Experimental manipulation of habitat structure: a retrogression of the small mammal succession


  • Barry J. Fox,

    Corresponding author
    1. School of Biological, Earth and Environmental Science, University of New South Wales, Sydney, Australia 2052
      Barry J. Fox, School of Biological, Earth and Environmental Science, University of New South Wales, Sydney, N.S.W., Australia 2052. E-mail: b.fox@unsw.edu.au
    Search for more papers by this author
  • Jennifer E. Taylor,

    1. School of Biological, Earth and Environmental Science, University of New South Wales, Sydney, Australia 2052
    Search for more papers by this author
  • Peter T. Thompson

    1. School of Biological, Earth and Environmental Science, University of New South Wales, Sydney, Australia 2052
    Search for more papers by this author

Barry J. Fox, School of Biological, Earth and Environmental Science, University of New South Wales, Sydney, N.S.W., Australia 2052. E-mail: b.fox@unsw.edu.au


  • 1The habitat accommodation model for animal succession proposed that animal species enter a succession when changes in the vegetation succession reach a threshold of habitat suitable to that species. As the vegetation succession moves on, the habitat becomes less suited to that species and it is competitively excluded by species better suited to the habitat.
  • 2The main species in the mammalian succession following fire for wet heath in eastern Australia are rodents, with Pseudomys gracilicaudatus (eastern chestnut mouse) being followed by Rattus lutreolus (swamp rat) which becomes dominant with increasing time since fire. The abundance of both species has been shown to correlate with vegetation density, but in markedly different ways, and asymmetric interspecific competition has been demonstrated using controlled, replicated removal experiments in the field.
  • 3We used this system to examine if vegetation density is causal, manipulating the habitat by clipping to remove 60–70% of the vegetative cover from the 10 m × 10 m area surrounding each of six trapping stations on each grid. There were four experimental plots clipped, each with two abutting grids, one clipped and one unclipped, and a further four control plots each with two abutting grids that remained undisturbed.
  • 4We monitored the effects on each species with three censuses in January 1993 (summer) before clipping in early February, and on five further censuses, three in February (late summer) to assess immediate, short-term effects, one in August (winter) and one in December 1993 (early summer).
  • 5The abundance of Rattus lutreolus was significantly reduced by clipping the vegetation, while the abundance of Pseudomys gracilicaudatus remained relatively unchanged by the clipping.
  • 6Two species that are abundant on early succession stages in dry heath succession, but occurred at very low abundance on these wet heath habitats before clipping, P. novaehollandiae (New Holland mouse) and Mus domesticus (house mouse), showed marked increases in abundance on clipped and control plots soon after clipping.
  • 7An abundance index based on the standardized difference between clipped and control plots, was used to assess responses to habitat manipulation. Rattus lutreolus demonstrated significant negative index values, Pseudomys gracilicaudatus had index values close to zero until the winter census, P. novaehollandiae and Mus domesticus showed positive abundance indices soon after clipping. The order of significant responses to the habitat manipulation was revealed as: Mus domesticusPseudomys novaehollandiae→ P. gracilicaudatus→ Rattus lutreolus.
  • 8The impact of a habitat manipulation experiment on these four species of rodents produced a retrogression of the small-mammal succession. This demonstrated a causal role for vegetation density, which provided insight into the mechanisms that operate during the succession following fire, and supported the habitat accommodation model for animal succession.


In this paper we manipulate habitat to test for a causal relationship between abundance in the small-mammal succession following fire and vegetation density, to provide better understanding of the mechanisms involved in mammalian succession.

A model for the succession of small mammals in heathland from eastern Australia has previously been proposed to explain the changes in small mammal abundance following a wildfire (Fox 1982, 1990). One of the main features of the model was that mammal succession did not respond to time per se but followed the change in vegetation. The model was framed with reference to the three models put forward by Connell & Slatyer (1977). However, because of the animals’ dependence on habitat, controlled by the external factor of vegetation change, the models assessed were: ‘habitat facilitation’, ‘habitat tolerance’ and ‘habitat inhibition’. As the small-mammal species contravened aspects of one or more of these models, a new model for animal succession was proposed to incorporate elements of both the habitat facilitation and habitat tolerance models, and the new model was termed ‘habitat accommodation’ (Fox 1982). In this model, species did not modify local conditions, but rather, these conditions changed in response to external factors (vegetation succession in the example used). Small mammal species entered the succession when the changing local physical conditions (habitat) first met their specific requirements (habitat facilitation model). But as conditions moved out of the species’ optimal range their ability to obtain an adequate share of resources was reduced by competition from species better suited to those conditions (habitat tolerance model). Animals were either (1) reduced in numbers or (2) left the succession (Fox 1982). For the example used, in the first 10 years following fire the first option was observed with changes in relative abundance only. In the next 5 years the second option came into play as two species became locally extinct in the postfire succession (Fox 1996: 485; B. J. Fox, unpublished data). The habitat accommodation model for animal succession has also been shown to apply to ant succession (Fox 1990). Interspecific competition also plays an important role in determining structure in these ant communities (Fox & Fox 1982; Fox, Fox & Archer 1985; Haering & Fox 1987).

The two main aspects of the habitat accommodation model for small-mammal succession – (1) the existence of a threshold level for habitat colonization and (2) the competitive superiority of succeeding species – have been tested experimentally for the mammalian succession following disturbance by sand mining. Several studies (Fox & Fox 1978, 1984; Twigg, Fox & Luo 1989) have shown that the New Holland mouse (Pseudomys novaehollandiae (Waterhouse, 1843)) was absent on the earliest stages of the succession following sand mining, that were occupied by the house mouse (Mus domesticus Rutty, 1772). Fox & Twigg (1991) used transplant experiments to demonstrate that P. novaehollandiae was unable to colonize successfully when added to sites with less than 4·8 years of vegetation regeneration, but M. domesticus readily colonized these sites. They concluded that the habitat was suitable for M. domesticus, but was not suitable for P. novaehollandiae until after 5 years of regeneration. These two species had already been shown to compete, with reciprocal removal experiments using a series of replicated, controlled field manipulations of species’ abundance (Fox & Pople 1984; Fox & Gullick 1989). These studies were conducted on a sand-mined area with regeneration age between 6 and 8 years when the studies took place. The demonstrated competitive superiority of P. novaehollandiae explained the reduced abundance of M. domesticus when P. novaehollandiae entered the succession.

Similar manipulation experiments have been conducted in three different ages of wet heathland recovering from disturbance by wildfire (Higgs & Fox 1993; Thompson & Fox 1993). In these experiments the swamp rat Rattus lutreolus (Gray, 1841) was shown to be competitively superior to the eastern chestnut mouse Pseudomys gracilicaudatus (Gould, 1845), with a shift in numerical dominance, reflecting this behavioural dominance, occurring as the heathland grew older. The competition was asymmetric, with body size playing a critical role in the outcome (Thompson & Fox 1993). Habitat selection by these species indicates some separation along a habitat gradient defined by correlation with small scale topography, total vegetation density and other habitat variables. For R. lutreolus there was a positive correlation with vegetation density in some layers, while P. gracilicaudatus correlated negatively with total vegetation (Haering & Fox 1995). The area studied by Thompson & Fox (1993) had last burned 28 February 1988, it showed much more rapid vegetation growth, and the mammalian succession proceeded more rapidly than that studied by Fox (1982). A recent study of succession following fire, that was very rapid, months rather than the years of the 1982 study, was able to reconcile these two events, successfully demonstrating that these mammalian successions related to vegetation density rather than elapsed time (Monamy & Fox 2000).

Some studies have examined the microhabitat use of small mammals and investigated differential habitat selection between sympatric species in eastern Australia (Braithwaite & Gullan 1978; Braithwaite, Cockburn & Lee 1978; Fox & Fox 1978, 1981; Newsome & Catling 1979; Haering & Fox 1995, 1997; Monamy 1995; Monamy & Fox 1999). While these studies have demonstrated correlations with habitat variables including vegetation density, they do not demonstrate causality. Here, we wish to examine the direct effect that manipulation of habitat structure has on the succession of small mammals after fire. We focus on the density of vegetation on an area of wet heath regenerating after fire and concentrate on the species dominant in the latter stages of that succession, R. lutreolus. We ask three specific questions:

  • 1Is the abundance of R. lutreolus reduced by removing 60–70% of the vegetation from 10 m × 10 m squares surrounding trap stations on experimental plots?
  • 2Can the mammal succession following fire be reset to an earlier stage by this habitat manipulation?
  • 3How would such a retrogression of the small-mammal succession compare to the small-mammal succession following fire?


study area and study species

The mammal succession for heathland regenerating after fire has been reported for a complex mosaic of wet and dry habitats in a 7-ha coastal heathland plot (SL1), in Myall Lakes National Park (32°28′S, 152°24′E), 300 km north of Sydney (Fox 1982). Four species of rodents in that succession are relevant here: two wet heath species, Rattus lutreolus, adults often 100–180 g; and Pseudomys gracilicaudatus, adults generally 45–100 g; one species more usually associated with dry heath habitats, and less often found in wet heath, P. novaehollandiae, adults generally 16–18 g; and an introduced opportunist species found in all types of habitat, Mus domesticus, with adults usually 12–14 g.

Twenty years of data, covering two fire events (August 1974 and August 1980) and subsequent postfire regeneration at this site, for two broadly similar successions, have been reviewed as part of a study of long-term ecological sites (Fox 1996). As an example, trapping data from 1980 to 1988 for wet heath habitats following the second fire (August 1980) show a replacement sequence. The abundance of M. domesticus reached 19 individuals at 2 years then rapidly decreased, such high abundance in the earliest stages of succession is a noted feature of this opportunistic species (Fox & Gullick 1989). The abundance of P. novaehollandiae was also greater in the early part of the succession (9 individuals by 3 years) and then decreased, a subdued response to changes in the wet heath vegetation, that is much more strongly expressed in its more favoured dry heath habitats (Fox 1982, 1996). Pseudomys gracilicaudatus reached its peak abundance of 16 individuals at 3 years and then decreasing as the abundance of R. lutreolus increases with time, to 14 individuals at 8 years.

Although this study was not focused on P. novaehollandiae and M. domesticus, which are found predominantly in dry heath, they played an interesting role in this experiment so that information about them should also be reviewed. Pseudomys novaehollandiae is often associated with microhabitats of high floristic diversity which is largely a function of its granivorous diet and recolonization of heath following disturbance in the early or middle seral stages is typical of this species (Fox & Fox 1978; Kemper 1990), abundance increases with regeneration age over this period. This is likely to be a function of its habitat requirements as the species has previously shown a positive association with vegetation density in the low shrub layer in regenerating heath (Fox & Fox 1978) and regenerating open forest (Fox & Fox 1984).

Mus domesticus is an early successional species that occupies a diverse array of habitats including heath, forests and agricultural lands. Numbers show substantial fluctuations that appear to be related to changes in their regional level of abundance, which is not merely a seasonal breeding response. Krebs et al. (1995) provide evidence that some changes in behaviour, movement and home range size in agricultural lands are related to breeding and non-breeding periods, with nomadic behaviour in the latter, contributing to high regional abundances related to the occurrences of plague populations in wheat-growing areas. When plague populations have not been present, the regional abundance of house mice may be extremely low or absent. These regional population fluctuations need to be taken into account (see Fox 1996). The opportunism of M. domesticus, shifting into vacant niches, has previously been described (Braithwaite et al. 1978), with a similar role proposed for M. domesticus at Myall Lakes (Fox 1982).

Our present study relates directly to the succession of P. gracilicaudatus and R. lutreolus and their coexistence has been shown to be mediated by their differing foraging abilities in different stages of the succession (Luo & Fox 1996). Their diets have been well studied (Luo & Fox 1994), and competition from R. lutreolus has been shown to influence the diet of P. gracilicaudatus (Luo & Fox 1995) and alter patterns of habitat use (Haering & Fox 1995). Data collected on vegetation cover and abundance of these two species from the SL1 site and surrounding area, covering young, mid and old stages of vegetation succession (Higgs 1989; Higgs & Fox 1993) and the wet heath site further south (PT 1–8; Thompson 1990; Thompson & Fox 1993) can be used to document the relationship between each species’ abundance and vegetation density. Vegetation density was measured, when viewed from above, as the percentage of vegetation cover in the first 50 cm above the ground. For R. lutreolus there is a significant positive linear relationship (y = 0·871 + 0·084x, P = 0·012) that explains 33·4% of the variance. However, a quadratic relationship is even more significant (y = −10·062 + 0·595x − 0·00623x2, P = 0·0007) and explains 62% of the variance in R. lutreolus abundance (Fig. 1a), and has additional biological relevance as censuses from very old wet heathland have indicated very low abundances (or absence) of R. lutreolus (Fox 1996), rather than the high values that a linear relationship would imply. For P. gracilicaudatus there is a significant negative linear relationship (y = 7·346 + 0·108x, P = 0·0021) that explains 45·5% of the variance. However, biologically there is a problem with such a relationship as observations of sites with very low vegetation cover (earliest stages after fire) indicate an absence of P. gracilicaudatus rather than an increased abundance that a negative relationship would imply. This is reinforced by observations that there were at least two monthly censuses when no individuals were captured on SL1 after the 1974 fire (Fox 1982). Hence, considering the origin as a logical zero and fitting a quadratic relationship with no intercept (y = 0·277x− 0·00447x2, P < 0·0001) 81·1% of the variance in P. gracilicaudatus abundance is explained. That provides a relationship that is biologically more realistic (Fig. 1b). Over the range of values shown for the vegetation index in Fig. 1 (20–65) it is clear that for either linear or quadratic relationships the abundances of these two species tend to move in opposite directions, with the age of the regenerating vegetation shown by the size of the symbols used (smaller = younger and larger = older), indicating that manipulation of vegetation cover is indeed the appropriate method to study the mechanism determining these two species’ position in the mammalian succession.

Figure 1.

The number of individuals on 1-ha plots as a function of the vegetation cover (x) below 50 cm. Plots were from sites with four different stages of vegetation succession indicated by the size of the symbols: large = 8y, a set of four old plots; medium = 3·5y, a set of four mid-aged plots; small = 1·5y, a set of four young plots (Higgs 1989) and smallest = 1·75y, a second set of six young plots (Thompson 1990). Quadratic regressions fitted for rodent abundance are: (a) Rattus lutreolus= −10·062 + 0·595x −0·00623x2 (P = 0·0007) that explains 62% of the variance, and (b) Pseudomys gracilicaudatus= 1·38 + 0·204x − 0·00362x2 (P = 0·0163) that explains 40% of the variance.


A set of eight replicate plots (PT 1–8) that had previously been used in a removal experiment (Thompson & Fox 1993) was selected for use in this experiment. Two of the plots (5 and 8) were moved slightly to correct for some structural and floristic differences in the vegetation that had been detected by Thompson & Fox (1993). The position of the eight plots used and the experimental treatment applied to each is shown in Fig. 2, together with the distribution of wet heath, dry heath and swamp habitats. All plots were separated by at least 100 m to reduce the likelihood of animals moving between plots.

Figure 2.

(a) The location on the study site of four experimental and four control plots interspersed in wet heath habitat, abutting dry heath and swamp habitats, three km south of SL1. (b) The layout of an experimental plot showing trap stations (•) and (c) stylized representations of areas 10 m × 10 m surrounding stations designated for clipping from which 60–70% of the vegetation was removed (shown with shading). Small clumps of unclipped vegetation remained, particularly in the immediate vicinity of the trap station (X). The abutting clipped (3 × 4 stations) and unclipped (3 × 4 stations) plots are labelled. Control plots had the same layout but the vegetation remained unaltered.

habitat manipulation

Each plot comprised 12 trap stations in a three × four pattern with 20-m spacing. These were subdivided into two abutting grids (2 × 3). On the experimental plots (1, 3, 4, 6) one grid was clipped and one left unclipped for a future treatment, while both grids were left unclipped on control plots (2, 5, 7, 8; Fig. 2b). Each set of four clipped plots and four unclipped plots comprised two at lower elevation and two at higher elevation. The four control plots each comprised two abutting grids that spanned the same range of elevation, vegetation and geographical variation (Fig. 2). Within the 10 m × 10 m around each of the six trap stations on each clipped grid 60–70% of the vegetation cover was cut to within 5 cm of the ground using a brush-cutter, and the cut vegetation was removed. A stylized version of how grids were clipped is given in Fig. 2(c), in each case some unclipped vegetation was left immediately surrounding each trap station (X), as well as a mosaic of patches making up the 30–40% of vegetation remaining after clipping. The clumps of vegetation left unclipped (approximately 80–100 cm high) were usually positioned to include a large shrub (Banksia oblongifolia Cav.) or a clump of button grass (Gymnoschoenus sphaerocephalus (R. Br.) Hook. f) that would provide some cover to rodents using that station. The clipping took place in late summer during the period from 2 to 6 February 1993. The amount of vegetation cover around each individual trap station was calculated by estimating the cover in the 10 m × 10 m area around each trap station, from 16 separate quadrats each 2·5 m × 2·5 m. A mean value for each plot was obtained by averaging values over the six trap stations on each grid. On clipped grids the cover at each trap station was averaged across both the areas cleared by brushcutter (nominally zero cover) and the patches of uncut vegetation remaining within the clipped areas. On control plots, the wet heath vegetation was so dense that all stations had close to 100% cover present.

mammal trapping

At each trap station, a single collapsible Elliott small-mammal trap (33 cm × 10 cm × 9 cm, Elliott Scientific Co., Upwey, Victoria) was set and baited with a combination of peanut butter, oil and rolled oats. Each trapping session extended over three nights and traps were checked, cleared and rebaited if necessary each morning at sunrise. Each captured animal was identified to species, weighed, marked with an individual toe clip and released. Preclipping abundance was assessed from three trapping sessions that were carried out in summer before the clipping took place: 6–8 January, 14–16 January and 30 January to 1 February 1993. The short-term impact of the treatment was determined from postclipping abundance assessed from three trapping sessions in late summer: 7–9 February, and 16–18 February and 8–10 March. The medium-term impact was monitored during trapping sessions in winter 12–14 July, 7–9 August and a final session in early summer, 12–14 December 1993. At the time of the first winter trapping session (12–14 July) there had been 143 mm of rain on the site during 12 raindays in June and there was torrential rain on the site on 11 July. The result was at least 5 cm of standing water at many of the stations over all plots, with the worst affected plot up to 15 cm deep. As trapping results for this session were much more likely to have been influenced by the standing water rather than the treatment effects alone, we have not included this session in our analyses. The remaining five postclipping sessions were then graphed at 34, 43, 63, 215 and 342 days after the beginning of the experiment (although they extended a day either side). The clipping was nominally at day 30 (although it extended from 28 to 32, i.e. 2–6 February).

data analysis

The number of individuals trapped on each six-station grid over the 3-day trapping session was used as the measure of abundance. Each of the treatment and control trapping grids also had trapping carried out on an abutting grid at the same time. Control plots were treated in the same way as experimental plots with means for each abutting grid calculated separately. Then the two grids on each control plot were averaged to give a value for that replicate plot (see Fig. 2). Mean values for the clipping and control treatments were calculated from the four replicate plots. Results for each species were analysed separately.

Analyses of variance

A two-way analysis of variance (anova) on repeated measures was carried out with the factors being treatment (clipped or control) and times (preclipping or postclipping) with the repeated measures being the sets of three trapping sessions before, and three trapping sessions after, the clipping. Contrasts between specific cells in the analyses were conducted where perusal of the cell means and standard errors indicated this was appropriate and Bonferroni corrections were applied to determine the appropriate significance level.

To assess the significance of changes with time, and to further investigate some of the significant interaction effects observed, the mean value for each plot over the three preclipping sessions was compared with the value for that plot in a single postclipping session, as a two-way anova instead of a repeated measure analysis. We conducted a separate analysis for each trapping session after the clipping and used contrasts to compare clipped and control plots.

Abundance index

To highlight the difference between the clipped and control plots for each trapping session, and examine their changes over time, a mean abundance index was devised. The index was calculated by subtracting the mean value for control grids from the mean value for clipped grids, with the difference then standardized by dividing by the mean value for control grids. To obtain error bars for this index an average was taken of the coefficient of variation (mean/standard deviation) for clipped and control plots, and this was applied to the mean value for the abundance index to obtain estimates for standard deviation, and then standard error by dividing by the square root of the sample size.


The vegetation structure on clipped plots did not remain static over the period of the experiment (February to December) as clipping was carried out once only as a ‘pulse’ rather than a ‘press’ experiment (Bender, Case & Gilpin 1984). This allowed examination of both the initial impacts of the manipulation over the first month as well as the response of rodents as the vegetation regrew and moved toward its previous structure. The growth of Xanthorrhoea fulva (A. T. Lee) D. J. Bedford and sedges was relatively rapid so by the July and August census sessions there was substantial regrowth present. However, at all times over the period of the experiment there was very substantially less vegetation present on the clipped experimental plots than on the control plots. Changes in the mean abundance (number of individuals on each three × two trap grid on 0·24 ha) as a function of time over the trapping sessions from January to December 1993 are shown in Fig. 3.

Figure 3.

The mean abundance (± SE from four replicates) on clipped and control plots showing an average for the preclipping, trapping sessions (6–8, 14–16 January and 30 January to 1 February), with individual values for three immediate postclipping, trapping sessions (7–9, 16–18 February and 8–10 March) and two monitoring sessions (7–9 August and 12–14 December) as a function of the number of days since the experiment began: for (a) Rattus lutreolus (swamp rat); (b) Pseudomys gracilicaudatus (chestnut mouse); (c) P. novaehollandiae (New Holland mouse) and (d) Mus domesticus (house mouse). Asterisks indicate where clipped plots differ significantly from control plots (see Table 2).

At the commencement of the experiment, before clipping, the most abundant species was R. lutreolus. Pseudomys gracilicaudatus was much less abundant, although it had been the most abundant on this area in previous years, at earlier successional stages (Thompson & Fox 1993). The remaining two species, P. novaehollandiae and M. domesticus, had been trapped only at low abundance on similar wet heath sites in previous years (Higgs & Fox 1993), and were not very abundant in the preclipping censuses.

The total abundance of R. lutreolus summed over all plots remained remarkably constant over the period from January to August 1993 (42 (24 days), 46 (34 days), 47 (45 days), 35 (62 days), 41 (191 days) and 39 (217 days)). This was highlighted by the recorded movement of individual animals off the clipped grids onto the abutting unclipped grids, which explained the relative stability of the total number of R. lutreolus, although 4 months later, during the final census in early summer, the abundance had dropped. The total abundance over all plots for each of the other three species showed marked increases over the period of the experiment, although for M. domesticus the total abundance dropped again in the following summer.

repeated measures anova

A summary of the probability values for the two-way anova with repeated measures is shown in Table 1 for each of the four species. For R. lutreolus there was a strong treatment effect (P = 0·0012). While the treatment × times interaction was not significant (P = 0·1415), and before clipping the contrast between control and clipped plots showed no significant difference, after clipping the contrast was highly significant (P = 0·0015, Bonferroni critical value P = 0·0125) and there was no interaction with repeated measures sessions. Clearly R. lutreolus abundance was significantly decreased by clipping vegetation.

Table 1.  Probability values for two-way anova (Treatment – clipped or control; and Times – pre- or post-) with repeated measures (Sessions – 1, 2 and 3) comparing results from three preclipping sessions 6–8 January, 14–16 January, 30 January to 1 February 1993 with results from three postclipping trapping sessions 7–9 February, 16–18 February, 8–10 March 1993 for four species of rodents. Significant values are shown in bold (Bonferroni critical value for contrasts P = 0·0125)
 Rattus lutreolusPseudomys gracilicaudatusPseudomys novaehollandiaeMus domesticus
Treatment × Times0·14150·79820·71060·4114
Sessions × Treatment0·22270·67820·08360·0097
Sessions × Times0. 17990·16600·30440·0002
Sessions × Treatment × Times0·26170·83380·02270·0388
Postcontrol vs. clipped0·00150·90400·60080·6154
Interaction with Sessions0·23710·90320·00680·0011

For P. gracilicaudatus there was a significant times effect (P = 0·0365) and the repeated session effect was marginal (P = 0·0574), but there were no significant contrasts. The abundance of P. gracilicaudatus on both control and clipped plots increased significantly with time, pre- to post-, and was close to significant from session to session.

For P. novaehollandiae the times effect was marginal (P = 0·0523), as was the repeated session effect (P = 0·0718) and the session × treatment interaction (P = 0·0836). However the session × treatment × times interaction was significant (P = 0·0227), hence requiring separate examination of each of the main effects and two-way interactions. Mus domesticus had a highly significant times effect (P = 0·0001) and the session effect and all session interactions were also significant, again requiring more detailed analyses.

two-way anova

The two-way anova for the mean of the preclipping plots compared with the mean for each separate postclipping session was conducted for each of the four species, and the contrasts between clipped and control plots are presented in Table 2.

Table 2.  Two-way anova comparing preclipping mean value (from sessions 6–8, 14–16, 30 January to 1 February 1993) with results at each single postclipping trapping session. Probability values for the postclipping contrasts between clipped and control plots are shown for four species of rodents. Significant values are shown in bold
Preclipping mean value 6–8, 14–16, 30/I to 1/II/93Short-termWinterSummer
4 days postclipping 7–9/II/9313 days postclipping 16–18/II/9333 days postclipping 8–10/III/93185 days postclipping 7–9/VIII/93312 days postclipping 12–14/XII/93
(A) R. lutreolus0·00010·00310·01140·00850·0226
(B) P. gracilicaudatus0·62831·0001·0000·12860·0188
(C) P. novaehollandiae0·01090·39330·09480·22900·2201
(D) M. domesticus0·14040·03350·02510·10490·3238

Rattus lutreolus shows a highly significant reduction in abundance for all sessions after clipping. The only significant interaction (P = 0·0125) was for the first postclipping trapping session and contrasts between the clipping and control plots were significant for all postclipping sessions (Fig. 3a and Table 2).

In comparison to this P. gracilicaudatus shows no significant treatment effects, although the interaction for the December session is marginal (P = 0·0545) and for this session the contrast shows the clipping plots are significantly lower than control plots (Table 2). The increased P. gracilicaudatus abundance on the control plots could reflect the decrease in R. lutreolus abundance on control plots between August and December. All postclipping trapping sessions, except the first, show significant increases over preclipping levels for both clipped and control plots (Fig. 3b).

Pseudomys novaehollandiae shows significant time effects with increases over preclipping levels on both clipped and control plots for the third short-term (summer) session, and the August and December sessions. The clipped plots contrast showed abundance significantly below that for control plots for the first postclipping session (Table 2), and abundance on the clipped plots continued to increase throughout the whole year (Fig. 3c).

Mus domesticus abundance increased over preclipping levels on both clipped and control plots for two short-term (summer) sessions and the winter session. There was a significant contrast for the short-term (summer) sessions (Table 2), with clipped plots exceeding control plots in February (13 days), but with control plots exceeding clipped plots in March (33d). There were marked decreases in abundance on all plots between August and December (Fig. 3d).

abundance index

The abundance index provides ways to look at direct differences between clipped and control plots, independent of differences in overall abundance between species. The value of this index as a function of time is shown for all four species trapped in Fig. 4. The zero line represents what would be expected if the clipping had no effect. Negative values indicate reductions in the abundance of the species on clipped plots, and positive values indicate increased abundance on clipped plots, relative to the control plots.

Figure 4.

An abundance index calculated as standardized difference in mean abundance {(clipped-control)/control} for R. lutreolus, P. gracilicaudatus, P. novaehollandiae and M. domesticus on 0·24-ha plots. The zero line represents zero difference between clipped and control plots. Asterisks indicate where Abundance Index values differ significantly from zero.

All species except M. domesticus display a strong negative effect in abundance on clipped plots immediately after clipping (within 4 days). However, from this point on, marked differences in response become evident. The response of M. domesticus is significantly positive in the first postclipping session (34 days), 4 days after clipping, then decreasing to 63 days, with another significant positive value in winter. Pseudomys novaehollandiae responds a little more slowly, increasing to a peak in the second and third session (63 days) and then dropping below equilibrium, before climbing again above the equilibrium line. Of the wet heath species the most neutral effect seems to be on P. gracilicaudatus remaining at equilibrium through the second and third sessions (43 and 63 days) before the difference increases to peak at the fourth session (215 days), then falling back below equilibrium and into negative impact. Rattus lutreolus remains detrimentally affected over the entire period of the experiment (with significant negative values at 34 and 342 days), and shows virtually no response as the clipped vegetation regrows. The order and pattern of the species’ response, as measured by this abundance index, is clear: M. domesticusP. novaehollandiaeP. gracilicaudatusR. lutreolus.

remaining vegetation

The amount of vegetation actually remaining in the 10 m × 10 m area around each trap station on clipped plots was measured soon after clipping. The mean values for each of the four clipped plots was 29·7%, 30·7%, 33·7% and 36·0%. But this represents an averaging of patches of vegetation that remained unclipped (with cover estimates 96·7 ± 0·6%) and the intervening matrix of clipped area with almost zero cover, so that the mean value is largely determined by the area of the unclipped patches remaining, as a percentage of the 10 m × 10 m area. Hence, the mean value may have a different meaning from the same percentage of unclipped vegetation more or less evenly distributed. For clipped plots, the abundance of each wet heath species is shown for the first two trapping sessions after clipping (mean abundance from sessions 4 days and 13 days after clipping) plotted against this estimate of remaining vegetation in Fig. 5. The abundance of R. lutreolus decreases regularly with decreasing cover, to reach zero at the two least dense clipped plots. The abundance of P. gracilicaudatus appears to have a peak at the second least dense clipped plot, and then appears to be decreasing towards zero at a value below those measured. These results indicate that P. gracilicaudatus can occupy areas with lower cover than can R. lutreolus, and also that at lower values of cover P. gracilicaudatus might also decrease toward zero abundance.

Figure 5.

Rodent abundance (number of individuals) on each of the four 0·24-ha replicate plots, as a function of the remaining vegetation cover (percentage) measured soon after clipping: (a) P. gracilicaudatus and (b) R. lutreolus shown as mean abundance for the first two trapping sessions following clipping.


Experimental manipulation of wet heath habitat by clipping 70% of the vegetation in 10 m × 10 m area surrounding trap stations produced significant decreases in the abundance of the most abundant rodent present (R. lutreolus). As there was little change in total abundance of R. lutreolus over the main period of the experiment (up until August), any differences in abundance between experimental and control plots must be due to individuals altering their use of habitat. We interpret the reduced use of the clipped experimental plots, as being caused by the change in habitat structure, because there was relatively little change in habitat use on the control plots, except for the session immediately after clipping. The reduction in the density of the wet heath vegetation led to a reduction in the density of swamp rats (R. lutreolus) on experimental plots, consistent with the relationship shown in Fig. 1, but with the density of vegetation now demonstrated to be the causal factor. Our interpretation is that this habitat manipulation causes individual animals to move immediately from the experimental clipped plots as they adjust to a new equilibrium level of abundance and this is consistent with the observed abundance on the four clipped replicates in the first two weeks after clipping (Fig. 5).

For P. gracilicaudatus the situation is less simple. There were virtually no changes in abundance or use of habitat on the experimental plots in relation to control plots until August (215 days; see Fig. 4) and December (352 days; see Fig. 3b). The increase in abundance on control plots between August and December we would interpret as competitive release caused by the reduction in R. lutreolus abundance on control plots over that period (see Higgs & Fox 1993; Thompson & Fox 1993). One interpretation could be that vegetation density is not a causal influence on the abundance of P. gracilicaudatus. A more likely interpretation would be that the alteration to the density of vegetation present in wet heath habitats was not sufficient to make the habitat unsuitable for P. gracilicaudatus, as it still lay within the range of habitat variables they encounter in early stages of the plant succession following fire. A more detailed approach to the second interpretation would take account of the observed relationship between P. gracilicaudatus abundance and vegetation cover (Fig. 1, see also comments in Study Area), that would imply that some change should occur. If the relationship was negative, linear and causal, then one would expect a decrease in the amount of vegetation cover of the magnitude in this experimental manipulation would produce an increase in the abundance of P. gracilicaudatus relative to that on control plots. However, if the relationship is indeed quadratic as we argued (Fig. 1), then one might expect that decreasing the amount of vegetation cover would first produce an increase in abundance that would then level off before beginning to decrease again. We would argue that this is what has occurred on these experimentally clipped plots, and that this interpretation is also consistent with our observations of the abundance of P. gracilicaudatus in the four clipped plots in the first 2 weeks after clipping (Fig. 5).

Had we reduced the vegetation cover by a smaller amount, by clipping a smaller proportion of the area on experimental plots, we may have detected a significant increase in P. gracilicaudatus abundance. We would further predict as a corollary, that a manipulation removing a sufficiently large proportion of the vegetation would reduce the habitat to below the threshold necessary for successful colonization and establishment by P. gracilicaudatus and we would see them become less abundant in the same way as R. lutreolus has been in this experiment. This latter experiment has been undertaken and has conclusively demonstrated a significant reduction on experimental plots from which 85% of the vegetation has been clipped (Monamy 1998). The results of that experiment are being prepared for publication (V. Monamy personal communication).

The marked increases observed in the abundance of P. novaehollandiae and M. domesticus were not an expected result from the habitat manipulation experiment. Although both species have been caught in wet heath habitats at other times (Fox 1982), the dense vegetation present in the lower layers in general make these habitats less suitable for these species (Fox & Fox 1981), but see Haering & Fox (1997) for an alternative view of juvenile and subadult P. novaehollandiae dispersing into wet heath habitats. With the abundance of R. lutreolus significantly reduced on clipped plots, and the abundance of P. gracilicaudatus showing little change for much of the time, there definitely was a reduced abundance of rodents on experimental plots, and perhaps a reduction in some resource levels as well, because of the habitat manipulation. However, for P. novaehollandiae increased amounts of open space, reduced vegetation in the lowest 20 cm and relatively ‘bare sand’, recognized as habitat requirements (Fox & Fox 1981), might be considered to increase their habitat resources, making the experimental plots suitable habitat for rapid colonization.

We would advance a similar argument for the opportunist M. domesticus which very often appears in habitats when previous occupants have been reduced to low abundance (Fox 1982; Fox & Pople 1984; Fox & Gullick 1989). However, there are some interesting differences between the colonization of these two species. Firstly, M. domesticus shows a rapid increase on control plots as well as experimental plots, and there is a marked reduction in abundance on all plots by the December census. This is consistent with regional fluctuations in the abundance of this species, with relatively short periods of often extremely high abundance interspersed with longer periods of low abundance or even absence (see Fox & Gullick 1989) that Krebs et al. (1995) link with nomadic behaviour driven by the flow on effects of their fluctuating abundance in distant agricultural areas. Secondly, M. domesticus shows the most significant differences between clipped and control plots immediately after the clipping (Fig. 4), while P. novaehollandiae abundance on experimental plots is significantly lower than on control plots at that time (Figs 4 and 5). For P. novaehollandiae the difference is not significantly positive until 63 days, in the third session after clipping (Fig. 4). The most likely interpretation is that the opportunist M. domesticus most rapidly capitalizes on the ‘new’ resource of unoccupied space created when R. lutreolus moves away from clipped plots and that P. novaehollandiae is a little slower in its opportunistic response.

The significant interactions for M. domesticus, three-way and two-way (Table 1), require reanalysis of the M. domesticus response, as they are clearly behaving differently in different sessions and treatments as well as times. The results of specific comparisons in Table 2 and Fig. 4 support this interpretation of the M. domesticus response. For P. novaehollandiae there is a significant three-way interaction (Table 1) with markedly different responses in each of the first three sessions after clipping, thus requiring separate analyses for each session for this species as well. These analyses indicate that there is a marked increase in the strength of the P. novaehollandiae response on clipped plots in comparison with control plots from the first to the third session after clipping (Figs 4 and 5).

While we cannot easily make a direct comparison of the two postfire successions on the SL1 area (1974–80 and 1980–88; see Fox 1996) with the observations from the clipping experiment over less than 1 year on the new area (Figs 4 and 5), it is possible to draw some inference from the patterns observed. To begin; the opportunistic behaviour of M. domesticus can be seen in the earliest stages in both cases, with the response of P. novaehollandiae seen as less opportunistic but still responding relatively quickly to the clipping disturbance. We should emphasize here that from 1974 to 1980 in the original analyses (Fox 1982) M. domesticus showed an early rapid response in the dry heath habits and did not shift to the wet heath habitats until later, when the abundance of P. novaehollandiae in the dry habitats was increasing rapidly. Hence, we view the early postfire succession in wet heath as rapid colonization and increasing abundance of M. domesticus followed by colonization and increasing abundance of P. novaehollandiae.

An early successional species in wet heath is the usual role attributed to P. gracilicaudatus and this is seen in the 8-year succession (1980–88), also reflected in its neutral response to clipping. The late successional role of R. lutreolus is also reflected in the detrimental effect of clipping. We would note that on heathland much further south, Catling (1986) found a somewhat different succession of small mammals following fire when Pseudomys was absent; R. lutreolus colonized at a much earlier time, but that study did not take account of the rate of vegetation succession. Careful measurement of the rate of vegetation succession at sites close to those used in the present study have demonstrated much more rapid rates of succession by R. lutreolus as well. These also documented even earlier rapid colonization by P. gracilicaudatus, reflecting a similarly rapid increase in vegetation density, clearly indicating that the mammalian response is more likely related to the rate of the vegetation succession, rather than time per se (Monamy 1998; Monamy & Fox 2000).

The question of scale must be addressed here as the perturbation by fire covered many thousands of hectares, and left only a small part of the 7-ha SL1 plot unburned (Fox 1982). On the other hand the disturbance simulation we achieve by clipping on these plots is on a much different scale, with four sets of six 10 m × 10 m clipped areas, each spread over approximately one-quarter of a hectare. Hence, animals are able to occupy the unclipped corridors between clipped areas, the abutting unclipped plots and the surrounding undisturbed vegetation including the control plots. This difference in scale meant that the response to the clipping perturbation should be much more rapid, as animals can remain adjacent in undisturbed areas rather than needing to disperse longer distances as was necessary to reach the burned areas of SL1. This is in agreement with what we observe. We also note that differences in the way the vegetation is distributed at the scale of 10 m × 10 m means that the amount of vegetation remaining, as shown on the x-axis for Fig. 5, cannot be directly equated with the vegetation density shown on the x-axis for Fig. 1. The relative position of the two species on the axis would be the same but the absolute value on the axis could be shifted, depending on the method used for calculation for Fig. 5.

We interpret our results to mean that the experimentally clipped plots represent an early stage in the vegetation succession. Clipped plots are not a suitable habitat for the late successional species R. lutreolus, as they moved from clipped stations to unclipped stations, reducing their abundance on clipped plots. The early successional wet heath species P. gracilicaudatus maintains abundance on the clipped ‘early successional stage’, which then fell within its preferred habitat, the early successional stage. Confirmation of the clipped plots as ‘early successional’ is provided by the fact they are clearly considered as suitable habitat for the very early colonizers, M. domesticus and P. novaehollandiae. We provide a stylized summary interpreting these results in Fig. 6 from which we conclude that clipping of this wet heath habitat with dominant R. lutreolus, and subordinate P. gracilicaudatus shifts it to an earlier successional stage and causes the subtraction of the former species without change to the latter, while at the same time leads to the addition of the early colonizers P. novaehollandiae and M. domesticus. We also note that the coexistence is likely to be influenced by recent observations of the dominant role that female R. lutreolus play in determining the habitat use of male R. lutreolus, and both male and female P. gracilicaudatus (Monamy 1997). Similar results have also been reported for a related congener and subspecies, P. higginsii and R. lutreolus velutinus in Tasmania (Monamy & Fox 1999), where competition was also confirmed (Luo, Monamy & Fox 1998). Our results demonstrate the causal link between habitat structure, habitat variables and the abundance of R. lutreolus occupying these wet heath habitats.

Figure 6.

A stylized summary of small-mammal succession on wet heath habitat altered by clipping, noting the species added and subtracted as a consequence of the clipping.

Foster & Gaines (1991) also used clipping (mowing in their case) in their habitat manipulation in Kansas, to produce interstitial areas between different sized patches of old field successional vegetation. Only one of the four species encountered in the patches of successional vegetation was trapped in large numbers on these mown interstitial areas, Peromyscus maniculatus (Wagner, 1845) (deer mouse, 15–25 g adult mass); this species can in some ways be regarded as an ecological analogue of Pseudomys novaehollandiae. The other three species traversed these areas to reach patches of successional vegetation. The largest species, Sigmodon hispidus Say & Ord, 1825 (cotton rat, 90–160 g) was never caught on the mown areas, and captures of Microtus ochrogaster (Wagner, 1842) (prairie vole, 28–55 g) averaged less than one individual per season in the three years of the experiment. The smallest species Reithrodontomys megalotis (Baird, 1858) (western harvest mouse, 8–15 g) provided 10% of its total captures from the mown area. While there is less evidence for ecological analogues with these three species, both R. lutreolus, and P. gracilicaudatus show some vole-like characters, particularly the use of runways in dense vegetation. The wet heath vegetation is clearly very different floristically from the prairie habitat in Kansas, but structurally both habitats comprise short, dense vegetation. Peromyscus maniculatus captures on the mown areas increased markedly, as the old field patches increased in age over the 3 years of the experiment, to provide more than 50% of captures for the latter half of the experiment.

Abramsky, Dyer & Harrison (1979) experimentally perturbed shortgrass prairie by the addition of water and nitrogen, separately and together, for comparison to control plots. They also found their experimental plots were ‘invaded’ by species usually considered rare on shortgrass prairie, while the three species normally found on shortgrass prairie appeared to avoid the experimentally supplemented plots. There was increased growth on the experimental plots, particular the water + nitrogen treatment. They interpreted these results as the ‘invading’ species selecting the more favourable habitat and excluding the species usually found there. By removing the ‘invading’ species from the water + nitrogen treatment Abramsky et al. (1979) were able to demonstrate significantly increased abundance in what they termed the ‘native’ species. They also made direct measurements of competition coefficients, as has been done for R. lutreolus and P. gracilicaudatus (Fox & Luo 1996).

Brown & Heske (1990) report on a manipulation experiment that removed large seed-eating Dipodomys spp. (kangaroo rats) to produce marked changes in vegetation type over 12 years, with marked increases in grass cover. Over this period they have shown a similar dependence by rodents on vegetation cover, in their case cover of tall perennial grasses. Specialized grassland species such as Sigmodon spp. showed a lower rate of increase in captures with increasing grass cover than did the rate of increase for all rodents including Reithrodontomys spp. They regard the guild of large seed-eating rodents as ‘keystone species’ because of the influence they exert on the entire ecosystem. Although on a much smaller scale, our study indicates there are some elements of a similar effect being exerted by Rattus lutreolus on all other rodent species in the wet heath habitat. However, we are not able to provide any information on possible feedback loops where the rodents themselves influence the vegetation succession.

The coexistence of the wet heath species, R. lutreolus and P. gracilicaudatus, has been shown to be mediated by their differing foraging abilities in different stages of the succession (Luo & Fox 1996). These two species have been shown to exhibit asymmetric interspecific competition (Higgs & Fox 1993; Thompson & Fox 1993). The diet of P. gracilicaudatus changes with season and stage of vegetation succession (Luo & Fox 1994) and these changes have been related to competition with R. lutreolus (Luo & Fox 1995). Although less well studied with regard to their coexistence, there are studies demonstrating the effects of interspecific competition on P. novaehollandiae and M. domesticus (Fox & Pople 1984; Fox & Gullick 1989), together with transplant experiments examining habitat suitability (Fox & Twigg 1991) and a study of their habitat selection (Haering & Fox 1997).

This experiment was designed specifically to test the effects of manipulating wet heath habitat on the two species of rodents normally found in that habitat, and to assess the roles of habitat and interaction in their replacement sequence as part of the mammalian succession. The opportunity to examine an expanded succession was fortuitously, and somewhat unexpectedly, provided by the arrival of two early colonists, more commonly found in dry heath habitat. Hence, we had the opportunity to examine when species entered the succession in direct response to our manipulation of the habitat. The use of newly clipped plots by these species, in relation to the percentage cover of remaining vegetation is consistent with the roles for habitat selection and species interaction set out in the habitat accommodation model for animal succession (Fox 1982, 1990). Removing 60–70% of the vegetation around trap stations on experimental plots effectively reset the mammalian succession to an earlier stage. This retrogression of the small-mammal succession has provided insight into the mechanisms that operate during the succession following fire and support the habitat accommodation model for animal succession.


We thank Peter Higgs, Miranda Gott and Anna Povey for their assistance in collecting field data, and New South Wales National Parks and Wildlife Service for permission to work in Myall Lakes National Park. Vaughan Monamy provided useful comments on this paper. The study was supported in part by the Australian Research Council (most recently on grants A18930327, A19330222 and A19700994 to BJF).