Plant responses to agricultural intensification

Authors

  • Josh Dorrough,

    Corresponding author
    1. Arthur Rylah Institute for Environmental Research, Department of Sustainability and Environment, Victoria 3084, Australia; and
    2. Future Farm Industries Co-operative Research Centre, 35 Stirling Hwy, Crawley, Western Australia 6009, Australia
      *Correspondence author. Josh Dorrough, CSIRO Sustainable Ecosystems, GPO Box 284, Canberra, ACT 2601, Australia. E-mail: josh.dorrough@csiro.au
    Search for more papers by this author
  • Michael P. Scroggie

    1. Arthur Rylah Institute for Environmental Research, Department of Sustainability and Environment, Victoria 3084, Australia; and
    Search for more papers by this author

*Correspondence author. Josh Dorrough, CSIRO Sustainable Ecosystems, GPO Box 284, Canberra, ACT 2601, Australia. E-mail: josh.dorrough@csiro.au

Summary

  • 1A large proportion of the world's land surface is extensively managed for livestock production. In areas where livestock systems are becoming more intensive, a major challenge is to predict those plant species likely to decline, persist or increase as a result of agricultural intensification.
  • 2Most analyses develop inferences for frequent or abundant species, or rely on intensive studies of single species. A promising approach is to identify plant traits related to disturbance to enable inference to be made about changes in plant community composition. We used a Bayesian hierarchical model to analyse the response to agricultural intensification of 494 plant species of pastures and woodlands in southern Australia, and to identify how simple species’ traits (life form, growth form and species origin) influence those responses.
  • 3The probability of occurrence of most species declined along the two intensification gradients, grazing intensity and soil phosphorous concentration, although the occurrence of a greater proportion of species was negatively correlated with soil phosphorous. Responses could be broadly predicted from both plant origin and plant traits, in particular growth form.
  • 4Native perennial geophytes, ferns and shrubs were most negatively affected by both gradients, while exotic annual grasses and forbs were more tolerant. Along the phosphorous gradient, 24 of the 30 most negatively affected plant species were native geophytes. Mean within-group responses masked considerable within- and between-species variation, particularly for the exotic species group which included species that responded both negatively and positively to intensification.
  • 5Synthesis and applications. The hierarchical model described here provides a powerful method for estimating individual plant responses and identifying how species’ traits influence those responses. Plant species native to southern Australia are sensitive to grazing and phosphorous apparently due to a shared evolutionary history of low grazing intensity and low phosphorous soils. Invading exotic plants have faced strongly contrasting ecological filters, leading to a greater diversity of responses. Where grazing systems have been most intense, a small suite of exotics dominate. Maintaining native and functional plant diversity will necessitate limits being placed on intensive livestock management systems.

Introduction

Grasslands and savanna woodlands support extensive grazing systems throughout the world. Although in some regions land abandonment is a dominant feature (MacDonald et al. 2000), in many extensive grazing systems there is a trend towards intensification with higher application of fertilizer and increased livestock densities (Tilman et al. 2001). While a trend towards intensification of grazing lands has been particularly notable in the developing world, it is not restricted to these areas (e.g. Donald et al. 2006; Dorrough, Moll & Crosthwaite 2007). Local, regional and global loss of plant diversity is predicted as a result of intensification (Tilman et al. 2001) but actual impacts are likely to vary under different scenarios. One of the major challenges facing conservation biologists is predicting which species will decline or increase in response to changes in land use, and to develop appropriate management strategies to maximize native plant species diversity at a range of spatial scales.

It is widely recognized that there is a need to improve our ability to predict species’ responses to human-induced global change, often based on limited data (Guisan et al. 2006). Long-term occurrence data, in relation to land use, are available for few species or localities and models are often developed based on presence–absence of information among areas with contrasting conditions. Predictions are typically limited to the most frequent species (Stockwell & Peterson 2002). However, it has long been recognized that within a given landscape, most plant species are infrequent and rare (Raunkiaer 1934; Preston 1948; Stohlgren et al. 2005) and conservation management often requires information about the responses of those species. Responses of abundant taxa may not reflect changes in less frequent species nor should abundant species necessarily be the most sensitive to changing land use (McIntyre & Lavorel 1994; Cao, Larsen & Thorne 2001). Indices of community composition and diversity (e.g. species richness or Simpson diversity index) are often used as surrogates, but are rarely satisfactory substitutes if species-specific information is required.

Estimates of the statistical distribution of individual species’ responses to land use within plant life history, functional or trait groups could be used to assist in the prediction of the responses of less frequent taxa. It is generally accepted that certain traits exhibited by plants influence their response to environmental or resource gradients (Lavorel et al. 1997; McIntyre et al. 1999). This has generated much interest in classification of individual plant species into syndromes or groups based on shared traits in an attempt to develop predictions of species’ responses to environmental change (e.g. McIntyre, Lavorel & Tremont 1995; Chapin et al. 1996; Diaz et al. 1999; Landsberg, Lavorel & Stol 1999). In addition to developing a general understanding of the unifying processes influencing species distributions, this approach is appealing because it could aid in estimating responses of rarely recorded taxa or predicting vegetation change in less-studied environments.

Rare and infrequent species also pose problems for estimating the distribution of disturbance responses within a given trait. This has been recognized in macroecology where exclusion of infrequent species has led to bias in estimates of trait distributions, e.g. body size and extinction probability (Blackburn & Gaston 1998). The distribution of plant responses within functional or trait groups could also be biased by a failure to include rare and infrequent species, particularly if rare species differ in either their responses or trait distributions. The responses of infrequent species to disturbance or environmental gradients are often uncertain, and thus, these species are typically ignored (e.g. Diaz & Marcelo 1997; Jauffret & Lavorel 2003).

Two factors likely to influence plant responses to intensification of livestock systems are increases in the total density of livestock and changes in the concentration and availability of plant nutrients. Numerous plant traits have been identified that are correlated with responses to livestock grazing (Diaz et al. 2007) and changes in soil productivity in agricultural landscapes (Kleyer 1999; McIntyre & Lavorel 2007). Increasing effort has been directed towards trying to identify how specific traits related to resource capture (leaf nitrogen, specific leaf area) and regeneration (seed mass, seed number) might influence how plants respond to disturbance and land use change (Lavorel et al. 1997; McIntyre et al. 1999; Garnier et al. 2007). However, trait values are often unknown for many species and the traits can vary within as well as among species. In contrast, simple life-history attributes can be obtained for whole regional flora and thus provide a useful starting point for the development of predictions (e.g. Mayfield, Ackerly & Daily 2006). Here we restrict our analyses to stable traits (sensu Garnier et al. 2007) that can be obtained from the literature. Our aim is not to test all likely traits related to intensification, rather it is to develop inference for a few easily accessible traits and then to use this information to assist in inference for individual species.

We develop a Bayesian hierarchical model of the responses to livestock density and available phosphorous of almost 500 plant species observed during systematic surveys in pasture and woodlands of south-eastern Australia. Using these data, we simultaneously estimate the distribution of responses to grazing and available soil phosphorous for (i) all observed plants species, and (ii) a range of simple plant traits (longevity and growth form) and species origins (native or exotic).

Prior to European settlement in the early 1800s the landscape supported extensive grassy woodland (savanna woodland) and grassy forests which have since been largely cleared to support livestock grazing and localized cropping. Pastures of introduced perennial grasses and annual legumes have been sown where soils and topography have allowed. Phosphorous fertilizer has been widely applied but at varying intensities. These management practices have created a mosaic of intensification with varying composition of associated plant species and traits (McIntyre & Lavorel 2007). There is a growing body of research examining grazing responses in these landscapes but few studies have considered responses of more than a few dominant species to fertilizer, although dramatic declines in native plant species richness have been reported (Dorrough et al. 2006). Because Australian vegetation has evolved with low nutrient soils (Beadle 1954) and a short history of ungulate grazing (Moore 1959; Milchunas et al. 1989), responses of native plants to intensification are predicted to be negative, although potentially modified by life history, growth form, and species-specific variation.

Study area and methods

The study was undertaken on the inland slopes and hills of central Victoria, Australia between 37°17′ S, 142°55′ E and 36°11′ S, 146°28′ E, an area of approximately 40 000 km2 (Fig. 1 in Dorrough & Moxham 2005). The study area covers an altitudinal range of 150–600 m above sea level and average annual rainfall ranges from a low of 530 mm year−1 at Maryborough in the central-west of the study area to 670 mm year−1 at Benalla in the central-east with approximately 60% of rain falling between May and October (Australian Government Bureau of Meteorology, http://www.bom.gov.au).

Figure 1.

Observed phosphorous and grazing intensity values at the study sites.

The region contains a range of land systems ranging from alluvial plains, originally dominated by open eucalypt savanna woodland (predominantly Eucalyptus camaldulensis, E. melliodora and E. microcarpa), to slopes and hills of sedimentary and granitic origin supporting dry forests and woodland (dominant eucalypts include E. goniocalyx, E. polyanthemos and E. macrorhyncha). Following European settlement in the early 19th century, much of the overstorey vegetation was cleared; as little as 3% tree cover persists in areas managed for livestock production (Dorrough & Moxham 2005). Exotic pastures have been sown on up to 60% of the grazed landscape (Dorrough et al. 2007). The primary land use is sheep grazing with some cattle grazing and cropping.

Five spatially separated regions were selected centred around the towns of Ararat, Maryborough, Broadford, Violet Town and Springhurst (Dorrough & Moxham 2005). Within each of the five regions, data on management, vegetation and soils were collated from three to four private farms (across 17 farms in total) and adjacent public reserves. The reserves had no history of direct fertilizer application and minimal livestock grazing.

Four hundred and twenty six 0·09-ha plots were pre-selected across the 17 farms using a stratified random design based on soils, topography and aspect, past land use (cultivation, grazing and fertilizer history) and tree cover. Plots on farms spanned an intensification gradient from little grazed, unfertilized and uncultivated woodlands to intensively grazed, fertilized and sown pastures. Between six and 88 plots were sampled on each farm depending on farm size and number of stratification levels encountered. An additional 70 plots were sampled on public lands, intentionally avoiding areas with obvious signs of soil disturbance or run-on from roads or adjacent fertilized pastures. Although an attempt was made to sample lightly stocked, high-phosphorous and heavily stocked, low-phosphorous areas, the two gradients positively co-vary to some extent (Fig. 1).

In an attempt to maximize the number of species observed at a particular site, we used a large plot (30 × 30 m) and varied our search time based on the number of species being encountered. A minimum of 30 min searching was conducted at each plot by two botanists, although much longer was taken in sites with high plant diversity (~90 min). Plots were searched until we were satisfied no new species could be added without considerable additional search time. All sampling was undertaken in late spring, between October and December of 2002 and 2003, when most plant species could be readily identified. In each plot, we recorded the identity of all observed plant species, although no abundance data were obtained.

Plants were either identified to species in the field or collected for later identification. Plants were identified using the Flora of Victoria (Walsh & Entwistle 1994, 1996, 1997) and nomenclature follows Ross & Walsh (2003). In a few cases, plants could only be identified to genus. Plants within some genera were rarely identified to species (e.g. Vulpia spp., Aira spp., Phalaris spp.), and in these cases, all occurrences were pooled to genera.

Within each plot, five random soil samples (depth 3–10 cm, after removal of surface litter), were collected and then bulked. Soil samples were analysed for Colwell available phosphorous (Colwell 1963). Extractable phosphorous is correlated with a history of fertilizer application in these low-phosphorous soils (Burkitt, Gourley & Sale 2002; Dorrough et al. 2006).

At each plot, recent livestock grazing density (stocking rate) was estimated as dry sheep equivalents (dse) per ha (a single non-lactating ewe is equivalent to one dry sheep). This was determined through consultation of recent stocking records and discussions between a consultant agronomist and the landholder.

Although there are many environmental and land-use gradients that could be considered, here we only explore available soil phosphorous, as an indicator of fertilizer history, and the density of livestock as measured in dse ha−1. A number of plots had been cultivated in the past and cultivation tends to be correlated with heavy grazing and fertilizer. Thus, phosphorous and livestock broadly represent an intensification gradient but they are not exhaustive in describing what can be a complex of land uses.

Plant origin and two stable traits, life history (longevity) and growth form, were obtained for all plants from published sources. The source and trait-specific predictions for each are briefly outlined below.

Plant origin, either native or exotic, was determined from the census of vascular plants of Victoria (Ross & Walsh 2003). The evolutionary history of ungulate grazing can influence how plant species and communities respond to livestock grazing (Milchunas et al. 1989). In an Australian context, it is generally assumed that native plant species respond more negatively to agricultural intensification than exotics. This is thought to be because native species did not co-evolve with hard-hoofed ungulates (Bennett 1999) or high soil phosphorous availability (Beadle 1954; Groves, Austin & Kaye 2003). Agricultural management, which now dominates the Australian land mass, would impose a major filter on those introduced species likely to successfully invade (Moore 1959; McIntyre & Martin 2001).

Plant life history (perennial or short lived annual/biennial) was obtained from the Flora of Victoria (Walsh & Entwistle 1994, 1996, 1997). Much research has indicated that longevity is negatively correlated with increasing intensity of livestock grazing (for review and meta-analysis see Diaz et al. 2007). Increasing soil fertility is expected to favour short-lived plant species with high relative growth rates as they may be more competitive under greater resource concentrations (Grime 1977; Chapin 1980).

Each plant species was attributed to one of five growth forms using published sources (Walsh & Entwistle 1994, 1996, 1997; Lunt & Morgan 1999; Parsons 2000; Williams et al. 2005); forb (herbaceous dicot), shrub (woody dicot), fern, non-geophyte graminoid (primarily Poaceae, Cyperaceae, Juncaceae but also some Liliacae and Iridaceae with persistent above-ground vegetation), geophyte monocots (mainly Liliaceae, Iridaceae and Orchidaceae). Variation in grazing response among growth forms has been observed, although responses can be inconsistent and regionally specific (Vesk, Leishman & Westoby 2004; Diaz et al. 2007). In a temperate Australian context, shrubs, forbs and geophytes are often sensitive to grazing while grasses tend to be tolerant (McIntyre, Lavorel & Tremont 1995; Pettit, Froend & Ladd 1995; Clarke 2003; Dorrough, Ash & McIntyre 2004; Leonard & Kirkpatrick 2004). Grasses and forbs are predicted to respond positively to increasing soil fertility, while responses of shrubs and geophytes, often highly dependent on mycorrhiza for phosphorous acquisition, are likely to be negative (Heddle & Specht 1975; Rasmussen 2002).

model description and data presentation

Numerous approaches have been used to explore species’ responses to environmental gradients and to develop inference about whether responses vary among species because of differences in their traits (e.g. Boutin & Keddy 1993; Ackerly & Cornwell 2007). Often, analyses use a two-step process: (i) ordination or regression to define plant responses to the environmental or disturbance gradients, and (ii) examination of the distribution of these responses in relation to traits (e.g. Landsberg et al. 1999). Individual species are often classified into response groups, e.g. increasers, decreasers, neutral, and the distribution of traits explored within each (e.g. Diaz, Noy-Meir & Cabido 2001). Such classification schemes often ignore both the variation among species within response groups and the variation within species (but see Vesk & Westoby 2001). One major limitation with classification and ordination approaches is the difficulty of objectively assessing uncertainty around predictions for individual species, groups of species or the whole assemblage.

An alternative approach, and the one we take here, is to model the occurrence of a species at a locality given its taxonomic identity, its traits and the environmental or disturbance values for the location (Gelfand et al. 2005). There are several advantages of this approach. First, it is possible to directly incorporate uncertainty about species’ responses and variation among species within trait groups. Secondly, all species observations are used to develop inferences for the effects of the intensification gradients on the occurrence of traits. This is particularly important for those trait groups that may be species-rich but where species turnover is high and few individual species are frequent (e.g. native perennial geophytes). Thirdly, parameter estimates for individual species are informed by their traits. For frequent species, this is likely to improve precision, and for infrequent species this information can be used to estimate response coefficients even where there are few observations.

We modelled the probability of occurrence of the ith plant species at the jth site (pij) as a function of the soil phosphorous and grazing intensity covariates using a standard logistic regression equation:

image(eqn 1)

Both covariates were standardized to have a mean of zero and variance of one. The parameters αi are the species-specific intercept terms, representing the log-odds of occurrence of each species at mean phosphorous and grazing levels. The coefficients β1i and β2i represent the species-specific responses to the phosphorous and grazing intensity gradients, respectively. The coefficients β1i and β2i are themselves decomposed into components representing the origin (native or exotic), life history (perennial or annual) and growth forms (forb, fern, geophyte, graminoid or shrub), through the use of binary dummy variables and associated coefficients (denoted as b and c) in the models:

β1i = b1 + b2 · nativei + b3 · annuali + b4 · ferni + b5 · geophyteib6 · graminoidi + b7 · shrubi + ɛ1i(eqn 2)
β2i = c1 + c2 · nativei + c3 · annuali + c4 · ferni + c5 · geophyteic6 · graminoidi + c7 · shrubi + ɛ2i(eqn 3)

In addition, the parameters β1i and β2i also include a set of random effect terms, ɛ1i and ɛ2i. The random effects are normally distributed, and represent the deviation of each individual species from the mean phosphorous and stocking responses that would be expected given their origin, life history and growth form.

The model was fitted to the data using Bayesian Markov chain Monte Carlo (MCMC) methods with the software package winbugs version 1·4·1 (Spiegelhalter et al. 2004). A primary benefit of the Bayesian approach was the ease with which the complex hierarchical error structure (i.e. the random effects terms in the model) could be accommodated in the face of a non-Gaussian error distribution and the unbalanced structure of the data. Vague (non-informative) priors were specified for all parameters. The series of b and c terms were given vague normal priors of N(0, 1000), as were the intercept terms α. The variance of the random effects terms ɛ1 and ɛ2 were given vague, inverse gamma priors.

A very small number of observed plants had known growth forms but unknown origins and life histories; i.e. we were unable to identify them to species level and ascertain whether they were of native or exotic origin, or whether they were perennial or annual. These missing values in the data set were estimated as part of the model-fitting process. Bernoulli (0·5) priors were used for these binary values, as we considered either origin or life history state for each of these species to be equally likely a priori.

A burn-in of 500 iterations of the MCMC algorithm provided adequate convergence, as assessed by calculating the Brooks–Gelman–Rubin diagnostic statistic for the output of three replicate Markov chains with overdispersed starting values (Brooks & Gelman 1998). Convergence was also assessed by visual inspection of the chain histories of selected parameters. Final inferences were based on 8000 iterations of a single chain, collected after the burn-in period of 500 iterations.

Samples were obtained from the posterior probability distributions of the individual parameters describing the effects of plant attributes (i.e. origin, life history and growth form) on plant responses to grazing and phosphorous. Similarly, we also sampled from the posterior probability distributions of the effects of grazing and phosphorous on each of the 12 major combinations of traits (e.g. exotic perennial graminoid, native perennial forb; hereon referred to as trait groups), and from responses of each of the individual plant species. Posterior probability distributions were summarized by their posterior means and 95% credible intervals (CI).

Results

species frequencies

Approximately half of the species observed were recorded in less than 1% of plots. An additional 30% of species were recorded in 1–5% of plots. Only two native plant species occurred in more than 30% of all plots, Lomandra filiformis (perennial graminoid) and Microlaena stipoides (perennial graminoid). Although most exotic plant species are also rare, seven exotic plant taxa were observed in more than 50% of plots. These were Aira spp. (annual graminoid), Trifolium subterraneum (annual forb), Romulea rosea (perennial geophyte), Arctotheca calendula (annual forb), Bromus hordeaceus (annual graminoid), Hypochaeris radicata (perennial forb) and Vulpia spp. (annual graminoid).

frequency of traits in response to grazing and phosphorous

The total number of species per plot declined along each of the intensification gradients but most sharply in response to phosphorous (Supplementary Material Fig. S1, and see Dorrough et al. 2006). With increasing grazing and phosphorous, a much smaller proportion of the observed species were native and perennial (Supplementary Material Fig. S1). Native and perennial plant species never dominated the plot-level species pool when available phosphorous was high. Considerable variation in the frequency of each of the growth forms was observed along the intensification gradients, particularly phosphorous availability. At low levels of phosphorous (between approximately 1 mg kg−1 and 20 mg kg−1), ferns, geophytes and shrubs often comprised 20% or more of all recorded species (Supplementary Material Fig. S2). At high levels of phosphorous most species were forbs or graminoids and other growth forms were rarely encountered.

analysis of responses to phosphorous and livestock within traits

On average, phosphorous and grazing have a general effect of decreasing the probability of species occurrence. This is particularly the case for available phosphorous, the effects of which are strongly negative (Table 1, Fig. 2). Although overall effects tend to be negative, responses vary owing to the trait attributes of the plants. Annuals tend to respond less negatively to phosphorous than perennials while natives have more strongly negative responses compared to exotics. Of the five growth forms, geophytes, ferns and shrubs tend to have more negative responses to phosphorous than either forbs or grasses. Responses to grazing are similar, although shrubs are weakly more sensitive than either geophytes or ferns, and grasses have a strongly positive response in contrast to the other growth forms. The probability of occurrence of native plants declines with increasing livestock density. Across the two intensification gradients, plant origin (native or exotic) and growth form most strongly modify the overall response to both livestock density and phosphorous, while the effects of plant life history are comparatively weak (Table 1).

Table 1.  Posterior means and 95% credible intervals for the parameters of the model of plant responses to phosphorous and grazing. The common effect is that expected for an exotic, perennial, forb. A response coefficient of ±1 indicates that an increase in either grazing or phosphorous of one standard deviation of the overall grazing or phosphorous distribution will increase/decrease the log-odds of occurrence of a plant group by one unit
 PhosphorousGrazing
2·5%Mean97·5%2·5%Mean97·5%
Common effect−0·59−0·33−0·07−0·31−0·140·06
Longevity – annual−0·240·030·30−0·140·070·27
Origin – native−1·14−0·86−0·59−0·84−0·62−0·44
Growth form – fern−2·30−0·790·61−1·23−0·370·49
Growth form – geophyte−2·06−1·35−0·64−1·02−0·63−0·23
Growth form – graminoid−0·26−0·060·130·060·220·40
Growth form – shrub−1·08−0·57−0·07−1·08−0·77−0·47
Figure 2.

Estimated mean response coefficients to both grazing and phosphorous for 12 major trait groups present in the grazing lands of southern Australia. Filled circles are posterior means and line segments denote 95% credible intervals.

The responses of the 12 trait groups reflect the patterns described above. None of the trait groups have 95% credible intervals for the response to phosphorous that include zero, although exotic perennial and annual graminoids and forbs have the least negative responses (Fig. 2). The phosphorous coefficient for native perennial geophytes is the most negative (Fig. 2). Although native perennial ferns have a similar posterior mean, the uncertainty around the estimate is considerable.

Native perennial shrubs have the most negative response to grazing, although only marginally less negative than either native perennial geophytes or ferns. Exotic graminoids and forbs, both annuals and perennials, have grazing response coefficients that approach zero or are weakly positive.

individual species responses

Individual trait and trait group posterior coefficients represent the expected mean response of species within each and tend to mask information about variation among individual species within groups. Sixty per cent of the species are predicted to respond negatively with certainty (i.e. 95% CI does not cross zero) to phosphorous in contrast to 38% to grazing (Fig. 3). Of the 30 most negative responses to phosphorous, 24 are native perennial geophytes and the remainder are native perennial ferns, shrubs, forbs or graminoids. The most negative responses to grazing are more broadly distributed among the trait groups, although few are exotic. Only 1·4% of species are predicted to have a certain positive response along the phosphorous gradient in contrast to 5% for grazing (Fig. 3). With the exception of one native perennial forb (Urtica incisa), all species responding positively to phosphorous were exotic annuals. The majority of species with positive responses to grazing were exotics, primarily annual grasses and forbs. There are also a large number of species with either strongly negative or positive mean responses but with credible intervals crossing zero (Fig. 3). The posterior distributions of the responses of each species to either gradient, based on the 8000 MCMC iterations, was used to provide an estimate of the overall proportion of species responding positively to each gradient. For phosphorous, the posterior mean proportion of species with positive responses is 11% (95% CI of 9–14%), while in response to grazing, 24% of species are estimated to have positive responses (95% CI of 21–27%).

Figure 3.

Estimated response coefficients to phosphorous and grazing for 494 individual plants species. Each panel includes a single-trait group. Line segments delimit the 95% credible intervals for the phosphorous and grazing responses of each species.

Variation among individual species within native plant life-history groups was generally small, particularly along the phosphorous gradient (Fig. 3). In contrast, there was much among-species variation within exotic life-history groups. In many cases, the exotic life-history groups included species with both strongly positive and strongly negative responses to both grazing and phosphorous.

Discussion

We have described an approach to model the responses of all observed plants to two dominant agricultural intensification gradients, while estimating how responses are modified by trait attributes. The estimates we present here indicate that within the grazing landscapes of southern Australia, few species or trait groups are tolerant of the intensification gradients represented by phosphorous availability and livestock density. Our prediction that native plant species would be most susceptible to both intensification gradients was upheld, although the strength of the response to the phosphorous gradient was less expected. The results suggest that native species are more susceptible to increasing soil fertility than to increases in livestock density, although due to co-variation in the intensification gradients, their specific effects are difficult to disentangle.

patterns within and among trait groups

Despite considerable within- and among-species variation, plant origin and growth form both proved to be useful predictors of plant responses to grazing and phosphorous. Life history (annual vs. perennial) proved to be less informative. Several prior studies have indicated that life history is often a more useful predictor than either origin or growth form, particularly when describing response to grazing (Dorrough et al. 2004; Diaz et al. 2007). Differences in grazing response among perennial and annual plant species is one of the most well-accepted and widely observed patterns (Noy-Meir, Gutman & Kaplan 1989; Fensham, Holman & Cox 1999). In our study, observed increases in the frequency of annuals with intensification were probably confounded with changes in the likelihood of occurrence of native plants, most of which are perennial, and of perennial-dominated growth forms (e.g. shrubs or geophytes).

Prior ecological and evolutionary processes within southern Australia should have favoured species capable of persisting under both low available phosphorous and low herbivore densities. Although an evolutionary history of aridity, particularly during the last glacial maximum (ca. 30 000–20 000 years bp), and fire could have conferred traits that provide resistance to intensive herbivory, these disturbances are likely to differ in their specific effects and subsequent plant responses. Resource-rich patches would have occurred in the landscape prior to European settlement, although they are likely to have been both temporally (e.g. nutrient pulses following rain, flood or fire) and spatially (e.g. basalt-derived soils) restricted. In contrast, exotic plant species have invaded, with both intentional and accidental human assistance, into landscapes now supporting strongly divergent phosphorous concentrations and livestock densities (McIntyre & Lavorel 2007). Our results suggest that this has selected for a range of species with varying responses, including species able to tolerate low-phosphorous soils and light grazing and those capable of persisting under the alternate extreme. The current vegetation communities now arguably comprise at least two distinct species pools. The first occurs in low phosphorous, lightly grazed environments and is dominated by natives with a small number of exotics sharing similar responses. The species pool in these environments is also primarily perennial and often supports a range of growth forms (Supplementary Material Figs S1 and S2). Areas with high soil phosphorous and grazing intensities support a contrasting compositionally depauperate pool of species primarily of exotic forbs and graminoids, many of which are annual. This supports the general prediction that in regions with a native flora that has evolved in the absence of ungulate grazing and with low soil nutrient availability, the impacts of agricultural intensification are devastating.

Geophytes, ferns and shrubs tended to be most intolerant of intensification via fertilizer or grazing. These observations, particularly the intolerance of shrubs to intensive grazing, are supported by previous studies in Australian temperate woodlands (Pettit et al. 1995; Clarke 2003). The strong negative relationship between probability of geophyte occurrence and phosphorous availability was as predicted. The geophyte flora is dominated by orchids and lilies, and terrestrial orchids in particular are dependent on symbiotic associations with mycorrhiza for germination and growth (Rasmussen 2002). In Australian soils low in plant-available phosphorous, many native plant species establish symbiotic mycorrhizal relationships to increase phosphorous uptake (Warcup 1980). Mycorrhizal abundance generally declines under phosphorous fertilization (Treseder 2004), and where the majority of the plant species depend on mycorrhizal fungal infection, loss of mychorrhiza owing to fertilization could be the proximate cause of significant losses in plant diversity (Hartnett & Wilson 2002). Other mechanisms, including interspecific plant competition, could also play a role in the observed negative correlation between plant occurrence and phosphorous availability.

analysis of species responses

Identifying how species are distributed along environmental or disturbance gradients continues to be a major area of research in ecology. While historically such research was directed at understanding how and why species are distributed in space and time, there has been increasing urgency to identify those species that are likely to be most impacted (positively or negatively) by rapidly changing land use and human-induced climate change. Although an understanding of the patterns exhibited by dominant and widespread species is useful, approaches to identifying the response of less-frequent species to changing land use are increasingly needed.

One potential application of trait analyses is to develop predictions about the likely responses to intensification of less frequently observed taxa that share traits with more abundant species. We know of no other studies that have explicitly attempted to develop predictions for less abundant species using this approach. In this current study, inferences were made about species observed in less than 1% of plots. To make this possible, (i) all species were included in model development and contributed to estimates of the trait effects, and (ii) trait information was used to improve estimates of the responses of individual species to the intensification gradients.

While estimates of the likely effects of either grazing or phosphorous on each species were made, precise prediction of the current spatial distribution of a particular rare species would require further data collection. However, spatial predictions would be improved by incorporating the results presented here through construction of informative priors, based on the species’ identity and trait group membership.

Few studies of functional or trait group responses to disturbance include information about infrequent species. This oversight could result in erroneous conclusions about the statistical distribution of responses within and among trait groups, particularly where dominant species differ in their responses to less-frequent species (Cao et al. 2001). Trait response groups derived from abundant species may describe emergent ecosystem function relatively well, but fail to predict changes in abundance and composition of less-frequent species, including native species or exotic invaders that may be of concern for conservation.

Our model relies on several assumptions and simplifications, and there is much room for improvement. Firstly, we have assumed species do not exhibit unimodal responses and can be adequately described using a simple logistic form. We recognize that unimodal responses are common (Austin 2002). Consideration of alternative response shapes would increase model complexity but could lead to significant insight. Secondly, we assumed that species responses were to the measured land-use gradients and a result of their various traits and individual species variances. Interactions among species, either competitive or facilitative, are likely to influence how plants respond (Guisan et al. 2006). We also assume that every species could potentially occur at every site. Species could have been excluded for numerous biological and biogeographical reasons unrelated to phosphorous or livestock and not explained by either their longevity, origin or growth form. We assume that these factors increase the uncertainty around a species’ inferred response and that better knowledge would improve these estimates. Finally, to minimize model complexity, we have not attempted to model interactive effects of phosphorous and grazing, although such interactions are likely.

implications for land management

Intensification of agriculture within the woodlands of southern Australia has caused dramatic declines in plant diversity and underlies significant changes in critical ecosystem functions (e.g. nutrient cycling, carbon storage and soil protection) as a result of changes in plant longevities and growth forms (McIntyre & Lavorel 2007). At a paddock scale, preventing nutrient enrichment and limiting livestock numbers will be required to maximize the diversity of trait groups and native plant species. Identifying those areas with a history of limited nutrient enrichment and light grazing should be a primary objective. At regional scales, controlling the extent of intensive agriculture will be important for conserving the maximum number of native plant species and limiting impacts on key ecological functions. Most plant species are infrequent, indicating the need to spread conservation management efforts among sites. Maintaining extensive grazing systems could have significant conservation benefits in these landscapes.

conclusion

As human demand for resources rapidly increase, so too will agricultural intensification and its impacts on native plant species. In some cases, abandonment of marginal grazing lands is occurring, with a legacy of raised soil nutrients and novel plant communities. While it is important to understand how dominant species will respond to these changes in management, such information may tell us little about the responses of the majority of species. Incorporating information about both infrequent and frequent species into models of vegetation change in response to land use and global climate change will continue to be a major challenge. We have shown that simple information about species’ life history, growth form and origin can be used to assist development of individual plant species predictions and could have significant benefit for the management of vegetation.

Acknowledgements

Claire Moxham, Geoff Sutter, Gary Cheers, Rhiannon Apted, Nathan Wong, Vivienne Turner, Marc Bellette and Dale Tonkinson assisted with data collection. Thanks to Geoff Sutter, David Cameron and Nathan Wong for plant species identification. MPS thanks Graeme Newell and Peter Griffieon for access to computing resources. Thanks to Peter Vesk, Sue McIntyre, Mike Smith, Julian Reid, Andy Sheppard and two anonymous referees for insightful comments on earlier drafts. Data were collected with funding support from the Land Water and Wool Native Vegetation and Biodiversity Sub-program and the Land & Water Australia Native Vegetation R&D Program. Analysis and manuscript preparation has been assisted by funding through Meat and Livestock Australia.

Ancillary