Plant functional types can predict decade-scale changes in fire-prone vegetation

Authors


*Author to whom correspondence should be addressed: David Keith. E-mail: david.keith@environment.nsw.gov.au.

Summary

  • 1Plant functional types (PFTs) are groups of species sharing traits that govern their mechanisms of response to environmental perturbations such as recurring fires, inundation, grazing, biological invasions and global climate change. The key components of a PFT approach are an underlying model of vegetation dynamics for a given system and a classification of functional types based on traits deduced from key processes in the model.
  • 2Prediction and generalization underpin the potential utility of the PFT approach for understanding ecosystem behaviour. For PFTs to be useful in ecosystem management, they (in concert with their underlying model) must reliably predict vegetation change under given environmental scenarios and they must produce robust generalizations across the species that are classified and the range of environments in which they occur.
  • 3The efficacy of plant functional types has been explored using various approaches in a wide range of ecosystems. However, very few studies have tested the accuracy and generality of PFT predictions against vegetation changes observed empirically over medium to long time scales.
  • 4We applied this approach to examine the predictive accuracy and generality of a PFT classification and an associated model of vegetation dynamics for a fire-prone, species-rich wet heathland in south-eastern Australia. We assigned each species to one of six PFTs derived using a deductive approach based on the vital attributes scheme. We measured their initial abundance at a set of sample sites distributed across local environmental gradients. We used the PFT traits and processes in the underlying model to predict qualitative changes in abundance in response to a fire regime scenario observed at the sample sites during a subsequent period of 21 years. We then re-surveyed the sample sites to compare predictions with observed changes in abundance.
  • 5The PFTs and their underlying model produced an accurate prediction of average vegetation responses over the 21-year period. The majority of species within each PFT exhibited the predicted response and few species had strongly opposing responses in different environments. However, not all species within a PFT underwent the predicted direction of change, and responses of individual species were not uniform across the environmental gradients.
  • 6Synthesis. We conclude that plant functional types based on vital attributes are very useful tools for prediction and generalization in ecosystem management, although interpretations need to be tempered by the fact that PFTs may not accurately predict responses of all species across all environments.

Introduction

Plant functional types (PFTs), groupings of plant species based on functional traits, have become widely used for understanding ecosystem change in response to environmental perturbation. The approach has been applied to predict ecosystem responses to recurring fires (Noble & Slatyer 1980; Pausas 1999), grazing (McIntyre et al. 1995; Díaz et al. 2007), inundation (van der Valk 1981; Shipley et al. 1989; Boutin & Keddy 1993), biological invasions (Noble 1989; Richardson et al. 1990) and global climate change (Skarpe 1996; Leemans 1997; Bond et al. 2005). While the PFT approach is sufficiently general to accommodate such a diverse range of applications, it is widely agreed that the approach must be applied in context-specific ways if it is to serve a particular purpose effectively (Woodward & Cramer 1996).

The development of theory for PFTs has focused on species classification for prediction and generalization (Noble & Slatyer 1980; Keddy 1992; Lavorel & Garnier 2002). Here, ‘prediction’ refers to the ability to determine species’ responses (within plausible bounds) to a scenario of environmental perturbation, given an underlying model of system dynamics. ‘Generalization’ refers to the ability to determine the responses of large numbers of species, given knowledge of functional traits that they share.

By defining a functional type as a group of species whose response to a particular perturbation is mediated by the same mechanism, Gitay & Noble (1997) recognize that ecological mechanisms underpin the predictive properties of PFTs. It is therefore axiomatic of the PFT approach that: (i) members of the same functional group share common features (traits) that govern their dynamics under a range of environmental scenarios, so that their responses to a given scenario are similar; (ii) different PFTs will be characterized by different combinations of features, which may result in different outcomes of change under a given scenario, or at least different mechanisms by which species reach similar outcomes; and (iii) PFTs and the species that they contain behave consistently across environmental gradients.

There has been a great deal of research on how functional classifications should be constructed to maximize the robustness of their predictive properties (Woodward & Cramer 1996; Díaz & Cabido 1997; Smith et al. 1997; Rusch et al. 2003). Functional classifications may be derived using either inductive or deductive methods. Inductive methods seek groupings of species by analysing an observed set of ‘relevant’ traits that were selected with a minimal level of subjectivity (e.g. Leishman & Westoby 1992). Deductive methods derive groupings of species directly from a set of explicit assumptions or models that describe how a system operates (e.g. Noble & Gitay 1996).

Empirical tests of how well functional classifications support prediction and generalization have taken a range of approaches (Gitay & Noble 1997). Many studies have examined the usefulness of functional classifications by characterizing the functional types that have persisted or become abundant at sites with known histories of environmental perturbation (e.g. Keith & Bradstock 1994; McIntyre & Lavorel 2001; Pausas et al. 2004). Other studies have tested the comparative performance (e.g. growth rates) of plants from different functional types exposed to experimentally manipulated conditions in field plots or the laboratory (e.g. Chapin et al. 1995; Grime et al. 1997). A third approach has been to explore changes in the occurrence of functional types under contrasting environmental scenarios using computer simulation (e.g. Burgmann 1996; Bradstock et al. 1998; Bond et al. 2005). However, there have been very few longitudinal studies that track the fate of PFTs within wild ecosystems through real environmental scenarios to determine whether predicted outcomes actually eventuate. We carried out such a study over two decades to determine the predictive value and generality of a PFT classification for understanding vegetation change.

Our study system was a fire-prone, species-rich, edaphically variable heathy wetland in south-eastern Australia. We used a deductive process to derive a minimum set of PFTs based on explicit assumptions about the key processes governing dynamics of the system. Building on previous work (Keith & Bradstock 1994; Tozer & Bradstock 2002), we developed a model of vegetation dynamics capable of predicting changes in the abundance of plant functional types in response to any given fire regime. We recorded the abundance of vascular plants at a set of sample sites, classified the sites according to their position on an edaphic gradient (Keith & Myerscough 1993) and documented the fires that burnt each site during the ensuing two decades. We applied the model to predict whether the plants within each functional type would increase, decline or remain stable in abundance over the observation period. Finally, we re-measured the abundance of each species at each site two decades after the original census and calculated the observed change in abundance to assess: (i) whether each PFT showed the expected direction of change predicted by the model for the observed fire regime scenario; (ii) whether species within each PFT showed consistent patterns (i.e. same direction) of change; and (iii) the variability of species responses across the edaphic gradient.

Methods

study area

The study was carried out within Dharawal Nature Reserve and State Conservation Area in the O’Hares Creek catchment (centred on 34°14′ S, 150°52′ E), 45 km south-west of Sydney's central business district (Fig. 1). The area is an inclined sandstone plateau, dipping from 450 m above sea level in the south-east to below 200 m elevation in the north-west. Average annual precipitation varies from 1550 mm in the east to 850 mm in the north-west. Within this area, treeless, species-rich, sclerophyllous wet heathlands and sedgelands, known locally as upland swamps (Keith & Myerscough 1993), occur in headwater valleys within a matrix of eucalypt woodlands (Keith 1994). There is considerable variation in the edaphic properties within upland swamps that is related to a hydrological gradient. With increasing distance from drainage zones, average water table depth declines, as does soil organic matter content, cation exchange capacity and conductivity of the soil solution (Keith & Myerscough 1993). Turnover in plant species composition and variation in vegetation structural properties are correlated with this edaphic gradient.

Figure 1.

Location of the study area and its 54 permanent transect monitoring sites. Darkes Forest is approximately 6 km west from the coast and 42 km south-south-west of Sydney central business district.

Fire history of the area, from the mid 1960s to the present, was compiled from records and maps held by Sydney Catchment Authority and the National Parks and Wildlife Service, including wildfires and prescribed burns. Since 1982, fire occurrences have been verified by personal observations (D.A.K.). During its recorded fire history, the area was burnt by three major wildfires in early 1968, November 1990 and January 2002. The 1968 and 2002 fires burnt the entire area, whereas the 1990 fire burnt large parts of the area in the north and west. There was also a record of an extensive fire in 1965, although its boundaries could not be verified. During the early 1970s and the mid 1980s several hazard reduction fires were lit between May and August. The majority of these burnt small areas. The largest hazard reduction fire was in August 1984, after the first field census. There was no overlap between this fire and the subsequent wildfire in 1990. The prevailing fire intervals overlapping the period between the two field surveys were thus 1968–84, 1968–90, 1984–2002 and 1990–2002, varying from 12 to 22 years’ duration, while the penultimate interval was either 12 or 18 years’ duration.

vegetation sampling

Sampling of vegetation was stratified along the soil moisture gradient and across the range of structural variability in the vegetation. Thus three categories of soil moisture and three categories of vegetation structure (based on the height and form of shrubs) were sampled in an orthogonal design at 60 sites in upland swamps scattered throughout the study area (Fig. 1, Keith & Myerscough 1993).

Each sample site was a belt transect of 60 contiguous 0.5 × 0.5 m quadrats, in which presence/absence of all vascular plant taxa was recorded (based on whether they were rooted within quadrats) and tallied to give a frequency out of 60. The sites were originally sampled in 1983, marked in the field by two wooden stakes at either end of the transects, and plotted on a 1 : 10 000 aerial photograph. By 2004, it was possible to relocate 54 of the 60 original transects and of these, at least one of the original markers was found for 20 transects. During the intervening years, some of the wooden markers had been consumed by fires, but based on the annotated aerial photograph, field notes and detailed recollections of the original observer (D.A.K.), transects were confidently re-established within approximately 10 m of their original location. These formed the basis for establishing permanent swamp monitoring sites, which are now marked in the field with stainless steel wire stakes, c. 1 m tall, and located on the Map Grid of Australia (Geodectic Datum of Australia) using a global positioning system.

Exploration and subsampling of the 1983 data showed that the floristic relationships between the 60 sites could be adequately retrieved if species frequencies were calculated from a sample of 30 quadrats within each transect (correlation of association matrices based on 60 and 30 quadrats yielded Mantel's R > 0.95). Therefore, only the first 30 quadrats of each transect were re-sampled in 2004.

At the time of the first survey in 1983, all but eight of the monitoring sites had a post-fire age of 15 years (i.e. last burnt in the 1968 fire), with the remainder having post-fire ages of 2–8 years. All monitoring sites had a post-fire age of 2–3 years at the time of the repeat survey, as the entire area was burnt in 2002.

To characterize the position of the 54 samples on local environmental gradients, each was assigned to one of the five groupings defined by Keith & Myerscough (1993). These represent plant communities with different edaphic and structural characteristics. Ti-tree Thicket (TT) occurs in the most waterlogged soils, Cyperoid Heath (CH) occurs in intermediate soils, while Sedgeland (SL), Restioid Heath (RH) and Banksia Thicket (BT) occur in the least waterlogged soils. These latter three communities exhibit unique structural characteristics (SL short stature and open cover, RH intermediate and BT tall stature and dense cover) and subtle differences in soil chemistry (Keith & Myerscough 1993).

derivation of plant functional types

We used a deductive approach based on the vital attributes scheme of Noble & Slatyer (1980) to derive functional traits for plant species in the study area. The salient processes that govern the response of plant populations to fire regimes are: the method of in situ persistence through fire (resprouting or seed banks) or means of dispersal from ex situ sources after fire; the effects of competitors in unburnt conditions; and the timing of crucial life-history stages (maturation and senescence).

Only a small number of uncommon species in the area had propagule dispersal modes (anemochory, exochory) that were likely to result in frequent movement of seeds or fruits over appreciable distances. This is typical of Australian heathland floras (Keith et al. 2002a). For the purpose of our study, we therefore assumed dispersal ability to be invariant among species (i.e. all species had localized propagule dispersal).

Resprouting from fire-resistant organs (lignotubers, non-woody tubers, rhizomes, etc.) is an important means of in situ persistence exhibited by a large proportion of plant species in the study area. We therefore classified species as resprouters if at least some (notionally > 10% of mature plants) resprouted after 100% canopy scorch. Our definition recognizes that resprouting ability varies within populations in response to plant developmental stage, fire intensity and other factors (Gill 1981).

Recruitment from seed banks is also an important means of in situ persistence within the study area. Following Keith & Bradstock (1994) and Keith et al. (2002a), we recognized three types of seed banks that vary in their response to disturbance and longevity. Canopy seed banks are retained in fire-protected woody fruits, which are totally exhausted after fire and do not persist after senescence of standing plants in the population (Lamont et al. 1991). Persistent soil seed banks are not retained in the plant canopy, as seeds are released when fruits are mature. These seeds typically remain dormant in the soil until stimulated to germinate by fire, but a portion of seeds may remain dormant until a future fire, and viable seeds may remain in the soil after senescence of standing plants in the population (Bell et al. 1993; Auld et al. 2000). Transient soil seed banks are also released into the soil when fruits are mature, but do not remain dormant, i.e. all seeds either germinate or die within 1 year of release (Lunt 1997; Morgan 1998). Consequently, their seeds do not persist after senescence of standing plants in the population.

To consider the effects of competitors we distinguished the immediate post-fire period (1–2 years) from the subsequent period, when resources are sequestered by established vegetation (Noble & Slatyer 1980). We considered competitive effects on two different life-cycle phases: the establishment of new individuals (recruitment); and the survival and reproduction of established individuals. For recruitment, species may be grouped according to three possible outcomes (Noble & Slatyer 1980): intolerance (I) of conditions in the established community, with recruitment restricted to the post-fire period; tolerance (T) of conditions in the established community, with recruitment occurring any time in the fire cycle; and requiring (R) of conditions in the established community, with recruitment occurring only after the post-fire period. In upland swamps, seedling emergence is abundant in the first post-fire year and declines to negligible levels after 2 years (Keith et al. 2002a). Seedlings that do emerge after the post-fire period are unlikely to grow to maturity due to slow growth rates. Vegetative recruitment in clonal species shows a similar pattern to seedling recruitment, but vegetative recruitment after the post-fire period may not decline to negligible levels if resources are transferred within genets from established ramets to new ramets. In our study, we therefore recognized two types of recruitment responses to competitors: tolerant (limited vegetative recruitment); and intolerant (negligible recruitment).

We adopted the model proposed by Moore & Noble (1990) to assess the effects of competitors on established plants. This assumes that the outcomes of competitive interactions are determined by the relative vertical stature of individuals. We therefore assigned species to one of three vertical strata: upper (tall shrubs reaching maximum heights of c. 3–5 m); mid (shrubs reaching maximum heights of c. 1–2 m); and lower (graminoids, herbs and ferns reaching maximum heights of c. 0.2–1 m).

Noble & Slatyer (1980) defined quantitative parameters representing the longevity of propagules and standing plants and for the timing of maturation. We adopted a simplified representation using categorical variables. Variation in propagule longevity is already accommodated in the classification of three seed bank types described above. We defined maturation as the time after fire required for at least some individuals in a population (notionally 10%) to produce viable seed, and recognized three categories: short (1–2 years); moderate (3–4 years); and long (≥ 5 years). Our definition does not distinguish primary and secondary juvenile periods (Gill 1981) because the source of seeds (new recruits or resprouting individuals) contributing to seed bank accumulation was unimportant for our purposes (Noble & Slatyer 1980). We recognized four categories of longevity for standing plants: < 5 years; 10–30 years; 30–50 years; and > 50 years.

If all character states defined above were combined in a factorial manner, there would be 432 functional groups of species (1 dispersal × 2 resprouting × 3 seed bank × 2 competitive effects on recruitment × 3 competitive strata × 3 maturation × 4 standing plant longevity). However, only 11 combinations of traits are represented by species that are relatively common within the study area. We further simplified the classification into six PFTs by combining closely related trait syndromes (Table 1). Specifically, woody resprouters with different types of seed bank were combined within PFT 3, while PFT 4 included short-lived non-resprouting herbs with persistent soil seed banks (Pate et al. 1985) and longer-lived resprouting herbs with transient soil seed banks (Coates et al. 2006). The resprouting members of PFT 4 mimicked the short-lived members, as they were generally reduced to dormant tubers or rhizomes after the post-fire period. PFT 5 included non-rhizomatous graminoids, herbs and ferns with either persistent or transient soil seed banks, while PFT 6 included rhizomatous graminoids, herbs and ferns with either type of seed bank. The six PFTs encompassed six of Noble & Slatyer's (1980) vital attribute (VA) functional groups (Table 1). Three of the PFTs corresponded directly with one VA group, while the others included two to three VA groups. The differences between classifications were partly due to lumping of the 11 factorial trait combinations, and partly due to the fact that competitive strata were only included in FATE (Moore & Noble 1990), a later embellishment of the VA model (Noble & Slatyer 1980).

Table 1.  Six plant functional types of the upland swamp vegetation, their characteristic traits, corresponding vital attributes group (Noble & Slatyer 1980) and predicted overall change in abundance under fire regimes during 1983–2004. *refers to senescence of above-ground portion of plants only
Plant functional typeDispersalResproutingSeed bankRecruitment in absence of fireCompetitive stratumTiming of maturationTiming of senescenceVital attributes group(s)Predicted overall change in abundance under observed fire regime
1. Serotinous obligate seeder shrubsLocalNon-resprouterCanopyNegligibleUpperLong (4–8 years)Moderate-long (30–50 years)CIIncrease
2. Non-serotinous obligate seeder shrubsLocalNon-resprouterPersistent soilNegligibleMidModerate (2–5 years)Moderate-long (10–30 years)SIIncrease/stable
3. Resprouter shrubsLocalResprouterCanopyNegligibleMidModerate (3–5 years)Long (> 50 years)VIDecline
3. Resprouter shrubsLocalResprouterPersistent soilNegligibleMidModerate (2–5 years)Long (> 50 years)ΣIDecline
3. Resprouter shrubsLocalResprouterTransient soilNegligibleMidShort-moderate (1–3 years)Long (> 50 years)UI, VIDecline
4. Fire ephemeral herbsLocalNon-resprouterPersistent soilNegligibleLowerShort (1 year)Short (< 5 years)SIIncrease
4. Fire ephemeral herbsLocalResprouterTransient soilNegligibleLowerShort (1 year)Short (< 5 years)*UIIncrease
5. Non-rhizomatous herbs & graminoidsLocalResprouterPersistent soilNegligibleLowerShort (1–2 years)Moderate-long (> 10 years)UI, ΣIDecline/stable
5. Non-rhizomatous herbs & graminoidsLocalResprouterTransient soilNegligibleLowerShort (1–2 years)Long (> 50 years)UI, VIDecline
6. Rhizomatous graminoids, herbs & fernsLocalResprouterPersistent soilLimited vegetativeLowerShort (1–2 years)Moderate-long (> 10 years)UTDecline/stable
6. Rhizomatous graminoids, herbs & fernsLocalResprouterTransient soilLimited vegetativeLowerShort (1–2 years)Long (> 50 years)UTDecline/stable

model predictions

The model of vegetation dynamics for the study system was described by Keith & Bradstock (1994), and is based on a variation of the vital attributes scheme proposed by Moore & Noble (1990). The model is capable of predicting qualitative changes in the abundance of PFTs under any fire regime. Here we outline predicted changes only for the fire regime observed in the study area (Fig. 2). The salient characteristics of this regime are that fire intervals were 12–22 years’ duration, that all fires resulted in crown scorch or consumption of vegetation, and that the post-fire age of vegetation was much younger in 2004 (all 54 sites c. 2 years old) than in 1983 (46 sites 15 years old, 8 sites 2–8 years old).

Figure 2.

(a) Timing of key life-history processes in relation to time since fire for six plant functional types (icons labelled 1–6 in left-hand column; for details of the six plant functional types see Table 1). Symbols represent the following processes: [, commencement of seedbank accumulation; Δ, height of PFT 1 exceeds height of other PFTs; +, senescence of standing plants; ], senescence of propagule bank. Multiple symbols for each process in each PFT indicate the range of variation between species and across environmental conditions. Dots indicate where the process may continue beyond 50 years post-fire. Unbroken vertical lines indicate length of fire intervals observed during the study period. Broken vertical lines indicate post-fire age at time of first (1983) and second census (2004), respectively. (b) Predicted change in relative abundance of PFTs under observed fire regimes showing community states at first census, just prior to most recent fire, and at second census, respectively. Other transitional states between first census and second census have been omitted for simplicity.

The serotinous obligate seeding shrubs (PFT 1, Fig. 2a) are key drivers of community dynamics, as they are the competitive dominants of the system (Keith & Bradstock 1994) and may be eliminated by short or long fire intervals, which exhaust their canopy seed banks (Lamont et al. 1991). By a post-fire age of 12–22 years, populations of these species have had up to 14 reproductive seasons and thus accumulated large seed banks (Fig. 2a). Consequently, fires of these intervals are predicted to increase the density of populations (Bradstock & O’Connell 1988), unless limited by density-dependence (Morris & Myerscough 1988). As only seven of 54 sites supported high densities of these species in 1983, the model predicted an average increase in their abundance between 1983 and 2004 (Table 1, Fig. 2b).

The serotinous obligate seeders begin to overtop other shrub species at a post-fire age of 5–6 years (Keith & Bradstock 1994; Fig. 2a). When their densities are high, the canopies of PFT 1 species curtail growth, seed production and survival of standing plants in subordinate strata during subsequent years. The model predicts that the expected increase in abundance of PFT 1 species will lead to a decline in abundance of understorey PFTs unless their populations are able to evade competition by rapidly accumulating a substantial persistent seed bank before canopy closure affects their standing plants (Fig. 2b). Non-serotinous obligate seeding shrubs (PFT 2) reach maturity at 2–5 years of age (Table 1, Fig. 2a), which is early enough to commence seed bank accumulation prior to canopy closure, allowing the species in this group to increase or maintain their abundance by evading competition later in the fire interval as a persistent soil seed bank. The model therefore predicted increased or stable mean abundance for PFT 2 (Table 1, Fig. 2b).

Resprouter shrubs (PFT 3) were predicted to decline under the observed fire regime (Table 1, Fig. 2b) because their slower growth rates, relatively low fecundity and dependence on standing plants for persistence through fire predisposes this group to competitive effects of the denser canopy of PFT 1 species.

Of all PFTs, the fire ephemeral herbs (PFT 4) were predicted to be least sensitive to competition from PFT 1 species because they essentially complete their standing-plant life phase and accumulate a persistent seed bank before canopy closure (Table 1, Fig. 2a). As a consequence of their short above-ground life phase, however, PFT 4 species were expected to show a strong response to time since fire (Pate et al. 1985). At a post-fire age of 2 years, standing plants would still be present in populations of PFT 4 species. By 15 years post-fire, however, these populations would be represented largely by dormant seeds, rhizomes or tubers due to early senescence of standing plants. Subsoil seed banks, tubers and rhizomes were not sampled in our survey. The model therefore predicts higher abundance of PFT 4 standing plants in 2004 than in 1983 (Table 1, Fig. 2b).

Unlike PFT 4, PFT 5 (non-rhizomatous herbs and graminoids) is dependent on persistent standing plants to maintain populations during intervals between fires. PFT 5 is therefore more sensitive than PFT 4 to competition from a dense canopy of PFT 1 species, although species with persistent, rather than transient, seed banks (Table 1) are expected to be more resilient to decline. This is because recruitment from dormant seed may at least partly compensate the loss of standing plants that succumb to competition. The model therefore predicts that abundance of PFT 5 species will decline or remain stable during 1983–2004 (Table 1, Fig. 2b).

Rhizomatous graminoids, herbs and ferns (PFT 6, Fig. 2a) also depend on persistent standing plants to maintain populations during intervals between fires, except that their propensity for lateral growth may allow them to exploit canopy gaps, thereby evading competition during the fire interval. PFT 6 was therefore predicted to decline or remain stable during 1983–2004 (Table 1, Fig. 2b).

data analyses

Species were selected for analysis to determine changes in abundance between 1983 and 2004 if they were recorded within at least 100 quadrats out of a possible total of 3240 across the two sampling times. Eighty species met this criterion, while a further 135 infrequently recorded species were excluded from analysis. For each of the 80 species selected for analysis, the change in abundance was calculated by subtracting its frequency recorded in 2004 from its frequency recorded in 1983. Each of these species was assigned to one of the six PFTs described in Table 1. The mean change in frequency and 95% credible intervals were estimated for each PFT by fitting a Bayesian hierarchical linear model using the WinBUGS package (Spiegelhalter et al. 2004). For each species, change in abundance was drawn from a Normal distribution for its PFT. Means of the PFT distributions were drawn from a Normal meta-distribution and standard deviations were drawn from vague uniform prior distributions. The mean and standard deviation of the meta-distribution were also drawn from vague priors. After burning in, the model was run over 100 000 iterations. Inspection of the residuals indicated a good fit to the data. The direction and magnitude of frequency change (based on the position of the 95% confidence intervals relative to zero) were examined to determine whether the abundance of each PFT increased, remained stable or declined during the period 1983–2004.

Bayesian linear models, based on normal error distributions, identity link functions and vague prior distributions, were also fitted to each of the 80 species to estimate their mean change in frequency (with 95% credible intervals) across all 54 sites to assess the variability of species responses within each functional group.

Finally, to assess variability in species responses across environmental gradients, linear models with normal error distributions and identity link functions, were fitted to each of the 80 species to estimate their mean change in frequency (with 95% credible intervals) across the five plant communities defined by Keith & Myerscough (1993).

Results

Temporal responses (1983–2004) varied between PFTs. Overall, three PFTs (1, 2 and 4) increased in abundance, while three groups (PFTs 3, 5 and 6) declined in abundance (Fig. 3). The confidence intervals on these changes indicated that the directions of these responses were unequivocal (more than 99.9% certain) for all groups except PFT 2, for which there was a 96.4% chance that abundance had increased. The directions of observed changes in mean abundance of all six PFTs were consistent with predictions of the underlying model (Fig. 3, cf. Table 1).

Figure 3.

Mean changes in abundance (number of quadrats occupied per transect) for six functional groups of plants: 1, serotinous obligate seeder shrubs; 2, non-serotinous obligate-seeder shrubs; 3, resprouter shrubs; 4, fire ephemeral herbs; 5, non-rhizomatous resprouting herbs and graminoids; and 6, rhizomatous resprouting graminoids, herbs and ferns. Error bars are 95% Bayesian credible intervals.

Table 2 reports the estimated mean changes in abundance (with 95% confidence intervals) for the species in each functional group overall, and within each of the five plant communities along the environmental gradients. Species were interpreted as having increased their abundance between 1983 and 2004 if the posterior Bayesian probability of increase P(increase) ≥ 0.9, as having declined if P(increase) ≤ 0.1, or as stable in abundance if 0.1 < P(increase) < 0.9 (Table 2).

Table 2.  Mean changes in abundance (with 95% credible intervals) during 1983–2004 for 80 plant species grouped by plant functional type (see Table 1). Estimates (left to right) are for overall change (mean across all samples), P(increase), the probability that the overall change was an increase in abundance, mean change in each of the five plant communities (with 95% credible intervals), the number of communities in which the mean change in abundance was positive and the number of communities in which the mean change in abundance was negative. Botanical nomenclature follows Harden (1990–2002) and http://plantnet.rbgsyd.nsw.com.au/
PFT. SpeciesOverall changeP (increase)Banksia Thicket (BT)Restioid Heath (RH)Sedgeland (SL)Cyperoid Heath (CH)Ti-tree Thicket (TT)Number of communities with increase in abundanceNumber of communities with decline in abundance
1 Serotinous obligate seeder shrubs
  Banksia ericifolia 9 (6 : 12.1)1.00 –1.9 (–7.9 : 4.1) 12.1 (7.5 : 16.7) 18.2(13.6 : 22.7) 8.6(4.5 : 12.7) 2.4(–3.2 : 8)41
  Hakea teretifolia 4.4 (1.3 : 7.4)1.00 –9.3 (–15.7 : –2.9) 10.8 (5.9 : 15.6)  5.3(0.4 : 10.1) 4.2(–0.2 : 8.5) 6.3(0.3 : 12.2)41
  Leptospermum squarrosum 0.7(–1.2 : 2.7)0.77  6(1.4 : 10.6)  1.8 (–1.8 : 5.3) –0.8(–4.3 : 2.7)–0.8(–3.9 : 2.3) 22
  Melaleuca squamea 0.6 (–0.1 : 1.3)0.95 –  –  –  2.3(1.2 : 3.4) 10
  Petrophile pulchella 2.2 (0.5 : 3.9)0.99  0.4 (–3.5 : 4.4)  4.9 (1.9 : 7.9)  3.5(0.5 : 6.5) 1.1(–1.6 : 3.8) 0.1(–3.6 : 3.8)50
2 Non-serotinous obligate seeder shrubs
  Almaleea paludosa 2.2 (0 : 4.3)0.98  2.6 (–2.7 : 7.8)  3.3 (–0.7 : 7.3)  2.3(–1.7 : 6.2) 1.8(–1.7 : 5.4) 0.9(–4 : 5.8)50
  Bauera microphylla 0.8 (–1.5 : 3.0)0.75 –2.6 (–8 : 2.7)  0.5 (–3.6 : 4.6)  2(–2 : 6.1) 1.8(–1.9 : 5.4) 0.4(–4.6 : 5.3)41
  Boronia parvifolia 1.1 (–0.4 : 2.7)0.92  0.9 (–2.8 : 4.5)  1 (–1.8 : 3.8) –1.4(–4.2 : 1.4) 4(1.5 : 6.5) 0.3(–3.1 : 3.6)41
  Cryptandra ericoides0.3(–1.3 : 0.7)0.28  2.3 (–0.1 : 4.7) –1.6 (–3.4 : 0.2) –0.9(–2.7 : 0.9)–0.2(–1.8 : 1.4) 13
  Dillwynia floribunda0.8 (–2.4 : 0.8)0.16 –8.5 (–11.8 : –5.2) –1.4 (–3.9 : 1.1) –0.3(–2.8 : 2.2) 2.3(0 : 4.5) 0.3(–2.8 : 3.3)23
  Epacris microphylla1.9(–4 : 0.2)0.04  2.6 (–2.2 : 7.3) –0.5 (–4.1 : 3.1) –3.4(–7 : 0.2)–4.9(–8.1 : –1.6)–0.2(–4.7 : 4.2)14
  Epacris obtusifolia 4.1 (1.3 : 6.9)1.00  2.6 (–3.5 : 8.8)  8.2 (3.5 : 12.8) –3.9(–8.6 : 0.8) 7.5(3.3 : 11.7) 5.5(–0.2 : 11.2)41
  Lepidosperma filiforme2.2 (–4.1 : –0.4)0.01 –  –0.2 (–3.3 : 2.9) –8.1(–11.2 : –5)–1.6(–4.4 : 1.2) 03
  Mirbelia rubiifolia 0.6 (–0.2 : 1.5)0.93  0.2 (–3.5 : 8.8)  2.6 (1.1 : 4.1)  0.2(–1.4 : 1.7)  0.1(–1.7 : 2)40
  Pultenaea aristata 0.4 (–1 : 1.9)0.73 –1.1 (–4.5 : 2.3)  2.8 (0.2 : 5.3)  0.9(–1.7 : 3.5)–0.8(–3.1 : 1.5) 0(–3.2 : 3.2)22
  Sphaerolobium vimineum 0.6 (–0.1 : 1.4)0.95  0.4 (–1.3 : 2.2)  0.8 (–0.6 : 2.1)  0.8(–0.5 : 2.2) 1.6(0.4 : 2.7)–1.4(–3 : 0.2)41
  Sprengelia incarnata 0.9 (–0.8 : 2.5)0.86  0.9 (–3.1 : 4.8)  1.1(–2 : 4.1)  1.1(–2 : 4.1) 1.1(–1.6 : 3.9) 40
  Symphionema paludosum0.9 (–2.2 : 0.5)0.10 –0.9 (–3.9 : 2.2) –3.8 (–6.1 : –1.4)  0.8(–1.5 : 3.1)–0.4(–2.4 : 1.7) 13
  Viminaria juncea 1.6 (–0.1 : 3.3)0.97  0.7 (–3.4 : 4.9)  0.7(–2.5 : 3.8)  1.8(–1.4 : 4.9) 3.3(0.5 : 6.2) 0.6(–3.3 : 4.4)50
3 Resprouter shrubs
  Baeckea imbricata2.4 (–4 : –0.8)0.00 –1.4 (–5.2 : 2.4) –2.3 (–5.2 : 0.7) –5.3(–8.2 : –2.4)–2(–4.6 : 0.6) 04
  Baeckea linifolia 1 (–0.2 : 2.3)0.94 –0.6 (–3.5 : 2.4)  1.9 (–0.4 : 4.2) –  0.7(–1.4 : 2.7) 3.4(0.6 : 6.2)31
  Banksia oblongifolia3.5 (–5.2 : –1.7)0.00 –6.7 (–9.7 : –3.7)–11.2 (–13.5 : –8.9) –0.7(–3 : 1.5)  03
  Banksia robur2.3 (–3.8 : –0.8)0.00  0 (–3.2 : 3.2) –  –0.2(–2.6 : 2.3)–5.3(–7.5 : –3.1)–5.5(–8.5 : –2.5)03
  Epacris paludosa 2 (0.7 : 3.3)1.00  1.7 (–1.5 : 4.9)  1.5 (–0.9 : 3.9)  2.4(0 : 4.8) 1.8(–0.3 : 4) 3.1(0.2 : 6.1)50
  Gahnia sieberiana0.7 (–1.8 : 0.4)0.11  1.6 (–0.4 : 3.6) –  –  0.7(–0.7 : 2)–7.4(–9.2 : –5.5)21
  Grevillea oleoides1 (–2 : 0.1)0.04 –4.2 (–6.6 : –1.8) –1.8 (–3.6 : 0.1)  0.1(–1.8 : 1.9)–0.2(–1.9 : 1.5) 13
  Hibbertia serpyllifolia0.8 (–2.2 : 0.7)0.14 –6.4 (–9.7 : –3.2) –0.5 (–3 : 2)  0.3(–2.1 : 2.8) 0.3(–1.9 : 2.5) 22
  Leptospermum grandifolium0.5(–2 : 0.9)0.23 –0.4 (–3.6 : 2.7)  3.1 (0.7 : 5.5) – –1.7(–3.9 : 0.4)–4.6(–7.6 : –1.7)13
  Leptospermum juniperinum0.7(–1.9 : 0.5)0.13  0 (–2.7 : 2.7)  0.5(–1.5 : 2.5)  1.3(–0.7 : 3.4)–1.4(–3.2 : 0.5)–5(–7.5 : –2.5)22
  Melaleuca squarrosa 0 (–1.3 : 1.3)0.52 –  –  – –1.4(–3.5 : 0.7) 2.9(0 : 5.7)11
  Xanthorrhoea resinifera5.6 (–7.5 : –3.7)0.00 –8.7 (–13.1 : –4.3) –6.8 (–10.1 : –3.4) –4.9(–8.3 : –1.6)–6.7(–9.7 : –3.7)–0.9(–5 : 3.2)05
4 Fire ephemeral herbs
  Actinotus minor 0.9 (–0.3 : 2.1)0.94 –0.9 (–3.7 : 1.9)  2.8 (0.7 : 5)  1.5(–0.6 : 3.6) 0.4(–1.5 : 2.3) 31
  Cassytha glabella 1.2 (–1.1 : 3.4)0.85  3 (–2.4 : 8.4)  3.2 (–0.9 : 7.3) –0.8(–4.9 : 3.3) 0.7(–2.9 : 4.4) 0.6(–4.4 : 5.7)41
  Drosera binata 0.8 (–1 : 2.6)0.81 –  –    4(0.9 : 7.2) 0.6(–2.2 : 3.4)–1.7(–5.6 : 2.2)21
  Gonocarpus micranthus 3.4 (1.1 : 5.7)1.00  4.9 (–0.5 : 10.2)  1.3 (–2.8 : 5.3) –0.2(–4.2 : 3.9) 5(1.4 : 8.6) 8.1(3.1 : 13.1)41
  Gonocarpus salsoloides 7 (4.1 : 10)1.00  4 (–2.8 : 10.8)  4.8(–0.4 : 10)  8.4(3.2 : 13.6)11.6(6.9 : 16.3) 3.3(–3.1 : 9.6)50
  Goodenia dimorpha ssp. dimorpha 4.2 (2.1 : 6.3)1.00  7.3 (2.4 : 12.2)  4.6(0.9 : 8.3)  7.3(3.5 : 11) 2.3(–1.1 : 5.6) 0.4(–4.2 : 4.9)50
  Mitrasacme pilosa 1.6 (–0.1 : 3.4)0.97 –0.3 (–4.2 : 3.7)  2.1(–0.9 : 5.1)  0.6(–2.4 : 3.6) 0.3(–2.4 : 3) 6.9(3.2 : 10.6)41
  Mitrasacme polymorpha 0.7 (–1.4 : 2.8)0.75  7.4(2.8 : 12) –0.6(–4.1 : 2.9)  1(–2.5 : 4.5)–2.7(–5.9 : 0.4) 2.8(–1.5 : 7)32
  Opercularia varia 1.9 (0.7 : 3.1)1.00  6.6 (4 : 9.3)  0.8(–1.2 : 2.9)  1.1(–0.9 : 3.1) 1.6(–0.2 : 3.4) 1.3(–1.2 : 3.7)50
  Stylidium lineare 0.4 (–1.4 : 2.1)0.67  3.9 (–0.2 : 8.1) –1.4(–4.6 : 1.7)  0.7(–2.5 : 3.8) 0.1(–2.7 : 2.9) 0.1(–3.7 : 4)41
  Xyris juncea 1.6 (0 : 3.2)0.97  0.8 (–2.8 : 4.4)  0.8(–1.9 : 3.6)  5.9(3.1 : 8.6) 0(–2.4 : 2.5) 30
5 Non-rhizomatous resprouting herbs and graminoids
  Blandfordia nobilis0.9 (–1.9 : 0)0.03  0.1 (–2.1 : 2.3) –3.1(–4.8 : –1.4)  0(–1.7 : 1.6)–1(–2.5 : 0.5) 12
  Burchardia umbellata 0.2 (–0.9 : 1.3)0.64  0.5(–2.2 : 3.2) –0.7(–2.7 : 1.4)  1.3(–0.7 : 3.3)  21
  Drosera peltata 0.4(–1.2 : 2)0.70 –  –3.1(–5.7 : –0.4)  5(2.4 : 7.6)  11
  Drosera spathulata4 (–6.3 : –1.6)0.00  2 (–2.7 : 6.7)  –4(–7.6 : –0.4)–12.3(–15.9 : –8.7)–2.4(–5.7 : 0.8) 13
  Entolasia stricta0.1 (–2.2 : 2)0.46 –1.4 (–6.2 : 3.4) –1.2(–4.8 : 2.5)  3.3(–0.3 : 7)–2.9(–6.1 : 0.4) 2.6(–1.8 : 7.1)23
  Gymnoschoenus sphaerocephalus0.6 (–2.1 : 0.9)0.22 –0.1 (–3.7 : 3.4) –0.2(–2.9 : 2.5)  0.1(–2.6 : 2.8)–2.8(–5.3 : – 0.4) 1.5(–1.8 : 4.8)23
  Haemodorum corymbosum 0.4 (–0.5 : 1.4)0.82  5 (3.1 : 6.9) –2.1(–3.5: – 0.7)  1(–0.5 : 2.4) 0.1(–1.1 : 1.4) 31
  Plinthanthesis paradoxa2.5 (–4.3 : –0.6)0.01 –0.7 (–4.9 : 3.4) –7.8(–10.9: –4.6) –0.4(–3.6 : 2.8)–2.2(–5.1 : 0.6) 0.1(–3.8 : 4)14
  Schoenus lepidosperma2.7 (–4.8 : –0.6)0.01  0.1 (–3.9 : 4.2) –4.5(–7.6: –1.4) –4.7(–7.7 : –1.6)–1.9(–4.7 : 0.8) 13
  Schoenus paludosus2.5(–4.2: –0.8)0.00  0.4(–3.8 : 4.7) –1.7(–4.9 : 1.6)–10.8(–14 : –7.6)–0.2(–3.1 : 2.7)–0.1(–4.1 : 3.8)14
  Sowerbaea juncea0.3 (–1.5 : 0.9)0.29  1 (–1.8 : 3.8) –1.8(–3.9 : 0.4)  1.1(–1 : 3.2)–1.3(–3.2 : 0.6) 0.4(–2.2 : 3)32
  Tetrarrhena turfosa2.9 (–5.6 : –0.1)0.02 –3.6 (–10.1 : 2.9) –2(–7 : 3) –0.2(–5.2 : 4.7)–3.2(–7.6 : 1.3)–7.3(–13.4 : –1.2)05
  Thysanotus juncea0.5 (–1.3 : 0.2)0.08  0.3 (–1.4 : 2) –2.3(–3.6 : –1) –0.2(–1.5 : 1.1)–0.1(–1.2 : 1.1) 13
  Xyris gracilis ssp. laxa0.5 (–1.6 : 0.5)0.16  0.4 (–2.1 : 3)  0.5(–1.4 : 2.4) –1(–2.9 : 0.9)–1.8(–3.5 : – 0.1) 22
  Xyris operculata0.2 (–2.4 : 2)0.44 –   4.8(1 : 8.7) –2.5(–6.4 : 1.3)–2.7(–6.1 : 0.7) 0.4(–4.3 : 5.1)22
6 Rhizomatous resprouting graminoids, herbs and ferns
  Baloskion gracile1.6 (–2.9 : –0.4)0.01 –1.9 (–4.7 : 0.8) –4.9(–7 : –2.8) –1.1(–3.2 : 1)–0.3(–2.2 : 1.6) 04
  Baumea acuta 3.2 (1.5 : 4.9)1.00  3.6 (–0.4 : 7.5)  1.2(–1.9 : 4.2)  6.2(3.2 : 9.2) 2.1(–0.6 : 4.8) 3.8(0 : 7.5)50
  Baumea rubiginosa 3.7 (1.6 : 5.7)1.00 –  –   2.1(–0.7 : 4.8) 3.5(1 : 5.9)15.5(12.1 : 18.9)30
  Baumea teretifolia2 (–4.1 : 0.2)0.04 –0.9 (–5.6 : 3.9) –  –1.2 (–4.8 : 2.5)–0.9(–4.1 : 2.4)–9.4(–13.9 : –4.9)04
  Chorizandra sphaerocephala0.2 (–2 : 1.7)0.43 –1.6 (–5.8 : 2.7)  0.7(–2.6 : 3.9)  0.9 (–2.4 : 4.1)–3.3(–6.2 : –0.4) 4.1(0.1 : 8.1)32
  Cyathochaeta diandra0.3 (–1.7 : 1.1)0.33 –2.6 (–5.9 : 0.8) –   0.1 (–2.5 : 2.7)  11
  Dampiera stricta3 (–4.7 : –1.3)0.00 –3.6 (–7.1 : –0.2) –8.1(–10.7 : –5.5)  1.2 (–1.4 : 3.8)–3.7(–6.1 : –1.4) 13
  Empodisma minus2.6 (–4.9 : –0.4)0.01 –8.8(–13.9 : –3.6) –0.7(–4.7 : 3.2)  0.5(–3.4 : 4.4)–3(–6.5 : 0.5)–4.3(–9.1 : 0.5)14
  Eurychorda complanata0.4 (–1.6 : 0.8)0.25  1 (–1.7 : 3.7)  0.9(–1.1 : 3) –0.5(–2.6 : 1.5)–2.9(–4.8 : –1.1) 1.3(–1.3 : 3.8)32
  Gleichenia microphylla 0.7 (–1.1 : 2.6)0.78  1.4 (–2.9 : 5.8) –   0.9(–2.4 : 4.2) 2.6(–0.4 : 5.6)–2.6(–6.7 : 1.4)31
  Gonocarpus tetragynus1.7 (–3.3 : –0.1)0.02 –3.4(–7.1 : 0.3) –4(–6.8 : –1.2) –2.4(–5.2 : 0.4) 0.9(–1.7 : 3.4)–1(–4.5 : 2.5)14
  Lepidosperma limicola1.4 (–3 : 0.2)0.05 –0.8 (–4.6 : 3)  0.7(–2.2 : 3.6) –2.8(–5.7 : 0.2)–1.2(–3.8 : 1.4)–3.4(–6.9 : 0.2)14
  Lepidosperma neesii0.8 (–2.8 : 1.3)0.24 –2.4(–7.3 : 2.4) –4.3(–7.9 : – 0.6) –0.8(–4.5 : 2.9) 2.1(–1.2 : 5.4) 0.6(–3.9 : 5.1)23
  Leptocarpus tenax4.4 (–6.6 : –2.2)0.00–12.9 (–17.6 : –8.2) –5.8(–9.3 : –2.2) –0.7(–4.3 : 2.8)–4.9(–8.1 : –1.7) 0.1(–4.3 : 4.5)14
  Lepyrodia scariosa4.1 (–6.5 : –1.6)0.00  3.4 (–2.1 : 8.8) –2.5(–6.6 : 1.6) –5.6(–9.8: –1.5)–8.7(–12.4 : –5)–2.4(–7.4 : 2.7)14
  Lindsaea linearis0.5 (–1.8 : 0.7)0.19 –5.1 (–7.8 : –2.3) –0.6(–2.7 : 1.5)  0.3(–1.9 : 2.4) 0.6(–1.3 : 2.5) 0.1(–2.4 : 2.7)32
  Lycopodium laterale 0.7(–0.5 : 2)0.88 –   1.1(–1.2 : 3.3)  1.7(–0.6 : 3.9) 0.5(–1.5 : 2.6)–0.1(–2.9 : 2.6)31
  Ptilothrix deusta 0.8 (–1.7 : 3.2)0.74 –2.7 (–8.5 : 3.1)  0.3(–4.1 : 4.8)  0.7(–3.8 : 5.1) 3.3(–0.7 : 7.2) 31
  Schoenus brevifolius1.1 (–3.8 : 1.7)0.22 –10 (–16.1 : –3.9) –1.1(–5.7 : 3.5) –1.4(–6 : 3.2)–0.7(–4.8 : 3.4) 6.4(0.7 : 12)14
  Selaginella uliginosa0.7 (–2.1 : 0.7)0.15 –1.4 (–4.7 : 1.9)  0.7(–1.9 : 3.2) –0.5(–3 : 2)–2.3(–4.6 : 0) 0.3(–2.8 : 3.3)23
  Tetraria capillaris0.1 (–1.7 : 1.6)0.46 –0.4 (–4.3 : 3.5) –0.1(–3.1 : 2.9) –2.4(–5.4 : 0.6) 0.3(–2.4 : 3) 3(–0.6 : 6.6)23

PFT 1, which showed a strong increase in mean abundance (Fig. 3), contained five species. Four of these increased in abundance, while the remaining species (L. squarrosum) was stable (P(increase) c. 0.77) (Table 2). In PFT 2, which showed a modest increase in abundance overall (Fig. 3), six species increased in abundance, four remained stable and three declined (Table 2). PFT 3, which declined in abundance overall (Fig. 3), included two species that increased in abundance, five species that remained stable and five that declined. Of the five that remained stable, three had P(increase) < 0.15 (Table 2), and hence a probability of decline greater than 85%. PFT 4, which increased in abundance overall (Fig. 3), included seven species that increased in abundance, four that remained stable and none that declined. All four stable species had P(increase) > 0.65 (Table 2). In PFT 5, which declined overall (Fig. 3), no species increased in abundance, eight remained stable and seven declined (Table 2). PFT 6, which also declined overall (Fig. 3), included two species that increased in abundance, 11 that remained stable and eight species that declined (Table 2). Thus, all species within PFTs 1, 4 and 5, either exhibited trends in the direction predicted by the model or were stable. The remaining PFTs included a minority of species that showed trends that were contrary to those predicted, and these few species included both uncommon and relatively common components of the community.

Plant species exhibited a range of different temporal responses in relation to environmental gradients. Twenty-seven of the 80 species analysed showed a consistent direction of change in abundance across all the plant communities in which they occurred (Table 2). However, many of these changes were relatively weak because the 95% confidence intervals for their estimated mean changes in abundance encompassed zero. For example, Almaleea paludosa exhibited a mean increase in abundance across all five plant communities in which it occurred, but in each case the confidence interval around the mean estimated change indicated a non-negligible chance that abundance may have declined.

Sixty-four of the 80 species analysed exhibited a strong change in abundance in at least one community (95% confidence interval for the mean change excluded zero) and, of these, 23 species exhibited a strong change in two or more communities. Only six of these 23 species had strongly conflicting directions of change in different communities, although each was relatively common in the system. These include: Hakea teretifolia, which underwent a strong decline in BT, but increased strongly in RH, SL and TT; Dillwynia floribunda, which underwent a strong decline in BT, but increased in CH; Leptospermum grandifolium, which declined strongly in TT and increased in RH; Drosera peltata, which declined strongly in RH and increased strongly in SL; Chorizandra sphaerocephala, which declined in CH and increased in TT; and Schoenus brevifolius, which declined in BT and increased in TT (all changes with > 95% confidence, Table 2).

Discussion

are plant functional types useful predictors of vegetation change?

Our results suggest that plant functional types based on vital attributes can provide a very useful framework for predicting vegetation change. Each of the six PFTs showed changes in abundance over 21 years that were consistent with the predicted direction of change for the observed fire regime scenario. Two main processes in the underlying model of vegetation dynamics were crucial to these successful predictions of vegetation change: seed bank accumulation and competition.

The model successfully predicted increases in abundance for PFTs 1, 2 and 4, principally on the basis of seed bank mechanisms. For PFT 1, fires occurred at times when large canopy seed banks had been accumulated, providing the basis for population increase through large post-fire recruitment events. For PFTs 2 and 4, relatively rapid accumulation of large persistent soil seed banks allowed their populations to evade the effects of competition, which becomes important later in the fire cycle (> 5–6 years post-fire) due to the development of tall, dense canopies of PFT 1 species.

The model also successfully predicted declines in the abundance of PFTs 3, 5 and 6, principally as a result of these competitive mechanisms because these PFTs are more heavily dependent on standing plants than seed banks for population persistence through successive fires. The model predicted that survival, growth and reproductive output of standing plants would be reduced by dense canopies of PFT 1 species, which increased in abundance under the observed fire regimes. Many of the species within PFTs 3, 5 and 6 were incapable of producing seed banks that persist on site longer than standing plants, further predisposing their populations to adverse effects of competition. Those species in PFTs 3, 5 and 6 that did have persistent seed banks were predicted to accumulate them more slowly than species in PFTs 2 and 4, due either to slow maturation, low fecundity or both. Consequently, the model predicted that seed banks did not provide an effective means for PFTs 3, 5 and 6 to evade competition from the canopy of PFT 1 species, as it developed with time since fire.

It is noteworthy that the three PFTs that increased in abundance were comprised almost entirely of obligate seeders, while the three PFTs that declined were exclusively resprouters. Bond & Midgley (2001) proposed that obligate seeders were dependent on ‘regeneration niches’, the range of conditions required for plant recruitment (Grubb 1977), while resprouters were dependent on ‘persistence niches’, the range of conditions required for persistence of established plants (Bond & Midgley 2001). Habitat destruction and disease have been identified as processes that are unfavourable for persistence niches (Bond & Midgley 2001; Keith et al. 2007). If our model of vegetation dynamics accurately explains the cause of declines in resprouters, then competition could be a third process that is unfavourable to persistence niches. The susceptibility of resprouters to competition is apparently a consequence of their slow growth, resulting from the allocation of resources to storage organs (Pate et al. 1991).

The vital attributes model, as originally formulated by Noble & Slatyer (1980), does not make qualitative predictions about change in abundance, but it successfully predicted the persistence of all six VA functional groups under the observed fire regimes. The VA model only predicts outcomes of persistence or local extinction (cf. qualitative changes in abundance) and, unlike its successor (Moore & Noble 1990), does not explicitly include the competitive relationships between different functional groups. Our results suggest a remarkable intolerance of many understorey species, particularly in PFTs 3, 5 and 6, to competition from overstorey species. While it is tempting to suggest that the dynamics of the system is more sensitive to competitive interactions than fire responses, it appears to be the interaction between disturbance regimes and competition that governs species’ responses. Fire responses determine which functional types are most sensitive to competition, and fire regimes determine when competition is sufficiently intense to influence species’ abundances by regulating the population density of overstorey species in PFT 1.

do plant functional types support robust generalizations across multiple species?

Species within each PFT showed reasonably consistent patterns (i.e. same direction) of change in abundance over the 21-year period of observation. For all PFTs except PFT 3, the predicted direction of change in abundance was observed in the majority of member species. In contrast, only seven of the 80 species examined showed a trend that was counter to that predicted for their PFT (three in PFT 2, two in PFT 3 and two in PFT 6). The lack of complete consistency of species’ responses within PFTs may be explained by a combination of statistical and mechanistic reasons.

A failure to detect predicted trends was in some cases due to insufficient statistical power to support a strong inference. Each PFT included one or more species that were inferred to have remained stable. In many such cases, the balance of probability suggested a change in the predicted direction, but with a non-negligible probability of a change in the direction contrary to prediction. For example, there was a 77% chance that the one ‘stable’ species in PFT 1 actually increased in abundance and more than 85% chance that three of the five ‘stable’ species in PFT 3 actually declined (see P(increase) in Table 2). Further sampling of these species may have permitted more certain inferences that their responses were consistent with predictions for their respective PFTs.

Deviations between observed and expected responses may also be attributable to heterogeneity of PFTs with respect to crucial life-history traits. To derive a parsimonious classification, we lumped 11 unique combinations of traits into six PFTs. PFT 3, for example, included species with three different kinds of seed bank (Table 1). Both species that increased in abundance, counter to the model's prediction, had persistent soil seed banks (Baeckea linifolia and Epacris paludosa, Table 2). These species were also the most fecund species in PFT 3, producing thousands of small seeds per plant each year (D. A. Keith, personal observation). Recruitment from large, persistent soil seed banks may be expected to compensate for predicted declines in standing plant numbers as a consequence of competition. Thus, heterogeneity in fecundity, a parameter not explicitly included within our PFT classification, may explain why the responses of a minority of species ran contrary to expectations for their PFT.

Similar explanations may account for the anomalous species responses within PFTs 2 and 6. Rates of seed bank accumulation are likely to have an important influence on the response of PFT 2 species (those with slower rates are more likely to decline under competition). Variation in the ability to spread laterally into canopy gaps or to spread rapidly in early post-fire years is likely to influence the response of PFT 6 species (those with greater propensity for spread are likely to increase under the observed fire regime). Neither of these characteristics was explicitly included in our PFT classification.

The consistency of PFT responses could be improved by incorporating more plant traits and recognizing more PFTs in our classification. However, there is a trade-off between greater detail in a PFT classification, perhaps leading to greater predictive accuracy, and increased numbers of PFTs in the classification, which increases its complexity and potentially reduces its utility as a practical tool for generalization (Gitay & Noble 1997). In our example, PFTs 5 and 6 could potentially be lumped, as their salient difference in life history (the propensity for lateral vegetative spread) apparently did not result in substantially different outcomes under the observed fire regimes. However, these PFTs may exhibit different responses under other scenarios that were not examined here. On the other hand, our classification of six PFTs was simplified by lumping some of the 11 factorial combinations of traits, and this probably increased the variability of responses within some PFTs.

Misclassification of species into PFTs is another potential explanation for anomalous responses of species within PFTs. We think this is unlikely to be an important explanation for the relatively few anomalies in our results because diagnoses of plant functional traits and assignment of study species and assigning species to PFTs were based on extensive field observations in both space and time. However, we note that uncertainty in PFT membership can be expected where data are scarce. Some forms of uncertainty, such as inherent vagueness and natural variability, may be difficult to reduce, even where abundant data are available (Regan et al. 2002). For example, in our study diagnosis of Hakea teretifolia (PFT 1) and Viminaria juncea (PFT 2) as non-resprouters (Table 1) was relatively uncertain because some individuals resprouted after 100% leaf scorch in some fires at some locations, even though death of standing plants was the most common response.

Our results are consistent with those of Keith & Bradstock (1994), who found no consistent pattern in the abundance of non-serotinous obligate seeder shrubs (PFT 2) in relation to the presence of a canopy of PFT 1 species, whereas serotinous resprouters (PFT 3) were consistently less abundant under a canopy than in the open. Tozer & Bradstock (2002) found that many understorey species (PFTs 2–6) were less abundant beneath a canopy of PFT 1 species, but did not show a strong pattern of relative abundance in relation to plant traits (resprouting ability, seed longevity or seed bank type). Their data show two PFT 3 species (Banksia oblongifolia and Xanthorrhoea resinifera) to be most abundant at sites that lacked a canopy of PFT 1 species, while two others (Aotus ericoides and Gahnia sieberiana) showed no such pattern. Five obligate seeding shrubs (PFT 2) in their study showed a wide range of abundance patterns in relation to a canopy of PFT 1 species, while the majority of resprouting herbs and sedges (PFTs 5 and 6) were less abundant under a canopy (Tozer & Bradstock 2002). Thus, their results are not in strong disagreement with our model. The variability of responses between species within PFTs found in previous studies is a likely result of similar causes to those outlined above. Sampling issues may be a further source of variation. Both of the previous studies were based on a space-for-time substitution sampling design, rather than a temporal comparison and, consequently, interpretation of the results is limited by unmeasured variation in initial abundance. Furthermore, both studies were based on less spatially extensive sampling than the current study.

do plant functional types support robust generalizations across different environments?

The responses of individual species were not invariant across the environmental variation represented by different plant communities. Nonetheless, the qualitative changes in abundance were reasonably consistent across the communities in which a given species occurred. More than one-third of the 80 species had the same qualitative responses (increase or decline) across all communities, although the trends were mostly weak. The remaining species had opposite responses in at least two of the communities in which they occurred, although again, many of these trends were weak. The apparent variability within species and weakness of trends suggests that there may be appreciable variation in species’ responses between sites within communities. However, this may be partially due to the limited number of samples per community, which had been pooled to calculate overall mean changes in abundance for comparisons between species within PFTs.

About one-quarter of those species that exhibited strong changes in abundance in more than one community had strongly opposing trends in different communities. For at least three of these six species with variable responses, the results suggest that competitive effects of overstorey species may vary between communities. Hakea teretifolia, Dillwynia floribunda and Schoenus brevifolius declined in Banksia Thicket, where PFT 1 species were most abundant (Keith & Myerscough 1993), but increased markedly in at least one other community. The density-dependence mechanisms potentially associated with these patterns warrant further investigation.

Conclusions

A temporal test of dynamic predictions, such as one described in this study, can provide powerful insights into the predictive utility and generality of the plant functional type approach. We conclude that PFTs are powerful tools for prediction and generalization. However, there are limits to the predictive properties and generality of the approach. Not all members of a PFT may show the expected response and species may not behave consistently in different environments. In this study, PFTs and their underlying model produced an accurate prediction of average vegetation responses over a 21-year period. The majority of species within each PFT exhibited the predicted response and few species exhibited strongly opposing responses in different environments. More elaborate models, which include processes that were not accommodated within our current model, may have generated more accurate predictions for the minority of species that behaved contrary to predictions and inconsistently across different environments. However, more complex models will trade off some generality and parsimony, which are important strengths of the PFT approach.

While our study has focused on fire regimes, it would be fruitful to examine PFTs in relation to other kinds of disturbance (e.g. inundation, grazing, climate change, etc.) by comparing predicted with observed changes over appropriate time scales to determine whether the PFT approach was similarly reliable in these situations.

Our results suggest that ecosystem managers may use PFTs based on vital attributes to reliably predict average changes in abundance for groups of species in response to particular scenarios of environmental perturbation. This makes the approach extremely useful in species-rich ecosystems, where management of all species is unlikely to be plausible (Keith et al. 2002b). Despite the overall accuracy and generality of PFTs for predicting change, however, managers should be cognisant of limitations that restrict accurate predictions about the response of each individual species across all environments. Management strategies will therefore need to complement PFT approaches by providing for management and monitoring of some individual species whose conservation or functional status warrants priority action.

Acknowledgements

Helen Jessup (National Parks and Wildlife Service) and George Williams (Sydney Catchment Authority) provided logistic support, access to the study area and information on its history and management. Belinda Pellow, Peter Myerscough, Jeannie Highet, Ted Jones, Jono Sanders, Ros Muston and Phil Ward assisted with field sampling. William Bond and Peter Myerscough provided helpful comments on a draft manuscript. This project was funded by the Special Areas Strategic Management Research and Data Program (project number RD07).

Ancillary