δ13C of wood in growth-rings indicates cambial activity of drought-stressed trees of Eucalyptus globulus

Authors


Abstract

1. Growth, density and δ13C of wood and leaf area were measured in two adjacent stands of 6 year-old Eucalyptus globulus growing in the 600–700 mm year–1 rainfall region of south-western Australia. Study sites were identical except for differences in the availability of water owing to physical properties of soil profiles and location of sites within the landscape.

2. Abundance of 13C (expressed as δ13C) in wood of trees growing on the drought-prone site (– 24·8‰±1·4) was greater than in other trees (– 25·8‰±1·2, P<0·001) throughout the 6 years and, with further development, the δ13C signatures of wood may become useful indices of drought-susceptibility in plantations within a few years of establishment. The seasonal pattern of δ13C of wood appeared to reflect seasonal variation in water availability and duration of cambial activity.

3. Basic density of wood of trees growing on the more drought-prone site (496±14·0 kg m–3) was reduced compared to other trees (554±5·3 kg m–3, P<0·001). δ13C of wood across boundaries of growth-rings suggested that drought stopped cambial activity resulting in less production of late wood and less dense wood.

4. The stand growing on the drought-prone site had reduced growth, wood yield and leaf area but identical specific leaf area. Annual growth was correlated with the previous season’s rainfall. Together, these results suggested that within the same evaporative climate, drought reduces growth primarily by reducing leaf area and that there is a lag between onset of drought and reduced productivity.

Introduction

The ratio of 13C:12C in wood (expressed as δ13C) from as few as four co-dominant trees of the same species can provide a retrospective record of drought within a stand (Leavitt 1993; Högberg et al. 1995). For example, the patterns of variation in δ13C of wood from stands of Pinus strobus and Acer saccharinum were consistent within years and within species (Leavitt 1993) but increased by 1‰ in years of drought. There is a significant, inverse relationship between annual productivity and the δ13C signature of wood and leaves for a number of species (Leavitt & Long 1988; Dupouey et al. 1993; McNulty & Swank 1995; Saurer, Siegenthaler & Schweingruber 1995), indicative of the strong general relationship between tree growth and availability of water, although δ13C may be a better indicator of inter- and intra-annual variability in water availability than annual productivity (Dupouey et al. 1993; Leavitt 1993; Livingston & Spittlehouse 1996; Walcroft et al. 1997).

While the development of the vascular elements is generally regarded as sensitive to the availability of water (Kozlowski 1982), δ13C of wood within growth-rings can vary up to 2–3‰ between trees of the same species at a site owing to different microenvironments, especially in closed-canopy sites (Leavitt & Long 1988). δ13C of wood may also vary with stem height owing to self-shading (Leavitt & Long 1986), because unshaded leaves fix CO2 more quickly than shaded leaves (Francey & Farquhar 1982), as well as to refixation by lower branches of CO2 respired from soil (Vogel 1978; Medina & Minchin 1980).

Furthermore, the effects of availability of water on tissue development will affect wood density (Kozlowski, Kramer & Pallardy 1991) and we might expect that the δ13C of wood could be related to its density. For example, in Douglas Fir from different climatic regions, δ13C of leaves was correlated with specific conductivity of branches (Panek 1996) which in turn is related to wood density via the thickness of walls and diameter of vascular elements (Tyree & Ewers 1991).

In the long term, shortages of water reduce the area and rate of photosynthesis of leaves, resulting in reduced productivity (Kozlowski 1982; Specht & Specht 1989; Gholz, Ewel & Teskey 1990). Specht & Specht (1989) have suggested that within the same evaporative climate, reduced leaf area may be more important to productivity than reduced leaf specific area (an indicator of resistance of leaves to evaporation). The δ13C of wood may help to differentiate between effects on productivity of changes in leaf area and anatomy and physiology because of its close relationship to stomatal conductance and the rate of photosynthesis (Farquhar et al. 1988).

In south-western Australia, poor availability of water to trees has long been associated with specific landforms and soils (e.g. McGrath et al. 1991; Dutkowski 1995). The landscape consists of a gently undulating surface of low relief (50–100 m) with numerous minor valleys dissecting the laterized granitic and gneissic parent materials (Churchward & Dimmock 1989). The soil pattern is complex with rapid lateral changes in soil attributes (McArthur 1991). Previous studies in the Blackwood catchment (site of the present study) highlighted the importance of soil type and position in the landscape to drought susceptibility and death of Pinus radiata (McGrath et al. 1991). Similarly, Adams (1996) emphasized the importance of landscape and soil factors to the natural distribution of tree species in Australia, including south-western Australia.

The Tasmanian Bluegum (Eucalyptus globulus Labill.) is grown in parts of south-western Australia with a Mediterranean climate with an annual rainfall of 600–900 mm (Bartle 1991). This region is subject to regular summer drought and occasional winter drought (Gentilli 1989). The climate and landscape of south-western Australia provide an opportunity to study the effect of drought stress on the production of wood and foliage, wood density and carbon isotope discrimination of wood in stands growing with the same rainfall and in the same evaporative climate, but with differing availability of water.

We aimed to test whether (1) δ13C of wood can be used as an indicator of drought in E. globulus plantations, (2) density and δ13C of wood vary between trees within stands, (3) density and δ13C of wood increase with increasing stem height and (4) δ13C of wood is correlated with wood density.

Materials and methods

STAND DESCRIPTIONS AND EXPERIMENTAL DESIGN

Experimental sites were located in the ‘Wheatley’ plantation of E. globulus operated by Bunning’s Treefarms Pty Ltd. The plantation is near Hilton (33·99°S, 116·32°E) 15km east of Bridgetown, Western Australia, and was established at a stocking rate of 1250 trees ha–1 on ex-pasture land in autumn–winter 1990. Average annual rainfall for Hilton from 1989 to 1996 was 640 mm year–1. The soils of the plantation belong to the Yalanbee and Pindalup associations which comprise mainly deep, sandy gravels over duricrust on crests and upper slopes, and sandy duplex soils on mid-slopes (Churchward & Dimmock 1989). Previous work has shown that marked differences in water availability within the plantation (which, at the extremes, had killed some trees) were the result of differences in soil type and position in the landscape (Dutkowski 1995; C. Shedley, Bunning’s Treefarms, personal communication).

We selected study plots on the basis of (1) inferred leaf area based on aerial photographs, (2) inferred rates of growth based on inventory records and (3) inferred rates of water infiltration and retention based on soil features (as determined by excavation and description of the soil profiles to a depth of 2 m), location in the landscape and knowledge from other studies (e.g. McGrath et al. 1991; Dutkowski 1995). Three plots (of 0·02 ha) were selected within ‘Site 1’ (crest/upper slope, greater water retention, growth and leaf area) and three were selected within ‘Site 2’ (mid-slope, lesser water retention, growth and leaf area).

TREE SAMPLING

Diameter at breast height (1·3 m) over bark (dbhob) of trees in all three sample plots at both sites was measured in April 1996. A stratified random sample based on dbhob of five trees from across the three plots at each site was felled on 15–16 May 1996. Felled trees were measured for total height and stem length (to a diameter of 40 mm). The stem was stripped of branches and 5 cm thick discs of stem-wood were sampled from between five and 12 locations depending on length and form of the stem, including a disc from the base (20 cm above ground level). Diameter under and over bark of each disc was measured and the over-bark and under-bark volumes calculated using Smalian’s formula (Husch, Miller & Beers 1982).

BASIC DENSITY AND MASS OF STEM WOOD AND BARK

Radial sections of wood comprising one quarter to one half of each disc and all of the bark from each disc were used to determine basic density as follows:

basic density

inline image

There was little variation in bark density within or between trees and the mass of bark was calculated from the mean density of bark for the whole tree. The mass of each section of stem wood was calculated from the volume of each section and the area-weighted mean basic density of the discs above and below it as follows:

mass of stem section

= volume of stem section

inline image

where density1, density2, area1 and area2 are the basic density and cross-sectional area of the discs above and below each stem section.

MASS OF TREE AND BRANCH COMPONENTS AND LEAF AREA

For each sample tree, the diameter of all live branches was measured 20 mm from where branches joined the stem and a stratified random sample of eight or nine branches > 5 mm diameter per tree was selected. The branches were stripped of leaves and the branch wood divided into that > 5 mm diameter and that < 5 mm diameter. Branch-wood components were oven dried at 65 °C to constant mass. Leaves were weighed fresh and subsampled (200–250 g) to determine leaf area and leaf mass.

Allometric relationships of the form: ln y = a + b ln x, where y is the mass or area of the branch component and x is the diameter of the branch, were used to determine the mass of branch components and leaf area for each sample tree. Separate regressions were developed for large (diameter > 18 mm) and small (diameter < 18 mm) branches to allow for the tendency of allometric regressions to over- or underestimate the dependent variable at large values of the independent variable (Whittaker & Marks 1975).

From the component totals for each tree, allometric relationships for all tree components and leaf area [using tree diameter (dbhob) as the independent variable], were used to calculate the total mass and area of leaves, the mass of branch wood, the mass and volume of stem wood and bark, and the total above-ground biomass for the three plots within each site from the dbhobs. Mean leaf area index (LAI) for each site was the mean of LAI of the three plots at each site which was calculated as follows:

leaf area index (LAI)

inline image

Mean specific area of leaves for each site was the mean of that for the sample trees from each site which was calculated as follows:

mean leaf-specific area (LSA)

inline image

Branch and tree regressions were corrected for proportional bias using Snowden’s (1991) ratio estimator for bias correction and tested for homogeneity of slope and intercept between stands using ANCOVA within the general linear model (Minitab Inc).

ANNUAL INCREMENTS OF WOOD GROWTH

Discs of wood were air dried and sanded to identify boundaries between growth-rings. The width of growth-rings was measured in four radial directions and the volume of each growth-ring in each tree calculated using Smalian’s formula (Husch et al. 1982). Basic density of growth-rings was determined and used to calculate the mass of annual increments of growth as previously described. Allometric regressions for volume and mass of each growth-ring against tree diameter (dbhob) were derived (similarly as for branch and tree regressions) and used to calculate annual increments of growth for the three plots within each site.

TREE SAMPLING FOR δ13C OF WOOD

Mean δ13C of wood from the growth-rings of basal discs of all 10 trees was analysed in order to compare between and within stands and between years. The smallest tree at each site was later removed from between-site and between-year comparisons owing to its different annual pattern of δ13C (Fig. 2). Mean δ13C of wood from the growth-rings of all the discs of the tallest tree from each site were analysed in order to examine variation with increasing height, as the tallest trees included both fully shaded and fully exposed parts of the canopy.

Figure 2.

.δ13C of wood in growth-rings at the base (20 cm) of 6 year-old E. globulus from Site 1 (a) and Site 2 (b) near Hilton, WA. Tree 1 is the smallest and tree 5 the largest tree at each site.

The seasonal variation of δ13C in wood was determined in one co-dominant tree from each site which had similar dbhob (13·0–13·6 cm) and, hence, similar growth rates. It was assumed that these trees were representative of the trees at each site based on the mean δ13C of wood across growth-rings (Fig. 2a,b). Seasonal variation of δ13C in wood was also determined in the basal discs of the smallest tree from each site for comparison to the representative, co-dominant tree from each site.

ANALYSIS OF δ13C OF WOOD

For determination of mean δ13C in growth-rings, a radial–tangential section was taken from each disc and a sample of wood from the whole growth-ring was oven-dried at 65 °C and then ground into a fine, homogeneous powder of which 2·0–2·5 mg per sample was analysed. Individual growth-rings from the wood section were then separated and their basic density determined as previously described.

For determination of seasonal variation of δ13C across growth-rings, a 1-mm-thick transverse section was cut from the radial–tangential section and the position of boundaries between growth-rings recorded. The section was divided into 1-mm-thick pieces (2·0–2·5 mg) from the pith to the cambium. Either the whole section or growth-ring was sampled or else sampling was intensive (every 1 mm) near the boundaries of growth-rings and less intensive (every 2–3 mm) within growth-rings.

δ13C was analysed in whole-wood samples. Nuclear Magnetic Reasonance analysis showed no difference in cellulose or lignin content of early wood and late wood (M. A. Adams & C. Macfarlane, unpublished data). On this basis and that of (1) the literature on the isotopic composition of different chemical constituents (including lignin, starch, cellulose and hemi-cellulose) of wood and other tissues (Park & Epstein 1961; Wilson & Grinstead 1977; Leavitt & Long 1986; Benner et al. 1987; Brugnoli et al. 1988; Schleser 1990, 1992; Gleixner et al. 1993; Livingston & Spittlehouse 1996; Walcroft et al. 1997), (2) the relatively low lignin content of E. globulus wood (21–23%, Ferrari & Mughini 1995) and (3) relatively low starch content of E. globulus wood (e.g. 2–3%, Tomkins, Kellas & Squire 1989), we expected that the δ13C of whole wood of E. globulus would be c. 1‰ more negative than that of cellulose and more likely to reflect climatic conditions than its chemical composition (which may be related, Schleser 1990).

δ13C in wood samples was analysed using an automated coupling of an ANCA SL gas chromatograph with a SIRA 9 mass spectrometer (Europa Scientific, Crewe, UK). The combusted samples were separated in the GC and bled directly into the MS. The system was calibrated using 2·0–2·5 mg of corn-flour standardized against PeeDee Belemnite (PDB) as a reference (Craig 1957). Abundance of 13C in wood was expressed in δ13C (thousandths, ‰) and was calculated as follows:

inline image

(Farquhar, O’Leary & Berry 1982).

Less negative δ13C indicates greater drought stress (Leavitt 1993).

Results

BIOMASS AND LEAF AREA

The mass of all above-ground components and the mean annual increment of stem wood of the three plots at Site 1 were nearly twice those at Site 2 (P < 0·005, Table 1). The mean stocking rate at Site 2 was 80% that of Site 1 and accounted for about 40% of the reduction in biomass (Table 1).

Table 1.  . Stocking rate, basal area, mass, volume and density of tree components, mean annual increments of growth (MAI), leaf area and leaf specific area of two stands of 6 year-old Eucalyptus globulus from ‘Wheatley’ plantation near Hilton, WA. Values are mean ± SE of three plots per stand, except for wood and bark density and leaf specific area which are the mean ± SE of five sample trees per site. All site means differ significantly (P < 0·005, one-way ANOVA) except for stocking rate, bark density and leaf specific area which were not different Thumbnail image of

The mean basic density of wood from the plots at Site 1 was greater than that from Site 2 (P < 0·005, Table 1) and the LAI at Site 1 was nearly twice that at Site 2 (P < 0·01, Table 1). Bark density and leaf specific area were similar at each site (Table 1).

ANNUAL GROWTH-RING ANALYSIS

The sixth annual growth-ring was almost complete at the time of sampling as indicated by the presence of late wood (identified by colour and texture). The width of late wood in growth-rings of trees at Site 1 was generally greater than that at Site 2. Annual increments of wood volume of the plots at Site 1 were greater than those at Site 2 (P < 0·01, one-way ANOVA, Fig. 1) and were greatest in the fourth year (Fig. 1). The maximum annual rate of wood growth followed a year of high rainfall (1993, 780 mm) and the subsequent decrease followed 2 years of low rainfall (1994, 510 mm and 1995, 528 mm; Fig. 1). Wood growth after year three was poorly and negatively correlated with the current season’s rainfall but highly and positively correlated with the previous season’s rainfall (Table 2). Wood growth could not be correlated with growth before year three owing to the substantial juvenile effect on growth rate.

Figure 1.

. Annual increments of wood volume for years ending autumn 1991–96 and annual rainfall (May–April 1990–96) in two stands of 6 year-old Eucalyptus globulus near Hilton, WA. Values are mean ± SE of three plots.

Table 2.  . Pearson correlation coefficients of total annual rainfall for the same year and the previous year with annual increments of wood growth for 1993–94, 1994–95 and 1995–96 in two stands of E. globulus near Hilton, WA: n = 3 Thumbnail image of

ANNUAL VARIATION OF δ13C OF WOOD

Trees within each site displayed similar patterns of δ13C of wood with the exception of the smallest trees from each stand which were atypical (Fig. 2) and the data for these two trees were not included for statistical comparisons within sites, or between sites or years. Mean δ13C of wood was not correlated with tree diameter (dbhob) and did not differ significantly among trees within either stand (Tukey’s test). Mean δ13C of wood in trees from Site 2 (– 24·8‰± 1·4) was greater than that in trees from Site 1 (– 25·8‰± 1·2, P < 0·001, one-way ANOVA nested rings). Mean δ13C of wood in the first and last growth-rings of the Site 2 trees was greater than that in the same growth-rings of Site 1 trees (P < 0·005, one-way ANOVA), and greater than that in the other growth-rings in Site 2 trees (P < 0·001, Tukey’s test).

δ13C of wood in growth-rings was negatively correlated with the previous season’s rainfall for the whole 6 years (Site 1, r2 = 0·18, P < 0·05, n = 24; Site 2, r2 = 0·29, P < 0·01, n = 24; data not presented) but there was no correlation between δ13C of wood and the current season’s rainfall.

Mean δ13C of wood in growth-rings of the two tallest trees from each site increased by 0·22‰ m–1 of height (P < 0·001, Fig. 3a). However, this trend was generally reversed in stem wood below canopy height (Fig. 3b).

Figure 3.

. (a) Correlation of δ13C of wood with stem height in growth-rings of the tallest E. globulus tree (tree 5) from two stands near Hilton, WA (n = 54). (b) δ13C of wood in the two outermost growth-rings at different stem heights of the tallest E. globulus (tree 5) from two stands near Hilton, WA. The first live branches were at heights of 3·5 m (Site 1) and 2·1 m (Site 2).

SEASONAL VARIATION OF δ13C OF WOOD

δ13C of wood in the representative trees from either stand varied little within each growth-ring except at clear (visible) boundaries between growth-rings (Fig. 4a). Generally, δ13C of wood was greater at Site 2 than at Site 1 except at boundaries where there were substantial increases in δ13C of wood from Site 1 (Fig. 4a). For the tree from Site 1, greater δ13C of wood at the boundaries of the 1992–93 and 1993–94 growth-rings followed 2 years of high rainfall (1992, 696 mm and 1993, 780 mm) while the lesser δ13C of wood at the boundaries at the end of the 1994–95 and 1995–96 growth-rings followed 2 years of lesser rainfall (1994, 510 mm and 1995, 528 mm). There was a postive but non-significant (r2 = 0·5, P = 0·119) correlation between the previous season’s rainfall and maximum δ13C of wood within each growth-ring of the tree from Site 1 but no correlation between the current season’s rainfall and maximum δ13C of wood (r2 < 0·2). δ13C of wood increased substantially in the 1994–95 and 1995–96 growth-rings of the tree from Site 2 (Fig. 4a).

Figure 4.

.δ13C of wood from a representative, co-dominant tree (a) Site 1, (dbhob = 13·0 cm; Site 2, dbhob = 13·6 cm) and the smallest tree (b) Site 1, (dbhob = 8·6 cm; Site 2, dbhob = 6·9 cm) from two stands of E. globulus measured from the pith to the cambium at the base (20 cm) of the tree. Arrows indicate autumn late wood–early wood boundaries.

The smallest trees from each stand did not display a regular seasonal pattern of δ13C of wood as did the other trees (Fig. 4b). δ13C of wood in both trees decreased during the last four growth-rings, particularly in the tree from Site 2. δ13C of wood in the smallest tree from Site 2 was generally less than that in the smallest tree from Site 1.

BASIC DENSITY OF WOOD

Mean basic density of wood in trees from Site 1 (516 kg m–3± 16·9) was greater than that in trees from Site 2 (473 kg m–3± 8·5, P < 0·05, one-way ANOVA nested rings, Fig. 5). Mean density of basal wood was 29–40 kg m–3 less than the mean for whole trees but the difference between sites was similar. There were no differences between density of individual growth-rings between (one-way ANOVA) or within stands (Tukey’s test).

Figure 5.

. Basic density of wood in growth-rings at the base (20 cm) of E. globulus from two stands near Hilton, WA. Values are mean ± SE of six rings from four trees per stand.

Basic density and δ13C of wood were not correlated (Fig. 6a) but were consistent within each site. Basic density of wood in growth-rings increased by nearly 10 kg m–3 m–1 of height (P < 0·001, Fig. 6b) irrespective of site. There was no correlation between wood density and rainfall.

Figure 6.

. (a) Relationship between basic density of wood and δ13C of wood in growth-rings at the base (20 cm) of E. globulus from two stands near Hilton, WA (n = 46). (b) Correlation of basic density of wood with stem height in growth-rings of the tallest E. globulus (tree 5) from two stands near Hilton, WA (n = 54).

Discussion

In general, the measured rates of growth, tree abundance and leaf area support the original contention that availability of water was greater at Site 1 than Site 2. The greater δ13C of wood from trees at Site 2 and the correlation between previous season’s rainfall and δ13C of wood, suggest that δ13C signatures have potential as broad indices of drought stress in stands of E. globulus. The inferred influence of water availability on δ13C of wood is supported by other studies (e.g. Leavitt 1993; McNulty & Swank 1995). However, greater replication of sampling may be necessary before δ13C measurements are adopted as predictive measures of water stress. For example, δ13C of wood from each site was different in the first and sixth year but was not in years two to five. When data for each site from all 6 years were pooled (increasing the number of observations from four to 24) the difference between sites was significant. The 1‰ greater δ13C of wood from trees at Site 2 was similar to the difference in δ13C of wood produced in drought or non-drought years by P. strobus and A. saccharinum (Leavitt 1993).

Trees within each stand had similar variation of δ13C of wood (except for the smallest trees) and δ13C of wood was not correlated with growth rate of individual trees suggesting that microenvironments of individual trees were similar within a stand and that any variations had little effect on the mean δ13C of wood. The atypical pattern of δ13C of wood in the smallest trees suggests that their microenvironment differed greatly from that of other trees and data from these trees were not included in statistical analyses. Decreasing δ13C of wood during the last 4 years suggests that the smallest trees became increasingly more shaded (Francey & Farquhar 1982). Leavitt (1993) and Panek & Waring (1995) have previously noted the importance of selecting co-dominant trees with similar shape and size from a stand for investigating relationships between carbon isotope discrimination and environmental effects.

δ13C of wood increased with increasing stem height at a similar rate to that observed in growth-rings of Pinus edulis (– 0·15‰ m–1, Leavitt & Long 1986) and leaves of Fagus sylvatica (– 0·12‰ m–1, Schleser 1992) but was poorly related to height below canopy height; a pattern also observed by Leavitt & Long (1986). Increased δ13C of wood with increasing height has been attributed to gradients in microclimatic conditions (e.g. shading of the lower leaves, Francey & Farquhar 1982; Leavitt & Long 1986) and greater refixation by lower foliage of CO2 respired from the soil (Vogel 1978; Medina & Minchin 1980). However, Schleser (1992) found no significant gradient of δ13C in stem wood of 140 year-old F. sylvatica with height and concluded that lower branches fixed insufficient carbon to fully support their own growth and that of the stem cambium. Hence, δ13C of all stem wood was similar to that of leaves in the top third of the canopy. The age of those trees and the size of their branches (up to 8 m length) would have produced a substantial requirement of carbohydrate for maintenance. In contrast, the trees in our study were young and the majority of branches were small (length 0·5–2 m). It seems reasonable to suggest that lower branches were self-sufficient and contributing carbohydrate to local cambial activity in the stem. The trend of increasing δ13C of stem wood below canopy height suggests that the contribution of carbohydrate fixed by lower branches to cambial activity is localized and that carbon fixed in the upper canopy is the major source of carbohydrate for cambial activity in the lower stem and the roots. Because longitudinal variation of δ13C of stem wood in trees is likely to depend on the size and distribution of branches along the stem, as well as microclimatic conditions, care must be taken in relating δ13C of wood from the lower stem (e.g. at breast height) to that of the canopy.

The poor relationship between wood density and δ13C may be the result of the different mechanisms by which they are influenced by water availability. Water availability influences δ13C of wood mainly by altering stomatal conductance and the concentration of CO2 in leaves (Farquhar et al. 1982) and influences wood density through the supply of photosynthates and auxins to the cambium (Kozlowski 1982), but stomatal conductance is generally more sensitive to water deficits than is photosynthesis (Teskey et al. 1986). We might speculate that the δ13C of wood is most influenced by frequent, relatively mild water deficits (which modify stomatal conductance) while cambial activity and wood density may be more influenced by less frequent but more severe water deficits (which reduce photosynthesis). Further investigation of the wood density–carbon isotope discrimination relationship with older stands is required to elucidate this relationship.

The representative, co-dominant tree from each site showed a clear seasonal pattern of δ13C of wood. δ13C was least during the main growth period when the availability of water was greatest and evaporative demand of the atmosphere was least, but increased greatly during the hot, dry summer–autumn period. The apparent relationship between δ13C of wood and soil and atmospheric water deficits observed in this study is similar to that observed in studies by Leavitt (1993), Livingston & Spittlehouse (1996) and Walcroft et al. (1997).

Our data show that while water was generally more available at Site 1 than at Site 2, there was a greater seasonal increase in δ13C of wood in the tree from Site 1. Together with our observations that (1) growth-rings in trees from Site 1 contained more late wood than those from Site 2, especially in the outer growth-rings, and (2) basic density of wood from Site 2 (496 ± 14 kg m–3) was less than from Site 1 (554 ± 5·3 kg m–3) and there was no relationship between wood density and growth rate of trees within either stand, the data suggest that cambial activity ceased earlier in the growing season at Site 2 than at Site 1. δ13C of wood only ‘records’ drought stress while the cambium is active and carbon fixed by leaves is assimilated into wood (Leavitt & Long 1989). As the growing season proceeds and the store of available soil water decreases, δ13C of wood increases but cambial activity slows and may eventually cease (Kozlowski 1982; Fig. 7). The earlier studies of Zahner (1968) showed that the onset of drought after late-wood formation had begun resulted in fewer late-wood fibres being formed and Larson (1967) argued on the basis of considerable evidence that trees can produce more late wood and more dense wood when planted as exotics in climates with an extended growing season. Similarly, in Western Australia, Hingston (1996) generally measured little growth in stem diameter of E. globulus during summer (and sometimes stem shrinkage) in low rainfall regions. Our results also support the view of Hillis (1978) that environmental conditions are more important than growth rate for determining wood density.

Figure 7.

. Stylized pattern of δ13C of wood of E. globulus during one season of growth. The first formed wood in the growth-ring (a) carries a ‘record’ of drought stress from the previous summer. δ13C of subsequently formed wood indicates relative drought stress during the main growth period (b) and the increase of δ13C in the last formed wood of the growth-ring (c) indicates relative drought stress during and the duration of cambial activity into the summer–autumn period.

Further support for our interpretation of the cause of the seasonal and site patterns of δ13C in wood comes from other studies of hardwoods. For example, vessel formation at the beginning of annual cambial activity in F. sylvatica was mainly controlled by ‘internal factors’ with little influence of rainfall, while towards the end of cambial activity vessel formation was strongly influenced by recent rainfall (Sass & Eckstein 1995). Our correlation of annual growth with the previous season’s rainfall is certainly partly owing to the small number of observations (3 years only), but is also evidence of ‘residual’ effects of drought. Reduced growth of new leaves and leaf abscission late in the previous growing season reduce the supply of photosynthates and auxins to the cambium in the following season and thereby reduce cambial activity and wood production (Kozlowski 1982; Aloni 1991). Residual effects of drought include the gradual decrease of δ13C in the first formed wood in each growth-ring (Fig. 7). In F. sylvatica, the mean area of the first vessels formed in each growth-ring was strongly influenced by water availability in the previous summer (Sass & Eckstein 1995). However, the gradual decrease of δ13C may also reflect mixing of labile carbon fractions across the growth-ring boundary (Tans, de Jong & Mook 1978) and analysis of the cellulose fraction rather than whole wood will be needed to determine the cause. The considerable ‘lag’ between the onset of drought conditions and significant changes to the pattern of distribution of carbon (e.g. to wood production) are some confirmation that E. globulus is poorly adapted to drought (Jacobs 1981).

Another plausible explanation of the observed seasonal patterns of δ13C in wood is that cambial activity ceased at the same time at both sites, but that the greater LAI at Site 1 (nearly twice that of Site 2) and associated transpiration late in the growing season, rapidly depleted the store of available soil water causing greater stomatal closure and a rapid increase in δ13C of wood. While this probably contributed to the increased δ13C of wood late in the growing season, E. globulus is known to effectively control water loss as the soil dries (e.g. White, Beadle & Worledge 1996; David et al. 1997). Cambial activity certainly continued during conditions of greater stomatal closure at Site 1 than at Site 2 and the weight of other evidence strongly suggests earlier cessation of cambial activity at Site 2.

Growth (e.g. mean annual increments) and leaf mass and area of the two stands measured here were commensurate with the availability of water. For example, LAI of Site 1 (4·6 ± 0·2 m2 m–2) was comparable to that measured at Mummbalup, WA, and Boola, Victoria (4·2 m2 m–2), where rainfall is close to 1000 mm year–1 while LAI of Site 2 (2·7±0·03m2 m–2) was comparable to that at Darkan, WA (2·9m2 m–2), and at Glencoe, Victoria (1·7–3·2 m2 m–2), where rainfall is close to 600 mm year–1 (Hingston 1996; Bennett, Weston & Attiwill 1997). The data also suggest that predicted rates of growth as great as 20–30 m3 ha–1 year–1 for Bluegum plantations are optimistic in medium- to low-rainfall regions of WA or areas where soil properties greatly limit the availability of water.

In contrast to the difference in total above-ground mass, leaf mass and leaf area between the sites, leaf specific area (3·5 m2 kg–1) was unaffected but was less than any of the E. globulus stands studied by Bennett et al. (1997) or Hingston (1996). We agree with Linder (1985) and Pereira et al. (1989) that productivity and leaf area are closely related and that the water balance of the site controls leaf area and productivity in the long term, as suggested by Grier & Running (1977). Leaf specific area seems better related to the evaporative climate than to water availability per se (Specht & Specht 1989).

This study revealed that δ13C of wood has potential as an early indicator of drought-stress in eucalypt plantations. It has also suggested that δ13C of wood can indicate changes in the nature and duration of cambial activity which may result in less production of late wood and less dense wood. These conclusions are supported by studies of other tree species (Zahner 1968; Kozlowski 1982; Leavitt 1993; Högberg et al. 1995).

Acknowledgements

We acknowledge financial support from Bunning’s Treefarms Pty Ltd for this project, and David Groom, Steve Grallelis and Mike Booth for assistance with selecting experimental stands and with felling of sample trees. We also thank Melanie Hendricks and Ben McMillen for assistance in the field, Peter Landman, David Arthur and Peter Kemp for advice and assistance with preparing wood samples for δ13C analysis, John Pate and Lidia Bednarek for access to and operation of the mass spectrometer, and Lauren Bennett for advice on statistical analyses. We thank an anonymous reviewer for direction to additional literature and constructive comments which greatly improved the manuscript.

Ancillary