Does a top-predator provide an endangered rodent with refuge from an invasive mesopredator?

Does a top-predator provide an endangered rodent with refuge from an invasive mesopredator? M. Letnic 1, M. S. Crowther 1,2 & F. Koch 1 1 Institute of Wildlife Research, School of Biological Sciences, University of Sydney, NSW, Australia 2 Department of Environment and Climate Change (NSW), Hurstville, NSW, Australia Animal Conservation. Print ISSN 1367-9430 Keywords Dingo; fox; mesopredator; Notomys fuscus; refuge; keystone species; conservation; arid. Correspondence M. Letnic, Institute of Wildlife Research, School of Biological Sciences, University of Sydney, NSW 2006, Australia. Email: mletnic@usyd.edu.au Received 10 November 2008; accepted 29 January 2009 doi:10.1111/j.1469-1795.2009.00250.x Abstract In arid environments, ecological refuges are often conceptualised as places where animal species can persist through drought owing to the localised persistence of moisture and nutrients. The mesopredator release hypothesis (MRH) predicts that reduced abundance of top-order predators results in an increase in the abundance of smaller predators (mesopredators) and consequently has detrimental impacts on the prey of the smaller predators. Thus according to the MRH, the existence of larger predators may provide prey with refuge from predation. In this study, we investigated how the abundance of an endangered rodent Notomys fuscus is affected by Australia s largest predator, the dingo Canis lupus dingo, introduced mesopredators, introduced herbivores, kangaroos and rainfall. Our surveys showed that N. fuscus was more abundant where dingoes occurred. Generalised linear modelling showed that N. fuscus abundance was associated positively with dingo activity and long-term annual rainfall and negatively with red fox Vulpes vulpes activity. Our results were consistent with the hypothesis that areas with higher rainfall and dingoes provide N. fuscus with refuge from drought and predation by invasive red foxes, respectively. Top-order predators, such as dingoes, could have an important functional role in broad-scale biodiversity conservation programmes by reducing the impacts of mesopredators. Introduction The ecological refuge model is a widely cited concept to explain the population dynamics of small- and mediumsized mammals in arid Australia (Morton, 1990; Dickman, Predavec & Downey, 1995). Ecological refuges are frequently conceptualised as places where animal species can persist through drought owing to the existence of relatively dependable supplies of moisture and nutrients (Morton, 1990; Morton, Short & Barker, 1995). Movement of animals among resource rich refuges is thought to increase the persistence of populations at a regional scale because refuges may act as source areas for emigrants to the surrounding landscape when favourable climatic conditions prevail (Carstairs, 1974; Dickman et al., 1995; Brandle & Moseby, 1999). Resource rich refuges are generally thought to result from geographical features such as rocky ranges and low-lying drainage basins where the availability of nutrients and moisture is greater than the surrounding landscape. Areas that have received localised rainfalls have also been postulated to serve as ephemeral refuge habitats (Newsome & Corbett, 1975). Since British colonisation of Australia in 1788, many animal species including the rabbit Oryctolagus cuniculus, red fox Vulpes vulpes, domestic cat Felis catus and domestic livestock have been introduced to Australia. The disruptions to ecological processes caused by these species have been devastating and have been implicated in the mass extinction and range reductions of marsupials and rodents, although the exact causes remain unknown (Burbidge & McKenzie, 1989; Morton, 1990; Dickman et al., 1993; McKenzie et al., 2006). Many explanations for the extinction of mammals from the Australian arid zone invoke hypotheses of multiple causation (Morton, 1990; Lunney, 2001; Johnson, 2006). For example, Morton (1990) proposed that the combination of drought and the degradation of resource rich refuge habitats by introduced herbivores contributed to the loss of mediumsized mammals (35 5500 g) from the Australian arid zone. A growing body of evidence showing inverse relationships between invasive predator abundance and the abundance of small- and medium-sized mammals (Newsome, Parer & Catling, 1989; Read, 1997; Risbey et al., 2000; Kinnear, Sumner & Onus, 2002) suggests that the refuge concept may be extended to include predation refuges. Such refuges are likely to exist where predation by invasive predators, is relaxed due to predator control or geographic isolation. The mesopredator release hypothesis (MRH) postulates that the elimination of top-order predators 302

results in an increase in the abundance and impact of smaller predators (mesopredators) due to a reduction in intraguild predation and competition within the predator community (Palomeres et al., 1995; Rogers & Caro, 1998; Crooks & Soulé, 1999). Thus according to the MRH, larger predators may provide refuge to the prey of mesopredators, by reducing the frequency of fatal encounters between mesopredators and their prey. Australia s largest predator, the dingo Canis lupus dingo (10 20 kg), was introduced 3000 5000 BP (Newsome & Coman, 1989; Corbett, 1995). In recent years there has been growing awareness of the role that dingoes may play in the conservation of Australian wildlife through its interactions with the red fox (3.5 7.5 kg, Newsome & Coman, 1989) and other invasive species (Glen et al., 2007; Johnson, Isaac & Fisher, 2007; Letnic, 2007). The MRH predicts that the removal of dingoes would result in an increase in red fox populations and subsequently, the negative impact of foxes on native prey species through increased predation. Consistent with the MRH, several studies have found negative relationships between dingo and fox abundance (Newsome et al., 2001; Letnic, 2007) and positive relationships between dingo abundance and the persistence of medium-sized marsupials and native rodents (Smith & Quin, 1996; Johnson et al., 2007; Southgate et al., 2007). In this study, we used an endangered rodent, the dusky hopping mouse Notomys fuscus, as a model species to investigate the relative role of dingoes, rainfall and purported threatening processes in determining the abundance of Australian rodents (Lee, 1995). Notomys fuscus is typical of many extant native rodents in arid Australia, as it displays irruptive population dynamics, is thought to be threatened by interactions with introduced predators and herbivores, and has undergone a major range reduction in the last 100 years (Lee, 1995; Moseby, Brandle & Adams, 1999; Dickman, Lunney & Matthews, 2000). Most recent museum records (o25 BP) of the species occur in areas where dingoes are present (Fig. 1a; Moseby et al., 1999). In addition, a study of the closely related Notomys alexis (body weight 25 45 g), a species of similar body size and dietary habits to N. fuscus (body weight 25 50 g), is vulnerable to predation by foxes (Sinclair et al., 1998). Given our knowledge of N. fuscus and the meso-predator release hypothesis, we predicted that N. fuscus may benefit from the presence of dingoes (Moseby et al., 2006). To test this prediction we conducted snapshot comparisons of N. fuscus abundance in areas where dingoes were common and in nearby areas where dingoes were rare within its historic range. We also used an information theoretic approach (Burnham & Anderson, 2002) to assess the relative support for competing hypotheses to explain the abundance of N. fuscus at the trapping locations. Methods Rationale and study area Notomys fuscus is most common in the sand dunes of the Strzelecki Desert, in the southern arid zone of Australia (a) N 0 200 400 600 (b) Dingo fence Alice Springs 800 1000 km Central, dingoes common South, dingoes common Adelaide Broken Hill North, dingoes common North, dingoes rare Central, dingoes rare South, dingoes rare N 0 100 200 km Figure 1 (a) Map showing locality records (circles) of Notomys fuscus, the dingo fence (hashed) and areas where dingoes are common (shaded). Black circles indicate records of N. fuscus made post-1970 and open symbols represent records made before 1970. Records of N. fuscus were obtained from the South Australian Museum, Victorian Museum, Queensland Museum, Australian Museum, Ellis (1993), Croft, Montague-Drake & Dowle (2007), Waudby & How (2008) and the present study. The rectangle shows the area presented in (b) showing the location of the sub-sites at north (circles), central (squares) and south sites (triangles) on either side of the dingo fence. (Moseby et al., 1999). To conduct our study, we sampled N. fuscus populations and measured indices of processes hypothesised to threaten N. fuscus (Lee, 1995; Moseby et al., 1999, 2006) on either side of the Australian dingo fence in the Strzelecki Desert (Fig. 1b). The sites were located within 303

Table 1 Predictor variables entered into generalised linear models and their hypothesised effect on the abundance of Notomys fuscus Variable Predicted response of N. fuscus abundance Description Units Dingo Positive due to suppression of mesopredators by dingoes Index of abundance at each sub-site Predator activity index Fox Negative due to predation by foxes Index of abundance at each sub-site Predator activity index Cat Negative due to predation by cats Index of abundance at each sub-site Predator activity index Rabbit Negative due to competition for food and habitat alteration by rabbits Dung count on each trapping grid (Piles/transect) Cow Negative due to competition for food and habitat alteration by cattle Dung count on each trapping grid (Piles/transect) Kangaroo Negative due to competition for food and habitat alteration by kangaroos Dung count on each trapping grid (Piles/transect) Sheep Negative due to competition for food and habitat alteration by sheep Dung count on each trapping grid (Piles/transect) Rain 6 Positive due to increased short-term availability of food resources Cumulative rainfall in the 6 months mm before trapping Cover Positive due to increased availability of food resources Percentage cover of grasses and forbs on each trapping grid % Mean rain Positive due to increased year to year availability of food resources Long-term mean annual rainfall mm 30 km of recent (o25 BP) records of N. fuscus obtained from museum collections (Fig. 1), the scientific literature (Fig. 1) and discussions with researchers and landholders. The dingo barrier fence (2 m high, 45000 km long) divides the rangelands of southern Australia (McKnight, 1969). Its function is to exclude dingoes from predominantly sheep grazing lands because dingoes frequently attack and kill sheep (Fleming et al., 2006). The dingo fence crosses the Strzelecki Desert and is impenetrable to dingoes, foxes, kangaroos and livestock. Dingoes are rare on one side of the fence, due to intensive control, and common on the other side where they are only subject to sporadic control (Newsome et al., 2001). Foxes were not subject to intensive control within the study area. Conducting experiments that manipulate large predator populations is rarely possible due to the large spatial extent required for manipulations, and legal, political and ethical considerations. The difference in dingo density on either side of the dingo fence allowed us to conduct a natural experiment with two dingo density treatments (Fig. 1). In order to make comparisons of N. fuscus population densities across the dingo fence we established three study sites, North, Central and South, in the Strzelecki Desert. Each study site consisted of a pair of sub-sites located on pastoral properties that were located inside (dingoes rare) and outside (dingoes common) the dingo fence (Fig. 1). Sub-site pairs were selected for similarity of long-term annual rainfall, landscape and vegetation type. Cattle grazing was conducted at all sub-sites where dingoes were common. At sub-sites where dingoes were rare, cattle were grazed at south site, sheep and cattle at central site and sheep only at north site. At each site we could not intersperse the dingo treatments (dingoes common, dingoes rare) on either side of the dingo fence. The reason for this was that dingoes were rare inside the dingo fence. Thus our treatments were spatially segregated and in a strictly statistical sense our experimental design does not allow inferences concerning the effect of dingoes (Hurlbert, 1984). Because of this potential confounding issue, we used two statistical approaches to test our a priori hypothesis that N. fuscus would be more abundant in the presence of dingoes. We replicated the study at three widely separated sites and pooled the results of our independent tests (Oksanen, 2001) to determine if the overall trend in N. fuscus abundance on either side of the dingo fence was consistent. We used generalised linear modelling to assess the relative support for competing hypotheses (Table 1) to explain the abundance of N. fuscus at the trapping locations and investigated the influence of spatial autocorrelation in our model (Burnham & Anderson, 2002; McAlpine et al., 2006). The Strzelecki Desert is topographically uniform and dominated by sand ridges 3 8 m high, running in a northeast to south-west direction. At our three study sites, the vegetation on dune crests and slopes was dominated by scattered trees and shrubs and perennial under storey vegetation was sparse. Following rain, many ephemeral herbs and forbs are present. The plains between the sand dunes have clay soils with vegetation dominated by ephemeral forbs and short-grasses. Callitris glaucophylla, Casuarina pauper, Acacia aneura and Acacia ligulata and Dodonaea viscosa were the dominant shrubs in the study area. The study area has an arid climate with a mean annual rainfall ranging between 180 and 220 mm (Table 1). The mean annual maximum temperature at the nearest weather station to the study area (Tibooburra, 29.431S, 142.011E) is 27 1C and the mean annual minimum temperature is 13 1C. Maximum temperatures in summer may be 445 1C and in winter minima may be o0 1C(source:AustralianBureauofMeteorology). Elevation above sea level (a.s.l.) in the study areas decreased from north to south and east to west, and ranged between 68 m a.s.l. at south site to 124 m a.s.l. at north site. Trapping At each paired sub-site (inside and outside the dingo fence), N. fuscus were live trapped on eight trapping grids. Simultaneous trapping of both sub-sites at each site was not possible because there are few places where it is possible to cross the dingo fence. Consequently sub-sites at each site were trapped sequentially with 1 or 2 days passing between trapping sessions. To avoid possible fence effects, sub-sites were situated at least 5 km from the dingo fence. Because the South Australian government intensively controls dingoes 304

in a buffer zone along the outside of the South Australian dingo fence, grids at south site located outside the dingo fence were situated at least 35 km from the dingo fence. Within each site, grids were situated at least 1 km apart and to reduce the influence of livestock grazing which is focused around sources of water (Landsberg et al., 2003), most grids were situated at least 2 km from artificial watering points. Each trapping grid covered 0.01 km 2 on the dune crest and adjacent swale and comprised five lines of five Elliott (box) live-capture traps, baited with a mixture of peanut butter, oats and treacle, and two lines of three pitfall traps. All traps were spaced 20 m apart. Pitfall traps consisted of a 10 m drift fence made of aluminium fly-wire, positioned over a length of PVC pipe (16 cm diameter, 60 cm deep), buried level with the ground. Fly-wire was placed underneath the traps to prevent animals from escaping. Trapping was conducted for four consecutive nights at each site. Trapping was conducted at each trapping grid in March/April 2007, September 2007 and November 2007 at north, central and south sites, respectively. A total of 1152 pitfall trap nights and 4800 Elliott trap nights were conducted. To prevent double counting, trapped N. fuscus were given a unique mark with a marker pen. For analyses, an index of N. fuscus abundance was calculated as the total number of captures recorded at each grid during each trapping session. Predictor variables We calculated an index of the activity of dingoes, cats and foxes at each sub-site by establishing and monitoring 25 30 track detection stations at 1 km intervals along low-use, single-lane dirt roads (Engeman et al., 2000). The tracking stations (2 m long) were raked and smoothed to produce a good tracking surface that spanned the road width (c. 3m on average). The tracking stations were monitored and swept daily for three consecutive nights. Each morning a unique mark was made in the corner of each tracking station. This mark was used to determine if wind, rain or vehicles would affect the observer s ability to interpret the plot the following morning. If the mark could not be clearly detected the next morning, the previous night s record from the station was considered invalid. A predator activity index (PAI) was calculated for each predator species (Dingo, Fox and Cat) and expressed as the number of predator track sets (number of incursions) detected divided by the number of nights that the plot was considered valid. The PAI value for each sub-site was calculated as the mean of all track stations. Differences in grazing activity by kangaroos, rabbits, domestic sheep and domestic cattle on each trapping grid were measured using dung counts (Landsberg & Stol, 1996). Groups of fresh herbivore dung (indicated by the presence of a black patina) were counted on three parallel 2m 100 m belt transects on each grid. The transects were spaced 30 m apart. For analyses, indices of kangaroo (Kangaroo), rabbit (Rabbit), cattle (Cattle) and sheep (Sheep) activity, respectively, on each grid were calculated as the mean of the three transects on each grid. Population fluctuations in Australian rodents appear to be driven by recent rainfall events rather than season per se (Dickman et al., 1999). Previous studies have shown that the abundance of N. fuscus and other species of arid dwelling rodents is associated with grass and forb abundance, which in turn reflects primary productivity and the magnitude of recent rainfall events (Letnic, Tamayo & Dickman, 2005; Moseby et al., 2006). We conducted surveys of ground cover vegetation using the step-point method (Landsberg et al., 2003). On each trapping grid, three 80 m transects were sampled at 1 m intervals giving 240 points on each grid. At each point, the ground cover was classified as either bare or vegetated. If vegetation was present, the species were identified and classified as live or dead. For analyses, vegetation cover was calculated as the percentage of points where plants were recorded (Cover). Data on the long-term average rainfall (mm) of each trapping grid (Mean rain) were obtained from a digital model of Mean Annual Rainfall for Australia (source: Australian Bureau of Meteorology). The dataset was constructed using data from the period 1961 1990 using 6000 rainfall stations across Australia and had a positional accuracy of c. 2 km. Cumulative rainfall (mm) received at each sub-site in the 6 months preceding trapping (Rain 6) were derived from a digital model of monthly rainfall for Australia (source: Australian Bureau of Meteorology). The resolution of the grid was c. 25 km. Analyses Comparisons of N. fuscus abundance, at each sub-site, on either side of the dingo fence were conducted using Mann Whitney U-tests. To provide an overall test of the response of N. fuscus to high and low dingo abundance the probability values of the three independent Mann Whitney U tests were combined using Fisher s test for combining probabilities from independent tests of significance (Sokal & Rolf, 1981). This approach is suitable for a small sample size. Previous studies have suggested that the processes threatening mammals in the Australian arid zone may act in concert with one another. Consequently, we analysed a set of predictor variables expected to influence the abundance of N. fuscus (Table 1). Before analyses, assessments of normality indicated that many of the datap had ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi a skewed distribution. A square root transformation ( ðx þ 1Þ) was applied to correct skewed distributions. To allow comparison of model parameter estimates, predictor variables were standardised to have a mean of 0 and a standard deviation of 1. We investigated collinearity of predictor variables using Spearman s rank correlation. If a pair of variables had a correlation coefficient 40.7 they were considered to be proxies of one another and were subsequently never included in the same model (Green, 1979). To reduce the number of predictor variables for a final model we first examined the relationship between predictor variables and N. fuscus abundance for each predictor variable, using univariate generalised linear models (GLM) 305

(McCullagh & Nelder, 1989) with a Poisson s distribution. We ranked each model based on the Akaike information criterion (AIC) (Akaike, 1983). We calculated AIC as AIC ¼ 2L þ 2K ð1þ where L is the marginal negative log-likelihood of the model and K is the number of parameters in the model. We reduced the number of model variables to a manageable subset (n=9) by ranking the univariate models of the predictor variables according to their AICs. We then applied a model averaging approach to account for model uncertainty (Burnham & Anderson, 2002) whereby we constructed a set of alternative models from all linear combinations of the subset of predictor variables. We then ranked these models by their AIC values and calculated the Akaike weight (w m ) for each model (Burnham & Anderson, 2002). We also constructed a 95% confidence set of models by starting with the model with the highest Akaike weight and repeatedly adding the model with the next highest weight until the cumulative sum of weights exceeded 0.95 (Burnham & Anderson, 2002). The direction and magnitude of the effect size of each explanatory variable was based on the model-averaged parameter estimates, calculated using the average of the coefficient estimates from all models, weighted by w m (Burnham & Anderson, 2002). We calculated the uncertainty of the parameter estimate based on the unconditional standard error of the parameter estimate (Burnham & Anderson, 2002). To quantify the relative importance of the predictor variables, we summed the Akaike weight (sw m ) from all model combinations where the variable occurred. We then ranked the predictor variables according to their Sw i, with the larger the weight value, the more important the variable is relative to the others. To determine the independent effect of the key predictor variables, we ranked the effect size of explanatory variables using hierarchical partitioning analysis using the R-package hier.part (Mac Nally & Walsh, 2004) to separate the percentage independent and joint contribution of each variable to the total explanatory power of the model (Mac Nally, 2000). The joint effect measures collinearity between model variables. Highly collinear variables have a large joint effect in relation to their independent effects. Spatial autocorrelation occurs when the value of a variable at any one location in space can be predicted by the values of nearby locations. The existence of spatial autocorrelation indicates that sampling units are not independent of one another. To investigate the influence of spatial autocorrelation in the model, we tested for spatial autocorrelation in the Pearson residuals of N. fuscus abundance at 10 km distance intervals within a 200 km neighbourhood using Moran s index calculated in ROOKCASE (Sawada, 1999), and plotted the output as a correlogram, following McAlpine et al. (2006). We tested each distance interval for significance, using a progressive Bonferroni s correction, whereby the value was calculated by dividing the nominal a of 0.05 by the distance class (Lichstein et al., 2002). The corrected a value was used to calculate a corrected z-value. If this corrected z-value was lower than the z-value of each Moran s I coefficient, then that distance class showed significant spatial autocorrelation. Results Notomys fuscus abundance A total of 118 N. fuscus were captured on 20 of a total of 48 trapping grids. Notomys fuscus were captured at all three sub-sites where dingoes were common and only at one subsite where dingoes were rare (Fig. 2). Comparisons of mean abundance on either side of the dingo fence at each site indicated that N. fuscus was more abundant outside the dingo fence at south (Mann Whitney U, U=4, P=0.001) and central sites (Mann Whitney U, U=7.5, P=0.009) but not at north site (Mann Whitney U, U=28, P=0.317). Overall, the abundance of N. fuscus was greater where dingoes were common (Fisher s test of combined probabilities w 2 =25.53, d.f.=6, Po0.01.). Correlations between predictor variables There was low to moderate collinearity between predictor variables (Table 3). Variables showing the high degree of collinearity were Dingo and Cat, Dingo and Rabbit, Rain 6 and Sheep (Table 2). After taking these collinearities into account, and the AIC value of univariate models, the final subset of predictor variables entered into the global model were: Dingo, Mean rain, Cow, Sheep Fox, Cattle, Rabbit and Kangaroo. Effects of explanatory variables in GLMs The GLM combinations revealed a moderate level of model uncertainty, with 68 models in the 95% confidence set (Sw i =0.95). There were also high levels of parameter uncertainty for several variables (Fig. 3). Despite these levels of uncertainty, there were parameters that had clearly stronger relationships with N. fuscus abundance than other parameters. Dingo and Mean rain had the strongest positive Capture per grid 14 12 10 8 6 4 2 0 North Central South Site Figure 2 Mean captures of Notomys fuscus at three sites contrasting areas where dingoes were present (shaded) and absent (open). Error bars indicate +1 standard error. Significance at the Po0.001 level. Significance at the Po0.01 level. 306

Table 2 Values of predator variables and cumulative rainfall in the 6 months before trapping (Rain 6) at each of the six sub-sites, and mean values ( 1 standard error) of herbivore variables, vegetation cover and long-term annual rainfall of the eight trapping grids at each sub-site Variable Dingo rare Dingo common Dingo rare Dingo common Dingo rare Dingo common Dingo (PAI) 0 0.083 0 0.1 0 0.19 Fox (PAI) 0.17 0.06 0.07 0 0.15 0.07 Cat (PAI) 0 0.06 0 0 0 0.23 Rabbit (piles/transect) 0.3 (0.2) 0.3 (0.2) 0.6 (0.5) 14.4 (4.2) 2.9 (1.2) 18 (4.0) Cow (piles/transect) 0 0.3 (0.3) 0.1 (0.1) 0 0.5 (0.3) 0 Kangaroo (piles/transect) 1.3 (0.9) 0.3 (0.3) 2 (0.9) 0 2.5 (0.6) 0 Sheep (piles/transect) 4.6 (1.9) 0 10.8 (3.9) 0 0 0 Rain 6 (mm) 85.9 63.4 104.8 85.72 58.3 71.5 Cover (%) 64 (3) 44 (2) 42 (3) 52 (5) 18 (2) 19 (2) Mean rain (mm) 182 (0.2) 184 (0.3) 215 (0.8) 210 (0.8) 202 (0.8) 220 (0.6) PAI, predator activity index. Site North Central South Dingo 0.97 ± 0.06 Cover 0.05 ± 0.02 Cow 0.09 ± 0.04 Sheep 0.25 ± 0.06 Rabbit 0.09 ± 0.02 Notomys fuscus abundance Mean rain 1.15 ± 0.08 Kangaroo 0.08 ± 0.03 Fox 0.77 ± 0.07 Figure 3 Path diagram showing the average parameter estimate and the standard error of these estimates for key explanatory variables derived from all combinations of the regression models (n=256). Dashed lines indicate a negative effect, and line width is weighted by the average parameter estimate. Cats were omitted from this model because their abundance was correlated positively with dingo abundance and dingo had a stronger relationship with Notomys fuscus abundance. effect on N. fuscus abundance (Fig. 3). Fox had a strong negative effect (Fig. 3). Cat was not included in the final analysis because it was correlated positively with dingo. Independent effects of explanatory variables Dingo had a strong independent (27.6% of total explained variance) effect on N. fuscus abundance, followed by Cat (27.1%), Rabbit (19%) and Mean rain (15.4%) (Fig. 4).The combined independent effects of the key explanatory variables accounted for 24% of the total explained variation with a 76% combined joint effect, reflecting a moderate level of collinearity between variables (Fig. 4). Of the explanatory variables, Dingo had the highest rank (w i =0.999), followed by Mean rain and Fox (Fig. 5). Spatial autocorrelation The graph of Moran s I plotted against distance showed no significant spatial autocorrelation at the scale at which the study was conducted (Fig. 6). These results indicate that the study grids were independent for the purposes of our analysis. Discussion Our results suggest that N. fuscus populations may benefit from the existence of both resource refuges (mediated by rainfall induced vegetation) and predation refuges (mediated by a top carnivore) in the Strzelecki Desert. Although our study was not experimental and relied upon replicated snapshot observations, our results were consistent with those of previous studies that have reported broadscale relationships between dingoes and the persistence of native rodents (Smith & Quin, 1996; Moseby et al., 1999, 2006). These speculations are discussed below in the context of inter-specific interactions between predators and the responses of rodents to rainfall and introduced herbivores. Overall, N. fuscus was not restricted to areas where dingoes occurred. However, in support of our prediction under the MRH its abundance showed a positive 307

Dingo 27.57% Cat 27.13% Fox 3.92 % Cow 0.07% Sheep 0.02 % Rabbit 18.98 % Kangaroo 1.21% Cover 5.51 % Notomys fuscus abundance Rain 6 15.39 % Mean rain 15.39 % Figure 4 Path diagram showing the independent effects of the key explanatory models on Notomys fuscus abundance. Dashed lines indicate a negative effect. Line width is weighted by the independent effect size. Dingo Mean rain Variable and rank Fox Sheep Rabbit Kangaroo Cow All cover 0 0.2 0.4 0.6 0.8 1 Akaike weight ( W i ) Figure 5 Ranking of the predictor variables to explain Notomys fuscus abundance according to the sum of Akaike weights (Sw i ) for each variable. relationship between long-term rainfall and dingo activity and a negative relationship with fox activity. Dingo activity was also associated with the activity of cats and rabbits but these latter variables were omitted from the final analysis because of their correlations with dingo activity (Table 3). Other variables hypothesised to influence the abundance of N. fuscus including recent grazing by cattle and sheep, kangaroos and recent rainfall (o6 months before trapping) had weak relationships with N. fuscus abundance. Previous studies, including a study from the Strzelecki Desert, have reported inverse relationships between the abundance of dingoes and foxes (Newsome et al., 2001; Letnic, 2007), and cats and foxes (Risbey et al., 2000), respectively. A likely cause for these relationships is that dingoes prey upon foxes and cats, and foxes prey upon cats (Marsack & Campbell, 1990; Paltridge, 2002). Interference competition, whereby smaller predators avoid interactions with larger predators, and thus change their use of habitat and foraging patterns, may also occur (Switalski, 2003; Berger & Gese, 2007). We speculate that a likely consequence of these predatory and behavioural interactions between predator species was the release of N. fuscus from predation by foxes. Studies that have experimentally investigated the role of red foxes in regulating populations of native and introduced native mammals in Australia have found strong evidence that red foxes can suppress populations of rabbits and native rodents (Newsome et al., 1989; Risbey et al., 2000). In this study, cats were only present at detectable densities at two sub-sites where dingoes were present (Table 2) and thus had a positive correlation with N. fuscus abundance. One possible explanation for this relationship between dingo, cat and N. fuscus abundance is that cat populations in the arid zone are driven by spatial and temporal variation in the availability of their primary prey species, rabbits and rodents (Read & Bowen, 2001; Holden & Mutze, 2002; Letnic et al., 2005), more so than the presence of dingoes. The strong relationship between N. fuscus abundance and long-term rainfall suggest that areas that on average receive more annual rainfall serve as resource refuges. In arid 308

environments, the scarcity of water generally limits primary productivity. Rainfall events trigger pulses of increased primary productivity (Ludwig et al., 1997); the magnitude of these growth pulses is dependent on the size of the rainfall event and the availability of seeds and nutrients (Ludwig et al., 1997). Thus areas which, on average, receive more rainfall are likely to provide N. fuscus with more dependable food resources than areas receiving less rain. In the present study, a gradient in mean annual rainfall may explain why the abundance of N. fuscus decreased from north to south at sites where dingoes were present (Fig. 2, Table 2). The population dynamics of arid zone small mammals including N. fuscus typically increase in response to rainfall driven resource pulses (Brown & Heske, 1990; Meserve et al., 2003; Letnic et al., 2005; Moseby et al., 2006). In the present study, recent rainfall (Rain 6) and primary productivity were poor correlates of N. fuscus abundance. Because the emphasis of our snapshot study was on associations between N. fuscus abundance, rainfall and invasive species Moran's I 0.2 0.1 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 Moran's I Z normal Z critical neg Z critical 10 20 30 40 50 60 70 80 90 100 Distance (km) Figure 6 Correlograms showing minimal spatial autocorrelation of Notomys fuscus captures as measured by Moran s index with increased size of the neighbourhood around each trapping grid. 3 2 1 0 1 2 3 Z activity at a large spatial scale, we can only speculate on the factors influencing the temporal dynamics of N. fuscus populations at smaller spatial scales. Four plausible explanations for this poor correlation are: (1) that the magnitude of variation in recent rainfall between sub-sites was small when compared with the temporal variation in rainfall that has been observed to drive rodent population dynamics in other studies (Predavec, 1994; Letnic et al., 2005); (2) the spatial resolution (25 km) of the recent rainfall variable may have been too coarse to identify finer scale patterns that may have existed; (3) predation by red foxes which were more abundant in the absence of dingoes (Newsome et al., 2001; Letnic, 2007) may have dampened the response of N. fuscus populations to recent rainfall events; (4) large herbivores (415 kg) which were abundant in the absence of dingoes (Caughley et al., 1980; Pople et al., 2000) may have depleted the food resources available for N. fuscus, and thus limited the response of N. fuscus populations to resource pulses. This last explanation was supported poorly by our model fitting process, which showed a weak association between N. fuscus abundance and large herbivore activity. Though our study suggests that predation by the red fox may be a more detrimental factor influencing N. fuscus populations than the activity of introduced herbivores, it may not have always been the case. Current rabbit populations in arid Australia are far lower than those reported 70 years ago and even 15 years ago due to the introductions of biological control agents (Wood, 1980; Mutze et al., 2002). Consequently, the possibility that dietary competition with rabbits, habitat degradation by rabbits or hyperpredation (Sinclair et al., 1998) resulting from predator populations being bolstered by abundant rabbits cannot be discounted (Courchamp, Langlais & Sugihara, 1999) and may have contributed to the decline of N. fuscus. It is also likely that rabbit populations may recover if their populations develop resistance to the biological control agents and thus may affect N. fuscus in the future. In addition, livestock densities are generally considered to have been much greater in the late 19th century and early 20th century than they are now (Letnic, 2007). Consequently, the impacts of livestock grazing on the food resources and habitats of N. fuscus may have been much greater in the past. Variation in the cumulative impact of livestock could also have contributed Table 3 Spearman s correlation matrix between predictor variables (n=48) except for paired correlations between dingo, fox, cat and mean rainfall when n=6 Dingo Fox Cat Rabbit Cow Sheep Kangaroo Cover Rain 6 Fox 0.65 Cat 0.72 0.26 Rabbit 0.63 0.31 0.29 Cow 0.19 0.07 0.08 0.28 Sheep 0.56 0.39 0.42 0.35 0.18 Kangaroo 0.59 0.42 0.41 0.20 0.19 0.25 Cover 0.22 0.05 0.38 0.33 0.11 0.38 0.19 Rain 6 0.21 0.06 0.37 0.15 0.19 0.72 0.06 0.56 Mean rain 0.51 0.33 0.32 0.51 0.01 0.10 0.15 0.47 0.15 309

to the patterns we observed (Landsberg et al., 2003). We attempted to reduce this variation between sites and grids by establishing most of our study grids 42 km from stock watering points. Nonetheless, we cannot account for historical differences in cumulative grazing impacts between grids. Dingoes and the conservation of threatened mammals in Australia Large mammalian carnivores are capable of structuring ecosystems through their predatory and competitive interactions with herbivores and smaller predators (Berger et al., 2001; Hebblewhite et al., 2005; Berger, Gese & Berger, 2008). This top-down control of trophic webs can have positive benefits for biodiversity conservation, by reducing the impacts of herbivores and mesopredators (Crooks & Soul e, 1999; Hebblewhite et al., 2005; Berger & Gese, 2007; Berger et al., 2008). In Australia, dingoes appear to be capable of structuring arid ecosystems at large spatial scales by regulating populations of kangaroos, emus (Caughley et al., 1980; Pople et al., 2000) and red foxes (Newsome et al., 2001; Letnic, 2007). Moreover, positive relationships between the presence of dingoes and persistence of mammal species vulnerable to predation by red foxes suggest that the structuring effects of dingoes may be beneficial for biodiversity conservation (Smith & Quin, 1996; Johnson et al., 2007; Southgate et al., 2007). A likely mechanism by which these benefits accrue is that by interacting with foxes, dingoes reduce the predatory impact of foxes on their prey (Johnson et al., 2007). The mass extinction of mammals from the Australian deserts that has occurred in the last 100 years, despite the presence of dingoes (McKenzie et al., 2006), makes it clear that dingoes are not a silver bullet for biodiversity conservation. Nonetheless, the positive effects for biodiversity conservation that have stemmed from the reintroduction of wolves into North American ecosystems (Beyer et al., 2007) or the maintenance of dingo populations (Johnson et al., 2007) in Australia indicate that the ecosystem structuring effects of top-order predators can be incorporated into biodiversity conservation management programmes (Soul e et al., 2003). Consequently, maintaining or restoring dingo populations may be a useful strategy to mitigate the predatory impacts of foxes on small and medium sized native mammals. Acknowledgements Ben Daly, Chris Gordon and George Madani assisted with the fieldwork. Rob and Stacy Davidson, Kelvin Westbrooke, Paul Jonas, Tony Connors, James Morgan and John Osmond provided access to study sites. Peter Bird and David Peacock provided useful discussions in the formulation of the project. Cath Kemper, David Stemmer, Sandy Ingleby, Steve van Dyck, Richard Longmore and Heather Janetzki provided museum records. Adele Haythornthwaite and two anonymous referees provided comments on an earlier version of the paper. This Research was funded by Australian Research Council grant DP0666574. References Akaike, H. (1983). Information measures and model selection. Int. Stat. Inst. 44, 277 291. Berger, J., Stacey, P.B., Bellis, L. & Johnson, M.P. (2001). A mammalian predator prey imbalance: Grizzly bear and wolf extinction affect avian neotropical migrants. Ecol. Appl. 11, 947 960. Berger, K.M. & Gese, E.M. (2007). Does interference competition with wolves limit the distribution and abundance of coyotes? J. Anim. Ecol. 76, 1075 1085. Berger, K.M., Gese, E.M. & Berger, J. (2008). Indirect effects and traditional trophic cascades: a test involving wolves, coyotes and pronghorn. Ecology 89, 818 828. Beyer, H.L., Merrill, E.H., Varley, N. & Boyce, M.S. (2007). Willow on Yellowstone s northern range: evidence for a trophic cascade? Ecol. Appl. 17, 1563 1571. Brandle, R. & Moseby, K.E. (1999). Comparative ecology of two populations of Pseudomys australis in northern South Australia. Wildl. Res. 26, 541 564. Brown, J.H. & Heske, E.J. (1990). Temporal changes in a Chihuahuan desert rodent community. Oikos 59, 290 302. Burbidge, A.A. & McKenzie, N.L. (1989). Patterns in the modern decline of Western Australia s vertebrate fauna: causes and conservation implications. Biol. Conserv. 50, 143 198. Burnham, K.P. & Anderson, D.R. (2002). Model selection and multimodel inference: a practical information theoretic approach. 2nd edn. New York: Springer. Carstairs, J.L. (1974). The distribution of Rattus villosissimus (Waite) during plague and non-plague years. Aust. Wildl. Res. 1, 95 106. Caughley, G., Grigg, G.C., Caughley, J. & Hill, G.J.E. (1980). Does dingo predation control the densities of kangaroos and emus? Aust. Wildl. Res. 7, 1 12. Corbett, L. (1995). The dingo in Australia and Asia. Sydney: University of New South Wales Press. Courchamp, F., Langlais, M. & Sugihara, G. (1999). Cats protecting birds: modelling the mesopredator release effect. J. Anim. Ecol. 68, 282 292. Croft, D.B., Montague-Drake, R. & Dowle, M. (2007). Biodiversity and water point closure: is the grazing piosphere a persistent effect? In Animals of arid Australia: out on their own: 143 171. Dickman, C.R., Lunney, D. & Burgin, S. (Eds). Sydney: Royal Zoological Society of New South Wales. Crooks, K.R. & Soul e, M.E. (1999). Mesopredator release and avifaunal extinctions in a fragmented system. Nature 400, 563 566. Dickman, C.R., Lunney, D. & Matthews, A. (2000). Ecological attributes and conservation of native rodents in New South Wales. Wildl. Res. 27, 347 355. Dickman, C.R., Mahon, P.S., Masters, P. & Gibson, D.F. (1999). Long-term dynamics of rodent populations in arid Australia: the influence of rainfall. Wildl. Res. 26, 389 403. 310

Dickman, C.R., Predavec, M. & Downey, F.J. (1995). Longrange movements of small mammals in arid Australia: implications for land management. J. Arid Environ. 31, 441 452. Dickman, C.R., Pressey, R.L., Lim, L. & Parnaby, H.E. (1993). Mammals of particular conservation concern in the Western Division of New South Wales. Biol. Conserv. 65, 219 248. Ellis, M. (1993). Extension to the known range of the Fawn hopping mouse Notomys cervinus in New South Wales. Aust. Zool. 29, 77 78. Engeman, M., Pipas, M.J., Gruver, K.S. & Allen, L. (2000). Monitoring coyote population changes with a passive activity index. Wildl. Res. 27, 553 557. Fleming, P.J.S., Allen, L.R., Lapidge, S.J., Robley, A., Saunders, G.R. & Thomson, P.C. (2006). A strategic approach to mitigating the effects of wild canids: proposed activities of the Invasive Animals Cooperative Research Centre. Aust. J. Exp. Agric. 46, 753 762. Glen, A.S., Dickman, C.R., Soul e, M.E. & Mackey, B.G. (2007). Evaluating the role of the dingo as a trophic regulator in Australian ecosystems. Aust. Ecol. 32, 492 501. Green, R.H. (1979). Sampling design and statistical methods for environmental biologists. New York: Wiley. Hebblewhite, M., White, C.A., Nietvelt, C.G., McKenzie, J.A., Hurd, T.E., Fryxwell, J.M., Bayley, S.E. & Paquet, P.C. (2005). Human activity mediates a trophic cascade caused by wolves. Ecology 86, 2135 2144. Holden, C. & Mutze, G. (2002). Impact of rabbit haemorrhagic disease on introduced predators in the Flinders Ranges, South Australia. Wildl. Res. 29, 615 626. Hurlbert, S. (1984). Pseudoreplication and the design of ecological field experiments. Ecol. Monogr. 54, 187 211. Johnson, C.N. (2006). Australia s mammal extinctions: a 50 000 year history. Cambridge, UK: Cambridge University Press. Johnson, C.N., Isaac, J.L. & Fisher, D.O. (2007). Rarity of a top predator triggers continent-wide collapse of mammal prey: dingoes and marsupials in Australia. Proc. Roy. Soc. Lond. Ser. B 274, 341 346. Kinnear, J.E., Sumner, N.R. & Onus, M.L. (2002). The red fox in Australia an exotic predator turned biocontrol agent. Biol. Conserv. 108, 335 359. Landsberg, J., James, C.D., Morton, S.R., Muller, W.J. & Stol, J. (2003). Abundance and composition of plant species along grazing gradients in Australian rangelands. J. Appl. Ecol. 40, 1008 1024. Landsberg, J. & Stol, J. (1996). Spatial distribution of sheep, feral goats and kangaroos in Woody Rangeland Paddocks. Rangel. J. 18, 270 291. Lee, A.K. (1995). The action plan for Australian rodents. Canberra: Australian Nature Conservation Agency. Letnic, M. (2007). The impacts of pastoralism on the fauna of arid Australia. In Animals of arid Australia: out on their own: 65 75. Dickman, C.R., Lunney, D. & Burgin, S. (Eds). Sydney: Royal Zoological Society of New South Wales. Letnic, M., Tamayo, B. & Dickman, C.R. (2005). The response of mammals to La Niña (ENSO)-associated rainfall, predation and wildfire in arid Australia. J. Mamm. 86, 689 703. Lichstein, J.W., Simons, T.R., Shriner, S.A. & Franzreb, K.E. (2002). Spatial autocorrelation and autoregressive models in ecology. Ecol. Monogr. 72, 445 463. Ludwig, J., Tongway, D., Freudenberger, D., Noble, J. & Hodgkinson, K. (1997). Landscape ecology function and management: principles from Australia s Rangelands. Canberra, Australia: CSIRO. Lunney, D. (2001). Causes of the extinction of native mammals of the western division of New South Wales: an ecological interpretation of the nineteenth century historical record. Rangel. J. 23, 44 70. Mac Nally, R. (2000). Regression and model-building in conservation biology, biogeography and ecology: the distinction between and reconciliation of predictive and explanatory models. Biodiver. Conserv. 9, 655 671. Mac Nally, R.M. & Walsh, C.J. (2004). Hierarchical partitioning public domain software. Biodiver. Conserv. 13, 659 660. Marsack, P. & Campbell, G. (1990). Feeding behaviour and diet of dingoes in the Nullarbor region, Western Australia. Aust. Wildl. Res. 17, 349 357. McAlpine, C.A., Rhodes, J.R., Callaghan, J.C., Bowen, M.E., Lunney, D., Mitchell, D.L., Pullar, D.V. & Possingham, H.P. (2006). The importance of forest area and configuration relative to local habitat factors for conserving forest mammals: a case study of koalas in Queensland, Australia. Biol. Conserv. 132, 153 165. McCullagh, P. & Nelder, J.A. (1989). Generalized linear models. 2nd edn. London: Chapman and Hall. McKenzie, N.L., Burbidge, A.A., Baynes, A., Brereton, R.N., Dickman, C.R., Gordon, G., Gibson, L.A., Menkhorst, P.W., Robinson, A.C., Williams, M.R. & Woinarski, J.C.Z. (2006). Analysis of factors implicated in the recent decline of Australia s mammal fauna. J. Biogeogr. 34, 597 611. McKnight, T.L. (1969). Barrier fencing for vermin control in Australia. Geogr. Rev. 59, 330 347. Meserve, P.L., Kelt, D.A., Milstead, W.B. & Gutierrez, J.R. (2003). Thirteen years of shifting top-down and bottom-up control. BioScience 53, 633 646. Morton, S.R. (1990). The impact of European settlement on the vertebrate animals of arid Australia: a conceptual model. Proc. Ecol. Soc. Aust. 16, 201 213. Morton, S.R., Short, J. & Barker, R.D. (1995). Refugia for biological diversity in arid and semi-arid Australia. Biodiversity series, paper no. 4. Canberra, Australia: Department of Environment Sport and Territories. Moseby, K.E., Brandle, R. & Adams, M. (1999). Distribution, habitat and conservation status of the rare dusky hopping mouse, Notomys fuscus (Rodentia: Muridae). Wildl. Res. 26, 479 494. 311

Moseby, K.E., Owens, H., Brandle, R., Bice, J.K. & Gates, J. (2006). Variation in population dynamics and movement patterns between two geographically isolated populations of the dusky hopping mouse (Notomys fuscus). Wildl. Res. 33, 223 232. Mutze, G., Bird, P., Kovaliski, J., Peacock, D., Jennings, S. & Cooke, B. (2002). Emerging epidemiological patterns in rabbit haemorrahagic disease, its interaction with myxomatosis, and their effects on rabbit populations in South Australia. Wildl. Res. 29, 577 590. Newsome, A.E., Catling, P.C., Cooke, B.D. & Smyth, R. (2001). Two ecological universes separated by the dingo fence in semi-arid Australia: interactions between landscapes, herbivory and carnivory, with and without dingoes. Rangel. J. 23, 71 98. Newsome, A.E. & Coman, B.J. (1989). Canidae. In Fauna of Australia. Vol. 1B: 993 1005. Walton, D.W. & Richardson, B.J. (Eds). Canberra: AGPS. Newsome, A.E. & Corbett, L.K. (1975). Outbreaks of rodents in semi-arid and arid Australia: causes, preventions, and evolutionary considerations. In Rodents in desert environments: 117 153. Prakash, I. & Ghosh, P.K. (Eds). The Hague: Dr W. Junk. Newsome, A.E., Parer, I. & Catling, P.C. (1989). Prolonged prey suppression by carnivores predator-removal experiments. Oecologia 78, 458 467. Oksanen, L. (2001). Logic of experiments in ecology: is pseudoreplication a pseudoissue? Oikos 94, 27 38. Palomeres, F., Gaona, P., Ferreras, P. & Delibes, M. (1995). Positive effects on game species of top predators by controlling smaller predator populations: an example with lynx, mongooses, and rabbits. Conserv. Biol. 9, 295 305. Paltridge, R. (2002). The diets of cats, foxes and dingoes in relation to prey availability in the Tanami Desert, Northern Territory. Wildl. Res. 29, 389 403. Pople, A.R., Grigg, G.C., Cairns, S.C., Beard, L.A. & Alexander, P. (2000). Trends in numbers of kangaroos and emus on either side of the South Australian dingo fence: evidence for predator regulation. Wildl. Res. 27, 269 276. Predavec, M. (1994). Population dynamics and environmental changes during natural irruptions of Australian desert rodents. Wildl. Res. 21, 569 582. Read, J. & Bowen, Z. (2001). Population dynamics, diet and aspects of the biology of feral cats and foxes in arid South Australia. Wildl. Res. 28, 195 203. Read, J.L. (1997). Stranded on desert islands? Factors shaping animal populations in Lake Eyre South. Global Ecol. Biogeogr. Lett. 6, 431 438. Risbey, D.A., Calver, M.C., Short, J., Bradley, J.S. & Wright, I.W. (2000). The impact of cats and foxes on the small vertebrate fauna of Heirisson Prong, Western Australia II, a field experiment. Wildl. Res. 27, 223 235. Rogers, C.M. & Caro, M.J. (1998). Song sparrows, top carnivores and nest predation: a test of the mesopredator release hypothesis. Oecologia 116, 227 233. Sawada, M. (1999). ROOKCASE: an Excel 97/2000 Visual Basic (VB) add-in for exploring global and local spatial autocorrelation. Bull. Ecol. Soc. Am. 80, 231 235. Sinclair, A.R.E., Pech, R.P., Dickman, C.R., Hik, D., Mahon, P. & Newsome, A.E. (1998). Predicting effects of predation on conservation of endangered prey. Conserv. Biol. 12, 564 575. Smith, A.P. & Quin, D.G. (1996). Patterns and causes of extinction and decline in Australian conilurine rodents. Biol. Conserv. 77, 243 267. Sokal, R.R. & Rolf, F.J. (1981). Biometry. New York, USA: Freeman and Company. Soulé, M.E., Estes, J.A., Berger, J. & Martinez Del Rio, C. (2003). Ecological effectiveness: conservation goals for interactive species. Conserv. Biol. 17, 1238 1250. Southgate, R., Paltridge, R., Masters, P. & Carthew, S. (2007). Bilby distribution and fire: a test of alternative models of habitat suitability in the Tanami Desert, Australia. Ecography 30, 759 776. Switalski, T.A. (2003). Coyote foraging ecology and vigilance in response to gray wolf reintroduction in Yellowstone National Park. Can. J. Zool. 81, 985 993. Waudby, H.P. & How, T. (2008). An additional record of the dusky hopping mouse Notomys fuscus in South Australia. Aust. Mammal. 30, 47 49. Wood, D.H. (1980). The demography of a rabbit population in an arid region of New South Wales, Australia. J. Anim. Ecol. 49, 55 79. 312