Burrowing by translocated boodie (Bettongia lesueur) populations alters soils but has limited effects on vegetation

Abstract Digging and burrowing mammals modify soil resources, creating shelter for other animals and influencing vegetation and soil biota. The use of conservation translocations to reinstate the ecosystem functions of digging and burrowing mammals is becoming more common. However, in an increasingly altered world, the roles of translocated populations, and their importance for other species, may be different. Boodies (Bettongia lesueur), a commonly translocated species in Australia, construct extensive warrens, but how their warrens affect soil properties and vegetation communities is unknown. We investigated soil properties, vegetation communities, and novel ecosystem elements (specifically non‐native flora and fauna) on boodie warrens at three translocation sites widely distributed across the species’ former range. We found that soil moisture and most soil nutrients were higher, and soil compaction was lower, on warrens in all sites and habitat types. In contrast, there were few substantial changes to vegetation species richness, cover, composition, or productivity. In one habitat type, the cover of shrubs less than 1 m tall was greater on warrens than control plots. At the two sites where non‐native plants were present, their cover was greater, and they were more commonly found on boodie warrens compared to control plots. Fourteen species of native mammals and reptiles were recorded using the warrens, but, where they occurred, the scat of the non‐native rabbit (Oryctolagus cuniculus) was also more abundant on the warrens. Together, our results suggest that translocated boodie populations may be benefiting both native and non‐native flora and fauna. Translocated boodies, through the construction of their warrens, substantially alter the sites where they are released, but this does not always reflect their historic ecosystem roles.


| INTRODUC TI ON
As ecosystem engineers (sensu Jones et al., 1994), digging mammals influence geomorphological processes (Butler, 1995) and soil resources, creating shelter for other animals (e.g., Dawson et al., 2019;Whittington-Jones et al., 2011) and affecting vegetation and soil biota (Davidson et al., 2012;Fleming et al., 2014). However, due to complex interactions between the digging species, habitat and soil characteristics, and the extent and longevity of the burrows or warrens, the effects of digging mammals on soil and vegetation properties vary (Mallen-Cooper et al., 2019). For example, vegetation cover, biomass, and species richness are higher on the burrows and warrens of some species, for example, pocket gophers Thomomys talpoides (Grant et al., 1980), pikas Ochotona pallasi (Wesche et al., 2007), and Artic ground squirrels Urocitellus parryii (Wheeler & Hik, 2013), but lower on those of others such as black-tailed prairie dogs Cynomys ludovicianus (Davidson & Lightfoot, 2006) and Daurian pikas Ochotona daurica (Komonen et al., 2003). How the magnitude and direction of the effects of digging mammals on soil resources and vegetation might be altered as environmental conditions change is largely unknown. Burrows could increase in importance as thermal refuges from climate change for other animals (Pike & Mitchell, 2013), and diggings may help rehabilitate degraded areas (Andersen & Macmahon, 1985;Munro et al., 2019). Conversely, digging mammals may accelerate ecosystem change by facilitating the spread or growth of non-native plants (Larson, 2003;Torres-Díaz et al., 2011). To effectively manage terrestrial ecosystems, we need to understand what roles digging mammals play in specific locations, and how they interact with novel (i.e., not previously occurring in that ecosystem) elements and conditions.
In an increasingly altered world, translocations are used to improve the conservation status of threatened species and to return species to areas from which they had become extirpated (IUCN/ SSC, 2013;Seddon et al., 2014). Translocations are also more frequently being used to restore ecosystems by reinstating ecosystem processes regulated by lost species (Palmer et al., 2020). Ecosystem engineers, such as digging and burrowing mammals, may be particularly useful for this kind of translocation because of their ability to fundamentally restructure ecosystems (Seddon, 2010). However, predicting the effects of translocating an ecosystem engineer will require a thorough understanding of the species' historical roles in regulating key processes, such as nutrient cycling and plant recruitment, and insight into how these may be altered by the presence of novel ecosystem elements. Since the majority of terrestrial habitats have been, or will be, altered by human activity (Ellis et al., 2010), understanding interactions between ecosystem engineers and novel ecosystem elements is particularly relevant within a translocation context.
Boodies (Bettongia lesueur) are medium-sized (~1 kg), Australian marsupials that forage for underground food resources, creating numerous, shallow digs in the process (Burbidge et al., 2008;Newell, 2008). Uniquely among macropods, boodies also construct complex warren systems (Sander et al., 1997). Boodies once occurred across most of Australia's semi-arid and arid zones but natural populations are now restricted to three Western Australian islands (Burbidge et al., 2014). Predation by the introduced red fox (Vulpes vulpes) and feral cat (Felis catus), human persecution and habitat degradation are major factors in their decline, which occurred following European occupation of Australia, with the last boodies recorded on the mainland in the 1960s (Burbidge et al., 2014). To redress their decline, boodies have been translocated to a number of additional islands and fenced, mainland reserves. Conservation concerns continue to drive translocation actions for boodies, but most future translocations are also aiming to restore the species' historic ecosystem functions (Palmer et al., 2020). However, our understanding of their roles within ecosystems is largely based on the effects of their foraging diggings. Information on the impacts of their warrens on soils and vegetation is limited to a pilot study (Chapman, 2015) and an investigation of long-abandoned "relict" warrens (Noble et al., 2007). Boodie foraging diggings typically cover less than half a square meter and may persist for several months or, sometimes, years (B. Palmer pers. obs.) while boodie warrens can be more than 90 m in diameter and can persist for decades after abandonment (Burbidge et al., 2007). Because diggings and warrens occur over different temporal and spatial scales, more detailed information about the impacts of their warrens is required to complete our knowledge of boodies' ecosystem functions.
To inform future translocations and to clarify if, and how, translocated boodies direct ecosystem change, we need to quantify how boodie warrens affect ecosystems, and how they interact with novel ecosystem elements. In this study we address four primary questions designed to elucidate the importance of boodie warrens in ecosystems: (a) How does the construction of boodie warrens alter soil properties and ground cover?, (b) How does the construction of boodie warrens alter vegetation communities and plant productivity?, (c) How do novel ecosystem elements, specifically non-native plants and animals, interact with boodie warrens?, and (d) How do the structural characteristics of boodie warrens vary with habitat, and how does this influence their effects on soils and vegetation communities? To answer these questions, we identified boodie warrens at three sites, and measured soil properties, ground cover, and vegetation species richness, cover, composition and productivity.
We also recorded evidence of other vertebrates using the warrens, and quantified warren density, size and activity levels.

| Study sites
Research was conducted at three sites, one in eastern South Australia and two in Western Australia, supporting translocated boodie populations: Yookamurra Wildlife Sanctuary (hereafter "Yookamurra"), Matuwa-Kurarra Kurarra Indigenous Protected Area ("Matuwa") and Faure Island Wildlife Sanctuary ("Faure"; Figure 1). The sites are all former pastoral leases currently managed for conservation and/or cultural purposes but have different climates, habitat types and boodie translocation histories (Table 1). The boodie populations at both Matuwa and Yookamurra inhabit 1,100 ha fenced reserves from which all mammalian predators have been removed; Faure is a 4,500 ha island which is also free from mammalian predators. Boodies formerly inhabited Matuwa and Yookamurra, and at Matuwa relict boodie warrens can still be observed throughout areas with calcareous soils. There is no evidence for the former presence of boodies on Faure, but they were present on the adjacent mainland at the time of European occupation.

| Warren identification
We walked transects to locate warrens constructed by the translocated boodie populations in August 2018 (Yookamurra), and in March (Matuwa) and September (Faure) 2019. An even number of transects, running at right angles from vehicle access tracks, were randomly located in the dominant vegetation types at each property: mallee woodlands on sandy soils ("mallee") and myoporum shrublands on calcareous soils ("myoporum") at Yookamurra, spinifex grasslands on red sands ("spinifex") and acacia shrublands on calcareous soils ("shrub") at Matuwa, and acacia shrublands on red sands ("acacia") at Faure. In total, approximately 20%, 30% and 1% of the areas inhabited by boodies were searched at Yookamurra, Matuwa and Faure respectively. Boodie warrens found along the transects were differentiated from the burrows (i.e., single entrance holes) or warrens (i.e., networks of interconnected burrows) of other co-occurring burrowing species using a combination of warren/burrow morphology and animal tracks and scats. Boodie warren locations were marked on a GPS.
We measured the diameter of each identified boodie warren along, and at right angles to, its longest axis. We also counted the number of active and inactive entrances on each warren. Active entrances showed evidence of recent use including freshly disturbed soil, tracks or fresh scats. Inactive entrances had no fresh animal sign, had debris blocking the entrance, or were partially or completely collapsed. Warrens selected for further study had at least three active burrow entrances, were undisturbed by human interference, were greater than 200 m away from another F I G U R E 1 Our study included three boodie translocation sites spanning a range of environmental conditions and translocation histories. Inset maps show the major habitat types (Faure: green-acacia, blue-all other habitats; Yookamurra: green-mallee, blue-myoporum; Matuwa: green-spinifex, blue-shrub) and location of the studied boodie warrens (black circles) at each site. Inset photographs (credit: B Palmer) show examples of the acacia habitat at Faure, the myoporum habitat at Yookamurra, and the spinifex habitat at Matuwa. Only the fenced enclosures at Matuwa and Yookamurra are shown selected warren and were evenly distributed among the dominant vegetation types at each site. This resulted in slightly different sample sizes for each site: Faure n = 15, Yookamurra n = 20, Matuwa n = 17. We established a paired control plot, of the same dimensions as the warren, 50-100 m from each warren in the same habitat type.

| Soil assessments
Soil assessments, that is, moisture and compaction measurements and sampling to assess nutrient content, pH, and conductivity, were conducted in August 2018 at Yookamurra, March 2019 at Matuwa, and September 2019 at Faure. On warrens, soils were assessed at three microsites: burrow entrances, spoil piles and the intervening matrix ( Figure 2b). Burrow entrances (hereafter "entrances") are the holes through which the boodies enter and exit the warren. Spoil piles ("spoils") are the mounds of soil that accumulate through the excavation of burrows. The intervening matrix ("matrix") is the relatively undisturbed area of ground interspersed between burrow entrances and spoil piles within the warren boundaries. We assessed three replicates for each microsite type at each warren and selected only active entrances and spoils. Three random locations were selected for assessment within each control plot ("controls") by blindly tossing a small pebble from the center of the plot.
We measured soil moisture (%) to a depth of 10 cm using a Hydrosense II soil moisture reader. We measured soil compaction (kg/cm 2 ) using a Geotester pocket penetrometer; we averaged the results of ten compaction readings from each microsite replicate to get the final compaction reading for that replicate, as per the manufacturer's recommendations. We collected a soil sample, from a depth of 10 cm, from each warren microsite and control replicate.
Replicates from each microsite or control were then pooled for analysis. The CSBP Soil and Plant Laboratory conducted standard tests for total carbon (%), nitrate nitrogen (mg/kg), ammonium nitrogen (mg/kg), phosphorus (mg/kg), potassium (mg/kg), sulfur (mg/kg), electrical conductivity (dS/m) and pH (CSBP Lab, 2019).  number and spacing of quadrats for control plots were matched to their paired warren. The area of ground disturbed by mammals and the cover of bare ground, leaf litter and cryptogamic crust were estimated to the nearest whole percent within each quadrat. Plants that had any portion of their foliage within a quadrat were identified to species or genus level where possible and their percent cover estimated. Species that had a total cover estimate of less than 1%

| Vegetation and ground cover sampling
were assigned a value of 0.5% for analysis. Ground and plant cover estimates summed to 100% for each quadrat but area disturbed by mammals was assessed separately (e.g., where leaf litter overlaid patches of cryptogamic crust the area was designated as leaf litter not cryptogamic crust, but where leaf litter overlaid areas of mammal disturbed ground the area was included in both the disturbed ground and leaf litter estimates). Animal scats found within each quadrat were counted and identified to species. While we were conducting the soil and vegetation surveys, we additionally recorded all vertebrate species we observed using a warren, that is, sheltering within, or entering or exiting a burrow. Orthomosaics and digital elevation models were generated from the calibrated images for each warren-control plot pair in Metashape Professional version 1.6.2. The resulting orthomosaics were converted into raster files, the red and near infra-red (NIR) bands separated and normalized difference vegetation index (NDVI) values were calculated for each pixel using the formula:

| Productivity estimates
A new raster file containing the NDVI values for each pixel was generated for each flight.
A k-means unsupervised classification was applied to the NDVI raster files to separate vegetated and nonvegetated pixels using the unsuperClass function in RStoolbox version 0.2.6 for R Statistical Software version 3.6.2 (Leutner et al., 2019). Manual inspection of the resulting geometries confirmed that the classification was highly accurate at assigning pixels containing green vegetation to the correct class. Some grass clumps were incorrectly assigned to the nonvegetated class, but as these were in a dormant state at the time of image collection due to drought conditions (i.e., the above-ground structures were not photosynthesizing), we did not attempt to correct this. Geometries containing only the vegetated pixels were generated for each flight and these were used for assessing the difference in the NDVI values between warren and control plots.

| Statistical analyses
We used nonmetric multidimensional scaling ( was also analyzed separately for the following functional types: large shrubs (shrub species > 1 m tall as adults, including small individuals of these species), small shrubs (shrub species < 1 m tall as adults), grasses, native species, non-native species and dead vegetation.
Differences in the percent cover of plant species that were significantly correlated with either the warren or control plots (see below), and the five most frequently recorded species at each property, were also tested using generalized linear mixed-effects models.
We used correlation indices to identify plant species preferences We conducted separate models for each site (i.e., Yookamurra, Matuwa and Faure) as, given their large geographical separation, we assumed that both soils and vegetation will vary. We used the mean values for each microsite or control for the soil properties, and the mean values for each warren or control for vegetation or ground percent cover.
Plot-pair was used as a random effect in all models to account for the nonindependence of the samples among the microsites (soil variables) and the paired nature of our warren and control plots. Habitat, and the interaction between habitat and microsite or plot, were included as explanatory variables in all models. Data were log-transformed where necessary to meet the model assumptions. Although percent data (i.e., moisture, carbon and percent cover variables) are often arcsine transformed, we found this inadequately transformed our data so used a log transformation after Warton and Hui (2011).

| Soil properties
We found clear distinctions between the soils of the warren and control microsites, and these differences were significant at all three sites (based on nMDS analyses; Faure r 2 = 0.294, p = 0.001; Matuwa r 2 = 0.295, p = 0.001; Yookamurra r 2 = 0.175, p = 0.001). All of the soil variables we assessed were significant contributors to the models for at least two of the three sites. Most soil variables differed between the warrens (particularly the entrance and spoil microsites) and the controls in at least one habitat at each site ( Figure 2).
Although we detected variation in the magnitude of these differences across our three sites, the direction of change was generally consistent, even when the difference was not significant. The warren microsites and the controls varied from each other in a generally consistent pattern: entrances and spoils were often significantly different from the matrix and control microsites, but these pairs were only occasionally significantly different from each other (e.g., matrix and controls rarely differed).
Moisture levels at Faure were very low (only 10% of readings returned a value > 0) and did not differ between the microsites on warrens and the controls (Figure 2, Table A1). At Matuwa, spoils had significantly higher moisture levels than the controls and the other warren microsites in the shrub but were only significantly higher than the matrix microsite in the spinifex habitat ( Figure 2, Table A1).
Moisture levels were significantly higher in the entrances compared to the control and matrix microsites in both habitats at Yookamurra ( Figure 2, Table A1).
Entrances and spoils generally had lower soil compaction than matrix microsites and the controls ( Figure 2, Table A1). This was significant between entrances and both controls and the matrix microsites in all habitats and sites except the mallee habitat at Yookamurra ( Figure 2). Ammonium nitrogen, nitrate nitrogen, potassium and sulfur were significantly higher on the entrances and spoils compared to the controls and the matrix microsites in most habitats and sites ( Figure 2, Table A1). Phosphorus and carbon were significantly  Table A1).

| Ground cover
The amount of ground disturbed by animals was significantly higher on boodie warrens at all three sites (Table A2). Boodie warrens also had more bare ground and this was significant in all habitats except the spinifex at Matuwa (Table A2). Warrens generally had less leaf litter, but this was significant only in the mallee at Yookamurra  Table A2). Yookamurra. The abundance of scat from these species did not differ significantly between the warren and control plots.

| Warren use by other vertebrates
While recording other data we observed several species using the

| Vegetation properties
The nMDS analyses showed that, overall, vegetation properties were similar on boodie warrens and controls at all three sites. All of the vegetation variables we assessed were significant contributors to the models for at least two of the three sites.

| Species richness
Vegetation species richness did not differ between warrens and controls at any of the three sites ( Figure 3, Table A4). We did not record any non-native species at Matuwa so we did not assess differences between native and non-native species for that site. Neither native nor non-native species richness differed between warrens and controls at Faure or Yookamurra (Table A4).

| Cover
There were few differences in the mean percent cover of most plant groups between warrens and controls at any of the three sites ( Figure 3, Table A5). Exceptions to this were significantly higher total vegetation (t 19.3 = −3.59, p = 0.009) and small shrub  Table A5).

| Correlation indices & individual species responses
We found substantial differences in the response of native and non-native plant species to warren presence at Faure and Yookamurra (no non-native species were recorded at Matuwa). A greater proportion of non-native species were positively associated with warrens compared to controls at both Faure (75% c.f. 25%) and Yookamurra (58% c.f. 33%; Figure 4). In contrast, native species did not respond to warren presence at Faure (the same proportions of native species (44%) were positively associated with warrens and controls), and a greater proportion of native species were positively associated with controls rather than warrens at Yookamurra (58% c.f. 33%; Figure 4).

F I G U R E 4
At both Faure and Yookamurra, more non-native plant species were positively associated with (i.e., more abundant and found more commonly at) boodie warrens than control plots. Percentages for each group do not always sum to 100 because some species were associated with both warrens and controls. Data from the two habitat types at Yookamurra are combined

F I G U R E 3 There were few differences in vegetation species richness, percent cover or composition between boodie warrens and undisturbed control plots. Plot shows the results (p values) of generalized linear mixed-effects models comparing vegetation properties between warrens and controls
Six species, three of which were non-native, were significantly associated with warrens (i.e., were more abundant and found more commonly on warrens) across the three sites. Herniaria cinerea, Schismus barbatus, and Sclerolaena obliquicuspis were found more commonly on warrens in the myoporum habitat at Yookamurra, Senna artemisioides and an unidentified grass were significantly associated with warrens in the shrub habitat at Matuwa, and Sisymbrium orientale was found more commonly on warrens at Faure (Table A6).
Native species were more likely to be significantly associated with controls. At Matuwa three species, two native acacias and an unknown grass sp., were found more commonly on the controls and, at Yookamurra, eight species, all native, were found more commonly on the controls (Table A6). No species were significantly correlated with the controls at Faure (Table A6).
Four species, one from Faure and three from Yookamurra, had significantly higher cover on the warrens compared to the controls: These included three annual, non-native species and one native, perennial shrub (Table A6). All four species are known to respond positively to disturbance. Five species, one each from Faure and Matuwa and three from Yookamurra, had significantly higher cover on the control plots. All five species were native species: two perennial shrubs, two annual herbs, and an unidentified grass (Table A6).

| Species composition
There were significant differences in species composition between warrens and controls in the mallee habitat at Yookamurra (r 2 = 0.16, p = 0.05) and the shrub habitat at Matuwa (r 2 = 0.16, p = 0.04; Figure 3). Species composition did not differ in the other habitat types at Matuwa and Yookamurra or at Faure. There were nine species that were recorded only on the warrens and five that were recorded only on the controls at Faure (Table A7). At Matuwa, seven species were recorded only on the warrens and nine only on the controls (Table A7). There were fourteen species recorded only on the warrens and seventeen species recorded only on the controls at Yookamurra (Table A7).

| Productivity
Plant productivity was significantly higher on boodie warrens compared to control plots at all three sites (Figure 3, Table A8). When a buffer zone around the edge of the warren was added, to encompass any potential effects of underground structures extending beyond the disturbance visible at the surface, the difference between the mean NDVI values of warrens and control plots remained significant for Faure (t 11270 = −7.05, p < 0.001) and Yookamurra (t 62060 = 9.73, p < 0.001) but became marginally nonsignificant at Matuwa.
However, the mean NDVI values of warren and control plots were all below zero (indicating most of the vegetation was not photosynthetically active) and differed by less than 0.01 in all cases (Table A8).
The NDVI values of vegetated pixels increased significantly with distance from the center of warrens at Yookamurra (t 31440 = −8.91, p < 0.001) but not at either Faure or Matuwa.

| Warren distribution, size, and activity level
Warren density was far higher at Faure (5.65 warrens/ha) than at either Yookamurra (0.3 warrens/ha) or Matuwa (0.1 warrens/ha; Table 2). The proportion of warrens with at least one active entrance was also highest at Faure (Table 2). Warrens at Matuwa were significantly larger and had significantly more entrances and significantly more active entrances than either Faure or Yookamurra (Table 2).
Warrens at Faure were the smallest, with significantly fewer total entrances than warrens at Yookamurra (Table 2).

TA B L E 2
The density, size, and activity levels of boodie warrens at three sites the shrub habitat at Matuwa, had higher cover on the warrens, and several native vertebrate species, other than boodies, were recorded using the warrens. However, novel ecosystem elements, such as non-native plant species and rabbits, also benefited from the conditions found on boodie warrens, indicating that boodies may contribute to ecosystem changes.

| Burrowing mammals alter soil properties
Overall, we found that soil properties and ground cover often differed between warrens and controls. Unsurprisingly, boodie warrens had significantly more bare and disturbed ground, and tended to have less leaf litter and cryptogamic crust. The magnitude of the differences in the soil properties across our three sites varied but the direction of change was generally consistent. Two of the microsites on boodie warrens, entrances and spoils, tended to have higher moisture and lower compaction levels than controls. Of the soil nutrients we assessed, nitrogen, sulfur and potassium levels tended to be higher, and phosphorus and carbon levels lower, on entrances and spoils compared to the controls. These results are largely consistent with those recorded for other burrowing mammals (Hagenah et al., 2013, e.g., Kerley et al., 2004, Wesche et al., 2007. However, previous studies of relict boodie warrens have recorded higher levels of phosphorus and carbon on warrens (Chapman, 2015;Noble et al., 2007). In contrast to the relatively inactive relict warrens, the high levels of soil turnover on the active warrens in our study most likely reduced soil carbon through increased decomposition rates or the deposition of subsurface soil containing little organic matter on the surface (Kurek et al., 2013;Sherrod & Seastedt, 2001).
We found that matrix microsites on the warrens were largely similar to the controls. This suggests that, rather than selecting warren locations based on their compaction, moisture, or nutrient levels, boodies cause the soil property differences we observed.
Compaction was most likely reduced through the reworking of the soil during burrow construction. Less compact soils may subsequently lead to higher moisture levels through increased infiltration rates (Fleming et al., 2014). Increased soil nutrient levels on burrows and warrens have previously been attributed to the accumulation of urine, feces and decomposing nesting material (Hagenah et al., 2013;Moorhead et al., 1988). We found that boodie scat was 2-14 times more abundant on warrens, and primarily concentrated around entrances and spoils, where nutrient concentrations were also higher. Boodies also accumulate nesting material within their warrens (Stodart, 1966); its decomposition may be enriching the sub-surface soil which is later brought to the surface. Through their deposition of organic material and their construction activities, boodies are significantly changing the compaction, moisture, and nutrient contents of the soils on their warrens.

| Vegetation communities on warrens were similar to undisturbed sites
Many Australian soils are poor in nutrients, especially phosphorus and nitrogen, and this, along with a lack of water, is a primary limiting factor for plants (Smith & Morton, 1990). The higher nutrient and moisture concentrations on boodie warrens may make them favorable sites for vegetation, but, surprisingly, we found that there were few differences between the plant communities on warren and control plots. Many other studies comparing plant communities between mammal burrows and control sites have found significant differences in vegetative cover (e.g., Haussmann et al., 2018;Parsons et al., 2016;Wheeler & Hik, 2013), species richness (e.g., Davidson & Lightfoot, 2008;Hagenah et al., 2013), and composition (e.g., Bryce et al., 2013;Davidson & Lightfoot, 2006;Wesche et al., 2007). Noble et al. (2007) found that uninhabited, Environmental conditions, particularly rainfall, are likely to affect the impact of boodie warrens on vegetation. Moorhead et al. (1988) found, for example, that the cover of spring annuals was significantly higher on banner-tailed kangaroo rat (Dipodomys spectabilis) mounds but only in years of sufficient rainfall. Arid zone Australia is characterized by highly variable rainfall resulting in distinct pulses in plant productivity. It is only in wet years that substantial amounts of annual vegetation are present and perennial plants establish (Morton et al., 2011). Our research was carried out in a time of extreme drought at all three sites, which is likely to have reduced vegetative responses and may have exacerbated trophic impacts.
Long-term monitoring of vegetation on and off boodie warrens may be the only way to obtain a complete understanding of their effect on plant communities.

| Boodie warrens benefit novel ecosystem elements
We found that non-native plant species appeared to benefit from the conditions found on boodie warrens more than native plant species. This may be because the amount of disturbed ground was significantly higher on the warrens and many non-native plants are adapted to disturbance (Liebman et al., 2001). Alternatively, the increased moisture and nutrient levels on the warrens may be more important for non-native species. For example, Ward's weed (Carrichtera annua), which was significantly more abundant on warrens at Yookamurra, will not germinate when soil moisture is very low (Facelli & Chesson, 2008). Browsing pressure on palatable, native species on the warrens could have decreased competition, allowing unpalatable, non-native species to increase. Preferential browsing by black-tailed prairie-dogs (Cynomys ludovicianus) has been shown to increase the cover of non-native forbs at four times the rate of native forbs (Beals et al., 2014). Because boodie warrens appear to be favorable locations for non-native plant species, in areas where there are species of management concern, it may be prudent to control or eradicate these prior to translocating boodies. In situations where this is not possible, boodie warrens could be used as focal points for control of problem species post-release.
At Matuwa, boodie warrens are used extensively by introduced rabbits. Boodie warrens are thought to have facilitated the rapid spread of rabbits throughout arid Australia in the late 19th and early 20th centuries (Parer & Libke, 1985); an example of how native ecosystem engineers can have detrimental effects under altered environmental conditions. While boodies appear to be largely unaffected by the presence of rabbits, more sensitive species that compete directly for food resources may be adversely affected. If translocations of boodies are planned to areas supporting rabbits, it is probably best to eradicate them prior to releasing boodies, to prevent further expansion of the rabbit population.
Boodie warrens are also used by native fauna (Read et al., 2008) and our study added to the number of native species known to benefit from their presence. We opportunistically

| Soil substrate influences warren characteristics
The translocated boodie populations in our study constructed large, complex warrens on the harder soils at Yookamurra and Matuwa (myoporum and shrub habitats) and smaller, simpler warrens in the softer sands (mallee, spinifex and acacia habitats) at all three sites. Warrens constructed by extant boodie populations are most commonly described as large structures created by boodies burrowing into or under caprock or calcrete lenses (Burbidge et al., 2007;Noble et al., 2007;Sander et al., 1997). In the coastal sand on Bernier and Dorre Islands, however, boodies also construct smaller, simpler warrens under shrubs and tussock grasses (Ride et al., 1962). It is likely that boodies once constructed warrens in softer sands throughout their former range, but these warrens would have disappeared not long after boodies were extirpated. Large warrens, such as those that have persisted long after boodies disappeared, appear to be restricted to locations where a calcrete layer or caprock provides long-term structural integrity.
Boodie warren density in our study was likely driven by substrate characteristics, though population density may have also been influential. At Matuwa, and to some extent Yookamurra, the calcrete layer prevents the construction of warrens in many areas but, where opportunities to dig under it exist, it provides structural integrity and allows warrens to expand into substantial structures.
As a result, Matuwa, where the calcrete layer is particularly thick and close to the surface, has the lowest warren density but the warrens there are significantly larger and have significantly more burrow entrances than the warrens at Yookamurra and Faure. A similar pattern of warren density, size and activity level has been recorded in southern hairy-nosed wombats (Walker et al., 2007).
Future boodie translocations should, therefore, consider substrate characteristics when making predictions about warren and population densities.
The impact of warren sizes and density on boodie population structure and dynamics will contribute to shaping the impact that boodies have on the wider landscape. Boodies tend to stay within a few hundred meters of their warren (Finlayson & Moseby, 2004;Short & Turner, 1999 If there is a consistent relationship between boodie warren density and population size it may be possible to use the presence of relict warrens to estimate historic population densities. However, because warrens constructed in sands are likely to degrade quickly, the density of relict warrens will most likely underrepresent historic warren densities. The warren densities we observed at Matuwa and Yookamurra are similar to those recorded in studies of relict warrens (Burbidge et al., 2007;Noble et al., 2007) suggesting Matuwa and Yookamurra boodie population densities may be lower than, or similar to, historic densities. At Faure, the population density is more than ten times that of Matuwa and Yookamurra, possibly because the sandy soils allow unconstrained warren construction. Although there is concern that the population density of several confined boodie populations is too high, early European reports suggest that boodies previously occurred in very high densities (Abbott, 2008) and we hypothesize that current population densities at our three sites are likely to fall within the range of densities previously exhibited by boodies in different habitat and soil types.  (Catano & Stout, 2015). While translocated boodies may not perfectly replicate the roles and effects of historic populations, they are likely to substantially alter the locations in which they are released.

| CON CLUS IONS
Our study shows that, through the construction of substantially sized, numerous, and sometimes long-lived warrens, translocated boodie populations play an important role in structuring ecosystems. Boodie warrens consistently modify soils across a range of habitats and sites by increasing soil moisture and many soil nutrients, and decreasing compaction. Although we did not find substantial impacts on vegetation communities, the higher cover of some plant groups and altered species compositions on warrens in two habitats indicate that warrens do have some effect. Due to the absence of data from prior to boodie population declines, it is impossible to know whether the effects we detected represent the restoration of boodies' historic roles or, conversely, novel impacts and interactions.
Regardless, boodies should be considered ecosystem engineers with the potential to alter soil resources for both plants and animals.

ACK N OWLED G M ENTS
We thank the Australian Wildlife Conservancy (AWC) and the and Nganguraku people as the traditional custodians of the lands on which we conducted our research.

CO N FLI C T O F I NTE R E S T
None declared.

DATA AVA I L A B I L I T Y S TAT E M E N T
Data used in this study are available for download on Dryad (https:// doi.org/10.5061/dryad.w3r22 80pp).

TA B L E A 1
The mean values of soil properties at four microsites on boodie warrens at three sites, Yookamurra, Matuwa, and Faure