Plant functional traits in Australian subtropical rain forest: partitioning within-community from cross-landscape variation

Authors

  • Robert Kooyman,

    Corresponding author
    1. Department of Biological Sciences, Macquarie University, North Ryde, NSW 2109, Australia
    2. National Herbarium of New South Wales, Botanic Gardens Trust, Mrs Macquaries Road, Sydney, NSW 2000, Australia
    Search for more papers by this author
  • Will Cornwell,

    1. Biodiversity Research Centre, University of British Columbia, 6270 University Blvd., Vancouver, BC V6T 1Z4, Canada
    2. Department of Integrative Biology, University of California, Berkely, CA 94720, USA
    Search for more papers by this author
  • Mark Westoby

    1. Department of Biological Sciences, Macquarie University, North Ryde, NSW 2109, Australia
    Search for more papers by this author

Correspondence author. E-mail: rkooyman@bio.mq.edu.au

Summary

1. Plant functional traits are dimensions of ecological strategy variation and provide insights into the assembly of plant communities. For woody rain forest vegetation of northern coastal New South Wales, Australia, we quantified four continuous traits (leaf size, seed size, wood density and maximum height) for 231 freestanding woody species and documented community composition for 216 plots. Using trait-gradient analysis, we partitioned species trait values between alpha (within-site) and beta (among-site) components. This allowed us to identify both trait shifts along gradients and variation among co-occurring species.

2. Alpha trait components consistently varied more widely than beta components, meaning that trait variation among species within plots was wider than variation in the mean trait values of plots where species typically grow.

3. Beta trait components covaried significantly among leaf area, seed size, wood density and maximum height. For example, species found in habitats with a large mean leaf size were consistently also found in plots with large mean seed size (r = 0.70). Beta correlations show that these leaf, wood and seed traits respond in parallel to the dominant abiotic gradients: soil types, topographic position, elevation and large-patch disturbances such as those caused by cyclones–storms, landslips or fires.

4. In contrast, the alpha components of traits were largely uncorrelated among species. Alpha leaf area was not associated with alpha larger seeds, meaning that leaf area and seed size act as independent axes of differentiation among coexisting species.

5.Synthesis. The different correlation structures for alpha and beta components of traits reflect community assembly processes at different scales. Within sites, assembly processes have not created strong linkages among these traits. But across different sites in the landscape, abiotic drivers have created strong linkages.

Introduction

Trait-gradient analysis (TGA) (Ackerly & Cornwell 2007) partitions species traits into within-site (alpha) versus between-site (beta) components, providing an improved approach to the question of how correlations between different traits across species arise. The utility of the method is that it arrays communities along a one-dimensional trait gradient reflecting the mean trait value of co-occurring taxa, and then partitions individual species mean trait values into alpha and beta components. The community ordination and the units of the resulting species parameters are explicitly framed in terms of trait values. The beta trait component is a measure of a species position across the range of sites in which it occurs, and the alpha trait component measures how the traits of each species differ from those of co-occurring species. In TGA, trait plot means provide an integrated measure of the abiotic and biotic interactions that may influence community assembly. The partitioning of species trait values into alpha and beta components is achieved by describing across-site variation in relation to the trait itself, rather than by reference to environmental variables. As a consequence, the analysis can be conducted even when environmental data are not available or the factors underlying gradients in particular traits remain unknown (Ackerly & Cornwell 2007).

Ecologists have long been interested in the role of functional traits in structuring communities and in sorting species along environmental gradients (Schimper 1898; Cowles 1899; Clark, Palmer & Clark 1999; Lavorel & Garnier 2002; Ackerly 2003, 2004; Phillips et al. 2003; Ackerly et al. 2006; Grime 2006; McGill et al. 2006; Westoby & Wright 2006; Mouillot, Mason & Wilson 2007). Traits reflect both ecological and evolutionary processes of community assembly (Cavender-Bares et al. 2009). This study uses TGA to partition trait variation and focuses on correlation structures among traits at different scales in order to elucidate plant functional diversity and community assembly.

As an example of correlation between traits, consider the ‘leaf economic spectrum’: where several leaf functional traits are correlated across species world-wide (Reich, Walters & Ellsworth 1997; Reich et al. 1998, 1999; Westoby et al. 2002; Wright et al. 2004). The leaf economic spectrum can be considered part of a spectrum from acquisitive (fast) to retentive (slow and persistent) strategies (Grime 1974, 1977, 1979; and see Grubb 1998). The question is how much of the trait correlation arises because different traits need to be coordinated for functional effectiveness, even within sites, versus how much arises because different traits are adjusting in parallel in response to physical geography, geomorphology and soils, and temperature and rainfall gradients?

In large data sets spanning a range of sites, strong correlations across species have been reported between leaf size, specific leaf area, and maximum height (e.g. Fonseca et al. 2000), and between leaf size, wood density, seed size and maximum height (Ackerly et al. 2002; Cornwell, Schwilk & Ackerly 2006; Cornwell & Ackerly 2009). These correlations could arise because: (i) traits respond in parallel to abiotic gradients across sites, while being decoupled within each site (correlated beta components, uncorrelated alpha components); (ii) traits may be correlated within sites, but respond independently to gradients across sites (correlated alpha, uncorrelated beta); or (iii) traits could be correlated at multiple scales (both beta and alpha correlated). In the particular case of the leaf economic spectrum, it is known that the traits involved are indeed correlated both across sites and across species within sites. More generally, robust quantitative answers are not yet available about the contributions of different spatial scales to trait correlation across species.

This study applied TGA to a large subtropical data set (231 species, 216 plots, 4 traits) that captures significant variation in rain forest communities across a regional scale. Our expectation was that partitioning trait components across these scales should identify the extent and strength of trait correlations in relation to both environmental gradients and interactions between co-occurring species. Our aim was to find which trait correlations were general across scales and which were a product only of shared responses to landscape-scale gradients. Traits chosen were seed size, leaf size, height and wood density (Cornelissen et al. 2003). These traits influence the structure, dynamics and assembly of communities (Ackerly et al. 2002; Cornwell, Schwilk & Ackerly 2006). Three of the traits have previously been linked to identified strategy dimensions: (i) seed size is correlated positively with fruit size and negatively with seed output (regeneration strategy); (ii) species with larger leaves tend to have larger twigs, less frequent branching and larger fruits (Corner’s Rules, Corner 1949); and (iii) typical maximum height is connected with light interception strategy (Westoby 1998; Westoby et al. 2002; Wright et al. 2007; Kraft, Valencia & Ackerly 2008; Poorter et al. 2008). Wood density is the fourth trait considered here. It is linked to several inter-related aspects of ecological strategy variation that reflect growth, mortality and succession (Chave et al. 2009). Contributing factors are thought to include allocation to growth versus strength and resistance to pathogen attack (Turner 2001; Falster & Westoby 2005; van Gelder, Poorter & Sterck 2006; Kooyman & Westoby 2009), and the hydraulic properties of species (Wright et al. 2007).

Materials and methods

Study system

The study area in eastern Australia (Fig. 2) has subtropical climate with high rainfall (>2000 mm yearly average) and generally mild temperatures (approx. mean max. 22 °C, mean min. 12 °C) (Bureau of Meteorology (Australia), n.d.). Mean annual precipitation does not vary substantially across the sample. The data represent two broad rain forest community types from the study area, Complex Notophyll Vine Forest (CNVF) and Simple Notophyll–Simple Notophyll Microphyll Vine Forest (SNVF–SNMVF) (nomenclature follows Webb 1978).

Figure 2.

 Scatterplots of species trait values versus abundance-weighted plot-mean trait values for: (a) log10 leaf area cm−2, (b) log10 seed size (mm), (c) (actual) wood density (kg m−3), and (d) log10 maximum height (m); in 216 woody plant assemblages from two rainforest communities in north-east New South Wales. The larger, solid black circles represent the intersection of the mean trait value for the plots occupied by each species (the species βi trait value on abcissa) and the species mean trait value (on ordinate). The small black symbols represent species occurrences. The vertical alignment of the latter represents the species in a plot (Fig. 1 shows this as a vertically aligned rectangle marked as ‘co-occurring species/plot sample’). The horizontal alignment joined by the grey lines represents a species occurrence across the range of plots in which it occurs. The distance of the large black symbol from the = X dashed line is the species αi trait value (that represents the difference between the species mean trait value and its beta value). The = X dashed line represents the slope of the trait gradient defined by the mean trait values of the species co-occurring in assemblages. The range of occupied plots on the x-axis, represented as the length of the grey line, is the species niche breadth. This represents the range and spread of plots occupied by the species along the gradient of trait variation represented in the sample.

Simple Notophyll Vine Forest and CNVF can reach similar stature under optimum conditions. SNMVF is a variant of SNVF that often occurs in more exposed mountain areas on shallow to skeletal soils. It is less species-rich and lower in stature. Most species in SNMFV are shared with SNVF, including all the dominant species. The SNVF data include some minor presence of adjacent eucalypt-dominated wet sclerophyll communities. Following large-scale disturbances that include mineral soil exposure, regeneration in both SNVF and CNVF can include long-lived sclerophyll species from genera such as Acacia (Fabaceae) and Eucalyptus (Myrtaceae).

The SNVF–SNMVF assemblage samples are from the southern flanks of the eroded Mt. Warning (Wollumbin) volcanic caldera and occur mostly on rhyolite-derived soils. The CNVF samples are drawn from a larger geographic area of the far north-east of NSW, and occur predominantly on basalt-derived soils. There is some minor overlap of these soil types (and communities) in the sample, and a number of species are shared between the communities (and across the range of abiotic variables). SNVF is often referred to as Warm Temperate Rainforest. It is dominated by just a few canopy species mostly in Cunoniaceae (dominant), Lauraceae, Myrtaceae and Atherospermataceae. Trees mostly lack buttresses, and vines are generally thin and wiry. CNVF is more structurally complex than SNVF, and has high floristic diversity in all strata. Canopy trees come from many families, notably Malvaceae (Sterculioideae), Meliaceae, Myrtaceae, Lauraceae, Rutaceae, Fabaceae, Sapindaceae, Sapotaceae, Elaeocarpaceae, and Ebenaceae. Many tree species have plank buttresses. Lianas (large woody vines) are common.

The volcanic activity (Focal Peak and Mt Warning) that gave rise to both basalt and rhyolite parent materials occurred c. 30–20 Ma (Stevens 1977). Soils derived from basalt are free-draining fine grained deep red earths (clay loam), and those from rhyolite generally form shallower brown clay loams (Appendix S1 in Supporting Information). The terrain is mountainous and the study sites ranged from c. 200 to 1000 m a.s.l. The upland plateaus are deeply dissected by numerous streams. The streams form a network of large river valleys. Cliffs and deep gorges feature in the headwaters of the major streams.

The most important environmental influences on communities in this landscape (see Appendix S1 and Baur 1957; Beckman & Thompson 1977; Turner & Kelly 1981; Floyd 1990; Kariuki & Kooyman 2005; Rossetto & Kooyman 2005; Kariuki et al. 2006) are (i) topographically mediated variation in soil depth on rhyolite substrates, with shallow to skeletal soils on upper slopes and crests; (ii) increasing soil depth and nutrient concentration moving from rhyolite to basalt parent materials; (iii) altitude; and (iv) disturbances that include wind and storm effects, logging history and fire.

Plot sampling

The 216 plots were established in 2000–05 as part of a larger project to sample regional rain forest diversity and dynamics. A complete species list was compiled (by R.M.K.) for all established plants taller than 1 m, and a Braun-Blanquet cover abundance rank was recorded (1 = < 5% cover and rare, 2 = < 5% cover and more common, 3 = 6–20%, 4 = 21–50%, 5 = 51–75%, 6 = 76–100%).

The TGA presented here deal with the 231 freestanding native tree and shrub species. Herbs, sedges, ferns, epiphytes, orchids, vines, palms and cordylines were removed because trait data for wood density and maximum height were not available or not comparable. Alpha diversity ranged from 18 to 78 species per plot, and there were a total of 7575 species-plot observations. Plots were 0.1 ha except for the geographically constrained SNMVF (100 samples), where plot size was 0.04 ha. In that case, increasing plot size (from 0.04 to 0.1 ha) made little difference to woody species diversity accumulation. Sampling for all plots proceeded from an initial (nested) subplot quadrat sample size of (20 × 20 m) 0.04 ha, and was expanded to 50 × 20 m for the SNVF and CNVF samples. Woody species accumulation with increasing plot size was negligible for SNMVF, low for SNVF, and high for CNVF. Information about abiotic gradients was collected as ranked environmental variables for each plot at the time of sampling. This included information for topographic position, altitude, slope, aspect, soil texture, soil depth and fire frequency (time interval). Altitude was obtained from topographic maps.

Trait data

Trait data for leaf area (LA), wood density (WD) and seed size (SS) were extracted from published floras and other sources including Bootle (1983); Stanley & Ross (1983–89, volumes 1–3); Floyd (1989); Ilic et al. (2000); Harden (1990–2002, volumes 1–4 with revisions). A limitation of this study was that a single value for each trait was allocated to each species, irrespective of site. Leaf size was for the whole area of simple leaves or phyllodes and for the leaflets of compound leaves. Juvenile leaf sizes were excluded from consideration. Leaf size (cm2) was estimated as maximum length × maximum width × 0.70. This formula has been shown to correlate well with photographic area estimates of rain forest tree leaves (e.g. Kraft, Valencia & Ackerly 2008). Leaf material was collected for > 100 of the species in this study and was consistent with dimensions reported in the floras. Seed size was estimated using maximum dimensions of embryo plus endocarp (length + width/2, in millimetres; average diameter). Wood density estimates (as dry weight in kg m−3) were taken from published sources. Estimated maximum height at maturity (Hmax) was based on field information previously collected by R.M.K., which reflected maximum canopy height (m) for species (at largest known diameters, at reproductive maturity). This provided a single value for maximum potential height for each species, irrespective of site.

Analysis

Leaf area, maximum height at maturity and seed size were log10-transformed to reduce skew. Wood density was not transformed. Analyses were carried out in r version 2.7 (R Development Core Team 2006)

Trait-gradient analysis (Ackerly & Cornwell 2007) decomposes trait values into alpha (within sites or plots) and beta (among sites or plots) components. A mean trait value for each plot is calculated across all species at the plot (eqn 1). The plots are then arranged along a spectrum or gradient according to their trait means, forming the x-axis of Fig. 1 (the ‘trait-gradient’ of TGA). Each species at each plot has a point-location in Fig. 1. The x-axis location is the plot mean for the trait, and the y-axis location is the trait value for the individual species. The vertically arranged points at a particular value of pj represent the species co-occurring in a plot (Figs 1 and 2). The ordinary least-squares regression line of tij versus pj (representing the X line) has, by definition, slope 1 and intercept 0.

Figure 1.

 Scatterplot of species trait values versus abundance-weighted plot-mean trait values for log10 leaf area cm−2 in 216 woody plant assemblages (representing two main rain forest communities in north-east New South Wales). Values for three species are highlighted for illustration: Argyrodendron trifoliolatum– Malvaceae (most common species in Complex Notophyll Vine Forest on basalt-derived soils); Cinnamomum oliveri– Lauraceae (occurs in almost all sites across all gradients); Acrotriche aggregata– Ericaceae (occurs only in lowest stature Simple Notophyll Microphyll Vine Forest on skeletal rhyolite-derived soils). Acrotriche aggregata is represented as a large open square, and the plots it occupies are shown as solid grey squares. Using only A. aggregata as an illustration, the large symbol represents the intersection of the mean trait value for the plots occupied by the species (the βi trait value on abcissa) and the mean species trait value (on ordinate). The distance of the large symbol from the = X dashed line is the species αi trait value. This represents the difference between the species mean trait value and its beta value. The = X dashed line represents the slope of the trait gradient defined by the mean trait values of the species co-occurring in assemblages. The range of occupied plots on the x-axis is the niche breadth, shown here as the labelled open arrow and representing the range of plots occupied by the species along and across the gradient of trait variation represented in the sample.

Abundance-weighted plot-mean trait values (eqn 1), species mean trait values (eqn 2), and mean of plot means for plots occupied by each species (eqn 3) are defined as follows:

image(eqn1)
image(eqn2)
image(eqn3)

where tij is the trait value and αij is the abundance for species i in plot j, the total number of plots in the study is P, and the species richness of plot j is Sj. The analyses presented here use cover abundance-weighted values. Analyses were repeated using presence/absence data only and the general patterns and results (not presented) were very similar, as also found by Ackerly & Cornwell (2007).

Beta components (βi, eqn 3) are the x-axis means for each species, illustrated in Fig. 1 by larger symbols (black outline square) for Acrotriche aggregata (R.Br.) Sprengel (Ericaceae) and (black outline triangle) Cinnamomum oliveri F.M. Bailey (Lauraceae). Beta components describe the habitats where each species occurs, in units of the mean trait value across all species at those sites. The niche breadth of a species is then characterized as the range of plots occupied (pj-values; illustrated in Fig. 1 for A. aggregata). Alpha components (αi) measure the deviation of species trait values from the cross-species mean at the site, that is, the deviation in the y dimension from the Y = X line in Fig. 1. Alpha components reflect the differentiation of each species from co-occurring taxa with regard to the trait. The partitioning into components in TGA is made possible by describing across-site variation by reference to the trait itself, rather than by reference to any single environmental variable, which might not be the sole or main cause of trait variation.

In some studies, trait values for each species may be measured separately at each site. Then a slope bi of tij versus pj can be calculated for each species, and this slope measures how closely phenotypic variation within the species is aligned with the trait gradient (Ackerly & Cornwell 2007). However, in other studies, including this one, only a single average trait value is available for each species, so point locations for a species across different sites all lay in a horizontal row (Figs 1 and 2). In situations where large geographic areas are sampled, trait measures from a range of sites that capture the extent of variation may be included.

Results

Across the 231 species sampled, traits ranged > 1000-fold for leaf area, almost 100-fold for seed size, > 20-fold for maximum height at maturity and 5-fold for wood density.

Niche breadths describe species distributions in units of plot means for the trait (Table 1, Fig. 1). Some species occurred only once and consequently had niche breadth measured as zero, while others spanned virtually the whole gradient (Figs 1 and 2). Correlation values were mostly low between niche breadth and species trait means except for wood density; generally negative and high between niche breadth and beta trait values; and low between niche breadth and alpha trait values (Table 2).

Table 1.   Summary statistics for four traits measured across 231 species and 216 plots in north-east New South Wales, Australia. (a) All results and measurements of the various components of TGA in log10-transformed units of the trait(s) except for WD; (b) back-transformed (actual) trait values
(a)Traits (units, transformations)
ParameterLA (cm2, log10)SS (mean) (mm, log10)Hmax (m, log10)WD* (kg m−3)
Species characteristics
 tis mean1.180.671.34724
 tis minimum–maximum−1.78, 2.34−0.30, 1.630.30, 1.65240, 1100
 βis minimum–maximum0.90, 1.340.59, 0.831.19, 1.52670, 790
 αis minimum–maximum−2.80, 1.03−0.98, 0.92−1.22, 0.36−483, 346
 Ris mean0.210.140.1677
 Ris minimum–maximum0, 0.540, 0.320, 0.330, 174
Plot characteristics
 pjs mean1.150.691.33739
 pjs minimum–maximum0.81, 1.350.54, 0.861.19, 1.50653, 827
(b)Traits (units, back-transformed log10)
ParameterLA (cm2)SS (mean) (mm)Hmax (m)WD (kg m−3)
  1. LA, leaf area; SS, seed size; Hmax, estimated maximum height; WD, wood density; tis, species trait mean; βis, beta trait value; αis, alpha trait value; Ris, niche breadth; pjs, plot-mean trait value.

  2. *A total of 206 species with data available for WD.

Species characteristics
 tis mean15.14.721.88724
 tis minimum–maximum0.02, 218.80.5, 42.72, 44.67240, 1100
 βis minimum–maximum7.9, 21.93.9, 6.815.49, 33.11670, 790
 αis minimum–maximum0.001, 10.70.1, 8.30.06, 2.29−483, 346
 Ris mean1.61.41.4577
 Ris minimum–maximum0, 3.50, 2.10, 2.140, 174
Plot characteristics
 pjs mean14.14.921.38739
 pjs minimum–maximum6.5, 22.43.47, 7.215.49, 31.62653, 827
Table 2.   Correlation coefficients across species among mean trait values, beta, alpha, and niche breadth (Rs) for four traits (leaf area; seed size; wood density; maximum height – estimated maximum height at maturity) in the northern New South Wales data set. Species that occurred only once (singletons) were removed from the analysis, as niche breadth values for such species using the trait gradient method are ‘0’; leaving 204 from the 231 species. *P > 0.05–0.1; **P < 0.001; ***P < 0.0001
  Species mean (ti)Beta trait (βi)Alpha trait (αi)
Maximum heightβi0.33***1 
αi0.98***0.131
Rs0.01−0.52***0.11
Leaf areaβi0.42***1 
αi0.98***0.26**1
Rs−0.10−0.62***0.01
Seed sizeβi0.37***1 
αi0.99***0.26**1
Rs0.08−0.40***0.14*
Wood densityβi0.82***1 
αi0.59***0.031
Rs0.39***0.48***−0.01

The range of alpha (αi) was much wider than the range of beta (βi) components for all traits (Fig. 3, Table 1), meaning that trait values varied much more across co-occurring species than across means for plots where they occurred. Nevertheless, the covariance between traits was stronger for the beta components (βi) (Fig. 3).

Figure 3.

 Scatterplots of (a) plot-mean trait values (pj), (b) species beta trait values (βi), (c) species alpha trait values (αi), and (d) species mean trait values for pairwise combinations of leaf size (log), wood density (actual), seed size (log) and maximum height (log). Correlation (r) values are shown.

Pairwise trait correlations

Correlations among plot means for traits (pj) (column a in Fig. 3) indicate whether a pair of trait shifts in parallel along the dominant environmental gradient. Plot means were negatively correlated between leaf area (LA) and wood density (WD) (= −0.52), positively correlated between leaf area and seed size (SS) (= 0.37) and uncorrelated between seed size and wood density (= 0.15) (Fig. 3a).

Beta trait components (βi) (among-site species trait values) were strongly correlated among the same trait pairs (leaf area and wood density = −0.75; leaf area and seed size 0.70) (Fig. 3b), while alpha components (αi) (within-site species trait values) showed much weaker correlations among these traits (Fig. 3c). The LA–WD and LA–SS trait correlations across species (Fig. 3d) therefore predominantly reflected the beta correlation.

For the relationship of wood density with seed size, the direction of the beta-component correlation was opposite to the alpha-component correlation. However, because both components were relatively weak, the overall correlation among species means was only very weakly positive.

Across plot means (Fig. 3a) and across species (Fig. 3d), Hmax was positively correlated with leaf area and seed size. Hmax was negatively correlated with wood density across plot means and uncorrelated across species. Species beta trait values (βi) were the most strongly correlated among the same traits (Fig. 3b), while species alpha trait values (αi) showed much weaker correlations (Fig. 3c). The species mean Hmax–LA–SS–WD trait correlations therefore predominantly reflected beta trait values (βi) (Fig. 3d).

Abiotic factors

Topographic positions lower downslopes tended to be associated with larger plot-mean leaf size and lower wood density (Table 3).

Table 3.   Correlation values for plot trait means (leaf area – tp LA; seed size – tp SS; maximum height – tp Hmax; and wood density – tpWD) for 216 plot-based samples in northern New South Wales, Australia, by eight environmental variables [Topog., topographic position; Altitud., altitude; Soil Te., soil texture; Soil Dep., soil depth; Fire, fire frequency (interval)]
 tp Hmaxtp LAtp SStp WDTopog.Altitud.SlopeAspectSoil Te.Soil Dep.Fire
tp LA0.35          
tp SS0.410.38         
tp WD−0.27−0.520.15        
Topog.−0.020.23−0.02−0.29       
Altitud.−0.41−0.08−0.310.10−0.28      
Slope−0.210.08−0.07−0.060.080.15     
Aspect−0.07−0.05−0.160.12−0.220.350.19    
Soil Te.−0.27−0.27−0.180.41−0.110.26−0.110.15   
Soil Dep.0.440.560.30−0.500.24−0.25−0.20−0.34−0.37  
Fire0.09−0.22−0.060.03−0.04−0.33−0.12−0.10−0.24−0.06 
Disturbance0.160.04−0.06−0.19−0.11−0.14−0.21−0.16−0.320.280.57

Higher altitude tended to be associated with lower height and seed size. Deeper soils and basalt soils were associated with taller plant heights, larger leaf size and larger seed size, and lower wood density (data not shown). All this is consistent with higher productivity sites in the study area (including sites in lower slope to gully topographic positions; and those with basalt-influenced soils) carrying taller species, with larger leaves and seeds, and lower wood density. Lower wood density on higher nutrient (more productive) sites has also been observed in the American tropics (Muller-Landau 2004). A trend to smaller leaves and higher wood density in upslope topographic positions has previously been noted in a study of an Amazonian forest plot (Kraft, Valencia & Ackerly 2008).

Discussion

Partitioning of species trait values

Here, as elsewhere, the range of trait variation within plots (αi) was wider than variation across-site means (βi), meaning that much of the trait variation between species was associated with different functional strategies within a shared environment (Wright et al. 2004; Westoby & Wright 2006). The emphasis here was on between species trait variation relative to community assembly. We did not include values for phenotypic (within-species) variation in this study.

Although alpha components contributed more variation than beta components, correlations between these traits arose predominantly from correlations between beta components. This means that despite the wide scatter among species within each site, correlations between these traits were predominantly driven by their tendency to vary in parallel along wider scale abiotic gradients. The coordinated shifts in mean trait values across abiotic gradients can be thought of as habitat filtering (Díaz, Cabido & Casanoves 1998; Ackerly 2004; Díaz et al. 2004; Cornwell, Schwilk & Ackerly 2006). In contrast, alpha components were largely uncorrelated between traits across species, suggesting that they act as independent axes of differentiation among coexisting species.

In community ecology, a restricted range of trait values is often viewed as evidence for habitat filtering (Cornwell, Schwilk & Ackerly 2006). The convergence of form and function in relation to edaphic and climatic conditions results in stronger trait correlations but reduced trait breadth at landscape (beta) scales. In essence, species that are more similar (i.e. with shared trait attributes) are filtered into habitats along abiotic gradients. In contrast, at local scales (alpha), a broader range of trait variation (and weaker trait correlations) may be interpreted as reflecting the signal of various mechanisms of species coexistence (e.g. Tilman 1994). The explicit partitioning of alpha and beta trait components in TGA provides a clear conceptual basis to interpret trait correlations at different scales, which reflect both trait shifts along gradients and variation among co-occurring species (Ackerly & Cornwell 2007).

Traits, abiotic gradients and habitats

In this Australian subtropical forest vegetation, there is a history of classifying communities by reference to leaf size (Webb 1978), giving confidence that the axis of plot-mean leaf area reflects a major environmental gradient related to community differentiation. Previous studies in these forests have confirmed the influence of soils, topography and altitude (e.g. Baur 1957; Horne & Gwalter 1987; Rossetto & Kooyman 2005; Kariuki et al. 2006; Rossetto et al. 2008), and the differentiation of the CNVF and SNVF communities along a productivity gradient that reflects soil type (Webb, Tracey & Williams 1972; Turner & Kelly 1981).

Across the abiotic gradients sampled, sites with taller species generally also had larger leaves and seeds, and lower wood density. Within each plot, these correlations were much weaker or absent. At alpha scale, no relationship was evident between wood density and height at maturity, indicating that coexisting species use a broad range of wood densities to reach a given height.

Within the whole data set, there was a weak positive correlation across species between seed size and leaf size (Fig. 3d). This might seem to contrast with earlier findings by Ackerly & Cornwell (2007), where seed size varied independently from leaf and wood traits in woody plant communities of coastal California, and by Rossetto & Kooyman (2005), where seed size varied independently from leaf and wood traits within SNVF–SNMVF rain forest. However, the correlation across species in this study arose predominantly from the beta component (Fig. 3b), with species having larger leaves and larger seeds tending to co-occur in the higher productivity community (CNVF). This illustrates how sensitive trait correlations can be to the spread of vegetation types included in a data set and highlights the value of separating alpha from beta components.

Species with larger leaves and lower wood density contributed to the spread of variation within most plots. At the same time, species with this combination of traits were more abundant at some plot-types than others. A partial explanation for this was that some of these species were pioneers (Bazzaz 1979) and were more abundant at recently disturbed plots. However, species with a similar combination of traits were also relatively common at undisturbed plots on more productive soils derived from basalt or deeper rhyolite, while few such species were found in the SNVF–SNMVF communities on poorer soils regardless of disturbance levels (Horne & Mackowski 1987).

The different correlation structures for alpha and beta components of traits reflect processes at different scales. Within sites, community assembly did not create strong linkages among traits. Across the landscape, abiotic drivers created strong linkages among traits. Grime (2006) suggested that the physical and chemical traits that drive ecosystems would in most cases be linked to the site productivity gradient, while disturbance-related traits would be less differentiated across wider scales. Our results were consistent with that prediction. Our study shows how partitioning variation in trait values across different scales can help clarify the linkages among traits that characterize plant functional diversity.

Acknowledgements

We thank NSW Department of Environment and Climate Change (National Parks Section) and Rous Water for access to the sites. The TGA methods were an output of a working group of the ARC-NZ Research Network for Vegetation Function, funded by the Australian Research Council. Comments on an earlier draft by Peter Wilf, Chris Lusk and two anonymous referees significantly improved the paper. R.M.K. is supported by an Australian Postgraduate Award. Thanks to Rainforest Rescue and Andrew Hall who helped support the original data collection and continuing research.

Ancillary