The magnitude of hydraulic redistribution by plant roots: a review and synthesis of empirical and modeling studies


Author for correspondence:
Zoe G. Cardon
Tel: +1 508 289 7473



II.Synthesis of the magnitudes of HR across ecosystems338
III.Hydraulic redistribution models339
IV.Methodological considerations affecting the magnitude of HR344
V.Site characteristics affecting the magnitude of HR346
VI.Plant characteristics affecting the magnitude of HR347


Hydraulic redistribution (HR) – the movement of water from moist to dry soil through plant roots – occurs worldwide within a range of different ecosystems and plant species. The proposed ecological and hydrologic impacts of HR include increasing dry-season transpiration and photosynthetic rates, prolonging the life span of fine roots and maintaining root–soil contact in dry soils, and moving rainwater down into deeper soil layers where it does not evaporate. In this review, we compile estimates of the magnitude of HR from ecosystems around the world, using representative empirical and modeling studies from which we could extract amounts of water redistributed by plant root systems. The reported average magnitude of HR varies by nearly two orders of magnitude across ecosystems, from 0.04 to 1.3 mm H2O d−1 in the empirical literature, and from 0.1 to 3.23 mm H2O d−1 in the modeling literature. Using these synthesized data, along with other published studies, we examine this variation in the magnitude of upward and downward HR, considering effects of plant, soil and ecosystem characteristics, as well as effects of methodological details (in both empirical and modeling studies) on estimates of HR. We take both ecological and hydrologic perspectives.

I. Introduction

Hydraulic redistribution (HR) – the movement of water from moist to dry soil through plant roots – was demonstrated in laboratory experiments as early as the 1930s (Breazeale, 1930; Breazeale & Crider, 1934). Half a century later, the ecological relevance of HR in the field was suggested by Mooney et al. (1980), who proposed night-time HR of groundwater, supporting next day transpiration, as an explanation for the survival of Prosopis tamarugo trees in the rainless Atacama Desert. The first field demonstration of night-time HR occurred half a decade later, in 1987, using thermocouple psychrometers installed around the root systems of sagebrush in the semiarid western United States (Richards & Caldwell, 1987). Since then, HR exploration has expanded from its initial focus on arid and semiarid landscapes into other regions, often with pronounced dry seasons, but not necessarily low total annual rainfall. It is now recognized that HR occurs worldwide within a range of different ecosystems and plant species (Caldwell et al., 1998; Horton & Hart, 1998; Jackson et al., 2000).

The proposed ecological and hydrologic impacts of HR are numerous, among them:

The occurrence of HR within a given environment requires the development of a water potential gradient within the soil, and plants with root systems that span the potential gradient, either vertically or horizontally. Many researchers have searched within and across ecosystems for commonalities and differences among soil, climate, and above- and below-ground plant characteristics that affect HR, and variation in the magnitude of HR can be substantial (Yoder & Nowak, 1999; Ishikawa & Bledsoe, 2000; Hultine et al., 2003; Espeleta et al., 2004; Meinzer et al., 2004; Zou et al., 2005; Warren et al., 2007; Scholz et al., 2008b, 2010). Ultimately, the hydrologic and ecological significance of HR depends on its timing as well as its magnitude. If HR contributes a considerable fraction of water necessary to satisfy the transpiration demand, its direct hydrologic effect may be large in ecosystems where transpiration is a significant contributor to the land–atmosphere water flux. If HR contributes only a very small percentage of transpirational water, it likely does not have a direct, short-term hydrologic impact. However, small upward HR at key times might still be ecologically relevant, reducing the development of root embolism during drought (Domec et al., 2004, 2006; Bauerle et al., 2008), supporting the establishment of seedlings or shallow-rooted competitors (Dawson, 1993; Moreira et al., 2003; Ludwig et al., 2003; Hawkins et al., 2009), or influencing microbial activity and nutrient availability (Snyder et al., 2008; Aanderud & Richards, 2009; Wang et al., 2009). And, in the long term, the ecological importance of small HR, manifested through plant community establishment and persistence, could well have large, landscape-level, hydrologic effects.

In this review, we compile information from a number of representative empirical and modeling studies that report amounts of water redistributed by plant root systems during HR. Using these synthesized data, along with other information from the literature, we investigate sources of variation in estimated magnitudes, considering methodological details (both empirical and modeling) as well as plant and environmental characteristics. We take both ecological and hydrologic perspectives, and we mostly limit our discussion to measurements and modeling of natural landscapes.

II. Synthesis of the magnitudes of HR across ecosystems

We used 29 published papers focused on 16 different ecosystems for which the magnitude of HR was explicitly stated or could be calculated from empirical data or model output. Fig. 1 presents the average and maximum magnitudes of HR determined for each study in each ecosystem, in order of increasing average magnitude. The magnitude represents the volume of water redistributed by the roots within the instrumented or modeled soil depth, either divided by the root area of influence (plant area basis) or divided by the root area of influence and multiplied by the fractional coverage of vegetation at the site (land area basis). The ‘root area of influence’ is the area of soil surface beneath which there are plant roots influencing the soil moisture measured to calculate HR.

Figure 1.

Average and maximum upward and downward hydraulic redistribution (HR; in mm H2O d−1), from both empirical (blue bars) and modeling (red bars) studies, expressed on a per plant area basis (bars without texture) or on a land area basis (stippled bars). Numbers on bars identify the source paper (first number) and substudy or experiment within that publication (second number, after hyphen), corresponding with entries in Table 1.

Table 1 identifies the source papers for the magnitudes of HR plotted in Fig. 1. The values plotted in Fig. 1 are included in the Supporting Information,Table S1, along with standard deviations in magnitudes of HR (where calculable), and more in-depth descriptions of the data sources, data processing, and information about transpiration rates in these ecosystems. Numbers superimposed on the bars in Fig. 1 refer to the number assigned to each paper in the tables; where two numbers are shown separated by a hyphen on a single bar, the first is that assigned to the paper and the second differentiates between empirical datasets (or modeling runs) within the paper.

Table 1.   Papers from which the magnitude of hydraulic redistribution (HR) was extracted,a grouped in gray and white by ‘sites and species’, and coded by ‘study #’ to correspond to the labels on bars in Fig. 1, with field studies at top and modeling studies below Thumbnail image of

Empirical estimates of the average magnitude of upward HR span more than an order of magnitude, from 0.04 mm H2O d−1 in Brazilian Savanna (Scholz et al., 2010) to 1.3 mm H2O d−1 in New England sugar maples (Emerman & Dawson, 1996), with a mean of 0.3 mm H2O d−1 (= 31 light blue bars, Fig. 1; upward HR moves water from deep, moist soil layers into shallow, dry soil layers). Downward HR ranged between 0.2 and 1.7 mm H2O d−1, with mean of 1.0 mm H2O d−1 (= 5 dark blue bars, Fig. 1; downward HR moves precipitation to deep soil). For those ecosystems where transpiration was also reported, the magnitude of upward HR represented c. 2–80% (mean of 15%, = 25) of the water lost by transpiration (Fig. 2). For the one ecosystem in Arizona where downward HR was recorded and seasonal transpiration rates were available, the total amount of downward HR represented 11–49% of total estimated dry-season transpiration (Scott et al., 2008). In this ecosystem and in Western Australia, downward HR supplied deep soil layers with c. 1 month’s worth of the water extracted by deep roots during the dry season (Burgess et al., 2001; Hultine et al., 2004). The approaches to modeling HR are described in the next section of this review, but in general, modeling studies (red bars) tended to produce HR estimates on the upper end of the spectrum, predicting 0.1–3.2 mm H2O d−1 of average upward and/or downward redistribution with a mean of 0.9 mm H2O d−1 (= 17). These magnitudes represent 2–143% (mean of 30%, = 17) of modeled transpiration rates (Table S1).

Figure 2.

Average hydraulic redistribution (HR) as a percentage of average transpiration for the studies presented in Fig. 1 that reported transpiration. Numbers on bars refer to publications listed in Table S1, which provides transpiration information.

What drives this variation in estimated HR across ecosystems? In the following sections, we outline the methodological differences (both empirical and modeling), site characteristics, and plant characteristics that likely contribute.

III. Hydraulic redistribution models

First, we summarize several mathematical models of HR, so that in subsequent sections we can discuss model results and compare them with empirical studies. Most of the model formulations use Richards’ equation (Richards, 1931) to describe water movement through unsaturated soil and include a source/sink term to describe uptake or release by plant roots. Variation among models largely exists in the formulation and complexity of the source/sink term. The overall goal of the models is to capture the dynamics and magnitude of HR as influenced by soil water contents (related to water potentials in soil-specific ways); by soil, soil–root, and root conductivities (the ease with which water flows); and by the forces driving water flow. The following modeling sections build upon a description of the first HR modeling effort (Ryel et al., 2002), outlining components of the soil–plant system that subsequent modeling approaches aimed to clarify.

1. Soil layer connection models

A majority of HR modeling studies use a Darcy’s law formulation as the source/sink term in the soil water balance equation (Richards’ equation). Water redistributes between soil layers based on differences in soil water potential, and redistribution is also influenced by soil-to-root conductances along the flow path. Ryel et al. (2002) were the first to develop this type of model, and their mathematical description of HR is the one most commonly incorporated into larger-scale models:

image(Eqn 1)

HRi describes the net water movement into soil layer i from all other layers (j). Ryel et al. (2002) use a constant, maximum soil–root conductance value for the entire soil–root system, Cmax. Cmax is reduced using an empirical relationship from van Genuchten (range 0–1, Simùnek et al., 1996) as soil water potential Ψ decreases (i.e. soil dries) in the source (cj) or the sink (ci) soil layers. Conductance is distributed among soil layers as a function (Rij) of root biomass distribution in the layers. Because this approach does not model flow within the root system itself, and therefore does not simulate root water potential, it cannot easily capture the competition for xylem water between atmospheric water demand (via transpiration) and dry soil layers. Ryel et al. therefore include a ‘on/off’ term, Dtran, that restricts redistribution to periods with low transpiration demand.

Zheng & Wang (2007), Baker et al. (2008), and Wang (2011) adopted Ryel et al.’s formulation. Scholz et al. (2010) slightly altered the effective conductance calculation to focus on the drying (water-receiving) soil layer’s control over flow. In many published modeling studies (Ryel et al., 2002; Zheng & Wang, 2007; Baker et al., 2008; Scholz et al., 2010; Wang, 2011), HR was turned off during the day (Dtran = 0) and on at night (Dtran = 1) as a simple switch.

In contrast to Ryel et al.’s (2002) constant maximum conductance approach, Lee et al. (2005) used a maximum soil–root conductance value that scales with leaf area index (LAI) and is reduced based on soil water potential following an empirical relationship distinct from van Genuchten. Using LAI captures the idea that root–soil conductivity must be sufficient to support the established plants’ maximum transpirational requirement. However, this link also results in modeled HR decreasing as canopy LAI decreases, and stopping when plants drop leaves or senesce (Lee et al., 2005; J-E. Lee, pers. comm.). In reality, HR can occur in senesced and dormant plants (Hultine et al., 2004; Leffler et al., 2005; Scholz et al., 2008b). Eqn (1) in Lee et al. (2005) also indicates that there is no ‘on/off’ term (analogous to Ryel et al.’s Dtran) included to restrict HR to periods with low transpiration. HR solely depends on water potential differences among soil layers and hydraulic conductance. Lee et al. (2005) assumed that all plants can carry out HR, and their model results ultimately suggested a substantial influence of HR on tropical convective climate.

2. Big root models

Mendel et al. (2002) and Amenu & Kumar (2008) built upon the soil-connection formulation by incorporating radial flow into and out of, and axial flow through, the root system itself. Roots are treated as a bundle of parallel, laminar-flow pipes with identical pressure gradients driving flow. The system is described by a set of coupled differential equations that solve for flow within the soil and root systems. The two domains are connected by flow into and out of the root system, which depends on the radial conductance of the root system and the water potential difference between the root and soil. Radial root conductivity scales with soil saturation, and both effective axial and radial root conductivity depend on root density. Transpiration is incorporated as the top boundary condition on the root domain. This approach inherently captures competition for plant water between the atmosphere and soil, and thus an analog of Dtran (the on/off term) is not required.

The two studies differ in their solutions to the developed coupled differential equations. Mendel et al. (2002) solve the equations for a radial domain and use a prescribed transpiration rate. Amenu & Kumar (2008) solve them for a one-dimensional vertical domain and use the Penman–Monteith equation to solve for transpiration.

3. Macro–meso scale models

Siqueira et al. (2008) developed a model that couples vertical macro-scale flow through the soil column (Richards’ equation solved for meter-scale, one-dimensional vertical flow) with radial, meso-scale flow towards individual rootlets (Richards’ equation solved for mm-scale, radial flow around roots). The formulation captures dynamic water gradients around individual roots as they deplete and refill the surrounding soil. Horizontal homogeneity is assumed such that all rootlets in a layer experience the same radial gradients. Uptake and release of water by roots depends on radial root conductance and the difference between root and soil water potentials. Unlike the ‘big root’ models described in the previous section, axial flow through the root system is not included. The authors argue that a representation of water potential gradients within root xylem is not necessary because water potential is hydrostatically distributed and simply adjusts to maintain the transpiration demand (i.e. no capacitance). (This simplification contrasts with literature describing water potential gradients within root systems (e.g. Landsberg & Fowkes, 1978; Doussan et al., 1998; Steudle & Peterson, 1998; Zwieniecki et al., 2002; Doussan et al., 2006) and is based on the idea that water potential gradients within the root system are small compared with those across the root–soil interface and within soil.) Thus, the entire root system has the same pressure, which depends on atmospheric evaporative demand, leaf water potential and root-to-shoot hydraulic resistance. Because soil, root and leaf water potentials are coupled, an iterative solution scheme is required.

4. Dynamic root profile model

Schymanski et al. (2008) explored how a dynamic root profile affects HR. In their model, root water uptake depends on conductance of the root–soil interface and on the potential difference between root and soil. Root water potential is determined from the suction force exerted by above-ground tissue after accounting for loss of hydrostatic head as water moves from the soil surface to the soil layer of interest. This suction force is controlled by the difference between root water uptake and transpiration. In contrast to the previous models, the root profile changes in response to soil moisture distributions, following an optimization that minimizes root system carbon cost while meeting the canopy water demand. The result is growth or loss of roots, distributed among soil layers, with the rate of change constrained by a maximum daily root growth or turnover rate.

IV. Methodological considerations affecting the magnitude of HR

The almost two orders of magnitude variation in HR shown in Fig. 1 could result, at least partially, from methodological differences. Below, we briefly explore this potential for models and empirical studies.

1. Methodological considerations – modeling

HR formulation  Few of the formulation variations described earlier have been applied and compared in the same ecosystems, although Lee et al. (2005, study #25), Zheng & Wang (2007, study #26) and Wang (2011, study #29) did all examine HR in the Amazon forest. All used the ‘soil layer connection’ formulation, but, as discussed earlier, Zheng & Wang’s (2007) and Wang’s (2011) model used a constant maximum root–soil conductance term and an ‘on–off’ switch for HR, whereas Lee et al. (2005) used a variable maximum root–soil conductance term linked to LAI and no ‘on–off’ switch for HR. Lee et al.’s model estimated substantially more HR (0.4 mm H2O d−1 more), on average, for the Amazon forest than that of Zheng & Wang (2007) and Wang (2011), but a concerted comparative modeling effort would be needed to identify the main drivers of this difference.

Modeled behavior of roots in the soil profile  The basic forms of the models described earlier, with the exception of Schymanski et al. (2008), consider the root system architecture to be static. Within this framework, Zheng & Wang (2007, #26) explored how a more physiologically dynamic depiction of water uptake affects the magnitude of HR.

Zheng & Wang (2007)modeled HR and its effects in the Amazon using two land surface models: the Community Land Model, version 3 (CLM3, Oleson et al., 2004) and the Integrated Biosphere Simulator, version 2 (IBIS2, Kucharik et al., 2000). To capture ‘dynamic root water uptake’ independent of HR, CLM3 and IBIS2 were modified so that photosynthesis and transpiration were not immediately diminished when any portion of the root system was water-stressed; instead, a threshold integrated soil water availability had to be reached before above-ground physiology was affected. Zheng & Wang (2007) also dynamically allocated plant water uptake among soil layers, limiting uptake from dry layers and promoting uptake from moist layers. (Plants might physiologically alter root water uptake rates in this way by differentially adjusting radial conductance via aquaporin expression in root cell membranes (Henzler et al., 1999; McElrone et al., 2007).) Water uptake from deep moist layers does occur in the Amazon (Nepstad et al., 1994; Davidson et al., 2011), though whether, and for how long during drought, deep root uptake can increase sufficiently to compensate for decreased water flow from drying upper soil layers remains unclear (Markewitz et al., 2010; Bleby et al., 2010). By comparing model runs incorporating HR alone with runs featuring both HR and dynamic root water uptake, Zheng & Wang (2007) demonstrated that the modeled amount of HR decreased substantially when dynamic root water uptake was included, for example, by 0.19 mm H2O d−1 in the IBIS2 runs (#26-3 vs #26-4), because more uniform soil moisture extraction along the root profile reduced soil water potential gradients. The ‘dynamic root profile’ formulation developed by Schymanski et al. (2008) also demonstrated that the modeled amount of HR decreased dramatically when a dynamic (rather than static) root system grew biomass in moist soil layers (#28-1 vs #28-2, Fig. 1).

Constraining modeled HR  The impetus for including HR in large-scale models of ecosystem function emerges from the potential for HR to improve the match between modeled and measured transpiration rates (Zheng & Wang, 2007; Amenu & Kumar, 2008), surface temperatures (Lee et al., 2005), or net ecosystem exchange (Baker et al., 2008). However, such an improved match is not necessarily proof that HR is occurring. For example, Zheng & Wang (2007) found that including HR in CLM3 and IBIS2 without including dynamic root water uptake (described earlier) produced similar improvement in the modeled transpiration rate for their study site, as did including dynamic root water uptake without HR. Markewitz et al. (2010) were able to model soil water content during Amazon forest drought accurately by including enhanced water uptake by deep roots, and did not include HR in their model. At Amenu & Kumar’s (2008) Sierra Nevada site, inclusion of HR improved the model’s match to measured transpiration rates by increasing dry-season transpiration, but, surprisingly, worsened the match to measured dry-season surface soil moisture because, presumably, the model plant extracted moisture out of the wrong soil layers. Together, these results advocate for constraining models with multiple data types, matching output to both transpiration rates and soil moisture profiles whenever possible.

2. Methodological considerations – empirical

The several approaches taken by empirical studies to estimate the magnitude of HR in ecosystems could also be partially responsible for some of the variation apparent in Fig. 1. Below, we consider effects of the time-frame over which measurements are made, and the sensors used to detect both upward and downward HR.

Time-frame of HR measurements  As noted previously, HR is often seen in systems where precipitation is seasonal, and the magnitude of HR varies as seasons progress. Several of the empirical studies that quantified larger amounts of average upward HR (c. 1 mm H2O d−1, studies #1, #2, #4, and #14) used data from ≤ 5 d to estimate HR. Such a short window cannot capture the full range of HR in the system, although it may catch the maximum. For example, based on 31 d of data collection, Z. G. Cardon et al. (unpublished, study #3) estimate that sagebrush in Utah can lift up to 1 mm H2O d−1, but an average of c. 0.1 mm H2O d−1 over the dry season, with 1 mm H2O d−1 more closely matching averages estimated during two other, shorter, sagebrush studies (#1 and #2). However, this time-frame issue cannot explain all the variation evident in Fig. 1. Four years of data were used to quantify HR for the old-growth Ponderosa pine and old-growth Douglas fir stands, and the maximum rate detected during these long-term studies never approached 1 mm H2O d−1. Thus, while a shorter data-collection period may shift an empirical estimate towards higher reported rates of HR, it does not explain the order-of-magnitude difference in maximum rates across ecosystems.

Sensors and upward HR Table 1 summarizes measurement techniques used to generate the estimates of HR in Fig. 1. A majority of the upward HR studies use either thermocouple psychrometers or soil water content sensors. Both sensor types are sensitive to temperature, and temperature corrections and expectations for behavior as a function of soil moisture and texture are available in the literature (e.g. Rawlins & Dalton, 1967; Brown & Bartos, 1982; Wraith & Or, 1999; Ishikawa & Bledsoe, 2000; Warren et al., 2011). Because of this sensitivity, sensors are usually not installed in the shallowest soil layers (< 20 cm, Table 1) where diel soil temperature fluctuations are largest. However, these are the same soil layers in which HR might be largest. Simulations of the Brazil savanna by Scholz et al. (2010) suggest that c. 50% of all redistributed water flows into the top 10 cm of soil, indicating that empirical estimates of HR in Fig. 1 may need to be considered lower estimates.

A cautionary note emerges, however, from recent work by Warren et al. (2011) analyzing and modeling data from the Ponderosa pine forest. They suggest soil temperature fluctuations can also drive night-time soil vapor transport, accounting for up to 40% of the nocturnal moisture increase in uppermost soil layers. Romero-Saltos et al. (2005) also proposed (though see Markewitz et al., 2010 for contrast) that in Tapajos National Forest in Brazil, hydraulic conductivity of clay soils may at times remain high enough for water to flow from moister deep to drier shallow soil, without entering roots. Meinzer et al. (2004) note, however, that because hydraulic conductivity of soil drops precipitously with drying, only rarely is conductivity likely to be high enough to support substantial unsaturated flow upward through drying soil outside of roots.

Many of the studies that estimated larger upward HR (Utah sagebrush #1–3, umbrella thorn #9, and sugar maple #4) used psychrometers to measure soil water potential changes within the root zone. By contrast, many of those that estimated the smaller amounts of upward HR (< 0.5 mm d−1 in the Brazilian savanna #6 and #11, Pacific Northwest #5–8 and #10, and Amazon #12) used soil moisture probes (capacitance or reflectometry) which measure the bulk dielectric constant of the soil (Table 1). However, this correspondence was not absolute; capacitance probes were used in loblolly pine in North Carolina (study #16) where max HR exceeded 1 mm d−1. And Warren et al. (2005, study #7) and Brooks et al. (2002, study #5) deployed thermocouple psychrometers along with capacitance sensors in their Pacific Northwest sites, documenting diel hysteresis in the relationship between soil water content (θ) and soil water potential (Ψ) but generally noting congruence in behavior of the two sensor types. It seems unlikely that sensor type, when each is properly deployed (where diel temperature gradients are minimal) and calibrated (to local soil conditions), would lead to systematically increased or decreased estimates of HR reflected in the variation observed in Fig. 1.

The way that sensor output is transformed into estimates of HR magnitude differs for capacitance probes and psychrometers. Briefly, capacitance probes measure the bulk dielectric constant of soil, which is highly sensitive to water content, and also varies with soil material characteristics, including soil texture (Starr & Paltineanu, 2002). To transform measures of bulk dielectric constant into absolute water contents, capacitance sensors must be calibrated to site soils, but relative changes in moisture content (such as those used to quantify HR) are not as dependent on soil-specific calibration (Baumhardt et al., 2000; Lane & Mackenzie, 2001; Morgan et al., 2001; Geesing et al., 2004; Starr & Paltineanu, 2002; and compare Warren et al., 2005; #7, with Meinzer et al., 2004, #6).

Psychrometers, by contrast, directly sense soil water potential, which must be transformed into water content using water-retention curves specific to soil around the sensors. Substantial soil heterogeneity, horizontal and vertical, can exist even within a single field site, potentially necessitating use of varied water-retention curves to translate water potential into soil water content. Two studies illustrate the point. Emerman & Dawson (1996, study #4) present parameters for diverse water retention curves determined from 24 soil cores collected at their New England site. Using minimum and maximum parameter sets with the psychrometer trace presented in their fig. 1, we calculate an average night-time HR of either 1.65 or 0.97 mm H2O d−1– a 0.68 mm H2O d−1 difference – depending on the parameter set. A similar exercise using vertical variation in soil water-retention curves across the top 60 cm of soil at the young Douglas fir, old Douglas fir and Ponderosa pine sites (Warren et al., 2005) results in calculated HR shifting by 0.03, 0.06 and 0.3 mm H2O d−1, respectively, assuming each site experiences a realistic night-time water potential change of 0.05 MPa (from −1.05 MPa to −1 MPa, Meinzer et al., 2004).

Sensors and downward HR  Downward redistribution has been quantified in three ecosystems (deep blue bars, Fig. 1). In three of four studies (#17–19), the average magnitude of downward redistribution was greater than the largest average magnitude of upward HR (> 1.3 mm H2O d−1). It is possible that the amount of water redistributed downward is, in general, greater than that redistributed upward (Burgess et al., 2001). However, some of the water may have moved downward along preferential flow paths in the soil (Burgess et al., 2001), including along the outside of roots. Downward redistribution was largely estimated from changes in deep soil water content (#17, #19) or water potential (#18) following a rainstorm. In the desert, dye tracers demonstrate that stem flow funnels rainwater to the base of plants where it can be preferentially transported into soil along channels formed by live or dead roots (Martinez-Meza & Whitford, 1996; Devitt & Smith, 2002; Li et al., 2009). Still, for the landscape hydrologic budget, the net movement of rainwater into deeper soil layers, where it cannot evaporate, is of main importance, whether through, or outside of, roots (Burgess et al., 2001; Ryel et al., 2003; Hultine et al., 2004; Scott et al., 2008; Lange et al., 2009). The one study (#20) that estimated smaller downward HR used sap flow measurements. While sap flow separates HR from preferential flow, quantification of HR using sap flow is challenging. There is the possibility that if plant capacitance is large, some sap flow through roots reflects refilling of a plant sink, rather than flow into soil. And, more broadly, using sap flow to estimate HR requires scaling up measurements from a few individual roots on a few plants.

V. Site characteristics affecting the magnitude of HR

Site characteristics also are critical to determining the magnitude of HR. Interestingly, the magnitude of HR reported in Fig. 1 is not related in a straightforward way to annual precipitation. For example, studies of old-growth Ponderosa pine, Brazilian cerrado, and old-growth Douglas fir (Fig. 1) report similar average and maximum HR, but mean annual precipitation of 550, 1500, and 2500 mm, respectively (Meinzer et al., 2004). These systems have a pronounced dry season in common, several months in duration, with only c. 100 mm of precipitation.

1. Access to groundwater

New England sugar maples, which lift > 1 mm H2O d−1 (Fig. 1), have access to a continuous water source, groundwater, beneath a 50 cm fragipan layer which creates a compartmented soil column (Dawson, 1993; Emerman & Dawson, 1996; study #4). Groundwater is also noted as important for strong upward HR in East African umbrella thorn (Ludwig et al., 2003), and oak in southern Portugal (Kurz-Besson et al., 2006). Using modeling, Ryel et al. (2002) demonstrated that for sagebrush growing in a silt loam, access to groundwater increased HR by 0.2 mm H2O d−1 during a simulated 100-d-long drought. Groundwater access may be particularly important in ecosystems where deep soil is sandy, since sandy soils have lower water contents even near field capacity.

2. Soil texture and conductivity

The existing literature suggests that soil texture influences the potential magnitude of HR, with sandier soils promoting less HR. To date, however, comparisons of HR across ranges of soil types within ecosystems are limited. In one comparative survey across multiple sites in the Mojave Desert, the frequency of detectable upward HR decreased as the percentage of sand in soil increased (Yoder & Nowak, 1999). However, Yoder & Nowak (1999) note that vegetation across the sites also varied, potentially confounding the relationship. In controlled laboratory experiments, sandier soils decreased the frequency of upward HR by cotton (Wang et al., 2009) and reduced the rate at which water was redistributed by bean seedlings (Schippers et al., 1967). Burgess et al. (2000) and Scholz et al. (2008b) concluded that the very high hydraulic conductivity of sandy soils in Western Australia, and well-drained oxisols of the Brazilian cerrado, cannot maintain the inverted water potential gradient required for downward HR. Prieto et al. (2010b) studied arid ecosystem shrubs in Chile and Spain, and used a suite of water retention curves to calculate that for the same change in water potential, sandy soils release and adsorb less water than soils with a finer texture. In modeling experiments, sandy soil promoted smaller amounts of upward HR than finer soils (Siqueira et al., 2008).

Soil texture can exert its effects on HR, at least in part, through texture-specific relationships between root and soil conductance and soil moisture. A range of empirical studies has demonstrated that, as surface soils dry, HR often initially increases (as the driving water potential gradient develops in the soil column), reaches a maximum and then either decreases or plateaus (Meinzer et al., 2004; Warren et al., 2005; Scholz et al., 2008b; Prieto et al., 2010a). Models reproduce this drop in HR as soils dry by reducing soil–root conductance. Several mechanisms likely underlie this reduction. The conductance of soils decreases as soils desaturate (Van Genuchten, 1980), and this decrease occurs more rapidly in coarser textured (sandier) soils (Bristow et al., 1984). Root–soil contact (and conductance) also diminishes as soils desaturate (Schröder et al., 2008). Roots shrink and fewer water bridges connect the root to soil grains (Nobel & Cui, 1992a,b), and root–soil contact may be more difficult to maintain in coarser textures soils because larger air-filled pore spaces develop (Li et al., 2005). Finally, axial root conductance decreases as roots cavitate in dry soil (i.e. air bubbles – embolisms – form within xylem, blocking water flow) and, as noted earlier, coarser soils hold less water than finer soils.

3. Soil moisture status

Soil moisture status in an ecosystem can influence the magnitude of HR. Most fundamentally, a soil water potential gradient must develop for HR to occur. Multiple field studies suggest that upward HR does not begin until shallow soil water potential falls approximately into the range −0.4 to −0.8 MPa (e.g. Ishikawa & Bledsoe, 2000; Domec et al., 2004; Meinzer et al., 2004). The reason for this initiation threshold is unclear, but at the old-growth Ponderosa pine stand, the two Douglas fir stands, and the Brazil savanna sites (Fig. 1), the initiation potential matched the predawn leaf water potential, indicating that the soil needed to dry out enough to effectively compete for water with the above-ground portion of the tree (Meinzer et al., 2004). However, soils cannot be so dry that they limit HR. Regional earth system modeling incorporating HR (Wang, 2011) illustrates the pattern; in South America, HR had the largest impact on transpiration rates at medium wetness levels and a negligible impact at the moisture extremes. Using modeling, Siqueira et al. (2008) further demonstrated that the mid-range soil moisture level that maximizes HR is lower in sandier and higher in loamier soils.

4. Transpiration demand

During rainy periods, transpiration demands on plants can be low, resulting in the atmosphere competing less effectively with dry soil for water within the plant, enabling even daytime HR (the opposite situation – high night-time transpiration reducing HR – is discussed in the next section). Empirical studies of HR that use sap flow sensors often demonstrate that roots can maintain flow away from the trunk toward soil (Smith et al., 1999; Scott et al., 2008) over multiple days, day and night, although the behavior is not reliable across all roots or time periods (Burgess et al., 1998; Smith et al., 1999; Oliveira et al., 2005). Furthermore, low transpiration demand during rainstorms can promote simultaneous upward and downward HR (Williams et al., 1993), from deep moist and shallow rain-soaked soil layers to mid-depth dry layers. Leffler et al. (2005) presumed that this type of convergent flow occurred in senesced cheat grass at their site in Utah after a rainstorm. Such subtleties of transpiration demand are not captured by models using the ‘soil layer connection’ formulation, with an ‘on–off’ (Dtrans) switch for HR, usually based on time of day.

VI. Plant characteristics affecting the magnitude of HR

Below, we briefly outline several candidate physiological and structural traits that have been linked to control of HR.

1. Rooting patterns

Most importantly, for HR to occur, plant root systems must bridge a soil water potential gradient large enough to drive flow, and the root biomass embedded in moister and drier soil volumes must be capable of absorbing or releasing water (e.g. not suberized, Rewald et al., 2011). Several papers have examined co-occurring species with contrasting root architectures and their capacity to conduct HR. In the Brazilian cerrado, Scholz et al. (2008b) compared nine co-occurring tree species for evidence of HR, and found that none of the three evergreen species facilitated HR because their root systems dominantly grew into, and absorbed water from, deep soil layers. By contrast, three deciduous and three brevi-deciduous species with both shallow lateral and deep roots tended to support HR, though not all individuals had root systems dimorphic enough to facilitate HR. In an earlier study, Hultine et al. (2003) used sap flow to assess whether three co-occurring Chihuahuan desert arroyo species all conducted HR, and found that one –Celtis reticulata, with deep sinker roots and large above-ground capacitance (water storage capacity) – did not. The other two –Fraxinus velutina and Juglans major– had dimorphic roots systems and could conduct HR. Siqueira et al. (2008) used modeling to demonstrate that upwards HR is enhanced with an increase in root distribution asymmetry. Given the large number of plants, of various functional types (i.e. grasses, trees, shrubs, forbs) that are known to conduct HR, clearly a very wide variety of root system architectures can span soil volumes of sufficiently different water potential for water flow to occur. However, the studies presented earlier show that even the largest, co-occuring plants in ecosystems do not necessarily all carry out HR when soil water potential gradients exist, and modeling efforts that assume all plants conduct HR are providing upper estimates.

2. Root survival, cavitation, and conductance

Not only do roots have to bridge water potential gradients, but HR also requires that the root system be capable of moving water into dry soils. A study of HR in three oak species, with natural distributions spanning xeric to subxeric Carolina sandhill habitats, demonstrated that Quercus margaretta (rare in the xeric habitat) did not redistribute any water (Espeleta et al., 2004, 2009). The authors hypothesize that fine root death in surface soil may have led to reduced root–soil contact and thus loss of conductivity to support HR. By contrast, the two species found in both subxeric and xeric sandhill habitats (Q. laevis and Q. incana) had lower rates of root mortality in dry surface soil and redistributed water.

The magnitude of HR can also be strongly influenced by decreases in axial root conductance resulting from xylem embolisms accumulated during dry seasons. Warren et al. (2007) found that observed seasonal trends of HR in Pacific Northwest coniferous forests could be matched by simply considering the soil water potential gradient driving flow and the percentage loss in root conductance as the dry season progressed and embolisms accumulated. In the Brazilian savanna, reverse sap flow decreased linearly with root conductance (Scholz et al., 2008b). And, modeling HR in the Sierra Nevada, Amenu & Kumar (2008) predicted twice as much redistribution during the wet season than during the dry season, attributing the difference to root conductance, which, in the model, scaled with soil moisture. Age of individuals within a species can also have ramifications for vulnerability to xylem embolism that can affect the magnitude of HR. Domec et al. (2004) determined that roots on young Douglas fir and Ponderosa pine stands in the Pacific Northwest were more embolized at the end of dry summer than roots on older trees. Old-growth stands of both species maintained moister surface soils during the dry season through more effective HR, and older Ponderosa pine experienced a more rapid rebound in root function than young Ponderosa pine upon the onset of rain.

3. Sensitivity of modeled HR to root and root–soil conductances

Modeling facilitates analysis of the sensitivity of HR to species characteristics that are difficult to measure. Three such studies considered the impact of root, or soil–root, conductance on the magnitude of HR. Mendel et al. (2002) used the ‘big root’ formulation to test the sensitivity of HR to the radial root conductivity (#23-1 vs 23-2, Table 1, Fig. 1), and found that over the range tested, with each order of magnitude increase in radial conductivity of rootlets, HR increased by a factor of c. 1.4. Siqueira et al. (2008), using the ‘macro-meso scale’ formulation, also found that an order of magnitude increase in radial conductance resulted in an approx. 1.4-fold increase in HR at mid-range water contents (#24-1 vs 24-2). Wang (2011), working with the ‘soil layer connection’ model formulation, explored the sensitivity of HR to the maximum soil–root conductance term (CRT); with a doubling of CRT, HR increased by a factor of c. 1.5–2 over the range of values explored (#29-2 vs 29-3). CRT will change depending on the ecosystem being studied, but it is often an unknown parameter, and an order of magnitude uncertainty in root or root–soil conductance is not uncommon (Mendel et al., 2002; Siqueira et al., 2008).

4. Night-time transpiration

As noted previously, competition between the atmospheric sink (via open stomata) and dry soil should decrease the magnitude of HR. The on–off (Dtrans) switch in ‘soil layer connection’ HR models thus usually restricts HR to the night time, assuming stomata are closed. However, recent reviews of night-time stomatal behavior and transpiration (Caird et al., 2007; Dawson et al., 2007) in C3 and C4 plants of multiple functional types demonstrate that night-time transpiration is a widespread phenomenon. Rates from woody plants surveyed by Dawson et al. (2007) ranged from c. 1 to 25% of daytime rates, and increased soil water availability and night-time vapor pressure deficit (VPD) increased transpiration. However, a species’ capacity to transpire at night does not necessarily negate its ability to conduct substantial HR. For example, in Dawson et al.’s (2007) survey, sugar maple was one of the two species for which night-time transpiration could reach 25% of daytime rates, but sugar maple is also the species with the largest empirically detected rate of upward HR (Fig. 1).

A number of studies have explored the impact that night-time transpiration has on the magnitude of HR. Caldwell & Richards (1989) and Bauerle et al. (2008) illuminated sagebrush (Artemisia tridentata) and grapevine (Vitis vinifera), respectively, during the night, forcing the plants to transpire, and successfully reduced HR. Other researchers have bagged plants during the night to raise humidity and thus diminish night-time transpiration. In a glasshouse study by Howard et al. (2009), HR increased with bagging for sagebrush (A. tridentata) and sunflower (Helianthus anomalus), but did not change for oak (Quercus laevis). The authors note that the young oaks had substantially lower shoot biomass (leaf area) to root biomass ratio than the other species, and they surmise that night-time transpiration was thus a negligible water sink competing with HR through large oak root mass, even though transpiration rates expressed per unit leaf area were higher in oak than in sagebrush. Working in a Chihuahuan desert arroyo, Hultine et al. (2003) found that reverse sap flow in lateral roots (indicative of HR) decreased as VPD (driving night-time transpiration) increased for ash (Fraxinus velutina). However, co-occurring walnut (Juglans major) did not host sufficient night-time transpiration to drive a similar relationship. Among the Brazilian cerrado trees, Scholz et al. (2008b) covered one deciduous species, Kielmeyera coriacea, at night to reduce transpiration, resulting in increased reverse sap flow. Prieto et al. (2010a) found that the magnitude of HR supported by the Mediterranean shrubby legume Ratama sphaerocarpa was large only when maximum night-time stem sap flow was small.

Taken together, these experiments support the importance of night-time transpiration for restricting HR, but also indicate that a high measured night-time transpiration rate per unit leaf does not necessarily signal that HR will be limited by that transpiration. The strengths of the whole shoot and root–soil system sinks for water determine the outcome of competition for water.

5. Plant capacitance

The whole shoot sink strength is also affected by shoot capacitance (capacity for storage of water), which may in some cases be a significant sink for water flow through roots at night. Scholz et al. (2008a) noted that among Neotropical trees, the percentage total daily water use contributed by water storage inside the plants can be large, for example, c. 16–33%. Hultine et al. (2003) reported that the tree C. reticulata did not carry out HR, although co-occuring other species did. Celtis had deep sinker roots, but also sap flow sensors on roots indicated that flow continued toward the shoot at night, regardless of night-time VPD, that is, independent of night-time transpiration. Large above-ground capacitance depleted during the day by transpiration may have served as a strong sink for water originating from deep, moist soil layers throughout the night.

6. Senescence and dormancy

Since senesced and dormant plants show limited transpiration, their root systems can continually redistribute water, supporting large rates of HR. For example, Hultine et al. (2004) demonstrated that the taproot of dormant velvet mesquite hosted continual reverse sap flow for weeks (see also Scott et al., 2008), and reverse sap flux was enhanced in the deciduous Brazilian cerrado tree K. coriacea when leafless (Scholz et al., 2008b). Senesced cheatgrass in Utah also supported substantial HR (Leffler et al., 2005). This continued ability of senesced vegetation to redistribute is not captured by regional-scale modeling where maximum conductance of the soil–root system supporting HR is determined by dynamic LAI (e.g. Lee et al., 2005).

VII. Conclusions

Empirical and modeling estimates of the average amount of water moved by hydraulic redistribution span nearly two orders of magnitude (Fig. 1), from c. 0.04 to 3.2 mm H2O d−1. Upward and downward HR clearly can be ecologically and hydrologically significant in many ecosystems, enhancing transpiration and photosynthetic carbon gain, and conducting precipitation to deep soil layers. The literature also shows, however, that upward HR is likely of rather small immediate hydrologic significance in some ecosystems where its contribution to transpiration is small (Fig. 2). In such ecosystems, small amounts of water moved by HR can still be ecologically significant for plant survival, for example, maintaining fine roots, mycorrhizal hyphae, and root–soil contact, or preventing embolism, and thus at a population and community scale may influence hydrology over the long term. Dynamic vegetation and plant community models offer potential for detecting such influences (e.g. Levis et al., 2004; Moorcroft et al., 2001) but are only rarely coupled with HR models (e.g. Wang et al., 2011).

Not surprisingly, both the empirical and modeling-based HR literatures demonstrate that root characteristics – their tendency to embolize, their ability to survive in dry soil, their radial and axial conductance, their architectures, and their capacity for dynamic adjustment of water uptake (through growth or physiology) from multiple soil layers – all exert strong influence over the magnitude of HR. ‘Big root’, ‘macro-meso’, and ‘dynamic root’ models, all more complex than the soil layer connection formulations, and including some aspect of flow through or to/from roots, would be best used, tested, and compared in field sites where extensive below-ground empirical information, including root information, is available to support parameterization. For example, taking advantage of any information about root architecture, root growth timing, root shrinkage away from soil contacts, branching orders and distributions of absorptive vs simply conductive (e.g. suberized) root biomass, and seasonality of root embolism could greatly constrain the models and potentially enhance their representations of HR. Of course, more general below-ground information is also key; psychrometer data should be calibrated to local soil. (Water content probes may be more forgiving if soil-specific calibration is impossible.) Also, the possibility that water vapor transport driven by soil temperature fluctuations can confound estimates of upward HR (Warren et al., 2011) should be further explored.

The various HR models reviewed here all highlight that it is the integration of plant architecture and function with the characteristics of the full soil column in which roots are embedded that determines the magnitude of HR. Plant root and soil characteristics are among the least well-known features of many ecosystems and thus make parameterization and validation of any HR model, and any larger land–atmosphere model (e.g. Mahfouf et al., 1996; Feddes et al., 2001), very challenging. But progress can still be made. For example, recent surveys of night-time transpiration, its dependency on soil moisture content, and its ability to restrict HR clearly point toward the necessity of modifying the ‘on–off’ switch commonly used in soil layer connection models (e.g. Dtrans in Ryel et al., 2002) to a more flexible mathematical form (as suggested, originally, by Ryel and colleagues). The form should be capable of capturing effects of night-time transpiration and daytime stomatal closure and/or diminution of the driving gradient for water loss to the atmosphere (e.g. during rainfall when air is saturated and leaf surfaces are wet). Incorporation of plant capacitance, too, will be important, for systems where empirical work demonstrates it is large and likely can compete for water with HR.

Systematic examinations along environmental gradients of key variables suspected of controlling the magnitude of HR are rare, probably because it is challenging to find a controlled enough gradient where only one, or a few, characteristic(s) are changing. The literature suggests, for example, that sandy soils tend to support less HR, but few studies have selected field sites along a soil texture gradient to characterize the dependency. Still, careful comparisons across diverse ecosystems have revealed commonalities in characteristics of HR that otherwise might have been missed. Meinzer et al. (2004), for example, comparing six Pacific Northwest and Brazilian savanna sites, discovered surprisingly similar patterns in soil water utilization and effects of soil water potential on development of HR, although plant species, root architectures, and soils were different. Biophysical constraints may have structured the patterns.

Finally, several recommendations for modeling at larger scales, and cautionary notes for scaling empirical measures, emerge directly from the reviewed literature. When HR is incorporated into larger-scale process models, for example to improve the match between modeled transpiration and fluxes measured at an eddy flux tower, the model should be constrained, and/or evaluated, with multiple data types if possible, matching output to soil moisture profiles as well as gas fluxes above ground. Also, landscape, regional, or global models cannot reliably assume all species in all ecosystems carry out HR. Some plant size-specific, species-specific, and functional type-specific differences in capacity to conduct HR have been characterized within individual ecosystems, and will help constrain the magnitude of HR at least locally. Above-ground characteristics of plant communities may also provide some insight. Conductivity of the whole below-ground (soil–root) system must, at a minimum, be sufficient to supply water to the maximum leaf area commonly observed in the ecosystem under common environmental conditions; night-time transpiration and high above-ground capacitance can effectively compete with dry soil and roots for water being lifted from moist layers, potentially constraining HR; and the seasonality of deciduousness in many ecosystems, captured at large scales by dynamic LAI, should lead to seasonality in the strength of these competing above-ground sinks. Such information may help to constrain modeled HR in ecosystems where investigators suspect HR strongly influences fluxes of water, energy, and carbon, improving the match between measured and modeled transpiration and photosynthetic rates, surface temperatures, runoff, and soil moisture throughout the soil column.


This work was supported by a NOAA Climate and Global Change Postdoctoral Fellowship to R.B.N., administered by the University Corporation for Atmospheric Research, and by NSF Ecosystems grant #0415938 to Z.G.C. Thanks to Suzanne Thomas for DataThief work, and to N. M. Holbrook, Harvard University, for hosting R.B.N. during manuscript preparation.