A depth-controlled tracer technique measures vertical, horizontal and temporal patterns of water use by trees and grasses in a subtropical savanna


Author for correspondence:
Andrew Kulmatiski
Tel: +1 907 786 1676
Email: afak@uaa.alaska.edu


  • As described in the two-layer hypothesis, woody plants are often assumed to use deep soils to avoid competition with grasses. Yet the direct measurements of root activity needed to test this hypothesis are rare.
  • Here, we injected deuterated water into four soil depths, at four times of year, to measure the vertical and horizontal location of water uptake by trees and grasses in a mesic savanna in Kruger National Park, South Africa.
  • Trees absorbed 24, 59, 14 and 4% of tracer from the 5, 20, 50, and 120 cm depths, respectively, while grasses absorbed 61, 29, 9 and 0.3% of tracer from the same depths. Only 44% of root mass was in the top 20 cm. Trees absorbed tracer under and beyond their crowns, while 98% of tracer absorbed by grasses came from directly under the stem.
  • Trees and grasses partitioned soil resources (20 vs 5 cm), but this partitioning did not reflect, as suggested by the two-layer hypothesis, the ability of trees to access deep soil water that was unavailable to grasses. Because root mass was a poor indicator of root activity, our results highlight the importance of precise root activity measurements.


As early as 1883, ecologists working in temperate ecosystems noted that woody plants can create deeper roots than grasses (Sperry, 1935; Albertson, 1937). Walter (1971) used this and similar observations to explain tree and grass coexistence in savanna ecosystems. He developed what is referred to as the two-layer hypothesis, which suggests that deep roots allow woody plants to escape competition from dense, shallow grass root mats (Walter, 1971; Scholes & Archer, 1997). This hypothesis has been integrated into models of species coexistence, hydrological cycles, biogeochemical cycles, and climate change (Foley et al., 1996; Daly et al., 2000; Seyfried et al., 2005; Newman et al., 2006; Weng & Luo, 2008). Inasmuch, the two-layer hypothesis has provided an important explanation of savanna structure and function for almost 40 yr (Scholes & Archer, 1997; Sankaran et al., 2005).

Several studies, however, have suggested that niche partitioning in savannas is either more complex than previously thought (Chimner & Cooper, 2004; Reynolds et al., 2004; Goldstein et al., 2008) or not important at all (Scheiter & Higgins, 2007). Conflicting conclusions probably reflect at least two factors. First, rooting strategies are likely to differ among savannas. Studies in temperate savannas have typically demonstrated deep rooting by trees (Weltzin & McPherson, 1997; Baldocchi et al., 2004; Asbjornsen et al., 2008), while studies in tropical and subtropical savannas have shown both deep (Jackson et al., 1999; Schenk & Jackson, 2002; Quesada et al., 2004; Goldstein et al., 2008) and shallow rooting by trees (Belsky, 1994; Leroux et al., 1995; Mordelet et al., 1997; Hipondoka et al., 2003; Rodriguez et al., 2007). Second, difficulties of belowground research have limited understanding of root uptake of soil resources. For example, measures of root mass are common, but are likely to be biased by the presence of large and inactive roots (Chen, 2004; Peek et al., 2005).

Measuring root activity remains one of the most challenging problems in plant ecology (Casper et al., 2003; Schenk, 2008b). Tracer techniques provide an important means for addressing this challenge (Schwinning et al., 2002; Ogle et al., 2004). Techniques that rely on natural variation in stable isotope concentrations typically only allow broad distinctions in water use (e.g. 0–50 vs 50–100 cm; Leroux et al., 1995; Jackson et al., 1999; Ludwig et al., 2004; Kulmatiski et al., 2006). Techniques that introduce tracers into the soil are more promising because they can constrain descriptions of water use to specific locations (i.e. vertical, horizontal, and temporal; Gottlein et al., 2005). While elements such as lithium (Li), strontium (Sr), and cesium (Cs) have been used as tracers (Mamolos et al., 1995; Bonanomi et al., 2005; Rodriguez et al., 2007), studies using deuterated water (D2O) show the most promise because plant roots typically do not distinguish D2O from H2O and, unlike some tracer elements, D2O is unlikely to accumulate variably in different plant parts (Dawson & Ehleringer, 1993; Takahashi, 1998; Webb & Longstaffe, 2003; but see Ellsworth & Williams, 2007).

D2O has been used in irrigation treatments, but the wide wetting front associated with irrigation only allows broad distinctions in water use (Schulze et al., 1998; Moreira et al., 2000; Schwinning et al., 2002; Sternberg et al., 2005). D2O injections offer a better approach for measuring root activity, but few studies have tested this technique. Those that have used the technique injected into only one or two shallow soil depths (Bishop & Dambrine, 1995; Plamboeck et al., 1999; Penuelas & Filella, 2003; Sternberg et al., 2005; Hawkins et al., 2009).

It was the aim of this study to test the hypothesis that tree and grass coexistence in savannas can, in part, be explained by vertical partitioning of the soil water resource. To test this hypothesis, we measured the timing, location, and extent of water use by trees and grasses in a mesic, subtropical savanna in Kruger National Park, South Africa. To do this, D2O was injected into the soil at four depths (5, 20, 50, and 120 cm) and during four sampling periods (Oct, Nov, Feb, and Apr 2007–08). Plant tissues were sampled following the tracer addition to determine the extent to which plants absorbed water from each soil depth. Plants were also sampled at several distances (0, 0–2, 2–3, and 3–5 m) from the tracer addition area to determine the horizontal foraging area of trees and grasses. Soil were sampled to determine how the tracer moved in the soil. Leaf area, leaf-level conductance, tree diameter over time, and soil water potential (ψ) were measured to assess whether tracer uptake was consistent with water availability.

Materials and Methods

Study site

Research was conducted 3 km south of Pretoriuskop, Kruger National Park, South Africa (−25.2076° N, 31.2832° E; elevation 655 m). The site is described as a broadleaf or woodland, mesic savanna (Sankaran et al., 2005; Archibald & Scholes, 2007). Dominant trees and shrubs are deciduous. The site receives mean annual precipitation of 746 mm, occurring primarily during the summer (Fig. 1). A drought occurred during the study; monthly precipitation was 11 and 51% of mean values in February and March 2008, respectively (Anonymous, 2010).

Figure 1.

 Mean (solid line) and observed (broken line) precipitation from July 2007 to Jan 2009, Pretoriuskop, Kruger National Park, South Africa. Open circles indicate dates for which missing data were inferred from nearby meteorological stations.

The study site was dominated by C4 grasses: Hyparrhenia filipendula (Hochst.) Stapf, Hyparrhenia dissoluta (Steud.) Hutch., Cenchrus ciliaris (L.) Link, Panicum maximum (Jacq.), and Setaria sphacelata (Schumach.) Stapf & C.E. Hubb); trees: Terminalia sericea (Burch. ex DC.) and Sclerocarya birrea (A.Rich.) Hochst.; and the erect shrub Dichrostachys cineria subsp. africana (Brenan & Brummitt). Leaf-out by trees is rather consistent each year and appears to be triggered by day length, while grass green-up is more variable and appears to be triggered by both day length and soil moisture (Archibald & Scholes, 2007). A dominant tree at the site, S. birrea, has a short, thick taproot that can grow to 2 m depth and has an average green-up date of 15 October ± 3 d (Archibold & Scholes, 2007). Terminalia sericea holds dead leaves through most of the dormant season and leaf-out typically occurs 1–3 wk after that of S. birrea, often before the first rains (Childes, 1988). Xylem vulnerability curves indicate that T. sericea is less vulnerable to cavitation than S. birrea, suggesting higher drought tolerance (R. J. T. Verweij, preliminary data). Mid-season ground cover was 57 ± 12% grasses, 19 ± 10% trees, 8 ± 5% shrubs, and 6 ± 2% forbs (mean ± standard deviation (SD)). Tree diameter at breast height (DBH) was 4 ± 4 cm (range 1–31 cm). Tree canopy radii were 3.9 ± 1.3 m (range 1.5–6.9 m). Soils are sandy loams (typic ustalfs) with little coarse fragment and a dense layer from 80 to 150+ cm.

D2O tracer experiment: the timing, location, and relative extent of water use

Eight circular (5 m radius) plots separated by 30 m were placed along each of four 240-m-long transects, located 60 m apart, for a total of 32 plots. Even-numbered plots were randomly assigned to a depth (5, 20, 50, or 120 cm) and sampling period (Oct, Nov, Feb, and Apr 2007–08) treatment (e.g. 20 cm in Nov). Each treated plot, therefore, was 60 m from any other treated plot. Odd-numbered plots were used as controls.

In a 1-m2 circle at the center of each treated plot, 100 pilot holes were drilled by hand using a 10-mm-diameter steel rod in a 10 cm × 10 cm grid pattern to the assigned depth. Two ml of 70% D2O followed by 2 ml of tap water was injected into each pilot hole using custom-made needles (16 gauge, regular width hypodermic tubing; Vita Needle, Needham, MA, USA) and syringes that could inject tracer directly to the bottom of the pilot hole. The tap water injected after the tracer was intended to expel all tracer from the syringe and needle, and prevent tracer contamination as the needle was removed from the hole. Holes were filled with soil when needles were removed to prevent evaporation of the tracer up the pilot hole. Assuming the resulting 4 ml of D2O + H2O injected into each point rapidly diffused into a prolate spheroid with a 2 cm horizontal radius and 3 cm vertical radius, the injected water produced a labeled soil volume of 5024 cm3 or 8.4% of the target soil volume. The 400 ml of added water represented only 0.05% of annual precipitation but was expected to increase volumetric soil water content by 7.9% in the 5024 cm3 volume. So, while the injection probably created a pulse of plant-available water around the injection point, it was not expected to increase plant growth.

One day following tracer injection, the first plant samples were removed from the 1-m2 tracer addition area and the 0–2-m area outside the tracer addition area. Two days following tracer injection, plant samples were removed from the 2–3-m area outside the tracer addition area. Three days following tracer injection, plant samples were removed from the 3–5-m area outside the tracer addition area. Samples removed from the 120-cm plots were removed 1 d later than samples from the 5-, 20- and 50-cm plots to allow time for xylem flow to move the tracer to the sampled plant materials. Similarly, tree samples removed from > 1 m above the ground were sampled 1 d later than other samples from the same distance from the tracer addition area. Xylem flow rates of 1–5 m d−1 were used to determine when samples should be taken (Fravolini et al., 2005; Meinzer et al., 2006), and this assumption was tested by repeat sampling of trees (0–2 m from center) and grasses (tracer addition area) in one plot (first 5-cm tracer injection) over 5 d.

We attempted to sample all individual trees and all grass species in plots. Samples of common species were composited to produce three to five replicate samples for each species in each distance from the pulse. Less common species were composited by functional type (i.e. grass or tree). As much plant material as could fit in the bottom 10 cm of the prepared sample vial was taken. All sampling tools (hands, trowels, clippers, and steel rod for filling samples into vials) were moved 20 m outside the plot and triple rinsed with tapwater before taking the next sample. For all plant samples, nontranspiring tissues were removed so that samples represented the mean water uptake by plant roots (Dawson & Ehleringer, 1993). Grasses were sampled from the root crown. Tree twigs or stems were sampled from below the height of the first leaves. Control samples were collected throughout the sampling period following the methods used for treated plots. At least three replicate samples of each species sampled in treated plots were also taken from control plots.

One week following tracer injection in November 2007, soil cores (5 cm diameter) were taken from treated and control plots to confirm the location of the added tracer. These soil cores were split into eight depths to identify the location of the pulse. This sampling design resulted in 505 plant and 67 soil samples. Plant and soil samples were immediately placed into borosilicate (19 mm outer diameter) tubes, sealed with parafilm, placed on ice, transported to a freezer within 6 h, and extracted within 2 wk.

Water samples were extracted from plant tissues and soils using a batch cryogenic distillation procedure (Vendramini & Sternberg, 2007) in Skukuza, Kruger National Park. Briefly, tubes with samples were placed in liquid nitrogen, secured to a vacuum extraction line with an Ultrator® fitting (Swagelok, Solon, OH, USA), evacuated to 10 milliTorrs pressure, flame sealed, and inserted sample-up in a custom-milled 12-cm-deep aluminum block. A matching aluminum block was placed above the samples and heated to 90°C. After 1 h of heating, liquid nitrogen was added to a foam tub that surrounded the bottom aluminum block. Samples were heated and frozen until ice placed against the sample tube created no condensation in the sample vial (typically 6 h). Sample vials were cracked open, samples were placed in labeled 2-ml vials and the vials were shipped to the University of Cape Town Stable Light Isotope Laboratory for the determination of deuterium to hydrogen ratios. Samples typically produced 1–4 ml of sample water, limiting concern regarding fractionation as a result of evaporation during sample transfer from sample vials to 2-ml vials. All isotope values are expressed in delta notation (δ) as the D : H ratio relative to a standard (Vienna standard mean ocean water). For clarification, inline image expressed as ‘parts per mil’ or ‘‰’ (R, the ratio of heavy to light isotope; sa, sample; std, standard; Rstd ≈ 1/6412). Analytical precision (2σ) was 0.9 ‰ for δD.

Tracer was assumed to be present in samples with δD values that were at least two SDs above mean concentrations in paired control samples. Because raw tracer concentrations are biased by differences in rooting area among species (i.e. tracer was added to a larger proportion of the rooting zone of a plant with a small rooting area relative to a plant with a large rooting area), tracer uptake was normalized by depth or distance to allow a direct comparison of water uptake between trees and grasses (Bishop & Dambrine, 1995; McKane et al., 2002; Schwinning et al., 2002) as follows: proportional uptake for a plant functional type by depth or distance = inline image (Sn, the mean δD value of samples from a treatment level n (e.g. 5 cm depth); C, the mean δD value of control samples for that functional type (e.g. trees)).

Differences in tracer uptake were tested using analysis of variance (ANOVA) computed using the MIXED procedure in sas/stat for Windows, Release 9.1.3 (SAS Institute Inc., Cary, NC, USA). To compare the proportion of tracer uptake between trees and grasses for each depth or distance from the tracer addition area, the fixed effect was plant type (trees and grasses). Month was used as a random effect. To compare tracer uptake for trees and for grasses among depths for each time period (month), the fixed effect was depth and replicate samples from each plot were used as random effects. Tukey post hoc pairwise comparisons were used to determine mean differences.

Soil water availability

To determine the timing, location, and extent of water availability and whether tracer uptake was only occurring in soils where soil water was plant available, soil water potential was measured. Soil moisture sensors were placed in a soil pit 500 m east of the study area. Soil water potentials were determined using an array of Campbell Scientific 229 heat dissipation sensors (Logan, UT, USA). Before installation, each sensor was calibrated using an endpoint test and by taking measurements in soils from one of three appropriate depth strata (0–30, 30–60, or 60–90 cm) that were equilibrated to each of five known water potentials for 16 h (Flint et al., 2002). Water potentials of the equilibrated soils were determined using the chilled-mirror technique (WP4T water potential meter; Decagon Devices, Pullman, WA, USA). Sensors were placed in pilot holes established in the undisturbed walls of the soil pits. A loop of sensor cable was placed below the sensor to prevent the creation of a preferential flow path to the sensor. Calibrated sensors were placed at 5, 10, 20, 30, 50, 75, 120 and 150 cm in the soil profile. Following sensor installation, the soil pit was filled and compacted and grass root mats were replaced. Measurements were recorded every 3 h on a multiplexed Campbell Scientific CR1000 datalogger.

Root mass, leaf-level conductance, leaf area and tree diameter

Root mass from 0 to 135 cm was determined by sieving, drying, and weighing all roots found in 16 soil cores (4 cm diameter) that were extracted from experimental plots throughout the growing season. Leaf-level conductance was measured using Decagon Devices porometers. For each plant species, at least three measurements were made on at least 10 individuals on at least 2 d for each of six sampling periods (Oct, Nov, Dec, Jan, Feb, and Apr) resulting in 1370 measurements. Measurements were paired so that all species were sampled within 15-min intervals before sampling was repeated. Adaxial conductance was not measurable on tree leaves, but was measured on grasses. As an indicator of tree and grass responses to water availability during the growing season, leaf-level conductance rates (mmol H2O m−2 leaf area) were multiplied by leaf area to determine tree and grass conductance on a m2 basis. Leaf area was determined by clipping all green vegetation in each of four randomly selected 1-m2 quadrats in each of four randomly selected plots located in control areas on each sampling date. Clipped vegetation was frozen within 6 h and scanned on a CID Inc. portable leaf area scanner (Camas, WA, USA) within 1 wk of sampling. Measured grass leaf area was doubled to account for adaxial leaf conductance. ANOVAs were used to test for differences in leaf area, leaf-level conductance, and aerial conductance between trees and grasses. In these analyses, plant type was used as a fixed effect and replicates were used as random effects.

To determine if trees used water stored in their tissues, dendrometers were placed on common trees to measure tree diameter over time. Changes in tree diameter were determined with biweekly, manual readings from dendrometer bands (Agricultural Electronics Corporation, Tucson, AZ, USA) at 0.1-mm increments on eight marula (S. birrea) and eight terminalia (T. sericea) individuals located 7 km from the study site.


Tracer detection

Of the 505 plant samples, 300 grass and 130 tree samples were extracted from experimental plots. Of these, 119 (40%) grass and 23 (18%) tree samples demonstrated isotope ratios that were 2 SDs or more above controls (i.e. received tracer; Table 1). Of the 75 control samples, only one plant sample demonstrated an isotope ratio that was 2 SDs above the mean (i.e. there was little to no contamination).

Table 1.   Tracer uptake by depth, distance, and plant type
DepthDistance from tracer addition (m)
  1. Each pair of values shows the number of samples analyzed and the number of samples with δD values that were 2 SD or more greater than mean δD values for control samples.

5 cm22/220/015/222/95/181/71/191/13
20 cm23/231/311/222/75/172/91/176/14
50 cm17/180/06/201/94/181/102/173/15
120 cm 3/170/03/192/120/161/121/150/10

Soil and plant samples were taken to confirm the location of the tracer and to determine how quickly tracer moved through soil and plants, respectively. Soil samples taken 1 wk following tracer injection showed clear differences in tracer concentration with depth among the treatments, although redistribution was also apparent (Fig. 2). Repeated vegetation sampling indicated that grasses realized a peak in tracer concentrations 1–2 d after tracer injection and trees realized a peak in tracer concentrations 1–3 d after tracer injection (Fig. 3).

Figure 2.

 Tracer concentration by depth (control, 5, 20, 50, or 120 cm) at 1 wk following tracer injection.

Figure 3.

 Tracer concentration in tree (a) and grass (b) samples taken by day following tracer injection. Grasses were sampled in the tracer addition area and trees were sampled in the tracer addition area and within 2 m of the tracer addition area. Error bars signify +/− SE.

Plant uptake

Trees absorbed 24, 59, 14 and 4% of tracer from the 5, 20, 50 and 120 cm depths, respectively, while grasses absorbed 61, 29, 9 and 0.3% of tracer from the same depths (Fig. 4a). Neither trees nor grasses, therefore, absorbed > 5% of their tracer from 120 cm depths. Grasses absorbed a greater proportion of water from the 5 cm depth than trees but there were no significant differences in the proportion of tracer uptake between trees and grasses at the other depths.

Figure 4.

 Proportional tracer uptake (± SE) by trees (open circles) and grasses (closed circles) by depth (a) and distance (b). Standard errors reflect variation between sampling dates. *, significant differences between trees and grasses within a depth or distance using P < 0.05; ns, no significant difference.

In the horizontal plane, there was no difference in tree uptake among distances from the pulse (Fig. 4b). By contrast, grasses foraged close to the stem. Almost all (98%) grass uptake occurred in the tracer addition area (Fig. 4b). As a result, trees absorbed a smaller proportion of tracer close to the stem than grasses, but absorbed a larger proportion of tracer 2, 3 and 5 m from the stem than grasses (Fig. 4b).

Trees absorbed more tracer in November and April than in October or February (F3,12 = 6.26, < 0.01; Fig. 5). Trees absorbed more tracer from 20 cm than from 5, 50 or 120 cm in November (Fig. 5b; F3,16 = 11.07, < 0.01), and more tracer from 20 cm than from 50 or 120 cm in April (Fig. 5d; F3,33 = 2.90, = 0.05).

Figure 5.

 Tracer concentrations in trees (open circles) and grasses (closed circles) in Oct (a), Nov (b), Feb (c), and Apr (d). Note that the top and bottom horizontal axes indicate grass and tree deuterium concentrations in, respectively delta notation. Standard errors reflect variation from replicate samples within a plot. Significant differences among depths are described in the text.

Grasses absorbed more tracer in November than April (F3,11 = 3.55, = 0.05; Fig. 5). In October, grasses absorbed more tracer from 20 cm than from 50 cm (Fig. 5a; F3,26 = 4.45, = 0.01). In November, grasses absorbed more tracer from 5 and 20 cm than from 50 and 120 cm, and more tracer from 50 cm than from 120 cm (Fig. 5b; F3,75 = 4.89, < 0.01). In February, grasses absorbed more tracer from 5, 20 and 50 cm than from 120 cm (Fig. 5c; F3,72 = 2.32, < 0.01). In April, grasses absorbed more tracer from 5 cm than from 50 or 120 cm (Fig. 5d; F3,67 = 4.48, < 0.01).

Soil water availability

Soil water potentials were largely consistent with plant water uptake (Fig. 6). Shallow (i.e. 5, 10, 20 and 30 cm) soil water was available to plants (e.g. Ψ > −3 MPa) on the October, November and April sampling dates, but not the February date (6 Feb). Shallow soil water became unavailable the day before the February pulsing began and remained unavailable until 13 February. Soil water at 50 cm remained available throughout the February pulse event. Soil water at 120 cm was available between 29 November 2007 and 9 March 2008. Soil water at 120 cm remained unavailable for the remainder of the growing season.

Figure 6.

 Mean weekly soil water potentials from 5 cm (a), 20 cm (b), 50 cm (c), 100 cm (d), and 120 cm (e) depths over the 2007–08 growing season, Pretoriuskop, Kruger National Park, South Africa.

Root mass, leaf area, leaf conductance, and tree diameter

Root mass decreased from the surface to 135 cm, with 39% of root mass in the top 20 cm (Fig. 7). Grass leaf area was 5–21 times greater than tree leaf area across sampling periods (Fig. 8a). Leaf-level conductance, however, was often two times greater for trees than grasses across sampling periods, except in October (Fig. 8b). Both tree and grass leaf-level conductance peaked in January. Leaf-level conductance for both trees and grasses was lower in February than January (F5,581 = 25.21, F5,779 = 12.29, <0.01). On an area (m2) basis, grass conductance was greater than tree conductance for each sampling period (Fig. 8c).

Figure 7.

 Root mass from 16 soil cores extracted from experimental plots throughout the 2007–08 growing season, Pretoriuskop, Kruger National Park, South Africa.

Figure 8.

 Leaf area (a), leaf-level conductance (b), and conductance on an aerial basis (c) for trees (open circles) and grasses (closed circles) through the 2007–08 growing season. All measures for trees and grasses were significantly different, except for leaf-level conductance in Oct 2007.

At the beginning of the 2006–07 growing season, S. birrea realized a decrease in stem diameter and produced leaves before summer rains. Terminalia sericea did not realize a decrease in stem diameter or produce leaves before the summer rains (Fig. 9). However, both S. birrea and T. sericea realized a decrease in stem diameter during the midseason drought in the 2006–07 season (Figs 1, 9). Neither tree species realized a decrease in stem diameter at the beginning of the 2007–08 growing season, which included early rains (Fig. 1). Again, both species realized a decrease in stem diameter during drought in February 2008.

Figure 9.

 Changes in tree diameter of Sclerocarya birrea (solid line) and Terminalia sericea (dotted line) in Pretoriuskop, Kruger National Park, South Africa (= 8 for each species).

Allometric relationships suggest that the 4-mm decrease in stem diameter in S. birrea (10 cm DBH; 8 m height) is associated with a 1.3-kg decrease in wood mass, a 0.80-kg decrease in dry leaf matter and a 4.9-m2 decrease in leaf area (Netshiluvhi & Scholes, 2001). Assuming that this decrease is only water loss and that a decrease in root mass may roughly double the amount of water loss, the observed decrease in diameter is estimated to produce 2.6 kg of water from an S. birrea tree.


Ecologists have long wondered how savanna trees and grasses access soil resources, but the difficulties associated with belowground research often prevent direct measurements of root activity. By injecting tracers at four depths up to 120 cm in a sandy, mesic savanna, we found that trees absorbed 83% and grasses absorbed 90% of their water from the top 20 cm of soil. This large proportion of tracer uptake from shallow depths was greater than the 44% of root mass found in the top 20 cm of soil or even the 61% of root mass found in the top 35 cm. Root mass, therefore, underestimated root activity during the study.

Results provided mixed support for the two-layer hypothesis. Trees and grasses did partition soil resources, but this partitioning did not reflect, as suggested by the two-layer hypothesis, the ability of trees to access deep soil water that was unavailable to grasses. More specifically, trees absorbed most of their soil water from 20 cm depths and grasses absorbed most of their soil water from 5 cm depths. Trees, however, did not absorb a larger proportion of water from 50 or 120 cm depths than grasses: trees did not have exclusive access to deep soil water. Because trees and grasses partitioned soil resources over a narrow vertical range, techniques that measure soil resource use must distinguish differences of a few centimeters (McKane et al., 2002; Ogle et al., 2004). Previous studies did not make such precise measurements and concluded that trees and grasses did not partition shallow soil resources (Leroux et al., 1995; Mordelet et al., 1997; Hipondoka et al., 2003).

Previous studies based on measurements of root mass suggest that the two-layer hypothesis will be most important in sandy, mesic savannas where deep infiltration provides a resource to deeply rooted trees (Schenk & Jackson, 2002). That trees and grasses in our sandy, mesic study site could partition resources within the top 20 cm of soil suggests that access to deep (e.g. > 1 m) soil water is not a prerequisite for resource partitioning among trees and grasses. Our results are consistent with the hypothesis that root activity occurs at the shallowest possible depths (Schenk, 2008a). This hypothesis and our results suggest that vertical niche partitioning may be important in savannas with or without deep soil water infiltration (i.e. any savanna type) and not just in sandy, mesic savannas.

The ability of trees to maintain leaf area longer than grasses has been used to support the two-layer hypothesis, but our measurements suggest two alternative explanations. Leaf production before seasonal rains (Archibald & Scholes, 2007) has been used to suggest that trees use deep soil water (Walter, 1971; Scholes & Archer, 1997). We found that, when there were no early-season rains, S. birrea used stored water to produce early leaves, as evidenced by decreases in stem diameter (as in Chapotin et al., 2006; Verweij, 2009). Leaf retention in the dry season has also been used to suggest that trees use deep soil water (Walter, 1971; Scholes & Archer, 1997). We found that leaves retained by T. sericea demonstrated little to no stomatal conductance, indicating that late-season leaves are not using water. When combined with our observation that trees absorbed little water below 20 cm and that during drought trees used stored water, as evidenced by decreases in stem diameter, our results suggest that trees did not use large amounts of deep soil water.

Trees and grasses showed large differences in horizontal foraging. Trees actively foraged for soil water at least 5 m from the tracer addition area, a distance beyond the width of their crowns (mean radius of 3.9 m; Sternberg et al., 2005). Surprisingly, several grass samples also received tracer up to 5 m from the stem, but the concentration of tracer absorbed from these distances was small. As a result, > 78% of tracer uptake by trees occurred outside the tracer addition area while < 2% of tracer uptake by grasses occurred outside the tracer addition area. In summary, trees showed the potential to integrate resource-rich patches over wide areas while grasses foraged immediately under their stems. These results are consistent with previous measurements of root mass (Schenk & Jackson, 2002) and root activity in savanna systems (Sternberg et al., 2005).

Plant access to soil resources is a fundamental aspect of terrestrial life that remains poorly understood. By quantifying the location of water uptake by different plants, the depth controlled tracer technique allowed a critical step to be taken towards understanding plant access to soil resources. Results showed that plants can partition soil resources over small spatial scales and that the tracer technique can be used to detect these differences. Future research that combines this technique with species-level estimates of transpiration offers the potential to produce quantitative and spatially explicit species-level water budgets. These water budgets are needed to predict how changes in precipitation and temperature are likely to change plant growth and feedback to influence hydrologic cycles (Casper et al., 2003; Li et al., 2007).


The Andrew W. Mellon Foundation funded this research. This research was also funded by Alaska EPSCoR NSF award #EPS-0701898 and the state of Alaska. We thank Shane Heath and Lauren Hierl for managing the field work. Thanks to SanParks and Kruger National Park for access to park lands. The experiments described here complied with the current laws of the Republic of South Africa.