Architectural plasticity in young Eucalyptus marginata on restored bauxite mines and adjacent natural forest in south-western Australia

The aboveground architecture of Eucalyptus marginata (Jarrah) was investigated in chronosequences of young trees (2.5, 5 and 10 m height) growing in a seasonally dry climate in a natural forest environment with intact soils, and on adjacent restored bauxite mine sites on soils with highly modified A and B horizons above an intact C horizon. Compared to forest trees, trees on restored sites were much younger and faster growing, with straighter, more clearly defined main stems and deeper, narrower crowns containing a greater number of branches that were longer, thinner and more vertically angled. Trees on restored sites also had a higher fraction of biomass in leaves than forest trees, as indicated by 20–25% thicker leaves, 30–70% greater leaf area, 10–30% greater leaf area to sapwood area ratios and 5–30% lesser branch Huber values. Differences in crown architecture and biomass distribution were consistent with putatively greater soil-water, nutrient and light availability on restored sites. Our results demonstrate that under the same climatic conditions, E. marginata displays a high degree of plasticity of aboveground architecture in response to the net effects of resource availability and soil environment. These differences in architecture are likely to have functional consequences in relation to tree hydraulics and growth that, on larger scales, is likely to affect the water and carbon balances of restored forest ecosystems. This study highlights substrate as a significant determinant of tree architecture in water-limited environments. It further suggests that the architecture of young trees on restored sites may need to change again if they are to survive likely longer-term changes in resource availability.


Introduction
Aboveground architecture is a major determinant of tree function through its effects on leaf-scale physiological processes such as photosynthesis and transpiration (e.g., Hubbard et al. 2001), which in turn contribute strongly to larger-scale ecosystem characteristics such as carbon and water balances.Tree architecture is a classical illustration of the 'genotype versus environment' interaction, where 'plasticity' in architecture might be thought of as the variation in architecture from a genetically pre-determined pattern of growth.One of the most plastic of all tree genera is Eucalyptus, and in wild populations in Australia, it is difficult to find even two eucalypts of the same species identical in shape and structure.Eucalyptus is a commercially and environmentally important genus worldwide and there is a growing need to understand the 'plasticity' of eucalypt architecture in natural and managed environments, especially as it relates to physiological processes such as water and carbon gain.
Most of our knowledge of Eucalyptus architecture comes from the studies of species in managed plantations, where both genotypic and environmental influences on architecture are routinely manipulated to meet commercial objectives.Control of genotype is achieved through selection and breeding, and control of environment is achieved through practices such as irrigation (Stape et al. 2008), nutrient addition (Smethurst et al. 2003), spacing (Neilsen and Gerrand 1999), pruning (Pinkard 2002) and thinning (Medhurst and Beadle 2001).Numerous species respond to management, and in many cases, plantation architectures are almost unrecognizable from native architectures.A prime example is Eucalyptus camaldulensis Dehn.(river red gum)-one of the most widely distributed of all eucalypts.In semi-arid riparian ecosystems, naturally occurring E. camaldulensis is characterized by a short, thick bole and a large, irregularly shaped crown (Boland et al. 1984), while in managed plantations, some provenances of this species can be grown with a tall and straight bole and a narrow crown (Akhtar et al. 2008).
Compared to managed plantations, our knowledge of 'natural' architectural plasticity in Eucalyptus remains poor.Anecdotal evidence suggests many, if not most, eucalypts vary in size and shape in response to differences in soil type, climate and the availability of key resources including light, water and nutrients (Bowman and Kirkpatrick 1986, Vander Willigen and Pammenter 1998, Li et al. 2000).However, the lack of quantitative data describing exact responses of architecture to environmental and climatic variables for Eucalyptus severely limits our ability to predict structural and functional outcomes of growth under any given set of conditions.Increasingly, there is a need to characterize the architectures of species grown on reclaimed land (e.g., saline landscapes, farmland and restored mine sites), where the ecohydrological consequences and sustainability of re-forestation are uncertain.
We took advantage of a 'natural experiment' provided by bauxite mining in the Jarrah forest of south-western Australia to examine the combined influences of resource availability, competition and soil on aboveground architecture of juvenile Eucalyptus marginata (Jarrah).We compared trees of the same genotype under the same climatic conditions growing in two contrasting environmentsundisturbed natural forest and adjacent restored mine sites-with markedly different soil profiles, age structures and availability of resources.
In native forests, Jarrah grows in a deeply weathered lateritic soil profile, consisting of sandy-gravel A and B horizons (< 0.1 m) above a caprock layer and friable bauxite ore, below which lie deep layers of clay (C horizon) up to 40 m depth (Churchward and Dimmock 1989).The C horizon stores a large amount of water, but the A and B horizons are frequently dry such that plants unable to penetrate the caprock layer typically suffer water shortages during the 6-month summer drought.The soils are also poor in most nutrients, even by comparison with other eucalypt forest soils (Hingston et al. 1981).Early growth of Jarrah is heavily influenced by neighbouring overstorey, and juveniles may persist for many years in a semi-dormant form until a gap forms in the canopy above (Abbott and Loneragan 1984).At maturity, the plasticity of the species is emphasized by its tall and straight habit (up to 40 m height) in higher rainfall areas (1200 mm year À1 ), compared to its reduced mallee form (< 5 m height at maturity with multiple stems arising from a lignotuber) in nearby low rainfall areas (< 300 mm year À1 ).
In contrast to undisturbed forest, restored mine sites are free of large trees and have highly modified upper soil profiles (Koch 2007a).During mining, all vegetation is cleared, and the A and B horizons and caprock are removed and stockpiled.Bauxite ore is removed and the stockpiled soil material is returned above the intact C horizon.The resto-ration process includes surface landscaping to ensure sites are internally drained, and a significant amount of water is stored in the upper profile over the mining period and during the early years of restoration.The nutrient capital of the soil is frequently augmented with phosphatic fertilizer and growth of N-fixing legumes.Emerging stands of E. marginata (from seed) on restored sites have full access to light and are notably homogeneous in age, size and tree density.
We studied chronosequences of trees in each environment to account for changes in resources availability over time with increasing age and size.Our hypothesis was that compared to forest sites, patterns of biomass distribution to leaves and woody tissue on restored sites would be consistent with putatively greater water and nutrient availability, and that differences in crown architecture would reflect these and greater light availability.A further aim of this study was to assess the architectural trajectory of trees growing on restored sites in relation to the long-term objective of mine restoration, which is to establish a self-sustaining forest ecosystem.

Study sites
This study was conducted in the northern Jarrah forest of southwest Western Australia.The region has a Mediterranean-type climate with hot, dry summers and cool, wet winters.Mean annual rainfall is around 1250 mm, with $ 80% falling between May and October.Mean annual pan evaporation is around 1450 mm.The Jarrah forest is a tall, dry sclerophyll open forest dominated by E. marginata (Jarrah) and Corymbia calophylla (Marri).Soil profiles of the northern Jarrah forest typically consist of coarse-textured, sandy gravel topsoil to 15-30 cm depth, overlying layers of lateritic gravels to a lateritic caprock at $ 1 m depth.Below the caprock lies a zone of mottled, friable loam rich in aluminium oxides (bauxite) to $ 4 m depth.The bauxite zone overlies deep layers of fine-textured kaolinitic clay down to bedrock at 10-40 m depth.
Study sites were located in a mixed-aged stand of Jarrah near the town of Jarrahdale (32°39 0 S and 116°01 0 E; elevation 250 m) that had remained largely undisturbed since heavy logging during the 1940s, but had been subject to bauxite mining since the early 1990s.Mining takes place in pods of several hectares leaving a mosaic of mined and unmined areas.We selected 'natural forest' sites in undisturbed patches of land that contained numerous juvenile Jarrah trees (up to 10 m height) shaded by older, taller Jarrah trees (20-30 m height).Overstorey crown cover at these sites was $ 50-80% and the understorey comprised numerous herbaceous species and small shrubs (including Hakea, Xanthorrhoea and Macrozamia spp.), but only a small number of woody species other than Jarrah (most notably Banksia, Allocasuarina and Persoonia spp.).The structure and growth dynamics of post-logging Jarrah forest stands are described in detail in Abbott and Loneragan (1984).
Restored mine sites were located adjacent to the natural forest sites.Sites were mined and restored by Alcoa World Alumina Australia according to the following general procedure described in detail by Koch (2007a).Briefly, areas of forest allocated for mining were cleared of vegetation, stripped of topsoil and mined to a depth of several metres.After mining, pit walls were pushed down and shaped to smoothly blend in with the surrounding topography, and pit floors were pre-ripped to break up compaction.Previously stockpiled sandy gravel (overburden) was then spread evenly over the pit to a depth of about 0.5 m, followed by fresh topsoil to a depth of about 0.1 m. 'Directly returned' topsoil was sourced from nearby pits being prepared for mining (returned topsoil has been shown to contain a large fraction of the forest seed bank, nutrients and microbes necessary for successful re-establishment of plant species and soil processes (see Jasper 2007, Koch 2007b)).Pits were then deep-ripped to create a friable rooting zone to 1.5 m depth.Ripping was contoured to allow water infiltration but prevent rainfall erosion.After site preparation, a seed mix of tree and understorey plants was broadcast at rates based on pre-mining vegetation surveys.Seeds were sourced from a defined zone up to about 20 km from the site to ensure the return of local genetic material.After seeding, sites were supplied with nitrogen and phosphorus fertilizers (500 kg ha À1 di-ammonium phosphate and micronutrients).Target tree densities for restored sites were $ 2500 stems ha À1 (Grant et al. 2007), and following several years of growth, established sites contained $ 80% Jarrah, 20% Marri, with a dense understorey layer dominated by N 2 -fixing Acacia spp.
Restored sites used in this study were similar in size (3-5 ha), topography and landscaping, plant density and composition, nutrient status and post-establishment practices, and they all had a clearly defined, three-tiered vegetation structure (understorey, midstorey and overstorey) with a notably homogeneous overstorey in tree age, size and canopy cover.General site characteristics illustrating the major differences in growth environment between the mine restoration sites and the natural forest are presented in Table 1.

Experimental design
We measured Jarrah trees from three size groups based on tree height: 2.5, 5 and 10 m (±0.5 m).Size groups were chosen to represent a chronosequence of tree development over time in each environment (natural forest versus restored sites).Comparisons between environments were made on the basis of size (height) rather than age, given that (i) Jarrah trees in undisturbed forest stands tend to develop very slowly and may remain dormant in the seedling stage for up to 15-20 years before responding to reductions in overstorey competition and (ii) for relatively young trees (< 20-30 years), size is clearly a more appropriate variable for linking tree structure (e.g., height, leaf area, sapwood area and crown architecture) with function (e.g., likely rates of transpiration, photosynthesis and growth).Clearly, this design also incorporates the influence of age on architectural development.
Measurements were obtained from five replicate trees per size group from each growth environment over a 2-week period in May 2001 (autumn).All trees selected for measurement were visibly healthy and not affected by disease or herbivory.Trees from restored sites were selected from 1 • 20 m diameter plots in sites established in 1998 (2.5 m), 1996 (5 m) and 1991 (10 m).Forest trees were selected from 1 • 20 m diameter plots in two patches of natural forest: one adjacent to the 1991 restored site, and the other adjacent to the 1996 restored site.Plots in natural forest sites contained a mix of study trees from all size groups (2.5, 5 and 10 m).Selected patches of natural forest were similar to each other in stand age, structure and soil type, and all sites were located within a 2 km radius.
Table 1.Descriptive characteristics of natural Jarrah forest and restored bauxite mine sites similar to those used in this study.Note: These data are sourced from the literature and are provided only as a general guide to differences in growth environment between undisturbed forest sites ($ 50 years postlogging) and restored sites ( 10 years of age).References are in brackets.

Site characteristics
Natural forest

Branch allometry and leaf area
All trees were felled at the base and all live branches extending from the main stem were measured for the following: basal diameter (D b ), length (L b ), distance from the lowest branch and angle of inclination above horizontal.Basal diameter was measured over bark at a point 30 mm from the main stem, above any basal swelling.Four branches from the upper two-thirds of the crown and two branches from the lower third of the crown (by length) were harvested, transported to the laboratory and stripped of leaves.About 10 to 20 representative leaves were sub-sampled from each branch and immediately measured for leaf area using a leaf area meter (Model AAM-7, Hayashi Denkoh Co. Ltd., Tokyo, Japan).Leaf subsamples were then oven-dried at 80 °C for 48 h, along with all other leaves.Specific leaf area (SLA, cm 2 g À1 ) of subsample leaves was determined from measurements of leaf area (A leaf , cm 2 ) and leaf dry mass (g).Leaf thickness (T leaf , mm) was measured to the nearest 0.01 mm using digital callipers.Allometric relationships for predicting branch leaf area (A Lb ) from branch basal cross-sectional area (A b ) were developed for upper and lower crown positions for each experimental treatment (2 crown positions • 3 size classes • 2 growth environments = 12 linear functions).These relationships were then applied to estimate the total tree leaf area (A L ) from the census of branch diameters.

Stem sapwood area
Stem sapwood area (A S ) was determined from wood discs (20-30 mm in thickness) severed from the main stem at 0.5 m height.Discs were wrapped in plastic and kept at 4 °C until processed in the laboratory within 1-2 weeks.
In the laboratory, wood discs were stripped of outer bark and upper and lower surfaces were cleanly shaved using a razor blade.Discs were then partially immersed in a solution of 0.1% basic fuschin dye, proximal side down.Staining of xylem vessels on the distal surface was observed following the movement of dye through open xylem vessels via capillary action.Wood discs were removed from dye and air-dried for 12 h.After drying, discs were re-shaved using a razor blade and freshly cut surfaces and thin sections of wood were examined for dye-stained and occluded xylem vessels using a light microscope to determine sapwood-heartwood boundaries.Cross-sectional outlines of sapwood area were carefully traced onto tracing paper, cut out and measured using a leaf area meter.Measurements of stem sapwood area were then used to quantify whole-tree leaf area to sapwood area ratios (A L /A S ).Unstained portions of wood discs from forest trees were further examined under the microscope to determine the approximate tree age (±2-3 years) from the number of growth rings.

Aboveground architecture
Measuring tapes were used to measure tree height (H), crown length (L c ) and crown diameter (D c ).Digital callipers were used to measure stem diameter at 0.1 m height (above any basal swelling), stem diameter at breast height, 1.3 m (DBH) and stem diameter at 0.5 m height increments from the base to the tip of the crown.All measurements of stem diameter were over bark.Total stem volume (V) was estimated as the sum of all segment volumes following division of the main stem into 0.5 m length segments.The volume of each segment was calculated as the product of segment length and the cross-sectional area mid-segment.Tree form was quantified using a form factor (F) describing the main stem volume relative to the volume of a reference cylinder of the same basal diameter (Philip 1994): where A 0.9 is the cross-sectional area of the stem at a proportion of 0.9 of the total height measured from the crown tip, i.e., 0.1 from ground level.Crown volume (V c ) was estimated using a model for young hardwoods, including Eucalyptus, based on the volume of a cone (Philip 1994): Other crown measurements included ratio-based indices describing live crown length (L c /H), apical dominance (H/D c ), crown shape (L c /D c ) and leaf area density (A L /V c ).Additional indices derived from branch measurements included the branch shape ratio (L b /D b ), the percentage of branches > 10 mm in basal diameter and the percentage of branches > 1 m in length.Branching architecture was further described using simple 'pipe-model' parameters, including the branch Huber value (A b /A Lb ), describing the investment in stem tissue per unit leaf area fed (Tyree and Ewers 1991), and a whole-tree 'branchiness' ratio, defined as the ratio of the total basal cross-sectional area of all branches to proximal stem sapwood area (RA b /A S ).

Statistical analyses
The effects of environment, tree size and environment-size interactions for selected variables were examined using ANOVAs.Where necessary, specific differences were examined using t tests.Branch-scale allometric relationships were developed from linear regression analyses, and where necessary, treatment differences were assessed using the 95% confidence intervals (CIs) of slopes and intercepts of regression functions.

Leaf architecture
Leaves of trees on restored sites were 20-25% thicker than leaves of forest trees, but growth environment had no effect on the size of individual leaves as measured by area (Figure 1A and B; Table 2).There was, however, a consistent difference in both leaf thickness and area related to tree height.In both environments, 2.5 m trees had leaves that were 20% larger in area and 20-30% thicker than leaves of 5 and 10 m trees (Figure 1A and B; Table 2).As a result of these differences, the SLA for restored sites was 15% less than that for forest sites (Figure 1C; Table 2).

Branch architecture
Mean branch diameter of taller trees was significantly greater than that of shorter trees (Figure 2A; Table 2) due to the fact that thicker branches (> 10 mm diameter) increased in number with increasing tree height (Figure 2B; Table 2).Overall, mean branch diameter increased by 20% with each doubling of tree height.Similarly, taller trees had significantly longer mean branch lengths than shorter trees (Figure 2C; Table 2) as a result of having more longer branches > 1 m length (Figure 2D; Table 2).
Growth environment had a small but significant influence on both mean branch diameter and mean branch length.On average, forest trees had slightly shorter, thicker branches, whereas trees on restored sites had slightly longer, thinner branches (Figure 2A-D; Table 2).This trend was confirmed by branch shape ratio data derived from individual branches, which showed that trees on restored sites had significantly greater length/diameter ratios than forest trees (Figure 2E; Table 2).These data also showed that branch shape remained more or less constant with increasing tree height.Although branches were different in shape, Huber values indicated that trees tended to allocate the same amount of stem tissue per unit leaf area (Figure 2F;  Table 2), except for F 2.5m trees, which allocated 40% more stem tissue per unit leaf area compared to all other trees (P < 0.05).

Branch allometry
Branch leaf area was strongly related to branch stem basal area in all cases (R 2 = 0.40-0.97,P < 0.01) (Figure 3; Table 3), confirming that whole-tree leaf area could be satisfactorily estimated from a census of branch diameters.On average, lower crown branches supported 40% less leaf area per unit basal area than upper crown branches.For upper crown branches, leaf area per unit basal area was not dependent on growth environment or tree height; however, for lower crown branches, shorter trees on restored sites (R 2.5m and R 5m ) supported significantly more leaf area   2.
per unit basal area than shorter forest trees (F 2.5m and F 5m ) (Figure 3; Table 3).

Crown architecture
Apical dominance (height growth relative to lateral spread of the crown) increased with increasing height in both environments, and taller trees on restored sites (R 5m and R 10m ) were $ 20% more apically dominant than forest trees of the same size trees (Figure 4A; Table 2).Trees on restored sites clearly had deeper crowns spanning a greater proportion of the main stem than forest trees, and crown depth decreased significantly with increasing tree height in both environments: from 90% (R 2.5m ) of the main stem to 60% (R 10m ) for trees on restored sites, and from 50% (F 2.5m ) of the main stem to 30% (F 10m ) for forest trees (Figure 4B; Table 2).Crown volumes increased threefold with each doubling of tree height (Figure 4C; Table 2), and although there were no significant differences in volume between forest trees and those on restored sites, there was a marked difference in crown shape.Crowns of trees on restored sites were half as wide as they were deep, whereas forest tree crowns were equally wide as they were deep (Figure 4D; Table 2).Differences in crown shape were clearly established at an early age and maintained over time.Within the crown, leaf area density (m 2 leaf area per m 3 crown volume) of trees on restored sites remained constant as crown volume and whole-tree leaf area increased with increasing tree height (Figure 4E; Table 2).In contrast, leaf area density of forest trees decreased significantly with increasing tree height, by a factor of 0.5 as trees doubled in height from 2.5 to 5 m, and then again by a factor of 0.25 as trees doubled in height from 5 to 10 m (Figure 4E; Table 2).
Branch frequency (# branches on main stem per m crown length) remained constant for trees on restored sites, but decreased significantly with increasing height for forest trees to the point where F 10m trees had 30% fewer branches per m crown length than F 2.5m trees (Figure 4F; Table 2).Despite these differences, all trees displayed a similar degree of 'branchiness', i.e., there was close agreement in the crosssectional area of sapwood invested in the main stem to feed the sum total basal cross-sectional area of all the branches of the trees (Figure 4G; Table 2).However, branches of trees on restored sites grew at a significantly steeper angle above horizontal (60°) than branches of forest trees (43°) (Figure 4H; Table 2).

Whole-tree architecture
Tree height varied little within each size class as per the experimental design, although F 10m trees were in reality 20% taller than R 10m trees (Figure 5A).As expected, forest trees were significantly (three times) older than trees of the same size on restored sites (Figure 5B; Table 2).Mean DBH of F 2.5m trees was nearly twice that of R 2.5m trees but there were no other differences in DBH related to environment (Figure 5C; Table 2).We note that breast height of 1.3 m was close to the crown base for F 2.5m trees but closer to the middle of the crown for R 2.5m trees, due to their significantly longer crowns (as shown in Figure 4B).Hence, the difference in DBH between forest trees and those on restored sites was largest for the 2.5 m size group, compared to the 5 and 10 m size groups where breast height lay below the crown.In terms of growth, DBH of trees on restored sites tripled with the first doubling of height from 2.5 to 5 m and then doubled with a second doubling of height from 5 to 10 m, whereas DBH of forest trees consistently doubled with each doubling of height.Shorter trees on restoration sites (R 2.5m and R 5m ) tended to have boles that were more tapered and conical in shape relative to all other trees, although form factors used to describe stem taper were not significantly different (Figure 5D; Table 2).
Whole-tree leaf area increased significantly with increasing height in both environments and trees on restored sites consistently had more leaf area than forest trees of the same size (Figure 5E; Table 2).Most notably, F 5m trees had 40% less leaf area than R 5m trees (P < 0.001), and F 10m trees Table 3. Linear regression statistics for branch basal area versus leaf area relationships shown in Figure 3.All regressions were highly significant (P < 0.01).Slope values within the same crown position indicated by asterisk (*) are significantly different from each other (P < 0.05).Intercept values indicated by asterisk are significantly different from zero (P < 0.05).

Crown position
Height   had 20% less leaf area than R 10m trees (P < 0.01).For trees on restored sites, leaf area tripled as height doubled from 2.5 to 5 m, and then doubled again with a further doubling in height from 5 to 10 m, whereas for forest trees, leaf area steadily increased by a factor of 2.5 with each doubling of height.Leaf area to sapwood area ratios (A L /A S ) of trees on restored sites were 1.3-1.7 times greater than those of forest trees, and ratios of 10 m trees were 0.7 times those of smaller trees in both environments (Figure 5F; Table 2).

Discussion
Experimental tests of the effects of different soil types on tree architecture are scarce.To our knowledge, ours is the first study to test the effects of substrate (extensively modified A and B soil horizons) on aboveground tree architecture under the same climatic conditions.The resultant plasticity in architecture clearly exemplifies the frequently made claim that knowledge of soil properties and their interaction with plant architecture is essential to improving analysis and prediction of plant architectures and therefore functional properties under differing environmental conditions (Magnani et al. 2002, Sperry et al. 2002, Li et al. 2005).
From a broad array of measured architectural parameters, we found many significant differences between trees of the same height class growing in undisturbed forest and those growing on restored sites.The most striking differences were related to whole-tree leaf area, the spatial arrangement of foliage and patterns of biomass distribution to leaves versus woody tissue.Compared to forest trees, trees on restored sites had deeper crowns (55-90% of the main stem), more branches and 30-70% greater leaf area than forest trees.The crowns of forest trees, on the other hand, extended no lower than the mid-point of the main stem and their lower branches had relatively few leaves.Compared to forest trees, trees on restored sites had longer, thinner, more vertically angled branches that protruded from straighter, more clearly defined main stems, which led to the formation of crowns that were narrower in shape and more apically dominant.Trees on restored sites had relatively more biomass distributed in leaves versus woody tissue compared to forest trees, as indicated by thicker leaves, greater leaf area to sapwood area ratios, and smaller branch Huber values and bole volumes in the early stages of growth.
The architectures described above are consistent with known net rates of growth of juvenile E. marginata in the high rainfall zone of the Jarrah forest in both undisturbed (Abbott and Loneragan 1984, Whitford 1991, Stoneman and Whitford 1995) and restored sites (Ward andKoch 1993, Koch andWard 2005).Growth in the forest was relatively slow and steady: trees took up to 15 years to reach sapling size (5 m in height), and a further 10-15 years to reach the pole (10 m in height) stage of growth.In comparison, growth on restored sites was fast and vigorous: trees took 2-3 years to reach sapling size, and only 8-10 more years to reach the pole size.While growth rate and tree age are intrinsic determinants of aboveground architecture, these factors alone do not fully account for the significant differences in structure and architecture observed in this study.For the most part, trees retained the same gross architecture at each stage of growth, and many of the differences we found supported our primary hypothesis that architectures and patterns of distribution of biomass to leaves and woody tissue reflected putatively greater availability of resources on restored sites.
It is reasonable to assume, albeit without quantitative data, that trees on open restored sites received substantially more light than forest trees in the shady understorey of the undisturbed, mixed-age forest.Different strategies for capturing light are thus reasonable explanations for the spatial arrangement of foliage.The arrangement of foliage in forest trees was high, broad and expansive, helping to optimize light capture from directly overhead rather than from low angles of incidence.This is entirely consistent with the relative direction and amount of light available to plants in the Jarrah forest understorey (Silberstein et al. 2001).In contrast, the crowns of trees on restored sites were long and narrow, well suited to the capture of direct and reflected light, including light at low angles of incidence, more consistent with light availability on open sites.These architectural trends were in broad agreement with the findings of other workers who have compared open-grown trees and forest-grown or shaded trees (O'Connell and Kelty 1994, Bartlett andRemphrey 1998, King 1998).
In addition to light, it is likely that crown architecture was shaped by tree density.Individual forest trees had few near neighbours of similar size and plenty of space for crown expansion.Even-age trees on restored sites were far more crowded, resulting in much stronger apical growth.This finding is in agreement with recent work on 18-year-old stands of E. marginata on similar restored sites by Grigg et al. (2008) who reported that crown and foliage cover generally decreased as tree spacing increased, with the crowns of trees in sparser stands resembling those of opengrown trees.
Structural properties related to the distribution of biomass to leaves and woody tissue represented the 'net effect' of access to other key resources in addition to light, and differences in relative growth rate.While it is difficult to attribute specific differences in architecture to specific properties of the soil profile, attributes such as greater leaf area and A L /A S ratios, and reduced branch Huber values of trees on restored sites are consistent with greater access to water and nutrients (White et al. 1998, Clearwater and Meinzer 2001, Pinkard et al. 2006).There is strong anecdotal evidence that neither summer drought nor competition for water or nutrients limited tree growth on restored sites.Analysis of pre-dawn leaf water potential of shoots suggested trees on restored sites were still moderately well supplied with water by the end of the dry season, if compared to forest trees (Bleby 2003).Also, trees on restored sites grew in soils augmented with fertilizer before establishment, and with a vigorous understorey of N-fixing legume species after establishment.At the time of harvest, the relatively 'resource-rich' restoration environment had encouraged substantial distribution of biomass to leaves.By comparison, trees growing in the relatively 'resource-poor' forest environment had relatively less biomass distributed to leaves.
To some extent, the high fraction of biomass in leaves of trees on restored sites at the time of harvest can be attributed to their young age and fast growth rate.Biomass distribution at any point in time is the result of the balance between leaf production and leaf shedding, which varies with age and growth rate.Trees on restored sites reached their allotted sizes in a relatively short time without a large amount of leaf shedding, such that smaller trees (R 2.5m and R 5m ) may have held half or more of their lifelong leaf production in their current leaves (leaf life span for E. marginata is $ 2 years).This contrasts with the older forest trees that took several more years to reach their allotted size.During this time, forest trees would have produced and shed several generations of leaves (and possibly also some branches), such that their current leaf mass represented a smaller, unknown fraction of their cumulative lifelong leaf production.
Although trees on restored sites held a greater fraction of leaf biomass, it is worth noting that these trees probably allocated a lesser fraction of their annual production (i.e., carbon) to leaves than forest trees.The fraction of annual production allocated to leaves versus stems in young trees is known to vary according to age, size and the light environment (King 1994), and allocational responses to light in older saplings with leaf turnover can be fundamentally different from those in younger saplings that still retain all of their leaves (King 2003).It was beyond the scope of this study to measure leaf turnover or annual production, but it is reasonable to suggest that forest trees allocated most of their annual production to leaves.High allocation to leaves is more likely in shade-grown saplings that lie close to a steady state in which photosynthate of current leaves is used mostly to replace old leaves and maintain leaf area on stems and branches that have already been constructed (King 2003).Trees on restored sites, on the other hand, almost certainly invested a substantial amount of photosynthate in stem growth (in addition to leaves) to support a growing mass of leaves and increase access to light in competition with rapidly growing neighbours.As discussed above, fast early growth on restored sites was encouraged by greater availability of light, water and nutrients, and the resulting high leaf biomass seems likely to have generated a positive feedback, further enhancing both stem and leaf growth.
Our data on SLA offer further insight into the influence of resource availability.SLA is a plastic trait that responds to environmental influences and serves to balance a number of competing needs (Eamus et al. 1999).Greater SLA maximizes the efficiency of light capture and N use, whereas low SLA improves structural strength and longevity and promotes greater water use efficiency.Species with high SLA are commonly found in resource-rich, low light, mesic growth environments where there is little need to conserve water.In contrast, low SLA species are commonly found in dry, resource-poor, high light environments.For Eucalyptus, SLA generally varies among genotypes across climatic gradients (Schulze et al. 2006), but within-genotype variation may or may not be evident (Macfarlane et al. 2004, Warren et al. 2006).We found that E. marginata growing on forest sites had significantly greater SLA than trees on restored sites, and the low SLA leaves of trees on restored sites were significantly thicker.Previous work on eucalypts suggests that increased leaf thickness may be due to increased thickness of palisade mesophyll, which in turn is associated with increased concentrations of leaf N and chlorophyll, and increased photosynthetic capacity per unit leaf area (Sefton et al. 2002).
In addition to affecting biomass distribution and growth, measured differences in architecture between forest trees and those on restored sites are likely to have a number of other functional consequences, particularly in relation to hydraulic architecture and water transport.Water availability plays a dominant role in shaping the hydraulic architecture of eucalypts in both natural and managed environments, and there is growing evidence that plasticity in hydraulic architecture in eucalypts is strongly linked to major changes in water availability or evaporative demand (Eamus et al. 2000, Li et al. 2000, Clearwater and Meinzer 2001, Whitehead and Beadle 2004).
The most relevant hydraulic architectural parameters measured in this study were leaf area (A L ) and leaf area to sapwood area ratio (A L /A S ).Trees on restored sites had significantly greater A L and A L /A S than forest trees, consistent with inferred differences in water availability and a general view that trees optimize their 'hydraulic equipment' for drawing water from their particular spatial and temporal niche in the environment (Sperry et al. 2002).Based on the hydraulic theory, particularly the requirement to maintain homeostatic water potential gradients in the stem during water transport (Tyree and Ewers 1991), there is a strong likelihood that greater A L and A L /A S of trees on restored sites were accompanied by a more hydraulically conductive sapwood.The literature suggests that A L /A S is positively correlated with stem hydraulic conductivity in situations where A L /A S is increased in response to increased resource availability, particularly soil water (Medhurst and Beadle 2002).On this basis, we hypothesize that relative to forest trees, trees on restored sites will exhibit faster rates of transpiration per unit leaf area, greater whole-tree hydraulic conductance and ultimately greater carbon assimilation under conditions of continued abundant water supply.
Interestingly, we note that A L /A S decreased significantly in larger trees in both environments, indicating that a significant change in the balance between leaf area and water transport capacity was required in response to increasing tree age and size.In this study, A L /A S (m 2 cm À2 ) was 0.9 for 5-year-old trees and 0.65 for 10-year-old trees on restored sites, and from other work by Grigg et al. (2008), 0.2 for 18-year-old trees.There is a suggestion that a decrease in A L /A S with increasing size may be a homeostatic mechanism that partially compensates for decreased stem hydraulic conductance as trees grow in height (McDowell et al. 2002).This phenomenon is worthy of further investigation, given conflicting reports of foliagesapwood interactions in larger Eucalyptus trees and the growing interest in using this genus to test the hydraulic limitation hypothesis (Barnard and Ryan 2003).
There is a growing literature on the many variables that interact to determine tree architecture under natural conditions (e.g., climate, habitat and site quality), but soil properties are poorly documented or ignored altogether in many studies.Substrate clearly regulated root growth and access to soil resources in this study.Eucalyptus marginata may take 15-20 years in undisturbed forests to grow a large lignotuber from which they gradually develop a specialized, two-tiered system of shallow lateral roots and deep sinker roots capable of accessing deep sources of water by exploring ancestral root channels and passing through fissures in the lateritic caprock (Doley 1967, Carbon et al. 1980, Dell et al. 1983).Recently, it has been confirmed that trees on restored sites develop extensive root systems from an early age in the modified A and B horizons.Szota et al. (2007) showed that on restored sites E. marginata has an 'opportunistic root system capable of responding to a variety of soil conditions encountered in the post-mining landscape'.They found that on poorer sites, young trees (13 years old) relied on a large number of lateral and small sinker roots to explore only the top 0.5 m of the profile.On better sites, they found that trees produced fewer roots but that some roots were able to penetrate further into the profile.In this study, we illustrate that the rate of growth, the architecture of the tree crowns and hydraulic properties such as A L /A S , SLA and Huber values are consistent with the patterns of root development on restored sites that allow trees improved access to soil resources.A general conclusion from our work and that of Szota et al. (2007) is that substrate has a major influence on aboveground architecture.
Finally, our results provide a useful indicator of the early trajectory of E. marginata trees on restored sites in relation to long-term management objectives (Koch 2007a).In this respect, it must be remembered that young E. marginata in 'resource-poor' natural forest grow slowly, developing their above-and belowground architectures gradually in response to incremental improvements in growing conditions.Aboveground growth only accelerates once root systems are established, and then only after there are gaps in the canopy large enough to facilitate increases in light and soil temperature (Stoneman et al. 1995).Eventually, trees develop a long bole with a large basal area, and a crown dominated by large limbs that are widely spread but support relatively few leaves.This architecture is optimized for survival and growth with the harsh soil condi-tions and drought-prone climate that typifies the Jarrah forest.The architecture of forest trees measured in this study embodies the first slow steps towards this end.
In contrast, relatively 'resource-rich' restored mine sites provide near-ideal growing conditions, and young E. marginata develop architectures to suit.In the first decade of growth, survival rates are good despite annual summer droughts, and trees grow rapidly in response to full interception of solar radiation and addition of both N and P fertilizers (Koch and Samsa 2007).Limited space ensures trees on restored sites grow straight and with better form than forest trees.The characteristics described above rarely accompany young trees in natural, uneven-aged forests of south-western Australia, but typically accompany eucalypts growing in managed plantations (Medhurst and Beadle 2001).Compared to forest trees, the aboveground architectures of trees on restored sites are clearly indicative of germination and growth in an environment that strongly buffers trees against resource limitations.The architecture of trees on restored sites also demonstrates that architectural plasticity is a useful trait that may allow this species to grow efficiently under the full range of conditions encountered naturally in the Jarrah forest, including vigorous regeneration on high-quality, resource-rich sites, or in dense, even-aged stands following severe wildfires.
The restoration of forests after mining provides an important, large-scale and on-going experiment.Eventually, trees on restored sites will experience limitations and it remains to be seen if the long-term architecture better reflects natural regimes of resource availability in the Jarrah forest.It is difficult to gauge when the architecture of trees on restored sites may converge with that of forest trees, or if this may be facilitated by management (e.g., thinning).This is an important topic for further research.

Funding
We thank the Australian Research Council and Alcoa World Alumina Australia for financial support.T.M.B. received an Australian Postgraduate Award (Industry) for part of the time during this study.

Figure 1 .
Figure 1.Bar plots of leaf measurements from 2.5, 5 and 10 m tall trees growing on restored mine sites (restored) and natural forest sites (forest), including: (A) average area of individual leaves; (B) leaf thickness; and (C) SLA.Data are mean ± SE, n = 5.ANOVA statistics are shown in Table 3.A leaf = average area of individual leaves, SLA = specific leaf area (i.e., A leaf divided by mean leaf mass) and T leaf = leaf thickness.

Figure 2 .
Figure 2. Bar plots of branch measurements from 2.5, 5 and 10 m tall trees growing on restored mine sites (restored) and natural forest sites (forest), including: (A) branch diameter; (B) % of branches with D b > 10 mm; (C) branch length; (D) % of branches with L b > 1 m; (E) branch shape ratio; and (F) branch Huber value.Data are mean ± SE, n = 5.ANOVA statistics are shown in Table 3. D b = branch diameter, L b = branch length, A b = branch basal crosssectional area and A Lb = branch leaf area.

Figure 3 .
Figure 3. Allometric relationships used to predict branch leaf area (A Lb ) from branch stem cross-sectional area (A b ) measured at the branch base (over bark).Separate relationships are shown for the upper (A, C and E) and lower (B, D and F) crowns of 2.5, 5 and 10 m tall trees growing on restored mine sites (restored) and natural forest sites (forest).'Upper crown' refers to the upper two-thirds and 'lower crown' refers to the lower third of the crown by length.Upper crown data: n = 20 branches (4 branches per tree • 5 replicate trees).Lower crown data: n = 10 branches (2 branches per tree • 5 replicate trees).Linear regression lines are shown (solid = restoration and dashed = forest), and log scales are used for clarity.Regression statistics are presented in Table2.

Figure 4 .
Figure 4. Bar plots of crown measurements from 2.5, 5 and 10 m tall trees growing on restored mine sites (restored) and natural forest sites (forest), including: (A) apical dominance ratio; (B) live crown ratio; (C) crown volume; (D) crown shape ratio; (E) leaf area density; (F) branch frequency (# branches per unit length on main stem); (G) 'branchiness' ratio; and (H) mean branch angle.Data are mean ± SE, n = 5.ANOVA statistics are shown in Table 3. H = height, D c = crown diameter, L c = crown length, V c = crown volume, A L = total tree leaf area, A S = stem crosssectional sapwood area at 0.5 m height and A b = branch basal cross-sectional area (RA b = sum of all branches).

Figure 5 .
Figure 5. Bar plots of whole-tree measurements from 2.5, 5 and 10 m tall trees growing on restored mine sites (restored) and natural forest sites (forest), including: (A) height; (B) age; (C) DBH; (D) form factor; (E) total leaf area; and (F) leaf area to sapwood area ratio.Data are mean ± SE, n = 5.ANOVA statistics are shown in Table3.H = height, DBH = diameter at breast height, F = form factor, A L = total tree leaf area and A S = stem cross-sectional sapwood area at 0.5 m height.

Table 2 .
ANOVA statistics showing the effect of environment, tree size and environment-size interactions on aboveground architectural parameters shown in Figures1, 2, 4 and 5. Environments were natural forest and restored mine sites.Tree size classes were 2.5, 5 and 10 m height.*P < 0.5; **P < 0.01; ***P < 0.001; and ns, not significant.