Reverse flow of sap in tree roots and downward siphoning of water by Grevillea robusta



1. Constant-power heat-balance sap flow gauges were used to compare sap flow in vertical and lateral roots of Grevillea robusta trees growing without access to ground water at a semiarid site in Kenya.

2. Reversal of sap flow occurred when root systems crossed gradients in soil water potential. Measurement of changes in the direction of flow was possible because of the symmetrical construction of the sap flow gauges; gradients in temperature across the gauges, and thus computed rates of sap flow, changed sign when reverse flow occurred.

3. Reverse flow in roots descending vertically from the base of the tree occurred, while uptake by lateral roots continued, when the top of the soil profile was wetter than the subsoil. The transfer of water downwards by root systems, from high to low soil water potential, was termed ‘downward siphoning’; this is the reverse of hydraulic lift.

4. Downward siphoning was induced by the first rain at the end of the dry season and by irrigation of the soil surface during a dry period.

5. Downward siphoning may be an important component of the soil water balance where there are large gradients in water potential across root systems, from a wet soil surface downwards. By transferring water beyond the reach of shallow-rooted neighbours, downward siphoning may enhance the competitiveness of deep-rooted perennials.


There is considerable evidence that water taken up by plant roots from moist zones of soil can be transported through the root system and released into drier soil. The process is thought to be driven passively by gradients in water potential between zones of soil joined by interconnected roots (Dawson 1993). Most commonly, this phenomenon has been studied in the context of ‘hydraulic lift’, where water is transferred into dry surface layers of the soil from deeper, wetter subsoil (Richards & Caldwell 1987; Caldwell & Richards 1989; Dawson 1993), but lateral transport across split root systems has also been observed (Hansen & Dickson 1979; Baker & van Bavel 1988). Regardless of the path of such transport, water must flow away from the stem in some roots; collectively therefore transport of water by plant roots from one soil zone to another can be called ‘reverse flow phenomena’. Here, to develop further the preliminary report made by Smith, Jackson & Roberts (1997), we utilize measurements of reverse flow in tree roots to demonstrate the opposite process to hydraulic lift: the siphoning of water downwards by root systems of trees spanning the gradient in water potential between a wet surface and dry subsoil.

Reverse flow phenomena have previously been investigated using two experimental approaches. In the first, fluctuations in the water content or potential of dry soil have been interpreted as the result of the nocturnal transport of water by roots from adjoining, more moist soil layers (van Bavel & Baker 1985). Richards & Caldwell (1987) were thus able to demonstrate in the field that water absorbed at night from moist soil by deeper roots of the desert shrub Artemisia tridentata was transported to, and lost from, roots in drier upper soil layers. Fluctuations in soil water potential resulted from the subsequent re-absorption of this water during the day for transpiration. Vetterlein & Marschner (1993) similarly found that water was transferred upwards into dry soil by the herbaceous root system of Pennisetum americanum.

In the second approach, radioactive or stable isotopes have been used as tracers of soil water movement. Hansen & Dickson (1979) found that 3H2O was transferred along a series of soil containers interconnected by split root systems. Caldwell & Richards (1989) fed 2H2O to the distal ends of deep roots of A. tridentata and found enhanced concentrations of 2H in neighbouring, shallow-rooted grasses over subsequent days; concurrent fluctuations in the water potential in the root zone of the grasses indicated that they had acquired 2H emitted from A. tridentata roots after hydraulic lift. Dawson (1993) used natural gradients in 2H/H ratios in soil water to demonstrate hydraulic lift by roots of Acer saccharum and subsequent uptake by understorey plants of the water released.

We applied a third technique to the study of reverse flow phenomena. We used heat-balance sap flow gauges to distinguish sap flow towards the trunk in woody roots of Grevillea robusta A. Cunn. from sap flow in the reverse direction, towards the root tips.

Materials and methods


Measurements of sap flow in roots of G. robusta trees were made at the Machakos Research Station of the International Centre for Research in Agroforestry (ICRAF), Kenya (1°33'S, 37°8'E; 1560 m above mean sea level). The site has a semiarid climate, with mean annual rainfall of 782 mm mostly occurring in two distinct rainy seasons; mean rainfall from March to June, the ‘long rains’, is 345 mm; from October to December, the ‘short rains’, mean rainfall is 265 mm. The trees were located on a south-facing, 20% slope with an alfisol (Khandic Rhodustalf), sandy clay loam in texture, overlying a hard gneiss bedrock at depths between c. 0·2 and 2·0 m. A water table was not observed at the site and excavations revealed that the bedrock was free of deep cracks and that roots of the trees did not penetrate more than 2–3 mm into the weathered surface of the rock.

The trees were arranged in a grid pattern with a 3 × 4 m spacing and were 5–6 years old and 8–10 m tall. The soil beneath the trees was bare and kept weed free.


Of the four widely used techniques for measuring sap flow in plant stems reviewed by Smith & Allen (1996), only the heat balance method is suitable for use on roots if sap flow can reverse direction. The symmetrical design of heat-balance sap flow gauges (Fig. 1) means that, when used with a constant power supply, output from the gauge simply changes sign if the direction of flow is reversed. When used with conventional, assymetric configurations of sensors, the heat-pulse or thermal-dissipation methods would fail if flow reversed direction, although modifications to the heat-pulse method can be made to enable measurement of flow in either direction (S. Burgess, unpublished data). The heat balance method would be similarly inappropriate for use on roots if used with a variable power supply that is regulated to maintain a constant temperature gradient across the gauge (Ishida, Campbell & Calissendorf 1991), as the power supply would merely increase if flow reversed, in an attempt to re-establish the temperature gradient.

Figure 1.

. Cross-sectional view of a heat-balance sap flow gauge installed on a root, showing the symmetrical arrangement of the thermocouple junctions used to measure the gradients in temperature across the heater, ΔTa and ΔTb; when the direction of flow is reversed, the mean temperature gradient (ΔT) changes sign.

When constant-power sap flow gauges are used, the circumference of a short, insulated section of stem or root is heated with an electric heater. The power supplied to the heater, output from a radial thermopile and the gradients in temperature across the heater, ΔTa and ΔTb (Fig. 1), are used to quantify components of the heat balance of the gauge; the residual of the heat balance is the heat absorbed by the moving sap stream (qf). The mass rate of sap flow (F) is finally calculated from qf and the mean change in temperature across the heater (ΔT) using (Sakuratani 1981; Baker & van Bavel 1987):


where cs is the specific heat capacity of sap and ΔT = (ΔTa + ΔTb)/2.


Constant-power sap flow gauges were installed on roots after excavating around trees up to c. 0·4 m from the base of the trunk. Roots that were of a suitable size and sufficiently straight to accommodate a sap flow gauge were then selected and excavations extended as required. Soil was removed to a depth of 0·2 m around lateral roots and a depth of 0·4 m around vertical roots.

Installations were made during three periods; a description of the roots and sizes of gauge used in each case are listed in Table 1. Installation 1 was made between 26 November and 17 December 1996. Rainfall ceased for the season on 29 November, when the seasonal total had reached only 157 mm. Installation 2 was made on roots of a second tree between 21 March and 15 April 1997, enabling changes in the dynamics of sap flow associated with the onset of rains to be recorded, as the dry season ended with 50 mm of rain on 30 March. The final period, Installation 3, was between 13 June and 1 July 1997, when gauges were installed on roots of a third tree. This period occurred after the end of the long rains, but 45 mm of irrigation was applied to a 6 × 8 m area centred on the tree between 23 and 25 June.

Table 1.  . Roots for which sap flow was measured during three periods of observation. The nominal diameter of the gauge used and the diameter of the root at the mid-point of the gauge are given, with the cross-sectional area (Ax) of each root at the base of the trunk and the summed Ax of all vertical or lateral roots of each tree Thumbnail image of

Shelters constructed from polythene prevented rain flooding the excavations and diverted stemflow away from the exposed roots. To minimize the influence of fluctuations in ambient conditions on the performance of the gauges, black netting was suspended over the shelter, shading the gauges from direct sunlight, and sections of roots left exposed after installation of the gauges were insulated with foam pipe lagging and covered with aluminium foil. With these precautions, problems identified by Lott et al. (1996) with the use of sap flow gauges on roots were not encountered.

Output from the gauges was recorded using an AM416 multiplexer and 21X data logger (Campbell Scientific Ltd, Shepshed, UK); measurements were made every 30 s and logged as 10 min averages. The temperature of the heated section of root was measured in all cases by placing a 0·2 mm-diameter copper-constantan thermocouple junction beneath the heater. This temperature was used to estimate a storage term in the heat balance of the gauge (Smith & Allen 1996).


To deduce radial heat fluxes from sap flow gauges, and thus enable determination of qf, a constant representing the thermal conductance of the gauge, often called the sheath conductance (Ksh), must be evaluated for each new installation. When gauges are used on stems, Ksh is commonly calculated using data from predawn periods when it can be assumed there is no sap flow (Smith & Allen 1996); however, when sap flow in both directions along a root is suspected this assumption cannot be made. As a consequence, values of Ksh for the gauges installed on roots were determined from data collected at the end of each period of measurement, after sap flow was stopped by severing the root below the distal end of the gauge.


Profiles of soil water content were measured over each period of observation using a neutron probe (IH II, Didcot Instrument Co., Abingdon, UK) and 45 mm-diameter aluminium access tubes, which had been in place since 1993. For Installation 1, these measurements were made in a plot with a similar soil depth and the same configuration of trees, c. 80 m from the tree instrumented with sap flow gauges. For Installations 2 and 3, measurements were made using access tubes 1·0 and 1·8 m upslope from the base of the instrumented trees; these tubes were situated beyond the influence of the shelters covering the excavated roots. Readings were taken regularly at depth increments of 0·2 m, between 0·2 m and the underlying bedrock.

For Installations 1 and 2, water potentials were calculated from measured soil water contents using water retention curves for the five soil horizons identified at the site. Pressure plate apparatus was used to determine the retention characteristics of undisturbed cores from each horizon and parameters of the retention curves were fitted by regression (Campbell 1985). The appropriate retention curve for each depth increment of each access tube was selected by identifying the extent of the different horizons in each profile; this was accomplished using results from particle size analyses of samples extracted during installation of the access tubes. For Installation 3, soil water potentials between the depths of 0·1 m and the bedrock were measured every 0·1 m using an array of tensiometers located 1 m upslope from the trunk of the tree.

During Installations 1 and 3, a pressure chamber was used to measure the water potentials of leaves from the crowns of the instrumented trees. Measurements were made at regular intervals between the predawn period and sunset on several days; mean leaf water potentials were determined from a minimum of three leaves at each time interval.

Data analysis



1 shows that when ΔT is zero or close to zero, F is undefined or implausibly high. For this reason, it is recommended that the power supplied to gauges is set so that, even at the highest rates of flow, ΔT is always higher than about 1·0°C. When the direction of flow reverses, however, ΔT must pass through zero, causing values of F calculated with equation 1 to become undefined. Thus, if flow in roots reversed direction, it was necessary to filter out data from the transition period when calculated flow rates became excessive.

Examination of the sap flow data collected revealed that the period of transition in the direction of flow could be distinguished on the basis of two criteria that were met concurrently only during a change in the direction of flow: F was set to zero if |ΔT|≤2·0°C and (ΔTbΔTa) ≥ 1·0 °C. The latter criterion was effective because it was not met when flow rates were truly high, as (ΔTbΔTa) is proportional to the conductive heat flux from the gauge (Smith & Allen 1996), which is always low when sap flow is rapid.

Results and discussion


The performance of a constant-power sap flow gauge during the reversal of flow in a vertical root is shown in Fig. 2. The key to identifying periods of reverse flow was the sign of ΔT (Fig. 2a), which was positive when sap warmed by the heater moved towards the trunk, but negative when sap flowed in the opposite direction, towards the root tips. Uptake of heat by the moving sap stream was substantial during both day and night (Fig. 2b), except during transitions between positive and negative flow. Flow was thus measured in both directions, although the filtering out of unreliable data near transition periods resulted in the loss of some resolution in rates of sap flow at these times (Fig. 2c).

Figure 2.

. Data from a sap flow gauge, installed on a vertical root of Grevillea robusta, recorded on 13 December 1996: (a) (ΔTbΔTa) and ΔT , where ΔT = (ΔTa + ΔTb)/2; (b) the ratio of heat absorbed by the moving sap stream (qf) to the power supplied to the gauge heater (P); (c) rates of sap flow, before and after filtering of data from the period of transition between positive and negative flow, where positive flow was towards the trunk.


The record of sap flow in roots from Installation 1 is shown in Fig. 3. For the entire period of observation, sap flow in the lateral root was positive, indicating uptake of water from the soil. Uptake continued at substantial rates throughout each night, although nocturnal flow rates declined as the period progressed. In the vertical root, sap flow was negative, or towards the root tips for much of the period. Rates of reverse flow were highest at night and declined as positive flow rates in the lateral root peaked during the day. Later, overnight rates of downward flow in the vertical root gradually declined and, after 11 December, sap flow became positive during the middle of the day.

Figure 3.

. Sap flow in (a) a 30 mm-diameter lateral root of Grevillea robusta and (b) a 15 mm-diameter vertical root during Installation 1; positive flow was towards the trunk. All lateral roots of the tree were cut on 15 December.

All lateral roots of the tree were severed on 15 December. The high rates of sap flow recorded in the lateral root on this day were probably the result of increased uptake compensating for the severing of other roots (Lott et al. 1996), as the gauged lateral was the last to be cut. The following day, sap flow was far higher in the vertical root than on previous days, with much lower rates of downward flow overnight; severing of the laterals thus forced uptake through the vertical roots of the tree. This response to the cutting of roots confirmed that the sap flow gauges were operating correctly.

Figure 4 shows profiles in soil water potential (ψs) during Installation 1. At the start of the period, because the rains had been insufficient to wet the whole profile, there was a strong gradient in ψs between the wet surface and drier soil below 0·4 m. Thus, the downward flow of water observed in the vertical root appears to have resulted from the transfer of water, after uptake by lateral roots, along a gradient in ψs between the surface layer and drier soil at the bottom of the profile. This is the process of downward siphoning of water by roots, the opposite of hydraulic lift.

Figure 4.

. Profiles of total soil water potential (ψs) on 28 November (□), 5 December (▵) and 12 December (○) 1996, during Installation 1.

Downward siphoning of water at night was apparently supplied by nocturnal uptake of water by lateral roots, although some of this uptake may have re-filled storage capacity in the trunk. Prior to 6 December, downward siphoning continued throughout the day, indicating that a proportion of the transpiration stream was diverted down into vertical roots. The decline in reverse flow in the vertical root over subsequent days and nights occurred as the surface dried after rainfall ceased on 29 November and the subsurface layers became wetter, causing the gradient in ψs to gradually weaken (Fig. 4).


The effects of the first rains at the end of the dry season on sap flow in a G. robusta root system are shown in Fig. 5. Before the storm on 30 March, rates of sap flow were low in both the vertical and lateral root, with higher uptake by the vertical root, probably because most soil water was available from between the depths of 0·6 and 1·2 m (Fig. 6). Within c. 12 h of the initial wetting of the soil surface by rain, which began near midday on 30 March, these patterns of sap movement changed markedly. Uptake of water by the lateral root increased rapidly, quickly exhibiting the same pattern of sap flow seen in the lateral root during Installation 1; rates of sap flow peaked during the day but uptake continued throughout the night. In the vertical root, sap flow reversed direction during the night, although negative flow always ceased during the day.

Figure 5.

. Sap flow in (a) a 17 mm-diameter lateral and (b) a 28 mm-diameter vertical root of Grevillea robusta during Installation 2; positive flow was towards the trunk. Rainfall over the period is shown in (c).

Figure 6.

. Profiles of total soil water potential (ψs) on 27 March (□), 31 March (▵), 6 April (○) and 9 April (◊) 1997, during Installation 2.

Thus, when the first rains after a long dry period wetted the soil surface and created a vertical gradient in ψs (Fig. 6), the root system of the tree quickly began siphoning water downwards from the surface. The gradient in ψs was about one order of magnitude smaller than during Installation 1, however, and so rates of downward flow in the vertical root were lower in this instance and reverse flow did not continue during the day.


Prior to the application of irrigation during Installation 3, there was uptake of water by each of the gauged roots during the day, but no measurable sap movement at night (Fig. 7). Wetting of the soil by irrigation was mostly confined to the top 0·3 m of the profile (Fig. 8) and the gradient in ψs created was small, because soil below this level had retained considerable amounts of water after the recently concluded long rains. The gradient in ψs was, however, sufficient to induce changes in the pattern of sap movement: during the nocturnal periods following irrigation, uptake of water by the near-surface lateral continued and sap flow was negative in both the subsurface lateral and the vertical roots. Wetting of the top of the soil profile thus caused downward siphoning of water by the tree, although the rates of downward flow were low because the gradient in soil water potential was small.

Figure 7.

. Sap flow in (a) an 18 mm-diameter near-surface lateral, (b) a 29 mm-diameter subsurface lateral and (c) a 22 mm-diameter vertical root of Grevillea robusta during Installation 3; positive flow was towards the trunk. Irrigation was applied to the soil surface during the period indicated by the shaded region of the x-axis.

Figure 8.

. Profiles of total soil water potential (ψs) on 18 June (□), 26 June (▵) and 29 June (○) 1997, during Installation 3.


The occurrence of reverse flow in vertical roots may depend on the water potential in xylem at the base of the trunk (ψb), where horizontal and vertical roots meet. Sap flowing towards the trunk in horizontal roots would then be diverted downwards and result in reverse flow whenever ψb remained higher than the potential of water at the tips of the vertical roots, which would be approximately equal to the potential of surrounding soil. The highest values of ψb would have occurred overnight, when transpiration was negligible and leaf water potentials (ψl) approached their predawn maxima. Thus, the gradient in potential across the vertical roots would tend to have been largest at night, resulting in the highest rates of downward flow.

Prior to 2 December 1996, during Installation 1, the minimum ψl recorded in the middle of the day was – 1·0 ± 0·07 MPa. Heterogeneity in soil water distribution near the trees may have caused variation in the gradient in ψs with depth, but such a decline in ψl over the day was probably not sufficient to overcome the gradient in ψs below 0·4 m (Fig. 4) and induce water uptake from these depths. However, ψb must have declined with ψl, causing the gradient in water potential across the vertical root to diminish during the day, when the tree was transpiring rapidly; hence, diversion of the transpiration stream downwards continued throughout the day, but the rate of reverse flow in the vertical root was lower during the day than at night (Fig. 3).

Later in the first period of observation, on 14 December, the minimum ψl measured was – 1·3±0·2MPa, which was probably low enough for uptake of water to occur from the bottom of the profile (Fig. 4); sap flow in the vertical root was consequently positive during the middle of the day (Fig. 3). As ψl increased overnight to the predawn maximum of – 0·4 ± 0·01 MPa, concurrent increases in water potential in lateral roots and in ψb would have re-established the downward gradient in potential across the root system, so that flow in vertical roots was negative at night (Fig. 3).

The water relations of downward siphoning during Installation 3 are less clear, as the mean predawn ψl was – 0·2 ± 0·7 MPa, which was lower than ψs at any depth (Fig. 8). Despite this, small downward flows were recorded during the night in the vertical root and the subsurface lateral. Rather than suggesting counter-gradient transport of water in the tree, however, the disparities in ψl and ψs most probably resulted from imprecision in the determination of ψl, heterogeneity in ψs and the height of the crown above the soil, which at c. 10 m would account for a potential difference of 0·1 MPa; ψb was probably intermediate to ψs near lateral and vertical roots, resulting in a gradient in potential (Fig. 8) sufficient to drive downward siphoning.


Reverse flow in roots does not appear to result in equilibration of water potentials across root systems and therefore the gradual elimination of such flow. The continual downward flow of water observed in the vertical root over many days during the initial period of Installation 1 suggests that water passed from the roots into the rhizosphere, as storage of such quantities of water in root tissue without rapid equilibration of water potentials does not seem possible. Wetting of the rhizosphere could ultimately have the same effect, because the low hydraulic conductivity of the drier bulk soil would slow the diffusion of water away from the root, causing the water potential in the rhizosphere to increase and rates of reverse flow in roots to decrease. However, when rates of reverse flow were highest, during nocturnal periods, they were nearly constant, with no indication that reverse flow slowed as nights progressed.

Emission of water from roots may not therefore have been limited by the conductivity of the sandy-clay-loam textured soil at this site. In addition, some water may have been emitted into predominantly dry soil, a possibility if emission was only from the tips of growing roots, as Jensen, Sterling & Wiebe (1961) observed when dye was used to trace the path of water drawn by suction in the reverse direction through solution-grown plants. Consequently, some water transported by reverse flow may be utilized in cell expansion and to wet the rhizosphere near root tips and thus ease the passage of growing roots through dry soil. A question for future research is, therefore: does reverse flow in roots support root growth into dry soil?


Approximation of rates of sap flow in whole root systems can be attempted by extrapolating from sap flow in single roots on the basis of the cross-sectional area of roots at the base of the trunk. When sap flow measured in Installation 1 was extrapolated to the whole root system on this basis, using root cross-sectional areas in Table 1, the mean proportion of water taken up by lateral roots each day that was diverted downwards into vertical roots was 0·26. During Installations 2 and 3, rates of flow in vertical roots were lower and so this fraction was smaller.

Such estimates likely contain large uncertainties, however, because of variation in the lengths of each major root exposed to different water potentials in each soil layer. Thus, to properly quantify the influence of downward siphoning on soil water balances, sap flow must be measured in more roots of each tree than was possible in this study. The observed rates of downward flow in roots indicate, however, that downward siphoning may be a substantial component of the soil water balance where vertical gradients in soil water potential are large. Emerman & Dawson (1996) reached a similar conclusion for hydraulic lift, as they estimated that c. 20% of daily uptake by an A. saccharum tree was transferred to surface soil layers by reverse flow.


Downward siphoning may have important consequences for interactions between trees and neighbouring, understorey plants with shallower root systems. Emission of water into soil may result in the storage of water, if it can be re-extracted at a later time, below the maximum rooting depth of neighbours. Storage of water in this way could increase the efficiency of water utilization by trees, as it could reduce the amount of water lost by evaporation from the soil, and reduce the availability of water to shallow-rooted neighbours. Downward siphoning of water could therefore give trees a competitive advantage over neighbours in dry environments where plants are reliant on seasonal rainfall for water.

If downward siphoning supports root growth, it may enable trees to maintain a network of deep roots during periods when seasonal rains are poor and only wet the surface layer of the soil. Such growth, which need only be sufficient to replace fine roots lost as a result of natural turnover, would enable trees to rapidly exploit water from below the root zone of shallow-rooted neighbours during seasons wet enough to recharge subsurface soil layers, thus ensuring the trees have a larger resource pool than neighbours and so a competitive advantage.

At sites in arid or semiarid regions where deep-rooted trees or shrubs utilize ground water, downward siphoning may enable roots to grow down through dry soil layers to the water table. Downward siphoning may therefore be an important adaptation allowing juvenile phreatophytic plants to make the transition from reliance on seasonal rainfall to exploitation of ground water.

Like downward siphoning, hydraulic lift has implications for resource utilization and interactions among coexisting species. Hydraulic lift is hypothesized to have a role in facilitating the acquisition of nutrients from dry, but fertile, surface soils, by mobilizing nutrient ions and prolonging the activity of soil micro-organisms (Caldwell, Richards & Beyschlag 1991). As the availability of nutrients from deeper soils is typically low, hydraulic lift may consequently play a key role in maintaining the nutritional health of deep-rooted plants in dry regions. Downward siphoning is unlikely to have a similar function, however, because it occurs when the soil is wettest near the surface, so that uptake of nutrients is not limited by water availability.

Caldwell et al. (1991) suggested that, like downward siphoning, hydraulic lift enables plants to store water for later use, when transpirational demand cannot be met by the sparse root network in the subsoil. In contrast to downward siphoning, however, hydraulic lift also makes water available to shallow-rooted neighbours (Corak, Blevins & Pallardy 1987; Caldwell & Richards 1989; Dawson 1993), creating a form of parasitism in which water resources captured by one species are transferred to competitors (Caldwell et al. 1991). While downward siphoning may enhance the competitiveness of deep-rooted trees and shrubs, the opposite process, hydraulic lift, may therefore improve the competitive ability of neighbours; thus, hydraulic lift may encourage close spatial relations between trees and grasses in water-limited environments, while downward siphoning encourages their separation.

Reverse flow phenomena hold similar ramifications for the design and management of agricultural systems which combine species, such as agroforestry. Storage of water through downward siphoning would enhance the competitiveness of trees, by transferring resources away from zones that are accessible to crops. Downward siphoning would therefore tend to reduce the extent to which the patterns of water uptake by crops and trees such as G. robusta are thought to be complementary (Lott et al. 1996; Howard et al. 1997). The competitiveness of trees would, in contrast, be reduced by hydraulic lift, if it enhanced the availability of water to crops (Emerman & Dawson 1996). Thus, while hydraulic lift might allow a higher density of trees to be grown without deleterious effects on crops, it could be necessary, for example, to increase the spacing between trees in agroforestry in conditions where downward siphoning may be substantial.


Downward siphoning may be an important, though previously missing, component of the soil water balance at locations where gradients in water potential across root systems, from a wet soil surface downwards, are large. These conditions are most likely to be found in arid or semiarid regions where there is no water table or only a very deep water table. Vertical gradients in water potential at such sites are probably largest, and therefore downward siphoning most important, when seasonal rains are poor, as was the case in this study during Installation 1, or when rainstorms bring an end to long dry spells, as occurred during Installation 2.

By transferring water beyond the reach of more shallow-rooted neighbours, downward siphoning by deep-rooted perennials may have a significant impact on the distribution of vegetation and interactions among species in natural ecosystems, such as savannas or desert-shrub communities. Downward siphoning is also likely to influence the partitioning of water resources between trees and crops in agroforestry; at sites where considerable downward siphoning is possible therefore this should be reflected in the strategies used to design and manage agroforestry systems.

Constant-power sap flow gauges provide an important new tool for the investigation of downward siphoning and reverse flow phenomena in general. They will enable hydraulic lift and downward siphoning to be quantified in the field, and when combined with measurement of soil water and the use of isotopic tracers, they will enable study of the role of reverse flow phenomena in resource acquisition strategies used by plants.


This publication is an output from a research project funded by the Department for International Development of the United Kingdom. However, the Department for International Development can accept no responsibility for any information provided or views expressed. The work described here was funded from project R6363 of the Forestry Research Programme. We are especially grateful for the able technical assistance given by staff at Machakos Research Station, in particular Mr Elijah Kamalu and Mr Patrick Angala. We also thank Dr Simon Allen for valuable advice given during the planning and implementation of this study.