Floristic diversity in fire-sensitive eucalypt woodlands shows a ‘U’-shaped relationship with time since fire

Authors


Correspondence author. E-mail: carl.gosper@dec.wa.gov.au

Summary

  1. Understanding ecosystem responses to disturbance is important for effective management of biodiversity. Observed relationships between time since disturbance and diversity have taken a variety of forms, only some of which are explicitly predicted in models of vegetation succession. This makes generalization and predictions for specific communities difficult.
  2. Negative relationships have been the predominant diversity response to time since fire in fire-prone Mediterranean-climate ecosystems; however, few studies have analysed responses in infrequently burnt ecosystems such as Mediterranean-climate woodlands dominated by fire-sensitive trees. We used a space-for-time approach and multiple stand-ageing techniques (Landsat imagery, growth ring counts and growth ring–size relationships) to characterize diversity and compositional changes with time since fire (3–370+ years) in fire-sensitive Eucalyptus salubris woodlands in south-western Australia.
  3. Species density and Pielou's evenness showed an overall ‘U’-shaped response to time since fire, although variability between plots was considerable. Plant functional type and species composition differed with time since fire, with greater richness and cover of ground layer, and ‘long dispersal potential’ functional types with increasing time since fire. Conversely, there was an early or intermediate peak in taller and ‘short dispersal potential’ functional types.
  4. We propose that the unusual ‘U’-shaped diversity–time since fire relationship is driven by competitively dominant tree and shrub layers having maximum cover at intermediate times since fire. Subdominant functional types were able to exploit lower levels of competition in the immediate post-fire period and after density-dependent thinning of the trees and shrubs.
  5. Synthesis and applications. Recurrent fire is not required to maintain diversity in these fire-sensitive woodlands as diversity reached a maximum in mature vegetation. Fire intervals of <c. 200 years are likely to have adverse consequences on diversity, which is of conservation concern given apparently high recent rates of occurrence of fire. Changes in diversity were not apparent when times since fire were truncated to those available from remote sensing, illustrating that space-for-time studies defined solely by remote sensing may obscure equivalent ‘U’-shaped diversity–time since fire relationships.

Introduction

Disturbances are important ecosystem processes affecting the balance of species dominance, availability of resources and opportunities for recruitment (Connell 1978; Pickett & White 1985). Patterns of species diversity (including species richness and evenness) and community composition are influenced by the type, frequency, scale and history of disturbances. Determining appropriate disturbance regimes for particular ecosystems is thus an important issue for natural resource management.

In a review of observed relationships between diversity and level of or time since disturbance, Mackey & Currie (2001) found the relationship can take a variety of forms. Non-significant, positive monotonic and negative monotonic relationships were most frequent. Peaked relationships were less frequent, and other shapes were rare. Only some of these relationship forms are explicitly predicted in models of vegetation succession, making generalization and predictions for specific communities difficult. Negative monotonic diversity changes are consistent with the initial floristic composition model of succession, where all species occur immediately after the disturbance and changes in diversity and composition over time are driven by the sequential loss of species from the above-ground vegetation through differential rates of growth, competition and longevity (Egler 1954). Peaked diversity patterns are consistent with the intermediate disturbance hypothesis, when species coexistence is maximized between disturbance-tolerant but competitively less dominant species and disturbance-intolerant but competitively dominant species (Connell 1978).

The other relationship forms are not strongly associated with specific models of vegetation succession. They could occur, however, under models that do not include explicit predictions of changes in diversity with time since disturbance, such as relay succession (e.g. the classical model of Clements (1916)) and non-equilibrium models (e.g. Cattelino et al. 1979). Relay succession predicts changes in community composition, with post-disturbance communities replaced by new communities in an ordered manner if undisturbed for a sufficient period. Non-equilibrium succession models also predict community composition change, via multiple succession pathways due to variation in disturbance characteristics or local environmental factors.

Plant functional types (PFTs) offer an alternative approach to predicting patterns of vegetation change under specific disturbance regimes (McIntyre, Lavorel & Tremont 1995; Keith et al. 2007; Gosper, Yates & Prober 2012a). This approach involves the grouping of plants based on traits relevant to processes of vegetation change under a specific vegetation assembly model (Keith et al. 2007). More generalized predictions about diversity and composition changes with time since disturbance may then be made based on PFT responses.

Recurrent fire is a dominant disturbance in landscapes across Mediterranean-climate regions (Cowling et al. 1996) and can impinge strongly on plant species composition and diversity (Bond & van Wilgen 1996). Negative relationships between time since fire and diversity, supporting the initial floristic composition model of succession, have been the predominant response form in the major vegetation associations of these regions, including heathlands in Australia and South Africa, and shrublands and forests in California and the Mediterranean (Gill 1999; Gosper et al. 2012b; Keeley et al. 2012). Cases of relay succession over long periods in the absence of fire have also occurred, with increasing abundance of interfire recruiters, such as Callitris, Allocasuarina and rain forest taxa, in Eucalyptus woodlands in Australia (Keeley et al. 2012). The species that contribute to these changes in diversity with time since fire are predictable based on their functional traits (Keith et al. 2007; Gosper, Yates & Prober 2012a). There have, however, been few studies of the response of plant species composition and diversity to time since fire in ‘fire-sensitive’ Mediterranean-climate woodlands, where long-lived dominant overstorey trees and shrubs are typically killed by stand-replacing canopy fires, resulting in dense post-fire recruitment (Keeley et al. 2012; but see Kutiel 1997).

The Great Western Woodlands (GWW) region of south-western Australia supports the world's largest remaining area of Mediterranean-climate woodland, which in mosaic with mallee, shrublands and salt lakes cover an area of 160 000 km2 (Watson et al. 2008; Prober et al. 2012). Eucalyptus woodlands in this region are typically fire sensitive and slow growing, and fire return intervals recorded over recent decades have been much shorter than the long-term average (O'Donnell et al. 2011a; Parsons & Gosper 2011). This has led to considerable conservation concern regarding the loss of mature woodlands (DEC 2010) and has highlighted a need to better understand impacts of fire on plant composition and diversity.

Towards this goal, we employed a space-for-time approach to examine the diversity and composition responses of plant species and PFTs to time since fire in gimlet Eucalyptus salubris F.Muell. woodlands. We predicted decreasing plant diversity with time since fire, consistent with the initial floristic composition model of succession that is most prevalent in other Australian and Mediterranean-climate ecosystems (Gill 1999; Keeley et al. 2012). Consistent with the PFT approach, we also predicted that change in community and PFT (defined by competitive stratum and dispersal potential) composition would coincide with the maximum competitive influence of the tree canopy (Scanlan & Burrows 1990; Keeley et al. 2012), as we expected the process of competition would be a significant driver of vegetation change over the interfire period. Contrary to our prediction, we found that diversity indices showed a ‘U’-shaped relationship with increasing time since fire, but this response was consistent with the competitive influence of overstorey dominants as predicted by the PFT approach.

Materials and methods

Survey plots

We established 72 plots along the western edge of the GWW, south-western Australia: near Karroun Hill (30o14′S, 118o30′E), Yellowdine (31o17′S, 119o39′E) and Parker Range (31o47′S, 119o37′E). This area has a semi-arid Mediterranean climate (see Gosper et al. 2013 for more climatic details). Plots had a dominant crown layer of E. salubris, sometimes in association with other Eucalyptus spp.

The sample area contained large recent (<10 years) and older (>38 years, but readily visible in 1972 Landsat imagery) fire scars and large areas with no evidence of contemporary fire. Plots were distributed in these age classes across the geographic spread of sampling, with additional plots 10–38 years post-fire sampled where such fires had occurred. Plot time since fire was determined through a combination of Landsat imagery analysis, growth ring counts and estimates based on E. salubris growth ring–size relationships (Gosper et al. 2013). We use the best two plausible models for estimating time since fire: (i) untransformed growth rings predicted by diameter at the base (D10) plus location (Model 2), and (ii) square-root-transformed growth rings predicted by D10 plus location (Model 5), acknowledging that the certainty of plot time since fire declines with increasing time, particularly beyond 200 years (Gosper et al. 2013). The sample age range was 2–370 (Model 2) or 2–1460 (Model 5) years post-fire (see Table S1, Supporting Information). Analyses were conducted using Model 2 times since fire, although for interpretation purposes, the equally valid alternative time-since-fire scale from Model 5 has been added to figures as a second x-axis or to text in parentheses. Information on other aspects of the fire regime other than time since fire was not available.

Plots were 50 × 50 m and placed >250 m apart (>500 m apart for plots of the same time since fire) in relatively uniform vegetation within 1 km of vehicular tracks on public land. As fires across this landscape can be very large, sometimes exceeding 100 000 ha (McCaw & Hanstrum 2003), it was often necessary to place multiple plots within the one fire scar. Having multiple samples within individual fires potentially creates problems in disentangling time-since-fire effects from other factors (Hurlbert 1994). In using a ‘space-for-time’ approach, we assume that each of the different plots is truly comparable and that fire event effects (Bond & van Wilgen 1996) do not confound time-since-fire effects. Although there is subsequently a degree of uncertainty in attributing differences to time since fire alone, we have taken a number of steps to minimize the difficulties for interpretation. First, we sampled a broad geographic area, so overall a number of different fires were sampled for each time-since-fire class. Second, many of our analyses concentrate on the detection of trends over time, rather than solely determining differences between individual fires, which would be confounded by sampling a limited number of fire events. Third, the spread of plots of each time since fire across the study area suggests that systematic location bias confounding time-since-fire effects are unlikely.

Vegetation sampling

Plots were sampled in spring 2010 or spring 2011, with each time-since-fire class sampled in each year. To explore the impact of sampling effort on rates of species accumulation between times since fire, we tallied the cumulative number of taxa in nested plots of 1, 5, 10, 25, 100, 500, 1000 and 2500 m2. Sampling plots over multiple sizes allow for an assessment of changes in the rates of species accumulation with area and help determine the appropriate scale at which comparisons should be made between times since fire.

To measure abundance, we systematically placed a 12·5-mm-diameter pole vertically at 50 points per plot, each 3 m apart, and recorded the identity of all plants with live parts intercepting the pole. Sixteen of these points were placed along each of the two sides of the plot commencing at the north-west corner, with the remaining 18 points placed along the diagonal starting in the same corner. This technique provided an objective measure of abundance reflecting but not equivalent to projective cover and is hereafter referred to as ‘cover’. Species that were present but not recorded at point intercepts were allocated a nominal proportional abundance of 1%.

A soil sample was taken at a subset of plots (the 52 sampled in 2010). There were no discernible differences in soils between time-since-fire age classes, indicating that soil factors can be discounted as confounding time-since-fire effects (Appendix S1, Supporting Information).

Plant functional types

PFT approaches aim to group species on the basis of traits that determine their response to specific disturbances under a specified model of vegetation change (Keith et al. 2007; Gosper, Yates & Prober 2012a). In studies of the response of species to disturbance by fire, traits used to define PFTs often include the capacity of individuals to persist after fire, the form and persistence of the seed bank, competitiveness and dispersal capacity (Keith et al. 2007; Gosper, Yates & Prober 2012a). In E. salubris woodlands, there is insufficient knowledge to classify a majority of species by these traits. However, groupings of species on the basis of traits relevant for vegetation change following other forms of disturbance (McIntyre, Lavorel & Tremont 1995; Pausas & Lavorel 2003; Meers et al. 2010) may still provide insights for fire management.

We defined PFTs on the basis of life-form and plant height (seven categories) and seed dispersal potential (long vs. short) (Table 1). Life-form and plant height are important traits with respect to the effects of time since fire through being proxies for competitive dominance during the interfire period (Pausas & Lavorel 2003; Keith et al. 2007) and plant longevity. Dispersal potential reflects the capacity for persistence at the landscape level (Pausas & Lavorel 2003) and provides an indication of the likelihood of dispersal between areas of different times since fire and perhaps the potential for interfire recruitment. Fire intervals in E. salubris woodlands can be relatively long (many hundreds of years; O'Donnell et al. 2011a; Gosper et al. 2013), and the climate is temporally variable (O'Donnell et al. 2011b); hence, opportunities may arise for significant interfire recruitment among those PFTs with the capacity for dispersal to suitable microsites. Of the 14 potential combinations of these traits, 12 were represented in the sampled flora of 270 species (Tables 1 and S2, Supporting Information).

Table 1. Traits used for plant functional type (PFT) classification. Life-form types were based on McIntyre, Lavorel & Tremont (1995) and Meers et al. (2010), except for some differences in the height divisions between life-forms and the addition of aerial parasites. Dispersal potential was based on putative dispersal mechanisms following assessment of dispersal unit morphology, with species aggregated into two trait groups on the basis of whether their seed dispersal mechanisms were likely to assist in long-distance dispersal or not. As we lacked seed size data, very small seeds (the ‘Mobile’ groups of McIntyre, Lavorel & Tremont 1995 and Meers et al. 2010) were included in barochory. Trait abbreviations are in parentheses, which for each combination of life-form and dispersal potential form PFTs (e.g. Tree-L). No representatives of Geo-L or Aerial-S were recorded
Life-forms
Aerial parasite (Aerial)
Phanerophytes – trees (>8 m height) (Tree)
Phanerophytes – shrubs (1–8 m height) (Shrub)
Chamaephytes (shrubs 0·1–1 m height) (Low shrub)
Hemicryptophytes (persistent buds at soil surface) (Hemicryp)
Geophytes (Geo)
Therophytes (annuals) (Annual)
Dispersal potential
Long (Endozoochory, Epizoochory, Anemochory) (L)
Short (Myrmecochory, Barochory) (S)

Statistical analyses

For the purpose of some analyses where categorical age classes were useful, we divided our sample plots into ‘young’, <18 years post-fire (= 19); ‘intermediate’, 35–120 (35–200) years post-fire (= 26); and ‘mature’, >140 (>260) years post-fire (= 27). The divisions between age classes reflected gaps in the range of times since fire sampled. Annual rainfall varied markedly between our two sample years, with rainfall in 2010 at Merredin being the 4th lowest of over 100 years of records, while in 2011, rainfall was well above average (Bureau of Meteorology 2012). In statistical analyses where we have incorporated a factor for the year of sampling, we have classified it as a fixed factor, meaning that our results should be interpreted as being for this combination of sampling years (rather than sampling year being random).

primer software (Version 6.1.11, PRIMER-E, Plymouth, UK) was used to derive diversity indices for floristic plot data. We tallied total species number per plot (species density) and, using cover data, calculated Pielou's evenness index for species and PFTs. To examine relationships between diversity indices and time since fire, simple regression models (linear or quadratic) were fitted using sigmaplot 10.0 (Systat Software Inc., Chicago, IL, USA), with model selection based on minimizing Akaike Information Criterion (AIC). To reduce the leverage of the few plots of longest time since fire, we applied square-root transformation to plot age. As sample plots were distributed over a wide geographic area (mostly north–south; Gosper et al. 2013), we tested the possibility that diversity indices differed in a consistent fashion with location by conducting regressions as above of indices against northing.

anova, using Statistica (Version 7.1, Statsoft, Tulsa, OK, USA), was used to test for differences in PFT species density and cover between time-since-fire age classes and sampling year. To homogenize variances, transformations were applied in some cases (Table 2).

Table 2. Species density and cover of PFTs between time-since-fire age classes (young, <18 years post-fire; intermediate, 35–120 (35–200) years; mature, >140 (>260) years). Table 1 has PFT abbreviations. Different-lettered superscripts indicate differences between time-since-fire age classes according to post hoc Newmann–Keuls tests
PFTTotal nYoungIntermediateMatureTime since fire (T)Sample year (S)× S
Mean species density ± SE F 2,66 F 1,66 F 2,66
  1. a

    Variances for Aerial-L were unable to be homogenized, so were not analysed with anova.

  2. b

    Square-root- (x + 1) or §log10(x + 1) transformation was applied to homogenize variances.

  3. *< 0·05; ** 0·01; *** 0·001.

Aerial-La20·00 ± 0·000·08 ± 0·050·30 ± 0·10   
Tree-L§41·00 ± 0·22a0·50 ± 0·15b0·11 ± 0·06b7·89***0·430·09
Tree-S163·47 ± 0·222·73 ± 0·212·85 ± 0·241·552·210·37
Shrub-L325·32 ± 0·574·81 ± 0·336·04 ± 0·301·991·840·29
Shrub-S8612·9 ± 0·78a12·2 ± 0·72a8·74 ± 0·59b8·29***0·040·06
Low shrub-Lb152·79 ± 0·53b2·12 ± 0·31b4·19 ± 0·46a5·91**4·17*0·34
Low shrub-Sb121·26 ± 0·35a0·77 ± 0·17a0·22 ± 0·10b7·40**0·061·39
Hemicryp-Lb142·47 ± 0·34b2·46 ± 0·48b3·85 ± 0·32a9·10***1·871·42
Hemicryp-Sb70·42 ± 0·18b0·31 ± 0·11b1·04 ± 0·18a5·55**0·010·28
Geo-S80·84 ± 0·241·62 ± 0·231·59 ± 0·272·471·440·22
Annual-Lb456·16 ± 0·954·88 ± 0·938·52 ± 1·352·431·270·86
Annual-Sb292·63 ± 0·532·54 ± 0·513·67 ± 0·690·8417·9***0·85
Total L11217·7 ± 1·30b14·8 ± 1·30b23·0 ± 1·61a9·59***2·071·85
Total S15821·5 ± 1·3720·2 ± 1·2018·1 ± 1·211·524·51*0·12
  Mean cover ± SE   
Aerial-La 0·00 ± 0·000·08 ± 0·050·30 ± 0·10   
Tree-L§ 1·21 ± 0·32a0·73 ± 0·36b0·11 ± 0·06b7·11**0·000·27
Tree-S 33·6 ± 1·92b45·2 ± 2·40a34. 7 ± 2·63b6·96**2·861·77
Shrub-L 10·2 ± 1·307·88 ± 0·8110·3 ± 0·930·690·030·76
Shrub-Sb 29·7 ± 2·67a30·9 ± 2·98a19·4 ± 1·66b5·58**0·000·83
Low shrub-L 2·74 ± 0·54b2·23 ± 0·36b4·63 ± 0·56a6·20**2·310·22
Low shrub-S§ 3·26 ± 1·19a3·54 ± 1·52a0·59 ± 0·44b6·07**0·170·53
Hemicryp-Lb 2·95 ± 0·63b2·54 ± 0·28b4·33 ± 0·48a5·74**1·100·89
Hemicryp-S§ 0·42 ± 0·16b0·31 ± 0·11b1·19 ± 0·22a7·16**0·360·66
Geo-S 0·84 ± 0·241·69 ± 0·241·74 ± 0·322·390·870·13
Annual-Lb 6·26 ± 0·96a,b5·04 ± 0·97b10·4 ± 1·93a3·46*1·971·11
Annual-Sb 2·74 ± 0·543·00 ± 0·794·70 ± 0·991·6221·6***1·40
Total L 23·4 ± 1·74b18·5 ± 1·45b30·0 ± 2·53a5·17**0·240·60
Total Sb 70·6 ± 2·43b84·6 ± 3·03a62·3 ± 3·00b13·8***0·130·72

primer was used for ordination analyses of floristic and PFT composition. Cover data were square-root-transformed to decrease the influence of species with high cover. We used non-metric multi-dimensional scaling, with the Bray–Curtis dissimilarity metric, and permanova and permdisp to test for differences in statistical location and dispersion, respectively, among vegetation age classes and year of sampling. Vectors of a range of extrinsic variables (location, elevation, soil parameters, land tenure and plot time since fire) were estimated in ordination space for the floristic data set. The SIMPER algorithm was used to determine which species and PFTs contributed most to similarity within and dissimilarity between times-since-fire classes.

Results

Floristic diversity

Species–area accumulation plots were similarly shaped regardless of time-since-fire age class (Fig. 1); hence, floristic comparisons are not substantially influenced by plot size. The rate at which new species were encountered declined as plot size increased, but did not reach a plateau. Consequently, we conducted all further floristic analyses at the 2500-m2 scale. Vegetation of an intermediate time since fire, 35–120 (35–200) years, had lower species density than either young or mature vegetation across all plot sizes.

Figure 1.

Species–area accumulation rates in Eucalyptus salubris woodlands, showing the mean ± standard error of all samples.

Species density and Pielou's evenness (species- and PFT-based) were best represented by a concave quadratic function with time since fire (AIC for species density, linear model = 360, quadratic = 352; species-based evenness, linear = −266, quadratic = −392; PFT-based evenness, linear = −357, quadratic = −377). All had higher values shortly post-fire (<c. 10 years) and when long unburnt (>c. 200 [>c. 450] years), and lower values over the intervening period (Fig. 2). No regressions of diversity indices against location were significant.

Figure 2.

Changes in (a) species density and (b) Pielou's evenness for Eucalyptus salubris woodlands with time since fire. The function and shape of curve for the relationship between plant functional type evenness and time since fire was similar, with years since fire again square-root-transformed (Evenness = 0·789 − 0·032+ 0·0021x2; Adj. r2 = 0·345, F2,69 = 19·7, < 0·0001). There were no significant relationships between species density (c) and Pielou's evenness with time since fire in analyses with times since fire truncated at 38 years as per Landsat imagery. Analyses (a) and (b) were conducted using the time-since-fire distribution from Model 2 of Gosper et al. (2013) (bottom x-axis), but the equally valid alternative distribution from Model 5 is shown (top x-axis) for comparison.

Floristic composition

permanova indicated that time since fire and year of sampling had significant effects on floristic composition (Table S3, Supporting Information). Mature vegetation was distinct in composition, while young vegetation and intermediate vegetation were similar (Fig. 3). Young vegetation was more variable than either intermediate or mature vegetation. Correlation coefficients of extrinsic variables (Fig. 3) indicated that plot location also strongly influenced species composition. These results were consistent when reanalysed with the numerically and competitively dominant Eucalyptus spp. omitted. Many of the species that contributed most to differentiation among times-since-fire classes were among those with the highest cover (Table S4, Supporting Information).

Figure 3.

Non-metric multi-dimensional scaling of Eucalyptus salubris woodland plots by cover of (a) plant species and (b) plant functional types. Three-dimensional final solutions, with the first and second dimensions shown. Vectors are of (a) extrinsic variables with Pearson's correlation coefficients >0·6; (b) PFTs with Pearson's correlation coefficients >0·7. Table 1 has PFT abbreviations.

PFT composition

Time since fire and year of sampling affected cover of PFTs in a similar way to their effect on species, with stronger distinction of mature woodland than between intermediate and young woodland (Table S3, Supporting Information; Fig. 3). The relative changes in PFT species density (richness) and cover with time since fire and year of sampling provide insights into the underlying processes driving these community-level changes (Table 2).

PFTs that decreased in species density and/or cover with time since fire tended to be from taller PFTs, with phanerophyte trees with long dispersal potential (e.g. Hakea francisiana; see Table S2, Supporting Information for taxonomy) and phanerophyte shrubs with short dispersal potential (e.g. Acacia hemiteles) displaying this pattern of change. Phanerophyte trees with short dispersal potential (e.g. E. salubris) had no change in species density but a peak in cover in vegetation of an intermediate age. The only ground-layer PFT to have declining species density and/or cover in mature vegetation was chamaeophytes with short dispersal potential (e.g. Microcybe multiflora subsp. multiflora) (Table 2).

Other ground-layer PFTs tended to increase with greater time since fire, including chamaeophytes with long dispersal potential (e.g. Sclerolaena diacantha), hemicryptophytes with long (e.g. Austrostipa elegantissima) and short (e.g. Lomandra effusa) dispersal potential, and therophytes with long dispersal potential (e.g. Calotis hispidula). A similar non-significant trend occurred in geophytes with short dispersal potential (e.g. Arthropodium curvipes) and therophytes with short dispersal potential (e.g. Hydrocotyle rugulosa). Aerial parasites with long dispersal potential (e.g. Lysiana casuarinae) were absent from young vegetation. Phanerophyte shrubs with long dispersal potential (e.g. Santalum acuminatum) appeared to be the PFT least responsive to variation in time since fire.

In addition to the trends associated with life-form and plant height, there were overall patterns of change apparent with dispersal potential. In aggregation, plants with long dispersal potential had increasing species density and cover in mature vegetation. Species with short dispersal potential showed no changes in species density with time since fire and peak cover in vegetation of an intermediate age (Table 2).

Year of sampling only affected species density and cover of a small number of PFTs (Table 2). In all cases, indicating a capacity for the community to respond to changing climatic conditions and consistent with increased rainfall, species density and/or cover in 2011 exceeded that in 2010.

Discussion

U’-shaped diversity responses

Contrary to our hypothesis that diversity would decline monotonically with time since fire, diversity indices (species density, species and PFT evenness) all reached a nadir at an intermediate time since fire, although variability between plots of similar time since fire was considerable. While similar ‘U’-shaped responses to time since disturbance do occur (e.g. Kutiel 1997; Langlands, Brennan & Ward 2012), they appear rare compared to other relationship forms (Mackey & Currie 2001). This leads to the question of what characteristics of E. salubris woodland promote ‘U’-shaped responses, and whether these characteristics might predict other cases.

We propose that the ‘U’-shaped response in E. salubris woodland is driven by changing levels of overstorey competition, consistent with observed inverse relationships between overstorey dominance and abundance of understorey species (Scanlan & Burrows 1990; Pekin et al. 2012). In young woodlands, resources are unlikely to be fully sequestered by the regenerating phanerophytes, allowing less competitive species to grow and reproduce (Keith et al. 2007). As in many fire-prone ecosystems, diversity then declines as the dense phanerophyte regeneration drives lower understorey diversity at intermediate times since fire. Unlike many other fire-prone ecosystems, in mature woodlands, the cover of the dominant tree and shrub layers then declined, presumably reducing competitive effects on understorey species (Pekin et al. 2012).

We predict that ‘U’-shaped responses in diversity indices to time since fire would most likely occur in plant communities where post-fire changes are driven by competitive hierarchies and were composed of species with the following combination of traits:

  1. Adults of the competitive dominants are killed by fire and recruit en masse post-fire from a persistent seed bank. This combination of traits allows opportunities for species to exploit a relatively long (c.f. competitive dominants that re-sprout following fire; Pausas 1999; Gosper, Yates & Prober 2012a) low-competition post-fire environment, thereby leading to high diversity indices shortly post-fire.
  2. Massive levels of recruitment set the scene for a protracted competition bottleneck as the seedlings of the competitive dominants grow, leading to substantial suppression of subdominant vegetation (Scanlan & Burrows 1990) at intermediate times since fire. Importantly, this high-density, high-competition environment must not facilitate the establishment of a new set of species (such as under classical succession models).
  3. Density-dependent thinning and an associated reduction in cover of the competitive dominants eventually result in reduced overstorey competition.
  4. A pool of species must exist that are able to colonize and exploit the lower levels of competition and/or enhanced facilitation (possibly via hydraulic redistribution of water under tree canopies; Brooksbank et al. 2011) that follow density-dependent thinning.
  5. The competitive dominants are long-lived and/or capable of sporadic interfire recruitment, to avoid the decline in diversity indices that often characterize senescent shrublands (Bond 1980; Gosper et al. 2012b).
  6. A subsequent set of community dominants does not become established and initiate a second wave of competition-driven diversity changes (e.g. when rain forest species become established under E. regnans; Jackson 1968).

This combination of traits appears restrictive, yet we propose that they are not unique to E. salubris woodlands and may also occur in (i) some boreal forests, which can have much higher tree density and canopy cover at intermediate times since fire (Alexander et al. 2012) and some individual species which show ‘U’-shaped responses in basal area with time since fire (Chen & Taylor 2012); (ii) some Mediterranean Pinus forests and woodlands, which often have dense regeneration after fires (Pausas et al. 2008) and have shown ‘U’-shaped diversity changes with time since crown fires (Kutiel 1997); (iii) Callitris woodlands in eastern Australia, which regenerate in dense stands after disturbances, and thinning regrowth increases cover of groundstorey vegetation (McHenry et al. 2006); and (iv) other fire-sensitive eucalypt communities in south-western Australia (e.g. mallet and marlock).

There are, however, alternative processes that could explain ‘U’-shaped diversity–time since fire relationships. For example, the abundance of many animal species changes with time since fire (Watson et al. 2012), and some animals are important seed dispersers. Long dispersal potential PFTs had greater richness and cover in mature woodlands; hence, if key animal seed dispersers, such as emus Dromaius novaehollandiae (Calviño-Cancela et al. 2006), prefer mature woodlands, this may lead to increased seed immigration and potential for interfire recruitment. There is, however, a lack of data on responses of local animal seed dispersers to time since fire that would lend support to this suggestion. Clearly, further work is needed to determine the relative roles that competition for resources and rates of seed immigration play in creating the ‘U’-shaped response to time since disturbance we observed in E. salubris woodlands.

While the ‘U’-shaped response has been rare in the literature, it is possible that the very long time frames over which such changes occur have led to them being poorly detected in the past, due to technical constraints on determining the times since disturbance of events pre-dating remotely sensed imagery. Indeed, if we truncated our time-since-fire distribution to that able to be determined from Landsat imagery, there were no significant relationships between diversity indices and time since fire (Fig. 2c). Basing interpretations of vegetation change with time since fire in infrequently burnt communities solely on the basis of remotely sensed data can thus lead to misleading findings.

Succession models and PFT groupings

Consistent with Gosper, Yates & Prober (2012a), effects of time since fire on species diversity and cover mirrored effects on PFT diversity and cover, despite the relatively rudimentary PFT classification available for this study. Further, similar to Kazanis & Arianoutsou (2004), who evaluated changes in PFT richness and abundance in fire-sensitive Mediterranean-climate forests, we found differential responses among PFTs to time since fire, illustrating the usefulness of PFTs for interpreting patterns of vegetation change after fire.

Increasing species density in three perennial PFTs with increasing time since fire (and chamaeophytes with long dispersal potential between years) provides evidence of biologically meaningful levels of interfire recruitment. Eucalyptus salubris woodlands, in contrast to some neighbouring communities (Gill 1999; Gosper et al. 2012b), thus do not adhere strictly to the predicted initial floristic composition model of community assembly. The intermediate disturbance hypothesis also does not apply, as diversity indices reached a nadir at intermediate times since fire. Although the stability of species density in dominant phanerophyte trees with short dispersal potential does not indicate classical relay succession, changes in lower vegetation layers may provide some support to this model of vegetation change.

Time since fire rarely affected PFTs with the same life-form (but different dispersal traits) in the same way, indicating that competitive dominance does not play a unilateral role in structuring E. salubris woodlands. Overall changes in long and short dispersal potential PFTs with time since fire suggest that dispersal mode plays a role in determining the ability of plants to respond to changing resources. Long dispersal potential by definition indicates a greater probability of seed immigration, with some of the dispersal modes within this trait potentially leading to directed dispersal to favourable microsites for recruitment or survival (Wenny 2001). Some hemicryptophytes may also be able to expand through vegetative growth and lateral spread (Keith et al. 2007).

Management implications

Species density and evenness reached a maximum in mature vegetation; hence, there is no support from these community-level measures for E. salubris woodlands requiring recurrent fire to maintain diversity. Notwithstanding, some of the species contributing to higher diversity in young and old woodlands differed. Long intervals between fires potentially adversely affect species or PFTs that mainly occur in the above-ground vegetation in young woodlands, although any declines are likely to be mediated through traits such as seed bank longevity, which are in general poorly quantified.

While there was evidence for species turnover with time since fire in E. salubris woodlands, the overall community type showed no evidence of change. Areas that were E. salubris woodlands before fire (as determined by post-fire stags; C. Gosper pers. obs.) regenerated with E. salubris as a dominant species, and this remained so over the whole chronosequence. Recognizing that time-sequence rather than space-for-time analyses would be more appropriate for testing this hypothesis, our results cast doubt on the proposal by Berry et al. (2009) that even single fires change woodlands into pyric successional stages of mallee and shrubland.

The number and size of recent wildfires have led to concerns being raised over the loss of mature Eucalyptus woodlands (Watson et al. 2008; DEC 2010). Recent fire return intervals across the same part of the GWW in which this study was conducted have been in the order of about every 40 years (Parsons & Gosper 2011). This interval is an aggregate across all vegetation communities and does not account for community-level differences (see O'Donnell et al. 2011a), but appears shorter than longer-term fire intervals in Eucalyptus woodlands (c. 400 years; O'Donnell et al. 2011a). If intervals <c. 200 years were to become widespread in E. salubris woodlands, species density and evenness would fail to increase to the community's maximum, and the mature vegetation community that appears distinct in species and PFT composition would not develop. Lower-intensity fires that do not entirely kill the dominant overstorey (e.g. in the early stages of fire development, or flanking or backing sections of larger fires) may have different outcomes for community diversity, but such fires have not been typical of recent GWW fires.

Acknowledgements

This study was supported by the Department of Environment and Conservation's (DEC) Great Western Woodlands Conservation Strategy and the Australian Supersite Network, part of the Australian Government's Terrestrial Ecosystems Research Network (www.tern.gov.au), a research infrastructure facility established under the National Collaborative Research Infrastructure Strategy and Education Infrastructure Fund – Super Science Initiative – through the Department of Industry, Innovation, Science, Research and Tertiary Education. The spatial distribution of sampling was based in part on remote sensing data provided by Glen Daniel (Fire Management Services, Regional Services Division, DEC). We thank Georg Wiehl for field assistance and Lachie McCaw, Margarita Arianoutsou and an anonymous reviewer for helpful comments.

Ancillary