Provenance variation of ecologically important traits of forest trees: implications for restoration



    1. School of Animal Biology (M092)and
    Search for more papers by this author

    1. Department of Environment and Conservation, Locked Bag 104, Bentley Delivery Centre, Western Australia 6983, Australia; and
    Search for more papers by this author

    1. School of Plant Biology, Faculty of Natural and Agricultural Sciences, The University of Western Australia, 35 Stirling Highway, Crawley, Western Australia 6009, Australia;
    2. Kings Park and Botanic Garden, Botanic Gardens and Parks Authority, Fraser Ave, West Perth, Western Australia 6005, Australia
    Search for more papers by this author

E. K. O’Brien, School of Animal Biology (M092), Faculty of Natural and Agricultural Sciences, The University of Western Australia, 35 Stirling Highway, Crawley, Western Australia 6009, Australia (fax +61 86488 1029; e-mail


  • 1The source of seed or plant material can have profound implications for the success of restoration efforts because most species exhibit adaptive genetic variation within their range. Understanding the geographical distribution of ecologically relevant genetic variation and the environmental factors driving adaptive divergence within species will help to ensure appropriate sourcing of material for ecological restoration.
  • 2We present a study of geographical variation of ecologically important traits of the forest tree jarrah Eucalyptus marginata from a 15-year-old provenance trial in south-western Australia. We assessed trait variation in association with rainfall, latitude and slope position at the site of origin.
  • 3Survival and stem diameter varied at the largest scale, between northern and southern jarrah forest provenances. Stem diameter also varied among rainfall zones, while latitude was a more important determinant of variation of reproductive traits (flowers and buds). None of the environmental variables accounted for significant variation of height, growth form or the presence of capsules. Slope position at the site of origin did not account for significant variation of any trait.
  • 4Trees from low rainfall sites had smaller stem diameters, possibly reflecting selection for slower growth. Such a strategy could prevent drought stress and may explain why trees from the high rainfall southern jarrah forest, which showed the fastest growth, had the poorest survival at the drier northern trial site.
  • 5Variation in the presence of buds and flowers among latitudinal divisions may be because of variation in flowering time, which has been observed previously among E. marginata populations. However, variation among replicate blocks within the trial suggests that the environment also strongly influences expression of these traits.
  • 6Synthesis and applications. We have demonstrated divergence of several ecologically important traits in association with different types of environmental variation. Our findings support an argument for ‘habitat matching’ when sourcing material for restoration; however, differences among trait types in the distribution of variation highlight the need to consider environmental variation at a range of geographical scales. Consideration of ecologically important genetic variation within species is important and this information should be integrated into seed collection strategies for ecological restoration.


Most plant species exhibit spatial structuring of genetic variation throughout their range (Hamrick 1990), which may arise through local adaptation or genetic drift. Genetic variation provides evolutionary flexibility and enables a response to environmental change (Lande & Shannon 1996; Booth & Grime 2003). It has therefore been suggested that preserving genetic variation within, as well as among, species should be a priority in restoration programmes (Crozier 1992; Moritz 1994). While translocation of individuals is sometimes proposed as a strategy to increase genetic variation within populations, individuals transferred across different environments may be poorly adapted to the new conditions. Furthermore, there is a risk of outbreeding depression or genetic swamping when divergent populations interbreed (Hufford & Mazer 2003; Potts et al. 2003; McKay et al. 2005). These consequences of seed transfer must therefore be weighed against the potential benefits of increased genetic variation within founding populations.

To conserve locally adaptive genotypes and prevent outbreeding depression, collection of source seed of the local provenance is frequently advocated for the restoration of plant populations (Keller, Kollmann & Edwards 2000; Jones, Hayes & Sackville Hamilton 2001; Sackville Hamilton 2001). However, such a strategy requires knowledge of the level and distribution of genetic diversity within the species of interest, and assessment of this can vary considerably, depending upon the technique used to detect it (Storfer 1999; Hedrick 2004). While molecular markers such as allozymes, AFLP (amplified fragment length polymorphism), RFLP (restriction fragment length polymorphism)and microsatellites provide powerful means of assessing genetic structuring within and among populations, such markers have repeatedly been shown to be poor predictors of variation of adaptive traits (Karhu et al. 1996; McKay et al. 2001; Reed & Frankham 2001). This disparity among molecular and quantitative measures of genetic variation may be attributed to the influences of different processes. While the effect of neutral genetic drift is consistent across all loci, natural selection affects loci controlling traits that confer adaptation, which may represent only a small proportion of the genome (Eriksson 1995). In the absence of barriers to gene flow (a necessary precursor to divergence by genetic drift), the likelihood of detecting genetic differentiation among populations using techniques that do not specifically target loci controlling adaptive traits is low, even if there is strong, divergent selection. In a restoration context, success of the restored population is contingent upon the survival, growth and reproduction of the founding individuals, hence consideration of adaptive variation is important.

The opportunity for divergence through drift vs. natural selection varies across landscapes and species, in response to factors such as life history, habitat and dispersal (Loveless & Hamrick 1984). In species that are highly outcrossing and continuously distributed, such as many forest trees, extensive gene flow is likely to prevent genetic divergence of neutral variation, although adaptive divergence may still arise through localized selection (Loveless & Hamrick 1984; Hamrick 2004). In such cases, reliance on molecular markers alone to guide decisions about where to source material for restoration may result in the introduction of poorly adapted genotypes.

Assessment of adaptive variation within a species’ range requires the measurement of ecologically important, heritable traits of plants from different source populations, grown at a common site to control for environment as a cause of phenotypic variation (Crandall et al. 2000; McKay et al. 2005). Provenance trials have been used in forestry since the early 19th century for detecting populations with economically desirable characteristics to be targeted for tree-breeding programmes (Guries 1990). Because the source of material included in such trials is known, existing trials may be utilized to study variation of ecologically important traits and identify appropriate source populations for restoration. For long-lived, slow-growing species, such as many forest trees, long-established trials provide a unique opportunity to examine later life-history traits, which may reveal important variation not evident in younger plants.

We assessed survival, growth and reproductive traits of jarrah Eucalyptus marginata Donn ex Smith, grown in a provenance trial in south-western Australia. Eucalyptus marginata is a dominant forest tree species, endemic to the south-west of western Australia. Restoration of E. marginata is needed primarily to replace populations following bauxite mining, which clears approximately 700 ha of E. marginata forest each year (Gardner 2001). In an effort to conserve locally adapted genotypes and maintain intraspecific genetic variation, bauxite mine restoration is undertaken using local provenance seed and plant material (Gardner 2001). However, delineation of appropriate seed collection boundaries would benefit greatly from improved understanding of the geographical distribution of ecologically important genetic variation within key species.

Eucalyptus marginata is long-lived, typically surviving for 300–400 years or more (Dell & Havel 1989). It exhibits substantial morphological variation throughout its distribution, ranging from a tall, straight tree to a shrubby plant with a mallee growth form (Dell & Havel 1989). In natural populations, the height and volume of E. marginata tend to be greatest in the southern jarrah forest and the high rainfall zone on the western edge of the northern jarrah forest, decreasing clinally to the north and east, where mean annual rainfall is lower (Dell & Havel 1989). Despite this apparent high intraspecific variability, analyses of nuclear and chloroplast RFLP variation revealed minimal genetic structuring within the jarrah forest (Wheeler, Byrne & McComb 2003; Wheeler & Byrne 2006) and it has been suggested that seed transfer could be undertaken at a regional scale (Wheeler, Byrne & McComb 2003). However, as much of the genetic variation detected by molecular markers is likely to be neutral, the absence of divergence at marker loci does not necessarily mean that seed collected from throughout the region will be equally well adapted to conditions at a particular site.

If local adaptation has driven genetic divergence of ecologically important traits of this species, it is expected that trait variation among seed sources will increase with environmental divergence. Given an understanding of this relationship, it may be possible to use characteristics of the site of origin to predict the performance of seed at restoration sites. Therefore, the objectives of the study were to: (i) assess whether seed source is associated with variation of survival, growth and reproductive traits of this species and (ii) determine how these trait types vary in association with environmental variation over different geographical scales.


study species

Eucalyptus marginata is distributed in Australia from 100 km north of Perth (31°27′S, 115°52′E) to Albany (34°19′S, 115°09′E) on the south coast, and from the west coast inland to the 600-mm rainfall isohyet (Churchward & Dimmock 1989). There are also outlying populations north and east of this main range at Mt Lesueur, Jilakin Rock and Katanning (Dell & Havel 1989). The climate of this region is mediterranean, characterized by hot, dry summers and cool, wet winters. There is a gradient of mean annual rainfall across the northern jarrah forest, exceeding 1200 mm in the west and declining to 650 mm near the eastern boundary (Dell & Havel 1989). Mean annual rainfall in the southern jarrah forest is higher, often exceeding 1400 mm (Gentilli 1989). Soils within the Darling Range are predominantly lateritic gravels overlying sandy clay subsoils (Churchward & McArthur 1980; McArthur 1991) but on the coastal plains to the west and south of this region E. marginata also occurs on pale sands and sandy gravels (McArthur 1991). Within the area of this study, the main source of edaphic variation is between the northern and southern jarrah forests. Temperature and evaporation rate vary most significantly with latitude, being highest at the northern end of the species’ distribution and declining further south.

sampling design

In 1986–87, seeds were collected from 250 E. marginata trees, 10 from each of 25 sites (provenances) (see Fig. S1 in the supplementary material). We wished to test whether geographical variation at different scales could be used to predict variation of ecologically important traits. Therefore sampling was designed to enable analysis of trait variation in association with site characteristics representing a range of distance classes: region (hundreds of kilometres), latitudinal division (tens of kilometres north–south), rainfall zone (tens of kilometres east–west) and slope position (less than 10 km). Each provenance belonged to either the northern or southern jarrah forest region and lay within one of five latitudinal divisions: Jarrahdale (latitude 32°24′S), Dwellingup (latitude 32°71′S), Collie (latitude 33°21′S), Pemberton (latitude 34°27′S) and Walpole (latitude 34°58′S). Within the northern jarrah forest, provenances were situated along transects running west to east, to encompass rainfall variation across the area. Therefore, rainfall at the site of origin was assigned to one of four categories: low (< 900 mm year−1), medium (900–1100 mm year−1, high (1100–1300 mm year−1) or southern jarrah forest (> 1300 mm year−1). Additionally, each northern jarrah forest provenance was classified as upper slope (high relief relative to immediately adjacent landscape) or lower slope (low relief relative to surrounding landscape). Characteristics of each provenance are provided in Appendix S1 (see the supplementary material).

Previous studies have shown eucalypts to be highly susceptible to inbreeding depression, with reduced growth and survival observed in populations with higher levels of inbreeding (Eldridge & Griffin 1983; Potts, Potts & Cauvin 1987). To avoid this potentially confounding source of variation among provenances, all seed was collected from trees within large stands, which have been shown to exhibit consistently high rates of outcrossing in this species (Millar et al. 2000). Within a provenance, seed was collected from trees spaced at least 100 m apart, to avoid sampling from close relatives. Seeds from different maternal trees were kept separate, to enable among-family trait variation within provenances to be calculated for estimation of trait heritability.

trial establishment

Seeds were germinated in the glasshouse and seedlings grown for 8 months prior to planting out in the field. The trial was established in July 1988 on a rehabilitation site at Alcoa's Jarrahdale bauxite mine (32°24′S, 116°05′E), approximately 50 km south-east of Perth, western Australia. This site lies within the high rainfall zone of the Jarrahdale division of the northern Jarrah forest (see Fig. S1 in the supplementary material).

Seedlings were planted in 75 plots of 100 trees, with each plot comprising trees from a single provenance. The trial was divided into three replicates of 25 plots, such that each provenance occupied exactly one plot in each replicate. In addition, each replicate was subdivided into five incomplete blocks of five plots each, with random allocation of provenances to plots within each incomplete block. Such a design accounts for spatial variation across the site to a greater degree than is permitted by a randomized complete block design (Williams, Matheson & Harwood 2002).

Within plots, 10 rows of 10 seedlings were planted, such that plants within a row were the progeny of a common maternal tree. In general each plot consisted of 10 maternal families but, where there were insufficient seedlings from a particular family, one family was represented twice within the plot. In total, 245 families were included in the trial. Family row positions were randomized within each plot.

Within rows, trees were spaced at 2-m intervals, with adjacent rows 4 m apart. Several measures were taken to maintain a consistent competitive pressure on all seedlings in the trial and prevent edge effects. (i) Plots were positioned such that the same distances separated adjoining plots as separated trees within plots. (ii) Where there were insufficient seedlings available from a particular maternal plant, because of a lower than expected germination or survival, a marri Corymbia calophylla Lindley seedling was planted in each gap. (iii) Two buffer rows of E. marginata (of mixed provenance) were planted around the perimeter of the trial, spaced at the same interval as trees within the trial.

trial assessment

The trial was assessed in November–December 2003, when the trees were 15 years old. Because of the potential for temporal variation of flowering, assessment of the three replicates was progressed evenly. Traits assessed were survival, stem diameter, height, growth form (single or multiple stems) and the presence of buds, flowers and fruits. Growth traits (height and stem diameter) were assessed for each tree. For remaining traits, the plot was the unit of analysis, with each trait expressed as the proportion of trees within each plot exhibiting the trait.

Runts (height < 2 m and without a dominant stem) were assessed for survival, but not for any other trait. For all other surviving trees, the number of stems at breast height (1·3 m above the ground) was recorded. For each stem, height to the top of the canopy was assessed using a vertex hypsometer, and stem diameter at breast height (d.b.h.) was measured over the bark using a diameter tape. For analysis, d.b.h. of each tree was taken to be the square root of the sum of each stem diameter squared, and height was taken as the height of the tallest stem.

For each tree, ‘growth form’ was assessed by scoring trees as having either a single or multiple stems at 1·3 m above the ground. The proportion of trees with multiple stems was determined for each plot. Buds, flowers and fruits were scored as present or absent for each tree. The proportion of trees bearing each of these was determined for each plot.

data analysis

Geographic distribution of trait variation

To assess whether seed source could explain variation of ecologically important traits of E. marginata, provenance variation of each trait was analysed as a mixed model, using the restricted maximum likelihood (REML) function in GenStat (Version 7·1). Growth traits (height and d.b.h.) were each analysed using individual tree data, with provenance and replicate included as fixed factors, and family and incomplete block included as random factors. To analyse provenance variation of survival, multiple stems and the presence of buds, flowers and fruits, the family term was omitted because data points represented a proportion of trees within each plot. For each of these traits, data were transformed (arcsine square root) to meet assumptions of normality and homogeneity of variances.

Two-sample t-tests were used to assess variation of each trait among provenances from the northern and southern jarrah forest regions. Provenance means obtained from the REML analyses of trait variation were used in one-way analyses of variance to investigate how each trait varied in association with slope position (northern jarrah provenances only), rainfall and latitudinal division. For northern jarrah forest provenances, three-way analyses of variance were also conducted to assess interaction effects among slope position, rainfall and latitudinal division on variation of each trait. Southern jarrah forest provenances were excluded from these analyses, because they all fell within one rainfall zone and slope position was not recorded.

Correlations among traits

Because of the potential for variation of tree density to influence growth, regression analyses were used to test for associations between the following combinations of traits: survival and mean d.b.h. (cm) within plots; survival and mean tree height (m) within plots.


Variance components obtained from the REML analysis of individual tree measurements were used to estimate heritability of traits. This was possible only for height and d.b.h., because all other traits were assessed for plots rather than individual trees. Given the small number of representatives of each family within each plot (at most 10), calculating heritability separately for each provenance would not be meaningful in this case. Consequently, a single heritability estimate was made for each trait, encompassing individuals across the entire trial.

Heritability (h2) of height and d.b.h. was calculated as:


where inline imagerefers to the variation amongst families within provenance, inline image is the total phenotypic variance and r is the coefficient of relatedness among progeny within a family. A coefficient of relatedness of 1/2·5 was used. This value has been derived for eucalypts from natural, open-pollinated stands, to take into account the rate of outcrossing generally observed in such systems (Williams, Matheson & Harwood 2002).


trait variation among seed sources

The provenance of seed was associated with significant variation of height and d.b.h. (Table 1) but not of any other trait. Trees from Centaur (upslope in the medium rainfall zone of the Collie division) had the greatest mean height (12·8 ± 0·51 m), 23% higher than trees from Saddleback (upslope in the low rainfall zone of the Dwellingup division), which had the lowest mean height (10·4 ± 1·13 m) (see Appendix S1 in the supplementary material). Trees from Dordagup (Pemberton division of the southern jarrah forest) had the largest mean d.b.h. (21·5 ± 1·38 cm), 22% greater than those from Amphion (upslope in the medium rainfall zone of the Dwellingup division), which had the smallest mean d.b.h. (17·2 ± 3·24 cm). The replicate–provenance interaction was associated with significant variation of both height (P < 0·001) and d.b.h. (P = 0·003) (Table 1), indicating within-provenance variation among different replicate blocks for these traits. Trait means and standard deviations of each trait for each provenance are provided in Appendix S1 (see the supplementary material).

Table 1.  REML analyses of the effect of provenance (Prov) and replicate (Rep) on variance of d.b.h. and height among provenances and replicates within the trial
TraitEffectd.f.Wald statisticP
D.b.h.Rep 2  0·030·986
Prov24 57·8< 0·001
Rep × Prov48 80·10·003
HeightRep 2  0·110·946
Prov24104·16< 0·001
Rep × Prov48129·90< 0·001

Trees from northern jarrah forest provenances had, on average, higher survival (t = 2·16, P= 0·04) and smaller d.b.h. (t = 2·20, P= 0·04) than those from southern jarrah forest provenances (Fig. 1). None of the other traits varied significantly between the two regions.

Figure 1.

Comparison of trees from northern and southern jarrah forest provenances for (a) mean (± 1 SE) survival (%) and (b) mean (± 1 SE) diameter at 1·3 m above the ground (d.b.h.) (cm). Survival rate was significantly higher for northern jarrah forest provenances (two-tailed t-test, t= 2·16, d.f. = 23, P= 0·04), while mean d.b.h. was significantly larger for trees from the southern jarrah forest (two-tailed t-test, t= 2·20, d.f. = 23, P= 0·04).

trait variation in association with environmental variables

Rainfall accounted for significant variation of d.b.h. (P = 0·002), while the proportion of trees bearing buds (P < 0·001) and flowers (P < 0·001) varied significantly among latitudinal divisions (Table 2). Slope position did not account for significant variation of any trait measured. Interactions among environmental variables did not explain significant variation of any trait.

Table 2.  Estimated variance components for random effects family (Fam) and incomplete block (IncBlock) from REML analyses of variance of d.b.h. and height
TraitRandom termVariance componentSE
D.b.h.Rep × IncBlock34·47
Prov × Fam 0·820·29
Rep × Fam 0·640·38
HeightRep × IncBlock 5·38
Prov × Fam 0·310·07
Rep × Fam 0·420·08

Trees from the high rainfall region of the northern jarrah forest and from the southern jarrah forest, where annual rainfall is greatest, had a significantly higher mean d.b.h. than trees from the low rainfall zone of the northern jarrah forest. Trees from the medium rainfall zone were intermediate for this trait (Fig. 2).

Figure 2.

Mean (± 1 SE) of stem diameter at breast height (d.b.h.) (cm) of trees from each rainfall zone (F = 5·08, P < 0·001). Different letters indicate statistically significant differences at α = 0·05, identified by Tukey's HSD post-hoc test. Rainfall zones refer to mean annual rainfall: low (< 900 mm), medium (900–1100 mm), high (1100–1300 mm), southern (> 1300 mm).

The mean proportion of trees within a plot bearing buds was significantly greater among Jarrahdale provenances than provenances from Collie, Dwellingup and Walpole. Pemberton provenances were intermediate for this trait (Fig. 3a). Jarrahdale and Dwellingup provenances had the highest mean proportion of trees bearing flowers, significantly greater than Walpole or Collie, with Pemberton provenances intermediate (Fig. 3b). Both of these traits also varied significantly among replicates within the trial. Replicate two had a greater proportion of trees bearing buds (16·1 ± 3·4%) and flowers (16·1 ± 3·4%) than the other two replicates (buds 5·93 ± 1·07% and flowers 6·36 ± 3·52%).

Figure 3.

Comparison of trees from different latitudinal divisions for mean (± 1 SE) percentage trees bearing (a) buds and (b) flowers. Different letters indicate statistically significant differences at α = 0·05, identified by Tukey's HSD post-hoc test. Differences among divisions were highly significant for both traits (buds, F= 19·23, P < 0·001; flowers, F= 12·15, P < 0·001). Dwell, Dwellingup; Wal, Walpole; Pemb, Pemberton; J'dale, Jarrahdale.

correlations among traits

Regression analysis revealed a significant, negative correlation between survival and mean d.b.h. within plots (R2 = 0·097, P= 0·007). However, the relationship was not significant at the provenance level (P = 0·07), suggesting it is a more important determinant of variation within, rather than among, provenances for either of these traits. Mean height also tended to have a negative correlation with survival, but this relationship was not significant (R2 = 0·0009, P= 0·06).


Variance components obtained from the REML analyses of trait variation (Table 3) were used to calculate heritability of height and d.b.h. Heritability of height (0·13 ± 0·03) was greater than that of d.b.h. (0·06 ± 0·02). However, both of these values were low compared with heritability of growth traits calculated for other forest tree species (Mazanec & Mason 1993).

Table 3.  One-way anova tables showing significance of variation of each trait as a result of (a) latitudinal division and (b) rainfall. With Bonferroni correction for multiple tests, differences were significant at α= 0·007
Source of variationTrait
(a) Rainfall 30·002 30·342 30·082 30·201 30·847 30·38330·411
     Error21 21 21 21 21 21 21 
(b) Division 40·169 40·485 40·081 40·872 4< 0·001 4< 0·00140·113
      Error20 20 20 20 20 20 20 


distribution of trait variation

We have demonstrated ecological trait variation in 15-year-old E. marginata trees over different geographical scales. Significant variation of growth traits (height and d.b.h.) was observed among trees from different provenances. Survival and d.b.h. varied at a regional scale, between trees from the northern and southern jarrah forests. When provenances were partitioned into zones on the basis of mean annual rainfall, average growth rate (at least for d.b.h.) was found to be greater in trees sourced from higher rainfall sites. For reproductive traits, variation was among latitudinal divisions, with trees from the local Jarrahdale provenances exhibiting the highest rates of flowering and bud production.

The geographical variation of traits of E. marginataobserved in the present study contrasts with the negligible genetic differentiation detected by Wheeler, Byrne & McComb (2003) using RFLP markers. Molecular markers have been shown to be poor predictors of quantitative trait variation in many plant species (Karhu et al. 1996; McKay et al. 2001; Reed & Frankham 2001), which may reflect the influence of different processes. Eucalyptus marginata is continuously distributed throughout much of its range, including the entire area represented in this trial. Such a distribution, coupled with highly mobile insect, bird and mammal pollinators, is likely to afford the opportunity for extensive gene flow, which would prevent genetic divergence via drift and may account for the lack of spatial genetic structure at RFLP loci. However, local adaptation through natural selection can drive adaptive trait divergence, even in the presence of gene flow (Endler 1977), and thus represents a potential mechanism to explain the distribution of trait variation observed in the present study. Possible alternative explanations include non-genetic maternal effects and variable levels of inbreeding.

The maternal environment is known to be a significant determinant of progeny variation, particularly for growth traits, in several plant genera (Schaal 1984; Roach & Wulff 1987), including eucalypts (Lopez et al. 2003). This effect therefore has the potential to produce trait variation that may be misinterpreted as having a genetic basis. However, maternal effects appear to exert their greatest influence on early life-history traits of progeny (seed mass, germination and early growth) and diminish considerably with age (Schaal 1984; Roach & Wulff 1987; Lopez et al. 2003), therefore they are likely to be less significant for measurements of adult trees. Furthermore, maternal effects produce high phenotypic correlation among individuals from a common maternal family. This was not observed in this trial, as evidenced by the low heritability estimates for height and d.b.h.

Variable inbreeding depression among populations is another potential determinant of trait variation. Inbreeding depression is associated with reduced growth and survival in numerous eucalypt species (Eldridge & Griffin 1983; Potts, Potts & Cauvin 1987). This is unlikely to be a significant factor in this case, however, because the seed used in the trial was collected from large stands. Outcrossing rates are uniformly high (mean outcrossing rate = 0·81) within large populations of E. marginata (Millar et al. 2000), giving reason to expect that levels of inbreeding, and therefore inbreeding depression, across the sampled populations were largely consistent.

The pattern of trait variation observed in this trial, in conjunction with the lack of spatial genetic structuring at RFLP loci reported by Wheeler, Byrne & McComb (2003), is therefore consistent with a hypothesis of local adaptation in E. marginata, with widespread gene flow limiting neutral genetic divergence. Implicit in an argument that spatial variation of traits results from local adaptation is an assumption that divergent selection pressures operate on these traits throughout the species’ range. It is therefore important to consider how this variation may confer adaptation within the different source sites represented in the trial. A discussion of the factors likely to be driving trait variation among populations needs to consider the different categories of traits separately because they exhibited differing patterns of geographical variation.

growth traits

Lower mean d.b.h. of trees sourced from sites of low mean annual rainfall may indicate greater allocation of resources to below-ground growth. Eucalyptus marginata has been found to maintain high transpiration rates throughout the dry summer months (Doley 1967), a strategy facilitated by the maintenance of an extensive root system that can access water from deep within the soil profile (Dell & Havel 1989). Selection on the development of deep roots, relative to above-ground biomass, is likely to be stronger in water-limiting environments and, indeed, plants are often reported to exhibit a higher root to shoot ratio at drier sites (Chapin, Autumn & Pugnaire 1993; Schenk & Jackson 2002). In the southern jarrah forest, high mean annual rainfall and low evaporation result in a greater supply of soil water, and selection may instead favour rapid above-ground growth, driven by competition for light and other resources. Consistent with this, trees originating from the southern jarrah forest exhibited faster growth rates than those from northern jarrah forest populations. More rapid growth is, however, likely to be less sustainable in the drier northern jarrah forest and may account for the lower survival of southern jarrah forest trees at the trial site.

Despite significant variation of growth traits (height and d.b.h.) among provenances and of stem diameter among rainfall zones, heritability estimates of height and d.b.h. were both very low, suggesting a strong, direct contribution of environment to phenotypic expression of these traits. This is corroborated by significant within-provenance variation of both traits across different replicates in the trial.

reproductive traits

Variation of the mean percentage of trees bearing buds and flowers among provenances from different latitudinal divisions may indicate variable reproductive output or temporal variation of flowering. The trial was assessed for all traits during a single 2-month period, therefore it is not possible to distinguish between these possibilities. However, the absence of significant variation among provenances or divisions in the proportion of trees bearing fruits (a measure of flowering the previous year) is evidence that variation of flowering time within the year may account for variation among different geographical regions to a greater degree than overall variation in flowering rate.

Such an explanation is also supported by published observations of E. marginata flowering in natural populations (Davison & Tay 1989). In a study of flowering at sites within the northern Jarrah forest, maximal flowering time was found to vary from October to January, occurring earliest in the most southerly populations and later in populations further north (Davison & Tay 1989). Furthermore, flowering in the Jarrahdale division was found to be maximal in November–December, which coincides with the measurement of this trial (Davison & Tay 1989).

Consistent with these findings, the mean proportion of trees per plot bearing flowers was highest for populations from Jarrahdale in our trial. If flowering is largely environmentally determined, we may have observed the maximal flowering rate for all populations, given that the trial site was located within the Jarrahdale division. If this is the case, the higher mean proportion of trees flowering from Jarrahdale populations may indicate local adaptation. Evidence for a strong influence of environment on flowering time is seen in the very high variation among replicates for all reproductive traits. Assessment of the trial was intentionally progressed evenly across the three replicates, so it is unlikely that time of measurement can account for this finding. Rather, it seems likely that fine-scale environmental variation across the site is influencing expression of these traits. Assessment of flowering rates throughout the year is necessary to resolve the relative contribution of genotype and environment as determinants of flowering time.

Regardless of the mechanism, variation of flowering rate at one point in time has significant consequences for reproductive isolation. Difference in flowering time among trees at a common site provides a barrier to gene flow, thereby limiting the potential outcrossing rate in the population and increasing biparental inbreeding and correlated paternity (Millar et al. 2000). This has important implications for restoration.

implications for restoration

The rehabilitation of native plant communities following bauxite mining in the northern jarrah forest of south-western Australia is currently undertaken using local provenance seed, which is sourced from within 20 km where possible (Krauss & Koch 2004). We did not detect significant quantitative trait variation below this scale, suggesting a 20-km radius is sufficiently ‘local’ to ensure conservation of adaptive genetic variation of E. marginata. However, we have identified several environmental variables that may be useful predictors of variation of ecologically important traits of this species.

Perhaps the most important finding was that different trait types varied over different geographical scales, highlighting the complexities involved in identifying the distribution of adaptive genetic variation within species. Given that this is likely to be a common observation in widespread, continuously distributed plant species occupying heterogeneous landscapes, the question of how such information may best be used in the development of seed collection practices warrants discussion. Arguments against the use of non-local seed for restoration centre on concerns about loss of local adaptation and outbreeding depression. Decisions about where to draw seed collection boundaries therefore need to consider the likely risk of each of these.

The results of this study suggest that, for E. marginata, seed transfer among rainfall zones may result in the introduction of poorly adapted genotypes, given the slower growth rate (at least of d.b.h.) of trees sourced from lower rainfall sites. Sourcing seed from low rainfall sites for deployment within the high rainfall zone has the potential to reduce mean growth rate at the site. Perhaps more importantly, seed transfer from high to low rainfall sites will be accompanied by increased mean growth rate for the site, which may be unsustainable under drier conditions, resulting in increased mortality. Evidence for this is seen in the relatively poor survival of individuals from the very high rainfall southern jarrah forest at the drier northern jarrah forest trial site.

Significant variation in rates of flower and bud production was observed among provenances from different latitudinal divisions, indicating differences in flowering propensity or flowering time. In either case, the result would be reduced potential for interbreeding among individuals from different divisions. In a restoration context, the implication of this is a diminished risk of genetic swamping or outbreeding depression with the introduction of non-local seed/plants. However, such a strategy is not recommended, because asynchronous flowering within a population reduces the number of potential mating partners at any point in time, which may result in increased selfing and inbreeding depression, threatening the long-term persistence of the restored population. Conversely, the lack of variation of flowering rate among trees from different rainfall zones within latitudinal divisions implies that seed transfer among these sites will provide opportunities for interbreeding, which may have detrimental consequences through genetic swamping or outbreeding depression.

While seed collection programmes should be designed to maintain adaptive genetic divergence among populations, we emphasize that this must be balanced against the need to ensure restored populations contain sufficient genetic variation to minimize the negative effects of inbreeding depression, facilitate future evolution and ensure long-term population persistence. This is of particular importance for species such as E. marginata, which are highly outcrossing and occur in large populations, because they are likely to be more susceptible to inbreeding depression, with an erosion of genetic variation. Assessment of the relative importance of genetic variation within and among populations as determinants of population performance is beyond the scope of this study. However, we emphasize that within the guidelines suggested here for conserving adaptive divergence among populations, efforts should be made to collect seed from as many individuals as possible.

It is also important to note that this trial was conducted at a single, highly disturbed site. Given that the genotype–environment interaction has been shown to exert a significant influence on variation of many plant traits (Williams, Matheson & Harwood 2002), we are unable to comment on how the patterns observed here may vary on other sites. If trait variation is driven by divergent natural selection, comparison of local and non-local provenances across a range of sites using reciprocal transplant experiments should reveal a consistent ‘home site advantage’. Furthermore, it may be that, on less disturbed sites, adaptive trait variation among populations would appear more pronounced, because conditions would resemble those under which selection has taken place. Indeed, it has been suggested that the practice of sourcing seed locally to conserve adaptive genotypes may be less beneficial on highly disturbed sites (Lesica & Allendorf 1999). Future studies should therefore examine performance of plants at multiple sites and at different disturbance levels to understand better these potential sources of variation.

summary and conclusions

We present strong evidence that the provenance of seed is associated with significant variation of ecologically important traits in E. marginata, a finding consistent with studies of adaptive variation in a broad range of plant species (Keller, Kollmann & Edwards 2000; Jones, Hayes & Sackville Hamilton 2001). The development of general guiding principles for sourcing material for restoration remains difficult because the distribution of adaptive genetic divergence varies among species and landscapes; however, our results do support an argument for ‘habitat matching’. That is, collecting from sites with similar environmental characteristics to the site being restored. We support previous suggestions that recording the source of seed deployed in restoration should become common practice (McKay et al. 2005) and, given our finding that different types of ecologically important traits of E. marginata varied at different geographical scales, we suggest that ongoing monitoring of the restored populations includes assessment of a variety of measures of performance, across different life-history stages. These practices will lead to greater understanding of the distribution of adaptive genetic variation within plant species and improved success of ecological restoration.


Many thanks to John Koch (Alcoa) and Stephen Vlahos (Worsley) for arranging site access, and to Nicole Johnson and Myles Menz for help with field measurements. Thank you to Alcoa for making establishment of this trial possible and for their ongoing commitment to genetic conservation in mine rehabilitation. Mike Johnson, John Koch and three anonymous referees provided helpful comments on this manuscript. Funding was provided by the WA Department of Conservation and Land Management, Alcoa World Alumina Australia, Worsley Alumina Pty Ltd and an Australian Research Council Linkage Grant ( LP0214131) to Mike Johnson (UWA) and Siegy Krauss. Eleanor O’Brien undertook this work with the support of an Australian Postgraduate Award.