Fire, cats and the decline of northern. Australia s mammals

Fire, cats and the decline of northern Australia s mammals By Hugh F. Davies ORCID: 0000-0002-8473-4540 Submitted in total fulfilment of the requirements of the degree of Doctor of Philosophy March 2018 School of BioSciences Faculty of Science The University of Melbourne

Hugh F. Davies, 2018. ii

Thesis abstract: In brief: Over the past 200 years, the Australian continent has suffered the highest rate of mammal extinction on Earth, and if current trends across northern Australia continue, this extinction rate will likely persist. In order to mitigate these declines, we need a better understanding of the drivers involved, and to ensure that the application of current management actions is as effective as possible. I first use newly developed survey techniques and statistical analyses to investigate the distribution and decline of mammal species on Melville Island, one of the few remaining areas in Australia to retain its complete mammal fauna. Secondly, I review the approach and application of the most widely used management tool for biodiversity conservation across northern Australian savannas: prescribed fire. In doing so, this thesis increases the capacity of management agencies to more effectively mitigate the decline of native mammals across the savannas of northern Australia, with concepts that can be applied in other fire-prone environments. The current global rate of species extinction is jeopardising the functionality of the ecosystems on which all life depends. Due partly to its long geographic isolation, the unique biodiversity of the Australian continent has proven particularly susceptible to disturbance. Australia has contributed a disproportionate number of the world s mammal extinctions over the 230 years since European settlement. While most of these mammal extinctions occurred from the mid 19 th to early 20 th Centuries and were concentrated in the southern and arid parts of Australia, over the past three decades severe declines have been recorded further north in the monsoonal tropics. Worryingly, the current decline of mammals in northern Australia is most evident in taxa similar to those driven to extinction and severe decline elsewhere in Australia (e.g. bandicoots, large rodents and dasyurids in the critical weight range [CWR] of body mass 35 5500 g). As such, if these trends continue, Australia will soon be facing more mammal extinctions. Multiple factors have been implicated in the current decline of native mammals across northern Australia, including the breakdown of traditional Aboriginal burning practices, predation by feral cats iii

(Felis catus), habitat degradation by large exotic herbivores and poisoning by the introduced cane toad (Rhinella marina, thought to mostly impact carnivorous mammals). However, these threats vary in their occurrence and magnitude across the landscape, and operate as complex interactions (both synergistic and antagonistic) with other threatening processes and abiotic factors. Hence, our understanding of the drivers involved, and our ability to effectively mitigate them, remains limited. If we are to effectively halt these declines and avoid further extinctions, we urgently need a better understanding of the drivers involved, as well as a clearer understanding of how best to apply the currently available management actions. This thesis has two main aims: 1) To further our understanding of the drivers associated with the ongoing decline of native mammals across northern Australia by utilising newly developed analytical and methodological techniques in one of the last remaining areas in Australia to support an intact mammal assemblage. 2) To review the current application of prescribed fire as the primary management tool for biodiversity conservation across northern Australia. These aims represent important knowledge gaps that have so far hindered the development and implementation of effective remedial conservation efforts for mammal species across northern Australia. Chapter 1 provides the general context for the thesis. Chapters 2 and 3 address the first aim. Specifically, Chapter 2 includes the use of motion-triggered camera traps to resurvey historical survey sites on Melville Island. I then use statistical analyses that specifically account for imperfect detection, to determine the environmental correlates of both current (2015) occupancy, as well as site-extinction over a 15-year period for the threatened brush-tailed rabbitrat (Conilurus penicillatus). I provide the first evidence that suggests predation by feral cats is driving the contraction of brush-tailed rabbit-rat populations on Melville Island, and hence has been a significant driver in the decline of this species across northern Australia more broadly. In Chapter 3, I investigate changes in the broader mammal community of Melville Island between 2000 and 2015, as iv

well as modelling the environmental correlates of current (2015) occupancy for an additional three species. I report further evidence of mammal decline on Melville Island (first reported in Chapter 2), demonstrating substantial decreases in site-level trap-success and species richness, as well as speciesspecific changes in trap-success for the northern brown bandicoot (Isoodon macrourus), brush-tailed rabbit-rat and black-footed tree-rat (Mesembriomys gouldii). The results from this chapter suggest that populations of native mammals on Melville Island are exhibiting similar patterns of decline to those recorded in Kakadu National Park two decades earlier, and across the northern Australian mainland more generally. Chapters 4 and 5 address the second aim. In Chapter 4, I use the species-specific responses of native mammals to a long-running, replicated fire experiment on Melville Island, to test the importance of pyrodiversity for maintaining native mammal diversity in northern Australian savannas. I demonstrate that the diversity of the entire mammal assemblage is positively associated with simulated pyrodiversity, but peaks at an intermediate value. Hence, in this case, maximising pyrodiversity would reduce native mammal assemblage diversity below its potential maximum. Importantly, I demonstrate a useful and flexible approach with which to develop fire management strategies based on the demonstrated requirements of target species. In Chapter 5, I then use this approach to investigate the importance of tailoring fire management to the target species, and demonstrate the potential ecological consequences of not doing so. My thesis supports the hypothesis that predation by feral cats has been a major proximate driver of native mammal decline across northern Australia, particularly in areas with a relatively open understorey. Native mammal populations in areas where cats are present should be considered at risk of population decline regardless of how resilient they appear to have been thus far. Until effective methods for the direct control of feral cats are developed, fire management is our best option for mitigating cat impacts at large spatial scales. Fire management for mammal conservation in northern Australian savannas will likely be most effective when spatially-tailored based of the fire-requirements of the target species. The effectiveness of such fire management could be increased when applied concurrently with other management actions (such as fenced predator-free exclosures). Characterising v

the relative effectiveness of fire management to mitigate cat impacts (alone and in conjunction with direct cat control) should be the focus of future research. vi

Declaration: This thesis contains no material which has been accepted for a degree or diploma by the University or any other institution, and to the best of my knowledge no material previously published or written by another person except where due acknowledgement is made in the Preface or in the text of the thesis. Hugh Davies March, 2018 vii

Preface: This thesis is a product of collaborations with numerous colleagues from multiple institutions. These include: my four PhD supervisors Brett Murphy (Charles Darwin University), Michael McCarthy (The University of Melbourne), Alan Andersen (Charles Darwin University) and Graeme Gillespie (The Northern Territory Government s Department of Environment and Natural Resources); the Tiwi Land Rangers Willie Rioli, José Puruntatameri, Willie Roberts, Colin Kerinaiua, Vivian Kerinauia, and Kim Brooks Womatakimi; as well other collaborators John Woinarski (Charles Darwin University), Ronald Firth (Strategen Environmental), Emily Nicholson (Deakin University) and Hayley Geyle (Charles Darwin University). Chapters 2 5 are written as independent scientific publications and therefore include adequate information to properly frame each body of work. As such, there is some unavoidable repetition and slight differences in formatting depending on journal preferences. Collaborator contributions for each chapter were: Chapter 2. This chapter was published as Top-down control of species distributions: feral cats driving the regional extinction of a threatened rodent in northern Australia in the journal Diversity and Distributions during my PhD candidature. The content of this chapter is nearly identical to the journal article, with the exception of minor edits to the figures and tables to be in accordance with the thesis structure. The idea for this paper was conceived by Brett Murphy, Michael McCarthy, Graeme Gillespie, Emily Nicholson and I. Two datasets were used for this chapter, the first of which was collected from 2000-02 by Ronald Firth and John Woinarski, and later formatted to allow occupancy modelling by Hayley Geyle. Willie Rioli, José Puruntatameri, Willie Roberts, Colin Kerinaiua, Vivian Kerinauia, Kim Brooks Womatakimi and I collected the second dataset in 2015. I completed all of the analyses, and wrote the first draft of the manuscript. Brett Murphy, Alan Andersen, Michael McCarthy, Graeme Gillespie, Ronald Firth, John Woinarski, Hayley Geyle and Emily Nicholson provided editorial comments on the chapter. viii

Chapter 3. This chapter was published as Declining populations in one of the last refuges for threatened mammal species in northern Australia in the journal Austral Ecology during my PhD candidature. The idea for this paper was conceived by Brett Murphy, Michael McCarthy, Graeme Gillespie, Alan Andersen and I. As in Chapter 2, two datasets were used for this chapter, the first of which was collected from 2000-02 by Ronald Firth and John Woinarski. Willie Rioli, José Puruntatameri, Willie Roberts, Colin Kerinaiua, Vivian Kerinauia, Kim Brooks Womatakimi and I collected the second dataset in 2015. I completed all of the analyses with guidance from Brett Murphy and John Woinarski. I wrote the first draft of the manuscript. Brett Murphy, Alan Andersen, Michael McCarthy, Graeme Gillespie, Ronald Firth and John Woinarski provided editorial comments on the chapter. Chapter 4. This chapter was published as An experimental test of whether pyrodiversity promotes mammal diversity in a northern Australian savanna in the Journal of Applied Ecology during my PhD candidature. The idea for this paper was conceived by Brett Murphy, Michael McCarthy, Alan Andersen and I. Willie Rioli, José Puruntatameri, Willie Roberts, Colin Kerinaiua, Vivian Kerinauia, Kim Brooks Womatakimi and I collected the data. I completed all of the analyses with guidance from Brett Murphy. I wrote the first draft of the manuscript. Brett Murphy and Alan Andersen provided editorial comments on the chapter. Chapter 5. This chapter will soon be submitted as The importance of tailoring fire management for target species: a case study from northern Australian savannas to the Journal of Applied Ecology. The idea for this paper was conceived by Brett Murphy, Michael McCarthy, Graeme Gillespie, Alan Andersen and I. This chapter utilised a large dataset collected and owned by the Northern Territory Government. I completed all of the analyses with guidance from Brett Murphy. I wrote the first draft of the manuscript. Brett Murphy provided editorial comments on the chapter. My PhD candidature and associated work was supported by an Australian Post-graduate Award from the Australian Federal Government and a University of Melbourne Research Scholarship. ix

First authored publications: 1. Hugh F. Davies, Michael A. McCarthy, Ronald S. C. Firth, John C. Z. Woinarski, Graeme R. Gillespie, Alan N. Andersen, Hayley M. Geyle, Emily Nicholson, Brett P. Murphy (2016) Top-down control of species distributions: feral cats driving the regional extinction of a threatened rodent in northern Australia. Diversity and Distributions, 23, 272-283. 2. Hugh F. Davies, Michael A. McCarthy, Ronald S. C. Firth, John C. Z. Woinarski, Graeme R. Gillespie, Alan N. Andersen, Willie Rioli, José Puruntatameri, Willie Roberts, Colin Kerinaiua, Vivian Kerinauia, Kim Brooks Womatakimi, Brett P. Murphy (2018) Declining populations in one of the last refuges for threatened mammal species in northern Australia. Austral Ecology, DOI: 10.1111/aec.12596. 3. Hugh F. Davies, Michael A. McCarthy, Alan N. Andersen, Willie Rioli, José Puruntatameri, Willie Roberts, Colin Kerinaiua, Vivian Kerinauia, Kim Brooks Womatakimi, Brett P. Murphy (2018) An experimental test of whether pyrodiversity promotes mammal diversity in a northern Australian savanna. Journal of Applied Ecology. 4. Hugh F. Davies, Michael A. McCarthy, Ronald S. C. Firth, John C. Z. Woinarski, Graeme R. Gillespie, Alan N. Andersen, Brett P. Murphy (in preparation) The importance of tailoring fire management for target species: a case study from northern Australian savannas. x

Other publications during candidature: 1. Hayley M. Geyle, Gurutzeta Guillera-Arroita, Hugh F. Davies, Ronald S.C Firth, John C.Z. Woinarski, Brett P. Murphy, Dale G. Nimmo, Euan G. Ritchie, Emily Nicholson, E. (in press) Towards meaningful monitoring: a case study of a threatened rodent. Austral Ecology. 2. Brett P. Murphy, Hugh F. Davies (2014) There is a critical weight range for Australia s declining tropical mammals. Global Ecology and Biogeography, 23, 1058-1061. xi

Acknowledgements: Finally (!!!), the time has come to reflect on the past four and a half years. Taking a moment to look back has turned out to be both a rewarding and humbling exercise. Rewarding, due to the knowledge, experiences and friendships I have gained. Humbling, due to the realisation that completing this thesis has only been possible due to the support of many amazing people, that I am massively indebted to and truly grateful are part of my life. Firstly, to my four supervisors Brett Murphy, Michael McCarthy, Alan Andersen and Graeme Gillespie, I simply can t thank you enough for all your guidance and patience over the years. Brett, it may sound cliché but I couldn t have had a better primary supervisor. I have the greatest respect for you, not only as an academic but as a genuinely good person. When you moved the 4000km to the other side of the country half way through my PhD I bet you thought you d see less of me, but no, like a bad smell I up and followed. My thanks also extend to your lovely (and accommodating) family, Birgit, Xavo and Elias. A massive thank you to Mick McCarthy the magnificent mathemagician for all you taught me about ecological modelling. Alan, without you, my PhD would have been a very different (and much more difficult) experience. In our first meeting, you suggested I become involved in the Tiwi Carbon Study. That was definitely a turning point in my PhD and I am so grateful for where that suggestion has led. Graeme, your support, especially your guidance on camera trapping techniques for small mammals in northern Australia ensured I was able to collect a powerful dataset, and for that I am extremely grateful. To the Tiwi Land Council, Tiwi rangers and the Tiwi people, I feel immensely lucky to have had the opportunity to spend so much time on your beautiful islands, and humbled from experiencing your amazing culture and community. A massive thank you to Willie Rioli, I could always count on you to help me out, you re an absolute legend. To Colin, Smokey, Kim, Willy Roberts and Viv thanks for all your help in the field and I apologise for how tough it got sometimes, especially in the build-up. Thank you to everyone at Tiwi College, a truly amazing place that sets the benchmark for remote education. A big thanks to Kate Hadden for all the support and advice over the years. I hope we can all continue working together to address some the environmental issues on the Tiwis. xii

To John Woinarski and Ron Firth, I feel privileged to have built on the work you started out on the Tiwis 15-years ago. Following in the footsteps of such respected ecologists is always inspiring and I hope I carried the torch well. Thanks for all the data, advice and feedback. To Jon Schatz and Anna Richards, thanks for helping me get my project off the ground, you both taught me a great deal about the ecology of tropical savannas. Next, I need to thank the most important people in my life, to the people that have always been there for me, and who are a never ending source of inspiration to me, my family. I dedicate this thesis to you. To my mum and dad, Margaret and John Davies and to my brothers, Michael and Chris, thanks for everything you ve all done for me; I owe all my happiness to you. Jaime, everything in my life is better with you. You even made the final stages of my PhD bearable. I promise to return the favour soon. I m looking forward to our lives post-phd. I love you all. Thanks to everyone in the Quantitative and Applied Ecology Group at the University of Melbourne. A truly amazing group, filled with seriously intelligent and genuinely friendly people. To Els, Micha, Daz, Freya, Matt Best (the conservation success story), Reid, Heini, Saras, Esti, Jose, Guru, Bron, Luke thanks for all the support, chats and coffee breaks. A special mention to the god-sends that are Pauline and Dolla, thanks for everything. Casey, I seriously couldn t have done this without you, thank you so much for your endless help, you re the best, around, nothing s gonna ever keep you down. To everyone at CDU, especially Alyson Stobo-Wilson, thanks for everything. Last but not least, a special thanks must go to my Tiwi twin, Michelle, in the wise and ever resonant words of Barry Manilow, it looks like we made it. xiii

Table of Contents: Thesis abstract... iii Declaration... vii Preface... viii Acknowledgements... xii List of Figures... xvi List of Tables... xvii Chapter 1: General Introduction... 1 Australia s historical mammal declines... 2 Mammal declines move north... 4 Fire management for biodiversity conservation: Going beyond pyrodiversity... 9 Thesis aims and scope... 11 References... 14 Chapter 2: Top-down control of species distributions: feral cats driving the regional extinction of a threatened rodent in northern Australia... 25 Abstract... 26 Introduction... 27 Methods... 29 Study site... 29 The study species... 30 Data collection... 30 Data analysis... 32 The probability of feral cat detection... 34 Correlates of C. penicillatus current distribution... 35 Correlates of C. penicillatus site extinction... 36 Results... 37 Correlates of C. penicillatus current distribution... 37 Correlates of C. penicillatus site extinction... 39 Discussion... 41 Conclusion... 45 References... 47 Supporting information... 55 Appendix S2.1... 55 Appendix S2.2... 56 Chapter 3: Declining populations in one of the last refuges for threatened mammal species in northern Australia... 57 Abstract... 58 Introduction... 58 Methods... 60 Study site... 60 Data collection... 61 Data analysis... 65 Trends in trap-success... 65 Correlates of mammal distribution in 2015... 65 Results... 67 Trends in trap-success... 67 Correlates of mammal distribution in 2015... 67 Discussion... 71 References... 75 Supporting information... 81 Appendix S3.1... 81 xiv

Appendix S3.2... 82 Chapter 4: An experimental test of whether pyrodiversity promotes mammal diversity in a northern Australian savanna... 83 Abstract... 84 Introduction... 85 Methods... 87 Study site... 87 Data collection... 87 Data analysis... 89 Results... 91 Discussion... 96 References... 103 Supporting information... 110 Appendix S4.1... 111 Appendix S4.2... 113 Chapter 5: Tailored fire management for target species: a case study from northern Australian savannas... 114 Abstract... 115 Introduction... 115 Methods... 117 Study site... 117 Data collection... 118 Data analysis... 121 Results... 123 Discussion... 128 References... 133 Chapter 6: General discussion... 139 An increased understanding of northern Australia s mammal decline... 140 Effective management for mammal conservation across northern Australia... 144 The importance of spatially tailored fire management... 145 Limitations and shortcomings... 146 Future research directions... 147 1. Determining why feral cat densities vary across northern Australia... 147 2. Linking the carbon economy with the conservation of northern Australia s declining native mammals... 149 A multi-faceted management approach... 150 Concluding remarks... 151 References... 152 xv

List of Figures: Chapter 1: General Introduction Figure 1.1: Mammal disturbance across Australia s 85 bioregions... 4 Figure 1.2: Fire frequency across the Australian continent 2000-2016... 7 Chapter 2: Top-down control of species distributions: feral cats driving the regional extinction of a threatened rodent in northern Australia Figure 2.1: Location of the 88 survey sites on Melville Island... 31 Figure 2.2: Model averaged regression coefficient estimates for Conilurus penicillatus occupancy... 38 Figure 2.3: Modelled relationship between a) the probability of feral cat detection b) the density of shrubs and the probability of site-occupancy by Conilurus penicillatus... 39 Figure 2.4: Modelled relationship between a) the probability of feral cat detection b) the density of shrubs and the probability of Conilurus penicillatus site extinction... 41 Chapter 3: Declining populations in one of the last refuges for threatened mammal species in northern Australia Figure 3.1: Location of the 88 survey sites on Melville Island... 63 Figure 3.2: The difference in native mammal trap-success recorded in 2000-02 and 2015 on Melville Island... 68 Figure 3.3: Model averaged regression coefficient estimates for Melville Island native mammals... 70 Chapter 4: An experimental test of whether pyrodiversity promotes mammal diversity in a northern Australian savanna Figure 4.1: Locations of the 18 experimental fire plots on Melville Island... 88 Figure 4.2: The predicted abundance of native mammals for each experimental fire treatment... 95 Figure 4.3: The relationship between simulated pyrodiversity and mammal diversity/extinction risk on Melville Island... 96 Chapter 5: Tailored fire management for target species: a case study from northern Australian savannas Figure 5.1: Location of the 388 sites surveyed for native mammals across northern Australia... 119 Figure 5.2: Model averaged regression coefficient estimates for mammal species richness... 125 Figure 5.3: Modelled relationship between the diversity of post-fire age classes and site-level mammal species richness on mainland northern Australia... 126 Figure 5.4: The predicted abundance of native mammals for each categorised level of annual proportion burnt... 127 Figure 5.5: The relationship between simulated pyrodiversity and mammal diversity/extinction risk on the northern Australian mainland and Tiwi Islands... 128 xvi

List of Tables: Chapter 2: Top-down control of species distributions: feral cats driving the regional extinction of a threatened rodent in northern Australia Table 2.1: Description and justification of the variables used in analyses to assess the drivers of Conilurus penicillatus distribution and site extinction on Melville Island... 33 Table 2.2: Model selection results for dynamic occupancy models fit to test competing hypotheses of the drivers of Conilurus penicillatus site extinction... 40 Chapter 3: Declining populations in one of the last refuges for threatened mammal species in northern Australia Table 3.1: Description and justification of the variables used in analyses to assess the correlates of Melville Island mammal distribution in 2015... 64 Table 3.2: Summary of the changes in the native mammal populations of Melville Island recorded with live-trapping between 2000 02 and 2015... 69 Chapter 4: An experimental test of whether pyrodiversity promotes mammal diversity in a northern Australian savanna Table 4.1: A summary of the model selection process for native mammal alpha diversity and speciesspecific abundance in 18 experimental fire plots across Melville Island, Northern Territory... 92 Table 4.2: Predicted abundance of native mammal species resulting from different approaches to fire management... 93 Chapter 5: Tailored fire management for target species: a case study from northern Australian savannas Table 5.1: Description and justification of the variables included in analyses... 120 xvii

Chapter 1: General Introduction Savanna sunrise. 1

The current global rate of species extinction and population decline is jeopardising the functionality of the ecosystems on which all life on Earth depends (Barnosky et al., 2011, Dirzo et al., 2014). With more than 320 vertebrate species driven to extinction within the last 500 years, and estimates that between 16 and 33% of all vertebrate taxa are threatened, the rate and magnitude of anthropogenically-driven biodiversity loss is comparable with that of the five previous mass extinction events of Earth s history (Hoffmann et al., 2010, Dirzo et al., 2014). It is evident that at broad scales, the extinction risk of species is both phylogenetically and geographically non-random (McKinney, 1997, Cardillo et al., 2008). For example, when comparing the extinction risk of terrestrial vertebrate classes, amphibians are currently considered the most threatened, followed by mammals and reptiles, with birds the least threatened (Stuart et al., 2004, Dirzo et al., 2014). Furthermore, studies across vertebrate taxa have identified that specific traits are associated with heightened extinction risk, including a large body size, low reproductive rates and large home ranges (Cardillo et al., 2005, Dirzo et al., 2014). Geographically, species have been more prone to extinction in tropical regions, and in areas subject to the most rapid human population growth due to the associated habitat loss and transformation, as well as active persecution (Ceballos and Ehrlich, 2002, Dirzo et al., 2014, Woinarski et al., 2015). While broad-scale studies of the correlates of extinction risk can help develop hypotheses of the underlying drivers (Fisher and Owens, 2004), they have been criticised for having little impact on conservation management as they are often ambiguous, context-specific and difficult to translate into effective on-ground policy (Cardillo et al., 2008, Cardillo and Meijaard, 2012, Dirzo et al., 2014). This is particularly evident when investigating the recent mammal extinctions on the Australian continent. Australia s historical mammal declines The recent history of mammal extinction on the Australian continent remains an anomaly, with the patterns of loss being markedly different to those elsewhere in the world (Woinarski et al., 2015). Given that the Australian continent is very sparsely populated, relatively little modified and home to a modern, wealthy society, one might expect somewhat secure biodiversity. However, almost paradoxically, the opposite is true. Since European settlement in 1788, Australia has lost 29 mammal species to extinction, 2

constituting more than 10% of the continent s endemic terrestrial mammal species and the highest rate of mammal extinction on Earth. These 29 extinctions comprise 35% of all 84 mammal extinctions recorded globally since 1500 (Woinarski et al., 2015). What makes this more tragic, is that many of the species driven to extinction have been unique and phylogenetically distinctive species. For example the thylacine (Thylacinus cynocephalus) and pig-footed bandicoot (Chaeropus ecaudatus), both the sole species in their respective families, have now been lost forever. Somewhat ironically, the long geographic isolation of the Australian continent (since becoming separated from Gondwana almost 40 million years ago) that gave rise to the unique, endemic mammal fauna, is also the reason for its remarkable susceptibility to extinction (McLoughlin, 2001, Short et al., 2002, Woinarski et al., 2015). Australia s mammal extinctions have been geographically and taxonomically uneven. The susceptibility of a particular suite of species, coupled with the geographic pattern of mammal extinction helped shape the early hypotheses of the drivers involved. Early studies identified that non-flying mammal species with a mean adult body-size of 35 g to 5500 g the so called critical weight range (CWR) were most likely to become extinct (Burbidge and McKenzie, 1989), and multiple theories were proposed to explain this pattern. Some early studies linked the loss of CWR mammals with habitat degradation due to the loss of traditional Aboriginal burning practices (Bolton and Latz, 1978, Kitchener et al., 1980). For example, Bolton and Latz (1978) noted the survival of the western hare-wallaby (Lagorchestes hirsutus) only in areas where Aboriginal people had maintained quality habitat with a tight mosaic of vegetation in various stages of fire succession. Results such as this gave rise to one hypothesis that suggested that the susceptibility to extinction of CWR mammals reflected their limited mobility and high daily metabolic requirements, and hence high vulnerability to the reduction in available resources associated with widespread habitat changes experienced since European arrival, including changed fire regimes, grazing by introduced herbivores and agriculture (Burbidge and McKenzie, 1989). Alternatively, while some studies suggested a minor role of introduced predators in CWR mammal extinctions (Bolton and Latz, 1978), the persistence of CWR mammals on islands not colonised by feral cats (Felis catus) and foxes (Vulpes vulpes) (Short and Turner, 1994, Dickman, 1996), the recovery of populations in response to fox baiting (Kinnear et al., 1988) and the absence of the fox 3

in areas where mammals have experienced the least decline (i.e. Tasmania and the far north of Australia) (Short and Smith, 1994) (Figure 1.1), strongly implicated introduced predators as a major factor in Australia s mammal extinctions (Johnson, 2006). a) b) Figure 1.1: a) the level of disturbance to the non-flying native mammal fauna in each of Australia s 85 bioregions, expressed here by the faunal attrition index (FAI) of Burbidge et al. (2009) b) the distribution of the non-native red fox. Grey shading indicates bioregions where the fox was considered to be unambiguously absent by Burbidge et al. (2009). The bold line indicates the northern range limit of the fox in a typical year, and the bold dashed line indicates the limit in a good season, according to van Dyk and Strahan (2008). In both a) and b) the Tropic of Capricorn is indicated by the thin dashed line. Figure from Murphy and Davies (2014). Mammal declines move north The decline of mammal communities in monsoonal northern Australian was first identified between 1986 and 1993, and was initially attributed to a reduction in groundwater levels (Braithwaite and Muller, 1997). However, by first demonstrating the continuation of these declines following years of above rainfall, Woinarski et al. (2001) suggested that the declines recorded earlier throughout southern and arid Australia were now being recorded in the monsoonal tropics. This proposition was supported by evidence that the mammal species exhibiting decline in northern Australia were strongly concentrated in the CWR, suggesting fundamentally similar drivers (Murphy and Davies, 2014). The realisation that the mammal fauna of the vast savannas of northern Australia may not be as secure as once thought stimulated efforts to identify and mitigate the drivers involved. The gradual westward-spread of the 4

highly toxic, introduced cane toad (Rhinella marina) has caused dramatic population declines in native frog-eating predators across northern Australia (Phillips et al., 2003, Letnic et al., 2008). However, the direct effect of cane toad ingestion is likely restricted to only carnivorous mammals, such as the northern quoll (Dasyurus hallucatus), and such populations were experiencing decline before the toad front arrived (Braithwaite and Griffiths, 1994, Ibbett et al., 2017). As such, the role of cane toad ingestion in the northern mammal decline is likely only part of the story. Given that the fox is absent from most of monsoonal northern Australia (Short and Smith, 1994) (Figure 1.1), and no evidence suggesting a recent increase in the abundance of the feral cat (which has been present since the late 19 th century) (Abbott, 2002, Abbott, 2008), the timing of these declines appeared at odds with the suggestion that introduced predators had been a primary driver of CWR mammal decline across northern Australia. Similarly, while these declines have been interpreted as implicating the breakdown of Aboriginal fire management across northern Australia as a major driver, there is also a temporal discrepancy of more than 80 years between these events (Ritchie, 2009). It is important to note that while only identified in second half of the 20 th century, mammal populations across northern Australia may have been declining for a considerably longer period. Unfortunately, there are no systematic baseline data to either support or reject this. However, the observations of extremely high mammal abundance by early naturalists in northern Australia (Dahl, 1897), support the proposition of a gradual mammal decline since European arrival, rather than a recent collapse. As such, we do not know what event(s) coincided with the initial decline of native mammals across northern Australia. Fire is a highly conspicuous ecological process in the savannas of northern Australia, which are among the most flammable landscapes on Earth (Figure 1.2). Fire has long been part of northern Australian landscapes, being present in the fossil record since at least 60 million years ago (Crisp et al., 2011). Fire has also played a central role in more than 50,000 years of Aboriginal history (Jones, 1969). In northern Australian savannas, traditional Aboriginal fire management was implemented by small bands of nomadic hunters and gatherers, and involved the lighting of small, low-intensity fires that maintained fine-scale, heterogeneous fire mosaics (Yibarbuk et al., 2001, Bowman et al., 2004). However, due predominantly to introduced disease, the population of the region s indigenous people was reduced to 5

a tenth of pre-settlement levels within two generations following the arrival of Europeans in the mid 19 th century (Ritchie, 2009). As a result, by the start of the 20 th century, traditional relationships with the land were severely disrupted, and fire regimes had fundamentally changed (Ritchie, 2009). The finescale, heterogeneous fire mosaics maintained by traditional Aboriginal fire management were progressively lost, and replaced with large tracts of continuous fuel, resulting in large-scale (> 1million ha -1 ) wildfires (Bowman et al., 2001). The incidence of high-intensity wildfires, such as those occurring after July 31 st each year (henceforth referred to as late dry season fires) is also thought to have increased since the breakdown of traditional Aboriginal fire management. These changes are thought to have resulted in significant structural and compositional changes in the vegetation across large areas, including a reduction in fleshy-fruit bearing shrubs, perennial grasses and tree hollows (Vigilante and Bowman, 2004a, Vigilante and Bowman, 2004b, Andersen et al., 2005, Russell-Smith et al., 2010). The decline of populations of the northern cypress pine (Callitris intratropica) across northern Australian savannas is thought to reflect an increase in the frequency of high-severity fires since the breakdown of traditional burning practices (Bowman and Panton, 1993). The decline of mammals in monsoonal northern Australia parallels the decline of a range of granivorous birds in the same region (Franklin, 1999). Due to their similar diet, the concurrent decline of granivorous birds and many rodent species was thought to reflect a reduction in common food resources associated with changed fire regimes (Franklin, 1999, Woinarski et al., 2001). 6

Figure 1.2: The frequency of fire across the Australian continent in a 16-year period (November 2000 November 2016), based on a MODIS satellite-derived fire history (Giglio et al., 2016). The solid black line indicates the southern boundary of Australia s tropical savanna landscapes (Fox et al., 2001). Multiple studies over the last two decades have since demonstrated a significant negative association between fire and CWR mammal populations in northern Australia, both correlatively (Legge et al., 2008, Firth et al., 2010, Woinarski et al., 2010, Lawes et al., 2015) and experimentally (Pardon et al., 2003, Woinarski et al., 2004, Andersen et al., 2005, Griffiths et al., 2015). Such studies have greatly increased our understanding and helped develop management priorities. For example, decreasing fire frequency, increasing the area of long-unburnt habitat, as well as establishing and maintaining a finegrained mosaic are key objectives of fire management agencies across northern Australia that emerged from such research (Andersen et al., 2005). However, in terms of biodiversity recovery, there has been variable success to such landscape-scale fire management. While the reduction of fire frequency over a 5 million hectare area of the Kimberley region in Western Australia resulted in an improvement in a 7

range of biodiversity indicators, including the abundance and diversity of small native mammals (Legge et al., 2011b), there has been little evidence of such a response from other areas under active fire management most notably Kakadu National Park. A plausible explanation for this varying success is that the current approach to fire management across northern Australian savannas remains limited as it is still based on a rudimentary understanding of fire biodiversity relationships. Fire is a form of disturbance that can operate in complex interaction (both synergistically and antagonistically) with numerous abiotic and biotic factors (Bowman et al., 2009). While studies have demonstrated a negative correlation between northern Australian native mammal diversity and the frequency (Woinarski et al., 2010, Griffiths et al., 2015), intensity (Legge et al., 2008, Firth et al., 2010), and size of fires (Lawes et al., 2015), few studies have quantified the proximate mechanism(s) driving these negative relationships. Legge et al. (2008) highlighted that in most cases, mammal declines in northern Australia, as in other parts of the world (Whelan, 1995, Torre and Díaz, 2004), often occur over the course of a year following fire. This suggests that the indirect effects of fire on survival and reproductive success are often more important than direct fire-related mortality. The indirect effects of fire thought to contribute to the decline of native species across northern Australia include changes in the availability of critical resources and predation pressure. Studies have started to demonstrate how the availability of critical resources for mammals (as well as other species) is influenced by fire, including perennial grasses (Woinarski et al., 2004, Andersen et al., 2005), fruit-bearing shrubs (Vigilante and Bowman, 2004b) and tree hollows (Woinarski and Westaway, 2008). Concurrently, robust evidence has also amassed demonstrating the significant impact of invasive predators on Australian native mammals. Studies have demonstrated that feral cats preferentially select mammals as prey (Kutt, 2012), and mammal populations can be rapidly extirpated by even low densities of cats (Frank et al., 2014). Predation by feral cats has also resulted in the failure of mammal reintroduction programmes (Hardman et al., 2016, Short, 2016). Furthermore, the response of native mammal populations within fenced predator-free areas is stark evidence of the significant topdown pressure that predators impose on their prey species (Dickman, 2012, Hayward et al., 2014). In northern Australian savannas, the predation pressure imposed by feral cats (and possibly other 8

predators) appears to be exacerbated by processes that simplify vegetation structure, such as frequent, intense fires, and grazing by feral herbivores (Legge et al., 2011a, McGregor et al., 2016, Leahy et al., 2016). However, while Firth et al. (2010) demonstrated that frequent, high-intensity fires reduce the survival probability of the threatened brush-tailed rabbit-rat (Conilurus penicillatus), they also predicted population extirpation for this species at unburnt sites. This finding is significant as it suggests that native mammal declines in northern Australia are exacerbated by, but not dependent on fire. While the decline of native mammals in northern Australian savannas is likely the result of a combination of drivers (including fire, predation and resource availability), the relative contribution of each of these factors remains unclear. As a result, our ability to effectively mitigate these declines remains severely limited. Fire management for biodiversity conservation: Going beyond pyrodiversity Fire is a global ecosystem driver (Bond et al., 2005, Bowman et al., 2009, Kelly and Brotons, 2017), with profound effects on the evolution of biological communities and ecological processes (Gill et al., 1981, Whelan, 1995, Bond and Van Wilgen, 1996). As a result, fire is an integral part in the functioning of most fire-prone biomes (Bowman et al., 2011). The clearing of vegetation, grazing, introduction of exotic plants and animals, alteration of ignition sources and patterns, and the active suppression of fires associated with the expansion of human society has disrupted fire regimes on a global scale, causing substantial ecosystem change and biodiversity loss (Bowman et al., 2011). Worryingly, the disruption of fire regimes is likely to become exacerbated by global climate change (Moritz et al., 2012). However, fire is also a useful management tool that is applied in flammable environments worldwide. The application of prescribed fire is controversial. The suppression of fires around urban settlements can create a dangerous juxtaposition of flammable vegetation, high densities of humans and infrastructure (Bowman et al., 2011). This can result in catastrophic fires with huge loss of human life and infrastructure, such as those that occur infrequently through the Mediterranean basin, California, South Africa, Russia and southern Australia (Bowman et al., 2011). In such areas, intensive fuel 9

reduction burning is often used to reduce fire hazard (Fernandes and Botelho, 2003) but there is often concern surrounding the impact of such burning on fire-sensitive biodiversity values (Thompson et al., 1989, Driscoll and Roberts, 1997). However, in flammable environments that are sparsely populated and have little infrastructure, prescribed fire can be applied solely for conservation. Savannas, which are characterised by the co-dominance of trees and grasses, are a globally extensive biome, covering around 33 million km 2 or 20% of the world s land surface (Sankaran et al., 2005, Lehmann et al., 2011). Globally, savannas are heavily exploited by humans, supporting a large proportion of the world s human population, and most of the world s rangeland, livestock and wild herbivore biomass (Sankaran et al., 2005). Given our reliance on savannas, and their predicted sensitivity to future change in land-use and climate change (Bond et al., 2003), a thorough understanding of the factors influencing the functioning of savanna ecosystems is needed to underpin sustainable management (Sankaran et al., 2005). Tropical savannas are the most frequently and extensively burnt ecosystem on Earth (Murphy and Bowman, 2012). For example, as around 50% of the tropical savannas of northern Australia is burnt every year, on average just 3% of the landscape will remain unburnt for longer than 5 years (Andersen et al., 2005). As such, it is not surprising that fire plays a vital role in the persistence and functioning. Hence, our ability to effectively manage fire will be vital to the ongoing functioning of savanna ecosystems. By global standards, the tropical savannas of northern Australia appear relatively intact, supporting a very low human population with few areas cleared for agriculture (Williams et al., 1996). The breakdown of Aboriginal burning practices has been implicated in the decline of a range of communities and taxa across northern Australia, including the northern cypress pine (Callitris intratropica) (Bowman and Panton, 1993), sandstone heathlands (Russell-Smith et al., 2002), monsoon rainforests (Russell-Smith and Bowman, 1992), granivorous birds (Franklin, 1999), and more recently the catastrophic collapse of native mammal populations (Firth et al., 2010, Woinarski et al., 2011, Ziembicki et al., 2014). Currently, prescribed fire is applied by the majority of land users across northern Australia, making it the most widely applied management action (Russell-Smith, 1995). Importantly, the motivation for applying prescribed fire varies between the different land tenures of northern 10

Australia (including fuel reduction and infrastructure protection, promotion of more palatable herbs and grasses, hunting and cultural practices, biodiversity conservation) (Williams et al., 1998). When applied in an attempt to mitigate faunal declines, fire management often follows the patch mosaic burning paradigm, which attempts to establish and maintain a fine-scale, heterogeneous mosaic of varying fire histories (Parr and Andersen, 2006), under the assumption that pyrodiversity begets biodiversity (Martin and Sapsis, 1992). However, despite being widely adopted (particularly across the Australian and African continents), this approach lacks a solid empirical or theoretical basis, and often lacks clear operational guidelines that specify which aspects of pyrodiversity should be maximised (Parr and Andersen, 2006, Taylor et al., 2012). This uncertainty has recently stimulated critical analyses of the relationship between pyrodiversity and biodiversity (Parr and Andersen, 2006, Taylor et al., 2012, Kelly et al., 2016). For example, while Tingley et al. (2016) demonstrated that pyrodiversity promotes bird diversity in Californian conifer forests, Taylor et al. (2012) found no such pattern in a semi-arid region of south-eastern Australia, and suggested that burning for fire-mediated heterogeneity could actually threaten the avian fauna in this system. The validity of patch-mosaic burning has also been questioned for reptiles (Nimmo et al., 2013) and mammals (Kelly et al., 2012) in semi-arid Australia. These inconsistent results highlight the context-specificity of pyrodiversity requirements, and the risks associated with the blanket application of a management paradigm focussed on maximising pyrodiversity. Instead, fire management for biodiversity conservation should be directly underpinned by the demonstrated context-specific fire requirements of the target species (Andersen et al., 2014, Swan et al., 2015, Kelly et al., 2016). This is likely to be particularly important across the vast savannas of northern Australian, where the decline of native mammals appears to be exacerbated by the synergism between fire and other factors that vary in both their occurrence and magnitude across the landscape. Thesis aims and scope: While we have come a long way in our understanding of Australia s high rate of mammal extinction and the current declines across northern Australia, uncertainty still surrounds the drivers involved and 11

the best approach to management. As the decline of mammal species currently being recorded across northern Australia shares many similarities with the earlier mammal extinctions recorded throughout southern and arid Australia, we may soon lose more species to extinction. Given that Australia has already lost numerous phylogenetically unique species, these declines are a matter of global significance. The effective mitigation of these declines will require an improved understanding of the drivers involved, as well as how best to apply the most widely used management tool for biodiversity conservation in northern Australia: prescribed fire. Importantly, reviewing the application of prescribed fire for biodiversity conservation in northern Australia will develop concepts that can be applied in other fire-prone environments. In this thesis, I have two broad aims: 1) To further our understanding of the drivers associated with the ongoing decline of native mammals across northern Australia by utilising newly developed analytical and methodological techniques in one of the last remaining areas in Australia to support an intact mammal assemblage. 2) To review the current application of prescribed fire as the primary management tool for the conservation of small mammals across northern Australia. These aims address important knowledge gaps that have hindered the development and implementation of effective remedial conservation efforts for mammal species across northern Australia. The structure of my thesis is as follows. Here in Chapter 1 (Introduction), I provide the general context for the entire thesis. In Chapters 2 and 3, I address the first aim. Specifically, in Chapter 2, I determine the environmental correlates of both current (2015) occupancy, and site-extinction over a 15-year period for the threatened brush-tailed rabbit-rat (Conilurus penicillatus). In Chapter 3, I model the environmental correlates of current (2015) occupancy for an additional three species and investigate changes in the broader mammal community of Melville Island between 2000 and 2015. 12

In Chapters 4 and 5, I address the second aim. In Chapter 4, I use the species-specific responses of native mammals to a long-running, replicated fire experiment on Melville Island, to investigate how fire management might maximise native mammal diversity in northern Australian savannas. I outline a useful and flexible approach with which to develop fire management strategies based on the demonstrated requirements of target species. In Chapter 5, I then use this approach to demonstrate the importance of tailoring fire management to the target species, and demonstrate the potential ecological consequences of not doing so. Lastly, in Chapter 6 (Discussion), I highlight the advances made by my research. I then discuss the broad implications of my findings and how they have contributed towards the effective conservation of native mammal populations across northern Australia. I then identify remaining knowledge gaps, and suggest important future research directions. 13

References: Abbott, I. 2002. Origin and spread of the cat, Felis catus, on mainland Australia, with a discussion of the magnitude of its early impact on native fauna. Wildlife Research, 29, 51-74. Abbott, I. 2008. The spread of the cat (Felis catus) in Australia: Re-examination of the current conceptual model with additional information. Conservation Science Western Australia, 7, 1-17. Andersen, A. N., Cook, G. D., Corbett, L. K., Douglas, M. M., Eager, R. W., Russell-Smith, J., Setterfield, S. A., Williams, R. J. & Woinarski, J. C. Z. 2005. Fire frequency and biodiversity conservation in Australian tropical savannas: implications from the Kapalga fire experiment. Austral Ecology, 30, 155-167. Andersen, A. N., Ribbons, R. R., Pettit, M. & Parr, C. L. 2014. Burning for biodiversity: highly resilient ant communities respond only to strongly contrasting fire regimes in Australia's seasonal tropics. Journal of Applied Ecology, 51, 1406-1413. Barnosky, A. D., Matzke, N., Tomiya, S., Wogan, G. O. U., Swartz, B., Quental, T. B., Marshall, C., McGuire, J. L., Lindsey, E. L., Maguire, K. C., Mersey, B. & Ferrer, E. A. 2011. Has the Earth s sixth mass extinction already arrived? Nature, 471, 51. Bolton, B. & Latz, P. 1978. The Western Hare-Wallaby Lagorchestes Hirsutus (Gould) (Macropodidae), in the Tanami Desert. Wildlife Research, 5, 285-293. Bond, W., Midgley, G. & Woodward, F. 2003. The importance of low atmospheric CO2 and fire in promoting the spread of grasslands and savannas. Global Change Biology, 9, 973-982. Bond, W. & Van Wilgen, B. 1996. Fire and Plants. Population and Community Biology, 263. Bond, W. J., Woodward, F. I. & Midgley, G. F. 2005. The global distribution of ecosystems in a world without fire. New Phytologist, 165, 525-538. 14

Bowman, D., Price, O., Whitehead, P. & Walsh, A. 2001. The 'wilderness effect' and the decline of Callitris intratropica on the Arnhem Land Plateau, northern Australia. Australian Journal of Botany, 49, 665-672. Bowman, D. M. J. S., Balch, J., Artaxo, P., Bond, W. J., Cochrane, M. A., D Antonio, C. M., DeFries, R., Johnston, F. H., Keeley, J. E., Krawchuk, M. A., Kull, C. A., Mack, M., Moritz, M. A., Pyne, S., Roos, C. I., Scoot, A. C., Sodhi, N. S. & Swetnam, T. W. 2011. The human dimension of fire regimes on Earth. Journal of Biogeography, 38, 2223-2236. Bowman, D. M. J. S., Balch, J. K., Artaxo, P., Bond, W. J., Carlson, J. M., Cochrane, M. A., D Antonio, C. M., DeFries, R. S., Doyle, J. C., Harrison, S. P., Johnston, F. H., Keeley, J. E., Krawchuk, M. A., Kull, C. A., Marston, J. B., Moritz, M. A., Prentice, I. C., Roos, C. I., Scott, A. C., Swetnam, T. W., Van Der Werf, G. R. & Pyne, S. J. 2009. Fire in the Earth System. Science, 324, 481-484. Bowman, D. M. J. S. & Panton, W. J. 1993. Decline of Callitris intratropica R. T. Baker & H. G. Smith in the Northern Territory: Implications for pre- and post-european colonization fire regimes. Journal of Biogeography, 20, 373-381. Bowman, D. M. J. S., Walsh, A. & Prior, L. D. 2004. Landscape analysis of Aboriginal fire management in Central Arnhem Land, north Australia. Journal of Biogeography, 31, 207-223. Braithwaite, R. & Griffiths, A. 1994. Demographic variation and range contraction in the northern quoll, Dasyurus hallucatus (Marsupialia : Dasyuridae). Wildlife Research, 21, 203-217. Braithwaite, R. W. & Muller, W. J. 1997. Rainfall, groundwater and refuges: Predicting extinctions of Australian tropical mammal species. Australian Journal of Ecology, 22, 57-67. Burbidge, A. A. & McKenzie, N. L. 1989. Patterns in the modern decline of western Australia's vertebrate fauna: Causes and conservation implications. Biological Conservation, 50, 143-198. Burbidge, A. A., McKenzie, N. L., Brennan, K. E. C., Woinarski, J. C. Z., Dickman, C. R., Baynes, A., Gordon, G., Menkhorst, P. W. & Robinson, A. C. 2009. Conservation status and biogeography of Australia s terrestrial mammals. Australian Journal of Zoology, 56, 411-422. 15

Cardillo, M., Mace, G. M., Gittleman, J. L., Jones, K. E., Bielby, J. & Purvis, A. 2008. The predictability of extinction: biological and external correlates of decline in mammals. Proceedings of the Royal Society B: Biological Sciences, 275, 1441-1448. Cardillo, M., Mace, G. M., Jones, K. E., Bielby, J., Bininda-Emonds, O. R. P., Sechrest, W., Orme, C. D. L. & Purvis, A. 2005. Multiple Causes of High Extinction Risk in Large Mammal Species. Science, 309, 1239-1241. Cardillo, M. & Meijaard, E. 2012. Are comparative studies of extinction risk useful for conservation? Trends in Ecology & Evolution, 27, 167-171. Ceballos, G. & Ehrlich, P. R. 2002. Mammal Population Losses and the Extinction Crisis. Science, 296, 904-907. Crisp, M. D., Burrows, G. E., Cook, L. G., Thornhill, A. H. & Bowman, D. M. 2011. Flammable biomes dominated by eucalypts originated at the Cretaceous-Palaeogene boundary. Nature Communications, 2, 193. Dahl, K. 1897. Biological notes on north-australian mammalia. Zoologist, 1, 189-216. Dickman, C. R. 1996. Overview of the impacts of feral cats on Australian native fauna, Australian Nature Conservation Agency, Canberra. Dickman, C. R. 2012. Fences or ferals? Benefits and costs of conservation fencing in Australia. In: Somers, M. J. & Hayward, M. W. (eds.) Fencing for conservation: Restriction of Evolutionary Potential or a Riposte to Threatening Processes? Springer, New York. Dirzo, R., Young, H. S., Galetti, M., Ceballos, G., Isaac, N. J. B. & Collen, B. 2014. Defaunation in the Anthropocene. Science, 345, 401-406. Driscoll, D. A. & Roberts, J. D. 1997. Impact of fuel-reduction burning on the frog Geocrinia lutea in southwest Western Australia. Australian Journal of Ecology, 22, 334-339. 16

Fernandes, P. M. & Botelho, H. S. 2003. A review of prescribed burning effectiveness in fire hazard reduction. International Journal of Wildland Fire, 12, 117-128. Firth, R. S. C., Brook, B. W., Woinarski, J. C. Z. & Fordham, D. A. 2010. Decline and likely extinction of a northern Australian native rodent, the Brush-tailed Rabbit-rat Conilurus penicillatus. Biological Conservation, 143, 1193-1201. Fisher, D. O. & Owens, I. P. F. 2004. The comparative method in conservation biology. Trends in Ecology & Evolution, 19, 391-398. Frank, A. S. K., Johnson, C. N., Potts, J. M., Fisher, A., Lawes, M. J., Woinarski, J. C. Z., Tuft, K., Radford, I. J., Gordon, I. J., Collis, M.-A. & Legge, S. 2014. Experimental evidence that feral cats cause local extirpation of small mammals in Australia's tropical savannas. Journal of Applied Ecology, 51, 1486-1493. Franklin, D. C. 1999. Evidence of disarray amongst granivorous bird assemblages in the savannas of northern Australia, a region of sparse human settlement. Biological Conservation, 90, 53-68. Fox, I. D., Neldner, V. J., Wilson. G.W. and Bannink, P.J. (2001) The Vegetation of the Australian Tropical Savannas. Environmental Protection Agency, Brisbane. Giglio, L., Schroeder, W. & Justice, C. O. 2016. The collection 6 MODIS active fire detection algorithm and fire products. Remote Sensing of Environment, 178, 31-41. Gill, A. M., Groves, R. H. & Noble, I. R. 1981. Fire and the Australian biota, Australian Academy of Science, Canberra. Griffiths, A. D., Garnett, S. T. & Brook, B. W. 2015. Fire frequency matters more than fire size: Testing the pyrodiversity biodiversity paradigm for at-risk small mammals in an Australian tropical savanna. Biological Conservation, 186, 337-346. Hardman, B., Moro, D. & Calver, M. 2016. Direct evidence implicates feral cat predation as the primary cause of failure of a mammal reintroduction programme. Ecological Management & Restoration, 17, 152-158. 17

Hayward, M. W., Moseby, K. & Read, J. L. 2014. The role of predator exclosures in the conservation of Australian fauna. In: Glen, A. & Dickman, C. R. (eds.) Carnivores of Australia: past, present and future. CSIRO Publishing, Melbourne, Australia. Hoffmann, M., Hilton-Taylor, C., Angulo, A., Bohm, M., Brooks, T. M., Butchart, S. H. M., Carpenter, K. E., Chanson, J., Collen, B., Cox, N. A., Darwall, W. R. T., Dulvy, N. K., Harrison, L. R., Katariya, V., Pollock, C. M., Quader, S., Richman, N. I., Rodrigues, A. S. L., ognelli, M. F., Vie, J.-C., Aguiar, J. M., Allen, D. J., Allen, G. R., Amori, G., Ananjeva, N. B., Andreone, F., Andrew, P., Ortiz, A. L. A., Baillie, J. E. M., Baldi, R., Bell, B. D., Biju, S. D., Bird, J. P., Black-Decima, P., Blanc, J. J., Bolanos, F., Bolivar-G., W., Burfield, I. J., Burton, J. A., Capper, D. R., Castro, F., Catullo, G., Cavanagh, R. D., Channing, A., Chao, N. L., Chenery, A. M., Chiozza, F., Clausnitzer, V., Collar, N. J., Collett, L. C., Collette, B. B., Fernandez, C. F. C., Craig, M. T., Crosby, M. J., Cumberlidge, N., Cuttelod, A., Derocher, A. E., Diesmos, A. C., Donaldson, J. S., Duckworth, J. W., Dutson, G., Dutta, S. K., Emslie, R. H., Farjon, A., Fowler, S., Freyhof, J., Garshelis, D. L., Gerlach, J., Gower, D. J., Grant, T. D., Hammerson, G. A., Harris, R. B., Heaney, L. R., Hedges, S. B., Hero, J.-M., Hughes, B., Hussain, S. A., Icochea M., J., Inger, R. F., Ishii, N., Iskandar, D. T., Jenkins, R. K. B., Kaneko, Y., Kottelat, M., Kovacs, K. M., Kuzmin, S. L., La Marca, E., Lamoreux, J. F., Lau, M. W. N., Lavilla, E. O., Leus, K., Lewison, R. L., Lichtenstein, G., Livingstone, S. R., Lukoschek, V., Mallon, D. P., McGowan, P. J. K., McIvor, A., Moehlman, P. D., Molur, S., et al. 2010. The Impact of Conservation on the Status of the World s Vertebrates. Science, 330, 1503-1509. Ibbett, M., Woinarski, J. C. Z. & Oakwood, M. 2017. Declines in the mammal assemblage of a rugged sandstone environment in Kakadu National Park, Northern Territory, Australia. Australian Mammalogy, https://doi.org/10.1071/am17011. Johnson, C. 2006. Australia's Mammal Extinctions: a 50,000-year History. Cambridge University Press, Melbourne. Jones, R. 1969. Fire-stick farming. Australian Natural History, 16, 224-8. Kelly, L. & Brotons, L. 2017. Using fire to promote biodiversity. Science, 355, 1264-1265. 18

Kelly, L. T., Brotons, L. & McCarthy, M. A. 2016. Putting pyrodiversity to work for animal conservation. Conservation Biology, 31, 952-955. Kelly, L. T., Nimmo, D. G., Spence-Bailey, L. M., Taylor, R. S., Watson, S. J., Clarke, M. F. & Bennett, A. F. 2012. Managing fire mosaics for small mammal conservation: a landscape perspective. Journal of Applied Ecology, 49, 412-421. Kinnear, J. E., Onus, M. L. & Bromilow, R. N. 1988. Fox control and rock-wallaby population dynamics. Wildlife Research, 15, 435-450. Kitchener, D. J., Chapman, A., Muir, B. G. & Palmer, M. 1980. The conservation value for mammals of reserves in the western Australian wheatbelt. Biological Conservation, 18, 179-207. Kutt, A. S. 2012. Feral cat (Felis catus) prey size and selectivity in north-eastern Australia: implications for mammal conservation. Journal of Zoology, 287, 292-300. Lawes, M. J., Murphy, B. P., Fisher, A., Woinarski, J. C., Edwards, A. & Russell-Smith, J. 2015. Small mammals decline with increasing fire extent in northern Australia: evidence from long-term monitoring in Kakadu National Park. International Journal of Wildland Fire, 24, 712. Leahy, L., Legge, S. M., Tuft, K., McGregor, H. W., Barmuta, L. A., Jones, M. E. & Johnson, C. N. 2016. Amplified predation after fire suppresses rodent populations in Australia s tropical savannas. Wildlife Research, 42, 705-716. Legge, S., Kennedy, M. S., Lloyd, R. A. Y., Murphy, S. A. & Fisher, A. 2011a. Rapid recovery of mammal fauna in the central Kimberley, northern Australia, following the removal of introduced herbivores. Austral Ecology, 36, 791-799. Legge, S., Murphy, S., Heathcote, J., Flaxman, E., Augusteyn, J. & Crossman, M. 2008. The short-term effects of an extensive and high-intensity fire on vertebrates in the tropical savannas of the central Kimberley, northern Australia. Wildlife Research, 35, 33-43. Legge, S., Murphy, S., Kingswood, R., Maher, B. & Swan, D. 2011b. EcoFire: restoring the biodiversity values of the Kimberley region by managing fire. Ecological Management & Restoration, 12, 84-92. 19

Lehmann, C. E. R., Archibald, S. A., Hoffmann, W. A. & Bond, W. J. 2011. Deciphering the distribution of the savanna biome. New Phytologist, 191, 197-209. Letnic, M., Webb, J. K. & Shine, R. 2008. Invasive cane toads (Bufo marinus) cause mass mortality of freshwater crocodiles (Crocodylus johnstoni) in tropical Australia. Biological Conservation, 141, 1773-1782. Martin, R. E. & Sapsis, D. B. 1992. Fires as agents of biodiversity: pyrodiversity promotes biodiversity. In: Proceedings of the conference on biodiversity of northwest California ecosystems. Cooperative Extension, University of California, Berkeley. McGregor, H. W., Legge, S., Jones, M. E. & Johnson, C. N. 2016. Extraterritorial hunting expeditions to intense fire scars by feral cats. Scientific reports, 6, 22599. McKinney, M. L. 1997. Extinction vulnerability and selectivity: Combining ecological and paleontological views. Annual review of ecology and systematics, 28, 495-516. McLoughlin, S. 2001. The breakup history of Gondwana and its impact on pre-cenozoic floristic provincialism. Australian Journal of Botany, 49, 271-300. Moritz, M. A., Parisien, M.-A., Batllori, E., Krawchuk, M. A., Van Dorn, J., Ganz, D. J. & Hayhoe, K. 2012. Climate change and disruptions to global fire activity. Ecosphere, 3, 1-22. Murphy, B. P. & Bowman, D. M. J. S. 2012. What controls the distribution of tropical forest and savanna? Ecology Letters, 15, 748-758. Murphy, B. P. & Davies, H. F. 2014. There is a critical weight range for Australia s declining tropical mammals. Global Ecology and Biogeography, 23, 1058-1061. Nimmo, D., Kelly, L., Spence-Bailey, L., Watson, S., Taylor, R., Clarke, M. & Bennett, A. 2013. Fire mosaics and reptile conservation in a fire-prone region. Conservation Biology, 27, 345-353. 20

Pardon, L. G., Brook, B. W., Griffiths, A. D. & Braithwaite, R. W. 2003. Determinants of survival for the northern brown bandicoot under a landscape-scale fire experiment. Journal of Animal Ecology, 72, 106-115. Parr, C. L. & Andersen, A. N. 2006. Patch mosaic burning for biodiversity conservation: a critique of the pyrodiversity paradigm. Conservation Biology, 20, 1610-1619. Phillips, B. L., Brown, G. P. & Shine, R. 2003. Assessing the potential impact of cane toads on Australian snakes. Conservation Biology, 17, 1738-1747. Ritchie, D. 2009. Things fall apart: the end of an era of systematic Indigenous fire management. In: Russell-Smith, J., Whitehead, P. J. & Cooke, P. (eds.) Culture, ecology and economy of fire management in North Australian savannas: Rekindling the Wurrk tradition. CSIRO Publishing, Melbourne, Australia. Russell-Smith, J. & Bowman, D. M. J. S. 1992. Conservation of monsoon rainforest isolates in the Northern Territory, Australia. Biological Conservation, 59, 51-63. Russell-Smith, J. 1995. Fire Management. In: Press, A. J., Lea, D. A. M., Webb, A. L. & Graham, A. D. (eds.) Kakadu. Natural and Cultural Heritage and Management. Australian Nature Conservation Agency/North Australian Research Unit, Darwin, Australia. Russell-Smith, J., Price, O. F. & Murphy, B. P. 2010. Managing the matrix: decadal responses of eucalypt dominated savanna to ambient fire regimes. Ecological Applications, 20, 1615-1632. Russell-Smith, J., Ryan, P. G. & Cheal, D. C. 2002. Fire regimes and the conservation of sandstone heath in monsoonal northern Australia: frequency, interval, patchiness. Biological Conservation, 104, 91-106. Sankaran, M., Hanan, N. P., Scholes, R. J., Ratnam, J., Augustine, D. J., Cade, B. S., Gignoux, J., Higgins, S. I., Le Roux, X., Ludwig, F., Ardo, J., Banyikwa, F., Bronn, A., Bucini, G., Caylor, K. K., Coughenour, M. B., Diouf, A., Ekaya, W., Feral, C. J., February, E. C., Frost, P. G. H., Hiernaux, P., 21

Hrabar, H., Metzger, K. L., Prins, H. H. T., Ringrose, S., Sea, W., Tews, J., Worden, J. & Zambatis, N. 2005. Determinants of woody cover in African savannas. Nature, 438, 846. Short, J. 2016. Predation by feral cats key to the failure of a long-term reintroduction of the western barred bandicoot (Perameles bougainville). Wildlife Research, 43, 38-50. Short, J., Kinnear, J. E. & Robley, A. 2002. Surplus killing by introduced predators in Australia evidence for ineffective anti-predator adaptations in native prey species? Biological Conservation, 103, 283-301. Short, J. & Smith, A. 1994. Mammal decline and recovery in Australia. Journal of Mammalogy, 75, 288-297. Short, J. & Turner, B. 1994. A Test of the Vegetation Mosaic Hypothesis: A Hypothesis to Explain the Decline and Extinction of Australian Mammals. Conservation Biology, 8, 439-449. Stuart, S. N., Chanson, J. S., Cox, N. A., Young, B. E., Rodrigues, A. S. L., Fischman, D. L. & Waller, R. W. 2004. Status and Trends of Amphibian Declines and Extinctions Worldwide. Science, 306, 1783-1786. Swan, M., Christie, F., Sitters, H., York, A. & Di Stefano, J. 2015. Predicting faunal fire responses in heterogeneous landscapes: the role of habitat structure. Ecological Applications, 25, 2293-2305. Taylor, R. S., Watson, S. J., Nimmo, D. G., Kelly, L. T., Bennett, A. F. & Clarke, M. F. 2012. Landscape-scale effects of fire on bird assemblages: does pyrodiversity beget biodiversity? Diversity and Distributions, 18, 519-529. Thompson, M., Medlin, G., Hutchinson, R. & West, N. 1989. Short-Term Effects of Fuel Reduction Burning on Populations of Small Terrestrial Mammals. Wildlife Research, 16, 117-129. Tingley, M. W., Ruiz-Gutierrez, V., Wilkerson, R. L., Howell, C. A. & Siegel, R. B. Pyrodiversity promotes avian diversity over the decade following forest fire. Proc. R. Soc. B, 2016. The Royal Society, 20161703. 22

Torre, I. & Diaz, M. 2004. Small mammal abundance in Mediterranean post-fire habitats: a role for predators? Acta Oecologica, 25, 137-142. Van Dyck, S. & Strahan, R. 2008. The mammals of Australia, New Holland Publishers, Sydney. Vigilante, T. & Bowman, D. M. J. S. 2004a. Effects of fire history on the structure and floristic composition of woody vegetation around Kalumburu, North Kimberley, Australia: a landscape-scale natural experiment. Australian Journal of Botany, 52, 381-404. Vigilante, T. & Bowman, D. M. J. S. 2004b. Effects of individual fire events on the flower production of fruit-bearing tree species, with reference to Aboriginal people's management and use, at Kalumburu, North Kimberley, Australia. Australian Journal of Botany, 52, 405-416. Whelan, R. J. 1995. The ecology of fire, Cambridge university press, Cambridge. Williams, R. J., Duff, G. A., Bowman, D. M. J. S. & Cook, G. D. 1996. Variation in the composition and structure of tropical savannas as a function of rainfall and soil texture along a large-scale climatic gradient in the Northern Territory, Australia. Journal of Biogeography, 23, 747-756. Williams, R. J., Gill, A. M. & Moore, P. H. R. 1998. Seasonal changes in fire behaviour in a tropical savanna in northern Australia, International Journal of Wildland Fire, 8, 227-239. Woinarski, J. & Westaway, J. 2008. Hollow formation in the Eucalyptus miniata E. tetrodonta open forests and savanna woodlands of tropical northern Australia. Final report to Land and Water Australia (Native Vegetation Program). Project TRC-14. Woinarski, J. C. Z., Armstrong, M., Brennan, K., Fisher, A., Griffiths, A. D., Hill, B., Milne, D. J., Palmer, C., Ward, S., Watson, M., Winderlich, S. & Young, S. 2010. Monitoring indicates rapid and severe decline of native small mammals in Kakadu National Park, northern Australia. Wildlife Research, 37, 116-126. Woinarski, J. C. Z., Burbidge, A. A. & Harrison, P. L. 2015. Ongoing unraveling of a continental fauna: Decline and extinction of Australian mammals since European settlement. Proceedings of the National Academy of Sciences, 112, 4531-4540. 23

Woinarski, J. C. Z., Legge, S., Fitzsimons, J. A., Traill, B. J., Burbidge, A. A., Fisher, A., Firth, R. S. C., Gordon, I. J., Griffiths, A. D., Johnson, C. N., McKenzie, N. L., Palmer, C., Radford, I., Rankmore, B., Ritchie, E. G., Ward, S. & Ziembicki, M. 2011. The disappearing mammal fauna of northern Australia: context, cause, and response. Conservation Letters, 4, 192-201. Woinarski, J. C. Z., Milne, D. J. & Wanganeen, G. 2001. Changes in mammal populations in relatively intact landscapes of Kakadu National Park, Northern Territory, Australia. Austral Ecology, 26, 360-370. Woinarski, J. C. Z., Risler, J. & Kean, L. 2004. Response of vegetation and vertebrate fauna to 23 years of fire exclusion in a tropical Eucalyptus open forest, Northern Territory, Australia. Austral Ecology, 29, 156-176. Yibarbuk, D., Whitehead, P. J., Russell-Smith, J., Jackson, D., Godjuwa, C., Fisher, A., Cooke, P., Choquenot, D. & Bowman, D. M. J. S. 2001. Fire ecology and Aboriginal land management in central Arnhem Land, northern Australia: a tradition of ecosystem management. Journal of Biogeography, 28, 325-343. Ziembicki, M. R., Woinarski, J. C., Webb, J. K., Vanderduys, E., Tuft, K., Smith, J., Ritchie, E. G., Reardon, T. B., Radford, I. J., Preece, N., Perry, J., Murphy, B. P., McGregor, H., Legge, S., Leahy, L., Lawes, M. J., Kanowski, J., Johnson, C., James, A., Griffiths, A., Gillespie, G. R., Frank, A., Fisher, A. & Burbidge, A. A. 2014. Stemming the tide: progress towards resolving the causes of decline and implementing management responses for the disappearing mammal fauna of northern Australia. Therya, 6, 169-225. 24

Chapter 2: Top-down control of species distributions: feral cats driving the regional extinction of a threatened rodent in northern Australia Clockwise from top left: Feral cat detected on camera trap; Brush-tailed rabbit-rat; Tiwi Island savanna. 25

Abstract: Aim: To investigate if feral cats influence the distribution of Australia s largest remnant population of the threatened brush-tailed rabbit-rat Conilurus penicillatus, and examine whether they influenced the extinction probability of C. penicillatus over a 15-year period (2000-2015). Location: Melville Island, northern Australia. Methods: In 2015, small mammal surveys were conducted at 88 sites across Melville Island, 86 of which had previously been surveyed in 2000-2002. We used single-season occupancy models to investigate correlates of the current distribution of C. penicillatus, and dynamic occupancy models to investigate correlates of C. penicillatus local extinction. Results: Our results show that C. penicillatus, which once occurred more widely across the island, is now restricted to parts of the island where feral cats are rarely detected and shrub density is high. Our results suggest that feral cats are driving C. penicillatus towards extinction on Melville Island, and hence have likely been a significant driver in the decline of this species in northern Australia more broadly. The impact of feral cats appears to be mitigated by vegetation structure. Main conclusions: The ongoing development and implementation of methods to effectively reduce feral cat densities, coupled with the management of landscape processes to maintain shrub density, through fire management and the removal of large exotic herbivores, will contribute substantially to conserving this threatened species. This study demonstrates that the distribution of species can be strongly influenced 26

by top-down factors such as predation, thereby highlighting the importance of including biotic interactions when investigating the distribution of predation-susceptible species. Keywords: Australia, Conilurus penicillatus, distribution, extinction, feral cats, predation Introduction: Knowledge of the current and predicted future distributions of species is paramount for biodiversity conservation (Austin, 2007). While species distribution modelling (SDM) can address various problems in applied ecology and conservation biology, the assumption that distributions are limited primarily by bottom-up factors, such as resource availability, often leads to the omission of potentially important causal factors, such as biotic interactions (Guisan and Thuiller, 2005). For example, introduced predators can exert a strong regulatory force on the distribution of their prey, and the omission of such drivers could severely limit the utility of SDM for species whose distributions are strongly influenced by top-down factors such as predation by introduced carnivores (Guisan and Thuiller, 2005). Globally, less than 2% of recent extinctions have been directly attributed to introduced species (Gurevitch and Padilla, 2004). However, in areas with diverse populations of evolutionarily naïve prey, such as those that often occur on islands, the impacts of introduced carnivores on native animal species have been severe (Nogales et al., 2004, Medina et al., 2011). For example, feral cats (Felis catus) have contributed to at least 14% of global mammal, bird and reptile extinctions on islands (Medina et al., 2011, Doherty et al., 2015). Island rodents are particularly susceptible to introduced predators (Nogales et al., 2004), being driven to or near extinction on many islands throughout the Caribbean (Clough, 1976), the Galapagos (Patton and Hafner, 1983, Dowler et al., 2000) and north-western Mexico (Tershy et al., 2002). Australia s long history of geological isolation makes many species particularly susceptible to predation by introduced mammalian predators (Short et al., 2002). Since European arrival, Australia has 27

experienced the highest rate of modern mammal extinction on Earth, with the introduced feral cat and red fox (Vulpes vulpes) being implicated as key factors in the majority of these (IUCN, 1996, Short et al., 2002, Woinarski et al., 2014). Since its introduction to Australia, the feral cat has been implicated in the extinction of 19 species, as well as the recent catastrophic declines recorded across northern Australia s savanna landscapes (Johnson, 2006, Woinarski et al., 2011, Fisher et al., 2013, Ziembicki et al., 2014, Woinarski et al., 2014). Over the past two decades, robust evidence demonstrating the significant impact of feral cats on Australian native mammals has emerged. Feral cats preferentially select small mammals as prey (Kutt, 2012), and native mammal reintroduction programmes have failed due to feral cat predation (Hardman et al., 2016, Short, 2016). Frank et al. (2014) demonstrated that even low densities of feral cats have the ability to rapidly extirpate populations of re-introduced rodents in northern Australia. The impact of feral cat predation on small mammal populations in northern Australia appears to be related to processes that simplify vegetation structure, such as high-frequencies of high-intensity fires and heavy grazing by feral herbivores (Legge et al., 2011a, Lawes et al., 2015a, Leahy et al., 2016). Recent studies have demonstrated that cats in northern Australia hunt more successfully in structurally simple compared with complex habitats (Oakwood, 2000, McGregor et al., 2015). Despite the evidence that predation by feral cats can strongly influence small mammal populations, no study has directly linked cats with the regional extinctions that have occurred for many small mammal species across northern Australia. Using two datasets collected 15 years apart, we applied occupancy modelling to investigate if feral cats influence the distribution and decline of the nationally threatened brush-tailed rabbit-rat (Conilurus penicillatus), and to investigate the role of habitat structure in its persistence. Our study was conducted on Melville Island (5786 km 2 ), which represents an excellent model system to study the drivers of small mammal distribution in northern Australia s savanna landscapes. It is one of the last areas in Australia that retains its complete pre-european assemblage of small mammals (Burbidge et al., 2009), despite the prevalence of the hypothesised drivers of small mammal decline elsewhere in northern Australia, including feral cats and large herbivores (water buffalo Bubalus bubalis and horses Equus caballus), and very high fire frequencies. Melville Island has one of the largest extant population of C. penicillatus 28

(Woinarski et al., 2014), which occurred across vast areas of northern Australia at the time of European settlement (Cramb and Hocknull, 2010), but is now almost extinct on the mainland (Firth et al., 2006a, Woinarski et al., 2014). While anecdotal evidence suggests that direct predation by feral cats is a major driver in the decline of C. penicillatus, other possible mechanisms related to feral cats, such as disease, may also be contributing to the decline of this species (Woinarski et al., 2014). We modelled both the current spatial patterns of site-occupancy, and change in site-occupancy by C. penicillatus over a 15 year period (2000-2015) on Melville Island, to test the hypotheses that feral cats are a primary driver of C. penicillatus distribution, and that their impact is mitigated by vegetation structural complexity. We predicted that the probability of feral cat detection would be negatively correlated with the current distribution of C. penicillatus on Melville Island, and positively correlated with the probability of local extinction of C. penicillatus at historic sites. We also predicted that the current distribution and persistence of C. penicillatus on Melville Island would be associated with extensive shrub cover. Methods: Study site: Melville Island, the larger of the two main Tiwi Islands, is located 80 km north of Darwin, in Australia s Northern Territory (Figure 2.1). The island is of low relief ( 103 m above sea level) and experiences a tropical monsoonal climate with a wet season (November April) in which over 90% of the annual rainfall occurs (Australian Bureau of Meteorology, 2015). There is a substantial rainfall gradient, from 1400 mm in the east, to 2000 mm in the north-west. The major vegetation types are savanna woodlands and open forests dominated by Eucalyptus miniata, E. tetrodonta and Corymbia nesophila, with a predominantly grassy understorey. Shrub density is highly variable, and studies on the mainland have shown that this is influenced by fire regime (Russell-Smith et al., 2003, Woinarski et al., 2004). Recent fire mapping of the Tiwi Islands from 2000-2013, estimated that on average, 54% of the open savanna 29

woodland burns annually, with the majority burning in late dry season (i.e. after July 31 st ) (Richards et al., 2015). The study species: C. penicillatus is a predominantly granivorous rodent that dens in tree hollows and hollow logs (Firth et al., 2005, Firth et al., 2006b). It forages mostly on the ground, exposing it to feral cat predation. Its adult body weight (150 g) falls within the so-called critical weight range, within which Australian mammals have experienced high-rates of decline (Cardillo and Bromham, 2001, Johnson and Isaac, 2009). Data collection: In 2015, small mammal surveys were conducted at 88 sites in savanna forest across Melville Island (Figure 2.1), 86 of which had previously been surveyed in 2000-2002 (see Firth et al. 2006a). The 2000-2002 surveys followed the now standardised protocol for live-trapping in northern Australia, which involved a 50 x 50 m quadrat with 20 Elliot traps (33 x 10 x 9 cm) spaced equidistantly around the perimeter and four cage traps (56 x 20 x 20 cm) located on each corner. Traps were baited with a mixture of peanut butter, oats and honey, and set for three consecutive nights. The 2015 surveys used both live-trapping and camera-trapping (at 82 of the sites), and only cameratrapping at the other six sites. Live-trapping followed the 2000-2002 protocol but was conducted over four consecutive nights instead of three, and used eight cage traps and 16 Elliot traps. Camera trapping involved five horizontally facing motion-sensor cameras that were deployed at each site for at least 35 consecutive days. To increase the likelihood of being triggered, each camera was carefully positioned to ensure the base of its bait station was in the centre of the field of view (Gillespie et al., 2015). Bait stations contained a mixture of peanut butter, oats and honey. Vegetation within each camera s field of view was cleared using a fire rake to reduce the chance of false triggers and to reduce the risk posed by 30

fire. Of the five cameras deployed at each site, two were Reconyx TM HC550 Hyperfire white flash cameras (Reconyx Inc., Holmen, USA), while the remaining three cameras were Reconyx TM PC800 Hyperfire Professional infra-red flash cameras (Reconyx Inc., Holmen, USA). All cameras could be triggered 24-hours a day and were set to take three image bursts per trigger, with a one second time delay interval between images. The sensitivity of each camera was set to high, with cameras re-arming instantly after being triggered. Figure 2.1: Location of the 88 sites surveyed for small mammals in 2015 on Melville Island, the largest of the Tiwi Islands. The location of Melville Island relative to mainland Australia is shown in the inset. The diameter of the circles is proportional to the predicted probability of detecting a feral cat, based on the most highly-ranked spatial generalised linear model, ranging from 0.03 to 0.71. Filled circles indicate sites where feral cats were detected, open circles indicate non-detection. 31

Data analysis: To model spatial patterns of site-occupancy by C. penicillatus, we selected predictor variables relating to dingoes, fire, rainfall, and habitat, in addition to the probability of feral cat detection, based on results from the literature and recent research on the Tiwi Islands (Geyle, 2015). Dingoes were included because of their potential role in suppressing cat activity (Johnson, 2006, Kennedy et al., 2012), as well as their potential as predators of C. penicillatus. Variables relating to sampling methods were also incorporated, including Julian day because of potential seasonal variation in C. penicillatus detectability (Geyle, 2015). A full description and justification of the variables included in our analyses can be found in Table 2.1. 32

Table 2.1: Description and justification of the variables used in analyses to assess the drivers of Conilurus penicillatus distribution and site extinction on Melville Island. Explanatory variable Fire activity Rainfall Description and justification for inclusion Following Lawes et al. (2015), a remote-sensed fire variable derived from fine-scale (30 x 30 m) LANDSAT satellite imagery, representing the proportion of the area surrounding each site that was burnt in each year, averaged over the five years preceding mammal sampling. Calculations were made using an area with a radius of 3.2 km as shown by Lawes et al. (2015b) to have the strongest influence on small mammal populations. Mean annual rainfall (Australian Bureau of Meteorology, 2015). Throughout Australia, feral cat densities tend to be lower in areas of high rainfall (S. Legge and J. Woinarski, unpublished data) and mammal species in areas of high rainfall have declined the least (Fisher et al., 2013). Variable used in analyses to predict: Probability of feral cat detection C. penicillatus distribution C. penicillatus extinction Probability of feral cat detection C. penicillatus distribution Dingo activity The proportion of nights that dingoes were recorded on camera at each site. This was taken as an approximation of dingo activity at each site. Included in analyses to investigate the potential negative influence of dingoes on feral cats and potential benefits for small mammal populations (Johnson, 2006, Kennedy et al., 2012). Probability of feral cat detection C. penicillatus distribution Shrub density Distance to water Coarse woody debris (CWD) Probability of feral cat detection Feral herbivore presence Julian day Number of cameras operating Year A count of the number of shrubs in a 1 x 100 m quadrat at each site. Shrubs were defined as anything taller than 20 cm but shorter than 1.3 m, or taller than 1.3 m with a diameter at breast height of less than 5 cm. Shrubs with multiple stems were counted as a single individual. Vegetation structure has been demonstrated to reduce feral cat hunting success, and therefore could influence the distribution of feral cats as well as the occupancy and detectability of small mammals (McGregor et al., 2015). A remote-sensed variable measuring the distance (m) from each site to the closest permanent water body. The distance to water was demonstrated by Firth et al. (2006a) to strongly influence a number of small mammals on Melville Island. A count of the number of logs with a diameter of greater than 5 cm that crossed a 200 m long transect at each site. Included in analyses due to Firth et al. (2006b) demonstrating the reliance of C. penicillatus on fallen logs as den sites. As an index of feral cat activity, we used the predicted probability of detecting feral cats at each site, derived from spatially explicit generalised linear models (Murphy et al., 2010). The probability of feral cat detection was included in the analyses as cats have been implicated as a major factor in the northern mammal decline (Woinarski et al., 2011, Ziembicki et al., 2014). A binary variable indicating the presence or absence of large feral herbivores at each site. Feral herbivores were the introduced water buffalo (Bubalus bubalis) and horse (Equus caballus). Feral herbivores potentially influence small mammal populations via impacts on vegetation structure (Legge et al., 2011a). The Julian day of the calendar year that sampling started at each site. Recent work by Geyle (2015) demonstrated that the detectability of C. penicillatus increases throughout the dry season. An observation level covariate to account for the variation in detectability arising from uneven numbers of cameras operating at different sites due to camera malfunction and destruction. Year was included in the dynamic occupancy models to account for the variation in detectability arising from the different survey methods used in the two years. Probability of feral cat detection C. penicillatus distribution C. penicillatus extinction C. penicillatus distribution C. penicillatus distribution C. penicillatus distribution C. penicillatus extinction C. penicillatus extinction C. penicillatus distribution C. penicillatus extinction C. penicillatus distribution C. penicillatus extinction 33

The probability of feral cat detection: There was clear spatial patterning of feral cat detections across Melville Island (Figure 2.1). While the spatial clustering of detections could reflect higher densities of feral cats, it is also possible that feral cats were simply more detectable at those sites due to site-specific variability in habitat, cat behaviour and/or sampling efficiency. Unfortunately we could not model the effect of habitat variables on siteoccupancy by feral cats due to low detectability. Instead we used a range of variables hypothesised to influence the detectability of feral cats to investigate if there was any evidence that spatial patterning of feral cat detections was likely due to higher detectability at those sites. To do this we used occupancy models assuming a constant probability of site-occupancy by feral cats across Melville (see Appendix S2.1 in Supporting Information). We found no evidence that the detectability of feral cats was influenced by any site-specific habitat or sampling characteristics (such as the time of year). The only variable to have a significant influence on feral cat detectability was the distance to the closest cat detection. It is important to note that this analysis does not completely rule out the potential influence of site-specific habitat and/or sampling characteristics on the detectability of feral cats as our data may simply have been too sparse to detect an effect. Whatever the mechanism driving the spatial patterning of feral cat detections, the significant effect of spatial autocorrelation on the detectability of feral cats needed to be accounted for. As such, to generate the best predictions of the probability of feral cat detection at each site, we used spatially explicit generalised linear models (Murphy et al., 2010), with binomial errors, which use an autocovariate to account for the significant effect of spatial autocorrelation on feral cat detectability. The models were implemented using the fields, ncf and raster packages in the program R (R Development Core Team, 2013). We used the Akaike Information Criterion (AIC) to rank and identify parsimonious models (Burnham and Anderson, 2002). Accounting for spatial autocorrelation improved the explanatory power of the most parsimonious model by 12%, which explained 16% of the variation in cat detections. The model performed well at predicting a higher probability of feral cat detection in and around the areas where cats were actually detected (Figure 2.1). The area under the receiver 34

operating curve (AUC), calculated using the proc package in R, was 0.77. Models with AUC scores greater than 0.75 are considered useful (Phillips and Dudík, 2008). Correlates of C. penicillatus current distribution: Single-season occupancy models, which explicitly account for imperfect detection, were used to investigate how each predictor variable (Table 2.1) correlated with site-occupancy by C. penicillatus. Single-season occupancy modelling was conducted using only the 2015 camera trapping data in the unmarked package in R. Prior to analysis, predictor variables were centred and standardised by subtracting the mean and dividing by the standard deviation (Gelman and Hill, 2006). Due to the many variables and therefore the large number of possible models, occupancy modelling was applied in a two-step process. First, we determined which variables best explained the detectability of C. penicillatus by running all combinations of the nine variables we hypothesised might influence detectability (i.e. 512 models). This was done with occupancy constrained to a saturated model of the seven variables we hypothesised might influence site-occupancy by C. penicillatus. Model selection based on AIC was then used to select the most parsimonious detectability model in the candidate set. The second step involved running all possible combinations of the seven occupancy variables (128 models) with detectability constrained to the most parsimonious model identified in step one. Model selection based on AIC was then used again to determine the best model in the candidate set. Where no single model arose as superior at explaining the distribution of C. penicillatus (i.e. ΔAIC <2), model averaging provided parameter estimates based on the results of multiple models (Burnham and Anderson, 2002). Once the most parsimonious model with only the main effects was identified, we investigated the possible effect of an interaction between the probability of feral cat detection and both fire activity and shrub density. This was done because processes that simplify the structure of vegetation (such as frequent fire) might amplify the impact of feral cats. 35

Accounting for imperfect detection provides more realistic, but more imprecise estimates of occupancy (Guillera-Arroita et al., 2014). To gauge how accounting for detectability influenced our occupancy estimates and hence the confidence in our conclusions drawn from these models, we also ran all combinations of the occupancy variables, assuming constant detectability. We assessed model fit with a goodness-of-fit test based on parametric bootstrapping and Pearson s chisquare statistic. This method repeatedly simulates datasets based on a fitted model, and then evaluates the probability that the observed history of outcomes has a reasonable chance of happening if the model assessed is assumed to be correct (MacKenzie and Bailey, 2004). Correlates of C. penicillatus site extinction: We used dynamic occupancy modelling to investigate correlates of the dynamic processes associated with changes in site-occupancy by C. penicillatus from 2000 and 2015 (MacKenzie et al., 2003). Dynamic occupancy modelling was conducted using the original live-trapping data and the 2015 camera trapping data in the unmarked package in R. Explanatory variables were centred and standardised prior to analysis. Recent work based on the original live-trapping data demonstrated that the detectability of C. penicillatus on the Tiwi Islands is strongly influenced by the time of year that mammal surveys were conducted and fire frequency, while C. penicillatus occupancy is mostly determined by vegetation structural elements that are strongly associated with high rainfall as well as fire frequency (Geyle, 2015). Based on these results we parameterised the sub-model of initial occupancy as: ψ ~ Rainfall + Fire activity. We accounted for variation in detectability due to different survey methods being used in the two years by including year in the sub-model of detectability which was defined as: p ~ Julian day + Year + Fire activity. Initial modelling revealed colonisation probabilities were effectively zero (i.e. sites with a low probability of occupancy in 2000 remained unoccupied in 2015), so the sub-model for colonisation was fixed at zero. The sub-model for site extinction was then parameterised to test the following hypotheses about the correlates of C. penicillatus site extinction: 36

Hypothesis 1: Null (extinction constant across environmental space) Hypothesis 2: Vegetation structure (Shrub density) Hypothesis 3: Fire activity Hypothesis 4: Large herbivores Hypothesis 5: Feral cats Hypothesis 6: Feral cats and fire activity Hypothesis 7: Feral cats and large herbivores Hypothesis 8: Feral cats and vegetation structure Hypothesis 9: Feral cats interacting with fire activity Hypothesis 10: Feral cats interacting with large herbivores Hypothesis 11: Feral cats interacting with vegetation structure Model selection based on AIC was used to select the most parsimonious model in the candidate set. Results: Correlates of C. penicillatus distribution: While no single model arose as superior (i.e. had a ΔAIC <2), model averaging demonstrated that the probability of feral cat detection was the strongest predictor variable for the distribution of C. penicillatus (Figure 2.2), with C. penicillatus detected only at sites at which the probability of detecting a feral cat was <30% (Figure 2.3). Where the probability of feral cat detection was <10%, the probability of site-occupancy by C. penicillatus tended to be >70% (Figure 2.3). Shrub density was the only other significant predictor variable (Figure 2.2), and was positively associated with site-occupancy by C. penicillatus (Figure 2.3). Nightly detectability for C. penicillatus was 6.1% at each 5-camera survey site, but due to the length of time the cameras were deployed (minimum of 35 nights), a very high (>93%) probability of overall detection was achieved. Given the overall high detectability, the estimated rate of occupancy by the best model was only slightly greater than the naïve occupancy rate (i.e. the proportion of sites where C. 37

Estimated regression coefficient penicillatus was detected) and null model occupancy estimate (see Appendix S2.2 in Supporting Information) and the predicted impacts of covariates on occupancy were the same regardless of whether or not the models included effects of covariates (including Julian day ) on detectability. There was no evidence that including an interaction between fire and shrubs with the probability of feral cat detection improved the model fit. 4 2 0-2 -4-6 -8 Fire activity Probability of feral cat detection* CWD Distance to water Dingo activity Rainfall Shrub density* Figure 2.2: Model averaged regression coefficient estimates for Conilurus penicillatus occupancy. Error bars indicate 95% confidence intervals; asterisks indicate where they do not overlap zero. 38

Probability of C. penicillatus occupancy Probability of C. penicillatus occupancy a) b) 1 1 0.9 0.9 0.8 0.8 0.7 0.7 0.6 0.6 0.5 0.5 0.4 0.4 0.3 0.3 0.2 0.2 0.1 0.1 0 0 0.2 0.4 0.6 0.8 0 0 50 100 150 200 250 300 Probability of feral cat detection Shrub density (100 m -2 ) Figure 2.3: Modelled relationship between a) the probability of feral cat detection b) the density of shrubs and the probability of site-occupancy by Conilurus penicillatus. Thin lines indicate 95% confidence intervals. Crosses indicate observed data. Correlates of C. penicillatus site extinction: The probability of feral cat detection, shrub density and large herbivores were clear correlates of C. penicillatus site extinction. The four best models (i.e. ΔAIC <2), which incorporated either the probability of feral cat detection and shrub density or the probability of feral cat detection and large herbivore presence, had overwhelming support (Table 2.2), far superior to the model parameterised with the single main effect of the probability of feral cat detection. The most parsimonious model demonstrated that C. penicillatus had virtually no chance of persistence where the probability of detecting a feral cat was >40% (Figure 2.4). The probability of C. penicillatus site extinction was strongly influenced by the density of shrubs; with high shrub density significantly reducing the probability of C. penicillatus site extinction (Figure 2.4). 39

Table 2.2: Model selection results for dynamic occupancy models fit to test competing hypotheses of the drivers of Conilurus penicillatus site extinction. K indicates the number of parameters; w i is the Akaike weight; ΔAIC represents the difference between the model s AIC value and that of the topranking model. ψ denotes occupancy, γ denotes colonisation, ε denotes extinction and p denotes detectability. Bold text indicates the four best models. Model K ΔAIC w i H1: Null extinction 9 34.45 0.00 ψ (Rainfall + Fire activity) γ (~1) ε (~1) p (Julian day + Year + Fire activity) H2: Vegetation structure 10 25.39 0.00 ψ (Rainfall + Fire activity) γ (~1) ε (Shrub density) p (Julian day + Year + Fire activity) H3: Fire activity 10 21.21 0.00 ψ (Rainfall + Fire activity) γ (~1) ε (Fire activity) p (Julian day + Year + Fire activity) H4: Large herbivores 10 20.59 0.00 ψ (Rainfall + Fire activity) γ (~1) ε (Large herbivores) p (Julian day + Year + Fire activity) H5: Feral cats 10 16.46 0.00 ψ (Rainfall + Fire activity) γ (~1) ε (Feral cat detection) p (Julian day + Year + Fire activity) H6: Feral cats interacting with fire 12 12.60 0.00 ψ (Rainfall + Fire activity) γ (~1) ε (Feral cat detection * Fire activity) p (Julian day + Year + Fire activity) H7: Feral cats and fire 11 12.38 0.00 ψ (Rainfall + Fire activity) γ (~1) ε (Feral cat detection + Fire activity) p (Julian day + Year + Fire activity) H8: Feral cats interacting with large herbivores 12 3.58 0.09 ψ (Rainfall + Fire activity) γ (~1) ε (Feral cat detection * Large herbivores) p (Julian day + Year + Fire activity) H9: Feral cats and large herbivores 11 2.33 0.17 ψ (Rainfall + Fire activity) γ (~1) ε (Feral cat detection + Large herbivores) p (Julian day + Year + Fire activity) H10: Feral cats interacting with vegetation structure 12 1.82 0.21 ψ (Rainfall + Fire activity) γ (~1) ε (Feral cat detection * Shrub density) p (Julian day + Year + Fire activity) H11: Feral cats and vegetation structure 11 0.00 0.53 ψ (Rainfall + Fire activity) γ (~1) ε (Feral cat detection + Shrub density) p (Julian day + Year + Fire activity) 40

Probability of C. penicillatus site extinction Probability of C.penicillatus site extinction a) b) 1 1 0.9 0.9 0.8 0.8 0.7 0.7 0.6 0.6 0.5 0.5 0.4 0.4 0.3 0.3 0.2 0.2 0.1 0.1 0 0 0.2 0.4 0.6 0.8 Probability of feral cat detection 0 0 50 100 150 200 250 300 Shrub density (100 m -2 ) Figure 2.4: Modelled relationship between a) the probability of feral cat detection b) the density of shrubs and the probability of Conilurus penicillatus site extinction. Thin lines indicate 95% confidence intervals. Discussion: In an attempt to better understand the decline of small mammals across northern Australia, we modelled the current spatial patterns of site-occupancy, as well as the change in site-occupancy by C. penicillatus over a 15 year period (2000-2015) on Melville Island in northern Australia. Our results demonstrate that C. penicillatus was once more widespread on Melville Island but is now restricted to areas with a low probability of feral cat detection and high shrub density, where predation by feral cats is expected to be significantly diminished (McGregor et al., 2015). This study provides the first corroboration from the mesic savannas of the Northern Territory of the hypothesis that a sparse understorey (i.e. lack of shrubs) influences predation by feral cats to drive the decline of small mammals in northern Australia (McGregor et al., 2014, Lawes et al., 2015a, Leahy et al., 2016). Despite being widespread at the start of the 20 th century, the distribution of C. penicillatus has now contracted dramatically across mainland northern Australia (Dahl, 1897, Firth et al., 2010, Woinarski et al., 2014). However, C. penicillatus has persisted on Melville Island in relatively high numbers, with the island supporting one of the largest remaining populations of this species (Firth et al., 2006a). Due to a lack of historical records (Abbott and Burbidge, 1995), the timing of the introduction of feral cats 41

on Melville Island remains equivocal. It is possible that feral cats arrived with the first British settlement on Melville Island, the short-lived Fort Dundas (1824-28) (Brocklehurst, 1998), however, they may have arrived appreciably later on Melville Island than in other parts of northern Australia. If feral cats were established on Melville Island by the mid-19 th century, this poses the question: If feral cats have been a significant driver of C. penicillatus decline across northern Australia, why has this species persisted on Melville Island but declined substantially on the mainland? The persistence of C. penicillatus on Melville Island can be explained if the predation pressure exerted by feral cats on populations of C. penicillatus is mitigated by other environmental factors that vary between Melville Island and mainland northern Australia. Species distribution modelling by Firth et al. (2006a) led to the suggestion that the persistence and unusually high abundance of C. penicillatus on the Tiwi Islands may be related to more benign fire regimes. Frequent and/or intense fires reduce the survival of C. penicillatus and pose a significant threat to the long-term persistence of this species (Firth et al., 2010). Recent fire mapping of the Tiwi Islands has estimated that an average of 54% of the open savanna woodland is burnt each year, with 65% of the burning occurring in the late dry season (Richards et al., 2015). Russell-Smith et al. (1997) quantified the fire history of Kakadu National Park, where C. penicillatus became extinct in the 2000s. They reported that from 1980-94, an average of 55% of Kakadu s lowland savannas was burnt each year, with the majority of burning occurring in the early dry season period. This is similar to the estimates by Russell-Smith et al. (2009) derived for Kakadu for the period 1995-2004, who estimated that on average 50% of the lowland savannas were burnt each year, with 28% of burning occurring in the late dry season. Therefore, at a macro-scale, there is no evidence that the dominant fire regimes on Melville Island have been any less severe than those in areas where C. penicillatus has experienced massive declines on the mainland. However, as Melville Island receives a greater amount of dry-season rain, fires regimes at a finer scale may have been significantly less severe than in areas on mainland Australia. Determining how fire regimes differ between Melville Island and mainland Australia warrants further investigation. Studies have demonstrated that small mammal declines occur in the months following a fire, due to a heightened rate of predation as a result of reduced vegetation cover 42

(Leahy et al., 2016). As such, the current persistence of C. penicillatus populations on Melville Island is possibly related to more structurally complex vegetation. Measurements of shrub density in 18 experimental fire plots across Melville Island in 2015, estimated an average of 8,306 stems per hectare for small shrubs (<50 cm), and 5,939 stems per hectare for medium shrubs (50 cm to 2 m) (Anna Richards, unpublished data). These figures dramatically exceed shrub densities reported from analogous parts of the adjacent mainland (Kakadu, Litchfield and Nitmiluk National Parks) of 3,258 stems per hectare for small shrubs (<50 cm) and 802 stems per hectare for medium shrubs (50 cm to 2 m) (Russell-Smith et al., 2010). The exceptionally high shrub density recorded on Melville Island likely reflects the high productivity of the ecosystem (Richards et al., 2012). Not only does Melville Island sustain a more structurally complex understorey, evidence suggests that the rate of vegetation recovery after a fire event may also be greater than in less productive areas (Michelle Freeman, unpublished data). The length of time C. penicillatus is exposed to the heightened risk of predation following a fire event is likely to be shorter on Melville Island than in less productive areas. Our results demonstrate that shrub density is negatively, and the presence of large herbivores positively, associated with the probability of C. penicillatus site extinction. These results support the hypothesis that feral cats hunt less efficiently in areas with a dense understorey. There is evidence that large herbivores in northern Australia may simplify vegetation structure in a way that exposes small mammals to a greater risk of predation (Legge et al., 2011a). In our study, large herbivores were more likely to be present in areas with low shrub density. The role that large herbivores play in influencing vegetation structure in northern Australia, and the possible flow-on effects on small mammal populations, should be the focus of future research. While the current persistence of C. penicillatus on Melville Island is likely due in part to high productivity mitigating the top-down impacts of feral cat predation through vegetation structure, the bottom-up influences of high productivity have also likely contributed to the persistence of C. penicillatus on Melville Island. Shrub density was positively associated with site-occupancy by C. penicillatus irrespective of the probability of feral cat detection, possibly indicating the importance of productivity driven resource availability. The high availability of critical resources on Melville Island 43

such as tree hollows, hollow logs and perennial grasses (Crowley, 2008, Woinarski and Westaway, 2008), may result in increased breeding rates and high abundance, thereby offering C. penicillatus populations greater overall resilience to feral cat predation. The resilience of small mammal populations to feral cat predation may also vary spatially across Melville Island itself. This not only explains the pattern of site persistence and extinction of C. penicillatus, but possibly explains the spatial pattern of feral cat detections. If feral cat predation has effectively reduced prey availability in areas with low population resilience, cats in these areas may now be required to increase their foraging activity thus making them more detectable. Evidence suggests that C. penicillatus was one of the first species to exhibit decline on mainland northern Australia, followed by declines of a wide range of small mammal species (Woinarski et al., 2010). It is possible that the change in C. penicillatus populations reported here for Melville Island reflects the pattern of earlier decline on mainland northern Australia. If this is the case, Melville Island may soon be facing the same pattern of severe declines of a range of small mammal species. Whether or not history repeats itself will likely depend on the management response and investment in ongoing monitoring and research. Currently there are few feasible methods for the direct management of feral cats at large spatial scales (Nogales et al., 2004, Hardman et al., 2016). The applicability and effectiveness of each method is greatly influenced by a range of factors depending on the environment in which cat control is being undertaken (see Hardman et al., 2016 and references therein). Globally, feral cats have been eradicated from at least 100 islands, with most successful eradications using a combination of control methods (Nogales et al., 2004, DIISE, 2015). The majority of islands from where feral cats have been eradicated have been small (<5 km 2 ), to date the largest island from which feral cats have been successfully eradicated is subantarctic Marion Island (290 km 2 ) (Bester et al., 2002, Nogales et al., 2004). Melville Island is an order of magnitude larger (5788 km 2 ), making complete feral cat eradication a formidable task. While the ongoing development of effective feral cat control measures may make feral cat eradication from Melville Island a more realistic objective in the future, reducing the adverse impacts of feral cats in areas of high conservation value (i.e. around remnant populations of C. penicillatus) 44

could be a feasible and effective immediate management option. The benefits to small mammal populations from any direct management of feral cats may be amplified when implemented concurrently with landscape management that enhances vegetation structure (McGregor et al., 2014). CSIRO s Tiwi Carbon Study (Richards et al., 2012), recorded an increase in shrub density in response to reduced fire frequency on Melville Island (Anna Richards, unpublished data). Fire management that reduces the frequency of fire could therefore be a viable management option to reduce the impact of feral cats by enhancing shrub-layer complexity on Melville Island. The EcoFire project established a strategic regional prescribed burning programme in the Kimberley region of Western Australia, which aimed to increase the heterogeneity of vegetation age and structure, and the amount of old-growth vegetation (3 years or older) (Legge et al., 2011b). Biodiversity indicators (including small mammal species richness and abundance) collectively showed an improvement in response to the applied fire management. Similar to fire, large introduced herbivores also have the potential to reduce the complexity of understorey vegetation in the mesic savannas of northern Australia, and studies have demonstrated the recovery of small mammal populations following de-stocking (Legge et al., 2011a). The ongoing development of effective direct cat control methods, coupled with the management of landscape processes to maintain shrub density, through fire management and the removal of large exotic herbivores, are likely to contribute substantially to conserving C. penicillatus on the Tiwi Islands. Whether such management can be useful across northern Australia s vast savanna landscapes is much less certain, given the complexity of interactions between fire, exotic herbivores and vegetation cover and that these processes are highly context-specific. There is unlikely to be a one-size-fits-all approach to managing landscape processes (e.g. fire, grazing) for small mammal conservation. Conclusion: Predation pressure by feral cats, and its association with vegetation structure, was the most plausible reason for explaining the persistence and decline of a mammal species of significant conservation concern. Our study has highlighted that the ongoing monitoring of threatened species is crucial for their 45

conservation. Ongoing monitoring greatly increases our understanding of the causal factors of decline, allowing the development of effective management strategies for species conservation. The identification of such causal factors also has important implications for the use of species distribution modelling in the context of biodiversity conservation, when such models assume a primacy of bottomup factors. Our findings reinforce the need to include all important factors based on current ecological theory when predicting species distributions (Austin, 2002, Guisan and Thuiller, 2005). 46

References: Abbott, I., & Burbidge, A. 1995. The occurrence of mammal species on the islands of Australia: a summary of existing knowledge. CALMScience, 1, 259-324. Austin, M. 2002. Spatial prediction of species distribution: an interface between ecological theory and statistical modelling. Ecological Modelling, 157, 101-118. Austin, M. 2007. Species distribution models and ecological theory: A critical assessment and some possible new approaches. Ecological Modelling, 200, 1-19. Australian Bureau of Meteorology 2015. Australian rainfall and temperature surface data. Available:http://www.bom.gov.au/climate/averages/climatology/gridded-data-info/gridded-climatedata. Bester, M., Bloomer, J., Van Aarde, R., Erasmus, B., Van Rensburg, P., Skinner, J., Howell, P. & Naude, T. 2002. A review of the successful eradication of feral cats from sub-antarctic Marion Island, Southern Indian Ocean. South African Journal of Wildlife Research, 32, 65-73. Brocklehurst, P. 1998. The History and natural Resources of the Tiwi Islands, Northern Territory. Parks and Wildlife Commission of the Northern Territory, Darwin, Australia. Burbidge, A. A., McKenzie, N. L., Brennan, K. E. C., Woinarski, J. C. Z., Dickman, C. R., Baynes, A., Gordon, G., Menkhorst, P. W. & Robinson, A. C. 2009. Conservation status and biogeography of Australia s terrestrial mammals. Australian Journal of Zoology, 56, 411-422. Burnham, K. P. & Anderson, D. R. 2002. Model Selection and Multimodel Inference: A Practical Information-Theoretic Approach. Springer-Verlag, New York. Cardillo, M. & Bromham, L. 2001. Body size and risk of extinction in Australian mammals. Conservation Biology, 15, 1435-1440. Clough, G. C. 1976. Current status of two endangered Caribbean rodents. Biological Conservation, 10, 43-47. 47

Cramb, J. & Hocknull, S. 2010. New Quaternary records of Conilurus (Rodentia: Muridae) from eastern and northern Australia with the description of a new species. Zootaxa, 2634, 41-56. Crowley, G. 2008. Cockatoo grass Alloteropsis semialata as a keystone species in northern Australia. Northern Territory Naturalist, 20, 58-63. Dahl, K. 1897. Biological notes on north-australian mammalia. Zoologist, 1, 189-216. DIISE 2015. The Database of Island Invasive Species Eradications, developed by Island Conservation, Coastal Conservation Action Laboratory UCSC, IUCN SSC Invasive Species Specialist Group, University of Auckland and Landcare Research New Zealand [Online]. Auckland. Available: http://diise.islandconservation.org. Doherty, T. S., Davis, R. A., Van Etten, E. J. B., Algar, D., Collier, N., Dickman, C. R., Edwards, G., Masters, P., Palmer, R. & Robinson, S. 2015. A continental-scale analysis of feral cat diet in Australia. Journal of Biogeography, 42, 964-975. Dowler, R. C., Carroll, D. S. & Edwards, C. W. 2000. Rediscovery of rodents (Genus Nesoryzomys) considered extinct in the Galapagos Islands. Oryx, 34, 109-118. Firth, R. S., Jefferys, E., Woinarski, J. C. & Noske, R. A. 2005. The diet of the brush-tailed rabbit-rat (Conilurus penicillatus) from the monsoonal tropics of the Northern Territory, Australia. Wildlife Research, 32, 517-523. Firth, R. S. C., Brook, B. W., Woinarski, J. C. Z. & Fordham, D. A. 2010. Decline and likely extinction of a northern Australian native rodent, the Brush-tailed Rabbit-rat Conilurus penicillatus. Biological Conservation, 143, 1193-1201. Firth, R. S. C., Woinarski, J. C. Z., Brennan, K. G. & Hempel, C. 2006a. Environmental relationships of the brush-tailed rabbit-rat, Conilurus penicillatus, and other small mammals on the Tiwi Islands, northern Australia. Journal of Biogeography, 33, 1820-1837. 48

Firth, R. S. C., Woinarski, J. C. Z. & Noske, R. A. 2006b. Home range and den characteristics of the brush-tailed rabbit-rat (Conilurus penicillatus) in the monsoonal tropics of the Northern Territory, Australia. Wildlife Research, 33, 397-407. Fisher, D. O., Johnson, C. N., Lawes, M. J., Fritz, S. A., McCallum, H., Blomberg, S. P., VanDerWal, J., Abbott, B., Frank, A., Legge, S., Letnic, M., Thomas, C. R., Fisher, A., Gordon, I. J. & Kutt, A. 2013. The current decline of tropical marsupials in Australia: is history repeating? Global Ecology and Biogeography, 23, 181-190. Frank, A. S. K., Johnson, C. N., Potts, J. M., Fisher, A., Lawes, M. J., Woinarski, J. C. Z., Tuft, K., Radford, I. J., Gordon, I. J., Collis, M. A. & Legge, S. 2014. Experimental evidence that feral cats cause local extirpation of small mammals in Australia's tropical savannas. Journal of Applied Ecology, 51, 1486-1493. Gelman, A. & Hill, J. 2006. Data analysis using regression and multilevel/hierarchical models. Cambridge University Press, Cambridge. Geyle, H. 2015. Survey design for detecting declines in the threatened brush-tailed rabbit-rat Conilurus penicillatus on the Tiwi Islands of the Northern Territory. Deakin University, Melbourne. Gillespie, G. R., Brennan, K., Gentles, T., Hill, B., Low Choy, J., Mahney, T., Stevens, A. & Stokeld, D. 2015. A guide for the use of remote cameras for wildlife survey in northern Australia. National Environmental Research Program, Northern Australia Hub, Charles Darwin University. Guillera-Arroita, G., Lahoz-Monfort, J. J., MacKenzie, D. I., Wintle, B. A. & McCarthy, M. A. 2014. Ignoring imperfect detection in biological surveys is dangerous: a response to fitting and interpreting occupancy models'. PLoS ONE, 9, e99571. Guisan, A. & Thuiller, W. 2005. Predicting species distribution: offering more than simple habitat models. Ecology Letters, 8, 993-1009. Gurevitch, J. & Padilla, D. K. 2004. Are invasive species a major cause of extinctions? Trends in Ecology & Evolution, 19, 470-474. 49

Hardman, B., Moro, D. & Calver, M. 2016. Direct evidence implicates feral cat predation as the primary cause of failure of a mammal reintroduction programme. Ecological Management & Restoration, 17, 152-158. IUCN 1996. 1996 IUCN Red List of Threatened Animals. IUCN, Cambridge, UK. Johnson, C. 2006. Australia's Mammal Extinctions: a 50,000-year History. Cambridge University Press, Melbourne. Johnson, C. N. & Isaac, J. L. 2009. Body mass and extinction risk in Australian marsupials: The Critical Weight Range revisited. Austral Ecology, 34, 35-40. Kennedy, M., Phillips, B. L., Legge, S., Murphy, S. A. & Faulkner, R. A. 2012. Do dingoes suppress the activity of feral cats in northern Australia? Austral Ecology, 37, 134-139. Kutt, A. S. 2012. Feral cat (Felis catus) prey size and selectivity in north-eastern Australia: implications for mammal conservation. Journal of Zoology, 287, 292-300. Lawes, M. J., Fisher, D. O., Johnson, C. N., Blomberg, S. P., Frank, A. S. K., Fritz, S. A., McCallum, H., VanDerWal, J., Abbott, B. N., Legge, S., Letnic, M., Thomas, C. R., Thurgate, N., Fisher, A., Gordon, I. J. & Kutt, A. 2015a. Correlates of recent declines of rodents in northern and southern Australia: Habitat structure is critical. PLoS ONE, 10, e0130626. Lawes, M. J., Murphy, B. P., Fisher, A., Woinarski, J. C., Edwards, A. & Russell-Smith, J. 2015b. Small mammals decline with increasing fire extent in northern Australia: evidence from long-term monitoring in Kakadu National Park. International Journal of Wildland Fire, 24, 712. Leahy, L., Legge, S. M., Tuft, K., McGregor, H. W., Barmuta, L. A., Jones, M. E. & Johnson, C. N. 2016. Amplified predation after fire suppresses rodent populations in Australia s tropical savannas. Wildlife Research, 42, 705-716. Legge, S., Kennedy, M. S., Lloyd, R. A. Y., Murphy, S. A. & Fisher, A. 2011a. Rapid recovery of mammal fauna in the central Kimberley, northern Australia, following the removal of introduced herbivores. Austral Ecology, 36, 791-799. 50

Legge, S., Murphy, S., Kingswood, R., Maher, B. & Swan, D. 2011b. EcoFire: restoring the biodiversity values of the Kimberley region by managing fire. Ecological Management & Restoration, 12, 84-92. MacKenzie, D. I. & Bailey, L. L. 2004. Assessing the fit of site-occupancy models. Journal of Agricultural, Biological, and Environmental Statistics, 9, 300-318. 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. Ecology, 84, 2200-2207. 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. McGregor, H. W., Legge, S., Jones, M. E. & Johnson, C. N. 2014. Landscape management of fire and grazing regimes alters the fine-scale habitat utilisation by feral cats. PLoS ONE, 9, e109097. Medina, F. M., Bonnaud, E., Vidal, E., Tershy, B. R., Zavaleta, E. S., Donlan, J. C., Keitt, B. S., Corre, M., Horwath, S. V. & Nogales, M. 2011. A global review of the impacts of invasive cats on island endangered vertebrates. Global Change Biology, 17, 3503-3510. Murphy, B. P., Paron, P., Prior, L. D., Boggs, G. S., Franklin, D. C. & Bowman, D. M. 2010. Using generalized autoregressive error models to understand fire vegetation soil feedbacks in a mulga spinifex landscape mosaic. Journal of Biogeography, 37, 2169-2182. Nogales, M., Martin, A., Tershy, B. R., Donlan, C. J., Veitch, D., Puerta, N., Wood, B. & Alonso, J. 2004. A review of feral cat eradication on islands. Conservation Biology, 18, 310-319. Oakwood, M. 2000. Reproduction and demography of the northern quoll, (Dasyurus hallucatus), in the lowland savanna of northern Australia. Australian Journal of Zoology, 48, 519-539. Patton, J. L. & Hafner, M. 1983. Biosystematics of the native rodents of the Galapagos Archipelago, Ecuador. Patterns of evolution in Galapagos organisms, 539, 568. 51

Phillips, S. J. & Dudik, M. 2008. Modeling of species distributions with Maxent: new extensions and a comprehensive evaluation. Ecography, 31, 161-175. R Development Core Team 2013. R: A Language and Environment for Statistical Computing. R Foundation for Statistical Computing, Vienna, Austria. Richards, A. E., Andersen, A. N., Schatz, J. O. N., Eager, R., Dawes, T. Z., Hadden, K., Scheepers, K. & Van Der Geest, M. 2012. Savanna burning, greenhouse gas emissions and indigenous livelihoods: Introducing the Tiwi Carbon Study. Austral Ecology, 37, 712-723. Richards, A. E., Liedloff, A. & Sschatz, J. 2015. Tiwi Islands CFI capability project report. Australia. CSIRO, Darwin. Russell-Smith, J., Edwards, A., Woinarski, J., McCartney, J., Kerin, S., Winderlich, S., Murphy, B. P. & Watt, F. 2009. Fire and biodiversity monitoring for conservation managers: a 10-year assessment of the 'Three Parks' (Kakadu, Litchfield and Nitmiluk) program. In: Russell-Smith, J., Whitehead, P. J. & Cooke, P. (eds.) Culture, Ecology and Economy of Fire Management in North Australian Savannas: Rekindling the Wurrk Tradition. CSIRO Publishing, Melbourne. Russell-Smith, J., Price, O. F. & Murphy, B. P. 2010. Managing the matrix: decadal responses of eucalypt dominated savanna to ambient fire regimes. Ecological Applications, 20, 1615-1632. Russell-Smith, J., Ryan, P. G. & Durieu, R. 1997. A LANDSAT MSS-derived fire history of Kakadu National Park, monsoonal northern Australia, 1980-94: seasonal extent, frequency and patchiness. Journal of Applied Ecology, 34, 748-766. Russell-Smith, J., Whitehead, P. J., Cook, G. D. & Hoare, J. L. 2003. Response of Eucalyptusdominated savanna to frequent fires: Lessons from Munmarlary, 1973 1996. Ecological Monographs, 73, 349-375. Short, J. 2016. Predation by feral cats key to the failure of a long-term reintroduction of the western barred bandicoot (Perameles bougainville). Wildlife Research, 43, 38-50. 52

Short, J., Kinnear, J. E. & Robley, A. 2002. Surplus killing by introduced predators in Australia evidence for ineffective anti-predator adaptations in native prey species? Biological Conservation, 103, 283-301. Tershy, B., Donlan, C., Keitt, B., Croll, D., Sanchez, J., Wood, B., Hermosillo, M., Howald, G. & Biavaschi, N. 2002. Island conservation in north-west Mexico: a conservation model integrating research, education and exotic mammal eradication. Turning the tide: the eradication of invasive species, 293-300. Woinarski, J., Burbidge, A. & Harrison, P. 2014. Action Plan for Australian Mammals 2012. CSIRO Publishing, Melbourne. Woinarski, J. & Westaway, J. 2008. Hollow formation in the Eucalyptus miniata E. tetrodonta open forests and savanna woodlands of tropical northern Australia. Final report to Land and Water Australia (Native Vegetation Program). Project TRC-14. Woinarski, J. C. Z., Armstrong, M., Brennan, K., Fisher, A., Griffiths, A. D., Hill, B., Milne, D. J., Palmer, C., Ward, S., Watson, M., Winderlich, S. & Young, S. 2010. Monitoring indicates rapid and severe decline of native small mammals in Kakadu National Park, northern Australia. Wildlife Research, 37, 116-126. Woinarski, J. C. Z., Legge, S., Fitzsimons, J. A., Traill, B. J., Burbidge, A. A., Fisher, A., Firth, R. S. C., Gordon, I. J., Griffiths, A. D., Johnson, C. N., McKenzie, N. L., Palmer, C., Radford, I., Rankmore, B., Ritchie, E. G., Ward, S. & Ziembicki, M. 2011. The disappearing mammal fauna of northern Australia: context, cause, and response. Conservation Letters, 4, 192-201. Woinarski, J. C. Z., Risler, J. & Kean, L. 2004. Response of vegetation and vertebrate fauna to 23 years of fire exclusion in a tropical Eucalyptus open forest, Northern Territory, Australia. Austral Ecology, 29, 156-176. 53

Ziembicki, M. R., Woinarski, J. C., Webb, J. K., Vanderduys, E., Tuft, K., Smith, J., Ritchie, E. G., Reardon, T. B., Radford, I. J., Preece, N., Perry, J., Murphy, B. P., McGregor, H., Legge, S., Leahy, L., Lawes, M. J., Kanowski, J., Johnson, C., James, A., Griffiths, A., Gillespie, G. R., Frank, A., Fisher, A. & Burbidge, A. A. 2014. Stemming the tide: progress towards resolving the causes of decline and implementing management responses for the disappearing mammal fauna of northern Australia. Therya, 6, 169-225. 54

Supporting Information: Appendix S2.1: Detectability correlates of feral cats on Melville Island The model: The following model assumes constant occupancy across Melville Island to investigate whether rainfall, fire, vegetation structure, dingoes, time of year, camera malfunctions or the distance to the nearest site with a cat detection had a significant influence on the detectability of feral cats through the data collection in 2015: Model: occu(~rainfall + Fire activity + Shrub density + Dingoes + Time of year + Number of camera s operating + Distance to nearest cat detection ~ 1) Results: Model: Occupancy: Estimated SE z P(> z ) 1.07 1.09 0.976 0.329 Detection: Estimate SE z P(> z ) Intercept -5.0717 1.894-2.6783 0.0074 Rainfall -0.0623 0.469-0.1328 0.8943 Fire activity 0.0475 0.232 0.2046 0.8379 Shrub density 0.1561 0.216 0.7215 0.4706 Dingoes -0.3100 0.381-0.8131 0.4161 Time of year -0.0365 0.400-0.0911 0.9274 Number of cameras operating 0.1342 0.389 0.3446 0.7304 Distance to nearest cat detection -0.9091 0.409-2.2236 0.0262* Conclusion: There is no evidence in our dataset that the detectability of feral cats was significantly influenced by site-specific habitat or sampling characteristics, however there is evidence of significant spatial autocorrelation. 55

Appendix S2.2: Occupancy and detectability estimates of Conilurus penicillatus on Melville Island Table S1: ΔAIC values for the null model (where occupancy and detectability parameters are assumed to be constant across all survey sites), and the most parsimonious models for Conilurus penicillatus. Estimates of occupancy, nightly detectability, and the probability of overall detection given the average number of camera trap nights are also shown. The naïve occupancy estimate (i.e. the proportion of sites where C. penicillatus was detected) is also shown. Species Model ΔAIC Occupancy (Ψ) (± SE) Nightly detectability (p) (± SE) Probability of overall detection Conilurus penicillatus Naïve - 0.23 - - Null model 88.6 0.23 (0.05) 0.11 (0.01) 0.99 Best model 0.0 0.27 (0.08) 0.06 (0.03) 0.93 56

Chapter 3: Declining populations in one of the last refuges for threatened mammal species in northern Australia Clockwise from top left: Black-footed tree-rat; Common brushtail possum; Northern brush-tailed phascogale 57

Abstract: Australia has contributed a disproportionate number of the world s mammal extinctions over the past 200 years, with the greatest loss of species occurring through the continent s southern and central arid regions. Many taxonomically and ecologically similar species are now undergoing widespread decline across the northern Australian mainland, possibly driven by predation by feral cats and changed fire regimes. Here we report marked recent declines of native mammal species in one of Australia s few remaining areas that support an intact mammal assemblage, Melville Island, the largest island off the northern Australian coast. We have previously reported a marked decline on Melville Island of the threatened brush-tailed rabbit-rat (Conilurus penicillatus) over the period 2000 to 2015, linked to predation by feral cats. We now report a 62% reduction in small mammal trap-success and a 36% reduction in site-level species richness over this period. There was a decrease in trap-success of 90% for the northern brown bandicoot (Isoodon macrourus), 64% for the brush-tailed rabbit-rat and 63% for the black-footed tree-rat (Mesembriomys gouldii), but no decline for the common brushtail possum (Trichosurus vulpecula). These results suggest that populations of native mammals on Melville Island are exhibiting similar patterns of decline to those recorded in Kakadu National Park two decades earlier, and across the northern Australian mainland more generally. Without the implementation of effective management actions, these species are likely to be lost from one of their last remaining strongholds, threatening to increase Australia s already disproportionate contribution to global mammal extinctions. Introduction: Australia has experienced the highest number and proportion of mammal extinctions of any continent over the past two centuries (IUCN, 1996), having lost around 10% of its native mammal species (Short et al., 2002, Woinarski et al., 2014). While most of these mammal extinctions occurred from the mid 19 th to early 20 th Centuries and were concentrated in the southern and arid parts of Australia, over the past three decades severe declines have occurred further north in the monsoonal tropics (Woinarski et al., 2001, Woinarski et al., 2010). Given that the current decline of mammals in northern Australia is 58

most evident in taxa similar to those driven to extinction and severe decline elsewhere in Australia (e.g. bandicoots, large rodents and dasyurids in the critical weight range (CWR) of 35 5500 g) (Burbidge and McKenzie, 1989, Woinarski et al., 2010, Murphy and Davies, 2014), similar factors may be responsible. There is compelling evidence that predation by the introduced red fox (Vulpes vulpes) and feral cat (Felis catus) was the driver of decline and extinction of many mammal species in temperate, arid and semi-arid Australia (Johnson, 2006, Hardman et al., 2016, Short, 2016). Red foxes do not occur in monsoonal Australia, but the feral cat has been present across the northern Australian mainland since the 19 th century (Abbott, 2002, Abbott, 2008). There is accumulating evidence that predation by feral cats is a key factor in the current declines across northern Australia. Recent studies have demonstrated extirpations of some native mammal species on islands recently colonised by cats, but persistence on islands without cats, with this pattern especially strong for mammal species that have declined extensively across mainland areas (Southgate et al., 1996, Woinarski et al., 1999, Woinarski et al., 2011b). An experimental reintroduction of the long-haired rat (Rattus villosissimus) on the northern Australian mainland failed outside predator exclosure areas, due primarily to predation by feral cats (Frank et al., 2014). At another mainland site, Leahy et al. (2016) demonstrated that predation by feral cats and dingoes (Canis dingo) was the primary cause of local population declines of two native mammal species over the 12-month study period. Fire was also a significant factor, but its influence was not through direct fire-related mortality, nor fire-induced reductions in food availability or reproductive success, or emigration. Rather, the influence of fire was related to cat predation, the effects of which were most severe in areas subject to high-intensity fire that removed a greater proportion of vegetation cover (Leahy et al., 2016). The synergistic relationship between fire and predation was further supported by McGregor et al. (2016), who demonstrated that feral cats concentrated their hunting activity to areas recently burnt by high-intensity fires. Despite the accumulating evidence that predation by feral cats is a key factor in the current declines across northern Australia, the apparent asynchrony between the establishment of the feral cat in monsoonal northern Australia and the late 20 th century mammal declines, suggests other factors may be involved. Frequent, high-intensity fires, such as those characterising the fire regime across northern 59

Australia since the breakdown of traditional Aboriginal burning practices, significantly alter the availability of critical resources, including fleshy-fruit bearing shrubs, logs and tree hollows (Russell- Smith et al., 2003b, Vigilante and Bowman, 2004, Firth et al., 2006b, Woinarski and Westaway, 2008). Any species with a strong reliance on such resources may be strongly disadvantaged by current northern Australian fire regimes, even in the absence of predation. However, species that are both dependent on these fire-mediated resources and also susceptible to predation are likely to be particularly threatened, and expected to be the first to exhibit decline and range contraction. Predation by feral cats and frequent high-intensity fire both occur throughout far northern Australia, and therefore are potential threats to areas that currently still retain their full mammalian fauna. One such area is Australia's second-largest island, Melville Island, where the threatened brush-tailed rabbitrat (Conilurus penicillatus) has recently contracted to areas where feral cats were rarely detected and shrub density was high (Davies et al. 2016). Here we build on that study by investigating changes in the broader mammal assemblage on Melville Island, which includes many species that have declined extensively across mainland northern Australia (Firth et al., 2006a, Woinarski et al., 2010, Firth et al., 2010) and several endemic subspecies. We predict that: (1) a range of native mammals on Melville Island, and not just the brush-tailed rabbit-rat, will be in decline; (2) declines will be most evident in those species that have declined most dramatically on the mainland; (3) the current distribution of small mammals on Melville Island will be inversely related to the presence of feral cats and frequent fire. Methods: Study site: Melville Island (5788 km 2 ) is the larger of the two main Tiwi Islands, located ~20 km off the coast of Australia s Northern Territory (Figure 3.1). The Tiwi Islands became separated from mainland Australia relatively recently (between 12,000 and 8,000 years ago) (Woodroffe et al., 1992), and hence have an environment and a mammal assemblage largely comparable to the mainland (with the notable absence of the northern quoll (Dasyurus hallucatus). The islands are of low relief ( 103 m above sea 60

level) and experience a tropical monsoonal climate with an intense wet season (November April) in which over 90% of the annual rainfall occurs. There is a substantial rainfall gradient on Melville Island, from 1400 mm in the east, to 2000 mm in the northwest. The major vegetation types are savanna woodlands and open forests dominated by Eucalyptus miniata, E. tetrodonta and Corymbia nesophila, with a predominantly grassy understorey. Shrub density is highly variable, and studies on the mainland have shown that it is negatively affected by frequent, high-intensity fires (Russell-Smith et al., 2003a, Woinarski et al., 2004). Fire mapping of the Tiwi Islands, has shown that an average of 54% of the savannas were burnt each year from 2000-2013, with 65% of this area burning in the late dry season (Richards et al., 2015). There is currently no evidence to suggest any recent change in fire intensity or frequency, feral animal densities or invasive weeds on the Tiwi Islands (see Woinarski et al. (2001) for further details on introduced species). Over the past two decades, the expansion of the forestry industry across the western half of Melville Island, has seen the replacement of around 30,000 ha of eucalypt tall open forest with short-rotation plantations of exotic Acacia mangium (Woinarski and Tiwi Land Council, 2001). Due to a lack of historical records (Abbott and Burbidge, 1995), the timing of the arrival of feral cats on Melville Island is unknown. While cats could have arrived as far back as the establishment of a British military outpost at Fort Dundas in 1824 (Brocklehurst, 1998), they also could have arrived appreciably later than in other parts of northern Australia (Davies et al. 2016). Data collection: From 2000 2002, small to medium-sized, non-flying mammals were sampled systematically at 351 sites as part of a general wildlife survey of the Tiwi Islands. These surveys involved a 50 50 m quadrat with 20 Elliott traps (33 10 9 cm) spaced equidistantly around the perimeter with one cage trap (56 20 20 cm) located on each of the four corners. Traps were baited with a mixture of peanut butter, oats and honey, and set for three consecutive nights. Four pitfall traps (two 20 L and two 10 L plastic buckets, each with 10 m of 30 cm high drift-line fence) were also used at each site over the 3-day 61

sampling period. Additionally, five 10-minute searches (two at night using spotlights and three during daylight hours) were conducted at each site. Between April and December 2015, 88 of the original 351 sites were revisited (Davies et al. 2016), all located in the savanna woodland and open forests of Melville Island (Figure 3.1). These 88 sites were chosen to capture the large variation in both annual rainfall and fire history on Melville Island. Eightytwo of these sites were surveyed using both live-trapping and camera-trapping, with the remaining six sites being surveyed using only camera-trapping. Live-trapping followed the 2000 2002 protocol but was conducted over four consecutive nights instead of three, and used eight cage traps and 16 Elliott traps. To avoid bias relating to possible seasonal differences in trap-success, re-visited sites were trapped at a comparable time of year to when they were originally surveyed. In 2015, no pitfall traps were used or site searches undertaken. Camera-trapping involved five horizontally facing motion-sensor cameras left continuously recording (24-hours per day) for a minimum 35 consecutive day period that overlapped with each site s live-trapping survey. Camera traps were baited with a mixture of peanut butter, oats and honey. To ensure maximum likelihood of being triggered, each camera was carefully positioned to ensure the bait was in the centre of the field of view (Gillespie et al., 2015).Vegetation within each camera s field of view was cleared to reduce the chance of false triggers and to reduce the risk posed by fire. Of the five cameras deployed at each site, two were Reconyx TM HC550 Hyperfire white flash cameras (Reconyx Inc., Holmen, USA), while the remaining three were Reconyx TM PC800 Hyperfire Professional infra-red flash cameras. All cameras were set to take three image bursts per trigger, with a 1-second delay between images. The sensitivity of each camera was set to high, with cameras re-arming instantly after being triggered. To allow for direct comparison of the live-trapping results, we excluded captures from the original surveys arising from methods not repeated in 2015 (i.e. pitfall traps and site searches). We acknowledge here a potential source of bias arising from the exclusion of the pitfall records. On any given night, a particular animal caught in a pitfall trap is no longer able to be trapped in an Elliott or cage trap. Therefore the number of animals caught in pitfall traps could have influenced the 2000-2002 trapsuccess. We highlight that this source of bias relates only to those species small enough to be caught in 62

pitfall traps (i.e. mice and dunnarts), and note that the number of mammals caught in pitfall traps in 2000-2002 was low and therefore not a large source of bias. As trap-success and species richness do not increase linearly with the number of nights that traps are deployed, we also excluded all captures recorded on the fourth night of live-trapping in 2015. The initial sampling derived an abundance measure from the number of captures and noted the possibility that multiple captures could include the same individual. For consistency we derived trap success the same way. Figure 3.1: Location of the 88 sites surveyed for CWR mammals in 2015 on Melville Island. Filled circles represent the 82 sites where both camera-trapping and live-trapping were conducted, and open circles represent the six sites where only camera-trapping was conducted. The location of Melville Island relative to mainland Australia is shown in the inset. 63

Table 3.1: Description and justification of the variables used in analyses to assess the correlates of Melville Island mammal distribution in 2015. Explanatory variable Fire activity Description and justification for inclusion Following Lawes et al. (2015), a remote-sensed fire variable derived from fine-scale (30 30 m) LANDSAT satellite imagery, representing the proportion of the area surrounding each site that was burnt in each year, averaged over the five years preceding mammal sampling. Calculations were made using an area with a radius of 3.2 km as shown by Lawes et al. (2015) to have the strongest influence on mammal populations. Variable used in analyses to predict: Feral cat activity Mammal occupancy and detectability Rainfall Mean annual rainfall (Australian Bureau of Meteorology 2015). Throughout Australia, feral cat densities tend to be lower in areas of high rainfall (Legge et al., 2016) and mammal species in areas of high rainfall have declined the least (Fisher et al., 2013). Feral cat activity Mammal occupancy and detectability Basal area Dingo activity Shrub density Distance to water Coarse woody debris (CWD) Feral cat activity Julian day Number of cameras operating Calculated as the sum of the basal area (m 2 /ha) of 50 trees (with a diameter at breast height greater than 5 cm) measured within a quadrat 5 m wide with a length equal to the distance to the 50 th measured tree or a maximum of 200 m. Firth et al. 2006a demonstrated the influence of basal area on the occurrence of mammals. The proportion of nights that dingoes were recorded on camera at each site. This was taken as an approximation of dingo activity at each site. Included in analyses to investigate the potential negative influence of dingoes on feral cats and potential benefits for mammal populations (Johnson, 2006, Kennedy et al., 2012). A count of the number of shrubs in a 1 100 m quadrat at each site. Shrubs were defined as anything greater than 20 cm in height with a diameter at breast height of less than 5 cm. Shrubs with multiple stems were counted as a single individual. Vegetation structure has been demonstrated to reduce feral cat hunting success, and therefore influence the distribution of feral cats as well as the occupancy and detectability of mammals (McGregor et al., 2015). A remote-sensed variable measuring the distance (m) from each site to the closest permanent water body. The distance to water was demonstrated by Firth et al. (2006a) to strongly influence a number of mammals on Melville Island. A count of the number of logs with a diameter of greater than 5 cm that crossed a 200 m transect at each site. Included in analyses due to Firth et al. (2006b) demonstrating the reliance of some mammals on fallen logs as den sites. As an index of feral cat activity, we used the predicted probability of detecting feral cats at each site, derived from spatially explicit generalised linear models (Murphy et al., 2010). The probability of feral cat detection was included in the analyses as cats have been implicated as a major factor in the northern mammal decline (Woinarski et al., 2011a, Ziembicki et al., 2014). See Davies et al. (2016). The Julian day of the calendar year that sampling started at each site. Recent work by Geyle (2015) demonstrated that the detectability of the brush-tailed rabbit-rat (Conilurus penicillatus) increases throughout the dry season (May November). An observation level covariate to account for the variation in detectability arising from uneven numbers of cameras operating at different sites due to camera malfunction and destruction. Mammal occupancy and detectability Feral cat activity Mammal occupancy and detectability Feral cat activity Mammal occupancy and detectability Mammal occupancy and detectability Mammal occupancy and detectability Mammal occupancy and detectability Mammal detectability Mammal detectability 64

Data analysis: 1) Trends in trap-success Mammal species >200 g, such as northern brown bandicoot (Isoodon macrourus), common brushtail possum (Trichosurus vulpecula) and black-footed tree-rat (Mesembriomys gouldii), were caught almost exclusively in cage traps, whereas smaller species such as the delicate mouse were exclusively caught in Elliott traps (See Appendix S3.1 in Supporting information). Therefore, prior to investigating changes in trap-success, we first had to account for the different ratio of cage traps to Elliott traps used at each site between sampling years, as this would strongly influence the recorded trap-success in each year. To account for this bias, we derived a species-specific effective trap-success based on the relative effectiveness of each trap type (cage vs. Elliott). This was done for each species that showed a strong bias for either trap type (See Appendix S3.2 in Supporting information). These scaling factors could only be determined from the 2015 live-trapping data as the original data did not consistently record the trap type. For example, although the trap effort for Elliott traps was twice that of cage traps in 2015, Elliott traps accounted for only 2.7% of black-footed tree-rat captures. Therefore the effective trapsuccess in each year for this species was calculated as: Effective trap-success = number of captures (number of cage trap nights + 0.027*number of Elliott trap nights) * 100 Wilcoxon matched-pairs tests were used to investigate changes in trap-success (calculated as the sum of all species effective trap-success at each site), site-level species richness and species-specific effective trap-success at the 82 sites where live-trapping was conducted in both 2000 2002 and 2015. 2) Correlates of mammal distribution in 2015 Davies et al. (2016) investigated a range of environmental correlates of the distribution of the brushtailed rabbit-rat on Melville Island. Here we use the same approach to investigate the 2015 distribution of three other mammal species that were detected by camera traps sufficiently to permit occupancy modelling: the northern brown bandicoot, black-footed tree-rat and common brushtail possum. Singleseason occupancy models were used to investigate how each predictor variable (Table 3.1) influenced 65

site-occupancy. For comparison, we also present results published by Davies et al. (2016) for the brushtailed rabbit-rat. Occupancy modelling was conducted using only the 2015 camera trapping data (88 sites) in the package unmarked in R (Fiske and Chandler, 2011). Explanatory variables were centred and standardised prior to analysis. Due to the many variables and therefore the large number of possible models, occupancy modelling was applied in a two-step process. First we determined which variables best explained the detectability of each species by running all combinations (512 models) of the nine variables we hypothesised might influence the detectability of small mammals. This was done with occupancy constrained to a saturated model of the seven variables we hypothesised might influence site-occupancy for the mammal species. Model selection based on AIC was then used to select the most parsimonious model in the candidate set. The second step involved running all possible combinations of the seven occupancy variables (128 models) with detectability constrained to the most important variables identified in step one. Model selection based on AIC was then used for the second time to determine the best model in the candidate set. Where no single model was clearly superior at explaining the distribution of a species (i.e. ΔAIC <2), we used model averaging to obtain parameter estimates (Burnham and Anderson, 2002). Once the most parsimonious model with only the main effects was identified, we investigated the possible effect of an interaction between the feral cat activity and both fire and shrubs. This was done because processes that simplify vegetation structure (such as frequent fire) might amplify the impact of feral cats. We also tested whether the inclusion of tree basal area increased the model fit. Accounting for imperfect detection provides more realistic, but less precise, estimates of occupancy (Guillera-Arroita et al., 2014). To gauge how accounting for detectability influenced our occupancy estimates and hence the confidence in our conclusions drawn from these models, we also ran all combinations of the occupancy variables but assuming constant detectability. We assessed the fit of the most saturated model for each mammal species with three goodness-of-fit tests based on parametric bootstrapping: Pearson s chi-square statistic, the sum of squared errors and the Freeman-Tukey chi-square statistic. These methods repeatedly simulate datasets based on a fitted 66

model, and then evaluate the probability that the observed history of outcomes has a reasonable chance of happening if the model assessed is assumed to be correct (MacKenzie and Bailey, 2004). Results: 1) Trends in trap-success A total of twelve mammal species were recorded, two of which do not have a mean body size within the CWR: delicate mouse and Butler s dunnart. Overall trap-success was 62% lower in 2015 (6.1 ± 0.8) than in 2000 02 (16.1 ± 1.5; Z = 5.6, p < 0.001). Three of the five species recorded from at least ten sites across the sampling periods, exhibited a significant decrease in trap-success, with northern brown bandicoot decreasing by 90% (Z = 5.42, p < 0.001), brush-tailed rabbit-rat by 64% (Z = 1.97, p < 0.05) and black-footed tree-rat by 63% (Z = 3.33, p < 0.001) (Figure 3.2). The remaining five species (mostly reported from few sites) showed no significant change (Table 3.2). Site-level species richness decreased by 36% between 2000-02 (1.6 ± 0.1) and 2015 (1.0 ± 0.1; Z = 3.82, p < 0.001). The proportion of sites where no native mammals were trapped doubled from 13% to 26%. Six species exhibited a large decrease in naïve occupancy (39 80%), while the common brushtail possum and delicate mouse were recorded at 29% more sites in 2015 than in 2000-02 (Table 3.2). 2) Correlates of mammal distribution in 2015 Given the overall high detectability for all species, the estimated rate of occupancy by the respective best model for each species was very similar to the naïve and null model estimates (See Appendix S3.2 in Supporting information). As such, the effect of each covariate on site-occupancy was similar regardless of whether the models included effects of covariates on detectability or not. Shrub density was a (borderline) significant predictor of site-occupancy by the black-footed tree-rat, as it was for the brush-tailed rabbit-rat, but not for the northern brown bandicoot or common brushtail possum (Figure 3.3). Unlike the situation for the brush-tailed rabbit-rat, feral cat detection was not a significant predictor variable for any of the other species (Figure 3.3). Fire was not a significant 67

Trap-success (% of trap nights) Trap-success (% of trap nights) Trap-success (% of trap nights) Trap-success (% of trap nights) predictor for any species, and neither the inclusion of tree basal area nor an interaction between fire and shrubs with feral cat activity improved the model fit for any species. a) Northern brown bandicoot* b) Black-footed tree-rat* 50 25 40 20 30 15 20 10 10 5 0 8 0 0 5 10 15 20 25 30 35 40 45 0 5 10 15 20 25 30 35 Ranked sites Ranked sites c) Brush-tailed rabbit-rat* d) Common brushtail possum 50 6 40 4 30 20 2 10 0 0 2 4 6 8 10 12 14 16 Ranked sites 0 0 5 10 15 20 25 30 35 Ranked sites Figure 3.2: The difference in live trap-success in 2000-02 (solid line) and 2015 (dashed line) for a) northern brown bandicoot b) black-footed tree-rat c) brush-tailed rabbit-rat and d) common brushtail possum. Asterisks indicate a statistically significant (p < 0.05) change in trap-success. 68

Table 3.2: Summary of the changes in the native mammal populations of Melville Island recorded with live-trapping between 2000 02 and 2015. Naïve occupancy was calculated as the percentage of the 82 live-trapped sites where a species was detected. * = p < 0.05, *** = p < 0.001. Decreases denoted by -, increases denoted by +. Species in bold indicate a body-size outside the critical weight range. Dashes indicate species for which a proportional change in trap-success could not be calculated. Species Northern brown bandicoot (Isoodon macrourus) Black-footed tree-rat (Mesembriomys gouldii) Common brushtail possum (Trichosurus vulpecula) Brush-tailed rabbitrat (Conilurus penicillatus) Grassland melomys (Melomys burtoni) Delicate mouse (Pseudomys delicatulus) Pale field-rat (Rattus tunneyi) Red-cheeked dunnart (Sminthopsis virginiae) Northern sugar glider (Petaurus breviceps) Northern brush-tailed phascogale (Phascogale pirata) Butler's dunnart (Sminthopsis butleri) Western chestnut mouse (Pseudomys nanus) Naïve occupancy 2000-02 (%) Naïve occupancy 2015 (%) Change in naïve occupancy (%) 49 12-75 38 23-39 27 38 + 29 17 9-50 7 4-50 7 12 + 29 6 1-80 5 1-75 2 0-2 0-1 0 - Trapsuccess 2000-02 (%) (±SE) 6.91 (0.99) 3.69 (0.59) 3.78 (0.88) 0.53 (0.15) 0.69 (0.31) 0.20 (0.09) 0.10 (0.05) 0.10 (0.06) 0.03 (0.02) 0.03 (0.02) 0.03 (0.02) 0 2-0 Trapsuccess 2015 (%) (±SE) 0.71 (0.23) 1.38 (0.33) 3.32 (0.73) 0.19 (0.08) 0.10 (0.06) 0.30 (0.10) 0.02 (0.02) 0.03 (0.03) Change in trapsuccess (%) - 90*** - 63*** - 12-64* - 88 + 33-80 - 70 0-0 - 0-0.03 (0.02) - 69

Estimated regression coefficient Estimated regression coefficient Estimated regression coefficient Estimated regression coefficient a) Northern brown bandicoot b) Black-footed tree-rat 2 0-2 -4-6 -8 2 0-2 -4-6 -8 c) Common brushtail possum d) Brush-tailed rabbit-rat 2 0-2 -4-6 -8 2 0-2 -4-6 -8 Figure 3.3: Model averaged regression coefficient estimates for a) the northern brown bandicoot, b) the black-footed tree-rat, c) the common brushtail possum and d) the brush-tailed rabbit-rat (first published in Davies et al. (2016)). Error bars indicate 95% confidence intervals; asterisks indicate where they do not overlap zero, i.e. a statistically significant effect. Data sourced from 2015 camera-trapping. 70

Discussion: Many native mammal species have recently experienced severe range contractions across northern Australia, and Melville Island is one of the few remaining areas to have retained an intact mammal fauna (Woinarski et al. 2010, Ziembicki et al. 2014). However, Davies et al. (2016) demonstrated that the Melville Island population of the brush-tailed rabbit-rat had retracted to areas where feral cats were rarely detected and shrub density was high. Here we build on that study to report evidence of broader decline in the mammal assemblage of Melville Island. We found that trap-success and species richness at the site-level decreased by 62% and 36% respectively from 2000-02 to 2015. As predicted, declines were most evident for three species that have suffered considerable declines on mainland northern Australia: northern brown bandicoot (90% decrease in trap-success), brush-tailed rabbit-rat (64%) and black-footed tree-rat (63%). These severe declines are particularly notable because the local subspecies of black-footed tree-rat (M. gouldii melvillensis) is endemic to the island, and the local subspecies of brush-tailed rabbit-rat (C. penicillatus melibius) is endemic to Melville Island and the adjacent Bathurst Island. However, inconsistent with our prediction, there was no such decline for the common brushtail possum, a comparably-sized species that has also exhibited marked decline on the northern Australian mainland (Woinarski et al., 2010). We acknowledge the limitations associated with inferring mammal decline with only two data points spaced 15 years apart. However, the pattern of decline observed on Melville Island is very similar to that recorded in Kakadu National Park between 2001 and 2009. Over this period in Kakadu, the brushtailed rabbit-rat and black-footed tree-rat were not recorded frequently enough to permit statistical analysis (despite both species being common 30 40 years previously) (Woinarski et al., 2010), and the abundance of the northern brown bandicoot and common brushtail possum decreased by 88 and 86% respectively (Woinarski et al., 2010). This suggests that in Kakadu, the brush-tailed rabbit-rat and blackfooted tree-rat were among the first mammal species to decline. It is therefore plausible that Melville Island is currently experiencing the pattern of decline that occurred in Kakadu National Park over a decade earlier. 71

Unlike the brush-tailed rabbit-rat (Davies et al. 2016) and despite substantial decreases in trap-success, neither the northern brown bandicoot nor black-footed tree-rat appear to have suffered marked range contractions on Melville Island. This suggests that their declines have not yet progressed to the point of influencing their distribution. The lack of change in the occurrence of these species (as opposed to abundance) can explain why we failed to detect relationships between environmental variables, including the presence of feral cats or frequent fire, and site-occupancy. There are a number of potential explanations for why small mammal declines on Melville Island have not progressed to the same extent as on mainland northern Australia. First, Melville Island is a highly productive area of monsoonal northern Australia (Richards et al., 2012), and receives the highest annual rainfall in the Northern Territory. This productivity likely results in high resource availability and high rates of survival and reproduction. As such, mammal populations on Melville Island might not only be more resilient than those on the mainland, but initial population sizes on Melville Island may have been higher. Either of these possibilities may result in a longer period of time being required to cause widespread contractions of species distributions. Second, and also related to the higher rainfall, much of Melville Island has a particularly dense understorey and midstorey of shrubs and small trees (Richards et al., 2012, Davies et al., 2016). As a result, the predation pressure imposed on mammal populations, particularly by feral cats, may be lower on Melville Island compared with the mainland (McGregor et al., 2015, Hohnen et al., 2016, Leahy et al., 2016). This is supported by evidence that site-occupancy by the brush-tailed rabbit-rat and the blackfooted tree-rat on Melville Island is positively related to shrub density (Davies et al., 2016). Third, it is possible that fire intensity is generally lower on Melville Island than on the mainland. Melville Island has a longer wet season and experiences a greater amount of dry-season rain than is the case for the mainland, and so dry fuel for fire is available for a shorter period of time. The amount of flammable grass-layer fuel loads may also be lower on Melville Island due to a higher density of trees and shrubs. As such, any fire-related depletion of resources on Melville Island may not have occurred to the same extent. If the decline of common brushtail possum populations on mainland northern Australia has been primarily driven by fire-driven resource depletion (especially tree 72

hollows)(woinarski and Westaway, 2008), the apparent stability of this species on Melville Island may be related to a more benign fire regime. There is currently no direct evidence demonstrating that fire regimes have been more benign on Melville Island than other areas. Addressing this knowledge gap should be the focus of future research. Finally, it is likely that cats have not been on Melville Island for as long as they have on the mainland (Abbott and Burbidge, 1995, Abbott, 2002, Abbott, 2008), and although there have been some anecdotal reports that cats were introduced to Tiwi islands only within the past few decades (Firth, 2010), definitive evidence of an introduction date is not available. It is also plausible that feral cat populations (and their impact on native mammals) on Melville Island have been suppressed by high dingo densities (Kennedy et al., 2012). Unfortunately, the validity of the above explanations are difficult to evaluate given a lack of relevant information on changes in native mammal populations, feral cat and dingo densities, as well as any changes in fire regimes. However, our finding of widespread declines in an area recently thought to be a refuge for mammals highlights the importance of differentiating between true refuges, where threatening processes are either absent or effectively mitigated, and areas that appear intact simply because declines have been delayed or have occurred relative to a higher initial density level. The latter appears to be the case on Melville Island. This distinction may help prioritise management actions and facilitate timely intervention. These results also have important implications for conservation management. Davies et al. (2016) suggested that fire management that enhances the density of the shrub layer could be a feasible management option to reduce the impact of feral cats on the threatened brushtailed rabbit-rat. Here, we have also demonstrated a significant, positive association between the density of shrubs and the probability of site-occupancy by the black-footed tree-rat. While this association might be due to increased food availability rather than shelter from predation (Friend, 1987), it appears that fire management that maintains a dense understorey could also benefit this species. Small mammal populations have been shown to respond positively to strategic fire management in the Kimberley region of Western Australia (Legge et al., 2011), and a similar approach to fire management could prove beneficial for Melville Island biodiversity. 73

In conclusion, we have shown that the severe population declines that have been documented for small mammals across mainland northern Australia appear to be underway on Melville Island. On the basis of current evidence, we cannot distinguish between a series of potential primary causal mechanisms, and hence cannot yet provide a tight focus for remedial management response. We recommend the following explicit research actions to tease apart the relative impacts of these putative causal factors: (1) quantify the availability of critical small mammal resources across Melville Island and determine how fire frequency and intensity influences their rate of depletion (2) determine how (and why) the density of mammalian predators varies across Melville Island (3) quantify the response of small mammal populations to a range of experimentally manipulated combinations of predation pressure (using fenced exclosures) and fire frequency. In 2001, Woinarski et al. (2001) warned that the mammal fauna of northern Australia may suffer the same fate as the decimated central Australian mammal fauna. Unfortunately, this appears to be coming to fruition, with many species suffering widespread contraction across mainland northern Australia (Woinarski et al. 2010). Here we reiterate this warning with increased urgency and highlight that the consequences of losing these species go beyond the ecological. Traditional food sources (including small mammals) are an important part of the Tiwi diet and provide vital healthy food options for communities. Hunting activities reinforce traditional authority structures, are an important way of passing on traditional knowledge, and form the basis for cultural land management. If Tiwi small mammal populations continue to decline, there will be significant impacts on the expression of Tiwi culture. As such, there is a critical need to improve our understanding of the factors driving these declines, and to implement management actions before these species are lost from one of the last remaining areas in Australia with an intact small mammal fauna. 74

References: Abbott, I. 2002. Origin and spread of the cat, Felis catus, on mainland Australia, with a discussion of the magnitude of its early impact on native fauna. Wildlife Research, 29, 51-74. Abbott, I. 2008. The spread of the cat (Felis catus) in Australia: Re-examination of the current conceptual model with additional information. Conservation Science Western Australia, 7, 1-17. Abbott, I. & Burbudge, A. 1995. The occurrence of mammal species on the islands of Australia: a summary of existing knowledge. CALMScience, 259-324. Brocklehurst, P. 1998. The History and natural Resources of the Tiwi Islands, Northern Territory. Parks and Wildlife Commission of the Northern Territory, Darwin, Australia. Burbidge, A. A. & McKenzie, N. L. 1989. Patterns in the modern decline of western Australia's vertebrate fauna: Causes and conservation implications. Biological Conservation, 50, 143-198. Burnham, K. P. & Anderson, D. R. 2002. Model Selection and Multimodel Inference: A Practical Information-Theoretic Approach. Springer-Verlag, New York. Davies, H. F., McCarthy, M. A., Firth, R. S. C., Woinarski, J., Gillespie, G. R., Andersen, A. N., Geyle, H., Nicholson, E. & Murphy, B. P. 2016. Top-down control of species distributions: feral cats driving the regional extinction of a threatened rodent in northern Australia. Diversity and Distributions, 23, 272-283. Firth, R. S. C. 2010. Population monitoring of a threatened species; the brush-tailed rabbit-rat Conilurus penicillatus in Kakadu National Park and Garig Gunak Barlu National Park (Cobourg Peninsula). Darwin, Australia: EWL Sciences. Firth, R. S. C., Brook, B. W., Woinarski, J. C. Z. & Fordham, D. A. 2010. Decline and likely extinction of a northern Australian native rodent, the Brush-tailed Rabbit-rat Conilurus penicillatus. Biological Conservation, 143, 1193-1201. 75

Firth, R. S. C., Woinarski, J. C. Z., Brennan, K. G. & Hempel, C. 2006a. Environmental relationships of the brush-tailed rabbit-rat, Conilurus penicillatus, and other small mammals on the Tiwi Islands, northern Australia. Journal of Biogeography, 33, 1820-1837. Firth, R. S. C., Woinarski, J. C. Z. & Noske, R. A. 2006b. Home range and den characteristics of the brush-tailed rabbit-rat (Conilurus penicillatus) in the monsoonal tropics of the Northern Territory, Australia. Wildlife Research, 33, 397-407. Fisher, D. O., Johnson, C. N., Lawes, M. J., Fritz, S. A., McCallum, H., Blomberg, S. P., VanDerWal, J., Abbott, B., Frank, A., Legge, S., Letnic, M., Thomas, C. R., Fisher, A., Gordon, I. J. & Kutt, A. 2013. The current decline of tropical marsupials in Australia: is history repeating? Global Ecology and Biogeography, 23, 181-190. Fiske, I. & Chandler, R. 2011. unmarked: An R Package for Fitting Hierarchical Models of Wildlife Occurrence and Abundance. Journal of Statistical Software, 43, 1-23. Frank, A. S. K., Johnson, C. N., Potts, J. M., Fisher, A., Lawes, M. J., Woinarski, J. C. Z., Tuft, K., Radford, I. J., Gordon, I. J., Collis, M. A. & Legge, S. 2014. Experimental evidence that feral cats cause local extirpation of small mammals in Australia's tropical savannas. Journal of Applied Ecology, 51, 1486-1493. Friend, G. 1987. Population ecology of Mesembriomys gouldii (Rodentia: Muridae) in the wet-dry tropics of the Northern Territory. Wildlife Research, 14, 293-303. Geyle, H. 2015. Survey design for detecting declines in the threatened brush-tailed rabbit-rat Conilurus penicillatus on the Tiwi Islands of the Northern Territory. Deakin University, Melbourne. Gillespie, G. R., Brennan, K., Gentles, T., Hill, B., Low Choy, J., Mahney, T., Stevens, A. & Stokeld, D. 2015. A guide for the use of remote cameras for wildlife survey in northern Australia. National Environmental Research Program, Northern Australia Hub, Charles Darwin University. 76

Guillera-Arroita, G., Lahoz-Monfort, J. J., MacKenzie, D. I., Wintle, B. A. & McCarthy, M. A. 2014. Ignoring imperfect detection in biological surveys is dangerous: a response to fitting and interpreting occupancy models'. PLoS ONE, 9, e99571. Hardman, B., Moro, D. & Calver, M. 2016. Direct evidence implicates feral cat predation as the primary cause of failure of a mammal reintroduction programme. Ecological Management & Restoration, 17, 152-158. Hohnen, R., Tuft, K., McGregor, H. W., Legge, S., Radford, I. J. & Johnson, C. N. 2016. Occupancy of the invasive feral cat varies with habitat complexity. PLoS ONE, 11, e0152520. IUCN 1996. 1996 IUCN Red List of Threatened Animals. IUCN, Cambridge, UK. Johnson, C. 2006. Australia's Mammal Extinctions: a 50,000-year History. Cambridge University Press, Melbourne. Kennedy, M., Phillips, B. L., Legge, S., Murphy, S. A. & Faulkner, R. A. 2012. Do dingoes suppress the activity of feral cats in northern Australia? Austral Ecology, 37, 134-139. Lawes, M. J., Murphy, B. P., Fisher, A., Woinarski, J. C., Edwards, A. & Russell-Smith, J. 2015b. Small mammals decline with increasing fire extent in northern Australia: evidence from long-term monitoring in Kakadu National Park. International Journal of Wildland Fire, 24, 712. Leahy, L., Legge, S. M., Tuft, K., McGregor, H. W., Barmuta, L. A., Jones, M. E. & Johnson, C. N. 2016. Amplified predation after fire suppresses rodent populations in Australia s tropical savannas. Wildlife Research, 42, 705-716. Legge, S., Murphy, B., McGregor, H., Woinarski, J., Augusteyn, J., Ballard, G., Baseler, M., Buckmaster, T., Dickman, C. & Doherty, T. 2016. Enumerating a continental-scale threat: How many feral cats are in Australia? Biological Conservation, 206, 293-303. Legge, S., Murphy, S., Kingswood, R., Maher, B. & Swan, D. 2011. EcoFire: restoring the biodiversity values of the Kimberley region by managing fire. Ecological Management & Restoration, 12, 84-92. 77

MacKenzie, D. I. & Bailey, L. L. 2004. Assessing the fit of site-occupancy models. Journal of Agricultural, Biological, and Environmental Statistics, 9, 300-318. 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. McGregor, H. W., Legge, S., Jones, M. E. & Johnson, C. N. 2016. Extraterritorial hunting expeditions to intense fire scars by feral cats. Scientific reports, 6, 22599. Murphy, B. P. & Davies, H. F. 2014. There is a critical weight range for Australia s declining tropical mammals. Global Ecology and Biogeography, 23, 1058-1061. Murphy, B. P., Paron, P., Prior, L. D., Boggs, G. S., Franklin, D. C. & Bowman, D. M. 2010. Using generalized autoregressive error models to understand fire vegetation soil feedbacks in a mulga spinifex landscape mosaic. Journal of Biogeography, 37, 2169-2182. Richards, A. E., Andersen, A. N., Schatz, J. O. N., Eager, R., Dawes, T. Z., Hadden, K., Scheepers, K. & Van Der Geest, M. 2012. Savanna burning, greenhouse gas emissions and indigenous livelihoods: Introducing the Tiwi Carbon Study. Austral Ecology, 37, 712-723. Richards, A. E., Liedloff, A. & Schatz, J. 2015. Tiwi Islands CFI capability project report. CSIRO, Darwin. Russell-Smith, J., Whitehead, P. J., Cook, G. D. & Hoare, J. L. 2003a. Response of Eucalyptusdominated savanna to frequent fires: Lessons from Munmarlary, 1973 1996. Ecological Monographs, 73, 349-375. Russell-Smith, J., Yates, C., Edwards, A., Allan, G. E., Cook, G. D., Cooke, P., Craig, R., Heath, B. & Smith, R. 2003b. Contemporary fire regimes of northern Australia, 1997-2001: change since Aboriginal occupancy, challenges for sustainable management. International Journal of Wildland Fire, 12, 283-297. Short, J. 2016. Predation by feral cats key to the failure of a long-term reintroduction of the western barred bandicoot (Perameles bougainville). Wildlife Research, 43, 38-50. 78

Short, J., Kinnear, J. E. & Robley, A. 2002. Surplus killing by introduced predators in Australia evidence for ineffective anti-predator adaptations in native prey species? Biological Conservation, 103, 283-301. Southgate, R., Palmer, C., Adams, M., Masters, P., Triggs, B. & Woinarski, J. 1996. Population and habitat characteristics of the golden bandicoot (Isoodon auratus) on Marchinbar Island, Northern Territory. Wildlife Research, 23, 647-664. Vigilante, T. & Bowman, D. M. J. S. 2004. Effects of individual fire events on the flower production of fruit-bearing tree species, with reference to Aboriginal people's management and use, at Kalumburu, North Kimberley, Australia. Australian Journal of Botany, 52, 405-416. Woinarski, J., Burbidge, A. & Harrison, P. 2014. Action Plan for Australian Mammals 2012. CSIRO Publishing, Melbourne. Woinarski, J. & Westaway, J. 2008. Hollow formation in the Eucalyptus miniata E. tetrodonta open forests and savanna woodlands of tropical northern Australia. Final report to Land and Water Australia (Native Vegetation Program). Project TRC-14. Woinarski, J. C. Z., Armstrong, M., Brennan, K., Fisher, A., Griffiths, A. D., Hill, B., Milne, D. J., Palmer, C., Ward, S., Watson, M., Winderlich, S. & Young, S. 2010. Monitoring indicates rapid and severe decline of native small mammals in Kakadu National Park, northern Australia. Wildlife Research, 37, 116-126. Woinarski, J. C. Z., Legge, S., Fitzsimons, J. A., Traill, B. J., Burbidge, A. A., Fisher, A., Firth, R. S. C., Gordon, I. J., Griffiths, A. D., Johnson, C. N., McKenzie, N. L., Palmer, C., Radford, I., Rankmore, B., Ritchie, E. G., Ward, S. & Ziembicki, M. 2011. The disappearing mammal fauna of northern Australia: context, cause, and response. Conservation Letters, 4, 192-201. Woinarski, J. C. Z., Milne, D. J. & Wanganeen, G. 2001. Changes in mammal populations in relatively intact landscapes of Kakadu National Park, Northern Territory, Australia. Austral Ecology, 26, 360-370. 79

Woinarski, J. C. Z., Palmer, C., Fisher, A., Brennan, K., Southgate, R. & Masters, P. 1999. Distributional patterning of mammals on the Wessel and English Company Islands, Arnhem Land, Northern Territory, Australia. Australian Journal of Zoology, 47, 87-111. Woinarski, J. C. Z., Risler, J. & Kean, L. 2004. Response of vegetation and vertebrate fauna to 23 years of fire exclusion in a tropical Eucalyptus open forest, Northern Territory, Australia. Austral Ecology, 29, 156-176. Woinarski, J. C. Z. & Tiwi Land Council 2001. Biodiversity Conservation on the Tiwi Islands: Plants, Vegetation Types and Terrestrial Vertebrates on Melville Island: a Report to the Tiwi Land Council. Parks and Wildlife Commission of the Northern Territory, Darwin. Woinarski, J. C. Z., Ward, S., Mahney, T., Bradley, J., Brennan, K., Ziembicki, M. & Fisher, A. 2011b. The mammal fauna of the Sir Edward Pellew island group, Northern Territory, Australia: refuge and death-trap. Wildlife Research, 38, 307-322. Woodroffe, C. D., Bryant, E. A., Price, D. M. & Short, S. A. 1992. Quaternary inheritance of coastal landforms, Cobourg Peninsula, Northern Territory. Australian Geographer, 23, 101-115. Ziembicki, M. R., Woinarski, J. C., Webb, J. K., Vanderduys, E., Tuft, K., Smith, J., Ritchie, E. G., Reardon, T. B., Radford, I. J., Preece, N., Perry, J., Murphy, B. P., McGregor, H., Legge, S., Leahy, L., Lawes, M. J., Kanowski, J., Johnson, C., James, A., Griffiths, A., Gillespie, G. R., Frank, A., Fisher, A. & Burbidge, A. A. 2014. Stemming the tide: progress towards resolving the causes of decline and implementing management responses for the disappearing mammal fauna of northern Australia. Therya, 6, 169-225. 80

Supporting information: Appendix S3.1: Shows the number of individuals of each species that were caught in each trap type in 2015. Species in bold are those that showed significant a bias for a specific trap type. Species # trapped in # trapped in cage traps Elliott traps Brush-tailed rabbit-rat 7 6 Northern brown bandicoot 21 0 Grassland melomys 1 6 Black-footed tree-rat 36 1 Delicate mouse 0 15 Western chestnut mouse 1 1 Pale field rat 2 0 Red cheeked dunnart 1 1 Common brushtail possum 76 2 81

Appendix S3.2: ΔAIC values for the null model (where occupancy and detectability parameters are assumed to be constant across all survey sites), and the most parsimonious models for each small mammal species. Estimates of occupancy, nightly probability of detection, and the overall probability of detection (i.e. over 43 nights) are also shown. The naïve occupancy estimate (i.e. the proportion of sites where each species was detected) is also shown. Data sourced from 2015 camera-trapping. Species Model ΔAIC Occupancy (Ψ) (± SE) Nightly probability of detection (p) (± SE) Overall probability of detection Brush-tailed rabbit-rat Naïve - 0.23 - - Null model 88.6 0.23 (0.05) 0.11 (0.01) 0.99 Best model 0.0 0.27 (0.08) 0.06 (0.03) 0.93 Black-footed tree-rat Northern brown bandicoot Common brushtail possum Naïve - 0.64 - - Null model 43.7 0.64 (0.05) 0.20 (0.01) 0.99 Best model 0.0 0.64 (0.08) 0.19 (0.02) 0.99 Naïve - 0.90 - - Null model 76.6 0.90 (0.03) 0.20 (0.01) 0.99 Best model 0.0 0.90 (0.04) 0.21 (0.02) 0.99 Naïve - 0.92 - - Null model 442.4 0.92 (0.03) 0.29 (0.01) 0.99 Best model 0.0 0.92 (0.04) 0.29 (0.02) 0.99 82

Chapter 4: An experimental test of whether pyrodiversity promotes mammal diversity in a northern Australian savanna Clockwise from top left: A low intensity prescribed burn; Smoke haze in tall savanna; Log being consumed by fire. 83

Abstract: 1. The increasing awareness that a fire regime that promotes biodiversity in one system can threaten biodiversity in another has resulted in a shift away from fire management based on vague notions of maximising pyrodiversity, towards determining the optimal fire regime based on the demonstrated requirements of target species. 2. Here, we utilised a long-running, replicated fire experiment on Melville Island, the largest island off the northern Australian coast, to test the importance of pyrodiversity for native mammals in a northern Australian savanna landscape. We first developed statistical models to determine how native mammal abundance has responded to nine years of experimentallymanipulated fire frequency. Next, given each species modelled response to fire frequency, we identified the level of pyrodiversity and optimal mix of fire frequencies that would be expected to maximise mammal diversity and abundance, and minimise extinction risk, This was done for both the entire mammal assemblage and for the mammal species currently declining on Melville Island. 3. Fire frequency was a significant predictor of abundance of the northern brown bandicoot (Isoodon macrourus), black-footed tree-rat (Mesembriomys gouldii), brush-tailed rabbit-rat (Conilurus penicillatus), grassland melomys (Melomys burtoni), pale field-rat (Rattus tunneyi), and mice/dunnarts but not of the common brushtail possum (Trichosurus vulpecula). 4. The geometric mean abundance (GMA) of the entire mammal assemblage was positively associated with pyrodiversity, but peaked at an intermediate value. Hence, maximising pyrodiversity would reduce native mammal assemblage GMA below its potential maximum. The fire history for an area that maximised the entire native mammal assemblage GMA consisted of 57% long-unburnt, 43% triennially burnt and <1% annually burnt. Pyrodiversity did not reduce the extinction risk, nor increase the GMA of declining mammals above that predicted in areas entirely annually or triennially burnt. 5. Synthesis and applications: We demonstrate a useful approach with which to develop fire management strategies based on the demonstrated requirements of target species. By comparing 84

the optimal fire regime identified for the conservation of threatened species and that identified for the entire mammal assemblage, we demonstrate the flexibility of this approach to tailor fire management to address specific management priorities in other fire-prone environments. Keywords: extinction risk, fire experiment, Melville Island, northern Australia, pyrodiversity, tropical savanna, native mammal diversity. Introduction: Fire is a global ecosystem driver (Bond et al., 2005, Bowman et al., 2009, Kelly and Brotons, 2017), with profound effects on the evolution of biological communities and ecological processes (Gill et al., 1981, Whelan, 1995, Bond and Van Wilgen, 1996). As a result, fire plays an integral part in the functioning of some biomes (Bowman et al., 2011). The clearing of vegetation, livestock grazing, introduction of exotic plants and animals, alteration of ignition sources and patterns, and the active suppression of fires associated with the expansion of human society has disrupted fire regimes on a global scale, causing substantial ecosystem change and biodiversity loss (Bowman et al., 2011). Worryingly, the disruption of fire regimes is likely to become exacerbated by global climate change (Bowman et al., 2009). Fire also represents one of the few tools for conservation management at the landscape scale. In many fire-prone environments, fire management follows the patch mosaic burning paradigm, which attempts to establish and maintain a fine-scale, heterogeneous mosaic of varying fire histories (Parr and Andersen, 2006), under the assumption that pyrodiversity begets biodiversity (Martin and Sapsis, 1992). However, this approach lacks a solid empirical or theoretical basis, and has also been criticised for lacking clear operational guidelines that specify which aspects of pyrodiversity should be maximised (Parr and Andersen, 2006, Taylor et al., 2012). This uncertainty has recently stimulated critical analyses of the relationship between pyrodiversity and biodiversity (Parr and Andersen, 2006, Taylor et al., 2012, Kelly et al., 2016). For example, while Tingley et al. (2016) demonstrated that pyrodiversity promotes bird diversity in Californian conifer forests, Taylor et al. (2012) found no such pattern in a semi-arid region of south-eastern Australia, and suggested that 85

burning for fire-mediated heterogeneity could actually threaten the avian fauna in this system. The validity of patch-mosaic burning has also been questioned for reptiles (Nimmo et al., 2013) and mammals (Kelly et al., 2012) in semi-arid Australia. These inconsistent results highlight the contextspecificity of pyrodiversity requirements, and the risks associated with the blanket application of a management paradigm focused on maximising pyrodiversity. Hence, fire management for biodiversity conservation must be directly underpinned by the demonstrated fire requirements of the target species (Andersen et al., 2014, Swan et al., 2015, Kelly et al., 2016). Fire has shaped Australian landscapes for millions of years (Kershaw et al., 2002, Crisp et al., 2011), and has been managed by humans for up to 50,000 years of Aboriginal history. The breakdown of traditional burning practices has been implicated in the decline of a range of taxa across northern Australia, including the native cypress pine (Callitris intratropica) (Bowman and Panton, 1993), granivorous birds (Franklin, 1999), and more recently the catastrophic collapse of native mammal populations (Firth et al., 2010, Woinarski et al., 2011, Ziembicki et al., 2014). In an attempt to mitigate these declines, prescribed management fires are widely applied across northern Australian savannas (Andersen et al., 2005). Creating a fine-scale fire mosaic and increasing the amount of long-unburnt vegetation are often key objectives of fire management for biodiversity conservation in northern Australia (Woinarski and Winderlich, 2014). However, with limited and sometimes conflicting empirical data relating Australian biota to specific fire patterns, this approach may be ineffective, and even potentially threaten important biodiversity values. Here, we utilise a long-running (9-year) fire experiment on Melville Island (Richards et al., 2012) to evaluate the relationship between pyrodiversity and native mammal diversity in a northern Australian savanna. We first examine the relationship between experimental fire treatments and mammal diversity and abundance. We then use specific estimates of species abundances in each fire treatment to investigate the relationship between simulated pyrodiversity and mammal diversity and extinction risk. In doing so, we identify the specific proportions of different fire treatments within a simulated landscape that maximises native mammal diversity and minimises extinction risk. 86

Methods: Study site: Melville Island is Australia's second largest island (5788 km 2 ), and the larger of the two main Tiwi Islands, located 20 km off the coast of Australia s Northern Territory (Fig. 4.1). The islands experience a tropical monsoonal climate, with a wet season (November April) in which over 90% of the annual rainfall occurs (Australian Bureau of Meteorology, 2015). There is a substantial annual rainfall gradient on Melville Island, from 1400 mm in the east, to 2000 mm in the northwest. The major vegetation types are savanna woodlands and open forests dominated by Eucalyptus miniata, E. tetrodonta and Corymbia nesophila, with a predominantly grassy understorey. Shrub density is highly variable, and studies on the mainland have shown that it is negatively affected by frequent, high-intensity fires (Russell-Smith et al., 2003, Woinarski et al., 2004). Fire mapping of the Tiwi Islands from 2000 2013 has shown that an average of 54% of the savannas were burnt each year, with 65% of this area burning in the late dry season when fire intensity tends to be highest (Richards et al., 2015). Despite no evidence of recent changes in fire intensity or frequency, feral animal densities or invasive weeds on the Tiwi Islands, native mammal populations appear to have declined significantly over the past 15 years, especially the brush-tailed rabbit-rat (Conilurus penicillatus), black-footed tree-rat (Mesembriomys gouldii) and northern brown bandicoot (Isoodon macrourus) (Davies et al., 2016, H. Davies et al., 2018). Data collection: In 2008, 18 experimental fire plots (each 50 100 ha) were established at four locations across the western half of Melville Island as part of the Tiwi Carbon Study (see Richards et al. 2012 for details) (Fig. 4.1). The experiment tests three contrasting fire regimes: (1) annual burning in the early dry season; (2) triennial burning in the early dry season; and (3) no burning (henceforth referred to as long- 87

unburnt). Each treatment is replicated six times, with one replicate of each treatment at Imalu, Taracumbi, and Pickertaramoor, and three replicates of each treatment at Shark Bay (Fig. 4.1). Prior to the experiment, all plots were burnt at intervals of 1 3 years (Richards et al., 2012). The mean Byram fire-line intensities of the annual and triennial experimental burns were 650 and 1850 kw m -1 respectively (A. Richards, unpublished data). In December 2013, four vertically downward-facing motion-sensor camera traps were established in each of the 18 fire plots (72 cameras in total) to monitor native mammals. These cameras were left operating continuously for two years, being removed during this period only when plots were burnt, with cameras re-deployed as soon as possible after burning. All cameras were Reconyx TM PC800 Hyperfire Professional cameras, with infra-red flash (Reconyx Inc., Holmen, USA). Cameras were deactivated between 8:00 and 18:00 daily, and were programmed to take ten image bursts per trigger. The sensitivity of each camera was set to high, with cameras re-arming five minutes after being triggered. Figure 4.1: Locations of the 18 experimental fire plots on Melville Island, northern Australia. The black, grey and white rectangles represent the six annually burnt, triennially burnt and long-unburnt plots, respectively. The location of Melville Island relative to mainland Australia is shown in the inset. 88

Data analysis: To investigate changes in native mammal populations in response to the experimental manipulation of fire frequency, we first derived response variables from camera images to characterise native mammal diversity and abundance in each of the 18 fire plots. Diversity was expressed as the mean number of species recorded per 100 trap nights, and abundance of each native mammal species was calculated as the proportion of total camera trap-nights the species was recorded. There is a positive relationship between the number of individuals of a species recorded at a site in live-traps and the proportion of nights the species was recorded on cameras traps on Melville Island (See Appendix S4.1 in Supporting information). We calculated this abundance metric for: the northern brown bandicoot, black-footed treerat, common brushtail possum (Trichosurus vulpecula), brush-tailed rabbit-rat, grassland melomys (Melomys burtoni) and pale field-rat (Rattus tunneyi). Species smaller than 50 g (native mice [Pseudomys spp.] and dunnarts [Sminthopsis spp.]) could not be reliably identified, and were combined as a single group referred to as mice/dunnarts. We note that this group could comprise of up to four individual species: delicate mouse (Pseudomys delicatulus), western-chestnut mouse (Pseudomys nanus), red-cheeked dunnart (Sminthopsis virginiae), Butler s dunnart (Sminthopsis butleri). The northern brush-tailed phascogale (Phascogale pirata) and northern sugar glider (Petaurus breviceps) were also recorded, but too infrequently for meaningful analysis of abundance. We used generalised linear models (GLMs) to investigate how native mammal diversity and abundance, as well as the abundance of each species, was related to fire frequency. To do this we compared four models: a null model, two separate models containing the single main effects of fire treatment and site, and a model containing both fire treatment and site. While time since fire inevitably fluctuated over the period of data collection, we consider our focus on fire frequency as warranted as it characterises a fire regime, not a single fire event. We included the site variable to account for spatial autocorrelation. Given the small sample size (n=18), site-specific variation in fire response could not be assessed. Model selection was based on a robust form of Akaike s Information Criterion, AIC c, an index that favours both model fit and model simplicity (Burnham and Anderson, 2002). Models with 89

lower values of AIC c are considered to have greater support relative to other models in the candidate set. Using the respective best model based on AIC c, we predicted the abundance of each species in each of the three fire treatments. Obtaining species-specific estimates of abundance in each fire treatment allowed us to explore the relationship between mammal diversity and simulated pyrodiversity. To do this, we generated 5000 hypothetical, spatially-dimensionless sites with varying levels of pyrodiversity. We randomly varied the proportion of each fire treatment (annually burnt, triennially burnt and longunburnt) at each of these sites, such that the sum of the proportions of the three fire treatments was equal to 1. Following studies of birds, mammals and reptiles in semi-arid southern Australia (Taylor et al., 2012, Kelly et al., 2012, Nimmo et al., 2013, Farnsworth et al., 2014), we calculated pyrodiversity using the Shannon s diversity index, scaled such that the maximum (i.e. a site with equal proportions of annually burnt, triennially burnt and long-unburnt vegetation) was equal to 1. As used here, Shannon s diversity (pyrodiversity) is maximised when fire treatments are equally represented at the simulated site, and this would hypothetically lead to a maximisation of overall biodiversity values only if each treatment had equal biodiversity value and we acknowledge here that this is a much simplified way in which to quantify pyrodiversity. We used the species-specific estimates of abundance (derived from GLMs) in each fire treatment to obtain an abundance estimate for each species at each simulated site and then calculated the geometric mean abundance (GMA) for each site as equation 1: n GMA = Pj 1 Pj 2 Pj 3 Pj n, eqn 1 where n is the number of species, and Pj is the predicted abundance for each species given the simulated fire history. GMA is an appropriate biodiversity index, having heuristic properties that capture a range of desirable criteria with which to assess biodiversity (Buckland et al., 2011, Van Strien et al., 2012, McCarthy et al., 2014). We scaled predicted GMA so the maximum value was equal to 1, and related it to each site s pyrodiversity index. A useful property of GMA is that it tends to be correlated with the proportion of species within an area that are likely to become extinct (McCarthy et al., 2014). Following Giljohann et al. (2015), we utilised 90

this relationship and quantified the change in extinction risk (ER) resulting from different levels of pyrodiversity at each site as equation 2: ER = 1 ( 1 ( ln (GMA) ) 1 ln (GMAmax) ) eqn 2 We then plotted the values of both GMA and ER against each site s pyrodiversity index. This was done separately for both the entire assemblage of mammals (seven species) and then for only the three species in significant decline on Melville Island (northern brown bandicoot, black-footed tree-rat, brush-tailed rabbit-rat) (Chapter 3). By doing so, we identified the specific proportions of the different fire treatments that maximised native mammal diversity and minimised the extinction risk. We conducted a sensitivity analysis to investigate how the identified optimal fire regime was influenced by the variability surrounding each species abundance estimate. This was done by randomly selecting values from the distribution surrounding each species predicted abundance and identifying the level of pyrodiversity that maximised the GMA of the entire native mammal assemblage. This was replicated 10,000 times. Results: The best model of diversity contained the term fire treatment, but this model was not significantly better (<2 AIC c units) than the similar model without this term (Table 4.1). Therefore, fire treatment had no clear effect on native mammal diversity. However, species composition clearly varied between fire treatments. For all seven species, the best model included the term fire treatment, and for all except the brushtail possum, this model was significantly better ( 2 AIC c units) than a similar model without the term fire treatment (Table 4.1) (for a comparison between the effect size of fire treatment and site on the predicted abundance of each species, see Appendix. S4.2 in Supporting information). Species varied in their preferred fire treatment: the abundances of the black-footed tree-rat, brush-tailed rabbit-rat and mice/dunnarts were highest in annually burnt plots; the abundance of the northern brown bandicoot and pale field-rat was highest in triennially burnt plots; and the abundance of the grassland 91

melomys was highest in long-unburnt plots (Figure 4. 2). Importantly, due to the logistical constraints of conducting a highly-replicated fire experiment, some of these estimates were uncertain. Table 4.1: A summary of the model selection process for native mammal alpha diversity and speciesspecific abundance in 18 experimental fire plots across Melville Island, Northern Territory. K indicates the number of parameters; ΔAIC c is the difference between the model s AIC c value and the minimum AIC c value in the candidate set; w i is the Akaike weight (the likelihood of the model being the best in the candidate set). Bold text indicates significant effect of fire treatment (ΔAIC c 2, relative to other models in the candidate set). Models with essentially no empirical support (ΔAIC c > 10) are not included in the table. Response Model K ΔAICc w i Native mammal alpha diversity ~ Fire treatment + Site 8 0.0 0.60 ~ Site 6 1.0 0.36 ~ Null model 1 6.0 0.03 ~ Fire treatment 3 8.5 0.01 Native mammal abundance ~ Fire treatment + Site 8 0.0 0.57 ~ Site 6 0.6 0.43 Northern brown bandicoot abundance ~ Fire treatment + Site 8 0.0 1.00 Common brushtail possum abundance ~ Fire treatment + Site 8 0.0 0.67 ~ Site 6 1.5 0.33 Black-footed tree-rat abundance ~ Fire treatment + Site 8 0.0 1.00 Brush-tailed rabbit-rat abundance ~ Fire treatment + Site 8 0.0 1.00 Grassland melomys abundance ~ Fire treatment + Site 8 0.0 1.00 Pale field-rat abundance ~ Fire treatment + Site 8 0.0 0.97 ~ Site 6 7.1 0.03 Mice/dunnart abundance ~ Fire treatment + Site 8 0.0 1.00 The geometric mean abundance (GMA) of the entire mammal assemblage was positively associated with pyrodiversity, but peaked at intermediate pyrodiversity (63% of maximum possible pyrodiversity) (Figure 4.3a). The fire history of an area that maximised native mammal GMA (i.e. GMA of 100%) consisted of 57% long-unburnt, 43% triennially burnt and <1% annually burnt vegetation. GMA was 94% for sites composed of entirely long-unburnt vegetation, 83% for entirely triennially burnt sites and 67% for entirely annually burnt sites (Figure 4.3a). Mammal GMA at the highest possible simulated pyrodiversity (i.e. equal proportions annually, triennially and long-unburnt) was 95%. Having sites consisting of entirely long-unburnt, triennially burnt or annually burnt vegetation (i.e. those with the lowest possible level of pyrodiversity) increased the average extinction risk by 1.5, 4.0 and 8.0%, 92

respectively (Figure 4.3a). Maximum pyrodiversity was associated with a 1.0% increase in extinction risk. Given species-specific fire preferences (Figure 4.2), each simulated fire management scenario represented a trade-off between the abundance of species. The predicted abundance of each species resulting from different approaches to fire management are summarised in Table 4.2. Table 4.2: Predicted abundance of each native mammal species resulting from different approaches to fire management. Approach to fire management Northern brown bandicoot Common brushtail possum Black-footed tree-rat Brush-tailed rabbit-rat Grassland melomys Pale field-rat Entirely annually burnt 2.52 2.94 3.62 1.96 0.13 0.01 1.35 Entirely triennially burnt 4.30 2.63 2.67 1.58 0.10 0.08 0.72 Entirely unburnt 2.61 2.44 2.97 1.11 1.01 0.02 1.29 Maximum pyrodiversity 3.15 2.67 3.09 1.55 0.43 0.04 1.12 Maximum GMA/ Minimum extinction risk 3.33 2.52 2.84 1.31 0.65 0.05 1.04 Mice/ dunnarts In marked contrast, when considering only those mammals which are currently declining on Melville Island, there was little evidence of a positive association between pyrodiversity and GMA (Figure 4.3b). GMA again peaked at 63% of the maximum possible value of pyrodiversity. However, the optimal balance of fire histories was markedly different compared with that for all mammals. The optimal regime consisted of no unburnt vegetation, and near equal parts annually burnt (51%) and triennially burnt (49%) (Figure 4.3b). Native mammal GMA was 99% for sites composed of entirely triennially burnt vegetation, 98% for entirely annually burnt sites and 77% for entirely long-unburnt sites (Figure 4.3b). Having a site consisting of entirely triennially burnt or annually burnt vegetation would increase the average extinction risk of the three declining species by less than 1%, while a site consisting of entirely long-unburnt would increase their average extinction risk by 7% (Figure 4.3b). Maximum pyrodiversity was associated with a 2.0% increase in the extinction risk for those species currently declining on Melville Island. Multiple combinations of the three fire treatments resulted in the same level of pyrodiversity, but varied in their predicted GMA and extinction risk i.e. those sites along the dotted black lines (Figure 4.3). For example, for the entire mammal assemblage, sites with a pyrodiversity value of 63% varied in their predicted GMA by 19% (representing a 4% range in extinction risk). This variability stemmed from the relative dominance of the three fire treatments at each simulated site, with an increasing proportion of 93

annually burnt resulting in lower GMA and higher extinction risk. For those species currently declining on Melville Island, an increasing proportion of long-unburnt resulted in lower GMA and higher extinction risk. From 10,000 replications of our sensitivity analysis, the level of pyrodiversity that corresponded to maximum native mammal GMA averaged 60%, with a 95% confidence interval of 13 98%. 94

Abundance Abundance Abundance Abundance Abundance Abundance Abundance 5 Northern brown bandicoot: 5 Common brushtail possum: 4 4 3 3 2 2 1 1 0 Annual Triennial Unburnt 0 Annual Triennial Unburnt 5 Black-footed tree-rat: 3 Brush-tailed rabbit-rat: 4 3 2 1 2 1 0 Annual Triennial Unburnt 0 Annual Triennial Unburnt 2 Grassland melomys: 0.15 Pale field-rat: 1 0.1 0.05 0 Annual Triennial Unburnt 0 Annual Triennial Unburnt Mice/dunnarts: 2 1.5 1 0.5 0 Annual Triennial Unburnt Figure 4.2: The predicted index of abundance (±SE) of native mammals for each experimental fire treatment. 95

a) Relative change in the index proportional to the extinction risk Geometric mean abundance 1 0.8 0.6 0.4 0.2 0 0 0.2 0.4 0.6 Pyrodiversity 0.8 Annual Triennial Unburnt 0.08 0.06 0.04 0.02 0 1 0 0.2 0 0.2 0.4 0.6 Pyrodiversity 0.8 1 b) Relative change in the index proportional to the extinction risk Geometric mean abundance 1 0.8 0.6 0.4 0.2 0 0 0.2 0.4 0.6 Pyrodiversity 0.8 1 0.08 0.06 0.04 0.02 0 0.4 0.6 0.8 1 Pyrodiversity Figure 4.3: The relationship between pyrodiversity and the geometric mean abundance (GMA) and the relative change in the index proportional to the extinction risk for (a) all ground-active native mammal species; (b) the three declining native mammal species, at 5000 simulated sites. The square, triangle and diamond intercept markers indicate the predicted mammal diversity and extinction risk at sites composed of entirely annually burnt, triennially burnt and long-unburnt vegetation, respectively. The dotted vertical lines indicate the level of pyrodiversity with the maximum predicted mammal GMA and minimum change in extinction risk. Discussion: Globally, there is a pressing need to identify fire regimes that maximise biodiversity. However, a fire regime that maximises biodiversity in one system may not necessarily do so in another (Farnsworth et al., 2014). As a result, fire management applied without context-specific empirical evidence may not only be a waste of resources but potentially threaten biodiversity (Taylor et al., 2012). In northern Australian savannas, determining the fire regime that most strongly promotes native mammal diversity is a focus of much research, as fire management is widely advocated as a key tool to prevent further 96

population declines (Woinarski and Winderlich, 2014, Davies et al., 2016). By first demonstrating the varied response of individual mammal species to experimentally-manipulated fire frequency, we have provided the first empirical evidence for the relationship between pyrodiversity and both native mammal gamma diversity, and extinction risk, in a northern Australian savanna. We found that predicted mammal diversity (based on geometric mean abundance; GMA) peaked at an intermediate level of pyrodiversity. Hence, maximising pyrodiversity would actually reduce native mammal GMA (albeit slightly) below its potential maximum. This is a similar result to that found for a range of bird and mammal species in the semi-arid Mallee region of south-eastern Australia (Kelly et al., 2012, Taylor et al., 2012, Farnsworth et al., 2014), which have particular requirements for longunburnt vegetation, rather than heterogeneous fire. Our results suggest that the optimal fire regime for the entire mammal assemblage would be dominated (>50%) by long-unburnt habitat, highlighting the importance of long-unburnt vegetation to maintain mammal diversity in northern Australian savannas. We demonstrate that the predicted mammal diversity and extinction risk at a particular pyrodiversity value depends on the relative dominance of each of the three fire treatments, suggesting that the initial fire state of an area dictates the potential benefits of applying fire management. Importantly, maximum pyrodiversity was associated with a higher predicted diversity of mammals (and lower extinction risk) than the lowest pyrodiversity (i.e. those sites entirely annually, triennially or unburnt). Hence, while pyrodiversity is clearly an important element of a fire regime that can help maintain high diversity, other components of the fire regime may be particularly important e.g. the proportion of the landscape in a particular desired (or undesired) state. Despite the majority of species exhibiting the highest abundance in the annually burnt fire plots, we identified the optimal fire regime for total native mammal diversity as an area composed of mostly longunburnt and triennially burnt vegetation. While seemingly counter-intuitive, this reflects the mathematical properties of the index we used to quantify native mammal diversity: the geometric mean abundance (GMA). GMA is a useful measure of biodiversity and reflects both evenness and abundance; however, as it works on the multiplicative scale, GMA is most sensitive to changes in the rarest species (Buckland et al., 2011). As the pale field-rat and grassland melomys were infrequently detected 97

compared to the other species, their habitat preferences for triennially burnt and long-unburnt vegetation were particularly influential, thus explaining the importance of these fire regimes for maintaining overall mammal assemblage diversity. Investigating the relationship between pyrodiversity and the diversity and extinction risk of the entire mammal assemblage allowed us to test the relevance of the pyrodiversity hypothesis for northern Australian mammals. However, conservation efforts are often focused on a small subset of species, which are either threatened or locally declining (Drummond et al., 2010). For any given area, there will be an optimal fire regime that maximises diversity and minimises overall biodiversity loss (Richards et al., 1999). However, even if the optimal fire regime is known, it will often not be feasible to implement due to resource constraints. As such, managers must balance alternative management strategies, each associated with different costs and benefits. As a result, management objectives targeted towards the conservation of threatened species are often prioritised over more common species. For example, the optimal fire regime identified here for the entire mammal assemblage suggested the optimal fire regime would be dominated (>50%) by long-unburnt habitat. Not only was this relationship primarily driven by the habitat requirements of two relatively stable mammal species (grassland melomys and pale field-rat) (IUCN, 1996, Woinarski et al., 2014), but it represents an impracticable target to achieve in the highly fire-prone mesic savannas of northern Australia, where on average over 50% of the landscape burns each year. As such, from a manager s perspective, this approach would be undesirable as it would involve a significant investment of resources for minimal benefit to species of conservation concern. The optimal fire regime identified for declining mammal species was markedly different, and more feasible, to that which would maximise the diversity of the entire mammal assemblage. Increasing pyrodiversity did not drastically reduce the extinction risk, nor increase the GMA of declining mammals above that predicted in areas entirely annually or triennially burnt. However, increasing pyrodiversity in suboptimal areas for these declining mammals (i.e. areas dominated by unburnt habitat) could result in a 7% reduction in extinction risk. Targeting certain areas in the landscape with an empirically based objective highlights how this method might be utilised by managers in the spatial prioritisation of prescribed fire. We acknowledge the 98

limitation that our fire experiment lacked a spatial component. As such, these results may only be relevant for prescribed fire applied at a similar spatial scale i.e. patches 50 100 ha. Importantly, these results demonstrate that once the species-specific fire requirements in any particular area are established, this approach can be utilised to optimise fire management to achieve explicit management priorities, such as the conservation of threatened species. Compared with invertebrates, there has been limited research relating savanna mammal diversity to pyrodiversity (Briani et al., 2004, Griffiths et al., 2015). While our results concur with those of Maravalhas and Vasconcelos (2014), who demonstrated the importance of pyrodiversity for ant diversity in Brazilian cerrado (savanna), they contrast with research of Australian and African savanna invertebrates, which exhibited a high-level of resilience to fire (Parr et al., 2004, Andersen et al., 2014). Again, while our results concur with some studies linking pyrodiversity to the diversity of faunal groups in biomes other than savanna (Tingley et al., 2016, Ponisio et al., 2016), they contrast with others (Taylor et al., 2012, Farnsworth et al., 2014). The inconsistent support for the pyrodiversity hypothesis between biomes and faunal groups emphasises the proposition of Kelly and Brotons (2017) of the need for fire management to be tailored to local conditions. While much of the biota in fire-prone environments has evolved to be remarkably resilient to fire, firesensitive elements often persist within the same landscapes (Kelly and Brotons, 2017). The native mammal fauna of northern Australia is a group widely reported to be highly responsive to fire; several studies have demonstrated strong, but marked variation in the response of different mammal species to both experimentally manipulated and naturally varying fire regimes (Corbett et al., 2003, Woinarski et al., 2004, Andersen et al., 2005). For example, Woinarski et al. (2004) demonstrated significant differences in the abundance of native mammal species between an annually burnt area and an area where fire had been excluded for 23 years. They showed that the black-footed tree-rat and common brushtail possum were more abundant in the long-unburnt area, while the northern quoll (Dasyurus hallucatus), northern brown bandicoot and pale field-rat were more abundant in the annually burnt area. However, the large-scale, replicated Kapalga fire experiment demonstrated a preference for unburnt areas for five out of seven mammal species, including the northern quoll, fawn antechinus (Antechinus 99

bellus), northern brown bandicoot, common brushtail possum and grassland melomys (Andersen et al., 2005). Given the variable requirements of northern Australian native mammal species in relation to fire, any area subject to a spatially homogeneous fire history will inevitably disadvantage some species, thus explaining the positive association between pyrodiversity and native mammal diversity demonstrated here. This is consistent with the predictions of the original pyrodiversity model proposed by Martin and Sapsis (1992). That is, when species have different preferences in relation to fire history, pyrodiversity is required to maximise the persistence of all species. However, in order to develop clear and effective fire management, and avoid the often vague operational guidelines based on pyrodiversity rhetoric, it is vital to determine the optimal level of pyrodiversity to achieve specific management objectives (Parr and Andersen, 2006). While we have suggested fire regimes that may support biodiversity on Melville Island, for a number of reasons this may not necessarily be the case for other areas of northern Australia s savannas. Fire is a stochastic, spatially complex form of disturbance and the fire experiment reported here manipulated only one aspect of the fire regime: fire frequency. As a result, we were unable to account for many aspects of the fire regime including the intensity and spatial patterning of fires. As such, our results are based on a much simplified application of fire, and future research should focus on incorporating other aspects of the fire regime into a similar analysis. It is also important to note that our burning treatments did not include high-intensity fires that typically occur late in the dry-season. For example, the intensity of experimental annual fires on Melville Island averaged just 650 kw m -1. By comparison, at Kapalga in Kakadu National Park, the Byram fire-line intensity of experimental annual early dry season fires averaged 2100 kw m -1 (Williams et al., 1998). High-intensity fires have been shown to have both direct (Firth et al., 2010) and indirect negative impacts (Legge et al., 2008, Leahy et al., 2016) on the survival of multiple species in northern Australian savannas. While Andersen et al. (2005) suggested that fires of very low-intensity (occurring in April/early May) could benefit ground-active native mammals, they pointed out that early dry season management fires are typically of higher intensity. As our results are based on small, low-intensity experimental fires, the suggested optimal fire regime may only be applicable in areas where such low-intensity fire regimes are achievable. 100

The effect of fire regimes on biodiversity can act synergistically with other threatening processes (Driscoll et al., 2010, Andersen et al., 2012). In northern Australian savannas, these include the density of exotic mega-herbivores (Legge et al., 2011), invasive grasses (Rossiter et al., 2003), as well as the density of mammalian predators including the dingo (Canis dingo) and feral cat (Felis catus) (Leahy et al., 2016). As these other threatening processes vary across the landscape, so too will the optimal fire regime for biodiversity conservation. For example, in areas with high predator densities, it may be that species that would otherwise inhabit more open areas, are forced to shelter in long-unburnt vegetation due to its mitigating effect on predation pressure (McGregor et al., 2015, McGregor et al., 2016, Leahy et al., 2016). The presence and density of the threatened native species for which management is commonly aimed at conserving, also vary across the landscape. Here we have identified the optimal fire regime for the ground-active native mammals of Melville Island. Again, given that a different suite of mammal species occurs in the savannas of mainland northern Australia, the optimal fire management for species conservation will likely differ to some extent. While our study investigated how different fire patterns influence native mammal diversity, future work should also incorporate the fire response of other non-mammal species, especially those thought to be declining and sensitive to fire regimes (e.g. partridge pigeon, Geophaps smithii; (Fraser et al., 2003)). Our results were sensitive to the variability in species specific abundance estimates (due to both the limited replication of our fire experiment and the sensitivity of GMA to rare species). As such, future work utilising this method should include a sensitivity analysis. Furthermore, the approach outlined in this manuscript may be strengthened by the incorporation of diversity metrics other than Shannon s diversity index and GMA. Specific targets that go beyond pyrodiversity rhetoric are necessary for fire management for biodiversity conservation to be operationally effective (Andersen et al., 2005). However, these targets are highly context specific and depend on a range of factors including management priorities, the fire requirements of the species present in a particular area, as well as the presence and severity of other threatening processes. The realisation that a fire regime that promotes biodiversity in one system is often not applicable to another, has resulted in a more concerted effort to develop fire management that is supported by ecological theory, but tailored to local conditions (Farnsworth et al., 2014, Kelly and 101

Brotons, 2017). By utilising a long-running fire experiment we have demonstrated not only the utility of first determining species-specific responses to fire with which to develop fire management, but the flexibility that this approach affords to develop and tailor fire management based on specific and changing management priorities in other fire-prone environments, i.e. conserving threatened species vs. conserving an entire mammal assemblage. While the feasibility of implementing replicated fire experiments is low, conducting correlative pilot studies that are specifically designed to relate biodiversity to fire regimes will greatly improve our ability to develop effective fire management strategies. 102

References: Andersen, A. N., Cook, G. D., Corbett, L. K., Douglas, M. M., Eager, R. W., Russell-Smith, J., Setterfield, S. A., Williams, R. J. & Woinarski, J. C. Z. 2005. Fire frequency and biodiversity conservation in Australian tropical savannas: implications from the Kapalga fire experiment. Austral Ecology, 30, 155-167. Andersen, A. N., Ribbons, R. R., Pettit, M. & Parr, C. L. 2014. Burning for biodiversity: highly resilient ant communities respond only to strongly contrasting fire regimes in Australia's seasonal tropics. Journal of Applied Ecology, 51, 1406-1413. Andersen, A. N., Woinarski, J. C. Z. & Parr, C. L. 2012. Savanna burning for biodiversity: Fire management for faunal conservation in Australian tropical savannas. Austral Ecology, 37, 658-667. Bond, W. & Van Wilgen, B. 1996. Fire and Plants. Population and Community Biology, 263. Bond, W. J., Woodward, F. I. & Midgley, G. F. 2005. The global distribution of ecosystems in a world without fire. New Phytologist, 165, 525-538. Bowman, D. M. J. S., Balch, J., Artaxo, P., Bond, W. J., Cochrane, M. A., D Antonio, C. M., DeFries, R., Johnston, F. H., Keeley, J. E., Krawchuk, M. A., Kull, C. A., Mack, M., Moritz, M. A., Pyne, S., Roos, C. I., Scoot, A. C., Sodhi, N. S. & Swetnam, T. W. 2011. The human dimension of fire regimes on Earth. Journal of Biogeography, 38, 2223-2236. Bowman, D. M. J. S., Balch, J. K., Artaxo, P., Bond, W. J., Carlson, J. M., Cochrane, M. A., D Antonio, C. M., DeFries, R. S., Doyle, J. C., Harrison, S. P., Johnston, F. H., Keeley, J. E., Krawchuk, M. A., Kull, C. A., Marston, J. B., Moritz, M. A., Prentice, I. C., Roos, C. I., Scott, A. C., Swetnam, T. W., Van Der Werf, G. R. & Pyne, S. J. 2009. Fire in the Earth System. Science, 324, 481-484. Bowman, D. M. J. S. & Panton, W. J. 1993. Decline of Callitris intratropica R. T. Baker & H. G. Smith in the Northern Territory: Implications for pre- and post-european colonization fire regimes. Journal of Biogeography, 20, 373-381. 103

Briani, D. C., Palma, A. R., Vieira, E. M. & Henriques, R. P. 2004. Post-fire succession of small mammals in the Cerrado of central Brazil. Biodiversity & Conservation, 13, 1023-1037. Buckland, S. T., Studeny, A. C., Magurran, A. E., Illian, J. B. & Newson, S. E. 2011. The geometric mean of relative abundance indices: a biodiversity measure with a difference. Ecosphere, 2, 1-15. Burnham, K. P. & Anderson, D. R. 2002. Model Selection and Multimodel Inference: A Practical Information-Theoretic Approach. Springer-Verlag, New York. Corbett, L. C., Andersen, A. N. & Muller, W. J. 2003. Terrestrial vertebrates. In: Andersen, A. N., Cook, G. D. & Williams, R. J. (eds.) Fire in tropical savannas: The Kapalga experiment. Springer- Verlag, New York. Crisp, M. D., Burrows, G. E., Cook, L. G., Thornhill, A. H. & Bowman, D. M. 2011. Flammable biomes dominated by eucalypts originated at the Cretaceous-Palaeogene boundary. Nature Communications, 2, 193. Davies, H. F., McCarthy, M. A., Firth, R. S. C., Woinarski, J., Gillespie, G. R., Andersen, A. N., Geyle, H., Nicholson, E. & Murphy, B. P. 2016. Top-down control of species distributions: feral cats driving the regional extinction of a threatened rodent in northern Australia. Diversity and Distributions, 23, 272-283. Davies, H. F., McCarthy, M. A., Firth, R. S. C., Woinarski, J. C. Z., Gillespie, G. R., Andersen, A. N., Rioli, W., Puruntatameri, J., Roberts, W., Kerinaiua, C., Kerinauia, V., Womatakimi, K. & Murphy, B. P. 2018. Declining populations in one of the last refuges for threatened mammal species in northern Australia. Austral Ecology, DOI: 10.1111/aec.12596. Driscoll, D. A., Lindenmayer, D. B., Bennett, A. F., Bode, M., Bradstock, R. A., Cary, G. J., Clarke, M. F., Dexter, N., Fensham, R., Friend, G., Gill, M., James, S., Kay, G., Keith, D. A., MacGregor, C., Russell-Smith, J., Salt, D., Watson, J. E. M., Williams, R. J. & York, A. 2010. Fire management for biodiversity conservation: Key research questions and our capacity to answer them. Biological Conservation, 143, 1928-1939. 104

Drummond, S. P., Wilson, K. A., Meijaard, E., Watts, M., Dennis, R., Christy, L. & Possingham, H. P. 2010. Influence of a Threatened-Species Focus on Conservation Planning. Conservation Biology, 24, 441-449. Farnsworth, L. M., Nimmo, D. G., Kelly, L. T., Bennett, A. F. & Clarke, M. F. 2014. Does pyrodiversity beget alpha, beta or gamma diversity? A case study using reptiles from semi-arid Australia. Diversity and Distributions, 20, 663-673. Firth, R. S. C., Brook, B. W., Woinarski, J. C. Z. & Fordham, D. A. 2010. Decline and likely extinction of a northern Australian native rodent, the Brush-tailed Rabbit-rat Conilurus penicillatus. Biological Conservation, 143, 1193-1201. Franklin, D. C. 1999. Evidence of disarray amongst granivorous bird assemblages in the savannas of northern Australia, a region of sparse human settlement. Biological Conservation, 90, 53-68. Fraser, B. F., Lawson, V., Morrison, S., Christophersen, P., McGreggor, S. & Rawlinson, M. 2003. Fire management experiment for the declining partridge pigeon, Kakadu National Park. Ecological Management & Restoration, 4, 94-102. Giljohann, K. M., McCarthy, M. A., Kelly, L. T. & Regan, T. J. 2015. Choice of biodiversity index drives optimal fire management decisions. Ecological Applications, 25, 264-277. Gill, A. M., Groves, R. H. & Noble, I. R. 1981. Fire and the Australian biota. Australian Academy of Science, Canberra. Griffiths, A. D., Garnett, S. T. & Brook, B. W. 2015. Fire frequency matters more than fire size: Testing the pyrodiversity biodiversity paradigm for at-risk small mammals in an Australian tropical savanna. Biological Conservation, 186, 337-346. IUCN 1996. 1996 IUCN Red List of Threatened Animals. IUCN, Cambridge, UK. Kelly, L. & Brotons, L. 2017. Using fire to promote biodiversity. Science, 355, 1264-1265. 105

Kelly, L. T., Brotons, L. & McCarthy, M. A. 2016. Putting pyrodiversity to work for animal conservation. Conservation Biology, 31, 952-955. Kelly, L. T., Nimmo, D. G., Spence-Bailey, L. M., Taylor, R. S., Watson, S. J., Clarke, M. F. & Bennett, A. F. 2012. Managing fire mosaics for small mammal conservation: a landscape perspective. Journal of Applied Ecology, 49, 412-421. Kershaw, A. P., Clark, J. S., Gill, A. M. & D Costa, D. M. 2002. A history of fire in Australia. In: Flammable Australia: the fire regimes and biodiversity of a continent, 3-25. Leahy, L., Legge, S. M., Tuft, K., McGregor, H. W., Barmuta, L. A., Jones, M. E. & Johnson, C. N. 2016. Amplified predation after fire suppresses rodent populations in Australia s tropical savannas. Wildlife Research, 42, 705-716. Legge, S., Kennedy, M. S., Lloyd, R. A. Y., Murphy, S. A. & Fisher, A. 2011. Rapid recovery of mammal fauna in the central Kimberley, northern Australia, following the removal of introduced herbivores. Austral Ecology, 36, 791-799. Legge, S., Murphy, S., Heathcote, J., Flaxman, E., Augusteyn, J. & Crossman, M. 2008. The short-term effects of an extensive and high-intensity fire on vertebrates in the tropical savannas of the central Kimberley, northern Australia. Wildlife Research, 35, 33-43. Maravalhas, J. & Vasconcelos, H. L. 2014. Revisiting the pyrodiversity biodiversity hypothesis: longterm fire regimes and the structure of ant communities in a Neotropical savanna hotspot. Journal of Applied Ecology, 51, 1661-1668. Martin, R. E. & Sapsis, D. B. 1992. Fires as agents of biodiversity: pyrodiversity promotes biodiversity. In: Proceedings of the conference on biodiversity of northwest California ecosystems. Cooperative Extension, University of California, Berkeley. McCarthy, M. A., Moore, A. L., Krauss, J., Morgan, J. W. & Clements, C. F. 2014. Linking Indices for Biodiversity Monitoring to Extinction Risk Theory. Conservation Biology, 28, 1575-1583. 106

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. McGregor, H. W., Legge, S., Jones, M. E. & Johnson, C. N. 2016. Extraterritorial hunting expeditions to intense fire scars by feral cats. Scientific reports, 6, 22599. Nimmo, D., Kelly, L., Spence-Bailey, L., Watson, S., Taylor, R., Clarke, M. & Bennett, A. 2013. Fire mosaics and reptile conservation in a fire-prone region. Conservation Biology, 27, 345-353. Parr, C. L. & Andersen, A. N. 2006. Patch mosaic burning for biodiversity conservation: a critique of the pyrodiversity paradigm. Conservation Biology, 20, 1610-1619. Parr, C. L., Robertson, H. G., Biggs, H. C. & Chown, S. L. 2004. Response of African savanna ants to long-term fire regimes. Journal of Applied Ecology, 41, 630-642. Ponisio, L. C., Wilkin, K., M'Gonigle, L. K., Kulhenek, K., Cook, L., Thorp, R., Griswold, T. & Kremen, C. 2016. Pyrodiversity begets plant pollinator community diversity. Global Change Biology, 22, 1794-1808. Richards, A. E., Andersen, A. N., Schatz, J. O. N., Eager, R., Dawes, T. Z., Hadden, K., Scheepers, K. & Van Der Geest, M. 2012. Savanna burning, greenhouse gas emissions and indigenous livelihoods: Introducing the Tiwi Carbon Study. Austral Ecology, 37, 712-723. Richards, A. E., Liedloff, A. & Sschatz, J. 2015. Tiwi Islands CFI capability project report. CSIRO, Darwin. Richards, S. A., Possingham, H. P. & Tizard, J. 1999. Optimal fire management for maintaining community diversity. Ecological Applications, 9, 880-892. Rossiter, N. A., Setterfield, S. A., Douglas, M. M. & Hutley, L. B. 2003. Testing the grass-fire cycle: alien grass invasion in the tropical savannas of northern Australia. Diversity and Distributions, 9, 169-176. 107

Russell-Smith, J., Whitehead, P. J., Cook, G. D. & Hoare, J. L. 2003. Response of Eucalyptusdominated savanna to frequent fires: Lessons from Munmarlary, 1973 1996. Ecological Monographs, 73, 349-375. Swan, M., Christie, F., Sitters, H., York, A. & Di Stefano, J. 2015. Predicting faunal fire responses in heterogeneous landscapes: the role of habitat structure. Ecological Applications, 25, 2293-2305. Taylor, R. S., Watson, S. J., Nimmo, D. G., Kelly, L. T., Bennett, A. F. & Clarke, M. F. 2012. Landscape-scale effects of fire on bird assemblages: does pyrodiversity beget biodiversity? Diversity and Distributions, 18, 519-529. Tingley, M. W., Ruiz-Gutierrez, V., Wilkerson, R. L., Howell, C. A. & Siegel, R. B. Pyrodiversity promotes avian diversity over the decade following forest fire. Proc. R. Soc. B, 2016. The Royal Society, 20161703. Van Strien, A., Soldaat, L. & Gregory, R. 2012. Desirable mathematical properties of indicators for biodiversity change. Ecological Indicators, 14, 202-208. Whelan, R. J. 1995. The ecology of fire, Cambridge university press, Cambridge. Williams, R., Gill, A. & Moore, P. 1998. Seasonal changes in fire behaviour in a tropical savanna in northern Australia. International Journal of Wildland Fire, 8, 227-239. Woinarski, J., Burbidge, A. & Harrison, P. 2014. Action Plan for Australian Mammals 2012. CSIRO Publishing, Melbourne. Woinarski, J. & Winderlich, S. 2014. A strategy for the conservation of threatened species and threatened ecological communities in Kakadu National Park 2014-2024. Woinarski, J. C. Z., Legge, S., Fitzsimons, J. A., Traill, B. J., Burbidge, A. A., Fisher, A., Firth, R. S. C., Gordon, I. J., Griffiths, A. D., Johnson, C. N., McKenzie, N. L., Palmer, C., Radford, I., Rankmore, B., Ritchie, E. G., Ward, S. & Ziembicki, M. 2011. The disappearing mammal fauna of northern Australia: context, cause, and response. Conservation Letters, 4, 192-201. 108

Woinarski, J. C. Z., Risler, J. & Kean, L. 2004. Response of vegetation and vertebrate fauna to 23 years of fire exclusion in a tropical Eucalyptus open forest, Northern Territory, Australia. Austral Ecology, 29, 156-176. Ziembicki, M. R., Woinarski, J. C., Webb, J. K., Vanderduys, E., Tuft, K., Smith, J., Ritchie, E. G., Reardon, T. B., Radford, I. J., Preece, N., Perry, J., Murphy, B. P., McGregor, H., Legge, S., Leahy, L., Lawes, M. J., Kanowski, J., Johnson, C., James, A., Griffiths, A., Gillespie, G. R., Frank, A., Fisher, A. & Burbidge, A. A. 2014. Stemming the tide: progress towards resolving the causes of decline and implementing management responses for the disappearing mammal fauna of northern Australia. Therya, 6, 169-225. 109

Individuals live-trapped Individuals live-trapped Supporting information: a) Common brushtail possum 9 8 7 6 5 4 3 2 1 0 y = 0.0947x + 0.0331 R² = 0.3136 0 10 20 30 40 50 Number of nights detected on camera traps b) Black-footed tree-rat 7 6 5 4 3 y = 0.067x + 0.1694 R² = 0.2109 2 1 0 0 5 10 15 20 25 30 35 Number of nights detected on camera traps c) Northern brown bandicoot 110

Individuals live-trapped 2.5 2 1.5 1 0.5 0 y = 0.0417x - 0.0473 R² = 0.2453 0 5 10 15 20 25 30 35 40 Number of nights detected on camera traps Appendix S4.1: The relationship between the number of nights detected on camera traps and the number of individuals live-trapped per site for a) common brushtail possum b) black-footed tree-rat c) northern brown bandicoot. Data collected at 84 sites across Melville Island in 2015. 111

Coefficient (±95% CI) Coefficient (±95% CI) Coefficient (±95% CI) Coefficient (±95% CI) Coefficient (±95% CI) Coefficient (±95% CI) a) Northern brown bandicoot b) Common brushtail possum 2.5 2 1.5 1 0.5 0 16 12 8 4 0 c) Black-footed tree-rat d) Brush-tailed rabbit-rat 3.5 3 2.5 2 1.5 1 0.5 0 1.8 1.5 1.2 0.9 0.6 0.3 0 e) Grassland melomys f) Pale field-rat 25 20 15 10 5 0 25 20 15 10 5 0 g) Mice/dunnarts 112

Coefficient (±95% CI) 1.4 1.2 1 0.8 0.6 0.4 0.2 0 Appendix S4.2: The effect size of the fire treatment and site variables in the respective best models predicting the abundance of a) northern brown bandicoot b) common brushtail possum c) black-footed tree-rat d) brush-tailed rabbit-rat e) grassland melomys f) pale field-rat g) mice/dunnarts. 113

Chapter 5: Tailored fire management for target species: a case study from northern Australian savannas Our ability to manage fire for biodiversity conservation remains limited. 114

Abstract: Despite being a globally significant issue, our ability to manage fire for biodiversity conservation remains limited. The realisation that a particular fire regime shown to promote biodiversity in one system may be detrimental in another has resulted in a more concerted effort to develop fire management that is tailored to meet the fire-requirements of the target species. Here we utilise a natural experiment in the savanna landscapes of northern Australia to investigate the potential ecological consequences of a blanket approach to fire management. We compare the relationship between fire regimes and native mammal diversity in two savanna landscapes: mainland northern Australia and the Tiwi Islands. While we demonstrate a positive association between the diversity of post-fire age classes with mammal alpha diversity on the mainland, we found no association between fire regimes and mammal alpha diversity on the Tiwi Islands. Our results suggest that the current approach to fire management on mainland northern Australia will likely conserve native mammal alpha diversity. Importantly, we demonstrate that while maintaining low levels of annual proportion burnt (APB) would likely maintain high native mammal gamma diversity (and low extinction risk) on mainland northern Australia, the same approach to fire management on the Tiwi Islands would increase the extinction risk of native mammals on the Tiwi Islands by more than 20%. In doing so, we demonstrate that a different approach to fire management is needed on the Tiwi Islands compared to mainland northern Australia. While developing the optimal fire regime for biological conservation requires an in-depth understanding of the fire response of all species in a given area, acquiring such an understanding is both a challenging and slow process. However, underpinning fire management with even rudimentary evidence linking some target species to particular fire patterns will likely be a safer option than applying fire management based on the fire diversity relationships demonstrated in other areas. Introduction: Fire profoundly influences vegetation diversity and structure in many biomes, and shapes entire ecosystems across the globe (Bowman et al., 2009, Kelly and Brotons, 2017). As a result, fire not only represents a potentially destructive form of disturbance, but a valuable management tool often applied 115

to achieve conservation outcomes (Driscoll et al., 2010). However, as fire acts synergistically with other environmental factors that vary across landscapes, a particular fire regime shown to promote biodiversity in one system may decrease it in another, and vice-versa (Driscoll et al., 2010, Taylor et al., 2012, Farnsworth et al., 2014). This realisation has resulted in a more concerted effort to develop fire management that while supported by ecological theory, is tailored to local conditions (Farnsworth et al., 2014, Kelly and Brotons, 2017). In northern Australian savannas, fire management is widely advocated as a key tool to prevent the further decline of native mammal populations, many of which have suffered severe declines over recent decades (Woinarski and Winderlich, 2014, Davies et al., 2016). The decline of native mammals across northern Australian savannas has been linked to the more frequent, high-intensity fires occurring since the breakdown of traditional Aboriginal fire management (Woinarski et al., 2010, Lawes et al., 2015). As such, determining the fire regime that most strongly promotes native mammal diversity has become a high priority of research. Currently, creating and maintaining a fine-scale fire mosaic, decreasing fire frequency and increasing the amount of long-unburnt vegetation are key objectives of fire management for small mammal conservation in northern Australia (Andersen et al., 2012, Woinarski and Winderlich, 2014, Ziembicki et al., 2014). This approach is advocated due to evidence that frequent, large, intense fires have a negative effect on small native mammals, most likely by increasing predation pressure while decreasing habitat quality (Andersen et al., 2005, Legge et al., 2008, Firth et al., 2010, Ziembicki et al., 2014, Griffiths et al., 2015, Lawes et al., 2015, Leahy et al., 2016, McGregor et al., 2016). However, it is unclear whether a uniform approach is suitable for all parts of northern Australian savannas. The negative effects of fire on biodiversity can act synergistically with other threatening processes (Driscoll et al., 2010, Andersen et al., 2012). In northern Australian savannas, these include exotic mega herbivores such as buffalo (Bubalus bubalis), cattle (Bos taurus) and horse (Equus caballus) (Legge et al., 2011a), invasive grasses (Rossiter et al., 2003), as well as mammalian predators including the dingo (Canis dingo) and feral cat (Felis catus) (Leahy et al., 2016, McGregor et al., 2016). As the occurrence and magnitude of these threatening processes vary across the landscape, so too will the optimal fire 116

regime for biodiversity conservation. For example, in areas with high predator densities, it may be that mammal species are forced to shelter in long-unburnt vegetation due to its mitigating effect on predation pressure (McGregor et al., 2015, McGregor et al., 2016, Leahy et al., 2016), whereas in areas with lower predator densities, the same species may actually prefer more frequently burnt areas. Not only does the natural distribution of mammal species vary across northern Australia s savanna landscapes, but the extent of decline has also not been uniform, with some areas (especially high rainfall areas) appearing more resilient than others (Davies et al., 2018). This further supports the proposition that the optimal fire regime for the conservation of mammal assemblages, as well as the potential benefits from the implementation of such management, also varies across the landscape. We utilise an ideal natural experiment in northern Australia to investigate how the relationship between mammal diversity and fire regimes differ between two savanna landscapes: the mainland Northern Territory and adjacent Tiwi Islands. We examine: (1) how the relationship between mammal alpha diversity and a range of fire characteristics differs between these two areas; (2) how the optimal fire regime that maximises mammal gamma diversity (and minimises extinction risk) differs between these two areas. Given the biophysical differences between the savanna landscapes of mainland Northern Territory and the Tiwi Islands, we hypothesised that the relationships between fire and mammal alpha diversity, as well as the identified optimal fire regime, would be significantly different between these two areas. Methods: Study site Between 1996 and 2011, native mammals were surveyed at 388 sites across Australia s Northern Territory. There were 123 sites on the mainland: (48 in Kakadu National Park [19,084 km 2 ], 35 in Nitmiluk National Park [2,947 km 2 ], 40 in Litchfield National Park [1,458 km 2 ]). There were 265 sites on the Tiwi Islands (8,320 km 2 ). All sites were located in savanna woodland dominated by Eucalyptus and Corymbia spp. with a predominantly grassy understorey (Figure 5.1). 117

The Tiwi Islands differ substantially from the mainland Northern Territory, both abiotically and biologically. While both areas are characterised by a tropical monsoonal climate with a distinct wet season (November April) in which over 90% of the annual rainfall occurs, the consistent formation of thunderstorms over the Tiwi Islands results in the highest annual rainfall across monsoonal northern Australia (> 2000 mm), a greater amount of dry season rain, and a reduced dry season duration compared to mainland Northern Territory (Brocklehurst, 1998, Crook, 2001). The Tiwi Islands are relatively flat and lack the rugged, rocky escarpments that characterise Kakadu, Nitmiluk and Litchfield National Parks. As a result, the Tiwi Island mammal fauna lacks the rock-adapted species that occur widely on the mainland, and has therefore been historically less diverse. However, the decline of native mammals on mainland northern Australia appears to have progressed to a much greater extent compared to the recently identified declines of Tiwi Island native mammals (Davies et al., 2018). Recent evidence suggests that feral cat densities are significantly lower on the Tiwi Islands compared to areas of mainland northern Australia (H. Davies, unpublished data), such as Kakadu National Park (Legge et al., 2017). Data collection: Surveys used a standard approach to general mammal surveys in northern Australia, involving a 50 50 m quadrat with 20 metal Elliott traps (33 10 9 cm) spaced equidistantly around the perimeter and one cage trap (56 20 20 cm) located on each of the four corners. Traps were baited with a mixture of peanut butter, oats and honey, and set for three consecutive nights. Four pitfall traps (two 20 L and two 10 L plastic buckets, each with 10 m of 30 cm high drift-line fence) were also used at each site over the 3-day sampling period. Additionally, five 10-minute searches (two at night using spotlights and three during daylight hours) were conducted at each site. 118

Figure 5.1: Location of the 388 sites surveyed for native mammals across northern Australia between 1996 and 2011: a) the Tiwi Islands b) Litchfield National Park c) Nitmiluk National Park d) Kakadu National Park. 119

Table 5.1: Description and justification of the variables included in analyses. Variables Description and justification for inclusion Diversity of post-fire age classes Diversity of fire-frequencies Proportion long-unburnt Patchiness Annual proportion burnt Rainfall Ruggedness Distance to water Underpinning the patch mosaic burning approach is the assumption that a heterogeneous fire regime will create a mosaic that simultaneously caters for the fire requirements of a greater number of species than a mosaic created by a more homogeneous fire regime. The Shannon Wiener diversity index is often used to describe the diversity of animal communities, and makes the distinction that a community dominated by only a few species is less diverse than a community of several different species of similar abundance. Following studies of birds, mammals and reptiles in semi-arid southern Australia (Taylor et al., 2012, Kelly et al., 2012, Nimmo et al., 2013, Farnsworth et al., 2014), we calculated the diversity of post-fire age classes for each site using the Shannon-Wiener diversity index. We derived the Shannon-Wiener diversity index for the proportion of the circular area (3.2 km radius) surrounding each site that belonged to six post-fire age classes: long-unburnt (> five years since fire); burnt five years prior; burnt four years prior; burnt three years prior; burnt two years prior and burnt one year prior. The premise behind this variable being that the more even the proportions of the post-fire age classes around each site, the greater the range of resources available for multiple species with differing requirements in regard to fire, compared to an area dominated by few post-fire age classes. We derived the Shannon Wiener index for the proportion of the circular area (3.2 km radius) surrounding each site that was burnt zero to five times over the five years preceding mammal sampling. The inclusion of this variable followed the same logic as our variable quantifying the diversity index of postfire age classes, i.e. as mammal species have specific preferences in relation to the frequency of fire, an area providing a greater range of fire frequencies will likely support a more diverse mammal assemblage. Given the detrimental effects of frequent, spatially extensive fires in northern Australia, increasing the proportion of relatively long-unburnt vegetation in the landscape is a common objective for management agencies. As such, we calculated the proportion of the circular area (3.2 km radius) surrounding each site that had not been burnt in at least the five years prior to mammal sampling. Creating, and maintaining a fine-scale patchy fire mosaic is also a common objective of management agencies in northern Australia (Andersen et al., 2005). A patchy mosaic may help break up the fuel load, thereby reducing the risk of large-scale, intense wildfires. Following Lawes et al. (2015), this metric of the spatial heterogeneity of fires was calculated by measuring the distance to the nearest burnt unburnt boundary at the end of each calendar year, within a circular area (radius of 3.2 km) surrounding each site. We then calculated the mean of all distance values to get an annual measure of patchiness for the area surrounding each site. We derived this measure for every site in each of the five years preceding mammal sampling and calculated the mean of these five values. Large values indicate areas of low patchiness i.e. areas dominated by large homogeneous patches of either burnt or unburnt vegetation. Frequent fires have been implicated in the decline of small mammals across northern Australia (Lawes et al. 2015). We calculated the proportion of the circular area (3.2 km radius) surrounding each site that was burnt in each of the five years preceding mammal sampling, and then took the average of these five values. As mammal species across northern Australia have been more resilient in areas with higher rainfall (McKenzie, 1981), we included mean annual rainfall in our models. Mammal species in northern Australia appear to have been more resilient in rugged areas (McKenzie, 1981, Hohnen et al., 2016). As such, we derived a measure of topographic complexity by calculating the variability of a digital elevation layer for the circular area (3.2 km radius) surrounding each site. The proximity of mammal sampling sites to permanent fresh-water systems is often a significant predictor of mammal occurrence. As such we included a remote sensed variable of the distance (m) of each site to the nearest permanent water body. 120

Data analysis: We used generalised linear models (GLMs) to investigate the association between our 8 explanatory variables (Table 5.1) and mammal species richness. To do this we ran all 256 combinations of the explanatory variables. This was done separately for mainland sites (123 sites), and then for the Tiwi Island sites (265 sites). The response variable of site-level species richness was a count of the species detected at each site, and was therefore modelled with a Poisson error structure and log-link function. Model selection was based on a robust form of Akaike s Information Criterion, AIC c, an index that favours both model fit and model simplicity (Burnham and Anderson, 2002). Where no single model arose as superior at explaining mammal species richness (i.e. ΔAIC < 2), model averaging provided parameter estimates based on the results of multiple models (Burnham & Anderson, 2002). To allow for the direct comparison of effect sizes, all explanatory variables were centred and standardised by subtracting the variable s mean value and dividing by the variable s standard deviation. To investigate how optimal fire management for mammal conservation may differ between mainland northern Australia and the Tiwi Islands, we followed a similar approach as Giljohann et al. (2015), Kelly et al. (2015) and Davies et al. (in press). That is, we used estimates of species abundances in response to fire regimes to identify the optimal mix of fire regimes within a landscape that maximises native mammal gamma diversity and minimises extinction risk. We did this for both mainland northern Australia, and the Tiwi Islands. First, we categorised the annual proportion burnt (APB) metric into high (sites with APB of 60%), medium (APB between 40% and 60%), and low (APB of 40%). We chose this fire metric as it combines both temporal (fire frequency) and spatial (fire size) aspects of the fire regime that can potentially be manipulated in northern Australian savannas (Legge et al., 2011b). To obtain species-specific estimates of abundance, we fitted a GLM (Poisson with log-link) for each species with the single main effect of APB categories. Obtaining species-specific estimates of abundance in each category of APB allowed us to investigate the optimal fire patterns for mammal gamma diversity on both the mainland northern Australia and Tiwi Islands. To do this, we generated 5000 hypothetical, spatially-dimensionless sites with varying proportions of each level of APB (low, medium and high) at each of these sites, such that the sum of 121

the three proportions was equal to 1. Following studies of birds, mammals and reptiles in semi-arid south-eastern Australia (Taylor et al., 2012, Kelly et al., 2012, Nimmo et al., 2013, Farnsworth et al., 2014), we expressed pyrodiversity for each simulated site using the Shannon diversity index, scaled such that the maximum (i.e. a site with equal proportions of low, medium and high APB) was equal to 1. We used the species-specific estimates of abundance (derived from GLMs) in each level of APB to obtain an abundance estimate for each species at each simulated site and then calculated the geometric mean abundance (GMA) for each site as: n GMA = Pj 1 Pj 2 Pj 3 Pj n, where n is the number of species, and Pj is the predicted abundance for each species given the simulated fire history. GMA is an appropriate biodiversity index, having heuristic properties that capture a range of desirable criteria with which to assess biodiversity (Buckland et al., 2011, Van Strien et al., 2012, McCarthy et al., 2014). We scaled predicted GMA so the maximum value was equal to 1, and related it to each site s pyrodiversity index. A particularly useful property of GMA is that it tends to be correlated with the proportion of species within an area that are likely to become extinct (McCarthy et al., 2014). Following Giljohann et al. (2015), we utilised this relationship and quantified the change in extinction risk (ER) resulting from different levels of pyrodiversity at each site as: ln (GMA) ) 1 ( ln (GMAmax) ) ER = 1 ( 1 We then plotted the values of both GMA and ER against each site s pyrodiversity index. This was done separately for both mainland northern Australia and for the Tiwi Islands. Five species were recorded frequently enough for statistical analysis on mainland northern Australia: fawn antechinus (Antechinus bellus); northern brown bandicoot (Isoodon macrourus); grassland melomys (Melomys burtoni); blackfooted tree-rat (Mesembriomys gouldii); pale field-rat (Rattus tunneyi). Six species were recorded frequently enough for statistical analysis on the Tiwi Islands: brush-tailed rabbit-rat (Conilurus 122

penicillatus); northern brown bandicoot; grassland melomys; black-footed tree-rat; pale field-rat; common brushtail possum (Trichosurus vulpecula). Results: While no single model was clearly superior in describing mammal species richness on both the Tiwi Islands and mainland northern Australia, model averaging demonstrated that on mainland northern Australia, site-level mammal species richness (alpha diversity) was significantly correlated with mean annual rainfall (positive), the distance to water (negative) and the diversity of post-fire age classes (positive) (Figure 5.2a). The associations between mammal species richness and our explanatory variables was less pronounced on the Tiwi Islands, with annual rainfall the only significant predictor of mammal species richness (negative) (Figure 5.2b). While we found a significant positive association between mammal species richness and the diversity of post-fire age classes at mainland northern Australian sites (Figure 5.3), we found no significant relationship between any fire variable and Tiwi Island mammal diversity. The estimated abundance of species varied between each level of APB (Figure 5.4). Importantly, the fire response of species that were recorded frequently enough for statistical analysis on both mainland northern Australia and the Tiwi Islands (northern brown bandicoot, grassland melomys, black-footed tree-rat and pale field-rat) varied between these two areas. As a result of this variation in species preferences in relation to fire, the optimal mix of fire regimes that maximised mammal diversity, and minimised extinction risk, differed substantially between mainland northern Australia and the Tiwi Islands. For mainland northern Australia, mammal diversity (GMA) was maximised (and extinction risk minimised) by an area composed of 51% low and 49% medium APB. However, it is important to note that this only slightly increased mammal diversity and reduced extinction risk compared to sites composed entirely of low or medium APB (Figure 5.5). A mainland site composed of entirely high APB (i.e. frequent, large fires) would have diversity levels almost half of the potential maximum, and be associated with a 35% increase in extinction risk (Figure 5.5). For the Tiwi Islands, mammal diversity 123

was maximised (and extinction risk minimised) by having a site of entirely medium APB. In strong contrast to mainland northern Australia where low levels of APB were associated with high levels of mammal diversity and low levels of extinction risk, sites on the Tiwi Islands composed entirely of low APB would have mammal diversity 23% below the potential maximum and an associated 23% higher extinction risk (Figure 5.5). 124

Estimated regression coefficient Estimated regression coefficient a) Mainland northern Australia 1.4 1.2 1 0.8 0.6 0.4 0.2 0-0.2-0.4-0.6 b) Tiwi Islands 1.4 1.2 1 0.8 0.6 0.4 0.2 0-0.2-0.4-0.6 Figure 5.2: Model averaged regression coefficient estimates for mammal species richness on a) mainland northern Australia b) the Tiwi Islands. Error bars indicate 95% confidence intervals; asterisks indicate where they do not overlap zero. 125

Site-level mammal species diversity 5 4 3 2 1 0 0 0.5 1 1.5 2 Diversity of post-fire age classes Figure 5.3: Modelled relationship between the diversity of post-fire age classes and site-level mammal species richness on mainland northern Australia. Thin lines indicate 95% confidence intervals. Open circles indicate observed data. 126

Abundance Abundance Abundance Abundance Abundance Abundance Abundance a) Fawn antechinus b) Brush-tailed rabbit-rat 0.7 0.6 0.5 0.4 0.3 0.2 0.1 0 Low Medium High c) Common brushtail possum d) Northern brown bandicoot 0.8 0.6 0.4 0.2 0 Low Medium High e) Grassland melomys f) Black-footed tree-rat 0.5 0.4 0.3 0.2 0.1 0 Low Medium High g) Pale field-rat 2 0.7 0.6 0.5 0.4 0.3 0.2 0.1 0 Low Medium High 1.2 1 0.8 0.6 0.4 0.2 0 Low Medium High 0.5 0.4 0.3 0.2 0.1 0 Low Medium High 1.5 1 0.5 0 Low Medium High Figure 5.4: The predicted abundance (±SE) of native mammals for each categorised level of average proportion burnt on mainland northern Australia (filled circles) and the Tiwi Islands (open circles). 127

1 0.5 0.8 0.6 0.4 L M H 0.2 0 0 0.5 1 Relative change in the index proportional to the extinction risk Geometric mean abundance a) Mainland northern Australia: 0.4 0.3 0.2 0.1 0 0 Pyrodiversity 0.5 1 Pyrodiversity b) Tiwi Islands: 0.5 Relative change in the index proportional to the extinction risk Geometric mean abundance 1 0.8 0.6 0.4 L M H 0.2 0 0 0.5 Pyrodiversity 1 0.4 0.3 0.2 0.1 0 0 0.5 1 Pyrodiversity Figure 5.5: The relationship between pyrodiversity and the geometric mean abundance (GMA) and the relative change in the index proportional to the extinction risk for ground-active native mammal species on a) northern Australian mainland and b) the Tiwi Islands at 5000 simulated sites. The square, triangle and diamond intercept markers indicate the predicted mammal diversity and extinction risk at sites composed of entirely low, medium and high annual proportion burnt, respectively. The dotted vertical lines indicate the level of pyrodiversity with the maximum predicted mammal GMA and minimum change in extinction risk. Discussion: Prescribed fire is routinely used to promote biodiversity in fire-prone landscapes across the globe. However, as fire biodiversity relationships are influenced by many ecological processes that vary between landscapes, a blanket approach to fire management may not only be a waste of resources but actually reduce biodiversity in some areas (Driscoll et al., 2010, Taylor et al., 2012, Farnsworth et al., 2014, Kelly and Brotons, 2017). In areas such as the savanna landscapes of northern Australia, where prescribed fire is the most widely applied management tool aimed at mitigating the ongoing decline of 128

native mammal species, it is especially important to determine how and where fire management is best applied. By demonstrating how fire biodiversity relationships differ substantially between two adjacent but ecologically and abiotically different savanna landscapes in northern Australia, we demonstrate the need for fire management for biodiversity conservation to be directly underpinned by the fire requirements of the target species. The decline of northern Australia s native mammal populations over recent decades has been anecdotally linked with the loss of traditional Aboriginal burning practices (Woinarski et al., 2011, Ziembicki et al., 2014). In an attempt to mitigate these declines, the current approach to fire management often aims to create and maintain a fine-scale fire mosaic, decrease fire frequency and increase the amount of long-unburnt vegetation (Andersen et al., 2012, Woinarski and Winderlich, 2014, Ziembicki et al., 2014). The positive association we identified between the diversity of post-fire age classes and mammal species richness suggests that on the mainland, native mammals have persisted better in areas with a heterogeneous fire history, and retracted from areas with a more homogeneous fire history. This is consistent with the earlier findings of Woinarksi et al. (2010) and Lawes et al. (2015). As such, our results validate the current approach to fire management on mainland northern Australia, i.e. creating a fine-scale mosaic of post-fire age classes should benefit native mammal populations. We show that fire history is not the only factor associated with native mammal diversity across mainland northern Australia. Our results suggest that areas with higher annual rainfall and closer to permanent water sources are important refuges for native mammal species. As refuges aid in the survival, persistence and longer-term re-establishment of populations in extensively burned landscapes (Robinson et al., 2013), a greater understanding of the attributes (number, size, location, demographic processes and connectivity) of such refuges may help facilitate the development of more spatiallytargeted fire management (Banks et al., 2011, Robinson et al., 2013). For example, in northern Australian savannas it may be beneficial for managers to prioritise the application of fire management (i.e. create a fine-scale mosaic of post-fire age classes) around refuge areas to facilitate the recolonisation, connectivity and dispersal of species from refuge source populations. Such an approach 129

to fire management has been suggested for the conservation of Mallee emu-wren (Stipiturus mallee) in south-eastern Australia (Brown et al., 2009), and future work should focus on integrating refuge connectivity into fire management in savanna landscapes. In contrast to the sites on mainland northern Australia, there was no clear association between Tiwi Island mammal species richness and fire. Davies et al. (2018) recently provided the first evidence of native mammal decline on the Tiwi Islands. By demonstrating a similar pattern of initial decline to that identified in Kakadu National Park over two decades earlier, they concluded that these declines had not yet progressed to the point of inducing the widespread range contractions experienced by many species across mainland northern Australia. There were a number of plausible explanations for the relative resilience of Tiwi Island native mammal populations including: higher rates of survival and reproduction, higher initial population densities, reduced predation pressure and relatively benign fire regimes. The relatively intact state of Tiwi Island mammal populations may explain the lack of any clear fire diversity relationships in the present study, as the distribution of Tiwi mammal populations may not yet have contracted. However, it is considered likely that without appropriate management intervention, Tiwi Island mammal populations will continue to decline similarly to the now highly degraded state of mainland populations (Davies et al., 2018). This highlights an important dilemma for developing fire management for relatively intact ecosystems: if we cannot confidently apply fire management based on the fire diversity relationships identified in another system, but we also can t identify those relationships within an intact system, on what basis do we develop and implement fire management for intact ecosystems? We have provided evidence that fire management for the conservation of native mammals in the savanna landscapes of mainland northern Australia should aim to create or maintain low or medium levels of annual proportion burnt (APB), and reduce the occurrence of areas with high APB by reducing the frequency and size of fires. However, for Tiwi Island savannas, fire management should aim to maintain medium levels of APB. High levels, as well as low levels of APB, are expected to be associated with reduced mammal diversity and increased extinction risk. Importantly, we demonstrate that while low APB on mainland northern Australia is associated with high levels of diversity (and low levels of 130

extinction risk), maintaining low APB on the Tiwi Islands would significantly reduce mammal gamma diversity, and increase extinction risk by more than 20%. Studies have recently warned that fire management strategies identified in one area are not necessarily applicable, without modification, to another (Taylor et al., 2012, Farnsworth et al., 2014). Here, we have demonstrated that a different approach to fire management may be needed on the Tiwi Islands compared to mainland northern Australia. The approach used here to characterise the fire regime that maximises native mammal gamma diversity (and minimises extinction risk) has no consideration of the spatial characteristics of applying prescribed fire, and does little to further our understanding of the underlying processes driving these relationships. Furthermore, our study was narrowly focussed on only one group of animals (mammals) in one habitat type (savanna woodland). Developing and incorporating a greater understanding of spatial fire characteristics (including size, patchiness, and connectivity of refuges), a mechanistic understanding of species responses to fire, as well as synergistic threatening processes into the development of fire management strategies currently represent significant shortcomings that should continue be the focus of future work (Driscoll et al., 2010). Ideally, management would maintain all species within a region. However, as the development of such management requires an in-depth knowledge of all species responses to fire (Dellasala et al., 2004, Clarke, 2008), in reality this is not possible (Driscoll et al., 2010). This shortcoming was apparent in the current study as we could not model the fire response of the rarer mammal species due to insufficient data. This is an important shortcoming as conservation efforts are often focused on the rarest species (Drummond et al., 2010), and particularly important in ecosystems likely to become progressively more disturbed, as determining species-specific responses to fire may become increasingly difficult as declines progress. Developing a functional-trait approach that groups all species based on their traits, like that has been proved moderately successful for plant species, could prove useful (Driscoll et al., 2010). However, an approach such as this would require a substantial increase in our current understanding of savanna fauna, including life-histories, dispersal, behaviour and resource use (Driscoll et al., 2010). 131

Despite being a globally significant issue, our ability to manage fire for biodiversity conservation remains limited (Bowman et al., 2009, Kelly and Brotons, 2017). For prescribed fire management for biodiversity conservation to be most effective, it needs to be tailored to specific areas based on an understanding of the fire responses of the species within that area (Taylor et al., 2012, Kelly and Brotons, 2017). However, given that characterising biodiversity responses to fire can be difficult and expensive (Driscoll et al., 2010), it is not surprising that current fire management is often based on a rudimentary understanding of the fire response of a handful of species in an area, or from fire biodiversity relationships demonstrated in entirely different areas. The value of the current study lies in demonstrating the potential benefits to biodiversity of developing fire management based on the fire requirements of the target species (and the potential risk of not doing so), and will hopefully serve as motivation to develop more effective, spatially-tailored fire management. 132

References: Andersen, A. N., Cook, G. D., Corbett, L. K., Douglas, M. M., Eager, R. W., Russell-Smith, J., Setterfield, S. A., Williams, R. J. & Woinarski, J. C. Z. 2005. Fire frequency and biodiversity conservation in Australian tropical savannas: implications from the Kapalga fire experiment. Austral Ecology, 30, 155-167. Andersen, A. N., Woinarski, J. C. Z. & Parr, C. L. 2012. Savanna burning for biodiversity: Fire management for faunal conservation in Australian tropical savannas. Austral Ecology, 37, 658-667. Banks, S. C., Dujardin, M., McBurney, L., Blair, D., Barker, M. & Lindenmayer, D. B. 2011. Starting points for small mammal population recovery after wildfire: recolonisation or residual populations? Oikos, 120, 26-37. Bowman, D. M. J. S., Balch, J. K., Artaxo, P., Bond, W. J., Carlson, J. M., Cochrane, M. A., D Antonio, C. M., DeFries, R. S., Doyle, J. C., Harrison, S. P., Johnston, F. H., Keeley, J. E., Krawchuk, M. A., Kull, C. A., Marston, J. B., Moritz, M. A., Prentice, I. C., Roos, C. I., Scott, A. C., Swetnam, T. W., Van Der Werf, G. R. & Pyne, S. J. 2009. Fire in the Earth System. Science, 324, 481-484. Brocklehurst, P. 1998. The History and natural Resources of the Tiwi Islands, Northern Territory. Parks and Wildlife Commission of the Northern Territory, Darwin, Australia. Brown, S., Clarke, M. & Clarke, R. 2009. Fire is a key element in the landscape-scale habitat requirements and global population status of a threatened bird: The Mallee Emu-wren (Stipiturus mallee). Biological Conservation, 142, 432-445. Buckland, S. T., Studeny, A. C., Magurran, A. E., Illian, J. B. & Newson, S. E. 2011. The geometric mean of relative abundance indices: a biodiversity measure with a difference. Ecosphere, 2, 1-15. Burnham, K. P. & Anderson, D. R. 2002. Model Selection and Multimodel Inference: A Practical Information-Theoretic Approach. Springer-Verlag, New York. Clarke, M. F. 2008. Catering for the needs of fauna in fire management: science or just wishful thinking? Wildlife Research, 35, 385-394. 133

Crook, N. A. 2001. Understanding Hector: The dynamics of island thunderstorms. Monthly weather review, 129, 1550-1563. Davies, H. F., McCarthy, M. A., Firth, R. S. C., Woinarski, J., Gillespie, G. R., Andersen, A. N., Geyle, H., Nicholson, E. & Murphy, B. P. 2016. Top-down control of species distributions: feral cats driving the regional extinction of a threatened rodent in northern Australia. Diversity and Distributions, 23, 272-283. Davies, H. F., McCarthy, M. A., Firth, R. S. C., Woinarski, J. C. Z., Gillespie, G. R., Andersen, A. N., Rioli, W., Puruntatameri, J., Roberts, W., Kerinaiua, C., Kerinauia, V., Womatakimi, K. & Murphy, B. P. 2018. Declining populations in one of the last refuges for threatened mammal species in northern Australia. Austral Ecology, DOI: 10.1111/aec.12596. Davies, H. F., McCarthy, M. A., Andersen, A. N., Rioli, W., Puruntatameri, J., Roberts, W., Kerinaiua, C., Kerinauia, V., Womatakimi, K. & Murphy, B. P. in press. An experimental test of whether pyrodiversity promotes mammal diversity in a northern Australian savanna. Journal of Applied Ecology. Dellasala, D. A., Williams, J. E., Williams, C. D. & Franklin, J. F. 2004. Beyond smoke and mirrors: a synthesis of fire policy and science. Conservation Biology, 18, 976-986. Driscoll, D. A., Lindenmayer, D. B., Bennett, A. F., Bode, M., Bradstock, R. A., Cary, G. J., Clarke, M. F., Dexter, N., Fensham, R., Friend, G., Gill, M., James, S., Kay, G., Keith, D. A., MacGregor, C., Russell-Smith, J., Salt, D., Watson, J. E. M., Williams, R. J. & York, A. 2010. Fire management for biodiversity conservation: Key research questions and our capacity to answer them. Biological Conservation, 143, 1928-1939. Drummond, S. P., Wilson, K. A., Meijaard, E., Watts, M., Dennis, R., Christy, L. & Possingham, H. P. 2010. Influence of a threatened-species focus on conservation planning. Conservation Biology, 24, 441-449. 134

Farnsworth, L. M., Nimmo, D. G., Kelly, L. T., Bennett, A. F. & Clarke, M. F. 2014. Does pyrodiversity beget alpha, beta or gamma diversity? A case study using reptiles from semi-arid Australia. Diversity and Distributions, 20, 663-673. Firth, R. S. C., Brook, B. W., Woinarski, J. C. Z. & Fordham, D. A. 2010. Decline and likely extinction of a northern Australian native rodent, the Brush-tailed Rabbit-rat Conilurus penicillatus. Biological Conservation, 143, 1193-1201. Giljohann, K. M., McCarthy, M. A., Kelly, L. T. & Regan, T. J. 2015. Choice of biodiversity index drives optimal fire management decisions. Ecological Applications, 25, 264-277. Griffiths, A. D., Garnett, S. T. & Brook, B. W. 2015. Fire frequency matters more than fire size: Testing the pyrodiversity biodiversity paradigm for at-risk small mammals in an Australian tropical savanna. Biological Conservation, 186, 337-346. Hohnen, R., Tuft, K., McGregor, H. W., Legge, S., Radford, I. J. & Johnson, C. N. 2016. Occupancy of the invasive feral cat varies with habitat complexity. PLoS ONE, 11, e0152520. Kelly, L. & Brotons, L. 2017. Using fire to promote biodiversity. Science, 355, 1264-1265. Kelly, L. T., Bennett, A. F., Clarke, M. F. & McCarthy, M. A. 2015. Optimal fire histories for biodiversity conservation. Conservation Biology, 29, 473-481. Kelly, L. T., Nimmo, D. G., Spence-Bailey, L. M., Taylor, R. S., Watson, S. J., Clarke, M. F. & Bennett, A. F. 2012. Managing fire mosaics for small mammal conservation: a landscape perspective. Journal of Applied Ecology, 49, 412-421. Lawes, M. J., Murphy, B. P., Fisher, A., Woinarski, J. C., Edwards, A. & Russell-Smith, J. 2015. Small mammals decline with increasing fire extent in northern Australia: evidence from long-term monitoring in Kakadu National Park. International Journal of Wildland Fire, 24, 712. Leahy, L., Legge, S. M., Tuft, K., McGregor, H. W., Barmuta, L. A., Jones, M. E. & Johnson, C. N. 2016. Amplified predation after fire suppresses rodent populations in Australia s tropical savannas. Wildlife Research, 42, 705-716. 135

Legge, S., Kennedy, M. S., Lloyd, R. A. Y., Murphy, S. A. & Fisher, A. 2011a. Rapid recovery of mammal fauna in the central Kimberley, northern Australia, following the removal of introduced herbivores. Austral Ecology, 36, 791-799. Legge, S., Murphy, S., Kingswood, R., Maher, B. & Swan, D. 2011b. EcoFire: restoring the biodiversity values of the Kimberley region by managing fire. Ecological Management & Restoration, 12, 84-92. Legge, S., Murphy, S., Heathcote, J., Flaxman, E., Augusteyn, J. & Crossman, M. 2008. The short-term effects of an extensive and high-intensity fire on vertebrates in the tropical savannas of the central Kimberley, northern Australia. Wildlife Research, 35, 33-43. Legge, S., Murphy, B., McGregor, H., Woinarski, J., Augusteyn, J., Ballard, G., Baseler, M., Buckmaster, T., Dickman, C. & Doherty, T. 2017. Enumerating a continental-scale threat: How many feral cats are in Australia? Biological Conservation, 206, 293-303. McCarthy, M. A., Moore, A. L., Krauss, J., Morgan, J. W. & Clements, C. F. 2014. Linking Indices for Biodiversity Monitoring to Extinction Risk Theory. Conservation Biology, 28, 1575-1583. 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. McGregor, H. W., Legge, S., Jones, M. E. & Johnson, C. N. 2016. Extraterritorial hunting expeditions to intense fire scars by feral cats. Scientific reports, 6, 22599. McKenzie, N. 1981. Mammals of the Phanerozoic south-west Kimberley, Western Australia: biogeography and recent changes. Journal of Biogeography, 263-280. Nimmo, D., Kelly, L., Spence-Bailey, L., Watson, S., Taylor, R., Clarke, M. & Bennett, A. 2013. Fire mosaics and reptile conservation in a fire-prone region. Conservation Biology, 27, 345-353. Robinson, N. M., Leonard, S. W. J., Ritchie, E. G., Bassett, M., Chia, E. K., Buckingham, S., Gibb, H., Bennett, A. F. & Clarke, M. F. 2013. Refuges for fauna in fire-prone landscapes: their ecological function and importance. Journal of Applied Ecology, 50, 1321-1329. 136

Rossiter, N. A., Setterfield, S. A., Douglas, M. M. & Hutley, L. B. 2003. Testing the grass-fire cycle: alien grass invasion in the tropical savannas of northern Australia. Diversity and Distributions, 9, 169-176. Taylor, R. S., Watson, S. J., Nimmo, D. G., Kelly, L. T., Bennett, A. F. & Clarke, M. F. 2012. Landscape-scale effects of fire on bird assemblages: does pyrodiversity beget biodiversity? Diversity and Distributions, 18, 519-529. Van Strien, A., Soldaat, L. & Gregory, R. 2012. Desirable mathematical properties of indicators for biodiversity change. Ecological Indicators, 14, 202-208. Woinarski, J. & Winderlich, S. 2014. A strategy for the conservation of threatened species and threatened ecological communities in Kakadu National Park 2014-2024. Woinarski, J. C. Z., Armstrong, M., Brennan, K., Fisher, A., Griffiths, A. D., Hill, B., Milne, D. J., Palmer, C., Ward, S., Watson, M., Winderlich, S. & Young, S. 2010. Monitoring indicates rapid and severe decline of native small mammals in Kakadu National Park, northern Australia. Wildlife Research, 37, 116-126. Woinarski, J. C. Z., Legge, S., Fitzsimons, J. A., Traill, B. J., Burbidge, A. A., Fisher, A., Firth, R. S. C., Gordon, I. J., Griffiths, A. D., Johnson, C. N., McKenzie, N. L., Palmer, C., Radford, I., Rankmore, B., Ritchie, E. G., Ward, S. & Ziembicki, M. 2011. The disappearing mammal fauna of northern Australia: context, cause, and response. Conservation Letters, 4, 192-201. Ziembicki, M. R., Woinarski, J. C., Webb, J. K., Vanderduys, E., Tuft, K., Smith, J., Ritchie, E. G., Reardon, T. B., Radford, I. J., Preece, N., Perry, J., Murphy, B. P., McGregor, H., Legge, S., Leahy, L., Lawes, M. J., Kanowski, J., Johnson, C., James, A., Griffiths, A., Gillespie, G. R., Frank, A., Fisher, A. & Burbidge, A. A. 2014. Stemming the tide: progress towards resolving the causes of decline and implementing management responses for the disappearing mammal fauna of northern Australia. Therya, 6, 169-225. 137

Chapter 6: General discussion Dry-season sunset from Pirlangimpi beach, Melville Island. 138