Propagule pressure, not fire or cattle grazing, promotes invasion of buffel grass Cenchrus ciliaris

Authors


Correspondence author. E-mail: rod.fensham@derm.qld.gov.au

Summary

  1. The invasion of exotic grasses into savannas is occurring globally. Despite threats to ecosystem processes and biodiversity, few studies have investigated factors that facilitate invasion success.
  2. The abundance of the exotic perennial Cenchrus ciliaris buffel grass was assessed across woodland clearings at 46 sites in Queensland, Australia. The influence of fire and grazing on the invasion success were also examined experimentally within 18 sampling stations (including cattle grazed and ungrazed treatment pairs) with unburnt, low-intensity and high-intensity fire treatments.
  3. Mixed-effects logistic regressions suggest that the probability of buffel grass occurring in remnants was predominantly a function of the abundance of buffel grass in adjoining paddocks. The abundance of buffel grass in remnants was greater where the boundaries with clearings faced the prevailing wind adjacent to paddocks with high buffel grass abundance.
  4. There was a significant positive association between buffel grass and tree canopies at approximately half of the sites, and the strength of this association was positively related to the amount of coarse sand in the surface soil. This association was interpreted as nutrient enrichment from litter fall on sandy infertile soils being a requirement for invasion. The invasion of buffel grass at the experimental site accelerated with above-average rainfall after a prolonged drought, and invasion continued even after imposing all combinations of fire and grazing.
  5. Cattle grazing modestly enhanced invasion, relative to the absence of grazing, but this difference was only significant without burning. The occurrence of fire in both the survey and experiment was not associated with enhanced buffel grass invasion.
  6. Invasion of buffel grass may be inevitable over large areas of savanna, especially when abundant rainfall follows periods of drought. However, invasion will be minimized in areas where buffel grass has not been planted, areas without cattle grazing and in areas with low soil fertility.
  7. Synthesis and applications. The invasion of buffel grass accelerates during abundant rainfall after drought and is enhanced by propagule pressure, dispersal trajectories and under trees, probably as a result of nutrient enrichment. Disturbance had little influence on invasion, and there is no support for fire-promoted invasion as predicted by the grass–fire cycle theory.

Introduction

The invasion of exotic grasses into natural ecosystems has the potential to alter fire regimes (D'Antonio & Vitousek 1992; Miller et al. 2010; Setterfield et al. 2010), displace native species (Fairfax & Fensham 2000; Jackson 2005) and disrupt nutrient cycles (Kirschbaum et al. 2008; Rossiter-Rachor et al. 2009). Despite a relatively advanced understanding of the impacts of exotic grasses, there is only a preliminary understanding of the processes that affect invasion success (Setterfield et al. 2005; Eschtruth & Battles 2009). This is a major knowledge gap impeding appropriate management actions to limit their spread.

It is well established that many exotic species are favoured by disturbance which creates niches for seedling recruitment (Seabloom et al. 2003). Although direct experimental evidence is lacking, the invasion of exotic perennial grasses in particular is thought to be promoted by fire (Brooks et al. 2004; Miller et al. 2010; Setterfield et al. 2010), assisted by a positive feedback that commences with persistence and spread after a fire event through competitive displacement of native species (D'Antonio & Vitousek 1992). Once established, high-biomass grasses have the capacity to alter fire regimes (Rossiter et al. 2003), further enhancing the invasion success of the grasses and resulting in increased grass biomass and fire intensity (Rossiter et al. 2003). The ‘grass–fire cycle’ often coincides with a decline of trees and shrubs (Brooks, 2004) and suggests that increases in light and release from other competitive influences of trees favours the spread of exotic grasses (Setterfield et al. 2005).

Grazing is another disturbance that can promote exotic species (Bock et al. 2007; Tozer et al. 2008), particularly in continents such as Australia that are assumed to be less resilient to managed grazing with domestic livestock given the relatively short and light exposure to grazing over evolutionary time frames (Milchunas et al. 1988). Other factors also influencing invasion success include the susceptibility of the site, the invasion potential of the species, the density of propagules, disturbance regimes and the capacity of the site to recover from disturbance (Lonsdale 1999). An integrated landscape-scale and experimental approach would allow the relative strengths of the factors relating to the susceptibility of tropical environments to invasive perennial grasses to be tested.

In the vast savanna biome of tropical latitudes, perennial exotic grasses are becoming a major problem (D'Antonio & Vitousek 1992; Daehler & Carino 1998; Franklin et al. 2006; Smyth et al. 2009) with invasion often facilitated by deliberate establishment for pasture development in conjunction with land clearing. There are very few empirical studies examining the conditions affecting disturbance success of invasive perennial grasses in tropical ecosystems (Setterfield et al. 2005).

Buffel grass Cenchrus ciliaris is native to southern Asia and eastern Africa and has been introduced to North America and Australia to improve pasture production (Williams and Baruch 2000). Buffel grass has a plumed seed surrounded by a fascicle of bristles suggesting wind dispersal and requires relatively small amounts of rainfall to germinate, with 50% emergence occurring with 17·4 to 19·9 mm of water under 25 °C (Ward, 2006). Buffel grass exhibits a rapid growth response to small rainfall events and is tolerant to drought and grazing pressure compared to native grass species (Hodgkinson et al. 1989; Williams & Baruch 2000).

Vast areas of north-eastern Australia have been converted to productive cattle pastures through clearing of native vegetation and sowing of buffel grass (Lawson et al. 2004; Bortolussi et al. 2005). Buffel grass is also invasive and has spread into natural areas in regions adjacent to developed pastures (Clarke et al. 2005; Eyre et al. 2009), as well as in regions that have never been cleared, such as the relatively arid reaches of central Australia (Lawson et al. 2004; Clarke et al. 2005).

For the invasive buffel grass, site characteristics are important determinants of invasion success, and spread occurs where summer rainfall is between 150 mm and 550 mm, winter rainfall is less than 400 mm, mean winter temperatures are rarely below 5 °C and soil texture is loamy (Cox et al. 1988). The spread of buffel grass is most likely to occur where site conditions are similar to the natural environment of the seed source (Ibarraf-F et al. 1995), and as with other tropical grasses, such as gamba grass Andropogon gayanensis (Setterfield et al. 2005), buffel grass (Butler & Fairfax 2003) can be inhibited by shade.

Propagule pressure may reflect the size of source populations, the remoteness from source populations and the factors influencing dispersal (Richardson & Pysek 2006; Eschtruth and Battles 2009). The amount of cleared habitat (where buffel grass is deliberately established) surrounding remnant patches of eucalypt woodland was the most important factor predicting buffel grass abundance in subtropical Eucalyptus savanna (Eyre et al. 2009). Further understanding of the factors impacting on propagule pressure could inform the most opportune circumstances to limit the deliberate planting of buffel grass, or to remove it in order to minimize its spread into the matrix of uncleared vegetation.

The current study was conducted in a landscape where buffel grass has been present for many decades, but where there are many areas that are not yet invaded. Buffel grass is deliberately established by bulldozing the eucalypt woodlands, burning the site and then seeding with buffel grass. This exposes adjacent areas of remnant woodland to buffel grass propagules.

This study uses a landscape-scale survey and a field experiment to address the following hypotheses: (1) buffel grass invasion will be more advanced as propagule pressure increases; (2) fire promotes the invasion of buffel grass; (3) cattle grazing promotes the invasion of buffel grass and (4) intercanopy areas will be more vulnerable to invasion than under tree canopies.

Materials and methods

Study area

The study was conducted in central Queensland (Fig. 1), a region characterized by annual rainfall averaging from 480 mm to 540 mm, with an average maximum temperature of 36 °C during summer and 8 °C during winter. The target vegetation is savanna woodland dominated by either Eucalyptus melanophloia silver-leaved ironbark or Eucalyptus populnea poplar box with an understorey of mixed native grasses (including Bothriochloa ewartiana, Chrysopogon fallax, Aristida jerichoensis, Triodia pungens and Themeda triandra), on flat to undulating plains. This is one of the most widespread vegetation types in north-eastern Australia encompassing 12·9 M ha (Wilson et al. 2002). The soils are well drained and nutrient-poor with sandy-loam A-horizons and variable sodicity at depth.

Figure 1.

The study area within Queensland representing remnant woodland (grey) and cleared areas (white). The points represent the study sites for the boundary survey and the cross marks the location of the field experiment in Corella Paddock.

Current land use is almost exclusively cattle grazing. In recent decades, clearing of the woody vegetation has been occurring at a rate of 1% per annum (Wilson et al. 2002), with the intention of increasing pasture production, but has been recently restricted by state-wide laws (Vegetation Management Act 1999). Burning to remove fallen timber and intentional sowing (aerial and hand sowing) of buffel grass is a ubiquitous practice that accompanies clearing in the region (Fairfax and Fensham 2000). Since the 1960s, buffel grass seed has also been deliberately spread from horseback or motorbikes in remnant savanna woodland, although the extent of this practice is unknown. Understorey fires in remnant woodland are moderately frequent (Russell-Smith et al. 2003; see also Results) and are fuel-limited during the long periods with relatively low rainfall.

Regional soil attributes of Eucalyptus woodland

To explore the relationship between sandiness and macronutrient content, a data base (SALI) of soil survey data was interrogated for a region (20·5˚–24·5˚S; 145·5˚–147·0˚E) centred on the study area (Fig. 1) using Eucalyptus woodland in the search field. Coarse sand content (0·02–0·2 mm) was determined by sieving and was correlated with total nitrogen (n = 59), and bicarbonate-extractable phosphorus (n = 61) determined using methods defined in Rayment and Higginson (1992). Relationships between these variables were explored using Spearman's rank correlation coefficient.

Boundary survey

With the use of satellite imagery (GIS), 46 accessible sites with a boundary between paddock and uncleared Eucalyptus woodland were selected (Fig. 2). There was a requirement that no two sites could occur in the same paddock, sites were more than 100 m from roads and the remnant vegetation was at least 500 m wide.

Figure 2.

Boundary between remnant vegetation and cleared paddock at a site for the boundary survey.

Surface soil (1–5 cm depth) samples were collected from each of the remnant vegetation transects by bulking three subsamples at the beginning, middle and end of transects. Particle size distributions were determined using laser diffraction (Mastersizer 2000, Malvern Instruments Ltd.), which is a cost-effective and reproducible technique (Arriaga et al. 2006), although not directly comparable to the traditional hydrometer and pipette method (Pieri et al. 2006; Eshel & Levy 2007). Soil samples were sieved (2 mm), and prior to measurement, samples were sonicated for one minute at 10 μm tip displacement to break up remaining aggregated particles. Absorption was maintained between 15–20% during particle size measurement. The output of continuous particle size distribution was segmented as clay (particles < 0·002 mm), silt (0·002–0·02 mm), fine sand (0·02–0·2 mm) and coarse sand (0·2–2 mm) and represented as percentages.

Buffel grass cover in the remnant patch was measured using 50-m line transects established 50, 100, 150, 200 and 250 m parallel to the remnant boundary (Fig. 3). The interception of buffel grass (represented by a convex polygon joining the extremities of the vegetative parts of the plant) was determined every metre along the tape (51 points). Tree canopy cover in remnant vegetation was assessed using the same method. The abundance of buffel grass in the paddocks was recorded using 1-m spaced intercepts along four 50-m line transects (Fig. 3). The proportion of buffel grass present along each transect was averaged to attain a single value approximating the buffel grass cover at each site. The deviation of the bearing perpendicular to the remnant paddock boundaries from the predominant wind direction (135˚) during April (when buffel grass seed production is assumed to be maximal; See Appendix S1, Supporting Information) was recorded. In practice most boundaries are orientated north-south or east-west and ‘boundary orientation’ was reduced to a binary index: facing wind (225–135˚), and not facing wind (135–225˚).

Figure 3.

The line transect layout for the boundary survey with line transects (dashed lines) laid parallel to the boundary between remnant vegetation and cleared paddock.

The occurrence of fire between 1997 and 2010 was determined for each remnant site using AVHRR 1-km2 scale fire maps (1997–2004) and MODIS 250-m2 scale fire maps (2004–2010) (http://www.firenorth.org.au/nafi2/). This was converted to a binary index of the presence (never more than one) or absence of fire (1997–2010), but was only included if the fire occurred after clearing of the adjacent paddock. Seventeen of the 46 sites were cleared before 1997 and may have been subject to other fires prior to that date.

Exploratory analysis of the boundary survey data set revealed that coarse sand alone satisfactorily represented the influence of other soil characteristics on buffel grass abundance in remnants, and this variable was used in the following analyses.

An information theoretical approach (Burnham & Anderson 2002) was adopted to test multiple a priori predictions relating explanatory variables to (1) the probability of occurrence of buffel grass in remnants and (2) buffel grass abundance in remnants (Table 1). For the first step, mixed-effects logistic regressions were used to model a binary response variable: 1 if some buffel grass was recorded along a given transect and 0 if no buffel grass was recorded. Second, we modelled abundance (proportion of cover) of buffel grass where it occurred, that is, we removed all transects from the data set where buffel grass was not recorded. The proportion response was logit-transformed and modelled using linear mixed-effects models. In all models, both logistic and linear, site was included as a random effect to account for clustered sampling within sites. Given that the observations (transects) at each site were located at regular intervals into the remnant, it is possible that spatial structure existed at the within-site level of these data. In other words, the within-site observations may not be statistically independent. To account for this possible dependency, an autoregressive (lag 1) serial structure was included in all linear mixed-effects models. For the logistic model, we assumed independence of within-site observations because a variable explicitly describing the distance from remnant edge was a very poor predictor. Two explanatory variables, age since clearing and average paddock buffel grass cover, were square-root transformed to ensure linearity of the modelled relationships for both responses. All linear models met the assumptions of normality and constant variance. Also, the random-site intercepts were approximately normally distributed in all mixed-effects models.

Table 1. A priori predictions for the enhancement of buffel grass invasion and the associated candidate models considered for the two response variables in the analysis (binary occurrence and logit-transformed proportion cover). Also shown are the corresponding AICc and Akaike weights for each response variable. Boldface values indicate the models with most support for each response
A priori predictionCandidate modelProbability of occurrenceAbundance
AICcAkaike WeightsAICcAkaike Weights
1. Independent of all measured factorsIntercept264·450·00478·620·001
2. Propagule supplyAverage paddock buffel grass 245·15 0·72 469·330·120
3. Time since introduction of propagule supplyTime since clearing261·710·00484·260·000
4. Distance from sourceDistance from the remnant boundary268·470·00479·510·000
5. Loamy more fertile soilSoil coarse sand266·740·00480·390·000
6. Longer exposure enhances propagule pressureAverage paddock buffel grass + time since clearing248·340·15476·050·004
7. Propagule supply is exaggerated with prevailing windAverage paddock buffel grass × boundary orientation249·410·09 465·84 0·700
8. Distance from the source is exaggerated with prevailing wind directionDistance from the remnant boundary × boundary orientation276·190·00479·140·000
9. Propagule pressure is exaggerated with decreased canopy coverRemnant canopy cover × average paddock buffel grass253·130·01472·440·030
10. Propagule pressure is exaggerated by fireBurnt since 1997 ×  average paddock buffel grass251·180·03469·020·140

For each response variable, 12 candidate models were considered (Table 1). We used the corrected Akaike Information Criterion (AICc, Hurvich and Tsai 1989) to assess model fit and subsequently calculated Akaike weights based on these AICc values. A null model (intercept only) was included as a candidate model for both logistic and abundance models of the survey data.

To explore the relationship between buffel grass occurrence and tree canopies, the strength of the association between buffel grass intercepts and tree canopy intercepts across the 254 points recorded on the five transects in the remnant vegetation was assessed using Chi-squared tests. After making the Chi-squared statistic negative when buffel grass was associated with intercanopies at a site and positive when buffel grass was associated with canopies, the statistic was plotted and correlated with the average coarse sand content at a site (average of five transects in remnant woodland).

Field experiment

The field experiment aimed to assess the influence of disturbance on the early stages of buffel grass invasion in the ecosystem targeted by the boundary survey. The study was conducted in Eucalyptus melanophloia woodland at Corella Paddock (Fig. 1), which has been grazed by cattle for at least 100 years prior to 2006. It is not known whether buffel grass seed has been deliberately spread in Corella Paddock, although this has certainly not occurred since 2004. Three 200-ha blocks were established for the following unreplicated burning treatments: low-intensity burn, high-intensity burn and no burn. Six sampling stations were established within each of three blocks and were located more than 500 m apart.

Surface soil samples were taken from 0–5 cm depth at five locations near the sampling stations. The replicates were amalgamated, mixed and then subsampled and analysed for soil particle size (clay, silt, fine sand, coarse sand) P, NO3 and NH4 content.

Over five years (2005–2011) and at three sample periods, the abundance of buffel grass was recorded in 20, 50 × 50 cm frames every five steps along two predetermined bearings (50 steps each) from a permanently marked location at each of the 18 grazed sampling stations. The frequency of buffel grass in the 20 frames will be referred to as the ‘frequency data’.

The high-intensity burn was implemented in September 2009, and the low-intensity burn in November 2009 (see Fig. S1, Supporting information). Prior to the burning treatments, exclosures (ungrazed plots) were established at each of the sampling stations varying in size from 81 m² to 170 m². Two weeks after the low-intensity burn, when buffel grass tussocks were identifiable in the exclosures, an equivalent area with similar densities of buffel grass plants was established adjacent to the exclosures to provide a grazed treatment. All plants were resprouting after the fire or, in the case of the unburnt area, were established tussocks that had predated the burning. Buffel grass plants were counted and mapped in all 36 plots (18 grazed and 18 ungrazed plots) in this initial November 2009 census and then finally in a March 2010 census. These count data relate to grazed, ungrazed and three burning treatment combinations and are distinct from the frequency data that were only collected outside the exclosures and over a broader area. The proportion of each plot covered by tree canopy was also mapped.

There was no or only occasional cattle grazing between 2006 and the application of the burning treatments. After the first burn, grazing in all treatments was at a rate of 1 adult bullock to 15 ha. There is a fence between the burnt and unburnt treatments, and the bullocks were managed separately in these paddocks. This was necessary because constant stocking across all treatments would not have resulted in even grazing across treatments because the cattle would have been attracted to the ‘fresh pick’ recovering on the burnt areas. Thus, the grazing rate is estimated as the average grazing between the date of the first burn in September 2009 and March 2010, but involved different densities and periods of exposure and rest within this period.

Parametric and nonparametric anova depending on the data distribution was employed to examine differences in soil attributes between burning treatments.

The effect of the burning treatment on the frequency of buffel grass (from 20 frames) was compared between the three burning treatments using Kruskal–Wallis tests. First, we tested for differences in buffel grass frequency prior to burning treatments, and none were detected. We then calculated the absolute change in frequency (Frequency after burning - Frequency before burning) for each of the 18 sampling stations and ran a second Kruskal–Wallis test assessing differences among burning treatments. Although these data were approximately normally distributed, we used nonparametric Kruskal–Wallis tests as a precautionary measure given the small sample sizes.

The response variable used for the count data was the change in the number of tussocks from the initial to the final (post-fire) measurements. This change was expressed as a response ratio; ln (n.tussockst0/n.tussockt1). Wilcoxon paired tests were used to compare the response ratios of paired grazed and ungrazed plots within each fire treatment.

Results

The soil survey data base revealed negative relationships between coarse sand and total nitrogen (< 0·01) and phosphorus (< 0·0001) in Eucalyptus woodlands.

In the remnants for the boundary survey, the range of transect-scale variables were coarse sand (mean: 26·8%; range: 5·2–50·5%), buffel grass cover (mean: 5·5%; range: 0·0–42·0%) and tree canopy (mean: 19·6%; range: 0·0–51·0%). Buffel grass was recorded in 66% of transects. The range of site-scale variables were average buffel grass in clearings (mean: 10·7%; range: 0·0–34·3%), age of clearing (mean: 14·9 years; range: 3·9–49·9 years), boundary orientation (43% boundaries face the wind, 57% face away from the wind) and incidence of fire in remnants (46% unburnt, remainder burnt once only in 13 years). The presence of buffel grass along a transect was strongly positively associated with the proportion of paddock buffel grass, and this simple model had the most support (75%) out of those considered (Table 1, Fig. 4).

Figure 4.

The probability of buffel grass occurring within a transect as a function of the average proportion of buffel grass in adjacent paddocks (from model 2 in Table 1). Points are means of 22 binned binary values, and bars are associated standard errors.

For the abundance of buffel grass (excluding transects where buffel grass was absent), the model with most support (84%) included paddock buffel, boundary orientation and their interaction (Table 1, Fig. 5). Buffel grass abundance increased with increasing paddock buffel grass cover, particularly where the remnant edge faced the prevailing wind. There was no support for the model that included the occurrence of fire to predict invasion of buffel grass.

Figure 5.

Buffel grass abundance (logit-transformed proportion cover) as a function of the average proportion of buffel grass in adjacent paddocks (from model 7 in Table 1). Separate slopes were modelled for sites facing and not facing the direction of the prevailing wind.

There was a significant association (P < 0·05) between the occurrence of buffel grass and tree canopies at 21 of the 44 sites where buffel grass was present in the remnant vegetation. The association was significant (P < 0·05) between buffel grass and intercanopies at only one site. The relationship between the strength of the association with tree canopies and coarse sand was positive and significant (R2 = 0·115, 0·01 < < 0·05).

Field experiment

There were no significant differences between any of the soil attributes between the treatment blocks prior to burning with the exception of NO3.N which was lower in the high-intensity burn block (Kruskal–Wallis χ2 = 6·43, d.f. = 2, = 0·040).

The monitoring of buffel grass abundance in Corella Paddock identified that buffel grass was in the early stages of invasion in 2004 (Fig. 6a). Prior to 2004, wet season rainfall had been less than average for every year since 1999. While buffel grass abundance was not measured between 2004 and 2008, observations suggest that it increased rapidly with above-average rainfall in the wet season of 2007–2008 (Fig. 6a and b).

Figure 6.

(a) A plot showing the average and standard error of the frequency of buffel grass from 20, 50 cm × 50 cm frames) within the three burning treatments (triangles, no burn; squares, low-intensity fire; circles, high-intensity fire). The timing of the burning between the second and third census is indicated. (b) Monthly rainfall (mm) with downward arrows indicating above-average annual rainfall, the equal signs indicating approximately average annual rainfall and the upward arrows indicating above-average annual rainfall. The average annual rainfall is 510 mm.

Buffel grass was present in all but one (grazed, low- intensity fire) of the 36 plots at the time of the initial measurement, with abundance ranging from one to 112 individual plants per plot. There was a tendency for plant numbers to increase most within grazed plots, with this effect being statistically significant in the no-burn fire treatment (Fig. 7). Plots within the low-intensity fire treatment showed the highest change in plant number for the ungrazed treatment (Fig. 7). The frequency data (frequency of buffel grass in 20 frames) supported these findings. The frequency data relate only to grazed areas and show similar trends to the count data within the plots. Buffel grass frequency tended to increase in all burning treatments but was most pronounced in the no-burn fire treatment (Fig. 6a); however, no significant differences in the absolute change in buffel grass frequency were detected among fire treatments (Kruskal–Wallis χ2 = 2·33, d.f. = 2, = 0·311).

Figure 7.

The response ratio of buffel counts to fire and grazing treatments. Positive values indicate an increase in the number of buffel grass tussocks relative to the baseline counts. Letters denote significantly different groups as determined from pairwise Wilcoxon rank sum tests run on each fire treatment separately. Letters compare within fire treatments, not among them.

Discussion

Buffel grass invasion in remnant woodland was enhanced by propagule pressure. This effect was apparent in relation to the increased likelihood of buffel grass occurring in remnants as the abundance of buffel grass increased in adjoining paddocks. The effects of dispersal were only moderately apparent with a slight increase when remnant boundaries face the prevailing wind. The age of clearing was not an important factor affecting the spread of buffel grass into remnants, suggesting that the amount of buffel grass in the clearing is not independent from clearing age.

Contrary to our second hypothesis, buffel grass was positively associated with tree canopies at more than half the sites. In other situations, tree canopies inhibit the spread of invasive perennial grasses (Setterfield et al. 2005), including buffel grass, (Butler & Fairfax 2003) presumably because of shade effects. The facilitation of buffel grass under the relatively diffuse-crowned eucalypts may be the result of the redistribution of nutrients via root uptake and leaf fall to surface soils as has been suggested in analogous semi-arid environments (Tiedeman & Klemmeds 1973).

Soils beneath tree crowns often have significantly higher concentrations of N, Ca, K and P, organic matter, reduced bulk density, higher microbial biomass, and increased water infiltration than open grassland soils (Callaway et al. 1991; Vetaas 1992; Belsky & Canham 1994; Tiessen et al. 2003; Riginos et al. 2009). The enrichment of soil and pasture productivity has also been demonstrated under Eucalyptus tree crowns where it is most pronounced on infertile soils (Jackson & Ash 1998, 2001). In our study, the association with tree crowns was greater on soils with high coarse sand content, and the regional analysis of soil attributes revealed a negative relationship between coarse sand and total N and total P content. This range of evidence is consistent with buffel grass having a stronger requirement for nutrient enhancement under tree crowns where soil fertility is especially low on relatively sandy infertile soils. However, the details of how trees enhance buffel grass invasion is equivocal. In an experimental situation, P, N and K have all been demonstrated to limit the growth of buffel grass on sandy soils (Christie 1975). In Eucalyptus savanna, tree canopies have substantially elevated N content and moderate differences in P content (Jackson and Ash 2001). It has been suggested that trees have a substantial influence on sandy soils through organic matter enrichment enhancing cation exchange capacity (Campbell et al. 1994). Comparison of savannas where Acacia trees exclude buffel grass and where Eucalyptus tree facilitate buffel grass invasion may enhance understanding of the role of trees in the invasion of buffel grass. Pasture productivity and nutrient uptake are particularly high close to killed trees (Jackson & Ash 1998), suggesting that buffel grass invasion will follow the dynamics of the savanna canopy, persisting where trees die and establishing under developing trees.

The spread of buffel grass in the paddock used for the field experiment appears to have followed a relatively constant trajectory (Fig. 6). However, field observations suggest that buffel grass did not expand between the relatively dry years 2004–2007, but expanded rapidly in the 2007–2008 wet season with the first above-average wet season rainfall for 10 years. Prior to the breaking of this drought period, fuel loads were measured three times and never exceeded 1·0 t. ha−1 (unpublished data). The coincidence of low biomass and abundant rainfall may have provided the opportunity for buffel grass to expand after 2007.

Evidence from the field experiment suggests that the invasion of buffel grass is modestly enhanced by cattle grazing, especially without burning. Buffel grass is better able to respond to grazing than native grass species (Hodgkinson et al. 1989), probably because it has a larger number of nodal tillers remaining after defoliation. With frequent defoliation, these horizontally orientated nodal tillers allow buffel grass to maintain a 10-fold greater area of green tissue below the cutting height (Hodgkinson et al. 1989). This not only allows buffel grass to persist with grazing but also provides a greater opportunity to invest energy into seeding than native species.

There is no evidence from the regional survey and the field experiment that fire enhances the invasion of buffel grass. Regular burning and no grazing may be the optimal strategy to minimize the invasion success of buffel grass, but this management or any other combination of cattle grazing and burning is unlikely to restrict its spread. Other exotic grasses have been shown to proliferate and become dominant under a range of disturbance regimes (Setterfield et al. 2005; Bock et al. 2007). Buffel grass invasion probably proceeds in bursts assisted by the availability of intertussock space after the breaking of drought. A study from central Australia in a drier environment than the current study verified that buffel grass enhances fuel loads and fire intensity, but could not confirm that buffel grass establishment is assisted by fire (Miller et al. 2010). At least in the infertile environnments of north-eastern Australia, our study provides experimental evidence that fire does not enhance buffel grass establishment. Without this critical positive feedback, it is not possible to invoke the grass–fire cycle (D'Antonio & Vitousek 1992) as a realistic model for the spread of buffel grass in the infertile landscapes of north-eastern Australia.

Management implications

This study demonstrates that buffel grass has the potential to monopolize remnant vegetation given the high cover values (up to 42%) observed in some transects of the boundary survey. Broad-scale disturbances such as fire and cattle grazing did not have powerful influences on the invasion of buffel grass, but the combination of no livestock grazing and fire seems to be optimal to minimize spread. This suggests that fire can be used for other management purposes without promoting buffel grass invasion. Buffel grass expanded after a return to above-average rainfall following a drought period, possibly facilitated by the low biomass of the existing grass sward. The evidence from this study demonstrating the importance of propagule pressure suggests that stopping deliberate dispersal and sewing of buffel grass would make the greatest contribution to limiting its invasion in remnant native Eucalyptus woodland. This is at odds with the desirability of buffel grass as a pasture species with most cattle producers in central Queensland (Marshall et al. 2011) but supports the need for dedicated conservation reserves with active management to eradicate buffel grass outbreaks. Across the vast expanses of semi-arid Australia that are suitable for buffel grass (Lawson, Bryant and Franks 2004), and probably on other continents where it has been introduced, effective large-scale management strategies are elusive. However, our results relating propagule pressure to invasion success suggest that strategic removal of key source populations, although requiring intensive and ongoing effort, could be effective for inhibiting the further spread of buffel grass, particularly on infertile soils.

The enemy release hypothesis is widely invoked to explain the potency of species that have migrated to new environments without natural enemies (Colautti et al. 2004). There is emerging evidence that the potency of invaders diminishes over time (Reinhart & Callaway 2006; Diez et al. 2010) as a result of enemies following the invaders or evolving within the new environment. This might alleviate long term impacts but in the meantime, native ecosystems are likely to undergo severe disruption with the advance and proliferation of buffel grass.

Acknowledgements

Thanks to Ian Hoch, Carl Hoch, Carl Rudd, Ian Herbert and Cathy Herbert for assistance with establishing the experiment and manipulating the fire and grazing treatments. A special thanks to Russell Fairfax and Chris Donald for assistance with fieldwork; Chris prepared Fig. 1 and Andrew Biggs provided the soil data from the Queensland Government data base.

Ancillary