Management legacies shape decadal-scale responses of plant diversity to experimental disturbance regimes in fragmented grassy woodlands

Authors


Correspondence author. E-mail: suzanne.prober@csiro.au

Summary

  1. Understanding multiple ecological determinants of plant diversity and composition underpins good vegetation management. In mesic ecosystems, ecological theory and empirical data predict that moderate to high disturbance promotes native plant diversity, but relationships between disturbance and other drivers of diversity are poorly understood.
  2. We examined local determinants of native plant diversity and composition through 12-year fire, mowing and grazing experiments in two mesic grassy woodland remnants in fragmented agricultural landscapes of south-eastern Australia. Remnants were representative of diverse woodlands, but had contrasting management histories. We hypothesized that (1) disturbance is a dominant driver of plant diversity and composition and (2) moderate to high disturbance promotes diversity independent of other drivers or disturbance agent.
  3. Contrary to our first hypothesis, rainfall explained an overwhelming 31–60% of variation in native plant richness and native forb cover (hereafter diversity), while aspects of disturbance regime accounted for a significant but moderate 3–8%. The magnitude of disturbance effects was often rainfall dependent, and similar to that of other local determinants such as spatial heterogeneity. Disturbance also influenced native forb composition at one site.
  4. The direction of disturbance effects on diversity did not depend on disturbance agent, but differed markedly between sites in accordance with their management history. At the cleared site with a history of frequent burning, diversity declined significantly with time since burning or mowing. By contrast, at the long-unburnt, wooded site, diversity declined with frequent fire and with lagomorph and macropod grazing.
  5. These contrasting responses indicate that moderate to high disturbance is not consistently beneficial for diversity in mesic grassy ecosystems, contradicting our second hypothesis. Instead, disturbance responses are dependent on interactions with other drivers, potentially including assemblage filtering associated with historical fire regime and increased understorey productivity associated with tree clearing.
  6. Synthesis and applications: The magnitude and direction of disturbance effects on diversity depend on interactions with other drivers, hence a single prescription for disturbance regime across different remnants of fragmented grassy ecosystems is not appropriate. Decadal-scale responses of diversity to disturbance may be shaped by historical disturbances, suggesting instead that management be guided by historical regime.

Introduction

A fundamental goal of management for biodiversity conservation is to maintain native species diversity and composition. At local scales, attention has typically been focused on disturbance regimes for managing plant diversity, which in turn has long been influenced by the ‘intermediate disturbance hypothesis’. This predicts that plant diversity (including species richness and evenness, Huston 1979) is maximized at moderate levels of disturbance, with a decline in intolerant species at high disturbance, and species losses due to competitive exclusion at low disturbance (Grime 1973; Huston 1979). An alternative hypothesis, the ‘initial floristic composition’ model, suggests that most plant species present during post-disturbance succession re-establish shortly after disturbance, and then differential growth rates and survivorship result in a gradual decline in richness with time (Egler 1954).

Tests of disturbance–diversity hypotheses have shown inconsistent results (Mackey & Currie 2001; Mittelbach et al. 2001). This has led to increasing recognition that effects of single drivers of diversity such as disturbance should not be viewed in isolation from other, potentially interacting drivers (Adler et al. 2011). For example, a number of analyses emphasize that responses of diversity to disturbance are dependent on interactions with ecosystem productivity or biomass (e.g. Al-Mufti et al. 1977; Grime 1979; Huston 1979, 2004). The importance of other drivers of diversity such as historical legacies, time-lags, species pools, disturbance agent and heterogeneity in space and time is also increasingly acknowledged (Zobel 1992; Noy-Meir 1995; Grace 1999; Adler et al. 2011).

Disturbance and its interactions with other potential drivers of diversity are particularly relevant to ecosystems dominated by herbaceous species (hereafter grassy ecosystems), where fire, grazing and mowing are widely used to manage diversity (Collins & Gibson 1990; Belsky 1992; Collins et al. 1998; Lunt, Prober & Morgan 2012). In south-eastern Australia, temperate lowland grasslands and grassy woodlands dominated by the perennial tussock grasses Themeda australis (R. Br.) Stapf (syn. T. triandra Forssk.) and/or Poa sieberiana Spreng. (hereafter Themeda and Poa) were once widespread and common in the more mesic regions of the wheat–sheep belt (>550 mm, Prober & Thiele 2005). Owing to their occurrence on prime agricultural land, they have been extensively cleared or substantially modified by livestock grazing and nutrient enrichment (Prober & Thiele 2005; McIntyre & Lavorel 2007). This has resulted in widespread replacement of the natural dominant grasses by other exotic and native grasses and decline of a wide range of grazing-sensitive forbs (Prober & Thiele 2005; Dorrough & Scroggie 2008). Less than 0·5% of diverse lowland grassy ecosystems still dominated by Themeda and Poa are estimated to remain, hence effective management is critical for the ongoing persistence of these communities and their component, often regionally declining, plant species (Prober & Thiele 2005).

Previous studies of patch-scale plant diversity and composition in remnant Themeda–Poa ecosystems have focused on a single or a small number of potential drivers. Prober, Lunt & Thiele (2002) showed higher species richness in sites of lower natural fertility, and beneath trees compared with that observed in gaps. Other studies have focused on disturbance regime using opportunistic surveys (Lunt & Morgan 2002; Prober & Thiele 2005). As found in grasslands elsewhere (Knapp & Seastedt 1986), lack of disturbance can lead to excessive biomass accumulation, suppressing growth and reproduction of subsidiary species (Stuwe & Parsons 1977; Morgan 1999). Morgan (1999) concluded that plant diversity is maximized under annual burning in productive Themeda grasslands, suggesting that dynamics are more consistent with the ‘initial floristic composition’ model (Egler 1954) than the intermediate disturbance hypothesis (Grime 1973). However, we are aware of no published studies in these ecosystems that have experimentally manipulated fire and other disturbances to directly measure outcomes for plant diversity or evaluate disturbance responses in relation to other drivers (Lunt & Morgan 2002).

While disturbance–diversity hypotheses do not distinguish between different types of disturbance, the agent of disturbance has also been shown to influence outcomes for plant diversity and composition in grassy ecosystems (Belsky 1992; Noy-Meir 1995; Collins et al. 1998). In contrast to the observed positive effects of fire (a natural disturbance), the exogenous disturbance of set stocking with sheep or cattle typically eliminates Themeda–Poa swards and reduces native species richness (Prober & Thiele 2005), so is not a viable management option for diverse lowland Themeda–Poa ecosystems. Dynamics of Themeda–Poa ecosystems in relation to disturbance agents other than fire and livestock grazing, such as mowing, feral lagomorph or native macropod grazing, are poorly documented, but potentially offer alternative management options (Lunt 1991; Allcock & Hick 2003; Martin 2003; Meers & Adams 2003; Prober, Thiele & Lunt 2007).

To help understand how drivers of plant species diversity and composition impinge on disturbance management, we undertook a 12-year experimental study in Themeda–Poa ecosystems of south-eastern Australia. We applied replicated burning, mowing, and macropod and lagomorph grazing regimes to two representative remnants with contrasting management histories and monitored these through two drought cycles. Earlier papers from these experiments have reported detrimental effects of biennial burning on Themeda swards and soil physical characteristics in interaction with drought (Prober, Lunt & Thiele 2008), and shifts from Poa dominance at low fire frequency to dominance by the more fire-resilient Themeda at higher fire frequency (Prober, Thiele & Lunt 2007). In this study, we evaluate impacts on native plant richness, native forb cover and native plant species composition in the context of the following hypotheses:

  1. Disturbance is a dominant driver of local-scale native plant richness, forb cover and composition.
  2. Native richness and forb cover are favoured by moderate (Grime 1973) or high (Egler 1954; Morgan 1999) disturbance, independent of other drivers or disturbance agents.

Materials and methods

Experimental trials were established in October 1993 at two 4-ha remnants of white box Eucalyptus albens Benth. and yellow box E. melliodora A. Cunn. ex Schauer grassy woodland with diverse Themeda–Poa understoreys, in central New South Wales, Australia (Monteagle, 34°12′ S, 148°21′ E and Woodstock, 33°46′ S, 148°51′). The remnants occur in rural cemeteries on arable midslopes between cropping and grazing paddocks. Both have a history of minimal livestock grazing or fertilization; consequently, they represent some of the least-degraded ground layers of this ecological community (Prober & Thiele 2005).

Average annual rainfall at both sites is approximately 600 mm, with a slight winter maximum and interannual variability of up to 50%. Mean maximum January temperature is c. 31°C and mean minimum July temperature is c. 1°C. Soils are deep red clays overlain by moderately–slightly acidic loams, derived from Devonian granodiorites at Monteagle and at Ordovician volcanics and limestones at Woodstock (Prober, Thiele & Lunt 2007).

While both sites represent high-quality native woodland ground layers, their management regimes differed in the 50 years prior to 1993. Monteagle had a history of frequent burning (c. 4–8 year intervals, last burnt c. 1990–1, K. Butt pers. comm.) and is largely treeless due to historical clearing. Woodstock retains a discontinuous canopy of mature trees and according to local knowledge had not been burnt for at least 50 years prior to our experiment. Aboriginal and other management over longer timeframes is unknown.

Experimental design

Plots of 5 × 5 m (separated by 2–5 m) were established in a treeless area at Monteagle and a wooded area at Woodstock. Experimental treatments were selected to reflect practical management options. Three burning frequencies (two, four or eight yearly in May/June) and an undisturbed treatment were trialled at both sites. At Monteagle, biennial mowing (to 10–15 cm without slash removal) was also included, leading to a five disturbance treatment × four replicate randomized complete block experiment. At Woodstock, an exclosure treatment (excluding feral rabbits Oryctolagus cuniculus Linnaeus and hares Lepus capensis Linnaeus and native kangaroos Macropus giganteus Shaw) was included in factorial combination with burning. This led to a four burning frequency (whole plot) × 2 fencing (subplot) × 4 replicate trial in a split-plot design with whole plots (10 × 5 m) in four blocks.

Swards at Monteagle burnt well, leaving minimal live vegetation and sparse litter. At Woodstock, burns were milder due to damp eucalypt litter, killing top-growth but leaving some litter unburnt. Burns were unsuccessful at Woodstock in 2004, thus in 2005, ‘2-year’ plots remained unburnt for a third year. Data from 2003 indicate vertebrate herbivores removed approximately one-third of standing biomass in grazed plots (Prober, Thiele & Lunt 2007).

Monitoring

Abundance of all vascular plant species in each plot was estimated using a point-intercept technique (Prober, Thiele & Lunt 2007), in October 1993 prior to treatments and every 1–3 years during October/November until 2005 (Table 1). An 8-mm dowel was placed vertically at each of 50 points spread regularly across each plot; abundance (‘points’) for each species was the number of points at which they intercepted the dowel or its projected extension to the canopy. Species present but not intercepting the dowel were allocated a score of 0·5. This technique provided an objective measure of abundance reflecting but not equivalent to projective cover (hereafter referred to as ‘cover’). Total richness, native richness and cumulative cover of native species other than the two dominant grasses (hereafter forb cover) were calculated from raw floristic data. In 2003, we spent longer preparing the species list at each plot (for other purposes), hence richness values were inflated. We accounted for this using a binary ‘sampling intensity’ variable in data analyses. Nomenclature follows NSW Flora Online (2012).

Table 1. Disturbance treatments and years in which monitoring took place (bold text) for Monteagle (cleared site) and Woodstock (wooded site)
 Monteagle and WoodstockMonteagleWoodstock
FencingUnfencedUnfencedUnfencedUnfencedUnfencedFencedFencedFencedFenced
Fire/mowBurntBurntBurntBurntBurntBurnt
Interval (year)Untreated248MownUntreated248
  1. a

    Burning was not successful at Woodstock.

1994  BurntBurntBurntMown BurntBurntBurnt
1995          
1996  Burnt  Mown Burnt  
1997         
1998  BurntBurnt Mown BurntBurnt 
1999          
2000 Burnt  Mown Burnt  
2001         
2002  BurntBurntBurntMown BurntBurntBurnt
2003          
2004 Burnta  Mown    
2005          

Monthly rainfall data were obtained from the nearest available weather stations (5 km from Monteagle and 21 km from Woodstock). May to October rainfall (reflecting the growing season prior to measurements) was used in analyses.

Data analysis

To investigate drivers of total richness, native richness and forb cover, we fitted linear mixed models employing restricted maximum likelihood estimation (REML, genstat 13.0, VSN International, Hemel Hempstead, UK). Cover data were transformed using logarithmic or square-root transformations as required to meet the assumptions of normality. Random models included terms for blocks, plots and years within plots, allowing for a different variance between plots for each year and covariance between years for each plot (power model). The random model was simplified where appropriate by removing terms that were not significant based on likelihood ratio tests.

Initial fixed models included, in order of addition, tree canopy cover (Woodstock only), pre-treatment (1993) values for the variable being analysed, potential determinants of interannual variation (sampling intensity, rainfall, rainfall2 and year) and a breakdown of disturbance regime into different components. The latter included terms for disturbance type (burnt vs. mown at Monteagle; burnt vs. unburnt and fenced vs. unfenced at Woodstock), and linear and quadratic terms for time since fire, number of fires from 1993 to the time of measurement (burnt plots only), time since mowing and number of mows from 1993 to the time of measurement (mown plots only). For Monteagle, untreated plots were classed as burnt because the whole site was burnt in c. 1990–1; at Woodstock, a distinction between burnt and unburnt plots was necessary in the initial model because unburnt plots could not be scored for time since fire. The variables ‘time since fire’ and ‘number of fires’ offered the best breakdown of burning treatments we could devise to minimize correlations between these explanatory variables (R2 = 0·41 at Woodstock, R2 = 0·30 at Monteagle). Time since fire was added prior to number of fires because effects of the former were expected to override the latter. Interactions between rainfall and disturbance variables, rainfall2 and disturbance variables, and fencing and burning (Woodstock) were included after main effects. To capture any residual treatment effects, fire and mowing treatment terms were included at the end of the models (but were not significant).

Once the fixed model had been built using all terms, nonsignificant fixed terms (P > 0·05) were iteratively removed on the basis of Wald statistics and F or chi-square tests while maintaining the marginality principle (i.e. retaining main effects for all variables included in interactions). To approximate the amount of variation explained by each term remaining in final models, we fitted multiple regression models with terms for block plus all remaining fixed terms in the final linear mixed model and calculated the percentage of the total sum of squares contributed by each term. To indicate the effects of key variables graphically, we removed the effects of noninteracting variables from the dependent variable (adjusted data) and then fitted a reduced linear mixed model containing the variable of interest and interacting variables. The fitted model was plotted with the adjusted data. Because total richness and native richness were highly correlated (r = 0·88 for Monteagle and r = 0·94 for Woodstock), terms remaining in their respective models were identical; thus, we present results only for native richness.

Native richness and forb cover were also analysed by anova with repeated measures and the Greenhouse–Geisser epsilon adjustment using genstat. This included initial data as a covariate, and Fischer's protected least significant differences to compare treatment means. These gave similar but usually less informative results, so are referred to only where they assisted interpretation (Figs S1 and S2 in Supporting Information).

Ordination and permutational analysis of variance (permanova) of 2003 native forb composition were used to indicate decadal-scale outcomes of disturbance regime. As 2-, 4- and 8-year plots were all burnt in 2002, 2003 data reflect the influence of fire frequency on burnt plots without the complication of differing time since fire. Quantitative data for all native species other than Themeda and Poa (which were strongly influenced by fire, Prober, Thiele & Lunt 2007) were square-root-transformed and used to produce a distance matrix using the Bray–Curtis coefficient of dissimilarity. Nonmetric multidimensional scaling (nMDS) was performed on the distance matrix using DECODA (Minchin 1989). Analyses were performed in one to five dimensions using 10 random starts; and three dimensional solutions were chosen. Vectors of maximum correlation (Rmax) of variables with the ordination were calculated using the vector-fitting procedure of DECODA (Minchin 1989) and used to order sites and species to produce a two-way table indicating species contributing to treatment effects.

permanova+ for PRIMER-E (Anderson, Gorley & Clarke 2008) was used to test the significance of differences between treatments in native forb composition, using 999 permutations and unrestricted permutation of raw data (preferable for small data sets, Anderson & Robinson 2001). To achieve the split-plot analysis for Woodstock, we conducted the analysis in two parts. To test for effects of fire treatments, centroids were calculated for pairs within whole plots, and block (random) and fire (fixed) main effects were tested. A second permanova was performed on subplots to obtain the significance of fencing and fencing × fire terms.

Results

We recorded 58 native plant species in Monteagle plots and 70 in Woodstock plots. These were typically herbaceous perennials (especially Asteraceae and Liliaceae sens. lat.), with occasional annuals and few shrubs.

The best-fitting linear mixed models for native richness and forb cover explained high levels of variation at both sites (64–82%, Table 2) and involved multiple interacting factors. Growing-season rainfall was overwhelmingly the strongest determinant, explaining an estimated 31–60% of the variation over 12 years. Relationships with rainfall were typically quadratic, with natives declining most rapidly when growing-season rainfall decreased below 300 mm (Figs 1 and 2). At Monteagle, there was also an unexpected trend of decreasing native richness and forb cover as growing-season rainfall increased above c. 400 mm (Fig. 1a,b).

Table 2. Significance of linear and square terms in the final linear mixed models for native richness and forb cover, for (a) Monteagle and (b) Woodstock. Terms that were not significant in either model at a site (see methods) are not shown. %SS (% of sum of squares explained) is approximated from multiple regression analyses (see methods). For Monteagle, years since burning and number of burns apply to untreated, 2-, 4- and 8-year plots, whereas at Woodstock, they apply to 2-, 4- and 8-year plots only. Time since fire = number of spring seasons since fire, trt = treatment, na = not tested, dash = > 0·05
 Native richnessa % SSLinear PSquare PNative forb cover % SSLinear PSquare P
  1. a

    Adjusted for higher sampling intensity in 2003.

  2. b

    Square term applies to fire variates.

(a) Monteagle
Block (spatial heterogeneity)10·6nana9·5nana
Initial values8·0<0·001na1·10·046na
Growing-season rainfall30·8<0·001<0·00143·7<0·001<0·001
Year (other interannual variation)7·4<0·001na10·4<0·001na
Mown/unmown0·20·325na2·30·068na
Time since fire (unmown plots)6·0<0·0030·8414·0<0·0010·032
Rainfall × time since fire1·50·2000·068b0·60·2720·006b
Total disturbance-related effects7·7  6·9  
Disturbance-related effects after main effects of rainfall excluded from total SS(11·0)  (12·2)  
Total 64·5    71·7   
(b) Woodstock
Block (spatial heterogeneity)3·2nana3·1nana
Tree coverna1·70·002na
Initial values5·20·009na3·5<0·001na
Growing-season rainfall57·7<0·001<0·00160·0<0·001<0·001
Year (other interannual variation)3·8<0·001na8·5<0·001na
Burnt/unburnt0·00·591na3·00·004na
Burnt/unburnt × rainfall0·40·044
Time since fire (burnt plots)0·80·0790·007
Number of fires (burnt plots)1·9<0·0011·00·0030·012
Fencing0·00·7200·80·060
Rainfall × fencing0·40·041
Total disturbance-related effects2·7  5·7  
Disturbance-related effects after main effects of rainfall excluded from total SS(6·4)  (14·2)  
Total 72·6    82·4   
Figure 1.

Estimated effects of growing season rainfall at differing times (spring seasons) since fire (a, b), and effects of time since fire at average growing season rainfall (380 mm; c, d) on floristic variables at Monteagle. Grey circles are adjusted data points. All models are adjusted for initial values; time since fire models are adjusted for rainfall terms, and exclude mown plots. The year term was excluded from simplified models.

Figure 2.

Estimated effects of growing season rainfall on burnt vs unburnt and fenced vs unfenced plots (a, b), and effects of number of fires/time (spring seasons) since fire at average growing season rainfall (335 mm; c, d) on floristic variables at Woodstock. Green circles or crosses refer to mean for untreated plots, grey circles or crosses are adjusted data points. All models are adjusted for initial values and tree cover if significant (see Table 2). Number of fires/years since fire models are adjusted for rainfall terms. The year term was excluded from simplified models.

A range of other variables explained significant but smaller amounts of variation in native richness and forb cover (Table 2). These included disturbance-related effects (summing to 3–8%), unexplained interannual variation (4–10%), spatial heterogeneity (block, 3–11%), positive effects of initial richness or cover (1–8%) and positive effects of trees on native forb cover at Woodstock (2%). There were also significant interactions between growing-season rainfall and aspects of fire regime at both sites, suggesting weaker effects of fire when rainfall was high (Table 2a,b; Figs 1 and 2).

Fire and other disturbance

The direction and nature of fire effects differed markedly between sites. At Monteagle, effects were typically positive, with native richness and forb cover declining quadratically with increasing time since fire (Table 2a, Fig. 1a–d). Models suggest that at average rainfall, this corresponded to a decline of about 0·5, one and four species per plot at 4, 8 and 15 years after fire, respectively (Fig. 1c). In years of high growing-season rainfall (c. 500 mm), time since fire effects were weaker, due to the decline in forb richness on burnt plots described above (Fig. 1a,b). Fire frequency effects were not detectable.

At Woodstock, natives were favoured by nil to low levels of disturbance and were influenced more by number of fires than time since fire (Table 2b, Fig. 2a–d). Patterns in native richness suggested a slight hump-shaped model at average rainfall, peaking at around the first burn and declining steadily by about one species per additional fire (Fig. 2c). The hump shape was supported statistically only in interaction with rainfall (after re-analysis to include untreated plots in the number of fires term; rainfall × number of fires2 = 0·014). Native forb cover declined substantially after the first fire and at average rainfall required about 3 years to recover to levels on unburnt plots (Fig. 2d). Models suggested this was countered by a trend for additional fires to promote forb cover (Fig. 2d), but the net outcome was significantly higher forb cover on unburnt than frequently burnt plots in most years, especially in the final year of measurement (Fig. S2b, Supporting information).

Disturbances other than fire affected native richness and cover in similar directions to fire. At Monteagle, biennial mowing enhanced natives compared with untreated plots, but usually to lesser extents than biennial burning (Fig. 1a,b). At Woodstock, forb richness and cover tended to be lower in grazed plots at low rainfall (Fig. 2a,b; Table 2), although effects for cover were only marginally significant (= 0·06).

Floristic composition

Native forb composition was most strongly correlated with spatial location (block) at both sites. At Woodstock, tree cover influenced composition, but effects of burning were only marginally significant (= 0·077), and no effects of fencing were evident (Table 3, Fig. 3). At Monteagle, burnt plots differed significantly from unburnt plots in native forb composition, although no effects of fire frequency were detectable. Differences included more graminoids (Carex spp. and Luzula meridionalis) in unburnt plots and greater cover or frequency of a wider range of species in burnt plots, particularly the annual forbs Sebaea ovata and Crassula sieberiana, and the perennial forbs Cynoglossum suaveolens, Wurmbea dioica, Drosera peltata and Tricoryne elatior (Table S1, Supporting information). Cumulative cover of Liliaceae sens. lat. increased in a direction consistent with the change from unburnt to burnt plots (Rmax = 0·67, = 0·016). Mown and unburnt plots were poorly distinguished.

Table 3. Vector correlations (Rmax) of treatment and environmental variables with ordinations of 2003 native forb composition
VariableMonteagle RmaxWoodstock Rmax
  1. na, not applicable.

  2. a

    < 0·001.

Block0·88***0·79***
Tree coverna0·80***
Burnt/unburnt0·87***0·38
Number of burns0·72***0·39
Fencingna0·32
Figure 3.

Distribution of fire and mowing treatments on three-dimensional ordinations of native forb composition. Axes were selected to best show treatment patterns; ***P < 0.001, **P < 0.01, ^P = 0.077, ns not significant for block and treatment main effects and interactions, derived from permanova. Different superscripts indicate significant differences (P < 0.05) between treatments. Note that analysis of initial (1993) composition confirmed no initial significant differences between treatments at either site.

Discussion

Contrary to our first hypothesis, disturbance was not the dominant driver of local-scale native richness and forb cover in our study. Instead, native richness and forb cover were influenced by multiple interacting factors, consistent with Adler et al. (2011). Interannual variability in growing-season rainfall was overwhelmingly the strongest determinant, similar to other studies of grassland dynamics in Australia, Africa and North America (Collins & Gibson 1990; Allcock & Hick 2003; Hobbs, Yates & Mooney 2007; Savadogo et al. 2008; Buitenwerf, Swemmer & Peel 2011). The contribution of disturbance was significant but moderate, comparable with that found for African grasslands (Fynn, Morris & Edwards 2004; Uys, Bond & Everson 2004; Savadogo et al. 2008) and similar to contributions from initial richness and cover (probably reflecting species pools and/or ecological inertia) and local-scale spatial variation.

Although effects of interannual rainfall variability were the most prominent, its effects are likely to be apparent or temporary, largely driven by dormancy of below-ground organs or limited growth in dry years (Hobbs, Yates & Mooney 2007). This raises the relative importance of drivers that are more amenable to management, in particular, the contribution of fire and other experimental disturbances rose from 3–8% to 6–14% after accounting for rainfall (Table 2). Regardless of other drivers, fire regime led to changes in native richness of sufficient magnitude to warrant management attention (e.g. at average rainfall, loss of four species per plot by 15 years after fire at Monteagle).

Are native richness and forb cover favoured by moderate or high disturbance?

Neither the intermediate disturbance hypothesis (Grime 1973) nor the initial floristic composition model (Egler 1954; Morgan 1999; Lunt & Morgan 2002) consistently predicted effects of disturbance regime on native richness or forb cover. Instead, responses were site dependent. This concurs with reviews indicating a wide range of diversity–disturbance responses (Mackey & Currie 2001; Mittelbach et al. 2001) and supports a multivariate approach to understanding diversity (Adler et al. 2011).

At Monteagle, the gradual decline in native richness and forb cover with time since fire most strongly supported the initial floristic composition model (Egler 1954), and hence predictions of Morgan (1999) and Lunt & Morgan (2002) that native richness is maximized under the most frequent fire regime. Biennial fires similarly promoted the highest richness and cover of soil crust flora at Monteagle (O'Bryan et al. 2009). However, significant mortality of Themeda tussocks occurred in 2-year burn plots (in interaction with drought, Prober, Thiele & Lunt 2007), and forbs had not declined substantially by 6 years after fire (at average rainfall), supporting a moderate fire interval at this site (Prober, Thiele & Lunt 2007).

At Woodstock by contrast, nil to low disturbance was more favourable for native richness and forb cover, inconsistent with hypotheses suggesting frequent or intermediate disturbance is preferable (Egler 1954; Grime 1973; Morgan 1999). Slow recovery of native forb cover after fire concurred with the response of the dominant grass Poa sieberiana, which also took at least three seasons to recover (Prober, Thiele & Lunt 2007). The dominant sward showed capacity to adapt to frequent fire through shifting dominance towards the more fire-resilient Themeda (Prober, Thiele & Lunt 2007). The positive relationship between forb cover and number of fires may similarly be indicative of an adaptive response, but forb cover was still lowest on 2-year burn plots by the final year of measurement (Fig. S2b, Supporting information), probably due to the dense Themeda sward that had developed. Similarly, forb richness had declined substantially after five fires, suggesting a lack of adaptation. Longer term data are needed to confirm these trends, but they are consistent with Collins & Gibson (1990) and Fynn, Morris & Edwards (2004), who showed that annual burning promotes increased dominance by matrix forming grasses and reduces richness in North American prairie and African grasslands.

We propose the contrasting responses at Monteagle and Woodstock were due to their differing management legacies. At Monteagle, the favourable response to disturbance concurred with its history of 4–8 year burning in the previous half-century. At Woodstock, the predominantly negative responses to disturbance concurred with absence of fire over this period. This is consistent with the concept of assemblage filtering (Keddy 1992), whereby historical regimes selected for different suites of species at each site. Replacement by species favoured under novel regimes may have subsequently been constrained by time-lags and limited propagule supply associated with isolated remnants within an agricultural matrix. This contrasts with Fynn, Morris & Edwards (2004, 2005), who found that fire tolerant forbs were replaced by other species with protection from disturbance in grasslands of southern Africa, leading to no change in total richness.

Historical tree clearing at Monteagle is also likely to influence ground-layer response to disturbance, particularly through effects on ground-layer productivity (Schultz, Morgan & Lunt 2011). Total ground-layer biomass recorded in 2003 on unburnt plots at the wooded site (Woodstock, 314 g m−2) was about half that recorded at the cleared site (Monteagle, 611 g m−2, Prober, Thiele & Lunt 2007). Hence, lower resilience to and need for disturbance at Woodstock is consistent with ecological theory proposing higher optimal disturbance levels for maintaining diversity in ecosystems with higher annual net primary productivity or total biomass (e.g. Al-Mufti et al. 1977; Grime 1979; Huston 1979, 2004; Schultz, Morgan & Lunt 2011; Lunt, Prober & Morgan 2012). Given the potential role of trees in mediating fire response at Woodstock, the role of trees in promoting native plant diversity (Prober, Lunt & Thiele 2002) and widespread tree regeneration in many remnant eucalypt woodlands (Lunt et al. 2010), interactions among trees, disturbance and diversity are worthy of further attention.

Influence of rainfall and disturbance type on disturbance responses

While the direction of disturbance effects at each site was largely consistent across different levels of growing-season rainfall, their strength was often dependent on rainfall. These effects indicate probable early implications of a drying climate for disturbance regimes, acknowledging that more complex interactions are likely under climate change than with interannual variation. Disturbance × rainfall interactions were relatively weak at Woodstock, but indicate slower recovery of native richness after fire in low rainfall years, consistent with lower disturbance needs at lower productivity (Huston 1979, 2004) and suggesting increasing vulnerability to fire in a drying climate (consistent with Lunt, Prober & Morgan 2012). At Monteagle, the most prominent interaction was, unexpectedly, a reduction in the positive impacts of disturbance at high rainfall. This may reflect interactions with the response of exotic annuals, which increased rapidly at high rainfall and may have suppressed forbs in high rainfall years (S. Prober unpubl. data). In this case, forb response to mild climate change may be tempered by impacts on exotics.

Finally, our hypothesis that responses of native richness and forb cover to disturbance are independent of disturbance agent was largely supported, with herbivores and mowing leading to effects similar (although lower in magnitude) to those of fire. This contrasts with other studies that have suggested differing effects of grazing compared with fire on these factors (Belsky 1992; Noy-Meir 1995; Collins et al. 1998) and support the argument that disturbance effects on diversity are driven through their impacts on biomass (Huston 1979; Lunt, Prober & Morgan 2012). Notwithstanding, the floristic composition of mown plots was more similar to unburnt than burnt plots, suggesting different species are contributing to effects. Further, in Themeda–Poa ecosystems, the dominant sward is quickly replaced by other grasses and a suite of native forbs is rapidly lost with introduction of novel (hoofed) grazers (Prober 1996; Dorrough & Scroggie 2008), hence positive effects of fire at Monteagle are unlikely to be mirrored by all disturbance types.

Management implications

Our study showed that moderate to high disturbance is not consistently beneficial for maximizing native richness and forb cover in grassy ecosystems, and hence that a single prescription for disturbance regime across different remnants is not appropriate. Instead, disturbance responses are dependent on interactions with other factors, which are likely to include the medium-term disturbance history. The latter supports a ‘status-quo’ approach to management, that is, maintenance of historical regimes as the lowest-risk management option to maintain existing floristic values in forb-rich ecosystems in fragmented landscapes. Specifically, this suggests continuing approximately 4–8-year fire intervals at Monteagle (with consideration for interactions with rainfall), and minimal fire combined with control of feral herbivores at Woodstock.

Notwithstanding, it may be acceptable to substitute fire with disturbance agents such as mowing if necessary. Furthermore, in a drying climate or as trees re-establish, historical regimes may need to be revised to account for lower resilience to fire at resultant lower levels of ground-layer productivity (Lunt, Prober & Morgan 2012).

Acknowledgements

This study was supported by the Australian government through the Grassy Box Woodlands Conservation Management Network, the Australian Research Council, the National Heritage Trust and Caring for our Country and the NSW government through its Environmental Trust. The Monteagle and Woodstock Bush Fire Brigades skilfully managed fires, and Laurie Adams and Chris Puttock helped build exclosures. Particular thanks go to the Butt family (Fairfields) and Hugh Jackson (Young Shire Council) for ongoing assistance at Monteagle; and Paul Bennett, Cowra Shire Council and Woods Flat Creek Landcare Group for support at Woodstock. Don Cram provided rainfall data for Monteagle and Marti Anderson and Bob Clarke assisted with split-plot Permanovas.

Ancillary