Dasyurus Hallucatus)

Quantifying the habitat requirements of an endangered marsupial predator, the northern quoll (Dasyurus hallucatus)

Harry Amos Moore BSc (Hons) Submitted in total fulfilment of the requirements of the degree of Doctor of Philosophy Institute of Land, Water, and Society School of Environmental Sciences Faculty of Science Charles Sturt University March 2021

Cover and interlude art by Nellie Pease

Certificate of authorship

a. I hereby declare that this submission is my own work and that, to the best of my knowledge and belief, it contains no material previously published or written by another person nor material which to a substantial extent has been accepted for the award of any other degree or diploma at Charles Sturt University or any other educational institution, except where due acknowledgment is made in the thesis. Any contribution made to the research by colleagues with whom I have worked at Charles Sturt University or elsewhere during my candidature is fully acknowledged.” b. I agree that this thesis be accessible for the purpose of study and research in accordance with the normal conditions established by the Executive Director, Division of Library Services or nominee, for the care, loan and reproduction of theses.

Harry Moore

23/02/2021

Table of Contents

Certificate of authorship ...... 2 Preface ...... 5 Approvals and funding ...... 5 Acknowledgments ...... 6 Publications resulting from the research ...... 10 Abstract ...... 13 Chapter 1 – Introduction ...... 16 Australian mammal declines ...... 16 Northern Australian mammal declines ...... 18 The northern quoll ...... 19 The Pilbara northern quoll ...... 21 Thesis objectives and structure ...... 22 Chapter 2 – Topographic ruggedness and rainfall mediate geographic range contraction of a threatened marsupial predator ...... 25 Abstract ...... 26 Introduction ...... 27 Materials and methods ...... 30 Results ...... 35 Discussion ...... 43 Acknowledgments ...... 49 Chapter 3 – The effect of camera orientation on the detectability of wildlife: a case study from north- western Australia ...... 51 Abstract ...... 52 Introduction ...... 53 Materials and methods ...... 55 Results ...... 59 Discussion ...... 64 Acknowledgments ...... 68 Chapter 4 – Habitat configuration is more important than habitat amount for a threatened marsupial predator in a naturally fragmented landscape ...... 69 Abstract ...... 71 Introduction ...... 72 Materials and methods ...... 74 Results ...... 81 Discussion ...... 84 Acknowledgements ...... 86

Chapter 5 – A rocky heart in a spinifex sea: occurrence of an endangered marsupial predator is multiscale dependent in naturally fragmented landscapes ...... 89 Abstract ...... 90 Introduction ...... 91 Materials and methods ...... 93 Results ...... 100 Discussion ...... 107 Acknowledgements ...... 111 Chapter 6 – Summary ...... 113 Broader implications ...... 118 Moving forward ...... 119 Appendix A – A brief history of the northern quoll (Dasyurus hallucatus): a systematic review ..... 123 Abstract ...... 124 Introduction ...... 125 Materials and Methods ...... 127 Results & Discussion ...... 131 Acknowledgments ...... 166 Appendix B – Spot on: Using camera traps to individually monitor one of the world’s largest lizards ...... 167 Abstract ...... 168 Introduction ...... 169 Materials and methods ...... 170 Results ...... 175 Discussion ...... 176 Acknowledgements ...... 186 Appendix C – Supplementary material for chapter 2...... 187 Appendix D – Supplementary material for chapter 3 ...... 191 Appendix E – Supplementary material for chapter 4 ...... 192 Appendix F – Supplementary material for chapter 5 ...... 198 References ...... 206

Preface

I have structured this thesis into six chapters, comprising an introductory chapter, four data chapters, and a summary chapter. The introductory chapter provides relevant background information, and introduces the four main objectives for this thesis which are addressed in the chapters that follow. Chapters 2–5 are manuscripts that have either been published (chapter 2, 3) or accepted (chapter 5) in peer reviewed science journals, or are in preparation for submission (chapter 4). As a result, there is some unavoidable repetition and slight differences in formatting between chapters. I am the principle contributor on all chapters in this thesis, and they are all co-authored by a minimum of three of my supervisors.

Approvals and funding

Charles Sturt University Animal Ethics Committee provided ethics approval (A17031, A18034) for all work involving animals. Regulation 17 Licences to take fauna for scientific purposes (08-002476-1, 08- 002475-1) were issued by the Western Australian Department of Biodiversity, Conservation and Attractions for all research conducted as part of this thesis. Permits to enter Yandeyarra Indigenous reserve (Reserve No. 31427, 31428) for the purposes of research were issued by the Western Australian Department of Aboriginal affairs (PER011786). Permission from appropriate land owners or managers was received prior to any work being conducted on private properties as part of this thesis. This project was supported by a scholarship from the Charles Sturt University Institute of Land, Water and Society. Other funding was provided by Roy Hill, the Holsworth Wildlife Research Endowment – Equity Trustees Charitable Foundation, the Ecological Society of Australia, the Western Australian Department of Biodiversity, Conservation and Attractions as well as environmental offsets and public good funding provided by BHP, Rio Tinto, Atlas Iron, Fortescue Metals Group, Roy Hill, Process Minerals International, Metals X and Main Roads Western Australia.

Acknowledgments

As the old saying goes, “It takes a village to raise a PhD Student”. Here I would like acknowledge and thank the village of people that helped me throughout my PhD journey, which I’m sure I will remember as some of the most enjoyable years of my life.

I would first like to acknowledge the Kariyarra and Nyamal people, whose land my field work was conducted on. I am privileged to have witnessed their connection with country first hand throughout my PhD, and I am determined to learn more about their past and present struggles. I would also like to acknowledge the Noongar people, whose land I have lived on for the past four years near the banks of the Derbarl Yerrigan (Swan River).

I would like to thank my primary supervisor Dale Nimmo. Dale took me under his wing as an Honours student in 2015, and since then he has invested an extraordinary amount of time and effort to bring out the best in me as a scientist and a person. Dale’s exploits as an ecologist are recognised globally, however I would like it known that he is equally proficient as a supervisor (skills that don’t always go hand in hand, so I’ve heard). Dale has read a truly sickening number of drafts in his time working with me, and yet never fails to provide comments that improve the manuscript in question, as well as my understanding of what good science is, and how its best communicated. Doing a PhD by distance means I can count the face-to-face PhD meetings I’ve had with Dale on one and a half hands, and yet somehow he has managed to be there whenever I have needed him – even if it’s 5PM AEST and he’s had a couple of beers. In summary, what Dale lacks in his AFL supporting proclivities, he makes up for in patience, hard-work, humour, and academic brilliance, and I am grateful to have learnt so much from him.

To Judy Dunlop. On my first quoll catching trip in 2016, I learnt that wherever Judy Dunlop is, adventure is always nearby. Judy is a true naturalist, and I learn more than I can remember any day I’m in the field with her. Judy’s logistical support made my project possible, but her friendship, down to earth advice, and shared opinion that a pub is as good a place as any to have a meeting, is a big part of what made it so enjoyable. Thank you Judes, for being a great supervisor, and a great role model. To Leonie Valentine, for giving me a home at UWA, and for providing all the positivity a PhD student could ask for. Whenever I kicked in Leonie’s door with a burning PhD question, I always left feeling optimistic and on track. A lot has

6 happened for Leonie in the time I’ve known her, and I can’t wait to watch young Rosie Wren grow up to be just like her mum.

Thanks to Euan Ritchie, who stuck with me through my honours and even managed to endure my PhD! Euan’s passion for conservation is inspiring and served as an excellent reminder for why our work is important. The world could do with a few more Euan Ritchie’s getting around I reckon, even if it meant more Carlton supporters. Last but not least to thank among my supervisory team is Dave Watson. Dave’s ecological wisdom contributed to a PhD design that went the distance, and I owe him for that. Hopefully I’ll catch him on his annual ESA pilgrimage one day soon so I can get an update on his fish tank of unusual size.

I’d like to give a massive thanks to the small army of field assistants I recruited throughout my PhD, consisting of Sian Thorn, Darcy Watchorn, Rainer Chan, Danito Bohorquez Fandino, Jacob Champney, Hannah Kilian, Mitch Cowan, and the Yandeyarra Indigenous Rangers. The Pilbara sun is a real bastard and the flies can nearly be worse, but you copped all of it just so that we could collect some great northern quoll data. Thanks for turning up with a smile in the morning, and for the beers and laughs in the evenings. Thanks to Colin, Betty and Graham from Indee station for putting up with me for many weeks. Their hospitality was lovely and their local knowledge was invaluable. Thanks also goes to Ben and Lindsey from Mallina station and Troy Eaton from Pippingarra station for allowing me access their land.

Thank you to Roy Hill for their huge amount of support. Without it, this project would not have been possible. Thanks especially to Harriet Davie, who played a big part in getting this project off the ground and who has been incredibly helpful the whole way through. She is always going out of her way to make my life easier, and I really appreciate it. A huge thanks to the Western Australian Department of Biodiversity, Conservation and Attractions for supporting this project. Special thanks to Stephen van Leeuwen, Alicia Whittington and Jo Williams for all their help, and a big thanks to Neal Birch, for assisting with logistics throughout my field work and for being a legend.

Thank you to UWA school of Biological Sciences and everyone from the Ecosystem Restoration & Intervention Ecology Lab for hosting me at UWA over the past four years. It’s honestly a huge part of why my PhD was so enjoyable. Special thanks to Richard Hobbs for being incredibly supportive the whole way, and also to Bec Campbell for helping me out whenever she got the chance. Thanks to all my great lab buddies; JP, Gab, Bryony, Stan (the man), Erika, Sian and Danito. It was a real pleasure getting to spend so much time with you. A notable thanks to everyone from the UWA beerclub – a place where I had some of my best and worst PhD ideas. I plan to keep attending.

To my fellow Conservation in Human Landscapes Lab members; Mitch Cowan, Leanne Greenwood, Chris Jolly, Grant Linley and Dylan Westerway. Thanks for setting such a cracking academic standard and for being great people. A massive thanks to Debra Noy from CSU, who saved me an enormous amount of administrative pain and was always so lovely to work with. Thanks to Simon McDonald and Deanna Duffy from the CSU Spatial Data Analysis Network, for all your spatial data assistance.

A gigantic thank you to all of my close friends and family for keeping me going. To my great house mates, Oli, Albert, Nestor and Celine for making my time at home so enjoyable, and offering scientific opinions, unscientific opinions, and cask wine as necessary (typically in that order). To all my close mates in Queensland and Victoria for constant support and hurtful insults. Special shout out to fellow PhD battlers, Darcy Watchorn, Dr. Kita Ashman and Dylan Westerway. It helped a heap to have people in my corner who know the highs and lows of a PhD. To my partner, Dr. Anna Cresswell, former PhD battler, now conqueror, who has done so much for me both inside and outside my life at uni. You made bad days bearable and the good days even better. Thank you for all of our adventures around Perth, Western Australia, and beyond, and for our never ending quests to find ducklings and greeblings. I’m looking forward to many more. Thanks also to Anna’s parents, Susan and George, who like Anna, edited many of my drafts. It was very much appreciated.

Finally, thanks to my family, for always fostering my love of nature, and backing me 100%. A huge thanks to Mum (Trace), and my younger sisters Meg and Georgia, for being on this journey with me. Each of them are now northern quoll experts, not by choice, but as a consequence of caring about what I care about. That is something I hope never to take for granted. Love you guys.

A petroglyph from the Pilbara bioregion depicting a northern quoll. Photo supplied by Judy Dunlop, with permission from Murujuga Aboriginal Corporation.

Publications resulting from the research

Chapter 2 Moore, H. A., Dunlop, J. A., Valentine, L. E., Woinarski, J. C.Z., Ritchie, E. G., Watson, D. M., and Nimmo, D. G. (2019). Topographic ruggedness and rainfall mediate geographic range contraction of a threatened marsupial predator. Published by Diversity and distributions in September 2019. Moore, H. A. Conceived and developed ideas, collected data, performed statistical analysis and wrote the manuscript. Dunlop, J. A. Contributed to manuscript preparation and developed ideas. Valentine, L. E. Developed ideas and contributed to manuscript preparation. Woinarski, J. C.Z. Contributed to manuscript preparation and developed ideas. Ritchie, E. G. Contributed to manuscript preparation. Watson, D. M. Contributed to manuscript preparation. Nimmo, D. G. Conceived and developed ideas, assisted with statistical analysis and contributed to manuscript preparation.

Senior author: Dale Nimmo

Chapter 3 Moore, H. A., Valentine, L. E., Dunlop, J. A., and Nimmo, D. G. (2020). The effect of camera orientation on the detectability of wildlife: a case study from north‐western Australia. Published by Remote Sensing in Ecology and Conservation in April 2020. Moore, H. A. Conceived and developed ideas, designed and conducted experiments, performed statistical analysis and wrote the manuscript. Dunlop, J. A. Conceived and developed ideas, assisted with designing and conducting experiments and contributed to manuscript preparation Valentine, L. E. Developed ideas and contributed to manuscript preparation. Nimmo, D. G. Conceived and developed ideas, designed experiments, assisted with performing statistical analysis and contributed to manuscript preparation.

Senior author: Dale Nimmo

Chapter 4 Moore, H. A., Dunlop, J. A., Valentine, L. E., and Nimmo, D. G. Habitat configuration is more important than habitat amount for a threatened marsupial predator in a naturally fragmented landscape. Manuscript in preparation for submission. Moore, H. A. Conceived and developed ideas, designed and conducted experiments, performed statistical analysis and wrote the manuscript. Dunlop, J. A. Conceived and developed ideas, assisted with designing and conducting experiments and contributed to manuscript preparation Valentine, L. E. Developed ideas and contributed to manuscript preparation. Nimmo, D. G. Conceived and developed ideas, designed experiments, assisted with performing statistical analysis and contributed to manuscript preparation.

Senior author: Dale Nimmo

Chapter 5 Moore, H. A., Michael, D.R., Ritchie, E. G., Dunlop, J. A., Valentine, L. E., Hobbs, R.J., and Nimmo, D. G. In the heart of a spinifex sea: habitat selection of an endangered marsupial mesopredator in naturally fragmented landscapes depends on within-patch, patch, and landscape variables. Accepted by Landscape Ecology in January 2021. Moore, H. A. Conceived and developed ideas, designed and conducted experiments, performed statistical analysis and wrote the manuscript. Michael, D.R. Contributed to manuscript preparation. Ritchie, E. G. Contributed to manuscript preparation. Dunlop, J. A. Conceived and developed ideas, assisted with designing and conducting experiments and contributed to manuscript preparation Valentine, L. E. Developed ideas and contributed to manuscript preparation. Nimmo, D. G. Conceived and developed ideas, designed experiments, assisted with performing statistical analysis and contributed to manuscript preparation.

Senior author: Dale Nimmo

Appendix A Moore, H. A., Dunlop, J. A., Jolly, C.J., Kelly, E., Woinarski, J. C.Z., Ritchie, E. G., Burnett, S.E., van Leeuwen, S., Valentine, L. E., Cowan, M.A., and Nimmo, D. G. A brief history of the northern quoll (Dasyurus hallucatus): a systematic review. Submitted to Australian Mammalogy in January 2021. Moore, H. A. Conceived and developed ideas, collected and filtered literature and wrote the manuscript. Dunlop, J. A. Developed ideas and contributed to manuscript preparation. Jolly, C.J. Contributed to manuscript preparation. Kelly, E. Contributed to manuscript preparation. Woinarski, J. C.Z. Contributed to manuscript preparation. Ritchie, E. G. Contributed to manuscript preparation. Burnett, S.E. Contributed to manuscript preparation. van Leeuwen, S. Contributed to manuscript preparation. Valentine, L. E. Developed ideas and contributed to manuscript preparation. Ritchie, E. G. Contributed to manuscript preparation. Cowan, M.A. Contributed to manuscript preparation. Nimmo, D. G. Conceived and developed ideas, contributed to manuscript preparation.

Senior author: Dale Nimmo

Appendix B Moore, H. A., Champney, J.L., Dunlop, J. A., Valentine, L. E., and Nimmo, D. G. (2020). Spot on: Using camera traps to individually monitor one of the world’s largest lizards. Published by Wildlife Research in May in 2020. Moore, H. A. Conceived and developed ideas, designed and conducted experiments, performed analysis and wrote the manuscript. Champney, J.L. Assisted with conducting experiment and analysis, and contributed to manuscript preparation. Dunlop, J. A. Conceived and developed ideas, assisted with designing and conducting experiments and contributed to manuscript preparation Valentine, L. E. Developed ideas and contributed to manuscript preparation. Nimmo, D. G. Conceived and developed ideas, designed experiments, assisted with performing statistical analysis and contributed to manuscript preparation.

Senior author: Dale Nimmo

Abstract

Australia’s mammalian fauna is in a state of precipitous decline, with at least 10% of the native endemic species declared extinct since 1788. While declines and extinctions were initially concentrated in southern arid and semi-arid regions of the continent, there is now growing recognition of a contemporary wave of declines occurring in Australia’s north. One species that has already suffered considerable declines is the northern quoll (Dasyurus hallucatus); an endangered marsupial predator that once occurred across much of Northern Australia, but is now restricted to a handful of isolated populations. The drivers of this decline are manifold, but undoubtedly include the cane toad, Rhinella marina, an introduced amphibian that is lethal when consumed by native predators, including the northern quoll. The cane toad now inhabits areas that overlap with the distributional range of northern quoll populations located in Queensland, the Northern Territory, and the Kimberley regions in Western Australia, but has not yet invaded the Pilbara bioregion; the most western northern quoll population, now considered a stronghold for the species. Here, identifying critical habitat for the northern quoll is a research priority. In this thesis, I set out to accomplish four research objectives: (i) Quantify changes in the northern quolls geographic range and ecological niche across the entire species distribution (ii) Develop appropriate and standardised methods for monitoring northern quolls using remote sensing cameras, (iii) Define characteristics of critical habitat for northern quoll within the Pilbara bioregion at the landscape scale, and (iv) Define characteristics of critical habitat for northern quoll within the Pilbara bioregion at the patch scale.

In chapter one, I provide a brief overview of the northern quoll in context to Australia’s other mammal declines. In my first data chapter (chapter two), I showed declines were largest in the eastern extent (Queensland) of the northern quoll range, and smallest in the West (Pilbara), likely reflecting the temporal spread of cane toads. Populations were most likely to persist in high-quality habitat, characterised by topographical ruggedness and high annual rainfall.

In chapter three, I compared the effectiveness of vertically and horizontally orientated camera traps as a survey tool for northern quolls. While studies have traditionally used horizontal camera traps to survey mammals, vertically orientated cameras may be advantageous when surveying quolls specifically, because they can capture high-quality images of spot patterning. Results indicated that vertically orientated cameras were able to reliably detect northern quolls when they were present and were equally as effective as horizontal cameras, but provided additional information at the individual level.

In chapter four, I examined factors influencing the occurrence and abundance of northern quolls in the Pilbara bioregion at the landscape scale, where both response and predictor variables were collected to represent entire landscapes. We used the vertically orientated camera traps piloted in chapter three, given they can collect data at the individual level. We found the spatial configuration of rocky habitats

13 were more important than the amount of habitat for predicting quoll occurrence and abundance at the landscape scale; northern quolls were less abundant in more fragmented landscapes, probably because they are required to spend more time moving between patches, where they are exposed to increased levels of predation. Supporting my findings in chapter two, this study also showed northern quolls favoured areas with high topographical ruggedness, and areas that received more rainfall.

Finally, in chapter five, I expand on the findings of chapter four by measuring northern quoll responses to habitat at additional scales: at the patch scale and within patch scale. Here I showed quolls are more likely to use rocky patches with smaller amounts of edge habitat relative to patch area, and those that comprise habitat with high amounts of vegetation cover and denning crevices. In the spinifex grasslands surrounding rocky patches, quolls were also more likely to use sites with high vegetation cover. Similar to chapter four, I suggest these results are likely driven by predation risk; northern quolls may be more exposed to predators at patches with high amounts of edge, low amounts of vegetation cover, and low den availability.

Collectively, the results from this thesis advance our understanding of northern quoll habitat requirements across its entire range, but particularly in the Pilbara, the last stronghold for the species. We demonstrate the value of large aggregated patches of rocky outcrops and suggest the division of such habitat, through either the construction of mining infrastructure (e.g roads, rail lines) or other means, is likely to have substantial negative impacts on northern quoll populations. In addition, our findings highlight the importance of managing vegetation within and surrounding rocky habitat, particularly in patchy landscapes, given doing so may have the potential to mitigate increased rates of predation in these areas. While the objectives of this thesis were mostly targeted towards the conservation of northern quolls, the methods I developed to achieve them have broad non-species specific applications, and our findings provide important theoretical insight relevant to the conservation of threatened species globally.

Chapter 1 – Introduction

Human activities have modified the planet, leading to increased global extinction rates (Barnosky et al. 2011). At least 69 vertebrate species have become extinct in the last century (Ceballos et al. 2015), and at least another 74 are on the brink of extinction (i.e. < 1000 individuals remaining) (Ceballos et al. 2020). Among the most affected taxa are terrestrial mammals (Visconti et al. 2011). While mammal declines have been global, the rate of loss in Australia has been disproportionately high; Australia has only 6% of the world’s mammal species, and yet one-third of mammal extinctions since the year 1500 have occurred on the island continent (Davidson et al. 2017). These extinctions are invariably linked to changes arising from European colonisation (Phillip 2015).

Australian mammal declines

In 1788, the ‘First Fleet’ of Europeans settlers to arrive in Australia landed in Botany Bay, New South Wales (Phillip 2015). Since then, Australia’s mammalian fauna, most of which is endemic, has suffered catastrophic declines either directly from human disturbance (e.g. land clearing), or indirectly, from the animals and diseases introduced by Europeans (Johnson 2006). A total of 34 Australian mammal species are now listed as extinct under Australian federal, state, or territory legislation, and/or under the International Union for Conservation of Nature (IUCN) Red List, with an average of 1-2 mammal extinctions per decade (since 1788) ― a rate higher than anywhere else globally (Woinarski et al. 2019b). Among those lost were some of the most phylogenetically distinct species on earth such as the thylacine (Thylacinus cynocephalus); a marsupial carnivore weighing up to ~ 35kg, and the oolacunta (desert rat-kangaroo; Caloprymnus campestris), a small arid dwelling macropod, both monotypic species (Woinarski et al. 2015) (Figure 1.1). The loss of Australia’s mammalian fauna continues; 37 non-marine species are currently listed as vulnerable, 13 as endangered, and 9 as critically endangered, according to the IUCN Red List (IUCN 2021). Seven Australia mammal species are forecast to go extinct by 2038 unless conservation efforts improve (Geyle et al. 2018).

Using a combination of subfossil records, journal accounts from early European explorers, and testimony from Indigenous people, it has been possible to reconstruct the historic geographic ranges of Australia’s mammalian fauna (Burbidge and Fuller 1979; Burbidge et al. 1988; Burbidge et al. 2009; Pearson et al. 2001). This evidence tells us that at the time of European settlement, many species now listed as extinct or threatened, occupied continental scale geographic ranges overlapping a broad spectrum of habitat types (Menkhorst and Knight 2011). For example, near the beginning of the 20th century, the numbat (Myrmecobius fasciatus) occurred across much of southern arid and semi-arid Australia as far east as the South Australia and New South Wales border (Friend 1990) (Figure 1.1).

The numbat’s range has since contracted west by at least 2200 km, and there are estimated to be less than 1000 mature individuals remaining in the wild (DBCA 2017). Similarly, many of Australia’s mammals previously occurred in densities far greater than current densities (Menkhorst 2009). For example, multiple historic accounts from the late 1800s in New South Wales and Victoria describe eastern quolls (Dasyurus viverrinus) appearing in the ‘thousands’, and ‘doing great injury to poultry… becoming quite a nuisance’ (Peacock and Abbott 2014) (Figure 1.1). This species is now extinct on mainland Australia, remaining only in Tasmania where it is listed as endangered and continues to decline (Burbidge and Woinarski 2016).

Figure 1.1 – Extinct or declining Australian mammals; A.) Tasmanian tiger, Thylacinus cynocephalus (Extinct), B.) Desert rat-kangaroo, Caloprymnus campestris (Extinct), C.) Numbat, Myrmecobius fasciatus (Endangered), D.) Eastern quoll, Dasyurus viverrinus, (Endangered). Artwork by Elizabeth Gould (Gould et al. 1863).

A number of studies, both early and recent, have shown Australian mammals most likely to have declined or become extinct since European settlement are those unable to fly, and with a mean adult body mass that lies within a ‘critical weight range’ (CWR) of 35g to 5500g (Burbidge and McKenzie 1989; Johnson and Isaac 2009; Murphy and Davies 2014). Among the most cited theories put forward to explain this trend is that non-volant species within the CWR bracket are targeted prey items by two recently introduced predators, the red fox (Vulpes vulpes) and the feral cat (Felis catus), and as such are likely to suffer comparatively greater rates of predation (Johnson and Isaac 2009). However, there are also examples of mammal declines occurring prior to feral predators invading (Krefft 1862), as well as examples of CWR mammals persisting in sympatry with feral predators while maintaining stable populations (Abbott 2002; Miritis et al. 2020; Morris et al. 2004). In reconciling with these observations, some studies have shown that a combination of factors are required in order to explain mammal attrition rates more fully (McKenzie et al. 2007; Woinarski et al. 2019b; Woinarski et al. 2015;

Ziembicki et al. 2013). These include habitat degradation associated with the expansion of cattle and sheep grazing industries, as well as the cessation of Indigenous burning practises, which had been used to manage the land for the 40 to 50 thousand years prior to European colonisation (Flannery 1990). Another factor often used to explain Australian mammal declines is rainfall; mammal attrition rates have been far greater in arid and semi-arid parts of Australia when compared to the tropics (Bennett et al. 1989; Burbidge and McKenzie 1989). On the basis of results such as these it was previously assumed that mammals occurring in Australia’s tropics were somewhat sheltered from declines occurring in dryer parts of the country, however this is no longer the case.

Northern Australian mammal declines

In recent decades there has been growing recognition of a contemporary wave of Australian mammal declines occurring in Australia’s monsoonal north (Davies et al. 2018; Fisher et al. 2014a; Ibbett et al. 2018; Woinarski et al. 2010a; Woinarski et al. 2011b; Ziembicki et al. 2013), where mammal abundance was historically noted as being very high (Woinarski 2000; Woinarski et al. 2011b). Here, mammal declines were first detected during the mid-1980s (Braithwaite and Muller 1997), and since then, an extensive catalogue of research has demonstrated declines in range and abundance across a broad suite of taxa (Cole and Woinarski 2000; Davies et al. 2018; Firth et al. 2010; Stobo-Wilson et al. 2019; von Takach et al. 2020a). For example, between 2001 and 2009, average species richness at sites in Kakadu National Park decreased by 54%, and the number of mammal individuals recorded decreased by 75% (Woinarski et al. 2011b). Species most affected appear morphologically similar to those which had previously declined in southern Australia, fitting within the 35g to 5500g CWR (Murphy and Davies 2014; Woinarski 2015), and, as such, similar drivers, including predation (Davies et al. 2017; Stokeld et al. 2018), over-grazing by introduced herbivores (Legge et al. 2011), and changing fire regimes (Russell-Smith et al. 2003) have been implicated.

An effective response to Australia’s contemporary wave of mammal declines requires enduring large scale management interventions (Scheele et al. 2018). Alas, funding allocations to achieve such a response are grossly inadequate ― the Australian government spends about 15% of what is needed to avoid extinctions and recover threatened species (Wintle et al. 2019). Maximising the efficacy of conservation budgets therefore requires that management efforts are as targeted as possible (Scheele et al. 2018). One way of achieving this is by channelling resources towards habitat where species are most likely to persist, and management interventions are most likely to be successful. Such habitats likely include areas where both threat impacts and species tolerance to these impacts are minimised, given each of these factors are determined by biotic and abiotic processes that vary across environmental space (Scheele et al. 2017). Identifying these habitats and quantifying their key characteristics is now

18 recognised as a research priority for many threatened and declining Australian mammal species (Woinarski et al. 2014).

The northern quoll

In this thesis, I quantify the multiscale habitat requirements that enable the persistence of a unique and endangered marsupial predator, the northern quoll (Dasyurus hallucatus Gould, 1842) (Figure 1.2). Northern quolls are the smallest of six Dasyurus species. Females typically weigh 350–700g, and males are larger (500-1000g). Male northern quolls are known to exhibit semelparity―a reproductive strategy where animals only breed once in their lifetime―characterized by increased levels of testosterone followed by rapid condition loss, with individuals rarely living longer than 11 months (Oakwood et al. 2001), although survival varies (Appendix 1). A consequence of large annual die-offs in the adult population of northern quolls is that the likelihood of population persistence is heavily reliant on offspring survivorship (Moro et al. 2019).

Prior to European colonisation, the geographic range of northern quolls incorporated much of northern Australia above the Tropic of Capricorn (90 ° South) within 200 km of the coastline (Braithwaite and Griffiths 1994), covering an area of over 1.2 million km2. This range is naturally structured into four largely spatially segregated units: Queensland, Northern Territory, the Kimberley region of Western Australia, and the Pilbara region (including the neighbouring Little Sandy and Great Sandy Desert bioregions) of Western Australia (Figure 1.3). Average annual rainfall across the northern quolls’ range varies substantially, ranging from 220 mm in the eastern extent of the Pilbara bioregion, to nearly 4500 mm in the Wet Tropics of northern Queensland (Australian Bureau of Meteorology 2020). Similarly, average maximum temperature in the warmest months ranges from 42.1°C in the Pilbara bioregion to 25.5°C in southern Queensland (Australian Bureau of Meteorology 2020).

Figure 1.2 – The northern quoll, Dasyurus hallucatus. Artwork by Elizabeth Gould (Gould et al. 1863).

Since European colonization of Australia (1788), and particularly within the past 50 years, the geographic range of northern quolls is suspected to have declined substantially (Braithwaite and Griffiths 1994). Consequently, the northern quoll is classified as ‘endangered’ by the IUCN, with a ‘declining’ population trend (IUCN, 2017). However, so far most evidence for the northern quolls declines has come from anecdotal sources, as well as localised studies across northern Australia (Burnett 1997; Ibbett et al. 2018; Woinarski et al. 2011b). For example, Burnett and Zwar (2009) found no evidence to suggest northern quolls persist in southern parts of Queensland where they were once known to occur. Populations in far north Queensland have also been subject to substantial decline. For example, Perry et al. (2015) and Burnett (1997) found northern quolls are largely absent from sizeable areas of Cape York Peninsula where they once occurred. In the Northern Territory (Ibbett et al. 2018) found northern quoll trap success in Kakadu National Park was significantly lower in 2002 than it was in 1980 (Begg 1981), and Woinarski et al. (2011b) found northern quoll abundance was lower at sites within Kakadu in 2007–2009 than it was in 2001–2004.

In addition to range contractions, northern quolls are also thought to have suffered a reduction in niche breadth ― the environmental and biotic factors that constrain a species distribution (Scheele et al. 2017) ― due to the disproportionate loss of local populations from particular environments. In Queensland, the Northern Territory, and the Kimberly, topographically simple landscapes that receive low rainfall have seen severe declines (Burnett 1997; Pollock 1999; Woinarski et al. 2008), and areas that would have historically been marginal habitat have seen the greatest declines (Moore et al. 2019).

Several threats are thought to be responsible for the northern quolls decline, most of which have been implicated in Australia broader mammal declines, including changing fire regimes (Griffiths and Brook 2015; Kerle and Burgman 1984), over-grazing by introduced herbivores (Braithwaite and Griffiths 1994), and predation by feral cats and dingoes/wild dogs (Cremona et al. 2017a; Jolly et al. 2018a; Oakwood 2000). However, the threat of cane toads (Rhinella marina) ― an introduced amphibian that is lethal when consumed by native predators including the northern quoll (Shine 2010) ― is an extinction factor unique to northern quolls among Australia’s threatened mammal species. In 1935, cane toads were introduced to a research station near Cairns, Queensland (17°04’S 145°47’E) (Lever 2001). From there, toads quickly expanded their distribution into other parts of Queensland. Cane toads first invaded the Northern Territory in the 1980s (Freeland and Martin 1985), reached Kakadu National Park in 2001 (Woinarski et al. 2002), and progressed through to Kimberley region in Western Australia around 2009 (Figure 1.3). They are predicted to reach the Pilbara (and potentially its offshore islands) between 2026 and 2064 (Kearney et al., 2008; Tingley et al., 2013).

Figure 1.3 – The northern quolls (Dasyurus hallucatus) historic range superimposed over the current distribution of the invasive cane toad (Rhinella marina). Cane toad data was extracted from Urban et al. (2008) and DBCA (2020).

The Pilbara northern quoll

The Pilbara bioregion remains the only major section of the northern quolls range yet to be invaded by cane toads, and, as such, preserving quolls there has been recognised as a national priority for the species (Hill and Ward 2010). However, much of the Pilbara northern quolls ecology remains poorly understood (Cramer et al. 2016), largely because it received little research attention until the mid-2000s, when additional fauna surveys were undertaken in the Pilbara region, corresponding with a boom in iron ore exploration that peaked around 2013 (ABS, 2021). Since then, genetic analysis has revealed that Pilbara northern quolls are distinct from all other major populations (Woolley et al. 2015), and that there has so far been limited evidence of decline (Cramer et al. 2016). It has also been established that rocky habitats, typically manifesting in the form of granite outcrops, or banded-iron-stone gorges and ridges, are likely to be important for northern quoll persistence in the Pilbara, providing shelter from predators, fire, and extreme summer temperatures (Cowan et al. 2020b) ― mean summer maximum temperatures in the Pilbara can exceed 40₀C (Australian Bureau of Meteorology 2020). Unfortunately, these rocky

21 features are also targeted by mining companies for iron-ore extraction, or quarried for road and rail bedding as part of mine infrastructure, posing yet another threat for the northern quoll (Cramer et al. 2016).

In 2013 and 2016, the Department of Biodiversity, Conservation and Attractions (DBCA), hosted workshops aimed at identifying key areas for future research to advance the conservation of Pilbara northern quolls. Workshop participants included scientists, state and federal members of government, as well as private environmental consultants and industry representatives. A final list of five research priorities were established:, (1) develop appropriate and standardised survey and monitoring methods; (2) define areas of critical habitat and better understand how disturbance affects habitat quality; (3) improve our understanding of population dynamics; (4) better understand the key threats to the northern quoll and the interactions between these threats in the Pilbara; and (5) determine whether the northern quoll will colonise restored areas or artificial habitat (Cramer et al. 2016).

In 2014, the Pilbara Northern Quoll Research Program (PNQRP) was established by the DBCA as part of efforts to address research priorities listed in Cramer et al. (2016). Since then, the PNQRP have developed a standardized live trap protocol for surveying northern quolls, established a series of long term monitoring sites which have been surveyed since 2014, collected hundreds of tissue samples for use in genetic analysis, summarised the Pilbara northern quolls diet, and conducted a number of exploratory surveys resulting in extensions of the Pilbara northern quolls known range by ~ 200 km.

Thesis objectives and structure

In 2017 this PhD was designed and implemented in collaboration with the PNQRP, Roy Hill (Iron Ore mining project located in the Pilbara), and the University of Western Australia to further address Pilbara northern quoll research priorities, along with conservation priorities for the species as a whole. Specific objectives were to:

1. Quantify changes in the northern quolls geographic range and ecological niche across the entire species distribution. 2. Develop appropriate and standardised methods for surveying northern quolls using remote sensing cameras. 3. Define characteristics of critical habitat for northern quoll within the Pilbara bioregion at the landscape scale. 4. Define characteristics of critical habitat for northern quoll within the Pilbara bioregion at the patch scale.

Each of these objectives is addressed sequentially in data chapters 2-5. In chapter two, I created a database containing all existing northern quoll records collected since European settlement (1788), and used it to map declines in the northern quolls geographic range and ecological niche across the whole of its distribution.

In chapter three, I compared the effectiveness of vertically and horizontally orientated camera traps as tools for monitoring northern quolls. Studies have traditionally used horizontal camera traps to survey mammals, largely because have a larger detection zone than vertically orientated camera traps, potentially giving them an advantage. Conversely, vertically orientated cameras may be advantageous when surveying quolls specifically, because they are able to capture high quality images of spot patterning located on the dorsal surface of animals, which can be used for identifying individuals. In chapter four, I used the vertically orientated camera traps piloted in chapter three, to examine habitat factors influencing the occurrence and abundance of northern quolls in the Pilbara bioregion at the landscape scale, where both response and predictor variables were collected to represent entire landscapes. In chapter five, I expand on the findings of chapter four by measuring northern quoll responses to habitat at the patch scale ― where predictor variables describe a discrete unit of habitat within a landscape.

Chapter 2 – Topographic ruggedness and rainfall mediate geographic range contraction of a threatened marsupial predator

Manuscript published: Moore, H. A., Dunlop, J. A., Valentine, L. E., Woinarski, J. C., Ritchie, E. G., Watson, D. M., and Nimmo, D. G. (2019). Topographic ruggedness and rainfall mediate geographic range contraction of a threatened marsupial predator. Diversity and distributions 25, 1818-1831.

Harry A. Moorea, b*, Judy A. Dunlopc, Leonie E. Valentinec, John. C.Z. Woinarskid, Euan G. Ritchiee, David M Watsona, Dale G. Nimmoa aInstitute for Land, Water and Society, School of Environmental Science, Charles Sturt University, Albury 2640, NSW, Australia bSchool of Biological Sciences, University of Western Australia, Crawley, WA 6009, Australia cDepartment of Parks and Wildlife, Science and Conservation Division, Dick Perry Ave, Kensington Western Australia, 6151 dThreatened Species Recovery Hub, National Environmental Science Program, Charles Darwin University, Darwin, NT 0909, Australia. eCentre for Integrative Ecology and School of Life and Environmental Sciences, Deakin University, Burwood, Victoria, Australia

Abstract

Aim

Species range contractions are increasingly common globally. The niche reduction hypothesis posits that geographic range contractions are often patterned across space owing to heterogeneity in threat impacts and tolerance. We applied the niche reduction hypothesis to the decline of a threatened marsupial predator across northern Australia, the northern quoll (Dasyurus hallucatus).

Location

Northern Australia

Methods

We assembled a data base containing 3178 historic and contemporary records for northern quolls across the extent of their distribution dating between 1778 and 2019. Based on these records, we estimated changes in the geographic range of the northern quoll using α-hulls across four main populations. We then examined how range contractions related to factors likely to mediate the exposure, susceptibility, or tolerance of northern quolls to threats.

Result

The extent of range contractions showed an east-west gradient, most likely reflecting the timing of spread of introduced cane toads (Rhinella marina). There were clear changes in environmental characteristics within the contemporary compared to the historic geographic range, with the most substantial occurring in populations that have suffered the greatest range contractions. The contemporary range comprises higher quality habitats (measured using environmental niche models), characterized by higher topographical ruggedness and annual rainfall, and reduced distance to water, compared to the historic range.

Main conclusions

Changes to range and niche likely reflect the capacity of complex habitats to ameliorate threats (namely predation and altered fire regimes), and access to resources that increase threat tolerance. This study highlights the multivariate nature of ecological refuges, and the importance of high-quality habitats for the persistence of species exposed to multiple threats. Our methods provide a useful framework which can be applied across taxa in providing valuable insight to management.

Introduction

The contraction of species’ geographic ranges is the result of population decline and local extinction, and a stark, prominent manifestation of adverse anthropogenic environmental change (Hobbs et al.

2017). Range contractions are increasingly common across the globe (Thomas et al. 2006)—for instance, 40% of 177 mammal species studied by Ceballos, Ehrlich & Dirzo (2017) have experienced range contractions of >80%. As range contractions increase extinction risk (Purvis et al. 2000), the extent of a species’ geographic range, as well as its population size, are considered key elements by the

International Union for Conservation of Nature (IUCN) in determining a species’ global conservation status (IUCN 2006). Range contractions entrain a raft of changes that occur when species are in decline, such as changes to species’ niches (Breiner et al. 2017). Small changes in a species’ geographic range size can correspond to substantial reductions in its niche breadth, and vice versa, due to local extinctions occurring in distinct environments (Breiner et al. 2017), leading to spatial patterning in range declines

(McDonald et al. 2015b)

To integrate the concepts of range decline and niche reduction, Scheele et al. (2017) introduce the ‘niche reduction hypothesis’. This hypothesis distinguishes between a species’ ‘historical niche’—defined as the realized niche of a species prior to decline—and the ‘contemporary niche’, a subset of the historical niche after a reduction in niche breadth due to novel threats (e.g., introduced species, habitat loss, disease). Three factors that shape the contemporary niche of northern quolls are (i) threat occurrence, which, when a threat operates variably across different populations or parts of a species’ range, may cause species to persist only in environmental refugia (ii) a species’ threat tolerance, the ability to persist despite threats, which can be increased within a subset of environmental conditions due to, for example, larger populations sizes, and (iii) geographic barriers that exclude threats from a part of a species range (Scheele et al. 2017). Understanding how threats shape species’ niches, and geographic ranges, is critical to conservation, as it can assist in identifying ecological refuges (Reside et al. 2019).

Further, basing conservation on the contemporary niche can lead to a narrow understanding of a species’ potential habitat, thereby limiting the range of options available for conservation, translocation, and restoration (Scheele et al. 2017).

Here, we apply Scheele’s (2017) niche reduction hypothesis by quantifying changes in the historical and contemporary range and niche of four populations of a marsupial predator—the northern quoll

(Dasyurus hallucatus). The northern quoll is the smallest of four Dasyurus species found in Australia, ranging from 300–1200 g, and males within the species often exhibit a semelparous life history, surviving for only a single breeding season (Oakwood et al. 2001) . The geographic range of northern quolls spans much of northern Australia (Figure 2.1), but is suspected to have declined considerably since European colonisation of Australia (1788), and particularly in the past 50 years (Braithwaite and

Griffiths 1994). Consequently, the northern quoll is classified as ‘endangered’ by the IUCN, with a

‘declining’ population trend (IUCN 2017). The main threats to northern quolls are predation by introduced animals such as feral cats, poisoning by the introduced cane toads (Rhinella marina), and habitat degradation caused by altered fire regimes and grazing by introduced herbivores (IUCN 2017).

Figure 2.1– Northern quoll presence records across each of the four populations. Blue markers represent historic presence records (prior to year 2001) and maroon markers represent contemporary presence records (post 2000). Populations are represented by varying line patterns.

The northern quoll offers an excellent case study of range contraction and niche reduction in a declining species for three reasons. First, declines in northern quoll range have occurred relatively recently (i.e., from the mid to late 20th century to the present day), and so an extensive catalogue of species records exists from within both its historic and contemporary range. This is not the case for many of the

Australian mammals that declined earlier following European colonisation (Woinarski et al. 2015).

Second, in addition to range contractions, northern quolls are thought to have suffered a reduction in niche breadth, due to the disproportionate loss of local populations from particular environments, such as lowland savanna and the arid fringe of their range (Braithwaite and Griffiths 1994; McKenzie 1981;

Oakwood 2000). Consistent with the niche reduction hypothesis, local extinctions are thought to have related to both threat occurrence and threat tolerance. For instance, topographically rugged areas (e.g., rocky outcrops) are considered to offer fixed refuges (sensu Reside et al. 2019) from predation, grazing, and fire (Burnett 1997), increasing the likelihood of persistence of northern quolls (Begg 1981). In addition, local extinctions are hypothesised to have occurred more often in marginal habitats (i.e., low rainfall and lower population size), that predispose populations to a lower threat tolerance due to smaller population sizes (Burnett 1997).

Third, range contractions of northern quolls likely differ across its four main populations (differentiated here by the state: Queensland, Northern Territory, Kimberley and Pilbara), due to geographic barriers that have so far prevented one primary threat—the cane toad (Rhinella marina)—from reaching parts of the northern quoll’s geographic range. Cane toads were introduced to Queensland in 1935 and spread westwards across the northern third of Australia, reaching the Northern Territory during the early 1970s

(Urban et al. 2008), and the Kimberley in 2010 (Doody et al. 2018) (although they have not yet colonised that region entirely), but have not yet reached the Pilbara. The primary mechanism by which northern quolls are impacted by cane toads is through lethal ingestion (Shine 2010). While there have been a number of reports documenting local northern quoll extinctions in Queensland (Burnett 1997;

Burnett and Zwar 2009; Woinarski et al. 2008) and the Northern Territory (Braithwaite and Griffiths

1994; Oakwood 2004b; Woinarski et al. 2010a; Ziembicki et al. 2013), fewer extirpations have been

29 recorded in the Kimberley (Hohnen et al. 2016b; Start et al. 2007), and none have been documented in the Pilbara (Cramer et al. 2016), a pattern of loss consistent with the sequential spread of toads.

We quantify and compare environmental conditions within the historic and contemporary geographic range of the northern quoll across its four major populations, compare range declines with niche reductions (i.e., declines in niche volume). We predicted that, range size and niche volume would be most reduced in Queensland and the Northern Territory, due to their long-term exposure to cane toads.

Consistent with the niche reduction hypothesis, we predicted that range declines within each population would relate to environmental variables that buffer (due to reduced co–occurrence) or increase the threat tolerance of quolls. Specifically, contemporary ranges would trend towards more topographically rugged areas with higher rainfall and greater protection from fire when compared to historic ranges due to extinctions occurring more often in more open, topographically simple and marginal habitats.

Materials and methods

Study area

The region over which northern quolls are believed to have occurred is ~2,634,641 km2 and stretches from south of Brisbane on Australia’s east coast, along northern Australia, to the Pilbara region in

Western Australia. This region includes four discrete populations of northern quolls: in Queensland, the

Northern Territory (including several islands), the Kimberley (including several islands), and the

Pilbara (Figure 2.1). We used previous estimates of northern quoll range to guide the placement of population boundaries (Braithwaite and Griffiths 1994; Oakwood et al. 2016) . Recent analysis suggests each of these populations function as distinct evolutionally significant units, and thus in the interests of genetic conservation, they should be considered separately (How et al. 2009).

The Queensland population spans substantial climatic variation with northern areas characterized by monsoonal wet seasons (December to March), while southern coastal areas receive rainfall that is less seasonal, and the western extent exhibits semi–arid conditions (Australian Bureau of Meteorology

2020) (Table 2.1). Vegetation varies significantly ranging from monsoonal rainforests to mixed eucalypt woodlands and tropical savannas (NVIS 2019). The Northern Territory population experiences a monsoonal tropical climate, with rainfall falling mainly in the period December to March (Australian

Bureau of Meteorology 2020) (Table 2.1). Vegetation is mostly characterized by eucalypt open woodlands and forests, spinifex grasslands and tropical savannahs (NVIS 2019). The Kimberley population also experiences a tropical monsoonal climate (Table 2.1). Vegetation is characterized by desert grasslands in the southern interior and tropical savanna in the north (DPIRD 2017). Climate in the Pilbara is characterized by extremely hot summers, mild winters and low rates of sporadic rainfall

(Australian Bureau of Meteorology 2020) (Table 2.1) . Vegetation is dominated by Acacia and

Eucalyptus low woodlands and hummock grasslands (NVIS 2019) .

Table 2.1 – Study area size and climatic variability in terms of mean coldest and warmest quarter temperature, as well as annual precipitation.

Annual Mean coldest quarter Mean warmest quarter precipitation Population Study area (km2) temp range (°C) temp range (°C) range (mm)

Queensland 989,833 8.6 – 25.4 19.9 – 30.4 407 – 3945

Northern Territory 619,058 17.9 – 24.3 28.2 – 31.6 340 – 1834

Kimberley 362,014 19.1 – 25.6 27.6 – 33.0 419 – 1453

Pilbara 663,736 13.2 – 21.9 28.6 – 33.3 216 – 465

Data collection

A data base containing 6391 northern quoll records was collected from national and state/territory fauna data bases, museum records, past publications, trap records as well as mining and government agencies, dated between 1788 and January 2019. The majority of records were sourced from online data bases the Atlas of Living Australia (ALA 2019), NatureMap (NatureMap 2019) and Wildnet (Wildnet 2019) as well as a 2008 Australian government report investigating northern quoll decline in northern

Australia (Woinarski et al. 2008). Duplicates were identified and removed. Records missing

31 information or with high locational inaccuracy were also removed. In the interest of controlling for threat absences on islands, all island records were removed from our analysis. To minimise the effect of localised survey effort on analysis, all records collected within the same year in the same 1 km*1 km grid cell were condensed to one.

Following Martínez‐Freiría et al. (2015), we separated our data set for each population into two: a historical data set, which sought to describe ranges prior to the year 2001, and a contemporary data set, which sought to describe the range of populations’ from 2001 onward. Also following Martínez‐Freiría et al. (2015), the historical data set included the contemporary observations (i.e. was ‘nested’), so that it included all populations that have likely been present (but may not have been detected) since

European colonisation. This approach was justified given the relatively sparse sampling effort in the historical period. To test the effect that nesting contemporary records within historic records had on our results, we ran a series of analyses (methods below) on both nested and un–nested datasets and compared the results. In Queensland and the Northern Territory, we found nesting had little impact on range contraction, niche volume or habitat suitability predictor importance (Figure S2.1, S2.2).

Conversely, in the Kimberley and the Pilbara, where historical sampling was less thorough, we found range size and niche volume increased through time in un–nested data sets — trends with no supporting evidence. Based on this comparison, we elected to only use nested data in further analysis.

We chose the year 2000 as the separation point between historic and contemporary data sets as it allowed us to have a large number of presences both before and after this point, facilitating the development of separate range estimates for each population during each of the two time periods

(historic and contemporary). While separating the data into more than two time periods would have been preferable, the amount of presences within finer temporal windows did not allow for the development of reliable models. Further, while we note that local extinctions have occurred post–2000

(Oakwood 2004b), these are likely minor compared to the reductions that occurred in the 20th century when cane toads and invasive predators spread across a large proportion of the quoll’s range

(Braithwaite and Griffiths 1994).

Range contractions

We measured change in geographic range using α–hulls, a standard IUCN measure representing a generalization of convex polygons that accounts for breaks in a species range (IUCN 2006). Breiner et al. (2017) found α–hulls closely tracked simulated extinctions and outperformed a range of alternative range quantification metrics (e.g., ENMs, convex polygons). Following IUCN recommendations

(IUCN 2006), we used an α value of two when calculating α–hulls for all datasets. α–hulls were created using ‘alphahull’ package in R (Pateiro-López and Rodrıguez-Casal 2009).

Niche change

We quantified niche volume (the space defined within the bounds of n independent environmental axes) for each population in each time period. To do this we used niche hypervolumes generated by a one– class support vector machine method (SVM), as described by Blonder et al. (2018). Hypervolumes were defined by the bounds of the five environmental variables (Table 2.2), and were scaled prior to analysis

(following Tingley et al. 2014). The SVM method was used as it is insensitive to outliers and generates a smooth boundary around the data, yielding binary predictions of niche volume, facilitating comparisons between data subsets. Once all n–dimensional hypervolumes had been assembled, niche volume was calculated and compared within populations to assess niche reduction. Hypervolumes were calculated in the R package ‘hypervolume’ (Blonder et al. 2014).

Table 2.2 –Variables included in northern quoll ecological niche model and hypervolume analysis accompanied by justification for inclusion and source.

Variable Description Justification Source

Topographical Calculated from the difference in Studies suggest quality northern quoll (Calculated ruggedness elevation between a cell and the habitat is often associated with rocky according to Riley eight cells surrounding it. rugged areas (Braithwaite and Griffiths (1999) (following Riley 1999). 1994; Burnett 1997; Hernandez-Satin (Geoscience 2016; Schmitt et al. 1989). Australia 2018) Elevation Metres above average sea level. Pollack 1999 found northern quolls in (Geoscience central Queensland were typically Australia 2018) found at lower elevation. Molloy 2017 found elevation was a strong contributor to MAXENT modelling for northern quolls in the Pilbara. Annual Derived from spatially interpolated Increased productivity as a result of (WorldClim 2018) precipitation monthly climate database based on high annual precipitation may also averages between 1950 and 2000. boost the capacity of northern quolls to tolerate threats (Burnett 1997; Hohnen et al. 2016b; McKenzie et al. 2007). Precipitation Average monthly variation in Northern quoll records in Queensland (WorldClim 2018) seasonality rainfall (1950 – 2000) expressed as associated with higher levels of rainfall a percentage ratio of the mean seasonality (Woinarski et al. 2008). monthly precipitation total and the standard deviation of the monthly precipitation. Distance to Distance to permanent water Studies suggest areas proximate to Derived from water measured in decimal degrees. permeant water are more likely to 1:100 000 provide high quality for northern quolls watercourse (Begg 1981; Braithwaite & Griffiths mapping provided 1994; Burnett 1997; Molloy et al. 2017) by Geoscience Australia (2018)

We created a layer of habitat quality of northern quolls across study populations using ecological niche models (ENMs). ENMs combine presence records with environmental data to predict habitat that is most likely to support populations of a given species (Guisan and Thuiller 2005). ENMs were developed using the MaxEnt algorithm (Elith et al. 2011). MaxEnt default output consists of a map with every cell assigned a log value representing relative probability of occurrence, ranging from 0 to 1. Cell size for all ENMs in this study was 1 km2 (1 km*1 km). To quantify habitat quality within the range of northern quolls, we incorporated five ecogeographical layers based on evidence within the literature supporting their relevance to northern quoll ecology (Table 2.2). As presence–only data is often subject to selection bias we included a bias grid within MaxEnt models, indicating sample biases across the study area

(Kramer‐Schadt et al. 2013). We used a “target group” background sampling approach to generate the

34 bias layers (Phillips et al. 2009). We defined our target group as all critical weight range (CWR) mammals (including northern quoll) within the study area (following Molloy et al. 2017). These species were selected as survey methods used to detect them would also detect the northern quoll (following

Molloy et al. 2017) — typically a highly detectable species (Austin et al. 2017) — thus ensuring both northern quoll presence data and model background data were drawn from a comparable sampling intensity. Target group species records were subjected to Point Density Analysis (PDA), masked to a 1 km2 grid layer using ArcGIS 10.3, creating a layer with cell values that accurately represent survey effort in relation to location. This layer was included as a bias grid in all MaxEnt models. The contribution of variables to each model was assessed using permutation importance values. Permutation importance is calculated by alternating the predictor values between presence and background points and recording the effect this has on model AUC (Elith et al. 2011).

To quantify changes in the environmental conditions within the contemporary range of quolls, we sampled each of our five environmental predictors 10,000 times within historic and contemporary α- hulls, across all populations. Finally, to quantify changes in habitat quality over time, we sampled the

MaxEnt habitat quality layer within both historic and contemporary α-hulls. We used Generalised

Linear Models (GLMs) to compare the average values of the five environmental predictors between the two time periods, using the historic period as the reference category. We considered there to be significant difference between time periods when the 95% confidence interval of the regression coefficients did not overlap zero. Violin plots were generated to visualized this data using the ‘plotly’ package in R (Sievert et al. 2017).

Results

The total data base consisted of 2025 historic records and 1153 contemporary records (Table S2.1).

Range size, as estimated by α–hulls, declined across all populations except for the Pilbara. Consistent with predictions, the largest absolute and proportional declines occurred in Queensland (areal reduction of 405,533 km2, 75.4% of the historic range), followed by the Northern Territory (115,024 km2, 57.7%) and the Kimberley (25,986 km2,16.9% )(Figure 2.2). Range declines in the Northern Territory were

35 dominated by the loss of record outliers from the semi–arid gulf region in the south–east of their range.

Total loss in range size across all populations was 546,886 km2 (45.2%) (Figure 2.2). In Queensland, the vast majority of persisting northern quoll populations are present on the Central Mackay Coast bioregion, the northern extent of the Northern Brigalow Belt, the Wet Tropics region (Cooktown, Cairns and Atherton Tablelands areas), and near Weipa on the Cape York Peninsula (Figure 2.2). In the

Northern Territory, persisting populations are mostly found in the Darwin Coastal bioregion, the northern extent of the Pine Creek bioregion, and the Tiwi Cobourg peninsula and the Victoria Bonaparte bioregion between the Victoria and Fitzmaurice river. For the Kimberley, contemporary records were mostly found in the King Leopold Ranges north of Derby as well as the Mitchell Plateau region, particularly between the Mitchell and Lawley River National Parks. Northern quolls are still distributed across the entire Pilbara bioregion and also enter the southern extent of the Great Sandy Desert and western extent of the Little Sandy Desert around Karlamilyi National Park.

Niche reduction, measured as hypervolumes, occurred differentially across all populations, closely matching range declines in the Northern Territory (60.8%), and the Kimberley (16.9 %). Marginal declines occurred in the Pilbara (0.1%). While substantial niche reduction was also observed in

Queensland (49.2%), it was a proportionally smaller reduction compared to range declines (Figure 2.3).

The performance of MaxEnt models were high across each of the four data sets, with AUC values averaging 0.92 (SD 0.02). Little change was observed in variable permutation importance between time periods, except for distance to water in the Northern Territory, which increased by 22 percentage points between historic and contemporary periods (Figure 2.4). Annual precipitation was the most important predictor in 6 of the 8 ENMs. Precipitation seasonality also contributed highly in Queensland (Figure

2.4).

There were clear differences between time periods in the average values of environmental variables within the geographic ranges of all four populations (Figure 2.5), and in habitat quality as measured by

MaxEnt, as indicated by 95% confidence intervals of coefficients not overlapping zero (Figure 2.5).

The largest changes occurred in Queensland and the Northern Territory, and smaller changes occurred in the Kimberley and the Pilbara (Figure 2.6). Consistent with predictions, values for topographic

36 ruggedness and annual precipitation increased in the contemporary range compared to the historic range in Queensland, the Northern Territory and the Kimberley. A large positive shift in precipitation seasonality was also recorded for Queensland. Finally, consistent with predictions, habitat quality (as measured by the historic MaxEnt model) was higher in the contemporary range, suggesting a loss of lower quality habitats (Figure 2.5, 2.6).

Figure 2.2– Northern quoll historic and contemporary predicted range across four study populations using α- hulls. Colouring within ranges represent outputs from MaxEnt ecological niche models. Values in the right-hand corner of plots represents α-hull area.

Figure 2.3 – Change in northern quoll range (α–hull), and niche volume (hypervolume) plotted against one another. Queensland (red), Northern Territory (green), Kimberley (blue), Pilbara (purple).

Figure 2.4 – Permutation importance of variables included in northern quoll MAXENT ecological niche models. Permutation importance is calculated by alternating the predictor values between presence and background points and recording the effect this has on model AUC.

Figure 2.5 – Violin plots comparing 10,000 random samples of habitat variables, within the historic (blue) and contemporary (maroon) range of northern quolls (a) annual precipitation, (b) precipitation seasonality, (c) elevation, (d) distance to water (e) topographical ruggedness, and (f) habitat quality derived from the MaxEnt model.

Figure 2.6 – Generalised linear model coefficients comparing 10,000 random samples of habitat variables from within the historic and contemporary ranges of northern quolls. (a) annual precipitation, (b) precipitation seasonality, (c) elevation, (d) distance to water, (e) topographical ruggedness, and (f) habitat quality derived from the historical MaxEnt model. Positive coefficients depict an increase in average values of the variable in the contemporary (cf historic) range. Negative coefficients depict a decrease in average values of the variable in the contemporary (cf historic) range.

Discussion

We applied the niche reduction hypothesis (Scheele et al. 2017) to quantify changes in the geographic range and ecological niche of a declining marsupial predator through time, across multiple populations.

Our results confirm a contraction in geographic range and niche volume for northern quolls in

Queensland, the Northern Territory, and the Kimberley (Braithwaite and Griffiths 1994; Woinarski et al. 2008). As predicted, range declines have been most severe in Queensland and the Northern Territory, less severe in the Kimberley, and negligible in the Pilbara. These findings are consistent with localised declines in each of these regions (Burnett 1997; Braithwaite and Griffiths 1994; McKenzie 1981), and at least partly reflect the incremental spread of toxic cane toads across the species’ range (Urban et al.

2008). Reductions in the northern quoll’s niche volume were > 20% in all populations other than the

Pilbara, and, consistent with the niche reduction hypothesis, we found evidence of environmental variables mediating niche reduction, potentially due to threat occurrence, species’ tolerances, and geographic barriers.

Niche volume—the array of environmental conditions within a species’ realized niche—is a predictor of extinction risk both now and over geological time scales (Saupe et al. 2015). Reductions in niche volume were largely proportionate to range contractions, except in Queensland, where proportional range declines exceeded proportional reductions in niche volume. This result aligns with theory suggesting species range and niche should be correlated for the most part (Slatyer et al. 2013); for example a recent analysis of 148 species using simulated data found niche volume and range size were mostly similar, although considerable residual variation did exist (Breiner et al. 2017). This variation was also detected in our analysis with range decline exceeding niche decline by 29% in Queensland, reaffirming that whilst a consistent relationship between niche volume and range size holds for the most part, one is not always fully reciprocal of the other (Colwell and Rangel 2009). Niche may provide a poor surrogate for range in Queensland given quolls have declined from large swathes of flatter, more open (and relatively homogenous) country, characterized by open savanna woodland vegetation and acacia forests, similar to declines from grasslands and savannah habitat observed in the Northern

Territory (Oakwood 2000). Whilst less diverse in terms of topography and vegetation than coastal

43 habitats, this habitat accounted for the majority of Queensland northern quoll historic range, and thus in disappearing from these areas, range size was reduced to a larger extent than niche volume.

Our finding that the Pilbara population has experienced less range contraction and niche reduction compared to other regions is consistent with expert opinion and the literature—the Pilbara is generally regarded as a stronghold for the northern quoll, due largely to the (current) absence of cane toads and abundance of topographically complex rocky outcrops (Cramer et al. 2016). However, we recognise that a smaller range contraction in the Pilbara may also be an artefact of low historic record availability.

Our final data base for this region contained 66 unique northern quoll records collected before 2001, and 740 unique records from 2001 onward—an 11–fold increase. The difference in sample size is due to a massive increase in survey effort as part of impact assessments and offsets related to a boom in iron ore exploration between 2002 and 2015 (ABS 2018). However, our findings are supported to some extent by genetic data from the Pilbara, which shows no signs of substantial range contraction (Spencer et al. 2013).

By comparing the average values of environmental variables that shape northern quoll distributions within the historic and contemporary niches, we assessed a series of hypotheses regarding the types of environments that would favour northern quoll persistence. We found that the contemporary niche of northern quolls is characterised by more topographically rugged areas, supporting the hypothesis that complex landscapes increase the resistance of northern quolls to threats (or because in such areas the threats are absent or of lower intensity) by acting as fixed refuges (Reside et al. 2019). We found support for this hypothesis in all populations other than the Pilbara, with the greatest increase observed in the

Queensland population. The increase in ruggedness within the Queensland contemporary range likely reflects the loss of populations that previously occupied topographically simple landscapes, thought to have been more exposed to the threats of livestock grazing (Kutt and Woinarski 2007), altered fire regimes (Ash et al. 1997; Murphy et al. 2014) and invasive predators (Hernandez-Satin 2016; Oakwood

2000). Similarly, central rock–rats (Zyzomys pedunculatus) have contracted from more simple to more complex rugged habitat where the impact of their primary threat, the feral cat, is reduced (McDonald et al. 2018). Rocky habitats are also important to northern quolls because they provide highly suitable

44 areas within which females are able to establish maternal denning sites (Oakwood 1997) —critical features in providing developing offspring with protection from predators (Oakwood 2000). The importance of suitable denning habitat is further compounded when considering the unusually short, semelparous life history of northern quolls, which makes population persistence particularly reliant on recruitment (Moro 2019) and thus the effectiveness of dens in providing young with protection. While there is no evidence that rugged habitats reduce the direct impacts of cane toads, it has been suggested that, by limiting the impacts of grazing, fire, and invasive predators, rocky habitats may indirectly buffer against toads by increasing population sizes that can offset mortality due to toads (Burnett 1997).

This finding is consistent with previous work showing that northern quolls are found at higher densities, and that individuals live longer, in rocky habitats (Begg 1981; Schmitt et al. 1989). A similar mechanism might explain why northern quoll contemporary range is characterized by a decreasing distance to permanent water when compared to the historic range, as northern quolls have been found in better reproductive condition closer to creek lines— a habitat associated with free standing water (Braithwaite and Griffiths 1994). The authors of that study suggest quolls closer to water are likely to engage in increased rates of reproduction, presumably because of an increased availability of resources, leading to increased recruitment and overall population size (Braithwaite and Griffiths 1994). This finding is also consistent with Reside et al. (2019), who identified water sources as important fixed refuges for threatened species.

Some of the largest changes in environmental values between the historic and contemporary niche were for annual rainfall, which was also the most important variable driving quoll distributions in three of the four regions. The contemporary niche is characterised by much higher average annual rainfall compared to the historic niche, particularly in the two regions that have experienced the greatest range declines (Queensland and the Northern Territory). The most likely explanation for this shift is that rainfall increases primary productivity (Pianka 2017), and food resources (Dickman et al. 1999), leading to improved body condition, increased offspring survivorship and population size (Meserve et al. 1996),

45 thereby offsetting increased mortality due to threats. Support for this hypothesis in northern Australia is strong, with native mammals typically persisting longer in areas that receive higher annual rainfall

(Fisher et al. 2014b; Start et al. 2012). Northern quolls also persist more in higher rainfall areas (Radford et al. 2014; Woinarski et al. 2008; Ziembicki et al. 2013), even though threats such as grazing, predation, and intense fire are present across both the arid and mesic extents of their range. Hohnen et al. (2016b) suggested one reason northern quoll populations may be more stable in high rainfall areas is because genetic connectivity is greater. They supported this hypothesis by showing quolls in wetter habitats were more closely related than quolls in drier habitats (Hohnen et al. 2016b). Here it is important to note that our analysis did not account for changes in rainfall patterns over time, a factor which has the potential to drive shifts in species range (Davis and Shaw 2001) and niche (Broennimann et al. 2007). Whilst the inclusion of such analysis was outside the scope of this study, we strongly suggest this topic be addressed in future studies as it has been in other species of quoll (Fancourt et al.

2015a), particularly in the context of the earth’s rapidly changing climate (Urban 2015).

We also observed a substantial increase in rainfall seasonality within the contemporary niche in

Queensland. As with the previous two examples, we suggest that this shift most likely reflects increased population size and output related to the timing in quoll breeding activity, where offspring dispersal

(November–February (Oakwood 2000) coincides with a boom in resource availability brought by summer monsoonal systems in northern Australia (Oakwood 2000). By synchronising their reproductive time line with highly seasonal rainfall patterns, northern quolls are likely able to increase offspring survivorship, allowing populations to be more likely to persist despite increased mortality due to threats (e.g., cane toads, invasive predators).

By comparing habitat quality derived from MaxEnt models within the historic and contemporary niche, we found that the contemporary geographic range of northern quolls comprises a subset of the historical niche that is of higher predicted habitat quality, highlighting the importance of high-quality habitats for providing refuge from stressors. Here it is important to recognise that refuges, like niches (Guisan and

Thuiller 2005), are multivariate, and can be defined by the bounds of more than one environmental parameter (Keppel et al. 2012). The most effective refuges should occur on the environmental

46 continuum where resource availability and buffering from threats permit the greatest chance of persistence (Keppel et al. 2012). In Queensland and the Northern Territory, where shifts to high quality habitat were greatest, habitat quality was best predicted by annual precipitation, distance to water and precipitation seasonality. This result conforms with research showing mammal attrition in Australia has been greatest in resource poor arid and semi–arid landscapes (Johnson 2006). Topographical ruggedness was also an important predictor for habitat quality in Queensland and the Kimberley, suggesting a combination of complex and high resource habitats may in some cases act as more effective refuge habitat than habitat that is simply high in resources.

Whilst effective refuge habitat is clearly critical to the persistence of species in the short-term , it can also be important at an evolutionary scale as refugium —a space within a species niche that can support populations over evolutionary time scales (Keppel et al. 2012; Reside et al. 2019). In the case of the northern quoll, habitat that limits but not fully eliminates exposure to key threats such as cane toads and feral predators may facilitate the development of both behavioural and phenotypic traits through behavioural learning or evolution that could potentially allow re-expansion into their historic niche.

Previous studies have found the timeframe for such changes to occur in response to threats (including cane toads) can be surprisingly short (<70 years) (Hudgens and Garcelon 2011; Phillips and Shine

2006). Further, recent studies have found that not only can northern quolls be trained to avoid cane toads in a single generation, but that the trained quolls offspring also avoided eating toads, suggesting quolls have the capacity to rapidly learn behaviour and distribute the information to their young

(Cremona et al. 2017b; Webb et al. 2015). These results suggest investment in protecting refuges (and populations within them) over a relatively short period of time has the potential to yield important conservation outcomes.

An important caveat of our study was that we did not explicitly account for threats such as cane toads, feral predators, or fire, in any of our models. While including threats as predictors would likely aid in determination of their relative influence in shaping the northern quoll’s contemporary range and niche, the scale and resolution of spatial layers required for this analysis do not currently exist and creating them was beyond the scope of our study. It may be possible to explicitly test the effect of threats on

47 quoll populations in future studies by limiting the scale of analysis to a regional level and linking changes in northern quoll occupancy with threat distributions (Hernandez-Satin 2016). For example, by measuring declines in range and niche in relation to fire, it may be possible to elucidate the value of logs and tree hollows — important refuge features for many northern boreal and arboreal mammal species (Goldingay 2012)— as quoll habitat.

Management implications

Previous research suggests cane toads have immediate and lasting detrimental impacts on northern quoll populations (Shine 2010). It is therefore likely that cane toads are a key culprit responsible for larger range and niche contractions in Queensland and the Northern Territory (Burnett 1997; Woinarski et al.

2008), given that cane toads have only very recently arrived in Western Australia and are yet to reach the Pilbara (Pizzatto et al. 2017). Cane toads are expected to colonise the remainder of the Kimberley within the next 5–10 years (Doody et al. 2018), and the Pilbara by 2037–2046 (Southwell et al. 2017).

In light of these predictions, several strategies have been developed to halt or slow the spread of cane toads from the Kimberley to the Pilbara (Phillips et al. 2016; Tingley et al. 2013). These include closing off water bodies (Tingley et al. 2013), as well as introducing toads with ‘less dispersive’ genes to the front of the invasion line to impede the progress of toads carrying ‘highly dispersive’ genes (Phillips

2016; Phillips et al. 2016). We suggest these strategies be given full consideration.

As with the northern quoll, previous studies show topographically rugged and high rainfall habitats also provide important refuge for a range of other threatened species of mammal (Southgate et al. 2007), and as such should form areas of focus for management (Reside et al. 2019). A first step in protecting refuge habitats is recognising where they are located, a process that has been enhanced by remote sensing technologies (Allan et al. 2018) . Once a refuge is located, a management priority should be to ameliorate any existing or emerging stressors to ensure they remain functional in facilitating species persistence (Reside et al. 2019). In the case of northern quolls, these may include limiting extensive grazing and burning, as well as implementing feral predator control programs (Hill and Ward 2010), and potentially applying cane toad aversion techniques (O’Donnell et al. 2010). Although we excluded islands from considerations in our analyses, the marked contractions in range and niche for mainland

48 quolls reinforces the importance of island populations and the need to maintain biosecurity for these islands to reduce the likelihood of toad or predator invasion. Finally, effective management of refuge systems requires high–quality ecological data collected over biologically meaningful timeframes

(Lindenmayer et al. 2012). Therefore, the implementation of sustained and effective monitoring programs inside and outside refuge habitats is highly recommended.

Acknowledgments

We thank all who contributed to record databases that were used in this study as well as authors of

Woinarski et al. (2008) for collating many of the Queensland records. We also thank Deanna Duffy and

Rachel Whitsed and Simon Mcdonald for their assistance with manipulating spatial data. H.A.M was supported by a scholarship from the Institute of Land, Water and Society operating funds from the

Faculty of Science at Charles Sturt University. L.E.V was funded by the Australian Government’s

National Environmental Science Program through the Threatened Species Recovery Hub. D.G.N. was supported by an Australian Research Council Early Career Researcher Award (DECRA). This project was supported by the Western Australian Department of Biodiversity, Conservation and Attractions as well as environmental offsets and public good funding provided by BHP, Rio Tinto, Atlas Iron,

Fortescue Metals Group, Roy Hill, Process Minerals International, MetalsX and Main Roads Western

Australia.

Chapter 3 – The effect of camera orientation on the detectability of wildlife: a case study from north-western Australia

Manuscript published: Moore, H. A., Valentine, L. E., Dunlop, J. A., and Nimmo, D. G. (2020). The effect of camera orientation on the detectability of wildlife: a case study from north‐western Australia. Remote Sensing in Ecology and Conservation.

Harry A. Moorea,b*, Leonie E. Valentineb, Judy A. Dunlopc, Dale G. Nimmoa

a Institute for Land, Water and Society, School of Environmental Science, Charles Sturt University, Albury 2640, NSW, Australia b School of Biological Sciences, University of Western Australia, Crawley, WA, Australia cDepartment of Biodiversity, Conservation and Attractions, Locked Bag 104, Bentley Delivery Centre, Perth, WA, Australia

Abstract

Camera traps are increasingly used to survey and monitor rare or cryptic species, yet few studies consider how camera orientation influences species detectability, among other metrics such as total detections and likelihood of missing detections. We used these measures to compare the performance of vertically and horizontally orientated camera traps at 46 sites spread over 10,000 km2 in north-west Australia. Data was collected for four taxa: northern quolls (Dasyurus hallucatus), Rothschild’s rock- wallaby (Petrogale rothschildi), feral cat (Felis catus) and varanids (Varanus spp.). Metrics compared included probability of species presence/absence, total detections recorded and likelihood of cameras missing or recording nightly detections. We found camera orientation did not impact camera performance across any metric for northern quolls. By contrast, we found horizontal cameras were more efficient at detecting feral cats and Rothschild’s rock wallabies. They also recorded more detections and were less likely to miss detections than vertical cameras for these species. For varanids, vertical cameras outperformed horizontal cameras across all metrics. Studies that use vertical cameras to collect image data better suited for species or individual identification should consider how target species detectability may be compromised by having a reduced detection zone size. However, horizontally orientated cameras may not always be superior to vertically orientated cameras in terms of species detectability, particularly for laterally compressed species such as lizards.

Introduction

Monitoring the occurrence and abundance of animal populations is a challenge in animal ecology (O'Connell et al. 2010; Pollock et al. 2002; Witmer 2005). An array of sampling techniques have been devised to survey animal populations, each varying in reliability and cost (Rabe et al. 2002; Thompson 2013). Among these, passive infrared triggered cameras—often referred to as camera traps—are one of the most popular (O'Connell et al. 2010; Rowcliffe and Carbone 2008), particularly for mammals (Burton et al. 2015). When compared to traditional trapping techniques such as live trapping, camera traps can be labour and cost effective (Welbourne et al. 2015), less invasive (Meek et al. 2014b) and more effective (De Bondi et al. 2010). However, the use of camera traps to survey an expanding array of fauna groups—now including birds (Brides et al. 2018), and reptiles (Welbourne 2013), has led to variability in the way cameras are deployed (O'Connell et al. 2010). For example, camera trigger type (Rovero et al. 2013) , trigger sensitivity (Heiniger and Gillespie 2018), flash (Meek and Pittet 2013) and focus settings (Welbourne et al. 2019) are all factors which can influence camera performance in terms of the quantity and quality of images collected, and each of these factors must be considered by the camera trap user with a specific research objective in mind. One of the most basic factors influencing camera trap performance is the direction that camera’s face (horizontal or vertical), here termed ‘orientation’ (Meek et al. 2012; Taylor et al. 2014). Aside from influencing the quantity and quality of images collected, camera orientation also affects the angle from which animals are photographed; hence it can affect user capacity to distinguish between individual animals of the same or different species (Muneza et al. 2019).

Individually identifying animals is important because most methods used to estimate animal abundance, such as mark-recapture methods, rely on repeated observations of individuals to derive abundance estimates (Robson and Regier 1964; Seber 1982; Wilson and Delahay 2001). The suitability of image data for individual identification relies on the way cameras are orientated in relation to features that are used to tell individuals apart. In the case of animals where individual markings are located on the dorsal surface, such as goannas (Varanids) (Moore et al. 2020b), numbats (Myrmecobius fasciatus) (Thorn per comms, 2019), and quolls (Dasyurus spp.) (Diete et al. 2016), orientating cameras in a vertical position (i.e., downward facing) above a target individual is likely to produce images that allow individual identification. Further, for species that can be identified from both a horizontal and vertical perspective (e.g. quolls, Dasyurus spp.), vertically orientated cameras may be preferable because they remove the need to capture both left and right lateral sides of an animals to confirm its identity, as is required for study designs using horizontally orientated cameras. However, given many camera traps are designed to be positioned horizontally, positioning them vertically can reduce the area within the ‘detection zone’ of sensors (Welbourne et al. 2016). For example, the detection zone of Reconyx cameras is made up of two horizontal bands, that cover roughly 30% of the cameras field of view

(Reconyx 2020) (Figure S1). When focused toward the horizon, Reconyx cameras are capable of detecting animals moving within these bands at a distance of 30.5 meters (Reconyx 2020), translating to a total detection zone of 324m2 (Meek et al. 2012) . When positioned vertically, the detection range of cameras is limited by the height above the ground at which cameras are positioned, and thus the total detection zone size is substantially reduced.

A reduced detection zone can reduce the number of detections in turn (Meek et al. 2012); for example, Nichols et al. (2017) found horizontally positioned cameras recorded 1.5 times more detections of feral cats (Felis catus) and mustelids (Mustela furo, M. erminea and M. nivalis) in New Zealand, and Taylor et al. (2014) found that horizontally oriented cameras had greater detectability than vertically orientated cameras for several medium-sized mammals in eastern Australia. Low detection rates can be problematic when attempting to estimate abundance, as it may result in sparse data, especially for rare or cryptic species (Thompson 2013). Hence, for species with dorsal markings, a trade-off may exist between i.) increasing detection rates and forfeiting the ability to identify individuals and ii.) risking lower detections rates but maintaining the ability to identify individuals. Studies to explore this trade- off already have so far been limited to few species, none of which include marsupial predators, rock wallabies or reptiles.

Here we compare the effectiveness of vertically and horizontally orientated camera traps at detecting three species of mammal as well as a group of reptiles using paired cameras in north-western Australia. Given horizontal cameras permit a larger detection zone than vertical cameras, for all species, we predicted;

(i) Horizontal cameras would have a higher nightly probability of detecting species than vertical cameras. (ii) Horizontal cameras would record more detections than vertical cameras. (iii) Horizontal cameras would be less likely to miss detections than vertical cameras.

Materials and methods

Study area

This study was carried out across four pastoral stations situated within the Pilbara bioregion in north- west Western Australia. Stations were Indee, Mallina, Pippingarra and Yandeyarra (Figure 3.1). Yandeyarra is also an Indigenous reserve. The study area encompassed the Karayarra and Nyamal Indigenous language groups. Vegetation across all study sites is dominated by hummock grasslands (ESCAVI 2003). This habitat class is characterized by open ground cover comprising spinifex (Triodia.spp) hummock grasses that cover roughly 30 – 50% of the ground surface. Geology is characterised by largely flat sand plains scattered with greenstone ridges and granite (Withers 2000). Climate within the study area is characterized by high temperatures and low annual rainfall. Average daily temperature maximums across the study period ranged from 28.4 C (August 2017) to 44.1C (December 2018) (Australian Bureau of Meteorology 2020).

Figure 3.1 – Study site locations on four pastoral stations, Indee, Mallina, Pippingarra and Yandeyarra, within the Pilbara bioregion in north-west Western Australia. Yandeyarra station is also an Indigenous reserve. The yellow rectangle within the continental inset represents the study area.

Taxa

Taxa targeted in this analysis included two native mammals; the northern quoll (Dasyurus hallucatus), Rothschild’s rock-wallaby (Petrogale rothschildi), one introduced mammalian predator, the feral cat (Felis catus), and four native varanids (Varanus giganteus, Varanus panoptes, Varanus gouldii and Varanus pilbarensis). Target taxa were selected to include species commonly recorded using camera traps within the Pilbara bioregion that vary in size, morphology, and behaviour. To increase statistical power, detections from all varanid species were pooled and classed as the group ’varanids’.

Survey methods

Forty-six camera sites were established within 23 study landscapes, with two sites, separated by at least 300 m, assigned to each landscape. Landscapes were 0.79 km2 in size. Six landscapes were established on Indee station, four on Pippingarra, four on Mallina, and nine on Yandeyarra (Figure 3.1). All study sites were positioned within rocky outcrops using ArcGIS 10.6 (ESRI 2011), as this habitat was most suitable for target taxa (Menkhorst and Knight 2011; Wilson and Swan 2017).

Two Reconyx PC900 Hyperfire covert cameras were deployed at each site. Camera one was attached to a wooden tree stake 1.5m above the ground and was orientated in a vertical position, with the camera lense, and PIR sensors focused directly at the ground surface using a bookshelf bracket (Figure 3.2). Camera two was attached to a second tree stake 50cm above the ground, located 2.5m north from camera one (as recommended by Meek et al. 2012 for medium sized mammals), and was orientated in a horizontal position. The camera lense and PIR sensors were focused at a 10° angle toward the ground surface, in the direction of camera one (South) (Figure 3.2). Horizontal cameras were positioned to face south to avoid intense sunlight exposure to the camera lense. A PVC canister containing 150g of pilchards (fish) was attached to the bottom of the tree stake supporting camera one, within the centre focus of both cameras (Figure 3.2).

Figure 3.2 – Diagram showing A.) vertically orientated camera, B.) horizontally orientated camera and C.) bait canister arrangements at each site.

Sites were sampled for 200 nights each. Twenty four sites were sampled between August 2017 and March 2018, and the remaining twenty two sites were sampled between August 2018 and March 2019. Cameras were set to high sensitivity, and five images were taken at one second intervals per trigger. To assess the rate at which a camera trap was visited by animals, we defined independent detection events for all taxa as detections separated by 15 minutes (Diete et al., 2016). Baits were replenished twice during sampling (approximately once every 70 nights). Five cameras failed, leaving 41 functioning sites. Failures were mainly due to cameras being knocked out of focus by animals. One camera failed as a result of battery failure.

Statistical methods

We examined the effect of camera orientation on three response variables: the nightly probability of detecting a given species, the total number of independent detections over the duration of sampling, and the paired detection likelihood (outlined below). In each case, we used generalized linear mixed models (GLMMs). Each taxa was modelled separately, and camera orientation was included as a fixed effect in all models. To account for non-independence between cameras positioned at the same site (pair), all models included ‘Site’ as a random variable. We report if p < 0.05. Models were fit using the lme4 package (Bates et al. 2007) in R version 3.5.3 (R Core Team 2020).

To model the nightly probability of detecting a species, our response variable was the proportion of nights a species was detected over the sample period, which we modelled as two vectored response variables indicating the number of nights each species was and was not detected over the first 60 nights of sampling. This type of proportion data can be modelled using a binomial distribution, with the number of nights detected/not detected modelled as the number of success and failures in a fixed number of Bernoulli trials (Crawley 2012) The output of these models is the probability that a given species will be detected using a given camera trap orientation on a randomly selected trap night. We restricted this analysis to the first 60 nights to account for declining species detectability over time as a result of reduced bait effectiveness.

We then generated cumulative nightly detectability curves for each taxa based on the detectability estimates derived from these models, using the following formula:

푃 = 1 − (1 − 푝1 ) × (1 − 푝2 ) × (1 − 푝3 ) … (1 − 푝푛 ) where 푃 is the cumulative nightly detection probability, n is number of nights and 푝 is the nightly detection probability. For each taxa, we then calculated the minimum number of nights per site necessary to be 80, 90 and 95% confident that the site is unoccupied: 푁 푚푖푛 = log(훼) /log (푝) where α =0.05 and confidence is equal to 1-p (Kery 2002). Models were fit using the lme4 package (Bates et al. 2007) in R version 3.5.3 (R Core Team 2020).

Next, we modelled the number of independent detections of each species in relation to camera trap orientation. We restricted this analysis to sites with at least one detection on either camera for two reasons. Firstly, because we had already modelled the probability of detecting a species and how it relates to camera orientation. Secondly, because data including all sites were zero inflated, and removal of sites without detections allowed the data to be modelled using a Poisson distribution.

Finally, to further compare the effect of camera orientation on detections, we modelled the likelihood of a horizontal or vertical camera detecting a species on a trap night, given its corresponding paired

58 camera detecting that species on the same trap night. We termed this measure ‘paired detection likelihood’. For each camera, we coded every night a species was detected as one, and every night a species wasn’t detected, but was detected by its corresponding paired camera, as zero. Next, we used a GLMM with a binomial distribution to model the probability that a detection occurred in one camera type (i.e., horizontal or vertical), given it was detected on the other camera type on the same night. Model outputs show the percentage likelihood that either horizontally or vertically orientated cameras will record a species on a given night, given that the camera it was paired with recorded a detection.

Results

A total of 6473 independent detection events were recorded for 105 species over 14,640 trap nights, including 1318 detections for study taxa. The most detected species was the northern quoll (820 detections, horizontal = 426, vertical = 394), followed by varanids (211 detections, horizontal = 80, vertical = 131), Rothschild’s rock-wallaby (165 detections, horizontal = 119, vertical = 46), and feral cats (127 detections, horizontal = 89, vertical = 38).

Detection probability

Camera orientation had a significant effect on nightly probability of detection for all species except for northern quolls (Figure 3.3). Horizontal cameras had higher nightly detection probabilities for feral cats and Rothschild’s rock wallabies, and vertical cameras had a higher nightly detection probability for varanids (Figure 3.3). For horizontal cameras, northern quolls required the least number of nights to be 95% confident of sites being unoccupied (24 nights, 95% CI 14 – 41 nights), followed by Rothschild’s rock wallabies (44 nights, 95% CI 50 –141 nights), feral cats (69 nights, 95% CI 44 –110 nights) and varanids (85 nights, 95% CI , 51– 141 nights) (Figure 3.4). For vertical cameras, northern quolls required the least number of nights to be 95% confident of absence (32 nights, 95% CI 19 –57 nights), followed by varanids (49 nights, 95% CI 32 – 78 nights), Rothschild’s rock wallabies (129 nights, 95% CI 67 – 248 nights), and feral cats (142 nights, 95% CI 78 – 258 nights) (Figure 3.4).

Figure 3.3 – Output from generalized linear mixed-effects models testing the effect of camera orientation on nightly detectability for northern quolls (Dasyurus hallucatus), feral cats (Felis catus), Rothschilds rock wallabies (Petrogale rothschildi), and varanids (Varanus spp.). Black points represent nightly detection probability and blue error bars represent 95% confidence intervals. Treatment (horizontal or vertical camera orientation) was used as the fixed effect. Site was included as a random effect. *** P < 0.001; ** P < 0.01; * P < 0.05.

Figure 3.4 – Cumulative nightly detection probabilities using horizontally orientated camera traps (red line, red shading=95% confidence intervals) and vertically orientated camera traps (blue line, blue shading=95% confidence intervals) for northern quolls (Dasyurus hallucatus), feral cats (Felis catus), Rothschilds rock wallabies (Petrogale rothschildi), and varanids (Varanus spp.). The x axis represents the number of consecutive trap nights at a site.

Total independent detections

Camera orientation had a significant influence on the total number of detections recorded at sites for all taxa except northern quolls (Figure 3.5). For feral cats, the predicted total number of independent detections for horizontal cameras (2.48, 95% CI 1.80 – 3.43) was over double the predicted total detections for vertical cameras (1.04, 95% CI 0.70 –1.57). Similarly, for Rothschild’s rock wallaby, the predicted total number of independent detections for horizontal cameras (3.81, 95% CI 2.36 – 6.15) was almost three times higher than the predicted total detections for vertical cameras (1.34, 95% CI 0.79 – 2.28). Total predicted detections for varanids was higher for vertical cameras (2.52, 95% CI 1.72 – 3.69) than for horizontal cameras (1.54, 95% CI 1.02 – 2.30) (Figure 3.5).

Figure 3.5 – Output from generalized linear mixed-effects models testing the effect of camera orientation on the total number of detections recorded for northern quolls (Dasyurus hallucatus), feral cats (Felis catus), Rothschilds rock wallabies (Petrogale rothschildi), and varanids (Varanus spp.). Black points represent total number of predicted detections and blue error bars represent 95% confidence intervals. Treatment (horizontal or vertical camera orientation) was used as the fixed effect. Site was included as a random effect. *** P < 0.001; ** P < 0.01; * P < 0.05

Paired detection likelihood

GLMM results indicated paired detection likelihood was significantly influenced by camera orientation for all species except for northern quolls (Figure 3.6). Horizontal cameras had a 82.1% (95% CI 73.2%, – 88.5%) probability of detecting feral cats if a feral cat was detected on a paired camera on the same night, and vertical cameras had a 35.6% probability (95% CI 26.5% – 45.9%). Similarly for Rothschild’s rock wallaby, we found horizontal cameras had a 89.2% probability (95% CI 81.7% – 93.8%) of detecting an animal if it was detected on a paired camera on the same night, but vertically orientated cameras had only a 34.6% probability of detecting an animal if it was detected on a horizontal camera (95% CI 25.2% – 45.3%). Horizontal cameras had a 51.9% probability (95% CI 43.5% – 60.3%) of detecting varanids if they were detected by a paired camera on the same night. Vertical cameras had a 75.0% probability (95% CI 67.1% – 81.5%) of detecting varanids if an animals was detected by a paired camera on the same night (Figure 3.6).

Figure 3.6 – Output from generalized linear mixed-effects models with a binomial distributions to test the effect of camera orientation on paired detection likelihood for northern quolls (Dasyurus hallucatus), feral cats (Felis catus), Rothschilds rock wallabies (Petrogale rothschildi), and varanids (Varanus spp.). Paired detection likelihood (one for detection recorded and zero for detection missed) was used as the response variable. Treatment (horizontal or vertical camera orientation) was used as the fixed effect. Site was included as a random effect. Model outputs show the probability a horizontally or vertically orientated camera will record a detection given the camera it was paired with records a detection. *** P < 0.001; ** P < 0.01; * P < 0.05.

Discussion

Camera traps are the primary means of surveying an increasingly diverse range of animal taxa (Burton et al. 2015; Rowcliffe and Carbone 2008; Trolliet et al. 2014), yet studies rarely consider how camera orientation affects survey data (Rovero and Zimmermann 2016). Our results suggest camera orientation does not significantly affect camera performance in terms of any of the detection metrics measured in this study for northern quolls. Conversely, we found horizontally orientated cameras were more efficient at detecting both feral cats and Rothschild’s rock-wallabies, and vertically orientated cameras were better at detecting varanids.

Species detectability

As horizontal cameras have a larger detection zone—an important factor in maximising species detectability (Fancourt et al. 2018; Meek et al. 2015)—we predicted they would have higher nightly detection probabilities than vertical cameras. Our results supported this prediction for feral cats and for Rothschild’s rock wallabies, concurring with findings of Nichols et al. (2017) who found a similar result using cats, and Taylor et al. (2014) using a smaller species of macropod.

In addition to detection zone size, attractiveness of different mammal species to bait canisters may also have influenced this result. Bait type has previously been shown to alter species detectability (Diete et al. 2016), and bait preferences vary between species (Paull et al. 2011). In our study, we used a meat- based bait and therefore vertical cameras may have been less likely to detect Rothschild’s rock-wallaby, which is less likely than carnivorous species to have been drawn to within close proximity of the bait. Cameras themselves may also act as a deterrent or attractant to species, by emitting novel visual and acoustic cues which can be received by animals (Meek et al. 2014a). While still poorly understood, the capacity of animals to receive and respond to these cues is likely to vary with species (Meek et al. 2016). Because vertical cameras are positioned directly above the bait lure, cues emitted by the vertical cameras may influence the likelihood of species entering the vertical detection zone, and ultimately the probability of presence being recorded on vertical cameras.

Interestingly, we found camera orientation did not significantly influence whether northern quolls were detected at sites or not, nor the number of nights required to be 95% confident of true absence. We suspect this result is reflective of bold foraging behaviour; that is, despite the novelty of the camera trap set-up, quolls readily approach bait canisters at close proximity, and thus are equally likely to breach both horizontal and vertical camera detection zones. This behaviour could be expected from northern quolls in the Pilbara, given they are generally not considered shy or neophobic, and are frequently reported entering people’s homes (Brierly per comms, 2019).

Previous studies using camera traps to detect lizards have employed horizontal (Ariefiandy et al. 2013) and vertical cameras (Moore et al. 2020b; Richardson et al. 2018; Welbourne et al. 2015), but our study is the first to compare how camera orientation influences lizard detectability. Contrary to our prediction, we found vertical cameras were more likely to record presence for varanids and required less survey effort to achieve a 95% probability of true absence. The most likely explanation for this result may be that vertically orientated cameras are more suited to detecting a lizard morphology. Squamates, including lizards, have a laterally compressed body type, with a larger dorsal, than lateral, surface area (Thompson and Withers 1997). This type of morphology is more visible from a top-down perspective when compared to a horizontal perspective, and thus likely more exposed to vertical camera sensors, despite the larger detection zone of horizontal cameras.

Total detections

Maximising total number of detections is a priority in many ecological studies, because the total number of species detections can determine the variety and power of techniques available for subsequent statistical analysis (Forcino et al. 2015; Foster and Harmsen 2012; Lashley et al. 2018). We predicted horizontal cameras would record a significantly greater number of total detections than vertical cameras for all species, and we found this to be true for feral cats and Rothschild’s rock-wallabies, concurring with the results of previous studies (Nichols et al. 2017; Taylor et al. 2014). Like other behaviourally cryptic species, rock-wallabies and feral cats can be difficult to detect using camera traps, even at sites where presence has previously been confirmed (Bluff et al. 2011; Comer et al. 2018; Glen et al. 2016; Turpin et al. 2018). Maximising total number of detections by orientating cameras appropriately may therefore be especially important for these species, given the risk of small samples sizes precluding the use of advanced analysis techniques already being high. For northern quolls and varanids, results were consistent with our probability of presence analysis ― camera orientation did not influence the total number of detections recorded for northern quolls, and vertical cameras recorded a higher number of varanid detections than horizontal cameras. For quolls, this result is again the likely product of bold foraging behaviour. For varanids, increased total detections on vertical cameras is likely due to their laterally compressed morphology.

Paired detection likelihood

We found that vertical cameras were significantly more likely to miss detections than horizontal cameras for feral cat and Rothschild’s rock wallabies (46.5% and 54.6% respectively). This result was unexpected, as we presumed that animals would not have been able to enter a vertical detection zone without having already entered the horizontal detection zone. A possible explanation for missed detections by horizontal cameras could be the distance at which they were placed from the bait cannister. For example, Taylor et al. (2014) positioned their horizontal cameras two metres away from bait cannisters and found they outperformed vertical cameras, whilst Smith and Coulson (2012) used a distance of three metres and recorded the opposite result using the same species . We used a distance of 2.5 meters, as recommended by Meek et al. (2012) for detecting medium sized mammals. Using a reduced distance may have limited the number of detections missed by horizontal cameras, although it may also have reduced total detections by reducing the detection zone immediately surrounding bait cannisters. Another explanation for missed horizontal detections may be that even though camera sensitivity was set as high as possible, camera sensitivity may still not have been high enough to always detect species moving within the detection zone. For example Heiniger and Gillespie (2018) found Reconyx PC850 cameras with similar sensor capabilities to cameras used in this experiment frequently missed detections when bait was being removed 1.5 metres away by medium sized species such as the northern quoll and northern brown bandicoot (Isoodon macrourus). We also cannot discount that animals evaded detection from horizontal, or vertical cameras, by moving under or over camera detection bands (Apps and McNutt 2018; Meek et al. 2014b), which cover less than half of the cameras field of view for Reconyx cameras (Reconyx 2020) (Figure S1).

Conclusion

We found that horizontally orientated cameras performed better than vertical cameras in terms of detecting feral cat and Rothschild’s rock wallabies, but no significant differences were observed for northern quolls. We also found vertically orientated cameras were better at detecting varanids. Given the clear differences in detectability performance, it is important to consider the practicality of using horizontal or vertical cameras on a species to species basis. For example, whilst orientation made no difference to the probability of cameras detecting northern quolls , wildlife managers may still be inclined to use vertically orientated cameras given they generally capture image data more suited to individually identifying northern quolls, which are most easily distinguished from their dorsal surface (although see Hohnen et al. 2013). By contrast, individual cats and possibly rock wallabies can be distinguished from their ventral surfaces, which are better captured using horizontal cameras (Bengsen et al. 2012; McGregor et al. 2015b), and thus there would be no reason to reduce detectability by using vertical cameras for these species. Vertical cameras were better at detecting Varanus spp, and have been shown to be well suited to species and individual identification for lizards (Moore et al. 2020b; Welbourne 2013; Welbourne et al. 2019), and therefore would be a more logical approach for lizards in most situations.

Overall our results demonstrate that neither horizontally or vertically orientated cameras are best suited to all taxa, and both orientations are susceptible to missing detections at times. We recommend future studies that use vertical cameras to collect image data better suited for species or individual identification should consider how target species detectability maybe compromised by having a reduced detection zone size.

Acknowledgments

Data collection was assisted by Sian Thorn, Darcy Watchorn, Rainer Chan, Daniel Bohorquez Fandino, Jacob Champney, Hannah Kilian, Mitch Cowan, Gasten Stewart, John Stewart and Nathaniel Herbert. Camera-trap deployment was also completed with the assistance of the Yandeyarra Indigenous ranger program facilitated by Pip Short and Greening Australia. Technical support was provided by Neal Birch, Brent Johnson, Hannah Anderson, Russell Palmer, Alicia Whittington and Jo Kuiper from the Western Australian Department of Biodiversity, Conservation and Attractions (DBCA), as well as Deb Noy from Charles Sturt University. Stephen van Leeuwen from DBCA provided assistance with the project conception and also provided support throughout. Equipment and operational costs were provided by DBCA, Roy Hill and Charles Sturt University. Roy Hill also covered the costs of flights, fuel and freight. Harriet Davie from Roy Hill provided technical support throughout, along with the Roy Hill rail team in Port Hedland. We thank Colin Brierly, Betty Brierly and Graham for providing access to Indee station, Ben and Lindsey for access to Mallina and Troy Eaton for access to Pippingarra. Belinda Barnett from BHP assisted with site access. H.A.M is supported by a scholarship from the Institute of Land, Water and Society and Charles Sturt University. L.E.V. was funded by the Australian Government’s National Environmental Science Program through the Threatened Species Recovery Hub. D.G.N. was supported by an Australian Research Council Early Career Researcher Award (DECRA). This project was supported by the Holsworth Wildlife Research Endowment – Equity Trustees Charitable Foundation and the Ecological Society of Australia, as well as the Western Australian Department of Biodiversity, Conservation and Attractions as well as environmental offsets and public good funding provided by BHP, Rio Tinto, Atlas Iron, Fortescue Metals Group, Roy Hill, Process Minerals International, Metals X and Main Roads Western Australia. Research ethics was granted through the Charles Sturt University Animal Ethics Committee (permit Number A17031), and the Western Australian Department of Biodiversity, Conservation and Attractions (permit Number 08- 002376-1). We are grateful to two reviewers and the topic editor who provided valuable and detailed comments on the manuscript and improved it to the version presented here.

Chapter 4 – Habitat configuration is more important than habitat amount for a threatened marsupial predator in a naturally fragmented landscape

Harry A. Moorea,b*, Damian R. Michaela, Judy A. Dunlopc, Leonie E. Valentineb, Dale G. Nimmoa aSchool of Environmental Science, Institute for Land, Water and Society, Charles Sturt University, Albury, NSW, Australia bSchool of Biological Sciences, University of Western Australia, Crawley, WA, Australia cDepartment of Biodiversity, Conservation and Attractions, Locked Bag 104, Bentley Delivery Centre, Perth, WA, Australia

Abstract

Context Habitat amount and configuration are tightly linked but nonetheless independent landscape characteristics that are often confounded in ecological studies. Differentiating the effects of each is critical for conservation, because the mechanisms by which they influence populations are distinct. While the preponderance of studies that have measured the effects of habitat amount and configuration separately have found the former to be more important, a subset of these studies suggest habitat configuration can increase in importance when habitat amount is low (< 20-30 %).

Objectives

We aimed to test the independent effects of habitat amount and configuration on the occupancy and abundance of an endangered marsupial predator, the northern quoll (Dasyurus hallucatus), which persists in fragmented rocky landscapes where habitat amount is naturally low. We also tested the influence of other important landscape properties, such as landscape configuration, and environmental gradients.

Methods

Northern quoll responses were tested across 22 study landscapes, deliberately selected such that measurements of habitat amount and configuration were not correlated. Northern quoll occupancy and abundance was estimated at each landscape using data collected from remote sensing cameras, and a combination of occupancy and n-mixture models.

Results

Spatial configuration of rocky habitats was more important than amount of habitat for predicting quoll occurrence and abundance at the landscape scale; northern quolls were less abundant in landscapes that were more fragmented. In addition, northern quolls favoured areas with high topographical ruggedness, and areas that received more rainfall.

Conclusions

Our results support the findings of studies which have found the effects of habitat configuration can be strongest when habitat amount is low, and suggest that the division of rocky habitat through either the construction of mining infrastructure (e.g road, rail lines) or other means is likely to have important negative impacts on northern quoll populations.

Introduction

Habitat loss and fragmentation are major threats to biodiversity worldwide (Crooks et al. 2017; Laurance et al. 2012). Although they often occur in tandem, and their impacts are frequently confounded, it is widely recognised that habitat loss and fragmentation are two separate phenomena (Fahrig 2017; Groombridge et al. 2002); habitat loss refers to the removal of habitat, while habitat fragmentation refers to the breaking apart of a given amount of habitat, and is conceptually independent of habitat loss (sensu Fahrig 2003). Studying the impacts of habitat loss and fragmentation as they occur is difficult due to (i) the large spatial and temporal scales over which both phenomena occur, and (ii) the confounded nature of habitat loss and fragmentation in real landscapes (Ewers and Didham 2006; Fahrig 2017; Villard and Metzger 2014). Therefore, in ecological research, the independent effects of habitat loss and fragmentation are often studied by measuring species or community responses to habitat amount and habitat configuration in replicate study landscapes (Bennett et al. 2006; Fardila et al. 2017), whereby differences in habitat amount act as a surrogate for habitat loss, and variation in habitat configuration for a given amount of habitat represent different levels of fragmentation (Villard and Metzger 2014).

Differentiating the importance of habitat amount and configuration to species is critical from a conservation perspective, because the mechanisms by which they influence populations are distinct (Villard and Metzger 2014). Habitat amount determines the amount of resources that are present within a landscape, and as such is often positively correlated with species occurrence and abundance (Bennett et al. 2006). By contrast, habitat configuration can determine the accessibility of resources (food, shelter, mates, ect.) to species (Fahrig 2003) ― in landscapes where habitat is disaggregated, an animal’s access to resources can be restricted to that contained by a single patch, which may be insufficient for species with large resource requirements (Zanette et al. 2000). Animals that are forced to travel to multiple patches to access adequate resources are likely to incur greater costs related to energy expenditure and predation (Andrén et al. 1985; Hargis et al. 1999). Habitat configuration also determines the extent to which species are exposed to edge effects (Laurance and Yensen 1991).

Most studies that have tested the independent effects of habitat amount and configuration in tandem suggest the former is stronger (Bennett et al. 2006; Fahrig 2017), yet few of these have been conducted in landscapes that are naturally fragmented― where habitat patches are embedded in a relatively unmodified matrix that is nonetheless unsuitable to a given species (Driscoll 2005). Here, habitat patches generally arise through geological or climatic processes that leave them exposed or restricted (e.g. rocky outcrops), similar to oceanic islands (Fitzsimons and Michael 2017). As such, the amount of habitat within these landscapes may be lower than is often observed in landscapes fragmented anthropocentrically. This has important implications for predicting species incidence and abundance in naturally fragmented landscapes as a consequence of the ‘fragmentation threshold’ effect (Bascompte

72 and Solé 1996; King 1999; Lande 1987)― where the impacts of landscape configuration are amplified when habitat cover is less than 20-30% of a landscape. For example, using simulations, Fahrig (1998) found fragmentation only had an important effect on species survival when breeding habitat cover was less than 20%. Improving our understanding of how the effects of habitat amount and configuration interact in naturally fragmented landscapes is likely to enhance our capacity to protect the species that occur here.

In addition to habitat amount and configuration, Bennett et al. (2006) outline two other important landscape properties that can influence species distributions: landscape composition and environmental gradients. Landscape composition refers to the types of elements present within a landscape and their relative proportions (Bennett et al. 2006). One important component of landscape composition is the type and structure of the matrix surrounding primary habitat within a landscape (Ewers and Didham 2006; Kupfer et al. 2006). This is because the matrix can influence species inter-patch movement (Bender and Fahrig 2005), as well as the availability of supplementary resources (Öckinger et al. 2012), and the intensity of edge effects (Haynes and Cronin 2006). Matrix effects may be particularly strong in fire-prone landscapes, where the structure and composition of matrix vegetation is in a constant state of flux as a result of being periodically incinerated (Nimmo et al. 2019). Environmental gradients include factors such as rainfall, and topography, both of which have been demonstrated to influence species landscape occupancy (McDonald et al. 2015a; Moore et al. 2019; Nimmo et al. 2013).

In this study, we test the independent effects of habitat configuration, in conjunction with habitat amount, landscape composition, and environmental gradients, on the occupancy and abundance of an endangered marsupial predator in naturally fragmented landscapes. The northern quoll (Dasyurus hallucatus) previously occurred widely across northern Australia, but has recently suffered broad scale declines across most of its range (Braithwaite and Griffiths 1994; Moore et al. 2019), largely as a result of changing fire regimes (Woinarski et al. 2011b), predation (Cremona et al. 2017a; Oakwood 1997), and the introduction of an invasive cane toad (Rhinella marina) that can be fatally toxic when consumed (Oakwood 2004c). Understanding the effects of habitat configuration on northern quolls is of critical importance for the species, particularly in the Pilbara bioregion ― the only section of the northern quolls range yet to be invaded by cane toads, and often regarded as a population stronghold (Cramer et al. 2016). Here, northern quolls occur within naturally fragmented landscapes comprising small patches of granite and iron-stone formations formed over 2.6–2.7 billion years (Withers 2000) , surrounded by a matrix of fire-prone spinifex grasslands. Sections of this rocky habitat are currently under threat, or have already been destroyed through iron ore mining activity (Cramer et al. 2016); a key threat for northern quolls in the region (Hill and Ward 2010; Woinarski et al. 2014). Defining areas of critical habitat is now a research priority for the Pilbara northern quoll in order to help target ongoing management efforts to conserve this population (Cramer et al. 2016).

Materials and methods

Study area

This study was carried out across a 6000 km2 area within the Pilbara bioregion in north-west Western Australia (Fig 4.1), encompassing the Kariyarra and Nyamal Indigenous language groups. The study area also overlapped with three cattle stations (Indee Station, Mallina Station, Pippingarra Station), and an Indigenous Reserve (Yandeyarra Indigenous Reserve). Vegetation across the study area is dominated by spinifex grasslands, with sparse tree cover comprised of mulga (Acacia aneura), snakewood (Acacia xiphophylla) and snappy gum (Eucalyptus leucophloia). Geology is characterised by largely flat sand plains scattered with banded ironstone ridges and granite outcrops (Withers 2000). Average daily temperature maximums across the study period ranged from 28.4 C (August 2017) to 44.1C (December 2018) (Australian Bureau of Meteorology 2019).

Study design and landscape selection

Measures of fragmentation, such as the amount of habitat edge or number of habitat patches within a landscape are typically correlated with the amount of habitat in a landscape (Fahrig 2003). Here, we aimed to capture non-correlated gradients of habitat amount and configuration (correlation coefficient < 0.5), such that they could be tested independently. We achieved this by selecting landscapes across the habitat amount gradient that varied in the configuration of habitat. First, we selected 60 candidate landscapes comprising rocky outcrops (habitat) embedded within a matrix of spinifex grasslands (non- habitat) from across the study area using QGIS 3.14.1 (QGIS 2020). Landscapes were circular and covered a total area of 75 hectares (1 km diameter), which is sufficient to cover the home range of multiple female northern quolls (Cowan et al. 2020a; Hernandez-Santin et al. 2020). We chose to use female home ranges as a guide in establishing study landscapes given they are far less variable in size than are male home ranges (Cowan et al. 2020a; Hernandez-Santin et al. 2020). Second, candidate landscapes were categorised based on the amount of rocky habitat within them. Landscapes with < 15% habitat coverage were classed as ‘low habitat amount’; landscapes with 15–25% habitat were classed as ‘medium habitat amount’, and; landscapes with >25% habitat coverage were classed as ‘high habitat amount’. The maximum habitat amount in any landscape was 37%. We acknowledge that our definition of high habitat amount (>25%) is lower than as defined in other studies (Fahrig 2017), however, given the sparsity of rocky habitat within the study area as a whole, we maintain this classing is justified.

Next, within each habitat amount class, a total of eight study landscapes were chosen to represent different configurations of habitat within that class, ranging from landscapes in which most habitat was

74 aggregated into a few patches, to landscapes in which habitat was dispersed among many patches, characterised by large amounts of edge habitat (Fig 4.2). This resulted in 24 landscapes that varied in habitat amount, and varied in configuration (amount of edge habitat) for a given category of habitat amount. Two landscapes were later excluded from further analysis (medium habitat amount =1, high habitat amount =1) because they were located in close proximity to the installation of a granite quarry, or a major river (factors which may bias results), leaving a total of 22 study landscapes.

Site selection

Within each landscape, we established 5 sites. Sites were chosen to represent gradients of outcrop size (basal area), shape (area-edge ratio), and geomorphology within landscapes. Sites were separated from one another by a minimum of 200m. At each site, we deployed a Reconyx PC900 Hyperfire passive infrared triggered camera trap (Reconyx 2020). Cameras were orientated vertically, and attached to a wooden tree stake 1.5m above the ground, with the camera lens and PIR sensors focused directly at the ground surface using a right-angle bracket (see Moore et al. 2020c). All vertical cameras were deployed in rocky habitat, given it is the preferred habitat of the Pilbara northern quoll (Cramer et al. 2016; Moore et al. 2021). Vertical cameras were ideal for capturing unique spot patterns located on the dorsal surface of northern quolls (Diete et al. 2016; Hohnen et al. 2013; Moore et al. 2021), which we used to identify individual animals (see ‘detection data’ section). These cameras were deployed as part of a broader monitoring program within which this study was embedded (Moore et al. 2021; Moore et al. 2020c), and such were not used in the current study. Hereafter, any mention of sites will only be in reference to sites where vertically orientated cameras were deployed. All cameras were set to high sensitivity, and five images were taken at one second intervals per trigger. Sites were sampled for 60 nights in the Pilbara dry season (August – November) and 60 nights in the wet season (December – March). We used 60 nights as this exceeded the sampling effort required to be 95% confident of northern quoll absence at sites using vertical cameras (Moore et al. 2020c).

Figure 4.1 – Study landscapes within the Pilbara bioregion in north-west Western Australia. A) location of the study area in north-west Western Australia and an image of a typical study landscape. B) aerial depictions of the original 24 study landscapes. Area’s shaded red represent rocky habitat, and white areas represent the spinifex grassland matrix. C) habitat amount and D) edge density in each of the original 24 study landscapes. * indicates landscapes that were excluded from final analysis.

Predictor variables

Study landscapes were characterized according to the four landscape properties outlined by Bennett et al. (2006) (Table 4.1); habitat amount, habitat configuration, landscape composition, and environmental gradients. Given the close association between rocky habitats and quoll occurrence and abundance (Hernandez-Santin et al. 2016; Moore et al. 2021), we considered the aerial extent of rocky habitat as a measure of habitat amount for the northern quoll. Habitat configuration was measured as the length of edge habitat per ha within each study landscape. While a number of more complex measures of landscape fragmentation were also considered (e.g., aggregation index, number of patches; (Kupfer 2012)), preliminary analysis revealed that edge density was the most parsimonious measure of configuration in terms of explaining the occupancy and abundance of the northern quoll based on Akaike Information Criterion (AIC). The ecological effects of habitat edges within fragmented landscapes have long been recognised (Fahrig 2003), and edge based-metrics provide an effective measure of landscape fragmentation (Zeng and Wu 2005). Further, our previous research in the same study region showed northern quolls avoid habitat with high amounts of edge (Moore et al. 2021).

Landscape composition was measured as the proportion of spinifex grassland comprising the landscape matrix that had been recently burnt. Fire has been shown to negatively influence northern quoll populations across their range, by reducing recruitment rates (Begg 1981; Griffiths and Brook 2015), and potentially increasing predation risk (Jolly et al. 2018a; Oakwood 1997). In the Pilbara, northern quolls have been shown to avoid areas that have been recently burnt (Hernandez-Santin et al. 2016). To measure the impact of fire in the matrix, we measured the proportion of landscapes burnt in the previous three years to sampling, using data collected from the Northern Australian Fire Information database (NAFI 2020). Three years was used as the cut-off for recently burnt spinifex based on a previous study from the nearby western desert which found post fire spinifex cover reach 20% after three years (Burrows et al. 2009).

We used topographical ruggedness and total rainfall in the previous Pilbara wet season (Dec–Mar) as measures of environmental gradients in this study. Topographical ruggedness is an important predictor of northern quoll occurrence across their range (Begg 1981; Braithwaite and Griffiths 1994; Pollock 1999), including in the Pilbara (Molloy et al. 2017; Moore et al. 2019), probably because topographically rugged habitat is visited less by predators like dingoes (Canis dingo) and feral cats (Felis catus) (Hernandez-Santin et al. 2016). To measure topographical ruggedness at each landscape, we used elevation data collected at 30*30m resolution (Geoscience Australia 2008). Then for each cell within a landscape, topographical ruggedness was measured as the difference in elevation between it and the eight cells surrounding it (following Reily et al. 1999). A landscapes topographical ruggedness value was measured as the mean ruggedness value of all cells within a landscape. Previous wet season rainfall was measured as rainfall during the months December – March prior to sampling. Rainfall data

77 was sourced from the Australia Bureau of Meteorology (Australian Bureau of Meteorology 2020). Northern quoll persistence is typically higher in areas that receive higher amounts of rainfall (Heiniger et al. 2020; Moore et al. 2019; Radford et al. 2014; Woinarski et al. 2008). We used previous wet season rainfall specifically because it is likely an important factor in determining the success of Pilbara northern quoll recruitment (occurring Jan–March), a factor that is critical to northern quoll population persistence (Moro et al. 2019).

Table 4.1 – Habitat variables used to characterize naturally fragmented landscapes occupied by northern quolls (Dasyurus hallucatus) in the Pilbara bioregion, Western Australia.

Habitat Landscape Description Justification variables property Habitat amount Habitat amount The proportion of landscapes Habitat amount can determine the amount of covered by rocky habitat resource available to species within a landscape, mapped using google satellite which can determine the total number of imagery in QGIS. Rocky individuals a landscape can support (Bennett et patches with an area < 50m2 al. 2006). were not included. Edge density Habitat The total perimeter of rocky Edge based metrics are strongly correlated with configuration habitat within study landscape fragmentation ― when habitat amount (fragmentation) landscapes (km2) divided by is constant, fragmented landscape have more landscape area (75 ha). habitat edge than non-fragmented landscapes Mapped using google satellite (Fahrig 2017). Increasing habitat fragmentation imagery in QGIS. Rocky can negatively influence species through ‘edge patches with an area < 50m2 effects’ (Laurance and Yensen 1991), as well as were not included. through increased exposure to the matrix, where prey specie can be more vulnerable to predation (Andrén et al. 1985). Proportion recently Landscape The proportion of landscapes Research has shown that fire can negatively burnt composition burnt within in the previous 3 impact northern quoll populations, by reducing years of sampling. Fire data recruitment rates (Griffiths and Brook 2015) and was collected from the potentially increasing predation rate (Oakwood Northern Australian Fire 1997). In the Pilbara, northern quoll are Information database (NAFI negatively associated with areas that have been 2020). recently burnt (Hernandez-Santin et al. 2016). Topographical Environmental Calculated from the difference High quality northern quoll habitat is often ruggedness gradients in elevation between a cell and associated with areas that are topographically the eight cells surrounding it rugged (Braithwaite and Griffiths 1994; Moore et (following Reily et al. 1999). al. 2019; Pollock 1999). Elevation data was sourced from Geoscience Australia, at 1 second resolution (~ 30m). Rainfall in previous Environmental Total rainfall recorded at Northern quoll persistence is typically higher in wet season gradients landscapes during the most areas that receive higher amount of rainfall recent wet season (December (Moore et al. 2019; Radford et al. 2014; – March). Rainfall data was Woinarski et al. 2008). Wet season rainfall is sourced from the Australia likely an important factor in determining the Bureau of Meteorology success of northern quoll recruitment in Pilbara, (Australian Bureau of which is critical to northern quoll population Meteorology 2020) persistence (Moro et al. 2019).

Detection data

Camera trap detections separated by at least 15 minutes were defined as independent detection events, following Diete et al. (2016), who found northern quoll detections separated by between 10 and 15 minutes were as likely to be a different individual as they were the previous individual. We attempted to individually identify quolls within each detection event, unless there were no images within a detection event that were suitable for individual identification. Images that contained less than roughly 30% of the northern quoll’s dorsal surface were immediately excluded from further image analysis unless they contained other distinctive features that could be used to identify individuals (e.g. unusual scarring patterns, missing ears etc.).We used unique spot patterning as well as scarring located on the dorsal surface of animals to identify individuals from camera trap imagery, following an identical process to that described in Moore et al. (2021) and Moore et al. (2020b). Data from all cameras within a landscape were collapsed for each trap night ― for each trap night within each landscape, we recorded if quolls were detected across all cameras, and the number of individual quolls recorded.

Statistical analysis

We used measures of occupancy and abundance to model northern quoll responses to landscape properties. At an appropriate scale, occupancy and abundance should be correlated (MacKenzie and Nichols 2004). However, it’s important to note that each measure distinct population dynamics ― occupancy measures the proportion of an area occupied by a species, while abundance measures the numbers of individuals in that area. As such, occupancy and abundance may not always covary in response to landscape properties (MacKenzie and Nichols 2004).

All statistical analysis was conducted in R version 4.0.2 (R Core Team 2020), using the unmarked package (Fiske and Chandler 2011). No variables included together in any model shared pairwise correlations > 0.6 (Table S4.1). All model fixed effects were scaled prior to analysis to improve model stability and allow direct comparisons between model coefficients (Harrison et al. 2018). Separate models were fit for the dry season and wet season to account for differences in occupancy and abundance resultant from varying levels of productivity between seasons and breeding events. Three cameras failed in the dry season sampling period, and three cameras failed in the wet season sampling period. To account for differences in survey effort as a result of camera trap failures, all models included total trap nights as an offset term, where the maximum was 300 (5 sites per landscape * 60 nights per season).

Occupancy

A total of 17 models were fit for each season. Given our primary focus on habitat amount and configuration, all models included habitat amount and edge density, either as additive terms or with an interaction term between the two. All other combinations of habitat variables were tested, including a null model (Table S4.2). No habitat variables were included in the detectability component of occupancy, firstly, because the influence of habitat variables on northern quoll detectability was not a focus in this study, and secondly to minimise the number of parameters in models given our low sample size. Models were ranked based on their AIC scores, corrected for small sample size (AICc) (Burnham and Anderson 2002). Models with ΔAICc <2 were considered useful (Burnham and Anderson 2004) . Goodness of fit was assessed for the most parsimonious models using a Chi-square test. Predictor variables were regarded as being influential if their 95% confidence intervals did not overlap zero (Nakagawa and Cuthill 2007).

Abundance

An identical process as described above was repeated using N-mixture models (Royle 2004), except with input data which comprised count data (number of northern quolls detected at a site for each survey night). We also tried to estimate quoll abundance using mark-resight models as part of our preliminary analyses. However, results from these models indicated that they did not perform well, probably as a result of the sparseness of the data. As such, we restricted our methods for estimating abundance to n- mixture models. N-mixture models are an extension of occupancy models that incorporate detection probabilities into abundance estimates using a combination of binomial and Poisson distribution models from count data (Royle 2004). We fit n-mixture models using the package unmarked. Preliminary analysis indicated that there were many nights across landscapes where quolls were not detected, leading to zero inflation. Models that do not account for zero inflation in count data can produce biased parameter estimates (Potts and Elith 2006). To overcome this constraint , we used a zero-inflated Poisson distribution in n-mixture models (Fiske and Chandler 2011). Model selection was again preformed using AICc.

Results

Across all landscapes and seasons, we recorded a total of 1004 independent northern quoll detection events, a slight majority of which were recorded in the wet season (56.4%). Most detection events (70.0%) contained images suitable for individual identification. In total, 136 individual northern quolls were identified. Individual locations where northern quolls were detected were uploaded to the publicly available NatureMap repository. No models tested for Goodness of fit showed signs of poor fit.

Table 4.2 – Summary tables from occupancy and n-mixture models, used to test northern quoll (Dasyurus hallucatus) responses to naturally fragmented landscapes in the Pilbara bioregion , Western Australia.

Variable Estimate SE lower CI upper CI Occupancy Intercept 9.76 4.68 -9.08 9.28

Area -1.49 0.49 -0.98 0.95 Edge -6.24 2.00 -3.98 3.86 Proportion recently burnt -5.61 1.21 -2.42 2.31 Topographical ruggedness 13.13 3.55 -6.83 7.09

Dry seasonDry

Intercept -6.72 0.88 -8.45 -5.00 Area 1.89 1.05 -0.16 3.94 Edge -2.24 1.01 -4.21 -0.27 Area*Edge 1.53 0.78 0.01 3.06

Wet season N-mixture model

Intercept -5.13 0.33 -5.78 -4.47 Area -0.16 0.18 -0.52 0.19 Edge -1.68 0.35 -2.38 -0.99 Topographical ruggedness 0.98 0.10 0.78 1.18

Dry seasonDry Intercept -4.62 0.43 -5.46 -3.78

Area 0.11 0.17 -0.21 0.44 Edge -1.81 0.34 -2.48 -1.14 Topographical ruggedness 0.52 0.16 0.21 0.84 Previous wet season 0.74 0.14 0.47 1.01

Wet season

Occupancy

In the dry season, the best occupancy model included habitat area and edge density as additive terms, as well as topographical ruggedness, and the proportion of landscape that has been recently burnt (Table S3.3), however no predictors had a significant effect on northern quoll occupancy (Table 4.2). In the wet season, the best occupancy model included an interaction term between habitat area and edge density, with edge density appearing to be the most important predictor (Table 4.2). When habitat area was low (6% coverage), predicted quoll occupancy was 94 times higher at landscapes with the minimum edge density (0.02 km/ha) when compared to sites with minimum edge density (0.19 km/ha). By contrast, when habitat area was high (37% coverage), predicted quoll occupancy was only 7.5 times higher at landscapes with the minimum edge density when compared to sites with maximum edge density (Fig 4.2).

Figure 4.2 – Predictions from wet season occupancy models, used to test northern quoll (Dasyurus hallucatus) responses to naturally fragmented landscapes in the Pilbara bioregion, Western Australia.

Abundance

In the dry season, the best n-mixture model included habitat area and edge as additive terms, as well as topographical ruggedness (Table S4.4). Habitat edge and topographical ruggedness were both important predictors (Table 4.2). Predicted quoll abundance was 24 when edge density was set to the minimum, and 0 when edge density was set to the maximum. Quoll abundance increased from one when topographical ruggedness was set to the minimum (0.32) to 20 when topographical ruggedness was set to the maximum (2.04) (Fig 4.3). In the wet season, the best model was identical to the best dry season model, with the addition of total rainfall in previous wet season. Habitat edge, topographical ruggedness, and total rainfall in previous wet season were all influential predictors of abundance. Predicted abundance was 6 times higher at landscapes that received the maximum amount of previous wet season rainfall (400mm) (6 CI 3–14) when compared to landscapes that received the minimum amount of previous wet season rainfall (160mm) (0 CI 0–2) (Fig 4.4). Average predicted landscape abundance was 8 (min = 0, max = 71) in the dry season and 10 (min = 0, max = 71) in the wet season (Table S4.5).

Figure 4.3– Predictions from dry season n-mixture models, used to test northern quoll (Dasyurus hallucatus) responses to naturally fragmented landscapes in the Pilbara bioregion, Western Australia. Northern quoll responses to A) edge density, and B) topographical ruggedness.

Figure 4.4 – Predictions from wet season n-mixture models, used to test northern quoll (Dasyurus hallucatus) responses to naturally fragmented landscapes in the Pilbara bioregion, Western Australia. Northern quoll responses to A) edge density, B) topographical ruggedness and C) previous wet season rainfall (mm).

Discussion

We examined the impacts of habitat amount and configuration on a globally endangered predator, the northern quoll, across a naturally fragmented landscape. We found that habitat configuration, represented as the density of ‘edge’ within a landscape, was more important for northern quolls than habitat amount: northern quolls were scarce in landscapes with higher amounts of edge. Our results contrast with a general view of the primacy of habitat amount for species conservation (Hodgson et al. 2011), and support the findings of studies which have found the effects of habitat configuration can be strongest when habitat amount is low (Andren 1994; Betts et al. 2006; Fahrig 2017).

In landscapes where habitat is patchy, animals may be forced to travel to multiple patches in order to acquire sufficient resources (Gehring and Swihart 2004). This can mean they spend more time travelling through the matrix (Forman 2014), where they may be more exposed to predation, negatively impacting populations (Driscoll et al. 2013; Kupfer et al. 2006). In landscapes where habitat amount is low, like naturally fragmented landscapes, the effect of patchiness is likely to be amplified, because the distance between patches is more likely to be high, further increasing the time animals must spend in the matrix (Andren 1994). This is one mechanism by which increased fragmentation (edge density) may have negatively influenced quoll populations in our study. Being predators, northern quolls require large amounts of habitat in order to meet their resource requirements ― in the Pilbara, average female home ranges can exceed 30 hectares, and males can exceed 300 hectares (Cowan et al. 2020a; Hernandez- Santin et al. 2020). In patchy landscapes, these requirements are often likely to exceed the habitat contained by a single patch, forcing quolls to spend large amounts of time traversing through the spinifex matrix to reach additional patches (Fahrig 2017). Here quolls are also likely to experience increased rates of predation by feral cats and dingoes, because both are known to favour less complex habitat (Hamer et al. 2021; McGregor et al. 2015a), more specifically spinifex grasslands in the Pilbara

(Hernandez-Santin et al. 2016). This hypothesis is evidenced by a previous study which found that female northern quoll mortality increases (mostly due to predation) when the proportion of rocky habitat within their home-range decreases (Oakwood 2000).

Another factor potentially useful in explaining the negative response of northern quolls to edge density are edge effects themselves. One of the most commonly observed edge effects for prey species is elevated rates of predation at or near edges, as a result of predators being attracted to edge habitats (Hartley and Hunter 1998; Lidicker 1999; Vernes et al. 2001). Predators of northern quolls, such as dingoes and feral cats, are known to target natural and manmade edges to enhance hunting success (Hansen et al. 2019; May and Norton 1996; McGregor et al. 2014). As such, northern quolls and other small-medium sized mammals may well encounter increased predation rates at the edges of rocky outcrops and the sandplain matrix.

An important factor mediating the strength of edge effects, as well as the vulnerability of prey species to predation in the matrix, is matrix composition (Franklin and Lindenmayer 2009). In this study, the landscape matrix was composed of spinifex grasslands ― a feature which is largely shaped by fire Australia (Allan and Southgate 2002; Haslem et al. 2011). Nonetheless, we were unable to show that fire (proportion of landscape recently burnt) had any effect on quoll occupancy or abundance, similar to the results of Cook (2010a) who found quolls failed to respond to fire. One explanation for this result could be that northern quolls only respond to fire for a very brief period which was not able to be captured in this study. Alternatively, it’s possible that quolls only respond to fire under certain conditions. For example, Radford et al. (2020) found northern quolls declined in abundance after fire in woodland habitats, but increased in abundance after fire in sandstone habitat. However, it’s important to note that matrix composition can also be influenced by additional factors, including grazing from introduced herbivores such as cattle― which are common within the study area (author personal observation). Previous studies have found cattle disturbance to have a negative effect on quoll abundance (Radford et al. 2015). As such, we recommend future studies investigating the occurrence and abundance of northern quolls in fragmented landscapes account for vegetation cover, potentially using high resolution drone imagery, as well as the grazing impacts of livestock.

In contrast to matrix composition, we found topographical ruggedness was an important factor in predicting quoll abundance, irrespective of season. This result aligns with a broad catalogue of literature that suggests an environmental gradients can play an important role in determining species distribution patterns in fragmented landscapes (Bennett et al. 2006). In addition, it supports previous studies which indicate that areas which are topographically rugged can provide high quality habitat for many of Australia’s declining mammal species (Einoder et al. 2018; von Takach et al. 2020b), including northern quolls (Braithwaite and Griffiths 1994; Moore et al. 2019; Woinarski et al. 2008), probably because the incidence of predators like dingoes (Stobo‐Wilson et al. 2020) and feral cats (McDonald et

85 al. 2020) are lower in these habitats. Similar to landscape composition, species can also respond to environmental gradients across fragmented landscapes (Bennett et al. 2006), and we found this to be true in our study; northern quoll wet season abundance was higher at sites that received higher amounts of rainfall in the previous wet season. In the Pilbara, northern quolls align their recruitment period with the peak in annual rainfall (Jan-March) (Dunlop per comms) when landscape productivity is at its highest for insects and other prey items (Dunlop et al. 2017) probably to increase recruit survival rate (Moro et al. 2019) . As such, this result is likely explained by a lag effect in recruitment survival ― juvenile survival the wet season before sampling was higher at sites that received more wet season rainfall, translating to an increased number of adults and recruits a year later.

Conservation and management implications

This study is unique within the current literature in that it investigates the independent effects of landscape fragmentation using threatened species in a non-temperate environment (Fardila et al. 2017), and as such our results provide useful theoretical insight. However, our findings also have important management implications. For example, a key result was that increasing landscape fragmentation was detrimental to quoll occurrence and abundance, likely because quolls were exposed to increased levels of predation in these landscapes. Firstly, this suggests that the division of rocky habitat in the landscape, through either the construction of mining infrastructure (e.g road, rail lines) or other means is likely to have important negative impacts on northern quoll populations, which should be considered as part of impact assessments. In the case of our study, we found predicted northern quoll abundance was high (>10) when edge density was less than 0.05 km/ha, and low (<2) when edge density was higher than 0.08 km/ha. Secondly, this result suggests that if predator densities within highly fragmented landscapes can be managed effectively, there is potential for quoll abundance to be maintained within fragmented landscapes. In addition to the negative influence of fragmentation, we found topographical ruggedness had a strong positive effect on quoll abundance ― predicted quoll abundance was high when topographical ruggedness was above 1.7, and low when ruggedness was 0.9. While supporting the findings of previous studies (Braithwaite and Griffiths 1994; Moore et al. 2019; Woinarski et al. 2008), this result also highlights the applicability of topographical ruggedness as an effective and easily acquired metric for determining northern quoll habitat quality, although we strongly advise that other metrics such as edge density, vegetation cover and den availability (Moore et al. 2021) are also incorporated.

Acknowledgements

Data collection was assisted by Sian Thorn, Darcy Watchorn, Rainer Chan, Daniel Bohorquez Fandino, Jacob Champney, Hannah Kilian, Mitch Cowan. Camera trap deployment was also completed with the assistance of the Yandeyarra Indigenous ranger program facilitated by Pip Short and Greening Australia. Technical support was provided by Neal Birch, Brent Johnson, Hannah Anderson, Russell Palmer, Alicia Whittington and Jo Williams from the Western Australian Department of Biodiversity, Conservation and Attractions (DBCA), as well as Deb Noy from Charles Sturt University. Stephen van Leeuwen from DBCA provided assistance with the project conception and also provided support throughout. Equipment and operational costs were provided by DBCA, Roy Hill and Charles Sturt University. Roy Hill also covered the costs of flights, fuel and freight. Harriet Davie from Roy Hill provided technical support throughout along with the Roy Hill rail team in Port Hedland. We thank Colin Brierly, Betty Brierly and Graham for providing access to Indee station, Ben and Lindsey for access to Mallina and Troy Eaton for access to Pippingarra. Belinda Barnett from BHP assisted with site access. H.A.M is supported by a scholarship from the Institute of Land, Water and Society and Charles Sturt University. L.E.V. was funded by the Australian Government's National Environmental Science Program through the Threatened Species Recovery Hub. D.G.N. was supported by an Australian Research Council Early Career Researcher Award (DECRA). This project was supported by the Holsworth Wildlife Research Endowment – Equity Trustees Charitable Foundation & the Ecological Society of Australia, as well as the Western Australian Department of Biodiversity, Conservation and Attractions as well as environmental offsets and public good funding provided by BHP, Rio Tinto, Atlas Iron, Fortescue Metals Group, Roy Hill, Process Minerals International, Metals X and Main Roads Western Australia. Research ethics was granted though the Charles Sturt University animal ethics committee (permit number A17031), and the Western Australian Department of Biodiversity, Conservation and Attractions (permit number 08-002376-1).

Chapter 5 – A rocky heart in a spinifex sea: occurrence of an endangered marsupial predator is multiscale dependent in naturally fragmented landscapes

Harry A. Moorea,b*, Damian R. Michaela, Euan G. Ritchiec, Judy A. Dunlopd, Leonie E. Valentineb, Richard J Hobbs b, Dale G. Nimmoa

Manuscript accepted: Moore, H., Michael, D., Ritchie, E., Dunlop, J., Valentine, L., Hobbs, R., and Nimmo, D. (2021). In the heart of a spinifex sea: habitat selection of an endangered marsupial predator is multiscale dependent. Landscape Ecology.

a School of Environmental Science, Institute for Land, Water and Society, Charles Sturt University, Albury, NSW, Australia b School of Biological Sciences, University of Western Australia, Crawley, WA, Australia c Centre for Integrative Ecology and School of Life and Environmental Sciences, Deakin University, Burwood, VIC, Australia d Department of Biodiversity, Conservation and Attractions, Locked Bag 104, Bentley Delivery Centre, Perth, WA, Australia

Abstract

Context

Research on the impacts of anthropogenic habitat fragmentation has dominated landscape ecology for decades, yet our understanding of what drives species’ distributions in naturally fragmented landscapes remains limited.

Objectives

We aimed to (i) determine whether rocky patches embedded within a ‘matrix’ of fire prone grasslands act as naturally fragmented landscapes for an endangered marsupial predator, the northern quoll (Dasyurus hallucatus), and (ii) reveal the extent to which within-patch, patch, landscape variables, and matrix condition drive the occurrence of northern quolls.

Methods

We deployed remote sensing cameras for a total of 200 nights, at 230 sites spanning rocky and grassland habitats across 6000 km2 of the Pilbara bioregion of Western Australia. We examined the influence of within-patch, patch, landscape variables, and matrix condition on northern quolls using Generalised Linear Mixed Models.

Results

We found strong evidence that northern quoll habitat is naturally fragmented, observing higher occurrence and abundance of quolls in rocky patches than the surrounding grassland matrix. Within rocky patches, quolls were more likely to use patches with higher vegetation cover and den availability (within-patch), lower amounts of edge habitat relative to patch area (patch), and larger amounts of surrounding rocky habitat (landscape). When quolls entered the matrix, they tended to remain in areas with high vegetation cover, close to rocky patches.

Conclusions

Species occurrence in naturally fragmented landscapes is influenced by factors operating at multiple scales. Rocky habitats are naturally fragmented and vital to the conservation of a range of taxa around the world, including the northern quoll.

Introduction

Species frequently inhabit naturally or human-caused fragmented landscapes, where patches of suitable habitat are embedded within less suitable or sometimes a hostile ‘matrix’ of unsuitable habitat (Fahrig 2003). Meta-population (Hanski 1994) and island biogeography theory (MacArthur and Wilson 2001) have shaped research on fragmented populations, whereby fragmented landscapes are viewed as consisting of habitat and non-habitat . ‘Non-habitat’ often arises in human modified landscapes through land clearing for agriculture and urbanisation (Hobbs and Saunders 1994). Landscapes that are naturally fragmented have habitat patches embedded in a relatively unmodified matrix that is nonetheless less suitable to a given species (Driscoll 2005). Examples of naturally fragmented landscapes include patches of humid high elevation forests surrounded by dryer vegetation at lower elevations for bird species in Mesoamerica (Watson and Peterson 1999), and isolated patches of suitable cliff habitat dispersed within an otherwise flat landscape for plant species in south-eastern Canada (Haig et al. 2000).

Within human-caused fragmented landscapes, habitat selection depends on variables operating at three spatial hierarchies: within-patch, patch, and landscape variables (Thornton et al. 2011). Within-patch variables are those that determine patch quality for a species, such as the abundance of food resources or the availability of shelter sites within the patch (Thornton et al. 2011). For example, tree hyraxes (Dendrohyrax arboreus) are less likely to occupy patches where high levels of wood removal has occurred, presumably due to a reduction in available denning hollows (Lawes et al. 2000). The size and shape of patches are characterised as patch variables (Thornton et al. 2011), each of which can influence species occurrence by limiting populations size (Bennett et al. 2006), and exposing patch-dependent species to edge effects (Andrén 1995). Finally, landscape variables relate to conditions surrounding a patch, such as the availability of further habitat and the condition of the matrix (Driscoll et al. 2013; Thornton et al. 2011)

While the amount of habitat in a landscape is a well-established predictor of species occurrence (Thornton et al. 2011), there is growing recognition of the important role of matrix condition in influencing species persistence within fragmented landscapes (Eycott et al. 2012; Ricketts 2001). For example, a meta-analysis of 1015 species comprising a range of taxa found the overall influence of patch area was significantly weaker in patches separated by a natural matrix when compared to anthropogenic matrixes (Prugh et al. 2008). Understanding processes influencing species use of the matrix can be vital for species conservation. Changes in fire management can promote collared lizard movement (Crotaphytus collaris collaris) through woodland matrices surrounded by rocky habitat, resulting in increased occupancy of rocky glades (Templeton et al. 2011).

In our study, we test if rocky patches embedded within fire prone grasslands act as naturally occurring fragmented landscapes for the largest marsupial predator in north-west Australia, the endangered

91 northern quoll (Dasyurus hallucatus). We then test how habitat variables measured at the within-patch, patch, and landscape scale drive northern quoll occurrence, and abundance. Northern quolls previously occurred widely in northern Australia, but have since undergone substantial declines (Braithwaite and Griffiths 1994; Moore et al. 2019), likely due to a combination of threats including altered fire regimes (Woinarski et al. 2011b), predation by feral predators (Oakwood 1997), and most recently, the arrival of cane toads (Rhinella marina); an invasive amphibian that can be fatally toxic when consumed by quolls (Oakwood 2004c). The Pilbara bioregion in north-west Western Australia is the only mainland section of the northern quoll’s range yet to be impacted by cane toads, and as such is regarded as a critical population for the species (Cramer et al. 2016). Here, northern quolls are known to rely on rocky habitat because they provide highly suitable areas within which females are able to establish maternal denning sites (Oakwood 1997) —critical features in providing developing offspring with protection from predators (Oakwood 2000). While most Pilbara northern quolls occurrences have been recorded on large, continuous rocky mesas, less is known about their patterns of occurrence in naturally fragmented rocky landscapes, typically comprising patches of smaller to medium-sized granite outcrops.

Within these landscapes, a range of factors have the potential to be important predictors of quoll patch suitability, yet few have been formally tested. For example, at the within-patch scale, an important predictor may be the availability of denning habitat (small caves, crevices), used for sheltering in during the day (northern quolls are nocturnal) and storing young inside during the breeding season (Cowan et al. 2020b; Oakwood 2000). At the patch scale, habitat suitability may be enhanced or reduced by factors such as the size and shape of rocky outcrops, as well as patch geomorphology, such has been demonstrated in other rock dependent species (Do Carmo and Jacobi 2016; Michael et al. 2008).

Further, given northern quolls use large home ranges that can exceed the size of a single rocky patch (Hernandez-Santin et al. 2020), the extent of rocky habitat surrounding patches may also be a determining factor in patch suitability. However, for quolls to access this additional rocky habitat they may be forced to traverse through a non-habitat matrix of spinifex grasslands, where they may be more likely to encounter predators (Hernandez-Santin et al. 2016), thus potentially creating a ‘landscape of fear’ (Laundré et al. 2001). The age and structure of spinifex vegetation―mostly determined by fire in northern Australia (Allan and Southgate 2002)― is therefore likely to be important in mitigating the risk of lethal encounters with predators such as dingoes (Canis dingo) and feral cats (Felis catus) for quolls, with the latter known to hunt more efficiently in open landscapes (McGregor et al. 2015a). Understanding the multi-scaled habitat requirements and constraints for northern quolls is particularly important in the Pilbara, as rocky habitats are increasingly being degraded or removed from the landscape as part of expanding mining activity (Majer 2014). Indeed, identifying patches of critical habitat that support northern quoll persistence has been identified as a key research and conservation priority (Cramer et al. 2016).

Materials and methods

Study area

This study was carried out across four properties situated within the Pilbara bioregion in north-west Western Australia. These were Indee Station, Mallina Station, Pippingarra Station and Yandeyarra Indigenous Reserve. Yandeyarra Indigenous Reserve is also a working cattle station. The study area encompassed the Karayarra and Nyamal Indigenous language groups. Vegetation across all study sites is dominated by hummock grasslands (Triodia spp.) that cover roughly 30 – 50% of the ground surface, with varying time since fire. Tree cover is sparse to non-existent across the study area, and is mostly comprised of mulga (Acacia aneura), snakewood (Acacia xiphophylla) and snappy gum (Eucalyptus leucophloia). Geology is characterised by largely flat sand plains scattered with greenstone ridges and granite. Climate within the study region is characterized by high temperatures and low annual rainfall. Average daily temperature maximums across the study period ranged from 28.4 C (August 2017) to 44.1C (December 2018) (Australian Bureau of Meteorology 2020).

Study design

We employed a nested experimental design by establishing a total of 230 sites across 23 pre-defined site clusters (~75 hectares in size) across a 6000 km2 study area in the Pilbara. Site clusters were chosen because they comprised rocky outcrops embedded within a matrix of spinifex grasslands, and thus constituted a likely naturally fragmented landscape for northern quolls. Within each cluster, we established ten sites: seven sites within rocky habitat and three sites within the spinifex grasslands surrounding the rocky outcrops (Figure 5.1). Rocky sites were chosen to represent gradients of outcrop size (basal area), shape (area-edge ratio), and geomorphology (see Table 5.1). Spinifex sites were positioned at either a small (~50m), medium (~100m), or large (~200m) distance from rocky habitat, and were chosen to capture a gradient of vegetation ages, from recently burnt (<3 year), to mid- successional (4-10 years) and long unburnt (>10 years) vegetation using ArcMap 10.3 (ESRI 2011) and data from the Northern Australian Fire Information database (NAFI 2020). All sites were separated by at least 200m.

Figure 5.1 – A.) Site clusters located within the Pilbara bioregion in north-west Australia. B.) An aerial view of a typical site cluster, showing the position of vertical and horizontal facing camera traps deployed to detect northern quolls (Dasyurus hallucatus). C.) A typical ‘site cluster’ surveyed as part of the current study, comprised granite outcrops surrounded by a matrix of spinifex (Triodia spp.) grasslands. D.) Horizontal camera trap set-up used at spinifex sites. E.) Vertical camera trap set-up used at rocky sites.

Table 5.1 – Summary of variables used to describe habitat properties within rocky and spinifex patches. R = rocky array, S = spinifex array, C = combined array. WP = within-patch, P = patch, L = landscape, M = matrix.

Habitat Description Justification Array Scale variable Habitat type Habitat classed as rocky outcrop or Rocky habitat is a preferred habitat of northern quolls C L (rock/spinifex) spinifex (Hernandez-Santin et al. 2016), and we expected northern quolls within site clusters to mostly occur in rocky habitat. To test this prediction we included habitat type as a predictor variables in combined site models. Den availability The availability of potential den sites Den sites are critical to sheltering northern quolls from R WP was measured along a 50m transect, predation, extreme heat and fire (Oakwood 1997). centred at camera locations. A sampling Dens are also used by female northern quolls to store point was marked at each end of the young in prior to their independence (Cowan et al. transect, and in the centre. From each 2020b). sampling point, 4 quadrants delineated by north, south, east and west and extending up to 10 metres away from the sample were marked out, and the presence or absence of a potential den was noted within each quadrant. Den availability was represented as the proportion of total quadrants for each transect (n= 12) within which potential dens were found. Vegetative Percent cover of vegetation less than Small to medium sized vertebrates often associate with R,S WP/M cover (<0.5m) 0.5m in height. Vegetation cover was habitats possessing high vegetation cover because it assessed visually within a 25m radius of can be more difficult for predators to detect them camera sites. (Doherty et al. 2015a; Kotler 1984; Loggins et al. 2019). For the northern quoll, vegetation low to the ground such as spinifex is likely to provide the most effective cover from predators. Patch Granite outcrops were classed into major Northern quolls may prefer some geomorphologies to R P geomorphology granite landform groups as described by others, based on physical characteristics that influence Withers (2000). Major granite landforms prey and den availability, as well as vulnerability to were nubbins, ridge line, and inselbergs. predation. Patch area Total surface area of rocky outcrop on Patch area has been recognised as an important factor R P which camera site was established. in driving species occurrence across a range a taxa (Thornton et al. 2011) Patch shape Total surface area/perimeter length of Decreasing patch area to edge ratio can have R P rocky outcrops on which camera sites detrimental consequences for prey species due to was established. increased vulnerability to predation (Andrén 1995) Landscape Total surface area of rocky habitat (ha) A decreasing amount of suitable habitat in the area R L extent within a 150m radius of camera sites, surrounding a patch may influence patch suitability by corresponding to the radius of median limiting the availability of nearby resources such as female northern quolls home estimates food and shelter (Bennett et al. 2006). collected using GPS collars within the study area (Moore, unpublished).

Distance to Distance in metres between sites and the Rocky habitat is a preferred habitat of northern quolls, S M rocky outcrop closest rocky outcrop over 50m2 in area. and northern quolls are likely less exposed to predation in these areas because they provide high densities of structural refuges (Hernandez-Santin et al. 2016). Northern quolls moving further away from rocky habitat into spinifex habitat where less structural refuges are available are likely exposed to increasing levels of predation risk. Years since Time in years between the time the site Vegetation structure in northern Australia is strongly S M burnt was last burnt and the beginning of the influenced by fire (Miller and Murphy 2017). Prey sampling period. species such as the northern quoll are likely to be more exposed to predation risk in habitat that have been recently burnt, because the structural complexity of vegetation is reduced (Nimmo et al. 2019).

In this study, a ‘site’ refers to a camera trapping station comprised a Reconyx PC900 Hyperfire passive infrared triggered camera trap (Reconyx 2020). Cameras were oriented in one of two ways (vertical or horizontal) and within one of three arrays (combined, rocky, or spinifex grassland), and orientation was consistent within each array. Cameras were oriented in two ways for two reasons. First, to compare the capacity of vertical and horizontal cameras to detect quolls, as part of the broader monitoring program within which this study was embedded (Moore et al. 2020c). Second, to account for predicted differences in detectability within the two habitat types. Horizontally oriented cameras (i.e., facing outwards) have a larger detection zone, which can increase the number of detections (Meek et al. 2012), and are often more effective at detecting predators (Moore et al. 2020c; Nichols et al. 2017). Consequently, all cameras deployed in the spinifex matrix (n = 3 per cluster, n = 69 overall) and a subset of cameras deployed on rocky outcrops (n = 2 per cluster, n = 46 overall) were positioned horizontally to improve detection rates, particularly in the spinifex matrix where we assumed quolls would be scarcer and less detectable.

Horizontal cameras were attached to a tree stake 50cm above the ground, with the camera lens and PIR sensors focused at a 10° angle toward the ground surface, facing south, and were baited using PVC canisters containing pilchards (fish). All other cameras on rocky outcrops (n = 5 per cluster, n = 115 overall) were positioned vertically. Vertically orientated cameras were attached to a right-angle bracket on a wooden tree stake 1.5m above the ground, with the camera lens and PIR sensors focused directly at the ground surface. Horizontal cameras deployed on rocky outcrops were positioned with vertical cameras (Figure 5.1), however because data from vertical and horizontal cameras are never combined, we refer to these as separate sites. All cameras were set to high sensitivity, and five images were taken at one second intervals per trigger. Sites were sampled for 100 days in the Pilbara dry season (August – November) and 100 days in the wet season (total nights = 200) (December – March) (Figure S5.1). We used a period of 100 days because it was sufficient to be 95% confident of northern quoll absence at a site within the site cluster using either a vertically or horizontally orientated camera (Moore et al. 2020c). Twelve site clusters were sampled over the 2017-2018 period, and the remaining eleven site clusters were sampled over the 2018-2019 period. Although subsequent analysis showed no difference in either nightly detection or the number of detections of northern quolls between vertical and horizontal cameras (Moore et al. 2020c), we do not combine cameras with different orientations into a single analysis at any point.

The three camera arrays allow us to ask different questions about quoll occurrence in both the rocky patches and the spinifex matrix. The first array, which we refer to as the combined array, grouped the horizontally orientated cameras at rocky sites with horizontally orientated cameras in the spinifex matrix (Figure 5.1). This array allows a comparison of northern quoll occurrence and activity between

96 rocky and spinifex habitat (i.e., to confirm that occurrence and abundance are higher in rocky habitats than spinifex grassland), to test whether these landscapes are naturally fragmented for quolls. The second array, which we refer to as the rocky array, included only the vertically orientated cameras on rocky outcrops. Vertical cameras were ideal for capturing unique spot patterns located on the dorsal surface of northern quolls, which we used to identify individual animals (see ‘detection data’ section below). The rocky array allowed us to ask questions regarding the extent to which within-patch, patch, and landscape variables affect quoll occurrence and abundance in rocky outcrops. Here, we employed a ‘focal patch’ design (sensu Kaplan and White 2002), where detections collected by each camera within the rocky array were used to measure northern quoll responses to habitat at all three levels (within- patch, patch and landscape) (Figure 5.2).

Figure 5.2 – An aerial view of the four scales (within-patch, patch, landscape, and matrix) at which northern quoll responses to habitat within fragmented landscapes were measured. Brown shading represents rocky habitat. The yellow dot represents a vertical camera site, and the blue dot represents a horizontal camera site. Bold outlines represent the scale in focus.

The final array, which we refer to as the spinifex array, was comprised of all horizontally orientated cameras deployed in the spinifex matrix. The spinifex array allowed us to ask questions regarding how matrix conditions affect quoll occurrence in the spinifex matrix.

Detection data

When northern quolls were detected on vertical cameras, we used unique spot patterning as well as scarring located on the dorsal surface of animals to identify individuals from camera trap imagery, following a similar process outlined in Moore et al. (2020a) (also see; Diete et al. 2016; Hohnen et al. 2013). Images from each detection event were catalogued into separate folders, and then within folders, we screened images in order to select those that represented individual spot patterns, preferably from multiple angles. Images that contained less than roughly 30% of the northern quoll’s dorsal surface were immediately excluded from further image analysis unless they contained other distinctive features that could be used to identify individuals (e.g. unusual scarring patterns, missing ears etc.). Most detection events had at least one image suitable for individual identification. At least one screened image from each detection event was then entered into I3S Spot (Den Hartog and Reijns 2016), a freely available software package that can be used to identify individual animals based on spot patterns. Recent studies have used I3S to assist users in identifying individual whale sharks (Araujo et al. 2019), killer whales (Denkinger et al. 2020), and lizards (Moore et al. 2020a). The package uses a two-dimensional linear algorithm to rank the likelihood of two images containing the same animal, based on the coordinates, shape, and size of an animal’s spots (Cerutti-Pereyra et al. 2018). Once images had been ranked in I3S, at least two observers confirmed if suggested matches were the same animal. If no match could be found, the image was marked as a new animal. If a consensus could not be reached between the observers, the individual was marked as ‘unidentifiable’. To assess the rate at which horizontal camera traps were visited by northern quolls, we defined independent detection events as detections separated by 15 minutes. This was because Diete et al. (2016) found consecutive northern quoll detections separated by between 10 and 15 minutes were as likely to be a different individual as they were the previous individual.

It is important to note that although we were able to individually identify animals in this study, our data was not suitable for mark-recapture analysis. The main reason for this was the sparseness of the data; sampling at patches was limited to a single camera, and thus the number of captures and recaptures at the majority of sites was insufficient to estimate abundance reliably in this way. Instead, we calculated the minimum number of animals known to be alive (MNA) (Krebs 1989) at each site; an abundance metric commonly used in ecology when more sophisticated measures are not available (Baker et al.

2017; Onodera et al. 2017; Wauters et al. 2000). MNA was calculated by summing the number of animals detected at each site in both seasons (dry, wet).

Statistical analysis

All data was modelled using generalized linear mixed effects models (GLMMs) in R version 3.6.2 (R Core Team 2020). These models allowed for the inclusion of ‘site cluster’ as random effect to account for spatial bias as a result of our nested experimental design. Preliminary analysis indicated that there were many sites across all arrays included in our analysis with no northern quoll detections, leading to zero inflation. Models that ignore zero inflation can produce biased parameter estimates (Potts and Elith 2006). To model zero-inflated data in this study, we used hurdle models fit using the package glmmTMB. Hurdle models are commonly used to model zero-inflated ecological data (Balderama et al. 2016; Brown et al. 2016; Cunningham et al. 2018) and consist of two components. The first component is the zero-inflated model, which describes the probability of species absence at a site (Potts and Elith 2006). To simplify the interpretation, we plotted model predictions as the probability of presence throughout (i.e., 1 - probability of absence). The second component is the conditional model, which uses a zero-truncated error distribution to describe the relationship between model predictors and non- zero count data. Given hurdle models operate using a two-step process, fixed effects can vary between binary and count models (Potts and Elith 2006). To account for over-dispersion, we fitted hurdle models with a zero-truncated negative binomial distribution (Lindén and Mäntyniemi 2011). To account for uneven sampling effort as a result of camera failures, all models included the number of nights cameras were operational as an offset term. All model fixed effects were scaled prior to analysis to improve model stability and allow direct comparisons between model coefficients (Harrison et al. 2018). We regarded fixed effects as being influential if the 95% confidence intervals of variable coefficients did not intercept zero (Nakagawa and Cuthill 2007).

We modelled data from the three camera arrays separately. Data from different seasons (dry, wet) was also modelled separately, to account for differences in habitat use and resultant varying levels of productivity between seasons. We found northern quoll detection data from the spinifex array were too sparse for hurdle models to be used effectively. To overcome this issue, we pooled data from the spinifex array across seasons (dry, wet), and converted count data to a binary format (presence/absence). We then modelled pooled spinifex array data using a GLMM fit with a binomial distribution in package lme4 (Bates et al. 2007).

To confirm that quolls were more common on rocky outcrops compared to spinifex grasslands, we used data from the combined array to build a univariate model of quoll presence/absence (zero-inflated response variable) and total number of independent northern quoll detections (conditional response

99 variable) in relation to habitat type; a categorical predictor variable with two levels indicating whether the site was located on a rocky outcrop or within spinifex grasslands. To examine the importance of within-patch, patch, and landscape variables for northern quolls, we used data from the rocky array to model quoll presence/absence (zero-inflated response variable) and total number of northern quoll individuals (conditional response variable) in relation to the six rocky array predictor variables (Table 1). Within-patch variables were den availability and vegetative cover, patch variables were patch size, shape and geomorphology, and the landscape variable was landscape extent (amount of rocky habitat within a 150m buffer of sites) (Table 5.1). Finally, using data from the spinifex array, we modelled the total number of independent northern quoll detections in relation to two variables describing matrix condition (vegetative cover less than 0.5m in height and years since fire) in addition to distance to nearest rocky patch (Table 5.1). No variables included together in any model shared pairwise correlations >0.50 (Table S5.2, S5.3).

To select the most parsimonious model for the rocky and spinifex array models, we built models with all possible subsets of habitat variables in both binary and count model components. We then used Akaike’s Information Criteria adjusted for small sample size (AICc) to rank zero-inflated and conditional models separately. All model selection was conducted using the dredge function in R package MuMIn (Barton 2020). A total of 64 rocky site models and 8 spinifex array models were compared. Estimates were selected from the most parsimonious model. To examine if our best models were biased by residual spatial autocorrelation, we calculated the correlogram of the model residuals based on Moran’s I, using the Pgirmess package (Giraudoux 2018). We found no evidence of spatial autocorrelation in the residuals of any of the models described in the Results section (Figure S5.3).

Results

Across all cameras and seasons, we recorded a total of 1814 independent northern quoll detections. The majority of detections were recorded within the rocky array (n = 1768, ~97.5%), and fewer detections were recorded in the spinifex array (n = 46, ~2.5%). A total of 837 (~46%) northern quoll detections were recorded during the dry season, and 977 (~54%) during the wet season. Using image data from vertical cameras, we were able to identify individual northern quolls in 70.2% of detection events, from which a total of 153 northern quolls were identified. We found there was a strong positive correlation between the number of northern quoll individuals detected at a site and the total number of independent detection events (Figure S5.2, Table S5.1). Three cameras failed in the dry season sampling period (spinifex array = 2, rocky array = 1, and three cameras failed in wet season (spinifex array = 3).

100

Combined array

The probability of northern quolls being detected within the rocky array (47.8%, CI 33.4% ‒ 62.3%) was three times higher than within the spinifex array (15.9%, CI 7.3% ‒ 24.6%) in the dry season, and also three times higher in the wet season (rocky array = 21.7%, CI 9.8% ‒ 33.7%, spinifex array = 7.2%, CI 1.1% – 13.4%), although lower overall (Table 5.2, Figure 5.3). The total number of predicted quoll detections was eight times higher within the rocky array (n = 8, CI 3 – 19) when compared to the spinifex array (n = 1, CI 0.2 – 4) in the dry season, and 20 times higher within the rocky array (n =10, CI 3‒33.3) when compared to the spinifex array (n=0.5, CI 0.1 – 2.6) in the wet season (Figure 5.3).

Figure 5.3 – Response of northern quoll (Dasyurus hallucatus) site use to habitat type in the Pilbara bioregion of northern Australia. Plots derived from hurdle models. A.) Changes in the probability of northern quoll presence at sites in response to habitat type in the dry season. B.) Changes in the total number of northern quoll detections at sites in response to habitat type in the dry season. C.) Changes in the probability of northern quoll presence at sites in response to habitat type in the wet season. D.) Changes in the total number of northern quoll detections at sites in response to habitat type in the wet season.

101

Table 5.2 – Response of northern quolls (Dasyurus hallucatus) to habitat variables at rocky, spinifex and combined (rocky and spinifex) sites in the Pilbara bioregion of Western Australia. Dry season models used data collected from August-November and wet season models used data collected from December to March. Significant results are shown in bold.

Array Season Variable Estimate CI Std. Error Conditional

Intercept) -2.01 -12.50 – 8.49 5.35

Habitat (rocky) 2.60 1.04 – 4.16 0.79 Dry Dry

Zero-inflated model Intercept 1.66 1.01 – 2.31 0.33 Habitat (rocky) -1.57 -2.44 – -0.70 0.44 Conditional

Combined Intercept -0.81 -2.40 – 0.77 0.80

Habitat (rocky) 2.90 1.74 – 4.07 0.59

Wet Zero-inflated model Intercept 2.55 1.64 – 3.46 0.46 Habitat (rocky) -1.27 -2.42 – -0.12 0.59 Conditional Intercept 0.52 0.04 – 0.99 0.24 Vegetation cover (<0.5m) 0.19 0.03 – 0.35 0.08

Patch area 0.23 -0.03 – 0.48 0.13 Zero-inflated model Dry Dry Intercept -0.19 -0.66 – 0.27 0.24 Den availability -0.63 -1.15 – -0.11 0.26 Vegetation cover (<0.5m) -0.79 -1.40 – -0.17 0.31 Patch shape -1.27 -2.30 – -0.24 0.53

Habitat extent -0.72 -1.38 – -0.06 0.33 Conditional Intercept 0.52 -0.20 – 1.24 0.91 Rocky Zero-inflated model

Intercept -0.82 -2.61 – 0.96 0.91

Geomorphology_Inselberg 1.33 -0.62 – 3.30 1.00

Wet Geomorphology_Nubbin 1.90 -0.01 – 3.82 0.98 Geomorphology_Ridge -1.15 -3.95 – 1.65 1.43 Vegetation cover (<0.5m) -0.94 -1.58 – -0.31 0.32 Patch shape -0.67 -1.23 – -0.11 0.28

Habitat extent -0.48 -1.05 – -0.07 0.28

Intercept -2.41 -3.66 – -1.52 0.53 Vegetation cover (<0.5m) 1.00 0.21 – 2.01 0.45

Distance to rocky patch -1.24 -2.49 – -0.35 0.53

Spinifex Dry + Wet + Dry

102

Rocky array

The most parsimonious model for the dry season rocky array included predictors measured at the within- patch scale (vegetative cover, den availability), the patch scale (patch shape) and the landscape scale (habitat extent) for the zero-inflated component, and the within-patch scale (vegetative cover) for the conditional component (Table 5.2). The probability of northern quoll occurrence increased significantly with increasing vegetation cover, den availability, and habitat extent, and decreased at sites with proportionally large amounts of edge habitat (patch shape) (Table 5.2, Figure 5.4). The most parsimonious model for the wet season rocky array model also included predictors measured at the within-patch, patch and landscape scale; vegetative cover, patch shape, patch geomorphology and habitat extent as predictors for the zero-inflated component. The most parsimonious model for the conditional component was the null model (Table 5.2). The probability of northern quolls being active at sites increased significantly with increasing vegetative cover (0.5 m), and habitat extent and decreased at sites with proportionally large amounts of edge habitat (patch shape) (Table 5.2, Figure 5.5).

103

Figure 5.4 – Probability of northern quoll (Dasyurus hallucatus) site use within rocky patches during the dry season in the Pilbara bioregion of northern Australia. Plots derived from hurdle models. A.) Vegetation cover (<0.5m) was measured as the proportion of ground covered with vegetation less than 50cm in height within a 25m radius of camera sites. B.) Den availability was measured as the availability of potential den sites at quadrats spread along a 50m transect, centred at camera locations (Table 5.1). C.) Outcrop shape was measured as the area to edge ratio of rocky outcrops at which a camera site was positioned. Larger values describe sites with larger amount of basal habitat area when compared to edge habitat. D.) Habitat extent was measured as the amount of rocky habitat (ha) within a 150m radius of camera sites. Dry season data was collected between August and November. Purple shading represents 95% confidence intervals.

104

Figure 5.5 – Probability of northern quoll (Dasyurus hallucatus) site use within rocky patches during the wet season in the Pilbara bioregion of northern Australia. Plots derived from hurdle models. A.) Outcrop shape was measured as the area to edge ratio of rocky outcrops at which a camera site was positioned. Larger values describe sites with larger amount of basal habitat area when compared to edge habitat. B.) Vegetation cover (<0.5m) was measured as the proportion of ground covered with vegetation less than 50cm in height within a 25m radius of camera sites. C.) Habitat extent was measured as the amount of rocky habitat (ha) within a 150m radius of camera sites. Wet season data was collected between December and March. Purple shading represents 95% confidence intervals.

Spinifex array

The most parsimonious model for the spinifex array included vegetative cover (0.5 m) and distance to rocky habitat (Table 5.2). The probability of northern quoll occurrence increased significantly with increasing vegetative cover, and decreased with increasing distance from rocky habitat (Table 5.2, Figure 5.6).

105

Figure 5.6– Response curves, derived from generalised linear models, showing changes in the probability of northern quoll (Dasyurus hallucatus) occurrence at spinifex sites in response to habitat variables in the Pilbara bioregion in Western Australia. A.) Vegetation cover (<0.5m) was measured as the proportion of ground covered with vegetation less than 50cm in height within a 25m radius of camera sites. B.) Distance from rocky patch was defined as the distance in metres between sites and the closest rocky patch over 50m2 in area. Purple shading represents 95% confidence intervals.

106

Discussion

The distribution and abundance of species that occur in landscapes fragmented by humans is known to be driven by within-patch variables, patch variables, and landscape variables (Thornton et al. 2011), but far less is known about the responses of species to such factors in naturally fragmented habitats. We found northern quolls were significantly more likely to use rocky patches when compared to the spinifex matrix, irrespective of season, suggesting these landscapes are naturally fragmented to northern quolls. Habitat selection by northern quolls occurred at multiple scales: i) at the within-patch scale, quolls were more likely to use patches with more denning crevices and vegetation cover, (ii) at the patch scale, quolls were more likely to use patches with smaller amounts of edge habitat relative to patch area, and (iii) at the landscape scale, quolls were more likely to use areas with higher rocky habitat extent. Use of the matrix by quolls tended to occur in close proximity to rocky habitat, and was more common in areas with high vegetation cover. Together, these results indicate that critical habitat for northern quolls in the Pilbara is likely defined by large areas of condensed, complex rocky habitat, with intact vegetation occurring within and in the areas surrounding. More generally, our results confirm the importance of understanding species’ habitat requirements at multiple scales and the importance of matrix condition within naturally fragmented landscapes.

An extensive catalogue of literature predicts the occurrence of species metapopulations within landscapes fragmented by human activities, but our study adds to a smaller body of work focused on landscapes that are naturally fragmented. Here, we found multiscaled responses by northern quolls aligned closely with species responses in human fragmented landscapes (Thornton et al. 2011). In addition, we found the response of northern quolls to naturally fragmented rocky habitat was similar to that of other species occurring in comparable naturally fragmented landscapes (Murray et al. 2008). For example, Carpentarian rock-rats (Zyzomys palatalis) are more likely to occur at rocky patches with higher vegetation cover, and within landscapes that have greater habitat extent (Trainor et al. 2000). These commonalties in species responses to landscapes fragmented through both natural and human mediated processes suggest factors determining species habitat selection may be similar in both. However, in order to explore these comparisons further in more detail, future research focused on the occurrence of species within naturally fragmented landscapes is required.

While this study is the first to quantify the spatial influence of rocky habitats on northern quolls, previous studies have shown rocky patches to be important sources of refuge for persisting populations across their range, likely for a combination of factors, including providing shelter from fire (Oakwood 2000), predators (Hernandez-Santin et al. 2016), and climatic extremes (Cowan et al. 2020b). Other species within the northern quolls range also rely on rocky patches (Hohnen et al. 2016a; Trainor et al. 2000), and at a global scale, rocky structures provide critical habitat for taxa ranging from apex predators (Bleich et al. 1996; Harrison et al. 2019) to predators (Mares 1997) and prey species (Kotler

107 et al. 1999; Metallinou et al. 2015). Nonetheless, rocky habitats are subject to a number of threats, including destruction for resource extraction and degradation by agriculture practises and recreation activities (Fitzsimons and Michael 2017). Further, species occurring in rocky habitats that are naturally fragmented, like northern quolls, have the potential to be impacted by additional threats that operate in the areas between rocky patch, such as fire and predation (Nimmo et al. 2019). This study adds to a growing amount of research highlighting the importance of rocky patches as species habitat (Fitzsimons and Michael 2017; Michael and Lindenmayer 2018), and as such the need to protect them.

Within-patch variables

Within-patch variables determine patch quality, and therefore are often strong predictors of species occurrence and abundance (Thornton et al. 2011). We found this to be true for northern quolls, which were more likely to use patches comprised of increased vegetation cover, regardless of season. Patches with increased vegetation cover probably provide increased food resources for northern quolls in the form of edible vegetation (seeds, flowers, fruits); an important component of the northern quolls diet in the Pilbara (Dunlop et al. 2017), while also supporting larger populations of prey species hunted by northern quolls, such as invertebrates, rodents, reptiles and smaller Dasyurids, given many of these species also rely on vegetation for both food and shelter (Menkhorst and Knight 2011; Wilson and Swan 2017). Increased vegetative cover may also provide northern quolls with increased cover from predators such as feral cats and dingoes, both of which are typically more common, and better hunters, in structurally simple habitats (Geary et al. 2020; McGregor et al. 2015a; Stobo‐Wilson et al. 2020). This was demonstrated during a recent reintroduction attempt, where northern quolls released into a recently burnt area within Kakadu national park were subject to extreme levels of dingo predation (Jolly et al. 2018a). The association of northern quolls with potential den availability underscores the importance of shelter sites for quolls to escape fire, climatic extremes, and predators, as well as providing thermally stable nursery sites for offspring (Oakwood 2000). We recommend measures of den availability, such as the one used here, and elsewhere (Hernandez-Santin et al. 2016; Oakwood 1997), be incorporated as part of future northern quoll habitat suitability assessments, given they are time efficient and require little training to complete.

Patch variables

Northern quolls also responded to patch shape, a patch-level variable; quolls were more likely to occur at patches that had less edge relative to patch size. This is a likely ‘edge effect’ (Andrén 1995) whereby animals that occur in patches with lower area to edge ratios are often more exposed to biological processes that are amplified around habitat transition zones , such as predation (Marini et al. 1995; Michel et al. 2016). For example, replica lizards positioned within edge habitats were significantly more likely to be attacked by predators when compared to replica lizards positioned in patches of remanent

108 vegetation (Hansen et al. 2019). In Australia, feral cats and dingoes are thought to use habitat edges to improve hunting success (Doherty et al. 2015b; McGregor et al. 2017), and thus the risk of northern quolls being predated upon is likely higher in habitat comprised of greater amounts of edge habitat. Further, the severity and type of edge effects can determine the effective habitat area or ‘core habitat’ remaining within a patch, which can be an important determinant of patch quality for species (Laurance and Yensen 1991).

Landscape variables

Habitat extent has previously been identified as a major driver of species occurrence in patchy landscapes (Thornton et al. 2011), and we found this to be true within the current study; northern quolls were more likely to use patches that were surrounded by greater extents of rocky habitat. A likely explanation for this result is that northern quolls are engaging in landscape supplementation (Dunning et al. 1992), where individuals increase their access to resources by visiting patches outside of their ‘home patch’. For example, Mexican mantled howler monkeys (Alouatta palliata Mexicana) living in fragmented forest landscapes are able to increase access to foods like fruits and flowers by visiting multiple patches (Asensio et al. 2009). This explanation may be particularly likely in the case of northern quolls given that they are predators, and as such may require access to areas of habitat larger than a single patch in order to meet their dietary requirements.

Matrix type can also influence species use and persistence within landscapes (Eycott et al. 2012; Ricketts 2001). Understanding factors that determine matrix permeability may be particularly important for northern quolls, because sub populations are prone to local extinction (Moro et al. 2019), and thus are dependent on dispersal for recolonization. Matrix type also plays a crucial role in maintaining functional connectivity between sub populations. For example, landscape heterogeneity is important for maintaining functional connectivity in the Allegheny woodrat (Neotoma magister), a small rodent that specialises on rocky outcrops (Kanine et al. 2018). We found that northern quolls tended to remain close to rocky habitats when entering the matrix, likely due to the quolls ‘boundary response’; the tendency of species to either advance or retreat upon encountering a patch boundary (Fahrig 2007; Nimmo et al. 2019). Species that occupy patches surrounded by high-risk matrix types should most often exhibit strong boundary responses that lead them back to habitat patches where risk is reduced (Fahrig 2007). For northern quolls, crossing the boundary from rocky patches into spinifex matrix is likely to substantially increase predation risk (Hernandez-Santin et al. 2016), and so remaining close to rocky refuges could allow a rapid retreat to escape a predator. Similar behaviour has been observed in caribou, which typically remain less than 500m from unburnt habitats (safer habitat) when moving within recently burnt habitat (riskier habitat) (Joly et al. 2003).

109

Differences in vegetation structure across the matrix can correspond to differences in predation risk, which can influence a species capacity to disperse through the matrix to other habitats (Nimmo et al. 2019). For example, white-shouldered fire-eye (Pyriglena leucoptera) survival was significantly lower when traversing through matrix with lower vegetation cover when compared to sites with high vegetation cover, likely due to higher predation risk (Biz et al. 2017). Hohnen et al. (2016a) found northern quoll populations separated by less topographically rugged areas were typically more closely related, suggesting that such habitats are easier for quolls to disperse through. Although the results of the current study were unable to demonstrate how differences in vegetation structure specifically impacted quoll dispersal, they do demonstrate northern quolls were generally less likely to use matrix sites with lower vegetation cover, potentially as a response to increased levels of perceived predation risk. This result suggests landscapes where matrix vegetation is structurally simple may be less favourable for quoll persistence, due to lower probabilities of colonization. Given vegetation structure within the northern quolls range is typically determined by fire (Miller and Murphy 2017) and grazing pressure (Liedloff et al. 2001), this result has important management implications for northern quoll populations in fragmented landscapes. For example, managing landscapes to create ‘corridors’ of intact vegetation running between rocky habitat patches could facilitate increased inter-patch movement, potentially benefiting northern quolls by (i) reducing the vulnerability of isolated populations to local extinction, and (ii) increasing the likelihood of recolonization should local extinctions occur (Bennett 1990).

It’s important to consider that factors such as sex may also influence the likelihood of quolls using matrix habitat. For example, a recent study focused on another short-lived marsupial, the southern brown bandicoot (Isoodon obesulus obesulus), found adults males were significantly more likely to use matrix habitat when compared to females, or juveniles of either sex (Maclagan et al. 2019). While we were unable to examine the influence of sex on matrix use as part of the current study, it’s possible sex may have a similar effect in northern quolls given males are known to move considerably larger distances than females, particularly during the mating season (Hernandez-Santin et al. 2020).

Conclusion

The results of our study indicate that granite outcrops nested within spinifex grasslands act as naturally fragmented landscapes for northern quolls in the Pilbara, and add to a growing body of literature in highlighting the importance of rocky habitat to a range of taxa (Fitzsimons and Michael 2017; Hohnen et al. 2016a). Further, we found northern quolls respond to habitat variables measured at the within- patch, patch and landscape scales, indicating species responses in naturally fragmented landscapes can mirror those and potentially aid predictions about the effects of fragmentation on species in more heavily-modified and human-dominated landscapes.

110

Acknowledgements

111

112

Chapter 6 – Summary

Australia’s ongoing mammal declines have attracted global attention as iconic species are pushed closer and closer to extinction (Woinarski et al. 2015). Aside from the uniqueness of the fauna at stake (Woinarski et al. 2019b), one of the most striking characteristics of these declines is the sheer scale at which they have occurred (Short and Smith 1994). This feature is typified by the decline of the northern quoll (Dasyurus hallucatus), a species that since 1788, has disappeared from an area equalling the size of France (Moore et al. 2019). The speed and magnitude of the northern quoll’s decline requires proportionate management interventions to help guarantee the persistence of this species (Appendix 1) (Hill and Ward 2010). It has long been recognised that the effectiveness of such interventions are largely determined by the foundation of ecological knowledge on which they are based (Burbidge et al. 2011).

In the previous chapters, I have enhanced our understanding of the northern quoll by quantifying its habitat requirements across three distinct scales: the range scale, the landscape scale, and the patch scale. As outlined in chapter one of this thesis my objectives were as follows;

1. Quantify changes in the northern quolls geographic range and ecological niche across the entire species distribution (chapter two). 2. Develop appropriate and standardised methods for surveying northern quolls using remote sensing cameras (chapter three). 3. Define characteristics of critical habitat for northern quoll within the Pilbara bioregion at the landscape scale (chapter four). 4. Define characteristics of critical habitat for northern quoll within the Pilbara bioregion at the patch scale (chapter five).

Collectively, my findings illustrate trends in the northern quoll’s decline, and highlight areas of most importance for their persistence, particularly in the Pilbara region, a last remaining stronghold for the species. Here, I contextualize the results of each chapter, and provide three important suggestions for future northern quoll research.

In chapter two, the first data chapter of this thesis, I assembled a database of all available northern quoll records (collected 1788 – 2019), and used them to estimate changes in the northern quoll’s geographic range between historic (prior to the year 2001) and contemporary (after the year 2000) time periods. By comparing the average values of environmental variables that shape northern quoll distributions within the historic and contemporary niches, I assessed a series of hypotheses regarding the types of environments that would favour northern quoll persistence. I found range declines were most severe in the eastern extent of the northern quolls range (Queensland), and became less severe

113 moving westward, corresponding with the spread of introduced cane toads (Rhinella marina) (Figure 6.1). I also found that the northern quolls contemporary range is comprised of proportionally higher quality habitat than its historical range, characterized by higher topographical ruggedness, annual rainfall, and rainfall seasonality (Figure 6.1). This result conforms with an extensive body of research which has found Australia’s mammal attrition rate to be greatest in areas that receive lower amount of annual rainfall, and that topographically rugged habitat provides important refuge for species threatened by predators including feral cats (Felis catus) and canids (Canis spp.).

Another key result from this study was that the Pilbara population has experienced less range contraction and niche reduction compared to other regions. This finding is consistent with both expert opinion and the current literature―though understudied, the Pilbara population is generally regarded as a stronghold for the northern quoll, due largely to the (current) absence of cane toads and abundance of topographically complex rocky outcrops (Cramer et al., 2016). Improving our understanding of the habitat requirements of northern quolls in the Pilbara is hence a priority, such that we can ensure the persistence of the species as a whole (Cramer et al. 2016). I addressed this research priority in chapters 4 & 5. However, I first needed to develop an effective means of monitoring northern quolls.

The cost and effort of surveying rare and cryptic species has been reduced substantially in many cases by the implementation of new technologies, such as camera traps (De Bondi et al. 2010; Meek et al. 2012; Welbourne et al. 2020). While vertically orientated camera traps have potential as a promising survey technique for northern quolls―given they collect high quality images of unique spot patterning located on the quolls dorsal surface, which can be used for individual animal identification (Diete et al. 2016; Hohnen et al. 2013) ― Previous studies using other species have found detectability can be lower than that of horizontal cameras, because the detection zone of vertically orientated cameras is smaller (Nichols et al. 2017; Taylor et al. 2014). In chapter three, I compared the effectiveness of vertical and horizontal camera trap designs for surveying northern quolls and other taxa, including Rothschild’s rock wallabies (Petrogale rothschildi), feral cats, and varanids (Varanus spp.) within our 6000 km2 study area in the Pilbara bioregion. While horizontal cameras were superior for sampling rock wallabies and feral cats, vertically orientated cameras were equally as effective as horizontal cameras, but provided additional information at the individual level.

114

Figure 6.1 – A conceptual diagram depicting the scale at which each of my thesis chapters where conducted, and habitat variables most important in predicting northern quoll responses within each of these scales.

115

In chapter four, I applied the vertically orientated camera trap design trialled in chapter three, to examine factors influencing quolls at the landscape scale―where both response and predictor variables were collected to represent entire landscapes. More specifically, I aimed to test the separate effects of habitat amount and habitat configuration on quoll occurrence and abundance in fragmented landscapes, in conjunction with other important landscape properties such as landscape composition and environmental gradients. Unlike much of the broader literature, I found habitat configuration, measured as edge density, was more important than habitat amount in influencing abundance (Figure 1, 2). I suggest this result is potentially explained by the ‘fragmentation threshold’ theory, which predicts that the importance of habitat configuration increases when habitat amount is low (<20–30%). I also found that quoll abundance increased with increasing topographical ruggedness. This was not surprising given I found similar results from our range scale analysis conducted as part of chapter 2 (Figure 6.1, 6.2). Topographical ruggedness is also mentioned frequently within the literature as an important characteristic defining northern quoll habitat, as was revealed within our systematic review in Appendix 1. During the wet season, I found northern quoll abundance increased with increasing rainfall in the previous wet season. This is likely explained by a lag effect in recruitment survival, where juvenile survival during the wet season before sampling was higher at sites that received more rainfall, translating to an increased number of adults and recruits a year later. Overall, the results from this study suggest remaining sections of northern quoll habitat which is topographically rugged and minimally fragmented holds the most value for conservation.

Finally, in chapter five, I expand on chapter four to determine the extent to which within-patch, patch, landscape variables, and matrix condition drive northern quoll habitat use and abundance within these landscapes. To do this, I used a focal patch design (sensu Kaplan and White 2002), where detections collected by each camera within the rocky patches were used to measure northern quoll responses to habitat at all three levels (within- patch, patch, and landscape); (i) at the within-patch scale, quolls were more likely to use patches with more potential denning crevices and increased vegetation cover, (ii) at the patch scale, quolls were more likely to use patches with smaller amounts of edge habitat relative to patch area, and (iii) at the landscape scale, quolls were more likely to use areas with a higher rocky habitat extent (Figure 1, 2). Use of the matrix by quolls tended to occur in close proximity to rocky habitat, and was more common in areas with high vegetation cover. Together, these results support the findings of chapter four, and indicate that critical habitat for northern quolls in the Pilbara is likely defined by large areas of condensed, complex rocky habitat. In addition, these results suggest that rocky habitat with intact vegetation occurring within and in surrounding area are likely to be most favourable.

116

Figure 6.2 – A conceptual diagram depicting key variables influencing northern quality habitat quality across multiple scales.

117

Broader implications

Collectively, the results from this thesis have improved our understanding of what characterises high quality northern quoll habitat, as well as how we can best detect and monitor northern quolls within it. Specific management implications arising from these findings are discussed in detail throughout. More broadly, a central theme from this research has been the importance of rocky habitat for the conservation of northern quolls. While this work is not the first to highlight the value of rocky habitat to northern quolls (Begg 1981; Braithwaite and Griffiths 1994; Hernandez-Santin et al. 2016; Oakwood 2002; Pollock 1999; Radford 2012), I was able to advance our understanding of the importance of specific rocky habitat features, such as the amount, configuration, as well as patch size, shape, and composition. Here, it is important to note that these findings may provide important theoretical insights relevant to additional species, particularly those occurring in rocky habitat. For example, within our study area, species such as Rothschild's rock wallaby (Petrogale rothschildi), common rock rat (Zyzomys argurus), Long-tailed dunnart (Sminthopsis longicaudata), and northern Pilbara rock monitor (Varanus pilbarensis) rely heavily on rocky habitat in order to persist (Menkhorst and Knight 2011; Wilson and Swan 2017). Like northern quolls, it’s possible each are impacted to some extent by factors such as fire, grazing, and predation, and as such, their responses to habitat characteristics tested in this thesis may be comparable. Lesson from this thesis could also be applied to species that occur beyond our study area, such as the Otago Skink (Oligosoma otagense), Andean cat (Leopardus jacobita), and long-tailed chinchilla (Chinchilla lanigera), all of which depend on rocky habitat, and are internationally recognised as endangered (IUCN 2021).

The methods I developed and used in this thesis also have broader applications. For example, the process I developed in chapter two to track northern quoll declines across Northern Australia has since been applied using data from an additional eight species from the Northern Territory (von Takach et al. 2020b): fawn antechinus (Antechinus bellus), brush-tailed rabbit rat (Conilurus penicillatus), black- footed tree-rat (Mesembriomys gouldii), pale field-rat (Rattus tunneyi), western chestnut mouse (Pseudomys nanus), northern brown bandicoot (Isoodon macrourus), common brushtail possum (Trichosurus vulpecula), and savannah glider (Petaurus ariel, previously Petaurus breviceps). Here, it was demonstrated that patterns in species decline were broadly similar to those I observed in the northern quoll; most species appear to have declined in range and niche, and have contemporary distributions characterised by high ruggedness and rainfall (von Takach et al. 2020b). I suggest these methods have useful applications for species elsewhere in Australia and internationally.In chapter three I demonstrated that vertically orientated cameras were an effective method for surveying northern quolls, however, it’s likely this method is also suitable for surveying other species of quoll, given all have unique spot patterning located on their dorsal surface. Further, I have since demonstrated that vertically orientated cameras have applications beyond quolls, by using them to survey perenties

118

(Varanus giganteus) ―Australia’s largest species of lizard (Moore et al. 2020a) . Here, I was not only able to individually identify animals, but also glean basic morphological data from images.

Moving forward

The work presented in this thesis addresses a broad scope of research objectives relevant to the ecology and conservation of northern quolls. Nonetheless, the northern quoll is still declining (Appendix 1), and the scale of its geographical range, combined with the diversity of threats impacting populations leaves many questions for future research to address. In Appendix 1, I provide a non-exhaustive list of 10 future research directions for northern quolls, which surfaced as part of our systemic review of literature relevant to the ecology and conservation of northern quolls. These include: (i) resolving the northern quolls taxonomy, (ii) clarifying the status of the Queensland northern quoll populations; (iii) understanding mechanisms allowing the persistence and resistance of northern quoll populations during cane toad invasions; (iv) quantifying the impacts of mining; (v) further investigating the threat of population isolation and feasibility of genetic rescue; (vi) unwinding interacting threats; (vii) predicting the impact of climate change; (viii) further incorporating Indigenous knowledge; (ix) harnessing the heritability of toad avoidance behaviour; and (x) further investigating the potential for artificial refuges to be used a conservation tool for northern quolls. Here, I summarise three of these future research directions most relevant to, and in context of the work presented as part of this PhD.

Unwinding interacting threats

Northern quolls face multiple co-occurring threats across their range (Appendix 1); however, limited research has investigated if and how these threats interact synergistically. While I was able to investigate the impact of fire as part of this thesis (chapters 4 & 5), our study design was such that our capacity to measure the effects of other threats simultaneously such as predation and overgrazing was limited. In the case of predation, this was because a large proportion (50%) of our camera traps were orientated in a way that was inappropriate for detecting important predators such as feral cats (chapter 3) and likely dingoes as well. Similarly, in the case of overgrazing, a reliable way of detecting herbivore activity was not included in our study design. Nonetheless, studies focused on other native mammals occurring in northern Australia demonstrate that threats may be compounding, and management that addresses only single threats may be ineffective (Legge et al. 2019). Better understanding northern quoll ‘threat webs’―a group of co-occurring threats that may have additive or non-additive impacts on each other―may improve the ability of land managers to focus efforts towards ultimate threats, rather than proximate threats, hopefully with improved conservation outcomes (Geary et al. 2019b).

Understanding the synergetic impacts of predation, fire, and grazing, on northern quolls is likely to be particularly important to their conservation, given each of these threats occurs in tandem across the

119 majority of their existing range. Untangling these interactions may also explain how the impact of dingo predation on northern quolls has been exacerbated since European arrival in Australia (Appendix 1), as well as how removing dingoes is likely to mediate the impact of feral cat predation on quolls. An important tool in untangling these threats is the use of manipulative experiments, where at least one threat within a system is artificially controlled (typically as part of conservation management activities), such that its relative impact on northern quoll populations can be measured in context with co-occurring threats. While experiments of this nature are already under way (eg. Palmer 2019), opportunity for further research in this area is likely to exist in areas where threat management (e.g., controlled burns, predator baiting) within northern quoll habitat is planned or actively occurring already.

Quantifying the impacts of mining

Mining activity occurs across the northern quolls entire range, including important cane toad-free populations such as Pilbara bioregion. Yet, few studies have investigated the impact of mining on northern quoll populations (Appendix 1). While I was not able to explore the impacts of mining related disturbance on northern quolls directly as part of this thesis, the results of chapter 4 and chapter 5 suggest habitat fragmentation that could potentially be caused by mining activity―more specifically, the creation of increased habitat edge―would likely have substantial adverse impacts. As such, determining the extent to which mining activity is likely to influence the persistence of remaining northern quoll populations through more targeted studies should be a priority for future research. This is particularly true for the Pilbara population, where overlap between the northern quolls geographic range and mining tenure is extremely high (DMIRS 2020). This is acknowledged by Cramer et al. (2016), who identify the impacts of linear infrastructure on northern quoll movement and the ability of the northern quolls to recolonise disturbed areas or artificial habitat as key research areas. In addition to these areas, I suggest future studies investigate secondary impacts of mining activity on northern quoll populations, such as increased predator densities surrounding mining camps, potentially as a result of increased resource subsidies (e.g., food, water).

The potential for artificial refuges as a conservation tool for northern quolls

In chapter 5, I demonstrated that while rocky habitat is important for northern quolls, the most favoured rocky habitat is that with high den availability. This is presumably because rocky dens provide a critical source of refuge for northern quolls to shelter during the day, and store young in during the breeding season (Oakwood 2000). When rocky habitat is degraded or removed, recent trials have shown that quolls will potentially use artificial refuges―man-made refuges created to replicate natural northern quoll dens (Cowan et al. 2020b). However, evidence that artificial refuges are being used as both temporary and natal den sites is currently lacking, as is our understanding of the dimensions that artificial refuges should be constructed to maximise northern quoll useability (Cowan et al. 2020b).

120

Future trials are required to address these knowledge gaps. Further, in light of results from chapter 5 which suggest vegetation cover is a major factor in predicting northern quoll habitat use, I suggest that future studies testing artificial refuges in disturbed habitats incorporate complementary management techniques such as the management of fire and exclusion of introduced herbivores.

121

122

Appendix A – A brief history of the northern quoll (Dasyurus hallucatus): a systematic review

Manuscript submitted: Moore, H., Dunlop, J., Jolly, C., Kelly, E., Woinarski, J., Ritchie, E., Burnett, S., van Leeuwen, S., Valentine, L., Cowan, M., and Nimmo, D. A brief history of the northern quoll (Dasyurus hallucatus): a systematic review. Australian Mammalogy.

Harry A. Moore1, 2, Judy A. Dunlop3, Chris J. Jolly 1,, Ella Kelly4, John C. Z. Woinarski5 , Euan G. Ritchie6, Scott Burnett7, Stephen van Leeuwen8, Leonie E. Valentine2, Mitchell A. Cowan1, Dale G Nimmo1

1Institute for Land, Water and Society, School of Environmental Science, Charles Sturt University, Albury, 2640, New South Wales, Australia 2School of Biological Sciences, University of Western Australia, Crawley, 6009, WA, Australia 3Department of Biodiversity, Conservation and Attractions, Bentley Delivery Centre, Locked Bag 104, Perth, WA, Australia 4Department of Environment, Land, Water and Planning, Nicholson Street, East Melbourne, 3002, VIC, 5Threatened Species Recovery Hub, National Environmental Science Program, Charles Darwin University, 0810, Darwin, NT, Australia 6Centre for Integrative Ecology and School of Life and Environmental Sciences, Deakin University, Burwood, 3125, VIC, Australia 7School of Science and Engineering, University of the Sunshine Coast, Sippy Downs, 4556, QLD, Australia. 8School of Molecular and Life Sciences, Curtain University, Bentley, 6102, WA, Australia

123

Abstract

In response to Australia’s current extinction crisis, substantial research efforts have been targeted towards some of the most imperilled species. One such species is the northern quoll (Dasyurus hallucatus); a marsupial predator that has recently suffered substantial declines in range and is now listed as endangered. We conducted a systematic review of all literature relevant to the conservation and ecology of northern quolls. We reviewed 137 publications, including research articles, government and industry reports, theses and books, and quantify research effort in terms of topic, location, and publication period. We then summarise research relevant to their taxonomy, genetics, distribution, habitat associations, diet, reproduction, movement, threats, management, and Indigenous knowledge.

Research effort was higher in the most recent decade than the previous four combined. Northern quolls in the Northern Territory were the most studied, followed by the Pilbara, the Kimberley, and

Queensland populations. The majority of publications focused on northern quoll distribution and habitat, management, and threats; primarily cane toads, predation, and fire. In concluding we provide a non-exhaustive list of 10 future research directions that if pursued, are likely to provide information critical to managing northern quolls in way that minimises future declines.

124

Introduction

Biodiversity conservation and management is likely to be most effective when derived from a robust evidence base (Salafsky et al. 2019). Despite an ever growing amount of ecological information

(Bornmann and Mutz 2015), there remains a considerable gap between the science of conservation biology and the on-ground implementation of evidence-based policy (Evans and Cvitanovic 2018;

Knight et al. 2008; Rose et al. 2019). One reason for this is that conservation practitioners can struggle to access, interpret, and synthesise an increasingly dispersed and voluminous body of scientific literature (Pullin and Knight 2005). Similar problems are encountered by researchers, leading to repetition and redundancy in ecological science (Pulsford et al. 2016). Systematic literature reviews are one means of overcoming these issues by identifying, selecting, and appraising research on a predefined topic, drawing out evidence-based management implications and critical knowledge gaps for further research (Moher et al. 2009).

Australia’s contemporary mammal extinction rate is the globe’s highest (Woinarski et al. 2019a). Yet more Australian mammals are forecast to become extinct within the next two decades (Geyle et al.

2018). In response to this crisis, increased research effort has been directed towards Australian species that have suffered population contractions over the last 200 years (Fleming and Bateman 2016). One such species is the northern quoll (Dasyurus hallucatus) ―a medium-sized marsupial predator (240–

1120 g) endemic to northern Australia. Prior to European colonisation in 1788, northern quolls were distributed across much of northern Australia (Braithwaite and Griffiths 1994), but have since suffered substantial declines (Fig A1) (Moore et al. 2019). As such, northern quolls are listed as Endangered both nationally (TSSC 2005) and globally (Oakwood et al. 2016). Due to their ongoing and precipitous decline, much of the research focused on the ecology and conservation of northern quolls has occurred within the last two decades. However, accessing this research can be challenging, largely because it is spread across a diverse and scattered literature (e.g., books, journals, government and consultant’s reports, and theses). To help overcome this barrier, we conducted a systematic review of all literature relevant to the ecology and conservation of northern quolls across their entire range.

125

Figure A1. The historic and contemporary range of the northern quoll (Dasyurus hallucatus). Figure adapted from (Moore et al. 2019). Note that islands occupied by the northern quolls (except Groote Eylandt) are not depicted: these are listed in S1.

The complexities of species conservation are such that multiple disciplinary approaches are often required in order to achieve meaningful progress (Dick et al. 2016). With this in mind, we take a deliberately broad approach to our review. We summarise research related to the northern quoll’s diet, distribution and habitat associations, genetics and taxonomy, reproduction, movement, and threats and management. We also summarise available Indigenous knowledge related to the northern quoll’s ecology. We identify and discuss knowledge gaps throughout, and provide a non-exhaustive list of future research directions for the northern quoll, which if applied, could be useful in providing knowledge critical to improving the conservation of northern quolls.

126

Materials and Methods

Database compilation

We followed the approach used by Ashman et al. (2019) to collate relevant literature for this review.

Three electronic databases were searched (Web of Science, Scopus and Google Scholar) on the 17th of

August 2020 using the search terms “northern quoll” OR “Dasyurus hallucatus”. Search terms were located in publication title, abstract and keywords. Searches retrieved publications from a range of categories including peer-reviewed literature, Master’ and PhD theses, government and publicly available industry reports, and government action and recovery plans. Publications were reviewed in a three-step process (Fig A2). First, duplicates were removed, then titles and /or abstracts were screened to detect the terms “northern quoll” or “Dasyurus hallucatus”, or a reference to a broader community of species that is likely to include northern quolls—for example ‘tropical mammals of Australia’. Last, full texts were reviewed. Publications were excluded if they were not immediately relevant to ecology and conservation of the species. Therefore, for example, publications that focused solely on the physiology or anatomy of northern quolls were removed, unless the publication made explicit reference to the relevance of their findings to ecology or conservation of wild populations (e.g., predator-aversion trials conducted in captivity etc).

127

Figure A2. Flow diagram of publications included and excluded from review.

128

Data retrieved

For each of the 137 publications that were included in the final analysis, we recorded date of publication, study population(s) (if applicable), and study site(s) (if applicable). Study populations were adapted from (Moore et al. 2019) and comprise of four largely spatially segregated units: Queensland, Northern

Territory, the Kimberley region of Western Australia, and the Pilbara region (including the neighbouring Little Sandy and Great Sandy Desert bioregions) of Western Australia. We also extracted home range estimates from publications that listed them. We pooled publication dates into five categories: prior to 1980, 1981–1990, 1991–2000, 2001–2010, 2011–2020. We used decadal increments to categorise publications given they were fine enough to detect changes in research effort through time, but also coarse enough to capture a substantial number of publications within each increment. We then categorised publications into one or more of nine topics, adapted from Ashman et al. (2019):

 Taxonomy and Genetics

 Distribution, declines and habitat associations

 Diet (prey choice, scat composition)

 Reproduction

 Movement

 Threats

 Management (direct management actions, legislation, action and recovery plans)

 Indigenous knowledge

 Miscellaneous (methods, reviews, albinism)

129

Threats

For publications that included the topic ‘threats’, we categorised the threats examined within the publication according to threat categories adapted from the National Recovery Plan (Hill and Ward

2010). Threat categories adapted from the recovery plan included the introduced and toxic cane toads

(Rhinella marina), feral cat (Felis catus) and dingo/dog (Canis spp.) predation, fire, grazing, habitat clearing, mining, disease, feral predator baiting, and vehicle strikes.

Topic summary and future research needs

We provide written summaries for all research topics included in the review, with the exception of

‘behaviour’, which is discussed within the management section, and ‘miscellaneous’, which is discussed throughout. Within summaries, we discuss important findings in context with the ecology and conservation of northern quolls. We also identify and discuss knowledge gaps to highlight areas that may require future research. We finish by providing a non-exhaustive list of ten future research areas that could be useful to refine management action(s) to conserve northern quolls.

130

Results & Discussion

The temporal distribution of the 137 publications shows a substantial increase in research effort in recent years (Fig A3). The earliest study included in our analysis was published in 1926 (Thomas 1926) and the most recent was published November 2020 (Ondei et al. 2020). Most publications were conducted between 2011–2020 (87), and few publications were conducted prior to 1981 (2) (Fig A3). The region with the most publications was the Northern Territory (76), followed by the Pilbara (47), the Kimberley

(44), and Queensland (29) (Fig A4). We identified 18 publications which were at least partly lab-based.

Each of the nine study topics were represented at least once within publications included in this review

(Fig A5). It is important to note that we found a limited amount of grey literature in this review and, as such, a number of environmental impact assessments (and the implications of their findings) were likely missed.

Figure A3 Research relevant to the conservation and ecology of northern quoll (Dasyurus hallucatus) categorised by publication date.

131

Figure A4. Research relevant to the conservation and ecology of northern quoll (Dasyurus hallucatus) mapped by population. Insets represent top three research topics within each populations. Dark grey area represents northern quoll contemporary distribution (prior to year 2000) adapted from (Moore et al. 2019).

132

Figure A5. Research relevant to the conservation and ecology of northern quoll (Dasyurus hallucatus) categorised by research topic and location.

133

Taxonomy and genetics

We identified 20 publications relevant to northern quoll taxonomy, covering all major populations.

Northern quolls (Dasyurus hallucatus Gould, 1842) were described by John Gould (1842) from two specimens collected at Port Essington in the Northern Territory. They are the smallest of six Dasyurus species: including the spotted-tailed quoll (Dasyurus maculatus), western quoll (Dasyurus geoffroii), eastern quoll (Dasyurus viverrinus), bronze quoll (Dasyurus spartacus), and New Guinean quoll

(Dasyurus albopunctatus). The latter two species are only found in Papua New Guinea and Indonesia.

Northern quolls. The current range of the northern quoll overlaps with the current range of the northern sub species of spotted tailed quoll (Dasyurus maculatus gracilis) in North Queensland. In the Pilbara, the northern quolls range overlaps with the historic distribution of the western quoll, as suggested by sub-fossil evidence (Baynes and McDowell 2010).

In 1926, northern quolls were separated into four subspecies, based mostly on the width of the skull at the nasals: D. h. hallucatus (Northern Territory to central Queensland); D. h. predator (Cape York,

Queensland); D. h. exilis (the Kimberley); and D. h. nesaeus (Groote Eylandt) (Thomas 1926). No

Pilbara specimens were included in this examination. While these subspecies are no longer recognised

(Jackson et al. 2015), there is now genetic evidence to suggest northern quolls in each of the four major populations (Queensland, Northern Territory, the Kimberley and the Pilbara) do represent distinct lineages. Firestone et al. (2000) suggest northern quoll cytochrome b sequences from Queensland and the Northern Territory are at least as divergent as those between western quolls and the bronze quoll.

Woolley et al. (2015) and Hohnen et al. (2016b) found that the Northern Territory and Kimberley northern quoll populations are genetically divergent, and How et al. (2009) found the same for

Kimberley and Pilbara populations. These results of more recent morphological examinations are mixed

― Umbrello (2018) found significant differences in skull size, dentition, and external characteristics between Queensland, Northern Territory, Kimberley, and Pilbara populations, while Viacava et al.

(2020) found few consistent differences. Some island populations separated from the mainland by permanent sea channels (Bigge, Boongaree, Koolan), as well as less permanent channels (Dolphin

Island) also appear genetically divergent from mainland populations (How et al. 2009). However, this

134 is not the case for the Groote Eylandt population, which genetically aligns with the mainland Northern

Territory population (Woolley et al. 2015).

The disjunct distribution of mainland northern quolls accompanied by the genetic and morphological differences between populations correspond to biogeographical barriers across northern Australia

(Bowman et al. 2010). Separating the Queensland and Northern Territory populations is the Carpentaria

Gap; a series of clay pans that limit dispersal between Cape York and the rest of the Australian monsoonal tropics for a range of taxa (Bowman et al. 2010). For example, the Carpentaria Gap separates a recently described species of glider, Petaurus ariel, from its sister species, Petaurus notatus (Cremona et al. 2020). Similarly, the gap between the Northern Territory and Kimberley populations aligns with the Ord Arid Intrusion, which divides sandstone blocks between Arnhem Land in the Northern Territory and the Kimberley (Bowman et al. 2010), and acts as a dispersal barrier for other species that use rocky habitat like rock-wallabies (Petrogale spp.) (Potter et al. 2012). The Kimberley and Pilbara populations of northern quoll are separated by the Great Sandy Desert―an extensive dune system often implicated in the isolation of both species and populations (Edwards et al. 2017).

The taxonomic status of the northern quoll may not yet be fully resolved. Lack of taxonomic clarity has previously hampered conservation efforts in other taxa. In the case of the northern quoll, understanding if populations should be treated as distinct taxonomic units or are genetically similar enough to be intermixed is likely to have important management implications, particularly in relation to cross- population translocations (as discussed in ‘management’ section). As such, further clarifying the extent to which northern quoll populations differ both genetically and morphologically should be a priority for future research. Other research related to northern quoll genetics are discussed within the ‘Management’ section.

Distribution, habitat associations, and geographic range contraction

135

A total of 74 publications were included in the ‘Distribution, declines and habitat associations’ topic, with the most focused on the Northern Territory population (n = 39), followed by the Kimberley (n =

27), Pilbara (n= 23), and Queensland populations (n = 11) (Fig A5).

Prior to European colonisation, the geographic range of northern quolls incorporated much of northern

Australia above the Tropic of Capricorn (90 ° South) and within 200 km of the coastline (but likely extended inward much further (Turpin and Bamford 2015)) (Braithwaite and Griffiths 1994), covering a total area of over 1.2 million km2. Average annual rainfall across this area varies substantially, ranging from 220 mm in the eastern extent of the Pilbara bioregion, to nearly 4500 mm in the Wet Tropics of northern Queensland (Australian Bureau of Meteorology 2020). Similarly, average maximum temperature in the warmest months ranges from 42.1°C in the Pilbara bioregion to 25.5°C in southern

Queensland (Australian Bureau of Meteorology 2020). Northern quolls naturally occur on at least 32 islands, mostly in the Northern Territory and Kimberley (S1). Many of these are relatively free from human disturbance and lack introduced predators and/or cane toads (How et al. 2009; Woinarski et al.

2007). In the Northern Territory, northern quolls are particularly associated with large, remote and rugged islands (Woinarski et al. 2007), although they are notably absent from two large Northern

Territory islands—Bathurst (2600 km2) and Melville (5786 km2). In 2003 and 2017, northern quolls were translocated to three islands in the Northern Territory outside of their historical geographic range

(Kelly 2018; Rankmore et al. 2008) (See management section).

Northern quolls occur across a broad range of habitat types—including tropical and monsoonal rainforest, Eucalyptus woodlands, Eucalyptus open forests, lowland savanna, vine thickets, on beaches, and amongst human settlements (Begg 1981; Braithwaite and Griffiths 1994; Oakwood 1997; Pollock

1980; Schmitt et al. 1989)—but appear most abundant in rugged and rocky landscapes, including rocky hills, patches of granite outcrops, boulder-strewn slopes, rocky creeklines, and gorges (Begg 1981;

Braithwaite and Griffiths 1994; Calaby 1973; Ibbett et al. 2018; Kitchener et al. 1981; McKenzie et al.

1975; Oakwood 1997; Olds et al. 2016; Pollock 1980; Schmitt et al. 1989) (Fig A6). This preference may be due to rocky habitats holding permanent water and potentially greater prey availability than surrounding habitats (Braithwaite and Griffiths 1994; Burnett 1997), however see Oakwood (1997).

136

Figure A6. Northern quoll habitat. A) Hazelwood Gorge, Queensland , B) East Alligator River, Northern Territory, C) Charnley River Wildlife Sanctuary, the Kimberley, Western Australia, and D) Indee station, the Pilbara, Western Australia. Image credit ― Dennis Jeffery, Catherine Marshall, Naomi Indigo, Daniel Bohorquez Fandino.

Another factor linking quolls to rocky habitats is the availability of shelter, particularly dens―small enclosed spaces that northern quolls use as either short-term shelter sites (temporary den) or semi- permeant dwellings used to raise offspring (natal den). Northern quolls den in tree hollows (both alive

137 and dead), termite mounds, logs, and goanna burrows (Oakwood 1997). In more arid regions that lack trees and logs (e.g., the Pilbara), rocky crevices within granite boulder piles and rocky mesas are critical for providing both temporary and natal den sites for northern quolls (Cowan et al. 2020b). In addition to providing protection from terrestrial predators such as dingoes and feral cats, rocky dens also provide important protection from climatic exposure. For example, in the Pilbara, Cowan et al. (2020b) found rocky dens sites were critical in sheltering from extreme external temperatures, which often exceeded safe temperatures for northern quolls (i.e., >36.5 °C) (Cooper and Withers 2010). In habitats where alternative den sites are available (e.g., tree hollows and logs), females selectively den in rocky habitat, and there is some evidence that females with more rocky habitat within their home range survive longer

(Oakwood 1997).

Since European colonization of Australia (1788), and particularly within the past 50 years (Braithwaite and Griffiths 1994), the geographic range of northern quolls has declined by at least 45.2%, and width of their ecological niche has also declined substantially (Moore et al. 2019). Declines have been most severe in Queensland, where >400,000 km2 of former habitat is now unoccupied, constituting a range contraction of >75% (Moore et al. 2019). The Northern Territory is the second most affected population, which has experienced a 58% range contraction (115,024 km2), mostly from the more arid southern extent of their historic distribution (Prior to the year 2000) (Braithwaite and Griffiths 1994;

Moore et al. 2019; von Takach et al. 2020b; Ziembicki et al. 2013). The Kimberley population has seen a 17% decline, equating to 25,986 km2 of lost range; (Moore et al. 2019), while the Pilbara―the most arid region currently supporting northern quoll populations―has so far seen little to no evidence of decline (Moore et al. 2019; Spencer et al. 2013).

These broadscale declines in geographic range arise through numerous local declines and extincitons that have been well documented. For example, Burnett and Zwar (2009) found no evidence to suggest northern quolls persist in the southern Mary River catchment north of Brisbane, despite the known occurrence of historic populations. Populations in Far North Queensland have also been subject to substantial decline. For example, Perry et al. (2015) and Burnett (1997) found northern quolls are largely absent from sizeable areas of Cape York Peninsula where they once occurred. In the Northern

138

Territory (Ibbett et al. 2018) found northern quoll trap success in Kakadu National Park was significantly lower in 2002 than it was in 1980 (Begg 1981), and Woinarski et al. (2011b) found northern quoll abundance was lower at sites within Kakadu in 2007–2009 than it was in 2001–2004.

There are clear spatial patterns in the decline and persistence of northern quoll populations. In

Queensland, the Northern Territory, and the Kimberly, topographically simple landscapes that receive low rainfall and are distant from permanent water have seen extremely severe declines (Burnett 1997;

Moore et al. 2019; Pollock 1999; Woinarski et al. 2008), and areas that would have historically been marginal habitat have seen the greatest declines (Moore et al. 2019). Hence, extant populations tend to occur in topographically rugged areas with high annual rainfall (Moore et al. 2019). The extent of decline across populations corresponds with the length of time that the population has co-occurred with the introduced cane toad (see threat section). Persisting mainland populations of northern quolls now exist on the Central Mackay coast, Wet Tropics, Einasleigh Uplands and Cape York Peninsula bioregions in Queensland, the Arnhem Plateau, Darwin Coastal Plain, Daly Basin, Pine Creek bioregions, and Groote Eylandt in the Northern Territory, the Central and North Kimberley bioregions, and the Pilbara bioregion.

Diet

We identified 27 publications that examined the diet of the northern quoll, representing all study populations, with the most publications conducted in the Northern Territory. A large number of these publications (n = 14) focused on the consumption of cane toads, or baits designed to kill cats and dingos

(n = 7) by quolls, and are discussed in the ‘threats’ section below. In general, northern quolls are omnivorous, opportunistic foragers, consuming a range of invertebrate, vertebrate, and plant species.

Research examining sex-based dietary differences is scarce; however, there is some evidence that females northern quoll consume less vertebrate prey items than males, probably because females are typically smaller in mass (Oakwood 1997).

Invertebrates are a dominant feature in the diets of northern quolls across all populations, with beetles

(Coleoptera), grasshoppers (Orthoptera), ants (Hymenopterans) and spiders (Arachnida) appearing

139 most frequently within scats (Dixon and Huxley 1985; Dunlop et al. 2017; Oakwood 1997; Pollock

1999; Radford 2012). Indigenous people from Arnhem Land in the Northern Territory also observed northern quolls feeding on wai (worms), grubs, and moths (Dixon and Huxley 1985). In Kakadu

National Park (Northern Territory), Oakwood (1997) found invertebrate consumption peaked in the early dry-season, coinciding with the arrival of juveniles into the population. In the Pilbara, invertebrate occurrence in the northern quoll diets decreases with the occurrence of rodents and plant material, potentially indicating invertebrates may be a staple food item, but not always preferred (Dunlop et al.

2017).

A diverse range of vertebrate prey also appear in the diet of northern quolls, including rodents (Melomys burtoni , Pseudomys delicatulus, Pseudomys hermannsbergensis, Rattus rattus, Rattus sordidus,

Zyzomys argurus), rabbits (Oryctolagus cuniculus), other dasyurids (Dasykaluta rosamondae, Ningaui timeleyai, Pseudantechinus sp., Sminthopsis macroura, Sminthopsis youngsoni), gliders (Petaurus spp.), possums (Trichosurus vulpecula), bandicoots (Isoodon auratus, Isoodon macrourus ), bats

(Nyctophilus spp., Rhinonicteris aurantia), birds, lizards (Scincidae spp., Agamidae spp., Varanidae spp., Gekkonidae spp.), snakes, and frogs (Dixon and Huxley 1985; Dunlop et al. 2017; Oakwood 1997;

Pollock 1999; Radford 2012). Larger mammals including kangaroos (Osphranter spp.), cows (Bos taurus), cats (Felis catus) and dogs/dingoes (Canis spp.) have also been recorded in northern quoll scats

(Dunlop et al. 2017), presumably consumed as carrion. In the Kimberley, northern quolls consume a larger proportion of larger prey, such as golden bandicoots (Isoodon auratus), in recently burnt habitats, potentially because hunting this prey is easier when vegetation cover is reduced (Radford 2012).

Plant material has been recorded in the diet of northern quolls in the Northern Territory (Oakwood

1997), the Kimberley (Radford 2012), and the Pilbara (Dunlop et al. 2017), and is typically comprised of fleshy fruits, seeds, and flowers. In the Northern Territory, Oakwood (1997) found fruits from wild grape plants (Ampellocissus acetose) were the most common plant material to appear in northern quoll diets, with peak consumption occurring in the late-dry to early-wet season. In the Pilbara, native figs

(Ficus spp.) were the most frequently recorded plant group within northern quoll scats, occurring within

16.1% of total scats measured (Dunlop et al. 2017).

140

Reproduction

We identified 17 publications related to northern quoll reproduction. The timing of mating varies among regions typically occurring between June and July in Queensland, May and June in the Northern

Territory, June to August in the Kimberley, and July to September in the Pilbara.

Male northern quolls are known to exhibit semelparity―a reproductive strategy where animals only breed once in their lifetime―characterized by increased levels of testosterone followed by rapid condition loss, with individuals rarely living longer than 11 months (Oakwood et al. 2001), although survival varies.

In the Northern Territory, most male quolls die within two weeks of mating (Dickman and Braithwaite

1992; Oakwood 2004a), although a small percentage may survive to a second breeding season (Begg

1981; Schmitt et al. 1989). Female survival is also very low in the Northern Territory (typically less than ~ 40% between years), although they can survive for up to three years (Braithwaite and Griffiths

1994; Cremona et al. 2017b; Oakwood 2000). In the Kimberley, Schmitt et al. (1989) found 4% of males and 37% of females survived to reach a second breeding season. In the Pilbara, just over 5% of males and 40% of females survive to a second breeding season (Hernandez-Santin et al. 2019).

A consequence of large annual die-offs in the adult population of northern quolls is that the likelihood of population persistence is heavily reliant on offspring survivorship. For example, in the Pilbara, an increase in mean juvenile mortality of 5% could potentially result in a 20% decline in overall population size (Moro et al. 2019). One strategy northern quolls may use to increase juvenile survival is polyandry―where females mate with multiple males to increase genetic diversity among offspring, conferring a group net fitness benefit. For example, a recent study found that 100% of examined northern quoll litters (n = 16) had young sired by multiple males, and in some litters, every offspring was fathered by a different male (Chan et al. 2020).

Following mating, females undergo a gestation period of between 21–25 days before giving birth

(Oakwood 2000). Mothers have between five and nine nipples (normally 8) (Begg 1981; Braithwaite

141 and Griffiths 1994). Although up to 17 young can be born (Nelson and Gemmell 2003), nipple number determines the maximum number of young that can be carried after birth. Braithwaite and Griffiths

(1994) found that, on average, females that lived closer to creeklines―areas that are more productive than surrounding habitat, and likely more conducive to reproduction―had more nipples than females that lived further from creeklines. Once young are attached to a nipple, mortality over the following three months ranges from less than 2% (Oakwood 2000) to 86% (Braithwaite and Griffiths 1994).

Young are deposited in dens at roughly two and a half months of age (Oakwood 2000), are independently foraging at four months, and are trappable by five months of age (Oakwood 1997).

Young are weaned by six months and disperse shortly thereafter (Oakwood 2000).

We found little information regarding northern quoll reproduction in Queensland. For example, no research included in this review documented whether quolls in Queensland show evidence of complete or partial annual male die-off, as observed in other regions (Oakwood et al. 2001). Documenting the post-breeding survival rate of northern quolls in Queensland will allow managers to better estimate the risk of Queensland sub-populations becoming locally extinct.

Movement

We identified 12 publications that investigated the movement ecology of northern quolls, which were spread across all regions (Figure A5). Seven of these publications included home range estimates, calculated using three different methods (Delimiting Supposed Home Range, Minimum Convex

Polygon, and 95% Minimum Convex Polygon) (Table A1). Across all home-range publications, a general trend appears to be that male home ranges are typically larger than female home ranges, especially in the breeding season, when males move large distances in search of females. In the Northern

Territory, Oakwood (2002) found that males occupied a much larger home range (84 ± 16 ha) than females (34.8 ± 6.4 ha). The difference was largest in the breeding season when males expanded their home-range to seek mating opportunities (Oakwood 2002). These males also travelled further between dens (average = 1.9 km) than females (average = 1.2 km). Similarly, on Groote Eylandt (Northern

Territory), Heiniger et al. (2020) found the average home range of male northern quolls (215 ± 58.4 ha) to be four times larger than that of females (53.1 ± 38.8 ha), although this was largely due to male quolls

142 expanding their home ranges by an average of 300% during the breeding season. Prior to the breeding season beginning, average home range sizes were larger for females (79.0 ± 58.8 ha) than they were for males (72.9 ± 24.4 ha) (Heiniger et al. 2020).

Table A1. Northern quoll (Dasyurus hallucatus) home range estimates (ha) sourced from publications included in review. MCP refers to minimum convex polygon. DSHR refers to Delimiting Supposed Home Range.

Author Location Female Male Data type Method Oakwood (2002) Northern Territory 34.8 84.1 VHF MCP ± 6.4 ± 16 n = 7 n = 8 Heiniger et al. 2020 Northern Territory 53.07 215.4 GPS MCP (Groote Eylandt) ± 38.77 ± 58.24 n = 10 n = 29 Schmitt et al. (1989) Kimberley 2.30 1.80 TRAP DSHM ± 1.20 ± 1.60 n = 7 n = 2 Cook (2010b) Kimberley 7 64 VHF MCP ± 2 ± 37 n = 11 n = 11 King (1989) Pilbara 168 464.75 VHF MCP ± 32.25 ± 200.245 n = 4 n = 4 Cowan et al (2020) Pilbara 13.8 301.4 VHF MCP (Red hill) ± 6.6 ± 108.9 n = 10 n = 10 Cowan et al (2020) Pilbara 32.5 931.1 VHF MCP (Yarraloola) ± 10.7 ± 259.9 n = 10 n = 11 Hernandez-Santin et Pilbara 34 193 GPS 95 % MCP al (2020) n = 1 ± 55 n = 8

In the Kimberley, (Cook 2010a) found home ranges were on average larger for males (64 ± 37 ha) than females (7 ± 2 ha). Maximum distance between dens was also greater for males (1.2 km) than females

(0.4 km). Schmitt et al. (1989) found quolls moved further between successive trap locations in the breeding season (104 ± 99 m) when compared to the non-breeding season (61 ± 82 m). In the Pilbara, average male home range estimates were between 2.7 and 28.6 times larger than female home range estimates (Table A1). Similar to the Northern Territory and Kimberly populations, male quolls in the

Pilbara appear to move further in the breeding season when compared to the non-breeding season

(Hernandez-Santin et al. 2020; Oakwood 2002; Schmitt et al. 1989). Although we did not find any direct measurements of home range for northern quolls in Queensland, Burnett et al. (2013) provides a

143 mean half maximum distance moved for 25 individuals (334.6 m), suggesting a crude circular home range estimate of 35 ha.

Lab-based publications investigating northern quoll locomotion have found northern quolls tend to sacrifice speed in favour of manoeuvrability in order to avoid making mistakes (Amir Abdul Nasir et al. 2017; Wynn et al. 2015). On Groote Eylandt, northern quolls with greater agility when moving around corners are more likely to survive their first 21 months of life than quolls that move slower around corners, potentially because they are better at avoiding predators such as dingoes, feral cats and birds of prey (Rew-Duffy et al. 2020).

Although male-biased dispersal is common in other dasyurids, evidence for this in northern quolls is limited (Oakwood 2002). Further research is required to better understand patterns in northern quoll dispersal. However, there is evidence that male northern quolls disperse further than females (Oakwood

2000) and male consecutive dens can be up to 4 km apart (Cook 2010b). In the Kimberley, genetic data suggests that habitats with higher annual rainfall and lower topographical ruggedness are likely to facilitate increased dispersal between sub-populations (Hohnen et al. 2016b).

Threats

A total of 66 publications included the topic ‘threats’, most of which were focused on northern quolls in the Northern Territory (n = 38), with substantially fewer publications focused on the Pilbara (n = 18), the Kimberley (n = 16) or Queensland (n = 9) (Fig A7).

144

Figure A7. Research relevant to threatening processes for northern quolls (Dasyurus hallucatus) categorised by threat and location. Threat categories were derived from the ‘National Recovery Plan for the Northern Quoll’ (Hill and Ward 2010).

Cane toads The most commonly investigated threat was cane toads (n = 22). In 1935, cane toads (Rhinella marina) were introduced to a research station near Cairns, Queensland, Australia (17°04’S 145°47’E) (Lever

2001). From there, toads quickly expanded their distribution into other parts of Queensland. Cane toads first invaded the Northern Territory in the 1980s (Freeland and Martin 1985), reached Kakadu National

Park in 2001 (Woinarski et al. 2002) and progressed through to Western Australia around 2009.

Like some other native predators, northern quolls that attempt to consume cane toads rapidly succumb to their novel and potent defensive toxins (Shine 2010). Oakwood (2004c) found 31% of radio-tracked quoll mortalities in Kakadu were likely caused by cane toads, while O’Donnell et al. (2010) found 29%, and Jolly et al. (2018a) found 85% of toad-naïve quolls and 18% of toad-trained quolls died as a result of consuming cane toads.

145

In Far North Queensland, Burnett (1997) presented anecdotal evidence that cane toads were the cause of northern quoll extirpation from 1983 and 1995. While northern quoll populations in the Northern

Territory exhibited declines prior to toad arrival (Ibbett et al. 2018; Woinarski et al. 2001; Ziembicki et al. 2013), rapid declines, often to extirpation, followed the invasion front (Woinarski et al. 2010a;

Woinarski et al. 2011b). In the Kimberley, Indigo (2020) found northern quoll populations declined by

86–96% following toad arrival, despite these populations being repeatedly exposed to TBZ-laced cane toad sausages―a technique shown to elicit toad-aversion in captive quolls (see management section).

Despite these declines, there is evidence that northern quolls can co-exist successfully with cane toads, and most of this evidence comes from Queensland, where toads and quolls have co-existed for over 80 years While it has been confirmed that Queensland northern quolls (and likely all other populations of northern quolls) are not physiologically resistant to cane toad toxins (Ujvari et al. 2013), natural variation in quoll behavioural responses to toads may have made some quolls less vulnerable to toad related mortality than others (see management section).

Predation Feral cats and dingoes are considered the most threatening predators to northern quolls across the majority of their range (Hill and Ward 2010), and this was reflected in the number of northern quoll publications that included reference to feral cat or dingo/dog predation (n = 15 and n = 21, respectively).

Feral cats were introduced onto the Australian mainland with the arrival of the ‘first fleet’ of British colonists in 1788 (Abbott 2008), and as such northern quolls have had limited evolutionary exposure in order to adapt to feral cats as predators (< 240 years). By contrast, northern quolls have co-existed with dingoes across their entire historical range for at least 3081 to 3348 years (Balme et al. 2018). This suggests the impacts of dingoes, and potentially their close relatives, domestic dogs, are unlikely to threaten the persistence of northern quolls on their own, but instead are amplified when acting in conjunction with other threats such as changing fire regimes and predation from feral cats (Geary et al.

2019b). Another introduced predator, the red fox (Vulpes vulpes), is also likely to predate on northern

146 quolls in some areas of their southern distribution (Cramer et al. 2016), however foxes are not present in the majority of the northern quolls range (Saunders et al. 2010).

In the Mackay Bowen region of Queensland, Pollock (1999) recorded eleven occurrences of quoll mortality as a result of domestic dogs and one record of predation by a black-headed python (Aspidites melanocephalus). This author also recorded predation by domestic or feral cats; but the number of quolls killed was not recorded. Cat predation on northern quolls in Queensland has also been recorded in the Rockhampton and Cape Upstart regions (Burnett pers. comm., 2020). In Kakadu National Park,

Northern Territory, Oakwood (2000) tracked 9 of 15 radio-tracked quolls to their death as a result of predation (dingo = 4, feral cat = 2, owl = 1, king brown snake, Pseudechis australis = 1, olive python,

Liasis olivaceus = 1). Similarly, Jolly et al. (2018a) found at least 7 of 19 quolls were killed by dingoes.

However, this number is likely to underrepresent the threat of dingo predation to quolls in this instance.

Most quolls that were tracked to their death (n = 10) died rapidly after release (< 4 days) due to toad consumption, reducing the opportunity for dingo predation to occur. Cremona et al. (2017a) found 3 of

4 quolls tracked to their death likely died of dingo/dog predation. In the Pilbara, Cowan et al. (2020a) found 6 of 41 collared northern quolls died as result of feral cat predation and two from dingo predation.

In a behavioural study, quolls from mainland Queensland were shown to recognise and avoid the scent of cats and dingoes, as did their captive born young, suggesting a genetic basis for predator recognition, including recognition of the introduced feral cat (Jolly et al. 2018b). However, quolls that had been translocated to Astell Island to conserve them against the impacts of cane toads appeared to have lost the ability to recognise dingoes and cats, after only 13 generations (Jolly et al. 2018b). The capacity of northern quolls to detect their predators could explain why Hernandez-Santin et al. (2016) found northern quolls avoided areas used by feral cats. The loss of antipredator traits observed on Astell Island may explain the role of dingo predation in the rapid extirpation of quolls during a reintroduction attempt

(Jolly et al. 2018a). Unfortunately, any attempts to train quolls to recognise dingoes as predators in captivity failed to impart predator aversion on quolls prior to reintroduction (Jolly et al. 2020).

Fire

147

A large number of publications (n = 24) considered the impacts of fire on northern quolls. We found no publications that recorded evidence of northern quoll mortality as a direct result of fire, however, we did find evidence from several populations that fire can have negative impacts on northern quoll populations. In the Northern Territory, Begg (1981) found that fire delayed northern quoll breeding and reduced the mean number of young that left the pouch, Oakwood (1997) found 55% of female northern quolls perished soon after fire in Kakadu, and Kerle and Burgman (1984) found that although northern quolls were common just after fire (< 1 year), they declined in the following year. In addition, Griffiths et al. (2015) predicted fire frequencies of over one fire per year would cause substantial decline in northern quoll populations, and Griffiths and Brook (2015) found modelled recruitment was 20% lower after a late season fire. In the Kimberley, Ondei et al. (2020) found northern quolls were only detected at rainforest sites, which burn less frequently than adjacent savanna sites, where no northern quolls were detected. Finally, in the Pilbara; Hernandez-Santin et al. (2016) found northern quolls were negatively associated with habitat that had been recently burnt.

There are multiple mechanisms that could explain observed negative impacts of fire on northern quolls.

It is possible that food may be scarce immediately following fire, which reduces overall habitat suitability. For example, in the Kimberley, Radford and Andersen (2012) found the total number of invertebrates―a key prey item for northern quoll―declined by 80–90% in the week following fire.

However, this same study found invertebrates were rapidly restored following the first wet season after fire, and there was no significant difference in invertebrate numbers between pre- and post-fire sessions within a year (Radford and Andersen 2012). Another probable mechanism is that fire increases predation risk, as a result of reduced ground-layer vegetation cover (Kerle and Burgman 1984;

Oakwood 1997). Interactions between fire and predation have previously been implicated in the decline of numerous Australian mammals (Hradsky et al. 2017; Leahy et al. 2016; McGregor et al. 2017;

Woinarski et al. 2010b), particularly for species including northern quolls that fall within a critical weight range (CWR) of species most vulnerable to predation from feral cats, foxes and dingoes

(Woinarski 2015).

148

It is important to note that some publications have not found fire to have a negative impact on northern quolls. Woinarski et al. (2004) found northern quolls near Darwin in the Northern Territory were more common at sites burnt annually, compared to sites that were long unburnt. In the Kimberley, Cook

(2010a) found fire had little impact on northern quoll home range size, even when an animal’s entire home range was burnt. These contrasting results highlight that the impact of fire on northern quolls is likely to be context specific, and might vary across their geographic range (Nimmo et al. 2012).

Grazing Overgrazing by feral and managed livestock has long been implicated in the decline of Australia’s mammal fauna (Woinarski et al. 2011b). Here, we identified 11 publications which discuss the impacts of over-grazing on northern quolls. The primary mechanism by which over-grazing is thought to impact northern quolls is through the removal of ground-level vegetation, which likely exposes quolls to increased levels of predation (Oakwood 1997). Additionally, factors such as water-hole contamination and soil-erosion caused by stock, as well as extensive tree-removal are likely to impact northern quolls.

Braithwaite and Griffiths (1994) suggest the combined impacts of grazing by feral ungulates are highly likely to have contributed to the decline of northern quolls across their range and similar assessments are also made elsewhere (Oakwood 1997; Radford et al. 2014; Woinarski et al. 2011b). However, it is important to note that quolls do appear to persist at some sites where heavy grazing occurs (Hill and

Ward 2010, Moore pers comms.; Woinarski et al. 2008). In these locations, it is possible that quolls may benefit from reduced fire frequency and intensities as a result of cattle minimising fuel loads (Hill and Ward 2010), and reduced predation as a result of dingo baiting.

Habitat clearing We identified six publications that discussed the impact of habitat clearing on northern quoll populations. Habitat clearing has been implicated as a factor in northern quoll declines prior to the year

2000, particularly in Queensland and the Northern Territory (Braithwaite and Griffiths 1994; Hill and

Ward 2010; Jones et al. 2014). However, despite the subsequent introduction of environmental legislation such as the Environment Protection and Biodiversity Conservation Act 1999 (EPBC Act) designed to protect species from such direct human impacts, habitat clearing is likely continuing to

149 threaten northern quoll populations. For example, in the period 2000–2017, an estimated 1.6 million hectares of potential northern quoll habitat was legally cleared (Ward et al. 2019). Northern quolls may be particularly sensitive to direct habitat loss because they require large areas (at least 220 km2) to maintain populations (Brook et al. 2011), they show strong negative responses to habitat fragmentation

(Rankmore 2006), and because hollows and logs used for shelter are removed during land clearing and take many years to form (Woinarski and Westaway 2008).

Mining There were few publications (n = 6) that investigated the response of northern quolls to disturbance as a result of mining/resource development. This was surprising, especially for the Pilbara population, where the scale of overlap between mining activity and northern quoll habitat is likely to be greatest―91% of the Pilbara bioregion is occupied by mining tenement (Environmental Protection

Authority 2014). Here, rocky ridges and mesas, which are thought to be important habitat for northern quolls, are frequently destroyed by mining companies targeting deposits of iron-ore and gold (Cramer et al. 2016). Surrounding granite outcrops are also quarried for rail formation ballast, rock armour for port infrastructure and basic raw materials for road construction. We found two publications from the

Pilbara which examined the impact of mining-related habitat clearing on quolls, including a government report that found northern quolls persist at two rocky sites located in close proximity to a recently installed rail line (Dunlop et al. 2015), and a Masters thesis with similar findings at the same sites

(Henderson 2015).

In addition to destroying habitat, mining activity can also impact species by introducing contaminants into the environment which can bioaccumulate, reducing the health of animals within affected areas

(Nawab et al. 2015). For example, Amir Abdul Nasir et al. (2018b) found airborne manganese dust from Groote Eylandt Mining Company (a BHP Billiton subsidiary) was absorbed by northern quolls living in close proximity to mining operations. Groote Eylandt quolls accumulated manganese within their hair, testes and brain. Quolls with higher manganese body burdens were slower at manoeuvring around corners than manganese free quolls, which may reduce their capacity to capture prey and escape predators (Amir Abdul Nasir et al. 2018a).

150

Other threats Other potential threats or sources of mortality for northern quolls are vehicle strikes (Oakwood 1997) and toxoplasmosis. However, Oakwood and Pritchard (1999) found no evidence of toxoplasmosis from

28 road-killed quolls in Kakadu National Park. Although it is clear from this review that northern quolls are exposed to multiple threats across their range, our understanding as to whether such threats to quolls are coincidentally or causally linked is poorly understood, and this can limit our capacity to manage threats effectively (Doherty et al. 2015b). For example, while targeted baiting for larger predators such as canids or foxes may result in short-term alleviation of pressure from one predator on the northern quoll (Jolly et al. 2018a), in the long-term it could potentially increase predation by mesopredators, such as feral cats (Marlow et al. 2015). Additionally, total isolation from predators (i.e. in havens) can cause rapid evolutionary loss of antipredator traits that may not be easily reinstated (Jolly and Phillips

2020; Jolly et al. 2018b). Similarly, while it has been suggested that fire may increase the susceptibility of northern quolls to predation (Oakwood 1997), our understanding as to how the timing, scale or intensity of fire influences predator-prey relationships remains limited for most ecosystems and species

(Geary et al. 2019a). Recognizing and understanding how these threats interact may facilitate more targeted management interventions potentially leading to more desirable conservation outcomes (see

Geary et al. 2019b).

Indigenous knowledge

Incorporation of Indigenous people and their knowledge into management actions is listed in three of eight recovery plan objectives for northern quolls (Hill and Ward 2010). Further, a recent study found

63% of the northern quolls current range, as defined by the IUCN, occurs within Indigenous peoples’ lands (O'Bryan et al. 2019). We identified eight publications that incorporated Indigenous knowledge as part of this review, two of which included Indigenous rangers or ranger groups as co-authors (Jolly et al. 2021; Kelly et al. 2020).

The name ‘quoll’, is derived from an Indigenous name for northern quolls, ‘Je-Quoll’, recorded by

Joseph Banks near Cooktown in North Queensland, 1770 (Beaglehole 1963). Other names were used

151 by Indigenous peoples to describe northern quolls across Northern Australia, and many of these are recorded by (Abbott 2013). Using local Indigenous vernacular names may help acknowledge the strong connections Aboriginal people share with Australia’s fauna and flora that were honed over thousands of years of coexistence.

The most comprehensive summary of Indigenous knowledge about northern quolls is provided by

Oakwood (1997). Like other quoll species (Attenbrow and Attenbrow 1987), northern quolls were consumed as food by Aboriginal people in Australia. However, there are mixed reports as to their palatability (Oakwood 1997). For example, northern quolls were considered "good tucker" by the

Gunwinggu people of West Arnhem Land (Goodfellow 1993). By contrast, barkuma (northern quolls) were not enjoyed as much by people of East Arnhem Land due to the buggan tumero (big smell) of the flesh (Dixon and Huxley 1985). Indigenous people of East Arnhem Land mentioned northern quolls as being numerous in diltji (open forest with grass) and also along beaches where cover is available, where they are said to shelter in hollow fallen logs and amongst stone.

In the Sir Edward Pellew Group of Islands of the Northern Territory, karnbulanyi (male northern quoll) or a-kaliba (female northern quoll) were also used as a food source (Bradley et al. 2006). Yanyuwa elders from the Pellew islands said northern quolls were once common, but younger people are less familiar with the species, and some of these island populations of quolls have been extirpated

(Woinarski et al. 2011a). Interestingly, the terms karnbulanyi and a-kaliba were later also used to describe feral cats (Bradley et al. 2006). Ziembicki et al. (2013) used Indigenous knowledge collated through a series of interviews across multiple communities in the Northern Territory to assess the extent and timing of regional mammal declines. This collation of information across communities found that range contractions were particularly pronounced for northern quolls, with the majority of decline recognised to have occurred in the 20 years prior to interviews taking place (1985–2009), and in the south of their range. Both cane toads and feral cats were implicated by observers as contributing factors in the declines, along with changing fire regimes associated with the cessation of Indigenous land management practises (Ziembicki et al. 2013).

152

There is significant Indigenous knowledge of the northern quoll across Australia (Abbott 2013, authors’ pers. obs.). While some traditional knowledge has been included in a handful of publications, there is a great potential for future integration into research and management of the remaining northern quoll populations. Recently, some research groups have worked very closely with Indigenous rangers and

Traditional Owners to improve the conservation of northern quolls in northern Australia. Despite our review surfacing limited publications including Indigenous knowledge, there are other ways in which

Indigenous people and their knowledge are actively involved in researching and conserving northern quoll populations, and it is important that this is acknowledged. One way is through Indigenous ranger programs, such as the Dhimurru, Jawoyn, Kenbi, Warddeken, and Marthakal ranger programs in the

Northern Territory, the Uunguu ranger program in the Kimberley and Martu ― Kanyirninpa Jukurrpa and Budadee programs in the Pilbara.. As part of these programs, Indigenous people play an important role in surveying and monitoring northern quolls, typically on lands with which they share a strong cultural connection, lands which are part of the Indigenous land estate (Jacobsen et al. 2020).

Management

In 1993, despite recognition of a potential national decline in geographic range of 10–50%, the conservation status of northern quolls was recognised as ‘apparently stable’ in an action plan for the conservation of Australian marsupials and monotremes (Kennedy 1992). Four years later, a similar action plan listed northern quolls as ‘lower risk (near threatened)’ (Maxwell et al. 1996). In 2005, the northern quoll was listed as Endangered under the EPBC Act, which enacts legal responsibility for consideration in environmental impact assessments, and management actions, such as the development of conservation advice and a recovery plan. Northern quolls were later listed as Critically Endangered in the Northern Territory, and Endangered in Western Australia, but have remained listed as ‘Least

Concern’ in Queensland. In 2010, a strategic set of management priorities for the northern quoll was outlined in the ‘National Recovery Plan for the Northern Quoll’ (Hill and Ward 2010), which included eight specific objectives: (1) prevent cane toads from reaching offshore islands where northern quolls are present; (2) foster the recovery of northern quolls in populations where they coexist with cane toads;

(3) halt northern quoll declines in areas where cane toads are present; (4) or absent; (5) maintain secure

153 populations for future translocations; (6) increase knowledge of disease; (7) reduce the impacts of feral predators on northern quolls; and (8) raise public awareness of the plight of the northern quoll.

We have identified four primary management actions which have so far been applied and documented in detail that aim to achieve these objectives. These are: 1) the use of islands as reserve populations through translocations, 2) cane toad control and aversion techniques, 3) feral predator control, and 4) the creation of artificial habitat.

Islands Islands can be important tools for species conservation by harbouring species at risk from threats present on the mainland (Ringma et al. 2018). The translocation of northern quolls to three islands (Astell

Island, Pobassoo Island, and Indian Island) off the coast of Northern Territory between 2003 and 2017

(Jolly and Phillips 2020; Kelly et al. 2020; Rankmore et al. 2008) has produced several positive outcomes. On Astell and Pobassoo islands, northern quoll populations increased from a total of 64 to

5600 individuals in the five years that followed translocations (Rankmore et al. 2008) and populations appeared to maintain genetic diversity at least in the short-term (Cardoso et al. 2009). Whilst both translocated populations exhibited some decline following their initial booms, both have now stabilised with high survival and recruitment rates compared to mainland populations (Griffiths et al. 2017).

In contrast to the Astell and Pobassoo island translocations, the Indian Island population declined sharply within a year of individuals being released and is now unlikely to be viable (Kelly et al. 2020).In addition to cane toad related mortality, the failure of the Indian Island translocation was likely contributed to by the extremely unfortunate timing of two major stochastic events (fire and cyclone) in the establishment year (Kelly et al. 2020). It is plausible, had the timing of these events been different, the introduced quoll population on Indian Island may have taken a different trajectory (Kelly et al.

2020). The primary purpose of the Indian Island translocation, rather than establishing an insurance population, was to experimentally measure a selection of toad-smart genes and, thus, test the effectiveness of targeted gene flow (see below) (Kelly et al. 2020). This also probably influenced the fate of the population, as the experiment required quolls to be released onto an island with a resident

154 toad population and a release cohort with appropriate population demographics (i.e., proportion of toad- smart individuals) meaning release numbers were small (n = 54).

Where quolls have been introduced to island arks for the purposes of setting up insurance populations, the success rate has been 100% (Griffiths et al. 2017). However, it’s important to consider that an objective of northern quoll island introductions is to create insurance populations which can be used for future mainland reintroductions (Hill and Ward 2010). As such, it will be important for those planning future island translocations to consider the impacts of isolation on quolls in terms of both genetic diversity (Cardoso et al. 2009) and predator naivety (Jolly and Phillips 2020) in order to maximise the success of future mainland reintroductions. Translocation planners should also consider the impact northern quolls may have on prey species occupying islands that quolls are translocated to, which, in the short-term, are unlikely to be equipped with appropriate behaviours to avoid being predated on by quolls (Jolly et al. 2021). Given the right circumstances, translocations to mainland locations within the northern quoll’s historic range could also be considered. This could include properties managed in terms of both predators and fire by conservations organisation such as the Australian Wildlife Conservancy and Bush Heritage Australia, or through the Australian Government’s Indigenous Protected Areas program

Cane toad control and aversion techniques Although established toad populations are largely impossible to eradicate with existing tools (Tingley et al. 2017), several publications have assessed the feasibility of controlling the spread of cane toads by capitalizing on their vulnerability to desiccation and blocking access to large artificial water bodies

(Brook et al. 2011; Gregg et al. 2019; Southwell et al. 2017; Tingley et al. 2013). Southwell et al.

(2017) suggest barriers blocking toad access to water between the Kimberley and the Pilbara could be constructed for $4.5 million AUD, with such a mechanism potentially capable of stopping toad invasion from the Kimberley to the Pilbara region. Initial field trials have been successful, with toads surviving a maximum of 5 days without access to surface water under the conditions where barriers would be installed (Gregg et al. 2019). However, it is important to note that the WA governments and pastoral industries desire to drought-proof the north-west pastoral industry through the increased development

155 of alternative agriculture practises (new crops and fodder) and intensification in pastoral diversification such a central pivot irrigation (e.g. La Grange region) will limit the effectiveness of strategies based on rendering water sources inaccessible to toads. Similarly, this method of restricting toad movement will not address breaches in biosecurity procedures which enable toads to reach the Pilbara with plant nursey stock and fresh food produce from the Kimberley and Northern Territory growing regions and as hitchhikers with tourists and industrial/mining equipment.

In populations where cane toads are already established, northern quolls are now ‘toad-smart’ and are less willing to depredate toads than quolls that have no previous exposure to toads (Kelly and Phillips

2017). This behaviour has been shown to be heritable, with offspring of toad-smart quolls being shown to innately avoid cane toads on their first encounter, suggesting rapid adaptive response in a small number of toad-impacted populations (Kelly and Phillips 2017; Kelly and Phillips 2019). To induce similar toad-smart behaviour in toad-naïve quolls, conditioned taste aversion (CTA) trials have recently been used to alter northern quoll predatory behaviour in captivity (Cremona et al. 2017b; Indigo et al.

2018; Jolly et al. 2018a; Kelly et al. 2018; O’Donnell et al. 2010; Webb et al. 2015) and have shown promise. CTA techniques used to train quolls to avoid toads typically use cane toad flesh laced with a nausea inducing dose of thiabendazole that deters quolls from subsequently eating cane toad flesh once released. Quolls trained with the toad sausages generally avoided the consumption of live and dead toads when tested in captivity (Indigo et al. 2018) and survived longer in the wild then toad-naïve quolls

(Jolly et al. 2018a; O’Donnell et al. 2010).

Despite the success of CTA trials on lab-trained quolls, buffering wild northern quolls against cane toads using CTA techniques has proved more difficult. For example, recent unpublished research found

CTA trials conducted within and adjacent to Mornington Wildlife Sanctuary in the Kimberley did not reduce toad impacts on the quoll population (Indigo 2020). Several contributing factors are thought to be potentially responsible for the failure of these trials, including: i) the decay of toad aversion with time since CTA exposure (Indigo et al. 2018); ii) ineffective delivery rates of the toad sausages; and iii) ineffective dose rates of thiabendazole within sausages (Indigo 2020). Although it has been demonstrated that intergenerational persistence of CTA trained quolls can occur in the wild following

156 translocation (Cremona et al. 2017b), the mechanism by which toad-avoidance behaviour is transmitted across generations (genetic or cultural) remains unclear. Decerning the transmission mechanisms is made considerably more difficult by the fact a certain proportion of quolls, irrespective of training, have a natural tendency to avoid attacking cane toads (Kelly and Phillips 2017). The efficacy of CTA as a management strategy is largely dependent on a single factor: whether quolls can confer the learnt toad- aversion lesson between generations via trained mothers teaching their young (cultural transmission).

Unfortunately, there is currently no evidence that quolls have the ability or tendency to train their young to avoid cane toads (Indigo et al. In Press). Recently, population viability models demonstrated that without a cultural transmission rate of >70%, for which there is no evidence, CTA is unable to prevent local extinction (Indigo et al. In Press).

Another cane toad mitigation strategy that could be implemented as part of future northern quoll management efforts is targeted gene flow, where quolls with heritable toad-smart genes are introduced into naïve populations to enhance their adaptive capacity (Kelly and Phillips 2016; Kelly and Phillips

2019). In 2017, the first and only trial of targeted gene flow in northern quolls involved releasing 54

CTA trained northern quolls onto toad infested Indian Island. The released quolls were composed of toad-smart genotypes from Queensland, hybrid toad-smart and toad-naïve genotypes (Qld ×NT) and toad-naïve genotypes from Northern Territory. The aim of the trial was to test if selection pressure in the form of cane toads would drive toad smart genes to spread throughout the introduced quoll population with each generation. Although northern quolls failed to establish a population on Indian

Island (details discussed above), genetic data collected the year after the translocation indicated selection toward toad-smarts had occurred after only a single generation (Kelly 2018; Kelly et al. 2020).

The study also demonstrated the successful hybridisation of Qld and NT northern quolls, with viable

F2 hybrids and backcrosses observed, suggesting outbreeding depression (a potential barrier to the success of targeted gene flow) is not an issue for this species (Kelly et al. 2020).

Feral predator control

157

Dingo and wild dog control has occurred across much of mainland Australia for well over a century

(Allen and Sparkes 2001). Dingo control is particularly common in pastoral areas and is mostly conducted via the aerial deployment of meat baits containing 1080 (sodium monofluoroacetate) (Twigg et al. 2000). Sodium monofluoroacetate is a poisonous compound produced naturally by plants in the genus Gastrolobium, which mostly occurs in the southwest region of Western Australia. While it is typically most lethal to animals without an evolved tolerance, this depends on both the dose rate and number of baits consumed (McIlroy 1981; McIlroy 1982). At sites where 1080 baits are deployed, dingo densities are often reduced (Thomson 1986; Twigg et al. 2000). Although northern quolls will eat 1080 baits (Calver et al. 1989), the baits themselves do not appear to negatively affect quoll population sizes

(King 1989). However, due to differences in their evolutionary history with plants containing sodium monofluoroacetate, eastern populations of northern quolls are likely to be more susceptible to 1080 than western populations (Twigg et al. 2003). Whilst dingo control may reduce dingo-related mortality in quolls, by potentially leading to mesopredator release (Crooks and Soulé 1999; Ritchie and Johnson

2009) of feral cats, it may also have complex and unintended indirect effects (Brook et al. 2012;

Dickman et al. 2009). However, there is conflicting evidence regarding whether feral cats are indeed released following dingo control in northern Australia (Brook et al. 2012; Kennedy et al. 2012; Leo et al. 2019; Stobo‐Wilson et al. 2020)

Similar to dingoes, cats can also be controlled using poison baits, albeit to date much less effectively

(Algar et al. 2007). New baits engineered specifically for cats such as EradicatTM , CuriosityTM and

HisstoryTM are now being trialled (Johnston et al. 2020; Johnston et al. 2013; Woinarski et al. 2019c) and have so far shown no signs of obvious impacts on northern quolls (Ranges 2017). Further, a five- year trial is currently underway investigating the impact of a large-scale EradicatTM baiting program on northern quolls, as well as other species vulnerable to cat predation such as Rothschild’s rock wallabies

(Petrogale rothschildi) and Pilbara olive pythons (Liasis olivaceus barroni) (Morris et al. 2015).

Preliminary results suggest feral cat baiting does have a positive effect on northern quoll populations

(Palmer 2019) without any direct impacts to the quolls themselves (Cowan et al. 2020a; Moro et al.

158

2019). However, confounding factors related to rainfall and fire justify the need for more sophisticated analyses to get an accurate measure of the impact cat baiting on quolls.

Artificial habitat Northern quolls, among many other CWR species in Australia, require structural complexity in the form of logs, tree hollows, termite mounds, or rocky outcrops to use as refuges (Oakwood 1997a), and reductions in habitat complexity as a result of fire or habitat clearing can expose these species to increased rates of predation (Oakwood 1997a, Palmer pers comm). One way of mitigating these effects may be through the use of artificial refuges (Cramer et al. 2016), which provide quolls with shelter from predation and climatic exposure where natural refuges have been removed. So far, artificial refuges designed for northern quolls have been deployed in several locations across the Pilbara bioregion

(Cowan et al. 2020b; Cramer et al. 2016) and have achieved mixed success. For example, Cowan et al.

(2020b) found that although artificial refuges closely replicated the thermal conditions created by natural dens, the surrounding environment was typically less complex, potentially contributing to greater feral cat visitation and lower prey availability. While it is possible that improving restoration efforts in the area surrounding refuges may increase their suitability for northern quolls, this has yet to be tested.

Future research directions

Northern quolls have been the subject of considerable research, much of which has improved our understanding of their threats and provided a useful basis for conservation management. However, the species remains threatened and continues to decline, with more resolute and strategic management required. Further research that addresses key knowledge gaps can contribute significantly to improving the effectiveness of conservation management for the northern quoll, and hence its overall conservation outlook. Here, we provide a non-exhaustive list of future research directions based on knowledge gaps evident from our review. If applied, each could be used to fine-tune and redirect management actions to improve conservation outcomes for northern quolls.

159

We acknowledge that a separate but overlapping set of research priorities have been identified for the

Pilbara population (Cramer et al. 2016), including: (i) develop appropriate and standardised survey and monitoring methods (ii) improve our understanding of habitat requirements, (iii) better understand the population dynamics of the northern quoll in the Pilbara, (iv) better understand key threats (cane toads, feral predators, mining infrastructure) and the interactions of these threats, and (v) determine the ability of the northern quoll to recolonise disturbed areas or colonise artificial habitat. We reiterate the importance of these research priorities to the conservation of Pilbara northern quolls.

1. Resolving taxonomy While no subspecies of northern quoll are currently recognised, several publications have found clear genetic distinctions among the four major populations based on microsatellite data, which are separated from one another by established biogeographic boundaries. Determining if major populations should be treated as distinct taxonomic units is likely to be critical in informing future management interventions such as targeted gene flow and genetic rescue, where quolls from one population are translocated to another (see management section). The use of more recently available genetic techniques, such as genome-wide single nucleotide polymorphisms (SNPs) analysis, is likely to be important in addressing whether these genetic divergences warrant taxonomic recognition and/or whether significant evolutionary units should be assigned to populations and managed differently

2. The status of Queensland northern quolls

The Queensland population of northern quoll previously occupied a larger area than any other northern quoll population, and has since undergone larger decline than any other northern quoll population

(Moore et al. 2019), yet we found a disproportionally small number of publications on Queensland northern quolls (although research in Queensland is ongoing). An explanation for this could be that northern quolls remain listed as ‘least concern’ under the Queensland Nature Conservation Act (1992), and therefore research funding provided by the state government for threatened species have so far not

160 been available. We suggest an increased research focus on northern quolls in Queensland is valid given the scale of declines that have occurred there, and further research may assist in elucidating if persisting northern quoll populations are stable, declining, or expanding.

3. Understand mechanisms allowing the persistence and resistance of northern quoll populations during cane toad invasions

In contrast to a trend of severe and rapid decline following the arrival of cane toads, some northern quoll populations do survive the initial wave of a toad invasion. However, apart from a small selection of broad-scale habitat predictors (Woinarski et al. 2008), we have little knowledge of population- specific characteristics that best predict a population’s short-term likelihood of surviving a toad invasion. Understanding the initial patterns of persistence may be useful in forecasting the probability of quoll population persistence in areas yet to be invaded by cane toads (southern Kimberley, Pilbara), potentially allowing us to prioritise the management of these areas. Such lessons are likely to both improve our understanding of the mechanisms and circumstances underlying a quoll populations’ persistence following cane toad invasion, but may also provide us with an increased mechanistic understanding of how these impacts and recoveries may play out in populations of other Australian predators that are threatened with extinction via the impacts of cane toads.

In addition to investigating factors facilitating quoll persistence through cane toad invasions, an important future research direction may be to assess if there are any signs of recovery. Have northern quolls returned to sites from which they were previously lost, and if so, what population/habitat characteristics have allowed for this return? Answering these questions is likely to provide information critical to the success of future assisted recolonizations―currently one of our most promising tools for conserving northern quolls (discussed above). Addressing these questions will first require the re- surveying of sites at which northern quolls have previously been confirmed to be absent following cane toad invasion (of which there are now many) to investigate whether recoveries have occurred. Secondly, the physical and genetic characteristics of the recolonising quolls, along with the make-up of the

161 recolonised habitat should be compared with sites where quolls have remained absent. Given northern quolls in Queensland have co-existed with cane toads for longer than any other northern quoll populations, it is likely here that post-cane toad recolonization events are most likely to occur. As such, we recommend future studies addressing this questions focus on the Queensland population.

4. Quantifying the impacts of mining

Mining activity occurs across the northern quolls entire range, including important cane toad-free populations, such as Groote Eylandt and the Pilbara bioregion. Yet, few publications (see threats section) have investigated the impact of mining on northern quoll populations. As such, determining the extent to which mining activity is likely to influence the persistence of northern quoll population should be addressed in future research. This is especially true for the Pilbara populations, where overlap between the northern quolls geographic range, and mining tenure is high (Cramer et al. 2016). In relation to mining activity specifically, Cramer et al. (2016) identify the impacts of linear infrastructure on northern quoll movement and the ability of the northern quoll to recolonise disturbed areas or artificial habitat as key research areas. In addition to these areas, we suggest future studies investigate secondary impacts of mining activity on northern quoll populations, such as increased predator densities surrounding mining camps, as a result of increased resource subsidies (e.g., food, water).

5. Population isolation and genetic rescue

A consequence of the northern quolls recent geographic decline is that many populations are now smaller and more isolated. Inbreeding and loss of genetic diversity is often unavoidable in small, isolated populations, increasing their extinction risk due to inbreeding depression (i.e. loss of fitness from low genetic diversity) and lowered adaptive potential (Frankham et al. 2017; Ralls et al. 2020).

Island populations of northern quolls are particularly exposed to these risks – and previous work has shown they have lower genetic diversity compared to populations on the mainland (Cardoso et al. 2009).

We recommend future studies expand on this work by including additional sites – both on islands as well as the mainland populations (Flanagan et al. 2018). Where isolated quoll populations are showing signs of genetic degradation, we would be wise to consider mixing populations via translocation, to

162 increase genetic diversity and adaptive potential of these degraded populations(Aitken and Whitlock

2013; Frankham 2015; Whiteley et al. 2015).

6. Unwinding interacting threats

Northern quolls face multiple co-occurring threats across their range; however, limited research has investigated if and how these threats interact synergistically. Publications of other native mammals occurring in northern Australia demonstrate that threats may be compounding and management that addresses only single threats may be ineffective (Legge et al. 2019). Better understanding northern quoll

‘threat webs’―a group of co-occurring threats that may have additive or non-additive impacts on each other―may improve land managers’ ability to focus efforts toward ultimate threats, rather than proximate threats, hopefully with improved conservation outcomes (Geary et al. 2019b). Understanding the synergetic impacts of predation, fire, and grazing, on northern quolls is likely to be particularly important to their conservation, given each of these threats occurs in tandem across the majority of their existing range. Untangling these interactions may also explain how the impact of dingo predation on northern quolls has been exacerbated post European arrival in Australia, as well as how removing dingoes is likely to influence the impact of feral cat predation on quolls. An important tool in untangling these threats is the use of manipulative experiments, where at least one threat within a system is artificially controlled (typically as part of conservation management activities), such that its relative impact on northern quolls populations can be measured in context to co-occurring threats. While experiments of this nature are already under way (eg. Palmer 2019), opportunity for further research in this area is likely to exist in areas where threat management (e.g., controlled burns, predator baiting) within northern quoll habitat is planned or actively occurring already.

7. Predicting the impact of climate change

Climate change poses an extreme risk to global biodiversity (Steffen 2009), yet we found little mention of the threat in literature related to northern quolls. Across the northern quolls geographic range, the

163 impacts of climate change are likely to include increased temperatures, rainfall variability, increased proportion of extreme rainfall events and less frequent cyclones (NESP 2018). While projected changes to total annual rainfall are uncertain, decreases in total rainfall are more likely than increases (NESP

2018). Measuring the extent to which changes in rainfall will impact resource availability and breeding success for northern quoll, among other northern Australian mammals, has obvious implications for the conservation of the species. In addition to changes in rainfall, average temperatures across Northern

Australia will continue to increase and there will be more days with extreme maximum temperatures

(NESP 2018). Understanding if these changes will lead to increased northern quoll mortality as a result of thermal stress, and the implications this would have for population persistence, should be a future research priority. More broadly, identifying potential climate refugia for northern quolls―areas that they will likely persist in under various climate change scenarios―where management efforts can be concentrated should also form a future research focus.

8. Further incorporation of Indigenous knowledge

Recognition of the knowledge held by the Indigenous people of Australia, often termed ‘two-way’ or

‘right-way’ science, has improve our ecological understanding of species on many occasions (Bohensky et al. 2013; Butler et al. 2012; Horstman and Wightman 2001; Telfer and Garde 2006). There is significant Indigenous knowledge of the northern quoll across northern Australia (Abbott 2013, authors personal communications). While some traditional knowledge has been included in a handful of publications (Dixon and Huxley 1985; Woinarski et al. 2011a; Ziembicki et al. 2013), there is a large potential for future integration into research and on-ground management of the remaining northern quoll populations, and potentially the detection of new isolated populations in remote areas of the Western

Desert.Given that such a large proportion of the known distribution of northern quolls is on

Traditionally Owned Indigenous land, future research and conservation endeavours should seek to be more inclusive of Indigenous stakeholders and aim to incorporate increased involvement of Indigenous people in such efforts.

164

9. Harnessing the heritability of toad avoidance behaviour

Individuals within quoll populations that survive the initial invasion of toads and continue to persist in sympatry with toads do so because they are innately unwilling to attack toads (Kelly and Phillips 2017;

Moore et al. 2019). Thus, if populations can avoid local extinction, natural selection should be rapidly acting upon heritable toad-averse traits. Harnessing the heritability of this behaviour forms the basis of targeted gene flow strategies (discussed above). Initial targeted gene flow trials using northern quolls have documented some encouraging results, however, additional research required to enhance this technique such that it can be applied on a broader-scale. This may be achieved by utilizing recent advancements in genetic technology to identify areas of the quolls genome that are most useful in predicting toad adverse behaviours in quolls. Future research in this area may also include attempting additional translocation trials, where captively bred quolls with genetic tendances to avoid toads are inserted into quoll populations predicted to be impacted by invading cane toads.

10. The role of artificial refuges

Initial trials have shown some potential for quolls to make use of artificial refuges in areas of disturbed or degraded habitat (Cowan et al. 2020b). However, this approach is not yet proven, and further research is required to determine whether artificial refuges can be a viable management tool for future northern quoll conservation. For example, evidence that northern quolls willingly use artificially constructed refuges as breeding habitat is currently lacking. Further trials, potentially incorporating differing complimentary actions (i.e., invasive predator control), are therefore required in order to address this knowledge gap. We also still have a limited understanding of the dimensions of artificially constructed den sites that maximise northern quoll useability. For example, existing artificial refuges range from 9–

165

150 m2 in size, but offer shallower crevices compared to those northern quolls use in natural habitat

(Cowan et al. 2020b). Future experiments should aim to test different sizes or arrangements of artificial refuges to determine how differences in these variables alter the thermal conditions of the refuges, and northern quoll use and survival. We also recommend that artificial refuges are trialled for northern quolls in habitats impacted by disturbances other than mining, such as fire and intense grazing, such has already been trialled for smaller dasyurids (Bleicher and Dickman 2020).

Acknowledgments

We thank Naomi Indigo for her assistance with this review. H.A.M. was supported by a scholarship from the Institute of Land, Water and Society operating funds from the Faculty of Science at Charles Sturt University. L.E.V. was funded by the Australian Government's National Environmental Science Program through the Threatened Species Recovery Hub. D.G.N. was supported by an Australian Research Council Early Career Researcher Award (DECRA).

166

Appendix B – Spot on: Using camera traps to individually monitor one of the world’s largest lizards

Harry A. Moorea*, Jacob L. Champneyb, Judy A. Dunlopc, Leonie E. Valentined, Dale G. Nimmoa

Manuscript published:

a School of Environmental Science, Institute for Land, Water and Society, Charles Sturt University, Albury, NSW, Australia b University of the Sunshine Coast, Sippy Downs, Qld, Australia c Department of Biodiversity, Conservation and Attractions, Locked Bag 104, Bentley Delivery Centre, Perth, WA, Australia d School of Biological Sciences, University of Western Australia, Crawley, WA, Australia

167

Abstract

Context Estimating animal abundance often relies on being able to identify individuals, but this can be challenging, especially when applied to large animals which are difficult to trap and handle. Camera traps have provided a non-invasive alternative by using natural markings to individually identify animals within image data. While camera traps have been used to individually identify mammals, they are yet to be widely applied to other taxa, such as reptiles. Aims We assessed the capacity of camera traps to provide images that allow for individual identification of the world’s fourth largest lizard species, the perentie (Varanus giganteus), and demonstrate other basic morphological and behavioural data that can be gleaned from camera trap images. Methods Vertically orientated cameras were deployed at 115 sites across a 10,000 km2 area in north-west Australia for an average of 216 days. We used spot patterning located on the dorsal surface of perenties to identify individuals from camera trap imagery, with the assistance of freely available spot ID software. We also measured snout-to-vent length (SVL) using image analysis software, and collected image time stamp data to analyse temporal activity patterns. Results Ninety-two individuals were identified, and individuals were recorded moving distances of up to 1975m. Confidence in identification accuracy was generally high (91%), and estimated SVL measurements varied by an average of 6.7% (min = 1.8%, max = 21.3%) of individual SVL averages. Larger perentie (SVL > 45cm) were detected mostly between dawn and noon and in the late afternoon and early evening, whereas small perentie (SVL < 30cm) were rarely recorded in the evening. Conclusions Camera traps can be used to individually identify large reptiles with unique markings, and can also provide data on movement, morphology, and temporal activity. Accounting for uneven substrates under cameras could improve the accuracy of morphological estimates. Given that camera traps struggle to detect small, nocturnal reptiles, further research is required to examine whether cameras miss smaller individuals in the late afternoon and evening. Implications Camera traps are increasingly being used to monitor reptile species. The ability to individually identify animals provides another tool for herpetological research worldwide.

168

Introduction

Estimating species’ abundance remains a key challenge within ecology and conservation biology (Volkov et al. 2003). One obstacle is the need to distinguish between individuals within a population in order to derive abundance estimates (Silver et al. 2004). For instance, mark-recapture analysis has traditionally required trapping and individually marking an animal, and then using the number of re- traps to estimate species density and abundance (Pradel 1996). However, physically capturing animals is not always practical (Sanecki and Green 2005) nor desirable (De Bondi et al. 2010), and therefore non-invasive methods that use unique morphological features to passively identify individuals are sometimes more appropriate (Brooks et al. 2010; Silver et al. 2004).

Camera traps provide a means of collecting image-based data that can be used to distinguish marked (scaring, natural markings) individuals from one another (Foster and Harmsen 2012; Higashide et al. 2012), and are increasingly used to estimate species density/abundance using mark-recapture analysis (Burton et al. 2015). So far, studies that have used camera trap imagery to identify individual animals and estimate density/abundance have been almost entirely limited to mammalian taxa (Burton et al. 2015), such as tigers (Panthera tigris)(Jhala et al. 2011), bobcats (Lynx rufus) (Alonso et al. 2015), leopards (Panthera pardus) (Rostro-García et al. 2018) and puma (Puma concolor)(Alexander and Gese 2018). Despite increasing recognition that camera traps provide a useful method with which to survey reptiles (Molyneux et al. 2018; Richardson et al. 2018; Welbourne et al. 2015; Welbourne et al. 2017), few studies have used camera trap imagery to identify individuals within a reptile species (Bennett and Clements 2014; Welbourne 2013), although other studies have identified individuals using manually operated cameras (Kellner et al. 2017; Moro and MacAulay 2014; Treilibs et al. 2016).

Species within the predatory lizard genus Varanus are notoriously difficult to trap and handle (Green and King 1978; Jessop et al. 2006), given they can weigh >90kg and measure three metres in length (Jessop et al. 2006), and yet the need to monitor these species has never been greater. Varanids (Varanus.spp) are widely distributed across Australia, Asia and Africa, and often fulfil the role of a mesopredator or apex predator (King and Green 1999). Multiple species of varanids are threatened from exploitation by humans (meat, leather) (Shine and Harlow 1998), the illegal pet trade (Ruxmoore and Groombridge 1990), habitat loss (Gibbons et al. 2000) and introduced species (Shine 2010). The removal of these important predators from ecosystems can have cascading impacts on a range of other species (Doody et al. 2006; Read and Scoleri 2015; Sutherland et al. 2010), highlighting the need for a more accurate understanding of their populations.

Here, we use camera trap imagery to identify individuals of the world’s fourth largest species of lizard, the Perentie (Varanus giganteus). Perentie measure in excess of 2 m long, weigh up 20 kg, and are

169 potentially venomous (Fry et al. 2006; Wilson and Swan 2017), however little is known about the ecology of this species. The use of non-capture methods to study perentie are therefore particularly desirable, especially given trapping these animals likely poses risks to both the animal and the handler. Further, their large size means that perentie tend to evade capture/detection using traditional reptile survey techniques such as pitfall trapping (Richardson et al. 2018). The scales of perentie are covered with lateral bands of large yellow spots that are unique to the individual (Moro and MacAulay 2014). In this study, we investigate the feasibility of individually identifying perentie at a landscape-scale using remote sensing cameras. We also use scaled camera trap images to estimate individual body lengths, and time stamp data to observe temporal trends — both techniques which have never been used on a varanid species and are very rarely applied to reptiles in general.

Materials and methods

Study site

This study was conducted across four cattle stations within the Pilbara bioregion in north-west Australia. This area also encompassed the Karayarra and Nyamal Indigenous language groups (Fig B1). Vegetation within this area is mostly dominated by Triodia grasslands (McKenzie et al. 2009). Geology in the Pilbara is characterised by largely flat sand plains and granite outcrops (Doughty et al. 2011; Withers 2000). Average annual rainfall within the study area is 339.5mm (Indee station)(Australian Bureau of Meteorology 2020).

Survey design

We used a whole-of-landscape experimental design, in which multiple sample sites are embedded within a heterogeneous area (Fahrig 2003), commonly termed a landscape (Bennett et al. 2006). Twenty three study landscapes were selected using ArcMap 10.3 (ESRI 2011). In our study, landscapes were circular with a diameter of 1 km (area =75 ha), and were selected to contain patches of rocky habitat dispersed within a matrix of spinifex grasslands. We deployed five camera sites within each landscape (115 sites in total), each separated by > 200 meters. All study sites were placed within rocky outcrops, as this habitat is known to be utilized by perenties and other species (e.g., northern quolls, Dasyurus hallucatus) that were the focus of the broader research program within which this study was embedded (Menkhorst and Knight 2011; Wilson and Swan 2017).

170

Figure B1 – Twenty three study landscapes spread across four cattle stations in the Pilbara bioregion — Pippingarra, Indee, Yandeyarra and Mallina. Grey lines represent cattle station boundaries. Five downward facing camera sites were deployed at each study landscape.

We used Reconyx™ PC900 Hyperfire covert cameras (Reconyx, Wisconsin, USA). These cameras use an infrared flash. Cameras were attached to a wooden tree stake 1.5m above the ground and were positioned to face downward, with the camera lens focused directly at the ground surface using a bookshelf bracket (Appendix 1). A PVC canister containing approximately 150g of bait (canned fish) was attached to the bottom of the tree stake supporting the camera. Twelve landscapes were sampled between August 2017 and April 2018. The remaining 11 landscapes were sampled between August 2018 and April 2019. Average camera deployment time was 216 (min =151, max = 245) days at each site. Bait was replenished twice at all sites during this period after roughly 70 and 140 days of camera deployment. At yandeyarra station, camera placement was partly guided by knowledge from Indigenous rangers as part of the Greening Australia ranger program. Cameras were set to high sensitivity with no delays between triggers, and five images were taken at one second intervals per trigger. We defined an independent detection event as consecutive triggers of the same individual separated by greater than 15 minutes.

171

Individual identification

We used spot patterning located on the dorsal surface of perentie to identify individuals from camera trap imagery (Figure B2). To do this, we first catalogued images from each detection event that best represented an individual’s unique spot pattern. Only images with at least 30% of the animal’s dorsal surface patterning visible were catalogued for analysis. We found that attempting to identify animals in images with less than 30% of the animal’s dorsal surface viable was generally unreliable, as fewer spots markings are available to distinguish between animals.

We then entered one or more images from each catalogued detection event into I3S (Den Hartog and Reijns 2016) —a freely available spot ID program that has previously been used to identify individuals in other spotted species such as spotted eagle rays (Aetobatus narinari) (Cerutti-Pereyra et al. 2018), Italian crested newts (Triturus carnifex) (Sannolo et al. 2016) and whale sharks (Rhincodon typus) (Rowat et al. 2009). I3S automatically compares a target individual’s spot pattern to a library of other individual spot patterns that have been pre-entered into the system by an observer. To do this, the I3S uses a two-dimensional linear algorithm, which compares spot coordinates, as well as spot shape and size between patterns. I3S then ranks which patterns within the existing library are most similar to the target spot pattern in the form of a pattern difference score, where lower pattern difference scores correspond to increasing pattern similarity (Cerutti-Pereyra et al. 2018; Speed et al. 2007). Once patterns had been ranked, at least two observers manually confirmed if the pattern suggested by the I3S program matched the target spot pattern. If no match could be found, the target individual was assigned a reference code and entered into the data base as a new animal.

To gauge observer confidence in individual identification accuracy, we marked each detection event with a confidence rating using an adapted identification protocol outlined in (Hohnen et al. 2013). Confidence rating were derived from the number of unique markings/features that could used to identify an animal. An observer had ‘high’ confidence that an individual has been accurately identified if at least five unique markings/features could be used to identify an animal. An observer had ‘medium’ confidence that an individual has been accurately identified if between two and five markings/features could be used to identify an animal. An observer had ‘low’ confidence that an individual has been accurately identified if only one marking/feature could be used to identify an animal. For individuals that were recorded at more than one site, we measured distance travelled using the measure function in ArcMap 10.3 (ESRI 2011).

We used an analytical Bayesian approach with a vague gamma prior to summarise the number of detections per perentie, as well as the number of perenties per site and per landscape. This approach produces upper and lower 95% credible intervals, which are comparable to confidence intervals produced when using a frequentist approach. We used credible intervals because standards estimates of

172 mean error (ie standard deviation) in count data where means are close to zero can produce nonsensical outputs, such as negative counts.

Figure B2 – An example of the spot patterning used to individually identify perentie (V. giganteus) within the Pilbara bioregion. The three images shown are of the same animal, and were taken over a period of 57 days (23/10/2017 – 19/12/2017 ) at three different camera sites. Red circles represent specific spots that were used to identify this animals.

173

Length measurement

We measured the snout-to-estimated vent length (SVL) of perentie individuals in camera trap images using the computer program imageJ (Abramoff et al. 2004). Because the vent was not visible from the dorsal surface, we consistently estimated the vent location as being at the base of the hind legs in the centre of the animal’s body, based on morphometric measurement protocol used by Thompson and Withers (1997). SVL measurements followed the animal’s spine from the snout to the vent to increase measurement accuracy. ImageJ allows users to measure distances within imagery by calculating a pixel- distance (mm) ratio from an object with a known length (scale) within an image. Once a pixel-distance ratio has been determined, objects with an unknown length can be measured. Here, we used brackets located at the base of tree stakes supporting cameras as the scale (40 mm). To test the accuracy of using image-j to measure animal sizes from camera trap imagery, we conducted a pilot test using Reconyx PC900 Hyperfire covert cameras and target objects with known dimensions. We found imageJ measurements varied from actual measurements (using a measuring tape) by an average of < 3%. To visualize perentie SVL data, we created histograms of SVL using ggplot in R version 3.5.3 (R Core Team 2020; Wickham and Wickham 2007). One SVL measurement was taken for each independent detection. When an individual was detected multiple times, and thus multiple SVL measurements were taken, we used the mean measurement.

Temporal activity

To further illustrate the range of data that can be collected for lizards using camera traps, we used time stamp data from detection images to fit non-parametric kernel density curves using the ‘overlap’ package in R (Ridout and Linkie 2009), giving a probability density distribution of perentie activity patterns .This method is commonly used to quantify species temporal activity patterns within ecological studies (Allen et al. 2018; Azevedo et al. 2018; Fancourt et al. 2015b; Lashley et al. 2018). We used a default smoothing parameter of 0.8 in our analysis, as recommended by Ridout and Linkie (2009) for small sample sizes.

Perenties being ectotherms, are reliant on external sources of heat to maintain a constant body temperature, which can determine when they are most likely to be active throughout the day. Because larger lizards are generally able to retain body heat longer than smaller lizards (Stevenson 1985), we expected body size to impact perentie temporal activity. To account for these differences, we analysed perentie temporal activity using three sizes classes; large (SVL >45 cm), medium (SVL = 30 to 45cm) and small (SVL < 30 cm). We determined sizes classes based on the distribution of SVL measurements collected in this study. To elucidate if perenties of different size classes were detected more frequently at different times, we measured overlap in temporal activity between large, medium, and small individuals, using an overlap coefficient (∆). Overlap coefficients are the product of a function which

174 describes the distance between two temporal activity density distributions, and can range from zero (no overlap) to one (complete overlap) (Ridout and Linkie 2009). We used the package overlap’ (Meredith and Ridout 2014) in R version 3.5.3 (R Core Team 2020) to calculate temporal overlap between perentie size classes, which produces three overlap coefficients for each temporal comparison. The most appropriate coefficient for estimating overlap is dependent on sample size (Ridout and Linkie 2009).

We used the overlap coefficient ∆1 as it is most appropriate for small datasets (Ridout and Linkie 2009). Overlap precision was measured using confidence intervals generated from 500 random bootstrapped samples after accounting for bootstrap bias (Meredith and Ridout 2014).

Results

We recorded 190 detections of perentie over 28,840 trap nights. Other varanid species detected included Varanus panoptes, V. pilbarensis, V. gouldii and V. acanthurus. Perentie were detected at 91.3% of landscapes and 48.6% of sites. Seventy-three percent of detection events were suitable for individual identification, and from these, we identified a total of 92 individuals using spot pattern analysis. Confidence in identification accuracy was generally high (91%) or medium (7%). Thirty-one (33.7%) of individuals were recaptured at least once. The number of detections per individual ranged from one to five (x̄ = 1.55, lower credible interval = 1.34, upper credible interval = 1.82), and the number of individuals detected within a single landscape ranged from zero to twenty (x̄ = 3.99, lower = 3.22, upper = 4.86). The number of individuals per site ranged from one to 13 (x̄ = 0.80, lower = 0.64, upper = 0.97) (Figure B3), and thirteen individuals were detected at more than one site (Figure B4,B5). The average total distance travelled between sites where the same individual was detected was 887 metres (sd = 632 m), ranging from 224 m to 1975 m (Figure B5). Average estimated SVL length was 37.22 cm (sd = 12.85 cm), ranging from 10.29 cm to 73.30 cm (Figure B6). When an individual was detected multiple times, length measurements varied by an average of 2.46cm (min = 0.64, max = 7.86), or 6.7% (min = 1.8%, max = 21.3%) of individual SVL averages (Appendix 2). SVL increased by 5cm and 7cm for two individuals that were detected 123 and 124 days respectively after they were initially detected, potentially as a result of growth. We found limited evidence of growth otherwise. Twenty-nine percent of individuals measured were classified as small (<30cm), 44% were classified as medium (30–45cm) and 29% were classified as large (>45cm).

Temporal data showed on average perentie were most active between dawn and noon, and around dusk (Figure B7). Overlap in activity patterns was highest between medium and small individuals (∆ = 0.80, lower CI = 0.63, upper CI = 0.89) and lowest between large and small individuals (∆ = 0.67, lower CI = 0.43, upper CI = 0.81). Overlap between large and medium individuals was 0.80 (lower CI = 0.55,

175 upper CI = 0.85). Large individuals were most active between dawn and noon but also active at dusk and during the night, whereas small individuals were almost exclusively active between dawn and noon.

Discussion

Camera trap imagery is commonly used to identify individuals within mammal populations (Burton et al. 2015), but few have been used to individually identify lizards (Bennett and Clements 2014; Welbourne 2013). In this study, we used camera traps to identify 92 individual perenties across 23 landscapes within the Pilbara bioregion, Australia. We found a large proportion of independent detections were suitable for individual identification, and observer confidence in identification accuracy was high, suggesting camera traps may offer a practical alternative to live trapping when monitoring some large reptile species. We also highlight that other useful data, such as movement data, body length, and temporal activity patterns can be derived from the same imagery. This research highlights the significant potential of camera traps for population monitoring of large reptiles across the world.

The capacity to individually identify animals from camera trap data unlocks the use of a variety of statistical analyses that can be used to estimate population densities, as well as survival probabilities and population spatial dynamics. For example, mark-recapture models rely on individual identification to estimate abundance (Pradel 1996). We show that collecting the data necessary to perform such analyses on large lizards is possible through the use of remote sensing cameras. However, it is important to note that in addition to the ability to identify individuals, other data assumptions must be met in order to reliably estimate abundance or density using mark-recapture methods (Pradel 1996). For example, for most mark-recapture models, accuracy is highest when populations are closed (Kendall 1999); that is no deaths, births, immigration or emigration occurs across sampling periods. Whilst it is possible individuals left or entered the study population during sampling periods in our study, the likelihood of this occurring can be reduced by shortening the sampling period, or using mark-recaptures models that account for open populations (Schwarz and Arnason 1996). Low recapture rates as a result of poor capture efficiency can also impact the reliability of mark-recapture estimates (Morton 1982; Pollock 1980). In our study, we found a third of individuals were recaptured at least once, and some individuals were recaptured up to four times. This result suggests capture efficiency may be adequate for mark- recapture methods to be applied on data collected in this study, however it would be possible to increase capture efficiency by increasing the number of cameras deployed per landscape.

Perentie were detected across most study landscapes, and some individuals were detected across multiple landscapes, moving distances of almost 2 km. Whilst the spatial ecology of perentie remains poorly understood, research focused on other varanids suggests movements of this distance are not unusual (Green and King 1978; Guarino 2002). We also found a number of individuals were recorded visiting the same camera sites repeatedly (up to five times), sometimes over a period of months. Whilst

176 we may only speculate where these individuals travelled in the time between detections, habitual visits to the same site could indicate some individuals may occupy a home range like other varanids (Lei and Booth 2018).

One disadvantage associated with using non-capture related techniques to survey wildlife is the lack of ability to collect morphological data. In this study, we were able to approximate perentie SVL by scaling images to a known distance. SVL length varied substantially between individuals suggesting camera traps were able to detect a range of demographics ranging from juveniles (SVL = 10.3 cm) to larger adults (SVL = 73.3 cm). Average SVL length was 37.22 cm (min = 10.3cm, max = 73.3 cm), which fits within the bounds of previous measurements that were recorded in situ on perentie in Western Australia. For example Thompson and Withers (1997) found average SVL length from 25 captured perentie was 44.2 cm (sd = 12.4 cm), ranging from 15.9 cm – 66 cm. Similarly King et al. (1989) found SVL length of captured perentie on Barrow Island ranged from 23 cm – 88 cm (no average available). Whilst estimating body length using scaled imagery is an accepted method within wildlife research (Weinstein 2018), in our study, uneven ground surfaces on top of which lizards were measured has the potential to distort estimates. Despite this, we found measurement consistency was relatively high (average variation from mean individual SVL = 6.7%). This result suggests that, although SVL measurements using camera trap imagery are obviously less accurate than measurements obtained with animals in hand, they may be able to provide reasonable estimates of animal size that could potentially be used to infer other individual attributes such as age and sex (although see Smith et al. 2007). A practical means of improving length estimates using camera traps in future studies may be to install a flat base plate underneath cameras (following Welbourne 2013; Welbourne et al. 2015), so that variation in the surface on which animals are measured is removed.

Behavioural data can be important to species conservation (French et al. 2019; Tyne et al. 2017), yet accessing this data usually involves tracking an animal using expensive satellite telemetry equipment (Bastille‐Rousseau et al. 2018; Hertel et al. 2019), or investing large quantities of time manually recording animal activity in the field (Brieger et al. 2017; Li et al. 2017). We used time stamp data imprinted on camera trap imagery to record perentie diel activity. Temporal activity varied between perentie size classes, however overlap between perentie size class combinations was mostly similar (∆0.67–∆0.80). The most notable differences in temporal activity were observed in temporal activity peaks. For example, activity peaks for large and medium sized individuals were spread across much of the 24-hour diel period, occurring at mid-morning, dusk and midnight, whereas activity for small individuals was mostly confined to mid-morning and early afternoon. Observed temporal differences could be related to individual physiology, where individuals with larger body mass are able to thermoregulate more efficiently than individuals with smaller body mass, and thus stay active longer (Garrick 2008). Alternatively, smaller individuals may avoid being active around dusk and at night to reduce their likelihood of encountering potential predators such as feral cats (Felix catus) and dingoes

177

(Canis dingo) which are generally most active at these times within the study area (GéCzy 2009; Hernandez-Santin et al. 2016; Johnson 1976). Whilst some varanids are known to predate on smaller members of their own species (Polis and Myers 1985), our results do not suggest smaller perenties are temporally avoiding larger perenties. This is because peak activity of large and small perenties overlaps considerably around 11:00am. Further, smaller perenties are not active at any time larger perenties are not active.

Another explanation is that cameras fail to detect smaller lizards at night due to a lack of thermal contrast. This is consistent with the findings of Richardson et al. (2018), who found that camera traps failed to detect small, nocturnal reptiles. Passive infrared triggered cameras, as used in this experiment, will detect an animal when a difference between the animals surface temperature and the background temperature is detected, (Welbourne et al. 2016), with optimal conditions for detection occurring when the temperature difference is at least 2.7° C (Meek et al. 2012). When the background temperature is similar to an animals surface temperature, the capacity of PIR cameras to detect the animal may be reduced (Rovero et al. 2013). On the basis of these facts and the laws of thermodynamics, the surface temperature of smaller perentie individuals maybe more similar to the background temperature — derived from the red granite on top of which cameras were positioned — because they have a smaller body mass, and therefore maybe less likely to be detected by PIR sensors. Contrast between the background temperature and surface temperature of smaller perenties may be larger in the day because animals are better able to thermoregulate by utilizing microhabitats that are cooler than the rest of their environment (Sears et al. 2016). To increase thermal contrast when using PIR cameras to detect lizards Welbourne (2013) successfully augmented background temperature using cork tiles, making the lizards more visible to sensors. Whilst implementing a similar technique would have been desirable in this study, unfortunately it was logistically not possible.

Recent improvements in remote sensing technologies have created opportunities for wildlife research to utilize non-invasive labour efficient techniques instead of methods that involve the live trapping of animals. Importantly, we recognise that although many large species of lizards bare natural markings that could potentially be used to tell animals apart (Chen et al. 2013; King and Horner 1987; Rodda et al. 1988), consistent individual identification may not be feasible for all species, especially those which lack spot or line based scale patterning. One potential objective of future research could be to explore the practicality of using camera traps to individually identify animals from other lizard taxa that exhibit visible scale patterning, such as species with the Agamidae and Scincidae families (Welbourne 2013). Another objective could be to trial alternative camera types and settings that may be better suited to detecting smaller species, whilst still providing high resolution imagery. Preliminary research in this area suggests cameras programmed to take images at standard intervals capture more images of squamates than cameras triggered by sensors, and cameras with a manually adjusted focus collect images better suited to identifying smaller lizards than cameras which use default focus settings

178

(Welbourne et al. 2019). In this study, we demonstrate how camera traps can be used to detect Australia’s largest species of lizard and provide imagery that can be used to differentiate individuals as well as observe temporal behaviour. Overall the findings of our study provide a promising indication of the potential uses for camera traps in studying large lizards and herpetological research in general.

Figure B3 – Boxplots depicting the total number of perentie (Varanus giganteus) individuals detected at study sites grouped within the 23 study landscapes within the Pilbara bioregion in Western Australia. Five study sites were within each study landscape.

179

Figure B4 – Perentie (Varanus giganteus) individuals detected at multiple study sites within the Pilbara bioregion in Western Australia. a.) Individual 02_YA_04 was detected at site Q0704 on 30/09/2018 before moving 508 metres to site Q0701 on 30/10/2018 and then 431 metres to site Q0703 on 05/01/2019. b.) Individual 01_MA_13 was detected at site Q6603 on 25/09/2017 before moving 1573 metres to site Q3205 on the 8/11/2017. Individual 01_MA_09 was detected at site Q3202 on 15/01/2018 before moving 1975 metres to site Q6601 on 10/03/2018. c.) Individual 02_YA_04 was detected at site Q0704 on 30/09/2018 before moving 508 metres to site Q0701 on 30/10/2018 and then 431 metres to site Q0703 on 05/01/2019. d.) Individual 01_ID_09 was detected at site Q2201 on 23/10/2017 before being detected 201 metres away at site Q2202 on 24/10/2017 and then 1245 metres away at site Q1905 on 19/12/2017.

180

Figure B5 – Histogram depicting total distances travelled (m) by the 13 Perentie (V..giganteus) individuals that were detected at more than one study landscape within the Pilbara bioregion .

181

Figure B6 – Histogram depicting the averaged snout to estimated vent length (cm) of 73 Perentie (V. giganteus) individuals that were measured from camera trap image using freely available image analysis software.

182

Figure B7 – Non-parametric kernel density curves depicting temporal activity for Perentie (Varanus giganteus) in the Pilbara bioregion recorded using camera traps. Grey sections represent overlap in temporal activity. Large individuals were individuals with an average SVL exceeding 45cm. Medium individuals were individuals with an average SVL between 30 cm and 45 cm. Small individuals were individuals with an average SVL less than 30cm. A.) Averaged temporal activity for all individuals B.) Overlap in temporal activity for large and medium individuals C.) Overlap in temporal activity for medium and small individuals D.) Overlap in temporal activity for large and small individuals.

183

Figure BS1– Schematic of camera trap design used to detect and identify Perentie (V. giganteus) individuals.

184

Figure BS2 – Boxplot depicting variation in the snout to estimated vent length (cm) of 15 Perentie (V. giganteus) individuals that were measured from camera trap image using freely available image analysis software

185

Acknowledgements

Data collection was assisted by Sian Thorn, Darcy Watchorn, Rainer Chan, Daniel Bohorquez Fandino, Jacob Champney, Hannah Kilian, Mitch Cowan. Camera trap deployment was also completed with the assistance of the Yandeyarra Indigenous ranger program facilitated by Pip Short and Greening Australia. Technical support was provided by Neal Birch, Brent Johnson, Hannah Anderson, Russell Palmer, Alicia Whittington and Jo Kuiper from the Western Australian Department of Biodiversity, Conservation and Attractions (DBCA), as well as Deb Noy from Charles Sturt University. Stephen van Leeuwen from DBCA provided assistance with the project conception and also provided support throughout. Equipment and operational costs were provided by DBCA, Roy Hill and Charles Sturt University. Roy Hill also covered the costs of flights, fuel and freight. Harriet Davie from Roy Hill provided technical support throughout along with the Roy Hill rail team in Port Hedland. We thank Colin Brierly, Betty Brierly and Graham for providing access to Indee station, Ben and Lindsey for access to Mallina and Troy Eaton for access to Pippingarra. Belinda Barnett from BHP assisted with site access. H.A.M is supported by a scholarship from the Institute of Land, Water and Society and Charles Sturt University. L.E.V. was funded by the Australian Government's National Environmental Science Program through the Threatened Species Recovery Hub. D.G.N. was supported by an Australian Research Council Early Career Researcher Award (DECRA). This project was supported by the Holsworth Wildlife Research Endowment – Equity Trustees Charitable Foundation & the Ecological Society of Australia, as well as the Western Australian Department of Biodiversity, Conservation and Attractions as well as environmental offsets and public good funding provided by BHP, Rio Tinto, Atlas Iron, Fortescue Metals Group, Roy Hill, Process Minerals International, Metals X and Main Roads Western Australia. Research ethics was granted though the Charles Sturt University animal ethics committee (permit number A17031), and the Western Australian Department of Biodiversity, Conservation and Attractions (permit number 08-002376-1).

186

Appendix C – Supplementary material for chapter 2

Table S2.1 – Presence records used in range, niche and habitat quality analysis for each of the four study populations. Historic records are dated from 2000 and earlier. Contemporary records are dated after the year 2000.

Population Historic Contemporary Queensland 544 165 Northern Territory 539 161 Kimberley 139 88 Pilbara 803 739

Comparison of nested and un–nested analysis To test the effect nesting contemporary records within historic records had on assessing decline in northern quolls, we compared nested and un–nested datasets across several analyses for each of our four study populations. Analyses included; (i) the number of records allocated to historic and contemporary periods (ii) geographic range and (iii) niche volume. Methods used to measure geographic range and niche hypervolume are listed in the main methods sections.

Differences in analysis results between nested and un–nested datasets were smaller in Queensland, and the Northern Territory, slight larger in the Kimberley, and much larger in the Pilbara. Larger differences between nested and un–nested results in the Kimberley and Pilbara are most likely caused by differences in available presence records before and after the year 2000. In the Pilbara, a large difference in the number of historical and contemporary presence records is related to massive increase in survey effort as part of impact assessments and offsets related to a boom in iron ore exploration between 2002 and 2015 (ABS 2018)

187

Figure S2.1– Historic and contemporary records collected from each northern quoll population. The top row of plots represents nested data — records collected after the year 2000 included in historic data set. The bottom row represents separated data —records collected after 2000 separated from records collected before 2001.

188

Figure S2.2– Historic and contemporary α-hull range area calculated for each northern quoll population. The top row of plots represents α-hull area derived from nested data — records collected after the year 2000 included in historic data set. The bottom row α-hull area derived from separated data —records collected after 2000 separated from records collected before 2001.

189

Figure S2.3– Historic and contemporary niche volume calculated for each northern quoll population. The top row of plots represents volume estimates derived from nested data — records collected after the year 2000 included in historic data set. The bottom row represents volume estimates derived from separated data —records collected after 2000 separated from records collected before 2001.

190

Figure S2.4– Histogram of filtered records collected from 1900 onward. Records are grouped into five year bins.

191

Appendix D – Supplementary material for chapter 3

A.) B.)

Figure S3.1 – The position of detection bands for ReconyxTM camera, adapted from (Reconyx 2020). A.) Image from vertically orientated camera, B.) horizontally orientated camera.

Appendix E – Supplementary material for chapter 4

192

Table S3.1 – Correlation matrix of predictors variables included in occupancy and n-mixture models.

Table S4.2 – Model format used in occupany and n-mixture models.

193

# model m0 null m1 habitat_amount*edge_density m2 habitat_amount*edge_density + proportion_recently_burnt m3 habitat_amount*edge_density + topographical_ruggedness m4 habitat_amount*edge_density + previous_wet_season_rainfall m5 habitat_amount*edge_density + proportion_recently_burnt + topographical_ruggedness m6 habitat_amount*edge_density + proportion_recently_burnt + previous_wet_season_rainfall m7 habitat_amount*edge_density + previous_wet_season_rainfall + topographical_ruggedness m8 habitat_amount*edge_density + proportion_recently_burnt + previous_wet_season_rainfall + topographical_ruggedness m9 habitat_amount + edge_density m10 habitat_amount + edge_density + proportion_recently_burnt m11 habitat_amount + edge_density + topographical_ruggedness m12 habitat_amount + edge_density + previous_wet_season_rainfall m13 habitat_amount + edge_density + proportion_recently_burnt + topographical_ruggedness m14 habitat_amount + edge_density + proportion_recently_burnt + previous_wet_season_rainfall m15 habitat_amount + edge_density + previous_wet_season_rainfall + topographical_ruggedness m16 habitat_amount + edge_density + proportion_recently_burnt + previous_wet_season_rainfall + topographical_ruggedness

194

Table S4.3 – Model selection table for occupancy models. We used Akaike’s Information Criteria adjusted for small samples (AICc) to rank models.

Model AICc delta habitat_amount+edge_density + proportion_recently_burnt + topographical_ruggedness 838.59 0 habitat_amount+edge_density + proportion_recently_burnt + previous_wet_season_rainfall + topographical_ruggedness 842.9 4.31 habitat_amount*edge_density + proportion_recently_burnt + topographical_ruggedness 842.99 4.4 habitat_amount+edge_density + proportion_recently_burnt 843.99 5.4 habitat_amount+edge_density 845.02 6.43 habitat_amount+edge_density + topographical_ruggedness 845.86 7.27 habitat_amount+edge_density + proportion_recently_burnt + previous_wet_season_rainfall 847 8.41 habitat_amount*edge_density + proportion_recently_burnt 847.13 8.54 habitat_amount*edge_density 847.62 9.03 habitat_amount+edge_density + previous_wet_season_rainfall 847.96 9.37 habitat_amount*edge_density + proportion_recently_burnt + previous_wet_season_rainfall + topographical_ruggedness 848.01 9.42 habitat_amount*edge_density + topographical_ruggedness 848.59 10 habitat_amount+edge_density + previous_wet_season_rainfall + topographical_ruggedness 849.66 11.06

habitat_amount*edge_density + proportion_recently_burnt + previous_wet_season_rainfall 849.98 11.39 habitat_amount*edge_density + previous_wet_season_rainfall 851.27 12.68 habitat_amount*edge_density + previous_wet_season_rainfall + topographical_ruggedness 852.98 14.38 null 854.37 15.78

Dry season Dry habitat_amount*edge_density 775.73 0 habitat_amount*edge_density + previous_wet_season_rainfall 777.02 1.29 habitat_amount+edge_density + proportion_recently_burnt + previous_wet_season_rainfall + topographical_ruggedness 777.41 1.68 habitat_amount+edge_density 777.55 1.82 habitat_amount+edge_density + previous_wet_season_rainfall 777.64 1.91 habitat_amount*edge_density + proportion_recently_burnt 779.56 3.83 habitat_amount*edge_density + topographical_ruggedness 779.57 3.84 habitat_amount+edge_density + topographical_ruggedness 779.84 4.11 habitat_amount+edge_density + previous_wet_season_rainfall + topographical_ruggedness 780.58 4.85 habitat_amount+edge_density + proportion_recently_burnt 781.2 5.48 habitat_amount*edge_density + proportion_recently_burnt + previous_wet_season_rainfall 781.31 5.58 habitat_amount+edge_density + proportion_recently_burnt + previous_wet_season_rainfall 781.35 5.62 habitat_amount*edge_density + previous_wet_season_rainfall + topographical_ruggedness 781.39 5.66

habitat_amount+edge_density + proportion_recently_burnt + topographical_ruggedness 783.62 7.89 habitat_amount*edge_density + proportion_recently_burnt + topographical_ruggedness 783.96 8.23 null 784.93 9.2 habitat_amount*edge_density + proportion_recently_burnt + previous_wet_season_rainfall + topographical_ruggedness 786.34 10.61

Wet season Wet

195

Table S4.4 – Model selection table for n-mixture models. We used Akaike’s Information Criteria adjusted for small samples (AICc) to rank models.

Model AICc delta habitat_amount+edge_density + topographical_ruggedness 1203.27 0 habitat_amount+edge_density + proportion_recently_burnt + topographical_ruggedness 1204.16 0.89 habitat_amount*edge_density + topographical_ruggedness 1207.21 3.94 habitat_amount+edge_density + previous_wet_season_rainfall + topographical_ruggedness 1207.5 4.23 habitat_amount+edge_density + proportion_recently_burnt + previous_wet_season_rainfall + topographical_ruggedness 1209.02 5.75 habitat_amount*edge_density + proportion_recently_burnt + topographical_ruggedness 1209.08 5.81 habitat_amount*edge_density + previous_wet_season_rainfall + topographical_ruggedness 1212.23 8.96 habitat_amount*edge_density + proportion_recently_burnt + previous_wet_season_rainfall + topographical_ruggedness 1214.87 11.6 habitat_amount+edge_density + proportion_recently_burnt 1252.89 49.62 habitat_amount*edge_density + proportion_recently_burnt 1255.79 52.52 habitat_amount+edge_density + proportion_recently_burnt + previous_wet_season_rainfall 1256.31 53.04 habitat_amount*edge_density + proportion_recently_burnt + previous_wet_season_rainfall 1259.44 56.17 habitat_amount*edge_density 1261.43 58.16

habitat_amount+edge_density 1261.8 58.53 habitat_amount*edge_density + previous_wet_season_rainfall 1262.75 59.48 habitat_amount+edge_density + previous_wet_season_rainfall 1263.48 60.21 null 1286.58 83.31

Dry season Dry habitat_amount+edge_density + previous_wet_season_rainfall + topographical_ruggedness 882.23 0 habitat_amount+edge_density + proportion_recently_burnt + previous_wet_season_rainfall + topographical_ruggedness 887.29 5.06 habitat_amount+edge_density + previous_wet_season_rainfall 887.67 5.43 habitat_amount+edge_density + proportion_recently_burnt + previous_wet_season_rainfall 890.52 8.28 habitat_amount*edge_density + previous_wet_season_rainfall 891.39 9.16 habitat_amount*edge_density + proportion_recently_burnt + previous_wet_season_rainfall + topographical_ruggedness 893.19 10.95 habitat_amount*edge_density + proportion_recently_burnt + previous_wet_season_rainfall 895.38 13.15 habitat_amount*edge_density 910.21 27.98 habitat_amount+edge_density 911.14 28.91 habitat_amount+edge_density + topographical_ruggedness 913.81 31.57 habitat_amount*edge_density + topographical_ruggedness 914.25 32.02 habitat_amount*edge_density + proportion_recently_burnt 914.49 32.26 habitat_amount*edge_density + previous_wet_season_rainfall + topographical_ruggedness 914.51 32.28

habitat_amount+edge_density + proportion_recently_burnt + topographical_ruggedness 917.56 35.32 habitat_amount*edge_density + proportion_recently_burnt + topographical_ruggedness 919.28 37.05 habitat_amount+edge_density + proportion_recently_burnt 932.12 49.88 null 941.65 59.41

Wet season Wet

196

Table S4.5 – N-mixture model abundance predictions for northern quolls

Landscape Season Dry Wet 1 1 (0-1) 1 (0-2) 2 4 (2-7) 2 (1-4) 3 9 (5-16) 5 (2-14) 4 71 (40-126) 71 (44-114) 5 16 (9-27) 33 (20-54) 6 0 (0-0) 0 (0-0) 7 12 (7-20) 22 (13-37) 8 6 (3-12) 4 (1-9) 9 0 (0-1) 1 (0-4) 11 4 (2-7) 2 (1-5)

12 11 (6-17) 19 (11-31) 13 10 (5-17) 5 (2-9) 14 0 (0-1) 0 (0-1) 15 0 (0-1) 0 (0-1) 16 9 (4-20) 15 (7-32) 17 0 (0-0) 0 (0-0) 18 0 (0-1) 0 (0-1) 19 0 (0-1) 0 (0-1) 21 8 (5-13) 18 (10-31) 22 1 (0-2) 1 (0-5) 23 20 (11-36) 28 (15-55) 24 1 (0-2) 1 (0-3)

197

Appendix F – Supplementary material for chapter 5

Figure S5.1 – Average monthly total precipitation (mm) for study area. Red square represent months in which data was collected to be used in dry season models, and blue square represent months in which data was collected to be used in wet season models. Data from weather station 004016 (Indee station).

198

Figure S5.2 – Predictions from linear models describing the relationship between the total number of northern quoll (Dasyurus hallucatus) individuals detected at a site and the total number of northern quoll independent detection events. Lines represent model predictions, points represent raw data.

199

Estimate SD CI Dry (Intercept) 0.59 0.13 0.33 – 0.83 Independent detections 0.28 0.01 0.25 – 0.30 Wet (Intercept) 0.49 0.13 0.24 – 0.74 Independent detections 0.20 0.01 0.18 – 0.21

Table S5.1 – Output from linear models describing the relationship between the total number of northern quoll (Dasyurus hallucatus) individuals detected at a site and the total number of northern quoll independent detection events.

200

Table S5.2 – Correlation matrix of continuous covariates included in rocky array models. Correlation between covariates was calculated using Pearson’s correlation coefficient

Covariates Den Vegetation Patch Patch Habitat availability cover(<0.5m) shape area Extent Den availability 1.00 -0.12 0.17 0.16 0.14 Vegetation cover (<0.5m) -0.12 1.00 0.05 0.18 0.02 Patch shape 0.17 0.05 1.00 0.44 0.48 Patch area 0.16 0.18 0.44 1.00 0.26 Habitat extent 0.14 0.02 0.48 0.26 1.00

Table S5.3 – Correlation matrix of continuous covariates included in spinifex array models. Correlation between covariates was calculated using Pearson’s correlation coefficient

Covariates Distance to Vegetation cover Years since rocky outcrop (<0.5m) burnt Distance to rocky outcrop 1.00 -0.08 -0.01 Vegetation cover (<0.5m) -0.08 1.00 -0.14 Years since burnt -0.01 -0.14 1.00

201

Table S5.4 – Hurdle model selection table for the dry season rocky sites model. We used Akaike’s Information Criteria adjusted for small samples (AICc) to rank zero-inflated and conditional models separately. Predictors included in top zero-inflated and conditional model were included in the final model. PG = Patch geomorphology, HE = Habitat extent (amount of rocky habitat within a 150m buffer of camera sites), DA = Den availability, PA = Patch area, PS = Patch shape (Patch area to edge habitat ratio), VC.0.5M = Vegetation cover (<0.5m).

Intercept PG HE DA PA PS VC.0.5M df AICc delta weight Zero inflated model -0.27 + + + + 8 384.23 0.00 0.16 -0.25 + + + + + 9 384.92 0.69 0.11 -1.58 + + + + + 11 385.94 1.71 0.07 -1.62 + + + + + + 12 386.13 1.91 0.06 -0.26 + + + 7 386.19 1.97 0.06 Conditional model 0.71 + + 6 405.02 0.00 0.15 0.75 + + + 7 406.12 1.11 0.09 0.67 + + + 7 406.75 1.73 0.06 0.81 + 5 406.93 1.92 0.06

202

Table S5.5 – Hurdle model selection table for the wet season rocky sites model. We used Akaike’s Information Criteria adjusted for small samples (AICc) to rank zero-inflated and conditional models separately. Predictors included in top zero-inflated and conditional model were included in the final model. PG = Patch geomorphology, HE = Habitat extent (amount of rocky habitat within a 150m buffer of camera sites), DA = Den availability, AER = Patch shape (Patch area to edge habitat ratio), VC.0.5M = Vegetation cover (<0.5m).

Intercept PG HE DA PA PS VC.0.5M df AICc delta weight Zero inflated model 0.76 + + + + 9 334.42 0.00 0.27 0.76 + + + + 10 335.62 1.20 0.15 0.76 + + + + 10 336.00 1.58 0.12 0.76 + + + + 10 336.04 1.62 0.12 Conditional model 0.76 4 365.99 0.00 0.07 0.72 + 5 366.57 0.58 0.05 0.88 + 5 366.99 1.00 0.04 0.83 + 5 367.01 1.02 0.04 1.39 + + + + 10 367.74 1.74 0.03 0.83 + + 6 367.81 1.81 0.03

203

Table S5.6 – Generalised linear mixed effects model selection table for the spinifex sites model. We used Akaike’s Information Criteria adjusted for small samples (AICc) to rank models. DR = Distance from rocky patch, VC.0.5M = Vegetation cover (<0.5m), YSF = Years since fire.

Intercept DR VC.0.5M YSF Df AICc AIC∆ wi -2.15 + + 3 57.50 0.00 0.62 -2.15 + + + 4 59.12 1.61 0.27

204

Figure S5.3 – Correlograms of hurdle model residuals based on Moran’s I. Models are A.) dry season rocky array, B.) wet season rocky array, C.) dry season combined array, D.) wet season combined array, and E.) spinifex array.

205

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.

Abbott, I. (2013). Extending the application of Aboriginal names to Australian biota:'Dasyurus'(Marsupialia: Dasyuridae) species. Victorian Naturalist, The 130, 109.

Abramoff, M. D., Magalhaes, P. J., and Ram, S. J. (2004). Image Processing with ImageJ. Biophotonics International 11, 36-42.

ABS, A. B. o. S. (2018). 8412.0 - Mineral and Petroleum Exploration, Australia.)

Aitken, S. N. and Whitlock, M. C. (2013). Assisted gene flow to facilitate local adaptation to climate change. Annual review of ecology, evolution, and systematics 44.

ALA (2019). Atlas of Living Australia.)

Alexander, P. D. and Gese, E. M. (2018). Identifying individual cougars (Puma concolor) in remote camera images–implications for population estimates. Wildlife research 45, 274-281.

Algar, D., Angus, G., Williams, M., and Mellican, A. (2007). Influence of bait type, weather and prey abundance on bait uptake by feral cats (Felis catus) on Peron Peninsula, Western Australia. Conservation Science Western Australia 6, 109.

Allan, B. M., Nimmo, D. G., Ierodiaconou, D., VanDerWal, J., Koh, L. P., and Ritchie, E. G. (2018). Futurecasting ecological research: the rise of technoecology. Ecosphere 9, e02163.

Allan, G. E. and Southgate, R. I. (2002). Fire regimes in spinifex landscapes. Flammable Australia: The Fire Regimes and Biodiversity of a Continent 1, 145-176.

Allen, L. and Sparkes, E. (2001). The effect of dingo control on sheep and beef cattle in Queensland. Journal of Applied Ecology 38, 76-87.

Allen, M. L., Peterson, B., and Krofel, M. (2018). No respect for apex carnivores: distribution and activity patterns of honey badgers in the Serengeti. Mammalian biology 89, 90-94.

Alonso, R. S., McClintock, B. T., Lyren, L. M., Boydston, E. E., and Crooks, K. R. (2015). Mark- recapture and mark-resight methods for estimating abundance with remote cameras: a carnivore case study. PloS one 10, e0123032.

206

Amir Abdul Nasir, A. F., Cameron, S. F., Niehaus, A. C., Clemente, C. J., von Hippel, F. A., and Wilson, R. S. (2018a). Manganese contamination affects the motor performance of wild northern quolls (Dasyurus hallucatus). Environmental Pollution 241, 55-62.

Amir Abdul Nasir, A. F., Cameron, S. F., von Hippel, F. A., Postlethwait, J., Niehaus, A. C., Blomberg, S., and Wilson, R. S. (2018b). Manganese accumulates in the brain of northern quolls (Dasyurus hallucatus) living near an active mine. Environmental Pollution 233, 377-386.

Amir Abdul Nasir, A. F., Clemente, C. J., Wynn, M. L., and Wilson, R. S. (2017). Optimal running speeds when there is a trade‐off between speed and the probability of mistakes. Functional Ecology 31, 1941-1949.

Andren, H. (1994). Effects of habitat fragmentation on birds and mammals in landscapes with different proportions of suitable habitat: a review. Oikos, 355-366.

Andrén, H. (1995). Effects of landscape composition on predation rates at habitat edges. In 'Mosaic landscapes and ecological processes' pp. 225-255. (Springer.)

Andrén, H., Angelstam, P., Lindström, E., and Widen, P. (1985). Differences in predation pressure in relation to habitat fragmentation: an experiment. Oikos, 273-277.

Apps, P. and McNutt, J. W. (2018). Are camera traps fit for purpose? A rigorous, reproducible and realistic test of camera trap performance. African journal of ecology 56, 710-720.

Araujo, G., Agustines, A., Tracey, B., Snow, S., Labaja, J., and Ponzo, A. (2019). Photo-ID and telemetry highlight a global whale shark hotspot in Palawan, Philippines. Sci Rep 9, 172-209. doi: 10.1038/s41598-019-53718-w.

Ariefiandy, A., Purwandana, D., Seno, A., Ciofi, C., and Jessop, T. S. (2013). Can camera traps monitor Komodo dragons a large ectothermic predator? PLoS One 8, e58800.

Asensio, N., Arroyo‐Rodríguez, V., Dunn, J. C., and Cristóbal‐Azkarate, J. (2009). Conservation value of landscape supplementation for howler monkeys living in forest patches. Biotropica 41, 768- 773.

Ash, A., McIvor, J., Mott, J., and Andrew, M. (1997). Building grass castles: integrating ecology and management of Australia's tropical tallgrass rangelands. The Rangeland Journal 19, 123-144.

Ashman, K. R., Watchorn, D. J., and Whisson, D. A. (2019). Prioritising research efforts for effective species conservation: a review of 145 years of koala research. Mammal Review 49, 189-200.

Attenbrow, V. and Attenbrow, V. (1987) 'The Upper Mangrove Creek catchment: a study of quantitative changes in the archaeological record.' (University of Sydney.)

207

Austin, C., Tuft, K., Ramp, D., Cremona, T., and Webb, J. K. (2017). Bait preference for remote camera trap studies of the endangered northern quoll (Dasyurus hallucatus). Australian Mammalogy 39, 72-77.

Australian Bureau of Meteorology (2020). Climate Data Online. (Australian Bureau of Meteorology.)

Azevedo, F., Lemos, F., Freitas‐Junior, M., Rocha, D., and Azevedo, F. (2018). Puma activity patterns and temporal overlap with prey in a human‐modified landscape at Southeastern Brazil. Journal of Zoology 305, 246-255.

Baker, M. A. A., Reeve, N., Conkey, A. A., Macdonald, D. W., and Yamaguchi, N. (2017). Hedgehogs on the move: Testing the effects of land use change on home range size and movement patterns of free-ranging Ethiopian hedgehogs. PloS one 12, 672-679.

Balderama, E., Gardner, B., and Reich, B. J. (2016). A spatial–temporal double-hurdle model for extremely over-dispersed avian count data. Spatial Statistics 18, 263-275.

Balme, J., O’Connor, S., and Fallon, S. (2018). New dates on dingo bones from Madura Cave provide oldest firm evidence for arrival of the species in Australia. Scientific reports 8, 1-6.

Barnosky, A. D., Matzke, N., Tomiya, S., Wogan, G. O., Swartz, B., Quental, T. B., Marshall, C., McGuire, J. L., Lindsey, E. L., and Maguire, K. C. (2011). Has the Earth’s sixth mass extinction already arrived? Nature 471, 51-57.

Barton, K. (2020). MuMIn: Multi-Model Inference. R package version 1.43.17.)

Bascompte, J. and Solé, R. V. (1996). Habitat fragmentation and extinction thresholds in spatially explicit models. Journal of Animal Ecology, 465-473.

Bastille‐Rousseau, G., Wall, J., Douglas‐Hamilton, I., and Wittemyer, G. (2018). Optimizing the positioning of wildlife crossing structures using GPS telemetry. Journal of Applied Ecology 55, 2055- 2063.

Bates, D., Sarkar, D., Bates, M. D., and Matrix, L. (2007). The lme4 package. In 'R package version' pp. 74.)

Baynes, A. and McDowell, M. C. (2010). The original mammal fauna of the Pilbara biogeographic region of north-western Australia. Records of the Western Australian Museum, 285-298.

Beaglehole, J. C. (1963) 'The Endeavour Journal of Joseph Banks, 1768-1771.' ([Sydney]: Trustees of the Public Library of New South Wales.)

Begg, R. J. (1981). The small mammals of Little Nourlangie Rock, NT III. Ecology of Dasyurus hallucatus, the northern quoll (Marsupialia: Dasyuridae). Wildlife research 8, 73-85.

208

Bender, D. J. and Fahrig, L. (2005). Matrix structure obscures the relationship between interpatch movement and patch size and isolation. Ecology 86, 1023-1033.

Bengsen, A., Butler, J., and Masters, P. (2012). Estimating and indexing feral cat population abundances using camera traps. Wildlife research 38, 732-739.

Bennett, A., Lumsden, L., and Menkhorst, P. (1989). Mammals of the mallee region of south-eastern Australia. Mediterranean landscapes in Australia: mallee ecosystems and their management. Melbourne, CSIRO, 191-220.

Bennett, A. F. (1990). Habitat corridors and the conservation of small mammals in a fragmented forest environment. Landscape Ecology 4, 109-122.

Bennett, A. F., Radford, J. Q., and Haslem, A. (2006). Properties of land mosaics: implications for nature conservation in agricultural environments. Biological conservation 133, 250-264.

Bennett, D. and Clements, T. (2014). The use of passive infrared camera trapping systems in the study of frugivorous monitor lizards. Biawak 8, 19-30.

Betts, M. G., Forbes, G. J., Diamond, A. W., and Taylor, P. D. (2006). Independent effects of fragmentation on forest songbirds: an organism‐based approach. Ecological applications 16, 1076- 1089.

Biz, M., Cornelius, C., and Metzger, J. P. W. (2017). Matrix type affects movement behavior of a Neotropical understory forest bird. Perspectives in ecology and conservation 15, 10-17.

Bleich, V. C., Pierce, B. M., Davis, J. L., and Davis, V. L. (1996). Thermal characteristics of mountain lion dens. The Great Basin Naturalist 56, 276-278.

Bleicher, S. S. and Dickman, C. R. (2020). On the landscape of fear: shelters affect foraging by dunnarts (Marsupialia, Sminthopsis spp.) in a sandridge desert environment. Journal of Mammalogy 101, 281-290.

Bluff, L. A., Clausen, L., Hill, A., and Bramwell, M. D. (2011). A decade of monitoring the remnant Victorian population of the brush-tailed rock-wallaby (Petrogale penicillata). Australian Mammalogy 33, 195-201.

Bornmann, L. and Mutz, R. (2015). Growth rates of modern science: A bibliometric analysis based on the number of publications and cited references. Journal of the Association for Information Science and Technology 66, 2215-2222.

Bowman, D. M., Brown, G., Braby, M., Brown, J., Cook, L. G., Crisp, M., Ford, F., Haberle, S., Hughes, J., and Isagi, Y. (2010). Biogeography of the Australian monsoon tropics. Journal of Biogeography 37, 201-216.

209

Bradley, J., Holmes, M., Marrngawi, D. N., Karrakayn, A. I., Wuwarlu, J. M., and Ninganga, I. (2006). Yumbulyumbulmantha ki-Awarawu: all kinds of things from country: Yanyuwa ethnobiological classification. Aboriginal and Torres Strait Islander Studies Unit Research Report Series 6.

Braithwaite, R. W. and Griffiths, A. D. (1994). Demographic variation and range contraction in the Northern Quoll, Dasyurus hallucatus (Marsupialia: Dasyuridae). Wildlife research 21, 203-217. doi: 10.1071/wr9940203.

Braithwaite, R. W. and Muller, W. (1997). Rainfall, groundwater and refuges: predicting extinctions of Australian tropical mammal species. Australian Journal of Ecology 22, 57-67.

Breiner, F. T., Guisan, A., Nobis, M. P., and Bergamini, A. (2017). Including environmental niche information to improve IUCN Red List assessments. Diversity and distributions 23, 484-495.

Brides, K., Middleton, J., Leighton, K., and Grogan, A. (2018). The use of camera traps to identify individual colour-marked geese at a moulting site. Ringing & Migration 33, 19-22.

Brieger, F., Hagen, R., Kröschel, M., Hartig, F., Petersen, I., Ortmann, S., and Suchant, R. (2017). Do roe deer react to wildlife warning reflectors? A test combining a controlled experiment with field observations. European Journal of Wildlife Research 63, 72.

Broennimann, O., Treier, U. A., Müller‐Schärer, H., Thuiller, W., Peterson, A., and Guisan, A. (2007). Evidence of climatic niche shift during biological invasion. Ecology Letters 10, 701-709.

Brook, B. W., Whitehead, P. J., and Dingle, J. K. (2011) 'Potential cane toad short to medium term control techniques: The biological feasibility and cost of exclusion as a mitigating control strategy.' (Department of the Environment and Heritage.)

Brook, L. A., Johnson, C. N., and Ritchie, E. G. (2012). Effects of predator control on behaviour of an apex predator and indirect consequences for mesopredator suppression. Journal of Applied Ecology 49, 1278-1286.

Brooks, K., Rowat, D., Pierce, S. J., Jouannet, D., and Vely, M. (2010). Seeing spots: photo- identification as a regional tool for whale shark identification. Western Indian Ocean Journal of Marine Science 9, 185-194.

Brown, C. J., Harborne, A. R., Paris, C. B., and Mumby, P. J. (2016). Uniting paradigms of connectivity in marine ecology. Ecology 97, 2447-2457. doi: 10.1002/ecy.1463.

Burbidge, A. and Woinarski, J. (2016). Dasyurus viverrinus. The IUCN Red List of Threatened Species 2016.

Burbidge, A. A. and Fuller, P. (1979). Mammals of the Warburton region, Western Australia. Records of the Western Australian Museum 8, 57-73.

210

Burbidge, A. A., Johnson, K. A., Fuller, P. J., and Southgate, R. (1988). Aboriginal knowledge of the mammals of the central deserts of Australia. Wildlife research 15, 9-39.

Burbidge, A. A. and McKenzie, N. (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., Brennan, K., Woinarski, J., Dickman, C., Baynes, A., Gordon, G., Menkhorst, P., and Robinson, A. C. (2009). Conservation status and biogeography of Australia’s terrestrial mammals. Australian Journal of Zoology 56, 411-422.

Burbidge, A. H., Maron, M., Clarke, M. F., Baker, J., Oliver, D. L., and Ford, G. (2011). Linking science and practice in ecological research and management: how can we do it better? Ecological Management & Restoration 12, 54-60.

Burnett, S. (1997). Colonizing cane toads cause population declines in native predators: reliable anecdotal information and management implications. Pacific Conservation Biology 3, 65-72.

Burnett, S., Shimizu, Y., and Middleton, J. (2013). Distribution and abundance of the Northern quoll (Dasyurus hallucatus) in far north Queensland. Univeristy of the Sunshine Coast. (Sunshine Coast.)

Burnett, S. and Zwar, A. (2009). Quolls (Dasyurus maculatus and D. hallucatus) in the southern Mary River catchment, south-east Queensland. Wildlife Preservation Society of Queensland, Sippy Downs.

Burnham, K. P. and Anderson, D. R. (2002). Model selection and.

Burnham, K. P. and Anderson, D. R. (2004). Multimodel inference: understanding AIC and BIC in model selection. Sociological methods & research 33, 261-304.

Burrows, N., Ward, B., and Robinson, A. (2009). Fuel dynamics and fire spread in spinifex grasslands of the Western Desert. Proceedings of the Royal Society of Queensland, The 115, 69.

Burton, A. C., Neilson, E., Moreira, D., Ladle, A., Steenweg, R., Fisher, J. T., Bayne, E., and Boutin, S. (2015). Wildlife camera trapping: a review and recommendations for linking surveys to ecological processes. Journal of Applied Ecology 52, 675-685.

Calaby, J. (1973). Mammals. In ‘Alligator Rivers Region Environmental Factfinding Study: Wildlife’. (CSIRO Division of Wildlife Research: Canberra.)

Calver, M., King, D., Bradley, J., Gardner, J., and Martin, G. (1989). An assessment of the potential target specificity of 1080 predator baiting in Western-Australia. Wildlife research 16, 625-638.

Cardoso, M. J., Eldridge, M. D., Oakwood, M., Rankmore, B., Sherwin, W. B., and Firestone, K. B. (2009). Effects of founder events on the genetic variation of translocated island populations: implications for conservation management of the northern quoll. Conservation Genetics 10, 1719- 1733.

211

Ceballos, G., Ehrlich, P. R., Barnosky, A. D., García, A., Pringle, R. M., and Palmer, T. M. (2015). Accelerated modern human–induced species losses: Entering the sixth mass extinction. Science advances 1, e1400253.

Ceballos, G., Ehrlich, P. R., and Dirzo, R. (2017). Biological annihilation via the ongoing sixth mass extinction signaled by vertebrate population losses and declines. Proceedings of the National Academy of Sciences 114, E6089-E6096.

Ceballos, G., Ehrlich, P. R., and Raven, P. H. (2020). Vertebrates on the brink as indicators of biological annihilation and the sixth mass extinction. Proceedings of the National Academy of Sciences.

Cerutti-Pereyra, F., Bassos-Hull, K., Arvizu-Torres, X., Wilkinson, K., García-Carrillo, I., Perez- Jimenez, J., and Hueter, R. (2018). Observations of spotted eagle rays (Aetobatus narinari) in the Mexican Caribbean using photo-ID. Environmental biology of fishes 101, 237-244.

Chan, R., Dunlop, J., and Spencer, P. (2020). Highly promiscuous paternity in mainland and island populations of the endangered Northern Quoll. Journal of Zoology 310, 210-220.

Chen, I.-P., Symonds, M. R., Melville, J., and Stuart-Fox, D. (2013). Factors shaping the evolution of colour patterns in Australian agamid lizards (Agamidae): a comparative study. Biological Journal of the Linnean Society 109, 101-112.

Cole, J. R. and Woinarski, J. (2000). Rodents of the arid Northern Territory: conservation status and distribution. Wildlife research 27, 437-449.

Colwell, R. K. and Rangel, T. F. (2009). Hutchinson's duality: the once and future niche. Proceedings of the National Academy of Sciences 106, 19651-19658.

Comer, S., Speldewinde, P., Tiller, C., Clausen, L., Pinder, J., Cowen, S., and Algar, D. (2018). Evaluating the efficacy of a landscape scale feral cat control program using camera traps and occupancy models. Scientific reports 8, 5335.

Cook, A. (2010a) Fire effect on habitat use and home-range of the northern quoll, Dasyurus hallucatus. (Masters of Science Thesis, University of Western Austalia.)

Cook, A. (2010b) 'Habitat use and home-range of the northern quoll, Dasyurus hallucatus: effects of fire.' (University of Western Australia.)

Cooper, C. E. and Withers, P. C. (2010). Comparative physiology of Australian quolls (Dasyurus; Marsupialia). Journal of Comparative Physiology B 180, 857-868.

Cowan, M., Moro, D., Anderson, H., Angus, J., Garretson, S., and Morris, K. (2020a). Aerial baiting for feral cats is unlikely to affect survivorship of northern quolls in the Pilbara region of Western Australia. Wildlife research.

212

Cowan, M. A., Dunlop, J. A., Turner, J. M., Moore, H. A., and Nimmo, D. G. (2020b). Artificial refuges to combat habitat loss for an endangered marsupial predator: How do they measure up? Conservation Science and Practice 2, 204.

Cramer, V. A., Dunlop, J., Davis, R., Ellis, R., Barnett, B., Cook, A., Morris, K., and van Leeuwen, S. (2016). Research priorities for the northern quoll (Dasyurus hallucatus) in the Pilbara region of Western Australia. Australian Mammalogy 38, 135-148.

Crawley, M. J. (2012) 'The R book.' (John Wiley & Sons.)

Cremona, T., Baker, A. M., Cooper, S. J., Montague-Drake, R., Stobo-Wilson, A. M., and Carthew, S. M. (2020). Integrative taxonomic investigation of Petaurus breviceps (Marsupialia: Petauridae) reveals three distinct species. Zoological Journal of the Linnean Society.

Cremona, T., Crowther, M., and Webb, J. K. (2017a). High mortality and small population size prevent population recovery of a reintroduced mesopredator. Animal conservation 20, 555-563.

Cremona, T., Spencer, P., Shine, R., and Webb, J. K. (2017b). Avoiding the last supper: parentage analysis indicates multi-generational survival of re-introduced ‘toad-smart’lineage. Conservation Genetics 18, 1475-1480.

Crooks, K. R., Burdett, C. L., Theobald, D. M., King, S. R., Di Marco, M., Rondinini, C., and Boitani, L. (2017). Quantification of habitat fragmentation reveals extinction risk in terrestrial mammals. Proceedings of the National Academy of Sciences 114, 7635-7640.

Crooks, K. R. and Soulé, M. E. (1999). Mesopredator release and avifaunal extinctions in a fragmented system. Nature 400, 563.

Cunningham, C. X., Johnson, C. N., Barmuta, L. A., Hollings, T., Woehler, E. J., and Jones, M. E. (2018). Top carnivore decline has cascading effects on scavengers and carrion persistence. Proceedings of the Royal Society B 285, 20181582.

Davidson, A. D., Shoemaker, K. T., Weinstein, B., Costa, G. C., Brooks, T. M., Ceballos, G., Radeloff, V. C., Rondinini, C., and Graham, C. H. (2017). Geography of current and future global mammal extinction risk. PloS one 12, e0186934.

Davies, H. F., McCarthy, M. A., Firth, R. S., Woinarski, J. C., Gillespie, G. R., Andersen, A. N., Geyle, H. M., Nicholson, E., and Murphy, B. P. (2017). 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., Woinarski, J. C., Gillespie, G. R., Andersen, A. N., Rioli, W., Puruntatameri, J., Roberts, W., and Kerinaiua, C. (2018). Declining populations in one of the last refuges for threatened mammal species in northern Australia. Austral Ecology 43, 602-612.

Davis, M. B. and Shaw, R. G. (2001). Range shifts and adaptive responses to Quaternary climate change. Science 292, 673-679.

213

DBCA (2017). Numbat (Myrmecobius fasciatus) Recovery Plan. Department of Biodiversity, Conservation and Attractions (Western Australia.)

De Bondi, N., White, J. G., Stevens, M., and Cooke, R. (2010). A comparison of the effectiveness of camera trapping and live trapping for sampling terrestrial small-mammal communities. Wildlife research 37, 456-465.

Den Hartog, J. and Reijns, R. (2016). Interactive Individual Identification Software (I3S).)

Denkinger, J., Alarcon, D., Espinosa, B., Fowler, L., Manning, C., Oña, J., and Palacios, D. M. (2020). Social structure of killer whales (Orcinus orca) in a variable low‐latitude environment, the Galápagos Archipelago. Marine Mammal Science 36, 774-785.

Dick, M., Rous, A. M., Nguyen, V. M., and Cooke, S. J. (2016). Necessary but challenging: multiple disciplinary approaches to solving conservation problems. Facets 1, 67-82.

Dickman, C. and Braithwaite, R. W. (1992). Postmating mortality of males in the dasyurid marsupials, Dasyurus and Parantechinus. Journal of Mammalogy 73, 143-147.

Dickman, C., Mahon, P., Masters, P., and Gibson, D. (1999). Long-term dynamics of rodent populations in arid Australia: the influence of rainfall. Wildlife research 26, 389-403.

Dickman, C. R., Glen, A. S., and Letnic, M. (2009). Reintroducing the dingo: can Australia’s conservation wastelands be restored. Reintroduction of top-order predators 238, 269.

Diete, R. L., Meek, P. D., Dixon, K. M., Dickman, C. R., and Leung, L. K.-P. (2016). Best bait for your buck: bait preference for camera trapping north Australian mammals. Australian Journal of Zoology 63, 376-382.

Dixon, J. M. and Huxley, L. (1985) 'Donald Thomson's mammals and fishes of northern Australia.' (Nelson.)

DMIRS (2020). Mining Tenements (DMIRS-003). (Ed. I. R. a. S. Department of Mines).)

Do Carmo, F. F. and Jacobi, C. M. (2016). Diversity and plant trait-soil relationships among rock outcrops in the Brazilian Atlantic rainforest. Plant and Soil 403, 7-20.

Doherty, T., Davis, A., Robert, and van Etten, E. (2015a). A game of cat-and-mouse: microhabitat influences rodent foraging in recently burnt but not long unburnt shrublands. Journal of Mammalogy 96, 324-331.

Doherty, T. S., Dickman, C. R., Nimmo, D. G., and Ritchie, E. G. (2015b). Multiple threats, or multiplying the threats? Interactions between invasive predators and other ecological disturbances. Biological conservation 190, 60-68.

214

Doody, J., Green, B., Sims, R., Rhind, D., West, P., and Steer, D. (2006). Indirect impacts of invasive cane toads (Bufo marinus) on nest predation in pig-nosed turtles (Carettochelys insculpta). Wildlife research 33, 349-354.

Doody, J. S., McHenry, C., Letnic, M., Everitt, C., Sawyer, G., and Clulow, S. (2018). Forecasting the spatiotemporal pattern of the cane toad invasion into north-western Australia. Wildlife research.

Doughty, P., Rolfe, J. K., Burbidge, A. H., Pearson, D. J., and Kendrick, P. G. (2011). Herpetological assemblages of the Pilbara biogeographic region, Western Australia: ecological associations, biogeographic patterns and conservation. Records of the Western Australian Museum, Supplement 78, 315-341.

DPIRD (2017). The Kimberley rangelands. Department of Primary Industries and Regional Development.

Driscoll, D. A. (2005). Is the matrix a sea? Habitat specificity in a naturally fragmented landscape. Ecological Entomology 30, 8-16.

Driscoll, D. A., Banks, S. C., Barton, P. S., Lindenmayer, D. B., and Smith, A. L. (2013). Conceptual domain of the matrix in fragmented landscapes. Trends in Ecology & Evolution 28, 605-613.

Dunlop, J., Johnson, B., Rayner, K., and Morris, K. (2015). Northern Quoll trapping surveys at Wall Creek and Mesa 228. Report prepared for Roy Hill Pty Ltd: Department of Parks and Wildlife, Kensington.

Dunlop, J. A., Rayner, K., and Doherty, T. S. (2017). Dietary flexibility in small carnivores: a case study on the endangered northern quoll, Dasyurus hallucatus. Journal of Mammalogy 98, 858-866.

Dunning, J. B., Danielson, B. J., and Pulliam, H. R. (1992). Ecological processes that affect populations in complex landscapes. Oikos 65, 169-175.

Edwards, R. D., Crisp, M. D., Cook, D. H., and Cook, L. G. (2017). Congruent biogeographical disjunctions at a continent-wide scale: Quantifying and clarifying the role of biogeographic barriers in the Australian tropics. PloS one 12, e0174812.

Einoder, L. D., Southwell, D. M., Lahoz-Monfort, J. J., Gillespie, G. R., Fisher, A., and Wintle, B. A. (2018). Occupancy and detectability modelling of vertebrates in northern Australia using multiple sampling methods. PloS one 13, e0203304.

Elith, J., Phillips, S. J., Hastie, T., Dudík, M., Chee, Y. E., and Yates, C. J. (2011). A statistical explanation of MaxEnt for ecologists. Diversity and distributions 17, 43-57.

Environmental Protection Authority (2014). Cumulative environmental impacts of development in the Pilbara region. Western Australia Government

215

ESCAVI, E. S. C. f. A. V. I. (2003). Australian Vegetation Attribute Manual: National Vegetation Information System Version 6.0. (Department of Environment Executive Steering Committee for Australian Vegetation Information, Canberra, Australian Capital Territory.)

ESRI (2011). ArcGIS Desktop: Release 10.)

Evans, M. C. and Cvitanovic, C. (2018). An introduction to achieving policy impact for early career researchers. Palgrave Communications 4, 1-12.

Ewers, R. M. and Didham, R. K. (2006). Confounding factors in the detection of species responses to habitat fragmentation. Biological Reviews 81, 117-142.

Eycott, A. E., Stewart, G. B., Buyung-Ali, L. M., Bowler, D. E., Watts, K., and Pullin, A. S. (2012). A meta-analysis on the impact of different matrix structures on species movement rates. Landscape Ecology 27, 1263-1278.

Fahrig, L. (1998). When does fragmentation of breeding habitat affect population survival? Ecological modelling 105, 273-292.

Fahrig, L. (2003). Effects of habitat fragmentation on biodiversity. Annual review of ecology, evolution, and systematics 34, 487-515.

Fahrig, L. (2007). Non‐optimal animal movement in human‐altered landscapes. Functional Ecology 21, 1003-1015.

Fahrig, L. (2017). Ecological responses to habitat fragmentation per se. Annual review of ecology, evolution, and systematics 48, 1-23.

Fancourt, B. A., Bateman, B. L., VanDerWal, J., Nicol, S. C., Hawkins, C. E., Jones, M. E., and Johnson, C. N. (2015a). Testing the role of climate change in species decline: is the eastern quoll a victim of a change in the weather? PloS one 10, e0129420.

Fancourt, B. A., Hawkins, C. E., Cameron, E. Z., Jones, M. E., and Nicol, S. C. (2015b). Devil declines and catastrophic cascades: is mesopredator release of feral cats inhibiting recovery of the eastern quoll? PloS one 10, e0119303.

Fancourt, B. A., Sweaney, M., and Fletcher, D. B. (2018). More haste, less speed: pilot study suggests camera trap detection zone could be more important than trigger speed to maximise species detections. Australian Mammalogy 40, 118-121.

Fardila, D., Kelly, L. T., Moore, J. L., and McCarthy, M. A. (2017). A systematic review reveals changes in where and how we have studied habitat loss and fragmentation over 20 years. Biological conservation 212, 130-138.

216

Firestone, K. B., Houlden, B. A., Sherwin, W. B., and Geffen, E. (2000). Variability and differentiation of microsatellites in the genus Dasyurus and conservation implications for the large Australian carnivorous marsupials. Conservation Genetics 1, 115-133.

Firth, R. S., Brook, B. W., Woinarski, J. C., and 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., Johnson, C. N., Lawes, M. J., Fritz, S. A., McCallum, H., Blomberg, S. P., VanDerWal, J., Abbott, B., Frank, A., and Legge, S. (2014a). The current decline of tropical marsupials in A ustralia: is history repeating? Global Ecology and Biogeography 23, 181-190.

Fisher, D. O., Johnson, C. N., Lawes, M. J., Fritz, S. A., McCallum, H., Blomberg, S. P., VanDerWal, J., Abbott, B., Frank, A., and Legge, S. (2014b). The current decline of tropical marsupials in Australia: is history repeating? Global Ecology and Biogeography 23, 181-190.

Fiske, I. and Chandler, R. (2011). Unmarked: an R package for fitting hierarchical models of wildlife occurrence and abundance. Journal of statistical software 43, 1-23.

Fitzsimons, J. A. and Michael, D. R. (2017). Rocky outcrops: a hard road in the conservation of critical habitats. Biological conservation 211, 36-44.

Flanagan, S. P., Forester, B. R., Latch, E. K., Aitken, S. N., and Hoban, S. (2018). Guidelines for planning genomic assessment and monitoring of locally adaptive variation to inform species conservation. Evolutionary Applications 11, 1035-1052.

Flannery, T. F. (1990). Pleistocene faunal loss: implications of the aftershock for Australia's past and future. Archaeology in Oceania 25, 45-55.

Fleming, P. A. and Bateman, P. W. (2016). The good, the bad, and the ugly: which Australian terrestrial mammal species attract most research? Mammal Review 46, 241-254.

Forcino, F. L., Leighton, L. R., Twerdy, P., and Cahill, J. F. (2015). Reexamining sample size requirements for multivariate, abundance-based community research: when resources are limited, the research does not have to be. PloS one 10, e0128379.

Forman, R. (2014). Land mosaics: The ecology of landscapes and regions 1995. The Ecological Design and Planning Reader; IslandPress: Washington, DC, USA, 217-234.

Foster, R. J. and Harmsen, B. J. (2012). A critique of density estimation from camera‐trap data. The Journal of wildlife management 76, 224-236.

Frankham, R. (2015). Genetic rescue of small inbred populations: Meta‐analysis reveals large and consistent benefits of gene flow. Molecular Ecology 24, 2610-2618.

217

Frankham, R., Ballou, J. D., Ralls, K., Eldridge, M., Dudash, M. R., Fenster, C. B., Lacy, R. C., and Sunnucks, P. (2017) 'Genetic management of fragmented animal and plant populations.' (Oxford University Press.)

Franklin, J. F. and Lindenmayer, D. B. (2009). Importance of matrix habitats in maintaining biological diversity. Proceedings of the National Academy of Sciences 106, 349-350.

Freeland, W. J. and Martin, K. C. (1985). The rate of range expansion by Bufo marinus in Northern Australia, 1980-84. Wildlife research 12, 555-559.

French, R. K., Muller, C. G., Chilvers, B. L., and Battley, P. F. (2019). Behavioural consequences of human disturbance on subantarctic Yellow-eyed Penguins, Megadyptes antipodes. Bird Conservation International 29, 277-290.

Friend, J. The numbat Myrmecobius fasciatus (Myrmecobiidae): history of decline and potential for recovery. 1990 pp. 369-377.

Fry, B. G., Vidal, N., Norman, J. A., Vonk, F. J., Scheib, H., Ramjan, S. R., Kuruppu, S., Fung, K., Hedges, S. B., and Richardson, M. K. (2006). Early evolution of the venom system in lizards and snakes. Nature 439, 584-588.

Garrick, D. (2008). Body surface temperature and length in relation to the thermal biology of lizards. Bioscience Horizons 1, 136-142.

Geary, W. L., Doherty, T. S., Nimmo, D. G., Tulloch, A. I., and Ritchie, E. G. (2019a). Predator responses to fire: A global systematic review and meta‐analysis. Journal of Animal Ecology.

Geary, W. L., Doherty, T. S., Nimmo, D. G., Tulloch, A. I. T., and Ritchie, E. G. (2020). Predator responses to fire: A global systematic review and meta-analysis. J Anim Ecol 89, 955-971. doi: 10.1111/1365-2656.13153.

Geary, W. L., Nimmo, D. G., Doherty, T. S., Ritchie, E. G., and Tulloch, A. I. (2019b). Threat webs: Reframing the co‐occurrence and interactions of threats to biodiversity. Journal of Applied Ecology 56, 1992-1997.

GéCzy, C. (2009). Cannibalism in captive Varanus timorensis. Biawak 3, 61-63.

Gehring, T. M. and Swihart, R. K. (2004). Home range and movements of long-tailed weasels in a landscape fragmented by agriculture. Journal of Mammalogy 85, 79-86.

Geoscience Australia (2008). GEODATA 9 second DEM and D8: Digital Elevation Model Version 3 and Flow Direction Grid 2008.

Geyle, H. M., Woinarski, J. C., Baker, G. B., Dickman, C. R., Dutson, G., Fisher, D. O., Ford, H., Holdsworth, M., Jones, M. E., and Kutt, A. (2018). Quantifying extinction risk and forecasting the

218 number of impending Australian bird and mammal extinctions. Pacific Conservation Biology 24, 157- 167.

Gibbons, J. W., Scott, D. E., Ryan, T. J., Buhlmann, K. A., Tuberville, T. D., Metts, B. S., Greene, J. L., Mills, T., Leiden, Y., and Poppy, S. (2000). The Global Decline of Reptiles, Déjà Vu Amphibians: Reptile species are declining on a global scale. Six significant threats to reptile populations are habitat loss and degradation, introduced invasive species, environmental pollution, disease, unsustainable use, and global climate change. BioScience 50, 653-666.

Giraudoux, P. (2018). pgirmess: Spatial Analysis and Data Mining for Field Ecologists.)

Glen, A., Anderson, D., Veltman, C., Garvey, P., and Nichols, M. (2016). Wildlife detector dogs and camera traps: a comparison of techniques for detecting feral cats. New Zealand Journal of Zoology 43, 127-137.

Goldingay, R. L. (2012). Characteristics of tree hollows used by Australian arboreal and scansorial mammals. Australian Journal of Zoology 59, 277-294.

Goodfellow, D. (1993) 'Fauna of kakadu and the top end.' (Denise Goodfellow.)

Gould, J. Characters of a new species of Perameles, and a new species of Dasyurus. 1842 pp. 41-42.

Gould, J., Dixon, J. M., Gould, J., Ornithologist, G. B., Gould, J., and Ornithologue, G.-B. (1863) 'Australian marsupials and monotremes.' (Macmillan.)

Green, B. and King, D. (1978). Home range and activity patterns of the sand goanna, Varanus gouldii (Reptilia: Varanidae). Wildlife research 5, 417-424.

Gregg, E. A., Tingley, R., and Phillips, B. L. (2019). The on‐ground feasibility of a waterless barrier to stop the spread of invasive cane toads in Western Australia. Conservation Science and Practice, e74.

Griffiths, A. D. and Brook, B. W. (2015). Fire impacts recruitment more than survival of small‐ mammals in a tropical savanna. Ecosphere 6, 1-22.

Griffiths, A. D., Garnett, S. T., and 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.

Griffiths, A. D., Rankmore, B., Brennan, K., and Woinarski, J. C. (2017). Demographic evaluation of translocating the threatened northern quoll to two Australian islands. Wildlife research 44, 238-247.

Groombridge, B., Jenkins, M. D., and Jenkins, M. (2002) 'World atlas of biodiversity: earth's living resources in the 21st century.' (Univ of California Press.)

219

Guarino, F. (2002). Spatial ecology of a large carnivorous lizard, Varanus varius (Squamata: Varanidae). Journal of Zoology 258, 449-457.

Guisan, A. and Thuiller, W. (2005). Predicting species distribution: offering more than simple habitat models. Ecology Letters 8, 993-1009.

Haig, A., Matthes, U., and Larson, D. (2000). Effects of natural habitat fragmentation on the species richness, diversity, and composition of cliff vegetation. Canadian journal of botany 78, 786-797.

Hamer, R. P., Gardiner, R. Z., Proft, K. M., Johnson, C. N., and Jones, M. E. (2021). A triple threat: high population density, high foraging intensity and flexible habitat preferences explain high impact of feral cats on prey. Proceedings of the Royal Society B 288, 20201194.

Hansen, N. A., Sato, C. F., Michael, D. R., Lindenmayer, D. B., and Driscoll, D. A. (2019). Predation risk for reptiles is highest at remnant edges in agricultural landscapes. Journal of Applied Ecology 56, 31-43.

Hanski, I. (1994). A practical model of metapopulation dynamics. Journal of Animal Ecology 63, 151- 162.

Hargis, C. D., Bissonette, J. A., and Turner, D. L. (1999). The influence of forest fragmentation and landscape pattern on American martens. Journal of Applied Ecology 36, 157-172.

Harrison, J. T., Kochert, M. N., Pauli, B. P., and Heath, J. A. (2019). Using motion-activated trail cameras to study diet and productivity of cliff-nesting Golden Eagles. Journal of Raptor Research 53, 26-37.

Harrison, X. A., Donaldson, L., Correa-Cano, M. E., Evans, J., Fisher, D. N., Goodwin, C. E. D., Robinson, B. S., Hodgson, D. J., and Inger, R. (2018). A brief introduction to mixed effects modelling and multi-model inference in ecology. PeerJ 6, e4794. doi: 10.7717/peerj.4794.

Hartley, M. J. and Hunter, M. L. (1998). A meta-analysis of forest cover, edge effects, and artificial nest predation rates. Conservation biology 12, 465-469.

Haslem, A., Kelly, L. T., Nimmo, D. G., Watson, S. J., Kenny, S. A., Taylor, R. S., Avitabile, S. C., Callister, K. E., Spence‐Bailey, L. M., and Clarke, M. F. (2011). Habitat or fuel? Implications of long‐ term, post‐fire dynamics for the development of key resources for fauna and fire. Journal of Applied Ecology 48, 247-256.

Haynes, K. and Cronin, J. (2006). Interpatch movement and edge effects: the role of behavioral responses to the landscape matrix. Oikos 113, 43-54.

Heiniger, J., Cameron, S. F., Madsen, T., Niehaus, A. C., and Wilson, R. S. (2020). Demography and spatial requirements of the endangered northern quoll on Groote Eylandt. Wildlife research 47, 224- 238.

220

Heiniger, J. and Gillespie, G. (2018). High variation in camera trap-model sensitivity for surveying mammal species in northern Australia. Wildlife research 45, 578-585.

Henderson, M. (2015) The Effects of Mining Infrastructure on Northern quoll Movement and Habitat (Edith Cowan University Perth.)

Hernandez-Santin, L., Dunlop, J. A., Goldizen, A. W., and Fisher, D. O. (2019). Demography of the northern quoll (Dasyurus hallucatus) in the most arid part of its range. Journal of Mammalogy.

Hernandez-Santin, L., Goldizen, A. W., and Fisher, D. O. (2016). Introduced predators and habitat structure influence range contraction of an endangered native predator, the northern quoll. Biological conservation 203, 160-167.

Hernandez-Santin, L., Henderson, M., Molloy, S. W., Dunlop, J. A., and Davis, R. A. (2020). Spatial ecology of an endangered carnivore, the Pilbara northern quoll. Australian Mammalogy.

Hernandez-Satin, L. (2016) Ecology and predator associations of the northern quoll ( Dasyurus hallucatus ) in the Pilbara. (University of Queensland Brisbane.)

Hertel, A. G., Leclerc, M., Warren, D., Pelletier, F., Zedrosser, A., and Mueller, T. (2019). Don't poke the bear: using tracking data to quantify behavioural syndromes in elusive wildlife. Animal behaviour 147, 91-104.

Higashide, D., Miura, S., and Miguchi, H. (2012). Are chest marks unique to A siatic black bear individuals? Journal of Zoology 288, 199-206.

Hill, B. M. and Ward, S. J. (2010). National recovery plan for the northern quoll Dasyurus hallucatus. Department of Natural Resources, Environment, The Arts and Sport, Darwin.

Hobbs, R. J. and Saunders, D. (1994). Effects of landscape fragmentation in agricultural areas. In 'Conservation biology in Australia and Oceania' pp. 77-95. (Surrey Beatty and Sons Pty Ltd.)

Hobbs, R. J., Valentine, L. E., Standish, R. J., and Jackson, S. T. (2017). Movers and stayers: novel assemblages in changing environments. Trends in Ecology & Evolution.

Hodgson, J. A., Moilanen, A., Wintle, B. A., and Thomas, C. D. (2011). Habitat area, quality and connectivity: striking the balance for efficient conservation. Journal of Applied Ecology 48, 148-152.

Hohnen, R., Ashby, J., Tuft, K., and McGregor, H. (2013). Individual identification of northern quolls (Dasyurus hallucatus) using remote cameras. Australian Mammalogy 35, 131-135. doi: 10.1071/am12015.

Hohnen, R., Tuft, K., Legge, S., Walters, N., Johanson, L., Carver, S., Radford, I. J., and Johnson, C. N. (2016a). The significance of topographic complexity in habitat selection and persistence of a declining marsupial in the Kimberley region of Western Australia. Australian Journal of Zoology 64, 198-216.

221

Hohnen, R., Tuft, K. D., Legge, S., Hillyer, M., Spencer, P. B., Radford, I. J., Johnson, C. N., and Burridge, C. P. (2016b). Rainfall and topography predict gene flow among populations of the declining northern quoll (Dasyurus hallucatus). Conservation Genetics 17, 1213-1228.

How, R., Spencer, P., and Schmitt, L. (2009). Island populations have high conservation value for northern Australia's top marsupial predator ahead of a threatening process. Journal of Zoology 278, 206-217.

Hradsky, B. A., Mildwaters, C., Ritchie, E. G., Christie, F., and Di Stefano, J. (2017). Responses of invasive predators and native prey to a prescribed forest fire. Journal of Mammalogy 98, 835-847.

Hudgens, B. R. and Garcelon, D. K. (2011). Induced changes in island fox (Urocyon littoralis) activity do not mitigate the extinction threat posed by a novel predator. Oecologia 165, 699-705.

Ibbett, M., Woinarski, J., and Oakwood, M. (2018). Declines in the mammal assemblage of a rugged sandstone environment in Kakadu National Park, Northern Territory, Australia. Australian Mammalogy 40, 181-187.

Indigo, N., Smith, J., Webb, J. K., and Phillips, B. (2018). Not such silly sausages: evidence suggests northern quolls exhibit aversion to toads after training with toad sausages. Austral Ecology 43, 592- 601.

Indigo, N. L. (2020) Safeguarding the northern quoll. Can we mitigate cane toad impacts through conditioned taste aversion? (PhD thesis, University of Technology Sydney.)

IUCN (2006). Guidelines for using the IUCN red list categories and criteria. International Union for Conservation of Nature, Gland, Switzerland.

IUCN (2021). The IUCN Red List of Threatened Species. Version 2021-1. (Ed. IUCN).)

Jackson, S., Jackson, S. M., and Groves, C. (2015) 'Taxonomy of Australian mammals.' (CSIRO publishing.)

Jacobsen, R., Howell, C., and Read, S. (2020). Australia’s Indigenous land and forest estate: separate reporting of Indigenous ownership, management and other special rights.

Jessop, T. S., Madsen, T., Sumner, J., Rudiharto, H., Phillips, J. A., and Ciofi, C. (2006). Maximum body size among insular Komodo dragon populations covaries with large prey density. Oikos 112, 422-429.

Jhala, Y., Qureshi, Q., and Gopal, R. (2011). Can the abundance of tigers be assessed from their signs? Journal of Applied Ecology 48, 14-24.

Johnson, C. (2006) 'Australia's mammal extinctions: a 50,000-year history.' (Cambridge University Press.)

222

Johnson, C. N. and Isaac, J. L. (2009). Body mass and extinction risk in Australian marsupials: the ‘Critical Weight Range’revisited. Austral Ecology 34, 35-40.

Johnson, C. R. (1976). Some behavioural observations on wild and captive sand monitors, Varanus gouldii (Sauria: Varanidae). Zoological Journal of the Linnean Society 59, 377-380.

Johnston, M., Algar, D., O'Donoghue, M., Morris, J., Buckmaster, T., and Quinn, J. (2020). Efficacy and welfare assessment of an encapsulated para-aminopropiophenone (PAPP) formulation as a bait- delivered toxicant for feral cats (Felis catus). Wildlife research 47, 686-697.

Johnston, M., O'Donoghue, M., Holdsworth, M., Robinson, S., Herrod, A., Eklom, K., Gigliotti, F., Bould, L., and Little, N. (2013) 'Field assessment of the Curiosity® bait for managing feral cats in the Pilbara.' (Arthur Rylah Institute for Environmental Research, Department of ….)

Jolly, C. J., Kelly, E., Gillespie, G. R., Phillips, B., and Webb, J. K. (2018a). Out of the frying pan: reintroduction of toad‐smart northern quolls to southern Kakadu National Park. Austral Ecology 43, 139-149.

Jolly, C. J. and Phillips, B. L. (2020). Effects of rapid evolution due to predator‐free conservation on endangered species recovery. Conservation biology.

Jolly, C. J., Smart, A. S., Moreen, J., Webb, J. K., Gillespie, G. R., and Phillips, B. L. (2021). Trophic cascade driven by behavioural fine-tuning as naïve prey rapidly adjust to a novel predator. Ecology.

Jolly, C. J., Webb, J. K., Gillespie, G. R., and Phillips, B. L. (2020). Training fails to elicit behavioral change in a marsupial suffering evolutionary loss of antipredator behaviors. Journal of Mammalogy 101, 1108-1116.

Jolly, C. J., Webb, J. K., and Phillips, B. L. (2018b). The perils of paradise: an endangered species conserved on an island loses antipredator behaviours within 13 generations. Biology letters 14, 20180222.

Joly, K., Dale, B. W., Collins, W. B., and Adams, L. G. (2003). Winter habitat use by female caribou in relation to wildland fires in interior Alaska. Canadian journal of zoology 81, 1192-1201.

Jones, M. E., Burnett, S., Claridge, A. W., Fancourt, B., Kortner, G., Morris, K., Peacock, D., Troy, S., and Woinarski, J. (2014). Australia’s surviving marsupial carnivores: threats and conservation. Carnivores of Australia: Past, Present and Future, 197-240.

Kanine, J. M., Kierepka, E. M., Castleberry, S. B., Mengak, M. T., Nibbelink, N. P., and Glenn, T. C. (2018). Influence of landscape heterogeneity on the functional connectivity of Allegheny woodrats (Neotoma magister) in Virginia. Conservation Genetics 19, 1259-1268.

Kaplan, D. M. and White, C. G. (2002) 'Integrating landscape ecology into natural resource management.' (Cambridge University Press.)

223

Kellner, C. J., Lawson, G. R., Tomke, S. A., and Noble, J. H. (2017). Computer-aided individual identification of Sceloporus consobrinus based on patterns of head scalation. Herpetol. Rev 48, 766- 769.

Kelly, E. (2018) Targeted gene flow for conservation: northern quolls and the invasive cane toad.

Kelly, E., Jolly, C. J., Indigo, N., Smart, A., Webb, J., and Phillips, B. (2020). No outbreeding depression in a trial of targeted gene flow in an endangered Australian marsupial. Conservation Genetics, 1-11.

Kelly, E. and Phillips, B. L. (2016). Targeted gene flow for conservation. Conservation biology 30, 259-267.

Kelly, E. and Phillips, B. L. (2017). Get smart: native mammal develops toad-smart behavior in response to a toxic invader. Behavioral Ecology 28, 854-858.

Kelly, E. and Phillips, B. L. (2019). Targeted gene flow and rapid adaptation in an endangered marsupial. Conservation biology 33, 112-121.

Kelly, E., Phillips, B. L., and Webb, J. K. (2018). Taste overshadows less salient cues to elicit food aversion in endangered marsupial. Applied Animal Behaviour Science 209, 83-87.

Kendall, W. L. (1999). Robustness of closed capture–recapture methods to violations of the closure assumption. Ecology 80, 2517-2525.

Kennedy, M., Phillips, B. L., Legge, S., Murphy, S. A., and Faulkner, R. A. (2012). Do dingoes suppress the activity of feral cats in northern Australia? Austral Ecology 37, 134-139.

Keppel, G., Van Niel, K. P., Wardell‐Johnson, G. W., Yates, C. J., Byrne, M., Mucina, L., Schut, A. G., Hopper, S. D., and Franklin, S. E. (2012). Refugia: identifying and understanding safe havens for biodiversity under climate change. Global Ecology and Biogeography 21, 393-404.

Kerle, J. A. and Burgman, M. (1984). Some Aspects of the Ecology of the Mammal Fauna of the Jabiluka Area. Northern Territory. Wildlife research 11, 207-222.

Kery, M. (2002). Inferring the absence of a species: a case study of snakes. The Journal of wildlife management, 330-338.

King, A. W. (1999). Dispersal success on fractal landscapes: a consequence of lacunarity thresholds. Landscape Ecology 14, 73-82.

King, D. and Green, B. (1999) 'Goannas: the biology of varanid lizards.' (UNSW Press: Sydney, 2052, Australia.)

224

King, D., Green, B., and Butler, H. (1989). The Activity Pattern, Temperature Regulation and Diet of Varanus-Giganteus on Barrow-Island, Western-Australia. Wildlife research 16, 41-47.

King, D. R. (1989). An assessment of the hazard posed to northern quolls (Dasyurus hallucatus) by aerial baiting with 1080 to control dingoes. Australian Wildlife Research 16, 569-574. doi: 10.1071/wr9890569.

King, M. and Horner, P. (1987). A new species of monitor (Reptilia: Platynota) from northern Australia and a note on the status of Varanus acanthurus insulanicus Mertens. Beagle: Records of the Museums and Art Galleries of the Northern Territory, The 4, 73-77.

Kitchener, D., Keller, L., Chapman, A., McKenzie, N., Start, A., and Kenneally, K. (1981). Observations on mammals of the Mitchell Plateau area, Kimberley, Western Australia. Biological Survey of Mitchell Plateau and Admiralty Gulf, Kimberley, Western Australia, 123-168.

Knight, A. T., Cowling, R. M., Rouget, M., Balmford, A., Lombard, A. T., and Campbell, B. M. (2008). Knowing but not doing: selecting priority conservation areas and the research–implementation gap. Conservation biology 22, 610-617.

Kotler, B., Brown, J., and Knight, M. (1999). Habitat and patch use by hyraxes: there’s no place like home? Ecology Letters 2, 82-88.

Kotler, B. P. (1984). Risk of predation and the structure of desert rodent communities. Ecology 65, 689-701.

Kramer‐Schadt, S., Niedballa, J., Pilgrim, J. D., Schröder, B., Lindenborn, J., Reinfelder, V., Stillfried, M., Heckmann, I., Scharf, A. K., and Augeri, D. M. (2013). The importance of correcting for sampling bias in MaxEnt species distribution models. Diversity and distributions 19, 1366-1379.

Krebs, C. J. (1989) 'Ecological methodology.' (Harper & Row New York.)

Krefft, G. (1862) 'On the vertebrated animals of the lower Murray and Darling, their habits, economy and geographical distribution.'

Kupfer, J. A. (2012). Landscape ecology and biogeography: rethinking landscape metrics in a post- FRAGSTATS landscape. Progress in physical geography 36, 400-420.

Kupfer, J. A., Malanson, G. P., and Franklin, S. B. (2006). Not seeing the ocean for the islands: the mediating influence of matrix‐based processes on forest fragmentation effects. Global Ecology and Biogeography 15, 8-20.

Kutt, A. S. and Woinarski, J. C. (2007). The effects of grazing and fire on vegetation and the vertebrate assemblage in a tropical savanna woodland in north-eastern Australia. Journal of Tropical Ecology, 95-106.

225

Lande, R. (1987). Extinction thresholds in demographic models of territorial populations. The American Naturalist 130, 624-635.

Lashley, M. A., Cove, M. V., Chitwood, M. C., Penido, G., Gardner, B., DePerno, C. S., and Moorman, C. E. (2018). Estimating wildlife activity curves: comparison of methods and sample size. Scientific reports 8, 4173.

Laundré, J. W., Hernández, L., and Altendorf, K. B. (2001). Wolves, elk, and bison: reestablishing the" landscape of fear" in Yellowstone National Park, USA. Canadian journal of zoology 79, 1401- 1409.

Laurance, W. F., Useche, D. C., Rendeiro, J., Kalka, M., Bradshaw, C. J., Sloan, S. P., Laurance, S. G., Campbell, M., Abernethy, K., and Alvarez, P. (2012). Averting biodiversity collapse in tropical forest protected areas. Nature 489, 290-294.

Laurance, W. F. and Yensen, E. (1991). Predicting the impacts of edge effects in fragmented habitats. Biological conservation 55, 77-92.

Lawes, M. J., Mealin, P. E., and Piper, S. E. (2000). Patch occupancy and potential metapopulation dynamics of three forest mammals in fragmented afromontane forest in South Africa. Conservation biology 14, 1088-1098.

Leahy, L., Legge, S. M., Tuft, K., McGregor, H. W., Barmuta, L. A., Jones, M. E., and 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., Murphy, S. A., and 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., Smith, J. G., James, A., Tuft, K. D., Webb, T., and Woinarski, J. C. (2019). Interactions among threats affect conservation management outcomes: Livestock grazing removes the benefits of fire management for small mammals in Australian tropical savannas. Conservation Science and Practice 1, e52.

Lei, J. and Booth, D. T. (2018). Intraspecific variation in space use of a coastal population of lace monitors (Varanus varius). Australian Journal of Zoology 65, 398-407.

Leo, V., Reading, R. P., Gordon, C., and Letnic, M. (2019). Apex predator suppression is linked to restructuring of ecosystems via multiple ecological pathways. Oikos 128, 630-639.

Lever, C. (2001) 'The cane toad: the history and ecology of a successful colonist.' (Westbury Academic & Scientific Pub.)

Li, D., Liu, Y., Sun, X., Lloyd, H., Zhu, S., Zhang, S., Wan, D., and Zhang, Z. (2017). Habitat- dependent changes in vigilance behaviour of Red-crowned Crane influenced by wildlife tourism. Scientific reports 7, 16614.

226

Lidicker, W. Z. (1999). Responses of mammals to habitat edges: an overview. Landscape Ecology 14, 333-343.

Liedloff, A. C., Coughenour, M. B., Ludwig, J. A., and Dyer, R. (2001). Modelling the trade-off between fire and grazing in a tropical savanna landscape, northern Australia. Environment International 27, 173-180.

Lindén, A. and Mäntyniemi, S. (2011). Using the negative binomial distribution to model overdispersion in ecological count data. Ecology 92, 1414-1421.

Lindenmayer, D. B., Likens, G. E., Andersen, A., Bowman, D., Bull, C. M., Burns, E., Dickman, C. R., Hoffmann, A. A., Keith, D. A., and Liddell, M. J. (2012). Value of long‐term ecological studies. Austral Ecology 37, 745-757.

Loggins, A. A., Shrader, A. M., Monadjem, A., and McCleery, R. A. (2019). Shrub cover homogenizes small mammals' activity and perceived predation risk. Sci Rep 9, 16857. doi: 10.1038/s41598-019-53071-y.

MacArthur, R. H. and Wilson, E. O. (2001) 'The theory of island biogeography.' (Princeton university press.)

MacKenzie, D. I. and Nichols, J. D. (2004). Occupancy as a surrogate for abundance estimation. Animal biodiversity and conservation 27, 461-467.

Maclagan, S., Coates, T., Hradsky, B., Butryn, R., and Ritchie, E. (2019). Life in linear habitats: the movement ecology of an endangered mammal in a peri‐urban landscape. Animal conservation 23, 260-272.

Majer, J. D. (2014). Mining and biodiversity: Are they compatible? In 'Resource curse or cure?' pp. 195-205. (Springer.)

Mares, M. (1997). The geobiological interface: granitic outcrops as a selective force in mammalian evolution. Journal of the Royal Society of Western Australia 80, 131-139.

Marini, M. A., Robinson, S. K., and Heske, E. J. (1995). Edge effects on nest predation in the Shawnee National Forest, southern Illinois. Biological conservation 74, 203-213.

Marlow, N. J., Thomas, N. D., Williams, A. A., Macmahon, B., Lawson, J., Hitchen, Y., Angus, J., and Berry, O. (2015). Cats (Felis catus) are more abundant and are the dominant predator of woylies (Bettongia penicillata) after sustained fox (Vulpes vulpes) control. Australian Journal of Zoology 63, 18-27.

May, S. A. and Norton, T. (1996). Influence of fragmentation and disturbance on the potential impact of feral predators on native fauna in Australian forest ecosystems. Wildlife research 23, 387-400.

227

McDonald, P. J., Griffiths, A. D., Nano, C. E., Dickman, C. R., Ward, S. J., and Luck, G. W. (2015a). Landscape-scale factors determine occupancy of the critically endangered central rock-rat in arid Australia: the utility of camera trapping. Biological conservation 191, 93-100.

McDonald, P. J., Luck, G. W., Dickman, C. R., Ward, S. J., and Crowther, M. S. (2015b). Using multiple‐source occurrence data to identify patterns and drivers of decline in arid‐dwelling Australian marsupials. Ecography 38, 1090-1100.

McDonald, P. J., Stewart, A., and Dickman, C. R. (2018). Applying the niche reduction hypothesis to modelling distributions: a case study of a critically endangered rodent. Biological conservation 217, 207-212.

McDonald, P. J., Stewart, A., Jensen, M. A., and McGregor, H. W. (2020). Topographic complexity potentially mediates cat predation risk for a critically endangered rodent. Wildlife research 47, 643- 648.

McGregor, H., Legge, S., Jones, M. E., and Johnson, C. N. (2015a). Feral Cats Are Better Killers in Open Habitats, Revealed by Animal-Borne Video. PloS one 10, e0133915. doi: 10.1371/journal.pone.0133915.

McGregor, H. W., Cliff, H. B., and Kanowski, J. (2017). Habitat preference for fire scars by feral cats in Cape York Peninsula, Australia. Wildlife research 43, 623-633.

McGregor, H. W., Legge, S., Jones, M. E., and Johnson, C. N. (2014). Landscape management of fire and grazing regimes alters the fine-scale habitat utilisation by feral cats. PloS one 9, e109097.

McGregor, H. W., Legge, S., Potts, J., Jones, M. E., and Johnson, C. N. (2015b). Density and home range of feral cats in north-western Australia. Wildlife research 42, 223-231.

McIlroy, J. (1981). The sensitivity of Australian animals to 1080 poison. II. Marsupial and eutherian carnivores. Wildlife research 8, 385-399.

McIlroy, J. (1982). The sensitivity of Australian animals to 1080 poison. III. Marsupial and eutherian herbivores. Wildlife research 9, 487-503.

McKenzie, N. (1981). Mammals of the Phanerozoic south-west Kimberley, Western Australia: biogeography and recent changes. Journal of Biogeography, 263-280.

McKenzie, N., Burbidge, A., Baynes, A., Brereton, R., Dickman, C., Gordon, G., Gibson, L., Menkhorst, P., Robinson, A., and Williams, M. (2007). Analysis of factors implicated in the recent decline of Australia's mammal fauna. Journal of Biogeography 34, 597-611.

McKenzie, N., Chapman, A., and Youngson, W. (1975). Mammals of the prince regent river reserve, north-west Kimberley, Australia. Wildlife Research Bulletin 3, 69.

228

McKenzie, N., Van Leeuwen, S., and Pinder, A. (2009). Introduction to the Pilbara biodiversity survey, 2002–2007. Records of the Western Australian Museum, Supplement 78, 3-89.

Meek, P., Ballard, G., Fleming, P., and Falzon, G. (2016). Are we getting the full picture? Animal responses to camera traps and implications for predator studies. Ecology and evolution 6, 3216-3225.

Meek, P. and Pittet, A. (2013). User-based design specifications for the ultimate camera trap for wildlife research. Wildlife research 39, 649-660.

Meek, P. D., Ballard, G.-A., and Fleming, P. J. (2015). The pitfalls of wildlife camera trapping as a survey tool in Australia. Australian Mammalogy 37, 13-22.

Meek, P. D., Ballard, G.-A., Fleming, P. J., Schaefer, M., Williams, W., and Falzon, G. (2014a). Camera traps can be heard and seen by animals. PloS one 9, e110832.

Meek, P. D., Fleming, P., and Ballard, G. (2012) 'An introduction to camera trapping for wildlife surveys in Australia.' (Invasive Animals Cooperative Research Centre Canberra, Australia.)

Meek, P. D., Fleming, P., Ballard, G., Banks, P., Claridge, A. W., Sanderson, J., and Swann, D. (2014b) 'Camera trapping: wildlife management and research.' (Csiro Publishing Melbourne, Australia.)

Menkhorst, P. and Knight, F. (2011) 'Field guide to the mammals of Australia.' (Oxford University Press: New York.)

Menkhorst, P. W. (2009). Blandowski’s mammals: Clues to a lost world. Proceedings of the Royal Society of Victoria 121, 61-89.

Meredith, M. and Ridout, M. (2014). Overview of the overlap package. In 'R Proj' pp. 1-9.)

Meserve, P. L., Gutiérrez, J. R., Yunger, J. A., Contreras, L. C., and Jaksic, F. M. (1996). Role of biotic interactions in a small mammal assemblage in semiarid Chile. Ecology 77, 133-148.

Metallinou, M., Cervenka, J., Crochet, P. A., Kratochvil, L., Wilms, T., Geniez, P., Shobrak, M. Y., Brito, J. C., and Carranza, S. (2015). Species on the rocks: Systematics and biogeography of the rock- dwelling Ptyodactylus geckos (Squamata: Phyllodactylidae) in North Africa and Arabia. Mol Phylogenet Evol 85, 208-212. doi: 10.1016/j.ympev.2015.02.010.

Michael, D. and Lindenmayer, D. (2018) 'Rocky Outcrops in Australia: Ecology, Conservation and Management.' (CSIRO PUBLISHING.)

Michael, D. R., Cunningham, R. B., and Lindenmayer, D. B. (2008). A forgotten habitat? Granite inselbergs conserve reptile diversity in fragmented agricultural landscapes. Journal of Applied Ecology 45, 1742-1752.

229

Michel, V. T., Jiménez‐Franco, M. V., Naef‐Daenzer, B., and Grüebler, M. U. (2016). Intraguild predator drives forest edge avoidance of a mesopredator. Ecosphere 7, e01229.

Miller, B. P. and Murphy, B. P. (2017). Fire and Australian vegetation. Australian Vegetation. Cambridge University Press, Cambridge 3, 113-134.

Miritis, V., Rendall, A. R., Doherty, T. S., Coetsee, A. L., and Ritchie, E. G. (2020). Living with the enemy: a threatened prey species coexisting with feral cats on a fox-free island. Wildlife research 47, 633-642.

Moher, D., Liberati, A., Tetzlaff, J., and Altman, D. G. (2009). Preferred reporting items for systematic reviews and meta-analyses: the PRISMA statement. Annals of internal medicine 151, 264- 269.

Molloy, S. W., Davis, R. A., Dunlop, J. A., and van Etten, E. J. (2017). Applying surrogate species presences to correct sample bias in species distribution models: a case study using the Pilbara population of the Northern Quoll. Nature Conservation 18, 25.

Molyneux, J., Pavey, C., James, A., and Carthew, S. (2018). The efficacy of monitoring techniques for detecting small mammals and reptiles in arid environments. Wildlife research 44, 534-545.

Moore, H., Champney, J., Dunlop, J., Valentine, L., and Nimmo, D. (2020a). Spot on: Using camera traps to individually monitor one of the world’s largest lizards. Wildlife research 47, 326-337.

Moore, H., Champney, J., Dunlop, J., Valentine, L., and Nimmo, D. (2020b). Spot on: Using camera traps to individually monitor one of the world’s largest lizards.

Moore, H., Michael, D., Ritchie, E., Dunlop, J., Valentine, L., Hobbs, R., and Nimmo, D. (2021). In the heart of a spinifex sea: habitat selection of an endangered marsupial predator is multiscale dependent. Landscape Ecology.

Moore, H. A., Dunlop, J. A., Valentine, L. E., Woinarski, J. C., Ritchie, E. G., Watson, D. M., and Nimmo, D. G. (2019). Topographic ruggedness and rainfall mediate geographic range contraction of a threatened marsupial predator. Diversity and distributions 25, 1818-1831.

Moore, H. A., Valentine, L. E., Dunlop, J. A., and Nimmo, D. G. (2020c). The effect of camera orientation on the detectability of wildlife: a case study from north‐western Australia. Remote Sensing in Ecology and Conservation.

Moro, D., Dunlop, J., and Williams, M. R. (2019). Northern quoll persistence is most sensitive to survivorship of juveniles. Wildlife research 46, 165-175.

Moro, D. and MacAulay, I. (2014). Computer-aided pattern recognition of large reptiles as a noninvasive application to identify individuals. Journal of applied animal welfare science 17, 125- 135.

230

Moro, D. D., J._Williams, M. R. (2019). Northern quoll persistence is most sensitive to survivorship of juveniles. Wildlife research 46, 165-175. doi: 10.1071/WR18010.

Morris, K., Cowan, M., Angus, J., Anderson, H., Garretson, S., Algar, D., and Williams, M. (2015). The Northern Quoll Cat Bait Uptake and Survivorship Study, Yarraloola Offset Area, Pilbara Region, WA: Yandicoogina JSW and Oxbow Project, Threatened Species Offset Plan.

Morris, K., Sims, C., Himbeck, K., Christensen, P., Sercombe, N., Ward, B., and Noakes, N. (2004). Project Eden--fauna recovery on Peron Peninsula, Shark Bay: Western Shield review--February 2003. Conservation Science Western Australia 5.

Morton, A. (1982). The effects of marking and capture on recapture frequencies of butterflies. Oecologia 53, 105-110.

Muneza, A. B., Ortiz-Calo, W., Packer, C., Cusack, J. J., Jones, T., Palmer, M. S., Swanson, A., Kosmala, M., Dickman, A. J., and Macdonald, D. W. (2019). Quantifying the severity of giraffe skin disease via photogrammetry analysis of camera trap data. Journal of wildlife diseases.

Murphy, B. P. and Davies, H. F. (2014). There is a critical weight range for A ustralia's declining tropical mammals. Global Ecology and Biogeography 23, 1058-1061.

Murphy, B. P., Lehmann, C. E., Russell‐Smith, J., and Lawes, M. J. (2014). Fire regimes and woody biomass dynamics in Australian savannas. Journal of Biogeography 41, 133-144.

Murray, J., Choy, S. L., McAlpine, C., Possingham, H., and Goldizen, A. (2008). The importance of ecological scale for wildlife conservation in naturally fragmented environments: a case study of the brush-tailed rock-wallaby (Petrogale penicillata). Biological conservation 141, 7-22.

NAFI (2020). Fire History. Northern Australia and Rangelands Fire Information.

Nakagawa, S. and Cuthill, I. C. (2007). Effect size, confidence interval and statistical significance: a practical guide for biologists. Biological reviews of the Cambridge Philosophical Society 82, 591-605. doi: 10.1111/j.1469-185X.2007.00027.x.

NatureMap (2019). Mapping Western Australia’s Biodiversity.)

Nawab, J., Khan, S., Shah, M. T., Khan, K., Huang, Q., and Ali, R. (2015). Quantification of heavy metals in mining affected soil and their bioaccumulation in native plant species. International journal of Phytoremediation 17, 801-813.

Nelson, J. E. and Gemmell, R. T. (2003). Birth in the northern quoll, Dasyurus hallucatus (Marsupialia: Dasyuridae). Australian Journal of Zoology 51, 187-198.

NESP (2018). Our changing climate: How will rainfall change in Northern Australia over this century?

231

Nichols, M., Glen, A. S., Garvey, P., and Ross, J. (2017). A comparison of horizontal versus vertical camera placement to detect feral cats and mustelids. New Zealand Journal of Ecology 41, 145-150.

Nimmo, D., Kelly, L., Spence‐Bailey, L., Watson, S., Taylor, R., Clarke, M., and Bennett, A. (2013). Fire mosaics and reptile conservation in a fire‐prone region. Conservation biology 27, 345-353.

Nimmo, D. G., Avitabile, S., Banks, S. C., Bliege Bird, R., Callister, K., Clarke, M. F., Dickman, C. R., Doherty, T. S., Driscoll, D. A., Greenville, A. C., Haslem, A., Kelly, L. T., Kenny, S. A., Lahoz- Monfort, J. J., Lee, C., Leonard, S., Moore, H., Newsome, T. M., Parr, C. L., Ritchie, E. G., Schneider, K., Turner, J. M., Watson, S., Westbrooke, M., Wouters, M., White, M., and Bennett, A. F. (2019). Animal movements in fire-prone landscapes. Biol Rev Camb Philos Soc 94, 981-998. doi: 10.1111/brv.12486.

Nimmo, D. G., Kelly, L. T., Spence‐Bailey, L. M., Watson, S. J., Haslem, A., White, J. G., Clarke, M. F., and Bennett, A. F. (2012). Predicting the century‐long post‐fire responses of reptiles. Global Ecology and Biogeography 21, 1062-1073.

NVIS, D. o. t. E. a. E. (2019). National Vegetation Information System, NVIS.)

O'Bryan, C. J., Garnett, S. T., Fa, J. E., Leiper, I., Rehbein, J., Fernández-Llamazares, Á., Jonas, H. D., Brondizio, E., Burgess, N., and Robinson, C. (2019). The importance of Indigenous Peoples' lands for the conservation of terrestrial vertebrates. bioRxiv.

O'Connell, A. F., Nichols, J. D., and Karanth, K. U. (2010) 'Camera traps in animal ecology: methods and analyses.' (Springer Science & Business Media.)

O’Donnell, S., Webb, J. K., and Shine, R. (2010). Conditioned taste aversion enhances the survival of an endangered predator imperilled by a toxic invader. Journal of Applied Ecology 47, 558-565.

Oakwood, M. (1997) The ecology of the northern quoll, Dasyurus hallucatus. Ph.D thesis Thesis. (Australian National University, Canberra.)

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.

Oakwood, M. (2002). Spatial and social organization of a carnivorous marsupial Dasyurus hallucatus (Marsupialia: Dasyuridae). Journal of Zoology 257, 237-248.

Oakwood, M. (2004a). Death after sex. Biologist 51, 5-8.

Oakwood, M. (2004b). The effect of cane toads on a marsupial carnivore, the northern quoll. Dasyurus hallucatus.

Oakwood, M. (2004c). The effect of cane toads on a marsupial carnivore, the northern quoll, Dasyurus hallucatus. Parks Australia.

232

Oakwood, M., Bradley, A. J., and Cockburn, A. (2001). Semelparity in a large marsupial. Proceedings of the Royal Society of London. Series B: Biological Sciences 268, 407-411.

Oakwood, M. and Pritchard, D. (1999). Little evidence of toxoplasmosis in a declining species, the northern quoll (Dasyurus hallucatus). Wildlife research 26, 329-333.

Oakwood, M., Woinarski, J., and Burnett, S. (2016). Dasyurus hallucatus. The IUCN Red List of Threatened Species: Retrieved from https://www. iucnr edlist. org/speci es/6295/21947321.

Öckinger, E., Lindborg, R., Sjödin, N. E., and Bommarco, R. (2012). Landscape matrix modifies richness of plants and insects in grassland fragments. Ecography 35, 259-267.

Olds, L. G., Myers, C., Reside, J., Madani, G., Dudley, A., Potter, S., Martin, R., Boona, E., Waina, T., and Taggart, D. A. (2016). Small terrestrial mammals on Doongan Station, in the Northern Kimberley bioregion, Western Australia. Australian Mammalogy 38, 164-176.

Ondei, S., Prior, L. D., McGregor, H. W., Reid, A. M., Johnson, C. N., Vigilante, T., Goonack, C., Williams, D., and Bowman, D. M. (2020). Small mammal diversity is higher in infrequently compared with frequently burnt rainforest–savanna mosaics in the north Kimberley, Australia. Wildlife research.

Onodera, R., Akimoto, Y., Shimada, T., and Saitoh, T. (2017). Different population responses of three sympatric rodent species to acorn masting—the role of tannin tolerance. Population ecology 59, 29- 43.

Palmer, R. (2019). Predator Control Baiting and Monitoring Program, Yarraloola and Red Hill, Pilbara Region, Western Australia. 2016 Annual Report - Year 2. Department of Biodiversity, Conservation and Attractions. (Perth.)

Pateiro-López, B. and Rodrıguez-Casal, A. (2009). alphahull: Generalization of the Convex Hull of a Sample of Points in the Plane. R package version 0.2-0. URL http://cran. at. r-project. org/web/packages/alphahull 15782.

Paull, D. J., Claridge, A. W., and Barry, S. C. (2011). There’s no accounting for taste: bait attractants and infrared digital cameras for detecting small to medium ground-dwelling mammals. Wildlife research 38, 188-195.

Peacock, D. and Abbott, I. (2014). When the ‘native cat’would ‘plague’: historical hyperabundance in the quoll (Marsupialia: Dasyuridae) and an assessment of the role of disease, cats and foxes in its curtailment. Australian Journal of Zoology 62, 294-344.

Pearson, S., Baynes, A., and Triggs, B. (2001). The record of fauna, and accumulating agents of hair and bone, found in middens of stick-nest rats (Genus Leporillus)(Rodentia: Muridae). Wildlife research 28, 435-444.

Perry, J. J., Vanderduys, E. P., and Kutt, A. S. (2015). More famine than feast: pattern and variation in a potentially degenerating mammal fauna on Cape York Peninsula. Wildlife research 42, 475-487.

233

Phillip, A. (2015) 'The Voyage of Governor Phillip to Botany Bay.' (Cambridge University Press.)

Phillips, B. (2016). Genetic. Australasian Science 37, 14.

Phillips, B. and Shine, R. (2006). An invasive species induces rapid adaptive change in a native predator: cane toads and black snakes in Australia. Proceedings of the Royal Society B: Biological Sciences 273, 1545-1550.

Phillips, B., Shine, R., and Tingley, R. (2016). The genetic backburn: using rapid evolution to halt invasions. Proceedings of the Royal Society B: Biological Sciences 283, 20153037.

Phillips, S. (2011). A brief tutorial on Maxent.)

Phillips, S. J., Dudík, M., Elith, J., Graham, C. H., Lehmann, A., Leathwick, J., and Ferrier, S. (2009). Sample selection bias and presence‐only distribution models: implications for background and pseudo‐absence data. Ecological applications 19, 181-197.

Pianka, E. R. (2017) 'Ecology and natural history of desert lizards: analyses of the ecological niche and community structure.' (Princeton University Press.)

Pizzatto, L., Both, C., Brown, G., and Shine, R. (2017). The accelerating invasion: dispersal rates of cane toads at an invasion front compared to an already-colonized location. Evolutionary Ecology 31, 533-545.

Polis, G. A. and Myers, C. A. (1985). A survey of intraspecific predation among reptiles and amphibians. Journal of Herpetology 19, 99-107.

Pollock, A. (1999). Notes on status, distribution and diet of northern quoll Dasyurus hallucatus in the Mackay-Bowen area, mideastern Queensland. Australian Zoologist 31, 388-395.

Pollock, K. H. (1980) 'Capture-recapture models: a review of current methods, assumptions and experimental design.' (Citeseer: North Carolina 27650, USA.)

Pollock, K. H., Nichols, J. D., Simons, T. R., Farnsworth, G. L., Bailey, L. L., and Sauer, J. R. (2002). Large scale wildlife monitoring studies: statistical methods for design and analysis. Environmetrics: The official journal of the International Environmetrics Society 13, 105-119.

Potter, S., Eldridge, M. D., Taggart, D. A., and Cooper, S. J. (2012). Multiple biogeographical barriers identified across the monsoon tropics of northern Australia: Phylogeographic analysis of the brachyotis group of rock‐wallabies. Molecular Ecology 21, 2254-2269.

Potts, J. M. and Elith, J. (2006). Comparing species abundance models. Ecological modelling 199, 153-163.

234

Pradel, R. (1996). Utilization of capture-mark-recapture for the study of recruitment and population growth rate. Biometrics 52, 703-709.

Prugh, L. R., Hodges, K. E., Sinclair, A. R., and Brashares, J. S. (2008). Effect of habitat area and isolation on fragmented animal populations. Proceedings of the National Academy of Sciences 105, 20770-20775. doi: 10.1073/pnas.0806080105.

Pullin, A. S. and Knight, T. M. (2005). Assessing conservation management's evidence base: a survey of management‐plan compilers in the United Kingdom and Australia. Conservation biology 19, 1989- 1996.

Pulsford, S. A., Lindenmayer, D. B., and Driscoll, D. A. (2016). A succession of theories: purging redundancy from disturbance theory. Biological Reviews 91, 148-167.

Purvis, A., Gittleman, J. L., Cowlishaw, G., and Mace, G. M. (2000). Predicting extinction risk in declining species. Proceedings of the Royal Society of London B: Biological Sciences 267, 1947- 1952.

QGIS (2020). QGIS geographic information system. Open source geospatial Foundation project.

R Core Team, R. (2020). R: A language and environment for statistical computing. R Foundation for statistical computing, Vienna.

Rabe, M. J., Rosenstock, S. S., and deVos Jr, J. C. (2002). Review of big-game survey methods used by wildlife agencies of the western United States. Wildlife Society Bulletin, 46-52.

Radford, I. J. (2012). Threatened mammals become more predatory after small‐scale prescribed fires in a high‐rainfall rocky savanna. Austral Ecology 37, 926-935.

Radford, I. J. and Andersen, A. N. (2012). Effects of fire on grass‐layer savanna macroinvertebrates as key food resources for insectivorous vertebrates in northern Australia. Austral Ecology 37, 733- 742.

Radford, I. J., Dickman, C. R., Start, A. N., Palmer, C., Carnes, K., Everitt, C., Fairman, R., Graham, G., Partridge, T., and Thomson, A. (2014). Mammals of Australia's tropical savannas: A conceptual model of assemblage structure and regulatory factors in the Kimberley region. PloS one 9, e92341.

Radford, I. J., Gibson, L. A., Corey, B., Carnes, K., and Fairman, R. (2015). Influence of fire mosaics, habitat characteristics and cattle disturbance on mammals in fire-prone savanna landscapes of the northern Kimberley. PloS one 10, e0130721.

Radford, I. J., Woolley, L.-A., Corey, B., Vigilante, T., Hatherley, E., Fairman, R., Carnes, K., and Start, A. N. (2020). Prescribed burning benefits threatened mammals in northern Australia. Biodiversity and Conservation 29, 2985-3007.

235

Ralls, K., Sunnucks, P., Lacy, R. C., and Frankham, R. (2020). Genetic rescue: A critique of the evidence supports maximizing genetic diversity rather than minimizing the introduction of putatively harmful genetic variation. Biological conservation 251, 108784.

Ranges, K. L. (2017). Assessment of the hazard that the Hisstory® bait for feral cats presents to a non-target species; northern quoll (Dasyurus hallucatus).

Rankmore, B., Griffiths, A., Woinarski, J., Ganambarr, B., Taylor, R., Brennan, K., Firestone, K., and Cardoso, M. (2008). Island translocation of the northern quoll Dasyurus hallucatus as a conservation response to the spread of the cane toad Chaunus (Bufo) marinus in the Northern Territory, Australia. Report to the Australian Government’s Natural Heritage Trust. Department of Natural Resources, Environment and the Arts, Darwin.

Rankmore, B. R. (2006) 'Impacts of Habitat Fragmentation on the Vertebrate Fauna of the Tropical Savannas of Northern Australia; with Special Reference to Medium-sized Mamals.' (Charles Darwin University.)

Read, J. L. and Scoleri, V. (2015). Ecological implications of reptile mesopredator release in arid South Australia. Journal of Herpetology 49, 64-69.

Reconyx (2020). Reconyx PC900 Hyperfire Professional Covert Camera.)

Reily, S. J., DeGloria, S. D., and Elliot, R. A. (1999). Terrain Ruggedness Index That Quantifies Topographic Heterogeneity. Intermountain. Journal of Science 5, 23.

Reside, A. E., Briscoe, N. J., Dickman, C. R., Greenville, A. C., Hradsky, B. A., Kark, S., Kearney, M. R., Kutt, A. S., Nimmo, D. G., and Pavey, C. R. (2019). Persistence through tough times: fixed and shifting refuges in threatened species conservation. Biodiversity and Conservation, 1-28.

Rew-Duffy, M., Cameron, S. F., Freeman, N. J., Wheatley, R., Latimer, J. M., and Wilson, R. S. (2020). Greater agility increases probability of survival in the endangered northern quoll. Journal of Experimental Biology 223.

Richardson, E., Nimmo, D. G., Avitabile, S., Tworkowski, L., Watson, S. J., Welbourne, D., and Leonard, S. W. (2018). Camera traps and pitfalls: an evaluation of two methods for surveying reptiles in a semiarid ecosystem. Wildlife research 44, 637-647.

Ricketts, T. H. (2001). The matrix matters: effective isolation in fragmented landscapes. Am Nat 158, 87-99. doi: 10.1086/320863.

Ridout, M. S. and Linkie, M. (2009). Estimating overlap of daily activity patterns from camera trap data. Journal of Agricultural, Biological, and Environmental Statistics 14, 322-337.

Ringma, J., Legge, S., Woinarski, J., Radford, J., Wintle, B., and Bode, M. (2018). Australia’s mammal fauna requires a strategic and enhanced network of predator-free havens. Nature ecology & evolution 2, 410.

236

Ritchie, E. G. and Johnson, C. N. (2009). Predator interactions, mesopredator release and biodiversity conservation. Ecology Letters 12, 982-998.

Robson, D. and Regier, H. (1964). Sample size in Petersen mark–recapture experiments. Transactions of the American Fisheries Society 93, 215-226.

Rodda, G. H., Bock, B. C., Burghardt, G. M., and Rand, A. S. (1988). Techniques for identifying individual lizards at a distance reveal influences of handling. Copeia 1988, 905-913.

Rose, D. C., Amano, T., González-Varo, J. P., Mukherjee, N., Robertson, R. J., Simmons, B. I., Wauchope, H. S., and Sutherland, W. J. (2019). Calling for a new agenda for conservation science to create evidence-informed policy. Biological conservation 238, 108222.

Rostro-García, S., Kamler, J. F., Crouthers, R., Sopheak, K., Prum, S., In, V., Pin, C., Caragiulo, A., and Macdonald, D. W. (2018). An adaptable but threatened big cat: density, diet and prey selection of the Indochinese leopard (Panthera pardus delacouri) in eastern Cambodia. Royal Society open science 5, 171-187.

Rovero, F. and Zimmermann, F. (2016) 'Camera trapping for wildlife research.' (Pelagic Publishing Ltd.)

Rovero, F., Zimmermann, F., Berzi, D., and Meek, P. (2013). " Which camera trap type and how many do I need?" A review of camera features and study designs for a range of wildlife research applications. Hystrix 24, 148-156.

Rowat, D., Speed, C. W., Meekan, M. G., Gore, M. A., and Bradshaw, C. J. (2009). Population abundance and apparent survival of the vulnerable whale shark, Rhincodon typus, in the Seychelles aggregation. Oryx 43, 591-598.

Rowcliffe, J. M. and Carbone, C. (2008). Surveys using camera traps: are we looking to a brighter future? Animal conservation 11, 185-186.

Royle, J. A. (2004). N‐mixture models for estimating population size from spatially replicated counts. Biometrics 60, 108-115.

Russell-Smith, J., Yates, C., Edwards, A., Allan, G. E., Cook, G. D., Cooke, P., Craig, R., Heath, B., and Smith, R. (2003). Contemporary fire regimes of northern Australia, 1997–2001: change since Aboriginal occupancy, challenges for sustainable management. International Journal of Wildland Fire 12, 283-297.

Ruxmoore, R. and Groombridge, B. Asian Monitor Lizards: A Review of Distribution, Status, Exploitation, and Trade in Four Selected Species. 1990. (Secretariat of the Convention on International Trade in Endangered Species ….)

237

Salafsky, N., Boshoven, J., Burivalova, Z., Dubois, N. S., Gomez, A., Johnson, A., Lee, A., Margoluis, R., Morrison, J., and Muir, M. (2019). Defining and using evidence in conservation practice. Conservation Science and Practice 1, e27.

Sanecki, G. M. and Green, K. (2005). A technique for using hair tubes beneath the snowpack to detect winter-active small mammals in the subnivean space. European Journal of Wildlife Research 51, 41- 47.

Sannolo, M., Gatti, F., Mangiacotti, M., Scali, S., and Sacchi, R. (2016). Photo-identification in amphibian studies: a test of I3S Pattern. Acta Herpetologica 11, 63-68.

Saunders, G. R., Gentle, M. N., and Dickman, C. R. (2010). The impacts and management of foxes Vulpes vulpes in Australia. Mammal Review 40, 181-211.

Saupe, E. E., Qiao, H., Hendricks, J. R., Portell, R. W., Hunter, S. J., Soberón, J., and Lieberman, B. S. (2015). Niche breadth and geographic range size as determinants of species survival on geological time scales. Global Ecology and Biogeography 24, 1159-1169.

Scheele, B. C., Foster, C. N., Banks, S. C., and Lindenmayer, D. B. (2017). Niche contractions in declining species: mechanisms and consequences. Trends in Ecology & Evolution 32, 346-355.

Scheele, B. C., Legge, S., Armstrong, D., Copley, P., Robinson, N., Southwell, D., Westgate, M. J., and Lindenmayer, D. B. (2018). How to improve threatened species management: an Australian perspective. Journal of Environmental Management 223, 668-675.

Schmitt, L., Bradley, A., Kemper, C., Kitchener, D., Humphreys, W., and How, R. (1989). Ecology and physiology of the northern quoll, Dasyurus hallucatus (Marsupialia, Dasyuridae), at Mitchell Plateau, Kimberley, Western Australia. Journal of Zoology 217, 539-558.

Schwarz, C. J. and Arnason, A. N. (1996). A general methodology for the analysis of capture- recapture experiments in open populations. Biometrics 52, 860-873.

Sears, M. W., Angilletta, M. J., Schuler, M. S., Borchert, J., Dilliplane, K. F., Stegman, M., Rusch, T. W., and Mitchell, W. A. (2016). Configuration of the thermal landscape determines thermoregulatory performance of ectotherms. Proceedings of the National Academy of Sciences 113, 10595-10600.

Seber, G. A. F. (1982) 'The estimation of animal abundance and related parameters.' (Blackburn press Caldwell, New Jersey.)

Shine, B. R. (2010). The Ecological Impact of Invasive Cane Toads (Bufo Marinus) in Australia. The Quarterly Review of Biology 85, 253-291. doi: 10.1086/655116.

Shine, R. and Harlow, P. S. (1998). Ecological traits of commercially harvested water monitors, Varanus salvator, in northern Sumatra. Wildlife research 25, 437-447.

238

Short, J. and Smith, A. (1994). Mammal decline and recovery in Australia. Journal of Mammalogy 75, 288-297.

Sievert, C., Parmer, C., Hocking, T., Chamberlain, S., Ram, K., Corvellec, M., and Despouy, P. (2017). plotly: Create Interactive Web Graphics via’plotly. js’. R package version 4.

Silver, S. C., Ostro, L. E., Marsh, L. K., Maffei, L., Noss, A. J., Kelly, M. J., Wallace, R. B., Gomez, H., and Ayala, G. (2004). The use of camera traps for estimating jaguar Panthera onca abundance and density using capture/recapture analysis. Oryx 38, 148-154.

Slatyer, R. A., Hirst, M., and Sexton, J. P. (2013). Niche breadth predicts geographical range size: a general ecological pattern. Ecology Letters 16, 1104-1114.

Smith, J., Brook, B., Griffiths, A., and Thompson, G. G. (2007). Can morphometrics predict sex in varanids? Journal of Herpetology 41, 133-141.

Smith, J. K. and Coulson, G. (2012). A comparison of vertical and horizontal camera trap orientations for detection of potoroos and bandicoots. Australian Mammalogy 34, 196-201.

Southgate, R., Paltridge, R., Masters, P., and Carthew, S. (2007). Bilby distribution and fire: a test of alternative models of habitat suitability in the Tanami Desert, Australia. Ecography 30, 759-776.

Southwell, D., Tingley, R., Bode, M., Nicholson, E., and Phillips, B. L. (2017). Cost and feasibility of a barrier to halt the spread of invasive cane toads in arid Australia: incorporating expert knowledge into model‐based decision‐making. Journal of Applied Ecology 54, 216-224.

Speed, C. W., Meekan, M. G., and Bradshaw, C. J. (2007). Spot the match–wildlife photo- identification using information theory. Frontiers in zoology 4.

Spencer, P., How, R., Hillyer, M., Cook, A., Morris, K., Stevenson, C., and Umbrello, L. (2013). Genetic analysis of northern quolls from the Pilbara region of Western Australia. Murdoch University: Perth.

Start, A., Burbidge, A., McDowell, M., and McKenzie, N. (2012). The status of non-volant mammals along a rainfall gradient in the south-west Kimberley, Western Australia. Australian Mammalogy 34, 36-48.

Start, A., Burbidge, A., McKenzie, N., and Palmer, C. (2007). The status of mammals in the north Kimberley, Western Australia. Australian Mammalogy 29, 1-16.

Steffen, W. (2009) 'Australia's biodiversity and climate change.' (Csiro Publishing.)

Stevenson, R. (1985). Body size and limits to the daily range of body temperature in terrestrial ectotherms. The American Naturalist 125, 102-117.

239

Stobo-Wilson, A. M., Murphy, B. P., Cremona, T., and Carthew, S. M. (2019). Contrasting patterns of decline in two arboreal marsupials from Northern Australia. Biodiversity and Conservation 28, 2951- 2965.

Stobo‐Wilson, A. M., Stokeld, D., Einoder, L. D., Davies, H. F., Fisher, A., Hill, B. M., Mahney, T., Murphy, B. P., Stevens, A., and Woinarski, J. C. (2020). Habitat structural complexity explains patterns of feral cat and dingo occurrence in monsoonal Australia. Diversity and distributions 26, 832- 842.

Stokeld, D., Fisher, A., Gentles, T., Hill, B., Triggs, B., Woinarski, J. C., and Gillespie, G. R. (2018). What do predator diets tell us about mammal declines in Kakadu National Park? Wildlife research 45, 92-101.

Sutherland, D. R., Glen, A. S., and de Tores, P. J. (2010). Could controlling mammalian carnivores lead to mesopredator release of carnivorous reptiles? Proceedings of the Royal Society B: Biological Sciences 278, 641-648.

Taylor, B. D., Goldingay, R. L., and Lindsay, J. M. (2014). Horizontal or vertical? Camera trap orientations and recording modes for detecting potoroos, bandicoots and pademelons. Australian Mammalogy 36, 60-66.

Templeton, A. R., Brazeal, H., and Neuwald, J. L. (2011). The transition from isolated patches to a metapopulation in the eastern collared lizard in response to prescribed fires. Ecology 92, 1736-1747. doi: 10.1890/10-1994.1.

Thomas, C. D., Franco, A. M., and Hill, J. K. (2006). Range retractions and extinction in the face of climate warming. Trends in Ecology & Evolution 21, 415-416.

Thomas, O. (1926). LXVII.—The local races of Dasyurus hallucatus.

Thompson, G. G. and Withers, P. C. (1997). Comparative morphology of western Australian varanid lizards (Squamata: Varanidae). Journal of Morphology 233, 127-152.

Thompson, W. (2013) 'Sampling rare or elusive species: concepts, designs, and techniques for estimating population parameters.' (Island Press.)

Thomson, P. (1986). The effectiveness of aerial baiting for the control of dingoes in north-western Australia. Wildlife research 13, 165-176.

Thornton, D. H., Branch, L. C., and Sunquist, M. E. (2011). The influence of landscape, patch, and within-patch factors on species presence and abundance: a review of focal patch studies. Landscape Ecology 26, 7-18.

Tingley, R., Phillips, B., Letnic, M., Brown, G. P., Shine, R., and Baird, S. J. (2013). Identifying optimal barriers to halt the invasion of cane toads Rhinella marina in arid Australia. Journal of Applied Ecology 50, 129-137.

240

Tingley, R., Vallinoto, M., Sequeira, F., and Kearney, M. R. (2014). Realized niche shift during a global biological invasion. Proceedings of the National Academy of Sciences, 201405766.

Tingley, R., Ward-Fear, G., Schwarzkopf, L., Greenlees, M. J., Phillips, B. L., Brown, G., Clulow, S., Webb, J., Capon, R., and Sheppard, A. (2017). New weapons in the Toad Toolkit: a review of methods to control and mitigate the biodiversity impacts of invasive cane toads (Rhinella marina). The Quarterly Review of Biology 92, 123-149.

Trainor, C., Fisher, A., Woinarski, J., and Churchill, S. (2000). Multiscale patterns of habitat use by the Carpentarian rock-rat (Zyzomys palatalis) and the common rock-rat (Z. argurus). Wildlife research 27, 319-332.

Treilibs, C. E., Pavey, C. R., Hutchinson, M. N., and Bull, C. M. (2016). Photographic identification of individuals of a free‐ranging, small terrestrial vertebrate. Ecology and evolution 6, 800-809.

Trolliet, F., Vermeulen, C., Huynen, M.-C., and Hambuckers, A. (2014). Use of camera traps for wildlife studies: a review. Biotechnologie, Agronomie, Société et Environnement 18, 446-454.

TSSC (2005). Northern Quoll (Dasyurus hallucatus).)

Turpin, J., White, N., Dunlop, J., and Bamford, M. (2018). New populations of the black-flanked rock-wallaby (Petrogale lateralis) from the Little Sandy Desert and Murchison, Western Australia. Australian Mammalogy 40, 234-242.

Turpin, J. M. and Bamford, M. J. (2015). A new population of the northern quoll (Dasyurus hallucatus) on the edge of the Little Sandy Desert, Western Australia. Australian Mammalogy 37, 86- 91.

Twigg, L. E., Eldridge, S. R., Edwards, G. P., Shakeshaft, B. J., depreu, N. D., and Adams, N. (2000). The longevity and efficacy of 1080 meat baits used for dingo control in central Australia. Wildlife research 27, 473-481.

Twigg, L. E., Martin, G. R., Eastman, A. F., and Kirkpatrick, W. E. (2003). Sensitivity of some Australian animals to sodium fluoroacetate (1080): additional species and populations, and some ecological considerations. Australian Journal of Zoology 51, 515-531.

Tyne, J. A., Johnston, D. W., Christiansen, F., and Bejder, L. (2017). Temporally and spatially partitioned behaviours of spinner dolphins: implications for resilience to human disturbance. Royal Society open science 4, 160626.

Ujvari, B., Oakwood, M., and Madsen, T. (2013). Queensland northern quolls are not immune to cane toad toxin. Wildlife research 40, 228-231.

Umbrello, L. S. (2018) Evolution and diversification of dasyurid marsupials of the Australian arid zone. (PhD thesis, University of Western Australia.)

241

Urban, M. C. (2015). Accelerating extinction risk from climate change. Science 348, 571-573.

Urban, M. C., Phillips, B. L., Skelly, D. K., and Shine, R. (2008). A toad more traveled: the heterogeneous invasion dynamics of cane toads in Australia. The American Naturalist 171, E134- E148.

Vernes, K., Dennis, A., and Winter, J. (2001). Mammalian Diet and Broad Hunting Strategy of the Dingo (Canis familiaris dingo) in the Wet Tropical Rain Forests of Northeastern Australia 1. Biotropica 33, 339-345.

Viacava, P., Blomberg, S. P., Sansalone, G., Phillips, M. J., Guillerme, T., Cameron, S. F., Wilson, R. S., and Weisbecker, V. (2020). Skull shape of a widely distributed, endangered marsupial reveals little evidence of local adaptation between fragmented populations. Ecology and evolution 10, 9707-9720.

Villard, M. A. and Metzger, J. P. (2014). Beyond the fragmentation debate: a conceptual model to predict when habitat configuration really matters. Journal of Applied Ecology 51, 309-318.

Visconti, P., Pressey, R. L., Giorgini, D., Maiorano, L., Bakkenes, M., Boitani, L., Alkemade, R., Falcucci, A., Chiozza, F., and Rondinini, C. (2011). Future hotspots of terrestrial mammal loss. Philosophical Transactions of the Royal Society B: Biological Sciences 366, 2693-2702.

Volkov, I., Banavar, J. R., Hubbell, S. P., and Maritan, A. (2003). Neutral theory and relative species abundance in ecology. Nature 424, 1035-1037. von Takach, B., Scheele, B. C., Moore, H., Murphy, B. P., and Banks, S. C. (2020a). Patterns of niche contraction identify vital refuge areas for declining mammals. Diversity and distributions 26, 1467- 1482. von Takach, B., Scheele, B. C., Moore, H., Murphy, B. P., and Banks, S. C. (2020b). Patterns of niche contraction identify vital refuge areas for declining mammals. Diversity and distributions.

Ward, M. S., Simmonds, J. S., Reside, A. E., Watson, J. E., Rhodes, J. R., Possingham, H. P., Trezise, J., Fletcher, R., File, L., and Taylor, M. (2019). Lots of loss with little scrutiny: The attrition of habitat critical for threatened species in Australia. Conservation Science and Practice.

Watson, D. M. and Peterson, A. T. (1999). Determinants of diversity in a naturally fragmented landscape: humid montane forest avifaunas of Mesoamerica. Ecography 22, 582-589.

Wauters, L. A., Lurz, P. W., and Gurnell, J. (2000). Interspecific effects of grey squirrels (Sciurus carolinensis) on the space use and population demography of red squirrels (Sciurus vulgaris) in conifer plantations. Ecological Research 15, 271-284.

Webb, J., Legge, S., Tuft, K., Cremona, T., and Austin, C. (2015). Can we mitigate cane toad impacts on northern quolls?| Final report.

242

Weinstein, B. G. (2018). A computer vision for animal ecology. Journal of Animal Ecology 87, 533- 545.

Welbourne, D. (2013). A method for surveying diurnal terrestrial reptiles with passive infrared automatically triggered cameras. PloS one 6, e18965.

Welbourne, D., Claridge, A., Paull, D., and Ford, F. (2019). Improving Terrestrial Squamate Surveys with Camera-Trap Programming and Hardware Modifications. Animals 9, 388.

Welbourne, D. J., Claridge, A. W., Paull, D. J., and Ford, F. (2020). Camera-traps are a cost-effective method for surveying terrestrial squamates: A comparison with artificial refuges and pitfall traps. PloS one 15, e0226913.

Welbourne, D. J., Claridge, A. W., Paull, D. J., and Lambert, A. (2016). How do passive infrared triggered camera traps operate and why does it matter? Breaking down common misconceptions. Remote Sensing in Ecology and Conservation 2, 77-83.

Welbourne, D. J., MacGregor, C., Paull, D., and Lindenmayer, D. B. (2015). The effectiveness and cost of camera traps for surveying small reptiles and critical weight range mammals: a comparison with labour-intensive complementary methods. Wildlife research 42, 414-425.

Welbourne, D. J., Paull, D. J., Claridge, A. W., and Ford, F. (2017). A frontier in the use of camera traps: surveying terrestrial squamate assemblages. Remote Sensing in Ecology and Conservation 3, 133-145.

Whiteley, A. R., Fitzpatrick, S. W., Funk, W. C., and Tallmon, D. A. (2015). Genetic rescue to the rescue. Trends in Ecology & Evolution 30, 42-49.

Wickham, H. and Wickham, M. H. (2007). The ggplot package.)

Wildnet (2019). WildNet wildlife records - published - Queensland.)

Wilson, G. J. and Delahay, R. J. (2001). A review of methods to estimate the abundance of terrestrial carnivores using field signs and observation. Wildlife research 28, 151-164.

Wilson, S. K. and Swan, G. (2017) 'A complete guide to reptiles of Australia.' (Reed New Holland Sydney.)

Wintle, B. A., Cadenhead, N. C., Morgain, R. A., Legge, S. M., Bekessy, S. A., Cantele, M., Possingham, H. P., Watson, J. E., Maron, M., and Keith, D. A. (2019). Spending to save: What will it cost to halt Australia's extinction crisis? Conservation Letters 12, e12682.

Withers, P. (2000). Overview of granite outcrops in Western Australia. Journal of the Royal Society of Western Australia 83, 103.

243

Witmer, G. W. (2005). Wildlife population monitoring: some practical considerations. Wildlife research 32, 259-263.

Woinarski, J. (2000). The conservation status of rodents in the monsoonal tropics of the Northern Territory. Wildlife research 27, 421-435.

Woinarski, J., Armstrong, M., Brennan, K., Fisher, A., Griffiths, A., Hill, B., Milne, D., Palmer, C., Ward, S., and Watson, M. (2010a). Monitoring indicates rapid and severe decline of native small mammals in Kakadu National Park, northern Australia. Wildlife research 37, 116-126.

Woinarski, J., Braby, M., Burbidge, A., Coates, D., Garnett, S., Fensham, R., Legge, S., McKenzie, N., Silcock, J., and Murphy, B. (2019a). Reading the black book: The number, timing, distribution and causes of listed extinctions in Australia. Biological conservation 239, 108261.

Woinarski, J., Burbidge, A., and Harrison, P. (2014) 'The action plan for Australian mammals 2012.' (Commonwealth Scientific and Industrial Research Organization Publishing ….)

Woinarski, J., Milne, D., and 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., Oakwood, M., Winter, J., Burnett, S., Milne, D., Foster, P., Myles, H., and Holmes, B. (2008). Surviving the toads: patterns of persistence of the northern quoll Dasyurus hallucatus in Queensland. Report to The Australian Government’s Natural Heritage Trust.

Woinarski, J., Rankmore, B., Fisher, A., Brennan, K., and Milne, D. (2007). The natural occurrence of northern quolls Dasyurus hallucatus on islands of the Northern Territory: assessment of refuges from the threat posed by cane toads Bufo marinus. Report to Natural Heritage Trust, 1-40.

Woinarski, J., Risler, J., and 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., Ward, S., Mahney, T., Bradley, J., Brennan, K., Ziembicki, M., and Fisher, A. (2011a). The mammal fauna of the Sir Edward Pellew island group, Northern Territory, Australia: refuge and death-trap. Wildlife research 38, 307-322.

Woinarski, J., Watson, M., and Gambold, N. (2002). Vertebrate monitoring and resampling in Kakadu National Park. Parks and Wildlife Commission of the Northern Territory, Palmerston.

Woinarski, J. and 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. Department of Natural Resources, Environment, The Arts and Sport, Darwin.

Woinarski, J. C. (2015). Critical‐weight‐range marsupials in northern A ustralia are declining: a commentary on F isher et al.(2014)‘The current decline of tropical marsupials in A ustralia: is history repeating?’. Global Ecology and Biogeography 24, 118-122.

244

Woinarski, J. C., Armstrong, M., Brennan, K., Fisher, A., Griffiths, A. D., Hill, B., Milne, D., Palmer, C., Ward, S., and Watson, M. (2010b). Monitoring indicates rapid and severe decline of native small mammals in Kakadu National Park, northern Australia. Wildlife research 37, 116-126.

Woinarski, J. C., Braby, M., Burbidge, A. A., Coates, D., Garnett, S. T., Fensham, R. J., Legge, S., McKenzie, N. L., Silcock, J., and Murphy, B. P. (2019b). Reading the black book: the number, timing, distribution and causes of listed extinctions in Australia. Biological conservation 239, 108261.

Woinarski, J. C., Burbidge, A. A., and 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.

Woinarski, J. C., Legge, S., Fitzsimons, J. A., Traill, B. J., Burbidge, A. A., Fisher, A., Firth, R. S., Gordon, I. J., Griffiths, A. D., and Johnson, C. N. (2011b). The disappearing mammal fauna of northern Australia: context, cause, and response. Conservation Letters 4, 192-201.

Woinarski, J. C., Legge, S. M., and Dickman, C. R. (2019c) 'Cats in Australia: Companion and Killer.' (CSIRO Publishing.)

Woolley, P. A., Krajewski, C., and Westerman, M. (2015). Phylogenetic relationships within Dasyurus (Dasyuromorphia: Dasyuridae): quoll systematics based on molecular evidence and male characteristics. Journal of Mammalogy 96, 37-46.

Wynn, M. L., Clemente, C., Nasir, A. F. A. A., and Wilson, R. S. (2015). Running faster causes disaster: trade-offs between speed, manoeuvrability and motor control when running around corners in northern quolls (Dasyurus hallucatus). Journal of Experimental Biology 218, 433-439.

Zanette, L., Doyle, P., and Trémont, S. M. (2000). Food shortage in small fragments: evidence from an area‐sensitive passerine. Ecology 81, 1654-1666.

Zeng, H. and Wu, X. B. (2005). Utilities of edge-based metrics for studying landscape fragmentation. Computers, Environment and Urban Systems 29, 159-178.

Ziembicki, M., Woinarski, J., and Mackey, B. (2013). Evaluating the status of species using Indigenous knowledge: Novel evidence for major native mammal declines in northern Australia. Biological conservation 157, 78-92.

245