Livestock grazing in subtropical pastures: steps in the analysis of attribute response and plant functional types


  • S. McIntyre,

    Corresponding authorSearch for more papers by this author
  • Sandra Lavorel

    1. CSIRO Sustainable Ecosystems, 120 Meiers Rd, Indooroopilly, Queensland 4068, Australia; and Centre d’Ecologie Fonctionnelle et Evolutive, CNRS UPR 9056 1919 route de Mende, 34293 Montpellier Cedex 5, France, and Ecosystem Dynamics Group, Research School of Biological Sciences, Australian National University, Canberra Australian Capital Territory, 0200, Australia
    Search for more papers by this author

Sue McIntyre, CSIRO Sustainable Ecosystems, 120 Meiers Rd, Indooroopilly, Queensland 4068, Australia (


  • 1In order to investigate the use of plant functional types as an alternative to floristic descriptions of response to grazing, we analysed plant communities from a cattle grazing experiment conducted in grassy eucalypt woodland in subtropical Queensland, Australia. The two variables analysed (landscape position and stocking rate) accounted for similar proportions of variation in the vegetation ground layer, although forbs were more sensitive to slope position and perennial grasses were more sensitive to stocking rate.
  • 2As grazing pressure increased, perennial grasses declined, while the relative proportion of forbs and annual grasses increased. Detailed functional group analyses were conducted for the perennial grass and forb life-forms. Annual grasses were represented by only two species, preventing identification of functional types within this life-form.
  • 3We conducted a five-step analysis for both perennial grasses and forbs as follows: (i) defining grazing-related species response groups; (ii) defining species groups based on natural attribute correlations; (iii) identifying attributes that changed significantly with grazing; (iv) identifying syndromes by relating (ii) and (iii); (v) describing functional types from (iii) and (iv), and assessing them against actual species response.
  • 4Eight grass and eight forb functional types were identified. Of the taxa that had an observed response to grazing, 54% of the grass taxa and 57% of the forb taxa corresponded to one of these functional types in terms of meeting both grazing response and trait criteria. The five-stage analysis provided a comprehensive but complex approach to functional type identification.
  • 5The functional types identified can be summarized as follows. Low levels of grazing were associated with more medium-sized, moderately leafy perennial grasses with wind-dispersed or adhesive seeds, and more erect or twining forbs with large to medium sized seeds. High levels of grazing were associated with: more annual grasses; low-growing leafy perennial grasses with small seeds having no dispersal appendages; mat-forming large-seeded forbs; and low-growing, scrambling, small-seeded forbs.


Plant functional classifications have been derived in a variety of ways for a variety of purposes. One specific aim is to identify response groups containing species which respond in similar ways to chosen environmental factor(s) (Lavorel et al. 1997), with the biological traits of the response groups providing the basis of the functional classification. In the grazing literature, response groups have been described as increaser, decreaser and neutral species (Noy-Meir et al. 1989). In order to identify a functional classification that relates to grazing response, two sets of data are essential: the species and its traits, and the species and its occurrence in relation to environment. A number of options for analysis of these data are then available. Firstly, species response groups can be identified and the trait characteristics of the groups can then be analysed (e.g. Noy-Meir et al. 1989). Alternatively, researchers may choose to identify natural attribute correlation patterns and then look for correlations between the emergent attribute groups and environmental factors (e.g. Friedel et al. 1988). A third approach is to derive a trait-based classification from an examination of the way in which the individual attributes of the measured traits vary with the environment (Trémont 1994; McIntyre et al. 1995). Most experimentally derived functional classifications involve these elements in various combinations (e.g. Díaz et al. 1992; Fernández Alés et al. 1993; Landsberg et al. 1999).

The use of different analytical procedures can make it difficult to compare the effectiveness and consistency of functional classifications across sites. Nonetheless these comparisons are necessary in order to identify robust classifications (Gitay & Noble 1997). In the analysis of grazing response presented in this paper we attempt to improve the transparency and comparability of our analysis by articulating a series of steps incorporating the analyses described above, together with some additional key linking steps. The analysis proceeds by answering five questions.

1.  Identifying species response groups. How does the abundance of a species vary in response to grazing? Species are classified by their response to grazing (e.g. increaser, decreaser).

2. Identifying emergent groups. For the traits measured, what are the natural attribute correlations among the plant species? Species are classified into groups that are determined by these attribute correlations. This classification accounts for correlations that might reflect physiological trade-offs and evolutionary (including phylogenetic) constraints. This approach cannot be substituted for phylogenetically independent contrasts (Harvey & Pagel 1991), but it makes it possible to take into account the linkage between traits rather than considering them as independent.

3. Identifying attribute responses. How does the frequency of attributes vary in response to grazing? Attributes associated with tolerance or intolerance of grazing are identified based on their distribution across grazing treatments.

4.  Identifying syndromes (relating steps 2 and 3). In what way do the emergent groups coincide with the individual attributes that change significantly in response to grazing? Step 3 identifies sets of attributes based on individual trait response, but how are these attributes combined in the vegetation? By comparing them with the actual attribute combinations found in the local flora (emergent groups) it is possible to identify syndromes that occur in real situations.

5.  Identifying functional types (relating steps 1, 3 and 4). This step summarizes the functional types identified in steps 3 and 4 and relates them to individual taxa, and examines how the attribute response groups match actual species response.

The linking stages (steps 4 and 5) are an attempt to explore thoroughly the concept of functional types by reconciling different aspects of vegetation dynamics, namely species’ response to grazing, distributions of attributes among species and the way the abundance of individual attributes changes in response to grazing. In this way, functionality (in our case, biological attributes that are associated with grazing tolerance) can be linked with the species that actually occur in the vegetation. In this study we apply the five-step analysis to data from a cattle grazing experiment in subtropical Queensland. We consider the utility of the analytical approach and of the functional types as descriptors of vegetation change.


Study site

The experiment was located on the property ‘Glenwood’ (60 km W of Mundubbera, Queensland, Australia 25°41′ S, 150°52′ E, average annual rainfall 708 mm). The study site supported open eucalypt woodland [mainly Eucalyptuscrebra, E. melanophloia and E. tereticornis (names sensuHenderson 1997)] with a mixed grass/forb ground layer. The ground layer was dominated by the perennial tussock grasses Heteropogon contortus, Melinis repens (an exotic species) and Aristida spp. Between these larger tussocks, are a number of subordinate grasses and sedges (notably Fimbristylis, Tripogon, Digitaria spp. and Eragrostis spp.) and forbs (most frequently Boerhavia, Evolvulus, Indigofera, Oxalis, Portulaca, Sida and Vernonia). The site had been subjected to commercial levels of livestock grazing and periodic tree control for about 100 years prior to the experiment. Soils are granite-derived and mainly comprise yellow podzolics with a coarse-textured surface. The treatments ran from July 1989 to March 1995. Twelve months prior to the commencement of the experiment, young tree regrowth (diameter < 50 mm) was killed, the site was burnt and left ungrazed until stocking rate treatments were imposed.

In all paddocks, six exotic legume species were oversown into the pasture using a technique that simultaneously drills seed, places fertilizer and sprays a 50-cm wide band of herbicide (glyphosate) (Cook et al. 1993). Fertilizer (molybdenized superphosphate; 0.02% Mo; 8.8% P) was applied at 45 kg ha−1. Legumes were sown in November 1989 and, because of poor establishment under the dry conditions, a second sowing was made in February 1993. Legumes sown (by seed weight, total = 3 kg ha−1) were 50% Siratro (Macroptilium atropurpureum), 17% Wynn cassia (Chamaecrista rotundifolia), 12% fine-stem stylo (Stylosanthes guianensis var. intermedia), 12% Seca stylo (Stylosanthes scabra), 5% Bargoo joint-vetch (Aeschynomene falcata) and 4% Lotononis (Lotononis bainesii). Recovery of the herbicide-treated areas was rapid and the sprayed rows were indistinguishable from the remaining pastures at the time of final floristic data collection (March 1995). The use of sown legumes was not core to the functional analysis (the experiment had several other objectives), but the use of these extra species did provide a better representation of some emergent types in the assemblage.

Experimental design and treatments

The experiment comprised 22 paddocks, each grazed continuously by two Belmont Red steers, which were introduced as yearlings and retained on the site for 2 years, with an annual changeover of the older animal. Paddock sizes varied (0.2–6.6 ha) to give four stocking rate treatments which were modified during the trial by removing or adding steers. The treatments were Very low SR (0.3/0.15 steers ha−1); Low SR (0.3/0.6 steers ha−1); Medium SR (0.6/0.6 steers ha−1); High SR (0.9/0.45 steers ha−1). The two stocking levels listed in brackets represent the regime in the first 4 years and the last 2 years, respectively. There were also some periods of halved stocking rates in years 1–4 that were associated with resting periods for legume establishment (detailed in MacLeod & McIntyre 1997). The six-year trial period comprised a long spell of moderate drought (39 months) followed by 21 months of severe drought. In the two highest stocking treatments, grazing pressures were extremely high and stock needed supplementary feeding to stay alive during the final year of the trial (MacLeod & McIntyre 1997).

The experimental design combined three landscape position treatments and the four grazing regimes. The undulating topography of the site produced a topo-sequence of lower, mid- and upper slopes; treatments were created by fencing the paddocks to include land from only one of these landscape positions. Grazing treatments were allocated randomly within the different landscape positions. Of the 12 possible factorial combinations of grazing regime and landscape position, two paddocks of each treatment except Very low SR on lower slopes were provided.

Field data collection

Floristic data were recorded in March 1989 immediately before the imposition of treatments, as well as at the end of the trial (March 1995). Individual species responses were analysed for the initial data set, and while some species varied significantly with landscape position, there were no species differences between paddocks allocated to the different grazing treatments. Only data for March 1995 are reported in this paper. Species were recorded in 1-m2 quadrats and abundance ranking was undertaken in a 0.25-m2 quadrat nested within the bigger one. Quadrats were located systematically in a grid pattern (25 × 50 m) over the study site and, as the size of paddocks differed, so did the quadrat numbers per grazing treatment (633 quadrats in total, range 34–111 quadrats per treatment). The floristic data set comprised the following:

  • 1. species richness –total number of all species per 1 m2;

  • 2. presence and abundance ranking of non-rare taxa (non-rare taxa were defined as those with a frequency of ≥ 2% over the entire sample). There were 58 non-rare taxa, of which two were annual grasses, 21 perennial grasses and 35 forbs. In each 0.25-m2 quadrat, abundance rankings were derived by assessing the sward visually and ordering the taxa according to the amount of biomass. The top four ranked taxa were given numerical abundance scores ranging from 5 (highest ranked) to 2 (fourth ranked), while remaining taxa (including those present in the larger quadrat) had an abundance of 1. The abundance scores were used for the calculation of attribute abundance.

Trait measurement

Our philosophy for the choice of traits and the hierarchical analysis is described in detail in McIntyre et al. (1999). It emphasizes structural-functional traits (sensuBox 1996) that are easily measured and that may act as surrogates for physiological traits which are more difficult to measure. The hierarchical analysis involves first analysing the response of major life-forms to grazing in terms of abundance (in this study major life-forms were forbs, perennial grasses and annual grasses) and then analysing the response of taxa within the major life-forms. Because of the limited number of annual grasses (two species), trait analysis was undertaken only for forbs and for perennial grasses/sedges. The latter group, referred to in the paper as ‘perennial grasses’, comprised 19 grass and 2 sedge taxa. Traits were measured on plants at the experimental site in the 1995–96 growing season. Measurements were made on 10 widely spaced clones or individuals. Healthy, adult, non-defoliated plants growing in the open were selected for measurement. In the case of traits describing morphological plasticity, grazed, shaded and crowded specimens were also observed between 1993 and 1996. Traits are listed and defined in Table 1. We only analysed traits that showed variation in the constituent attributes.

Table 1.  Traits measured for the two major life-form groups perennial grasses (including perennial sedges) and forbs (non-phanerophyte dicots, monocots that are not graminaceous). For quantitative traits, attribute ranges are given in brackets. Italicised traits were excluded from further analysis
Trait% of taxa with attributeTrait% of taxa with attribute
  • a

    Height: grass/sedge table height = height of bulk of leaves at vegetative stage; forb height: measurement of rosettes did not include inflorescences, partial rosette height did include inflorescences.

  • b

    Lateral spread: diameter of spread of leaves in rosette or clone.

  • c Canopy structure: analogous to subdivisions within hemicryptophytes (Raunkiaer 1934) but applied to annuals and forbs.

  • d ACD (above ground cover density): the degree to which above-ground biomass fills the projected canopy area (McIntyre et al. 1999).

  • e

    Plasticity under grazing: degree to which leaves and stems change shape, orientation and size in response to defoliation (orientation is most relevant to grasses).

  • f

    Fecundity: qualitative assessment of total numbers of seed produced by open-growing plants.

  • g

    Seed dispersal: vector inferred from seed/fruit morphology. ‘Undefined’ species had no identifiable mechanism that could be inferred from morphology.

  • h

    Habit: twiner/scrambler = decumbent, ascending and climbing plants; mat = prostrate stems.

  • i

    Plasticity of ACD: the degree to which plant structure is modified under crowding (e.g. node elongation).

  • j Dormant bud position: height of overwintering buds (sensuRaunkiaer 1934).

  • k

    Active bud position: length of shoot from highest actively growing buds to point of attachment to the ground.

  • l

    Life cycle: annual includes annuals, facultative annuals and semelparous species.

  • Definitions from McIntyre et al. (1999).

Table height (cm)aa Lateral spread (cm)b 
Short (< 20)43Narrow (0–10) 10
Medium (20–40)43Medium (11–25) 48
Tall (> 40)14Wide (26–100) 33
  Very wide (> 100) 10
Canopy structurec ACDd 
Rosette90Low  0
Partial rosette10High100
Protohemicryptophyte 0  
Leaf width (mm) Inflorescence 
Narrow (< 4)57Exposed 81
Wide (> = 4)43Indeterminate 19
Stem/leaf ratio Plasticity under grazinge 
Stemmy19Low 71
Moderately leafy48High 29
Seed length (mm) Fecundityff 
Small (< 1.5)33Low 33
Medium (1.5–3)33Medium 43
Large (> 3)33High 24
Seed dispersalg Vegetative reproduction 
Adhesion29Yes 14
Wind33No 86
Height (cm)a Lateral spreadb 
Low (< 20)71Narrow (0–10) 17
High (= > 20)29Medium (11–25) 29
  Wide (26–50) 23
  Very wide (> 50) 31
Canopy structurec Habith 
Rosette (erect, partial)23Erect 46
Protohemicryptophyte77Scrambler/twiner 31
  Mat 23
Plasticity of ACDi ACDd 
Low63Low 51
High37High 49
Root morphology   
Simple tap root51  
Other type49  
Inflorescence Dormant bud positionj 
Exposed37Therophyte 11
Along leafy stems63Hemicryptophyte 71
  Geophyte  6
  Chamaephyte 11
Active bud positionk   
Low (< 10)17  
Medium (11–30)46  
High (> 30)37  
Fecundity Life cyclel 
Low51Perennial 71
High49Annual 29
Seed length (mm) Dispersal 
Small (< 1)23Adhesion 6
Medium (1–2)37Wind11
Large (> 2)40Ballistic23

Data analysis

Overview of data: relative abundance of major life-forms and species richness

A table of life-form abundance in each quadrat was calculated as the matrix product of the table of species composition of the quadrats (quadrat × species) with a table formed by the binary records of life-forms for each species. A table of life-form relative abundance was then obtained by calculating for each quadrat the total abundance of all life-forms, and dividing the abundance of each life-form by this total abundance. Variations in species richness and life-form relative abundance with stocking rate and landscape position were analysed using generalized linear models (software GLM 3.77; Baker & Nelder 1978).

Characterization of functional types: steps 1–5

The data used to identify functional types consisted of three sets of tables.

  • 1. A table of species composition listing abundance ranks for the 58 most frequent species (columns) over the 633 quadrats (rows). This table was split into two separate tables recording composition of (i) perennial grasses/sedges (21 species); and (ii) forbs (35 species). The two annual grass species were not included in this analysis.

  • 2. A table of environmental variables recording landscape position (LP) and stocking rate (SR) for each of the 633 quadrats. Environmental variables were coded as categories, with three categories for landscape position and four categories for stocking rate.

  • 3. Six tables of species biological attributes (see Table 1 for attributes used), representing the morphological, grazing-related attributes, and regeneration attributes, for the grass and forb groups.

Data analysis proceeded in five steps described below. These were applied separately to the grass and forb data sets.

Step 1. Identifying species response groups (CCA analysis).

First we examined how species composition within each life-form changed in response to the combined effects of landscape position and stocking rate. This was undertaken using a canonical correspondence analysis (CCA, Ter Braak 1987) of the species composition table with respect to a treatment table representing all categorical combinations of landscape position and stocking rate. The amount of variation explained by these variables is calculated as the ratio of the inertia (the sum of eigenvalues) of that analysis over the inertia for a correspondence analysis (CA, Greenacre 1984) of the species composition table (Table 2). This multivariate correlation ratio (MCR, Sabatier et al. 1989) is compared with the expected value under the null hypothesis of no relationship, which is calculated as the ratio of the number of explanatory variables over the number of independent samples (see Lebreton et al. 1991). Groups of species associated with different levels of stocking rate were then identified graphically using a partial CCA (Ter Braak 1988) of the effects of stocking rate within landscape position (denoted SR/LP). Group membership for species marginal to the main groups was verified by directly examining species abundance distributions.

Table 2.  Results of separate canonical correspondence analyses (CCA) on forb and perennial grass data sets. The multivariate correlation ratio (MCR) represents the proportion of variation explained by the individual factors stocking rate (SR) and landscape position (LP), their interaction, and the effects of SR within LP. See text for details
 Perennial grassesForbs
AnalysisInertiaMCR (%)Expected value (%)InertiaMCR (%)Expected value (%)
  • Calculated as the ratio of the number of explanatory variables over the number of sampling units (633 quadrats).

  • CA, correspondence analysis; NA, not applicable.

LP0.13  2.90.470.26  5.30.47
SR0.17  3.90.630.13  2.60.63
LP × SR0.34  7.81.730.46  9.31.73
SR/LP0.22  4.91.730.19  3.91.73
Step 2. Identifying emergent groups (CA analysis).

Emergent groups resulting from the natural correlation patterns among attributes were identified using correspondence analysis (CA), with group membership and the defining attributes being identified graphically from ordinations.

Step 3. Identifying attribute response (GLM analysis).

For each attribute, we fitted a general linear model of the variation in its relative abundance (calculated using the same approach as for life-form distributions) with landscape position and stocking rate as explanatory variables (software GLM 3.77; Baker & Nelder 1978). We were then able to identify lists of attributes that were significantly associated with each stocking rate. Where there were significant interactions between stocking rate and landscape position, the grazing response was checked for consistency of response (in terms of overall trend) within each landscape position. A grazing response was only reported if it was consistent in trend across landscape positions. If different landscape positions were associated with contradictory grazing responses, no grazing response was reported.

Step 4. Identifying syndromes.

The identification of syndromes involved looking at the degree to which trait-emergent groups (CA results) could be related to the sets of attributes that were significantly associated with grazing (GLM results). This was undertaken on a species-by-species basis, with taxon membership of low or high grazing regime (GR) groups being assessed according to the attributes each possessed. A species was labelled as a ‘Low GR’ (or High GR) taxon if it possessed a majority of attributes associated with Very Low SR and/or Low SR (High GR = Medium SR and/or High SR). Species with equal numbers of low and high stocking rate attributes were labelled as ‘not consistent’. Congruence between emergent CA groups and these GLM response groups was then successively examined for morphology, grazing-related and regeneration traits by simple counts of the number of taxa classified as ‘Low GR’, ‘High GR’ or not consistent within each natural group.

Step 5. Identifying functional types.

Functional types were summarized in two stages. First, all the significant attributes that were important in grazing response were collated to describe a grass and a forb ‘ideal’ functional type for each GR level. Species that conformed to that type, and that actually had a grazing response that was consistent with that type were then identified. Very few species conformed to these ‘ideal’ types. Second, we identified a less strictly defined set of functional types that possessed a subset of the significant attributes. These were identified from taxa that had an actual grazing response (CCA analysis) that was consistent with the response predicted according to the syndromes (identified in Step 4) to which they belonged. This broader set of functional types therefore describes sets of species that are consistently associated with a particular grazing response and simultaneously with significant attributes. They represent a fuzzy set around the ‘ideal’ functional types, where membership is based on response and presence of a majority of significant attributes. Each of these functional types represents an alternative combination of attributes associated with the specified response.


Major life-forms and species richness – response to grazing

Perennial grasses and forbs had similar relative abundance, whereas annual grasses form a minor component of the vegetation under each grazing regime (Fig. 1). Annual grasses (χ32 = 149.9, P < 0.001) and forbs (χ32 = 144.8, P < 0.001) increased with increasing stocking rate, whereas perennial grasses (χ32 = 281.3, P < 0.001) declined in relative abundance.

Figure 1.

Mean relative abundance of the three major life-forms: forbs, perennial (Per.) grasses and annual (Ann.) grasses (groups as defined in Table 1) in relation to grazing regime (Very low, Low, Medium and High stocking rates).

Species richness varied significantly with both stocking rate (F3,630 = 18.8, P < 0.001) and landscape position (F2,630 = 12.6, P < 0.001), but the effects of stocking rate were stronger. Mean richness increased as stocking rate intensified from 11.3 to 13.7 spp. per 0.25 m2 in the lowest and highest stocking rates, respectively. Mean richness on mid-slopes was lower than on either upper or lower slopes, where values were similar.

Analyses of perennial grasses

Step 1. CCA analysis of perennial grass composition

Overall, stocking rate and landscape position accounted for similar proportions of the floristic variation in perennial grasses (see MCR in Table 2). The analysis of stocking rate effects within landscape positions (SR/LP) accounted for 4.9% compared to an expected value of 1.7%. The SR/LP ordination graph indicated that the strongest effects of grazing on perennial grasses were evident on the mid-slopes, followed by the upper slopes (Fig. 2). Grazing effects were least evident on the lower slopes. This could be explained by the lack of a Very Low stocking rate treatment on this landscape position. Of the 21 perennial grasses in the data set, 15 showed a stocking rate position response on axes 1 and 2 of the ordination (Fig. 2). Nine taxa were interpreted as being associated with Low GR, and four with High GR, while two grasses had complex interactions with landscape position and were not categorized (see right-hand column in Table 3).

Figure 2.

Step 1: identifying species response groups for perennial grasses. Factor and species ordination in the plane of the first two axes of the canonical correspondence analysis (CCA) for stocking rate within landscape position (SR/LP). Six taxa which showed no response to the factors are marked as unlabelled points (●). Axis 1 captured most of the variance (55%), while axes 2 and 3 accounted for only 15 and 13%, respectively. Allocation of taxa to response groups is summarized in the CCA column of Table 3.

Table 3.  Summary of results of generalized linear modelling (glm), correspondence analysis (CA) and canonical correspondence analysis (CCA) to examine congruency for perennial grasses and sedges (S). Plant names are as in Henderson (1997). The use of Aristida Section (§) names and species therein is described in McIntyre & Filet (1997)
 Morphological traitsGrazing traitsRegeneration traits  
Species/taxon name (% frequency)CAGLMCAGLMCAGLMGLM SummaryCCA†† Result
  • †Group to which the taxon is allocated according to the emergent CA classification (Table 4).

  • ‡How a species would be classified (associated with high or low grazing) according to the number of significant attributes (GLM Table 5) it possesses.

  • Assessment of the species’ grazing tolerance based on an overview of GLM results for all three trait types.

  • ††

    Low GR includes species abundant in the Very Low and/or Low stocking rate, High GR refers to Medium and High in the same way.

  • ‡‡

    Species with grazing/landscape position interactions and inconsistent grazing responses.

Aristida§Aristida (10)LargeInconsistentStemmy non-plasticInconsistentAdhesive seedsLow GRNoneLow GR
Aristida§Arthratherum (18)LowInconsistentModerately leafy non-plasticLow GRAdhesive seedsLow GRLow GRLow GR
Aristida§Calycinae (16)MediumLow GRStemmy non-plasticInconsistentAdhesive seedsLow GRLow GRLow GR
Aristida§Streptachne (9)MediumInconsistentLeafyHigh GRAdhesive seedsLow GRNoneLow GR
Arundinella nepalensis (4)LargeLow GRStemmy non-plasticLow GRLow fecundityLow GRLow GRNone
Bothriochloa decipiens (12)MediumInconsistentStemmy non-plasticLow GRAdhesive seedsLow GRLow GRNone‡‡
Chloris truncata/divaricata (9)LowHigh GRModerately leafy plasticLow GRWind-dispersedLow GRNoneHigh GR
Chrysopogon fallax (15)LowHigh GRLeafyHigh GRLow fecundityInconsistentHigh GRNone‡‡
Cymbopogon refractus (13)MediumInconsistentLeafyHigh GRLow fecundityInconsistentNoneLow GR
Cyperus fulvus (S) (2)MediumLow GRModerately leafy non-plasticLow GRMedium/High fecundityInconsistentLow GRLow GR
Digitaria ammophila (14)MediumLow GRLeafyInconsistentWind-dispersedLow GRLow GRLow GR
D. brownii (27)LowInconsistentModerately leafy non-plasticLow GRWind-dispersedInconsistentNoneNone
D. divaricatissima (41)MediumLow GRModerately leafy non-plasticLow GRWind-dispersedLow GRLow GRNone
Enneapogon arenicola (3)LowInconsistentModerately leafy non-plasticLow GRWind-dispersedInconsistentNoneLow GR
Eragrostis spp. (41)LowInconsistentLeafyHigh GRMedium/High fecundityHigh GRHigh GRHigh GR
Fimbristylis dichotoma (S) (75)LowHigh GRLeafyHigh GRMedium/High fecundityHigh GRHigh GRNone
Heteropogon contortus (70)LargeInconsistentModerately leafy plasticLow GRAdhesive seedsLow GRLow GRNone
Melinis repens (41)MediumLow GRModerately leafy plasticLow GRWind-dispersedLow GRLow GRLow GR
Panicum effusum (19)LowInconsistentModerately leafy non-plasticLow GRWind-dispersedInconsistentNoneHigh GR
Sporobolus creber (2)MediumInconsistentModerately leafy plasticLow GRMedium/High fecundityHigh GRNoneNone
Tripogon loliiformis (28)LowHigh GRLeafyHigh GRLow fecundityHigh GRHigh GRHigh GR

Step 2. CA analysis to identify emergent attribute groups among perennial grasses

Morphologically, the perennial grasses were categorized as large, medium or short plants, reflecting primarily height and, to a lesser extent, lateral spread (Table 4a). Of the grazing-related traits, stem/leaf ratio was the primary defining trait and plasticity under grazing was used to divide the moderately leafy group of species (Table 4b). Classification by regeneration traits resulted in groups being defined by seed dispersal. One of these groups (‘undefined’) was further divided to separate out the ‘low fecundity’ grasses (Table 4c).

Table 4.  Step 2: identifying emergent groups in perennial grasses based on correlations of (a) morphological; (b) grazing and (c) reproductive attributes. Natural attribute combinations were identified from species ordinations using correspondence analysis. The range of attributes represented in the groups is listed below each trait type. Italicized attributes are those possessed by all species in the group. Traits and attributes are described in Table 1 and classification of individual taxa is described in Table 3
(a) Morphological traits characterizing the group
Name of groupTable heightLateral spreadTaxa (n)
Large plantsTallWide–Very wide3
Medium plantsMediumMedium–Wide9
Low plantsShortNarrow–Very wide9
(b) Grazing-related traits characterizing the group
Name of groupLeaf widthInflorescence prominenceStem/leaf ratioPlasticity under grazingTaxa (n)
Moderately leafy, non-plasticNarrow–WideExposed-indeterminateModerately leafyLow6
Moderately leafy, plasticNarrow–WideExposed-indeterminateModerately leafyHigh4
Stemmy non-plasticNarrow–WideExposed-indeterminateStemmyLow4
(c) Regeneration traits characterizing the group
Name of groupSeed lengthFecunditySeed dispersalTaxa (n)
Adhesive seedsMedium–LargeLow–HighAdhesive6
Wind-dispersed seedsSmall–MediumMedium–HighWind7
Medium/high fecunditySmall–MediumMedium–HighUndefined4
Low fecunditySmall–LargeLowUndefined4

Step 3. GLM analysis of individual attribute response in perennial grasses

The majority of attributes varied significantly with both landscape position (LP) and stocking rate (SR), and there were significant interactions between LP/SR for about two-thirds of attributes (i.e. those listed as significant in Table 5). Although many attributes had a significant response to SR, the amplitude of the response was sometimes small. Height, lateral spread, stem/leaf ratio, seed length and seed dispersal had attributes that changed substantially in their relative abundance (right-hand column, italicized text, Table 5).

Table 5.  Step 3: identifying attribute response groups in perennial grasses. Results of univariate generalized linear modelling (GLM) analyses of attributes in relation to stocking rate and landscape position. Attribute response lists the stocking rate(s) at which the attribute was at the highest relative abundance. Italics indicate attribute responses that were of highest amplitude. Asterisks indicate probability (*** P < 0.001; ** P < 0.01; * P < 0.05; NS, not significant)
 Landscape position (LP)Stocking rate (SR)LP × SR 
 χ2 (d.f. = 2)Pχ2 (d.f. = 3)Pχ2 (d.f. = 5)PSR level(s) at which attribute was most abundant
Short 45.1***123.0*** 6.7NSMedium and High SR
Medium471.0***  8.0*15.2**V. low SR
Tall 38.3***135.0*** 9.5**Low SR
Lateral spread       
Narrow 20.1*** 39.6*** 9.8NSMedium and High SR
Medium  2.9NS165.0*** 8.4NSLow and V. low SR
Wide 46.4*** 64.2***25.1***Low, Medium, High SR
Very wide103.0*** 22.9***40.9***Inconsistent with LP
Leaf width       
Narrow leaves119.0*** 33.4***13.8*V. low, Med., High SR
Exposed  6.3*  4.9NS 3.3NSNo pattern
Stem/leaf ratio       
Stemmy 75.4*** 17.6***77.6***Inconsistent with LP
Moderately leafy204.0*** 52.4***17.4**Low SR
Leafy 91.2*** 51.3*** 6.8NSHigh SR
Grazing plasticity       
Low 85.7*** 27.3***19.8**V. low, Low, Med. SR
Seed length       
Small seeds 31.8*** 94.4*** 1.8NSMedium and High SR
Medium seeds 39.6***130.0***19.1**V. low and Low SR
Large seeds 21.8*** 13.0***22.4***Low SR
Low106.0*** 68.3*** 7.2NSV. low SR
Medium 13.8** 32.9*** 9.4NSLow, Med., High SR
High 91.7*** 21.5***14.7*V. low, Low, High SR
Seed dispersal       
Adhesive 38.4*** 77.6***35.6***Inconsistent with LP
Wind 40.1***107.0***26.9***V. low SR
Undefined139.0*** 55.9***14.7*Medium and High SR

More intense grazing was associated with a higher proportion of short plants, leafy plants, small seeds, medium fecundity and/or ‘undefined’ dispersal mechanism. Low levels of grazing were associated with medium to tall plants, plants with medium lateral spread, moderate leafiness, wind dispersal and/or medium length seeds (Table 5).

Step 4. Identifying perennial grass syndromes (relating CA to GLM results)

For each taxon, Table 3 lists the emergent (CA) group to which each grass belongs, as well as a prediction of the stocking rate response based on the significant GLM attributes shown by that taxon. For morphological traits, all the grasses that were classified by GLM attributes as High GR belonged to the ‘Low’ emergent group. All the grasses with Low GR attributes (GLM) were shared between ‘Large’ and ‘Medium’ emergent groups. Nearly half the grasses could not be classified as they were inconsistent with respect to the grazing-related attributes they possessed. They were spread over all three morphological emergent groups.

Amongst the grazing traits, grasses that were classed as ‘High GR’ by their GLM attributes were all in the ‘Leafy’ emergent group. In the regeneration traits there was a high correlation between ‘Low GR’ taxa and two of the emergent groups. The ‘Adhesive’ group were all ‘Low GR’, and the remaining ‘Low GR’ grasses belonged to the ‘Wind-dispersed’ group. The ‘Medium-High fecundity’ emergent group contained all but one of the ‘High GR’ taxa. In summary we were able to relate our lists of response attributes to a set of emergent groups. These represent the response syndromes.

Step 5. Perennial grass functional types

The groups of grass taxa linked to Low GR (nine taxa), High GR (four taxa) and no pattern (eight taxa) identified in the CCA were compared for congruence with the overall attribute-based (GLM) classification (two right-hand columns, Table 3). Of these 21 taxa, 8 were classified congruently and none were classified in a conflicting way (e.g. Low GR vs. High GR). Of the three trait categories, morphology and regeneration most successfully predicted the CCA classifications (50% congruence in each group), and grazing traits were the weakest (28% congruence). When the GLM univariate analysis was used to describe an ‘ideal’ functional type, Melinis repens was the only species matching all the attribute criteria for Low GR, and there was no species that matched the attributes for High GR (Table 6). Six additional functional types were identified as syndromes that had a majority of significant attributes and an actual grazing response as predicted from these attributes (i.e. taxa which showed congruence between two right-hand columns of Table 3). Four of these types were associated with Low GR and two with High GR (Table 6).

Table 6.  Grass and forb functional types derived from (i) univariate trait analysis (Tables 5 and 9) constrained by actual plant response (canonical correspondence analysis) (bold typeface) and (ii) univariate trait analysis broadly related to natural attribute combinations and constrained by actual plant response (see Tables 3 and 7)
Perennial grasses (and sedges) – Low GRForbs – Low GRPerennial grasses – High GRForbs – High GR
Functional typeTaxa matching criteriaFunctional typeTaxa matching criteriaFunctional typeTaxa matching criteriaFunctional typeTaxa matching criteria
Medium to tall grass of medium to wide lateral spread, moderately leafy, with medium to large wind-dispersed seedsMelinis repensTall erect, scrambling or twining forb with low plasticity, exposed inflorescence, low fecundity and medium to large seedsTricoryne elatiorLow-growing, leafy grass of narrow or wide lateral spread with small seeds and no dispersal appendagesLow growing mat-forming forbs of high or low plasticity, high fecundity, producing flowers along leafy stems and small-seedsPortulaca oleracea Portulaca pilosa Epaltes australis
Medium-sized stemmy grass with non-plastic grazing response and adhesive seedsAristida§CalycinaeErect, medium-seeded hemicryptophyte forbsVernonia cinerea Chrysocephalum apiculatumLow-growing, leafy grass with no dispersal appendageEragrostis spp.Mat-forming, large-seeded hemicryptophyte forbRichardia brasiliensis
Medium-sized, moderately leafy sedge, with non-plastic grazing responseCyperus fulvusTwining, large-seeded hemicryptophyte forbsMacroptilium atropurpureum Glycine tabacinaLow-growing, leafy grass with low fecundityTripogon loliiformisMat-forming, large-seeded chamaephyte forbChamaecrista rotundifolia
      Scrambling, small-seeded therophyte forbChenopodium cristatum
Medium-sized, leafy grass with wind-dispersed seedsDigitaria ammophila    Scrambling, small-seeded hemicryptophyte forbChamaesyce drummondii
Low-growing, moderately leafy grass with a non-plastic grazing response and adhesive seedsAristida§Arthratherum      

Analyses of forbs

Step 1. CCA analysis of forb composition

Forbs were more sensitive to landscape position than to grazing (Table 2). The analysis of stocking rate effects within landscape positions (SR/LP) accounted for 3.9% of total variation. The SR/LP ordination graph showed similar general patterns to those of grasses, i.e. the strongest effects of grazing were evident on the mid-slopes, followed by the upper and lower slopes (Fig. 3). Of the 35 forb taxa in the data set, 21 revealed a stocking rate response on axes 1 and 2 of the ordination (Fig. 3, right-hand column in Table 7). Seven taxa were associated with low grazing (i.e. Very Low and/or Low SR) and 14 with higher grazing (Medium to High SR). Two taxa (Sida and Brunoniella) had stocking rate responses that varied with landscape position, and Glossocardia had an intermediate response (more abundant in Low and Medium grazing regimes).

Figure 3.

Step 1: identifying species response groups for forbs. Factor and species ordination in the plane of the first two axes of the canonical correspondence analysis (CCA) for stocking rate within landscape position (SR/LP). Twenty-four of the 35 species studied are plotted on the ordination. The remaining species (●) showed no response to the factors (located close to the origin). Axis 1 captured most of the variance (40%), while axes 2 and 3 accounted for only 17 and 13%, respectively. Allocation of taxa to response groups is summarized in the CCA column of Table 7 (names sensuHenderson 1997).

Table 7.  Summary of results of generalized linear modelling (GLM), correspondence analysis (CA) and canonical correspondence analysis (CCA), which will be used to examine congruency for forbs (names sensuHenderson 1997)
 Morphological traitsGrazing traitsRegeneration traits  
Species/Taxon name (% frequency)CAGLMCAGLMCAGLMGLM summaryCCA†† result
  • †Group to which the taxon is allocated according to the emergent CA classification (Table 8).

  • ‡How a species would be classified (associated with high or low grazing) according to the number of significant attributes (GLM Table 9) it possesses.

  • Assessment of the species’ grazing tolerance based on an overview of GLM results for all three trait types.

  • ††

    Low GR includes species abundant in the Very Low and/or Low stocking rate, High GR refers to Medium and High in the same way.

  • ‡‡

    Species with grazing/landscape position interactions and inconsistent grazing responses.

Alternanthera nana (7)Scrambling/TwiningInconsistentMedium HemisInconsistentMedium-seededLow GRNoneNone
Boerhavia dominii (27)MatInconsistentHigh HemisLow GRLarge-seededLow GRLow GRNone
Brunoniella australis (4)ErectHigh GRLow HemisInconsistentLarge-seededLow GRNoneNone‡‡
Chamaecrista rotundifolia (32)MatHigh GRChamaephytesHigh GRLarge-seededHigh GRHigh GRHigh GR
Chamaesyce drummondii (2)Scrambling/TwiningHigh GRLow HemisInconsistentSmall-seededHigh GRHigh GRHigh GR
Cheilanthes sieberi (5)ErectInconsistentLow HemisInconsistentSmall-seededHigh GRNoneLow GR
Chenopodium cristatum (4)Scrambling/TwiningHigh GRTherophytesHigh GRSmall-seededHigh GRHigh GRHigh GR
Chrysocephalum apiculatum (24)ErectLow GRMedium HemisLow GRMedium-seededLow GRLow GRLow GR
Desmodium varians (8)Scrambling/TwiningHigh GRMedium HemisLow GRLarge-seededLow GRNoneNone
Epaltes australis (7)MatHigh GRMedium HemisInconsistentSmall-seededHigh GRHigh GRHigh GR
Evolvulus alsinoides (33)Scrambling/TwiningHigh GRMedium HemisInconsistentMedium-seededLow GRNoneLow GR
Glossocardia bidens (11)ErectLow GRLow HemisLow GRLarge-seededLow GRLow GRIntermediate
Glycine clandestina (9)Scrambling/TwiningInconsistentHigh HemisInconsistentMedium-seededLow GRNoneNone
G. tabacina (19)Scrambling/TwiningInconsistentHigh HemisInconsistentLarge-seededLow GRNoneLow GR
G. tomentella (8)Scrambling/TwiningInconsistentHigh HemisInconsistentLarge-seededLow GRNoneHigh GR
Gomphrena celosioides (29)Scrambling/TwiningHigh GRHigh HemisLow GRMedium-seededHigh GRNoneHigh GR
Goodenia glabra (17)MatHigh GRHigh HemisInconsistentLarge-seededLow GRNoneNone
Indigofera linnaei (27)MatHigh GRHigh HemisInconsistentMedium-seededLow GRNoneHigh GR
Macroptilium atropurpureum (32)Scrambling/TwiningInconsistentHigh HemisLow GRLarge-seededLow GRLow GRLow GR
Murdannia graminea (14)ErectLow GRGeophytesLow GRMedium-seededLow GRLow GRNone
Oxalis radicosa (25)ErectLow GRMedium HemisLow GRMedium-seededLow GRLow GRNone
Phyllanthus virgatus sens lat (25)ErectInconsistentMedium HemisInconsistentSmall-seededLow GRNoneHigh GR
Portulaca filifolia (17)ErectHigh GRTherophytesHigh GRMedium-seededHigh GRHigh GRNone
P. oleracea (8)MatHigh GRTherophytesHigh GRSmall-seededHigh GRHigh GRHigh GR
P. pilosa (72)MatHigh GRTherophytesHigh GRSmall-seededHigh GRHigh GRHigh GR
Richardia brasiliensis (16)MatHigh GRMedium HemisInconsistentLarge-seededHigh GRHigh GRHigh GR
Sida subspicata (31)Scrambling/TwiningInconsistentChamaephytesHigh GRMedium-seededLow GRNoneNone‡‡
Stylosanthes guianensis (17)ErectLow GRChamaephytesHigh GRLarge-seededHigh GRNoneNone
S. scabra (41)ErectLow GRChamaephytesHigh GRLarge-seededHigh GRNoneNone
Tricoryne elatior (18)ErectLow GRGeophytesLow GRLarge-seededLow GRLow GRLow GR
Vernonia cinerea (29)ErectLow GRMedium HemisLow GRMedium-seededHigh GRLow GRLow GR
Vittadenia pustulata (11)ErectInconsistentMedium HemisLow GRLarge-seededHigh GRNoneNone
Wahlenbergia queenslandica (7)ErectLow GRMedium HemisLow GRSmall-seededHigh GRLow GRHigh GR
Zornia dyctiocarpa (6)ErectHigh GRMedium HemisLow GRMedium-seededLow GRNoneHigh GR
Z. muriculata (25)ErectInconsistentMedium HemisInconsistentMedium-seededLow GRNoneHigh GR

Step 2. CA analysis to identify emergent attribute groups among forbs

The species groups defined by natural correlations of morphological attributes were determined by habit (Table 8a). A group of 11 scramblers/twiners was characterized by leafy stems, low height and canopy density (ACD), and wide lateral spread. The erect group included most of the rosette species, and tended to be non-plastic and compact in area (narrow with high ACD). The third group comprised mat plants, low and spreading with leafy stems. Of the grazing-related traits, dormant bud position was the primary defining trait and active bud position subdivided the largest group, the hemicryptophytes (Table 8b). The groups identified through correlations of regeneration attributes were defined by seed length, with a slight tendency for annual species to be concentrated in the small-seeded group (Table 8c).

Table 8.  Step 2: identifying emergent groups in forbs based on correlations of (a) morphological; (b) grazing and (c) reproductive attributes. Natural attribute combinations were identified from species ordinations using correspondence analysis. The proportion of taxa in each group with the attribute is indicated in brackets. Italicized attributes are those characterizing each group. Traits and attributes are described in Table 1. See Table 7 for classification of individual taxa
(a) Morphological traits characterizing the group
Name of groupHabitHeightPlasticity of ACDCanopy structureACDLateral spreadRoot morphology
Scrambler/TwinersScrambler/Twiner (11/11)Low (9/11)High (8/11)Leafy stems (11/11)Low (9/11)Wide–Very wide (9/11)Tap–Other
Erect plantsErect (16/16)Low to HighLow (13/16)Rosette (6/16)High (10/16)Narrow–Medium (13/16)Tap–Other
Mat plantsMat (8/8)Low (8/8)Low (6/8)Leafy stems (6/8)Low to highWide–Very wide (7/8)Tap–Other
(b) Grazing-related traits characterizing the group
Name of groupDormant bud positionActive bud positionInflorescence prominence
TherophytesTherophyte (4/4)Medium (3/4)Along leafy stems (4/4)
GeophytesGeophyte (2/2)Low (2/2)Exposed (2/2)
Low-growing hemicryptophytesHemicryptophyte (4/4)Low (4/4)Along leafy stems–Exposed
Medium-growing hemicryptophytesHemicryptophyte (13/13)Medium (13/13)Along leafy stems–Exposed
High-growing hemicryptophytesHemicryptophyte (8/8)High (8/8)Along leafy stems–Exposed
ChamaephytesChamaephyte (4/4)High (4/4)Along leafy stems (4/4)
(c) Regeneration traits characterizing the group
Name of groupSeed lengthLife cycleFecundity
Small-seededSmall (8/8)Annual (5/8)Low/High
Medium-seededMedium (13/13)Perennial (10/13)Low/High
Large-seededLarge (14/14)Perennial (12/14)Low/High

Step 3. GLM analysis of individual attribute response in forbs

Many attributes of forbs varied significantly with grazing and landscape position, and interactions were also frequent (Table 9). As in the case of grasses, the significant attribute lists were aggregated into those associated with two lower and two higher stocking rates (SR). Low and Very Low SR were associated with significantly higher abundance of the following attributes: tall height, medium lateral spread, erect or rosette habit, tap roots, hemicryptophytes, exposed inflorescences, perennials, medium length seeds and/or low fecundity. Attributes associated with Medium and High SR were low height, wide lateral spread, mat habit, fibrous roots, flowers along leafy stems, small seeds, high fecundity and annual life-cycle (Table 9).

Table 9.  Step 3: identifying attribute response groups in forbs. Results of univariate generalized linear modelling (GLM) analyses of attributes in relation to stocking rate and landscape position. Attribute response lists the stocking rate(s) at which the attribute was at the highest relative abundance. Italics indicate attribute responses that were of highest amplitude. Asterisks indicate probability (*** P < 0.001; ** P < 0.01; * P < 0.05; NS, not significant)
 Landscape position (LP)Stocking rate (SR)LP × SR 
 χ2 (d.f. = 2)Pχ2 (d.f. = 3)Pχ2 (d.f. = 5)PSR level(s) at which attribute was most abundant
Height      Abundance increases as SR increases
Low 12.2**165.3***41.3***
Lateral spread       
Narrow 16.7***  2.8NS 3.9NSNo pattern
Medium 38.3*** 57.3*** 6.1NSVery low SR
Wide 51.4*** 21.3***10.1NSMedium and High SR
Very wide 80.0***  9.5*14.6*Low SR
Canopy structure       
Rosette 39.7*** 28.0***13.7*Very low and Low SR
Erect 41.0*** 53.6***10.6NSVery low SR
Scrambler/twiner122.9*** 54.3***41.6***Very low and Low SR
Mat142.8***184.2***18.4**Medium and High SR
Low 56.1*** 26.3***26.3***Very low, Low and High SR
Plasticity of ACD       
Low106.0*** 46.9***45.3***Very low, Low and Medium SR
Root morphology       
Tap root 50.8*** 35.3***26.1***Very low and Low SR
Dormant bud position       
Therophyte  6.6NS  0.5NS 0.8NSNo pattern
Geophyte115.6*** 38.0***33.2***Very low SR
Hemicryptophyte  7.9* 20.9*** 1.6NSVery low and Low SR
Chamaephyte  7.5*  5.5NS 2.7NSNo pattern
Active bud position       
Low 71.2*** 22.9***29.2***Very low SR
Medium  9.0*  1.1NS11.8*No pattern
High 44.9***  8.8*15.3*Low, Medium, High SR
Exposed  4.7NS 25.8*** 3.0NSVery low and Low SR
Seed length      Abundance increases as SR increases
Small  1.2NS 34.5***25.2***
Medium173.3*** 25.6***41.5***Very low SR
Large185.8*** 23.4***83.5***Low SR
Low 69.2*** 61.0***39.7***Very low and Low SR
Life cycle       
Annual  1.1NS 23.3*** 3.9NSMedium and High SR

Step 4. Identifying forb syndromes (relating CA to GLM results)

Most ‘High SR’ taxa belonged to the ‘scrambler/twiner’ or ‘mat’ emergent groups (Table 7). ‘Low SR’ taxa all belonged to the ‘erect’ group. For the grazing category, all the ‘High SR’ taxa were distributed amongst ‘chamaephyte’ and ‘therophyte’ emergent groups. The emergent groups defined for regeneration were generally inconsistent with the membership of the GLM-defined response groups, apart from ‘High SR’, which was strongly associated with the ‘small-seed’ emergent group. In summary, as for grasses, we were able to find response syndromes within the emergent groups, although the patterns were not as tightly bound for the forbs. At least two reasons can be suggested for this. First, the forb group contained more species than the perennial grass group. Second, forbs had a much greater taxonomic diversity (e.g. number of families) than the grasses and sedges.

Step 5. Forb functional types

The groups of forb taxa linked to Low GR (6 taxa) and High GR (15 taxa) and no pattern (14 taxa) identified in the CCA were compared for congruence with the overall attribute-based (GLM) classification (two right-hand columns, Table 7). Of the 35 taxa, 57% were classified congruently and less than 3% were classified in a conflicting way (e.g. Low GR vs. High GR). Of the three trait categories, morphology and regeneration most successfully predicted the CCA classifications (48% and 40% congruence, respectively), and grazing traits were again the weakest (34% congruence). When the GLM univariate analysis was used to describe an ‘ideal’ functional type, Tricoryne elatior matched all attribute criteria for Low GR, while three species (Portulaca oleracea, P. pilosa and Epaltes australis) matched the attributes for High GR (Table 6). Six additional functional types were identified as syndromes that had a majority of significant attributes and an actual grazing response as predicted from these attributes (i.e. taxa which showed congruence between two right-hand columns of Table 7). Two of these types were associated with Low GR and four with High GR (Table 6).


Overview of vegetation response

At the lower grazing pressures, the vegetation is typical of perennial grasslands as described by Grubb (1986), Trémont & McIntyre (1994) and Grime (1998) with medium–large perennial grasses dominating the sward and a greater number of subordinate grass and forb species. Grazing pressure increased over time at the higher stocking rates. The drought conditions that affected the entire site limited plant growth and would have contributed to higher grazing pressures over all treatments, but would not have changed the relative impact of the stocking rates. The impacts of grazing pressure were reflected in changes in relative abundance of the major life-forms: there was a reduction in the proportional abundance of the perennial grass matrix and an increase in the dominance of forbs and annual grasses as stocking rate increased. These changes are consistent with the majority of other studies (e.g. Friedel et al. 1988; Facelli et al. 1989; Noy-Meir et al. 1989; Trémont 1994; Lavorel et al. 1997). This pattern is also linked with the responses of individual species. A greater proportion of grass than forb species were decreasers when grazing increased, and vice versa.

Within the broad pattern just described, other more specific responses were also occurring. The major dominant, Heteropogon is a perennial grass that was recorded as non-responsive to grazing (CCA result Table 3). However more detailed analysis of this species, taking into account frequency of dominance, suggests an intermediate response, with highest abundance in Low SR but lowest abundance in High SR (MacLeod & McIntyre 1997). This is consistent with its trait characteristics (GLM results Table 3).

The second major variable built into the experiment was landscape position. Although the topography is relatively gentle, lower slopes are associated with more prolonged frosts and accumulated water and sediments from upper slopes. Upper slopes are drier. Despite these differences, most species occur on all landscape positions, but some vary in their relative abundance at different positions. Compared to perennial grasses, forbs appeared to be more sensitive to the effects of landscape position than to grazing. This may relate to differences in the relative importance of underground organs between the two groups. The majority (70%) of forb species were perennial, typically with die-back of shoots and reliance on rootstocks for survival in winter, while grasses maintain both root and shoot biomass in winter. During winter therefore, perennial grasses may be more vulnerable to grazing, while forbs may be more sensitive to soil drainage conditions, etc.

Patterns of diversity

Although non-dominant species account for less total biomass than the dominant species, they formed the bulk of species richness at the sites, as in most herbaceous vegetation (Grime 1998). Species richness increased significantly with grazing. There was no intermediate disturbance response as in long-grazed tropical grasslands (McIvor 1998), and unlike the hump-backed response to grazing reported for temperate grasslands (McIntyre & Lavorel 1994). The latter study can be noted for its inclusion of sites with a continuous history of extremely light or intermittent livestock grazing (McIntyre & Lavorel 1994). In the study reported here, and in McIvor (1998), the low stocking rate treatments were imposed on grassland that had been subjected to commercial grazing pressures over the previous century. It is likely that filtering of the local species assemblage had already occurred, and that many grazing-sensitive species had been eliminated during this history of commercial grazing (similar to a continuous Low SR but with some periods of much higher grazing pressure). For example, phanerophyte and chamaephyte life-forms were grazing-sensitive elements of grasslands recorded in the temperate study (McIntyre et al. 1995) but virtually absent from the native component of the ‘Glenwood’ grasslands described here. Had they been better represented in the ‘Glenwood’ assemblage, grazing-sensitive species may have contributed to higher total species richness in the Very Low and Low SR. They may also have contributed to floristic changes of greater magnitude than were observed over the grazing gradient.

Which traits were important to grazing response?

Most of the attributes we analysed varied significantly with grazing intensity, but fewer were associated with a large response in terms of changes in attribute relative abundance (Tables 5 & 9). As in other studies (Noy-Meir et al. 1989; Díaz et al. 1992; Fernández Alés et al. 1993; Lavorel et al. 1999), more intense grazing generally favoured smaller species at the expense of tall species. In our study, by uncoupling vertical height from width, we were able to detect differences in response between grasses, in which short, narrow grass tussocks were grazing-tolerant, and forbs, in which low-growing but wide-spreading, long-stemmed species were grazing-tolerant, consistent with the observations of Trémont (1994) in temperate grasslands. A forb trait that was important was plasticity in the density of above-ground biomass (ACD). Low plasticity was associated with the lowest stocking rate. Similarly, high plasticity of ACD was shown to be significantly associated with grazing tolerance in forbs in Portuguese annual pastures (Lavorel et al. 1999). This is contrary to expectations that species that are able to etiolate would be favoured under conditions of increased biomass. This result suggests that, in our flora, plasticity of ACD may be a grazing-related rather than competition-related trait, and that an adaptive response to grazing is to produce many small-leafed shoots with short internodes, as in grasses (Briske 1996). Grasses were not scored for this trait because of invariance in ACD. However, a related trait was plasticity under grazing, which was also associated with heavier grazing.

In both perennial grasses and forbs, intense grazing generally favoured species that produce smaller seeds. In other studies, seed size has been identified as having functional significance in relation to disturbance (e.g. Grime 1979; Westoby et al. 1992), although few studies have examined patterns in response to grazing intensity. Landsberg et al. (1999) found no consistent seed size patterns in response to grazing in semi-arid woodlands. There was no clear picture regarding dispersal mode, and forbs were not analysed as too many species had undefined mechanisms. In grasses, wind dispersal was associated with low grazing intensity, while the absence of dispersal appendages was more common with intense grazing.

Although less important than the morphological and regeneration traits, two grazing-related traits were of interest. In forbs, exposed inflorescences were replaced by flowers located along leafy stems as grazing increased. In the same way that the morphological attributes of grazing tolerance were associated with reduced accessibility of biomass to grazers, reduced inflorescence prominence presumably protects reproductive investment. In grasses, stemmier plants were associated with lower stocking rates. This is a more complex trait to interpret. Stemminess reduces cattle bite size and intake (Stobbs 1973) and therefore acts as a grazing defence mechanism. However, under extremely high, non-selective grazing pressures, avoidance of grazing (through small stature) rather than defence (through the provision of tough stems) may be the dominant survival mechanism (Rosenthal & Kotanen 1994). At the experimental site, stemmy plants such as Aristida§Aristida and Arundinella were initially avoided by cattle where grazing was selective, but as grazing pressure increased in the Medium and High SR treatments, they were progressively grazed to ground level, and declined in frequency.

In the analysis of attribute response, there were many significant interactions between stocking rate and landscape position (30 attributes; Tables 5 & 9). For the majority of these (27 attributes), the general direction of attribute response was consistent across SR levels, with the response exaggerated or flattened across SR levels on different landscape positions. However, for three grass attributes (very wide lateral spread, stemmy plants and adhesive seeds) the responses were inconsistent in different landscape positions and the attributes were therefore discarded as being significant in the identification of syndromes. These three attributes were linked with sets of grass species that had different landscape preferences and different grazing responses.

How do the significant grazing traits align with trait combinations occurring in the assemblage?

The identification of response syndromes requires us to consider whether attributes associated with High and Low GR (GLM analysis) segregate into the different emergent groups. The general pattern was one of segregation. Sets of attributes associated with high and low grazing were predictably contained within one or more CA-defined emergent group. For example, for grasses, all the taxa which were labelled as ‘High GR’ as a result of their morphological attributes belonged to the ‘Low’ (growing) emergent group (Table 3). This pattern of containment was common to the majority of grass and forb taxa. Membership of an emergent group reliably predicts the functional trait categorization (GLM analysis), if one exists. As a corollary, for species with a mixture of High and Low GR attributes (‘Inconsistent’), membership of an emergent group could not be predicted. However, no taxon was inconsistent for all three trait types (morphological, grazing and regeneration).

The functional types

The detailed analysis within the major life-forms was devoted to identifying functional types amongst perennial grasses and forbs. For these groups, the strictest interpretation of a functional type was the combination of all the attributes that showed a significant response to stocking rate (Table 6, bold typeface). However, there is also a need to relate functional types to actual syndromes represented in the flora, rather than using unconstrained lists of attributes (Montalvo et al. 1991; Lavorel et al. 1999) and these are presented in Table 6 (normal typeface). Together, these methods identified eight grass and eight forb functional types, which corresponded in description to 7 grass (no species matched the ‘ideal’ functional type) and 12 forb taxa (Table 6). These taxa had both the appropriate grazing response and functional attributes. Of all the taxa that had an observed (i.e. detected in the CCA) grazing response, 54% of grasses and 57% of forbs fitted the functional type descriptions. However, most functional types corresponded to one taxon or, at most, three actual taxa.

To summarize across all functional types, low levels of grazing were associated with more: (i) medium-sized, moderately leafy, perennial grasses with wind-dispersed or adhesive seeds; (ii) erect or twining forbs with large to medium seeds. High levels of grazing were associated with more: (a) annual grasses; (b) low-growing, leafy, perennial grasses with small seeds having no dispersal appendages; (c) mat-forming, large-seeded forbs; (d) low-growing, scrambling, small-seeded forbs.

Does the methodology work?

The initial analysis of major life-forms provided an essential broad overview and allowed us to account for annual grasses without having to analyse this small group in detail. The independent analyses within perennial grasses and forbs meant that traits and attribute ranges could be tailored for each of the two groups. This approach was validated by the finding of different responses between grasses and forbs for the same trait (e.g. see lateral spread, Tables 5 and 9). Other traits of significance were relevant to one major life-form but not the other, e.g. ACD and canopy structure were invariant in grasses but significant traits for forbs. Splitting the data set into grass and forb groups therefore enabled functional types to be described more effectively.

The five-stage analysis was comprehensive, it enabled the data to be examined from three perspectives and the components of the analysis to be assembled with a reasonable depth of understanding. A drawback of the analysis is the complexity of data presentation that is required to render the process transparent. A problem also remains in that we are dealing with a continuum of species trait combinations, which we are attempting to couple with a continuum of species response (Austin & Gaywood 1994). Therefore any attempt to classify or delineate types involves the arbitrary drawing of boundaries, which may, in turn, make it difficult to compare the results of different classifications precisely.

A final issue relates to functional classifications in general, rather than the five-stage analysis specifically. Has the range of traits tested included all those of major relevance to grazing tolerance? The answer is almost certainly no, and the search for traits with better explanatory power will continue. The search is complicated by the fact that grazing involves many extrinsic mechanisms that have roles beyond the intrinsic features that a plant might possess. Hendon & Briske (1997) argued the need to also consider selective herbivory, herbivore-mediated competitive interactions and drought–herbivory interactions. However, the existence of these factors, amongst the many other complex community interactions, does not invalidate the search for functional types. It just makes it more difficult, and points to the need for continued explicit analyses of different components, as in the case of competition (Keddy et al. 1998).


This paper contributes to the Global Change and Terrestrial Ecosystems Core Research Programme (Task 2.2.1), a project of the International Geosphere-Biosphere Programme. The stocking rate trial was supported by Meat and Livestock Australia North Australia 2 Program (Project CS195). Travel to Australia (S. Lavorel) was supported through C.N.R.S. PICS 495 and the French Ministry of Foreign Affairs. Many people have contributed to the setting up and conduct of the experiment and the collection and processing of data. Particular acknowledgement is due to J.A. Taylor, J.J. Hodgkinson, J.A. Ogden, C.K. McDonald, S.J. Cook, B.R. Smith, G.C. McDowall, D. Waters, L.V. Wallace, P.N. Jones, K.F. Gould, G.K.W. Gerike. Assistance with the collection of trait data was given by K. Best. R.M. Jones and M. Stafford Smith provided comments on earlier drafts of the manuscript.

Received 15 February 2000 revision accepted 25 September 2000