Abundance and composition of plant species along grazing gradients in Australian rangelands


J. Landsberg, School of Tropical Biology, James Cook University, PO Box 6811, Cairns Queensland 4870, Australia (fax + 61 74042 1284; e-mail Jill.Landsberg@jcu.edu.au).


  • 1The widespread provision of livestock drinking water in previously dry Australian rangelands has supported concomitant increases in cumulative grazing pressure. While the associated impacts on plants of pastoral importance have been well documented, far less is known about the rest of the flora.
  • 2To address this deficiency, we measured the frequency of occurrence of all plant species at sites along water-centred grazing gradients in commercial paddocks located in the rangelands of central and southern Australia. Four gradients were in chenopod shrubland vegetation, and four in acacia woodlands. Each gradient extended to a reference site remote from all waters, where grazing by stock was minimal. Five further sites were sampled along each gradient, at locations progressively closer to the gradient's watering point.
  • 3Ground-layer species far outnumbered those in upper layers (466 and 134, respectively). Most were geographically localized (72% found at one gradient only) and locally uncommon (46% per gradient occurred with a frequency < 5% per site).
  • 4Species showing trends of decreasing frequency with proximity to water significantly outnumbered those showing increasing trends. Significantly more species that were recorded only once occurred at the sites furthest from water, where long-term grazing pressure was least. Some gradients also showed an overall decline in species richness with increasing proximity to water.
  • 5Most species were native, and among these there were no clearly identifiable ‘global winners’ (i.e. no widespread species advantaged by grazing and associated disturbances). In contrast, the majority of the few exotic species, including two of the three most widespread species found, showed increasing abundance with proximity to watering points.
  • 6Synthesis and applications. The results from this study indicate consistent and substantial changes in plant composition that are probably related to the accumulated long-term impacts of water-centred grazing. The consequences are potentially severe, because artificial watering points are now extremely widespread in the Australian rangelands. Identification and protection of representative, water-remote refugia for the most grazing-sensitive species should therefore be a high priority for regional conservation.


Rangelands are systems in which the production of domestic livestock is based on natural or semi-natural plant communities that, although they may be modified, are not deliberately destroyed. Hence they are systems where there is excellent potential for sustainable production to coexist with biodiversity conservation (Morton et al. 1995). Yet there is considerable debate about the extent to which this potential is achieved (West 1993; Fleischner 1994). This is partly because different authors focus on different aspects of biodiversity, and partly because grazing effects are inherently variable, depending on the nature of the environment and production systems and the spatial and temporal scale of measurement (Milchunas, Sala & Lauenroth 1988; Milchunas & Lauenroth 1993; Olff & Ritchie 1998; Landsberg, O’Connor & Freudenberger 1999). For plant diversity, several syntheses have been proposed for predicting how grazing effects should vary with evolutionary history and environment, but most empirical tests of these syntheses have relied on relatively short-term, small-scale, grazing experiments conducted in America, Europe and Africa. For example, of the 236 site data sets used by Milchunas & Lauenroth (1993) to test their synthesis, all consisted of grazed–ungrazed experimental treatments applied, on average, over 8–27 years, and nearly all were from America.

Rangelands also occupy vast areas of Australia (about 6 million km2 or 75% of the continent) and have been used for extensive production of livestock (sheep and cattle) since Europeans began to settle the interior, around 150 years ago (Harrington, Wilson & Young 1984). Australian rangelands differ from those in other continents in at least one important aspect: Australia has a relatively light evolutionary history of grazing by large herbivores, and thus many of its biota may be poorly equipped to cope with the elevated and continuous grazing that now predominates across most of its rangelands (Freeland 1990; Landsberg, O’Connor & Freudenberger 1999). Much of the research that has been done has focused on plants, and much is now known about how the current grazing regime affects many important pasture species, which are usually widespread and often abundant (Harrington, Wilson & Young 1984). Much less is known of the impact of livestock grazing on the less common species that comprise the bulk of plant diversity (Walker, Kinzig & Langridge 1999). However, because of their relative rarity these uncommon species may be those most prone to local extirpation (Pimm, Jones & Diamond 1988) and therefore the highest priority for conservation.

The study described in this paper aimed to address this knowledge deficiency by investigating the effect of livestock grazing on the full range of flora represented in eight widely scattered places in the central and southern rangelands of Australia. Our approach was to focus on changes in the abundance of all plant species observed along grazing gradients reflecting a realistic range of long-term grazing impacts. Our main emphasis was on examining a number of gradients spread across a broad geographical range, rather than providing definitive descriptions of localized responses within any single gradient.

We did not measure grazing impact directly, but instead used distance from water to provide a spatial gradient in the accumulated impact of long-term exposure to livestock. Many studies have convincingly demonstrated the existence of such water-centred grazing gradients under broad-scale commercial grazing in Australian arid rangelands (Lange 1969; Andrew 1988; Pickup, Bastin & Chewings 1994; James, Landsberg & Morton 1999; Hunt 2001). When stock watering points are artificially created using bores, as is common through much of inland Australia (Landsberg & Gillieson 1996), there is little confounding between grazing and other environmental patterns. Thus we were confident that carefully selected gradients of increasing distance from bore-fed water points should reflect gradients in long-term grazing activity. In contrast, grazing exclosures or experimental trials are frequently compromised by small size, poor replication, edge effects, unique local conditions and poor representation of larger scale processes (Stohlgren, Schell & Vanden Heuvel 1999). In addition, most manipulated grazing experiments are maintained for periods that are too short to separate long-term trends from short-term change due to rainfall variability, or to capture many of the processes (soil changes, competitive interactions, event-driven stochasticity) that lead to biotic change in arid rangelands. This was much less of a problem with water-centred gradients, which represent the accumulated impact of their history of pastoral use.

We used a number of approaches to assess changes in plant diversity. We analysed changes in species ‘richness’ (numbers of species per site) because it is the most commonly used single measure of biodiversity (World Conservation Monitoring Centre 1992). We also recognized the limitations of species richness as a biodiversity metric, particularly its lack of sensitivity to systematic changes in the abundance or identities of the individual species that constitute an assemblage (Cousins 1991; McKinney & Lockwood 1999). As our primary concern was identifying whether there were any systematic changes in abundance of species about which we had little prior knowledge, the main focus of our analyses was classifying and interpreting general patterns of abundance. We did this by identifying relative proportions of species showing local increases and decreases in abundance along individual grazing gradients. In grazing parlance these species represent local ‘increasers’ and ‘decreasers’ (sensuDyksterhuis 1949); in the context of global biodiversity change they may also indicate prospective ‘winners’ and ‘losers’ (sensuMcKinney & Lockwood 1999) along each of the gradients.

The specific hypotheses we posed were as follows.

  • 1In terms of individual species’ responses to grazing (sensuDyksterhuis 1949): (i) some species will be advantaged by the local disturbances associated with heavy to moderate grazing, and will thus show patterns of increasing abundance with proximity to water; and (ii) some species will be disadvantaged and will thus show patterns of decreasing abundance with proximity to water and its associated grazing impacts.
  • 2Overall, because the native species pool is likely to include species with low tolerance of grazing owing to the environment's low natural abundance of large grazing mammals (Milchunas, Sala & Lauenroth 1988; Olff & Ritchie 1998; Landsberg, O’Connor & Freudenberger 1999): (i) there will be relatively more native decreasers than native increasers; (ii) there may be relatively more native species restricted to water-remote reference sites than grazed sites; and (iii) there may be an overall decline in species richness (numbers of native species per site) with proximity to water.
  • 3In terms of global impacts on biodiversity (McKinney & Lockwood 1999): (i) increaser species (winners) are more likely to be widespread and relatively common because they are more likely to possess traits promoting opportunism; and, conversely, (ii) decreaser species (losers) are more likely to be relatively restricted in their geographical distributions; and (iii) there may be an influx of non-native, colonizing species at the most heavily grazed sites, because they are better able to tolerate direct grazing impacts, and increasing global transport promotes widespread, broadly tolerant, non-native species.

In this paper we report on our findings for the standing vegetation at study sites. Information on patterns in the vegetation represented in soil seed banks is reported in Landsberg et al. (1999).


Nomenclature for plants follows appropriate state flora and species lists (Black 1965; Jessop 1981; Boyland 1984; Western Australia Department of Agriculture 1986; Harden 1993; Mitchell & Wilcox 1994; Queensland Herbarium 1994), except for Chenopodiaceae where names follow George (1982), and Poaceae where names follow Simon (1990) and Jacobs & Everett (1996). Chapman (1991) was used for cross-border clarifications. Nomenclature for mammals follows Strahan (1992). Unless stated otherwise all statistical analyses were performed using the Genstat 5 computer program (Genstat 5 Committee 1993).

survey design and site location

The study was based on field surveys of the standing vegetation at eight grazing gradients, each located on different commercial pastoral enterprises in the acacia- or chenopod-dominated vegetation of Australia's central and southern rangelands (Fig. 1), in arid landscapes receiving 300 mm or less annual median rainfall (Table 1). Each gradient was located in a large, fenced paddock that was grazed year-round by sheep Ovus aries (seven properties) or cattle Bos taurus (one property). Historical records of stocking rates were not available but contemporary rates are generally conservative. All but one of the paddocks are likely to have been stocked for at least 100 years (Table 1).

Figure 1.

Diagrammatic representation of study gradients. Each gradient consisted of six sites within a paddock arrayed at increasing distances from a water point, with sampling within sites stratified as shown to take account of vegetation patchiness. Eight gradients (one per paddock) were located in widely dispersed locations across central and southern Australia. See Tables 1 and 2 for gradient names and descriptions.

Table 1.  Biophysical characteristics of the grazing gradients (see Fig. 1 for gradient design, locations and codes)
Gradient codeGradient name*LandformsVegetation structureProminent upper- layer speciesProminent ground- layer speciesAnnual rainfall(mm)Temperaturerange (°C)Seasonal rainfallprior to survey§
  • *

    The prefixes on gradient names refer to the states where they were located: NT, Northern Territory; Qld, Queensland; NSW, New South Wales; WA, Western Australia; SA, South Australia. The suffixes refer to their vegetation type: mulga, Acacia aneura woodland; chenopod, chenopod shrub steppe; mixed, acacia woodland interspersed with chenopod shrub steppe or herbland.

  • Mean (NT mulga, SA mixed, SA chenopod) or median (remainder) annual rainfall at the nearest recording station.

  • Mean monthly maximum temperatures in coldest (July) and hottest (January) months.

  • §

    Seasonal conditions were estimated by comparing monthly rainfall for the 6 months prior to the survey with long-term rainfall statistics for the district in which the gradient was located, using standard terminology based on decile ranges (Bureau of Meteorology 1986).

ANT mulgaUndulating sand plainPatchy mosaic of woodland groves and grassy open areasAcacia aneuraEragrostis eriopoda, Monochather paradoxa19620–38Average
BNSW mulgaUndulating sand plain with broad dendritic drainage patternsBanded mosaic of woodland groves and grassy open areasAcacia aneuraEragrostis eriopoda29019–35Average
CQld mulgaBroad alluvial plains bordering residual tablelandsBanded mosaic of woodland groves and moderately grassy open areasAcacia aneura, Acacia stowardiiEragrostis spp., Aristida spp.28520–37Above average
DQld mixedMantled, undulating pediments bordering dissected tablelandsPatchy mosaic of sparse woodland groves and chenopod herb landAcacia cambagei, Maireana spp., Atriplex spp.Sclerolaena spp.30820–37Above average
EWA mixedBroad flat plains with shallow accumulations of surface sand and an underlying hardpanPatchy mosaic of tall shrubland and chenopod shrub steppeAcacia sclerosperma, Acacia ramulosa, Acacia tetragonophylla, Maireana spp., Atriplex spp.Cenchrus ciliaris, various species in the family Asteraceae20322–33Above average
FSA mixedUndulating sandy plains interspersed with low hillsPatchy mosaic of chenopod shrub steppe and low woodlandsAcacia papyrocarpa, Maireana spp., Atriplex spp.Stipa scabra, Enneapogon spp., Rhodanthe floribunda18517–34Average
GSA chenopodUndulating stony slopes and hills with gilgai and drainage depressionsChenopod shrub steppeAtriplex vesicariaSclerolaena spp., Dissocarpus spp., Eragrostis eriopoda, various Asteraceae18217–34Above average
HWA chenopodExtensive flat karst plainChenopod shrub steppeMaireana sedifolia, Atriplex vesicariaStipa spp., Danthonia caespitosa, various Asteraceae28018–32Average

Each gradient consisted of six sites at varying distances from a livestock watering point (in each case a trough supplied by piped bore water) and included a reference site as remote from all waters as possible, where livestock impacts were minimal (Fig. 1). Distances to reference sites were chosen to lie outside the normal grazing range of the livestock using the paddock, and close to the limit of the water-centred grazing activities of free-ranging kangaroos Macropus spp. and feral goats Capra hircus (Landsberg & Gillieson 1996). The five other sites along each gradient were located progressively closer to the watering point (Table 2). The sites along the cattle gradient were spaced further apart to reflect the greater grazing range of cattle. All six sites on each gradient were located in the same land system, paddock and position in the landscape, so that they were as similar as possible in everything except distance from the watering point and its associated grazing activity. Each gradient was surveyed once only, the first four in spring 1994 (a year of average annual rainfall for all four) and the second four in spring 1995 (a year of above-average annual rainfall). Rainfall descriptors are based on annual decile ranges (Bureau of Meteorology 1986). More detailed descriptions of the gradients are published in Landsberg et al. (1999).

Table 2.  Characteristics of gradient paddocks and distances (in km) of sites from livestock watering points. Stock numbers are those given by the current property manager; long-term records were not available (see Fig. 1 and Table 1 for gradient design and descriptions)
Gradient codeGradientDomestic livestock grazing paddockApprox. date of settlement of pastoral lease*Paddock size (ha)Approx. number of stockOther grazing animals observed in paddockSite 1Site 2Site 3Site 4Site 5Site 6
  • *

    Note that development of gradient paddocks (fencing and installation of water points) may not have occurred for some years (possibly decades) after first settlement.

  • Macropus rufa.

  • Macropus robustus erubescens.

  • §

    Site 6 at the Qld mulga gradient was within the sphere of influence of two water points. The water point to the south-east, which was the one influencing sites 1–5 along this gradient, was 8·9 km distant. There was, however, another water point 7·0 km to the east of site 6. Few animals travelled from that water point to site 6, because it entailed crossing a stony plateau; thus site 6 was actually 8·9 km from the nearest readily accessible water point.

  • Capra hircus.

  • **

    Macropus fuliginosus.

  • ††

    Oryctolagus cuniculus.

ANT mulgaCattle1880–189041 000UnknownFew0·52·03·55·99·015·1
BNSW mulgaSheep/cattle1880–1900 5 000600/80Red kangaroos0·71·02·03·56·1 8·9
CQld mulgaSheep1880–1890 8 6001000Euros0·51·12·03·76·0 7·0–8·9§
DQld mixedSheep1880–190045 4003000Euros and red kangaroos0·71·21·83·75·5 8·9
EWA mixedSheep1890s 6 000400–600Goats0·60·92·13·66·0 8·3
FSA mixedSheep187812 000700–900Western grey** and red kangaroos0·51·02·03·66·010·2
GSA chenopodSheep1868 9 300700–900Red kangaroos0·51·12·23·66·4 8·2
HWA chenopodSheep196413 0002000Red kangaroos and rabbits††0·51·12·03·86·4 8·9

We assessed ground cover at each site to provide a check that grazing impacts varied as expected between sites. Many studies have demonstrated the nature and extent of cover change along grazing gradients and the high degree of spatial variability that makes definitive detection of this change difficult (Pickup, Bastin & Chewings 1998). Thus we did not attempt to provide a definitive description, but merely to check that livestock impacts on ground cover generally declined with distance from water. We used the step-point method of estimating cover, recording the nature of the ground cover between a well-marked point on a boot toe and a point 50 cm vertically above it. This was done at 1-m intervals along the 20 10-m sample transects used for locating quadrat arrays for vegetation sampling (described below).

vegetation sampling

The sampling strategy was designed to detect trends in relative abundance of all species present in the standing vegetation at each site along a gradient; it did not attempt to be comprehensive with respect to the total flora in each paddock. Two levels of quadrat sampling were used: 1-m2 quadrats for ground-layer vegetation (defined as all species with the bulk of their foliage generally within 50 cm of the ground) and 100-m2 quadrats for the upper layer (defined as all species taller than 50 cm). Each site was stratified according to the major structural elements of the vegetation present, which was usually tree groves intermingled with open areas. Species abundance was assessed as frequency of occurrence of all species rooted within quadrats. Ground-layer species were sampled in 80 1-m2 quadrats. These were located in 10 groups of eight contiguous quadrats, with each group in a separate vegetation patch representing alternate structural strata, and each group aligned along the long axis of a patch as illustrated in Fig. 1. Upper-layer plants were assessed in 20 10 × 10-m quadrats. Half of these were co-located with the ground-layer quadrats, and half in similar, neighbouring patches. (This division was necessary because individual patches were seldom large enough to accommodate more than one upper-layer quadrat.) Sites varied in size depending on the configuration of the replicate patches, but were generally around 5 ha in extent.

In addition to plants identified in quadrats, the presence of any additional species was recorded during a timed, 30-min, search of the site by the two people who had already assessed plant frequency in the quadrats. Species found only by searching were assigned a nominal frequency of 0·25% for the site where they were found.

Comprehensive voucher collections were made to confirm identifications. For most field-identified species at least one specimen was mounted in a field herbarium, even if reproductive material was not available, and all subsequent herbarium identifications were cross-referenced with this collection. Where good-quality reproductive specimens were available duplicates were lodged with the relevant state and national herbaria. Full species lists and identification details are available in Landsberg et al. (1999).


Geographic and local distribution of species

In order to detect any general patterns in species abundance relationships with sites and gradients, we undertook detrended correspondence analysis (DCA) of the site frequency data for the 720 records of ground-layer species detected on gradients. We also classified the abundance of each species individually, in order to examine interactions between local rarity and grazing responses. To do this we devised a simple four-way classification based on frequency of occurrence at individual sites and total number of sites occupied, where the distribution of each species on a gradient was classified as: uncommon, if its frequency at each site was less than 5%; common, if its frequency at any site was 5% or more; restricted, if it was recorded only at one or two sites; locally widespread, if it occurred at three or more of the six sites sampled along the gradient.

Ground cover

Ground cover data were amalgamated into four mutually exclusive categories: bare ground, litter, grass and forbs (non-grassy herbs and shrubs). The average cover per site in each category was then analysed separately to determine if it was significantly affected by distance zone (a categorical variable with six levels representing sites along a gradient), using analysis of variance. Because of obvious differences between chenopod- and acacia-dominated gradients, distance zone was nested within a higher-order factor ‘biome’ with two levels, chenopod and acacia. Variations among gradients and within sites were treated as nested blocking factors, with the block structure defined as site stratum (a two-level factor) nested within gradient (an eight-level factor). Standard diagnostic plots indicated no violation of anova assumptions.

Identifying increasers and decreasers

To determine how individual species were affected by variable distance from water on individual gradients, we identified a limited set of hypothetical trends we could realistically test in the context of our sample design (Fig. 2). We then classified the data for each species according to which of these trends best described its distribution on a particular gradient.

Figure 2.

Classification scheme used for determining species’ responses from their patterns of abundance with proximity to a water point. Separate classifications were undertaken for each species on each study gradient where it occurred. (See Fig. 1 and Tables 1 and 2 for gradient definition and descriptions.) When species occurred at three or more sites on a gradient, the nature and significance of their response trends were determined by fitting linear, quadratic and exponential regressions to frequency data. When species were restricted to one or two sites on a gradient they were assigned to response groups according to the location of the sites where they occurred.

For the locally widespread species (i.e. those occurring at three or more sites on a gradient) we used statistical models to fit regressions describing the set of hypothesized trends. We tested linear, quadratic and exponential regressions, with species frequency (%) as the dependent variable and distance from water (in km) as the independent variable, so that n= 6 for each regression. The model of best-fit was deemed to be that which explained the largest percentage of variation, provided that it was significant overall (P < 0·05) and that, if curvilinear, it explained significantly more variation (P < 0·05) than the linear model. Graphs showing fitted curves for all three models were also generated, and used to determine whether the trend of the best-fit model was increasing, decreasing or medial, as illustrated in Fig. 2.

Because these procedures represented a large number of independent tests there was a high probability of type I errors, i.e. false positives. Standard techniques for controlling this type of error involve reducing the minimum critical P-value in proportion to the total number of tests conducted. This reduces the likelihood of type I errors but at the cost of increasing the likelihood of type II errors, i.e. of wrongly rejecting significant test results when they are not due to chance. The standard Bonferroni correction for multiple tests reduces the critical P-value for significance testing of k multiple tests from α to α/k (Rice 1989). For the 500 or so tests we performed this would mean dropping the critical P-value from the conventional 0·05 to 0·0001, which would result in a very high type II error rate because any regression for which the fit was less than a near-perfect would be likely to be rejected as not significant.

Thus we had to choose between risking type I (false positive) errors with a conventional critical P-value, or type II (false negative) errors with a conventionally lowered critical P. We chose to accept the risk that type I errors could inflate the number of regressions deemed significant, because we reasoned that errors of this sort were likely to have a similar inflationary effect on numbers of species assigned to any of the response groups. We therefore tested our regressions against a conventional critical P-value of 0·05, and focused our conclusions on relative differences between response groups rather than absolute numbers of species so assigned. We used paired t-tests to test the significance of the differences between increaser and decreaser groups, for the number of species per gradient in linear, curvilinear and combined categories.

For the species with restricted distributions along a gradient (i.e. those occurring at one or two sites only) regressions were not fitted because the large number of zero values violated model assumptions. For these species we used a simple classification procedure whereby species were designated: restricted increasers if they occurred only at site 1, or only at sites 1 and 2; restricted decreasers, if they occurred only at site 6 or only at sites 6 and 5; restricted medials, if they occurred only at site 3 and/or site 4.

Again, we used paired t-tests to determine the significance of differences between numbers of species per gradient assigned to increaser and decreaser categories. The null hypothesis was that the number of restricted increasers was not significantly different from the number of restricted decreasers on each gradient.

Changes in species richness

We used the same set of linear, quadratic and exponential regression models to explore relationships between numbers of species per site and the proximity of that site to water. The models were tested separately for each gradient, using number of species per site as the dependent variable and site distance from water as the independent variable, and following the same procedure for determining which, if any, was the best-fit model.

Species restricted to a single site

Many plant species were singletons, i.e. were found at only one site along a gradient. In many cases this was likely to have been the result of chance detection of locally rare species. However, if the number of singletons found at the sites furthest from water was more than expected by chance, it could indicate an underlying trend of net species loss from all but these sites. We undertook a statistical analysis to determine whether there was such a pattern using a subset of species’ presence data from uniformly spaced sites.

We restricted the analysis to uniformly spaced sites to minimize spatial confounding. Spatial confounding arises because species are more likely to occur on adjacent sites if those sites are close together, and, conversely, species are less likely to occur on adjacent sites if those sites are far apart. It follows, therefore, that the probability of finding singletons at a site increases with increasing isolation of that site from its neighbours. On most of the study gradients the most water-remote site (site 6) was around 2·5 km from its nearest neighbour, while the site closest to water (site 1) was only 0·5 km from its neighbour (Table 2). Thus there was a higher probability of finding singletons at site 6 then there was of finding them at site 1, simply because site 6 was more isolated. However, for seven of the eight gradients (not NT mulga) sites 2, 4, 5 and 6 were all around 2·5 km apart (Table 2) and thus less subject to spatial confounding. Sites 2 and 6, the sites at the end of this reduced gradient, were still more isolated than the sites in the middle, but at least site 6 was no more isolated than site 2.

Thus our analysis of singletons was restricted to these four sites and the seven gradients where they were equidistant (i.e. the NT mulga gradient was excluded). First, a subset of species’ presence data was compiled, then the number of species found at just one of the four sites was calculated. Generalized linear models were then undertaken to test the significance of differences among sites in the number of singletons identified in this way. Variation among gradients was treated as a blocking factor, by first adding a term for gradient (a factor with seven levels), followed by a term for site, which was treated as a factor with four levels. Numbers of singletons were expressed as proportions of the total number of species found at all four sites and analysed assuming binomial errors and a logit-link function.


distribution patterns

Much of the plant diversity resided in the ground layer, in which we detected 478 species, of which 466 could be identified with confidence, compared with only 134 species identified in the upper layers of vegetation. As most upper-layer species were also relatively infrequent in occurrence we concentrated much of our analysis on the ground layer.

DCA (Fig. 3) showed tight associations between many ground-layer species and particular gradients, and also between species and sites within some of the gradients. A linear trend in species–site associations was apparent for gradient E, and separations between species associated with sites close to water (sites 1 and 2) and those remote from it (sites 5 and 6) were apparent in gradients A, G and H.

Figure 3.

Detrended correspondence analysis based on frequency of ground-layer plants (not transformed) at the sites along the study gradients. (See Fig. 1 and Tables 1 and 2 for gradient definition and descriptions.)

Associations between some gradients were also apparent, most strongly between gradients A and B, gradients C and D, and gradients F and G. These associations were consistent with biogeographical patterns: gradients A and B were both mulga gradients and were both assessed after seasons of average rainfall (Table 1). It is noteworthy that this occurred despite the fact that gradient A was grazed by cattle and gradient B by sheep, suggesting that geographical proximity and grazing pressure had more influence than livestock type on the ground-layer composition of these gradients. Gradients C and D were close together in geographical space (Fig. 1), as well as sharing some overlap in vegetation structure (Table 1); they were also both measured after above-average rainfall (Table 1). Gradients F and G likewise shared spatial proximity, some structural elements and average rainfall prior to sampling, while gradient E (the WA mixed gradient) was isolated in both geographical (Fig. 1) and analysis space (Fig. 3).

Not all species were closely associated with individual gradients. The DCA (Fig. 3) also indicated the existence of a set of more widespread species that showed little affinity with any particular gradient.

However, most species were not geographically widespread: 72% of the 466 ground-layer species identified were found on only one gradient, and none on all eight gradients (Fig. 4). Even within individual gradients 45% of species occurred on only one or two of the six sites that constituted the gradient (restricted distribution; Table 3). In addition to being restricted in their distributions, many species (46%) were also locally uncommon (Table 3).

Figure 4.

Numbers of ground-layer species found at different numbers of gradients. (See Fig. 1 and Tables 1 and 2 for gradient definition and descriptions.)

Table 3.  Distribution patterns of ground-layer species along individual gradients (see Fig. 1 and Tables 1 and 2 for gradient design and descriptions). Uncommon species were defined as occurring at less than 5% at any site, restricted species as occurring at one or two sites only, and common and locally widespread species as those in complementary categories
GradientUncommon and restricted speciesCommon but restricted speciesUncommon but locally widespread speciesCommon and locally widespread speciesTotal
NT mulga 15 2 6 31 54
NSW mulga 22 6 7 19 54
Qld mulga 391711 60127
Qld mixed 39 516 53113
WA mixed 4912 6 53120
SA mixed 531315 40121
SA chenopod 26 3 9 33 71
WA chenopod 16 4 2 38 60
Total259 (36%)62 (9%)72 (10%)327 (45%)720

Trends in ground cover were weak, but those that were detected were consistent with the expected trend in grazing impact. Both bare ground and grass cover were significantly affected by distance zone (Table 4), with bare ground tending to increase, and grass cover to decrease, with increasing site proximity to the water point (Fig. 5). Litter cover differed significantly between biomes (36% and 23% in acacia and chenopod biomes, respectively, with P= 0·023, F= 9·10, stratum d.f. = 1,6) but not with distance from water within vegetation type (P = 0·210, F= 1·37, stratum d.f. = 10,70). Forb cover did not vary significantly with biome or distance zone.

Table 4. anova results showing significant differences (bold font) in ground cover along study gradients (see Fig. 1 for gradient design)
Source of variationd.f.Sums of squaresMean squaresVariance ratioP
(a) Variate: cover of bare ground
Gradient stratum
Biome (chenopod/acacia) 1 2932·342932·34 2·80·143
Residual 6 6190·921031·82 0·23 
Gradient × site stratum 836174·534521·8247·26 
Gradient × site × units stratum
Biome × distance zone10 2386·97 238·70 2·500·013
Residual70 6696·87  95·67  
(b) Variate: grass cover
Gradient stratum
Biome (chenopod/acacia) 1  846·09 846·09 1·590·254
Residual 6 3189·90 531·65 8·36 
Gradient × site stratum 8  508·92  63·61 2·68 
Gradient × site × units stratum
Biome × distance zone10  657·77  65·78 2·770·006
Residual70 1661·06  23·73  
Total95 6863·74   
Figure 5.

Contributions of bare ground, litter, grass and other plant material (shrubs, forbs and wood) to total ground cover at the six sites per gradient. Average values for the acacia-dominated gradients (gradients A–D in Fig. 1 and Table 1) are shown in the upper half of the diagram and averages for the chenopod-dominated gradients (E–H in Fig. 1 and Table 1) in the lower half. Trends in bare ground (decreasing with distance zone) and grass cover (increasing with distance zone) were significant (Table 4).

response types

Of those ground-layer species classified into the response types illustrated in Fig. 2, decreasers significantly outnumbered increasers (Table 5). This was due mainly to significantly more species showing trends of linear decrease (4·3% of all species, compared with 1·1% classed as linear increasers), and restricted decreaser distributions (16·8%, compared with 8·3% classed as restricted increasers). Numbers of species per gradient showing exponential trends were similar for both decreasers (3·6%) and increasers (4·9%).

Table 5.  Numbers of ground-layer species classified in the response groups illustrated in Fig. 2. Differences between numbers of decreasers and increasers were significant for classes with linear trends (paired t-value = 3·29, d.f. = 7, P= 0·013) and restricted distributions (paired t-value = 2·83, d.f. = 7, P= 0·025) and for all classes combined (paired t-value = 2·59, d.f. = 7, P= 0·036), but not for classes showing exponential trends (paired t-value = 0·87, d.f. = 7, P= 0·411) (see Fig. 1 and Tables 1 and 2 for gradient design and descriptions)
GradientDecreaser classesIncreaser classesMedial classesExtremist classesNo patternGradienttotal
Linear trendExponential trendRestricted distributionLinear trendExponential trendRestricted distributionQuadratic trend (inverted-U)Restricted distributionQuadratic trend (U-shaped)Restricted distribution
NT mulga 1 0  80 2 1 0 300 39 54
NSW mulga 1 5 122 0 9 2 601 16 54
Qld mulga 7 5 282 3 7 31110 60127
Qld mixed 3 4 171 9 4 21510 57113
WA mixed 4 3 150 512 02101 59120
SA mixed 8 5 281 415 31304 40121
SA chenopod 3 1 101 7 6 01110 31 71
WA chenopod 4 3  31 5 6 1 600 31 60
Class total312612183560118636333720
% all species 4·3 3·6 16·81·1 4·9 8·3 1·511·90·40·8  46·3100
 Decreaser total = 24·7%Increaser total = 14·3%Medial total = 13·4%Extremist total = 1·2%  

Those species that were not classified as increasers or decreasers were mainly neutral in their response to distance from water, with nearly half (46·3%) showing no discernible pattern (Table 5). The only other moderately common pattern was among species with restricted distributions, with 11·9% of species showing a restricted, medial response pattern. However, this may have been because the classification scheme was biased in favour of this class. There were three potential configurations that could be classified as restricted medial (species at site 3 only, site 4 only, or sites 3 and 4 only), compared with only two for configurations for restricted increasers (site 1 only, or sites 1 and 2 only) and decreasers (site 6 only, or sites 6 and 5 only) and only one (sites 1 and 6 only) for restricted extremists.

In the upper layers of vegetation, numbers of increasers and decreasers were not significantly different (P = 0·54, paired t-value = 0·64, d.f. = 7). Numbers of non-native species were low at all gradients (Table 6), but of those that were recorded significantly more were increasers than decreasers.

Table 6.  Numbers of non-native species classified in the response groups illustrated in Fig. 2. Differences between numbers of decreasers and increasers were marginally significant (paired t-value = 2·26, d.f. = 7, P= 0·059) (see Fig. 1 and Tables 1 and 2 for gradient definition and descriptions)
Number of speciesNT mulgaNSW mulgaQld mulgaQld mixedSA mixedWA mixedSA chenopodWA chenopod
Total increasers00116426
Total decreasers00201100

species richness

The lowest numbers of ground-layer species were recorded at two of the mulga gradients and the highest numbers at the third (Table 3). This probably reflects differences in seasonal conditions: the gradient with highest species numbers (Qld mulga) had experienced above average rainfall prior to sampling (Table 1), in contrast to the NSW and NT mulga gradients where rainfall had been average and species numbers were low (Tables 1 and 3). The next highest species numbers were recorded at the three gradients with mixed vegetation (Table 3), which had areas of woodland interspersed with areas of open shrub steppe (Table 1). The relatively high numbers of species at these gradients was probably due to their mixed vegetation rather than rainfall, as two had experienced above-average rainfall (Qld mixed and WA mixed) but the other (the SA mixed) had not. Numbers of species at both chenopod gradients were similarly low (Table 3), despite contrasting rainfall prior to sampling (Table 1).

For ground-layer species at three of the gradients there was a significant trend of decreasing species richness with site proximity to water (Fig. 6). The relationships were: curvilinear decrease at the Qld mulga gradient (exponential regression with P= 0·034, F= 12·82, d.f. = 2,5); linear decrease at the Qld mixed gradient (linear regression with P= 0·013, F= 18·08, d.f. = 1,5); and curvilinear decrease at the WA chenopod gradient (exponential regression with P= 0·028, F= 14·5, d.f. = 2,5). However, this trend was largely driven by a drop in species numbers at site 1.

Figure 6.

Change in richness of ground-layer species along gradients. (See Fig. 1 and Tables 1 and 2 for gradient definition and descriptions.)

There was also a weak decreaser trend for ground-layer species at the SA mixed gradient that approached significance (linear regression with P= 0·08, F= 5·47, d.f. = 1,5). Species richness in the upper-layer vegetation did not vary significantly with proximity to water.

potential losers: species restricted to a single site

Of the ground-layer species restricted to just one of the subset of equidistant sites, significantly more were found only at the reference site than at any other (Table 7; P= 0·03 that site 6 values are different). No such effect was apparent among upper-layer plants, but numbers of restricted species in this layer were generally much lower (2 ± 1 per site per gradient).

Table 7.  Numbers of ground-layer species found at one site only, for the subset sites that were equal distances apart (see Fig. 1 and Tables 1 and 2 for gradient definition and descriptions)
GradientSite 2Site 4Site 5Site 6 (reference site)
NSW mulga 3 6 4 7
Qld mulga 4 51013
Qld mixed 5 8 714
WA mixed10 8 810
SA mixed 911 620
SA chenopod 311 4 5
WA chenopod 6 4 1 3
Mean ± standard error5·7 ± 1·17·6 ± 1·05·7 ± 1·110·3 ± 2·2

potential winners: species found at more than one gradient

Only 27 of the 132 species found at more than one gradient were classified as increasers or decreasers more than once (Table 8). Decreaser responses were predominant, with 52% (14 out of the 27 species) showing consistent decreaser responses. Only three species (11%) were consistently classified as increasers. Two of these, Carrichtera annua (Ward's weed) and Schismus barbatus (Arabian grass) are rapid-growing, free-seeding, annual exotics of little or no pastoral value (Cunningham et al. 1992).

Table 8.  Ground-layer species classified as increasers (I, i) or decreasers (D, d) at more than one gradient. Upper case indicates responses based on significant trends (linear or exponential); lower case indicates classifications based on restricted distribution patterns. The specific response groups are illustrated in Fig. 2 (see Fig. 1 and Tables 1 and 2 for gradient definition and descriptions)
SpeciesFamilyResponse at each gradientConsistency between gradients
  • *

    Non-native species.

Abutilon fraseriMalvaceaeD at Qld mulga; d at Qld mixedConsistent decreaser responses
Aristida obscuraPoaceaed at NT mulga; d at NSW mulgaConsistent decreaser responses
Atriplex vesicariaChenopodiaceaeD at SA mixed; D at WA chenopodConsistent decreaser responses
Calotis cuneifoliaAsteraceaed at Qld mulga; D at NSW mulgaConsistent decreaser responses
Daucus glochidiatusApiaceaed at Qld mixed; d at WA mixed; d at SA mixedConsistent decreaser responses
Enneapogon polyphyllusPoaceaeD at Qld mulga; D at SA mixedConsistent decreaser responses
Erodium cygnorumGeraniaceaed at SA mixed; d at SA chenopodConsistent decreaser responses
Goodenia berardianaGoodeniaceaed at Qld mulga; d at Qld mixedConsistent decreaser responses
Goodenia cyclopteraGoodeniaceaed at Qld mulga; d at Qld mixedConsistent decreaser responses
Portulaca filifoliaPortulaceaed at NT mulga; D at Qld mulgaConsistent decreaser responses
Rhagodia spinescensChenopodiaceaed at Qld mulga; d at NSW mulgaConsistent decreaser responses
Rhodanthe floribundaAsteraceaeD at Qld mixed; d at SA chenopod; D at WA chenopodConsistent decreaser responses
Salsola kaliChenopodiaceaeD at WA mixed; d at SA chenopod; D at WA chenopodConsistent decreaser responses
Sclerolaena decurrensChenopodiaceaed at SA mixed; D at SA chenopodConsistent decreaser responses
Aristida contortaPoaceaeI at NSW mulga; D at WA mixed; D at SA mixedMixed, but predominantly decreaser responses
Erodium crinitumGeraniaceaeD at Qld mixed; D at WA mixed; D at SA mixed; i at SA chenopodMixed, but predominantly decreaser responses
Eragagrostis dielsiiPoaceaeI at Qld mulga; D at SA mixedMixed responses
Gnephosis arachnoideaAsteraceaed at SA mixed; i at SA chenopodMixed responses
Lepidium phlebopetalumBrassicaceaei at Qld mulga; D at SA chenopodMixed responses
Maireana tomentosaChenopodiaceaed at NT mulga; i at SA mixedMixed responses
Solanum ellipticumSolanaceaeD at NSW mulga; I at SA mixedMixed responses
Brachycome ciliarisAsteraceaeD at SA mixed; I at WA chenopodMixed responses
Calotis hispidulaAsteraceaeI at Qld mulga; D at SA mixed; I at SA chenopodMixed, but predominantly increaser responses
Chenopodium melanocarpumChenopodiaceaeI at NT mulga; I at WA mixed; d at SA mixedMixed, but predominantly increaser responses
*Carrichtera annuaBrassicaceaeI at SA mixed; I at WA chenopodConsistent increaser responses
Chenopdium cristatumChenopodiaceaei at Qld mulga; i at NSW mulgaConsistent increaser responses
*Schismus barbatusPoaceaeI at SA mixed; I at SA chenopodConsistent increaser responses

There were also moderate numbers of inconsistent or mixed patterns for the same species on different gradients. Some may have been due to type I errors (i.e. false positives). This was particularly likely for classifications based on restricted distributions that, because they could not be tested for significance, were very sensitive to chance occurrences and type I errors. This might explain why mixed responses were recorded for Erodium crinitum, for example. At the only gradient where it was classified as an increaser its distribution was restricted (Table 8), making its classification particularly sensitive to type I error. In contrast, it was locally common and widespread at the three gradients where it was classified as a decreaser, and the probabilities of the trend occurring by chance were very low (P-values were 0·005, 0·005 and 0·008).

For other species mixed responses may represent real environmental differences. For example, Aristida contorta and Eragrostis dielsii, the only two grasses showing mixed responses, were classed as increasers on a mulga gradient and decreasers on mixed gradients (Table 8). The ground layer was predominantly grassy on the mulga gradient but grasses were far less prominent on the mixed gradients (Table 1). Neither Aristida contorta nor Eragrostis dielsii is regarded as particularly palatable among grasses (Cunningham et al. 1992) but sheep and cattle generally prefer grasses over shrubs and forbs (Wilson & Harrington 1984). Thus, the different grazing responses identified in different environments for these species could reflect differences in relative palatability, whereby animals tend to avoid them when other grasses are common but select them when the choice of grasses is more limited.


nature and magnitude of change

Most of the significant trends we recorded, in cover, species richness and frequency of individual species, were apparent in the ground layer but not amongst taller shrubs or trees. This does not mean that upper layers of vegetation are unaffected by grazing. Intensive life-history studies have demonstrated that long-term grazing by sheep has significant negative effects on many species of taller shrubs and trees during recruitment and regeneration (Tiver & Andrew 1997). Our results indicate that ground-layer vegetation may be even more sensitive.

However, the composition of the ground layer is highly variable in space: 72% of the species we recorded were restricted to one gradient and only 13% occurred in more than two gradients. Even on individual gradients more than half the species were uncommon and/or restricted in their distributions.

Yet despite this compositional variation, we found solid support for many of the hypotheses we originally posed. Regarding hypothesis 1, we were able to identify definite groups of species showing patterns of increasing and decreasing abundance (frequency) with proximity to water.

Regarding hypothesis 2, we found that decreasers significantly outnumbered increasers for both locally widespread species showing linear trends of response (nearly four times as many were decreasers) and for species that were found at one or two sites only (twice as many were decreasers). This occurred despite the spacing between sites, which resulted in more intense sampling near water and therefore a sampling bias toward overestimating numbers of increasers. Under the classification scheme we used, nearly a quarter (24·7%) of the species per gradient were designated decreasers, compared with 14·3% as increasers. We recognize that these values may be inflated by type I errors, but there is little doubt of the significance of the difference between the two groups.

In further support of hypothesis 2, we found that significantly more singletons (species restricted to a single site) occurred at the site furthest from water, where the long-term pressure of grazing was least. We also found some indication of a decline in species richness (numbers of species per site) with increasing site proximity to water, but only for three of the eight gradients.

We found mixed support for hypothesis 3, however. With respect to non-native species, numbers were low but those we did find provided support for (iii). Non-native species were more likely to be increasers, and two of the three widespread increasers we detected were non-native weeds that were most abundant at sites closest to water.

Most of the species we recorded were natives, and among these there were no clearly identifiable global winners (in the sense of being widespread species advantaged by grazing and associated disturbances). Most species, increasers and decreasers alike, were neither geographically widespread nor locally common on individual gradients. Among the 132 ground-layer species that were sufficiently widespread and common to be found on more than one gradient, only 27 (those listed in Table 8) showed discernible patterns of response to grazing. There was no evidence that these more widespread species were losing ground, nor that they were advantaged in any way by the disturbances associated with grazing. Among the relatively small number of widespread species that did show a grazing response, more tended to be decreasers than increasers.

Taken together, these results indicate consistent changes in plant composition that are likely to be related to the accumulated long-term impacts of water-centred grazing. They are also likely to be to the long-term detriment of the native flora, particularly as the available pool of native species may be relatively depauperate in species capable of expanding or thriving (winning) under the current grazing regime.

research issues

There are several caveats and assumptions associated with the present study. It did not measure grazing pressure directly, different mixtures of animal species grazed each of the paddocks, plant species were surveyed once only per site, and gradients were not replicated within study paddocks. However, the study design was the result of deliberate trade-offs. It was not aimed at determining autecological responses of plants to grazing, but at a biogeographical analysis of the collective impacts of pastoralism. We chose distance from water as a surrogate for long-term grazing because of its reliability and generalizability (Lange 1969; Andrew 1988; Pickup, Bastin & Chewings 1994, 1998; James, Landsberg & Morton 1999). Our decision to sample each gradient only once allowed us to allocate greater sampling effort to a larger number of more widely spaced gradients, in keeping with our study goals. We recognized that once-off sampling would miss some of the more ephemeral species, particularly at those gradients sampled when seasons were dry. We reasoned, however, that relative differences between sites within gradients should be retained, regardless of season. We also took soil samples to investigate those species represented primarily as soil-stored seed. Preliminary results suggest that the abundance of species in the seed bank (measured as counts of individual germinants) shows trends similar to those apparent in the standing vegetation (Landsberg et al. 1999).

Lastly, our decision not to replicate gradients within paddocks was partly to do with sampling effort, but it was also dictated by what was available: the configuration of the paddocks in which we worked did not allow for replication of gradients. This was because water points were usually close to fences (thereby restricting the number of directions in which gradients could run) and paddocks usually contained a mosaic of different land systems, most of which did not extend for the required 10 km from water.

Rather than detracting from the reliability of the trends we measured, these imperfections indicate the likely strength of the influence of distance from water and associated grazing. Despite the differences between gradients in location, vegetation, management, seasonal conditions and the identities of the species growing there, they show consistent trends in the response of their plant diversity to the current grazing regime. It is also noteworthy that relative differences in proportions of increasers and decreasers provided a more consistent basis than species richness for showing general patterns of biodiversity change under grazing.

However, inconsistencies in responses of some of the individual species found at more than one gradient were evident. Some of these inconsistencies may be sampling artefacts, but others are likely to represent real variation in the ways individual species are affected by grazing in different regions. Cross-regional inconsistency in the responses of some species to grazing is well recognized (Noy-Meir, Gutman & Kaplan 1989) and has led to calls for caution in using indicator species outside the vegetation types in which their responses to grazing have been observed (Landsberg, Lavorel & Stol 1999; Stohlgren, Schell & Vanden Heuvel 1999; Vesk, Leishman & Westoby 2004). Inconsistency in plant species’ responses to grazing appears to be particularly common in Australian rangelands. Meta-analysis of data compiled from 35 published studies (Vesk & Westoby 2001) showed that, in Australian rangeland, for 324 species classified as increasers or decreasers in at least two studies, 41% responded inconsistently, increasing at least once and decreasing at least once.

Most of the species we recorded were found at only one gradient. This limits the extent to which we can draw inferences about their regional conservation status. It also highlights the need for regional studies in areas where a detailed understanding of patterns of species distribution is required. Even without type I errors, our estimates of the proportions of species in local decline may overestimate the proportion of species in decline regionally. It is possible that some species identified as decreasers along one gradient may be more abundant or secure in other parts of the same landscape or other paddocks with different management histories. For example, some species identified as locally rare and in decline in the more productive parts of paddocks may be more secure in unproductive areas or protected areas such as rock outcrops, which may be less heavily grazed even when close to water. Other locally rare species may be more common, and thus more secure, in other parts of their geographical range. We found some support for these possibilities in a more recent study of grazing gradients within a single region (Landsberg et al. 2002), where relatively fewer species showed regional trends of decrease or increase.

implications for conservation management

Nevertheless, the relative prominence of decreaser species across so much of the Australian rangelands is of concern, because they are not balanced by a comparable number of native increaser species and most rangeland habitats are now grazed. Prior to European settlement most of the arid interior of Australia was so lacking in permanent surface water that it could have experienced only light or intermittent grazing by kangaroos, the only indigenous large grazing mammals (Ratcliffe 1951). Since Europeans moved into the rangelands in the mid- to late 1800s, the provision of artificial sources of drinking water has resulted in a major transformation. Now nearly everywhere outside the great deserts is within 10 km of an artificial water source such as a bore or earthen tank (Landsberg & Gillieson 1996). Hence domestic and feral livestock, kangaroos and other animals that need to drink regularly now have almost continuous access to most rangeland areas (Noble et al. 1998; James, Landsberg & Morton 1999). As a consequence, there are now very few areas of productive rangeland remaining sufficiently far from water to provide potential refuges for grazing-sensitive biota.

The most grazing-sensitive species are clearly at most risk. Our results indicate that this may be to the detriment of a substantial proportion of the native flora. For example, restricted and exponential decreasers (Fig. 2) together represent up to 20% of the flora (Table 3) across a huge geographical range (Fig. 1). While this value is probably an overestimate, it nonetheless gives some indication of the scale of the potential threat, if development of new water sources proceeds without incorporating specific measures to safeguard biodiversity.

Fortunately, our study design also lends itself to potential management strategies for reducing the threat to species in decline. First, it confirms what many land managers have long suspected: that those areas remote from water and its associated grazing are valuable reservoirs of plant diversity (Ratcliffe 1951; Noy-Meir 1996). Many managers may be receptive to the idea of retaining some of their water-remote areas as refuges from sustained grazing. Secondly, selective reduction in the number of artificial water sources in more heavily developed regions offers prospects as a strategic tool for improving the status of those species negatively affected by the water-centred activities of livestock and other grazing animals.

Unfortunately, the potential conservation benefits of reducing the spatial coverage of water-sources conflict with proven environmental and production benefits associated with expanding the water network. These benefits are twofold. First, by spreading grazing pressure more evenly across the landscape, localized degradation around existing watering points is reduced. Secondly, by expanding the area of grazing land accessible to livestock, greater numbers of stock can be produced without any increase in average stocking pressures. Hence, despite the already high densities of artificial water sources across much of Australia's rangelands, provision of new water sources remains a favoured management option for improving sustainability and increasing production (for a review see James, Landsberg & Morton 1999). There is increasing recognition of the risks that provision of new waters may pose to grazing-sensitive biota, however, and a number of studies are being undertaken to investigate strategies for minimizing negative impacts on biodiversity (James et al. 2000; Landsberg et al. 2002; Hunt et al. 2003).

The situation may be less threatening in other parts of the world, where there has been a longer evolutionary history of sustained grazing by large mammals. The relative paucity of widespread native increaser species identified in our study supports the contention that the Australian native flora may not be well adapted to the current grazing regime. Still, even in Africa, with its extended evolutionary history of grazing by large mammals, current grazing regimes frequently differ from historical ones, and there are many examples where livestock grazing has caused major changes in the composition of plant communities (Illius & O’Connor 1999; Tobler, Cochard & Edwards 2003). There is also increasing recognition that the effects of herbivores on plant diversity depend not only on the type and density of herbivores, but also on the nature of the abiotic environment, particularly soil moisture and fertility (Milchunas & Lauenroth 1993; Olff & Ritchie 1998). Our results are consistent with predictions by Olff & Ritchie (1998) that, in dry environments, large herbivores could be expected to have negative impacts on plant ‘diversity’ (species numbers), by increasing rates of local extinction. Olff & Ritchie (1998) further suggest that negative impacts of large herbivores might be exacerbated if soils are also infertile, because this reduces the likelihood that grazing might promote local colonization. The Australian arid zone is unusual by world standards in the infertility of its soils and unpredictability of its rainfall, but it is not unique (Stafford Smith & Morton 1990). However, relatively low natural productivity coupled with light evolutionary exposure to grazing may mean that the native flora of the Australian arid zone is particularly sensitive to the pressures of the current livestock regime. Thus it may also be in particular need of conservation management, to ensure protection of at least some areas from the otherwise pervasive influence of elevated, year-round grazing.


This study would not have been possible without the willing co-operation of the owners and managers of the pastoral properties on which we worked. We are deeply grateful for their interest, hospitality and many kindnesses. Ted Moore, Jo Palmer, Mike Lazarides, Lyn Craven, Lawrie Adams and other staff members at the National Herbarium, Centre for Plant Biodiversity Research, Canberra, provided facilities and much helpful advice for identifying plant specimens. Expert identifications of contentious species were provided by Terena Lally, Paul Wilson and Ray Cranfield, Western Australian Herbarium; Phillip Short, Parks and Wildlife Commission of the Northern Territory; and Robyn Barker, Botanic Gardens of Adelaide and State Herbarium, South Australia. Expert field identifications at several gradients were provided by Des Nelson, CSIRO (retired) Alice Springs, and John Stretch, WA Agriculture, Carnarvon. Alex Drew, Trevor Hobbs, Maire Bryannah, Jaimie Cook, David Jensen, Terena Lally, Carolyn Maxfield, Steve Marsden and Hugh Pringle provided valuable field assistance. Trevor Hobbs also managed the database, and Hugh Pringle provided helpful comments on an early draft of the paper. The study was funded by CSIRO and the Biodiversity Group of Environment Australia.