Carbon isotope discrimination by a sequence of Eucalyptus species along a subcontinental rainfall gradient in Australia


  • J. M. MIller,

    1. Environmental Biology Group,
    Search for more papers by this author
  • R. J. Williams,

    1. CSIRO Division of Wildlife and Ecology, and Co-operative Research Centre for Tropical Savannas, PMB Box 44, Winnellie, NT 0822, Australia
    Search for more papers by this author
  • G. D. Farquhar

    1. Environmental Biology Group,
    2. Co-operative Research Centre for Greenhouse Accounting, Research School of Biological Sciences, Institute of Advanced Studies, Australian National University, GPO Box 475, Canberra, ACT 2601, Australia, and
    Search for more papers by this author
    • Author to whom correspondence should be addressed.


1. The 13C/12C discrimination (Δ) by a series of co-occurring and replacement Eucalyptus species was measured across an eightfold decrease in rainfall. As Δ is a measure of the stomatal limitation on photosynthesis, it should provide a subcontinental scale measure of water-limited plant physiological performance.

2. Leaf Δ of five of 13 species decreased with decreasing rainfall, seven exhibited no trend, and one increased. Wood Δ decreased in eight species, showed no trend in four, and increased in one species.

3. Species replacements were marked by a shift in Δ reflecting greater stomatal limitation on carbon assimilation.

4. Wood Δ was less variable than leaf Δ.

5. There was a non-linear response of the multispecies average leaf and wood Δ to decreasing total annual rainfall. This response reflected the spatial pattern of the sensitivities of Δ to decreasing rainfall of the individual species. It was not the result of a proposed emergent behaviour where the trend in the multispecies average differed from that of the individual species.

6. Patterns of Δ across the distributions of species (reflecting increasing stomatal limitation on assimilation) did not provide a simple measure of the physiological limits of the distribution of eucalypts in north-western Australia.


North-western Australia has a pronounced rainfall gradient from the mesic coastal and subcoastal landscapes to the semiarid and arid interior. The dominant vegetation is tropical savanna trees, predominantly Eucalyptus spp., over a grassy understorey (Williams et al. 1996). Many of the dominant Eucalyptus species in the savannas are distributed in bands perpendicular to the rainfall gradient (Brooker & Kleinig 1994), and there is a pronounced pattern of species replacements amongst the eucalypts across the rainfall gradient. The ability of these species to withstand decreasing water availability must play a role in determining their distributions. By examining broad-scale patterns of 13C/12C discrimination (Δ) in the eucalypts, as a long-term integrative measure of stomatal limitation on photosynthesis, the link between species distributions and the physiological performance of adults was investigated.

The Δ recorded in plant tissues measures the internal partial pressure of CO2 (pi) during photosynthesis (A), reflecting the balance between rates of CO2 supply to, and demand from, the intercellular air spaces (Farquhar et al. 1982). Δ is negatively related to A/gc where gc, the conductance of CO2 into a leaf, depends on the plant being able to sustain the co-occurring transpiration which requires the acquisition and transport of water at rates equal to the evaporative demand. Due to the saturating response of A to increasing pi, Δ provides a non-linear measure of plant performance. With a fixed photosynthetic capacity, changes in Δ due to changes in gc at high Δ reflect less of a change in A than similar changes at low Δ. Decreases in Δ can arise from either decreases in gc or increases in photosynthetic capacity. While both reflect increased stomatal limitation on A, the two sources of variation in Δ may result in opposite effects on plant performance, especially when coupled with simultaneous changes in leaf area. A high-Δ plant may be more productive (per unit leaf area) than a co-occurring low-Δ plant if the low Δ is due to small gc rather than large photosynthetic capacity. The amount of photosynthate produced at a given pi will weight the subsequent Δ value of the tissues. Periods with sufficient water availability, and therefore large A and Δ, may dominate the signal recorded during dry periods when there is both a low A and low Δ.

It seems reasonable that plants should respond to decreasing moisture availability (either in time or space) by decreasing stomatal conductance relative to photosynthetic capacity, generating patterns of decreasing Δ (Comstock & Ehleringer 1992; Ehleringer 1994; Ehleringer & Cooper 1988; Schulze et al. 1998; Williams & Ehleringer 1996). At some point this increasing stomatal limitation on carbon gain could place a species at a competitive disadvantage compared to other species with morphological characteristics better suited to the acquisition of water under conditions of increasing aridity. These morphological features could include less allocation to leaf area and trunk height, and more to root length and depth. The difference in allocation patterns could segregate the alternative growth strategies along a moisture gradient, with the species with more root allocation experiencing less stomatal limitation on carbon gain in drier areas, but being unable to compete for light in wetter areas. Where the two species co-occur there could be a clear difference in Δ, with the species from the drier areas (xeric species) having a higher Δ than those from the wetter area (mesic species).

This pattern of higher Δ in plants from dry environments has been found in common garden experiments comparing a range of closely related species and ecotypes (Hubick & Gibson 1993; Lauteri et al. 1997; Read & Farquhar 1991; Zhang & Marshall 1995; Zhang, Marshall & Jaquish 1993). However, the opposite pattern, with species from arid areas maintaining lower Δ when grown in common gardens – reflecting a genetic selection for a large stomatal limitation on assimilation – has been found for ecotypes of a single species (Comstock & Ehleringer 1992), for Eucalyptus microtheca in dry soil (Li et al. 2000), and for several other Eucalyptus species (Anderson et al. 1996).

If each species along a gradient of decreasing rainfall exhibits declining Δ, and species with more xeric distributions exhibit less stomatal limitation on assimilation where species overlap, then Δ will have a ‘saw-toothed’ pattern (Fig. 1). The pattern of the multispecies average Δ with decreasing rainfall would then differ from the responses of the individual species. The average Δ would show less sensitivity to environmental conditions than the Δ of individual species, supporting the idea that biodiversity stabilizes land–surface gas exchange relationships.

Figure 1.

The hypothesized response of leaf (open symbols, dashed lines) and wood (closed symbols, solid lines) Δ of three sequentially distributed species along a gradient of decreasing rainfall. Lines connecting symbols give the ranges of expected Δ for each species; bold lines give the growth-form averages. It is hypothesized that there will be consistent decreases in Δ across the distribution of each species, reflecting a decline in the ability to maintain stomatal conductance with decreasing moisture availability. Leaf Δ will record plant responses during a narrow subannual optimal period when the leaves are formed (here a 2‰ decrease across the distribution of each species). Wood Δ values, integrating many annual signals of production-weighted Δ, will exhibit stronger responses than leaf Δ (here a 3‰ decreases across the distribution of each species). Species replacements are marked by an increased ability to acquire water, leading to an increased ability to maintain conductance relative to photosynthetic capacity. The growth-form average response to decreasing rainfall will mask the responses of individual species, and will show little trend across the transect as species with higher Δ replace those with lower values until no further replacement species are available.

In our earlier study in the Northern Territory (NT) along a rainfall gradient from 1800 to 200 mm year−1 the average leaf Δ of the sampled overstorey species exhibited little change until rainfall dropped below 475 mm year−1, when Δ declined with decreasing rainfall (Schulze et al. 1998). This contrasted with the pattern along a rainfall gradient in south-east Queensland, Australia, where there was a linear decline (with a slope of 3·3‰ m−1 long-term average total annual rainfall) of a community-averaged leaf Δ with decreasing rainfall between 1700 and 350 mm year−1 (Stewart et al. 1995). We proposed that the lack of response of the average Δ of overstorey tree species to decreasing rainfall was due to the pattern of species replacements outlined above, and to the failure of the trees to survive under the most arid conditions.

As Δ in tissues records the 13CO2 discrimination that occurred during the original assimilation events, leaf Δ and wood Δ may record different subannual periods. Leaf Δ reflects conditions during a short period when they are formed, assuming they are constructed exclusively of current photosynthate. Eucalypts of the NT typically have leaf longevities of 1 year or less, and in some species leaf flushing occurs just before the wet season when vapour pressure deficits drop and predawn water potentials are high (Myers et al. 1997; Myers et al. 1998; Williams et al. 1997). These conditions could cause little stomatal closure, and therefore little pattern in leaf Δ across the distribution of a species. On the other hand wood, assumed to be produced throughout the year, may show a stronger response than leaves, as it would include a production-weighted isotopic signal that includes the 4-month dry season when almost no rain occurs in Australian savannas.

The aim of this research was therefore to see if the environmental limits on the distribution patterns of a sequence of dominant Eucalyptus species could be inferred from patterns of Δ. We expected that a key element defining the pattern of species distributions would be the maintenance of conductance relative to photosynthesis during water stress. We hypothesized that Δ would decrease with decreasing total annual rainfall across each species’ distribution. Species from more arid areas were expected to have higher Δ where the species overlapped with those from wetter areas. Interspecific differences in Δ at each point along the rainfall gradient were expected to contribute to an emergent behaviour of the multispecies average that was different from that of each species. We expected a non-linear relationship between Δ and total annual rainfall. Δ would change little per unit decrease in rainfall at wetter sites, but would change more with the same decrease in rainfall at drier sites because we expected Δ to be influenced more by gc than by photosynthetic capacity. We tested this by measuring Δ in leaves and wood from the dominant and subdominant eucalypts across a subcontinental rainfall gradient. Leaf Δ was examined to see if it reflected an optimal, subannual period, while wood Δ was measured to see if it reflected a stronger stress signal encompassing growth over the full season.

Materials and methods

We measured Δ along the Northern Australian Tropical Transect, one of a series of the continental-scale transects of the International Geosphere-Biosphere Programme (IGBP) core project ‘Global Change and Terrestrial Ecosystems’ (Koch et al. 1995), established to examine the responses of ecosystems to global environmental change. The transect runs from the Darwin region on the north coast of the NT (≈12.5 °S) to the Tanamai Desert region at 18–19 °S. Across the transect, and further south to the southern border of the NT (≈25 °S) there is an eightfold decrease in total annual rainfall from 1600 to 200 mm year−1 (Fig. 2a). Around 80% of the rain occurs during December to March (the wet season) in the northern half of the transect. As latitude increases, the rainfall becomes less seasonal and more variable, and at the southern edge of the transect only 50% of the rainfall occurs during the wet season (Cook & Heergeden 1997; Heerdegen & Cook 1999). The annual average mid-afternoon saturation vapour pressure deficits (SVPD) increased during the wet season from 12 °S latitude to around 20 °S (where rainfall is ≈400 mm year−1), but were relatively uniform further south into the drier areas (Fig. 2b). The sampling protocol attempted to space sample collections evenly along the Stuart Highway (Fig. 2c). The collection intervals in terms of changes in rainfall between zones were therefore greater in the north than in the south.

Figure 2.

(a) Predicted total annual rainfall in the 33 zones where samples were collected (anuclim 1·8, Centre for Resource and Environmental Studies, Australian National University). (b) Calculated annual average mid-afternoon vapour pressure deficits (●), and during the wet season (○). (c) Sequence of species collected along the transect. ○, zones within the range of a species where it could not be found. Spacing between sample locations varies because not all species co-occurred as a mixed forest within each zone.

Dominant Eucalyptus species with mapped distributions (Brooker & Kleinig 1994) were selected a priori to include those restricted to the wetter, northern end of the transect (E. miniata A. Cunn. Ex. Schauer; E. tetrodonta F. Muell.; and E. tectifica F. Muell.), the middle of the transect (E. chlorophylla Brooker & Done; E. confertiflora F. Muell.; E. pruinosa Schauer; E. coolabah Blakely & Jacobs ssp. coolabah;E. leucophloia Brooker ssp. QQ; and E. dichromophloia F. Muell.), and the drier southern end of the transect (E. terminalis F. Muell.; E. odontocarpa F. Muell.; E. pachyphylla F. Muell.; and E. gamophylla F. Muell.). Species with restricted distributions (possibly linked to specific soil types) were avoided, although E. tectifica, E. chlorophylla and E. odontocarpa have relatively narrow distributions. Eucalyptus pruinosa, E. leucophloia and E. dichromophloia have wide distributions, and E. terminalis was selected because it has an extremely wide distribution. Most of the species are evergreen, although E. confertiflora is completely dry-season deciduous. Adult E. confertiflora and E. pruinosa retain the juvenile leaf-form, while the other species switch to the adult leaf-form at maturity. Eucalyptus. odontocarpa, E. pachyphylla and E. gamophylla are multistemmed mallee shrubs ≈4 m tall, while the other species are single-stemmed trees.

We sampled along 1750 km of the Stuart Highway between Darwin and the southern border of the NT. The distance was divided into 33 zones (each ≈50 km long). The species that occurred in each zone were sampled once per zone at locations where they were locally abundant. Not all species within a zone occurred as a mixed stand at the same location, so samples were collected at 97 locations. Most of the locations were flat, and we sampled on loams and sandy soils as the species of interest were not common on clay soils (Williams et al. 1996). There was no stratification in sampling for soil type. Sampling was conducted at locations accessible from the highway at the end of the dry season between 16 September and 18 October 1996.

Sun-exposed leaves from the upper part of the canopy and wood samples were collected from five individuals of each species per zone. The material from each individual was analysed separately to record variability within the population. Ten leaves from one branch were bulked for each individual tree. Wood material was collected as shavings from a hand drill from two holes per individual, or as a basal disk for the smallest trees and the mallee growth forms. The drill penetrated ≈100 mm after the bark had been removed. The exact number of years each wood sample represented could not be determined. In the wetter areas, where the annual growth increments are larger, fewer years are represented than in the drier areas.

Samples were air-dried in the field and redried to a constant weight at 80 °C in the laboratory, and were finely ground. Sample 13C/12C was measured by an Isochrom (Micromass, Middlewich, UK) continuous-flow mass spectrometer after combustion in a Carlo Erba (Milan, Italy) elemental analyser. 13C/12C ratios were converted to discriminations (Δ in ‰) by assuming a 13C composition for air of −7·8‰ relative to PDB (Farquhar et al. 1982). Repeated measures of the reference and selected Eucalyptus samples usually had standard deviations of ≈0·1‰.

From the latitudes and longitudes recorded in the field with a hand-held global positioning system and the elevations predicted from a digital elevation map, mean climatological parameters were predicted for the 97 sampling locations using anuclim 1·8 (Centre for Resource and Environmental Studies, Australian National University, Canberra [ANU]). Linear regressions of leaf and wood Δ on total annual rainfall were calculated for the full data set and plotted with the zone mean and standard errors (SAS 6·12, SAS Institute Inc.; Microcal Origin 5·0). The differences between species within a zone, and overall, were tested with a one-way anova followed by a Student–Newman–Kuels multiple range test (proc glm, SAS).


The response of both leaf and wood Δ to declining rainfall varied among the 13 species (Fig. 3). Leaf Δ decreased (P < 0·05) in five species (E. miniata, E. pruinosa, E. leucophloia, E. dichromophloia and E. terminalis), showed no response in seven (E. tetrodonta, E. tectifica, E. confertiflora, E. coolabah, E. odontocarpa, E. pachyphylla and E. gamophylla), and increased in one (E. chlorophylla). Wood Δ decreased in eight species (E. miniata, E. tetrodonta, E. pruinosa, E. leucophloia, E. coolabah, E. dichromophloia, E. terminalis and E. odontocarpa), showed no response in four (E. tectifica, E. chlorophylla, E. confertiflora and E. gamophylla), and increased in one (E. pachyphylla). The changes in leaf Δ with decreasing rainfall varied between +1‰ and −3‰, and the change in wood Δ between +0·5‰ and −2·5‰. The species with the greatest ability to withstand increases in the stomatal limitation on photosynthesis, E. dichromophloia and E. terminalis (i.e. those with the most negative changes in Δ) had the widest distributions. The two species that had positive changes in Δ with decreasing rainfall (E. chlorophylla and E. pachyphylla) had narrow distributions. The direction and magnitude of the response of both leaf and wood Δ did not relate to the location of the species along the transect.

Figure 3.

Responses of individual species leaf and wood Δ (x  ± SE; n = 5) to decreasing total annual rainfall. Significant trends (α = 0·05) in leaf Δ, dashed lines; in wood Δ, solid lines.

Eucalyptus miniata and E. tetrodonta are the co-dominant trees defining many of the vegetation types north of ≈15 °S along this transect, and E. miniata showed patterns of Δ that indicated a greater sensitivity than E. tetrodonta to the decreasing rainfall from north to south. The only completely dry-season deciduous species sampled, E. confertiflora, showed no trend in Δ with decreasing rainfall across its distribution. The ranges in leaf Δ of the tree species overlapped, and there were no clear break points separating significantly different subgroups. Leaf Δ in the three mallee shrubs was smaller than for all the tree species. The species rank order of wood Δ was not the same as for leaf Δ, but the range across each species’ distribution meant most of these differences were not significant.

Mean leaf Δ was between 0·9 and 1·9‰ greater than wood Δ, except for E. pachyphylla where the leaves averaged only 0·1‰ higher and were not significantly different (P < 0·05). In E. pruinosa and E. pachyphylla, the difference between leaf and wood Δ decreased with decreasing rainfall, and in E. gamophylla it increased, while the other species showed constant differences between leaf and wood Δ across their distributions.

With decreasing rainfall, the leaf Δ of new species (with a more xeric distribution than those already being sampled) was similar to the top-ranked species within the zone in one case, similar to the middle-ranked species in five cases, and similar to the bottom-ranked species in eight cases (Table 1). The wood Δ of the new species was similar to the top-ranked species in three cases, the middle species in two cases, and the bottom species in five cases. At the xeric edge of the distributions of the species, the leaf Δ of the departing species was similar to the highest ranked species in four cases, the middle-ranked species in four cases and the lowest ranked species in four cases. The wood Δ of the departing species was similar to the top ranked species in five cases, the middle ranked species in three cases, and the lowest ranked species in four cases. Where species were sampled at exactly the same location (as opposed to within the 50 km zone) there were 28 possible pairwise comparisons of mesic and xeric species. In seven cases the xeric species had significantly higher leaf Δ, and in three cases significantly higher wood Δ. In four cases the mesic species had higher leaf Δ, and in six cases higher wood Δ. In the remaining contrasts the values were not significantly different.

Table 1.  Rank order of species mean leaf and wood Δ within 33 zones (Δ decreases down the column)
  1. Species: (A) E. miniata; (B) E. tetrodonta; (C) E. tectifica; (D) E. chlorophylla; (E) E. confertiflora; (F) E. pruinosa; (G) E. coolabah; (H) E. leucophloia; (I) E. dichromophloia; (J) E. terminalis; (K) E. odontocarpa; (L) E. pachyphylla; (M) E. gamophylla. Groups of species within a zone (within each column) with the same superscript were not significantly different (SNK test, α = 0·05). The first occurrence of a species when moving from the wetter north (zone 1) to the drier south (zone 33) is highlighted with a bold letter; the last occurrence of a species is in lower case.

 B1B2C2C1I2A1E12I12A1F1D2I12G1e1F2J12F1I1g1J12J1h12L1L1L2I2 J1  M2M1 
     B3E2I12C12c2B1b23F12E12 G2H2G1G2H23H34H1L2           
       B2E2D2A1a23H12I2   H1H2K3K4             
         E2E2I23E23J2    K2F3              
 B2B2C2C1E1A2E1E2c12D1F1F1F12I1G1H2G1G1F12H2H1h12J2J1J1J2 M1  M2j1 
     B1E2A1B2A12I1b12I1E12 I1I3H2J1g3K23K1K2           
       B2I2E12E1E12H1J12   I2K1H3L3             
         I2B1I23J1I2    H1K4              

Our proposed saw-toothed pattern of the individual species responses (Fig. 1) was not consistently evident (Fig. 4). The non-linear response of the multispecies mean Δ to decreasing total annual rainfall reflected the spatial differences in sensitivity of the response of Δ to decreasing rainfall of individual species (Fig. 5). Species at the wetter, northern end of the transect exhibited less change in both leaf and wood Δ per unit decrease in rainfall than those in the drier south. Although the slopes of the responses of both measures of Δ to decreasing rainfall became more negative with decreasing rainfall, two of the three positive slopes occurred in E. pachyphylla and E. gamophylla, which were restricted to the driest areas of the transect.

Figure 4.

Relationship between leaf and wood Δ values predicted from total annual rainfall for each species. End points of lines give the range of rainfalls where the species was sampled, as shown in Fig. 1.

Figure 5.

Response of the zone mean (x  ± SE) leaf and wood Δ values to decreasing total annual rainfall. Regression lines from the full set of values (n = 645) are: leaf Δ = 14·29 + 18·15 × PPTm − 17·74 × PPTm2 + 5·76 × PPTm3 (r2 = 0·45); wood Δ = 13·26 + 16·88 × PPTm − 16·28 × PPTm2 + 5·24 × PPTm3 (r2 = 0·59) where rainfall is expressed in m (PPTm). ○, Leaf Δ; ●, wood Δ.

Across the transect, the responses of leaf and wood Δ to decreasing total annual rainfall were described by third-order polynomials. These explained 45% of the variability in the leaf Δ of the full data set (n = 645), and 59% of the variability in wood Δ. Near the mid-point of the transect, where rainfall was ≈700 mm year−1, the species restricted to the wetter end of the transect were replaced by those with intermediate distributions, and there was a distinct step decrease in the mean wood Δ. Where rainfall was <400 mm year−1 there were usually only two species present, the isolated tall tree E. terminalis and either of the multistemmed mallee shrubs E. pachyphylla or E. gamophylla. The variability in zone means reflects both the variability in Δ of E. terminalis and the differences between the two growth forms.


patterns of Δ across the distribution of species

This study is to our knowledge the first to systematically sample Δ across the distributions of replacement species along an environmental gradient. It views variation in Δ as resulting primarily from variation in CO2 supply into, rather than demand from, the intercellular air spaces within leaves, and tests a theory that explains both species and growth-form patterns. Unfortunately, using Δ as an index of stomatal limitation on assimilation is plagued by the ambiguity of the signal resulting from movements of both gc and photosynthetic capacity. Increased whole-plant carbon gain could occur by increased gc raising pi and Δ with a fixed photosynthetic capacity, or increased photosynthetic capacity with a fixed gc decreasing pi and Δ. The ratio of leaf area to root area will determine the water supply per unit leaf area, which will strongly determine gc, and the division of root allocation for water versus nutrient uptake will strongly determine leaf N contents and photosynthetic capacity. Across a spatial gradient in water availability, morphological plasticity could maintain consistent physiological relationships in leaves, or decreasing water availability could be expected to decrease gc per unit photosynthetic capacity and decrease Δ. Increased photosynthetic capacity per unit gc, which would also result in patterns of decreasing Δ, would indicate an adaptation to increase water-use efficiency with decreasing water availability.

The response of leaf and wood Δ to decreasing total annual rainfall varied for each species. Neither the total change in Δ from the wet to the dry edge of each species’ distribution, nor the sensitivity of Δ to changes in rainfall, reflected common physiological tolerances. The species ranking was therefore not conserved across their distributions, unlike in other species assemblages under varying environmental conditions (Ehleringer 1993; Sandquist et al. 1993).

patterns of Δ with species replacements

With decreasing rainfall, the new xeric species encountered did not consistently have higher Δ (indicating greater ability to maintain gc relative to photosynthetic capacity) compared to the mesic species already present. The replacing species generally had lower Δ than those already present on the transect, and then maintained even lower long-term Δ as they extended into drier areas. An experiment in a well watered common garden found a genetically determined low Δ in Eucalyptus species from dry areas (Anderson et al. 1996). Correlations between Δ and rainfall at the sites of origin have also been found for populations of a single species in a water-limited common garden (Li et al. 2000).

The response of the multispecies mean Δ to decreasing total annual rainfall reflected the patterns of responses of the individual species. The species in the wetter areas were less sensitive to decreases in rainfall than those in the dry areas. Although species such as E. miniata (found at the wetter end of the transect) exhibited a shift in Δ across its distribution similar to that of E. odontocarpa (found at the dry end of the transect), the slope of the response of E. miniata was much less than that for E. odontocarpa. As the proposed ‘saw-tooth’ pattern of species was not observed, the multispecies mean did not describe emergent behaviour at a higher level of organization that could not be predicted from the behaviour of the individual components of the system (O’Neill et al. 1986). The lack of an emergent behaviour means this study does not present evidence that tree species diversity is important in stabilizing ecosystem fluxes of compounds such as CO2, H2O and N in response to factors such as water availability (Schulze & Mooney 1993; Schulze et al. 1998).

patterns of Δ and mean annual rainfall

The leaf and wood Δ of the dominant eucalypts in northern Australia decreased non-linearly in response to decreasing total annual rainfall. We found a stronger decline in Δ where rainfall was between 1600 and 400 mm year−1 (≈1·5′) than was seen by Schulze et al. (1998), who found no change, and both studies found a threshold below which Δ declined sharply. The pattern of Δ in the NT overstorey trees differs from the linear decline in community-averaged Δ found with decreasing rainfall in south-east Queensland (Stewart et al. 1995). Several other studies have found non-linear responses of Δ across rainfall gradients when the species sampled were from a single genus or growth form. Lajtha & Getz (1993) found that the decrease in Δ of two New Mexican conifers occurred only at the drier end of their environmental gradient. Korol et al. (1999) found little change in Δ of Pinus radiata at sites with rainfalls >750 mm year−1, but Δ decreased as rainfall approached 500 mm year−1. Although the community averages of Stewart et al. (1995) may have included a shifting ratio of growth forms with inherently different Δ (Brooks et al. 1997; Pate et al. 1998), it is unlikely that the changing ratio of growth forms would have exactly offset non-linear responses within growth forms.

The monsoonal climate in north-western Australia has highly seasonal rainfall, but the occurrence of the rainfall is regular from year to year. In south-east Queensland, by comparison, the rainfall is more evenly distributed throughout the year. In areas with highly seasonal environmental conditions, indices of water availability other than annual totals are probably necessary (Walker & Langridge 1997). Patterns of evaporative demand (estimated by long-term averages of SVPD), which should influence patterns of Δ (Lloyd & Farquhar 1994), did not appear to be important in determining the patterns of Δ across the NT. Mid-afternoon SVPD, and SVPD weighted by moisture availability (Comstock & Ehleringer 1992), did not correlate with Δ (data not shown).

The reliability of the monsoon and the concentration of the total annual rainfall into a 4-month period have probably been features of northern Australian ecosystems for the past 10 million years (Pole & Bowman 1996). The predictable summer rainfall belt, where there is virtually no winter rain, extends south to 19–20 °S where rainfall is ≈450 mm year−1. However, at ≈16 °S the rainfall regime changes from one where most of the rain comes via the direct influence of the monsoon, to one where most of the rain comes primarily as smaller, isolated convective storms (Cook & Heergeden 1997; Heerdegen & Cook 1999). Major floristic boundaries occur at ≈16 °S and at ≈19 °S (Egan & Williams 1996; Williams et al. 1995). The 16 °S floristic boundary was approximately where the species typical of the wet end of the transect were replaced by those typical of the mid-section of the gradient. Around 16 °S hummock grasses, a feature of the Australian arid environment, begin to co-occur with tussock grasses (Cook & Heergeden 1997; Heerdegen & Cook 1999). The 19 °S winter rainfall boundary marked the transition from species characteristic of the mesic savannas to those characteristic of the xeric savannas.

Because of the intensity of the monsoon-driven rainfall north of ≈16 °S, the soils over much of the transect may become saturated during the wet season, and the rainfall gradient largely contributes to variations in stream flow rather than in soil storage. The water storage in the subsoil layers may provide an adequate water supply over much of the transect for the deep-rooted trees. Although this would explain the lack of variation in Δ at higher rainfalls, Williams et al. (1996) found that tree basal area, height and cover all increased with total annual rainfalls from 400 to 1800 mm year−1, showing that water availability was still determining community structure. At the wetter sites, trees could be growing to heights where internal hydraulic conductivity determines leaf water status (Ryan & Yoder 1997). If this were so, then Δ would not be linked to soil water availability, but this hydraulic limit to tree growth may not be universal (Becker, Meinzer & Wullschleger 2000), especially in angiosperms (Becker, Tyree & Tsuda 1999; Leavitt & Newberry 1992).

Sap-flow and eddy-flux measurements of canopy gas exchange in the wet and dry seasons at three sites along this transect (with total annual rainfalls of 1873, 870 and 520 mm year−1) have shown that the transpiration rates per unit ground area of the overstorey canopy trees are remarkably consistent along the transect between wet and dry seasons (Hutley, O’Grady & Eamus 2001; O’Grady, Eamus & Hutley 1999). Partial leaf shedding during the dry season increased leaf area-based transpiration rates during the dry season, and this rate did not vary consistently with rainfall across the gradient, matching patterns reported for Eucalyptus species along other gradients (Hatton et al. 1998). Overstorey leaf-area index (LAI) dropped from ≈0·9 in the wet season at the wettest site to 0·05 at the driest site, and decreased to ≈60% of the wet season LAI during the dry season (Hutley et al. 2001).

Changes in LAI may ameliorate the impact of decreased rainfall on stomatal conductance. The lack of pattern in Δ across many of the species’ distributions indicates that the trees either shed leaves before the onset of water stress, or shed them at similar intensities of stress. A decline in leaf life-span with decreasing water availability, rather than the length of time with partially open stomata, may be the primary indicator of plant performance and distributions. The infertility of Australian soils may limit tree LAI to 1–1·5 in the wet season, and there appears to be little difference in the leaf-level profit margins (the carbon return per unit invested) between fully deciduous and evergreen growth forms (Eamus et al. 1999). Intra-annual reduction in canopy leaf area, which also occurs in evergreen species (Williams et al. 1997), would increase the water supply to the remaining leaves and maintain high Δ. As the strength of the Δ signal also includes the amount of assimilation occurring, no signal will be recorded when water stress completely closes the stomata. Species from arid areas may continue assimilation with partially open stomata (recording low Δ), while those from the wetter areas close their stomata (recording no Δ signal). The completely dry-season deciduous E. confertiflora exhibited little change in Δ across its distribution in response to decreasing rainfall.

leaf Δ versus wood Δ

Leaf Δ was higher than wood Δ in all but one species. The differences between leaf and wood did not change across the range of most species. The hypothesis that leaf Δ might show little response to decreasing total annual rainfall (because the leaves were grown during a subannual period with little water limitation), while the wood Δ would show a stronger response (reflecting the full annual realized production and stress), was not supported. The parallel response of leaf and wood Δ to decreasing total annual rainfall could be due to leaf and wood growth occurring primarily at the same time during the year (Prior, Eamus & Duff 1997; Williams et al. 1997), or due to leaf production relying primarily on reserves accumulated throughout the year. The variation among species in this leaf to wood Δ difference probably reflects differences in tissue chemistry (Park & Epstein 1961), and may not exist in comparable cellulose fractions.

Wood Δ was less variable than leaf Δ within and between zones. Wood Δ is therefore a better measure of plant responses over large distances where long-term integrated signals are important. The variability in leaf samples between adjacent zones could reflect details of the environmental conditions for the specific year of sampling. As all leaves from a single tree were collected from a single branch, within-canopy differences (Waring & Silvester 1994) were not measured except as differences in leaf Δ between trees within a zone.


The proposed mechanism that would stabilize the multispecies mean Δ across a gradient of decreasing rainfall (Fig. 1) was not found. Species Δ did not decrease consistently with decreasing rainfall. Replacement species did not show an increased ability to maintain stomatal conductance relative to photosynthetic capacity when compared to species at the arid edge of their distributions. Instead, Δ of xeric species was often less than in co-occurring mesic species. As there were no consistent patterns of Δ across the species’ distributions, the relative Δ values of two co-occurring species (or across the distribution of a single species) do not provide a simple measure of plant performance. Other impacts on the total carbon budget of adult trees along this gradient, such as the changes in LAI in space and the variation in leaf life spans, appear to be more important than partial stomatal closure in determining species distributions. The nature of the selection of plant traits in response to decreasing water availability, and the impact of these characters on Δ in leaves or wood, are still not completely resolved and will require measurements that distinguish changes in photosynthetic capacity from those of stomatal conductance. A better understanding of the actual spatial and temporal soil-water availability should reconcile the differences found in Δ between the Northern Territory and Queensland.


We would like to thank Jack Cusack for his invaluable assistance in the field, as well as Lins Vellens and Margi Böhm. E.-D. Schulze and Lindsay Hutley provided clear, insightful comments on early drafts of the manuscript.

Received 23 March 2000; accepted 23 October 2000