Functional traits and prior abundance explain native plant extirpation in a fragmented woodland landscape


*Correspondence author. E-mail:


  • 1Habitat destruction and the resulting fragmentation increase the probability of local extirpation of plants but the process of community disassembly may take many years. There are very little direct data on the rate of this process. Revisitation studies using historical data offer the potential to quantify the rate and direction of change in remnant vegetation and can contribute to an improved understanding of the long-term species responses to habitat fragmentation. This study examined the changes that occurred in 12 grassy woodland remnants in rural southern Australia over three decades by comparing comprehensive site-level species lists collected in 1975 and 2006.
  • 2The probability of local population extirpation was assessed in relation to functional- and abundance-based mechanisms (i.e. plant functional traits, initial population abundance) and extrinsic threats (i.e. remnant area, remnant shape) using chi-square analysis and Bayesian logistic regression modelling.
  • 3Of the 775 populations of 177 perennial native species present in 1975, 199 populations (26%) of 98 species were not relocated in 2006 and are presumed to be locally extinct.
  • 4Extrinsic factors such as habitat area and shape provided little predictive power to understanding local population extirpation. Intrinsic factors (i.e. initial population abundance, seed dispersal and plant height, but not the plant traits clonality and life form) were better predictors. Populations of short species at low abundance in 1975 were most likely to become locally extinct due to inter-specific competition in the absence of disturbance whereas tall, abundant species have the least likelihood of population extinction.
  • 5Synthesis. The sensitivity of populations to local extirpation in fragmented landscapes is likely driven by a number of factors. The lack of substantial influence of site area and shape and some plant life-history traits suggest that simple predictions about population extinction are unlikely. Although population size was an important determinant of population loss, it is likely that other factors will interact to determine such outcomes (e.g. disturbance regimes, non-native plant invasions, edge effects) by affecting habitat quality, and put populations at greater risk of extinction than the life-history traits of plants alone.


Despite a general consensus that ongoing landscape transformation due to agriculture and urbanization is a major threat to species diversity, we are still far from knowing the mechanisms that underpin species decline and loss in remnant habitats (Harrison & Bruna 1999; Eriksson & Ehrlen 2001; Brook et al. 2008). Much research has focused on the effects of spatial landscape structure (i.e. area and isolation effects), as well as on the influence of more local processes on habitat quality (e.g. edge effects, non-native plant invasion, disturbance). Species may respond to such changes quickly (e.g. Leach & Givnish 1996), but there may also be a long time-lag in response to fragmentation (i.e. the extinction debt sensu Tilman et al. 1994). Hence, a critical issue for conservation biology is not only to assess the effect of spatial structure but to also include a temporal scale of the biodiversity response to landscape transformation (Brys et al. 2005; Helm et al. 2006; Vellend et al. 2006).

In fragmented landscapes, remnant vegetation is considered to be undergoing a long-term process of community disassembly or ‘relaxation’ (Diamond 1972) with local extirpation the dominant population process (Andren 1994). The time-dependent nature of the relaxation process necessitates that there will be a period after fragmentation when relaxation has yet to take effect (Tilman et al. 1994), but there are very little direct data on the rate and form of this (Gonzalez 2000; Davies et al. 2004). Revisitation studies enable the direction and rate of vegetation change to be quantified and are being increasingly used to assess long-term changes in remnant vegetation (e.g. Kirkpatrick 1986, 2004; Leach & Givnish 1996; Fischer & Stöcklin 1997; Duncan & Young 2000; DeCandido 2004; van Calster et al. 2008). When revisitation studies are coupled with a plant functional trait analysis of the compositional changes, a functional understanding of vegetation dynamics in fragmented landscapes can be more fully elucidated (Robinson et al. 1994; DeCandido 2004; Williams et al. 2005; Wiegmann & Waller 2006; van Calster et al. 2008). Such an understanding may improve the conservation management strategies aimed at encouraging the persistence of native flora in highly modified landscapes (Kirkpatrick 2004) and may identify those species (or groups of species) that are at greatest risk of extinction (Davies et al. 2000; Williams et al. 2005; Hoehn et al. 2007).

Revisitation studies compare historical and current records of the occurrence of plants (or animals) to assess local persistence and hence appear relatively straightforward measures of change. Changes in the vegetation can be assessed at the scale of the permanent plot (local-scale); however, investigations are likely to be more informative when undertaken at the scale of the site (large-scale) because population extinction rates, rather than plot extinction rates, can be determined. For revisitation studies to usefully inform our understanding of remnant vegetation dynamics, including local extinction rates, several important factors need to be considered. One of the greatest limitations of revisitation studies is that they can over-estimate extinction rates (Alpizar-Jara et al. 2004; Kery 2004; Shaw 2005). This occurs because plants, despite their sessile nature, vary in their detectability (i.e. detection probability is <1; Kery & Gregg 2003; Garrard et al. 2008) and hence, it is likely that not all extant individuals are detected when a site is revisited (i.e ‘false absences’ are recorded). Species may vary in their detectablilty due to differences in population size (Alpizar-Jara et al. 2004), plant size (Brown et al. 2004), phenological state and habitat type (i.e. dense vs. open vegetation) (Kery 2004), whereas observer (in)experience also introduces a level of uncertainty into detection rates (Kery & Gregg 2003). Additionally, climatic conditions in the year of observation (e.g. rainfall amount and timing) can affect the likelihood of detecting seasonal species such as annuals (Kirkpatrick 2004; Dostal 2005) and geophytes (Lesica & Steele 1994; Morgan 1998). Many of these inherent limitations, however, can be reduced when vegetation is sampled at multiple times across a year.

In southern Australia, natural woodland vegetation has been extensively cleared since European settlement in the late-19th century (Lunt & Bennett 2000) and remnants now sit in a landscape matrix dominated by agriculture (Prober & Thiele 1995; Fischer & Lindenmayer 2002). Despite the importance of the remaining woodland remnants for the conservation of regional biota (Prober & Thiele 1995; Radford & Bennett 2007), remarkably few studies have assessed how fragmentation affects long-term species persistence in these landscapes (but see Williams et al. (2005, 2006) for an obvious exception in native tussock grasslands). Hence, the aim of the current study was to use a revisitation approach to assess the changes that have occurred in the floristic composition of remnant grassy woodlands in southern Australia over the last three decades. Using species lists collected in 1975 and 2006, the study aimed to document the probability of local extinction of native plant species populations and whether site spatial characteristics or plant functional traits could explain the changes that have occurred. Additionally, we aimed to determine whether population abundance in 1975 affected the probability of local extinction of populations by 2006. Despite a general consensus that small populations are prone to local extinction more so than larger populations, this has rarely been quantified (Fischer & Stöcklin 1997; Davies et al. 2004; Matthies et al. 2004; Dostal 2005; Brook et al. 2008; van Calster et al. 2008).


study area

The study was conducted in remnant grassy woodland vegetation on the Brim Brim Plateau and Dundas Tablelands in western Victoria, Australia, encompassing an area of approximately 700 km2. The topography is undulating and soils are shallow, acidic (pH 5.0) clays and sandy loams (Gibbons & Downes 1964). Mean minimum temperature is 3.3 °C (July) and mean maximum temperature is 27.5 °C (January), while rainfall averages 675 mm per annum, with a distinct winter-spring peak (Bureau of Meteorology, unpublished data 2007).

The native vegetation is likely to have been dominated by grassy woodlands (a type of savanna) at the time of European settlement (Blackburn & Gibbons 1956; Willis 1971). Tree dominance (Eucalyptus camaldulensis, E. viminalis, E. ovata) varies depending on soil moisture and nutrient availability (Gibbons & Downes 1964). The understorey was mostly herbaceous, dominated by the tussock grasses Themeda triandra, Austrostipa spp. and Poa spp. with minimal occurrences of tall shrubs in the natural ecosystem (Blackburn & Gibbons 1956; Gibbons & Downes 1964; Willis 1971; Barlow 1996).

The region was one of the earliest settled by Europeans in Victoria (in the 1850s) and was initially used primarily for sheep grazing (Spreadborough & Anderson 1983). After World War II, soldier re-settlement reduced the average size of landholdings and the use of superphosphate fertilizers and improved pastures intensified land use (Blackburn & Gibbons 1956; Gibbons & Downes 1964). This led to substantial conversion of the native woodland vegetation to a land use of intensive agriculture. By 1959, about 60% of the pastures had been ‘improved’ by the addition of fertilizers and the clearing of trees (Gibbons & Downes 1964). Hence, habitat destruction due to land-use intensification, with consequent fragmentation and isolation of remnant native vegetation, is likely to have occurred in the district by the 1950s and has steadily increased thereafter. Remnant grassy woodland vegetation with an intact understorey is now very restricted in the district (<5% of the area; Barlow 1996). Most high-quality (i.e. species-rich) remnants are confined to very small, isolated areas that have not been developed for agriculture, for example, linear railway verges, wide roadside verges (initially used as travelling stock routes), creeklines, cemeteries and occasional unimproved paddocks. In general, stock grazing in these remnant areas has been minimal for decades. Although many roadside and railway verge remnants in the study area were burnt frequently during the 1960s to 1990s to reduce fuel loads (Stuwe & Parsons 1977; Williams et al. 2006), most remnants have received little active management for over a decade due to rural population decline (Williams 2007).

floristic inventory data – 1975 and 2006

In October 1975, the floristic composition of 27 high-quality native vegetation remnants was surveyed by Cliff Beauglehole using a random walk methodology (of variable duration) as part of regional surveys of native vegetation on Crown land (Beauglehole 1984; Wannon Conservation Society, unpublished data). Sites were deliberately sampled because their floristic integrity was considered high in a landscape dominated by exotic pasture grasses. Beauglehole ranked the abundance of all species observed using a four-point scale: 1 =‘Very Rare’ (less than two dozen individuals seen across the site), 2 = ‘Rare’ (appearing in dozens), 3 = ‘Common’ (appearing in hundreds), 4 = ‘Very Common’ (appearing in thousands). Beauglehole was a biologist of high repute (Corrick 2002) and it is likely that his site-level surveys are botanically accurate. However, it is clear from his original notes that the surveys were not comprehensive site-level species lists, that is, he did not identify all species. Hence, while the fate of all native species populations identified in 1975 can be determined in subsequent surveys, changes in species richness and floristic composition of individual sites cannot be inferred between surveys. Therefore, we focused on the population process of most interest in fragmented ecosystems (e.g. Davies et al. 2000; Vellend et al. 2006; Williams et al. 2006), the probability of local extinction in the native species first observed in 1975.

In 2006, we tried to relocate all 27 remnants using the original maps and descriptions of site locations made by Beauglehole. Four sites had been destroyed, 11 could not be accurately relocated (i.e. site boundaries were unclear) while 12 sites were relocated and remained largely intact. We examined these 12 sites (see Table S1 in Supporting Information). On three occasions in austral spring 2006, the 12 sites were searched systematically by the authors and all vascular species observed were recorded. The time spent in each site was proportional to its size and heterogeneity; typically, survey effort varied from a total of 1.5–6 h per site per survey. Our surveys, similar to those employed by Williams et al. (2005) in nearby remnant native grasslands, were much more comprehensive than the original work of Beauglehole and hence we believe that we have minimized the number of false absences in the data set. The area and perimeter to area ratio of each remnant were determined from site boundary measurements. Taxa difficult to identify in the field were collected as herbarium specimens and identified using the botanical keys in Walsh and Entwisle (1995, 1996, 1998). Each species was given an abundance score to indicate its relative abundance across each site using the four point scale used by Beauglehole. Voucher specimens of all species recorded in the surveys were submitted to the La Trobe University Herbarium in December 2008.

All taxa recorded in either year of survey (1975, 2006) were classified according to origin (native, non-native) and life form (using the categories of McIntyre et al. 1995; see Table S2) based on information provided in the local floras of Walsh and Entwisle (1995, 1996, 1998) (Appendix S1). As 2006 was a severe drought year in south-eastern Australia, all annual species were omitted from the analysis because it is likely that moisture limitation strongly limited their abundance (Levine et al. 2008). Native species were also classified according to three plant functional traits that may predict local persistence (Williams et al. 2005; see Table S2; see Supporting Information in Appendix S1): (i) capacity for vegetative spread (>10 cm lateral spread of perennial species), (ii) method of seed dispersal (ant, ingestion, adhesion, wind, mobile, undefined) and (iii) height. Traits were assigned using McIntyre et al. (1995), Lunt (1997) and Williams et al. (2005) as a guide, supplemented by reference to local floras, field observations and expert botanical opinion. In this article, we focus only on native species responses to fragmentation.

A vexing and difficult problem during the assembly of the botanical database to assess changes over time is the need to resolve ambiguities due to differences in taxonomic nomenclature (Robinson et al. 1994). The original lists of Beauglehole contain species which have undergone substantial taxonomic revision (e.g. the orchid genus Caladenia). The approach used in the current study was to aggregate the observed taxa to the degree necessary to ensure comparability at the two sampling times, with nomenclature following Walsh and Stajsic (2007). This approach will underestimate current (and previous) levels of species richness at the sites and hence probability of local extinction, but is largely an unavoidable product of conducting long-term research in a taxonomically uncertain flora.

data analysis

The change in abundance of plant populations from 1975 to 2006 was compared using a chi-square analysis. The influence of area and shape, and life form, height, dispersal mode and ability to grow vegetatively on the probability of local extinction were analysed using Bayesian logistic regression with Monte-Carlo Markov Chain (MCMC) random sampling in the WinBUGS Version 1.4.1 program (Spieglehalter et al. 2004). Bayesian analyses are being used more often in ecology as they have many advantages over frequentist statistics that rely on null hypothesis significance testing (Stephens et al. 2007). Bayesian statistics allow several competing hypotheses to be compared simultaneously. The hypotheses are modelled and then the degree of belief is calculated for each model. This degree of belief is called the Deviance Information Criterion (DIC) and is penalized for having additional parameters (McCarthy & Masters 2005). The DIC is the Bayesian analogue of Akaike's Information Criterion (AIC) and allows comparison of each hypothesis and selection of the most supported hypothesis (lowest DIC value). A difference in the DIC of < 2 indicates the two models are indistinguishable, a difference of 4–7 indicates the poorer model has considerably less support, and a difference of 10 or more indicates the poorer model has virtually no support (Burnham & Anderson 2002). Bayesian analyses are also useful as they allow existing data (if any) to be incorporated into the model, are able to handle missing data, can deal with complex ecological problems such as population dynamics and species distribution, and allow trends to be seen before results become significant which allows earlier conservation action to take place (Williams et al. 2005; Stephens et al. 2007).

For the site area and shape analysis, three logistic regressions were fitted to the data modelling the effects of area and shape alone, and then together. Both parameters were continuous. The Deviance Information Criterion (DIC) was used to compare models. For the plant traits analysis, height was the only continuous parameter, all other parameters were categories. All parameters (area, shape, life form, height, seed dispersal and vegetative spread) were assigned uninformative prior distributions. Parameter estimates for all models were based on 100 000 sampling iterations after a ‘burn in’ of 10 000 iterations. To ensure the accuracy of the MCMC random sampling from the target posterior distribution, the ‘burn in’ was discarded and the regression coefficients for each sample were visually inspected to ensure the absence of any trends and low autocorrelation. The mean, standard deviation and 95% Bayesian credible interval (2.5th and 97.5th percentiles) for each parameter were calculated. The effect of each variable was assessed using the mean and standard deviation of the posterior distribution of the parameters. For each parameter, the posterior distribution was plotted against the probability of local extinction.


overall changes

When the grassy woodland landscape was first surveyed for its floristic composition in 1975, the 12 remnants contained 177 native species comprising 755 populations. By 2006, the vegetation remnants were still rich in native plant species, but there had been a substantial loss of local populations; 199 populations (25.6%) of 98 species are considered locally extinct. In total, 27 native species no longer occurred in any of the sites sampled, although most of these species (67%) occurred at only one site in 1975. The average percentage of native plant species per site to become locally extinct was 25.9 ± 8.8% (mean ± 1 SD).

initial abundance and local extinction

Local extinction by 2006 was significantly associated with initial abundance in 1975 (χ2 = 115.35, d.f. = 12, P < 0.0001). Thirty-four percent of populations in the ‘Very Rare’ category in 1975 were extinct by 2006. Twenty-nine percent of populations in the ‘Rare’ category were extinct in 2006, whereas 25% of ‘Common’ and 15% of ‘Very Common’ populations were not recorded in 2006. Forty-four percent of all populations were considered as, or more, abundant than when first surveyed (Fig. 1).

Figure 1.

Changes in population abundances of native plants species from 1975 to 2006 in grassy woodlands in western Victoria. Very Rare ≤ 24 individuals; Rare = 24–100; Common = in the 100 s; Very Common = in the 1000 s.

Populations that were ‘Very Rare’ in 1975 tended to stay ‘Very Rare’ or became ‘Extinct’ by 2006 (Fig. 1), whereas disproportionately few became ‘Common’ or ‘Very Common’. Populations that were ‘Rare’ or ‘Common’ in 1975 increased or decreased into any abundance category, but disproportionately few ‘Rare’ populations became ‘Very Common’. Finally, populations that were ‘Very Common’ in 1975 tended to remain ‘Very Common’ or ‘Common’, and fewer than statistically expected became ‘Very Rare’ or ‘Extinct’.

extrinsic effects on local extinction

Of the site spatial characteristics, shape was the model with the best predictive power to explain local extinction (b = −0.4195, SD = 0.2639, DIC = 80.25). However, the DIC values of the remaining two models are very close to that of the best model (i.e. 1.94 difference for area and shape together, and 2.06 difference for area alone). This indicates the model using area and shape together (area: b = 0.0033, SD = 0.0073; shape: b = −0.4025, SD = 0.2681, DIC = 82.07) is indistinguishable from that of shape alone, and the model using area alone is only slightly less supported. Therefore, the model containing both parameters was used.

Increasing remnant area had a minor, but slightly positive influence on the probability of local extinction of native plant populations (Fig. 2). However, the 95% Bayesian credible interval illustrates large uncertainty around this prediction for larger sites. This suggests local extinction probability could increase or decrease with increased area. Species with populations in long and thin remnants (i.e. high perimeter to area ratio) had a decreased probability of extinction (Fig. 2) compared to sites with a lower perimeter to area ratio. However, again the 95% Bayesian credible interval illustrates large uncertainty around this prediction.

Figure 2.

The effect of (a) area and (b) shape (perimeter to area ratio) on probability of local extinction of native plant populations in grassy woodlands in western Victoria. For perimeter to area ratio, sites with less edge (e.g. squares) have a lower ratio value, and sites with more edge (e.g. long, thin roadsides) have a larger ratio value. The mean (solid line) and 95% Bayesian credible interval (dashed lines) are shown. Each model was plotted while holding the other parameter at its mean.

growth-form and plant functional trait effects on local extinction

The effect of each trait on the probability of local extinction is measured by the size of the mean of its regression coefficient (b) (Table 1). This influences how much the probability of local extinction changes across the range of the data. The standard deviation (SD) expresses the uncertainty around the estimated effect of the posterior distribution (Table 1). If the magnitude of the mean is more than twice the standard deviation of the regression coefficient, the effect of the variable is positive or negative (with 97.5% confidence). This is because the regression coefficients were approximately normally distributed. If the 95% Bayesian credible interval (Table 1) for the parameter crosses zero, the parameter does not have a substantial effect on the extinction probability of native plant species. Alternatively, if the 95% Bayesian credible intervals are all positive or all negative, the parameter substantially increases or decreases the probability of local extinction, respectively. With these considerations in mind, the effect of each variable is described below.

Table 1.  Mean (b), standard deviation (SD) and 95% Bayesian credible intervals of the posterior distribution for each parameter. Distribution is based on the regression model for local extinction probability of native plant species in fragmented grassy ecosystems. Phanerophyte and Ant dispersal were used as reference classes for the categorical variables life form and dispersal, respectively
Life form
 Within Hemicryptophyte
  Hemi Flat−1.2450.824−2.886−1.2350.345
  Hemi Erect−1.5190.691−2.898−1.508−0.196
  Hemi Partial−2.8710.774−4.426−2.858−1.402
  Hemi Proto−2.5180.741−4.001−2.504−1.101
Dispersal mode
Vegetative spread−0.2770.210−0.688−0.2780.136

Chamaephytes, geophytes and hemicryptophytes with either all leaves cauline (hemi-proto), all leaves basal and erect (hemi-erect) or basal and cauline leaves (hemi-partial) had a substantially lower probability of local extinction than species with other growth forms. For all other life forms, however, their roles were not substantial as there were large amounts of variation for the probability of local extinction (Table 1; Fig. 3a). Species with wind- or mobile-dispersed seeds had a substantially greater probability of going locally extinct. The remaining seed dispersal modes did not have a substantial effect on the probability of local extinction. This is because large amounts of variation surrounded their predictions, which was especially true for species with ingested seeds (b and SD; Table 1; Fig. 3b). The height of a species greatly influenced the probability of local extinction; however, its effects were highly variable for species approximately < 5 m tall. Tall species had a very low probability of local extinction and short species had a much higher probability of local extinction (Fig. 4). The ability of a species to spread vegetatively did not substantially influence the probability of local extinction (Table 1).

Figure 3.

The influence of (a) life form and (b) seed dispersal mode on the predicted probability of local extinction of native plant species in grassy woodlands in western Victoria. The mean predicted probability of extinction (ball) and the 95% Bayesian credible intervals (bars) are shown. The credible intervals encompass all uncertainty within the model; refer to Table 1 for the uncertainties associated with each parameter.

Figure 4.

The influence of height on the predicted probability of local extinction of native plant species in grassy woodlands in western Victoria. The mean predicted probability of extinction (solid line) and the 95% Bayesian credible intervals (dashed lines) are shown. The credible intervals encompass all uncertainty within the model; refer to Table 1 for the uncertainties associated with each parameter.


In the face of increasing human impacts, a major goal for conservation biology is to provide principles by which biological diversity can be preserved. The revisitation methodology employed in this study enabled us to observe three major outcomes of the effects of fragmentation over the last three decades. First, despite >50 years since agricultural intensification of the district in the 1950s, which likely precipitated the first wave of local population extinction, native species populations continue to be lost from remnant woodlands. Second, extrinsic factors (remnant area and remnant shape) provided low predictive power for understanding these ongoing effects, whereas some intrinsic factors (e.g. plant height, seed dispersal) had greater explanatory value. Hence, fragmentation effects have both a spatial and biological component. Finally, the population abundance of species in 1975 was a good predictor of local extinction by 2006, indicating that small populations are more likely to disappear over time. Hence, species at low abundance are most prone to local extirpation due to demographic and environmental stochasticity.

extrinsic factors

Site area and shape did not substantially explain the probability of local population extirpation of native plant species in this fragmented landscape, contrary to theoretical expectations (Fahrig 2003; Ewers & Didham 2006; Lindenmayer & Fischer 2006). Typically, species in the smallest remnants with the highest edge to area ratio are considered to be at most risk of extinction. However, Bayesian logistic regression suggests that these factors have rather weak explanatory value (and wide variability) in grassy woodlands. Perhaps this is because all the sites are small (<40 ha in area) and mostly linear, hence obscuring the true effects of these factors. Additionally, site isolation, rather than size may be a more important determinant of species persistence (Thomas et al. 2001). However, isolation distances between remnants was not quantified in this study as all sites occur in a matrix of agriculture with scattered paddock trees, making estimates of true isolation difficult.

Alternatively, the lack of distinct effects of remnant size and shape on persistence of native species may be because each of the sites are under different pressures in addition to fragmentation, that act in a non-uniform way to affect habitat quality (Thomas et al. 2001; Ewers & Didham 2006; Brook et al. 2008). These pressures include inter-specific competition, decreased endogenous disturbances, increased exogenous disturbances, edge effects and invasions of both plant competitors and animal herbivores (McIntyre et al. 1995; Davies & Margules 1998; Hobbs & Yates 2003; Lunt & Spooner 2005; Williams et al. 2005). Habitat deterioration can affect the population viability and persistence of plants (Brys et al. 2005) and, in most cases, degradation of habitat quality occurs due to the impact of the surrounding landscape matrix on habitat patches.

Habitat deterioration is likely to have played a role in the loss of local populations in this study. Although detailed site management histories were not available for any of the sites, information given by local residents is illuminating. For example, the site with the lowest proportional loss of species (Lawson's Lane; 13%) was reported to have been annually burnt until approximately 8 years before the 2006 surveys were conducted (unnamed local farmer, pers. comm., Nov 2006). In contrast, the site with the largest proportion of species lost (Balmoral; 44%) used to be burnt regularly (approximately annually) when the railway line was in use (Agnes Coxon, pers. comm., Jan 2007). However, as the railway line became unused three to four decades ago, burning also ceased at that time (Agnes Coxon, pers. comm., Jan 2007). This has resulted in many species being lost from the site, as has been found by Williams et al. (2006) for formerly burnt but now long-unburned grasslands. Many ‘new’ species to this site since 1975, including native shrubs (e.g. Astroloma conostephioides, Grevillea aquifolium) are now encroaching, which also suggests a more infrequent burning regime (Roques et al. 2001; Heisler et al. 2003). These changes are unlikely to represent natural succession because most new shrubs observed are not considered a component of the native vegetation (using the descriptions of Blackburn & Gibbons (1956), Gibbons & Downes (1964) and Willis (1971) as a guide to the pre-fragmented woodland state), but rather have encroached from nearby windbreaks and apparently via seed dispersal by roadside maintenance machinery. Their establishment in the absence of fire, due to land use change, might therefore be considered opportunistic. Hence, within-site habitat management could play an under-appreciated role in maintaining native species in this fragmented landscape, regardless of the size or spatial context of remnants. This is because of its direct impacts on habitat quality and, presumably, habitat heterogeneity. Despite the apparent importance of maintaining habitat quality for species persistence in fragmented landscapes, it remains poorly recognized relative to the fragmentation paradigm that focuses on the spatial patterns of habitat vs. non-habitat (Harrison & Bruna 1999; Thomas et al. 2001; Radford & Bennett 2007).

intrinsic factors

No life form appeared to predispose plant species to local population extinction due to habitat fragmentation in this grassy woodland landscape. Chamaephytes, geophytes and hemicryptophytes, whether they were clonal or not, all appeared equally capable of persistence in the small remnants. However, shorter species, in general, had a much higher probability of local extirpation than taller species. Taller species generally have a longer life span than shorter species (Duncan & Young 2000) and hence do not need to successfully reproduce as frequently. In addition, taller species are generally more competitive (Duncan & Young 2000; Kolb & Diekmann 2005). It is likely, however, that the lower probability of local extinction for taller species (i.e. trees) may simply reflect a time-lag in response to fragmentation (i.e. the extinction debt sensu Tilman et al. 1994), where long-lived adult individuals persist but conditions are no longer suitable for their recruitment.

Species with wind- and mobile-dispersed seeds had a higher probability of local extirpation, a finding at odds with much fragmentation literature. The ability to persist in a fragmented landscape has been considered to, in part, depend on the dispersal abilities of species (e.g. Hoehn et al. 2007). Those species that disperse readily and frequently will more likely persist (as populations rather than individuals) in habitat remnants though time than species that do not. For plants, however, dispersal by wind may ‘waste’ many propagules as they are more likely to be lost in the relatively unfavourable matrix of agricultural lands (Holt et al. 1995; Roy & de Blois 2006). Alternatively, the very small seeds of wind-dispersed species may render them poor competitors in the recruitment phase of their life cycle (Ehrlen & van Groenendael 1998; Soons & Heil 2002), making them susceptible to changes in the landscape caused by non-native plant invasions and changed disturbance regimes. Hence, it would appear that population persistence may not necessarily be enhanced by seed dispersal between remnants but rather, by other plant traits such as an ability to maintain site occupancy by longevity or a persistent soil seed bank, or site factors themselves. From a conservation perspective, differences in the colonization capacity among species imply that maintenance of plant diversity must not only focus on conditions within patches, but also consider the spatial arrangement of patches in order to enable plants to bridge gaps in time and space (Ehrlen & van Groenendael 1998).

By contrast, the probabilities of species with animal-dispersed seeds to go locally extinct were lower. This may be because sheep, cattle and other introduced animals (e.g. rabbits, cats, foxes) act as vectors for dispersal of some species (Hobbs & Yates 2003) or because some fauna may be able to cross the rural matrix, increasing the biotic dispersal between remnants.

Despite its theoretical significance, very few studies compare the past abundance of plant populations to their present abundance and how this may affect the probability of local extinction (but see Robinson & Quinn 1988; Fischer & Stocklin 1997; Davies & Margules 1998; Davies et al. 2000; Matthies et al. 2004; van Calster et al. 2008). Those that have used this approach tend to focus on rare species, neglecting the more abundant populations and species. Yet, it is implicit in MacArthur & Wilson's (1967) theory of island biogeography that extinction rates due to fluctuations in populations should decline with increasing population size, regardless of their ecological status. In this study, the past abundance (1975) of a species was a good predictor of local extinction risk for small populations when populations were revisited 30 years later. Small populations tended to have an increased risk of local extirpation in comparison to populations classed as ‘Very Common’ in 1975. Disappearance events in grassland annuals have previously been observed to be determined by population size in the previous year; small-sized populations were more prone to be missing (Dostal 2005), as have small populations elsewhere (Fischer & Stöcklin 1997; Duncan & Young 2000; Matthies et al. 2004; Walker & Preston 2006; Wiegmann & Waller 2006; van Calster et al. 2008). This is likely to be a result of their increased vulnerability to localized catastrophic events and genetic erosion caused by a reduced gene pool and inbreeding (Fischer & Stöcklin 1997; Matthies et al. 2004). In addition, their ability to recruit may be inhibited due to strong competition for space or perhaps seed limitation due to lack of seed production (Hobbs & Yates 2003). Interestingly, some populations classed as ‘Very Rare’ in 1975 became ‘Common’ by 2006, hinting that populations can be very dynamic in remnant woodlands and this may be influenced by temporal rescue effects (Piessens et al. 2004), climatic variability (Levine et al. 2008) or relaxation of previous fire regimes (Ross et al. 2002; Briggs et al. 2005).

Populations with a large past abundance varied in their response to fragmentation. Most initially large populations remained abundant, hence supporting the few studies that show common species have a decreased risk of extinction (e.g. Fischer & Stöcklin 1997; Duncan & Young 2000; Matthies et al. 2004; Walker & Preston 2006; Wiegmann & Waller 2006; van Calster et al. 2008). Importantly, not all large populations in 1975 were immune from local extinction by 2006; this has rarely been reported in the literature over such short timescales (Hooftman & Diemer 2002). Hence, this suggests that species can be at risk in fragmented landscapes for reasons other than having small populations.

Three potential explanations might be invoked for the observed declines in formerly common populations. First, woodlands in southern Australia are exposed to annual summer drought, the severity of which is linked to El Niño Southern Oscillation events. Severe drought can cause high species-specific mortality in grasslands (Williams 1968; Austin et al. 1981) and woodlands (Pook et al. 1966; Ashton & Spalding 2001) and hence, infrequent but catastrophic declines of common or dominant species due to environmental stochasticity are probably ‘normal’ in such ecosystems at decadal to centennial timescales (Nicholls 1991; Lande 1993; Brook et al. 2008).

Second, it is very likely that woodland habitat quality changed in different or non-uniform ways across the 12 sites over the 31 years. It is possible that what was initially good habitat that supported large populations of plants has become poorer habitat in some, but not all cases. Although there is empirical evidence that habitat modification and degradation negatively affect the biota of fragmented landscapes (Harrison & Bruna 1999), there is no theoretical framework for evaluating the effects of habitat modification and how this interacts with population dynamics (Davies et al. 2000). Clearly, it is important to determine what drives habitat quality. This will likely be edge effects, disturbance and invasions by non-native species.

Finally, another potentially important explanation for rapid and profound population loss and decline is considered to involve recent changes in disturbance regimes. Competition from dominant native tussock grasses and exotic species is known to underpin patterns of diversity in grassy ecosystems in temperate Australia (e.g. Stuwe & Parsons 1977) and, where there is a lack of disturbance by grazing or fire (Stuwe & Parsons 1977; Williams et al. 2006), these processes are likely to be intense. In the absence of fire, habitat heterogeneity can also decline, leading to reductions in species richness of plant communities (Collins 1992; Ross et al. 2002). Similarly, Leach and Givnish (1996) postulated that the changes in fire regimes that accompany habitat subdivision of tallgrass prairie were responsible for substantial losses of plant species. This suggests that fragmentation might often promote plant population loss in flammable ecosystems that are no longer exposed to fire, sentiments echoed more recently by Briggs et al. (2005).


We believe that because several factors other than fragmentation are impacting on the survival of native plant populations in this fragmented landscape (e.g. edge effects, changed and introduced disturbance regimes, interspecific competition, non-native plant invasions, drought), no single site spatial characteristic or plant trait showed to be a good predictor of local extinction. This indicates that in this agricultural landscape where remnant vegetation is restricted to roadsides and small reserves, fragmentation is not the only dominant driver of extinction. Management history may play an important role in local extinction because it can affect habitat quality. As a result, biodiversity will slowly and inconspicuously continue to be lost unless appropriate management is implemented on the remaining native vegetation fragments. Unfortunately, studies have shown that once native species richness is lost from a site and/or shrubs have encroached due to long fire-free intervals, increasing the frequency of fires does not restore the vegetation back to its original, high-quality state. This highlights the importance of implementing appropriate management practices and burning regimes to minimise the further loss of species from grassy ecosystems.


Field assistance was provided by Jenny Kane, Aggie Stevenson, Agnes Coxon, Hilary Turner, Lauren Brown, Kirsten Brunt, Carla Wilson, Vanessa Carnegie, Lauren McLees and Kelly Mether. David Cameron updated the original Beauglehole lists to current nomenclature and checked all our difficult specimens and we thank him for his enthusiasm and ongoing support. Neville Walsh and Graeme Lorimer provided additional taxonomic support. Nick Williams and Michael Smith provided help with WinBugs code. Bob Parsons, Michele Kohout, Scott Wilson and two anonymous referees improved early versions of the article.