Journal of Ecology
Volume 88, Issue 6 p. 964-977
Free Access

Shifts in trait-combinations along rainfall and phosphorus gradients

Carlos Roberto Fonseca, 

Carlos Roberto Fonseca

Present address: Departamento de Zoologia, Universidade Estadual de Campinas, C.P. 6109, CEP 13081-970, Campinas, São Paolo, Brazil.

Search for more papers by this author
Jacob McC. Overton, 

Jacob McC. Overton

Present address: Landcare Research, Private Bag 3127, Hamilton, New Zealand.

Search for more papers by this author
Bronwyn Collins, 

Bronwyn Collins

Present address: Centre for Plant Biodiversity Research, CSIRO Division of Plant Industry, GPO Box 1600, Canberra, ACT 2601, Australia.

Search for more papers by this author
Mark Westoby, 

Mark Westoby

Department of Biological Sciences, Macquarie University, Sydney, NSW 2109, Australia

Search for more papers by this author
First published: 24 December 2001
https://doi.org/10.1046/j.1365-2745.2000.00506.x
Citations: 2
Mark Westoby (e-mail: [email protected]).

Summary

1 If different factors inhibiting plant growth, e.g. low rainfall or low soil nutrients, were to select for species that have similar constellations of traits, then the unfavourable factors might usefully be grouped together as ‘stress’.

2 We offer a method for assessing this issue. A species mixture at a site is described by a point on a plane with two traits as axes. Change along an environmental gradient is then represented as a trajectory across the trait-plane. Trajectories along different environmental gradients are compared.

3 We measured leaf width, specific leaf area (SLA) and mature canopy height for the 386 perennial species found at 46 sites spread along rainfall and soil total phosphorus gradients in south-eastern Australia. Each trait was lognormally distributed across species within sites, hence the mean of log10(trait) satisfactorily described the species mixture at each site.

4 Combinations of assemblage-mean leaf width with SLA followed similar trajectories as rainfall and soil total P decreased. For these traits in this setting, the method indicated that low rainfall and low soil P favour similar trait-combinations.

5 Mature plant height also decreased along both rainfall and soil P gradients, and thus was positively correlated with leaf width and SLA at the level of assemblage means. The rainfall trajectories involving height behaved differently from the soil P trajectories, especially at rainfalls below c. 400 mm year−1, where assemblage-mean height declined much further than at low soil P.

6 Across all species, traits were only very loosely correlated (r2 from 0.04 to 0.17). For leaf width and SLA, evolutionary divergences were positively correlated, both before and after cross-correlation with divergence in rainfall and soil P was removed. This latter measures evolutionary divergence correlation within habitat. For height the picture was more complicated. Considering these within-habitat divergence correlations, species that were taller at maturity tended to have lower SLA and leaf width. This pattern is the reverse of the broad geographical correlation of assemblage means, showing that the patterns across assemblages result from species being selectively sifted from the regional flora into sites, not from evolutionary or cross-species correlations.

7 The trait-combination trajectory approach showed some commonalities between low soil nutrient and low rainfall habitats with regard to traits favoured in species occurring there, but also some differences. The approach has potential for clarifying which environmental factors can usefully be grouped together as ‘stress’, and which trait combinations can usefully be regarded as part of a syndrome favoured by stress.

Introduction

A long-standing issue in plant ecology is the ‘stress debate’. Some have argued that ecological responses to different factors inhibiting plant growth, such as low rainfall, low soil nutrients, low temperature, salinity or waterlogging, will have much in common (Grime 1977, 1979, 1989). Although the physiological pathways through which these factors have immediate impacts on plant growth are clearly different (Osmond et al. 1987; Taiz & Zeiger 1991; Larcher 1995; Lambers et al. 1998), ecological success under these varied conditions might require certain common traits, such as slow turnover of leaves and other plant parts, and slow growth response when favourable growth conditions are provided (Grime et al. 1988; Chapin 1991; Chapin et al. 1993). Those emphasizing these common traits see it as worthwhile to set up a category of stress-tolerating plant strategies. At the other end of the spectrum of views, it is argued (e.g. Grubb 1977, 1985, 1992, 1998; Harper 1981) that many adaptations are specific to different stress factors, and that plant strategy schemes are oversimplifications.

The present paper proposes a method for constructively addressing this issue. The method is implemented for three traits, for perennial species along rainfall and soil nutrient gradients in New South Wales, Australia. The essence of the approach is to characterize the response of pairs of traits, rather than of one trait at a time. If two traits are plotted against each other, then shifts along the environmental gradients towards lower rainfall or lower soil nutrient can be expressed as trajectories through the two-trait space (Fig. 1).

Details are in the caption following the image
Figure 1
Open in figure viewerPowerPoint

Schematic illustrating possible outcomes for trajectories along gradients of decreasing rainfall and soil nutrient through a plane described by assemblage means for two hypothetical traits, A and B, both of which are known to decline along both gradients. Both trajectories travel therefore from upper right towards lower left (arrows), beginning from the same high-rainfall, high-nutrient location. Outcome (a): the trait-combinations follow similar trajectories. In this sense low rainfall and low soil nutrient favour species having similar trait combinations. Outcome (b): the trait-combination trajectories diverge, with low rainfall favouring different trait-combinations from low soil nutrients.

That particular traits tend to change along geographical and environmental gradients is perhaps the oldest observation in ecology. Specifically, each of the three traits considered in this study − mature canopy height, leaf width and specific leaf area − tends to decline towards both lower rainfall and lower soil nutrients (e.g. Cain & de Oliveira Castro 1971; Werger & Ellenbroek 1978; Cowling & Campbell 1980; Dolph & Dilcher 1980; Givnish 1987; Wolfe 1993; Smith et al. 1998; Cunningham et al. 1999).

The question whether an individual trait responds similarly along rainfall and soil nutrient gradients has no quantitative answer, since the gradients are in different units (mm year−1 of rainfall vs. μg g−1 of soil P concentration, for example). A question that is capable of being answered, is whether trait shifts in response to rainfall gradients and soil nutrient gradients follow similar trajectories through a two-trait space. The effect of this formulation is that one of the traits is being used to calibrate the two environmental gradients against each other. Suppose one travels a distance along each gradient sufficient to reduce one of the traits by 50%: is the change in the other trait then similar (Fig. 1a) or different (Fig. 1b)?

Response of plants to environmental factors can be construed at a range of time-scales, including growth responses over weeks to months, morphogenetic responses over months to years, population-dynamical responses over years to centuries, and evolutionary-divergence responses over millions of years. Differences in trait-mixtures between present-day local assemblages are shaped primarily by the sifting process that over time-scales of perhaps 102−104 years permits a small minority of species with particular traits to establish populations at a site, while other species with other traits are excluded. For the results reported in this paper, shifts in the value of a trait value along an environmental gradient are mainly the product of such species-sifting.

The focus on trait-mixtures in local assemblages also explains why, by way of environmental factors, we have preferred to measure those long-term properties of the site that are determined by climate (in this case mean annual rainfall) and substrate (in this case total soil P) and that are independent of the species-mixture that has assembled at the site. For physiological or growth responses it might be preferable to know about currently available soil nutrients or even about tissue nutrient concentrations, but these are much influenced by the vegetation present at a site. Similarly, we have regarded soil P as a more fundamental site descriptor than soil nitrogen, despite the importance of nitrogen for plant functioning on time-scales of weeks to years, since nitrogen pools are products of the existing vegetation.

Physical setting and choice of traits

This study was conducted in New South Wales (NSW), south-eastern Australia. Within NSW, rainfall declines from c. 1500 mm year−1 at the eastern seaboard to c. 200 mm year−1 in the western arid zone. A range of rock types can be found within each part of this rainfall gradient, offering contrasts in soil nutrients. To reduce the influence of temperature, only sites between 30 and 35°S latitude and below 600 m altitude were included. Vegetation types included temperate rainforest and tall eucalypt forest towards the high-rainfall, high-soil-P corner of the environmental space, sclerophyll heaths with open eucalypt overstorey towards the high-rainfall, low-soil-P corner, and open grassy woodlands and chenopod shrublands towards the low-rainfall end. With the exception of rainforest and chenopod shrublands, natural vegetation in the region has developed under fire regimes with average between-fire intervals of between c. 5 and 100 years. Most species have life histories that include resprouting or seedling establishment after fire. There is not much species turnover during intervals between fires, although the relative percentage cover contributed by different species changes. Sites that were less than 5 years post-fire were avoided.

The three traits recorded for each species were leaf width, specific leaf area (SLA, leaf area per unit leaf dry mass) and potential height of the canopy of the plant. These traits were chosen for their ecological significance (outlined below) but also because it was practical to estimate them for a large number of species at a single visit to a site. Differences in these traits between species are expected to be meaningful, even though each is modulated to some degree in response to the environment in which the leaf or the plant develops. It is well known from general observation that species occurring at lower rainfall and on lower-nutrient soils tend to be lower in height and to have narrower and thicker or denser (more sclerophyllous) leaves.

Reasons for regarding SLA and height as ecologically important have recently been reviewed by Westoby (1998). Higher SLA (thinner lamina or lower tissue density or both) reflects a faster potential rate of return on investment in leaves. Lower SLA leaves are more robust and their revenue stream lasts longer. Mature canopy height refers to the top of the general foliage canopy of the species, considering full-grown individuals. Potential height has been widely recognized as an important trait when comparing species (e.g. Hubbell & Foster 1986; Grime et al. 1988; Keddy 1989; Bugmann 1996; Chapin et al. 1996).

With regard to leaf width, theory has been developed by Parkhurst & Loucks (1972), Givnish & Vermeij (1976) and Givnish (1978, 1979, 1984). Broader leaves have thicker boundary layers of still air. Consequently their convective heat loss is slower, and they will tend to be heated above air temperature by a wider margin. This may be a disadvantage because with increasing temperature respiration rates increase more than rates of photosynthesis. This effect will be more important for leaves in a stronger radiation stream, and where water is in short supply (since transpiration is also effective in shedding heat). Givnish (1978, 1979, 1984, 1987) expressed the costs of slow convective heat loss in terms of the carbon expenditure in the root system that would be needed to supply the leaf with sufficient water for cooling by transpiration. These costs were balanced against photosynthetic carbon gain in assessing optimal leaf size. As rainfall decreases, increasing water acquisition costs are expected to favour smaller leaves. As soil nutrient content decreases, mass per leaf area increases and the concentration of photosynthetic machinery per mass of leaf decreases. This flattens the response to temperature of potential photosynthesis per mass. Since respiration losses remain unaffected (in Givnish's treatment), optimal temperature for net carbon gain drops, favouring narrow leaves. Empirically, it is known that leaf size tends to decrease toward dry, sunny, or nutrient-poor habitats and is responsive to elevation (Schimper 1898; Raunkiaer 1934; Shields 1950; Webb 1968; Walter 1973; Hall & Swaine 1981; Givnish 1984, 1986, 1987).

Although much of the literature discusses leaf size in terms of the area of individual leaves, we chose to measure leaf width, and measured this as the diameter of the largest circle that can be fitted inside the leaf's outline. Leaf width measured in this way takes account of any lobing or indentations that have the effect of reducing the boundary layer, and therefore is more closely related to boundary layer effects.

Methods

Site selection and description

Forty-six sites were chosen in relation to rainfall and geological criteria (Table 1). Sites were spread from east to west across the rainfall gradient. Within a rainfall zone (as defined in Table 2), sites were required to be either in a different geology (categories from 1 : 1000 000 Australian geological series maps, Geological Survey of NSW, Department of Mines) or, if in the same geology, then at least 30 km apart. Efforts were made to locate sites both on rocks with high content of phosphorus and other nutrients (e.g. basalts) and on rocks with low phosphorus content (e.g. sandstones). Sites that were clearly run-on or run-off were avoided. Soil water potentials were not measured because measurements over many years would have been necessary to obtain a long-term site average, and in any event soil water potentials reflect water extraction by the vegetation as well as rainfall, whereas our aim was to use site descriptors that were independent of vegetation. Sites greater than 600 m a.s.l. were avoided in order to stay within a modest temperature band (c. 15–18 °C mean annual temperature).

Table 1. Description of the 46 field sites selected to represent the natural variation of annual rainfall and total soil phosphorus within New South Wales, south-eastern Australia.
Site
no.		 Site
name		 No. of
perennial
species		 Geology Latitude
(dec. °S)		 Longitude
(dec. °E)		 Elev.
(m a.s.l.)		 Soil
total P
(µg g−1)		 Estimated
rainfall
(mm year−1)		 Mean log10
leaf width
(mm)		 Mean log10
canopy
height (m)		 Mean
log10 SLA
(mm2 g−1)
1 Balranald 8 Qa 34.356 143.596 60 242 288 0.326 − 0.32 3.73
2 Bolo Creek 10 Rh 34.150 151.033 50 650 1229 1.515 0.58 4.05
3 Burbie 21 Tb 31.291 148.979 550 316 630 0.631 0.11 3.72
4 Cox's Gap 18 Pus 32.450 150.267 300 247 575 0.654 0.16 3.59
5 Danu 17 Tb 31.317 148.994 600 179 639 0.755 0.28 3.65
6 Honeysuckle 14 Rs 32.399 150.268 230 246 552 0.615 0.23 3.66
7 Kuringai Chase 36 Rh 33.586 151.284 140 95 1231 0.315 − 0.24 3.59
8 Mount Kiera 14 Pui 34.405 150.842 330 725 1692 1.472 0.09 4.27
9 Mungo 1 6 Qa 33.750 143.086 100 365 271 0.285 − 0.33 3.64
10 Rosewood 23 Slc 32.948 146.280 200 196 381 0.587 − 0.09 3.74
11 Weddin Mtns 15 Duh 33.893 148.001 400 99 589 0.326 − 0.03 3.72
12 Yathong 27 Qd 32.590 145.437 150 54 320 0.398 − 0.21 3.67
13 Cedar Brush 18 Tl 31.851 150.686 800 2063 863 1.101 0.18 4.06
14 Mungo4 14 Qd 33.712 143.175 100 111 270 0.396 − 0.49 3.62
15 Mungo3 9 Qd 33.785 143.128 100 216 272 0.313 − 0.10 3.64
16 Dolo Hills 15 Co 31.726 142.675 150 253 222 0.396 − 0.01 3.65
17 Little Topar 9 Qd 31.796 142.161 150 173 208 0.346 − 0.53 4.01
18 Acacia Creek 6 Qd 31.817 142.071 150 330 206 0.236 − 0.63 4.01
19 Netallie Hill 8 Dum 31.567 143.260 100 259 226 0.212 − 0.17 3.72
20 Purnawilla 10 Ts 31.355 143.467 100 269 230 0.222 − 0.12 3.79
21 Hermidale 12 θg 31.562 146.504 250 454 389 0.598 0.46 3.80
22 Cumberland 22 Rw 33.746 151.039 120 216 1064 1.171 0.46 4.11
23 Cobar 10 Slc 31.510 145.798 250 407 341 0.290 0.06 3.67
24 Nyngan 10 Qa 31.521 147.209 170 488 404 0.304 0.14 3.71
25 Kalyanka 5 Qa 31.460 143.503 80 212 229 0.425 − 0.31 4.00
26 Trangie 16 Ts 31.973 148.037 220 408 479 0.54 − 0.04 3.85
28 Emmdale 8 Qd 31.682 144.225 80 123 250 0.151 0.19 3.68
29 Brummagen 16 Qa 32.239 148.363 300 431 527 0.537 − 0.01 3.76
30 Kayrunnera 5 εzb 30.673 142.539 200 185 237 0.186 − 0.65 3.74
31 Copper Mine 9 Co 30.852 142.585 300 276 241 0.580 − 0.61 3.88
32 W Copper Mine 10 εzb 30.858 142.579 250 275 232 0.425 − 0.41 3.85
33 S White Cliffs 13 Ti 31.084 143.089 80 234 216 0.334 0.12 3.67
34 White Cliffs 6 Klr 30.907 143.062 150 382 228 0.256 − 0.41 3.78
35 Matakana 12 Qd 33.020 146.032 160 142 369 0.415 0.35 3.57
36 Narriah Mtns 7 θu 33.882 146.702 350 401 475 0.720 0.47 3.84
37 Lake Mere 14 Qd 30.271 144.907 100 228 269 0.423 − 0.54 3.88
38 Tundalya 14 Qa 30.731 144.834 90 407 262 0.557 − 0.77 4.08
42 Myall Lakes 17 Qa 32.514 152.327 15 22 1374 0.616 0.15 3.73
43 Copeland 21 Do 31.982 151.816 520 753 1291 1.051 0.16 4.00
44 Strickland 25 Rh 33.373 151.333 120 278 1196 1.478 0.44 4.10
45 Macquarie Pass 21 Pui 34.567 150.656 265 754 1578 1.388 0.49 4.18
46 Ourimbah 29 Qa 33.264 151.292 290 40 1251 0.607 0.05 3.72
47 Coffs Harbour 30 Db 30.321 152.972 195 249 2034 1.272 0.12 4.11
48 Nambucca Heads 39 Pls 30.636 153.002 30 152 1429 1.100 0.13 4.13
49 Stuart Point 12 Qa 30.837 152.988 10 24 1327 0.142 0.11 3.87
50 Morton 39 Ps 34.778 150.336 315 42 1821 0.399 − 0.20 3.83
  • Abbreviations for geology are: co, Carpentarian – Phyllite, shale, chert, quartz muscovite schist; Db, Devonian – Siliceous argilite, slate, minor siliceous greywacke; Do, Devonian – Laminated siltstone, sandstone, minor limestone; Duh, Devonian – Sandstone with siltstone, minor conglomerate; Dum, Devonian – Quartzite and sandstone, pebbly to conglomeratic in part, shale ans siltstone; εzb, Cambrian – Diorite, dolerite, basalt; Klr, Cretaceous – Siltstone, sandy shale, minor sandstone, some thin coal seams; Pls, Perminan – Slate, phyllite, schistose sandstone, schistose conglomerate; Ps, Permian – Lithic sandstone, feldspathic sandstone, sandy mudstone, shale, quartz sandstone, conglomerate, pebbly siltstone, latite flows; Pui, Permian – Lithic sandstone, shale, carbonaceous shale, coal conglomerate tuff; Pus, Permian – Sandstone, shale, conglomerate, coal; Qa, Quaternary – Alluvial and riverine plain deposits of gravel, sand, silt and clay; claypans and outwash areas of black and red clayed silt and sand; coastal sand dunes and beach deposits; Qd, Quaternary – Flat to gently undulating plains and dunes of red and brown clayed sand, loam and lateritic soils; largely aeolian; θg, Ordovician – Schist, phyllite, quartz greywacke, quartzite, slate, minor altered basic volcanics; θu, Ordovician – Quartzose greywacke, siltstone, slate chert, quartzite, phyllite, hornfels, schist, sanstone, mudstone, shale; Rh, Triassic – Massive quartz sanstone, minor shale lenses; Rs, Triassic – Massive quartzose sandstone, flaggy sandstone, siltstone, shale; Rw, Triassic – Shale, lithic sandstone; Slc, Silurian – Feldspathic and lithic greywacke, siltstone, mudstone, argillaceous and quartzitic sandstone, basalt quartz pebble conglomerate; Tb, Tertiary – Basaltic dolerite, minor trachyte, teschenite, phonolite, andesite flows, plugs and sills, minor tuff, breccia, diatomite; Ti, Tertiary – Silcrete, ferricrete, porcellanite; Tl, Tertiary – Basalt, dolerite, polymictic conglomerate, quartzose sandstone, shale, bole; Ts, Tertiary – Gravel, sand, clay, poorly consolidated conglomerate, sandstone and siltstone.
Table 2. Definition of rainfall and soil P gradients. The rainfall gradient was arbitrarily divided into six zones of decreasing yearly rainfall (see Fig. 3) but presenting similar soil P levels (significance tests by one-way anova). The soil P gradient was arbitrarily divided into three zones of decreasing soil P concentration but presenting similar rainfall levels
Zone Rainfall gradient Soil P gradient
No. of
sites		 Log10 rainfall
(mm year−1)		 Log10 soil P
(µg g−1)		 sites No. of
Log10 rainfall
(mm year−1)		 Log10 soil P
(µg g−1)
1 4 3.22 ± 0.03 2.58 ± 0.17 4 3.16 ± 0.03 2.86 ± 0.02
2 5 3.08 ± 0.01 2.49 ± 0.17 5 3.13 ± 0.05 2.27 ± 0.09
3 8 2.74 ± 0.02 2.42 ± 0.08 4 3.16 ± 0.04 1.49 ± 0.07
4 5 2.58 ± 0.01 2.48 ± 0.11
5 6 2.44 ± 0.01 2.38 ± 0.08
6 12 2.36 ± 0.01 2.38 ± 0.04
F-ratio 630.5 2.57 0.001 173.4
P < 0.001 NS NS < 0.001

For each locality, sites were selected where vegetation reflected species composition prior to habitat modification by people of European origin. At each site, five soil cores of 10-cm depth were collected and pooled. Subsamples were air-dried and ground to 0.2 mm then digested with sulphuric acid and hydrogen peroxide (Anderson & Ingram 1989). Total soil P was determined colourimetrically by an adaptation of Murphy & Ridley's (1967) single digestion method (Anderson & Ingram 1989). For each site an estimate of mean annual rainfall was obtained by interpolation between neighbouring rainfall stations with the use of the software ESOCLIM (CSIRO Division of Wildlife and Ecology, Canberra). The database and interpolation algorithm is as described for the related software BIOCLIM (Hutchinson & Bischof 1983).

Total soil phosphorus was adopted as the indicator of soil nutrients at each site since: (i) phosphorus is provided from the parent rock that characterizes the site, in contrast to nitrogen, which is substantially provided by nitrogen fixation (that is, by the vegetation itself); (ii) although only some soil phosphorus is readily available for plant growth and other short-term physiological processes, the total represents a pool of resource that has been available over the hundreds to thousands of years during which the present-day species mixture at the site has been assembled; (iii) other rock-derived nutrients, such as magnesium, are strongly correlated with phosphorus in rock, being influenced mainly by the proportion of nutrient-bearing minerals vs. quartz in the rock; and (iv) phosphorus is regarded as the most important single nutrient in shaping vegetation in temperate Australia (Beadle 1954; 1966).

Field sampling

At each site, a single quadrat of 50 × 20 m (0.1 ha) was surveyed with the aim of finding all vascular species apparent above-ground. Vouchers collected for subsequent identification are deposited in Macquarie University herbarium. Percentage canopy outline cover of each species within the plot was estimated by eye. The present paper reports results for perennials only. Ephemerals were excluded because it was not possible to sample them consistently (they were not growing at all sites at the time of visit). In addition, their traits would not relate to year-round rainfall in the same way as for perennials. (Nearly all perennials are evergreen within the study area). Epiphytes, parasites and vines were excluded on the basis that canopy height estimates do not reflect the same height-support trade-off as for self-supporting species. Also, parasites like mistletoes do not pay the root-construction costs for supporting their transpiration needs.

The height of uppermost leaves was recorded for mature individuals of each species. If only seedlings were present within the quadrat, full-grown individuals were sought nearby. Because heights were log-scaled (see below), results were not strongly affected if species had the potential to gain 10–20% over their measured height. Five leaves were collected and pressed for measurements of width, area and SLA. Leaves were taken from full-light situations and, where possible, from random branches on separate individuals (see Westoby 1998, appendix, for discussion of protocols for characterizing leaves and height). Leaf outline images were captured by HP DeskScan II scanner (Hewlett-Packard, Palo Alto, USA) and quantified by Delta-T SCAN image analysis software (Delta-T Devices, Cambridge, UK).

Data analysis: cross-assemblage and cross-species

SLA, leaf width and canopy height were log10-scaled and are presented as such, except when stated otherwise. Traits were log-scaled to reflect relative or proportional difference between species (Westoby 1998). For instance, the competitive relationship between canopy heights of 0.1 and 0.2 m would be more similar to that between 1 and 2 m than to that between 1.1 and 1.2 m. Also, log-scaled frequency distributions were generally symmetrical, with mean close to median.

For analyses presented in this paper, assemblages were characterized by the mean across all species occurring at each site, excluding ephemerals, epiphytes, parasites and vines. Data are presented to show that the mean of log(trait) reflects the median of a symmetrical distribution, and correspondingly that differences between sites are well captured by means of log(trait). For the purpose of comparing categories of sites, replication across sites was relevant, rather than variation between species within sites.

This paper focuses on cross-site patterns, but species-level patterns are also briefly described, to demonstrate that the cross-site patterns are not forced by the availability of only certain trait-combinations in species. Species-level analyses are reported as correlation patterns across all present-day species in the dataset, and also as correlations of evolutionary divergences. To examine evolutionary divergences, species data are re-expressed as a dataset in which each row represents a node, radiation or divergence-point in the phylogenetic tree. Each column is a measure of divergence from that node for a trait or an environmental variable (Ridley 1983; Felsenstein 1985; Grafen 1989; Harvey & Pagel 1991). This method is sometimes called ‘the comparative method’ or ‘phylogenetically independent contrasts’, but we prefer the term correlations of evolutionary divergences (Westoby 1999), which reminds the reader what is being done. ‘Evolutionary’ rather than ‘phylogenetic’ refers to the species traits being described from the locations where species actually occur, rather than in a common garden, such that environmental influences on development may potentially contribute to divergence, in addition to genetically fixed developmental differences. There has been debate (summary and references in Westoby 1999) about the interpretation of correlations of divergences. Our view is that rather than superseding correlations across present-day species, they provide complementary information.

Working phylogeny down to order or family level was taken from Bremer's ‘Uppsala phylogeny’. This phylogeny is based on Chase et al. (1993) and subsequent molecular phylogenies, conservatively interpreted by a well-informed group of systematists. It is updated from time to time and maintained at http://www.systbot.uu.se/classification/overview.html. We used the version of September 1996. Phylogeny within families was presumed to follow current systematic treatments of each family. Phylogeny within legumes was drawn from Crisp & Doyle (1995). Species were excluded if they had not been identified reliably enough to be positioned in the tree. Where a species occurred at more than one site, the log rainfall and log soil P associated with that species were averages across sites. The working phylogenetic tree had 180 radiations. Correlations of divergences were calculated using the phylogenetic regression (Grafen 1989, 1992), default method, implemented through PHYLO.GLM version 1.03 (Grafen 1991). The general linear modelling capacities of phylogenetic regression were able to calculate a correlation between divergences for a pair of traits after first extracting the correlation with divergences in soil P and rainfall – in effect, a within-habitat correlation of divergences – and these within-habitat correlations are presented for comparison with correlations that include across-habitat variation.

Results

The NSW dataset included a total of 386 perennial species in 720 records taken from the 46 study sites. Thus each species appeared on average at fewer than two sites, and differences in mean trait values between sites were overwhelmingly due to differences between species present rather than to plasticity within species. Further, most species overlap was between communities that were quite close on the rainfall-soil P plane (details not presented).

First, we examine variation in each trait within local assemblages in order to show that assemblage mean of log10(trait) is a satisfactory descriptor for sites. We then come to the core results, which are about trajectories of assemblage-mean trait-combinations along rainfall vs. soil P gradients. Following these core results, we summarize relationships between traits at the species level, both cross-species and as evolutionary divergences. The account of species-level relationships is limited to demonstrating that the assemblage-mean patterns cannot be understood as generated by species-level relationships.

Trait variation

All three traits showed wide between-species variation. Across the whole study, SLA varied between species about 50-fold, leaf width about 240-fold, and mature canopy height about 1700-fold. The species frequency distribution of SLA was log-normal (Kolmogorov-Smirnov, Dmax = 0.02, P > 0.05; Fig. 2a). Mean log-transformed SLA was 3.84 ± 0.29 (SD), with means for individual species ranging from 3.00 to 4.67. Log10 leaf width was not quite normally distributed (Dmax = 0.08, P = 0.02; Fig. 2b), having a tendency to be platykurtic (g2 = − 1.07, t = − 4.30, P < 0.001) or weakly bimodal, with means for individual species ranging from − 0.52 to 1.86. Canopy height was lognormally distributed (Dmax = 0.04, P > 0.05; Fig. 2c) with a mean log10 canopy height of 0.03 ± 0.62 (SD), species means ranging from − 1.70 to 1.52.

Details are in the caption following the image
Figure 2
Open in figure viewerPowerPoint

Frequency distributions for (a) specific leaf area, (b) leaf width, and (c) canopy height across 386 perennial species from New South Wales. All histograms span 3.5 orders of magnitude, while each bar corresponds to one quarter of an order of magnitude.

The traits were lognormally distributed across species within assemblages (for SLA and canopy height at all 46 studied sites, for leaf width at 45 out of 46 sites, Kolmogorov-Smirnov one-sample tests, 5% acceptance level). Accordingly, the mean of the log-scaled traits across species within each assemblage provided a good measure of the central tendency, and we used this mean to describe changes of assemblages along rainfall and soil P gradients. Shifts in the mean of logs between communities expressed the essentials of shifts in the frequency distributions. The shifts in means did not result from minority outliers or from shifting shapes of the frequency distributions.

We also investigated the trajectories of assemblage means in which each species was weighted by its estimated percentage cover. These showed qualitatively similar behaviour to unweighted means. Assemblage-weighted mean heights were shifted upwards relative to unweighted means, because high percentage cover species mostly occupied the tallest layer in each assemblage. These weighted means are not discussed further.

Trajectories of assemblage-mean trait-combinations

Although the design of the study had aimed to spread sites as widely as possible in rainfall and soil phosphorus dimensions, the observed range of soil phosphorus was somewhat wider at higher rainfall (Fig. 3). Consequently if one were to compare (say) the highest P-sites across the rainfall gradient, the comparison would actually be somewhat confounded, with P tending to decline in conjunction with rainfall. Therefore, to obtain the fairest comparison, we described trajectories by taking sites in bands, one band extending as far as possible in the rainfall dimension while showing no trend in the P dimension, and the other extending widely in the P dimension while showing no trend in the rainfall dimension (Fig. 3; Table 2). Because the widest spread of soil P occurred at higher rainfall, the lower rainfall sites were not included in the P trajectory. These two trajectories were characterized by grouping sites into zones along each band (Table 2; Fig. 3). The rainfall and the soil P trajectories included 40 and 13 sites, respectively (Table 2).

Details are in the caption following the image
Figure 3
Open in figure viewerPowerPoint

Study sites arranged on the environmental plane defined by mean annual rainfall and total soil phosphorus. Sites were grouped into zones (Table 2) along trajectories for rainfall and soil P gradients.

In the leaf width × SLA plane, both trajectories started at high width and high SLA, and crossed the plane diagonally towards low width and low SLA (Fig. 4a). Along the rainfall trajectory, neither width nor SLA underwent substantial further decline below zone 3 (about 400–500 mm, Table 2). The trajectory did not travel in any distinct direction across the last three zones (Fig. 4a). This lower terminus of the trajectory was in the same width-SLA region as arrived at by the soil P trajectory. Thus assemblage mean width-SLA combinations behaved similarly along rainfall and soil P trajectories.

Details are in the caption following the image
Figure 4
Open in figure viewerPowerPoint

Assemblage-mean trajectories in attribute space along the rainfall (○) and total soil phosphorus (□) gradients: (a) leaf width with specific leaf area, (b) mature plant height with specific leaf area, and (c) mature plant height with leaf width.

Given that SLA and leaf width changed in concert along the two gradients (Fig. 4a), differences in Fig. 4(b–c) arose from the behaviour of canopy height. In the height × SLA (Fig. 4b) and height × width (Fig. 4c) assemblage-mean planes, the most conspicuous difference between the two trajectories was that height declined from zone 4 to zones 5–6 along the rainfall trajectory (below about 300 mm). Height did not decline this far along the soil P gradient. Since SLA and leaf width had stopped declining below zone 4 along the rainfall gradient, the rainfall trajectories swing downwards in Fig. 4(b–c). Zones 1–4 of the rainfall gradient follow a quite similar trajectory to the soil P gradient. There is a suggestion that height declines along the soil P gradient but not the rainfall gradient, so that the trajectories cross over to some extent, but this is not clear-cut.

Trait relationships at the species level

At the same time as mean SLA, leaf width and plant height varied significantly between assemblages, variation between species within assemblages was also substantial. Standard deviations of log10(trait) across species within each site were calculated. Averages across sites for these SDs were 0.22 log10 units for SLA, 0.42 for leaf width and 0.53 for canopy height. For a normal distribution, 95% of the observations lie within two SDs each side of the mean. Consequently, a span of four SDs gives a quantification of the typical spread between species within sites, and can be considered in relation to the spread across all species in the dataset, shown in Fig. 2. For SLA, the 4-SD span within sites averaged c. 0.9 orders of magnitude (log10 units), and comparing this to Fig. 2(a), it can be seen that the within-site spread spans a substantial proportion of the total spread across all species at all sites. Similarly, for leaf width (4-SD span c. 1.7 orders of magnitude, compare to Fig. 2b) and for canopy height (4-SD span c. 2.1 orders of magnitude, compare to Fig. 2c), variation between species within sites was very substantial compared to variation between sites. In hierarchical anova, variation within assemblages accounted for 54%, 57% and 78% of the variation of log leaf width, log SLA and log canopy height across all records, respectively.

Trait relationships considered across all the species in the study included both cross-assemblage effects and within-assemblage effects. Consequently, the cross-assemblage patterns need not necessarily mirror the species-level patterns, and in fact the height-SLA relationships did not.

Across all species in the study, SLA, leaf width and canopy height were only weakly (although significantly) correlated with each other (Fig. 5a–c; cross-species r2 between 0.04 and 0.17, Table 3). For SLA with width and height with width, the correlation was positive (cross-species β in Table 3), which is the same direction as suggested by assemblage means in Fig. 4. For height with SLA, the cross-species correlation was negative, opposite to the pattern of assemblage means in Fig. 4. (The actual β value should not be interpreted too exactly because with these low r2 values the slope estimated by model I regression is considerably shallower than the true model II slope.)

Table 3. Summary of correlations between traits or evolutionary divergences of traits. Calculated by model I regression or by phylogenetic regression. *** P < 0.001, ** P < 0.01, NS = not significant
r 2 Slope (β) d.f. F
Cross-species correlation
Width vs. SLA 0.12 0.67 1, 384 50.2***
Height vs. SLA 0.04 − 0.43 1, 384 15.8***
Height vs. width 0.17 0.45 1, 384 76.5***
Evolutionary divergence correlation
Width vs. SLA 0.15 0.52 1, 178 30.6***
Height vs. SLA 0.006 − 0.11 1, 174 1.11NS
Height vs. width 0.20 0.38 1, 174 42.5***
Evolutionary divergence after removal of correlation with rainfall and soil P divergence
Width vs. SLA 0.05 0.29 1, 176 8.7**
Height vs. SLA 0.10 − 0.48 1, 175 19.8***
Height vs. width 0.08 0.24 1, 176 15.5***

Evolutionary-divergence correlations were quite similar to cross-species correlations, for SLA with width and height with width (Table 3). The evolutionary-divergence correlations after removing correlation with divergence in rainfall or in soil P (Table 3, third group) − in effect, the within-site evolutionary-divergence correlations − were similar but weaker for these trait-combinations.

For height with SLA, the evolutionary-divergence correlation was weaker than the cross-species correlation, to the point of being non-significant (Table 3). On the other hand, the evolutionary-divergence correlation after removing correlation with divergence in rainfall or in soil P (Table 3, third group) was stronger than the cross-species correlation. Thus, there were opposing effects with regard to the relationship between height and SLA, which we interpret as follows. Within sites, taller species tended to have lower SLA, both cross-species and as evolutionary divergences. On the other hand, sites with lower mean height tended also to have lower mean SLA. This effect resided mainly in the lowest rainfall sites (Fig. 4b–c). These low rainfall sites did not contribute a large proportion of species, hence the correlation across all species reflected within-site patterns more than between-site patterns. On the other hand, differences between families and genera were large contributors to the within-site pattern, with trees and herbs usually drawn from different clades. In the evolutionary-divergence correlation these differences between major clades contributed few degrees of freedom compared to the many differences between species within genera. Hence, in the evolutionary-divergence correlation, the within-site negative correlation was weakened compared to the cross-site positive correlation, and the two more or less cancelled out. In the within-habitat evolutionary-divergence correlation (after extracting effects of rainfall and soil P), the cross-site positive correlation had been removed, and the within-site negative correlation re-emerged.

The purpose of the present paper is to examine patterns across sites, specifically trajectories along rainfall and soil P gradients. The above brief consideration of cross-species patterns is not intended to explore them in depth, only to demonstrate that the cross-site patterns were not forced by cross-species relationships, but rather arise from the sifting process whereby only small subsets of the regional flora maintain populations at each site. This is apparent in the weak cross-species correlations (Table 3) such that each trait varies widely at any given level of any other trait (Fig. 5), with almost all possible combinations of two traits available in some species or other. It is illustrated also by the height–SLA relationship, where correlation across site means is positive but correlation across species is negative.

Details are in the caption following the image
Figure 5
Open in figure viewerPowerPoint

Trait combinations at species level across 386 perennial species from New South Wales: (a) leaf width and specific leaf area (r2 = 0.12), (b) mature plant height and specific leaf area (r2 = 0.04), and (c) mature plant height and leaf width (r2 = 0.17).

Discussion

Leaves were described in the present study by leaf width and SLA. Leaves with narrow width were smaller overall, or needle-shaped. Leaves with low SLA had thicker leaf lamina or higher tissue density within the leaf. Smaller, narrower, thicker and denser leaves have all been regarded as aspects of sclerophylly. Some researchers have interpreted sclerophylly as mainly an adaptation for tolerating water shortage (e.g. Shields 1950; Levitt 1972). However, sclerophylly is also found in some wetlands, and in high-rainfall but infertile-soil environments including those on the east coast of Australia, leading other workers to interpret sclerophylly as an adaptation for nutrient-use efficiency (e.g. Beadle 1954, 1966; Loveless 1961; Monk 1966; Chabot & Hicks 1982). Interactions with herbivory may also be important (Turner 1994). The very existence of the term sclerophylly indicates that botanists have recognized a syndrome or constellation of traits; although exact definitions have not been agreed (Turner 1994) because different elements of the constellation often vary independently. Some authors argue that either low nutrients or low water availability can be important in favouring different aspects of sclerophylly (Beard 1983; Mooney 1983; Specht & Moll 1983; Turner 1994).

Our results showed that assemblage means for SLA and leaf width marched together, and more especially, they marched together in the same direction towards decreasing rainfall as towards decreasing soil P (Fig. 4). Thus, at the whole-vegetation level, our results show leaf narrowing and increased leaf mass per area behaving as a consistent syndrome. On the other hand, at the level of species, a given leaf width was associated with a wide range of SLA values (Fig. 5). The species-level scatterplots, and the phylogenetic regressions, contradict any hypothesis that tight linkage between the two traits is forced by developmental processes or by the engineering exigencies for supporting a leaf. Rather, the assemblage-level outcomes must be generated by the processes that sift particular species out of the regional pool into local assemblages.

The trajectories for assemblage-mean trait-combinations of canopy height with SLA and canopy height with leaf width did not behave with the same consistency as the SLA-leaf width combination. From about 400 mm year−1 down to 200 mm year−1 rainfall, community-mean canopy heights continued to decline, to lower levels than reached along the soil P trajectory, and without continuing decline in leaf width or SLA.

In summary we found, as anticipated, some support for each side of the long-standing difference of opinion over ‘stress’. The combinations of leaf width with SLA that were successful at low rainfall were very similar to those that were successful at low soil P, but the combinations of either leaf width or SLA with canopy height were different, at least at the lowest rainfall levels around 200 mm year−1. We hope our approach in this paper can advance the stress debate by illustrating a method for testing whether particular environmental gradients favour similar or different shifts in particular trait-combinations.

Finally, some comment should be made about the very substantial variation in each trait between species within each assemblage. The fact that the assemblage mean shifted in response to physical conditions of the site indicates that the species sifted into the site are drawn selectively from the regional flora with regard to the value of the trait. But if one compares two sites with different assemblage means, the site with lower mean will include some species with a higher trait value than the assemblage mean of the other site. This contradicts the idea that physical properties of the two sites determine in a simple way what trait values are permitted. Some process spreads out the trait frequency distribution within a site into a broad mixture, at the same time as physical properties of the site somehow position the whole frequency distribution. We think it likely that some game theoretic or frequency dependent process is the main cause of the wide spread of species traits within a site. However, as a matter of logic, a wide spread might also arise from a broad frequency distribution of physical conditions across microsites within each site, or from continuing immigration from other sites that have different physical conditions. Data are not available to partition the contributions of these three forces to the spreading out the within-site trait frequency distributions.

Acknowledgements

We are grateful to B. Rice for her good will in helping with plant identifications, sometimes faced with plant materials lacking flowers or seeds. We would like to thank S. Cunningham, G. Ganade, M. Leishman, I. Wright, B. Rice and B. Murray for comments on previous versions of this manuscript and for informal suggestions during the Macquarie Ecology Discussion Group. G. Ganade, S. Cunningham and J. Pickard helped in field work. Soil analyses were conducted by the Soil Laboratory of James Cook University (Robert Congdon). Climate estimates for sites were kindly provided by CSIRO (Margaret Cawsey). The research was supported by an Australian Research Council grant to M.W. This is contribution number 311 from the Research Unit for Biodiversity and Bioresources, Macquarie University.

Received 1 July 1999 revision accepted 4 May 2000