REFUGES FOR DECLINING MAMMALS IN DRYLAND AUSTRALIA
Peter McDonald
A thesis submitted to fulfil the requirements for the degree of Doctor of Philosophy
Faculty of Science University of Sydney 2018
DECLARATION
This thesis is submitted to the University of Sydney in fulfilment of the requirement for the degree of Doctor of Philosophy.
The work presented in this thesis is, to the best of my knowledge and belief, original except as acknowledged in the text. I hereby declare that I have not submitted this material, either in full or in part, for a degree at this or any other institution.
Signed:
Peter James McDonald
Date: 29 October/2018
Cover photos: The rugged Mt Sonder in the MacDonnell Ranges (left), and the critically endangered central rock-rat (Zyzomys pedunculatus) (right). Photos by P. McDonald.
ii
AUTHOR ATTRIBUTION STATEMENT FOR PUBLISHED CHAPTERS
Chapter 2. McDonald, P.J., Luck, G.W., Dickman, C.R., Ward, S.J. and Crowther, M.S. (2015). Using multiple‐ source occurrence data to identify patterns and drivers of decline in arid‐dwelling Australian marsupials. Ecography 38 1090-1100.
Chapter 3. McDonald, P.J., Nano, C.E., Ward, S.J., Stewart, A., Pavey, C.R., Luck, G.W. and Dickman, C.R. (2017). Habitat as a mediator of mesopredator‐driven mammal extinction. Conservation Biology 31, 1183-1191.
Chapter 4. McDonald, P.J., Brim-Box, J., Nano, C.E., Macdonald, D.W. and Dickman, C.R. (2018). Diet of dingoes and cats in central Australia: does trophic competition underpin a rare mammal refuge?. Journal of Mammalogy 99, 1120-1127.
Chaper 5. McDonald, P.J., Stewart, A. and Dickman, C.R. (2018). Applying the niche reduction hypothesis to modelling distributions: A case study of a critically endangered rodent. Biological Conservation 217, 207-212.
Chapter 6. McDonald, P.J., Griffiths, A.D., Nano, C.E., Dickman, C.R., Ward, S.J. and Luck, G.W. (2015). Landscape-scale factors determine occupancy of the critically endangered central rock-rat in arid Australia: the utility of camera trapping. Biological Conservation 191, 93-100.
Chapter 7. McDonald, P.J., Stewart, A., Schubert, A.T., Nano, C.E., Dickman, C.R. and Luck, G.W. (2016). Fire and grass cover influence occupancy patterns of rare rodents and feral cats in a mountain refuge: implications for management. Wildlife Research 43, 121-129.
Author contributions: For all data chapter papers I was responsible for the chapter concept, data analysis, and writing, in consultation with my co-authors. For Chapter 3 I co-led the fieldwork with C. Nano and was assisted by many others. For Chapter 4 I oversaw scat collection and undertook cat scat analysis (G. Story undertook dingo scat analysis). For Chapter 5-7 I led the fieldwork with assistance from A. Stewart and many others.
Peter McDonald 29 October/2018
As supervisor for the candidature upon which this thesis is based, I can confirm that the authorship attribution statements above are correct.
Prof. Chris Dickman 29 October/2018
iii
CONTENTS
Acknowledgments...... vi Abstract ...... vii Chapter 1: Introduction ...... 1 1.1 Dryland mammal research ...... 2 1.2 Mammal decline in dryland Australia...... 3 1.3 The refuge concept ...... 5 1.4 Thesis aims ...... 6 1.5 Thesis structure ...... 7 Chapter 2: Using multiple-source occurrence data to identify patterns and drivers of decline in arid- dwelling Australian marsupials ...... 9 2.1 Abstract ...... 10 2.2 Introduction ...... 10 2.3 Methods ...... 11 2.4 Results ...... 14 2.5 Discussion ...... 15 2.6 References ...... 19 Chapter 3: Habitat as a mediator of mesopredator-driven mammal extinction ...... 21 3.1 Abstract ...... 22 3.2 Introduction ...... 23 3.3 Methods ...... 23 3.4 Results ...... 25 3.5 Discussion ...... 26 3.6 Acknowledgments ...... 29 3.7 Supporting information ...... 29 3.8 Literature cited ...... 29 Chapter 4: Diet of dingoes and cats in central Australia: does trophic competition underpin a rare mammal refuge ...... 31 4.1 Abstract ...... 32 4.2 Introduction ...... 32 4.3 Materials and methods ...... 33 4.4 Results ...... 34 4.5 Discussion ...... 36 4.6 Acknowledgments ...... 38 4.7 Supplementary data ...... 38 4.8 Literature cited ...... 38
iv
Chapter 5: Applying the niche reduction hypothesis to modelling species distributions: A case study of a critically endangered rodent ...... 40 5.1 Abstract ...... 41 5.2 Introduction ...... 41 5.3 Methods ...... 42 5.4 Results ...... 43 5.5 Discussion ...... 43 5.6 Acknowledgments ...... 46 5.7 Literature cited ...... 46 Chapter 6: Landscape-scale factors determine occupancy of the critically endangered central rock-rat in arid Australia: The utility of camera trapping ...... 47 6.1 Abstract ...... 48 6.2 Introduction ...... 48 6.3 Methods ...... 49 6.4 Results ...... 51 6.5 Discussion ...... 52 6.6 Acknowledgments ...... 54 6.6 Appendix A. Supplementary data ...... 54 6.7 References ...... 54 Chapter 7: Fire and grass cover influence occupancy patterns of rare rodents and feral cats in a mountain refuge: implications for management ...... 56 7.1 Abstract ...... 57 7.2 Introduction ...... 57 7.3 Methods ...... 58 7.4 Results ...... 60 7.5 Discussion ...... 61 7.6 Acknowledgments ...... 64 7.7 References ...... 64 Chapter 8: Synthesis, gaps and further research ...... 66 8.1 Synthesis and gaps ...... 67 8.2 Further research ...... 71 8.3 References ...... 72 Appendix 1 ...... 81 Appendix 2 ...... 86 Appendix 3 ...... 91 Appendix 4 ...... 110 Appendix 5 ...... 117 Appendix 6 ...... 125
v
ACKNOWLEDGMENTS
Thank you to my supervisor team, Chris Dickman, Gary Luck and Simon Ward, for all your guidance and support and for being genuinely nice people. Chris, thanks for your amazing insights on arid ecology and mammalogy, it has been a privilege studying under your supervision. Gary, you never failed to provide prompt and considered feedback on my questions and chapter drafts and for this I am extremely grateful. Simon, thank you for juggling my work and thesis supervision and for recognising the value of this research for mammal conservation in the NT.
Many people contributed to the fieldwork components of this thesis, thanks for the help and good times (alphabetical order): Eli Bendall, Gareth Catt, Jeff Cole, Edward Connellan, Hollie Gooden, Deon Grantham, Daniel Johansen, Kelly Knights, Daniel McCormack, Catherine Nano, Christie Stenhouse, Alistair Stewart, Claire Treilibs, Mark Trower, and Simon Ward. Thank you Alaric Fisher for supporting my PhD project and seeing the value of this work in providing conservation outcomes for the NT. I acknowledge Chris Day for all the logistic and fieldwork support provided by the Parks and Wildlife Commission of the NT. I also acknowledge the traditional owners of Tjoritja/West MacDonnell National Park for their support of the project.
Thanks friends and colleagues for the stimulating discussion on conservation and other less serious matters, including: Mark Carter, Alistair Stewart, Lauren Young, Edward Connellan, Andrew Schubert, Catherine Nano, Chris Pavey, Hugh McGregor, Brydie Hill, Jayne Brimbox, Tony Griffiths, Eli Bendall, Simon Ward, Glenn Edwards, Jenny Molyneux, and John Read.
Finally, thanks to my family Kelly, Ollie and Fern for the unconditional love and for giving my life balance and perspective, and to my parents Bob and Evelyn for the unwavering support and for introducing me to nature.
vi
ABSTRACT
Australia has a highly distinctive mammal fauna that has been severely impacted by novel threats and disturbances since European colonisation in 1788. Extinctions and declines of mammals have been most pronounced in Australia’s vast drylands, where 11 species no longer occur and many others have contracted to small portions of their pre-1788 distributions. Australian mammals weighing 35-5550 grams have been most vulnerable to extinction and range decline; indeed, all extinctions of dryland mammals have been in this ‘critical weight range’ (CWR). The CWR corresponds with Australia’s larger rodent and mid-sized marsupial species, including conilurine (old endemic) rodents and species in all extant marsupial orders. Ongoing declines demonstrate that Australia’s mammal fauna has not reached a state of 'equilibrium' with the conditions of post-European Australia and suggest that further extinctions are possible.
My research focuses on refuges for declining mammals in dryland Australia. I define refuges as the parts of historic distribution and habitat (i.e. realised niche) that still support declining species. Working under this definition I sought to understand the physical and biological characteristics that create refuges for declining dryland mammals. My thesis comprises six data chapters ordered along a gradient of spatial scale, beginning with multi- species analyses with large geographical extent and coarse grained analyses, through to single- species analyses performed at finer spatial grains.
In chapter 2, I sought to understand how refuges for dryland mammals operate across vast spatial extents. First, I used fauna atlas records to compare the distributions of all currently extant CWR marsupials in Australia’s dryland Northern Territory between pre- and post-1975. For species with evidence of range contraction, I then evaluated alternative hypotheses to explain this contraction (i.e. competition, predation, productivity, climate) using several techniques to improve our confidence in the data. Despite a substantial increase in the number of mammal records across the study region post-1975, the greater bilby (Macrotis lagotis) and desert form of the common brushtail possum (Trichosurus vulpecula) have contracted in distribution by 25 and 40%, respectively. These changes in distribution were best explained by hypotheses of competition and climate-change, respectively. The greater bilby was more likely to occur on land without a history of cattle grazing and with low rabbit densities, while the common brushtail possum has contracted to parts of its distribution that experience cooler
vii maximum temperatures over the hottest months of the year. For five other taxa (including the vulnerable black-footed rock-wallaby Petrogale lateralis) I recorded increases in distribution post-1975, probably reflecting increased survey effort rather than actual range expansion. I concluded that models using atlas data can provide key insights into the patterns and drivers of contemporary species’ declines across vast spatial scales, and thus represent useful tools for conservation.
Through the analyses in chapter 2 I identified the MacDonnell Ranges region as an important refuge for declining mammals. In chapter 3 I focused on this region and sought to understand the factors protecting declining mammals and to understand the drivers of extant mammal assemblages. First, I used sampling data from 90 sites to examine the drivers of extant mammal species richness. I then reconstructed historic mammal assemblages to determine proportional losses of mammal species across broad habitat types (landform and vegetation communities). Predation was supported as a major driver of extant mammal richness, but its effect was strongly mediated by habitat. Specifically, areas that were rugged or had dense grass cover supported more mammal species than the more productive and topographically simple areas. Twelve CWR species were extirpated from the MacDonnell Ranges, and the severity of loss of species correlated negatively with ruggedness and positively with productivity. Based on previous studies, I expect that habitat limits predation from red foxes and feral cats because it affects these species’ densities and foraging efficiency. I concluded that large areas of rugged terrain provided vital refuge for Australian dryland mammals and predicted such areas will support the persistence of CWR species in the face of ongoing mammal declines elsewhere in Australia.
In chapter 4 I consider an alternative hypothesis for the MacDonnell Ranges mammal refuge – that trophic competition between the dingo (Canis familiaris/Canis dingo) and smaller cat (Felis catus) can create refuge from predation for small mammalian prey. To do this I analysed the diets of the two predator species for evidence of potential competition. There was no evidence of exploitation competition between the two carnivores – cats consumed mostly small mammals and particularly larger rodents, whereas the diet of dingoes was dominated by one species of large macropod. There was also no evidence of a shift in the diet of cats, as their diets in refuges and non-refuges were highly overlapping. Consistent with interference competition, cats were the third most frequently consumed mammal species by dingoes. While predation by dingoes could therefore limit densities of cats across the MacDonnell Ranges, this alone does not explain why the most rugged habitats in the region are a refuge for rare
viii mammals. I concluded that habitat complexity most likely underpins the refuge and that possible effects of dingo predation on the cat population would be of secondary importance.
In chapters 5-7 I focused on the refuge requirements of the critically endangered central rock-rat (CRR; Zyzomys pedunculatus). In chapter 5 I took a refuge approach to species distribution modelling by predicting the historic and modern distributions of the CRR. My habitat suitability maps confirmed a dramatic range contraction for this species over the last 100 years. The current association of CRRs with extreme landscape ruggedness supported the hypothesis that the impact of a key threat to the species – cat predation – is mediated by habitat complexity. I detected no CRRs in five new locations predicted to be highly suitable in the refuge model. This highlights the need for in-situ threat management at the three known subpopulations, one of which may already have been extirpated. My map of the CRR's historic distribution identifies potential areas for translocation, including the site of a current translocation proposal into an area surrounded by a predator-proof fence.
My main objective in chapter 6 was to evaluate the utility of camera trapping for sampling CRRs and other small mammals in the rugged mountain refuge. I installed baited camera traps at 50 sites across 1795 ha of core CRR refuge habitat. I recorded all six species of small mammals known previously from this area, including the highly detectable CRR at five sites. I thus established the effectiveness of camera trapping for sampling the CRR and sympatric small mammals, and predicted that camera trapping will become a standard tool for sampling rare small mammals in difficult-to-access environments.
Following the success of camera trapping in chapter 6, I used this sampling tool to investigate the roles of fire history and hummock grass cover on the occurrence of cats and rare rodents in CRR refuge habitat in Chapter 7. I installed 76 camera traps across four sites in the Tjoritja/West MacDonnell National Park and used occupancy modelling to evaluate the influence of recent fire (within five years), hummock grass cover and fine-scale ruggedness on feral cat and rodent occupancy. I found that central rock-rat occupancy was positively associated with areas burnt within the past 5 years – a relationship probably driven by increased food resources in early succession vegetation. In contrast, the desert mouse (Pseudomys desertor) was detected at locations with dense hummock grass that had remained unburnt over the same period. Feral cats were widespread across the study area, although my data suggest that they forage less frequently in areas with dense hummock grass cover. These results suggest that fire management and grass cover manipulation can be used as a tool for rodent conservation in this environment and potentially elsewhere in arid Australia. Specifically,
ix creating food-rich patches within dense hummock grasslands may allow central rock-rats to increase occupancy while simultaneously affording them protection from predation.
In chapter 8 I provide a synthesis of my main findings within the context of ecological theory and provide a methodology on using the refuge approach for declining fauna. I also identify data gaps and key areas for further research. The refuge approach to understanding mammal decline takes the perspective of a declining species and can help to understand drivers of decline. A fundamental component of the refuge approach is the comparison of historical and recent distributions and niche spaces of declining fauna. The approach offers several advantages over analyses of single time periods. First, the comparison of the two time periods can provide inference on drivers of decline in distribution and niche space and provide data for conservation assessment. Understanding drivers of decline is a prerequisite for implementing actions to reverse this decline. Second, predictions about the historical distribution can guide conservation practitioners on where to direct habitat restoration and reintroduction programs. Finally, models of modern distribution and niche space can be used to direct surveys for locating new populations of declining species. I hope that the information and framework developed in my research can be used to improve the status and outlook for my study species and be applied to declining fauna outside of dryland Australia.
x
Chapter 1. Introduction
My primary study area in the MacDonnell Ranges west of Alice Springs (photo by P. McDonald)
1 This thesis describes the biology, status and threats to selected threatened species of Australian desert-dwelling mammals. These species are confined to a small number of sites, or refuges, that are discussed further, in general terms, below. The importance of refuge use by threatened species has been emphasised in recent studies in Australia and overseas, and the present work has been designed to further our understanding of this key topic area.
1.1 Dryland mammal research
About 45% of the Earth’s land surface is characterised by low rainfall and high evaporation, and these arid or dryland regions together comprise the Earth’s largest biome (Schimel 2010). There are two characteristics associated with precipitation that are largely unique to dryland ecosystems. First, the amount of precipitation is so low, and evaporation so high, that the resultant lack of free water controls biological processes and the availability of resources (Noy-Meir 1973). Second, precipitation is highly variable in time and space, with a random or stochastic element to this variability (Noy-Meir 1973). These characteristics have a profound influence on the biota of dryland systems (Ward 2016), especially mammals, which have relatively high water turnover and energetic demands that result from homeothermy and—with the exception of bats—limited ability to track and exploit shifts in resources across variable dryland landscapes (Degen 1997; Ward 2016).
Due to the extreme environmental conditions of dryland systems, a considerable amount of mammalian research has focused on physiology. Researchers have long been interested in understanding how dryland mammals – ranging from those weighing just a few grams through to 600 kg camels – cope with extreme temperatures, water loss and unpredictable water supply. Research has uncovered a range of physiological and behavioural adaptations that permit dryland survival, including high dehydration and temperature tolerance, kidney adaptations that enable the production of highly concentrated urine, reduced evaporative water loss, torpor, burrowing and nocturnality (Schmidt-Nielsen 1964; Walsberg 2000).
Ecological research in dryland systems has made substantial contributions to ecological theory. This includes pioneering work and research demonstrating the common theoretical ground between dryland systems and more predictable environments. Most notably, research on rodents in the deserts of North America and Israel has advanced our understanding of habitat selection (Rosenzweig and Winakur 1969; Brown 1988), foraging theory (Brown et al. 1994), competition and mechanisms for species coexistence (Brown and Lieberman 1973; Price 1978;
2 Kotler and Brown 1988), and predation risk (Kotler 1984; Price et al. 1984). Long-term ecological studies in North America’s deserts have added a temporal element to this early fundamental research and have demonstrated complex relationships between rainfall, vegetation and rodent assemblages (Brown and Ernest 2002; Ernest et al. 2016). Granivorous rodents have been the focus of much of this fundamental research because they provide a model system for assessing the effects of food limitation and interspecific competition within guilds of ecologically similar species.
As a result of unpredictable resource availability, a substantial component of the world's dryland faunas is characterised by irruptive population dynamics. Again, dryland rodents have been a focus of much research and are among the best-known irruptive fauna within the vertebrates. Much of our understanding of irruptive population dynamics comes from long- term studies in Chile and Australia. In arid systems on both continents, irruptions of rodents predictably follow irregular large rainfall events and the resulting pulses of primary productivity (Dickman et al. 1999; Lima et al. 1999; Letnic and Dickman 2010; Meserve et al. 2016). Also common to both systems are population increases of predators in response to the increased availability of rodent prey (Lima et al. 2002; Pavey et al. 2008). However, an important distinction between Australian and Chilean dryland systems – and indeed, between Australian and dryland systems globally – is that much of the predator numerical response in Australia comprises responses by three introduced carnivores; this has had a particularly profound impact on the Australian dryland mammal fauna (see below).
1.2 Mammal decline in dryland Australia
Australia has a highly distinctive mammal fauna that has been severely impacted by novel threats and disturbances since European colonisation in 1788. At least 31 endemic mammal species have become extinct and a further 56 species qualify as threatened under International Union for Conservation of Nature (IUCN) Red List criteria (Woinarski et al. 2015; Travouillon and Phillips 2018). Extinctions and declines of mammals have been most pronounced in Australia’s vast drylands (~70% of mainland Australia), where 11 species no longer occur and many others have contracted to small portions of their pre-1788 distributions (McKenzie et al. 2007; Woinarski et al. 2015).
Australian mammals weighing 35-5550 grams have been most vulnerable to extinction and range decline (Burbidge and McKenzie 1989; Chisholm and Taylor 2007); indeed, all
3 extinctions of dryland mammals have been in this ‘critical weight range’ (CWR) (Woinarski et al. 2015). The CWR corresponds with Australia’s larger rodent and mid-sized marsupial species, including conilurine (old endemic) rodents and species in all extant marsupial orders. Johnson (2006) identified a series of decline and extinction events following European colonisation. Relevant to dryland Australia, these include: 1880-1920 when several rodent species were last recorded from the drylands; 1930-1939 when small wallabies, large rodents and at least one small bandicoot disappeared from southern inland regions; and post-1939, which saw the demise of rat-kangaroos, bandicoots and small wallabies from dryland regions.
The post-1939 declines of dryland mammals have continued until very recent times. Losses include the extirpation of the last wild mainland population of the mala (Lagorchestes hirsutus) from the Tanami Desert in 1991 (Johnson et al. 1996), the ongoing range contraction and regional extirpation of the common brushtail possum (Trichosurus vulpecula vulpecula) (Kerle et al. 1992), and a site extirpation of the central rock-rat (Zyzomys pedunculatus) in 2012 (McDonald et al. 2015). These ongoing declines demonstrate that Australia’s mammal fauna has not reached a state of 'equilibrium' with the conditions of post-European Australia, and suggest that further extinctions are possible.
The primary factor implicated in the decline and extinction of Australia’s dryland mammal fauna is predation by the introduced cat (Felis catus) and European red fox (Vulpes vulpes) (Woinarski et al. 2015). This contrasts with the global situation, where habitat loss and hunting are the main factors threatening mammals (Hoffman et al. 2011). Correlative and experimental evidence supporting the predation hypothesis includes: the CWR corresponds to the preferred prey size range for cats and foxes (Dickman 1996; Pearre and Maass 1998; Radford et al. in press); geographic regions with a lower proportion of arboreal, rock-dwelling or burrowing taxa (i.e. taxa whose habits reduce their accessibility to introduced predators) have lost relatively more species than other regions (McKenzie et al. 2007); ground-dwelling CWR mammals have been more likely to decline than CWR arboreal species (Johnson and Isaac 2009); experimental removal of foxes consistently results in the recovery of black-footed rock- wallaby (Petrogale lateralis) colonies (Kinnear et al. 2010) and in increases in populations of many other similar sized marsupials (Dexter and Murray 2009; Saunders et al. 2010); cat and fox predation frequently results in the failure of attempts to reintroduce CWR mammals (Moseby et al. 2011); and CWR mammals invariably thrive within predator-free fenced exclosures (Hayward and Kerley 2009; Dickman 2012). While dryland Australia has not been subject to broad scale land-clearing and habitat loss, it is likely that there have also been
4 important interactions between cat and fox predation and wildfire (Leahy et al. 2015; McGregor et al. 2015), as well as with competition from introduced livestock and rabbits (Johnson 2006).
1.3 The refuge concept
In biology and ecology there has been considerable conflation of the terms ‘refuge’ and ‘refugium’ (or its plural ‘refugia’). However, a commonly applied temporal distinction is that refuges operate over decadal scales, while the term refugium is applied to evolutionary and climatic processes operating over millennia (Keppel et al. 2012; Davis et al. 2013; Pavey et al. 2017). Based on this temporal distinction, the term ‘refuge’ has been variously defined in biology and ecology and examples of definitions have included: ‘enemy-free-space’, microhabitat shelter and cover for individual animals, den sites for aestivation and hibernation, fire ‘shadows’ that escape wildfire, and locations where pest species do not occur (Berryman and Hawkins 2006; Robinson et al. 2013). Given that the ‘refuge’ concept encompasses many divergent phenomena, Berryman and Hawkins (2006) advocated that it be considered an important integrating concept in ecology.
A frequent application of the refuge concept is with respect to predator-prey dynamics. Specifically, in population ecology the predation refuge (or ‘enemy-free-space’) is a stabilising force that enables predator-prey coexistence (Rosenzweig and MacArthur 1963; Hassell 1978; Hochberg and Holt 1995). The predation refuge was probably first demonstrated empirically in Gause’s (1934) classical protozoan predator-prey experiments. Gause observed that Paramecium prey could enter the sediment at the base of his cultures, whereas the Didinium predator could not – thus the sediment provided refuge for the prey.
In dryland Australia the predation refuge concept has been applied to several species of threatened mammals. For example, it has been observed that the endangered dusky hopping- mouse (Notomys fuscus) occurs at higher abundance where fox activity is low (Letnic et al. 2009). The refuge concept has also been extended beyond the predator-prey relationship to include, for example, locations where moisture and nutrient availability enable the persistence of animal species through drought (Morton et al. 1995). In contrast, Pavey et al. (2017) approached the refuge concept from the perspective of irruptive species, specifically dryland rodents, and defined the refuge as: “the subset of the potential range of an irruptive species where a viable population persists during the low phase of a population cycle (i.e., bust phase)”
5 (p. 650). The authors defined two main types of refuge for irruptive species: ‘fixed refuges’ and ‘shifting refuges’, differentiated on whether a species' use of a refuge is stable or unstable between irruptions and within the bust phase. Pavey et al. (2017) noted that different species use refuges in different ways, and advocated building refuge models that were focused around knowledge of species' spatial occurrence and autecology.
1.4 Thesis aims
Many of those concepts explored by the pioneering dryland rodent researchers in North America – including habitat selection, foraging theory, competition, and predation – are central to my refuge definition (see below) and are examined throughout my thesis chapters. I also consider how ‘boom’ and ‘bust’ population dynamics operate in the context of refuges for a critically endangered irruptive rodent. Therefore, my thesis integrates various aspects of ecological theory within the urgent context of understanding how refuges operate for declining mammals and what can be done to expand or protect their populations.
Here, taking a similar spatial approach to that used by Pavey et al. (2017) but focusing on declining mammals, I define refuges as ‘the subset of distribution and habitat occupied by a species at the time of European colonisation that is currently occupied’. Importantly, this definition does not predefine a mechanism(s) for explaining species' persistence, leaving the door ajar for testing multiple hypotheses and multi-causality. I do, however, assume that refuge habitat meets all those fundamental requirements for species’ persistence (e.g., thermal limits, energy requirements, shelter), as well as providing a buffer from those factors that have driven decline and extirpation elsewhere. In this way my definition of refuges builds on Hutchinson’s (1957) seminal work on the ecological niche and the concept that a species current ‘realised’ distribution is the product of both physiological tolerances and biotic interactions.
My study region was dryland Australia, the vast region of the continent where mammal declines have been particularly severe. Within this region most of my thesis is focused on the MacDonnell Ranges in the southern Northern Territory. These ancient and rugged mountain ranges contrast starkly with the surrounding sandy deserts and had already been identified as a significant evolutionary refugium for plants and snails (Morton et al. 1995). While reasonably well-sampled biologically, the mammal fauna of the MacDonnell Ranges has received comparatively little research attention, particularly compared with the Simpson Desert (see Dickman 2013) or Strzelecki Desert (Gordon et al. 2017) further to the east. At the commencement of my PhD there were indications that the MacDonnell Ranges were an
6 important refuge for mammals; it was the last region known to support the critically endangered central rock-rat (Zyzomys pedunculatus), as well as the last population of the common brushtail possum in dryland Australia. However, there had been no investigation into the physical and biological characteristics that enable these species to persist in the MacDonnell Ranges but not elsewhere in the central arid region. More broadly, there had been no attempts to model which factors define refuges for declining mammals across the vast regions of dryland Australia. The overall aims of my thesis are therefore to:
1) Understand the characteristics that define refuges for declining mammals across the broad, vast spatial scales of Australia’s drylands;
2) Determine the physical and biological characteristics that have led to the MacDonnell Ranges being a refuge for a suite of declining mammals; and
3) Model and define the refuge habitat characteristics for the central rock-rat and uncover the drivers of its ongoing decline.
In each of the thesis chapters containing primary field data or other results (see 1.5, below), I use these broad objectives to define specific, testable hypotheses to guide my investigations.
1.5 Thesis structure
My thesis comprises six data chapters (Chapters 2-7), all of which have been published recently in peer-reviewed scientific journals. I have ordered these chapters along a gradient of spatial scale, beginning with multi-species analyses with vast geographical extent and coarse grained analyses, through to single-species analyses performed at finer spatial grain (Figure 1.1). Two data chapters have a primarily applied focus (Chapters 6 and 7), whereas the remainder contribute more subtantially to ecological theory (Chapters 2-5). After presentation of the data chapters, I include a concluding chapter that synthesizes and provides perspective on the major findings of the study, as well as suggestions for future research, conservation and management.
Please note that all references for this chapter and Chapter 8 are listed at the end of Chapter 8. All other references are included at the ends of each chapter as part of the paper in which they were published.
7
Fig. 1.1. Overview and spatial scale of the six data chapters in this thesis.
8 Chapter 2. Using multiple-source occurrence data to identify patterns and drivers of decline in arid-dwelling Australia marsupials
A storm over Tnorola on the Missionary Plain, the transition between the mountain ranges and sandy deserts (photo by P. McDonald)
Author contribution: I was responsible for the chapter concept, data analysis, and writing, in consultation with my co-authors.
9 Ecography 38: 1090–1100, 2015 doi: 10.1111/ecog.01212 © 2015 The Authors. Ecography © 2015 Nordic Society Oikos Subject Editor: Hector Arita. Editor-in-Chief: Miguel Araújo. Accepted 6 January 2015
Using multiple-source occurrence data to identify patterns and drivers of decline in arid-dwelling Australian marsupials
Peter J. McDonald, Gary W. Luck, Chris R. Dickman, Simon J. Ward and Mathew S. Crowther
P. J. McDonald ([email protected]), C. R. Dickman and M. S. Crowther, Desert Ecology Research Group, School of Biological Sciences, Univ. of Sydney, NSW 2006, Australia. Present address of PJM: c/o Flora and Fauna Division, PO Box 1120, Alice Springs, NT 0871, Australia. – S. J. Ward and PJM, Flora and Fauna Division, Dept of Land Resource Management, Alice Springs, NT 0870, Australia. – G. W. Luck, Inst. for Land, Water and Society, Charles Sturt Univ., Albury, NSW 2640, Australia.
Many of the world’s terrestrial mammal species are imperilled, but recent extinctions and declines have been most severe in Australia. In particular, arid-dwelling marsupials in a critical weight range (35–5500 g) have declined dramatically following European settlement. In the absence of long-term monitoring, documenting these declines or distribution shifts and their causes often relies on occurrence data from multiple sources. Using atlas records, we compared the distributions of all currently extant marsupials in the critical weight range in Australia’s arid Northern Territory pre- and post-1975. For taxa with evidence of range contraction, we evaluated alternative hypotheses to explain this contraction (i.e. competition, predation, productivity, climate) using several techniques to improve our confidence in the results. Despite a substantial increase in the number of mammal records across the study region post-1975, the bilby Macrotis lagotis and desert form of the common brushtail possum Trichosurus vulpecula appear to have contracted in distribution by 25 and 40%, respectively. These changes in distribution were best explained by hypotheses of competition and climate-change, respectively.Macrotis lagotis was more likely to occur on land without a history of cattle grazing and with low rabbit densities, while T. vulpecula has contracted to parts of its distribution that experience cooler maximum temperatures over the hottest months of the year. For five other taxa (including the vulnerable black-footed rock-wallaby Petrogale lateralis) we recorded increases in distribution post-1975, probably reflecting increased survey effort rather than actual range expansion. We conclude that models using multiple-source occurrence data can provide key insights into the patterns and drivers of contemporary species’ declines, and represent useful tools for conservation.
Many of the world’s terrestrial mammals have undergone vulnerability of mammals in a ‘critical weight range’ (CWR) contemporary range contractions and are imperilled with of 35–5500 g is well established (Burbidge and McKenzie extinction (Ceballos and Ehrlich 2002). Twenty-seven per- 1989, Chisholm and Taylor 2007). The CWR applies partic- cent of land mammal species are classified as threatened, ularly to terrestrial arid-dwelling species and may not extend with 3.5% of these critically endangered and facing a high into closed forest environments or Australia’s tropical north probability of imminent extinction (Schipper et al. 2008). (Johnson and Isaac 2009, Fisher et al. 2014). The loss of The dominant threats to land mammals are habitat loss or CWR species supports the hypothesis that predation by the degradation and hunting (Schipper et al. 2008), although introduced domestic cat Felis catus and red fox Vulpes vulpes invasive species are also an important threat, particularly is a factor contributing to the declines. The cat and fox are for smaller mammals that occur on islands (Clavero and important predators of mammals, and the upper end of Garcia-Berthou 2005, Fisher and Blomberg 2011). the CWR coincides with the upper limit of prey taken by Australia has seen the most dramatic of recent terrestrial V. vulpes (Dickman 1996). Further support for the preda- mammal declines, losing up to 29 species since European tion hypothesis includes the observation that arboreal and colonisation in 1788, including several taxa from offshore rock-dwelling mammals within the CWR have been more islands (Woinarski et al. 2014). At least 39 of the country’s resilient to decline, with their favoured habitats probably remaining mammal species occupy small fractions of their affording some level of protection from predation (McKenzie original distribution, and eight are now restricted to offshore et al. 2007, Johnson and Isaac 2009). islands (Burbidge et al. 2008). Within Australia, mammal In arid Australia, another major factor implicated in declines have been most pronounced in the arid regions, the decline of endemic terrestrial mammals is a reduction with arid endemics comprising the majority of extinct taxa in habitat suitability and productivity associated with the (Short and Smith 1994, McKenzie et al. 2007). introduction of livestock (sheep and cattle) and European Although the drivers of Australia’s contemporary rabbits Oryctolagus cuniculus (Burbidge and McKenzie 1989, mammal declines have not been conclusively established, the Morton 1990). In a framework for arid Australian ecology,
1090 10 Stafford-Smith and Morton (1990) proposed that medium- ‘arid zone’ definition of Williams and Calaby (1985), which sized native mammals were largely restricted to more pro- includes areas that receive 250 mm of annual rainfall in the ductive landforms and, because these areas are also typically south and 500 mm in the north, where rainfall is highly favoured by rabbits and for livestock production, they are seasonal. We chose the NT because it encompasses all major the most degraded environments in arid regions. Although landform types within Australia’s arid zone (Morton et al. the broad applicability of this framework has been criticised 2011) and includes the core areas of distribution for (Southgate et al. 2007), subsequent continent-scale studies several threatened marsupials in the CWR, including the bilby have found that environmental change associated with live- Macrotis lagotis, the desert form of the common brushtail stock and rabbits was important in explaining variation in the possum Trichosurus vulpecula vulpecula (hereafter referred to attrition of Australian mammals, with increased attrition cor- as Trichosurus vulpecula), and the black-footed rock-wallaby relating with grazing by livestock and higher rabbit densities Petrogale lateralis (Kerle et al. 1992, Gibson 2000, Southgate (Johnson et al. 2007, McKenzie et al. 2007). Another factor et al. 2007). The NT also has a government-administered, sometimes implicated in the demise of Australia’s mammals is multiple-source, occurrence fauna database, the NT Fauna altered fire regimes (Burbidge and McKenzie 1989, Burbidge Atlas, comprising vertebrate fauna records from museum et al. 2008), although there is little evidence in support of this collections, government departments, ecological consultants, (Clarke 2008). Climate change also has been implicated, but community groups, and the public (< www.lrm.nt.gov.au/ evidence again is speculative (Letnic and Dickman 2006). plants-and-animals/information-and-publications/nt- An important reason why the drivers of mammal declines fauna-observations >). Records are vetted for reliability by in Australia remain uncertain is that long-term monitor- experienced NT government ecologists before being added ing data are scant, a situation mirrored in many locations to the database. worldwide (Magurran et al. 2010, Lindenmayer et al. 2012). The database contains 903 291 fauna records dating back In the arid zone, the longest-running mammal monitoring to the mid-1800s that documents the occurrence of 984 programs were established following the most recent extinc- vertebrate animal species. It is the only source of information tion events and are located within a single biome, the hum- for long-terms trends in mammal distribution in the region. mock grasslands (Dickman 2013). Although these programs have produced substantial insight into arid-zone ecology and mammal population dynamics (Dickman 2013), many Evaluating range contraction, and data preparation arid-dwelling taxa occur outside the hummock grasslands and there are limits on the inferences that can be made We included marsupial occurrence records dating back beyond this biome. to 1901 in our analysis. These records were split between Information on changes in the distribution of mammals pre-1975 (including the year 1975) and post-1975 (Fig. 1b). can be obtained from museum collections and multiple-source We used 1975 as a cut-off point for comparing the histori- occurrence databases (Shaffer et al. 1998, Griffioen and Clarke cal and recent distributions of marsupials because 1975 just 2002, Luck et al. 2010). Such databases contain errors (e.g. precedes the last of arid Australia’s contemporary mammal geographic bias, observer and spatial errors), although many of extinctions (Burbidge et al. 1988), coincides with a shift these can be accounted for and minimised (McCarthy 1998, towards increasing temperatures across much of arid Australia Ponder et al. 2001, Aizpurua et al. 2015). Importantly, because (Collins et al. 2000, Hughes 2003) and is just after publi- these databases are frequently the only sources of records span- cation of the first checklist of terrestrial mammals for the ning the timeframes and geographic extent required to char- Northern Territory that brought together much of the infor- acterise temporal and spatial changes in species’ distributions, mation on known locations of species at the time (Parker they are currently the best means of potentially uncovering the 1973). Further, the date is close to 1974 which equalises drivers of these trends (Shaffer et al. 1998). time averaging (1956) and median sampling effort (1992). Here, we focus on arid Australian marsupials in the CWR Seven marsupial species in the CWR are currently extant in and use a multiple-source occurrence fauna database to: 1) the arid NT (Table 2). Occurrence data for the two mulgara determine whether any species have undergone recent con- species (Dasycercus blythi and D. cristicauda) were pooled tractions in distribution or range shifts, and 2) identify the due to the taxonomic uncertainty of many of the records broad-scale drivers of any contractions or shifts. Our work (Pavey et al. 2012, Woolley et al. 2013). For all species, we includes occurrence records spanning 11 decades from 1901 summed the number of records pre-1975 and post-1975 and to 2013, and draws on long-term occurrence data to exam- estimated the minimum extent of occurrence for each spe- ine contemporary mammal declines. The results have impli- cies in ArcMAP 10.2 by creating a polygon encompassing all cations for land management, especially livestock grazing records, using the Convex Hull in the Minimum Bounding in arid regions, and uncover potentially negative impacts of Geometry tool. Polygons were projected to Lamberts GDA94 climate change for a particular mammal species. prior to area calculation and all subsequent spatial work was in this datum. We undertook further analysis individually on any species Methods whose distribution covered less area (contraction) or more area (expansion) area post-1975 compared with pre-1975, Study region and atlas data applying a minimum threshold of 10%. We produced a grid layer of the arid NT with each grid square covering 625 We used the 993 729 km2 arid area of Australia’s Northern km2. This resolution was used to overcome finer-scale loca- Territory (NT) as our study region (Fig. 1a). We follow the tion errors present in the fauna atlas (e.g. location estimates
1091 11 Figure 1. (a) Location of the study area, the arid Northern Territory (grey), in relation to major roads and towns, and (b) the change in the total number of mammal records in the arid NT between the periods pre- and post-1975. from historical accounts, pre-GPS records) and because we equal or greater number of records of background species were interested in the broad-scale drivers of range contrac- pre-1975 than post-1975, and the ‘higher’ confidence level tion (Table 1). Grid cells were assigned the binary form ‘pres- was as per the lower confidence level but with a minimum ent’ (1) or ‘absent’ (0). Present was defined as any grid cell of five background records pre-1975. We applied a threshold within which the species was recorded post-1975. Absent whereby a minimum of 15 grid cells had to meet our require- was defined as any cell within which the species was recorded ments for ‘lower’ confidence data. If data did not meet this pre-1975 but was not recorded in the same cell post-1975. threshold for a species, then we interpreted the change in To improve our confidence in assigning a species as absent distribution as a potential artefact of differences in sampling from a grid cell, we summed records of ‘background’ fauna effort between the two periods. recorded in both time periods (Table 1). These included On reviewing the literature on mammal declines in arid records of those species likely to be detected with the same Australia, we developed six hypotheses to explain changes methods used to target the focal species (Supplementary in distributions pre- versus post-1975: productivity refuge, material Appendix 1, 2; Ponder et al. 2001). For range predation refuge, competition refuge, climate refuge, pre- contraction a focal species was considered absent from a dation–productivity interaction refuge, and competition– cell post-1975 when the number of records of background productivity interaction refuge (Table 3). The productivity species post-1975 equalled or exceeded the number of refuge relates to early hypotheses for arid Australian mammal records pre-1975, but the focal species was not recorded. For decline that predict species will contract or shift to the most range expansion, a focal species was considered absent from productive parts of their distributions, with higher fecundity a cell pre-1975 when the number of records of background and fitness buffering losses, particularly through drought species pre-1975 equalled or exceeded the number of records (Morton 1990, Stafford-Smith and Morton 1990). We esti- post-1975, but the focal species was not recorded. We also mated productivity using a normalised difference vegetation assigned two levels of confidence for designating a species index (NDVI) variable, with the median value calculated for as absent from a grid cell (Table 1). For range contraction, each cell from 250 m resolution MODIS satellite imagery the ‘lower’ confidence level comprised those cells with an (Table 3). The predation refuge predicts that native mam- equal or greater number of records of background species mals will contract or shift to places where there are either post-1975 than pre-1975, and the ‘higher’ confidence level fewer introduced predators or where habitat affords some was as per the lower confidence level but with a minimum level of protection from these predators (Johnson et al. 2007, of five background records post-1975. For range expansion, Letnic et al. 2009). We used predator indices and a rugged- the ‘lower’ confidence level comprised those cells with an ness index to evaluate this hypothesis. The predator indices
1092 12 Table 1. Methods of dealing with data uncertainty in the NT Fauna Atlas used for modelling drivers of range contraction or expansion.
Issue Method used to reduce uncertainty Fine-scale accuracy errors – Use database with a high level of quality control (e.g. vetted by biologists). – Convert point occurrence data to binary presence/absence using a coarse 25 25 km resolution. Type I errors (false positive – Use database with a high level of quality control (e.g. vetted by biologists). post or pre-1975) – Sample randomly from all ‘presence’ cells (e.g. 10) and run separate models with each set of random samples. This works for our data where the presence cells outnumber the absence cells. In other situations this could be achieved by randomly sampling a subset of presence or absence cells. Type II error (false negative – For modelling range contraction, only assign cells as absent post-1975 and present pre-1975 where there post or pre-1975) are records of ‘background’ fauna (likely to be detected when targeting the focal species) post-1975. – For modelling range expansion, only assign cells as present post-1975 and absent pre-1975 where there are records of ‘background’ fauna (likely to be detected when targeting the focal species) pre-1975. – Use two levels of confidence; lower 1–4 records, higher 5 records. were calculated as the % of records that feral cats, red fox et al. 1996). We removed one of a pair of variables where and dingo comprise of the total fauna records in each cell the correlation coefficient was 0.6 or where the variance (Table 3). The ruggedness index was calculated as the stan- inflation factors were 4. A variable was retained if it had a dard deviation of mean elevation in each cell using a 250 m low AICc (from a univariate logistic regression model using resolution digital elevation model (Table 3). The competition all presence and absence grids) and/or were biologically more refuge predicts that native mammals will contract or shift to meaningful (see Results). areas of their distribution that are less degraded by intro- To test for spatial autocorrelation, we ran a saturated duced herbivores (Morton 1990, McKenzie et al. 2007). We binary logistic regression model for each species (pres- used a land tenure variable and rabbit abundance index to ence 1, absence 0), which included all non-interaction evaluate this hypothesis (Table 3). For land tenure we used a variables that were used in the final models. We used the categorical variable with land designated as either cattle pro- deviance residuals from these models and the datapoint x duction or ‘other’. Cattle production represents cells where and y coordinates to plot correlograms using the R package the dominant tenure has been cattle production at any stage ‘ape’ (Paradis et al. 2014). We also used ArcMAP 10.2 to prior to 1975. Because cattle leases were all established in calculate a global measure of autocorrelation using the spatial the period 1901–1975, and mostly in the second half of that autocorrelation (Moran’s I) tool. Where spatial autocorrela- period, we consider it an appropriate variable for assessing tion was detected, through inspection of the correlogram and the influence of the cumulative and interactive effects of a significant Moran’s I (p 0.05), we created an autologistic grazing pre- versus post-1975. The ‘other’ land tenure pri- variable that was added to all models included in the final marily consists of Aboriginal freehold, government reserve set (Dormann et al. 2007). A criticism of autologistic regres- or vacant crown land. The rabbit abundance variable was sion is that the effect size of other predictor variables may calculated as per the predator indices. The climate refuge be underestimated, particularly where covariates are highly hypothesis is based on the global trend of species to shift or correlated with the autologistic variable (Dormann 2007). contract to cooler areas, particularly along elevation gradi- However, since our objective was to compare the relative ents, in response to increases in global surface temperature parsimony of several alternative explanatory models, rather or extreme heatwaves (Moritz et al. 2008). To evaluate this than to produce species distribution models, we believe that hypothesis we used a modern temperature layer, calculating autologistic regression was appropriate for accounting for the median value in each cell from mean maximum summer spatial autocorrelation. We also screened for pairwise cor- (December–January) temperature (°C) on an interpolated relations between the autocovariate variable and the other 2 km grid (Table 3). A contraction or shift towards areas predictor variables using Spearman’s correlation. that experience more benign maximum temperatures in the Because there were typically more presence than absence hottest months of the year would be considered support for cells, and in order to reduce the likelihood of false positives our hypothesis. Because the impacts of disturbance may (use of erroneous presence records), we randomly sampled interact with other factors such as habitat quality (Anson from all presence cells to match the number of absence cells et al. 2014), we also modelled for interactions between (Table 1). This was repeated 10 times for both the lower and predation and productivity and competition and productivity higher confidence sets of absence cells (i.e. 10 sets of ran- (Table 3). domly sampled presence cells matched in turn against the two sets of absence cells). We then ran binary logistic regression models for each of the 10 randomly selected lots of presence Data analysis cells (1), matched against the absence cells (0). Models were ranked in each set using an information theoretic approach All analyses were run in R (ver. 3.0.3) unless otherwise with Akaike’s information criterion (manually adjusted for stated. In order to compare estimates of model parameters, small sample size; AICc) (Burnham and Anderson 2002). we standardised predictor variables to have a mean of 0 and We calculated the difference in criterion values between the a standard deviation of 1. We assessed the predictor variables best ranked model and model i(Δi), where the model with for pairwise correlations using Spearman’s coefficient and for the smallest AICc has a Δi value of 0. For subsequent models, multi-collinearity by calculating variance inflation factors, Δi AICci – AICcmin, where AICci is the AICc of the model produced by regressing each variable against all others (Neter being compared with the best ranked model. Models with Δi
1093 13 2 were considered to have substantial empirical support, western arid zone, mostly within the Tanami Desert values of 2 to 4 to have some support, and values 4 to (Fig. 2a), and records of T. vulpecula were limited to the des- have little support from the models considered (Burnham ert uplands in the southern parts of the study area (Fig. 2b). and Anderson 2002). We also calculated Akaike weights (wi) Our best models for M. lagotis supported the to indicate the probability that a model is the best given the competition refuge hypothesis positing that cattle grazing available data (Burnham and Anderson 2002). We present was associated with the post-1975 contraction in distribu- all models in at least one of the ten 95% confidence sets tion. We also found some support for the role of higher (summed wi across models 0.95) and include the model- rabbit densities in their decline. There was spatial autocor- averaged coefficient estimate and standard error values for relation in the M. lagotis data (Moran’s I 0.29, z 4.13, all variables, the number of times the model was present in p 0.001), so we ran all models with and without an autol- a 95% confidence set, as well as median values for AICc, Δi ogistic variable (neighbourhood size 4 decimal degrees) 2 and wi. We also present the median Nagelkerke R model fit (Dormann et al. 2007), as well as a univariate model with for each model, using the R package ‘rms’ (Harrell 2014). just the autologistic variable (Table 4). There were no high correlations ( 0.6) between the autologistic variable and the other predictors. Although the autologistic variable was Results in all the 95% confidence set models, the addition of other variables, particularly ‘land tenure’ (TENURE), substantially Evaluating range change improved model performance (Table 3). Macrotis lagotis was For all CWR marsupial species in the arid NT, there was an less likely to occur in grid cells where the dominant tenure increase in the number of records post-1975 versus pre-1975 was cattle grazing (Fig. 2a). This variable was present in all and, for at least five species, an increase in the minimum 95% confidence sets of the lower and higher confidence extent of occurrence (Table 2). However, the data for these data, and was the only one present in the autologistic model five species did not meet the minimum confidence require- with the lowest median AICc (71.88 or 47.03 for the lower ments for modelling colonisation of cells (15 ‘lower’ confi- and higher confidence data, respectively), median iΔ (0.00 or 0.00), highest median wi (0.47 or 0.34) and second highest dence absence cells); with 10 of 123 cells for the mulgaras 2 Dasycercus spp., 7 of 164 cells for spectacled hare-wallaby median R (0.734 or 0.755) (Table 4). Lagorchestes conspicillatus, 4 of 80 cells for black-footed The only other autologistic model present in all 95% rock-wallaby Petrogale lateralis, and 1 of 27 cells for Itjarijari confidence sets of lower and higher confidence data included Notoryctes typhlops having an equal or greater number of also the variable ‘rabbit abundance’ (RABBIT; Table 4). The background records pre- versus post-1975. occurrence of M. lagotis was related negatively to rabbit The minimum extent of occurrence of the greater bilby abundance, although high standard errors relative to model Macrotis lagotis and common brushtail possum Trichosurus coefficient estimates suggest uncertainty in the strength of vulpecula decreased by 25.8 and 38.4%, respectively the effect (Table 4). This model had a slightly higher median (Table 2). The data for both species met our minimum AICc (73.61 or 47.74 for lower and higher confidence data, requirements for modelling range contraction with 15 respectively) median Δi (1.74 or 0.70), a lower or marginally lower median wi (0.21 or 0.28) and a marginally improved absence cells having an equal or greater number of back- 2 ground records pre- versus post-1975. median R (0.737 or 0.766), compared with the higher ranked model. Two other models were present in more than half the 95% confidence sets of the lower and higher confi- Drivers of range contraction in the bilby and dence data, including an autologistic model with an inter- common brushtail possum action term (land tenure: productivity; TENURE:NDVI) and an autologistic model with NDVI (Table 4). Macrotis Both M. lagotis and T. vulpecula occurred widely across lagotis occurrence was related negatively to productivity the arid NT prior to 1975 (Fig. 2). After 1975, records of (Table 4). The interaction term received more support than M. lagotis were restricted largely to an area in the north the productivity-alone model, with a lower median Δi (3.39
Table 2. Comparisons of the number of records and known areas of the distributions of all extant mammals in the critical weight range in the arid Northern Territory pre- and post-1975.
Recorded No. fauna No. fauna extent of Recorded extent Change in atlas records atlas records occurrence of occurrence recorded extent Species Weight (g) pre-1975 post-1975 pre-1975 (km2) post-1975 (km2) of occurrence (%) Brush-tailed and crest-tailed mulgara 60–185 169 743 417 394 575 216 37.8 Dasycercus blythi and D. cristicauda Greater bilby Macrotis lagotis 800–2500 232 797 702 629 521 483 –25.8 Common brushtail possum Trichosurus 1200–4500 89 253 516 391 317 953 –38.4 vulpecula vulpecula Spectacled hare-wallaby Lagorchestes 1600–4750 36 430 340 531 580 470 70.5 conspicillatus Black-footed rock-wallaby Petrogale 2800–5200 119 672 280 881 370 065 31.8 lateralis Itjaritjari Notoryctes typhlops 40–70 77 83 124 345 330 381 165.7
1094 14 Table 3. Hypotheses and model variables used to explain the contraction or shifts in the distribution of marsupials in the arid Northern Territory.
Predictor variable code and description (median, percentage or dominant categorical value Hypothesis Theoretical basis calculated within each 25 25 km grid scale). 1. Productivity refuge Due to a combination of stresses, species will NDVI1: Median of average greenness 2000–2007 contract or shift to the most productive parts of (MODIS 250 m resolution using 16-d normalised their distributions, with higher fecundity and vegetation index mean) fitness buffering losses, particularly through periods of drought (Morton 1990, Stafford- Smith and Morton 1990). 2. Predation refuge Predation by introduced predators is the primary CAT2: % feral cat Felis catus records from total fauna driver of range contraction and species will records contract or shift to places with fewer predators FOX2: % red fox Vulpes vulpes records from total or where dingoes (apex predator) or habitat fauna records (e.g. ruggedness) afford protection (Johnson DINGO2: % dingo Canis dingo records from total et al. 2007, Letnic et al. 2009). fauna records RUGGED3: Ruggedness index calculated as the standard deviation of mean elevation from Australia-wide 9 second (approx. 250 m resolu- tion) Digital Elevation Model (DEM) 3. Competition refuge Competition and habitat degradation associated TENURE4: Categorical value indicating dominant with introduced herbivores are the primary land-tenure of grid cell retrieved from cadastre drivers of range contraction or shifts (Morton layer. Category 1 various not cattle production 1990, McKenzie et al. 2007). (e.g. Aboriginal land, reserve), category 2 cattle production (including any time in last 50 yr) RABBIT2: % European rabbit Oryctolagus cuniculus records from total fauna records 4. Climate refuge Consistent with global warming, species will SUM_TEMP1: Median of mean maximum daily contract or shift to the cooler parts of their surface temperature (°C) for December–February distribution due to increased stresses associ- averaged over the period 1980–1999. Australian ated with higher temperatures (Burbidge et al. Bureau of Meteorology data, interpolated to daily 2008). time step and spatial grid of approx. 2 km 5. Predation–productivity Species will contract or shift to the more NDVI.CAT: Interaction term refuge productive parts of their distribution where NDVI.FOX: Interaction term there is protection from predation NDVI.DINGO: Interaction term NDVI.RUGGED: Interaction term NDVI.TENURE: Interaction term 6. Competition–produc- Species will contract or shift to the more NDVI.RABBIT: Interaction term tivity refuge productive parts of their distribution where there is less competition from introduced herbivores
1Australian Natural Resource Data Library ( http://asdd.ga.gov.au/asdd/tech/node/brs-4.html ). 2NT Fauna Atlas ( www.lrm.nt.gov.au/plants-and-animals/information-and-publications/nt-fauna-observations ). 3National Elevation Data Framework Portal ( http://nedf.ga.gov.au/geoportal/ ). 4Cadastre layer available to NT Government staff ( http://lrm.nt.gov.au/lrm ). versus 3.82 and 1.23 versus 4.63 for the lower and higher Although predator variables were largely absent from the confidence data, respectively), higher medianw i (0.08 versus best-ranked models (Table 4), we note that records for the 0.04 and 0.17 versus 0.04), and improved median R2 (0.702 three mammalian predators were relatively patchy across our versus 0.670 and 0.738 versus 0.726) (Table 4). study area. In the Macrotis lagotis data (total n cells used for Our best models for T. vulpecula supported the climate modelling 287) there were Felis catus records in 148 cells, refuge hypothesis with a post-1975 contraction in distribu- Canis lupus records in 110 cells, and Vulpes vulpes records in tion to the cooler parts of its distribution. In analyses for 46 cells. In the T. vulpecula data (total n cells used for mod- T. vulpecula, ‘summer temperature’ (SUM_TEMP) and elling 83) there were F. catus records in 41 cells, C. lupus ‘ruggedness’ (RUGGED) were highly correlated (Spearman’s records in 43 cells, and Vulpes vulpes records in 16 cells. r –0.729) and we removed the latter due to it having a higher AICc in a univariate logistic regression (86.81 versus 71.87), as well as it being one of several variables in the Discussion predation hypothesis, while summer temperature was the only variable in the climate change hypothesis (Table 3). Using fauna records from a multiple-source occurrence There was no evidence of spatial autocorrelation (Moran’s database, we identified substantial recent contractions in I 0.08, z 1.45, p 0.15) so we did not include an the distribution of two arid-dwelling CWR marsupial spe- autologistic variable in our modelling for this species. Only cies across a vast region of arid Australia. We also produced the summer temperature variable was present in the 95% robust models that offer insights into the drivers of these confidence set of models for the lower and higher confidence declines and we take confidence in the similarity of our data, and T. vulpecula occurrence was related negatively to results between the low and high confidence sets of data. mean maximum summer temperature (Table 4; Fig. 2b). Our results highlight the potential utility of multiple-source
1095 15 Table 4. Summary of the 95% confidence sets of models explaining the contraction in distribution of Macrotis lagotis and Trichosurus vulpecula in the arid Northern Territory.
No. times model present in 95% Median confidence set Median Median Akaike Median 1 2 Species Model (from 10 sets) AICc Δi weight (wi) Nagelkerke R Macrotis lagotis Lower confidence set (112 data points) AUTO (1.72; 0.37) TENURE (–0.99; 0.30) 10 71.88 0.00 0.47 0.734 AUTO (1.65; 0.38) TENURE (–1.02; 0.30) 10 73.61 1.74 0.21 0.737 RABBIT (–0.16; 0.33) AUTO (2.19; 0.35) TENURE:NDVI (–0.50; 0.21) 6 80.01 3.39 0.08 0.702 AUTO (2.17; 0.35) NDVI (–0.60; 0.26) 6 83.31 3.82 0.04 0.670 AUTO (1.68; 0.33) TEMP (0.92; 0.37) 4 82.77 4.13 0.06 0.673 AUTO (1.62; 0.28) CAT (1.93; 1.80) 4 89.13 6.94 0.02 0.634 AUTO (1.68; 0.28) DINGO:NDVI (1.10; 0.51) 2 87.31 4.18 0.05 0.579 AUTO (1.86; 0.32) FOX (2.04; 1.14) 2 89.47 5.54 0.03 0.630 AUTO (1.90; 0.34) CAT (1.47; 1.33) FOX 2 81.835 6.42 0.02 0.691 (1.82; 1.24) AUTO (2.05; 0.35) RABBIT (–0.08; 0.36) 1 75.98 1.58 0.16 0.667 AUTO (2.13; 0.38) FOX (2.47; 1.22) DINGO 1 81.32 6.92 0.01 0.694 (0.07; 0.30) AUTO (1.61; 0.27) RABBIT:NDVI (1.46; 0.86) 1 98.48 5.02 0.03 0.574 AUTO (1.58; 0.26) 1 100.19 6.72 0.01 0.547 null model (–2ll) 155.27 Higher confidence set (74 data points) AUTO (1.62; 0.46) TENURE (–1.39; 0.39) 10 47.03 0 0.34 0.755 AUTO (1.40; 0.47) TENURE (–1.49; 0.41) 10 47.74 0.70 0.28 0.766 RABBIT (–0.91; 1.81) AUTO (2.50; 0.52) TENURE:NDVI (–0.78; 0.31) 7 49.16 1.23 0.17 0.738 AUTO (2.3; 0.47) NDVI (–0.97; 0.41) 7 50.76 4.63 0.04 0.726 AUTO (1.79; 0.47) TEMP (2.16; 0.51) 6 51.28 2.31 0.09 0.721 AUTO (2.18; 0.47) RABBIT:NDVI (2.69; 1.27) 6 52.36 4.27 0.04 0.712 AUTO (2.26; 0.49) DINGO:NDVI (1.14; 0.59) 3 51.90 4.46 0.03 0.650 AUTO (1.76; 0.38) CAT (2.15; 1.65) 2 55.98 5.87 0.02 0.680 AUTO (1.92; 0.40) 2 60.97 6.03 0.02 0.615 AUTO (2.60; 0.55) FOX:NDVI (4.30; 1.67) 1 44.44 4.46 0.03 0.775 AUTO (2.26; 0.48) CAT:NDVI (0.80; 0.54) 1 54.40 6.47 0.01 0.693 AUTO (1.55; 0.33) CAT (1.94; 1.49) DINGO 1 62.24 7.30 0.01 0.643 (0.15; 0.44) null model (–2ll) 102.59 Trichosurus vulpecula Lower confidence set (76 data points) SUM_TEMP (–1.76; 0.44) 10 68.91 0.00 1.00 0.450 null model 91.495 Higher confidence set (58 data points) SUM_TEMP (–1.75; 0.46) 10 57.54 0.00 1.00 0.495 null model 80.405
1See Table 2 for variable definitions. occurrence databases for documenting and understanding and T. vulpecula also is probably very low, further enhanc- mammal declines and distribution shifts, particularly where ing confidence in our use of multiple-source data. With there are no alternative long-term datasets. increased restrictions on collecting vertebrate specimens Our approach is similar to that in previous studies using (Remsen 1995), we argue that multiple-source occurrence occurrence data from natural history collections (Pergams databases will become an increasingly important source for and Nyberg 2001, Jeppsson et al. 2010, Pyke and Ehrlich documenting faunal distribution shifts. Further, we suggest 2010). Although collection databases may have fewer errors our extension of the use of background records to assign than occurrence data from multiple sources (e.g. misiden- multiple levels of confidence is an effective way to improve tifications), by using multiple-source data we were able to the inferential value of models using multiple-source draw on more occurrence records collected over a larger area. occurrence data. For example, had we relied on museum collection data for The apparent increase in extent of occurrence we recorded M. lagotis, we would have had 23 records post-1975 rather for five CWR marsupials post-1975 is likely associated than the 797 records from multiple sources. The chance of with increases in human population density and biological misidentifying highly distinctive species such as M. lagotis survey effort in the study region in that period, rather than
1096 16 Figure 2. (a) Locations in the Northern Territory of pre- and post-1975 records of Macrotis lagotis in relation to history of cattle production, and (b) location of pre- and post-1975 records of Trichosurus vulpecula in relation to maximum summer temperature. representing actual range expansions. This is reflected in ably improve their resilience to above-ground threats. The the substantially higher number of records for most species ground-dwelling macropod Lagorchestes conspicillatus has post-1975 and the lack of background records to indicate fared better than its congeners, all of which are extinct in the equal or greater sampling effort pre-1975. In this context, wild, but it may have suffered a more recent decline than our the contractions in distribution of M. lagotis and T. vulpecula analysis would have detected (Burbidge and Johnson 2008). likely reflect actual declines, and are consistent with anec- Although the contraction of M. lagotis to the Tanami dotal reports of the disappearance of these species from sites Desert region of the NT has been reported previously throughout the study region (Burbidge et al. 1988, Kerle (Southgate 1990), the potential drivers of this contraction et al. 1992). had not been explored. We found that the decline of this Although we acknowledge the potential influence of species was consistent with a shift away from areas used increased sampling effort post-1975, no reduction in distri- for cattle production and possibly also from areas of high bution was recorded for five of the other CWR marsupi- rabbit density. These results support the competition-refuge als. For four of the five other species, we suggest that this hypothesis, with competition for food and shelter occur- demonstrates relative resilience and that this may be ring with the arrival of cattle and rabbits to the arid NT in explained though aspects of their ecology and life-history. the first half of the 20th century. The results are consistent The two carnivorous Dasycercus species are the only CWR also with correlative studies showing a relationship between marsupials in the study region known to enter torpor, and CWR mammal decline and environmental change (Burbidge the ability to substantially reduce energy expenditure may be and McKenzie 1989, Johnson et al. 2007, McKenzie et al. an important factor in their persistence (Geiser and Masters 2007), as well as with studies from Australia’s tropical north 1994, Masters and Dickman 2012). Rock-dwelling mam- and semi-arid South Africa, which demonstrate negative mal species, such as the macropod Petrogale lateralis, may impacts on terrestrial mammal species in association with a be more resilient than ground-dwelling species (McKenzie reduction in vegetation cover and food resources from live- et al. 2007, Johnson and Isaac 2009). This is attributed stock grazing (Eccard et al. 2000, Legge et al. 2011). We to greater habitat complexity in rocky areas providing acknowledge that our competition variables may be corre- more benign microclimates and greater protection from lated with additional unrecorded factors and that alternative predation. Notoryctes typhlops and the closely related or interactive mechanisms may have driven range contrac- N. caurinus are Australia’s only truly fossorial mammals and, tion. For example, cattle grazing and the loss of grass cover although they have been recorded in the diets of both the can improve the hunting efficiency of feral cats (McGregor cat and fox (Paltridge 2002), their underground habits prob- et al. 2014), while fire suppression on cattle stations may have
1097 17 reduced the availability of grass seeds favoured by M. lagotis The distribution of T. vulpecula appears to have (Southgate and Carthew 2006). Only field experiments contracted to the cooler, higher elevation regions of the arid controlling for alternative mechanisms would clarify the role NT, consistent with our climate change hypothesis. Global of competition in the persistence of M. lagotis (see below). surface temperatures have increased by 0.85°C and there has The uncertainty in the strength of the effect of rabbit been a global trend of increasing monthly heat records, both abundance on M. lagotis may be a consequence of the patchy attributed to anthropogenic climate change (Coumou and nature of rabbit occurrence across the arid NT. Although Rahmstorf 2012, IPCC 2013). In Australia the magnitude there is a general north-south shift from low to high rabbit of increasing temperatures has been highest in inland areas, abundance in the atlas data, there were many grid cells in the with median annual temperatures in Alice Springs (the major south without any rabbit records, and vice versa, presumably population centre in the south of the study region) increas- reflecting variation in sampling effort. This is not surprising ing by up to 2°C since the 1970s (Hughes 2003, Davis et al. given the absence of roads in many areas and the generally 2013). There has also been an overall positive trend in the low human population density across the arid NT. This is number of hot days ( 40°C) across arid Australia (Collins a significant problem when attempting to derive predictor et al. 2000). With a predominately temperate distribution variables from multiple-source occurrence databases. In this (Kerle and How 2008), arid-dwelling T. vulpecula may context, the relative unimportance of predator variables in already have been on the edge of its thermal limits. Some the M. lagotis and T. vulpecula models must be viewed with support for this hypothesis includes an observation of sub- caution. In particular, red fox records were extremely patchy stantial mortality of T. vulpecula during a recent heatwave across our grid cells. Predation by this species should not be in a temperate region of South Australia (Bird 2009). Given discounted as a potential driver of the northward contrac- the genetic similarity between T. vulpecula from arid central tion of M. lagotis (Southgate 1990), particularly given the Australia and temperate South Australia (Foulkes 2001), it is positive correlation between rabbit and red fox records in plausible that arid-dwelling populations are also vulnerable the bilby data (Spearman’s r 0.34) and elsewhere in to extreme weather events. As well as causing direct mortal- Australia (Read and Bowen 2001, Johnson et al. 2007). An ity, increasing temperatures may reduce the availability of alternative to deriving our own predator and rabbit indices suitable den sites for arid T. vulpecula. In the tropical north, would have been to source species distribution model (SDM) T. v. arnhemensis preferentially selected cooler tree hollows layers for these taxa. However, such layers are currently not among potential den sites (Isaac et al. 2008). Arid-dwelling available for large areas of Australia (Newsome et al. 2013) T. vulpecula is now known to occur only at sites without tree and we note the challenges associated with quantifying the hollows where the only available den sites are rock crevices. occurrence or density of cats, foxes and dingoes (Hayward These crevices offer a stable thermal buffer against extreme and Marlow 2014). temperature fluctuations (Geiser and Pavey 2007) and are Given the highly clustered M. lagotis occurrences likely to provide a superior buffer than tree hollows (Isaac post-1975, it was not surprising that model residuals were et al. 2008). spatially autocorrelated. The Tanami Desert is a vast sandy Given that temperature and ruggedness were highly region characterised by relatively homogenous landforms correlated in the T. vulpecula dataset, ruggedness should be and vegetation types (Southgate et al. 2006). Therefore, considered an alternative factor in explaining the persistence it seems likely that the spatial autocorrelation was the of the species in the desert uplands. Rugged rocky areas may product of an additional environmental determinant(s) offer increased protection from predation by cats, foxes and inherent to this region (Dormann et al. 2007). Southgate dingoes, with rock-dwelling mammals having been more et al. (2007) modelled the factors determining habitat suit- resilient than terrestrial species across Australia (McKenzie ability for M. lagotis across the Tanami Desert and found a et al. 2007, Johnson and Isaac 2009). However, tree- positive relationship with mean annual rainfall, substrate type dwelling species have also fared better than terrestrial species (including laterite/rock features) and probability of dingo and thus the hypothesis does not neatly explain the apparent occurrence. The authors conceded that there were probably disappearance of T. vulpecula from tree-lined watercourses additional ecological processes that were not accounted for and alluvial woodlands, where there are many opportuni- in the models, and this was evident in the weak predictive ties to forage above ground or to climb to evade terrestrial performance of the independent evaluation data. We found predators (Kerle et al. 1992). It may now be too late to a negative relationship between bilby occurrence and NDVI further explore the factors driving the contraction in distri- (itself highly correlated with rainfall across the study region) bution of T. vulpecula. The species appears to have continued and a weak positive relationship between bilby occurrence to decline in distribution over the last two decades and in and dingo records; finer-scale substrate variables could not the arid zone is now known only from a handful of sites in a be accounted for in our analysis. Regardless, we suggest that restricted area of mountainous terrain west of Alice Springs none of these variables adequately differentiate the Tanami (McDonald et al. unpubl.). Desert from other regions with few or no records post-1975. Our results for M. lagotis suggest that competition with The refuge status of the Tanami may reflect its vast area, cattle may have been a factor contributing to the species’ contiguous suitable habitat, low rabbit densities, absence of retraction from productive rangeland areas. However, it cattle grazing in most areas, lower fox densities than in areas is unclear whether removal of cattle could improve habi- further south, and potential fox-suppressive effects by the large tat suitability to the point that these areas could support dingo populations in certain parts of the region (Newsome M. lagotis again. Experimental removal of cattle from prop- et al. 2013). Further, substantial portions of the northern and erties neighbouring the Tanami Desert would allow testing southern Tanami Desert are reserved in Indigenous Protected of the role of competition on habitat suitability, but would Areas and managed by Indigenous rangers. need to account for potential confounding differences in 1098 18 vegetation type (e.g. spinifex grasslands versus tussock grass- Burbidge, A. A. and Johnson, P. M. 2008. Spectacled hare-wallaby. lands), predator density and interactions between the two – In: Van Dyck, S. and Strahan, R. (eds), The mammals of factors. However, given the importance of beef production to Australia, 3rd ed. New Holland Publishers, pp. 314–316. the economy of the NT, together with the complex potential Burbidge, A. A. et al. 1988. Aboriginal knowledge of the mammals of the Central Deserts of Australia. – Aust. Wildl. Res. 15: 9–39. interactions between competition and predation, the imme- Burbidge, A. A. et al. 2008. Conservation status and biogeogra- diate objective for the conservation of M. lagotis should be to phy of Australia’s terrestrial mammals. – Aust. J. Zool. 56: ensure its continued persistence in the Tanami Desert. This 411–422. will require an improved understanding of the factors driving Burnham, K. P. and Anderson, D. 2002. Model selection and population or occupancy dynamics within this region, and multimodel inference: a practical information-theoretic is likely to be best achieved through a systematic monitoring approach, 2nd ed. – Springer. program with appropriate intervention points for manage- Ceballos, G. and Ehrlich, P. R. 2002. Mammal population losses ment actions. Our data on Macrotis lagotis range contrac- and the extinction crisis. – Science 296: 904–907. Chisholm, R. and Taylor, R. 2007. Null hypothesis testing and the tion support the species’ current listing as vulnerable under critical weight range for Australian mammals. – Conserv. Biol. National (EPBC) and Territory (TPWCA) legislation. 21: 1641–1645. Although any intervention may be too late for Clarke, M. F. 2008. Catering for the needs of fauna in fire manage- arid-dwelling T. v. vulpecula, which now likely qualifies as ment: science or just wishful thinking? – Wildl. Res. 35: critically endangered in the NT (under IUCN Red List 385–394. criteria), there is an urgent need to better understand how Clavero, M. and Garcia-Berthou, E. 2005. Invasive species are a increased temperatures will affect the persistence of this leading cause of animal extinctions. – Trends Ecol. Evol. 20: species across Australia. This is important given the ongo- 110. Collins, D. A. et al. 2000. Trends in annual frequencies of extreme ing declines observed both for this subspecies in other parts temperature events in Australia. – Aust. Meteorol. Mag. 49: of its distribution and for T. v. arnhemensis in the Top End 277–292. of the NT (Kerle and How 2008, Woinarski et al. 2011). Coumou, D. and Rahmstorf, S. 2012. A decade of weather In addition to undertaking physiological experiments, the extremes. – Nat. Clim. Change 2: 491–496. experimental release of T. vulpecula into predator-free areas Davis, J. et al. 2013. Evolutionary refugia and ecological refuges: of arid Australia (e.g. predator-proof enclosures) would help key concepts for conserving Australian arid zone freshwater to establish the role of increasing temperatures on its survival biodiversity under climate change. – Global Change Biol. 19: in the absence of predation pressure. Our results for T. vul- 1970–1984. Dickman, C. R. 1996. Impact of exotic generalist predators on the pecula demonstrate the importance of thinking beyond the native fauna of Australia. –Wildl. Biol. 2: 185–195. ‘typical’ factors when attempting to understand drivers of Dickman, C. R. 2013. Long-haul research: benefits for conserving decline. Recent investigations into the decline of Australia’s and managing biodiversity. – Pac. Conserv. Biol. 19: 10–17. mammals have ignored climate change as a potential driver Dormann, C. F. 2007. Effects of incorporating spatial autocorrela- (McKenzie et al. 2007, Woinarski et al. 2011, Fisher et al. tion into the analysis of species distribution data. – Global 2014), although Burbidge et al. (2008) predicted future Ecol. Biogeogr. 16: 129–138. climate-driven range shifts for Australia’s terrestrial mam- Dormann, C. F. et al. 2007. Methods to account for spatial mals. This is despite temperature-driven range contractions autocorrelation in the analysis of species distributional data: a review. – Ecography 30: 609–628. already having occurred along an elevation gradient in the Eccard, J. A. et al. 2000. How livestock affects vegetation structures USA (Moritz et al. 2008). Our results suggest that climate and small mammal distribution in the semi-arid Karoo. – J. change and its interactive effects should be considered among Arid Environ. 46: 103–106. other hypotheses for recent and ongoing mammal declines Fisher, D. O. and Blomberg, S. P. 2011. Correlates of rediscovery in Australia. We conclude that models using multiple-source and the detectability of extinction in mammals. – Proc. R. Soc. occurrence data can provide important new insights into the B 278: 1090–1097. patterns and drivers of species’ declines and assist in directing Fisher, D. O. et al. 2014. The current decline of tropical marsupi- experimental management aimed at ameliorating these drivers. als in Australia: is history repeating? – Global Ecol. Biogeogr. 23: 181–190. Foulkes, J. N. 2001. The ecology and management of the common Acknowledgements – CRD was supported by grants from the brushtail possum Trichosurus vulpecula in central Australia. Australian Research Council. – Unpubl. PhD thesis, Univ. of Canberra, Canberra. Geiser, F. and Masters, P. 1994. Torpor in relation to reproduction in the mulgara, Dasycercus cristicauda (Dasyuridae: Marsupi- References alia). – J. Thermal Biol. 19: 33–40. Geiser, F. and Pavey, C. R. 2007. Basking and torpor in a Aizpurua, O. et al. 2015. Optimising long-term monitoring rock-dwelling desert marsupial: survival strategies in a resource- projects for species distribution modelling: how atlas data may poor environment. – J. Comp. Physiol. 177: 885–892. help. – Ecography 38: 29–40. Gibson, D. F. 2000. Distribution and conservation status of the Anson, J. R. et al. 2014. Effects of multiple disturbance processes black-footed rock-wallaby, Petrogale lateralis (MacDonnell on arboreal vertebrates in eastern Australia: implications for Ranges race), in the Northern Territory. – Aust. Mammal. 21: management. – Ecography 37: 357–366. 213–236. Bird, P. 2009. Heatwave mammal deaths. – Mammal Soc. Griffioen, P. A. and Clarke, M. F. 2002. Large-scale bird-movement Newslett. Oct. 2009: 22. patterns evident in eastern Australian atlas data. – Emu 102: Burbidge, A. A. and McKenzie, N. L. 1989. Patterns in the mod- 99–125. ern decline of Western Australia’s vertebrate fauna: causes and Harrell, F. E. 2014. Package ‘rms’ v 4.2-0. – R package, http:// conservation implications. – Biol. Conserv. 50: 143–198. cran.r-project.org/web/packages/rms/index.html .
1099 19 Hayward, M. W. and Marlow, N. 2014. Will dingoes really improved resource selection modelling approach. – PLoS One conserve wildlife and can our methods tell? – J. Appl. Ecol. 51: 8: e63931. 835–838. Paltridge, R. 2002. The diets of cats, foxes and dingoes in relation Hughes, L. 2003. Climate change and Australia: trends, projections to prey availability in the Tanami Desert, Northern Territory. and impacts. – Austral Ecol. 28: 423–443. – Wildl. Res. 29: 389–403. IPCC 2013. Climate change 2013: the physical science basis. Paradis, E. et al. 2014. Package ‘ape’ v3.1. – R package, http:// Contribution of Working Group I to the Fifth Assessment cran.r-project.org/web/packages/ape/index.html . Report of the Intergovernmental Panel on Climate Change. Parker, S. A. 1973. An annotated checklist of the native land mam- – Cambridge Univ. Press. mals of the Northern Territory. – Rec. S. Aust. Mus. 16: Isaac, J. L. et al. 2008. Microclimate of daytime den sites in a 1–57. tropical possum: implications for the conservation of tropical Pavey, C. R. et al. 2012. Habitat use, population dynamics marsupials. – Anim. Conserv. 11: 281–287. and species identification of mulgara, Dasycercus blythi and Jeppsson, T. et al. 2010. The use of historical collections to estimate D. cristicauda, in a zone of sympatry in central Australia. population trends: a case study using Swedish longhorn beetles – Aust. J. Zool. 59: 156–169. (Coleoptera: Cerambycidae). – Biol Conserv. 143: 1940–1950. Pergams, O. R. and Nyberg, D. 2001. Museum collections of Johnson, C. N. and Isaac, J. L. 2009. Body mass and extinction mammals corroborate the exceptional decline of prairie habitat risk in Australian marsupials: the ‘Critical Weight Range’ in the Chicago region. – J. Mammal. 82: 984–992. revisited. – Austral Ecol. 34: 35–40. Ponder, W. F. et al. 2001. Evaluation of museum collection Johnson, C. N. et al. 2007. Rarity of a top predator triggers con- data for use in biodiversity assessment. – Conserv. Biol. 15: tinent-wide collapse of mammal prey: dingoes and marsupials 648–657. in Australia. – Proc. R. Soc. B 274: 341–346. Pyke, G. H. and Ehrlich, P. R. 2010. Biological collections and Kerle, J. A. and How, R. A. 2008. Common brushtail possum. – In: ecological/environmental research: a review, some observations Van Dyck, S. and Strahan, R. (eds), The mammals of Australia, and a look to the future. – Biol. Rev. 85: 247–266. 3rd ed. New Holland Publishers, pp. 274–276. Read, J. and Bowen, Z. 2001. Population dynamics, diet and Kerle, J. A. et al. 1992. The decline of the brushtail possum, aspects of the biology of feral cats and foxes in arid South Trichosurus vulpecula (Kerr 1798), in arid Australia. – Rangeland Australia. – Wildl. Res. 28: 195–203. J. 14: 107–127. Remsen, J. V. 1995. The importance of continued collecting of Legge, S. et al. 2011. Rapid recovery of mammal fauna in the bird specimens to ornithology and bird conservation. – Bird central Kimberley, northern Australia, following the removal of Conserv. Int. 5: 145–180. introduced herbivores. – Austral Ecol. 36: 791–799. Schipper, J. et al. 2008. The status of the world’s land and marine Letnic, M. and Dickman, C. R. 2006. Boom means bust: interac- mammals: diversity, threat, and knowledge. – Science 322: tions between the El Niño/Southern Oscillation (ENSO), rain- 225–230. fall and the processes threatening mammal species in arid Shaffer, H. B. et al. 1998. The role of natural history collections Australia. – Biodivers. Conserv. 15: 3847–3880. in documenting species declines. – Trends Ecol. Evol. 13: Letnic, M. et al. 2009. Does a top-predator provide an endangered 27–30. rodent with refuge from an invasive mesopredator? – Animal Short, J. and Smith, A. 1994. Mammal decline and recovery in Conserv. 12: 302–312. Australia. – J. Mammal. 75: 288–297. Lindenmayer, D. B. et al. 2012. Improving biodiversity monitor- Southgate, R. I. 1990. Distribution and abundance of the greater ing. – Austral Ecol. 37: 285–294. bilby Macrotis lagotis Reid (Marsupialia: Peramelidae). – In: Luck, G. W. et al. 2010. What drives the positive correlation Seebeck, J. H. et al. (eds), Bandicoots and bilbies. Surrey Beatty between human population density and bird species richness and Sons, pp. 293–302. in Australia? – Global Ecol. Biogeogr. 19: 673–683. Southgate, R. and Carthew, S. M. 2006. Diet of the bilby Magurran, A. E. et al. 2010. Long-term datasets in biodiversity (Macrotis lagotis) in relation to substrate, fire and rainfall research and monitoring: assessing change in ecological com- characteristics in the Tanami Desert. – Wildl. Res. 33: munities through time. – Trends Ecol. Evol. 25: 574–582. 507–519. Masters, P. and Dickman, C. R. 2012. Population dynamics of Southgate, R. et al. 2006. An examination of the Stafford Dasycercus blythi (Marsupialia: Dasyuridae) in central Australia: Smith-Morton ecological model: a case study in the Tanami how does the mulgara persist? – Wildl. Res. 39: 419–428. Desert, Australia. – Rangeland J. 28: 197–210. McCarthy, M. A. 1998. Identifying declining and threatened Southgate, R. et al. 2007. Bilby distribution and fire: a test of species with museum data. – Biol. Conserv. 83: 9–17. alternative models of habitat suitability in the Tanami Desert, McGregor, H. W. et al. 2014. Landscape management of fire and Australia. – Ecography 30: 759–776. grazing regimes alters the fine-scale habitat utilisation by feral Stafford-Smith, D. M. and Morton, S. R. 1990. A framework cats. – PLoS One 9: e109097. for the ecology of arid Australia. – J. Arid Environ. 16: McKenzie, N. L. et al. 2007. Analysis of the factors implicated in 255–278. the recent decline of Australia’s mammal fauna. – J. Biogeogr. Williams, O. B. and Calaby, J. H. 1985. The hot deserts of Aus- 34: 597–611. tralia. – In: Evenari, M. et al. (eds), Hot deserts and arid shrub- Moritz, C. et al. 2008. Impact of a century of climate change in lands. Elsevier, pp. 269–312. small-mammal communities in Yosemite National Park, USA. Woinarski, J. C. Z. et al. 2011. The disappearing mammal fauna – Science 322: 261–264. of northern Australia: context, cause, and response. – Conserv. Morton, S. R. 1990. The impact of European settlement on Lett. 4: 192–201. the vertebrate animals of arid Australia: a conceptual model. Woinarski, J. C. Z. et al. 2014. The action plan for Australian – Proc. Ecol. Soc. Aust. 16: 201–213. mammals 2012. – CSIRO Publishing. Morton, S. R. et al. 2011. A fresh framework for the ecology of Woolley, P. A. et al. 2013. Past and present distribution of arid Australia. – J. Arid Environ. 75: 313–329. Dasycercus: toward a better understanding of the identity of Neter, J. et al. 1996. Applied linear statistical models, 4th ed. – Irwin. specimens in cave deposits and the conservation status of Newsome, T. M. et al. 2013. Anthropogenic resource subsidies the currently recognised species D. blythi and D. cristicauda determine space use by Australian arid zone dingoes: an (Marsupialia: Dasyuridae). – Aust. J. Zool. 61: 281–290. Supplementary material (Appendix ECOG-01212 at www.ecography.org/readers/appendix ). Appendix 1–2. 1100 20 Chapter 3. Habitat as a mediator of
mesopredator-driven mammal extinction
The habitat mosaic of the MacDonnell Ranges region, west of Alice Springs (photo: P. McDonald)
Author contribution: I was responsible for the chapter concept, design and implementation of fieldwork (co-led by C. Nano and assisted by many others), data analysis, and writing, in consultation with my co-authors. Supporting information is provided for this chapter in Appendix 4.
21 Contributed Paper Habitat as a mediator of mesopredator-driven mammal extinction
Peter J. McDonald,1,2 ∗ CatherineE.M.Nano,1 Simon J. Ward,1 Alistair Stewart,1 Chris R. Pavey,3 Gary W. Luck,4 andChrisR.Dickman2 1Flora and Fauna Division, Department of Environment & Natural Resources, Alice Springs, Northern Territory, 0870, Australia 2Desert Ecology Research Group, School of Life and Environmental Sciences, University of Sydney, New South Wales, 2006, Australia 3CSIRO Land and Water, P.O. Box 2111, Alice Springs, Northern Territory, 0871, Australia 4Institute for Land, Water and Society, Charles Sturt University, Albury, NSW, 2640, Australia
Abstract: A prevailing view in dryland systems is that mammals are constrained by the scarcity of fertile soils and primary productivity. An alternative view is that predation is a primary driver of mammal assemblages, especially in Australia, where 2 introduced mesopredators—feral cat (Felis catus) and red fox (Vulpes vulpes)— are responsible for severe declines of dryland mammals. We evaluated productivity and predation as drivers of native mammal assemblage structure in dryland Australia. We used new data from 90 sites to examine the divers of extant mammal species richness and reconstructed historic mammal assemblages to determine proportional loss of mammal species across broad habitat types (landform and vegetation communities). Predation was supported as a major driver of extant mammal richness, but its effect was strongly mediated by habitat. Areas that were rugged or had dense grass cover supported more mammal species than the more productive and topographically simple areas. Twelve species in the critical weight range (CWR) (35–5500 g) that is most vulnerable to mesopredator predation were extirpated from the continent’s central region, and the severity of loss of species correlated negatively with ruggedness and positively with productivity. Based on previous studies, we expect that habitat mediates predation from red foxes and feral cats because it affects these species’ densities and foraging efficiency. Large areas of rugged terrain provided vital refuge for Australian dryland mammals, and we predict such areas will support the persistence of CWR species in the face of ongoing mammal declines elsewhere in Australia.
Keywords: feral cat, mammal decline, predation, red fox, refuge, ruggedness
El H´abitat como Mediador de la Extincion´ Causada por Mesodepredadores Resumen: Una vision´ prevalente en los sistemas secos es que los mam´ıferos estan´ restringidos por la escasez de suelos f´ertiles y productividad primaria. Una vision´ alternativa es que la depredacion´ es un conductor primario de los ensamblajes de los mam´ıferos, especialmente en Australia, en donde dos mesodepredadores introducidos – el gato feral (Felis catus) y el zorro rojo (Vulpes vulpes) – son los responsables de declinaciones graves de los mam´ıferos des´erticos. Evaluamos a la productividad y a la depredacion´ como conductores de la estructura de ensamblaje de mam´ıferos nativos en los desiertos de Australia. Utilizamos datos nuevos de 90 sitios para examinar a los conductores de la riqueza de especies de mam´ıferos existentes y reconstru- imos los ensamblajes historicos´ de mam´ıferos para determinar la p´erdida proporcional de las especies de mam´ıferos a lo largo de los tipos generales de habitat´ (accidente geografico´ o vegetacion).´ La depredacion´ fue respaldada como el principal conductor de la riqueza de mam´ıferos existentes, pero su efecto estuvo mediado considerablemente por el habitat.´ Las areas´ que eran escabrosas o que ten´ıan una cobertura densa de pasto pudieron mantener a mas´ especies de mam´ıferos que las areas´ mas´ productivas o mas´ simples topograficamente.´ Doce especies en el rango cr´ıtico de peso (RCP) (35 – 5500 g) que son mas´ vulnerables a la depredacion´ por mesodepredadores fueron extirpadas de la region´ central del continente, y la gravedad de la p´erdida de especies estuvo correlacionada negativamente con la escabrosidad y positivamente con la
∗email [email protected] Paper submitted August 6, 2016; revised manuscript accepted February 3, 2017. 1183 Conservation Biology, Volume 31, No. 5, 1183–1191 C 2017 Society for Conservation Biology 22 DOI: 10.1111/cobi.12905 1184 Mesopredator-Driven Mammal Extinction productividad. Con base en estudios previos, esperamos que el habitat´ medie la depredacion´ por parte de los zorros rojos y los gatos ferales porque afecta a la densidad de estas especies y su eficiencia para buscar comida. Las areas´ extensas de terreno escabroso proporcionan refugio vital para los mam´ıferos des´erticos de Australia y pronosticamos que dichas areas´ podran´ mantener la persistencia de las especies RCP de frente a las declinaciones de mam´ıferos en otras partes de Australia.
Palabras Clave: declinaciones de mam´ıferos, depredacion,´ el gato feral, la escabrosidad, refugio, zorro rojo
Introduction et al. 2015). Given the severity of these mammal declines, we hypothesized that Australia’s assemblages of extant Spatiotemporal variability in water and nutrient availabil- dryland mammals strongly reflect the effects of intro- ity defines and shapes the world’s drylands (Schwinning duced predators and that these effects may be masking & Sala 2004). Temporally, extreme rainfall events and preexisting productivity-driven relationships. subsequent pulses in primary productivity trigger irrup- Cats and red foxes are regarded as habitat gener- tions of lower order consumers and their predators (Lima alists and, although widespread across Australia’s dry- et al. 2002; Letnic et al. 2005). These pulses provide lands (Legge et al. 2017), may have nonuniform im- key resources for future production and can influence pacts on prey in different landform and vegetation types. disturbance processes in the following months or years For example, McKenzie et al. (2007) found that geo- (Holmgren et al. 2006). Spatially, localized rainfall events graphic regions with a lower proportion of arboreal, can redistribute resources across the landscape; the re- rock-dwelling, or burrowing taxa lost more species, sponse of dryland vegetation to such rainfall varies with whereas Johnson and Isaac (2009) demonstrated that season, soil texture, and nutrient availability (Schlesinger ground-dwelling CWR mammals were more likely to & Pilmanis 1998; Nano & Pavey 2013). decline and become extinct than similar-sized arboreal Due to the spatial variability of nutrients in dryland sys- species. Therefore, it is possible that land-cover type and tems, a prevailing view is that consumer populations are structure could mediate the likelihood of mesopredator- constrained by the scarcity of fertile soils (Whitford 2002; driven extinction. Morton et al. 2011). For example, the ecological-refuge We evaluated productivity (bottom up) versus preda- model for Australian deserts predicts that native medium- tion (top down) as drivers of the structure of extant as- sized mammals and introduced large herbivores are re- semblages of native mammals in dryland Australia. Our stricted to islands of richer soil, where plant production study area, the MacDonnell Ranges Bioregion (Thackway is more continuous and plants are more nutritious and & Cresswell 1995), is ideal for such an investigation be- digestible (Stafford Smith & Morton 1990; Southgate et al. cause it has a gradient of productivity and a diversity of 2007). For small mammals, high-quality patches of habitat landform and vegetation types (hereafter habitat)—from can act as drought refuges, buffering against extended rugged mountain tops with heath-like vegetation and periods of low rainfall and enabling species persistence skeletal soils to ephemeral rivers and alluvial woodlands between resource pulses (Pavey et al. 2015). Although with deeper, richer soils. Based on our hypothesis that there is some evidence supporting these bottom-up pro- predation is the dominant driver of extant native mam- ductivity hypotheses for mammals in drylands (Free et al. mal assemblages, we predicted that the most species-rich 2015), few researchers have evaluated alternatives. mammal assemblages would be associated with habitat One alternative hypothesis is that top-down predation that provide effective protection from predators, irre- is a primary driver of mammal occurrence in drylands spective of spatial or temporal productivity, and these (Lima et al. 2002; Letnic et al. 2009). Australia is a partic- habitats would also support relatively intact mammal as- ularly appropriate system in which to test this hypothe- semblages with fewer extinct species in the CWR. We sis because 2 introduced mesopredators (feral cat [Felis evaluated our predictions based on new data we col- catus]andredfox[Vulpes vulpes]) are responsible for a lected from 90 sites stratified across 4 dominant habi- globally exceptional rate of recent mammal extinctions tats in the bioregion and on published data sets on sub- (Woinarski et al. 2015). Species that weigh 35–5550 g fossils, indigenous ecological knowledge, and historical have been most vulnerable to extinction and are referred accounts. to as being in the critical weight range (CWR) (Burbidge & McKenzie 1989; Chisholm & Taylor 2007). The mag- nitude of declines of CWR species has been most pro- Methods nounced in Australia’s drylands, where 11 species no long occur and many others exist in populations that have con- Study Area and Site Selection tracted to small portions of their ranges occupied before Our study was conducted in the 2592-km2 Tjoritja (West European colonization (McDonald et al. 2015; Woinarski MacDonnell) National Park, situated in the MacDonnell
Conservation Biology 23 Volume 31, No. 5, 2017 McDonald et al. 1185
Ranges Bioregion, southern Northern Territory (Support- spaced traps was set across each quadrat, and traps ing Information). Climate of the region is typical of semi- were open for 3 consecutive nights. The presence of arid Australia: low and highly variable rainfall (mean medium to large mammals, including common wallaroo annual rainfall at Alice Springs Airport 283.7 mm) and (Osphranter robustus) and black-footed rock-wallaby temperatures ranging from hot in the austral summer (Petrogale lateralis) (both macropods) and the short- (daytime maximum frequently > 40 °C) to cool in win- beaked echidna (Tachyglossus aculeatus), was deter- ter (overnight minimum frequently < 0 °C) (Australian mined by detection of fresh scats within each site quadrat. Bureau of Meteorology Climate Data). The study area Scats of each species are discernible in the field based on comprises a highly variegated landscape dominated by shape, size, and color (Triggs 2004), and we considered 3 main landform types: high-elevation (900–1389 m) them fresh if they had a glossy black appearance (macrop- quartzite ridges and mountains; medium-elevation (600– ods) or were mostly intact (echidna). 900 m) rocky ridges, foothills, stony flats, and rises on var- Fieldwork was carried out with approval from Charles ious metamorphic and sedimentary geologies; and river Darwin University (ethical approval A06028) and under channels and alluvial flats with deep sandy or loamy a scientific permit from the Northern Territory Parks and soils (<600 m). Vegetation is diverse, and the high- Wildlife Commission (permit 44636). elevation ridges and mountains typically have ground- cover dominated by spinifex grasses (Triodia spp.) with scattered low shrubs and mallees (Eucalyptus). The Predictor Variables for Native-Mammal Richness lower rocky ridges and stony flats are dominated by either spinifex grasslands or acacia (principally Acacia For productivity, we predicted that the habitats with the aneura) shrublands, creating a mosaic of shrubland and deepest soils and run-on hydrology (alluvial and acacia) grassland. River channels are lined by river red gums would be the most productive (i.e., highest primary pro- (Eucalyptus camaldulensis), and mixed woodlands ductivity). Therefore, we assigned sites to categorical (ironwood [Acacia estrophiolata], corkwood [Hakea variables representing habitats (e.g., alluvial [1] or other divaricata],andpricklyacacia[Acacia victoriae]) pre- [0]). We estimated site primary productivity with the nor- dominate on the alluvial flats, where groundcover is malized difference vegetation index (NDVI) derived from dense tussock grass, including the invasive couch (Cyn- MODIS imagery (http://remote-sensing.nci.org.au/u39/ odon dactylon) and buffel grasses (Cenchrus ciliaris) public/html/modis/lpdaac-mosaics-cmar/) and used 2 (Supporting Information). continuous variables: 16-day NDVI averaged across all We used fine-scale biophysical mapping geographic periods of mammal sampling (NDVI mean), and 16-day information system (GIS) layers to select 90 sites across NDVI at the end of the wettest cumulative 3 months the length of the park (Supporting Information) based of sampling (NDVI wet). To account for temporal varia- on the availability of landform and vegetation types. Sites tion in productivity, we derived cumulative rainfall val- were stratified across the 4 dominant habitats: moun- ues for the 12 months prior to sampling at each site tain, high-elevation quartzite ridges and mountains domi- (rainfall). Rainfall data were sourced from the nearest nated by spinifex grasslands and sparse heath-like shrubs; weather station at Alice Springs airport (number 015590) spinifex, Trioidia grasslands; acacia, Acacia shrublands (http://www.bom.gov.au/climate/data/?ref=ftr). (spinifex and acacia sites in the medium-elevation rocky For the predation hypothesis, we predicted that hills and stony flats); and alluvial, river channels, and allu- mountain and spinifex sites would provide the most vial plain woodlands (Supporting Information). At each protection from predation because they are relatively site, we defined an 80 × 80 m quadrat within a contin- more topographically and vegetatively complex. uous patch of homogenous vegetation. We maintained Mountain sites were characterized by high levels of a minimum distance of 250 m between sites to increase ruggedness, abundant rock crevices, and frequently a the likelihood of independence. dense groundcover of spinifex (Supporting Information). Although less rugged, spinifex on relatively low rocky hills was often characterized by dense groundcover of Tri- Mammal Sampling odia spp. (Supporting Information). The spiny, densely We sampled each site over one 4-day period between packed leaves and hummock formation of spinifex grass September 2011 and October 2013. Clusters of sites provide shelter for small mammals from large predators. were associated with regions of the study area and in- Mountain and spinifex sites were also included in models cluded the range of habitats in that region (Support- as categorical variables. The 3 remaining variables ing Information). The exception to this was the moun- were ruggedness (rugged 100), calculated as standard tain stratification, where 16 sites were sampled, all in deviation of mean elevation within a 100-m radius of September 2011. each site from a 1 s digital elevation model (https:// We sampled small mammals with Elliott box traps researchdata.ands.org.au/national-elevation-framework- baited with peanut butter and oats. An array of 25 evenly web-portal/574736); percentage of cover of all grasses
Conservation Biology 24 Volume 31, No. 5, 2017 1186 Mesopredator-Driven Mammal Extinction
Table 1. Generalized linear models explaining current species richness of native mammals in the MacDonnell Ranges Bioregion, Australia.
Model (coefficient, SE)a Hypothesis AICb Deviance explained (%)
Rugged 100 (0.43, 0.09) + grass (0.32, 0.11) predation refuge 208.7 31 Rugged 100 (0.45; 0.09) predation refuge 215.2 23 Wet NDVI (−0.37, 0.12) + rain (0.31, 0.10) productivity 217.6 23 Wet NDVI (−0.44, 0.12) productivity 224.9 14 Rain (0.38, 0.11) productivity 226.4 12 Grass (0.33, 0.10) predation refuge 228.8 10 aVariable definitions: rugged 100, standard deviation of mean elevation within a 100-m radius of each site; wet NDVI, 16-day NDVI at the end of the wettest cumulative 3 months of sampling; rain, cumulative rainfall values for the 12 months prior to sampling at each site; grass, grass cover measured using point-intersect methods along 2 × 50 m lengths of tape within each site. bAkaike’s information criterion.
(grass all); and percentage of cover of spinifex grass published and unpublished data collected within the (grass spinifex). Grass cover was measured using point- bioregion to develop a total extant assemblage for each intersect methods along 2 × 50 m lengths of tape within habitat (Supporting Information). Records in transitional each 80 × 80 m quadrat. We did not consider fire-history areas (e.g., gorges between the alluvial and mountain variables because these are highly correlated with grass habitats) were not included. We reconstructed historic cover in our study area (McDonald et al. 2016). (at the time of European colonization) native mammal assemblages for each habitat based on subfossil remains collected from the eastern part of the study area (Baynes Modeling Extant Species Richness & Johnson 1996) and based on indigenous ecological We used generalized linear models with a Poisson dis- knowledge sourced from aboriginal traditional owners tribution to assess the influence of predation protection (Burbidge et al. 1988) and historical accounts (e.g., Gould and productivity on site-scale mammal species richness. 1863) (Supporting Information). Because na¨ıve species richness counts can be biased by detectability, we calculated richness estimates for the trapping data with the jacknife2 procedure (Smith & van Belle 1984) in EstimateS version 9.1.0. Na¨ıve species Results richness was highly correlated with predicted species Drivers of Current Mammal Species Richness richness (Kendall rank correlation = 0.91 [Supporting Information]), so we built the model with na¨ıve rich- We recorded 10 native mammal species in our survey ness. We screened standardized continuous covariates sites (Supporting Information). The predation refuge hy- for collinearity with variance-inflation factors (VIFs). The pothesis explained richness in that the 2 highest Akaike’s 2 pairs of NDVI and grass-cover variables had VIFs ࣙ 3, information criterion–ranked models supported this hy- so we removed one of each (NDVI and grass spinifex). pothesis and species richness was related positively to Collinearity between habitats and the remaining continu- ruggedness and grass cover (Table 1; Fig. 1). In the ous covariates was examined with conditional box plots. productivity models, the relationship between species These revealed evidence of collinearity, so we removed richness and productivity was in the opposite direction the categorical habitat factors from the analyses. We predicted by the productivity hypothesis: richness was tested for spatial autocorrelation with a saturated model related negatively to NDVI (Table 1 & Fig. 1). Although with all remaining variables. Using the residuals from this richness and antecedent rainfall were positively related model, we ran Moran’s I test in ArcMAP version 10.2 and (Table 1 & Fig. 1), this was likely an artifact of sam- found no evidence of spatial autocorrelation (z =−0.01, pling effort because we sampled the most rugged sites P = 0.55). For each hypothesis, we included all model following the wettest 12 months. The 6 native mammal combinations of covariates and ran models in R version species recorded from mountain sites are resident in this 3.3.0 (R Development Core Team 2016). Dispersion fac- habitat, and none of the 3 potentially irruptive species tors in all models did not differ substantially from 1.0, (Pseudantechinus macdonnellensis, Pseudomys deser- and model validation indicated that model assumptions tor,andZyzomys pedunculatus) were irrupting at the were met. time of sampling. This suggests that rainfall had little ef- fect on mammal assemblages at this time. Specifically, we trapped P. macdonnellensis and Z. pedunculatus from Comparing Extant and Historical Mammal Assemblages 4 of 16 sites, which was much lower than or similar to We compiled data on native mammal occurrence col- their occupancy recorded in 2015 (preceded by 3 years lected for this study (last 5 years) and from our own of below-average rainfall in the region), respectively. We
Conservation Biology 25 Volume 31, No. 5, 2017 McDonald et al. 1187
Figure 1. Relationship between site-scale native mammal species richness and standardized (a) ruggedness, (b) grass cover, (c) normalized difference vegetation index (NDVI), and (d) rainfall from univariate generalized linear models (lines, fitted estimates; shading, 95% CIs). trapped P. desertor from only 1 mountain site. This was species (Antechinomys laniger, Dasycercus cristicauda, a slightly lower occupancy level than for this species in and Sminthopsis ooldea) below the CWR were not found 2015 (McDonald et al. 2016). in the study area, although all were extant in nearby regions (Fig. 2). All species above the CWR occurred in their historic habitat types. The proportional loss of Loss of CWR Mammals CWR species was correlated negatively with rugged- ness and positively with productivity (wet period NDVI) Four native mammal species not detected from the 90 sur- (Fig. 3). Across the 4 habitats, there was an overall signif- vey sites have been recorded in Tjoritja National Park in icant difference in ruggedness (one-way ANOVA F3,86 = the last 5 years (regarded as resident). CWR species were 60.27, P ࣘ 0.001) and wet NDVI (one-way ANOVA F3,86 extirpated from all habitats, and at least 12 species were = 20.52, P ࣘ 0.001). extirpated from the region (Fig. 2 & Supporting Informa- tion). Losses were most pronounced in alluvial habitat, at least 12 CWR species. In contrast, the mountain habitat Discussion supported the most intact mammal assemblage (retained 4 of 6 CWR species) (Fig. 3). In spinifex habitat, 1 of Although a previous continent-scale analysis showed that 6 CWR species was extant (Fig. 1). Acacia habitat lost geographic regions with fewer arboreal, rock-dwelling, all 7 of its original CWR species. Three small mammal or burrowing taxa lost more species than other regions in
Conservation Biology 26 Volume 31, No. 5, 2017 1188 Mesopredator-Driven Mammal Extinction
Figure 2. Historic and extant occurrence of nonvolant native mammals in the 4 main habitats (landform or vegetation community) of the MacDonnell Ranges Bioregion, Australia (dots and lines, median weights, and weight ranges for individual species, respectively; shaded area, critical weight range [defined in text] of 35–5500 g).
Australia (McKenzie et al. 2007), our results revealed the for arboreal, burrowing, and ground-sheltering fauna— degree to which habitat variability can mediate extirpa- likely supported the highest number of mammal species tion and extinction and shape an extant dryland mammal (n = 19), whereas the other 3 habitat types probably fauna. In the MacDonnell Ranges, CWR mammals have supported fewer species (n = 10–13). Therefore, it seems been decimated in all but the most rugged habitat type; more likely that present-day richness is a legacy of mam- severity of loss was correlated negatively with ruggedness mal range contraction and extinction. and positively with primary productivity. Most species There is compelling evidence that predation is the that occurred only in alluvial or acacia habitats have been dominant driver of extirpation and extinction of CWR extirpated. These results suggest that extant mammal as- species across Australia, particularly in the relatively semblages are largely the product of differences across undisturbed dryland regions (reviewed in Woinarski et al. habitats in predation impact from the feral cat and red 2015). Through our reconstruction of historic mammal fox (see below) and challenge the prevailing view that assemblages, we identified at least 12 CWR species that higher order consumers are constrained primarily by spa- no longer occur in the region, whereas most species be- tiotemporal variation in productivity (Morton et al. 2011; low and all species above the CWR were extant. Although Free et al. 2015). we acknowledge the potential for interactive effects be- Although our models of site-scale mammal richness are tween invasive predators and habitat disturbances (e.g., consistent with the top-down predation hypothesis, they fire, herbivore grazing [Doherty et al. 2015]), we inter- did not causally implicate cats and red foxes in mammal preted our data on the loss of CWR species as consistent declines in the MacDonnell Ranges. For example, it is with the role of cats and foxes in Australia’s high rate of possible that the most species-rich and complex moun- contemporary mammal extinctions. tain habitats simply offered more ecological niches for Habitat may mediate the impact of predation by in- native mammals (e.g., Freeland et al. 1988). However, our troduced mesopredators by governing predator density. conservative reconstruction of historical mammal assem- Supporting this, red foxes were absent from mountain blages showed that alluvial habitats—which offer niches habitat (McDonald et al. 2016) and cats occurred at lower
Conservation Biology 27 Volume 31, No. 5, 2017 McDonald et al. 1189
The relatively lower density of cats in mountain habitat may relate to decreased foraging efficiency associated with increased ruggedness and grass cover. In support of this hypothesis, McGregor et al. (2015) found that cat hunting success in northern Australia increased from 17% in habitats with complex rock structure or dense grass to 70% in flatter, open areas. Relatively low prey densities or reduced access to prey can translate to larger home ranges (Herfindal et al. 2005) and therefore low cat densities (Bengsen et al. 2016). Based on predator– prey theory, lower levels of predation could enable the persistence of CWR species at low densities (Sinclair et al. 1998). An alternative explanation for the higher propor- tional loss of mammal species in productive habitats is that these habitats were the most heavily degraded by introduced rabbits and livestock (Stafford Smith & Mor- ton 1990; Southgate et al. 2007). However, 6 of the 12 CWR species that historically occurred in alluvial habi- tats (Dasyurus geoffroii, Phascogale calura, Chaeropus ecaudatus, Isoodon auratus, Bettongia lesueur, Lepo- rillus apicalis) also occurred in other habitats , including those relatively unaffected by rabbits and livestock (e.g., mountains, sandy deserts) (Burbidge et al. 1988). All 6 of these species are extirpated or extinct from Australia’s arid zone, and the implication is that a pervasive causal mechanism or mechanisms are responsible (e.g., feral cats [Legge et al. 2017]). Among the remaining species, Macrotis lagotis has contracted northward away from areas of high rabbit and livestock density (McDonald et al. 2015), although predation by cats and foxes has been im- plicated in failed reintroduction attempts (Moseby et al. Figure 3. Relationship between the proportion of 2011); Trichosurus vulpecula persisted in alluvial sites mammal species lost and mean ruggedness and mean until at least the mid-1990s and did not recolonize this productivity (wet period NDVI) measured in the 4 habitat after the occurrence of rabbit hemorrhagic dis- main habitat types of the MacDonnell Ranges ease or since the destocking of national parks and re- Bioregion, Australia (error bars, 1 SD from the mean). serves over the last 2 decades (Foulkes 2001; McDonald Sample size for each point (left to right): 16 et al. 2015); Notomys longicaudatus favored clay soils, (mountain), 32 (spinifex), 29 (acacia), and 13 which were generally subject to high levels of degrada- (alluvial). tion from livestock (Gould 1863); and Rattus tunneyi was probably also affected by livestock because reduc- tion in grass cover amplified predation by feral cats and densities in this habitat relative to alluvial habitat (data other predators (Leahy et al. 2016). Therefore, although included in Legge et al. 2017). A similar pattern was habitat degradation was probably a factor in the disap- observed in a region of northern Australia, where feral pearance of some CWR mammals from the more produc- cat occupancy was significantly lower in topographically tive habitats of Australia’s arid zone, available evidence complex habitats than in neighboring simple habitats implicates predation by introduced mesopredators in the (Hohnen et al. 2016). Further, in the same region, Mc- extinction of most species. Gregor et al. (2014) found that GPS-collared feral cats Irrespective of the mechanisms underpinning differ- avoided rocky hills and areas of dense grass cover. In ences in extirpation and extinction across habitat types, our study area, the absence of foxes may relate to the our results highlight the importance of large areas of unavailability of den sites (Carter et al. 2012); absence rugged terrain—some of which are already identified of the introduced European rabbit (Oryctolagus cunicu- as significant evolutionary refugia (Pepper et al. 2013; lus), an important food source (Read & Bowen 2001); and Oliver & McDonald 2016)—as refuges for CWR mammals. competitive exclusion from dingoes (Canis lupus dingo), Although it is unclear whether other mountain systems the apex predator in this system (Letnic et al. 2009). in Australia’s drylands share the refuge characteristics of
Conservation Biology 28 Volume 31, No. 5, 2017 1190 Mesopredator-Driven Mammal Extinction the MacDonnell Ranges, there is evidence that rugged Free CL, Baxter GS, Dickman CR, Lisle A, Leung LP. 2015. Di- regions of northern Australia have also protected CWR versity and community composition of vertebrates in desert mammals. Most notably, the highly dissected plateaus of river habitats. PLOS ONE 10 (e0144258) https://doi.org/10.1371/ journal.pone.0144258. the northern Kimberley support an intact fauna, where Freeland WJ, Winter JW, Raskin S. 1988. Australian rock-mammals: a no CWR mammals have been extirpated since European phenomenon of the seasonally dry tropics. Biotropica 1:70–79. colonization (Start et al. 2007). In the face of severe Gould J. 1863. The mammals of Australia. Volume 3.Theauthor,Lon- and ongoing mammal declines across northern Australia don. (Woinarski et al. 2011), we predict that other areas of Herfindal I, Linnell JD, Odden J, Nilsen EB, Andersen R. 2005. Prey den- sity, environmental productivity and home-range size in the Eurasian high topographic complexity (e.g., Arnhem escarpment) lynx (Lynx lynx). Journal of Zoology 265:63–71. also may provide important refuges for CWR species. Hohnen R, Tuft K, McGregor HW, Legge S, Radford IJ, Johnson CN. 2016. Occupancy of the invasive feral cat varies with habi- tat complexity. PLOS ONE 11(e0152520) https://doi.org/10.1371/ journal.pone.0152520. Acknowledgments Holmgren M, et al. 2006. Extreme climatic events shape arid and semiarid ecosystems. Frontiers in Ecology and the Environment The Traditional Owners of Tjoritja NP gave us permis- 4:87–95. sion to undertake fieldwork. Rangers from the Parks and Johnson CN, Isaac JL. 2009. Body mass and extinction risk in Australian marsupials: the ‘Critical Weight Range’ revisited. Austral Ecology Wildlife Commission of the NT, Tjuwanpa Ranger Group, 34:35–40. and Ingkerreke Outstations Resource Services provided Leahy L, Legge SM, Tuft K, McGregor HW, Barmuta LA, Jones ME, assistance in the field. Johnson CN. 2016. Amplified predation after fire suppresses ro- dent populations in Australia’s tropical savannas. Wildlife Research 42:705–716. Legge S, et al. 2017. Enumerating a continental-scale threat: How many Supporting Information feral cats are in Australia? Biological Conservation 206:293–303. Letnic M, Crowther MS, Koch F. 2009. Does a top-predator provide Further details on the sampling sites, habitat characteris- an endangered rodent with refuge from an invasive mesopredator? Animal Conservation 12:302–312. tics, and the historical and extant mammal assemblages Letnic M, Tamayo B, Dickman CR. 2005. The responses of mammals are available online (Appendix S1). The authors are solely to La Nina (El Nino Southern Oscillation)–associated rainfall, pre- responsible for the content and the functionality of these dation, and wildfire in central Australia. Journal of Mammalogy materials. Queries (other than the absence of the mate- 86:689–703. rial) should be directed to the corresponding author. Lima M, Stenseth NC, Jaksic FM. 2002. Food web structure and climate effects on the dynamics of small mammals and owls in semi-arid Chile. Ecology Letters 5:273–284. McDonald PJ, Luck GW, Dickman CR, Ward SJ, Crowther MS. 2015. Literature Cited Using multiple-source occurrence data to identify patterns and drivers of decline in arid-dwelling Australian marsupials. Ecography Baynes A, Johnson KA. 1996. The contributions of the Horn Expedition 38:1090–1100. and cave deposits to knowledge of the original mammal fauna of McDonald PJ, Stewart A, Schubert AT, Nano CE, Dickman CR, Luck central Australia. Pages 168–186 in Morton SR, Mulvaney DJ, editors. GW. 2016. Fire and grass cover influence occupancy patterns of Exploring central Australia: society, the environment and the 1894 rare rodents and feral cats in a mountain refuge: implications for Horn expedition. Surrey Beatty & Sons, Sydney. management. Wildlife Research 43:121–129. Bengsen AJ, et al. 2016. Feral cat home-range size varies predictably with McGregor HW, Legge S, Jones ME, Johnson CN. 2014. Land- landscape productivity and population density. Journal of Zoology scape management of fire and grazing regimes alters the fine- 298:112–120. scale habitat utilisation by feral cats. PLOS ONE 9 (e109097) Burbidge AA, Johnson KA, Fuller PJ, Southgate RI. 1988. Aboriginal https://doi.org/10.1371/journal.pone.0109097. knowledge of the mammals of the central deserts of Australia. McGregor H, Legge S, Jones ME, Johnson CN. 2015. Feral cats are better Wildlife Research 15:9–39. killers in open habitats, revealed by animal-borne video. PLOS ONE Burbidge AA, McKenzie NL. 1989. Patterns in the modern decline of 10 (e0133915) https://doi.org/10.1371/journal.pone.0133915. Western Australia’s vertebrate fauna: causes and conservation impli- McKenzie NL, et al. 2007. Analysis of factors implicated in the recent de- cations. Biological Conservation 50:143–198. cline of Australia’s mammal fauna. Journal of Biogeography 34:597– Carter A, Luck GW, Wilson BP. 2012. Ecology of the red fox (Vulpes 611. vulpes) in an agricultural landscape. 1. Den-site selection. Australian Morton SR, et al. 2011. A fresh framework for the ecology of arid Mammalogy 34:145–154. Australia. Journal of Arid Environments 75:313–329. Chisholm R, Taylor R. 2007. Null-hypothesis significance testing and Moseby KE, Read JL, Paton DC, Copley P, Hill BM, Crisp HA. 2011. the Critical Weight Range for Australian mammals. Conservation Predation determines the outcome of 10 reintroduction attempts in Biology 21:1641–1645. arid South Australia. Biological Conservation 144:2863–2872. Doherty TS, Dickman CR, Nimmo DG, Ritchie EG. 2015. Multiple Nano CE, Pavey CR. 2013. Refining the ‘pulse-reserve’model for arid threats, or multiplying the threats? Interactions between invasive central Australia: seasonal rainfall, soil moisture and plant produc- predators and other ecological disturbances. Biological Conserva- tivity in sand ridge and stony plain habitats of the Simpson Desert. tion 190:60–68. Austral Ecology 38:741–753. Foulkes JN. 2001. The ecology and management of the common brush- Oliver PM, McDonald PJ. 2016. Young relicts and old relicts: a novel tail possum Trichosurus vulpecula in central Australia. PhD disser- paleoendemic vertebrate from the Australian Central Uplands. Royal tation. University of Canberra, Canberra. Society Open Science 3:160018.
Conservation Biology 29 Volume 31, No. 5, 2017 McDonald et al. 1191
Pavey CR, Addison J, Brandle R, Dickman CR, McDonald PJ, Moseby Southgate R, Paltridge R, Masters P, Ostendorf B. 2007. Modelling in- KE, Young LI. 2015. The role of refuges in the persistence of troduced predator and herbivore distribution in the Tanami Desert, Australian dryland mammals. Biological Reviews. https://doi.org/ Australia. Journal of Arid Environments 68:438–464. 10.1111/brv.12247. Stafford Smith DM, Morton SR. 1990. A framework for the ecol- Pepper M, Doughty P, Keogh JS. 2013. Geodiversity and endemism in ogy of arid Australia. Journal of Arid Environments 18:255– the iconic Australian Pilbara region: a review of landscape evolution 278. and biotic response in an ancient refugium. Journal of Biogeography Start AN, Burbidge AA, McKenzie NL, Palmer C. 2007. The status of 40:1225–1239. mammals in the North Kimberley, Western Australia. Australian R Development Core Team. 2016. R: a Language and Environment for Mammalogy 29:1–16. Statistical Computing. R Foundation for Statistical Computing, Vi- Thackway R, Cresswell ID. 1995. An interim biogeographic regional- enna. isation for Australia: a framework for setting priorities in the na- Read J, Bowen Z. 2001. Population dynamics, diet and aspects of the tional reserves system cooperative. Australian Nature Conservation biology of feral cats and foxes in arid South Australia. Wildlife Re- Agency, Canberra. search 28:195–203. Triggs B. 2004. Tracks, scats, and other traces. Oxford University Press, Schlesinger WH, Pilmanis AM. 1998. Plant-soil interactions in deserts. Melbourne. Pages 169–187 in Van Breeman N, editor. Plant-induced soil Whitford WG. 2002. Ecology of desert systems. Academic Press, Cali- changes: processes and feedbacks. Springer, The Netherlands. fornia. Schwinning S, Sala OE. 2004. Hierarchy of responses to resource pulses Woinarski JC, Burbidge AA, Harrison PL. 2015. Ongoing unraveling of in arid and semi-arid ecosystems. Oecologia 141:211–220. a continental fauna: decline and extinction of Australian mammals Sinclair ARE, Pech RP, Dickman CR, Hik D, Mahon P, Newsome AE. since European settlement. Proceedings of the National Academy of 1998. Predicting effects of predation on conservation of endangered Sciences 112:4531–4540. prey. Conservation Biology 12:564–575. Woinarski JCZ, et al. 2011. The disappearing mammal fauna of northern Smith EP, van Belle G. 1984. Nonparametric estimation of species rich- Australia: context, cause, and response. Conservation Letters 4:192– ness. Biometrics 40:119–129. 201.
Conservation Biology 30 Volume 31, No. 5, 2017 Chapter 4. Diet of dingoes and cats in central Australia: does trophic competition underpin a
rare mammal refuge?
Counts Point in the MacDonnell Ranges, an important refuge for the central rock-rat and other threatened mammals (photo: P. McDonald)
Author contribution: I was responsible for the chapter concept, overseeing scat collection, cat scat analysis (G. Story undertook dingo scat analysis), data analysis, and writing, in consultation with my co-authors. Supporting information is provided for this chapter in Appendix 5.
31 Journal of Mammalogy, 99(5):1120–1127, 2018 DOI:10.1093/jmammal/gyy083 Published online July 20, 2018
Diet of dingoes and cats in central Australia: does trophic competition underpin a rare mammal refuge? Downloaded from https://academic.oup.com/jmammal/article-abstract/99/5/1120/5056485 by Burkitt-Ford Library user on 24 October 2018 Peter J. McDonald,* Jayne Brim-Box, Catherine E. M. Nano, David W. Macdonald, and Chris R. Dickman Flora and Fauna Division, Department of Environment and Natural Resources, Alice Springs, Northern Territory 0870, Australia (PJM, JB-B, CEMN) Desert Ecology Research Group, School of Life and Environmental Sciences, University of Sydney, New South Wales 2006, Australia (CRD) Wildlife Conservation Research Unit, Department of Zoology, The Recanati-Kaplan Centre, University of Oxford, Oxford OX13 5QL, United Kingdom (DWM) * Correspondent: [email protected]
We investigated the hypothesis that trophic competition between a top predator and a smaller predator can create refuge from predation for small mammalian prey, using the dingo (Canis lupus dingo) and feral cat (Felis catus) in the MacDonnell Ranges of dryland Australia as a case study. We analyzed the diets of the 2 predator species for evidence of potential competition. There was no evidence of exploitation competition between the 2 carnivores— cats consumed mostly small mammals and particularly larger rodents, whereas the diet of dingoes was dominated by 1 species of large macropod. There was also no evidence of a shift in diet of cats, as their diets in refuges and non-refuges were highly overlapping. Consistent with interference competition, cats were the third most frequently consumed mammal species by dingoes. Although predation by dingoes could limit densities of cats across the MacDonnell Ranges, this alone does not explain why the most rugged habitats in the region are a refuge for rare mammals. We conclude that habitat complexity most likely underpins the refuge and that possible effects of dingo predation on the cat population would be of secondary importance.
Key words: activity, arid, Canis lupus dingo, competition, diet, Felis catus, mesopredator
Australia has a highly distinctive mammal fauna that has on the Australian mainland has also been linked to variation been severely impacted since European colonization in 1788. in habitat complexity. For example, habitat refuges for threat- At least 30 endemic mammal species (> 10% of the original ened mammals are typically associated with complex terrain or mammal fauna) became extinct in this period and a further vegetation (Hernandez-Santin et al. 2016; Davies et al. 2017; 56 species meet the IUCN criteria for listing under one of the McDonald et al. 2017). threatened categories (Woinarski et al. 2015). In contrast to the The quartzite mountains of the MacDonnell Ranges in cen- global situation, where habitat loss and hunting are the main tral Australia have been identified as an important refuge for factors threatening mammals (Hoffman et al. 2011), Australia’s small to medium-sized threatened mammals (McDonald et al. mammal extinctions and declines have probably been driven 2015, 2017). These mountains support the most intact mam- primarily by predation from 2 introduced mesopredators: the mal fauna in central Australia and several species are now feral cat (Felis catus; hereafter referred to as “cat”) and red fox regionally or globally restricted to this refuge. In contrast to (Vulpes vulpes—Woinarski et al. 2015). the quartzite mountains, the surrounding landforms (includ- While Australia’s mammal extinctions and declines have ing lower-elevation rocky hills, valleys, and alluvial plains) are been exceptional on a global scale, declines have not been geo- characterized by relatively simple topography (McDonald et al. graphically uniform. For example, 3 species of native mammal 2017). McDonald et al. (2017) hypothesized that the rugged that once occurred widely on mainland Australia are now con- and structurally complex quartzite geology mediates preda- fined to Australia’s largest fox-free island, Tasmania (Woinarski tion from cats by affecting their foraging efficiency and den- et al. 2015). Variation in mesopredator-driven mammal decline sity. While ruggedness was found to be a more important driver
© 2018 American Society of Mammalogists, www.mammalogy.org
1120 32 MCDONALD ET AL.—CAT AND DINGO COMPETITION 1121 of mammal assemblages than productivity in the MacDonnell dingoes and cats (Paltridge 2002; Pavey et al. 2008; Spencer Ranges (McDonald et al. 2017), there remains an additional et al. 2014), no dietary research has been undertaken in the bio- possible explanation for why this region is a refuge for rare logically distinct central Australian uplands. The MacDonnell mammals—top-down suppression of cats by dingoes (Canis Ranges differ from the sandy deserts in their complex topogra- lupus dingo). phy (McDonald et al. 2015), variegated and well-defined veg- Dingoes and cats are the 2 largest mammalian predators res- etation communities (Nano and Clarke 2008), and abundant ident in the MacDonnell Ranges. Red foxes are infrequently natural surface water (Box et al. 2008). This environment sup- Downloaded from https://academic.oup.com/jmammal/article-abstract/99/5/1120/5056485 by Burkitt-Ford Library user on 24 October 2018 recorded in the region and are absent from the core area of ports a substantial population of a large macropod, the euro or upland terrain (see McDonald et al. 2017). While cats are ubiq- hill kangaroo (Osphranter robustus—McDonald et al. 2017), uitous throughout dryland Australia, available data suggest that which thrives here because of access to abundant areas of shade densities of cats are lower in the quartzite refuge than in nearby (afforded by caves, overhangs, and vegetation) used as shelter topographically simple habitats (Legge et al. 2017). The dingo, during the day and surface water (Ealey et al. 1965). Unlike Australia’s apex mammalian predator, occurs throughout the cats, dingoes are large enough to capture and subdue large MacDonnell Ranges probably as a consequence of the presence mammal prey, particularly when hunting in packs (Corbett of extensive protected areas with abundant surface water and 1995). If euros dominate the diet of dingoes in the MacDonnell the absence of lethal control. Given the widespread reporting Ranges, this would suggest that dingoes and cats have highly of suppression of cats by dingoes (e.g., Kennedy et al. 2012; divergent dietary ecologies and render competition between the Moseby et al. 2012; Greenville et al. 2014), there could be 2 predators less likely (Keddy 2001). important interactions between the 2 predators in this system. Here, we investigated the hypothesis that dingoes are an The theoretical mechanisms for suppression of cats by important trophic regulator that suppress cats, and thus help to dingoes are exploitation and interference competition. sustain a refuge for rare mammals, in the MacDonnell Ranges. Exploitation competition occurs between 2 species when We examined the diets of both predators from scats collected there is high niche overlap (Wiens 1993). For example, when inside and outside the refuge. Based on a scenario of exploi- there is high dietary overlap between a pair of species, one tation competition, if dingoes outcompete cats for prey, we species will outcompete the other in times of food shortage predicted either: 1) high overall dietary overlap between the (Korpimäki 1987). Exploitation competition may also drive 2 predators and thus the potential for fitness impacts on the niche shift in one species, forcing its increased use of a sub- subordinate predator during times of food shortage, or 2) com- optimal niche (Bonesi et al. 2004; Harrington et al. 2009). petition with dingoes would force cats to consume increased Interference competition occurs when one species limits quantities of suboptimal prey in the refuge (dietary niche shift). another’s use of resources (Wiens 1993). In carnivores, this However, if dingoes consumed mostly large mammals, we process includes intraguild predation and a fear of predation expected low dietary overlap between the 2 predators and thus that drives spatial and temporal avoidance of a larger carni- low potential for exploitation competition. Based on a scenario vore (Fedriani et al. 2000; Linnell and Strand 2000). These of interference competition, we expected evidence of intraguild phenomena are demonstrably important in intraguild relation- predation with a high proportion of cats in dingo scats. ships among carnivores, often with consequences for conser- vation (e.g., Herteinsson and Macdonald 1992; Sidorovich et al. 1999). Materials and Methods Understanding the trophic ecology of dingoes and cats is a Study region.—We conducted our study in the 2,592-km2 prerequisite to uncovering potentially important competitive Tjoritja−West MacDonnell National Park (referred to here- interactions between the 2 predators that help to maintain the after as “Tjoritja NP”) in the MacDonnell Ranges Bioregion mammal refuge. While studies from Australia’s sandy desert (Thackway and Cresswell 1995), southern Northern Territory, systems have found moderate to high dietary overlap between Australia (Fig. 1). Climate is typical of semiarid Australia, with
Fig. 1.—a) Location of the study area in the Northern Territory, Australia. b) Study area enlarged with the locations of cat (Felis catus; n = 74) and dingo (Canis lupus dingo; n = 98) scats collected in Tjoritja National Park (park boundary indicated by black line). Maps generated in ArcMap 10.2 (www.esri.com). Background imagery courtesy of Geoscience Australia (www.ga.gov.au). 33 1122 JOURNAL OF MAMMALOGY highly irregular rainfall (mean annual rainfall at Alice Springs (Plymouth Marine Laboratory—Anderson et al. 2008; Clarke Airport = 283.7 mm) and temperatures ranging from hot in and Gorley 2015) to explore dietary differences between the summer (daytime maxima frequently > 40°C) to cool in win- 2 species. We first used the similarity percentage (SIMPER) ter (overnight minima frequently < 0°C) (Australian Bureau of analysis procedure to determine within- and between-group Meteorology Climate Data, http://www.bom.gov.au/climate/ diagnostic food categories for each species. We then used dis- data/). The main landforms in the park are rugged quartzite tance-based linear models (DISTLMs) to analyze and model mountains and ridges (to 1,389 m elevation), lower rocky hills the relationship between the adjusted Bray–Curtis similarity Downloaded from https://academic.oup.com/jmammal/article-abstract/99/5/1120/5056485 by Burkitt-Ford Library user on 24 October 2018 and flats of varying geology, and ephemeral rivers and alluvial resemblance matrix of untransformed food category data and plains. Vegetation communities are generally well defined, with 6 categorical and continuous predictor variables. We sought to hummock grasslands (Triodia spp.) and Acacia (e.g., Acacia determine the effect of species versus a range of environmental aneura) shrublands dominating the rocky landforms, while parameters (refuge versus non-refuge, rainfall, and season of rivers and alluvial plains support river red gum (Eucalyptus collection) on food content. We used the Draftsman Plot tool camaldulensis) and ironwood (Acacia estrophiolata) wood- to test for a skewed distribution in the explanatory variables lands. We defined the rugged quartzite mountain ranges as a (indicating a requirement for transformation) and for colline- refuge because they support several species of mammals that arity among the variables. Redundant variables, those strongly no longer occur, or are very rare, outside of this landform correlated with other variables (r > 0.95), were removed from (McDonald et al. 2017). the analysis. Following this, we used the Forward Selection Collection and analysis of fecal remains.—Cat and dingo procedure on the basis of the adjusted R2 selection criterion and fecal remains (hereafter “scats”) were collected opportunisti- then carried out constrained ordination using distance-based cally throughout the study area (Fig. 1b). Cat and dingo scats redundancy analysis (dbRDA). were collected between 2011 and 2013 and cat scats were Because dietary overlap is frequently reported using Pianka’s also collected in 2015–2016. Only intact scats judged to be < index, we also calculated this for prey frequency occurrence 6 months old were used for analysis. We identified scats as cat and volume using the equation: or dingo according to size, shape, and smell (Triggs 1996). We 05. placed scats individually into paper bags and then into an oven Op= pp22p jk ååij ik ()ijå ik at 70°C for > 10 h to kill parasites. We then washed samples through a series of sieves that left only indigestible fragments where j and k are the 2 species being compared, and p is the of prey. We placed fragments into sorting trays divided into 4 i equal sections for inspection and visual estimation of percent- frequency of occurrence (or volume) of the ith food type. age volume of prey categories. Mammals were identified to the Overlap ranges from 0 (no overlap) to 1 (complete overlap). We lowest possible taxonomic level by inspection of hair remains computed Pianka’s index for cats and dingoes from all scats. using cross-section and whole-mount techniques, and jaw and Because cats have smaller home ranges than dingoes in dryland skull fragments (Archer 1981; Watts and Aslin 1981; Brunner Australia (Corbett 1995; Edwards et al. 2001), we expected and Triggs 2002). Mammals were classified into size catego- their scats deposited in the refuge were more likely to include ries of small (< 500 g), medium (500–6,999 g), and large (≥ prey consumed in the refuge, so we also calculated Pianka’s 7,000 g). All other prey items were categorized as arthropod, index for cats separately for refuge and non-refuge locations. reptile or frog, bird, vegetation, or rubbish. We compared Pianka’s index values to linear null models, using Analysis.—To determine whether our scat sample sizes were the randomization algorithm RA3 with 10,000 runs in EcoSim sufficient for capturing mammal species and dietary diversity, Professional Version 1 (Entsminger 2014). we plotted the cumulative diversity of all mammal species and the other food categories against the number of scats exam- ined for both cats and dingoes. We calculated diversity with the Results Brillouin index: Diet.—We collected and analyzed 98 dingo and 74 cat scats lnNn!! ln from across the study area (Fig. 1; Supplementary Data SD1). -å i H = Cumulative diversity of mammal species and other prey cat- N egories reached asymptote for dingoes and cats, indicating that where H is the dietary diversity of the predator, N is the total sampling was sufficient to reliably describe the diets of the number of individual prey recorded, and ni is the number of predators (Supplementary Data SD2). individual prey items of the ith type (Brillouin 1956). To test Similarity percentage analysis revealed that diets of dingoes for dietary overlap or partitioning between cats and dingoes, and cats were highly divergent in their primary prey consump- we constructed a scat by food category matrix that was based tion. The diet of cats was characterized primarily by small on the untransformed volumetric contribution of each category mammals (80.4% within-group similarity), while the diet of (Klare et al. 2011). For this, we used 172 scats and the 8 food dingoes was characterized by large mammals (78.4% within- categories. group similarity). Birds and arthropods were 2nd- and 3rd- We used a range of multivariate techniques available in the order contributors for cats, followed by reptiles or frogs and PRIMER 7 software package with PERMANOVA + add-on medium-sized mammals. Vegetation (mostly masticated grass
34 MCDONALD ET AL.—CAT AND DINGO COMPETITION 1123 likely consumed by euros and other large mammal prey) and of small mammals distinguished the diet of cats from that of medium-sized mammals were 2nd- and 3rd-order contributors dingoes; large mammals distinguished the diet of dingoes from for dingoes. Between-group dissimilarity analysis showed that that of cats; and while medium-sized mammals were present diets of cats and dingoes were distinguishable primarily on the in the diet of both species, they were more prevalent in the diet basis of the 3 mammal size classes (Table 1). A high proportion of dingoes. Vegetation and rubbish were more closely associ- ated with the diet of dingoes, while a higher content of birds,
reptiles, and arthropods distinguished the diet of cats from that Downloaded from https://academic.oup.com/jmammal/article-abstract/99/5/1120/5056485 by Burkitt-Ford Library user on 24 October 2018 Table 1.—Similarity percentage (SIMPER) results showing the average abundance (Av ab), average dissimilarity (Av diss), percent- of dingoes (Table 1). age contribution to overall dissimilarity (% cont), and cumulative per- The patterns in the SIMPER analysis were supported by the centage (Cum %) for cat (Felis catus) and dingo (Canis lupus dingo) DISTLM and the dbRDA (Fig. 2; Table 1). The Draftsman Plot dietary comparison in the MacDonnell Ranges, Northern Territory, tool revealed that past annual rainfall and past winter rainfall central Australia. were collinear and we removed the latter from the model. The marginal tests showed that 2 explanatory variables had a highly Food category Av ab cat Av ab dingo Av diss % cont Cum % significant relationship (P < 0.001, species and past annual Small mammal 54.66 2.8 26.93 28.57 28.57 rainfall), and 1 variable had a significant relationship (P < 0.01, Large mammal 0 49.74 24.87 26.39 54.95 position) with the multivariate dietary data cloud. These 3 vari- Medium mammal 8.45 15.36 10.53 11.17 66.12 Vegetation 1.08 18.29 9.37 9.94 76.06 ables produced the most parsimonious model; species (i.e., cat Bird 15.47 5.51 9.32 9.89 85.96 versus dingo) explained most of the variation in scat compo- Reptile 8.18 6.63 6.64 7.04 93 sition (adjusted R2 = 0.247) and the addition of position and Arthropod 12.16 0.87 6.2 6.57 99.57 then past annual rainfall resulted in a marginal increase in Rubbish 0 0.81 0.4 0.43 100 explanatory power (adjusted R2 = 0.253). The first 2 dbRDA axes captured 99.4% of the variability in the fitted model, but
Fig. 2.—a) Distance-based redundancy analysis (dbRDA) of food category volume data from the most parsimonious model with 3 explanatory variables, and b) the same dbRDA model with a vector overlay of food category abundance Pearson correlations with the dbRDA axes. L mammal = large mammal; M mammal = medium-sized mammal; S mammal = small mammal; Annual_rain = rainfall (mm) in 12 months prior to scat collection. 35 1124 JOURNAL OF MAMMALOGY
Table 2.—Observed (O) and expected (E, simulated mean) Pianka’s index for temporal, spatial, and dietary overlap. One-tailed P-values are based on 10,000 randomizations, and a priori predictions were based on competition theory. Frequency = % frequency occurrence of prey items in diet; Volume = % volumetric representation of prey items in diet.
Species Overlap type Location Predicted overlap Observed Expected P-value Dingo/cat Frequency All High 0.343 0.549 0.899 Dingo/cat Volume All High 0.133 0.376 0.933 Cat/cat Frequency Refuge/non-refuge Low 0.962 0.542 0.002 Downloaded from https://academic.oup.com/jmammal/article-abstract/99/5/1120/5056485 by Burkitt-Ford Library user on 24 October 2018 Cat/cat Volume Refuge/non-refuge Low 0.993 0.356 0.000 only 26.4% of the total variation in the data cloud (Fig. 2). Axis 1 explained most of the variation in scat composition and it was most strongly related to species and to a lesser extent past annual rainfall (multiple partial correlations = −0.931 and 0.353, respectively). The 2nd axis was most strongly related to landscape position and then to past annual rainfall (multi- ple partial correlations = −0.831 and −0.544, respectively). The vector overlay of food categories showed that large mammal and small mammal components were most strongly dissoci- ated along axis 1, aligning with dingo and cat, respectively. The same relationship was apparent with vegetation and medium- sized mammals (less so) versus arthropods and birds (less so), though the correlation was not as strong. Small mammals, arthropods, and birds (less so) were also weakly positively associated with past annual rainfall along axis 1. Along axis 2, medium-sized mammals were weakly associated with high landscape position. Consistent with the SIMPER and dbRDA analyses, dietary overlap between cats and dingoes was not higher than expected by chance (Table 2). Dietary overlap for cats between the ref- uge and non-refuge was significantly higher than expected by chance based on the broad food categories (Table 2). However, there were substantial differences in the proportions of mam- malian prey consumed by cats between the refuge and non- refuge. Within the refuge, the critically endangered central rock-rat (Zyzomys pedunculatus) was the dominant diet item (22.8% of total scat volume; 25% by frequency of occurrence), Fig. 3.—a) % volume of small mammal species in cat (Felis catus) followed by the fat-tailed antechinus (Pseudantechinus mac- scats in non-refuge (dark gray) and refuge habitats (light gray), and b) donnellensis; 19.0% vol.; 25% freq.), and house mouse (Mus % frequency of occurrence of small mammal species identified from cat scats collected in non-refuge (dark gray) and refuge habitats (light musculus; 2.8% vol.; 5.6% freq.; Fig. 3). Outside the refuge, gray) in the MacDonnell Ranges, Northern Territory, central Australia. the desert mouse (Pseudomys desertor) was the dominant diet item (26.2% vol.; 42.1% freq.), followed by the house mouse Discussion (7.6% vol.; 10.5% freq.), fat-tailed pseudantechinus (4.2% vol.; 10.5% freq.), and long-haired rat (Rattus villosissimus; 2.6% We investigated the hypothesis that dingoes suppress cats vol.; 2.6% freq.; Fig. 3). through trophic competition mechanisms and that this sup- Within the large and medium-sized mammal prey categories pression helps to sustain a refuge for rare mammals in the for dingoes, the euro dominated (30.2% vol.; 49.0% freq.), fol- MacDonnell Ranges. We found no evidence consistent with lowed by cattle (Bos taurus; 8.1% vol.; 10.2% freq.), cat (6.7% exploitation competition between the 2 predators and some vol.; 9.2% freq.), short-beaked echidna (Tachyglossus acu- evidence consistent with interference competition. Although leatus; 5.7% vol.; 7.1% freq.), horse (5.3% vol.; 8.2% freq.), predation by dingoes could limit densities of cats across the red kangaroo (Macropus rufus; 4.8% vol.; 5.1% freq.), rabbit region, it is hard to see how this could explain why the most (2.0% vol.; 2% freq.), common brushtail possum (Trichosurus rugged habitats in the region are a refuge for rare mammals. vulpecula vulpecula; 0.9% vol.; 1.0% freq.), and dingo (0.2% Cats and dingoes had highly divergent diets in the vol.; 1.0% freq.; Fig. 4). We found no incidence of dingo pre- MacDonnell Ranges, suggesting limited potential for exploita- dation on the central rock-rat or fat-tailed pseudantechinus and tion competition during periods of food shortage (Wiens 1993). only 1 incidence of predation on the desert mouse (0.0% vol.; We found that cats fed mostly on small mammals and particu- 1% freq.). larly rodents. Globally, cats are exceptional hunters of rodents 36 MCDONALD ET AL.—CAT AND DINGO COMPETITION 1125
mouse (McDonald et al. 2015, 2016). The preference for larger rodents presumably confers an energetic advantage for cats targeting these species (MacArthur and Pianka 1966) and pro- vides some support to the idea that predation by feral cats is an important factor in the ongoing declines of the central rock-rat and other critical weight range rodent species (McDonald et al.
2015, 2017; Davies et al. 2017). In the face of targeted preda- Downloaded from https://academic.oup.com/jmammal/article-abstract/99/5/1120/5056485 by Burkitt-Ford Library user on 24 October 2018 tion by cats, the persistence of the central rock-rat and desert mouse could be facilitated by the fine-scale protection afforded by rockiness and dense spinifex grass, respectively (McGregor et al. 2015; McDonald et al. 2016). Consistent with interference competition, cats were the third most frequently consumed mammal species by dingoes (6.7% vol.; 9.2% freq.). To our knowledge, this is the highest inci- Fig. 4.—% volume (dark gray) and frequency occurrence (light gray) dence of cat consumption by dingoes thus far recorded for dry- of medium and large mammal species identified from dingo (Canis land Australia (Paltridge 2002; Pavey et al. 2008; Doherty 2015) lupus dingo) scats in the MacDonnell Ranges, Northern Territory, cen- tral Australia. and possibly the highest incidence of canid consumption of a felid globally (Macdonald and Sillero-Zubiri 2004). While this suggests that predation by dingoes could maintain lower densi- and rabbits (Pearre and Maass 1998), which is also consistent ties of cats in the MacDonnell Ranges, even a high incidence with most studies from dryland Australia (Pavey et al. 2008; of intraguild predation may not have population-level impacts. Spencer et al. 2014; Doherty 2015). In contrast, dingoes are For example, in Tanzania, predation by African lions (Panthera highly flexible predators capable of consuming small, medium- leo) was the leading cause of juvenile mortality in cheetahs sized, and large mammal prey (Corbett 1995). This flexibility (Acinonyx jubatus—Laurenson 1994). However, despite a tri- relates to body size and sociality—dingoes (~13–15 kg) are pling of the lion population over 3 decades, the cheetah popula- large enough to capture and subdue large mammals, particu- tion remained relatively stable in the study area (Swanson et al. larly when hunting in packs (Corbett 1995), but small enough 2014). Similarly, in South Africa, lions accounted for > 20% of that they are not constrained by the energy requirements for leopard (Panthera pardus) mortality but did not suppress their large prey imposed on large carnivores (> 21.5 kg—Carbone population or distribution (Balme et al. 2017). Determining et al. 1999). The diet of dingoes in the MacDonnell Ranges whether dingo predation on cats is compensatory or additive was dominated by 1 species of large kangaroo, the euro, which will require manipulation of densities of dingoes (Newsome was also the most widely detected mammal species in our et al. 2015). Previous experimental studies (Allen et al. 2013, study area in 2011–2013 (McDonald et al. 2017). Therefore, 2018) have been unable to address this question because they the availability of a stable population of large kangaroos prob- could not effectively or consistently reduce dingo populations ably underpins the low likelihood of exploitation competition (Johnson et al. 2014). between cats and dingoes in the MacDonnell Ranges. We were unable to evaluate evidence for an additional We found no evidence that competition with dingoes has potential mechanism for suppression of cats by dingoes, that driven a dietary niche shift in the refuge—there was high over- foraging behavior or densities of cats are influenced by a “land- lap between the refuge and non-refuge diets of cats. The high scape of fear” associated with avoidance of dingoes (Kennedy incidence of central rock-rat and fat-tailed pseudantechinus et al. 2012; Greenville et al. 2014). For a “landscape of fear” remains in cat scats collected in the refuge supported our a to negatively influence cats at the population level, avoidance priori split of cat scats into refuge and non-refuge categories; of dingoes by cats must have an energetic cost. However, in these small mammal species are restricted to or more wide- the MacDonnell Ranges, even if cat activity was influenced by spread within the refuge, respectively (McDonald et al. 2015; dingoes, our data demonstrating that cats consumed their pre- McDonald et al. 2017). Our dietary data, together with previ- ferred rodent prey throughout the study area suggest that forag- ous data on the occurrence of small mammals in the study area, ing strategies of cats are not strongly influenced by dingoes in suggest that cats preferentially hunt larger rodents. Specifically, refuge or non-refuge locations. in non-refuge habitats the desert mouse (25 g), a specialist In summary, we found no evidence that dingoes affect cats inhabitant of dense spinifex grasslands (Letnic and Dickman through exploitative trophic competition in the MacDonnell 2005; McDonald et al. 2016), was the dominant small mam- Ranges; diets of cats and dingoes were highly divergent and mal prey, yet its occurrence is highly restricted compared with cats targeted their preferred small mammal prey in refuge and the smaller (12 g), habitat-generalist house mouse (McDonald non-refuge habitats. While we found a relatively high inci- et al. 2017). The house mouse was rarely consumed by cats. dence of dingo predation on cats, we do not know whether pre- Similarly, in the refuge, the central rock-rat (65 g) was domi- dation was compensatory or additive. Regardless of whether nant in the diet of cats despite having a more restricted occu- dingo predation influences densities of cats in the MacDonnell pancy than both the fat-tailed pseudantechinus (25 g) and house Ranges, predation does not explain why the most rugged
37 1126 JOURNAL OF MAMMALOGY habitats in the region are a refuge for rare mammals. We there- Box, J. B., et al. 2008. Central Australian waterbodies: the impor- fore conclude that habitat complexity, and its effect on forag- tance of permanence in a desert landscape. Journal of Arid ing efficiency of mammalian predators, remains the most likely Environments 72:1395–1413. mechanism underpinning the refuge (McDonald et al. 2017). Brillouin, L. 1956. Science and information theory. Academic Press, Dingo predation of cats is either of secondary importance (if New York. Brunner, H., and B. Triggs. 2002. Hair ID: an interactive tool predation is additive) or is not a factor (if predation is compen- for identifying Australian mammalian hair. CSIRO Publishing, Downloaded from https://academic.oup.com/jmammal/article-abstract/99/5/1120/5056485 by Burkitt-Ford Library user on 24 October 2018 satory) in contributing to the maintenance of the refuge for rare Collingwood, Victoria, Australia. small mammals. Carbone, C., G. M. Mace, S. C. Roberts, and D. W. Macdonald. 1999. Energetic constraints on the diet of terrestrial carnivores. Nature 402:286–288. Acknowledgments Clarke, K. R., and R. N. Gorley. 2015. Getting started with PRIMER We thank the traditional owners of Tjoritja/West MacDonnell v7. PRIMER-E, Plymouth, United Kingdom. NP for allowing access to the survey locations. Park rangers Corbett, L. K. 1995. The dingo in Australia and Asia. UNSW Press, from the NT Parks and Wildlife Commission assisted with the Sydney, New South Wales, Australia. collection of predator scats, particularly C. Stenhouse. G. Story Davies, H. H., et al. 2017. Top-down control of species distributions: feral cats driving the regional extinction of a threatened rodent in analyzed the dingo scat remains. E. Connellan from Mengel’s northern Australia. Diversity and Distributions 23:272–283. Heli Services flew us into all remote sites. Doherty, T. S. 2015. Dietary overlap between sympatric dingoes and feral cats at a semiarid rangeland site in Western Australia. Australian Mammalogy 37:219–224. Supplementary Data Ealey, E. H. M., P. J. Bentley, and A. R. Main. 1965. Studies on Supplementary data are available at Journal of Mammalogy water metabolism of the hill kangaroo, Macropus robustus (Gould), online. in northwest Australia. Ecology 46:473–479. Supplementary Data SD1.—Percent occurrence (% occur- Edwards, G. P., N. De Preu, B. J. Shakeshaft, I. V. Crealy, and rence) and percent volumetric (% volume) composition of prey R. M. Paltridge. 2001. Home range and movements of male feral types in dingo (Canis lupus dingo) and cat (Felis catus) scats col- cats (Felis catus) in a semiarid woodland environment in central lected in the MacDonnell Ranges, central Australia. The dingo Australia. Austral Ecology 26:93–101. Entsminger, G. L. 2014. EcoSim Professional: null modeling soft- data include all scats collected across the study area, whereas ware for ecologists, version 1. Acquired Intelligence Inc., Kesey- the cat data are separated into refuge and non-refuge locations. Bear, and Pinyon Publishing, Montrose, Colorado. See “Materials and Methods” section for more information. Fedriani, J. M., T. K. Fuller, R. M. Sauvajot, and E. C. York. 2000. Supplementary Data SD2.—Accumulation of prey diversity Competition and intraguild predation among three sympatric carni- with increasing scat sample size for the dingo (Canis lupus vores. Oecologia 125:258–270. dingo) and feral cat (Felis catus) in the MacDonnell Ranges, Greenville, A. C., G. M. Wardle, B. Tamayo, and C. R. Dickman. 2014. Northern Territory, central Australia. Bottom-up and top-down processes interact to modify intraguild inter- actions in resource-pulse environments. Oecologia 175:1349–1358. Harrington, L. A., et al. 2009. The impact of native competitors Literature Cited on an alien invasive: temporal niche shifts to avoid interspecific Allen, B. L., L. R. Allen, R. M. Engeman, and L. K. Leung. 2013. aggression. Ecology 90:1207–1216. Intraguild relationships between sympatric predators exposed to Hernandez-Santin, L., Goldizen, A. W., and D. O. Fisher. 2016. lethal control: predator manipulation experiments. Frontiers in Introduced predators and habitat structure influence range con- Zoology 10:39. traction of an endangered native predator, the northern quoll. Allen, B. L., A. Fawcett, A. Anker, R. M. Engeman, A. Lisle, Biological Conservation 203:160–167. and L. K. Leung. 2018. Environmental effects are stronger than Herteinsson, P., and D. W. Macdonald. 1992. Interspecific competi- human effects on mammalian predator-prey relationships in arid tion and the geographical distribution of red and arctic foxes Vulpes Australian ecosystems. The Science of the Total Environment vulpes and Alopex lagopus. Oikos 1992:505–515. 610:451–461. Hoffman, M., et al. 2011. The changing fates of the world’s mam- Anderson, M., R. N. Gorley, and R. K. Clarke. 2008. Permanova+ mals. Philosophical Transactions of the Royal Society of London for Primer: Guide to Software and Statistical Methods. Primer-E B: Biological Sciences 366:2598–2610. Limited, Plymouth. Johnson, C. N., et al. 2014. Experiments in no-impact control of din- Archer, M. 1981. Results of the Archbold Expeditions. No. 104. goes: comment on Allen et al. 2013. Frontiers in Zoology 11:17. Systematic revision of the marsupial dasyurid genus Sminthopsis Keddy, P. A. 2001. Competition. Springer, Dordrecht. Thomas. Bulletin of the American Museum of Natural History Kennedy, M., B. L. Phillips, S. Legge, S. A. Murphy, and 168:61–224. R. A. Faulkner. 2012. Do dingoes suppress the activity of feral Balme, G. A., T. Pitman, H. S. Robinson, J. R. Miller, P. J. Funston, cats in northern Australia? Austral Ecology 37:134–139. and L. T. Hunter. 2017. Leopard distribution and abundance is Klare, U., J. F. Kamler, and D. W. Macdonald. 2011. A comparison unaffected by interference competition with lions. Behavioral and critique of different scat‐analysis methods for determining car- Ecology 28:1348–1358. nivore diet. Mammal Review 41:294–312. Bonesi, L., P. Chanin, and D. W. Macdonald. 2004. Competition Korpimäki, E. 1987. Dietary shifts, niche relationships and reproduc- between Eurasian otter Lutra lutra and American mink Mustela tive output of coexisting kestrels and long-eared owls. Oecologia vison probed by niche shift. Oikos 106:19–26. 74:277–285. 38 MCDONALD ET AL.—CAT AND DINGO COMPETITION 1127
Laurenson, M. K. 1994. High juvenile mortality in cheetahs for understanding mulga‐spinifex coexistence. Austral Ecology (Acinonyx jubatus) and its consequences for maternal care. Journal 33:848–862. of Zoology 234:387–408. Newsome, T. M., et al. 2015. Resolving the value of the dingo in eco- Legge, S., et al. 2017. Enumerating a continental-scale threat: logical restoration. Restoration Ecology 23:201–208. how many feral cats are in Australia? Biological Conservation Paltridge, R. 2002. The diets of cats, foxes and dingoes in relation to 206:293–303. prey availability in theTanami Desert, Northern Territory. Wildlife Letnic, M., and C. R. Dickman. 2005. The responses of small mam- Research 29:389–403. Downloaded from https://academic.oup.com/jmammal/article-abstract/99/5/1120/5056485 by Burkitt-Ford Library user on 24 October 2018 mals to patches regenerating after fire and rainfall in the Simpson Pavey, C. R., S. R. Eldridge, and M. Heywood. 2008. Native and intro- Desert, Australia. Austral Ecology 30:24–39. duced predator population dynamics and prey selection during a Linnell, J. D. C., and O. Strand. 2000. Interference interactions, rodent outbreak in arid Australia. Journal of Mammalogy 89:674–683. co‐existence and conservation of mammalian carnivores. Diversity Pearre, S., and R. Maass. 1998. Trends in the prey size-based trophic and Distributions 6:169–176. niches of feral and house cats Felis catus L. Mammal Review MacArthur, R. H., and E. R. Pianka. 1966. On optimal use of a 28:125–39. patchy environment. The American Naturalist 100:603–609. Sidorovich, V. E., H. Kruuk, and D. W. Macdonald. 1999. Body size, Macdonald, D. W., and C. Sillero-Zubiri. 2004. Wild canids—an and interactions between European and American mink (Mustela introduction and dramatis personae. Pp. 3–36 in Biology and con- lutreola and M. vison) in Eastern Europe. Journal of Zoology servation of wild canids (D. W. Macdonald and S. Sillero-Zubiri, 248:521–527. eds.). Oxford University Press, Oxford, United Kingdom. Spencer, E. E., M. S. Crowther, and C. R. Dickman. 2014. Diet and McDonald, P. J., et al. 2017. Habitat as a mediator of mesopredator- prey selectivity o three species of sympatric mammalian predators driven mammal extinction. Conservation Biology 31:1183–1191. in central Australia. Journal of Mammalogy 95:1278–1288. McDonald, P. J., A. D. Griffiths, C. E. M. Nano, C. R. Dickman, Swanson, A., et al. 2014. Cheetahs and wild dogs show contrasting S. J. Ward, and G. W. Luck. 2015. Landscape-scale factors deter- patterns of suppression by lions. The Journal of Animal Ecology mine occupancy of the critically endangered central rock-rat in arid 83:1418–1427. Australia: the utility of camera trapping. Biological Conservation Thackway, R., and I. D. Creswell.1995. An interim bioregion- 191:93–100. alisation for Australia. Australian Nature Conservation Agency, McDonald, P. J., A. Stewart, and C. R. Dickman. 2018. Applying Canberra, Australian Capital Territory, Australia. the niche reduction hypothesis to modelling distributions: a case Triggs, B. E. 1996. Tracks, scats and other traces: a field guide study of a critically endangered rodent. Biological Conservation to Australian mammals. Oxford University Press, Melbourne, 217C:207–212. Victoria, Australia. McDonald, P. J., A. Stewart, A. T. Schubert, C. E. M. Nano, Watts, C. H. S., and H. J. Aslin. 1981. The rodents of Australia. C. R. Dickman, and G. W. Luck. 2016. Fire and grass cover influ- Angus and Robertson, Sydney, New South Wales, Australia. ence occupancy patterns of rare rodents and feral cats in a moun- Wiens, J. A. 1993. The ecology of bird communities: processes tain refuge: implications for management. Wildlife Research and variations. Cambridge University Press, Cambridge, United 43:121–129. Kingdom. McGregor, H., S. Legge, M. E. Jones, and C. N. Johnson. 2015. Woinarski, J. C. Z., A. A. Burbidge, and P. L. Harrison. 2015. Feral cats are better killers in open habitats, revealed by animal- Ongoing unraveling of a continental fauna: decline and extinction borne video. PLoS One 10:e0133915. of Australian mammals since European settlement. Proceedings of Moseby, K. E., H. Neilly, J. L. Read, and H. A. Crisp. 2012. the National Academy of Sciences of the United States of America Interactions between a top order predator and exotic mesopreda- 112:4531–4540. tors in the Australian rangelands. International Journal of Ecology 2012:250352. Submitted 14 December 2017. Accepted 3 July 2018. Nano, C. E. M., and P. J. Clarke. 2008. Variegated desert vegeta- tion: covariation of edaphic and fire variables provides a framework Associate Editor was Chris Pavey.
39 Chapter 5. Applying the niche reduction hypothesis to modelling distributions: A case study of a critically endangered rodent
The critically endangered central rock-rat (Zyzomys pedunculatus) at Counts Point in the MacDonnell Ranges (Photo: P. McDonald)
Author contribution: I was responsible for the chapter concept, fieldwork (assisted by A. Stewart), data analysis, and writing, in consultation with my co-authors.
40 Biological Conservation 217 (2018) 207–212
Contents lists available at ScienceDirect
Biological Conservation
journal homepage: www.elsevier.com/locate/biocon
Applying the niche reduction hypothesis to modelling distributions: A case MARK study of a critically endangered rodent ⁎ Peter J. McDonalda,b, , Alistair Stewartb, Chris R. Dickmana a Desert Ecology Research Group, School of Life and Environmental Sciences, University of Sydney, New South Wales 2006, Australia b Flora and Fauna Division, Department of Environment and Natural Resources, Alice Springs, Northern Territory 0870, Australia
ABSTRACT
The ‘niche reduction hypothesis’ (NRH) is based on the idea that the realized niche of a declining species is reduced by threats that are mediated by environmental, biotic and evolutionary processes. The hypothesis was promoted to identify locations and interventions most likely to benefit declining species. We used a niche re- duction approach to species distribution modelling by predicting the historic and current distributions of a critically endangered Australian rodent, the central rock-rat (CRR). Our habitat suitability maps confirm a dramatic range contraction for this species over the last 100 years. The current association of CRRs with extreme landscape ruggedness supports the hypothesis that the impact of a key threat to the species—cat predation—is mediated by habitat complexity. We detected no CRRs in five new locations predicted to be highly suitable in the current distribution model. This highlights the need for in-situ threat management at the three known sub- populations, one of which may already have been extirpated. Our map of the CRR's historic distribution iden- tifies potential areas for translocation, including the site of a current translocation proposal into a predator-proof fence. We conclude that the NRH provides a useful framework for modelling the change in distributions of declining species in order to prioritise locations and interventions for management.
1. Introduction Hutchinson's seminal work (Hutchinson, 1957), Scheele et al. (2017) advanced these concepts into conservation biology by proposing the Species distribution models (SDMs) relate species occurrences to ‘niche reduction hypothesis’; this is based on the idea that the current environmental data in order to predict distributions (Elith and realized niche of a declining species is reduced from the historic rea- Leathwick, 2009). The use of SDMs has increased dramatically over the lized niche by threats that are mediated by environmental, biotic, last decade, and they are now the primary means of predicting en- geographic and evolutionary processes. The authors argue that by fo- vironmental suitability for species (Guisan et al., 2013). Proponents of cusing on how threats shape the current realized niche, management of SDMs claim they have broad utility to help solve a range of environ- declining species can be improved by identifying where to prioritise mental problems (Guisan and Thuiller, 2005; Elith and Leathwick, conservation actions (Scheele et al., 2017). 2009). However, much of the SDM research has been directed at re- One declining species that will benefit from an improved under- fining and assessing modelling methods (e.g. Bean et al., 2012; Crase standing of changes in the realized niche is the critically endangered et al., 2012) or predicting future climate-driven shifts in distributions central rock-rat (Zyzomys pedunculatus; referred to hereafter as ‘CRR’). (e.g. Kearney et al., 2010; Franklin et al., 2013). Examples of SDMs Historically, CRRs occurred over a vast area of dryland Australia, but applied to conservation management are rare in peer reviewed papers are currently known from only three small sub-populations in the and are mostly restricted to the grey literature (Cayuela et al., 2009; rugged MacDonnell Ranges of the Northern Territory (McDonald et al., Guisan et al., 2013). 2015a). Predation by feral cats (Felis catus; referred to hereafter as Two key concepts in SDM theory are the ‘fundamental niche’– ‘cats’) is regarded as the main threat to the CRR, and rugged terrain is defined as the area of potentially suitable habitat for a species, and the predicted to provide an environmental refuge by reducing the impact of ‘realized niche’–defined as the portion of the fundamental niche oc- feral cat predation (McDonald et al., 2015a; McDonald et al., 2017). cupied by the species due to biotic interactions (Guisan and Thuiller, The largest remaining sub-population of CRRs is currently being man- 2005). Although firmly entrenched in ecological theory following aged through in situ activities to reduce the threat from cats, and is also
⁎ Corresponding author at: c/o Flora and Fauna Division, PO Box 1120, Alice Springs, NT 0871, Australia. E-mail address: [email protected] (P.J. McDonald). https://doi.org/10.1016/j.biocon.2017.10.002 Received 17 July 2017; Received in revised form 21 September 2017; Accepted 2 October 2017 0006-3207/ © 2017 Elsevier Ltd. All rights reserved.
41 P.J. McDonald et al. Biological Conservation 217 (2018) 207–212 the subject of a translocation proposal (Australian Wildlife McDowell, 2010). Predation by feral cats (Felis catus) is regarded as the Conservancy, 2017). main driver of range decline, and experimental feral cat control is Here, we modelled the change in distribution of the CRR based on a currently being trialled at the largest known sub-population. There are framework of the niche reduction hypothesis. While it has been sug- also plans to translocate CRRs into a predator-proof fence on Newhaven gested that only mechanistic models – based on physiological or be- Reserve, in the southern Tanami Desert, Northern Territory, where havioural data – can predict the fundamental niche (Kearney and there are areas of potentially suitable rocky habitat. Our study area of Porter, 2004), correlative SDMs may also predict the fundamental niche ~365,000 km2 covers the historic known distribution of the CRR in the when the environmental data in the models correspond to the under- drylands of the Northern Territory (Fig. 1). lying processes constraining distribution (Dormann, 2007). Using measures of landscape-scale ruggedness and other environmental 2.2. Distribution modelling variables, we modelled the CRR's current distribution to identify po- fi tentially suitable areas and we surveyed ve locations predicted to be We used the software package MaxEnt to estimate the historic and highly suitable. We also modelled the CRR's historic distribution to current distribution of the CRR. MaxEnt is a presence-only model that identify suitable sites for translocation and we compared the two minimises the relative entropy of estimated probability densities be- models to gain insights into the processes driving the dramatic range tween species presences and the background landscape (Elith et al., contraction in this species. 2011). MaxEnt frequently outperforms other modelling techniques and is typically robust to small samples sizes (Hernandez et al., 2006; Wisz 2. Methods et al., 2008). For the historic distribution, we compiled the 32 CRR records spanning 2.1. Focal species and study area the period from firstcollectionin1894toitsrecenttemporarydis- appearance in 2002 (Nano, 2008). The species has not been detected at The CRR is a rock-dwelling, granivorous rodent endemic to the any of these locations since 2002, despite considerable search effort Northern Territory, Australia, listed as critically endangered by the (McDonald et al., 2013). For the current distribution we compiled the 36 IUCN (Woinarski and Burbidge, 2016). Once widespread in rocky records made from 2010 to 2016. We selected six environmental raster ranges and mountainous areas across the Northern Territory's drylands, layers predicted to influence CRR distribution and represent the environ- the species is known currently from three sub-populations in the rugged mental variables restricting the fundamental niche. These included two MacDonnell Ranges (Fig. 1). There are also sub-fossil records from ruggedness layers based on the prediction that ruggedness and habitat Western Australia's drylands, although it is unknown whether the CRR complexity mediates predation by feral cats and red foxes: ruggedness_- was extant there at the time of European colonisation (Baynes and coarse, the standard deviation of mean elevation within a 1 km radius of
Fig. 1. Location of the study area in dryland Australia. Red triangles are historic (1884–2002) records and blue triangles are current (2010–2016) records for the central rock-rat. Background 9-second digital elevation model courtesy of Geoscience Australia. (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.)
208 42 P.J. McDonald et al. Biological Conservation 217 (2018) 207–212 eachpixelina3s(~90m)digitalelevationmodel(availablehttp:// Table 1 www.ga.gov.au/metadata-gateway/metadata/record/gcat_72760); and Survey locations and effort based on the current distribution model for the central rock- ruggedness_fine, terrain ruggedness index (TRI) calculated from 3 s rat.
(~90 m) DEM. The TRI is calculated as the sum change in elevation be- Locationa Geology No. Trap nights per Central rock- tween a raster cell and its eight neighbour cells (Riley, 1999). The four camera camera (total rat detections remaining layers were: soils, Digital Atlas of Australia Soils at 1:2,000,000 traps deployment) scale (available http://www.asris.csiro.au/themes/Atlas.html); vegeta- i - Mt. Zeil Granite 7 57 (399) nil tion, broad vegetation communities mapped from Landsat imagery at ii - Mt. Razorback Quartzite 5 93 (465) nil 1:1,000,000 across the Northern Territory; productivity, median of iii - Mt. Chapple Granite 6 57 (342) nil average greenness 2000–2007 (MODIS 250 m resolution using 16-d nor- iv - Eastern Quartzite 9 30 (270) nil malised vegetation index mean) (available http://data.daff.gov.au/anrdl/ Chewings Range php/anrdlSearch.html); and temperature, mean monthly maximum tem- v - Georgina Quartzite 12 46 (552) nil perature for January (hottest month of the year) at 0.05 degree resolution Range for the period 1980–1999 (available http://data.daff.gov.au/anrdl/php/ anrdlSearch.html). Soil and vegetation were classified as categorical a See Fig. 3. variables and the other layers as continuous, and all were standardised to 90 m resolution. (contribution = 72.9%), followed by fine-scale ruggedness (contribu- Prior to running models we screened continuous variables for col- tion = 17.3%), temperature (contribution = 5.7%), and productivity linearity using Pearson's correlation coefficient; no variable pairs had a (contribution = 4.1%) (Fig. 3). Predicted suitability for the CRR was correlation coefficient > 0.7 so we retained all for initial modelling. 0.5 (i.e. mid-level suitability) with ~25 m of coarse-scale ruggedness We ran historic and current distribution models in MaxEnt v.3.4.1 using (SD of mean elevation within 1 km radius) and a fine-scale ruggedness the default regularization, background points, auto features and model index of ~100 (Fig. 3). The mean AUC for the training data was iterations. To account for variation in small mammal sampling across 0.953 ± 0.024 and the AUC for the test data ranged from the study area we applied a bias layer to the historic and current dis- 0.788 ± 0.093 to 0.983 ± 0.009. The historic distribution model tribution models (Phillips et al., 2009); comprising raster layers of the indicated that there is potentially suitable habitat at a proposed re- sum of small mammal records (NT Fauna Atlas, available https://nt. introduction site in the southern Tanami Desert; however, an area of gov.au/environment/environment-data-maps/fauna-atlas) collected in rugged hills to the north-east may be more suitable (Fig. 2). each period within each bioregion of the study area (Thackway and The current distribution model supports the assumption that the Cresswell, 1995). We evaluated model performance by withholding CRR has undergone a dramatic range contraction over the last 25% of occurrence records for testing using the bootstrap technique 100 years (Fig. 3). Fine-scale ruggedness was the most important vari- with 10 replicates. We reported mean area under the ROC curve (AUC) able (contribution = 65.5%), followed by coarse-scale ruggedness for the training and test data from the 10 replicates. Models with AUC (contribution = 31.1%), temperature (contribution = 3.3%), and pro- values > 0.75 are considered potentially useful (Elith, 2000). Our final ductivity (contribution = 0%). Predicted suitability for the CRR was predictive maps were based on mean values of suitability in logistic 0.5 (i.e. mid-level suitability) with ~65 m of coarse-scale ruggedness form which we interpreted as relative habitat suitability (Phillips and (SD of mean elevation within 1 km radius) and a fine-scale ruggedness Dudík, 2008). We reported the percentage contribution of each vari- index of ~350 (Fig. 3). The mean AUC for the training data was able; calculated in each iteration of the training algorithm as the in- 0.998 ± < 0.001 and the AUC for the test data ranged from crease in regularized gain added to, or subtracted from (in the case of a 0.998 ± 0.001 to 0.999 ± 0.001. negative lambda value), the contribution of the corresponding variable, We undertook 2028 camera trap nights of survey effort across four and averaged across the 10 replicate runs. The historic and current locations predicted to be highly suitable for CRRs based on the current distribution model with all variables indicated that low-lying areas of distribution model (Table 1). We detected no central rock-rats at any of salt lake were highly suitable, however these topographically feature- these locations but recorded several other native and introduced less landforms are not known to be inhabited by any mammal species. mammal species: Tachyglossus aculeatus, Pseudantechinus macdonnel- This predicted associated was driven by the soil and vegetation layers lensis, Petrogale lateralis, Pseudomys desertor, Mus musculus, and Felis which otherwise contributed little to the models (< 5% contribution), catus. so we omitted them prior to running the final model for each time period. 4. Discussion
2.3. Targeted survey We used a niche reduction hypothesis framework to consider the historic and current distributions of the critically endangered CRR. The We selected four locations predicted to be highly suitable for the predicted output maps from these models confirm a dramatic range CRR from the current distribution map. At each site we deployed contraction for this species over the last 100 years, and the current camera traps (Reconyx Hyperfire HC500 or HC550) orientated verti- association of the CRR with extreme ruggedness supports the hypoth- cally and baited with peanut butter and oats (McDonald et al., 2015a). esis that habitat complexity is needed to mediate the impacts of a key Camera traps were spaced a minimum of 250 m apart in the most sui- threat to the species: cat predation (McDonald et al., 2016). Our results, table locations – with recent fire history (burnt within 10 years) and/or demonstrating a shift over time in the relationship between habitat high rock complexity (McDonald et al., 2016). We deployed camera suitability and ruggedness characteristics, also support the idea that traps for periods of 46–93 consecutive nights in 2015–16 which should correlative SDMs can predict the realized niche in situations where have yielded a > 95% detection probability for CRRs (Table 1; environmental variables correspond to the underlying processes driving McDonald et al., 2015a, 2016). suitability (Dormann, 2007). We suggest that the application of the niche reduction hypothesis framework to SDMs can help to improve the 3. Results management of declining species. Specifically, alternative SDM ap- proaches that consider only historic, recent or combined species oc- The historic distribution model suggested that CRRs once occurred currence data may under or overestimate distributions and are less on any rocky areas with some level of ruggedness throughout the study likely to uncover the processes driving reductions in the realized niche. area (Fig. 2). Coarse-scale ruggedness was the most important variable While our two ruggedness variables were dominant in both
209 43 P.J. McDonald et al. Biological Conservation 217 (2018) 207–212
Fig. 2. Historic predicted distribution – predicted suitability for the central rock-rat from the historic occurrence data (1884–2002) in the study area in dryland Australia. Inset: proposed translocation site is at Newhaven Wildlife Sanctuary in the southern Tanami Desert. Current predicted distribution – suitability for the central rock-rat from the recent occurrence data (2010–2016) in the study area. White-outlined polygons are recently extant locations (west to east; Mt. Edward, Mt. Sonder, western Chewings and Heavitree Ranges). Black-outlined polygons are new locations surveyed in this study (i - Mt. Zeil, ii - Mt. Razorback, iii - Mt. Chapple, iv - eastern Chewings Range, v - Georgina Range).
distribution models, we found a change in the relationship between contraction. The concept of environmental conditions mediating threat suitability and ruggedness over time. Specifically, areas were predicted severity is a fundamental process identified in the niche reduction hy- to be more suitable at lower levels of coarse- and fine-scale ruggedness pothesis, and provides one mechanism to explain how declining species in the historic model. This was evident in the suitability maps – the can persist in a subset of their historic distributions (Rieman et al., historic model predicted that CRRs occurred widely on hilly terrain 2006; Scheele et al., 2017). McDonald et al. (2017) proposed that throughout the study area, while the current model predicted that CRRs rugged areas of dryland Australia offer refuge for the CRR and other are now confined to areas of extreme ruggedness. We also found a shift rare mammals by mediating predation from cats. This hypothesis is in the importance of ruggedness type – coarse-scale ruggedness domi- supported by our data here and also by studies showing that cat hunting nated the historic model while fine-scale ruggedness dominated the success, density and occupancy are lower in complex terrain than in current model. From these results we infer that topographic complexity topographically simple habitats (Hohnen et al., 2016; McGregor et al., has mediated the threat(s) responsible for the CRR's dramatic range 2015; Legge et al., 2017).
210 44 P.J. McDonald et al. Biological Conservation 217 (2018) 207–212
Fig. 3. Upper panels show predicted suitability (logistic output) for the central rock-rat in rela- tion to coarse-scale ruggedness from (a) the his- toric and (b) the current MaxEnt models, and predicted suitability in relation to fine-scale ruggedness from the historic (c) and the current MaxEnt models (d). Lower panel (e) shows the contribution of each environmental variable to the historic and current MaxEnt models.
Our historic distribution model identified potentially suitable re- protecting declining Australian mammals from threats and, in the ab- introduction sites for the CRR throughout the study area. However, sence of cats and foxes, native mammals invariably thrive (Hayward reintroductions to sites with low levels of fine-scale ruggedness within and Kerley, 2009; Dickman, 2012; Woinarski et al., 2015). A current the historic niche are likely to fail unless the threat of predation by cats proposal to translocate CRRs into a ~ 9000 ha predator-free fenced is mitigated (Moseby et al., 2011). There are currently two methods area in the southern Tanami Desert (Australian Wildlife Conservancy, available for mitigating the impact of feral cats on a landscape scale: 2017) was supported by our historic distribution model – a small area of aerially-distributed poison baiting, and predator-proof fences. Else- potentially suitable habitat was predicted within the fenced area. We where in dryland Australia, poison baiting has resulted in substantial also found areas of higher predicted suitability near the proposed fence reductions in feral cat abundances, but its effectiveness was strongly and these could be ‘seeded’ by dispersing animals if threats can also be influenced by the availability of small mammal prey and it is not yet mitigated outside of the fenced area (Dickman, 2012). In addition to clear whether baiting translates to conservation benefits for native providing an insurance population for the CRR, the proposed translo- mammals (Burrows et al., 2003; Christensen et al., 2013). Poison cation would provide a useful experiment for assessing the role of baiting trials are currently underway at the largest remaining CRR sub- predation-protection versus other factors (e.g. climate; McDonald et al., population in Tjoritja National Park, and the results of these will be 2015b) in CRR persistence, and for assessing the potential of our his- used to evaluate the efficacy of baiting for reducing the impacts of feral toric distribution model for predicting the historic realized niche. cats in CRR habitat. Predator-proof fences are an established method for We surveyed four locations predicted to be highly suitable for CRRs
211 45 P.J. McDonald et al. Biological Conservation 217 (2018) 207–212 from the current distribution model and failed to detect CRRs at any of climate projections need to be? Glob. Chang. Biol. 19, 473–483. these locations. We used best-practice detection methods that should Guisan, A., Thuiller, W., 2005. Predicting species distribution: offering more than simple – fi habitat models. Ecol. Lett. 8, 993 1009. have detected CRRs with > 95% con dence (McDonald et al., 2015a, Guisan, A., Tingley, R., Baumgartner, J.B., Naujokaitis-Lewis, I., Sutcliffe, P.R., Tulloch, 2016); therefore it is likely that the species was absent rather than not A.I., Regan, T.J., Brotons, L., McDonald-Madden, E., Mantyka-Pringle, C., Martin, detected. The absence of CRRs in locations of high predicted suitability T.G., 2013. Predicting species distributions for conservation decisions. Ecol. Lett. 16, 1424–1435. may be explained by environmental variation not accounted for in our Hayward, M.W., Kerley, G.I., 2009. Fencing for conservation: Restriction of evolutionary spatial layers. For example, even our fine-scale ruggedness layer (TRI potential or a riposte to threatening processes? Biol. Conserv. 142, 1–13. index at a 90 m scale) was relatively coarse, however no spatial layers Hernandez, P.A., Graham, C.H., Master, L.L., Albert, D.L., 2006. The effect of sample size ff are available for measuring ruggedness at a scale likely to correspond to and species characteristics on performance of di erent species distribution modeling methods. Ecography 29, 773–785. predation risk for individual CRRs (e.g. < 1 m scale). Therefore, de- Hohnen, R., Tuft, K., McGregor, H.W., Legge, S., Radford, I.J., Johnson, C.N., 2016. veloping fine-scale elevation data (e.g. using Light Detection and Ran- Occupancy of the invasive feral cat varies with habitat complexity. PLoS One 11, ging; Reutebuch et al., 2003) could improve the predictive accuracy of e0152520. Hutchinson, G.E., 1957. The multivariate niche. Cold Spring Harb. Symp. Quant. Biol. 22, a current CRR distribution model. As well as the potential limitations of 415–421. environmental variables in our distribution models, we note that most Kearney, M., Porter, W.P., 2004. Mapping the fundamental niche: physiology, climate, – of our survey locations were relatively small and isolated from extant and the distribution of a nocturnal lizard. Ecology 85, 3119 3131. Kearney, M.R., Wintle, B.A., Porter, W.P., 2010. Correlative and mechanistic models of sub-populations by a matrix of less rugged landforms. Small and iso- species distribution provide congruent forecasts under climate change. Conserv. Lett. lated sub-populations are most prone to extinction, particularly when 3, 203–213. the intervening matrix poses a barrier to dispersal (O'Brien et al., 2008; Legge, S., Murphy, B.P., McGregor, H., Woinarski, J.C.Z., Augusteyn, J., Ballard, G., Baseler, M., Buckmaster, T., Dickman, C.R., Doherty, T., Edwards, G., Eyre, T., Prugh et al., 2008). In support of this idea, the smallest of the recently Fancourt, B.A., Ferguson, D., Forsyth, D.M., Geary, W.L., Gentle, M., Gillespie, G., extant sub-populations (Mt. Sonder; Fig. 2) may have been extirpated in Greenwood, L., Hohnen, R., Hume, S., Johnson, C.N., Maxwell, M., McDonald, P.J., 2012 (McDonald et al., 2015a). Irrespective of the factors driving the Morris, K., Moseby, K., Newsome, T., Nimmo, D., Paltridge, R., Ramsey, D., Read, J., Rendall, A., Rich, M., Ritchie, E., Rowland, J., Short, J., Stokeld, D., Sutherland, D.R., current distribution of the CRR, our surveys suggest that our current Wayne, A.F., Woodford, L., Zewe, F., 2017. Enumerating a continental-scale threat: model has overestimated the area of suitable habitat and indicate that How many feral cats are in Australia? Biol. Conserv. 206, 293–303. in-situ management of all known CRR sub-populations may be neces- McDonald, P.J., Pavey, C.R., Knights, K., Grantham, D., Ward, S.J., Nano, C.E., 2013. sary to avert extinction. Extant population of the Critically Endangered central rock-rat Zyzomys pedunculatus located in the Northern Territory, Australia. Oryx 47, 303–306. McDonald, P.J., Griffiths, A.D., Nano, C.E., Dickman, C.R., Ward, S.J., Luck, G.W., 2015a. Acknowledgements Landscape-scale factors determine occupancy of the critically endangered central rock-rat in arid Australia: the utility of camera trapping. Biol. Conserv. 191, 93–100. McDonald, P.J., Luck, G.W., Dickman, C.R., Ward, S.J., Crowther, M.S., 2015b. Using We thank the traditional owners of Tjoritja NP and the Hayes and multiple-source occurrence data to identify patterns and drivers of decline in arid Connellan families for allowing access to the survey locations. Edward dwelling Australian marsupials. Ecography 38, 1090–1100. Connellan from Mengel's Heli Services flew us into the survey locations. McDonald, P.J., Stewart, A., Schubert, A.T., Nano, C.E., Dickman, C.R., Luck, G.W., 2016. Fire and grass cover influence occupancy patterns of rare rodents and feral cats in a Park rangers from the NT Parks and Wildlife Commission assisted on mountain refuge: implications for management. Wildl. Res. 43, 121–129. some surveys. Fieldwork was carried out with ethics approval from McDonald, P.J., Nano, C.E., Ward, S.J., Stewart, A., Pavey, C.R., Luck, G.W., Dickman, Charles Darwin University (approval no. A06028) and a scientific C.R., 2017. Habitat as a mediator of mesopredator-driven mammal extinction. Conserv. Biol. 31, 1183–1191. permit from the NT Parks and Wildlife Commission (permit no. 44636). McGregor, H., Legge, S., Jones, M.E., Johnson, C.N., 2015. Feral cats are better killers in open habitats, revealed by animal-borne video. PLoS One 10, e0133915. References Moseby, K.E., Read, J.L., Paton, D.C., Copley, P., Hill, B.M., Crisp, H.A., 2011. Predation determines the outcome of 10 reintroduction attempts in arid South Australia. Biol. Conserv. 144, 2863–2872. Australian Wildlife Conservancy, 2017. Newhaven Endangered Wildlife Restoration Nano, T., 2008. Central rock-rat Zyzomys pedunculatus. In: Van Dyck, S., Stahan, R. (Eds.), Project. Available from. http://www.australianwildlife.org/media/275206/2017- The mammals of Australia. Reed New Holland, pp. 658–660. newhaven-endangered-wildlife-restoration-project-mr-21-4-17.pdf, Accessed date: O'Brien, C.M., Crowther, M.S., Dickman, C.R., Keating, J., 2008. Metapopulation dy- 23 June 2017. namics and threatened species management: why does the broad-toothed rat Baynes, A., McDowell, M.C., 2010. The original mammal fauna of the Pilbara biogeo- (Mastacomys fuscus) persist? Biol. Conserv. 141, 1962–1971. graphic region of north-western Australia. In: Records of the Western Australian Phillips, S.J., Dudík, M., 2008. Modeling of species distributions with Maxent: new ex- Museum. 78. pp. 285–298. tensions and a comprehensive evaluation. Ecography 31, 161–175. Bean, W.T., Stafford, R., Brashares, J.S., 2012. The effects of small sample size and sample Phillips, S.J., Dudík, M., Elith, J., Graham, C.H., Lehmann, A., Leathwick, J., Ferrier, S., bias on threshold selection and accuracy assessment of species distribution models. 2009. Sample selection bias and presence-only distribution models: implications for Ecography 35, 250–258. background and pseudo-absence data. Ecol. Appl. 19, 181–197. Burrows, N.D., Algar, D., Robinson, A.D., Sinagra, J., Ward, B., Liddelow, G., 2003. Prugh, L.R., Hodges, K.E., Sinclair, A.R., Brashares, J.S., 2008. Effect of habitat area and Controlling introduced predators in the Gibson Desert of Western Australia. J. Arid isolation on fragmented animal populations. Proc. Natl. Acad. Sci. U. S. A. 105, Environ. 55, 691–713. 20770–20775. Cayuela, L., Golicher, D.J., Newton, A.C., Kolb, M., de Alburquerque, F.S., Arets, Reutebuch, S.E., McGaughey, R.J., Andersen, H.E., Carson, W.W., 2003. Accuracy of a E.J.M.M., Alkemade, J.R.M., Pérez, A.M., 2009. Species distribution modeling in the high-resolution lidar terrain model under a conifer forest canopy. Can. J. Remote. tropics: problems, potentialities, and the role of biological data for effective species Sens. 29, 527–535. conservation. Trop. Conserv. Sci. 2, 319–352. Rieman, B.E., Peterson, J.T., Myers, D.L., 2006. Have brook trout (Salvelinus fontinalis) Christensen, P.E., Ward, B.G., Sims, C., 2013. Predicting bait uptake by feral cats, Felis displaced bull trout (Salvelinus confluentus) along longitudinal gradients in central catus, in semi-arid environments. Ecol. Manag. Restor. 14, 47–53. Idaho streams? Can. J. Fish. Aquat. Sci. 63, 63–78. Crase, B., Liedloff, A.C., Wintle, B.A., 2012. A new method for dealing with residual Riley, S.J., 1999. Index that quantifies topographic heterogeneity. Int. J. Therm. Sci. 5, spatial autocorrelation in species distribution models. Ecography 35, 879–888. 23–27. Dickman, C.R., 2012. Fences or ferals? Benefits and costs of conservation fencing in Scheele, B.C., Foster, C.N., Banks, S.C., Lindenmayer, D.B., 2017. Niche contractions in Australia. In: Somers, M.J., Hayward, M. (Eds.), Fencing for Conservation. Springer, declining species: mechanisms and consequences. Trends Ecol. Evol. 32, 346–355. New York, pp. 43–63. Thackway, R., Cresswell, I.D., 1995. An Interim Biogeographic Regionalisation for Dormann, C.F., 2007. Promising the future? Global change projections of species dis- Australia: A Framework for Establishing the National System of Reserves, Version tributions. J. Basic Appl. Ecol. 8, 387–397. 4.0. Australian Nature Conservation Agency. Elith, J., 2000. Quantitative methods for modeling species habitat: comparative perfor- Wisz, M.S., Hijmans, R.J., Li, J., Peterson, A.T., Graham, C.H., Guisan, A., 2008. Effects of mance and an application to Australian plants. In: Ferson, S., Burgman, M. (Eds.), sample size on the performance of species distribution models. Divers. Distrib. 14, Quantitative Methods for Conservation Biology. Springer, New York, pp. 39–58. 763–773. Elith, J., Leathwick, J.R., 2009. Species distribution models: ecological explanation and Woinarski, J., Burbidge, A.A., 2016. Zyzomys pedunculatus. The IUCN Red List of prediction across space and time. Annu. Rev. Ecol. Evol. Syst. 40, 677–697. Threatened Species 2016, e.T23324A22456932. Elith, J., Phillips, S.J., Hastie, T., Dudík, M., Chee, Y.E., Yates, C.J., 2011. A statistical Woinarski, J.C.Z., Burbidge, A.A., Harrison, P.L., 2015. Ongoing unraveling of a con- explanation of MaxEnt for ecologists. Divers. Distrib. 17, 43–57. tinental fauna: decline and extinction of Australian mammals since European set- Franklin, J., Davis, F.W., Ikegami, M., Syphard, A.D., Flint, L.E., Flint, A.L., Hannah, L., tlement. Proc. Natl. Acad. Sci. U. S. A. 112, 4531–4540. 2013. Modeling plant species distributions under future climates: how fine scale do
212 46 Chapter 6. Landscape-scale factors determine occupancy of the critically endangered central rock-rat in arid Australia: The utility of camera trapping
Author contribution: I was responsible for the chapter concept, fieldwork (assisted by others), data analysis, and writing, in consultation with my co-authors. Supporting information is provided for this chapter is Appendix 6.
47 Biological Conservation 191 (2015) 93–100
Contents lists available at ScienceDirect
Biological Conservation
journal homepage: www.elsevier.com/locate/bioc
Landscape-scale factors determine occupancy of the critically endangered central rock-rat in arid Australia: The utility of camera trapping
Peter J. McDonald a,b,⁎,AnthonyD.Griffiths c, Catherine E.M. Nano b, Chris R. Dickman a, Simon J. Ward b, Gary W. Luck d a Desert Ecology Research Group, School of Biological Sciences, University of Sydney, New South Wales 2006, Australia b Flora and Fauna Division, Department of Land Resource Management, Alice Springs, Northern Territory 0870, Australia c Flora and Fauna Division, Department of Land Resource Management, Palmerston, Northern Territory 0831, Australia d Institute for Land, Water and Society, Charles Sturt University, Albury, NSW 2640, Australia article info abstract
Article history: The challenges of sampling rare fauna limit efforts to understand and mitigate the factors that restrict their dis- Received 4 March 2015 tribution. Camera traps have become a standard technique for sampling large mammals, but their utility for sam- Received in revised form 11 June 2015 pling small, rare species remains largely unknown. The central rock-rat (Zyzomys pedunculatus) is critically Accepted 15 June 2015 endangered and restricted to rugged range country in central Australia. Using Z. pedunculatus as a focal species, Available online xxxx we sought to evaluate the effectiveness of camera trapping for sampling small mammals in this environment and to better understand the factors driving the occurrence of this species. We installed baited camera traps at Keywords: Rodent 50 sites across 1795 ha of core refuge habitat for Z. pedunculatus. We recorded all six species of small mammals Arid known previously from this area, including the highly detectable Z. pedunculatus at five sites. Occupancy model- Camera-trapping ling showed that distance to the nearest occupied site was the most important predictor of Z. pedunculatus occur- Occupancy rence, suggesting that this rodent occurs in discrete sub-populations within the matrix of refuge habitat. Fire history and ruggedness may also influence occupancy of Z. pedunculatus at the landscape-scale and could assist in locating additional sub-populations. At the site-scale, occupancy of Z. pedunculatus was high and there was no clear influence of any site-scale variables. Management of Z. pedunculatus will require protection and expan- sion of known sub-populations. We conclude that camera trapping provided useful and cost-effective insight into the factors limiting rock-rat distribution, and predict that it will become a standard tool for sampling rare small mammals in difficult-to-access environments. © 2015 Elsevier B.V. All rights reserved.
1. Introduction monitoring of a location for weeks or months, without the presence of a human observer (Vine et al., 2009). In addition, disturbance and associ- The challenges of sampling rare or geographically restricted fauna ated stress to wildlife is likely to be minimal as cameras remove the are a major limitation to better understand the factors that limit their need to detain and handle wildlife, as is required when live-trapping occurrence (Dixon et al., 1998; Gu and Swihart, 2004). This is a particu- (De Bondi et al., 2010). Further, for uniquely patterned fauna, images lar problem for threatened species, where lack of information can result from camera traps allow for the use of capture–recapture analysis in inefficient allocation of resources and hamper efforts to identify and (Karanth, 1995; Soisalo and Cavalcanti, 2006), while developments in mitigate threatening processes (McDonald-Madden et al., 2011). occupancy modelling mean that robust estimates of site occupancy Given that funding for conservation programmes is limited globally can be calculated with detection/non-detection data derived from (James et al., 1999), sampling threatened species relies on the develop- camera images (Linkie et al., 2007; Sollmann et al., 2013). ment and implementation of cost-effective methods that balance effi- Although predominantly used for medium-to-large mammals, cam- ciency with accuracy and precision (Silveira et al., 2003). era traps are showing promise for sampling small mammals. By moving One important development in recent decades is the use of camera from the typical horizontal mount to a vertical position (camera facing traps for sampling wildlife. Depending on the objectives, camera traps the ground), maintaining a standard height from the ground, and may offer advantages over traditional sampling methods, such as live- using an attractant bait, numerous small mammal species can be readily trapping and direct observation. For example, they allow for continuous detected and reliably identified to species level (De Bondi et al., 2010; Rendall et al., 2014). To date, though, there have been very few direct tests of the efficacy of camera trapping for small mammal detection, ⁎ Corresponding author at: Flora and Fauna Division, PO Box 1120, Alice Springs, NT 0871, Australia. and it remains unclear how readily the method can be applied across E-mail address: [email protected] (P.J. McDonald). different habitats and land-use types. A recent study by De Bondi et al.
http://dx.doi.org/10.1016/j.biocon.2015.06.027 0006-3207/© 2015 Elsevier B.V. All rights reserved.
48 94 P.J. McDonald et al. / Biological Conservation 191 (2015) 93–100
(2010) showed that, for an area in temperate southern Australia, cam- quartzite ridges and mountain peaks (McDonald et al., 2013). However, era trapping was a more cost-effective method for small mammal detec- attempts to further refine population and habitat parameters are at tion than live-trapping. These authors concluded that camera traps had present severely constrained because standard live-trapping methods the potential to increase the replication and spatial coverage of small- are too labour- and resource-intensive to generate sufficient informa- mammal sampling programmes. Similarly, Rendall et al. (2014) found tion within the context of current budgetary constraints (McDonald camera trapping to be an efficient method of detecting invasive rodents et al., 2013). As a result, the factors limiting the distribution and size in an area of high conservation value. They reported advantages of cam- of Z. pedunculatus populations remain unknown. eras for detecting trap-averse species and in limiting disturbance to tar- Recent trials using camera traps have provided an indication that get and non-target species. Importantly though, the efficacy of camera cameras may have high utility for the study of this and other sparsely- trapping for small-mammal surveys has not been tested in challenging occurring small mammals in arid settings (McDonald et al., 2015a). environments such as arid systems where rare small-mammals are These trials have, however, been opportunistic and small-scale, and often very sparsely distributed, or in situations where the terrain im- more rigorous assessment of the technique is required. Here, we assess poses logistic challenges. the utility of camera traps for detecting and understanding the factors The central rock-rat (Zyzomys pedunculatus) exemplifies the chal- that limit occupancy of Z. pedunculatus in the West MacDonnell National lenges associated with sampling rare species in difficult-to-access envi- Park in the Northern Territory, Australia. Representing the core habitat ronments. One of Australia's rarest mammals, this endemic and of Z. pedunculatus and some of the most rugged and remote terrain in critically endangered (IUCN red list; Woinarski and Morris, 2008) ro- Australia, this park is an ideal location to evaluate the potential utility dent has contracted in extent of occurrence by N90% in the last of camera traps where access is difficult and there are size and weight 100 years and is currently known from a limited area of mountainous restrictions on equipment. Focusing on core Z. pedunculatus refuge hab- terrain west of Alice Springs in the Northern Territory (McDonald itat, and using Z. pedunculatus as our focal species, we aimed to: 1) eval- et al., 2013, 2015a). The central rock-rat exhibits classic ‘boom-bust’ uate the potential of camera traps for detecting Z. pedunculatus and population dynamics that are directly underpinned by extreme fluctua- other small mammals, 2) better understand the factors influencing oc- tions in primary productivity in time (Edwards, 2013a; Letnic and cupancy of Z. pedunculatus, and 3) provide management direction for Dickman, 2010). Following well above-average rainfall events, this the conservation of Z. pedunculatus. species, as with many desert rodents, may undergo population irrup- tions and become locally abundant in a variety of rocky situations 2. Methods (e.g. see Edwards, 2013a). However, with the inevitable return to more typical dry and low-resource conditions, this species contracts to 2.1. Study area core-habitat patches (refuges) during its non-irruptive population phases (McDonald et al., 2013; Pavey et al., 2014). Our study was located in the 2592 km2 West MacDonnell National Recently, significant progress has been made in the circumscription Park (NP) in the MacDonnell Ranges bioregion of the Northern Territory of core-habitat variables for the dry-time refuges of Z. pedunculatus, (NT), Australia (Fig. 1). This iconic national park extends 160 km west allowing for increased predictability of occupancy area and population from Alice Springs and includes the most rugged and high elevation size (McDonald et al., 2013). Specifically, it is now known that central (to 1500 m) uplands in arid Australia. Climate is semi-arid with highly rock-rat refuges are effectively confined to high-elevation (N1000 m) variable rainfall (mean annual rainfall at Alice Springs Airport =
Fig. 1. Maps showing: a) the study area in the Northern Territory, Australia; b) the locations of the 50 camera trap sites on the higher elevation (N950 m) quartzite landforms of the West MacDonnell NP (sites not to scale); and c) the locations of the 20 camera traps in the intensively sampled area. White and black squares represent sites where Zyzomys pedunculatus was recorded to be present or absent, respectively. Background digital elevation model and satellite imagery courtesy of Geoscience Australia.
49 P.J. McDonald et al. / Biological Conservation 191 (2015) 93–100 95
284 mm) and temperatures ranging from hot in summer (daytime max- vegetation and installed a plastic jar with 12 holes punched into it and ima frequently N 40 °C) to cool in winter (overnight minima frequently fixed to the ground with tie wire and a steel peg. Jars were baited b0 °C) (Australian Bureau of Meteorology Climate Data). Landforms are with cotton wool soaked in peanut butter and oats, an established bait diverse and include the rugged quartzite mountain ranges that charac- for Z. pedunculatus and other small mammals in the region (Edwards, terise the region. Vegetation on these ranges typically includes a 2013a). Cameras were set to run 24 h per day and take three consecu- groundcover layer dominated by hummock grasses (Triodia spp.) or tive images with a 1 min rest period. The ‘high’ sensitivity setting was tussock grasses (e.g. Eriachne mucronata, Neurachne tenuifolia), with applied to all cameras, and two layers of masking tape were added to forbs (e.g. Goodenia ramelli, Solanum quadriloculatum), sub-shrubs the Bushnell cameras to reduce flash brightness for close-range opera- (e.g. Gastrolobium brevipes, Ptilotus sessilifolius, Pomax rupestris), and a tion. Cameras were installed over several trips between 19 May and sparse overstory of heath-like scattered low shrubs and mallees (e.g. 20 June 2013 and remained in the field until 22 August 2013. Small Acacia macdonnellensis, Hakea grammatophylla, Prostanthera schultzii). mammal species were identified in images based on size (relative to the bait device) and tail shape (see S1). Detection/non-detection data 2.2. Site selection for each species were assigned initially on a per-night basis and then collapsed into 8-night periods (ranging from 8 to 12, 8-night periods Z. pedunculatus is currently known only from higher elevation per camera) to provide a better fit for occupancy analysis and a detec- (N1000 m) quartzite mountains and ridges in the western parts of the tion period more appropriate to the length of time the cameras West MacDonnell NP and on a quartzite range outlier 70 km west of remained in the field. the park boundary, on Haast's Bluff Aboriginal Land Trust (McDonald Based on the results of the landscape-scale camera deployment, we et al., 2013, 2015a). Using ArcGIS 10.2, we created an 80 × 80 m grid undertook an intensive follow-up camera trap survey of one area where across the West MacDonnell NP, between Ellery Creek in the east and Z. pedunculatus was located. We did this to test whether the area sup- Mt Razorback in the west. This area covers all recent records (post- ported a resident sub-population (rather than isolated or nomadic oc- 2002) of the species in the park and includes the majority of high eleva- currences), and to explore the factors influencing occupancy at a finer tion (N1000 m) quartzite habitat in the NT. Although home range data scale. From 3 September 2014 to 18 December 2014 we deployed a sin- are not available for Z. pedunculatus, we predicted that 80 × 80 m gle camera each at 20 sites, randomly selected within a 3 km2 area would be an appropriate grain size based on relatively high detectability where Z. pedunculatus had been recorded (Fig. 1). Because we set no and recapture rates at this scale using live-trapping (McDonald et al., minimum distance between sites (actual minimum distance = 2013). Using a 1-s resolution digital elevation model (Geoscience 100 m), our interpretations of occupancy are considered accordingly Australia; http://nedf.ga.gov.au/geoportal/catalog/main/home.page), (see below). we clipped the grid to N 950 m elevation. After removing partial cells At each site we measured a series of habitat variables considered on the edge of the grid, we randomly selected 50 cells, each a minimum likely to influence the presence of Z. pedunculatus. These habitat vari- distance of 400 m from the nearest other cell. ables were subsequently converted into derived site-scale variables for inclusion in occupancy modelling analyses (Table 2). 2.3. Camera trapping 2.4. Occupancy modelling We constructed lightweight camera posts using pairs of 900 mm al- uminium fence spacers (Fig. 2). The second spacer improves rigidity and We examined the influence of covariates on detection and occupan- reduces wind-induced camera shake and the incidence of false images. cy of Z. pedunculatus at the landscape- and site-scale, using single- We used 33 Reconyx Hyperfire cameras (Model # H500) and 17 season occupancy models in PRESENCE v6.9 (Hines, 2006). This analysis Bushnell Trophy cameras (Model # 119456). For the Reconyx cameras uses maximum likelihood to estimate site occupancy when the study we bolted an aluminium bracket through one end of the fence spacers species may be imperfectly detected (i.e. detection probability b1) and fixed the camera in the vertical position with a short bolt screwed (MacKenzie et al., 2002). The analysis enables detection and occupancy into the thread on the housing (Fig. 2). We accessed sites on foot and to be a function of covariates and has three key assumptions: 1) target by helicopter (Robinson 44). Pilot trials indicated that Z. pedunculatus species occupy sites for the duration of sampling, 2) target species are would be highly detectable with a single camera (McDonald et al., identified correctly, and 3) the probability of detecting a species at a 2015a), so we installed one camera within 20 m of the centre of each site is independent of its detection at another site (MacKenzie et al., 80 × 80 m grid. Posts were hammered vertically into rock crevices or 2002). Because the second survey may not have met assumption 3, loose substrate to a depth of 100–200 mm, resulting in a camera height we interpret occupancy from this dataset as the proportion of sites above ground of 700–800 mm. Beneath each camera we cleared all that contained Z. pedunculatus at a given point in time (see Robley et al., 2010). We derived three variables predicted to influence the probability of detecting Z. pedunculatus in our study area: time since deployment of camera (added as a linear change in probability increasing with each 8-night sampling period) (Watkins et al., 2010), camera model (cate- gorical; Reconyx or Bushnell) (Swann et al., 2004), and relative bright- ness of the moon (categorical, based on dominant phase over 8- night sampling period; light = first quarter to third quarter, dark = waning crescent to waxing crescent) (Kotler et al., 1991). We derived 10 landscape-scale and seven site-scale variables predicted to influence the occupancy of Z. pedunculatus (Tables 1 and 2). The variables were assigned to competing hypotheses based on existing literature on the decline and persistence of small mammals in Australia (Tables 1 and 2). The two sets of variables were considered separately; the landscape-scale variables were applied to camera-trap data from the initial deployment and the site-scale variables were applied to Fig. 2. Camera trap configuration used to sample small mammals in the MacDonnell camera-trap data from the intensive follow-up sampling. We screened Ranges, Australia. each set of variables for pair-wise collinearity using Spearman's
50 96 P.J. McDonald et al. / Biological Conservation 191 (2015) 93–100
Table 1 Landscape-scale variables predicted to influence occupancy of the central rock-rat (Zyzomys pedunculatus) on quartzite mountains and ridges in the MacDonnell Ranges, Australia.
Hypothesis Explanation Variables (codes) Measurement References
Metapopulation Z. pedunculatus persists in Distance to nearest occupied Distance (km) to nearest occupied site, O'Brien et al., 2008 metapopulations rather than site (dist_occ) measured using Google Earth. at low densities across entire Continuous range or isolated Categorical: continuous range (0) or study area. mountain (isolation) isolated mountain (1). Predation Z. pedunculatus persists where Ruggedness (rugged_500m; Ruggedness index: standard deviation Luque-Larena et al., 2002 there is more topographic rugged_1 km; rugged_2km) of mean elevation within 500 m, 1 km Sappington et al., 2007 complexity that provides refuge and 2 km radius of each site (based on from predation by feral cats. 1 s digital elevation model). Climate change Z. pedunculatus persists at Elevation (elevation_500m; Mean elevation (m) within 500 m, 1 km Moritz et al., 2008; higher elevations. elevation_1 km; elevation_2km) and 2 km radius of each site (based on McDonald et al., 2015b 1 s digital elevation model). Fire Z. pedunculatus persists where fire Area burnt (burn_500m; Area burnt (ha) in last four years within Kelly et al., 2011; history provides increased food burn_1 km; burn_2km) 500 m, 1 km and 2 km radius of each site. Letnic and Dickman, 2005 resources or protection from predation.
correlation coefficient in R (v 3.0.3) and removed one of each pair where the site-scale data using the single variables, limiting the number of pa- the correlation coefficient was ≥ 0.6. From the landscape-scale variables rameters to reduce the possibility of over-fitting models (Tabachnick we removed two of the elevation variables (retaining the variable at the and Fidell, 2001). Models were ranked using AICc (Symonds and 500 m scale, assuming that this would more accurately reflect the po- Moussalli, 2011). We assessed the relative strength of models subse- tential influence of climate), and two fire variables (retaining the mid- quent to the best-ranked model by comparing the difference in criterion dle variable at the 1 km scale). From the site-scale variables we values of the best ranked model with model i(Δi). Δi values b2havesub- removed the fire variable, which was correlated with spinifex and stantial support, Δi values of 4–6 should be considered plausible, Δi tussock grass cover, and the forb variable, which was correlated with values 7–10 have minimal support, and Δi values N 10 should be tussock grass cover. We assessed the remaining variables for multi- rejected (Burnham and Anderson, 2002; Symonds and Moussalli, collinearity, calculating variance inflation factors by regressing each 2011). We also calculated Akaike weights (wi) as a way of assessing variable against the others using the R package ‘car’ (Fox et al., 2014). the relative strength of evidence in support of a model, varying from 0 There were no variance inflation factors N 4, and thus we retained all re- with no support to 1 with complete support. We present the 99% confi- maining variables. dence set of models where the summed wi = 0.99. To examine the ef- Using all retained variables, we ran global, single-season occupancy fects of individual covariates on Z. pedunculatus occupancy, we plotted models to assess model fit of the landscape- and site-scale the fitted relationship from the best-ranked model, estimating the Z. pedunculatus data, separately (MacKenzie and Bailey, 2004). Using lower and upper 95% confidence intervals using the delta method (Ver 10,000 bootstrap iterations, the landscape-scale data exhibited no evi- Hoef, 2012). dence of overdispersion (p = 0.55, c-hat = 0.67). Site-scale data showed lack-of-fit and overdispersion (p b0.01, c-hat = 6.35), so we 3. Results used the small-sample corrected Quasi-Akaike Information Criterion (QAICc) and applied a c-hat value of 6.35 to the model ranking for 3.1. Camera trap results these data (Burnham and Anderson, 2002). To identify the most parsi- monious detection model for Z. pedunculatus, we ran a single-season oc- For the landscape-scale survey, three cameras failed from the cupancy model with all three occupancy variables and a null model with day of installation. From the remaining 47 cameras we recorded no covariates. Detection models were ranked using Akaike's Informa- Z. pedunculatus at five sites in two areas, a northern area near Mt Giles tion Criterion (AICc or QAICc), and the model with the lowest AICc (1389 m) and a southern area near Counts Point (1117 m) (Fig. 1). value was applied to all subsequent occupancy models (Symonds and Adult Z. pedunculatus were recorded from all five sites and a juvenile Moussalli, 2011). We then ran occupancy models for the landscape- was recorded from one site. We also recorded five additional species scale data using all single and paired variable combinations and for of small mammals previously known from the higher elevation
Table 2 Site-scale variables predicted to influence occupancy of the central rock-rat (Zyzomys pedunculatus) within a known sub-population in the MacDonnell Ranges, Australia.
Hypothesis Explanation Variable(s) Measurement References
Den Rock-crevices are a limiting factor for Crevice 0 = nil or sparse rock crevices on site Walker et al., 2003 Z. pedunculatus 1 = moderate to abundant rock crevices on site Food Tussock grasses and forbs/low shrubs Tussock, % cover summed from 50 m point intersect Edwards, 2013b are an important source of food for Forbshrub transect through site Nano et al., 2003 Z. pedunculatus Seeds produced by tussock grasses, forbs, Seed 0 = b20% of tussock grass, forb or low shrub has Edwards, 2013b and low shrubs are an important food source little or no seed Nano et al., 2003 for Z. pedunculatus 1=b20% of tussock grass, forb or low shrub has high amounts of seed or N20% tussock grass, forb or low shrub has low amounts of seed Predation Small mammals persist where steep terrain Slope Average slope (°) across site measured with Luque-Larena et al., 2002 provides refuge from predation a clinometer Small mammals persist close to cliff lines that Distance to cliff Distance (m) to nearest cliff (≥3 m vertical) Luque-Larena et al., 2002 provide refuge from predation Small mammals persist where dense Spinifex % cover summed from 50 m point intersect transect Letnic and Dickman, 2005 hummock grass provides refuge from predation through site
51 P.J. McDonald et al. / Biological Conservation 191 (2015) 93–100 97
Table 3 Top-ranked detection models for small mammal species recorded in camera trapping on quartzite mountains and ridges in the MacDonnell Ranges, Australia.
Species No. Naïve Best detection sites ψ model
Dasyuridae Pseudantechinus macdonnellensis 34 0.72 ψ(.), p(time) Sminthopsis longicaudata 11 0.23 ψ(.), p(.) Muridae Mus musculus 23 0.49 ψ(.), p(.) Notomys alexis 1 0.02 – Pseudomys desertor 5 0.11 ψ(.), p(camera) Zyzomys pedunculatus 5 0.11 ψ(.), p(time)
quartzite landforms: fat-tailed pseudantechinus (Pseudantechinus macdonnellensis) from 34 sites, long-tailed dunnart (Sminthopsis longicaudata) from 11 sites, house mouse (Mus musculus)from23 sites, spinifex hopping-mouse (Notomys alexis) from one site, and desert mouse (Pseudomys desertor)fromfive sites (Table 3). (See Table 4.) For the second site-scale survey, we targeted the Counts Point area where Z. pedunculatus was recorded from three sites during the first survey (Fig. 1). Cameras at all 20 sites here remained operational during the entire sampling period and we recorded Z. pedunculatus from 13 sites (Fig. 1). Cameras recorded images of multiple adult individuals from at least eight of these sites (as determined by body and tail length) and juveniles at seven sites.
3.2. Occupancy modelling
At the landscape-scale, the most parsimonious detection model for ‘ ’ Z. pedunculatus included the covariate time . There was a negative rela- Fig. 3. Relationship between: a) time since camera installation and probability of detec- tionship between detection probability and time since camera deploy- tion; and b) distance to occupied site and occupancy probability. 95% CI included. ment (Fig. 3). The occupancy covariate ‘dist-occ’ was present in the first seven best-ranked models (Table 4). This variable alone was the highest ranked model and had a wi of 0.28 (Table 4). The variable was In the subsequent site-scale survey, the most parsimonious detec- also the only covariate for which the 95% confidence interval for the tion model was the null model. However, there was a decrease in detec- slope of the logit covariate did not include zero. Occupancy probability tion probability with time since camera deployment and the 95% declined steeply with increasing distance to nearest occupied site, confidence interval for the slope of the logit ‘time’ did not include with a predicted probability of 0 at ≥5km(Fig. 3). ‘Presence’ sites zero. The addition of occupancy covariates to the null model resulted – ‘ ’ ‘ ’ were 0.5 1.5 km from the nearest other presence site while absence in marginal additional explained variance and the QAICc values for sites ranged from 0.5–42.3 km from the nearest ‘presence’ site (S3). these models were higher than for the null model (S3). All 95% confi- The inclusion of the covariates ‘rugged_1km’ and ‘burn_1km’ resulted dence intervals for the slopes of the occupancy logit variables over- in minor improvements in deviance explained (Table 4). There was a lapped zero. Z. pedunculatus was recorded on sites across the full positive relationship with ruggedness and area burnt within 1 km range of most variables, including: 12–40° slope, 0–100% burnt within over the last 5 years (S2). The only model without ‘dist_occ’ in the 5 years, 0–44% spinifex cover, 2–18% tussock grass cover, 0–12% forb 99% confidence set was the linear term ‘burn_1km’ and its squared cover, nil to abundant seed, and 10–240 m from the nearest cliff line. 2 term ‘burn_1km ’. From this polynomial model, predicted probability However, 12 of the 13 ‘presence’ sites had abundant rock crevices. of occupancy peaked with approximately 70% of the surrounding 1 km burnt (S2). 4. Discussion
4.1. Camera trapping small mammals in rugged environments Table 4 Best ranked occupancy models (99% confidence set) explaining the occurrence of Zyzomys pedunculatus at the landscape-scale on quartzite mountains and ridges in the MacDonnell Our study confirms the usefulness of camera traps for sampling Ranges, Australia. small mammals in rugged environments and when targeting rare spe- cies. The results support previous research showing that the level of rep- Model K −2ll AICc ΔAICc wi lication and spatial coverage of small mammal surveys can be increased ψ(dist_occ), p(time) 4 87.47 96.42 0 0.28 ψ(dist_occ + rugged_1 km), p(time) 5 85.73 97.17 0.77 0.19 substantially by using camera traps (De Bondi et al., 2010). Using baited ψ(dist_occ + burn_1 km), p(time) 5 86.13 97.59 1.17 0.13 camera traps and lightweight posts, we were able to sample many more ψ(dist_occ + rugged_500m), p(time) 5 86.89 98.35 1.93 0.09 sites concurrently – and over a larger area – than would have been pos- ψ(dist_occ + elevation_500m), p(time) 5 87.03 98.49 2.07 0.08 sible by live-trapping. Using a small helicopter (Robinson R44), two pas- ψ(dist_occ + isolation), p(time) 5 87.46 98.92 2.50 0.07 sengers were able to install up to 20 camera traps in a day with all ψ(dist_occ + rugged_2km), p(time) 5 87.47 98.93 2.51 0.07 ψ(burn_1 km + burn_1km2) 5 90.66 102.12 5.70 0.01 equipment loaded on initial departure. This included 15 min at each null model 0 106.30 site for installation and site descriptions, substantial walking between
K = number of model parameters, −2ll = −2 log-likelihood output from occupancy some sites, as well as a transit time of one hour to reach and leave the model, AICc = Akaike's Information Criterion, adjusted for small sample size, ΔAICc = area. Based on previous experience in the study area, the maximum difference in AICc from the best-ranked model, wi =relativemodelweight number of live-trapping sites that we could run concurrently with a
52 98 P.J. McDonald et al. / Biological Conservation 191 (2015) 93–100 single helicopter is 6–8. However, even this number of sites would po- (though not significant) relationship with ruggedness and surrounding tentially require animals to remain in traps for several hours after sun- area burnt within the last five years. Although models require validation rise, reinforcing the established ethical advantages of using camera via further sampling, they suggest that rock-rats may occupy those parts traps. While we acknowledge that the initial cost of acquiring large of the landscape where terrain provides some protection from preda- numbers of camera traps may be prohibitive, we predict that camera tion by feral cats (Felis catus)(e.g.Doherty et al., 2015)andwherelow traps will become the standard sampling technique for rare small shrubs and forbs – rather than spinifex grasses – typically dominate mammals (as they are already for large mammals) in a range of differ- the groundcover and provide reliable food resources (Nano et al., ent environments globally. 2003; Edwards, 2013b). Although our cameras were positioned lower on posts than in previ- Intensive sampling of one of the sub-populations located in the ous studies (De Bondi et al., 2010; Rendall et al., 2014), we could identify landscape-scale survey failed to identify any site covariates that influ- most small mammals to species level with high confidence. Images from enced occupancy of Z. pedunculatus. It is possible that additional, unre- the Reconyx cameras in particular were consistently clear, and we rec- corded variables, may influence occupancy at this scale. For example, ommend using cameras with similar depth-of-field when logistic con- we were not able to evaluate space use by feral cats in the area, and pre- straints dictate shorter posts. However, we note that the small dation pressure may be an important driver of occupancy at this scale. mammal species in our study area are readily distinguishable based However, previous research has shown that feral cats select for recently on size, head shape, tail shape, and tail length (S2). Sampling in rugged burnt open areas, probably because hunting success is higher in areas areas where there are morphologically similar species will require an with less grass cover (McGregor et al., 2014). By contrast, we found assessment of the minimum suitable height-above-ground and may that Z. pedunculatus was spread evenly across longer-unburnt densely need to consider using cameras with a white flash if colouration is an grassed sites and more open recently-burnt sites, perhaps because it important diagnostic feature (Glen et al., 2013; Meek et al., 2013). was able to access rock crevices for shelter in all sites. We also expect We recorded the critically endangered Z. pedunculatus at five sites. that occupancy may vary temporally in response to variation in food re- Although the per-night detectability of Z. pedunculatus could probably sources. For example, all five sites where we located Z. pedunculatus be increased by installing more cameras at each site, by collapsing during the landscape-scale survey had been totally or partially burnt nights into eight-night periods we were able to construct occupancy within the previous five years. At the time, the recently burnt areas pro- models appropriate to the long time-frames that the cameras were in vided substantial food resources, with prolific flowering and seeding of place. For Z. pedunculatus, occupancy modelling showed initially high forbs and low shrubs at many sites, whereas the dominant spinifex detectability and then a decline in detectability with time-since- grasses were not seeding on the longer-unburnt sites. In contrast, dur- installation. This decline suggests either that individuals lose interest ing the intensive follow-up sampling, spinifex grasses were seeding in the bait (from which they receive no food reward) or that bait- and food was abundant across sites with both fire histories. Regardless scent attractiveness itself declines with time. This is an important find- of any temporal variation, our results demonstrate flexible fire-history ing for camera trapping small mammals generally, and we recommend preferences in Z. pedunculatus and provide a contrast with the sympatric including time-since-installation as a sampling covariate when desert mouse (P. desertor), a typically rare but widespread rodent that assessing the detectability of small mammals, particularly when cam- requires long-unburnt vegetation (Letnic and Dickman, 2005). eras are operating for long periods. Regardless of the factors driving the decline in detectability, our results suggest that the duration of sam- 4.3. Conservation assessment and recommendations pling could be reduced when targeting Z. pedunculatus,andthehighde- tectability of this species means that camera traps are likely to be an Small and isolated sub-populations are usually most prone to effective monitoring tool (MacKenzie et al., 2003). extinction (Hanski et al., 1996). Consistent with this, we recorded In addition to Z. pedunculatus, we recorded all other small mammal Z. pedunculatus only from the two largest and most contiguous moun- species previously known from the study area, establishing the effec- tain ranges in our study region and failed to detect the species from tiveness of camera trapping as a general sampling tool for small mam- the isolated Mt. Sonder ridge, where this species was recorded from mals in this environment. Of particular significance, we recorded the conventional trapping surveys in 2010 and 2012 (McDonald et al., NT-listed vulnerable S. longicaudata from 11 sites. This species was pre- 2013). Even on the two contiguous ridges, we located only a single viously known from only five sites in the Northern Territory and is dif- sub-population on each. Given the high detectability of Z. pedunculatus ficult to live-trap. Our results suggest that camera trapping provides and the intensive sampling effort that we expended, our results confirm an effective means of better defining the distribution and conservation that this species occurs in only a small proportion (c. 10%, or 180 ha, of status of this and other trap-shy or cryptic small mammal species. the study area) of the ‘core’ quartzite mountainous habitat that is poten- tially available, most of which occurs in the study area. The IUCN listing 4.2. Factors influencing Z. pedunculatus occupancy of critically endangered is therefore well supported by our data (Criteria B2b(i,ii,iv)c(i,ii,iii)) and the current Federal (EPBC) and NT (TPWCA) While we recorded Z. pedunculatus from only five sites, we were able listings of endangered should be revised. to run occupancy models that provided insight into the factors driving The most urgent conservation priorities for wild Z. pedunculatus are occupancy. It could be argued that five sites are too few to make strong to 1) ensure the continued persistence of the two known sub- inferences about drivers of species' occurrence, but we suggest that this populations, 2) locate and protect additional sub-populations (includ- is likely the reality for the world's rarest fauna and that available data ing within and outside our study area), and 3) increase the area should be used conservatively to guide research and provide prelimi- occupied. Cats are likely to be an important factor limiting the occur- nary management directions. At the landscape-scale, distance to nearest rence of Z. pedunculatus (Woinarski et al., 2014)andpreliminarydietary occupied site was the most important covariate influencing occupancy data suggest they may preferentially hunt Z. pedunculatus over other of Z. pedunculatus and the species was unlikely to be recorded N5km small mammal species (McDonald et al., 2015a; McDonald, unpubl. from the nearest occupied site. This suggests that Z. pedunculatus occurs data). Therefore, the control or eradication of this invasive predator is in localised and isolated sub-populations within the greater matrix of likely to help achieve all three objectives. Although cat control is potentially suitable high-elevation quartzite habitat. In support of this, problematic, poison baiting (e.g. 1080) has proven effective at we found high levels of occupancy for Z. pedunculatus when we inten- appropriate times and is planned to be trialled in the study area sively sampled one of the two areas where this rodent had been located (Burrows et al., 2003; Christensen et al., 2012). This will be carried out during the landscape-scale sampling. Ruggedness and recent fire histo- under an experimental framework, with appropriate control areas and ry may also influence occupancy at the landscape-scale, with a positive bait deployment aimed at minimising non-target uptake by dingoes
53 P.J. McDonald et al. / Biological Conservation 191 (2015) 93–100 99
(Canis dingo), which may have some role in suppressing cat numbers Kelly, L.T., Nimmo, D.G., Spence-Bailey, L.M., Haslem, A., Watson, S.J., Clarke, M.F., Bennett, A.F., 2011. Influence of fire history on small mammal distributions: insights from a and activity (Wang and Fisher, 2012; Gordon et al., 2015). Given the 100-year post-fire chronosequence. Divers. Distrib. 17, 462–473. flexible use of areas with different fire-histories by Z. pedunculatus,it Kotler, B.P., Brown, J.S., Hasson, O., 1991. Factors affecting gerbil foraging behavior and is not clear whether longer-unburnt vegetation provides important rates of owl predation. Ecology 72, 2249–2260. Letnic, M., Dickman, C.R., 2005. The responses of small mammals to patches regenerating structural protection from cat predation in this environment. As a con- after fire and rainfall in the Simpson Desert, Australia. Austral Ecol. 30, 24–39. servative measure, we recommend that long unburnt (N15 years since Letnic, M., Dickman, C.R., 2010. Resource pulses and mammalian dynamics: conceptual fire) spinifex communities be retained in and around the known sub- models for hummock grasslands and other Australian desert habitats. Biol. Rev. 85, – populations. Research into the spatial ecology of feral cats and 501 521. Linkie, M., Dinata, Y., Nugroho, A., Haidir, I.A., 2007. Estimating occupancy of a data defi- Z. pedunculatus in this environment should be a high priority. Topo- cient mammalian species living in tropical rainforests: sun bears in the Kerinci Seblat graphically rugged sites should be targeted in surveys to locate new region, Sumatra. Biol. Conserv. 137, 20–27. sub-populations, with presence-only predictive modelling that is robust Luque-Larena, J.J., López, P., Gosálbez, J., 2002. Microhabitat use by the snow vole Chionomys nivalis in alpine environments reflects rock-dwelling preferences. Can. to small numbers of presence sites considered as a potential location J. Zool. 80, 36–41. method (e.g. MaxEnt; Pearson et al., 2007). Camera traps then should McGregor, H.W., Legge, S., Jones, M.E., Johnson, C.N., 2014. Landscape management of fire be deployed as the most useful and cost-effective survey tool at poten- and grazing regimes alters the fine-scale habitat utilisation by feral cats. PLoS ONE 9, fi e109097. tial new sites when these have been identi ed. MacKenzie, D.I., Bailey, L.L., 2004. Assessing the fit of site-occupancy models. J. Agric. Biol. Environ. Stat. 9, 300–318. MacKenzie, D.I., Nichols, J.D., Lachman, G.B., Droege, S.J., Royale, A.J., Langtimm, C.A., 2002. Acknowledgements Estimating site occupancy rates when detection probabilities are less than one. Ecology 83, 2248–2255. We acknowledge the assistance with fieldwork provided by NT MacKenzie, D.I., Nichols, J.D., Hines, J.E., Knutson, M.G., Franklin, A.B., 2003. Estimating site occupancy, colonization, and local extinction when a species is detected imperfectly. Parks and Wildlife rangers: Mark Trower, Daniel McCormack, and Hollie Ecology 84, 2200–2207. Gooden, and volunteer Eli Bendall. Edward Connellan from Mengel McDonald, P.J., Pavey, C.R., Knights, K., Grantham, D., Ward, S.J., Nano, C.E.M., 2013. Extant Helicopter Services safely and expertly negotiated numerous challeng- population of the Critically Endangered central rock-rat Zyzomys pedunculatus locat- ed in the Northern Territory, Australia. Oryx 47, 303–306. ing mountain-top landings. Work was carried out with ethics approval McDonald, P.J., Brittingham, R., Nano, C., Paltridge, R., 2015a. A new population of the from Charles Darwin University (approval no. A06028) and a scientific critically endangered central rock-rat (Zyzomys pedunculatus)discoveredinthe permit from the Northern Territory Parks and Wildlife Commission Northern Territory. Aust. Mammal. 37, 97–100. McDonald, P.J., Luck, G.W., Dickman, C.R., Ward, S.J., Crowther, M.S., 2015b. Using (permit no. 44636). multiple-source occurrence data to identify patterns and drivers of decline in arid- dwelling Australian marsupials. Ecography. http://dx.doi.org/10.1111/ecog.01212. McDonald-Madden, E., Chadès, I., McCarthy, M.A., Linkie, M., Possingham, H.P., 2011. Appendix A. Supplementary data Allocating conservation resources between areas where persistence of a species is uncertain. Ecol. Appl. 21, 844–858. fi Supplementary data to this article can be found online at http://dx. Meek, P., Vernes, K., Falzon, G., 2013. On the reliability of expert identi cation of small- medium sized mammals from camera trap photos. Wildl. Biol. Pract. 9, 1–19. doi.org/10.1016/j.biocon.2015.06.027. Moritz, C., Patton, J.L., Conroy, C.J., Parra, J.L., White, G.C., Beissinger, S.R., 2008. Impact of a century of climate change in small-mammal communities in Yosemite National Park, USA. Science 322, 261–264. References Nano, T.J., Smith, C.M., Jefferys, E., 2003. Investigation into the diet of the central rock-rat (Zyzomys pedunculatus). Wildl. Res. 30, 513–518. Burnham, K.P., Anderson, D., 2002. Model Selection and Multimodel Inference: A Practical O'Brien, C.M., Crowther, M.S., Dickman, C.R., Keating, J., 2008. Metapopulation dynamics Information-Theoretic Approach. 2nd ed. Springer Verlag, NewYork. and threatened species management: why does the broad-toothed rat (Mastacomys Burrows, N.D., Algar, D., Robinson, A.D., Sinagra, J., Ward, B., 2003. Controlling introduced fuscus) persist? Biol. Conserv. 141, 1962–1971. predators in the Gibson Desert of Western Australia. J. Arid Environ. 55, 691–713. Pavey, C.R., Cole, J.R., McDonald, P.J., Nano, C.E.M., 2014. Population dynamics and spatial Christensen, P.E.S., Ward, B.G., Sims, C., 2012. Predicting bait uptake by feral cats, Felis catus, ecology of a declining desert rodent, Pseudomys australis: the importance of refuges in semi-arid environments. Ecol. Manag. Restor. 14, 47–53. for persistence. J. Mammal. 95, 615–625. De Bondi, N., White, J.G., Stevens, M., Cooke, R., 2010. A comparison of the effectiveness of Pearson, R.G., Raxworthy, C.J., Nakamura, M., Peterson, A.T., 2007. Predicting species camera trapping and live-trapping for sampling terrestrial small-mammal communi- distributions from small numbers of occurrence records: a test case using cryptic ties. Wildl. Res. 37, 456–465. geckos in Madagascar. J. Biogeogr. 32, 102–117. Dixon, P.M., Olsen, A.R., Kahn, B.M., 1998. Measuring trends in ecological resources. Ecol. Rendall, A.R., Sutherland, D.R., Cooke, R., White, J., 2014. Camera trapping: a contempo- Appl. 8, 225–227. rary approach to monitoring invasive rodents in high conservation priority ecosys- Doherty, T.S., Davis, R.A., van Etten, E.J.B., Algar, D., Collier, N., Dickman, C.R., Edwards, G., tems. PLoS ONE 9, e86592. Masters, P., Palmer, R., Robinson, S., 2015. A continental-scale analysis of feral cat diet Robley, A., Ramsey, D., Woodford, L., Lindeman, M., Johnston, M., Forsyth, D., 2010. in Australia. J. Biogeogr. 42, 964–975. Evaluation of detection methods and sampling designs used to determine the Edwards, G.P., 2013a. Relative abundance of the central rock-rat, the desert mouse, and abundance of feral cats. Technical Report Series 181. Arthur Rylah Institute for the fat-tailed pseudantechinus at Ormiston Gorge in the West MacDonnell Ranges Environmental Research. National Park, Northern Territory. Aust. Mammal. 35, 144–148. Sappington, J.M., Longshore, K.M., Thompson, D.B., 2007. Quantifying landscape rugged- Edwards, G.P., 2013b. Temporal analysis of the diet of the central rock-rat. Aust. Mammal. ness for animal habitat analysis: A case study using bighorn sheep in the Mojave De- 35, 43–48. sert. J. Wildl. Manag. 71, 1419–1426. Fox, J., Sanford, W., Adler, D., Bates, D., Baud-Bovy, G., Ellison, S., Firth, D., Friendly, M., Silveira, L., Jacomo, A.T.A., Diniz-Filho, J.A.F., 2003. Camera trap, line transect census and Gorjanc, G., Graves, Sp, Heiberger, R., Laboissiere, R., Monette, G., Murdoch, D., track surveys: a comparative evaluation. Biol. Conserv. 114, 351–355. Nilsson, H., Ogle, D., Ripley, B., Venables, W., Zeileis, A., R-Core, 2014. Package ‘car’ Soisalo, M.K., Cavalcanti, S., 2006. Estimating the density of a jaguar population in the v2.0–22. R package. http://cran.r-project.org/web/packages/car/index.html. Brazilian Pantanal using camera-traps and capture–recapture sampling in combina- Glen, A.S., Cockburn, S., Nichols, M., Ekanayake, J., Warburton, B., 2013. Optimising camera tion with GPS radio-telemetry. Biol. Conserv. 129, 487–496. traps for monitoring small mammals. PLoS ONE 8, e67940. Sollmann, R., Mohamed, A., Samejima, H., Wilting, A., 2013. Risky business or simple Gordon, C.E., Feit, A., Grüber, J., Letnic, M., 2015. Mesopredator suppression by an apex solution — relative abundance indices from camera-trapping. Biol. Conserv. 159, predator alleviates the risk of predation perceived by small prey. Proc. R. Soc. B 405–412. 282. http://dx.doi.org/10.1098/rspb.2014.2870. Swann, D.E., Hass, C.C., Dalton, D.C., Wolf, S.A., 2004. Infrared-triggered cameras for Gu, W., Swihart, R.K., 2004. Absent or undetected? Effects of non-detection of species detecting wildlife: an evaluation and review. Wildl. Soc. Bull. 32, 357–365. occurrence on wildlife-habitat models. Biol. Conserv. 116, 195–203. Symonds, M.R.E., Moussalli, A., 2011. A brief guide to model selection, multimodel Hanski, I., Moilanen, A., Pakkala, T., Kuussaari, M., 1996. The quantitative incidence inference and model averaging in behavioural ecology using Akaike's Information function model and persistence of an endangered butterfly metapopulation. Conserv. Criteria. Behav. Ecol. Sociobiol. 65, 13–21. Biol. 10, 578–590. Tabachnick, B.G., Fidell, L.S., 2001. Using multivariate statistics. 5th ed. Pearson, New York. Hines, J.E., 2006. PRESENCE4—software to estimate patch occupancy and related parame- Ver Hoef, J.M., 2012. Who invented the delta method? Am. Stat. 66, 124–127. ters USGS–PWRC Available at http://www.mbr-pwrc. usgs.gov/software/presence. Vine, S.J., Crowther, M.S., Lapidge, S.J., Dickman, C.R., Mooney, N., Piggott, M.P., English, html ([verified 21 January 2015]). A.W., 2009. Comparison of methods to detect rare and cryptic species: a case study James, A.N., Gaston, K.J., Balmford, A., 1999. Balancing the Earth's 634 accounts. Nature using the red fox (Vulpes vulpes). Wildl. Res. 36, 436–446. 401, 323–324. Walker, R.S., Novaro, A.J., Branch, L.C., 2003. Effects of patch attributes, barriers, and Karanth, K.U., 1995. Estimating tiger Panthera tigris populations from camera-trap data distance between patches on the distribution of a rock-dwelling rodent (Lagidium using capture-recapture models. Biol. Conserv. 71, 333–338. viscacia). Landsc. Ecol. 18, 185–192.
54 100 P.J. McDonald et al. / Biological Conservation 191 (2015) 93–100
Wang, Y., Fisher, D.O., 2012. Dingoes affect activity of feral cats, but do not exclude them Woinarski, J., Morris, K., 2008. Zyzomys pedunculatus. IUCN Red List of Threatened Species from the habitat of an endangered macropod. Wildl. Res. 39, 611–620. v. 2012.2 http://www.iucnredlist.org [accessed 21 Jan 2014]. Watkins, A.F., McWhirter, J.L., King, C.M., 2010. Variable detectability in long-term Woinarski, J.C.Z., Burbidge, A.A., Harrison, P.L., 2014. The Action Plan for Australian population surveys of small mammals. Eur. J. Wildl. Res. 56, 261–274. Mammals 2012. CSIRO Publishing, Melbourne.
55 Chapter 7. Fire and grass cover influence occupancy patterns of rare rodents and feral cats in a mountain refuge: implications for management
Lightning is the main ignition source for wildfire on the mountains of the MacDonnell Ranges (Photo: P. McDonald)
Author contribution: I was responsible for the chapter concept, fieldwork (assisted by A. Stewart and others), data analysis, and writing, in consultation with my co-authors.
56 CSIRO PUBLISHING Wildlife Research, 2016, 43, 121–129 http://dx.doi.org/10.1071/WR15220
Fire and grass cover influence occupancy patterns of rare rodents and feral cats in a mountain refuge: implications for management
Peter J. McDonald A,B,E, Alistair Stewart A, Andrew T. Schubert C, Catherine E. M. Nano A, Chris R. Dickman B and Gary W. Luck D
AFlora and Fauna Division, Department of Land Resource Management, PO Box 1120, Alice Springs, NT 0871, Australia. BDesert Ecology Research Group, School of Biological Sciences, University of Sydney, New South Wales 2006, Australia. CDesert Wildlife Services, PO Box 3321, Alice Springs, NT 0871, Australia. DInstitute for Land, Water and Society, Charles Sturt University, Albury, NSW 2640, Australia. ECorresponding author. Email: [email protected]
Abstract Context. Feral cats (Felis catus) are implicated in the ongoing decline of Australian mammals. New research from northern Australia suggests that predation risk from feral cats could be managed by manipulating fire regimes to increase grass cover. Aims. We investigate the role of fire history and hummock grass cover in the occurrence of feral cats and rare rodents, including the critically endangered central rock-rat (Zyzomys pedunculatus), in a mountain refuge in central Australia. Methods. We installed 76 camera stations across four sites in the West MacDonnell National Park and used occupancy modelling to evaluate the influence of recent fire (within 5 years), hummock grass cover and ruggedness on feral cat and rodent occupancy. Key results. Occupancy of the central rock-rat was positively associated with areas burnt within the past 5 years – a relationship probably driven by increased food resources in early succession vegetation. In contrast, the desert mouse (Pseudomys desertor) was detected at locations with dense hummock grass that had remained unburnt over the same period. Feral cats were widespread across the study area, although our data suggest that they forage less frequently in areas with dense hummock grass cover. Conclusions. Our results suggest that fire management and grass cover manipulation can be used as a tool for rodent conservation in this environment and potentially elsewhere in arid Australia. Implications. Creating food-rich patches within dense hummock grasslands may allow central rock-rats to increase occupancy while simultaneously affording them protection from predation. Landscape-scale wildfire resulting in a single post-fire vegetation age class is likely to be unfavourable for native rodents in this environment.
Received 30 November 2015, accepted 23 February 2016, published online 2 May 2016
Introduction Australia has suffered the highest rate of mammal extinction Fire is a key disturbance factor in ecosystems worldwide (Bond globally in the last two centuries (Woinarski et al. 2014), and and Keeley 2005) and in many countries, land managers invest altered fire regimes are one of the potential drivers of mammal substantial resources to mitigate and suppress wildfire (Dombeck population decline (e.g. Burbidge and McKenzie 1989). While et al. 2004). One of the key conservation objectives of fire predation by the introduced domestic cat (Felis catus) (hereafter management is to enhance fauna populations; however, a lack referred to as ‘cat ’) and red fox (Vulpes vulpes) is a primary of ecological data on fauna responses to fire means that land driver of declines of Australian mammals (Woinarski et al. managers typically rely on guidelines based on maintaining 2015), new research from northern Australia has identified an plant community diversity (Clarke 2008). However, there is interaction between these two factors. In the southern Kimberley growing evidence to suggest that these guidelines are region, cats selected areas recently burnt by intense wildfire frequently unfavourable for fauna (Taylor et al. 2012; Griffiths (where fire reduced grass cover) and cat hunting success was et al. 2015; Kelly et al. 2015). higher in areas with sparse groundcover (McGregor et al. 2014;