spatially homogeneous model input. Soil moisture observation in the upper 2–6 cm (S. Pariente, personal communication, 2007) at each site at up to three locations under and between bushes were used to test the model performance. The depth of the upper soil layer was 10–20 cm and 80–90 cm for the lower layer to account for different rooting depths of grasses and shrubs.

[28] The dominant vegetation type at the sites is described by Holzapfel et al. [2006] with closed oak maquis to more open garrique dominated vegetation with coexisting herbaceous vegetation at Ein Ya'aqov, dwarf shrubs and herbaceous plants at Matta and Lahav, and small shrubs and sparsely growing desert annuals and geophytes at Sde Boqer. A frequently arising difficulty when performing spatially distributed simulations is to assess the actual vegetation distribution in the field, whereas estimations of the total vegetation cover are often available. Although vegetation distribution is strongly related to topography and soils, we tested model performance for randomly distributed vegetation that follows a normal distribution with a mean cover lying within a range that was visually estimated (M. Köchy, personal communication, 2007) and a standard deviation of mean value divided by two. This random distribution was chosen to disentangle the effects of vegetation cover and topography on soil moisture.

[29] We assessed the importance of spatial water redistribution by evaluating the mean annual soil moisture dependent on the flow accumulation index (fai) of each cell. We parameterized the model with data from Matta (climate data: year 2002) and assumed an artificial underlying elevation model of 100 by 100 grid cells that is described in Text S1. Within this grid vegetation was distributed randomly as described above. To analyze the influence of vegetation cover on soil moisture and processes we performed additional multiple simulations with each grass and shrub cover ranging from 0% to 60%.

3.2. Simulated and Observed Soil Moisture

[30] The overall accordance of simulated with measured soil moisture values in the upper layer is, especially for the three northern sites Ein Ya'aqov, Matta and Lahav, surprisingly good with high coefficients of determination, R², for nearly all years (Figure 3). Most of the peaks after rainfall and water losses due to evapotranspiration processes can be reproduced for different climatic regimes and different soil types. Nonetheless, peak soil moisture values after heavy rain events are underestimate for the most southern site Sde Boqer, where the underlying geology is quite complex, and soils are heterogeneous. Obviously, the model underestimates infiltration during heavy events, and as a result soil moisture is too low. Part of this underestimation can be explained by the inconsistency of rainfall and soil moisture data due to their spatial distance. Nonetheless, we must admit that the model has clear limitations when applied to sites with heterogeneous soils and complex bedrock.

Figure 3

Comparison between simulated and measured soil moisture data. For each site (Ein Ya'aqov, Matta, Lahav, and Sde Boqer), measured volumetric soil moisture in the upper soil (gray) was compared with simulated moisture (black), and the corresponding R² was calculated. Missing data are indicated by gaps in the lines. For each site and year, the corresponding precipitation is given below the soil moisture values.

[31] Soil moisture for the lower soil layer was unfortunately not available. It exhibits a lower temporal variability, both at the daily as well as at the long-term scale. Plant available soil water in deeper layers remains high during summer, when water has already evaporated from the surface layer.

[32] The presented evidence suggests that the model suites well to is purpose, which is the prediction of soil moisture dynamics at the long-term scale as a primer resource for plant growth. It corroborates furthermore that estimated values for the model parameters d and F_L2,frac, F_L2,max and F_L2,bare allow a consistent performances.

3.3. Model Sensitivity and Lateral Redistribution due to Runoff

[33] Figure 4 quantifies the effect of vegetation cover on annual average soil moisture (Figure 4a) as well as lateral redistribution of soil water due to surface runoff (Figure 4b). The flow accumulation index (fai) reflects the number of grid cells that potentially contribute runoff water to a respective cell. Soil moisture of the upper layer slightly and soil moisture of the lower layer strongly increases with increasing log(fai) from an average water content of 16.6 vol% at cells with almost no runon to about 27 vol% at cells with a potentially high amount of runon. Since we distributed vegetation randomly and therefore disentangled the effects of vegetation cover and topography, this structured variability in soil moisture can largely be attributed to local water redistribution. Thus, we can state that the model can account for the important process of water redistribution in semiarid areas.

Figure 4

Influence of grass and shrub cover and water redistribution on the soil moisture regime. The results are presented for the site Matta (year 2002) and show (a) the mean annual soil moisture in both layers dependent on grass cover c_g and shrub cover c_s. The cover of the respective other plant type is kept at 30%. (b) The spatial water redistribution, where mean soil moisture is evaluated dependent on the logarithmic flow accumulation index fai of each grid cell.

[34] As suggested by Figure 4a grass and shrubs have a rather different effect on mean annual moisture content of the upper soil layer. Increasing the cover of both plant types separately and keeping the cover of the respective other plant type fixed at 30% leads to a slight increase in moisture for grasses and to clear a decrease for shrubs. Structured variability of shrubs (for instance in a banded form) would therefore yield in a different pattern of runoff, than the observed pattern for randomly distributed vegetation cover.

[35] The opposite effects of grass and shrub cover on average annual soil moisture may be further illuminated when looking at how vegetation cover affects different water fluxes in Figure 5. On the one hand, grasses and shrubs both exhibit a positive influence on soil moisture because they increase infiltration, decrease runoff, and decrease evaporation of surface water. However, shrubs increase preferential infiltration into the lower layer as well, therefore less water infiltrates into the upper layer. Despite this increased infiltration, soil moisture in the lower layer decreases with shrub cover (Figure 4a) because of a strongly increased transpiration rate at high shrub cover that exceeds the gain due to increased infiltration (Figure 5f). Thus, the water amount in the lower soil layer reaches higher peak values at a higher shrub cover, but the water cannot be retained over a longer time period. Increasing grass cover has little effect on the soil moisture content of the lower layer. The diffusive flux between the upper and the lower layer increases with vegetation cover, which is caused by the increased difference in soil water content.

Figure 5

Influence of grass and shrub cover on water fluxes. The results are presented for the site Matta (year 2002) and show (a) infiltration, (b) the diffusive flux between the upper and the lower layer (positive values indicate a flux from the upper to the lower layer), (c) total runoff from the border of the grid, (d) evaporation from the surface, (e) evaporation from the upper soil layer, and (f) transpiration from the lower soil layer. The cover of the respective other plant type is kept at 30%.

[36] Hence, we can summarize that the soil moisture regime in upper layer benefits more from increasing the vegetation cover than the lower layer. This can be explained by the reduced evaporation in the upper layer and the increased transpiration losses from the lower layer. A further elaboration of the model sensitivity with respect to the selected depth of the soil layers is presented in Text S2 and Figure S1.

[37] Figure 6 condenses effects of vegetation on different processes in terms of the annual water balance. Although shrubs as well as grasses lead to a slightly higher total infiltration, water losses by evapotranspiration exceed these positive effects for shrub cover. In contrast, evapotranspiration remains constant for higher grass cover; that is, additional water can be stored more efficiently. Water losses by surface evaporation and runoff are affected similarly by grasses and shrubs: both surface evaporation and runoff decrease with vegetation cover. Thus, we may state that grass and shrub cover exert a strong control on subsurface water stocks and lateral redistribution of water in the landscape, and therefore on their primer resource. Simulations with a coupled dynamic vegetation model can therefore be expected to shed light on important feedbacks in dry lands.

Figure 6

Key terms in the annual water balance for Matta (year 2002) for varying shrub and grass cover. The cover of the respective other plant type is kept at 30%. Evapotranspiration is shown as subunit of infiltration (indicated by the cross pattern).

4. Simulating Climate Change

4.1. Setup of Simple Climate Change Scenarios

[38] We assessed the possible effects of climate change on soil moisture dynamics and on fluxes as an example for the site Matta with the same soil properties and vegetation cover as given in Table 2. For this, we generated a climate reference scenario with the model ReGen [Köchy, 2006] that was developed to simulate daily precipitation along a climate gradient in Israel. In ReGen, the probability of rainfall on a given day of the year is described by a regular Gaussian peak curve function. The amount of rain is then drawn randomly from an exponential distribution whose mean is the daily mean rain amount (averaged across years for each day of the year) described by a flattened Gaussian peak curve. We set the mean annual precipitation to 537 mm and used the default ReGen parameter set to gain precipitation for 100 years. Daily precipitation data were disaggregated to hourly data with the distribution shown in Figure 7.

Figure 7

Distribution of daily precipitation. Daily precipitation data are disaggregated to hourly precipitation with the indicated fractions.

[39] Daily and hourly temperature data (T_day, T_hour) were fitted to measured values from 2002 to 2004 with the following equation:

with mean yearly temperature T_year = 17.9°C, mean yearly fluctuations in temperature T_diff,year = 10.0°C and mean daily fluctuations T_diff,day = 7.5°C.

[40] Additionally to the reference climate scenario, we generated climate change scenarios with (1) a differing amount of total annual mean precipitation, (2) altered daytime temperatures, and (3) a varying number of extreme events. We run our model with these climate data and evaluated the effect of the changes on soil moisture dynamics and on selected fluxes. The total annual precipitation was changed by multiplying hourly rain in the reference scenario by 0.6 to 1.4; that is, we generally gained the same precipitation events, each with less or more rain, respectively. To simulate raised temperature values, we increased the mean annual temperature by 0.75 to 3.0°C by altering the daytime temperature. Extreme events were simulated by varying the optional model parameter influencing the amplitude of a single rain event (ΔDMR) in ReGen from −0.2 (less variable rain) to 0.2 (more variable rain), with the same annual mean precipitation as in the reference scenario.

4.2. Results

[41] The effects of the simulated climate change scenarios are shown in Figure 8 with changed total precipitation (Figures 8a and 8d), increased temperature (Figures 8b and 8e), and altered number of extreme events (Figures 8c and 8f). Increased total annual precipitation generally leads to higher soil moisture and especially to more days per year, in which the soil moisture lies above given thresholds (Figure 8a). In particular, the lower soil layer benefits strongly from increased precipitation. However, the additional water does not fully recharge soil moisture, since runoff increases faster than linearly (Figure 8d).

Figure 8

Effects of possible climatic changes on soil moisture conditions and on selected fluxes at Matta. (a–c) The annual mean number of days where soil moisture of the upper and lower layer exceeds a certain threshold and (d–f) the effect on important water fluxes. Figures 8a and 8d show the effect of changed annual mean precipitation, Figures 8b and 8e show the effect of increased daytime temperature, and Figures 8c and 8f show the effect of an altered number of extreme events. The parameters for the respective climate scenarios are listed in the legend.

[42] The predicted temperature increase in the course of climate change leads to a strong decrease in soil moisture (Figure 8b). Because of an enhanced potential evapotranspiration (Figure 8e), the soil dries out faster and moisture cannot be held in the soil for a long time period. This effect is more intense for the lower layer.

[43] The effect of altered rainfall variability that leads to a change in the number of extreme events is ambiguous (Figure 8c): the lower layer generally benefits from more variable rain, while the upper layer becomes slightly drier. As for the increase in the total precipitation amount, runoff also increases with a higher number of extreme events (Figure 8f).

5. Discussion

[44] Model results show a close link between vegetation cover and soil moisture. Generally speaking, grasses exert a positive influence on soil moisture and shrubs decrease the overall moisture. This becomes especially important if the model is coupled with a dynamic vegetation model, where vegetation cover responds to water availability and varies over time. In this case, additional positive feedback loops between water availability in one cell as a result of topography as shown in Figure 4 and vegetation cover would arise: vegetation would accumulate in flow paths, as it can be observed in many dry lands and would further influence the local water availability. The strong linkage in the model of processes to vegetation cover and local water redistribution is an important feature of the hydrological model that allows for pattern formation in the landscape. Additionally, coupling the hydrological model with a dynamic vegetation model and comparing the arising vegetation formation with field data would lead to further confidence in the choice of model parameters.

[45] Increased precipitation intensity of single events as it is expected in the future [Easterling et al., 2000; Dore, 2005] leads to a higher soil moisture content in the lower soil layer. This is caused by increased preferential flow and increased drainage from the upper to the lower layer. Because of this increased infiltration into the lower layer, soil moisture is generally higher and can be longer maintained for both scenarios: increased total precipitation and increased number of extreme events. At the same time, more intense rain events lead to enhanced Hortonian runoff, since the hydraulic conductivity limits infiltration speed. This results in a lower soil moisture content of the upper layer.

[46] Additionally, water in the upper layer experiences little protection against evaporation loss. Especially at a high soil moisture content, e.g., after highly intense rainfall events, water evaporates quickly and the soil moisture level cannot be maintained long. Contrary to this, water in the lower layer is stored longer: once water has entered the lower layer it is well protected against evaporation. However, as Figure 8b shows this protection is strongly reduced by higher air temperature, since the potential evapotranspiration is increased. This implies that the expected future temperature increase will strongly decrease the time span that water remains in the soil.

[47] The results of the climate change scenarios show that multiple factors determine soil moisture patterns under climate change. Apart from temperature and the mean annual precipitation, especially its intra-annual distribution is of great importance. This contrasts the assumptions of many present models of vegetation dynamics in semiarid and arid systems, which include precipitation on a yearly basis and simply assume a linear correlation between annual precipitation and water availability to vegetation (recent review by Tietjen and Jeltsch [2007]). Additionally, the effects of increasing temperature are often neglected in these models.

[48] The predicted changes in soil moisture patterns in the upper and lower layer have strong implications for vegetation growth and composition. Dependent on the stage of the life cycle, plants have different water requirements in different soil depths. Moisture of the upper soil layer is for example crucial in seed germination of annuals. Several studies have been conducted on the germination success of Israeli annual species dependent on the prevailing soil moisture or on precipitation events (e.g., Hordeum spontaneum [Gutterman and Gozlan, 1998]; Blepharis spp., Anastatica hierochuntica [Gutterman, 2000]; and various species [Köchy and Tielbörger, 2007]). The general outcome of these studies is that many species need a species- and soil-type-dependent amount of soil moisture maintained for a certain amount of time to germinate successfully. Our model results indicate a decrease in the amount of moisture in the upper soil layer under future climate conditions. According to the studies, both trends will worsen general conditions for germination. Friedman et al. [1981] additionally analyzed the effect of seed hydration prior to dehydration on the survival of seeds and seedlings in the desert annual Anastatica hierochuntica as well as the effect of the duration of storage of the dehydrated seeds/seedlings. Both treatments markedly influenced the survival probability of seeds/seedlings relative to control conditions: Furthermore, the later the drought occurred in their germination process and the longer they were stored after dehydration, the higher the mortality. This indicates that there is a “point of no return” after which sustained soil moisture becomes critical [see also Gutterman and Gozlan, 1998]. In the light of our model results, species that managed successful germination and establishment under previous conditions could face a riskier future under climate change.

[49] A changing water content of both soil layers will also affect shrubs. Lower soil moisture in the upper layer here as well reduces the probability of successful germination and establishment. This can lead to a decrease in species diversity as Lloret et al. [2004] showed in a study in a Mediterranean shrubland. In another evaluation, Henkin et al. [1998] analyzed the biomass production of herbaceous species in a Mediterranean ecosystem and came to the conclusion that not only annual precipitation is crucial, but mainly its conversion into effective growing days. On the other hand, increased soil moisture in the lower layer, as it can be expected from more extreme precipitation events, can improve the conditions for shrubs once they have access to the lower soil layer, which often occurs within the first year. This can change competitive ability and thus lead to a shift in the species composition between those plant types with varying access to lower soil layers. Additionally, changed soil moisture patterns can lead to increased vulnerability to invasive species. A review by Dukes and Mooney [1999] highlights how long-term establishment of alien plant species may be facilitated by a short-term increase in water availability. Other studies report higher risks of invasion caused by the expected increase of water use efficiency of invasive species due to higher atmospheric CO₂ concentrations [Polley et al., 1997; Kriticos et al., 2003]. In this case, lower soil moisture conditions harm native plant species more, while more water efficient invasive species take over. Once alien species have invaded, they can markedly change the conditions for surrounding species, e.g., by further changing the water availability [Melgoza et al., 1990; D'Antonio and Mahall, 1991]. In this study, we did not evaluate seasonal short-term conditions explicitly, but these kinds of analyses are easily feasible with the introduced model.

6. Conclusions and Outlook

[50] The presented examples stress the importance of exploring the linkage of hydrology and vegetation under climate change. A thorough investigation of the interdependence requires a coupled model of soil moisture and dynamic vegetation, which will be a future task. This coupled model will account for full bidirectional feedbacks between water and, such as facilitation of plant growth by local water redistribution [Ludwig et al., 2005], which can lead to pattern formation.

[51] Concerning our model we think we made a considerable steep toward a model that is at a minimal acceptable complexity level for hydrologists and at the same time is a parsimonious hydrological tool for ecologists to simulate plant water availability for fragile semiarid ecosystems. We estimated most of the parameters in this study on the basis of simple soil survey and literature data. Parameters that were not directly observable in the field where chosen at least consistent as they allow a good reproduction of observed soil moisture time series at several climates. A successful reproduction of observed vegetation patterns with dynamic vegetation at new locations will be the next benchmarks to show that the model works for the right reasons. Thus, on the long-term also hydrology could largely benefit from simulations with dynamic vegetation, e.g., to test feasibility and consistency of model parameters that are difficult to observe but have a crucial influence on plant water availability.

Acknowledgments

[52] This study was supported by the Heinrich-Böll-Foundation and by the German Federal Ministry for Education and Research (BMBF) through the framework of the GLOWA Jordan River project (01LW0306) and BIOTA Southern Africa (01LC0024). We thank the editor, the associate editor Andrew Guswa, and three anonymous referees for their valuable comments; Sarah Pariente and Eyal Sachs for provision of data; Martin Köchy, Kai Lessmann, Dan Malkinson, and Katja Schiffers for fruitful discussions; and Richard Walters for proofreading the manuscript.