1. Using three independent methods (eddy covariance, heat pulse and open-top chambers), diurnal and seasonal measurements of evapotranspiration were made in a wet–dry Eucalypt savanna of the Northern Territory, Australia.
2. Total annual dry-canopy water loss was estimated to be 870 mm and understorey evapotranspiration contributed 557 mm to this flux. Understorey evapotranspiration occurred predominantly during the wet season as bare soil evaporation and transpiration of Sorghum spp., a C4 grass.
3. Annual transpiration from trees was 313 mm, significantly less than the grassy understorey. Despite a very high degree of seasonality in distribution of rainfall and large changes to soil and atmospheric water content, water use by the trees did not differ between wet and dry seasons. This suggests that mature trees exploit a large soil volume and this may include extraction from the capillary fringe of the shallow water table (2–10 m below the ground surface).
4. The open canopy created an aerodynamically rough surface well coupled to the atmosphere with the coupling coefficient, Ω, ranging from 0·40 to 0·11 over a wet–dry cycle.
5. Leaf area index (LAI) of the overstorey was typically 1·0 in the wet season and 0·65 in the dry season. The decline in tree LAI occurred when evaporative demand showed a similar proportional increase. Consequently overstorey water use remained relatively unchanged throughout the year.
6. Given the very high rainfall intensities of the monsoonal climate and low LAI of the site, canopy interception was set at 5% of rainfall. Including this amount gives an annual evapotranspiration of 958 mm for this savanna.
If you can't find a tool you're looking for, please click the link at the top of the page to "Go to old article view". Alternatively, view our Knowledge Base articles for additional help. Your feedback is important to us, so please let us know if you have comments or ideas for improvement.
The wet–dry tropics of northern Australia are characterized by the occurrence of predictable wet and dry seasons (Taylor & Tulloch 1985; Duff et al. 1997). Vegetation of the region is dominated by Eucalypt open forests and woodlands, characterized by widely spaced trees with low overstorey leaf area index (LAI) and a seasonal grassy understorey with bare soil patches. Globally, savannas cover an area equal to that of tropical forests (Andreae 1991) and within Australia cover c. 25% of the land area. As such they contribute significantly to the regional hydrological cycle.
The phenology of savannas in northern Australia differs from savannas elsewhere in the world. African and Indian savannas are dominated by deciduous species (Menaut & Cesar 1979; Yadava 1990). In contrast, Llanos of South America are dominated by evergreen species (Sarmiento, Goldstein & Meinzer 1985). Savannas of northern Australia contain an approximately equal number of species in each of four guilds (evergreen, brevi-deciduous, semi-deciduous and deciduous; Williams et al. 1997), although evergreen Eucalypt species account for more than 80% of the standing tree biomass and canopy cover (O’Grady, Eamus & Hutley 1999). Grasses in these savannas are mostly perennial and absent during the dry season.
Seasonal changes in soil and atmospheric water content are large and have significant impacts on daily and seasonal patterns of leaf water potential (Ψw), stomatal conductance (gs), transpiration (E) and photosynthesis in all species examined to-date (Eamus & Cole 1997; Fordyce, Duff & Eamus 1997; Myers et al. 1997; Prior, Eamus & Duff 1997a,b). However, determinations of Ψw, gs and E have thus far been confined to leaf and tree scale rather than canopy-scale measurements. Predicting canopy and stand behaviour from leaf-scale measurements is notoriously difficult (Jarvis & McNaughton 1986).
Sustainable management of ground-water is becoming increasingly important globally. A key issue for such management is an understanding of the role of vegetation in the hydrological cycle. This is especially important in seasonally dry climates where the timing of peak demand for water by humans is likely to coincide with the largest dependence of vegetation on a gradually declining store of soil water. Given the dominance of evergreen trees in the overstorey canopy, it is possible that canopy water use in the dry season is significant.
For a typical catchment of the humid coastal zone of the Northern Territory we have addressed the following questions: what are the daily and seasonal patterns of water use by a Eucalypt savanna in northern Australia; what are the relative contributions of the soil, understorey and tree canopy to total evapotranspiration in the wet and dry seasons; does the dry season (with attendant drying of the upper soil profile) significantly reduce canopy water use? Such data will enable parameterization of models describing surface fluxes of mass and energy from tropical savannas and contribute to our ability to sustainably manage ground-water in the future.
Materials and methods
Study site and climate
The study site was located at Howard Springs (130° 45′E, 12° 30′S), c. 35 km south-east of Darwin in the coastal, humid zone of the Northern Territory, Australia. The wet season occurs from December to March inclusive and accounts for c. 95% of the 1750 mm mean annual rainfall. This is followed by an essentially rainless dry season, lasting from May to September (Taylor & Tulloch 1985). The months of October and November are transitional to the wet season and are characterized by increases in humidity and temperature, occasional thunderstorms and the onset of canopy flushing by many of the tree and shrub species (Williams et al. 1997). Figure 1 shows daily rainfall for the Howard Springs station (Bureau of Meteorology, Station no. 14149) 1996–1998, the period of the study. During December to February, monsoonal troughs and tropical cyclones result in sustained high intensity rain events. Total rainfall of the 1996/1997 water year was 2834 mm, the highest on record, with particularly high totals recorded in early January 1997, as a result of Cyclone Rachel.
Mean daily maximum temperatures at the Darwin Airport range from 30·4 °C (July) to 33·1 °C (October and November). Day-time values of vapour pressure deficit range from 1·5 kPa (wet season) to 3·1 kPa (June and July). Maximum and minimum temperatures have a range of 7 °C (wet season) and 11 °C (dry season) (McDonald & McAlpine 1991). Mean annual pan and potential evaporation for Darwin Airport (35 km from the field site) is c. 2700 mm and 2259 mm per annum, respectively. Penman potential evaporation is largely aseasonal (Fig. 7) relative to non-monsoonal climates of Australia, with dry season ranges from 6·0 to 7·3 mm per day, peaking during September and October when solar inclination and radiation loads are highest (Vardavas 1987). Wet season rates are lower at 5·1–6·0 mm day−1 because of increased cloud cover and water vapour pressure.
Vegetation of the catchment and region is a mosaic of Eucalypt-dominated woodlands, open forests, closed forests (sensuSpecht 1981), seasonally flooded swamps and wetlands. In the Howard River catchment Eucalypt open-forest dominates and all measurements were made in this community type. The overstorey is dominated by two evergreen species, Eucalyptus tetrodonta (F. Muell.) and Eucalyptus miniata (Cunn. Ex Schauer) which form a canopy of about 50% cover. These two species account for c. 80% of the tree basal area of 8–10 m2 ha−1. Overstorey LAI varies seasonally owing to the presence of brevi-, semi- and fully deciduous tree species and ranges from 0·6 during the dry season to 0·95 during the wet season. (O’Grady et al. 1999). Evergreen canopy fullness varies little seasonally (Williams et al. 1997). Sub-dominant tree species included Erythrophyleum chlorostachys (F. Muell.), Terminalia ferdinandiana (F. Muell.) and Eucalyptus porrecta (S. T. Blake). The understorey consists of semi-deciduous and deciduous small trees and shrubs with a seasonally continuous cover of annual and, to a small extent, perennial grasses. Understorey LAI changes dramatically over the wet season (a value of 2–3) with the C4 grass Sorghum spp. dominating. This grass senesces early in the dry season and understorey LAI remains low throughout the dry season (about 0·2).
Soils are derived from the Koolpinyah surface, a Late Tertiary 30–40 m deep sediment mantle extending from the Darwin region to the Arnhem escarpment of Kakadu National Park, NT. Soils are extensively weathered and laterised, weakly acidic and low in nutrient status (Russell-Smith, Needham & Brock 1995). Eucalyptus tetrodonta and E. miniata dominated open-forests and are commonly associated with lateritic red and yellow earths (Cole 1986), which tend to have A-horizons of well drained, highly weathered sandy loams with a massive and earthy structure. Transition at 15–30 cm to a sandy clay loam B-horizon is gradational and can extend up to 1–2 m, where ferricrete boulders occur in a matrix of mottled, heavy clays forming a duricrust of low permeability and variable depth. Prominent macropores, often containing tree roots, are found in this layer. Rounded ferricrete gravels occur on the soil surface and throughout the profile, and are between 20 and 50% by volume. The large seasonal amplitude in soil water store water for the top 3 m is shown in Fig. 2 for the 1996/1997 water year. Data for this figure were calculated from volumetric soil water content measurements to 3 m depth, using a capacitance soil moisture system, EnviroScan RT5 (Sentek Pty Ltd, Kent Town, Australia). Storage properties of these soils are poor, with only 0·08 cm3 cm−3 released between field capacity and wilting point (Cook et al. 1998). Plant available water for upper soil horizons is c. 70 mm m−1 and for the subsoil (1·3–3·9 m) is 90 mm m−1, reflecting a small increase in clay content (Cook et al. 1998). These soils overlie a surface aquifer with a seasonal amplitude of 8–10 m, with the water table rising to within 2 m of the land surface during the wet season (Pidsley et al. 1994).
Flux estimates were made for understorey, tree and whole-canopy strata. Seven field campaigns, lasting from 8 to 14 days were made between October 1996 and October 1998 (Table 1). Measurements were made of whole canopy flux using eddy-covariance techniques. Tree transpiration rates were also measured using heat pulse sensors. During 1998, understorey transpiration and evaporation measurements were made using open-top chambers.
Table 1. Measurement programme and techniques used at the Howard River experimental site
Time of year
Total forest flux
The site had adequate (1 km) fetch in all directions, slopes of less than 1° and an open canopy structure (mean tree-stem to tree-stem distance was 4·5 m), creating an aerodynamically rough surface. Eddy covariance and associated instruments were mounted on a pole supported by a tower at a height of 18 m. Tree canopy height was 14–15 m. We used a Campbell Scientific (Logan, UT, USA) eddy-covariance system, consisting of a three-dimensional sonic anemometer (model CSAT3) and a krypton hygrometer (model KH20) interfaced to a 21X datalogger. All data were collected at 10 Hz with latent energy (LE) and sensible heat flux (H) calculated at 30 min intervals. Corrections for oxygen absorption by the krypton hygrometer (Tanner, Swiatek & Greene 1993), co-ordinate rotation and corrections accounting for air density changes (Webb, Pearman & Leuning 1980) were applied to raw fluxes.
Measurements of meteorological conditions were also made, with mean values collected every 30 min. Measurements of air temperature, relative humidity, wind speed and direction (model 03001–5, RM Young Wind Sentry), and net radiation, Rn (model Q7·1, Radiation and Energy Balance Systems, Seattle, WA, USA) were made simultaneously at the same reference height as the eddy-covariance instruments (18 m). Soil heat flux, G, was estimated using a combination of four soil heat flux plates (model HFT-3, Radiation and Energy Balance Systems, Seattle, WA, USA) buried at 8 cm with an averaging soil thermocouple (model TCAV, Campbell Scientific, Logan, UT, USA) located at 2 and 6 cm depth in the soil. The four heat flux plates were randomly located within a 10 m plot with mean flux calculated from the four plates. Soil samples were also taken for water content estimates. The complete system provided data on all components of the forest energy balance, with the exception of biomass storage, which was assumed to be small on a diurnal basis.
Water use by trees was estimated using the heat-pulse technique (Swanson 1972). During each eddy-covariance measurement period, sapflow sensors (Greenspan Technology, Warwick, Queensland, Australia) were installed in trees of varying sizes and species. Heat pulse velocity measurements were scaled to tree water use using the weighted averages technique of Hatton, Catchpole, & Vertessy (1990). The installation at our site is described by O’Grady et al. (1999). Tree water use was measured in eight to 10 randomly selected trees within a 100 m radius of the eddy-covariance tower. For each instrumented tree, d.b.h. and leaf area were recorded. Tree leaf area was estimated using the ‘Adelaide technique’ which involved estimating leaf area visually by counting the number of leaf modules on a tree (Andrews, Noble & Lange 1979). Reference leaf modules were collected in the field and their total area was determined using a Delta-T leaf area meter (Delta-T Devices, Cambridge, UK).
Daily totals of water use by trees were regressed for each day against d.b.h. (cm), cross-sectional area (m2) and tree leaf area (m2). These relationships were used to express tree water use on an areal basis in three permanent plots (30 m × 30 m). Each plot was surveyed for tree d.b.h./basal area and leaf area. Plot water use was the sum of all trees within the plot divided by the plot area to give tree water use, Et, in mm day−1.
Open-top chamber measurements
Estimates of understorey evapotranspiration were made using open-top chambers (OTC) at the end of the wet season and during the late dry season of 1998 (see Table 1). Measurements were made on bare soil, grasses and other common understorey shrubs and saplings. Evaporation from the substrate enclosed by the chamber was calculated by the difference between vapour density of air entering and exiting the chamber. The chamber consisted of two sections – a lower cylindrical base of 0·77 m diameter and 1·23 m height, made from clear acrylic plastic and a metal frame supporting a clear plastic cone mounted on top of this base. The upper frame had an exit port of 10·5 cm diameter, with the tapering of this upper section designed to improve mixing as air exited the chamber. Total chamber height was 2 m and the volume enclosed was 0·78 m3.
Air was pumped into the chamber using an inlet fan mounted at the base of the chamber and flow rate measured at the exit port using a propeller anemometer. Typical exit flow rates were 4–6 m s−1 depending on canopy drag, giving volumetric flows of 0·035–0·052 m3 s−1. Water vapour densities of air entering and leaving the chamber were measured using a LiCor CO2/H2O analyser (model LI 6262, Lincoln, NE, USA). Air streams were ducted to the analyser using Beva-Line tubing at flow rates of 8 litres min−1 and controlled by a mass flow controller (model FC 280, Tylan General, Torrence, CA, USA). Output from the analyser was recorded using a datalogger (21X, Campbell Scientific, UT, USA) logging at 1 s intervals. The air stream to the analyser was switched between inflow and then outflow, with the measurement of each lasting between 3 and 5 min until stable. Following measurement the chamber was removed from the plot to reduce its influence on plot microclimate (Denmead et al. 1993). Air temperature of the out-going air stream was between 2 and 3 °C above ambient during the wet season (March), although this chamber effect was not observed during the dry season (September/October). Plot evapotranspiration was calculated as follows:
where Eu is the understorey evapotranspiration rate, ρout and ρin are the vapour densities (g m−3) of the out-going and in-coming air streams, respectively, V is the volumetric flow rate (m3 s−1) and A is the chamber area (m2). Rates measured over the course of a day were integrated to give estimates of evaporation in mm day−1.
The OTC was tested by measuring evaporation from a pan placed within the chamber. 2·5 litres of water was added to the pan (20 cm × 30 cm) and hourly rates of evaporation from the pan were measured using the OTC. Integration of the chamber-based estimates gave a daily loss from the pan of 667 ml, compared to the actual loss of 661 ml.
A fractional cover estimate for each substrate type (bare earth, grass, sapling) was used to weight estimates of Eu derived from the OTC. Line intercept methods were used to estimate fractional cover occupied by each substrate type on 10, 50 m transects (Whitehead et al. 1994). These fractions were combined with chamber data to give spatially averaged Eu in mm day−1. During the wet season, six plots were measured (20–22 March 1998), including three of mixed Sorghum spp. grass clumps, bare soil, plus common understorey species Planchonia careya (F. Muell.) R. Knuth and T. ferdinandiana. During the late dry season measurement (29–30 September 1998), four plots were monitored and included bare soil with a small amount of dead Sorghum spp. cover, one E. tetrodonta sapling, one with Cycad armstrongii (Miq.) and one with P. careya saplings. These substrates represented more than 95% of the cover in the understorey at each measurement time. During the week prior to the dry season chamber measurement, the site received 14·2 mm of rainfall.
Fluxes as estimated using the eddy-covariance system are given in Fig. 3 for days representing seasonal extremes. Energy-balance closure ranged from 8 to 15% when integrated over a day. Plots of LE, H, Rn and G describe typical diurnal patterns of sensible and latent heat transfer, with maximal fluxes 1–2 h after solar noon. Fluxes of LE declined as the dry season progressed and were associated with a shift in the partitioning of H, LE and G. The Bowen ratio (β = HLE−1) was calculated from daily integrals of H and LE and on these days and ranged from 0·27 for the wet season to 3·4 for late dry season. Soil heat flux, G, was a significant component of the energy balance and accounted for 14 and 20% of Rn during the wet and late dry seasons, respectively. During the wetter months G was of similar magnitude to H (Fig. 3a).
Seasonal measurements of water use at the Howard Springs site, partitioned into total, tree and understorey fluxes are described by Fig. 4. Mean Ea ranged from 3·5 mm day−1 during the late wet season to a minimum of 1·2 mm day−1 by the end of the dry season. In contrast to the seasonality of the total forest flux, scaled estimates of tree transpiration were remarkably constant, varying between 0·83 and 0·9 mm day−1 for the duration of the study, except for the late dry season estimate for 1998, which was lower at 0·57 mm day−1.
Chamber-based estimates of the weighted mean Eu during March of 1998 were c. 2·8 mm day−1. During March, the understorey at Howard Springs was characterized by high levels of Rn (12·6 MJ day−1, 80% of above canopy flux), air temperature (36 °C maximum) and soil moisture (Fig. 2). Under these conditions, tall (1·8 m), dense Sorghum spp. tussocks with a plot LAI of 2·5 evaporated water at rates of 6 mm day−1, at or near potential rates of atmospheric demand. Bare-soil evaporation was 2·2 mm day−1. By the late dry season, bare-soil evaporation was reduced to 0·52 mm day−1, a rate affected by rainfall of the previous week (14·2 mm), which was the first of the pre-monsoonal storms. Plots of litter and saplings were evaporating at c. 1·5 mm day−1 and the weighted average Eu was 0·55 mm day−1.
Meteorological and energy-balance data collected during each eddy-covariance measurement period were used to calculate equilibrium evaporation (Eeq, Priestley & Taylor 1972) and Penman potential evaporation (Eo). At Howard Springs, rates of Ea and Et were well below Eeq and Eo. The seasonal patterns of Ea/Eeq and Ea/Eo are given in Fig. 5a.
Estimates of annual evapotranspiration
Annual evapotranspiration was estimated using a combination of measured rates of canopy flux (Ea) and scaled estimates based on long-term, mean monthly Epan records collected by the Bureau of Meteorology at the Darwin Airport, the most reliable evaporation record for the region. Estimates of Ea for March, April, July, September and October were calculated from mean daily rates as measured during these months (Table 1). The evaporative fraction, Ea/Epan was also calculated for these periods. For periods when no Ea data were available this ratio was linearly interpolated to give a value of Ea/Epan for each month of the year (Fig. 5b). This fraction was multiplied by the corresponding monthly Epan to give an estimate of Ea for non-measured periods. These estimates, expressed as mm day−1 are given in Fig. 7. No eddy covariance measures were conducted during December, January or February, the wettest months, as site access was not possible. It is likely that rates during March are similar to January and February as evaporative demand, soil water store and LAI are similar.
As all measurements were made under dry canopy conditions, Ea should approximately equal Et plus Eu. Understorey evaporation measurements were undertaken to compare and verify estimates derived from the eddy-covariance and sap-flow measurements. Chamber-derived estimates of Eu appear to be reasonable and within the range of values given for other savannas (de Jager & Harrison 1982; Scholes & Walker 1993) and Eucalypt forests of Australia (Greenwood et al. 1985; McJannet et al. 1996). During March 1998, the scaled estimate of Eu was 2·8 mm day−1 and Et was 0·9 mm day−1 giving a total of 3·9 mm day−1. This is in reasonable agreement with the mean Ea of 3·5 mm day−1 for these same days (20–23 March 1998). During clear sky conditions (21 March), Ea reached 3·94 mm day−1 in excellent agreement with mean value of Eu plus Et. During the late dry season, comparisons are less favourable, with Eu plus Et equalling 1·11 mm day−1 compared to the eddy-covariance estimate of 1·52 mm day−1. The value for Et for September/October 1998 was the lowest of any measurement period. This is difficult to explain and appears to be an underestimate. Other late dry season estimates were 0·83 and 0·89 mm day−1 for October 1996 and September 1997, respectively, a mean of 0·86. Using this mean value for Et for the dry season gives a value of 1·43 mm day−1, in reasonable agreement with the estimate from eddy covariance. Given the error involved in the three methods (in the order of 10–20%) these comparisons suggest that some faith can be placed in the longer term measures using sap flux and eddy covariance and an approximation of Eu can be obtained as the difference between Ea and Et. Estimates of these components plus Eo are given in Fig. 7 and seasonal and annual totals for Et, Eu and Ea are given in Table 2.
Table 2. Wet and dry season totals of tree water use (Et), understorey evaporation (Eu) and actual evapotranspiration (Ea) for Howard Springs. All values are in mm
Riley, Waythe, & Goss (1995) compared rainfall catches beneath Eucalypt trees, grass patches and shrubs with rainfall received on adjacent bare soil patches at sites located in Kakadu National Park, c. 200 km east of the Howard Springs site. Vegetation of the Kakadu study is similar to savanna found at the Howard Springs and the results are applicable to our site. A regression of rainfall catch beneath individual tree canopies (throughfall, TF) vs bare soil catch (incident rainfall, RF) was TF = (0·99 × RF) − 0·44, with an R2 = 0·97 (Riley et al. 1995). For the grass canopy the regression was TF = (0·96 × RF) + 0·02, and for individual shrubs, TF = (0·79 × RF) − 0·25. These relationships suggest that interception by trees and grasses is very low but from individual shrubs it could approach 20%. However, cover of shrubs in the understorey at our site was c. 5%, being dominated by bare soil patches and Sorghum spp. clumps. These regressions were applied to our site for 82 wet-season storm events from 1997/1998, which ranged from 0·5 to 141 mm. The regressions used relate to isolated trees and shrubs and throughfall totals were multiplied by our measured fractional cover (0·5 for trees, 0·05 for shrubs and 0·65 for grasses) to express interception at a stand scale. Total interception was 80 mm, 4·5% of received rainfall (1780 mm).
Given these simple analyses, rainfall characterized by high intensities (up 95 mm h−1) with large drop sizes (Calder, Wright & Murdiyarso 1986) and a canopy of low LAI, it is likely that interception loss at the Howard Springs site would be very low. In the absence of a detailed study (e.g. contribution from stemflow), interception was set at 5% of annual rainfall, i.e. 88 mm. Annual total ET is then Ea (870 mm) plus interception, 958 mm. Including the approximation for interception, evapotranspiration estimates for Howard Springs appear to be within the range given for other humid savannas with rainfall above 1000 mm (Table 3).
Table 3. Evaporation estimates from tropical savannas. All values are in mm per annum
Other water-balance/evaporation studies conducted in the northern Australia include the wetland study of Vardavas (1988, 1989) and that of Cook et al. (1998). Cook et al. (1998) also estimated Ea during periods of climatic extremes, namely April and September 1994. Mean Ea for April was 3·8 mm day−1 and for September 2·0 mm day−1, comparable to our April and September 1997 estimates and their rates are higher. This may represent year to year and within catchment variability of water use as their site had a differing fetch. There was excellent agreement between the two studies for annual estimates of Eu, with the Cook et al. (1998) estimate being 562 mm (calculated as total evaporation minus tree transpiration) compared with 557 mm for this study. However, quoted values for Ea/Eeq differed, with their April value being 0·91 for and September value of 0·68, significantly higher than values found during this study (Fig. 5). As a result, scaled annual totals of evapotranspiration differ by 15% (Table 3).
Evaporation and savanna structure
This study has described a system whose patterns of energy balance and vapour exchange are typical of savanna systems, which function neither as grassland nor as a forest, with tree and grass components behaving in a contrasting manner. Widely spaced trees and low LAI has significant influence on the flux of mass and energy from the understorey (Eastham et al. 1988). The seasonality of Ea at Howard Springs is the result of the marked changes in flux from the understorey, which is determined by soil water supply (Fig. 6). Increased atmospheric demand from the mid-dry (May) to the late dry season (October) corresponded to periods of steadily declining rates of Ea (Fig. 7). While evergreen tree species dominated site biomass, their contribution to wet season total evaporation was less than 30% with the remainder arising from the understorey (Fig. 7). As the dry season progressed and soil water content decreased, Eu declined and by the late dry season Et accounted for 75% or more of the total evaporative flux. Even during periods of very high soil moisture, the evaporative fraction (Ea/Eeq) of the forest never exceed 0·7 (Fig. 5a).
The decoupling coefficient, Ω (Jarvis & McNaughton 1986), was calculated for the Howard Springs site using bulk canopy-conductance values estimated from an inversion of the Penman–Monteith equation. Low values of Ω indicate a close coupling of transpiration rates to atmospheric VPD and that changes in both stomatal conductance and VPD have a large impact on transpiration rate. Mean midday values ranged from 0·40 to 0·11 for the wet and dry seasons, respectively. These are in agreement to values given by Miranda et al. (1997), who calculated Ω at 0·32 and 0·17 for the wet and dry seasons, respectively, for Brazilian cerrado. Such ranges of Ω suggest coupling to the atmosphere, with transpiration rates largely controlled by stomatal aperture in response to changes in VPD.
Extremes in tree leaf water stress are not experienced at this site, despite 6 months of no rainfall (Duff et al. 1997). Canopy flushing of evergreen tree species occurs during the late dry season, a period characterized by maximal atmospheric demand and minimal upper-profile soil water availability (Fig. 2). This fact and the seasonally constant Et (Fig. 4) suggest that trees were extracting moisture from the subsoil (> 3 m) and capillary fringe of the water table as it receded over the dry season. Excavation of tree roots suggested that there was an extensive development of root biomass in the sandy loams of the top 1 m of soil. Large-diameter roots (> 3 cm) of large trees were able to extend through weaknesses in the duricrust and access water stored in the heavier clay sediments of the subsoil. This is similar to rooting patterns of Jarrah forest (Eucalyptus marginata) in Western Australia, which are also supported by lateritic sandy loams overlying a duricrust at c. 1–2 m depth (Kimber 1974; Crombie, Tippett & Hill 1988).
The distinguishing structural characteristic of savannas is the coexistence of grassy and woody components, possibly resulting from partitioning of soil water resources between the two competing life forms (Scholes & Archer 1997). During the wet season, tree and understorey vegetation at Howard Springs both utilize the upper profile (top 1 m), but given wet season rainfalls of up to 1700 mm, there may be little competition for water. Trees may utilize both upper and lower profiles, with the subsoil extraction dominating during the dry season. The extent of this extraction, which results in dry season discharge of the surface aquifer (Vardavas 1988, 1989), is a key management question, as significant interaction of vegetation with the surface aquifer suggests possible vulnerability to any human-induced alteration in ground-water dynamics.
This work was supported by the Land and Water Resources Research & Development Corporation (LWRRDC) and the Cooperative Research Centre for the Sustainable Development of Tropical Savannas (TS-CRC), Darwin NT. We are grateful to Errol Kearle and Ian Lancaster of the Natural Resources Division of the Department of Lands, Planning and Environment, NT Government, for the provision of soil moisture and climate data. We also thank Dr Tom Hatton and Dr Alex Held of CSIRO Division of Land and Water for the use of dataloggers and associated meteorological sensors.