Daniel Oscar Hunter

Top predators can induce ecological state-shifts over large spatio-temporal scales in Australian forest ecosystems

Daniel Oscar Hunter

A thesis in fulfilment of the requirement for the degree of Doctor of Philosophy

School of Biological, Earth and Environmental Sciences University of New South Wales, Australia

November 2018

1 2 3 4 INCLUSION OF PUBLICATIONS STATEMENT

UNSW is supportive of candidates publishing their research results during their candidature as detailed in the UNSW Thesis Examination Procedure.

Publications can be used in their thesis in lieu of a Chapter if: • The student contributed greater than 50% of the content in the publication and is the “primary author”, ie. the student was responsible primarily for the planning, execution and preparation of the work for publication • The student has approval to include the publication in their thesis in lieu of a Chapter from their supervisor and Postgraduate Coordinator. • The publication is not subject to any obligations or contractual agreements with a third party that would constrain its inclusion in the thesis

Please indicate whether this thesis contains published material or not. This thesis contains no publications, either published or submitted for ☐ publication (if this box is checked, you may delete all the material on page 2)

Some of the work described in this thesis has been published and it has ☐ been documented in the relevant Chapters with acknowledgement (if this box is checked, you may delete all the material on page 2)

This thesis has publications (either published or submitted for publication) ☒ incorporated into it in lieu of a chapter and the details are presented below

CANDIDATE’S DECLARATION

I declare that:

• I have complied with the Thesis Examination Procedure • where I have used a publication in lieu of a Chapter, the listed publication(s) below meet(s) the requirements to be included in the thesis. Name Daniel Hunter Signature Date 28/07/18

5 Postgraduate Coordinator’s Declaration (to be filled in where publications are used in lieu of Chapters)

I declare that:

• the information below is accurate • where listed publication(s) have been used in lieu of Chapter(s), their use complies with the Thesis Examination Procedure • the minimum requirements for the format of the thesis have been met. PGC’s Name PGC’s Signature Date (dd/mm/yy)

6 For each publication incorporated into the thesis in lieu of a Chapter, provide all of the requested details and signatures required

Details of publication #1: Full title: Not all predators are equal: a continent-scale analysis of the effects of predator control on Australian mammals Authors: Daniel Hunter, Malgorzata Lagisz, Viyanna Leo, Shinichi Nakagawa, Mike Letnic Journal or book name: Mammal Review Volume/page numbers: 48: 108-122 Date accepted/ published: 28 November 2017 Status Published x Accepted and In In progress press (submitted) The Candidate’s Contribution to the Work Study was conceptualised by Daniel Hunter and Mike Letnic. Data collection was performed by Daniel Hunter and Viyanna Leo. Data analysis were performed by Daniel Hunter, Malgorzata Lagisz, Shinichi Nakagawa and Mike Letnic. Daniel Hunter and Mike Letnic wrote the manuscript. Malgorzata Lagisz contributed to writing the results section. Location of the work in the thesis and/or how the work is incorporated in the thesis: This publication forms all of chapter 2 in my thesis. The publication forms the foundation for the rest of my thesis. Primary Supervisor’s Declaration I declare that: • the information above is accurate • this has been discussed with the PGC and it is agreed that this publication can be included in this thesis in lieu of a Chapter • All of the co-authors of the publication have reviewed the above information and have agreed to its veracity by signing a ‘Co-Author Authorisation’ form. Supervisor’s name Supervisor’s signature Date 31/07/2018 Mike Letnic

Details of publication #2: Full title: Reintroducing Tasmanian devils to mainland Australia can restore top-down control in ecosystems where dingoes have been extirpated. Authors: Daniel Hunter, Thomas Britz, Menna Jones, Mike Letnic

7 Journal or book name: Biological Conservation Volume/page numbers: 191: 428-435 Date accepted/ published: 24 July 2015 Status Published x Accepted and In In progress press (submitted) The Candidate’s Contribution to the Work Study was conceptualised by Daniel Hunter and Mike Letnic. Data collection was performed by Daniel Hunter. Data analysis were performed by Daniel Hunter, Thomas Britz and Mike Letnic. Daniel Hunter and Mike Letnic wrote the manuscript. Menna Jones and Mike Letnic provided guidance on manuscript structure as co-author and supervisor respectively. Location of the work in the thesis and/or how the work is incorporated in the thesis: This publication forms all of chapter 5. This work is pitched as a hypothetical solution to restoring top predator ecological function in areas where dingoes remain persecuted and/or functionally extinct. Primary Supervisor’s Declaration I declare that: • the information above is accurate • this has been discussed with the PGC and it is agreed that this publication can be included in this thesis in lieu of a Chapter • All of the co-authors of the publication have reviewed the above information and have agreed to its veracity by signing a ‘Co-Author Authorisation’ form. Supervisor’s name Supervisor’s signature Date 31/07/2018 Mike Letnic

Add further boxes if required. You may delete any boxes not used

8 Abstract

Top predators are recognised as important ecological constituents because empirical studies demonstrate that they can supress populations of herbivores and smaller predators. Despite their important roles, top predators are subject to human-induced population declines globally. The main drivers of decline include habitat loss, habitat modification and human persecution resulting from conflict for resources such as livestock. One outcome of their declines has been a widespread interest in documenting the ensuing ecological effects owing to the fact that top predators are recognised as keystone species. In terrestrial ecosystems, research has shown that top predators can influence ecosystems and their species composition owing to direct effects on smaller predators and their prey species. These direct effects have been linked to far-reaching indirect effects at multiple trophic levels, that extend to vegetation and even the physical attributes of the landscape. The role of top predators is increasingly being viewed as an important component of healthy and functional ecosystems around the world.

This thesis explores the hypothesis that Australia’s top terrestrial predator, the dingo

(Canis dingo), induces shifts in mammal populations and vegetation in Australian forest ecosystems. The novelty of this thesis is that I test this hypothesis using longitudinal methodologies through time as opposed to snap-shot surveys. In my first chapter, I provide a detailed overview of the current state of knowledge regarding the role of top predators around the world. Specifically, I detail how top predators drive trophic cascades via direct and indirect effect pathways. My second chapter contains comprehensive meta-analyses which investigate the effects of dingo and fox control on mammal species. This provides a robust foundation for identifying knowledge

9 gaps. My third chapter builds on the meta-analysis by conducting a multi-year occupancy analysis in forest ecosystems to determine how dingo control affects the occupancy of invasive mesopredators and native mammals. In chapter 4, I perform generalised linear modelling and quantile regression to improve our understanding of how dingoes supress foxes and to determine if environmental predictors have any bearing on this interaction. Finally, I test if reintroducing a marsupial predator back to mainland Australia can act as a surrogate for the dingo in some parts by restoring some lost ecological interactions to areas where dingoes have been extirpated.

10 Acknowledgements

I acknowledge the Traditional Owners of the land on which I conducted all of my field work and I pay my respects to their Elders, past and present and future, and any

Elders from any community who may read this thesis.

I would like to sincerely thank my primary supervisor Associate Professor Mike

Letnic. Your patience and lateral thinking has made me a better scientist. I have learned so much from you and I honestly couldn’t have asked for a more brilliant and creative mind to help guide my learning. Thanks Mike, I hope I have provided a decent return on investment!

I’m very grateful to Dr. Rosalie Chapple and the Blue Mountains World Heritage

Institute. Rosalie helped me find this PhD and the BMWHI provided me with a top-up scholarship which greatly helped me to focus on my research. Thank you very much

Rosalie and I hope the research in this thesis helps realise some of your organisation’s goals.

Thanks to the UNSW, the School of Biological, Earth and Environmental Sciences and the Centre for Ecosystem Science for their support during my candidature. The

UNSW is a world-class research institution and I feel privileged to have undertaken my PhD here. I would especially like to thank Professor Richard Kingsford, Sharon

Ryall and Jono Russell – all of you have helped me along my extended PhD journey in many ways. I’d especially like to thank you for your patience and support as I juggled my PhD and my professional work as a filmmaker – permitting me space and time has helped me realise my ultimate goal of making wildlife and conservation films for the BBC, National Geographic, Plimsoll and the ABC – so thank you very

11 much. A huge thanks to all my lab mates who helped make my transition to Sydney from Jan Juc a smooth one. Thanks also to all of the volunteers that helped me during the fieldwork phase of my PhD.

My friends and family are the most important part of my life. My closest friends and family have all supported me in one way or another and I am thankful to all of you. I would especially like to thank my mum and my dad, Karen and Jeff Hunter. Thank you also to my brother, Jordan Hunter. Finally, thanks to my dog, Luna. No single being spent more time at my side throughout this whole process than my furry little friend.

I dedicate this thesis to my late grandma, Margaret May Watson. I wouldn’t have even been able to attend university if it wasn’t for your help – this one is for you,

Gran!

12 Preface

My thesis contains four chapters (chapters 2 to 5) that have been published or are in review in peer-reviewed academic journals. Each chapter is a standalone manuscript.

There is one reference list for all chapters at the end of this thesis. Animal ethics permits were obtained for all of the research conducted in this thesis in accordance with Australian laws under Animal Research Authority: University of New South

Wales.

This thesis is a compilation of my own work, with guidance from my primary supervisor Associate Professor Mike Letnic (UNSW). All chapters were conceptualised jointly with my primary supervisor, Mike Letnic. I wrote the manuscripts and conducted the majority of the data analysis. See below for specific contributions for each chapter.

Chapter 2: Hunter DO. Lagisz M. Nakagawa S. Leo V. Letnic M. (2018) Not all predators are equal: a continent-scale analysis of the effects of predator control on

Australian mammals. Mammal Review 48: 108-122.

Study was conceptualised by Daniel Hunter and Mike Letnic. Data collection was performed by Daniel Hunter and Viyanna Leo. Data analyses were performed by

Daniel Hunter, Malgorzata Lagisz, Shinichi Nakagawa and Mike Letnic. Daniel

Hunter and Mike Letnic wrote the manuscript. Malgorzata Lagisz contributed to writing the results section.

Chapter 3: Top predator removal affects mammal species differently at multiple trophic levels in forest ecosystems.

13 Study was conceptualised by Daniel Hunter and Mike Letnic. Data collection was performed by Daniel Hunter. Data analyses were performed by Daniel Hunter, Ross

Goldingay and Mike Letnic. Ross Goldingay and Mike Letnic provided guidance on manuscript structure as co-author and supervisor respectively.

Chapter 4: Activity indices of a top predator and mesopredator respond in opposite ways to predator control programs.

Study was conceptualised by Daniel Hunter and Mike Letnic. Data collection was performed by Daniel Hunter, James Rees and Nick Colman. Data analyses were performed by Daniel Hunter. Daniel Hunter and Mike Letnic wrote the manuscript.

Chapter 5: Hunter DO. Britz T. Jones M. Letnic M. (2015). Reintroduction of

Tasmanian devils to mainland Australia can restore top-down control in ecosystems where dingoes have been extirpated. Biological Conservation 191, 428–435.

Study was conceptualised by Daniel Hunter and Mike Letnic. Data collection was performed by Daniel Hunter. Data analyes were performed by Daniel Hunter, Thomas

Britz and Mike Letnic. Menna Jones and Mike Letnic provided guidance on manuscript structure as co-author and supervisor respectively.

14 Table of Contents

Abstract ...... 9 Acknowledgements ...... 11 Preface ...... 13 Table of Contents ...... 15 List of Figures ...... 18 List of Tables ...... 21

1. General Introduction ...... 24 1.1 Mammal conservation in Australia ...... 25 1.2 The dingo and its ecology ...... 26 1.3. Human-dingo conflict ...... 31 1.4 Vacant research niches and thesis objectives ...... 33

2. Not all predators are equal: a continent-scale analysis of the effects of predator control on Australian mammals...... 36 2.1 Abstract ...... 37 2.2 Introduction ...... 38 2.3 Methods ...... 42 2.3.1 Literature search ...... 42 2.3.2 Study selection and eligibility criteria ...... 43 2.3.3 Data extraction and coding ...... 43 2.3.4 Statistical analysis ...... 44 2.3.5 Sensitivity analyses ...... 46 2.3.6 Publication bias ...... 47 2.4 Results ...... 49 2.4.1 Effects of dingo removal on the abundance of mammals ...... 52 2.4.2 Effects of fox removal on the abundance of mammals ...... 56 2.4.3 Publication bias ...... 60 2.5 Discussion ...... 62 2.6 Supplementary material ...... 67 2.7 Acknowledgments ...... 71

15 3. Top predator removal affects mammal species differently at multiple trophic levels in forest ecosystems ...... 72 3.1 Abstract ...... 73 3.2 Introduction ...... 74 3.3 Methods ...... 76 3.3.1 Study area ...... 76 3.3.2 Camera survey design ...... 77 3.3.3 Occupancy analysis ...... 78 3.3.4 Small mammal abundance ...... 79 3.4 Results ...... 81 3.4.1 Occupancy modelling ...... 81 3.4.2 Small mammals ...... 85 3.5 Discussion ...... 88

4. Activity indices of a top-predator and mesopredator respond in opposite ways to predator control programs ...... 93 4.1 Abstract ...... 94 4.2 Introduction ...... 95 4.3 Methods ...... 99 4.3.1 Study sites ...... 99 4.3.2 Sampling ...... 101 4.3.3 Predictor variables ...... 102 4.3.4 Statistical analyses ...... 104 4.4 Results ...... 105 4.4.1 Quantile regression ...... 105 4.4.2 Correlations between predictor variables ...... 106 4.4.3 Generalised linear modelling ...... 106 4.5 Discussion ...... 109 4.6 Supplementary material ...... 114

5. Reintroducing Tasmanian devils to mainland Australia can restore top-down control in ecosystems where dingoes have been extirpated ...... 117 5.1 Abstract ...... 118 5.2 Introduction ...... 119

16 5.3 Methods ...... 124 5.3.1 Reintroduction viability using species distribution modelling (SDM) ...... 124 5.3.2 The fuzzy cognitive model (FCM) ...... 125 5.3.3 The agents (species and vegetation) ...... 128 5.3.4 Data extraction ...... 128 5.3.5 Model scenarios ...... 131 5.4 Results ...... 134 5.4.1 Habitat suitability of devils on mainland ...... 134 5.4.2 FCM model ...... 134 5.4.3 FCM scenarios ...... 134 5.5 Discussion ...... 136 5.5.1 An ecological case for reintroducing devils to mainland Australia ...... 140 5.6 Supplementary Material ...... 142 5.7 Acknowledgements ...... 148

6. General Discussion ...... 149 6.1 Synthesis ...... 150 6.2 Chapter 2: Not all predators are equal: a continent-scale analysis of the effects of predator control on Australian mammals...... 150 6.2.1 Limitations and recommendations ...... 152 6.3 Chapter 3: Top predator removal affects mammal species differently across multiple trophic levels in forest ecosystems...... 154 6.3.1 Limitations and recommendations ...... 155 6.4 Chapter 4: Activity indices of an top predator and mesopredator respond in opposite ways to predator control programs...... 157 6.4.1 Limitations and recommendations ...... 158 6.5 Chapter 5: Reintroduction of Tasmanian devils to mainland Australia can restore top-down control in ecosystems where dingoes have been extirpated ...... 160 6.5.1 Limitations and recommendations ...... 161 6.6 Conclusion ...... 163 Literature cited ...... 166

17 List of Figures

Figure 2.1 Effects of dingo removal (A) and fox removal (B) experiments, for all species excluding target species, and for native species only. Overall estimates represent results from meta-analytic models and phylogenetic meta-analytical models (phylo). Full models are phylogenetic meta-regressions with moderators added as fixed effects. Point estimates represent mean intercepts, unless slope is indicated in the brackets. Whiskers represent 95% Confidence Intervals. Stars indicate estimates that are significantly different from zero (95% Confidence Intervals not spanning zero)...... 50

Figure 2.2. Effect sizes plotted against animal weights for dingo removal (A) and fox removal (B) experiments. A – when dingoes were removed, effect sizes are negative for light species and positive for heavy species, indicating decreased abundance of the former and increased abundance of the latter. B – when foxes were removed, there is no clear linear relationship between effect sizes and body weight, the overall positive effect is mainly driven by species of intermediate weight, indicating a positive effect of fox removal on their abundance. Sizes of the circles are related to the precision of effect sizes, smaller circles bear less weight in the analyses. All species apart from the target species are included in the datasets...... 51

Figure 2.3 Effects of dingo removal (A) and fox removal (B) experiments by species, including target species. Phylogenetic trees of relationships among the species are annotated with shapes representing whether each species is native (circles) or non-native to Australia (squares). The size of each shape is proportional to the natural logarithm of species body weight. Forest plots represent mean estimates for each species, along with 95% Confidence Intervals. Stars indicate estimates that are significantly different from zero (95% Confidence Intervals not spanning zero). k – number of effect sizes...... 52

Figure 2.4 Funnel plots used to estimate publication bias in the dingo and fox removal data sets. A – effect size estimates from the dingo removal dataset

18 plotted against their precision, B – residual effect sizes from the intercept model (phylogenetic meta-analysis) for the dingo removal dataset, C – effect size estimates from the fox removal dataset plotted against their precision, D – residual effect sizes from the intercept model (phylogenetic meta-analysis) for the fox removal dataset. Dashed vertical line indicates no effect...... 62

Supplementary Figure 1. PRISMA diagram showing the process of discovery and elimination of publications for the fox removal dataset. N – number of papers. 67

Supplementary Figure 2. PRISMA diagram showing the process of discovery and elimination of publications for the fox removal dataset. N – number of papers. 68

Supplementary Figure 3. Dingo distribution in Australia. Open circles indicate fox study locations and crosses indicate dingo study locations. In some instances, where multiple sites were within <50km, only a single point is used for clarity. Two markers close together indicate the same (or similar, <50km) study site was used but in a different publication...... 69

Figure 3.1 Total average dasyurid capture rate for each treatment. Total average is derived from the number of captures per grid divided by the number of nights traps were out at each grid...... 86

Figure 3.2 Total average bush rat (R. fuscipes) capture rate for each treatment. Total average is derived from the number of captures per grid divided by the number of nights traps were out at each grid...... 86

Figure 4.1 (a) the relationship between dingo and fox activity from Johnson & VanDerWal 2009, the solid line indicates the regression line for the 0.9 quantile (P = 0.02) (b) relationship between dingo and fox activity from Letnic et al. 2011, the solid line indicates the regression line for the 0.9 quantile (P = <0.05) (c) Scatterplot of our field data showing the relationship between quantile regression lines fitted to the 0.95 (P = <0.001) and 0.9 (P = 0.043) quantiles, open circles represent unbaited sites and closed circles are baited sites...... 98

Figure 4.2. Solid points depict site locations used for this study...... 100

Figure 4.3 Average activity indices (± 1 SE) for both dingoes and foxes in unbaited and baited study sites. Results from t-tests indicate that there is a statistically

19 significant difference (<0.05) between dingo and fox activity at unbaited sites...... 108

Figure 5.1 a) Species distribution model of potential Tasmanian devil (Sarcophilus harrisii) distribution on the mainland under the current climate scenario, b) distribution of the dingo (Canis dingo) and hybrids (Canis dingo × Canis familiaris) in Australia. Modified from Letnic et al. (2011)...... 125

Figure 5.2 The activation function f...... 127

Figure 5.3 Percentage shifts in abundance, expressed as the proportional change, for each agent under all scenarios. More extreme shifts are equivalent to strong cascading effects. Dark grey bars represent the initial capped abundance value. Black bars represent Tasmanian devil abundance...... 133

Figure 5.4 Predicted abundance for agents under each of the 5 scenarios. White bars are the status quo (scenario 1), black bars = dingo and fox eradication (scenario 2), light grey bars = no canid control with devils reintroduced (scenario 3), dark grey bars = dingo eradication (scenario 4) and freckled bars = dingo eradication with devils reintroduced (scenario 5)...... 137

20 List of Tables

Table 2.1 Studies included in the meta-analysis. For each study, the target predator removal species and number of effect sizes extracted per study is provided...... 48

Table 2.2 Meta-analytic and meta-regression models of effects of dingo removal on other species (native and non-native), excluding dingo. M – point estimates (mean), CI.lb – 95% Confidence Interval lower bounds, CI.ub – 95% Confidence Interval upper bounds, I2 – total heterogeneities for each model. Estimates represent intercepts, unless slope is specified in the brackets. Estimates that are significantly different from zero (95% Confidence Intervals not spanning zero) are highlighted in bold...... 52

Table 2.3 Meta-analytic and meta-regression models of effects of dingo removal on native species, excluding dingo. M – point estimates (mean), CI.lb – 95% Confidence Interval lower bounds, CI.ub – 95% Confidence Interval upper bounds, I2 – total heterogeneities for each model. Estimates represent intercepts, unless slope is specified in the brackets. Estimates that are significantly different from zero (95% Confidence Intervals not spanning zero) are highlighted in bold...... 54

Table 2.4 Meta-analytic and meta-regression models of effects of dingo removal on all species, including dingo. M – point estimates (mean) for intercepts, CI.lb – 95% Confidence Interval lower bounds, CI.ub – 95% Confidence Interval upper bounds, I2 – total heterogeneities for each model. Estimates that are significantly different from zero (95% Confidence Intervals not spanning zero) are highlighted in bold...... 55

Table 2.5 Meta-analytic and meta-regression models of effects of fox removal on other species (native and non-native), excluding fox. M – point estimates (mean), CI.lb – 95% Confidence Interval lower bounds, CI.ub – 95% Confidence Interval upper bounds, I2 – total heterogeneities for each model. Estimates represent intercepts, unless slope is specified in the brackets. Estimates that are significantly different from zero (95% Confidence Intervals not spanning zero) are highlighted in bold...... 57

21 Table 2.6 Meta-analytic and Meta-regression models on effects of fox removal on native species. M – point estimates (mean), CI.lb – 95% Confidence Interval lower bounds, CI.ub – 95% Confidence Interval upper bounds, I2 – total heterogeneities for each model. Estimates represent intercepts, unless slope is specified in the brackets. Estimates that are significantly different from zero (95% Confidence Intervals not spanning zero) are highlighted in bold...... 59

Table 2.7 Meta-analytic and meta-regression models of effects of fox removal on all species, including fox. M – point estimates (mean) for intercepts, CI.lb – 95% Confidence Interval lower bounds, CI.ub – 95% Confidence Interval upper bounds, I2 – total heterogeneities for each model. Estimates that are significantly different from zero (95% Confidence Intervals not spanning zero) are highlighted in bold...... 60

Supplementary Table 1. Excluded full-text studies – fox removal dataset ...... 70

Table 3.1 Model outputs for single-season occupancy models for each species...... 84

Table 3.2 The best occupancy model for each species is shown and the corresponding occupancy values are included...... 85

Table 3.3 F values for generalized linear mixed effects models investigating the effects that baiting, occasion and location had on the abundances of dasyurids and rodents...... 88

Table 4.1 Candidate generalised linear models analyzing the competing variables for dingo activity. Models were analyzed using a negative binomial distribution. The best fitting model and those with <2 AIC units of the top model are indicated with the dotted line. Models containing both baiting and distance to freehold were omitted as these parameters are correlated...... 107

Table 4.2. Candidate generalised linear models analysing the competing variables for fox activity. Models were analysed using a negative binomial distribution. The best fitting model and those with <2 AIC units of the top model are indicated with the dotted line. Models containing both baiting and distance to freehold were omitted as these parameters are correlated...... 109

22 Supplementary Table 1 Description of important information relating to each site used in the analysis. G – ground, T – trapping, A – aerial and S – shooting. .... 114

Supplementary Table 2 Top competing models describing dingo and fox activity. Mean values (95% CIs) are shown for all parameters...... 116

Table 5.1 Interaction matrix populated with interaction strengths (r) ...... 130

Table 5.2 Output values from FCM displayed here as percentage shift from null state (scenario 1)...... 135

Supplementary Table 1 Fox literature cited to populate FCM ...... 143

Supplementary Table 2 Dingo literature cited to populate FCM ...... 144

Supplementary Table 3 Devil literature cited to populate FCM ...... 145

Error! Bookmark not defined.

Supplementary Table 4 Diet data derived for Tasmanian devil used to populate FCM ...... 146

Error! Bookmark not defined.

Supplementary Table 5 Justification and reference information for inferred values ...... 147

23 1. General Introduction

24 1.1 Mammal conservation in Australia

In recent times, Australia’s mammal fauna has experienced more extinctions than any other continent on Earth (Bilney et al. 2010, Woinarski et al. 2015). Since the arrival of Europeans to Australia over the last ~200 years, 29 terrestrial mammal species have become extinct and 56 more are currently facing extinction (Woinarski et al.

2014). It is not clear in every instance what caused each extinction event but often a significant amount of evidence suggests that introduced mesopredators the red fox and feral cat are to blame (Johnson 2006). Key evidence supporting the hypothesis that foxes and cats are the primary drivers of mammal extinctions is that on some offshore islands where foxes and/or cats have not colonised, many small and medium- sized native mammal species remain extant despite being completely or locally extinct on the mainland (Kinnear et al. 2002). Furthermore, intensive and long-term lethal control programs have shown that targeting these invasive mesopredators can cause native mammal communities to recover (Dexter et al. 2007, Robley et al. 2014).

Cats are distributed throughout the Australian mainland and on the island of Tasmania

(Doherty et al. 2015) and foxes are found in all parts of Australia excluding the northern tropics (Dickman 1996). Given the widespread distribution of both invasive predators, and the naivety of Australian mammal fauna to these predators (Carthey &

Banks 2012, Steindler et al. 2018), the threat they pose to Australia’s small and medium-sized native mammal species is ongoing (Woinarski et al. 2015).

In Australia, considerable resources are spent on lethal control programs for both foxes and cats in an attempt to mitigate further extinctions (Doherty & Ritchie 2016).

However, the assumption that lethal control programs always benefit biodiversity has been challenged on the basis that the efficacy of these programs is not always

25 monitored (Reddiex et al. 2007). Further, lethal control raises important concerns such as the ethical treatment of animals and unintended ecological outcomes (Doherty &

Ritchie 2016). In order to benefit native mammal conservation, one possible solution to these problems in Australia could be to explore the application of biological control using top predators (Letnic et al. 2012b).

1.2 The dingo and its ecology

Australia’s largest terrestrial mammalian predator is the dingo (Canis dingo, body mass 11-22kg) (Letnic et al. 2014). Dingoes are relatively recent arrivals to Australia

(3500 – 5000 yBP) and mitochondrial DNA (mDNA) studies have revealed that dingoes diverged from their Asian ancestors <5000 yBP (Letnic et al. 2014). The geographical proximity of Asia (especially south-east Asia) and the recent timing of this divergence event lends support to the theory that humans transported dingoes to

Australia from New Guinea (McNiven & Hitchcock 2004) or from the Indonesian

Archipelago (Corbett 1995).

After their arrival to Australia, dingoes would have had to compete for resources with marsupial carnivores the thylacine (Thylacinus cynocephalus) and the Tasmanian devil (Sarcophilus harrisii). However, the prevalence of the dingo across almost all of the Australian mainland, and the subsequent extirpation of both thylacines and

Tasmanian devils from the mainland around the approximate time of dingo arrival, lends support to the hypothesis that dingoes were able to outcompete these marsupial carnivores (Letnic et al. 2012a). However, recent analysis has revealed that a severe drought event from 3000-5000 ybp coincided with the decline of devil populations

(Brüniche-Olsen et al. 2014). Thus it is plausible that extreme drought conditions may

26 have pushed devils to low population levels such that a “predator pit” (Sinclair et al.

1998) ensued which may have exacerbated the effects of human and dingo predation on devils.

Today, dingoes are the top mammalian predator in Australian ecosystems and have subsequently been shown to play a keystone role within these ecosystems (Colman et al. 2014, Johnson et al. 2007, Letnic et al. 2009a). Field studies from both arid and forested areas support this theory and provide important evidence that suggest dingoes shape and maintain ecological processes through predator-prey and predator-predator dynamics (Colman et al. 2014, Gordon et al. 2015, Letnic et al. 2012b).

Current ecological theory describes the important role that top predators, like dingoes, play in maintaining healthy and resilient ecosystems (Ripple et al. 2014). The effects of top predators on their prey can often be conspicuous and these effects manifest from the direct killing of their prey species (Glen et al. 2007). However, more subtle, non-lethal effects can be observed on prey and competitors through fear, for example, which can ultimately alter the prey or competitor’s behaviour or physiology (Kuijper et al. 2013, Ripple & Beschta 2004). These suppressive, top-down effects on the prey species often have cascading effects which ripple down through the food web affecting multiple trophic guilds (Ripple et al. 2014). Effects of this nature, whereby predator-prey effects influence more than one trophic level within a food web are known as trophic cascades (Pace et al. 1999).

Trophic cascade theory states that the removal of a top predator can result in the irruption of herbivores and a subsequent reduction in vegetation biomass and complexity (Beschta & Ripple 2009, Newsome & Ripple 2014, Ripple & Beschta

2011). In the case of the dingo, it has been shown that its removal from the ecosystem

27 can increase the abundance of herbivores and this, in turn, decreases the amount and complexity of the vegetation due to increased grazing pressure (Letnic et al. 2017,

Morris & Letnic 2017). Thus, the maintenance of dingo populations is shown to benefit vegetation via the trophic cascade mechanism.

Trophic cascades have now been documented in all of the world’s major biomes, both aquatic and terrestrial (Estes et al. 2011). Most of the research investigating these effects has been focussed on cascades which are induced by top predators (Ripple et al. 2014). The most well-known study of this process are cascades initiated by grey wolves in North America. Research shows that cervids, the main prey of the wolf, were on average almost six times more abundant in areas where wolves had been removed (Ripple & Beschta 2012). Wolves were, and continue to be, heavily persecuted in North America and the irruptions and contractions of cervid populations following wolf extirpation and their subsequent reintroductions induced changes in the composition of vegetation mostly resulting from differences in browsing activity from cervids (Beschta & Ripple 2009).

However, trophic cascades can also manifest from the bottom-up. Known as bottom- up cascades (Kagata & Ohgushi 2006), these occur when abiotic factors such as changes in nutrient supply results in similar changes at all other trophic levels. For example, rainfall places a limit on the primary productivity of an ecosystem which, in turn, constrains populations of herbivores and, ultimately, predators (Letnic et al.

2005). Additionally, bottom-up cascades may also affect predators by influencing the structure of habitat in which both the prey and predator exist (Ritchie & Johnson

2009). Understanding how these bottom-up cascades affect top predators has received attention mainly in European systems (Elmhagen & Rushton 2007) but more research is needed to understand the nature of this phenomenon in Australian systems.

28 A related concept to trophic cascade theory is the mesopredator release hypothesis

(MRH) (Prugh et al. 2009). A mesopredator is an intermediate sized predator occupying the trophic level below the top predator. Some examples of mesopredators are wild dogs in Africa (Creel & Creel 1996), coyotes in North America (Berger &

Gese, 2007) and red foxes in Australia (Johnson & VanDerWal 2009). It is important to note that a mesopredator isn’t defined by size per se but, rather, as being the middle-ranked predator in the system (Prugh et al. 2009). Thus, a mesopredator can become a top predator if a larger predator is removed.

The MRH proposes that removal or absence of a top predator from a system can alter mesopredator abundance by releasing it from suppression by the larger, more dominant top predator. The release from suppression of the mesopredator allows it to increase in abundance and activity (Prugh et al. 2009, Soulé et al. 1988). Importantly, a mesopredator that finds itself as the new top predator following the removal of the original top predator isn’t always ecologically similar to the previous top predator

(Prugh et al. 2009). For example, in many parts of Australia where dingoes have been extirpated foxes have become the top predator and their effects on native species, especially those in the critical weight range, are more detrimental and widespread than those of the dingo (Hunter et al. 2018). The subsequent population declines of mesopredator prey species following mesopredator release has been observed in many other instances in Australia and abroad (Crooks & Soulé 1999, Gordon et al. 2017,

Ripple et al. 2013, Terborgh et al. 2001).

Prugh et al. (2009) describes how the removal of top predators from ecosystems as a result of natural extirpation or anthropogenic persecution can simplify diversity and structure of ecosystems. A seminal paper providing support for the mesopredator release (Crooks & Soule 1999), describes how in the absence of coyotes, domestic

29 cats increased and this subsequently reduced the abundance of small scrub-breeding birds in southern California. One interesting aspect of this study is that it was carried out in a fragmented system on the edge of suburbia. Fragmented habitats have since been described as areas where mesopredator outbreaks are most common (Prugh et al.

2009).

Increased abundances and outbreaks of mesopredators in fragmented systems is likely attributed to three main reasons (Prugh et al. 2009). The first being that fragmented systems are often affiliated with excess resources available for mesopredators such as pet food, rubbish, livestock prey, crops and crop pests (Crooks & Soulé 1999). The possible reason for this affiliation between mesopredators and fragmented landscapes is that top predators require large home ranges, larger than those required by mesopredators, and so fragmented habitats rarely facilitate them (McNab 1963).

Additionally, the presence of top predators can instigate conflict with humans in areas of fragmentation because these areas are created by humans for residential, industrial or agricultural purposes (Woodroffe et al. 2005). Top predators are inevitably the overall losers from conflict between humans (Bruskotter 2013, Treves & Karanth

2003, Treves et al. 2007, Van Eeden et al. 2018). The result is often an increase in abundance of mesopredators resulting from the top predator decline and the ecosystem is subsequently destabilised and dramatic shifts in plant and animal assemblages often occur.

Ecosystem shifts following top predator removal are likely to continue as extirpation and persecution of top predators continues all around the world. In light of this, conservation of top predators is now a widely researched topic globally and many researchers agree that mitigating further top predator declines will help to maintain ecosystem integrity and this includes the maintenance of mesopredator suppression so

30 excessive predation of their prey does not occur (Ripple et al. 2014). However, the maintenance and reintroduction of top predator populations doesn’t come without consequences.

1.3. Human-dingo conflict

Thus far, evidence demonstrating the important role of dingoes in maintaining the structure and function of Australian ecosystems has done little to prevent their continued persecution (Fleming et al. 2004). As such, the maintenance or indeed reintroduction of dingo populations in many areas is deemed untenable for some livestock owners because controversy over the impact of dingoes on livestock production in Australia has persisted since the introduction of European agriculture at the end of the 18th century (Allen & West 2013, Allen & Sparkes 2001). The main driver of controversy is because dingoes have enormous detrimental impacts on the sheep farming industry (Allen & West 2013). However, in depth studies suggest that dingoes have no measurable effect on calf loss in arid rangeland and tropical savannah habitats (Allen 2014, Campbell 2018) and indeed a cost benefit study suggests that dingoes could have positive effects on cattle producers by reducing the impacts that kangaroos have on pastures (Prowse et al. 2015).

The threat of dingoes to sheep production was deemed so great that a 5631km long dingo-proof fence was constructed in the 20th century to keep dingoes out of

Australia’s prime sheep grazing country in the east. Dingoes remaining inside this fence continued to face intense persecution in the form of poison baiting, shooting and trapping. Today, poison baiting dingoes with 6 mg of the toxin sodium fluroacetate, commonly known as 1080 is the status-quo dingo management practice (Fleming et

31 al. 2001). Lethal control campaigns against dingoes, especially inside the fence, has meant that they have been completely extirpated from many areas in South Australia,

Victoria, New South Wales and Queensland (Letnic et al. 2012b).

The conflicting economic (Allen & Sparkes 2001, Forsyth et al. 2014, Prowse et al.

2015) and ecological (Allen et al. 2013, Colman et al. 2014, Glen et al. 2007) outcomes reported to result from dingo control ensure that the dingo remains one of the most divisive animals in Australia. On one hand, a growing body of observations demonstrate that dingoes structure ecological communities and are positively associated with the abundance of many species of small and medium-sized native mammals (Colman et al. 2014, Robertshaw & Harden 1986, Newsome et al. 1983).

On the other hand, the maintenance of dingo populations is perceived to be at odds with livestock production because dingoes predate on livestock. It has been shown that the functional extinction, and in many places the actual extinction of dingoes from areas can result in a shift in ecosystem states (Colman et al. 2014). One negative outcome that has been reported from these shifts in ecological states is an increase in populations of introduced predators and overabundant grazing species (Hunter et al.

2018) which has been linked to declines of Australia’s most vulnerable small and medium-sized native mammal species.

Novel solutions are required in order to appease both livestock producers and conservation managers simultaneously. A solution would need to not only provide protection to livestock owners from stock losses while, at the same time, allowing dingoes to persist at ecologically functional densities in the landscape in order for their reported ecological benefits to manifest.

32 1.4 Vacant research niches and thesis objectives

Previous studies report that removal and absence of dingoes leads to ecosystems states dominated by invasive mesopredators and overabundant grazing mammals which appear to be linked to declines of small and medium-sized native mammal species of conservation concern (Caughley et al. 1980, Newsome et al. 2001, Pople et al. 2000,

Wallach et al. 2010). However, some research reports different outcomes of dingo removal, particularly following their removal from lethal control programs (Allen et al. 2013, Allen et al. 2014). For instance, studies investigating the impacts of canid predator control on prey species in Australia have reported unexpected outcomes, such as an increase in the abundance of cats or certain prey species, due presumably to the existence of interference effects between predators or because of differences in the prey preferences of predator species (Hunter et al. 2018).

To date no studies have carried out meta-analyses to improve our understanding of the effects of both dingoes and foxes in both arid and mesic Australian systems. This is imperative for conservation managers because in order to improve conservation programs they require unbiased and clear information about the effects of removal of both dingoes and foxes given that canid baiting with 1080 poison is the most widespread management technique used across the continent and that despite claims that baiting with 1080 is a targeted control measure, both dingoes and foxes are affected regardless of the purported target canid.

In chapter 2, I conduct a meta-analysis of the effects that removal of dingoes and foxes has on Australian mammal species with the purpose of improving the information on the effects and “side-effects” of baiting programs that target these predators. A shortcoming of meta-analyses in this context is that they generally

33 cannot provide useful information about competition between dingoes and foxes.

Given that previous studies show that there is considerable dietary overlap between dingoes and foxes it is imperative to parse out their respective impacts on each other whilst taking into account environmental factors such as the productivity of the ecosystem, distance to freehold land and the presence of 1080 baiting. Indeed, previous studies have shown that dingoes do limit the abundance or supress the activity of foxes at smaller spatial scales (Johnson & VanDerWal 2009) but it is not known how environmental factors influence this interaction. Foxes are implicated in the endangerment of many native Australian animals, especially mammals (Johnson

2006, Woinarski et al. 2015), it is therefore imperative that research aims to improve management outcomes by describing how dingoes may supress foxes whilst incorporating the effects of different environmental factors into the interaction, and I address this question in a later chapter.

In my third chapter, I conduct a multi-year study investigating the role of dingoes at multiple trophic levels at two forest National Parks in New South Wales, on the east coast of Australia. I use occupancy analysis informed by camera trap data and generalised linear mixed models informed by Elliot trap surveys to search for an effect of dingoes on various mammal species.

In chapter 4, I build on my previous chapter and apply a different methodology to investigate a potential relationship between dingoes and foxes. I use track plot surveys to inform quantile regression analysis to determine the nature of the relationship between the activity indices of both dingoes and foxes. Then, I applied generalised linear models to my data to see if and how environmental variables such as season, distance to freehold land and baiting affected fox and dingo activity indices. If it can be shown that dingoes can mitigate the effects of foxes, which may occur via

34 exploitation competition or interference competition for example, it could therefore be beneficial for biodiversity conservation to allow dingoes to persist in the landscape. It may even be viable from a conservation point of view to reintroduce them into areas where they have been extirpated.

However, these options are currently untenable owing to the dingoes’ negative effects on livestock, especially sheep (Allen & West 2013). One possible solution to this issue may be to fill the vacant top predator niche with the Tasmanian devil. While devils, owing to their smaller body size and slower metabolism (Dawson & Hulburt

1970), are unlikely to occupy exactly the same niche as the dingo, theory and evidence from Tasmania where they remain extant suggests that devils have similar positive effects on small and medium-sized mammal species as dingoes do (Hollings et al. 2014). Despite speculation about a mainland devil reintroduction being a potential solution to help mitigate effects of invasive mesopredators on native mammal species (Ritchie et al. 2012), no research exists to support the theory.

I test the idea that reintroducing Tasmanian devils back to mainland Australia can restore some missing ecological interactions associated with top predator extirpation in chapter 5. I performed species distribution modelling (SDM) to see if suitable climatic conditions for devils do exist on the mainland. Second, I ran customised fuzzy cognitive models to predict the effects of a devil introduction on other species and understorey vegetation. As a side note, a possible mainland devil reintroduction also has the added benefit of repatriating a large and ultimately self-sustaining population of devil facial tumour disease-free animals back to the Australian mainland.

35 2. Not all predators are equal: a continent- scale analysis of the effects of predator control on Australian mammals.

36 2.1 Abstract

Introduced predators pose threats to biodiversity and are implicated in the extinction of many native species. In Australia, considerable effort is spent controlling populations of introduced predators, including the dingo Canis dingo and the red fox

Vulpes vulpes, in order to reduce their effects on native species and livestock. Studies describe different outcomes of controlling dingo and fox populations on native species, making biodiversity management decisions difficult for conservation managers.

We conduct a meta-analysis to compare the impacts that control programs targeted towards dingoes and foxes in Australia have on introduced predators and on other mammal species, including native species and prey species.

Our results provide evidence that lethal control of dingoes and foxes has different outcomes for different mammalian species. Dingo removal had a negative effect on the abundance of native mammals weighing less than the critical weight range (CWR) of 30 – 5500 g, and a positive effect on the abundance of mammals above the CWR.

Fox abundance increased in response to dingo control, but confidence intervals were large. Fox removal had strong positive effects on ground-dwelling and arboreal mammals. Lethal control of dingoes did not have a significant effect on cats, but where dingoes were removed there was a tendency for foxes to increase, and where foxes were removed there was a tendency for cats to increase.

Our results highlight unintended and perverse outcomes of lethal predator control on

Australian mammals. Lethal control of dingoes significantly increases abundances of above CWR mammals and significantly decreases abundances of under CWR

37 mammals. Lethal control of foxes significantly increases the abundances of CWR mammals. These findings show how removing dingoes and foxes alters mammal assemblages, and provide comprehensive and objective information for conservation managers.

Based on the results of this study we recommend that land management agencies take into consideration these results when planning lethal control programs targeting dingoes and foxes because depending on the target canid these programs result in different outcomes for other mammal species. Removal programs targeting dingoes and/or foxes need to be aware that this can result in increased abundances of introduced predators and, ultimately, have far-reaching effects on many mammal species.

2.2 Introduction

Globally, introduced predators rank as one of the greatest threats to biodiversity (King

1984, Savidge 1987, Biggins et al. 1999, Johnson 2006, Doherty et al. 2016).

Introduced predators have greater impacts on prey species than native predators, and in some cases have driven prey species to extinction because prey species with which they have not coevolved may lack appropriate adaptations to detect and escape novel predators, and thus are particularly vulnerable to predation (Salo et al. 2007, Carthey

& Banks 2012). Another reason put forward to explain the severe impacts of introduced predators is that they may thrive in their new environments and thus occur at remarkably high population densities (Moseby et al. 2015, Legge et al. 2017).

There are several potential drivers of high population densities of introduced predators. These include release from constraints on population growth posed by

38 competitors and diseases, as well as facilitation that can occur when populations of introduced predators benefit from the presence of high densities of prey (MacDonald

& Harrington 2003, Saunders et al. 2010, Sih et al. 2010, Letnic et al. 2012). High rates of encounters between prey and over-abundant introduced predators can have catastrophic effects on prey populations (Sinclair et al. 1998).

In many regions, multiple introduced species coexist (Ruscoe et al. 2011, Woinarski et al. 2015). In such circumstances it can be difficult to parse out the relative importance of their effects on prey species (Ruscoe et al. 2011, Wayne et al. 2017).

Coexisting introduced predator species may have additive effects on native prey populations if they have similar prey preferences. However, as a general rule, the impacts that predators have on prey species tends to scale with both the body sizes of the predator and prey species (Sinclair et al. 2003, Letnic et al. 2009a). This is because smaller predators tend to prefer smaller prey than larger predators and vice versa. Thus, we might expect that, in environments where multiple species of introduced predators coexist, their impacts on prey species vary according to their prey preference.

Another factor that can moderate the strength of introduced predators’ effects on prey species is the presence of competitive interactions between species. Interactions between coexisting predators may be particularly strong owing to competition for food and, as has been demonstrated in earlier studies, larger predators frequently kill smaller predators (Ripple et al. 2014), suppressing their abundances (Donadio &

Buskirk 2006). According to the mesopredator release hypothesis, larger predators can provide a net benefit for populations of the prey of smaller predators

(mesopredators) if intra-guild killing and interference competitions results in a decrease in the predatory impact of the mesopredator (Letnic et al. 2009a, Read &

39 Scoleri 2015). Such indirect effects can ripple along multiple interaction pathways, and, in doing so, can have profound effects on the composition of species assemblages (Colman et al. 2014).

In order to mitigate the threats that introduced predators pose to native species and livestock, introduced predators in many parts of the world are subjected to population control programs (Reynolds & Tapper 1996, Gillies & Pierce 1999, Robley et al.

2014). Such control programs may use selective techniques to remove introduced predators, such as shooting or trapping (Holbrook et al. 2016), or non-selective techniques, such as fencing and the distribution of poison baits (Miller & Miller 1995,

Hayward & Kerley 2009). However, the impacts that predator removal programs have on prey species can be difficult to assess in situations where multiple predators coexist (Ruscoe et al. 2011, Marlow et al. 2015a, Wayne et al. 2017), due to differences in the susceptibility of predator species to the control techniques employed, differences in the prey preferences of predator species (Letnic et al. 2009a) and indirect effects (Ruscoe et al. 2011, Colman et al. 2014).

In Australia, considerable effort is spent on controlling populations of introduced predators: the dingo Canis dingo, the red fox Vulpes vulpes and the feral cat Felis catus. Dingoes (15-25 kg) were introduced to Australia 3500-5000 yBP (years before present), and are likely to have contributed to the extinction of the thylacine

Thylacinus cynocephalus and the Tasmanian devil Sarcophilus harrisii from mainland

Australia (Letnic et al. 2012, Crowther et al. 2014). Because dingoes kill livestock, their populations are controlled, primarily by distributing meat baits containing the poison 1080 sodium fluroacetate (Claridge et al. 2010, Allen 2015). Red foxes (5-7

40 kg) were introduced to Australia around 1870 and are thought to be one of the major drivers of the endangerment of native mammals weighing 30 – 5500 g, also known as the critical weight range (CWR), and ground-nesting birds throughout the continent

(Johnson 2006). Like dingoes, foxes are controlled using poisoned 1080 baits to protect native fauna and livestock (Gentle et al. 2007). Dingo and fox populations are also controlled in some areas by using a combination of baiting, shooting, trapping and exclusion fencing. Cats were introduced to Australia in the late 18th century and have been implicated in the endangerment of mammals weighing less than 3000 g

(Fisher et al. 2014). Cat populations are difficult to control because they do not readily take most types of meat bait (Algar et al. 2002), but cats have been eradicated from relatively small areas by using a combination of exclusion fencing, shooting, trapping and poisoning (Moseby et al. 2009).

Because both dingoes and foxes readily take meat baits, many control programs target both species under the assumption that the impacts of the two predators on prey populations are similar (Claridge et al. 2010, Allen et al. 2013). However, studies investigating the impacts of canid predator control on prey species in Australia have reported unexpected outcomes, such as an increase in the abundance of cats or of certain prey species, due presumably to the existence of interference effects between predators or because of differences in the prey preferences of predator species (Risbey et al. 2000, Colman et al. 2014, Marlow et al. 2015a).

We conduct meta-analyses to compare the impacts that control programs targeted towards dingoes and red foxes in Australia have had on introduced predators and on other mammal species. We refer to the control programmes as fox/dingo removal. Our specific aims were: (1) to determine the direction and magnitude of the effect that the lethal control of dingoes (mainly via baiting) has on abundance indices of mammals

41 within three weight ranges: under CWR (<30 g), CWR (30 g-5500 g), and above

CWR (>5500 g), (2) to measure direction and magnitude of the effect that lethal control of dingoes has on abundance indices of dingoes, foxes and cats, (3) to measure the effects that lethal control of foxes has on abundance indices of mammals within the three weight ranges, and (4) to measure the direction and magnitude of the effect that lethal control of foxes has on abundance indices of foxes and cats. The findings are used to make recommendations for predator control based on a knowledge of how it alters mammal assemblages, with the goal of improving the information available to conservation managers.

2.3 Methods

2.3.1 Literature search

We followed the review approach as outlined in the PRISMA statement as far as possible for our study methodology (Moher et al. 2009). Our study search was conducted on ISI Web of Science and Google Scholar using the keywords ‘red fox baiting Australia’, ‘red fox predation Australia’, ‘dingo baiting’ and ‘dingo predation’

(see Supplementary Figures 1 and 2). Based on the results from these searches, we were able to use citations to trace back to other appropriate studies. We only included studies that were published in peer-reviewed journals and that included quantitative data from field-based surveys. Database searches for studies pertaining to foxes yielded 196 studies, searches for dingoes yielded 154 studies. From these we performed backward searches accordingly. Studies were initially collected up until 1

June 2015 and a search update was conducted in July 2017, following the same

42 procedures. A list of studies excluded at the full-text assessment stage is provided (see

Supplementary Table 1).

2.3.2 Study selection and eligibility criteria

The eligibility criteria were applied to studies retrieved by independent searches by two of the study authors. For the dingo and fox datasets, we initially included data for analysis if the study passed a set of criteria: (1) the study must have involved the attempted suppression or removal of dingoes or foxes (according to which dataset was being populated) and the response of mammalian species (often prey species) to predator removal must have been quantified, (2) studies must have involved before/after or paired control/treatment experimental designs, and (3) dingoes and foxes must have been removed predominantly by using 1080 sodium fluroacetate poison baiting or have been naturally absent from the treatment area (i.e. the treatment area was an island, see Kinnear et al. 2002). We included all terrestrial mammal species where we could extract an effect regardless of whether it was introduced or native.

Studies were classified according to the experimental design used by the researchers

(before/after or control/treatment). Studies that included data for the same species, but at different study sites, were treated as independent datasets. Wherever possible we requested additional information about studies from the study authors.

2.3.3 Data extraction and coding

43 Two authors independently extracted data and the extracted data were checked and discussed by all five authors. We used electronic callipers and also GraphClick

(Arizona Software, Switzerland) to extract relevant effect sizes from published studies reporting effects of dingo or fox baiting on other mammal species or on the target predator itself. Researchers usually reported measures of the abundance or activity of the animal of interest at both the removal and non-removal sites (or before/after removal). If multiple values were reported for the same species, we collected all values and then took the mean of these values for removal and for non-removal. We initially calculated the mean, standard deviation and sample size from data for each species under removal and non-removal conditions. We used log response ratio (lnRR, the natural logarithm of the ratio between the two means) as our effect size, because it is a more suitable measure of effect than Hedges’ d when sample sizes are small

(Friedrich et al. 2008, Hedges et al. 1999).

We also collected other information from the studies to use as moderators of potential heterogeneity in the data: (1) publication year, (2) taxonomic information (species, genus, family), (3) sample size of the control and treatment groups, (4) type of mammal (placental, marsupial or monotreme), mean animal body weight (kg,

Menkhorst & Knight 2001) and weight category (under CWR, CWR, above CWR),

(5) ecological type (herbivore, arboreal, mesopredator, ground-dwelling), (6) what type of data were reported in the study (index of abundance or true count data), and

(7) experiment type (control/treatment or before/after).

2.3.4 Statistical analysis

We performed all statistical analyses within R statistical software, version 3.2.4 (R

Studio version 0.98.5, http://www.rstudio.com). For meta-analytical and meta-

44 regression models, we used the R package metafor (Viechtbauer 2010), for phylogenetic tree construction and plotting, we used the R package ape (Paradis et al.

2004). We created a phylogenetic tree for the mammalian species included in the dataset, basing it on the mammalian supertree (Bininda-Emonds et al. 2007). We used study identity as a random factor in multilevel meta-analysis to control for non- independence arising from multiple effect sizes coming from single studies. We also accounted for species non-independence (via shared evolutionary history) by running phylogenetic meta-analytical and meta-regression models (Hadfield & Nakagawa

2010).

To quantify the overall effects of dingo removal on other species of mammal, we first constructed meta-analytic models (intercept-only, with and without phylogeny). We then ran phylogenetic meta-regression that included species average body weight and ecological type (herbivore, arboreal, mesopredator, or ground-dwelling) as moderators. Effects of fox removal on other species were assessed with analogous models: two meta-analytic (intercept-only models, with and without phylogeny) models and a phylogenetic meta-regression model with species mean body weight and ecological type (herbivore, arboreal, mesopredator, or ground-dwelling) as moderators. Body weight of animals was log-transformed and z-transformed before the analyses, so it had a mean of 0 and a standard deviation of 1. Positive estimates of the slope of body weight effect can be interpreted as increased abundance indicators with increasing body weight of species.

We quantified overall heterogeneity for the meta-analytic models using modified I2 statistics (total variance excluding sampling error variance divided by total variance,

Nakagawa & Santos 2012). Values of I2 above 75% are considered as high levels of heterogeneity (Higgins et al. 2003) and suggest that most of the variability across

45 studies is due to true heterogeneity rather than sampling noise, thus warranting examination of the potential sources of such heterogeneity.

We report mean effect sizes as our point meta-analytic estimates and 95% Confidence

Intervals (CI) of the estimated meta-analytical effects. We considered the point effect estimates statistically significant when their CI did not cross zero.

2.3.5 Sensitivity analyses

To extend our analyses and test robustness of our conclusions, we performed additional analyses on both dingo and fox removal datasets. For the dingo removal experiments we first looked at the native species (excluding dingo) data subset: we ran meta-analysis, phylogenetic meta-analysis, and then meta-regression with animal weight category (above CWR, CWR, under CWR). We then used data from all species, including dingo, to perform meta-analysis (with and without phylogeny) and a meta-regression using species identity as a categorical predictor. Such “species model” allows the identification of the most distinct species-specific effects and also allows the assessment of the influence of dingo removal experiments on the abundance indicators of dingo populations. We performed analogous analyses on the data from fox removal studies. We first created a data subset including only native species. For this data subset, we assessed two meta-analytical models and a meta- regression with animal weight category (CWR, under CWR). We then performed meta-analysis on the full dataset, including fox, and we also constructed a species meta-regression model using species identity as a moderator.

2.3.6 Publication bias

Typical publication bias arises when non-significant results are missing from the collected data due to studies reporting statistically non-significant results being less likely to be published (Rothstein et al. 2006). We assessed publication bias in dingo and fox removal datasets using three methods: (1) visual inspection of symmetry of funnel plots for the raw data and for the residual effect sizes from the meta-analytical models, (2) Egger’s regression (Egger et al. 1997), and (3) trim-and-fill analyses

(Duval & Tweedie 2000) on ‘meta-analytic’ residuals, as described by Nakagawa and

Santos (2012), and as implemented in the trimfill function in metafor R package.

47 Table 2.1 Studies included in the meta-analysis. For each study, the target predator removal species and number of effect sizes extracted per study is provided.

No. Paper reference Removal Number of Effect Sizes 1 Allen 2015 targetDingo species 3 2 Allen et al. 2013 Dingo 3 3 Allen et al. 2014 Dingo 5 4 Brook et al. 2012 Dingo 2 5 Burrows et al. 2003 Dingo 3 6 Caughley et al. 1980 Dingo 2 7 Coleman et al. 2014 Dingo 9 8 Gordon et al. 2015 Dingo 3 9 Gordon et al. 2017 Dingo 3 10 Letnic and Dworjanyn 2011 Dingo 3 11 Letnic et al. 2009 Dingo 8 12 Letnic et al. 2017 Dingo 2 13 Morris and Letnic 2017 Dingo 2 14 Newsome et al. 2001 Dingo 6 15 Pople et al. 2000 Dingo 1 16 Robertshaw and Harden 1986 Dingo 5 17 Wallach et al. 2010 Dingo 11 18 Banks 1999 Fox 1 19 Banks 2000 Fox 1 20 Coates 2008 Fox 4 21 Davey et al. 2006 Fox 5 22 Dexter and Murray 2009 Fox 5 23 Dexter et al. 2007 Fox 7 24 Kinnear et al. 1988 Fox 1 25 Kinnear et al. 1998 Fox 1 26 Kinnear et al. 2002 Fox 8 27 Kovacs et al. 2012 Fox 1 28 Marlow et al. 2015a Fox 2 29 Marlow et al. 2015b Fox 1 30 Molsher al. 2017 Fox 2 31 Pickett et al. 2005 Fox 1 32 Risbey et al. 2000 Fox 1 33 Robley et al. 2014 Fox 4 34 Towerton et al. 2011 Fox 9

48 2.4 Results

We extracted 125 effect sizes from 34 papers published between 1980 and 2017. This dataset was split into dingo removal and fox removal datasets, according to the target removal species of each paper. These two datasets were analysed independently.

The dingo removal dataset comprised 56 effect sizes, reporting effects on 22 species

(additionally there were 15 effect sizes for the dingo itself). There were similar numbers of effect sizes for native species (31 effect sizes) and non-native species (25 effect sizes), and for marsupials and monotremes (26 effect sizes) and placental species (30 effect sizes). The included species represented a broad range of average body weights, spanning 0.015 to 150 kg (mean ± SD: 20.5 ± 30.6). When the animals were placed in the three weight categories in relation to CWR, there were almost equal numbers of effect sizes for the above CWR (26 effect sizes) and CWR (23 effect sizes) categories, and only 7 effect sizes in the under CWR category.

The fox removal dataset comprised 47 effect sizes. The effects of fox removal were reported for a total of 21 different species (additionally, there were 7 effect sizes for the fox itself). In contrast to the dingo dataset, most effect sizes were for the native species (37 effect sizes), and only a few effect sizes came from the non-native species

(10 effect sizes, including the only mesopredator in the dataset – the cat). The dataset was dominated by marsupial species (33 out of 46 effect sizes). Also, the included species represented a narrower range of body weights than those in the dingo dataset, spanning 0.039 to 51.5 kg, and including lighter animals on average (mean ± SD: 7.2

±12.3). Thus, the dataset was dominated by CWR species (36 effect sizes), and only

11 effect sizes came from above CWR species. There were no species from the under

CWR category. The results (parameter estimates) of the meta-analytic and meta-

49 regression models for dingo removal and fox removal studies are presented in Figs 2.1

– 2.3 and Tables 2.2 – 2.7.

Figure 2.3 Effects of dingo removal (A) and fox removal (B) experiments by species, including target species. Phylogenetic trees of relationships among the species are annotated with shapes representing whether each species is native (circles) or non- native to Australia (squares). The size of each shape is proportional to the natural logarithm of species body weight. Forest plots represent mean estimates for each species, along with 95% Confidence Intervals. Stars indicate estimates that are significantly different from zero (95% Confidence Intervals not spanning zero). k – number of effect sizes.

2.4.1 Effects of dingo removal on the abundance of mammals

Dingo removal, overall, had negligible effect on other species of mammal (lnRRoverall

= -0.049, 95% CI = -0.594 to 0.496, lnRRoverall(phylo) = -0.023, 95% CI = -0.795 to

0.750, Fig. 2.1A, Table 2.2). We observed high total heterogeneity in the meta-

2 analytic models (with and without phylogeny I total = 98.6%, Table 2.2), which justified exploration of potential moderators with our meta-regression approach.

52 unless slope is specified in the brackets. Estimates that are significantly different from zero (95% Confidence Intervals not spanning zero) are highlighted in bold.

Model M CI.lb CI.ub I2 [%]

Meta-analytic mean -0.049 -0.594 0.496 98.6

Phylogenetic meta-analytic mean -0.023 -0.795 0.750 98.6

Meta-regression with animal weight Animal weight (slope) 0.856 0.167 1.545 and ecological type category: Arboreal -0.225 -1.741 1.291 Ground-dwelling 0.246 -0.905 1.397 Herbivore -0.149 -1.442 1.144 Mesopredator 0.455 -0.622 1.531

In our main meta-regression analysis on the dingo dataset, we investigated potential effects of animal weight and ecological type, representing the main ecological groups and key dingo competitors. Heavier species tended to respond to dingo removal by increasing their abundance, while lighter species tended to decline in abundance

(ßanimal weight = 0.946, 95% CI = -0.016 to 1.908, Fig. 2.1A, Fig. 2.2, Table 2.2). We found no evidence that ecological types, as defined in this study (herbivore, arboreal, mesopredator, ground-dwelling), were related to the magnitude of effects of dingo removal (Table 2.2). Particularly, there was no statistically significant effect on cat and fox abundance indices. However, foxes (lnRRVulpes_vulpes = 0.68, 95% CI = -0.271 to 1.632) did show a positive increase in abundance following dingo removal, which is evidenced as the third highest positive response after red kangaroos Macropus rufus and short-beaked echidnas Tachyglossus aculeatus.

Additional analyses performed on native species data subset (excluding dingo) revealed a very similar picture to that from the dataset including all species: there was no overall impact of dingo removal on native species (lnRRoverall = -0.004, 95% CI = -

53 0.996 to 0.987, lnRRoverall(phylo) = -0.003, 95% CI = -1.296 to 1.291, Fig. 2.1A, Table

2.3).

Model M CI.lb CI.ub I2 [%]

Meta-analytic mean -0.004 -0.996 0.987 99.0

Phylogenetic meta-analytic mean -0.003 -1.296 1.291 99.1

Meta-regression with animal weight category: Above CWR 1.658 0.525 2.791 CWR -0.007 -1.202 1.187 Under CWR -1.413 -2.716 -0.110

However, the effect on native species was again strongly influenced by species body weight, indicating opposite responses of lighter and heavier species in the dataset.

Indeed, in meta-regression models, where we used animal weight categories as a predictor, we revealed a large decline of under CWR species (lnRRunderCWR = -1.413,

95% CI = -2.716 to -0.110, Fig. 2.1A, Table 2.3) and also large increase of above

CWR species (lnRRaboveCWR = 1.658, 95% CI = 0.525 to 2.791, Fig. 1A, Table 3) in response to dingo removal. There was a large difference in mean effect magnitude between under and above CWR animals (lnRRaboveCWR-underCWR = -3.071, 95% CI = -

4.489 to -1.654).

Species meta-regression model on the full dingo dataset, i.e. including dingo, indicated that some species responded more strongly to dingo removal than others

(Fig. 2.3A, Table 2.4).

54 Table 2.4 Meta-analytic and meta-regression models of effects of dingo removal on all species, including dingo. M – point estimates (mean) for intercepts, CI.lb – 95% Confidence Interval lower bounds, CI.ub – 95% Confidence Interval upper bounds, I2 – total heterogeneities for each model. Estimates that are significantly different from zero (95% Confidence Intervals not spanning zero) are highlighted in bold.

Model M CI.lb CI.ub I2 [%]

Meta-analytic mean -0.132 -0.749 0.486 98.8

Phylogenetic meta-analytic mean -0.145 -1.051 0.761 99.0

Meta-regression with species: Tachyglossus aculeatus 1.330 -0.544 3.205 Notomys alexis -0.389 -3.035 2.257 Notomys fuscus -2.192 -3.735 -0.648 Mus musculus -2.158 -4.796 0.479 Rattus fuscipes -0.960 -3.601 1.682 Oryctolagus cuniculus 0.306 -1.032 1.644 Capra aegagrus hircus 0.401 -2.245 3.047 Sus scrofa -0.678 -3.329 1.973 Equus africanus asinus 0.008 -2.638 2.654 Canis dingo -1.373 -2.067 -0.678 Vulpes vulpes 0.680 -0.271 1.632 Felis catus 0.077 -0.819 0.973 Petaurus breviceps -0.650 -3.291 1.992 Trichosurus vulpecula -0.919 -2.461 0.622 Macropus rufogriseus 0.338 -1.583 2.259 Macropus rufus 2.464 1.457 3.471 Macropus giganteus 0.519 -0.711 1.749 Wallabia bicolor 0.645 -2.080 3.371 Antechinus stuartii -0.866 -3.507 1.776 Dasyuroides byrnei -0.251 -2.897 2.395 Sminthopsini spp -1.181 -3.827 1.465 Sminthopsis macroura -0.240 -2.876 2.395 Perameles suta -0.976 -3.620 1.669

Particularly, dingo removal was linked to a dramatic increase in red kangaroo abundance (lnRRMacropus_rufus = 2.464, 95% CI = 1.457 to 3.471) and to a decrease in the abundance of the dusky hopping-mouse Notomys fuscus (lnRRNotomys_fuscus = -

55 2.192, 95% CI = -3.735 to -0.648). There was also a tendency towards increased abundance of foxes (lnRRVulpes_vulpes = 0.680, 95% CI = -0.271 to 1.632, if 83% CI was used, it would not cross zero). Finally, our meta-analysis confirmed that dingo removal was effective at reducing the abundances of dingoes (lnRRCanis_dingo = -1.373,

95% CI = -2.067 to -0.678).

2.4.2 Effects of fox removal on the abundance of mammals

In contrast to results from the dingo dataset, fox removal, overall, resulted in increased abundance indices of other species present in the dataset (lnRRoverall = 0.737,

95% CI = 0.271 to 1.202, lnRRoverall(phylo) = 0.704, 95% CI = 0.198 to 1.210, Fig. 2.1B,

Table 2.5). We also observed high total heterogeneity in the meta-analytic models

2 (with and without phylogeny I total = 87.5 to 87.7%, Table 2.5), validating exploration of potential moderators via meta-regression models.

56 Table 2.5 Meta-analytic and meta-regression models of effects of fox removal on other species (native and non-native), excluding fox. M – point estimates (mean), CI.lb – 95% Confidence Interval lower bounds, CI.ub – 95% Confidence Interval upper bounds, I2 – total heterogeneities for each model. Estimates represent intercepts, unless slope is specified in the brackets. Estimates that are significantly different from zero (95% Confidence Intervals not spanning zero) are highlighted in bold.

Model M CI.lb CI.ub I2 [%]

Meta-analytic mean 0.737 0.271 1.202 87.7

Phylogenetic meta-analytic mean 0.704 0.198 1.210 87.5

Meta-regression with animal weight Animal weight (slope) 0.284 -0.138 0.706 and ecological type category: Arboreal 1.168 0.499 1.836 Ground-dwelling 0.673 0.090 1.256 Herbivore 0.671 -0.062 1.405 Mesopredator 0.659 -0.246 1.564

Analogous to the meta-regression model for the dingo dataset, we included animal weight and ecological type (herbivore, arboreal, mesopredator, and ground-dwelling) as moderators in meta-regression model. In contrast to the results of dingo removal, the effect of fox removal on other mammals was not related to their weight range

(ßanimal weight = 0.284, 95% CI = -0.138 to 0.706, Fig. 2.1B, Table 2.5). When effect sizes were plotted against animal weights, the largest positive effect sizes were clumped around intermediate values of species body weights (Fig. 2.2B). Arboreal and ground-dwelling animals responded to fox removal with increased abundance

(lnRRarboreal = 1.168, 95% CI = 0.499 to 1.836, lnRRground-dwelling = 0.673, 95% CI =

0.090 to 1.256, Fig. 2.1B, Table 2.5), while there were no statistically significant effects on herbivores and mesopredators (cat data only). However, the mean effect on cats was medium-large and had a wide CI (lnRRcat = 0.612, 95% CI = -0.246 to

1.564), suggesting that fox removal is likely to result in increased cat abundance

57 indices, although this effect varies between locations. Notably, if 79% CI was used, it would not cross zero.

Fox removal resulted in overall increases in abundance indices when analyses were performed on native species only (lnRRoverall = 0.758, 95% CI = 0.298 to 1.218, lnRRoverall(phylo) = 0.760, 95% CI = 0.276 to 1.243, Fig. 2.1B, Table 2.6).

58 Table 2.6 Meta-analytic and Meta-regression models on effects of fox removal on native species. M – point estimates (mean), CI.lb – 95% Confidence Interval lower bounds, CI.ub – 95% Confidence Interval upper bounds, I2 – total heterogeneities for each model. Estimates represent intercepts, unless slope is specified in the brackets. Estimates that are significantly different from zero (95% Confidence Intervals not spanning zero) are highlighted in bold.

Model M CI.lb CI.ub I2 [%]

Meta-analytic mean 0.758 0.298 1.218 83.0

Phylogenetic meta-analytic mean 0.760 0.276 1.243 83.1

Meta-regression with animal weight Above CWR 0.571 -0.134 1.276 category: CWR 0.841 0.296 1.387

The species included in this data subset represented only two out of three weight categories, there were no under CWR species. The increased species abundance was especially clear for native species from the CWR category (lnRRCWR = 0.841, 95% CI

= 0.296 to 1.387, Fig. 1B, Table 2.6), and was smaller and not statistically significant for the above CWR category species (lnRRaboveCWR = 0.571, 95% CI = -0.134 to 1.276,

Fig. 2.1B, Table 2.6), although the difference between these two weight categories was small and statistically similar to zero (lnRRaboveCWR-CWR = 0.270, 95% CI = -0.469 to 1.009).

Lastly, we noted three species with the strongest increase in abundance indices in the fox removal dataset: the common brushtail possum Trichosurus vulpecula, the long- nosed potoroo Potorous tridactylus and the black-flanked rock wallaby Petrogale lateralis, all with a mean effect size per species of above 1.3 (Fig. 2.3B, Table 2.7).

Fox removal treatments effectively reduced the abundance indices of foxes

(lnRRVulpes_vulpes = -1.116, 95% CI = -1.833 to -0.398).

59 Table 2.7 Meta-analytic and meta-regression models of effects of fox removal on all species, including fox. M – point estimates (mean) for intercepts, CI.lb – 95% Confidence Interval lower bounds, CI.ub – 95% Confidence Interval upper bounds, I2 – total heterogeneities for each model. Estimates that are significantly different from zero (95% Confidence Intervals not spanning zero) are highlighted in bold.

Model M CI.lb CI.ub I2 [%]

Meta-analytic mean 0.537 0.096 0.978 90.4

Phylogenetic meta-analytic mean 0.440 -0.142 1.023 90.3

Meta-regression with species: Tachyglossus aculeatus 1.471 -0.527 3.469 Rattus fuscipes -0.023 -1.152 1.107 Rattus rattus -1.324 -3.338 0.690 Lepus europaeus 0.343 -1.655 2.341 Oryctolagus cuniculus 0.004 -0.978 0.987 Vulpes vulpes -1.116 -1.833 -0.398 Felis catus 0.619 -0.328 1.567 Petaurus breviceps 0.105 -1.909 2.119 Pseudocheirus peregrinus 0.843 -0.524 2.210 Trichosurus vulpecula 1.547 0.788 2.306 Bettongia ogilbyi 1.271 -0.231 2.774 Potorous tridactylus 1.364 0.063 2.666 Macropus eugenii 1.561 0.059 3.064 Macropus rufogriseus 0.666 -1.333 2.664 Macropus robustus 0.735 -1.264 2.733 Macropus giganteus 0.584 -0.763 1.932 Wallabia bicolor 0.458 -0.674 1.590 Petrogale lateralis 1.501 0.360 2.642 Vombatus ursinus -0.257 -2.279 1.765 Antechinus stuartii 0.098 -1.916 2.112 Isoodon obesulus 1.033 -0.068 2.135 Perameles nasuta 0.283 -1.084 1.650

2.4.3 Publication bias

We found no evidence of publication bias in dingo removal or in fox removal datasets, when we considered three different approaches to assessing publication bias:

60 1) the funnel plots (Fig. 2.4) bear no distinct visual asymmetry, 2) Egger’s regression tests did not identify significant asymmetry in the dingo funnel plots of the residuals

(t69 = -1.252, P = 0.215) and provided evidence for asymmetry in the fox removal dataset (t52 = 2.895, P = 0.005), and 3) the trim-and-fill analyses found no missing effect studies on the right side of the fox data distributions and 18 for the dingo dataset. Such a pattern may not be indicative of publication bias in highly heterogeneous datasets, such as ours.

61 Figure 2.4 Funnel plots used to estimate publication bias in the dingo and fox removal data sets. A – effect size estimates from the dingo removal dataset plotted against their precision, B – residual effect sizes from the intercept model (phylogenetic meta-analysis) for the dingo removal dataset, C – effect size estimates from the fox removal dataset plotted against their precision, D – residual effect sizes from the intercept model (phylogenetic meta-analysis) for the fox removal dataset. Dashed vertical line indicates no effect.

2.5 Discussion

The results of our meta-analysis provide evidence that lethal control of dingoes and lethal control of foxes have different outcomes for other mammal species. The results showed that the effects of removing dingoes and foxes both scaled with the body size of potential prey species, but in different ways. Removal of dingoes had a negative effect on the abundance of native mammals weighing less than the CWR, but mammals weighing more than the CWR increased in abundance where dingoes were removed. On average, dingo removal had no effect on CWR mammals. In contrast, fox removal had a strong positive effect on the abundance of CWR mammals.

Arboreal and ground-dwelling mammals responded positively to fox control. Other key findings were that lethal control of dingoes did not have a significant effect on the abundance of cats, but where dingoes were removed there was a tendency for fox abundance to increase, and where foxes were removed there was a tendency for cat abundance to increase.

Our finding that the abundances of mammal species respond in different ways to the lethal control of dingoes and foxes has important implications for biodiversity conservation programs. In much of Australia, CWR mammals and mammals weighing less than the CWR have become endangered or extinct due to predation by introduced predators (Woinarski et al. 2015, Doherty et al. 2016). To counter the threat posed by introduced predators for native mammals, wildlife agencies have

62 invested considerable effort and funds in programs in which poison baits are distributed in order to control populations of foxes and dingoes (Fleming 1996,

Robley et al. 2014). However, our results suggest that population-control programs directed towards dingoes and foxes have different outcomes for mammal assemblages. It is important to note that most fox control programs were undertaken in areas where dingoes were rare or absent (Supplementary Figure 3) due to the persecution of dingoes by people. The absence of dingoes from sites where poison- baiting was targeted towards foxes was evidenced in our results by the absence of dingo abundance as a response variable in studies reporting the effects of fox control programs. Thus where fox control programs were conducted, foxes were the top predator.

On average, the abundance of CWR mammals increased in response to fox control, but CWR mammals showed on average no response to the removal of dingoes. In addition, the abundance of heavy mammals increased and that of light mammals decreased in response to dingo control. We suggest that the disparate responses of

CWR mammals and above CWR mammals to dingo and fox removal programs reflect the different dietary preferences of the predators. CWR mammals are within the optimal prey weight range for foxes (Johnson et al. 2007, Dexter & Murray 2009), but are preyed on less frequently by dingoes (Cupples et al. 2011, Spencer et al. 2014,

Davis et al. 2015), and thus may be expected to benefit from fox removal (Saunders et al. 2010, Robley et al. 2014). Similarly, above CWR prey are within the preferred prey weight range for dingoes (Newsome et al. 1983), and thus may be expected to benefit from dingo removal (Colman et al. 2014). Indeed, our results demonstrated that above CWR mammals occur at greater abundances where dingoes have been controlled. Increased abundance of kangaroos and wallabies is an unintended and

63 perverse outcome of dingo control programs, because these animals compete with livestock for pasture (Caughley 1987, Prowse et al. 2015). To reduce the impacts of kangaroos and wallabies on livestock, culling is undertaken throughout Australia, particularly in areas where dingo populations have been suppressed (Gilroy 1999,

Fillios et al. 2010). We recommend that further studies are undertaken to compare the net cost to graziers resulting from controlling dingoes to mitigate stock losses with the cost associated with the irruption of native grazing species resulting from dingo control.

The results of our meta-analysis showed that below CWR mammals responded negatively to the removal of dingoes, but no studies were available to examine the response of below CWR mammals to fox removal. Previous studies have attributed declines in small mammal abundances in areas where dingo populations were controlled to indirect effects resulting from dingo control (Letnic et al. 2009a, Letnic et al. 2009b). It has been theorised that red foxes and cats increase in abundance where dingoes are controlled and, as a consequence, their predatory impacts on small mammals increase (Ritchie and Johnson 2009, Letnic and Dworjanyn 2011). It has also been theorised that small mammals benefit from dingoes’ suppressive effects on kangaroos, and that when dingoes are controlled, over-abundant kangaroo populations destroy habitat and food resources required by small mammals (Colman et al. 2014).

The findings of our meta-analysis, that removal of dingoes did not have a significant effect on fox or cat abundance, but did have a positive effect on kangaroo abundance, lends support to the hypothesis that negative effects of dingo control on small mammal populations are due more to indirect effects arising from increased abundance of large herbivores than to mesopredator release (Morris & Letnic 2017,

Rees et al. 2017).

64 Counter to the results of some field studies (Letnic et al. 2009b, Brook et al. 2012), our meta-analysis showed that, on average, dingo control had no significant effect on the abundances of foxes or cats. However, in the case of foxes, there was a moderate increase in abundance in response to dingo control that was significant if an 83% CI was used as the critical alpha value. One explanation put forward to explain the lack of responses by foxes to dingo control is that the primary method of control used in these studies was baiting, and it is well known that foxes also take poison baits laid for dingoes (Fleming 1996, Twigg et al. 2000). Thus, dingo removal may have no net effect on fox populations because lower fox mortality where dingoes have been removed may be offset by fox mortalities resulting from bait consumption (Fleming

1996). However, Johnson and VanDerWal (2009) demonstrated that for dingoes to suppress foxes effectively, dingo abundance needed to exceed a certain threshold level above which their populations are ecologically effective. The existence of such a threshold could help explain the large confidence limits around the effects of dingoes on foxes, and may be due to some studies being conducted in areas where dingo abundances in baited and unbaited areas were either both above or below this threshold value, so that the authors did not make comparisons of ecologically effective versus ecologically ineffective dingo populations.

Previous studies have found cats to show both positive and negative responses to dingo control, which was reflected in our meta-analyses as an overall neutral effect for cats. One explanation put forward to explain the pattern of cat responses to dingo control in previous studies is that cats may be subject to top-down effects by both dingoes and foxes and, thus, may benefit little if dingoes are removed, because foxes are still present in the landscape (Letnic et al. 2009a, Gordon et al. 2015). This explanation is supported by the results of our meta-analysis, which show that cat

65 abundance tended to increase in response to programs aimed at controlling foxes, which are generally conducted in areas where dingoes have been extirpated. This finding highlights a perverse outcome of fox control and adds weight to the idea that it may be necessary to implement cat control programs at the same time and place as fox control programs (Marlow et al. 2015a, Wayne et al. 2017).

In summary, our meta-analysis demonstrates that lethal control of dingoes and of foxes have different and unintended consequences for native and introduced mammals. The goal of dingo control is often to protect livestock, and sometimes

CWR mammals (Twigg et al. 2000, Letnic et al. 2012), but it results in increased abundance of large mammals, a tendency for fox abundances to increase, declines in the abundance of small mammals, and it does not, on average, benefit CWR mammals. Fox control, which is primarily undertaken in areas where dingoes were rare or absent, on the other hand, benefits CWR mammals, but has the unintended effect of increasing the abundance of feral cats, which is an issue of great concern as cats have been highlighted as a major threat to many native mammals and birds

(Woinarski et al. 2015, Gordon et al. 2017, Wayne et al. 2017).

Our results demonstrate different outcomes for other mammals depending on whether baiting is targeting dingoes or foxes. Land management agencies need to consider whether the goal of their baiting program is for species recovery or livestock protection because our results show that removing both dingoes and foxes has far- reaching impacts for multiple mammal species.

66 2.6 Supplementary material

Supplementary Figure 1. PRISMA diagram showing the process of discovery and elimination of publications for the fox removal dataset. N – number of papers.

Search terms:

1. Red fox predation Australia 2. Red fox control Australia

Web of Science / Google Scholar search

N = 196

Rejected based on title N = 172

No relevant fox effect data N = 12

Additional sources found from backwards search

N = 5

Papers containing relevant data

N = 17

Included papers

N = 17

67 Supplementary Figure 2. PRISMA diagram showing the process of discovery and elimination of publications for the dingo removal dataset. N – number of papers.

68 Supplementary Figure 3. Dingo distribution in Australia. Open circles indicate fox study locations and crosses indicate dingo study locations. In some instances, where multiple sites were within <50km, only a single point is used for clarity. Two markers close together indicate the same (or similar, <50km) study site was used but in a different publication.

69 Supplementary Table 1. Excluded full-text studies – fox removal dataset

No. Study Reason of exclusion Baiting target species

1. Kortner et al. 2003 Measures survival of quolls Vulpes vulpes

2. Claridge et al. 2010 Not enough data points Vulpes vulpes

3. Hayward et al. 2005 No before/control data Vulpes vulpes

Excluded full-text studies – dingo removal dataset

No. Study Reason of exclusion Baiting target species

Inconsistent baiting/lack 1. Kennedy et al. 2012 baiting information Canis dingo

2. Kortner 2007 No control data Canis dingo

3. Eldridge et al. 2002 Data not presented Canis dingo

70 2.7 Acknowledgments

Funding for this research was provided by the Australian Research Council.

71 3. Top predator removal affects mammal species differently at multiple trophic levels in forest ecosystems

72 3.1 Abstract

Globally, top predators are increasingly being recognised for the important role they play in maintaining healthy and stable ecosystems. In Australia, the top predator is the dingo (Canis dingo) and recent research suggests the role dingoes play in Australian landscapes may be similar to top predators elsewhere. However, human activities such as livestock farming are in conflict with dingoes because dingoes predate on sheep and calves. In response, dingoes are lethally controlled across the continent.

Here, we investigate how the lethal control of Australia’s top predator affects multiple mammal species in forest ecosystems. I used camera surveys to perform occupancy analyses to determine if baiting had an effect on mammal species including introduced fox and cat mesopredators, bandicoots and macropods. Additionally, I captured small mammals using Elliot traps and this data informed generalised linear mixed models to determine how dingo baiting, vegetation and location (site) affected these species. Consistent with ecological theory, the occupancy of dingoes and critical weight range bandicoots and dasyurids was lower in baited than unbaited sites.

However, the occupancy of foxes, macropods and brushtail possums was similar across baited and unbaited sites. For all species, except foxes and possums, the detection model including different seasons had the greatest support. These results provide mixed support for the effect of baiting on the species I measured. Designing experiments around pre-existing manipulation components such as dingo baiting programs limits the amount of control we have over other aspects of the methodology.

I acknowledge that we were unable to control for every possible source of environmental variation that may have affected the occupancy/abundance of the mammal species. In addition, small sample sizes may have contributed to some of our confounding results. This study does confirm earlier findings that dingo control

73 methods are effective at reducing populations of dingoes and that this control reduces the occupancy and abundances of some small, ground-dwelling mammals in forest ecosystems.

3.2 Introduction

Globally, populations of top predators continue to decline and are increasingly threatened with extinction (Estes et al. 2011). Top predators are especially susceptible to population declines owing to their slow life histories and competition with humans for prey species (Cardillo et al. 2004). Additionally, range contraction, habitat fragmentation and illegal hunting of predators also contributes to global predator population declines (Ripple et al. 2014). These declines have been linked to ecosystem simplification and destabilisation in many habitat types around the world

(Wallach et al. 2010).

Top predators may influence the structure and functioning of ecosystems as a result of the direct effects they exert on their prey species and smaller predators through predation, competition and fear (Schmitz 2008, Brook et al. 2012, Creel & Creel

1996). Top predators can also induce cascades of indirect effects on species at lower trophic levels which benefit from the direct effects that top predators have on populations and the behaviour of herbivores and mesopredators (Estes et al. 2011,

Ripple et al. 2014, Donadio & Buskirk 2006, Ford et al. 2014). Thus, top predators’ effects on ecosystems may be far-reaching because they can “ripple” along multiple food-web pathways.

One possible reason why mammalian top predators are often very effective at regulating the populations of their prey and competitors is because they have high

74 energy requirements (Williams et al. 2004). The need for mammalian predators to consume large quantities of food, and their requirement for large home ranges contributes to their high incidence of conflict with humans (Ripple et al. 2014). In

Australia, the top predator is the dingo (Canis dingo). Dingoes are persecuted in many places because they attack livestock, especially sheep. In the forests of south-eastern

Australia, dingoes are subject to wide-spread lethal control programmes which involve distributing meat baits that have been injected with poison in areas that dingoes frequent (Letnic et al. 2012b).

Poison baiting programmes targeted towards dingoes can have unintended consequences on ecosystems (Colman et al. 2014) due to relaxation of their direct and indirect effects on other species (Hunter et al. 2018). Studies investigating the impacts of dingo control programmes in both arid and forested systems have shown that areas where dingo populations are controlled tend to have higher abundances of kangaroos and wallabies and red foxes, lower abundances of small and medium-sized native mammals and sparser understorey vegetation than sites where dingo populations are not controlled.(Colman et al. 2014, Hunter et al. 2018., Letnic et al. 2012, Morris &

Letnic 2017). However, unlike foxes, cats do not display consistent responses to dingo removal programs (Hunter et al. 2018).

Although there is evidence that dingoes can structure ecological assemblages in the forests of south-east Australia, studies to date have largely been short-term snap-shot studies of differences in widely separated sites where dingo populations were controlled or not controlled (Colman et al. 2014, Colman et al. 2015, Robertshaw &

Harden 1986) rather than longitudinal studies. Thus, it remains possible that the differences in ecosystem attributes between baited and unbaited sites reported were anomalies that may not persist through time.

75 In this study we investigated the effects that dingo control programs have on mammal assemblages in two forested conservation reserves in eastern Australia on three sampling occasions over two years. Dingo populations are controlled by distributing poisoned meat baits in many conservation reserves in south-east Australia in order to reduce the incidence of dingo attacks on livestock in neighbouring farms. Within each conservation reserve we established “sub-sites” where dingo baiting occurred and where it did not occur. Our specific aims were to determine how occupancy of macropods, introduced mesopredators and small and medium-sized mammal occupancy and abundance differed between baited and unbaited sites. Based on previous studies that have examined the effects of supressing dingo populations

(Colman et al. 2014, Hunter et al. 2018, Letnic et al. 2009b), we predicted that we would observe: 1) higher occupancy of macropods and introduced mesopredators in baited areas owing to potential predation and competition, 2) reduced vegetation cover at baited sites owing to higher macropod (grazing species) occupancy, 3) higher occupancy of brushtail possums in baited areas because they are frequently preyed on by dingoes, and 4) higher occupancy of bandicoots and small mammals at unbaited sites because they benefit from reduced predation pressure from introduced mesopredators and higher vegetation cover.

3.3 Methods

3.3.1 Study area This study was conducted in two conservation reserves, Myall Lakes National Park and Wollemi National Park in the Australian state of New South Wales. Within each

National Park we established a sub-site in an area where the dingo population was controlled by distributing poison baits and a sub-site where the dingo population was

76 not controlled. The Wollemi baited sub-site was baited for 5 years while the Myall

Lakes baited sub-site was baited for 15 years. The sub-sites within conservation reserves were located at least 25 km apart and were matched for underlying geology and vegetation (dominant Eucalypt spp.). Using Keith (2004) as a guide, the vegetation of Myall Lakes sites are described as Hunter-Macleay dry sclerophyll forest dominated by an overstorey of spotted gum (Eucalyptus maculata) and grey gum (Eucalyptus punctata). The sites at Wollemi NP are described as Sydney

Hinterland dry sclerophyll associations with overstorey which may be dominated by grey gum (Eucalyptus punctata), narrow-leaved apple (Angophora bakeri) and hard- leaved scribbly gum (Eucalyptus sclerophylla). Average annual rainfall since 1980 at

Myall Lakes NP was 1455 mm whereas Wollemi NP was 621 mm (Bureau of

Meteorology 2018). The main method of lethal control at baited sites was the distribution of poisoned meat baits containing 6 mg of sodium fluoroacetate

(compound 1080) which is a widespread and effective lethal control method for dingoes (Fleming et al. 2001).

3.3.2 Camera survey design We conducted camera surveys at each sub-site on three occasions. On each survey occasion, cameras were deployed for between four and 35 nights. The large discrepancy between some survey efforts was owing to logistical limitations. For any

Myall Lakes or Wollemi survey occasion there was no more than three nights difference in survey length between baited and unbaited sites (Myall Lakes, August

2014; baited nights 7, unbaited nights 10). Overall, the total number of survey nights was 70 at baited sites and 71 at unbaited sites.

77 Camera survey nights Unbaited site Baited site Wollemi, August 2013 8 nights 8 nights Wollemi, January 2014 5 nights 3 nights Wollemi, July 2014 5 nights 6 nights

Myall, June 2013 7 nights 10 nights Myall, November 2013 36 nights 36 nights Myall, August 2014 10 nights 7 nights

Surveys had between three and 16 functional cameras per survey (number of cameras deployed were always the same but technical issues meant the number functional varied). Cameras were attached to trees at the side of fire tracks and positioned at a height of approximately 1m from the ground. Cameras were angled slightly downwards and along the track at a 45 ° angle. The course of the sun was taken into account when setting up each camera to minimise the effects of glare on the cameras.

We used a combination of Scout Guard (SG550V), Bushnell (Trophy Cam HD) and

Reconyx (RC60) cameras and set these to take a single photo with a one-minute delay between photos. All camera models share very similar core features including 0.3 second trigger, illuminate subjects using infra-red lighting which causes minimal disturbance to the subject and silently cache images to media. Cameras used for each survey were selected at random from the pool of available cameras so that it was equally likely for any camera to be used at any site.

3.3.3 Occupancy analysis Site (Myall Lakes NP/Wollemi NP) and baiting (baited/unbaited) were used as predictor variables in occupancy analysis. Models were constructed and tested in program PRESENCE version 12.1 (USGS Patuxent Wildlife Research Centre, Laurel,

MD, USA). Detection histories were recorded as either present (1), absent (0) and if a camera failed this was marked as (-). A single-season occupancy model was used for

78 these analyses and an assumption of this modelling is that the population is demographically closed during sampling. We ran models using an individual site covariate and a combination of both covariates as well as a constant model with no covariates. For each of our species or species group (i.e. ‘medium macropod’ includes

Wallabia bicolour and Macropus rufogriseus), we test multiple models and use the process of elimination to remove models which do not offer any insight into our model selection. Models were ranked in the program Presence using the Akaike

Information Criterion (AIC) to identify the most parsimonious models with the model with the lowest AIC ranked highest. The differences in AIC of any model and the top model (AIC) suggests the plausibility of competing models. A process guided by

Burnham & Anderson (2004) states that where AIC is <2 there is strong support for the model and it is equally plausible as the top model with the lowest AIC. Models with AIC of 4-7 have some support but are not considered to be very important.

We used Presence to test the goodness of fit for all our models for all our target species. We used our fully parameterised model ((baited+site), p(seasonal)) to assess model fit. We ran 10000 bootstraps to whether this model showed a significant lack of fit to the data. We found that none of our data sets showed a significant lack of fit (dingo, P = 0.143, macropod, P = 0.861, bandicoot, P = 0.131, cat, P = 0.217, possum, P = 0.594, fox, P = 0.371).

3.3.4 Small mammal abundance The abundance of small mammals was assessed using Type A Elliott traps (Elliott

Scientific Equipment, Upwey, Australia) which were deployed for three consecutive nights on three sampling occasions at each sub-site. At each sub-site we established 8

79 grids and laid out 25 traps in 5 x 5 grid spaced 10 m apart. The same trapping grids were used at each sub-site on each of the three trapping occasions. Each trap was baited with a mixture of peanut butter, oats and golden syrup. We calculated indices of rodent (Rattus fuscipes) and dasyurid (Antechinus stuartii, Antechinus swainsonii and Antechinus flavipes) abundance for each grid and this was expressed as the number of captures per hundred trap nights.

Owing to restricted access to locations, my trapping duration was often limited.

Restricted access meant I was not able to trap at the exact same time at each location or between locations. In order to understand if this may have affected my data I have included overall trapping data (Table 3.4) which shows total number of small mammals captured (excluding recaptures) as well as the date and location of each survey. To mute any seasonal discrepancies between locations, I made sure to conduct surveys at each location in the same season.

The complexity of the understorey vegetation (<100 cm) was assessed at each small mammal trapping grid at every sub-site. At each trapping grid, four 5 m quadrats were created and at each quadrat the percentage of a 20 x 50 cm chequered coverboard obscured by vegetation was recorded. We used three strata (0-20, 20-50 and 50-100 cm) to measure the obscurity of vegetation and the percentage values were averaged to provide a single vegetation cover percentage for each trapping grid.

We assessed species accumulation curves to determine if our sampling was adequate.

Our results from these demonstrated that our sampling effort as sufficient. Over the total sampling period spanning approximately 18 months, we captured just five different species. One of those species was represented by a single capture of a house

80 mouse (M. mus). The remaining four species were consistently captured relative to their population composition within each ecosystem.

We used 3 factor generalised linear mixed models with a negative binomial distribution to investigate the relationships between the abundances of dasyurids and rodents and predictor variables. The predictor variables were baiting

(baited/unbaited), occasion (Winter 2013/Summer 2013-2014/Winter 2014) and location (Myall Lakes NP/Wollemi NP). Because there were multiple grids at each sub-site, sub-site was included in the model as a random variable. Because sampling was conducted on multiple occasions on the same grids, occasion was treated as a repeated measure with an AR1 error structure. Analyses were conducted using SPSS version 24.0.

3.4 Results

3.4.1 Occupancy modelling The dingo detection model that included different seasons had the greatest support compared to one in which detection was constant or time-varying. Detection varied from 0.05±0.02 to 0.19±0.03. The occupancy model that included baiting and location had the greatest support with a model weight of 0.77 (Table 3.1). The parameter estimates for occupancy for unbaited sites was 0.43–0.78 and for baited sites was

0.05–0.19 (Table 3.2) which provides support for the idea that baiting and locations with lower ecosystem productivity can reduce occupancy of dingoes.

Initially for macropods models were not converging. To remedy this, I had to collapse the data so that I had only three pooled occasions. Following this my models showed that the model where detection was estimated as constant across occasions showed a

81 better fit to the data than allowing detection to vary by season (Table 3.1). The overall occupancy value for macropods is 0.85 (Table 3.2).

The cat detection model that included seasons was better than one where detection was constant or time-varying. Detection spanned from 0.04±0.01 to 0.08±0.05. The occupancy model containing location only (wi – 0.95) was best. Adding baiting to the location model gave a poorer fit to the data and because the combined model is within

2∆AIC of the top model, ‘baiting’ is viewed as an uninformative parameter (Arnold

2010) so that model was deleted from the final set.

For possums the detection model that included different seasons had the greatest support. The lowest detection value was 0.02±0.01 and the highest as 0.06±0.02.

There appeared to be no strong effect of baiting or location covariates on possums as the constant model has more than three times the support (wi – 0.61) of the next best model containing site or baiting. However, there did appear to be an effect of season.

The bandicoot detection model that included seasons was better than one where detection was constant or time-varying. Detection values spanned from 0.01±0.0 to

0.05±0.02. Comparison of different models suggests that baiting does reduce bandicoot occupancy (Table 3.2). After the uninformative parameter (baiting+site) is removed from the model set (Arnold 2010), baiting has more than two times as much support as site according to weights. Despite that, the best model is only 1.53 AIC better than ‘site’ model. The parameter estimates for occupancy for unbaited sites was

0.72 and for baited sites was 0.26 (Table 3.2).

For the fox, a model with seasonal detection differed from one with detection constant across occasions differed by only 1.26 AIC. The detection value was 0.02±0.01.

Therefore, site covariates were modelled with detection constant across occasions.

82 The inclusion of baiting or location covariates did not lead to a better supported site model than one where occupancy was modelled as constant across sites (Table 3.1). A model that included both baiting and location did not converge.

83 Table 3.1 Model outputs for single-season occupancy models for each species.

Species and model AIC  AIC wi k -2l Macropod (.), p(.) 168.30 0.00 0.76 2 164.30 (.), p(seasonal) 170.62 2.32 0.24 4 162.62 Dingo (baiting+site), p(seasonal) 343.98 0.00 0.77 6 331.98 Dingo(baiting), p(seasonal) 346.40 2.42 0.23 5 336.40 (site), p(seasonal) 356.53 12.55 0.01 5 346.53 (.), p(seasonal) 358.51 14.53 0.01 4 350.51 Cat (site), p(seasonal) 356.27 0.00 0.95 5 346.27 Cat (.), p(seasonal) 362.96 6.69 0.03 4 354.96 (baiting), p(seasonal) 364.73 8.46 0.01 5 354.73 Brushtail possum (.), p(seasonal) 272.18 0.00 0.61 4 264.18 Brushtail(site), possum p(seasonal) 274.12 1.94 0.19 5 264.12 (baiting), p(seasonal) 274.18 2.00 0.18 5 264.18 (.), p(.) 276.08 3.90 0.07 2 272.08

Bandicoot (baiting), p(seasonal) 225.44 0.00 0.56 5 215.44 (site), p(seasonal) 226.97 1.53 0.26 5 216.97 (.), p(seasonal) 227.56 2.12 0.19 4 219.56

Fox (.), p(.) 137.39 0.00 0.38 2 133.39 (baiting), p(.) 137.52 0.13 0.35 3 131.52 (habitat), p(.) 138.03 0.64 0.27 3 132.03

84 Table 3.2 The best occupancy model for each species is shown and the corresponding occupancy values are included.

Animal Best model Baiting = 0, Baiting = 0, Baiting = 1, Baiting = 1,

Site = 0 Site = 1 Site = 0 Site = 1 Macropod (.), p(.)  0.85, SE 0.19  0.85, SE 0.19  0.85, SE 0.19  0.85, SE 0.19

Dingo (baiting+site),  0.78, SE 0.10  0.43, SE 0.14  0.19, SE 0.10  0.05, SE 0.04 p(seasonal)

Cat (site),  1.00, SE 0.0  0.45, SE 0.12  1.00, SE 0.0  0.45, SE 0.12 p(seasonal)

Possum (.), p(seasonal)  0.65, SE 0.16  0.65, SE 0.16  0.65, SE 0.16  0.65, SE 0.16

Bandicoot (baiting),  0.72, SE 0.17  0.72, SE 0.17  0.26, SE 0.15  0.26, SE 0.15 p(seasonal)

Fox (.), p(.)  0.63, SE 0.35  0.63, SE 0.35  0.63, SE 0.35  0.63, SE 0.35

3.4.2 Small mammals Dasyurids showed a significant response to baiting (F = 15.92, d.f. = 1, p = < 0.01 and were on average more abundant at unbaited sites (Fig. 3.1). Location, occasion and understorey vegetation density had no effect on dasyurid abundance and no interaction terms were significant (Table 3.3). The abundance of rodents differed between locations and was on average greater at Myall Lakes NP than Wollemi NP (Fig. 3.2). Rodent abundance was not affected by baiting but they did respond significantly to location (Table 3.3). Total small mammal captures (excluding recaptures) across all baited and unbaited sampling sub-sites at each location are listed in Table 3.4.

85 Figure 3.1 Total average dasyurid capture rate for each treatment. Total average is derived from the number of captures per grid divided by the number of nights traps were out at each grid. 1

0.9

0.8

0.7

0.6

0.5

0.4

Grid captures/night Grid 0.3

0.2

0.1

0 Unbaited Baited

0.6

0.5

0.4

0.3

0.2 Grid captures/night Grid

0.1

0 Unbaited Baited

87 Table 3.3 F values for generalized linear mixed effects models investigating the effects that baiting, occasion and location had on the abundances of dasyurids and rodents.

Factor df F

Dasyurid Rodent Baiting 1,78 4.538* 0.057

Occasion 2,78 0.351 0.675 Location 2,78 1.099 15.642*** Baiting x Occasion 1,78 1.639 1.597 Location x Baiting 1,78 2.115 2.045 Location x Occasion 2,78 1.181 1.555 Baiting x Occasion x Location 2,78 1.748 0.007 *P<0.05, ***P<0.01

Table 3.4 Total small mammal captures (excluding recaptures) across all baited and unbaited sampling sub-sites at each location.

Total small mammal caps Unbaited Baited

Wollemi, August 2013 49 21 Wollemi, January 2014 28 14 Wollemi, July 2014 13 17

Myall, June 2013 104 18 Myall, November 2013 38 21 Myall, August 2014 71 29

3.5 Discussion

The results show differences in the occupancy/abundance of dingoes, bandicoots and dasyurids that are generally consistent with theory and the results of previous studies investigating the responses of mammals to dingo control programs in forest ecosystems (Colman et al. 2014, Johnson & VanDerWal 2009, Robertshaw & Harden

88 1986). However, not all the mammals whose occupancy/abundance we monitored responded to baiting as we predicted. Modelling suggests that foxes were equally distributed across sites whether baited or not (constant model was top). Brushtail possums and macropods were also best explained by the constant model. Possums detection was best explained by season. Brushtail possum occupancy was constant across sites and seasonal variations were detected for this species. The abundance of rodents was unaffected by baiting but showed a strong response to location. Dasyurids responded negatively to dingo control. It seems that the total duration of small mammal trapping at each sub-site and location was suffice as the same species were consistently captured with no new species being discovered after the first survey at each site.

A potential short-coming of our study is that we did not manipulate dingo abundance but rather took advantage of a natural experiment made possible by pre-existing differences in dingo baiting regimes within two National parks. During the selection of our study sites, we attempted to control for environmental variation between baited and unbaited sites within each National Park by matching sub-sites for vegetation type and recent fire history. We acknowledge that we were unable to control for every possible source of environmental variation that may have affected the occupancy/abundance of the mammal species. Thus, we cannot rule out the possibility that the patterns in mammal occupancy/abundances that we report could be due to factors other than the suppression of dingo populations. However, given that our findings for dingoes, bandicoots and dasyurids accord with theory and the results of previous studies (Robertshaw & Harden 1986, Colman et al. 2014), we argue that our results provide some, albeit mixed, support for the idea that suppression of dingo populations engenders shifts in forest mammal assemblages. We highlight that further

89 studies which manipulate dingo abundance are required to confirm or refute the patterns in mammal occupancy/abundance in baited and unbaited areas which we report.

Consistent with previous studies (Colman et al. 2014, Fleming et al. 1996) and indeed the aim of baiting programs, our results demonstrate that 1080 baiting has a strong effect on the occupancy of dingoes in forest ecosystems. Given that the most parsimonious model for dingo occupancy included the terms baiting and location (i.e.

National Park), our results showed that there were considerable differences in dingo occupancy between location. We suggest that these differences may have been due to higher ecosystem productivity and presumably prey abundance at the Myall Lakes NP site which was situated on richer soils (Keith 2004) and received higher rainfall than the Wollemi NP site (Bureau of Meteorology 2018).

Our results showed that dingo suppression was not linked to an increase in macropod occupancy at either site as the constant model was best. Most studies conducted in temperate and arid ecosystems show that macropod abundance typically increases at sites where dingo populations are suppressed (Hunter et al. 2018). The complete model shows a subtle effect of dingo control on macropod occupancy but this was not strong enough to outcompete the constant model. These results reporting no difference in occupancy between sites are inconsistent with previous studies and suggest that irruption of macropod populations in response to dingo population control may not be a universal outcome. Both of the pairs of treatment and control sites at each National Park were in close proximity to freehold grazing land (<5 kms) and this may have facilitated rapid recruitment to baited sites from adjacent freehold pastures, potentially confounding results. We recommend that further research is required to elucidate in further detail the response of macropods to dingo suppression

90 programs because some effect on macropods is almost ubiquitous in the literature

(Hunter et al. 2018).

Our findings for the effects of dingo control on foxes suggested that fox occupancy was equivalent across sites. The high degree of variation in this estimate (i.e. high SE) suggests that making any certain conclusions about the effect of dingo control on foxes from this study is not possible. Our occupancy models showed that the best predictor for cat occupancy was location and that dingo baiting had little effect on cat occupancy. This finding is consistent with a recent meta-analysis which shows that on average, dingo control has little effect on cat populations (Hunter et al. 2018). While this conclusion is inconsistent with the mesopredator release hypothesis (Crooks &

Soulé 1999), previous studies suggest that factors influencing occupancy/abundance of cat populations may be due to a combination of factors including prey abundance and the relative strength of top-down effects exerted on their populations by both dingoes and foxes (Gordon et al. 2015, Hunter et al. 2018).

Brushtail possum occupancy showed no response to baiting as occupancy is estimated to be similar across both baited and unbaited sites. Dexter et al. (2013) hypothesised that brushtail possum abundance should increase where dingo populations are controlled because they are an important prey item of dingoes, and indeed this hypothesis was supported by the results of Colman et al. (2014) who surveyed their populations by conducting spotlight surveys. Although possums frequently forage on the ground, they are primarily arboreal. Thus, a potential short-coming of our study was that trail cameras set to detect animals using roadsides may not have been the most effective methodology to gauge the occupancy of brushtail possums.

91 Bandicoot occupancy was greater in unbaited than baited sites. This finding is consistent with previous studies which have showed positive relationships between abundance indices of dingoes and ground dwelling mammals (Colman et al. 2014,

Colman et al. 2015). Colman et al. (2014) found that dingo baiting per se had no effect on bandicoot abundance. They found evidence for an indirect positive effect of dingoes on bandicoots mediated by dingoes’ positive effect on understorey vegetation density, which they in turn attributed to a reduction in macropod grazing impacts in areas where dingoes were abundant. Given that fox and cat occupancy remained constant across treatments it is reasonable to suggest that bandicoot occupancy is facilitated indirectly by dingoes because their presence in this study was linked to increased vegetation cover.

The abundances of dasyurids and rodents (small, critical weight range mammals) displayed mixed responses to dingo control. Consistent with the findings of Colman et al. (2014), dasyurids were less abundant at sites where dingoes were baited. However, at odds with our predictions and the findings of Colman et al. (2014), rodent abundance was unaffected by baiting but was strongly influenced by location. Colman et al. (2014) attributed negative responses of small mammals to dingo control to a reduction in understorey habitat complexity due to higher grazing pressure by kangaroos and increased predation pressure by foxes. We recommend that a more in depth study is undertaken to investigate the effects that dingo control has on small mammals and the factors that influence their abundances.

Removal of dingoes has been shown to have profound effects on desert ecosystems and snapshot studies conducted in forested systems have demonstrated similar findings also. The results of this study provide mixed support for our a priori predictions. Possible reasons could be that dingo effects on other mammal species are

92 weaker than previous studies have suggested or that our analyses were constrained by low replication given that we only sampled four sub-sites in total. Camera failure during sampling may have also contributed to high error values in some instances.

This study does confirm earlier findings that dingo control methods are effective at reducing populations of dingoes and that this control reduces the occupancy and abundances of some small, ground-dwelling mammals in forest ecosystems.

4. Activity indices of a top-predator and mesopredator respond in opposite ways to predator control programs

93 4.1 Abstract

The mesopredator release hypothesis predicts that abundance of smaller predators should increase in the absence of larger predators due to release from direct killing and competition. However, top predators’ effects on mesopredators are unlikely to operate in isolation but interact with other factors such as the ecosystem productivity and human activities. Here, we investigate factors influencing abundance indices of an top predator, the dingo, and an introduced mesopredator, the red fox, in forests of south-eastern Australia. We used quantile regression to determine if the relationship between indices of dingoes and foxes was concordant with triangular, negative relationships reported in previous studies. We used generalised linear models to investigate the effects that net primary productivity, proximity to freehold land and poison baiting campaigns directed at dingoes had on fox and dingo activity. Quantile regression indicated that fox activity had triangular, negative relationships with dingo activity. Dingo activity was lower in baited than unbaited sites. Of the predictor variables tested, baiting was the best predictor of abundance for both dingoes and foxes. Baiting was in the best model set as was net primary productivity. Positive responses of foxes to dingo control are consistent with the mesopredator release hypothesis and suggest that dingoes have a greater suppressive effect on fox populations than poisoning campaigns directed towards dingoes. We speculate that divergent responses of indices of dingo abundance in baited and unbaited sites along a gradient of primary productivity may result from people investing greater effort towards dingo control in more productive landscapes. Fox activity was greater in areas where poison baiting was undertaken. Our results suggest that mesopredator and top predator abundances respond in different ways to top-down and bottom-up drivers

94 and that removal of dingoes may be counter-productive for biodiversity conservation because it may lead to higher activity of foxes.

4.2 Introduction

Recent studies have demonstrated the extent to which top predators can shape ecosystems due to the strong interactions they have with other species (Estes et al.

2011, Ripple et al. 2014). The most conspicuous effects top predators have on ecosystems are the suppressive and aversive “direct” effects that they have on populations and behaviour of their prey and smaller predators (mesopredators) that result from killing, the fear they instill and, in the case of mesopredators, competition for resources (Schmitz 2008, Brook et al. 2012, Creel & Creel 1996). However, a growing body of evidence demonstrates that top predators can facilitate smaller predators through carrion provisioning (DeVault et al. 2003, Sivy et al. 2018) and this can, in some cases, offset the net effect of suppression (Wilson & Wolkovich 2011).

Top predators can also propagate cascades of “indirect” effects on species at lower trophic levels which benefit from top predators’ strong suppressive effects on the populations and behaviour of herbivores and mesopredators (Estes et al. 2011, Ripple et al. 2014). For example, increased abundance of mesopredators that frequently follows the removal of top predators can result in the decline of small prey populations owing to elevated rates of predation by mesopredators (Ritchie &

Johnson 2009).

Although larger predators often exert top-down effects on populations of mesopredators, these effects are unlikely to operate in isolation but interact in complex ways with other factors such as the productivity of ecosystems, the diversity

95 of species within them and human land activities (Greenville et al. 2014, Hollings et al. 2013, Pasanen-Mortensen et al. 2013, Périquet et al. 2015). Theoretical frameworks guided by empirical studies predict that the effects of top predators on smaller predators should scale with primary productivity because all predators require energy and the availability of energy may result in bottom-up limitation when population growth is more constrained by resources than by predation (Elmhagen et al. 2010, Elmhagen & Rushton 2007, Letnic et al. 2010). Another reason put forward to explain why top predators’ effects on other species within ecosystems can wane with increasing primary productivity and associated species diversity, is that their effects become dampened as the number of interaction pathways increases (Finke &

Denno 2004, Ripple et al. 2014).

Human activities are a particularly important driver of predator abundances and behaviour (Ripple et al. 2014). Humans often persecute predators that pose a threat to livestock or human well-being and may also suppress the abundances of some predators by destroying their habitat. However, human activities may facilitate increases in the populations of some predator species by increasing the availability of food and habitat (Elmhagen & Rushton 2007) or if persecution of large predator species releases smaller predators from top-down control imposed by larger predators

(Newsome & Ripple, 2014). Because agriculture tends to focus on more productive landscapes (Haberl et al. 2007) the effects of human activities on predator abundance and behaviour may not be independent of ecosystem productivity (Elmhagen &

Rushton 2007).

Several studies have reported “triangular” relationships between indices of top predator and mesopredator abundance whereby the highest abundance of mesopredators occur where top predator abundance is lowest and the lowest

96 abundance of mesopredators occur where top predator abundance is highest.

However, at intermediate levels of top predator abundance there is no clear relationship between top predator and mesopredator abundance (Johnson &

VanDerWal 2009, Letnic et al. 2011, Bagniewska & Kamler 2014, Swanson et al.

2014) (Figure 4.1). In the case of dingoes and foxes in Australia, such relationships have been investigated using quantile regression to demonstrate that the edge of the distribution of fox abundance versus dingo abundance was a more useful predictor of fox abundance than least squares linear regression (Johnson & VanDerWal 2009). In such cases the interpretation of the triangular relationship has been that top predators place an upper limit on mesopredator abundance, but is not the sole determinant of mesopredator abundance (Johnson & VanDerWal 2009, Letnic et al. 2011).

Consequently, there has been considerable interest in the idea dingoes’ fox- suppressive effects could be harnessed to mitigate the devastating predatory impacts that foxes have on small and medium-sized mammals (Letnic et al. 2012, Newsome et al. 2015).

97 Figure 4.1 (a) the relationship between dingo and fox activity from Johnson & VanDerWal 2009, the solid line indicates the regression line for the 0.9 quantile (P = 0.02) (b) relationship between dingo and fox activity from Letnic et al. 2011, the solid line indicates the regression line for the 0.9 quantile (P = <0.05).

In this study we investigate the factors influencing abundance indices of red foxes and dingoes in the forests of south-eastern Australia. Because previous landscape-scale studies investigating dingo-fox dynamics have not accounted for confounding environmental variables, the relative importance that dingoes and other factors such as primary productivity and human land-use have in influencing fox abundance remains unclear (Allen et al. 2013a/b, Letnic et al. 2012). To elucidate factors limiting the activity of dingoes and foxes, we quantified dingo and fox activity in forested landscapes of south-eastern Australia. In our analyses we first investigated the relationship between indices of dingo and fox activity using quantile regression to determine if a relationship between indices of dingoes and foxes was concordant with the triangular, negative relationships reported in previous studies. We then further explored the data using generalised linear models (GLM) to investigate the effects that net primary productivity, proximity to freehold land and baiting history had on the activity indices of foxes and dingoes. In the case of foxes, dingo activity was also included as a predictor variable in the GLMs.

We predicted that (i) lethal control of dingoes would have a positive effect on fox activity owing to the mesopredator release hypothesis and under these conditions fox activity would increase with increasing productivity in the environment due to a greater availability of food resources, (ii) where dingo control was not conducted we predicted that foxes would not show marked increase in activity as environmental productivity increased owing to the fact that foxes would be suppressed from the top- down by the presence of dingoes, (iii) dingo activity would be negatively correlated

98 with lethal control and that in areas where there was no lethal control, dingoes would increase as environmental productivity increased due to greater availability of food resources.

4.3 Methods

4.3.1 Study sites The study was conducted in the Eucalyptus spp. dominated forest ecosystems of New

South Wales, south-eastern Australia (Figure 4.2)(Keith 2004). The main technique used by land management authorities to suppress dingo populations is the distribution of poisoned meat baits containing 6 mg of the toxin sodium fluoroacetate (compound

1080) (Fleming et al. 2001). The baits are typically distributed along unsealed dirt roads or from the air via helicopter or light aeroplane. In some places trapping and shooting of dingoes complements baiting.

Figure 4.2. Solid points depict site locations used for this study.

100 Of the 27 study sites, 13 were subjected to poison baiting at least once a year for at least two years prior to survey (Supplementary Table 1). The remaining 14 sites were not baited (Supplementary Table 1).). Each of the sites were situated within conservation reserves (n=26) or forestry areas (n=1) managed by the New South

Wales National Parks and Wildlife Service and the New South Wales Forestry

Corporation, respectively.

4.3.2 Sampling Carnivore abundance is often estimated using track-based indices (Catling & Burt

1995, Allen et al. 1996, Funston et al. 2010). While track based indices tend to be correlated with predator abundance it is important to note that they do not provide an estimate of predator abundance (Funston et al. 2010). In the case of dingoes and foxes, track based indices have been shown to correlate with other abundance indices

(Mahon et al. 1998) and to decline with the implementation of poison baiting campaigns intended to suppress their populations (Twigg et al. 2000, Kortner et al.

2003). These correlations suggest that track based indices provide a useful index of predator abundance (Mahon et al. 1998).

We used sand-plot surveys to calculate an activity index for both dingoes and foxes at baited and unbaited sites (Figure 4.2). The Myall Lakes and Wollemi National Park locations which featured as the primary locations in the Chapter 3 study were also used in this study. However, I was not able to conduct simultaneous occupancy/camera trap surveys and track plot surveys at all the same locations. This is because about half of the locations surveyed for this study were opportunistically sourced from colleagues who sampled locations before my study began. However, I contend that future track plot surveys should be carried out in conjunction with other survey methods measuring abundance as a means to further compare and contrast the

101 accuracy of activity indices as a proxy for abundance.

We constructed sand plots across small, unsealed vehicle roads by spreading sand 1 m wide and thick enough to ensure animal tracks could be identified. Owing to the fact that sand plots used in this study came from three different projects, each study area had between 10 and 28 track plots spaced between 500m and 750m apart, depending on the study. Each study placed sand plots the same distance apart. We checked sand plots each morning for between four and eight nights and raked them over once data was collected. A unique mark was placed on each sand plot every morning after checking each plot. This mark served as an indicator of disturbance to the plot by rain, wind or vehicles that might have erased tracks. If the mark was indecipherable the next morning, the plot was omitted from the previous night survey. We calculated indices of predator activity by dividing, for each species, the number of tracks by the number of nights and standardised them, so that values fell within the range of 0 to 1

(Johnson & VanDerWal 2009).

4.3.3 Predictor variables For each of the 27 survey locations an estimate of net primary productivity (NPP) was sourced from the dynamic global vegetation model MC1 (Bachelet et al. 2001). To extract estimates of NPP we used 0.5 latitude x 0.5 longitude size grid squares. The

MC1 model simulates monthly carbon and nutrient dynamics for the given ecosystem

(Bachelet et al. 2001). The model incorporates both aboveground and belowground processes to generate an NPP estimate where NPP is the maximum plant production rate constrained by certain effects such as soil nutrients, atmospheric CO2 concentration among other things (Bachelet et al. 2001). All NPP values were obtained in August 2015 once all data was collected and compiled. We use this NPP

102 value from each study location as a proxy for the overall productivity of each site as has been done before in Letnic & Ripple (2017).

Distance to freehold land has been demonstrated as a strong predictor of red fox abundance in forests of south-eastern Australia (Catling & Burt 1995). We included this in our models as a predictor for both fox and dingo activity and this was calculated by measuring the distance in kilometres from the closest sand plot to the closest boundary point on freehold land using the ruler tool in GIS.

Baiting is included in our analyses as we hypothesised it would be an important predictor variable owing to previous studies which demonstrate that it often results in significant reductions in abundances of both foxes and dingoes. We acknowledge that the responses to baiting are not binary and do in fact span a continuum from 0-100.

However, baiting is a widespread and significant process that has been previously demonstrated to affect both dingo and fox activity and thus its presence or absence in the landscape is included here as a binary function (Hunter et al. 2018).

Dingoes are included as a predictor variable in the fox analysis because ecological theory suggests that top predators, such as dingoes, can limit the populations of smaller mesopredators such as red fox (Colman et al. 2014). However, foxes are not included as a variable in the dingo models because the outcomes of studies investigating relationships between dingoes and foxes show no support for the notion that foxes affect dingo activity or abundance via processes such as facilitation or kleptoparatisism (Hunter et al. 2018). Moreover, foxes were not included in the dingo models because results from our quantile regression and those of previous studies do

103 not suggest the presence of a facilitative effect of dingoes on foxes (Johnson &

VanDerWal 2009, Coleman et al. 2014, Hunter et al. 2018).

Season is included as a variable in both dingo and fox models. In this study, season refers to which of the four seasons sand plot surveys took place at each of the 27 survey sites. The justification for including season is that previous studies have shown that both dingo and fox activity or abundance is influenced by seasonality. Year on the other hand is not included in the models because it simply refers to which year surveys were conducted and this is correlated with location owing to logistical constraints of data collection. Null models were included in the analyses for both dingo and fox.

4.3.4 Statistical analyses Some studies have applied ordinary least squares (OLS) regression to analyse the relationship between dingo and fox activity (Catling & Burt 1995, Newsome 2001), but these analyses have been criticised for violating OLS assumption of homogeneity of variance (Johnson & VanDerWal 2009).

In our study, we tested the relationship between fox and dingo activity indices using quantile regression, which is an appropriate technique for threshold-type relationships and does not assume homogeneity of variance (Cade & Noon 2003, Johnson &

VanDerWal 2009). Quantile regression allows the modeller to define quantiles of interest for a response variable and use these to provide a more thorough understanding of the relationship between variables in ecosystems (Cade & Noon,

2003). We plotted multiple quantile regression lines at the 0.95, 0.9, 0.75 and 0.5 quantiles using R statistical software (R Studio version 0.98.5, http://www.rstudio.com).

104 To determine factors that influence fox and dingo activity we constructed and tested

GLMs using R statistical software (R Studio version 0.98.5, http://www.rstudio.com).

We specified a negative binomial error structure and converted sand plot data to a count structure whereby data was expressed as the number of plots disturbed per 100 plot nights. We adjusted the Akaike Information Criterion (AICc) due to our small sample size (Burnham & Anderson 2002). To determine if any of the predictor variables were correlated we conducted a Spearman rank correlation analysis using the software SPSS. Highly correlated variables (r > ± 0.5) were not included in the same model.

4.4 Results

4.4.1 Quantile regression

Scatterplots of dingo versus fox activity revealed a “triangular” spread of the data. In baited areas (Figure 4. 1c) dingo activity was generally low and fox activity high and at unbaited areas (Figure 4.1c) dingo activity was mostly high and fox abundance was always low (<0.2 foxes-1 plot-1 night-1). The scatterplot showed that fox activity was low when dingo activity exceeded approximately 0.25 dingoes-1 plot-1 night-1 and the relationship between dingo and fox activity indices was significant at the 0.95 and 0.9 quantiles (P < 0.0001 and P = 0.043, respectively, Figure 4.1c). It appears that the

0.95 quantile line is driven by the presence of two outliers in the dataset. The significant upper quantiles suggested that dingoes place an upper limit of fox activity, but the spread of the data suggest that dingo activity was not the only variable influencing fox activity, particularly at sites where both dingo and fox activity were low.

105

Figure 4.1 (c) Scatterplot of our field data showing the relationship between quantile regression lines fitted to the 0.95 (P = <0.001) and 0.9 (P = 0.043) quantiles, open circles represent unbaited sites and closed circles are baited sites.

4.4.2 Correlations between predictor variables

There was a negative correlation between baiting and distance to freehold land

(spearman’s r=0.526, n=27, P=0.005) indicating that baiting was less likely to be conducted at sites further from freehold land than those closer to freehold land. The correlations between baiting and npp (r=-0.072, n=27, P=0.723) and distance to freehold land and npp (r=0.206, n=27, P= 0.302) were negligible.

4.4.3 Generalised linear modelling

The most parsimonious model for predicting dingo activity was baiting (Δi =0.00, wi

= 0.39, Table 4.1). On average dingo activity was greater at unbaited sites (0.20 ± SE

= 0.03) than baited sites (0.07 ± SE = 0.032) (Fig. 4.3). The next best model in the top model set was distance to freehold land (Δi = 0.6, wi = 0.28, parameter intercept

106 estimate 0.101 (0.021, 0.181) (Supplementary Table 2). The null model did not appear in the list of best competing models in either of the dingo or fox analysis.

Table 4.1 Candidate generalised linear models analysing the competing variables for dingo activity. Models were analysed using a negative binomial distribution. The best fitting model and those with <2 AIC units of the top model are indicated with the dotted line. Models containing both baiting and distance to freehold were omitted as these parameters are correlated.

Residual Competing models AICc Δi w i deviance

Baiting 13.63 28.1 0.0 0.39

Dist_freehold 26.73 28.7 0.6 0.28

NPP 26.82 28.8 0.7 0.27

NPP + Baiting 13.54 32.8 4.7 0.04

NPP + Dist_freehold 14.24 33.5 5.4 0.03

Null 14.32 37.8 9.7 <0.01

NPP + Baiting + Distfreehold 13.53 40.0 11.9 <0.01

107 0.3

0.25

0.2

0.15

0.1 Activity (tracks/nights) Activity index 0.05

0 Dingo Fox

Unbaited Baited

Figure 4.3 Average activity indices (± 1 SE) for both dingoes and foxes in unbaited and baited study sites. Results from t-tests indicate that there is a statistically significant difference (<0.05) between dingo and fox activity at unbaited sites.

The most parsimonious generalized linear model predicting fox activity was also the effect of baiting (Δi = 0.00, w i =0.33, Table 4.2). Distance to freehold land (Δi = 0.8, w i =0.22), dingo (Δi = 1.2, w i =0.18) and NPP (Δi = 1.5, w i =0.16) were all also in the best model set. On average fox activity was greater at baited (0.17 ± SE = 0.06) than unbaited sites (0.02 ± SE = 0.008) (Fig. 4.3).

108 Table 4.2. Candidate generalised linear models analysing the competing variables for fox activity. Models were analysed using a negative binomial distribution. The best fitting model and those with <2 AIC units of the top model are indicated with the dotted line. Models containing both baiting and distance to freehold were omitted as these parameters are correlated.

Residual Competing models AICc Δi w i deviance

Baiting 9.85 21.8 0.0 0.33

Dist_freehold 10.67 22.7 0.8 0.22

Dingo 11.05 23.0 1.2 0.18

NPP 11.39 23.4 1.5 0.16

NPP + Baiting 9.76 26.6 4.7 0.03

Baiting + Dingo 9.85 26.6 4.8 0.03

Null 10.32 27.0 5.2 0.02

NPP + Dist_freehold 10.51 27.3 5.5 0.02

Dist_freehold + Dingo 10.57 27.4 5.5 <0.01

4.5 Discussion

Our results provide support for the prediction of the mesopredator release hypothesis that suppression of top predator populations results in an increase in the abundance of mesopredators. Quantile regression indicated that fox activity was correlated negatively with dingo activity. Generalized linear modelling indicated that dingo activity was lower in areas where poison baits were laid and conversely, fox activity was higher in areas where poison baits were laid and dingo activity was lower.

However, in addition to baiting, the predictor variables distance to freehold land and

NPP were within the best model set for both dingo activity and fox activity and

109 dingoes were also in the top model set for foxes. However, for each of these predictor variables the polarity of the model coefficients was different for dingoes and foxes.

Below we discuss our results in the context with previous research on the mesopredator release hypothesis and interactions between coexisting canids.

A caveat of our study was that we necessarily incorporated data derived from different search efforts such that some transects were longer than others and sandplot numbers and survey season was varied. We acknowledge these methodological discrepancies and note that they were unavoidable in order to obtain sufficient quantity of data at the spatial scale required to test our hypotheses. It is worth noting also, that manipulative experiments investigating the effects that top predators

(animals with large home-ranges) have on mesopredators have rarely been conducted because of logistical and ethical issues (see Mao et al. 2005, Moseby et al. 2018,

Ripple & Beschta 2012). Taking advantage of existing manipulations of top predator populations such as I have done in this study provides one way to ask questions about top predator-mesopredator interactions.

A further limitation of our study is that because the relationship between track-based index of abundance we used and actual abundance is unknown is that drawing conclusions about population processes require making the assumption that there is a positive correlation between the track-based index and abundance. However, previous studies investigating the utility of track-based indices of dingo and fox abundance suggest that this assumption is valid because they found positive correlations with other abundance indices and declines in track based indices following the implementation of control operations (Allen et al. 1996, Körtner et al. 2003, Mahon et al. 1998, Twigg et al. 2000), however not all studies report clear negative effects of control programs (Claridge et al. 2010). Thus, we emphasise, further studies are

110 needed to understand the relationship between track-based indices and the abundance of dingoes and foxes.

The negative relationship observed between dingo and fox activity indices in our quantile regression analyses are consistent with the mesopredator release hypothesis and previous studies showing that larger predators suppress the abundances of smaller predators (Crooks & Soulé 1999, Elmhagen et al. 2010, Newsome & Ripple 2014).

When plotted against each other a “triangular” relationship was evident between indices of dingo and fox activity such that when dingo activity was high, fox activity was always low. Conversely the highest levels of fox activity occurred at sites where dingo activity was low, but at some sites where dingo activity was low fox activity was low also. Previous studies have interpreted these triangular relationships between indices of dingo and fox abundance as evidence that dingo abundance/activity may place an upper limit on fox abundance/activity but is not the sole determinant of fox abundance/activity (Johnson & VanDerWal 2009, Letnic et al. 2011). Unlike previous studies examining the numerical relationships between indices of top predator and mesopredator abundance, primary productivity was in the best model set but it was not so strong as to be in the top three predictor variables that effect fox activity in our generalised linear models (Elmhagen & Rushton 2007, Elmhagen et al. 2010,

Newsome & Ripple 2014).

In our study, baiting was the strongest predictor of dingo activity. The parameter estimates from the most parsimonious model indicated that baiting had a negative effect on dingo activity. Baiting is widely regarded as an effective strategy to mitigate dingo activity at small spatio-temporal scales and this is reflected in our results (Allen

2009). The distance to freehold land was also a useful predictor of dingo activity and was correlated with the occurrence of baiting. This is most likely because the areas of

111 bushland where dingoes are deemed to be a threat to livestock and require control were baited and those are nearest to freehold land. The response of dingoes to NPP was not significant but it was within <2 AICc of the baiting model.

The environmental variable that had most influence on fox activity in our GLMs was baiting, which had a positive effect on fox activity. The positive response of fox activity to dingo control is consistent with the results of previous studies that have investigated the effects that suppression of dingo populations using 1080 baiting, exclusion fencing, shooting and trapping has on abundance indices of foxes (Colman et al. 2014, Hunter et al. 2018, Johnson & VanDerWal, 2009, Letnic et al. 2011,

Letnic et al. 2009, Newsome et al. 2001, Wallach et al. 2010). These studies provide evidence that fox abundance/activity increase in areas subject to intensive dingo control. However, the positive response of foxes to 1080 baiting is counter-intuitive in some respects because 1080 baits are toxic to foxes, and 1080 baiting is widely used as a tool to suppress fox populations, particularly in places where dingoes are rare

(Dexter et al. 2007, Robley et al. 2014).

A hypothesis to explain the positive response of foxes to 1080 baiting is that baiting which is primarily intended to kill dingoes has less of an effect on fox populations than dingoes do. Such a situation could conceivably occur in places where dingo populations are suppressed by baiting, if fox populations are released from direct killing and interference competition from dingoes, and the per-capita rate of killing by dingoes exceeds that of poison baiting (Leo et al. 2015, Moseby et al. 2012). In the scenario just described, fox populations would receive a net benefit from dingo control.

112 An alternative hypothesis to explain the positive response of foxes to 1080 baiting is that fox populations are more resilient than dingo populations to 1080 baiting campaigns, because they are subordinate to dingoes and are released from competition following 1080 baiting campaigns (Marks & Bloomfield 1999, Saunders et al. 1995).

It is plausible that dingoes may monopolise baits which are normally distributed along dirt roads so that baits are taken more so by dingoes than foxes. Such a scenario is supported by studies showing that foxes avoid areas frequented by dingoes such as dirt roads, animal carcasses and sources of free water (Forsyth et al. 2014, Mahon et al. 1998, Mitchell & Banks 2005, Moseby et al. 2012). Another explanation is that because dingoes and foxes have high dietary overlap (Cupples et al. 2011, Mitchell &

Banks 2005) removing dingoes may increase the availability of prey for foxes and thus release foxes from competition for resources with dingoes. Studies comparing the demography, causes of mortality and movements of sympatric dingoes and foxes in baited areas are required to test these hypotheses. Our results provide no basis for the mesopredator facilitation hypothesis (see DeVault et al. 2003, Sivy et al. 2018) because foxes respond positively to dingo control and we would expect the opposite if there was notable relationship between the two predators.

Higher levels of fox activity in areas where dingoes are controlled has implications for the conservation of native wildlife. Foxes have been highlighted as a threatening process to many ground-nesting birds and mammals in south-eastern Australia

(Saunders et al. 2010). If as our results suggest, removal of dingoes results in higher activity/abundance of foxes, then dingo control may be counter-productive from a biodiversity conservation perspective, because it may lead to higher levels of fox predation on native wildlife. Indeed, previous studies have linked dingo control in the

113 forests and deserts of Australia to higher activity levels of foxes and lower abundances of native mammals (Colman et al. 2014, Letnic & Dworjanyn 2011).

4.6 Supplementary material

Supplementary Table 1 Description of important information relating to each site used in the analysis. G – ground, T – trapping, A – aerial and S – shooting. Site Baited? Control Years Year Sampling Plots Plot NPP type baited sampled season (n) nights value 1 Y G 2 2010 Spring 14 98 443 2 Y G 2 2010 Spring 10 70 443 3 Y G 3 2010 Spring 19 76 569 4 Y G, 15 2014 Summer 12 36 194 5 Y G 15 2013 Multiple 28 336 453 6 Y G, T 5 2011 Spring 16 48 509 7 Y G 5 2012 Spring 20 60 206 8 Y G, A 6 2014 Winter 21 252 200 9 Y A 6 2013 Autumn 20 60 523 10 Y G 13 2013 Autumn 20 60 539 11 Y G, S 5 2013 Autumn 20 60 437 12 Y G 10+ 2012 Autumn 21 63 585 13 Y G 15 2014 Winter 12 60 453 14 N NA 0 2013 Multiple 21 252 569 15 N NA 0 2014 Winter 15 150 453 16 N NA 0 2012 Winter 22 66 453 17 N NA 0 2011 Spring 17 51 509 18 N NA 0 2014 Winter 18 216 206 19 N NA 0 2013 Autumn 20 60 511 20 N NA 0 2013 Autumn 20 60 511 21 N NA 0 2013 Autumn 20 60 437 22 N NA 0 2012 Autumn 21 60 585 23 N NA 0 2010 Spring 14 70 569 24 N NA 0 2010 Spring 14 70 688 25 N NA 0 2014 Winter 21 168 200 26 N NA 0 2014 Winter 19 152 200 27 N NA 0 2014 Summer 11 33 221

114

115 Supplementary Table 2 Top competing models describing dingo and fox activity. Mean values (95% CIs) are shown for all parameters. Dingo

Model Model type Intercept (±95% CI) Parameter 1 (±95% CI) Parameter 2 (±95% CI)

Baiting GLM -1.59 (-1.81, -1.37) Baiting -1.06 (-1.49, -0.63) NA Distance to freehold GLM -2.16 (-2.41, -1.91) Dist 0.101 (0.021, 0.181) NA NPP GLM -1.58 (-2.14, -1.02) NPP <-0.001 NA

Fox

Model Model type Intercept (±95% CI) Parameter 1 (±95% CI) Parameter 2 (±95% CI)

Baiting GLM -3.75 (-4.41, -3.09) Baiting 1.98 (1.26, 2.70) NA Distance to freehold GLM -1.75 (-2.17, -1.33) Dist_freehold -0.53 (-0.88, -0.18) NA Dingo GLM -1.96 (-2.37, -1.55) Dingo -3.55 (-6.45, -0.65) NA NPP GLM -2.85 (-4.06, -1.64) NPP <0.001 NA

116 5. Reintroducing Tasmanian devils to mainland Australia can restore top-down control in ecosystems where dingoes have been extirpated

117 5.1 Abstract

Restoring missing ecological interactions by reintroducing locally extinct species or ecological surrogates for extinct species has been mooted as an approach to restore ecosystems. Australia’s top predator, the dingo, is subject to culling in order to prevent attacks on livestock. Dingo culling has been linked to ecological cascades evidenced by irruptions of herbivores and introduced mesopredators and declines of small and medium sized mammals. Maintenance of dingo populations is untenable for land-managers in many parts of Australia owing to their depredations on livestock.

However, it may be possible to fill the top predator niche with the Tasmanian devil which has less impact on livestock. Devils once occurred throughout Australia, but became extinct from the mainland about 3000 years ago, but are now threatened by a disease epidemic in Tasmania. To explore the feasibility of reintroducing devils to mainland Australia we used species distribution models (SDM) to determine if suitable climatic conditions for devils exist and fuzzy cognitive mapping (FCM) to predict the effects of devil reintroduction. Based on devils’ current distribution, our

SDM indicates that suitable areas for devils exist in south-eastern Australia. Our FCM examined ecosystem responses to predator-management scenarios by manipulating the abundances of devils, dingoes and foxes. Our FCMs showed devils would have cascading effects similar to, but weaker than those of dingoes. Devil introduction was linked to lower abundances of introduced mesopredators and herbivores. Abundances of small and medium sized mammals and understorey vegetation complexity increased with devil introduction. However, threatened species vulnerable to fox predation benefited little from devil introduction. Our study suggests that reintroducing ecological surrogates for top predators may yield benefits for biodiversity conservation.

118 5.2 Introduction

All species interact with other species both directly and indirectly, through mechanisms such as predation, competition, mutualism, facilitation and herbivory.

The outcomes of such interactions are among the major factors that shape ecological communities. Species are described as strongly interactive when their absence leads to significant changes in ecosystems and as keystone species when the strength of their interactions is disproportionate to their abundance (Soule et al. 2003).

Increasingly, ecologists are realizing that the effects of strongly interactive species could be used to manipulate ecological processes and achieve biodiversity conservation goals (Seddon et al. 2014, Ritchie et al. 2012). Understanding the important role that strongly interactive species can play in shaping ecosystems has prompted debate about the merits of restoring missing ecological interactions by reintroducing locally extinct species or ecological surrogates for now extinct species

(Donlan et al. 2006). Reintroducing strongly interactive species may also be desirable to curb the effects of invasive species, or to fill vacant niches in novel ecosystems that are dominated by suites of invasive species.

Large carnivores are often keystone species because they typically occur at low densities and may exert top-down control on ecosystems through their predatory and competitive interactions with herbivores and smaller predators (Ripple et al. 2014).

Because top predators kill their prey and frequently kill their competitors as well, the most obvious population-level effects of top-predator removal are increases in the abundance of large herbivores and smaller predators (mesopredators) (Ripple et al.

2014).

119 While the direct predatory effects of top predators are relatively easily observed, they typically have strong indirect effects as well. For example, numerous studies have found that, in the absence of a top-predator, herbivores increase in abundance, reducing the biomass of palatable plants (Ripple et al. 2014). A similar cascade of effects is predicted by the mesopredator release hypothesis (MRH). The MRH postulates that reduced abundance of top-order predators results in an increase in the abundance and predatory impact of smaller predators (Soulé et al. 1988).

Consequently, small prey species that are the preferred prey of mesopredators may decline in abundance (Prugh et al. 2009). Owing to the amplified interactions of herbivores and mesopredators, terrestrial ecosystems from which top predators have been removed tend to be characterized by reduced diversity of small vertebrates and shifts in vegetation composition (Letnic et al. 2012, Ripple et al. 2014), but these trends are not evidenced in all cases (Letnic et al. 2012).

Australia’s largest mammalian carnivore is the dingo (Canis dingo). Dingoes arrived on mainland Australia 5000-3500 years before present (ybp) (Letnic et al. 2014).

Around the time that the dingo arrived on mainland Australia, two marsupial predators, the Tasmanian tiger (Thylacinus cynocephalus) and Tasmanian devil

(hereon in devil) (Sarcophilus harrisii) became extinct from mainland Australia but persisted on the island of Tasmania which dingoes never colonised (Letnic et al.

2014). The coincidence in the timing of the arrival of the dingo and extinction of the

Tasmanian tiger and devil, coupled with the fact that larger dingoes were likely to be superior competitors in one on one agonistic interactions lends support to the idea that dingoes likely contributed to the extinction of these marsupial predators from

120 mainland Australia (Letnic et al. 2012). This hypothesis is supported by observations that devils are easily killed by domestic dogs (Jones et al. 2003).

It has also been proposed that increasing human population densities and or climate, not dingoes were the main driver of the extinction of the Tasmanian tiger and devil from mainland Australia (Johnson & Wroe 2003). Recent analysis of mitochondrial

DNA reveals that drought associated with a severe El Niño Southern Oscillation climatic event from 3000-5000 ybp coincided with the decline of devil populations

(Brüniche-Olsen et al. 2014). Extreme drought may have pushed devil populations to sufficiently low levels that a “predator pit” (Sinclair et al. 1998) ensued whereby predation by humans and dingoes prevented the recovery of devil populations.

Because they kill livestock, dingoes are widely perceived as a pest and their populations are controlled in many parts of Australia using poisoned meat baits, trapping and shooting (Allen et al. 2015). However, there is evidence also that dingoes are a strongly interactive top predator and that suppression of dingo populations by humans can drive shifts in the abundances of other species (Letnic et al. 2012). Sustained efforts to eradicate dingoes have resulted in irruptions of herbivores such as kangaroos and wallabies (macropods) (Robertshaw & Harden

1986, Colman et al. 2014, Allen 2015) and have, in some but not all cases, been linked to increases in the abundance of invasive mesopredators, the red fox (Vulpes vulpes) and feral cat (Felis catus) (Gordon et al. 2015, Allen et al. 2015, Colman et al.

2014, Johnson & VanDerWal 2009, Letnic et al. 2011). In turn, predation by foxes and depletion of understorey vegetation by macropods have been identified as drivers of population decline in small and medium sized mammals in forest ecosystems

(Robley et al. 2014, Colman et al. 2015, Foster et al. 2015).

121 Here we explore the feasibility of reintroducing devils to forest ecosystems in south- eastern Australia in order to stem the changes to ecosystem structure associated with lethal control of dingoes. Because dingoes are a major pest to livestock producers the maintenance of dingo populations is an untenable option for wildlife managers in many regions of south-eastern Australia. However, another option available to wildlife managers may be to fill the top predator niche with the devil. While devils, owing to their smaller body size and slower metabolism, are unlikely to occupy exactly the same niche as the dingo, theory and evidence from Tasmania where they remain extant suggests that devils have similar positive effects on small and medium sized mammal species as dingoes do (Hollings et al. 2013). This is because devils can have suppressive effects on the populations and activity patterns of native herbivores and invasive mesopredators (Hollings et al. 2013, Jones et al. 2007, Fancourt et al.

2015). Indeed, top down effects of devils has been mooted as one of the reasons why red foxes have not flourished in Tasmania despite several introductions (Jones et al.

2007). In addition, the failure of foxes to establish is Tasmania is thought to be one of the reasons why small and medium sized native mammals are so abundant there today

(Johnson 2006).

Importantly, devils are not as great a threat to livestock as are dingoes, although their predatory impact is at least similar to that of foxes with young sheep at risk (Jones et al. 2003), and thus their reintroduction is likely to obtain more support from livestock producers than would the maintenance of dingo populations. Since the turn of the 21st century, devils have undergone massive population decline due to an epidemic of devil facial tumour disease (DFTD) (Hollings et al. 2013). Thus, a mainland population of DFTD-free devils would also function as an insurance population in the event that the Tasmanian population became extinct as a result of DFTD or another

122 threatening process. It is conceivable also, that a reintroduction of devils to the mainland may succeed because two of the purported drivers of devil decline from the mainland during the Holocene, dingoes and hunting by Aboriginal people, are now diminished across much of the mainland and thus may not pose a threat to reintroduced devil populations.

We use two approaches to evaluate the potential to reintroduce devils to south-eastern

Australia. Because devils are now restricted to the island of Tasmania which has a temperate climate, it is possible that extant devils may be poorly adapted to the warmer environments of mainland Australia. Given this possibility, it makes sense to initially reintroduce devils to areas that are climatically similar to Tasmania. Thus, we used species distribution models (SDM) to identify areas that are climatically suitable for devils based on their existing distribution in Tasmania.

Our second aim was to use fuzzy cognitive maps (FCM) to predict the ecological outcomes of predator management scenarios. FCMs predict the outcomes of interactions among multiple input species (Dexter et al. 2009) and are a risk-free, low input method of testing novel management scenarios. The scenarios we investigated involved reintroducing devils in the absence of dingoes and reintroducing devils with dingoes present. We also modelled presently existing scenarios such as intensive dingo and fox baiting to use as comparative scenarios.

123 5.3 Methods

5.3.1 Reintroduction viability using species distribution modelling (SDM)

We used SDM to determine if climatic conditions are favorable for devil reintroduction to the south-eastern Australian mainland. We used 1469 presence records for S. harrisii in Tasmania sourced from the Atlas of Living Australia

(Accessed 10 August 2013).

Current climate data was sourced from the WorldClim website which is free to access

(www.worldclim.org). Current data ranges for WorldClim layers span from the years

1950 to 2000 and are based on Bureau of Meteorology (BOM) records. We used

Maxent (Maxent ver. 3.3.1) to generate an output of suitable devil locations on the mainland.

Outputs were restricted to the Australian mainland as it is the focus area of this study

(Fig. 5.1). We used recent climate scenarios only (1950-2000) to determine suitable areas of devil reintroduction.

124

We initially ran the model with all 19 bioclimatic variables and we then evaluated the output to determine which layers contributed most to the model. Bioclimatic variable number 1 (mean annual temperature) contributed strongly to the model. We then ran variable 1 with a combination of other variable layers and found number 1 best served our purpose on its own.

The fit of our model was evaluated using the AUC (area under curve), where the closer the value is to 1 the better the fit. A value of 0.5 indicates a model fit no better than random (Elith et al. 2006).

5.3.2 The fuzzy cognitive model (FCM)

Quantitative information about a species’ ecology or natural history is often unknown.

Nevertheless, scientists may know whether the effect of one species on another is

125 strong, moderate or weak, and whether it is a positive or negative effect. Although not numerically precise, this information can be harnessed to produce meaningful ecological information for land managers (Ramsey & Veltman 2005). When used in conjunction with fuzzy cognitive models (FCM), this type of data can provide powerful qualitative modelling of ecological systems that are highly complex or where data are uncertain (Dambacher et al. 2003).

Our particular model is a FCM implemented stochastically as a fuzzy Markov-type chain. FCMs represent concepts and their pair-wise interactions as the nodes and edges of mathematical graphs (Kosko 1986). In our study, the nodes are species and edges represent interaction strengths between -1 and 1. These strengths can occasionally be determined with high precision but are more often mapped fuzzily by the linguistic estimates “very weak”, “weak”, “moderate”, and “high” to the numeric strengths 0.1, 0.2, 0.5, and 0.8, respectively. The graph can be represented by a matrix

푬 in which each entry 푒푖푗 indicates the interaction strength that species 푗 has on species 푖. Blank spaces are interaction strengths that we have not been able to determine. In the model, these spaces are set to 0, with small stochastic perturbations applied at each step. As demonstrated in Ramsey & Veltman (2005) and further developed in Ramsey & Norbury (2009) and Dexter et al. (2012), this matrix representation can be used to add predictive power to the FCM. We further refined our model, making it more precise, by allowing each entry to assume a set of values such as a single value or an interval of values, depending on the data available.

Representing the species population distribution as a vector s_i at time step i, normalised relatively to the interval [0,1] with each species population represented by their percentage of the sum of populations, it is simple to calculate the relative

126 population distribution 풔풊+ퟏ at time step 푖 + 1 as a Markov chain process. As in the studies cited above, our study contains inherent uncertainties and these are represented in the model as randomising effects to be applied at each iteration. For these effects, we extended the lower and upper bounds of each interaction strength interval 푒푖푗 by 0.1. At each iteration, an element of this interval is chosen at random to represent 푒푖푗, and performing this randomisation to each matrix entry gives a randomised effect matrix 푅(푬). Our iteration is then given explicitly as

풔풊+ퟏ = 휆푓(푅(푬)풔풊) + (1 − 휆)풔풊 where 휆 is a relaxation parameter to speed up the asymptotic convergence of the population distributions. As the activation function f, used to limit the relative population numbers to the interval [0,1], (Dexter et al. 2012) suggest using a logistic function. We chose a logistic function that approximates the identity function on the interval [0,1], namely the function below which is illustrated in Fig. 2:

1 푓(푥) = 1 −4(푥− ) 1+푒 2

For each of the 5 scenarios, we calculated a sequence of 1000 iterations 100 times and found the 1000 iteration values, averaged over the 100 runs following Dexter et al.

(2012). However, we found that this was not required in our study.

Figure 5.2 The activation function 푓.

127 5.3.3 The agents (species and vegetation)

The SDM identified suitable habitats for devils exist in south-eastern Australia (Fig.

5.1). The areas identified in the models consists primarily of forest habitats dominated by Eucalyptus spp. (Keith 2004) and lands that have been cleared for agricultural purposes. Because the most suitable habitats for devils existed in forested regions, we calculated interaction strength for our FCMs using data from published studies that have been conducted in this region.

The agents chosen for modelling in this study include species that were likely to be strongly influenced by devil reintroduction and fox and dingo predation. To improve the quality of the model we set an initial abundance value for every agent.

Initial abundances of agents were calculated using data from a study reporting the effects of dingo control on forested ecosystems (Colman et al. 2014) by multiplying interaction strengths by abundance data. Abundance was standardised to a per kilometer metric.

We grouped all palatable vegetation below waist height as “vegetation complexity” for our analysis. Multiple small dasyurid species were frequently captured (e.g. A. flavipes and A. stuartii) and because of their similarity we have grouped these as

‘Antechinus’. Similarly, red-necked wallaby (Macropus rufogriseus) and swamp wallaby (Wallabia bicolor) are grouped as ‘wallaby’. We included two bandicoot species in this study, southern brown bandicoot (Isoodon obesulus) and long-nosed bandicoot (Perameles nasuta), and these have been grouped as ‘bandicoot’.

5.3.4 Data extraction

128 Most of the interaction strengths we used to populate our FCM were gleaned from the literature using a meta-analytic approach comparing the effects of predation on other species in the forested ecosystems of south-eastern Australia (Supplementary material). Raw data was used to extract t or f values if they were not readily available.

All interaction values were then converted to product moment correlation (r) values in

MetaWin (http://www.metawinsoft.com). Each r value represents the interaction strength between a pair of agents (Table 5.1). For instances where published data was scant or absent between a pair of agents we inferred interaction strengths using the literature or expert opinion (Supplementary material).

129 Table 5.1 Interaction matrix populated with interaction strengths (r)

130 Where information was especially scant for the effect of devil on bush rat (Rattus fuscipes) and Antechinus spp. we used published information on the diet of devils to infer weak interaction strengths based on the notion that predation on these species may have negative effects on their populations (Supplementary information). We converted the frequency of occurrence of small mammal in devil diet (<1%) to a correlation (r) value of 0,-0.1 (Jones & Barmuta, 1998).

All intraspecific interactions were set at 0.1 which assumes that all conspecifics have a slightly positive effect on one another at normal population densities based on the notion that conspecifics are required for reproduction.

5.3.5 Model scenarios

Baseline Scenario - No dingo control, no fox control

This scenario is the null scenario against which the effects of all other scenarios were compared (Fig. 5.3). It represents areas of National Parks that do not conduct any dingo baiting and we assume dingoes exist at ecologically effective densities. We modelled this scenario by not capping the abundance of any species and allowing them to ‘naturally’ interact.

Scenario 2 – Complete dingo eradication and complete fox eradication

This scenario simulates intensive control of both dingoes and foxes such that their populations are below densities at which they are ecologically effective (sensu Soule et al. 2003). In a field situation this would be achieved for dingoes by targeted leg hold trapping and baiting with 1080 (sodium fluroacetate) dosed meat baits. Nominal dose for dingoes is 6 mg of pure 1080 per bait (Fleming & Parker 1991). Intensive fox control is achieved with 1080 meat baiting, nominal dose 3 mg of 1080 per bait

131 (Claridge et al. 2010, Marks & Wilson, 2005). In the model we simulate this scenario by holding dingo and fox abundance at 0.

Scenario 3 – No dingo or fox control, devil reintroduced

This scenario simulates introducing devils into ecosystems where neither dingoes or foxes are controlled and assumes that devils can co-exist with dingoes. In the model we achieved this scenario by not capping dingo or fox abundance and by adding devils to the model.

Scenario 4 – Complete dingo eradication, reduced fox abundance

This scenario simulates control activities directed towards dingoes but not foxes. Such situations can arise when dingo control programs are undertaken with the purpose of protecting livestock but foxes are not targeted. Under this scenario dingo control is intensive as outlined in scenario 2, however, some fox mortality is expected to occur as a result of foxes consuming baits laid for dingoes (Fleming & Parker 1991). To create this scenario in the FCM we limited dingo abundance to 0 and fox abundance to 50% of unbaited levels.

Scenario 5 – Complete dingo eradication, reduced fox abundance and devil reintroduced

This scenario explores the effects of introducing devils to areas where dingoes have been removed and some foxes are killed by dingo control activities. It is the same as scenario 4 but with no cap on devil abundance. We’ve assumed minimal unintended devil fatalities as a result of dingo baiting since devils are 38 times more resistant to

1080 than dingoes (Mcilroy 1981). Additionally, burying baits to 15 cms reduces the likelihood of devils removing baits (Hughes et al. 2011).

132

133 5.4 Results

5.4.1 Habitat suitability of devils on mainland

The final SDM model AUC value was 0.927, representing a good fit. The model indicates that under current conditions habitats of high suitability occur mainly in south-eastern Australia (Fig. 5.1). Many areas where devils are climatically suited coincide with areas where dingoes are rare or absent (Fig. 5.1b).

5.4.2 FCM model

The convergence of each thousand-value set was rapid and stable, even for high values of 휆, with the maximal average absolute difference between two of any sequential species populations being less than 0.8%, and most differences being far smaller. This convergence and stability was due to the high ratio of matrix entries in 푬 that we determined.

5.4.3 FCM scenarios

Scenario 2 – Complete dingo eradication and complete fox eradication

The predicted effect of removing dingoes reduced vegetation complexity by 25% while wallaby numbers grew 95% (see Table 5.2 for species response values).

Antechinus and bush rat numbers fell 31% and 34% respectively. Brush-tailed rock wallaby (Petrogale pencillata) numbers increased (119%) while Spotted-tailed quolls

(Dasyurus maculatus) (hereon in quolls) also increased markedly (68%). Macropod species benefited the most under this scenario (Table 5.2, Fig. 5.4).

134 Table 5.2 Output values from FCM displayed here as percentage shift from null state (scenario 1).

Scenario 3 – No dingo or fox control, devil reintroduced Predator Scenario Scenario 1 Scenario2 Scenario 3 Scenario 4 Scenario 5 Baited dingo * * * Baited fox * Reduced fox * * Tasmanian devil * * Model Agents Veg. complexity 0% -25% +7% -18% -6% Kangaroo 0% +73% -10% +38% +13% Wombat 0% -27% -25% -15% -46% Wallaby 0% +95% -4% +64% +43% Rock wallaby 0% +119% -1% -1% 0% Fox 0% -100% -14% +124% +68% Cat 0% 0% -13% +5% -19% Brushtail possum 0% +56% 0% +4% +6% Spotted-tailed quoll 0% +68% +9% +5% +19% Greater glider 0% -9% +10% +10% +26% Bandicoot 0% -31% +15% -39% -18% Ringtail possum 0% +23% +17% +3% +33% Bush rat 0% -34% +10% -50% -35% Antechinus 0% -31% +8% -47% -33% Dingo 0% -100% -9% -100% -100% Tasmanian devil 0% 0% +62% 0% +100% This scenario exhibited an overall reduction of extreme cascades evident in the model output by low shifts in the abundance of the model agents (Table 5.2, Fig. 5.4). Cat and fox abundance decreased in the presence of the devil by 13% and 14% respectively (Table 5.2, Fig. 5.4). Greater gliders (Petauroides volans), bandicoots and ringtail possums (Pseudocheirus peregrinus) all increased by at least 10% (Table

5.2, Fig. 5.4). Rock wallabies and wallabies showed only negligible decreases (1-4%).

135 Vegetation complexity benefits most under this scenario likely attributed to lower abundances of medium and large herbivores .

Scenario 4 – Complete dingo eradication, reduced fox abundance

Removing dingoes resulted in population growth of most macropod species (Table

5.2, Fig. 5.4). Despite applying a 50% constraint on fox growth, numbers increase markedly in this scenario (Table 5.2, Fig. 5.4). Cat abundance increases 5% (Table

5.2, Fig. 5.4). All terrestrial small and medium sized mammal species experience negative growth.

Scenario 5 – Complete dingo eradication, reduced fox abundance and devil reintroduced

Macropods mostly respond positively to this scenario (Table 5.2, Fig. 5.4). Cats declined 14% compared to the same scenario without devils. Fox abundance almost halved compared to the same scenario without devils. Bandicoot, ringtail possum,

Antechinus, bush rat and greater glider abundances increased compared to the same scenario without devils (Table 5.2, Fig. 5.4).

5.5 Discussion

Our SDM model suggests that it would be feasible to reintroduce devils to the forests of south-eastern Australia and establish a disease-free “insurance” population because climatically suitable areas exist there (Fig. 5.1). FCM modelling suggests that once established, devils could assume a similar ecological role to dingoes in places where dingoes have been eradicated (Fig. 5.1) by suppressing the abundances of invasive mesopredators and wallabies (Fig. 5.4).

136

137 A scenario that we did not consider in our FCM exercise was that devils may be pre- adapted to the climate of mainland Australia. Such a scenario is not unrealistic because devils were found throughout mainland Australia during the Holocene and only became extinct less than 3000 years ago (Letnic et al. 2014). Further research could use mechanistic models to identify areas on the mainland where devils could thrive

(Kearney et al. 2010). If devils are pre-adapted to the mainland climate, we would expect them to have little difficulty becoming established in the areas of south-eastern and south-western Australia where dingoes have been eradicated and hunting of native mammal species by humans is highly regulated. However, if dingoes were the cause of devil extinction from mainland Australia, devils may be unlikely to establish populations in regions where dingo populations remain (Letnic et al. 2012). If a formal devil introduction was to be undertaken, a reintroduction strategy would need to consider ways to minimize potential competitive effects of canids on reintroduced devils, particularly for founding populations of devils.

Testament to the strength of our FCM is its accurate emulation of empirical field observations for the currently existing and testable scenarios 2 and 4. In accord with field observations, our model demonstrates that cascading effects ensue once dingoes are completely eradicated. Notably, foxes became more common as expected under the

MRH (Brook et al, 2012, Johnson & VanDerWal 2009). In accord with trophic cascade theory and field studies, complete eradication of dingoes and foxes was associated with increased abundance of macropods and a reduction in vegetation complexity (Colman et al. 2014, Dexter et al. 2013).

In our modelling the complete eradication of dingoes and foxes increased abundances of threatened species, the spotted-tailed quoll and rock wallaby, which is consistent with field studies reporting the effects of fox baiting on rock wallabies (Kinnear et al. 1988). Although canid baiting is the status quo management option for threatened species management in much of south-eastern Australia, our modelling suggests that canid eradication can result in the decline of some native mammal species. In our FCM canid eradication resulted in decreased abundances of small and medium sized mammals such as bush rats, Antechinus spp. and bandicoots due to increased impacts of introduced mesopredators and macropods. However, we caution, that although our models do suggest shifts in mammal abundances consistent with theory and previous field studies, a shortcoming of our FCM is that not all strengths of interactions are known and predictions are limited to the species included in our models.

One way to restore top-down control of herbivores and introduced mesopredators in areas where dingoes have been extirpated may be by reintroducing devils (Ritchie et al.

2012). Under scenario 5 where we simulated the reintroduction of devils under the assumption that dingoes and devils could not coexist, the abundances of red foxes, macropods and wombats (Vombatus ursinus) decreased. Conversely, the abundances of

Antechinus, ringtail possums, bandicoots, greater gliders, brushtail possums

(Trichosurus vulpecula) and vegetation complexity all increased relative to the same scenario without devils.

Our results suggest that devils can fulfill a similar ecological function to dingoes and are consistent with field studies from Tasmania where devils have been reported to suppress the abundances of feral cats and small macropods (Hollings et al. 2013, but see also Fancourt et al. 2015). However, devils’ overall effects on ecosystems are expected to be weaker than those of dingoes due to their smaller body size and lower metabolic demands (Letnic et al. 2014). Nonetheless, the failure of red foxes to establish large populations in Tasmania despite similar climatic conditions to mainland areas where foxes have thrived in the absence of dingoes suggest that there is considerable merit in

139 the idea of using devils to suppress the abundance and impacts of red foxes (McCallum

& Jones 2006, Jones et al. 2003). However, it is important to note that the idea that devils suppress foxes is not based on observations between the two species in the field, but is rather the most plausible hypothesis mooted so far to explain why foxes have not thrived in Tasmania (Jones et al. 2007, Ritchie et al. 2012). Experiments could be conducted to explore devil and fox introductions but would need to overcome prohibitive ethical, legal and logistic impediments.

Under scenario 3 where we assumed that dingoes and devils can coexist, the changes in the abundance of the model agents were not as extreme as under other scenarios (Fig.

5.4) which may suggest overall cascade strength is dampened with increasing diversity within the predator guild. The importance of maintaining predator diversity (Finke &

Denno 2004) and effective population densities of predators (Soule et al. 2003) in order to mitigate ecological degradation, although often discussed, has not been thoroughly tested in Australia. One explanation for dampening of trophic cascades with increasing predator diversity in our simulations is that more predator diversity increases competition and predation among guild members and this competition nullifies their direct effects on prey species (Finke & Denno 2004).

5.5.1 An ecological case for reintroducing devils to mainland Australia

Our modelling is consistent with results of field studies which suggest that manipulating the abundances of large carnivores can shift mammal assemblages to alternate states dominated by herbivores and mesopredators (Estes et al. 2011, Ripple et al. 2014).

Which of these states is most desirable is subjective. For example, scenarios 2 and 4 represent predator management regimes undertaken in some National Parks of south- eastern Australia where the aim of predator control is protecting threatened fauna

140 (Robley et al. 2014, Kovacs et al. 2012) and controlling dingoes to protect livestock on nearby properties (Colman et al. 2014), respectively. Unfortunately, both of these scenarios are characterised by extremes in species abundances at various trophic levels demonstrating poor equitability. The threatened species included in our simulations, quolls and rock wallabies, both responded positively to intensive predator control.

However, under this scenario cats and wallabies increased in abundance and small mammals declined directly as a result of predation from cats but also indirectly, since wallabies decrease available cover for small mammals (Dexter et al. 2013).

From a biodiversity conservation perspective scenario 3 benefits many species since dingoes are not eradicated, devils are present in the ecosystem and foxes and cats are suppressed and there is positive population growth for most small and medium native mammal species and vegetation complexity. Despite positive biodiversity outcomes, scenario 3 provides nearby livestock producers with limited protection from dingo attacks because dingoes remain present. Scenario 5 on the other hand sees devils introduced into areas with few canids. The addition of devils is associated with increased abundances of most small and medium mammal species and vegetation complexity increases also (Fig. 5.4). Under scenario 5, the effects of foxes, cats and macropods are ameliorated without ceasing dingo control. However, the most threatened species in our simulated system, quolls and rock wallabies, differ little from scenario 1 because these species are in the optimal prey size range for both foxes and devils (Jones, 1997, Jones & Barmuta 2000). Scenario 5 may also be desirable to livestock farmers because dingo spill-over from public to private land is managed.

Our modelling demonstrates some of the cascading effects of dingo removal may be mitigated by restoring lost predator function with devils. Scenarios with devils tended to have the most equitable assemblages exhibiting less extreme cascades. However,

141 introducing devils to the system provided little benefit for the most threatened mammal species, because devils also predate on many of these species (Jones 1997, Jones &

Barmuta 2000). Hence from a management perspective, devil introduction appears likely to provide considerable benefits for many but not all species subject to predation by red foxes and cats. When contemplating devil reintroduction managers need to be mindful of whether the purpose of their predator management regime is to establish a mainland insurance population of disease-free devils, protect livestock while restoring top-down control or the recovery of threatened species.

5.6 Supplementary Material

142 Supplementary Table 1. Fox literature cited to populate FCM

Supplementary Table 2. Dingo literature cited to populate FCM

144 Supplementary Table 3. Devil literature cited to populate FCM

145 Supplementary Table 4. Diet data derived for Tasmanian devil used to populate FCM

146 Supplementary Table 5. Justification and reference information for inferred values

147 5.7 Acknowledgements

The authors would like to thank Dr Adam Pérou Hermans who proofread the manuscript, Dr

Nick Murray and Kimberly McCallum who provided some interesting insights with SDM modelling and the reviewers whose reviews greatly improved the manuscript. ML and MJ were supported by funding from the Australian Research Council. DOH was also supported by the Blue Mountains World Heritage Institute top-up scholarship.

6. General Discussion

149 6.1 Synthesis

My thesis provides evidence to support the notion that lethal dingo control may initiate state shifts within forest ecosystems. Shifts in mammal assemblages can manifest both directly and indirectly with many small and medium-sized threatened native mammal species thriving in areas of no dingo control. My findings provide support for the notion that dingoes play an important role as a keystone species in forest ecosystems. Accurately elucidating all of these effects and how they manifest requires more research.

My meta-analysis chapter demonstrates that overall, dingo control programmes increase abundances of macropods but decrease abundances of many small and medium-sized native mammal species. Given these findings, I contend that, in many situations, dingo control programmes are counter-productive for biodiversity conservation efforts. Further, I show that dingoes appear to alter understorey vegetation complexity <100 cm, providing support for the notion that dingo presence/absence drives a top-down trophic cascade in my study systems.

Finally, I present a case for the reintroduction of the Tasmanian devil, a potential top- predator surrogate, to suitable forested areas of Australia. The aim of this is to provide a last resort example of how we may potentially restore missing ecological interactions resulting from intense dingo persecution.

6.2 Chapter 2: Not all predators are equal: a continent-scale analysis of the effects of predator control on Australian mammals.

The first chapter of this thesis describes ecological trends that have been documented following lethal control of dingoes from forest ecosystems. One of the most important potential benefits of maintaining top predator populations is their beneficial effects on

Australia’s small and medium-sized native mammal species. Interestingly, some studies report very different outcomes for mammal species following lethal dingo control (Allen et

150 al. 2014, Colman et al. 2014). Foxes are also the subject of lethal control in many areas of

Australia, especially where dingoes have been removed (Robley et al. 2014), because they are known predators of many of Australia’s most threatened native mammal species. In order to elucidate the outcomes of lethal canid control we conducted meta-analyses to provide objective information for conservation managers.

Data were gleaned from studies reporting responses of mammals to either dingo or fox poison baiting control programs. This is important because it can be difficult to parse out the relative importance of co-existing introduced predators’ effects on their prey species. Our study contained 125 effect sizes from 34 papers published between 1980 and 2017. We used log response ratio (lnRR, the natural logarithm of the ratio between the two means) as our effect size, because it is more suitable measure of effect than Hedges’ d when sample sizes are small (Friedrich et al. 2008, Hedges et al. 1999).

My results provide evidence that lethal control of dingoes and lethal control of foxes have different outcomes on other mammal species. The effects of removing both dingoes and foxes scaled with the body size of prey species, but in different ways. Dingo removal resulted in increased abundance of mammals weighing more than the CWR. However, the opposite was true for mammals weighing less than the CWR, their abundances declined significantly when dingoes were removed. In contrast, fox removal had strong and significant positive effects for mammals in the CWR. Our analyses show no significant effect of dingo removal on cats or foxes but where dingoes were removed there was a trend for foxes to increase in abundance. In instances where foxes were removed, cats tended to increase in abundance which suggests they may be suppressed not only by dingoes but foxes as well and this is corroborated in our results. This study demonstrates that lethal control of dingoes and of foxes have different and unintended consequences for native and introduced mammals.

151 6.2.1 Limitations and recommendations

Meta-analyses provide objective and useful information to end users but they do have limitations. One limitation of our study was associated with the contrast which we tested

(dingo/fox removal), as it assumes that the lethal control was always effective. In many cases where results reported the effect of baiting on the target predator, we were able to demonstrate efficacy. However, this was not always the case and thus raises concerns about how reliable the data for the effect of predator removal on prey species is because there is no data demonstrating how effective baiting was. Therefore, I recommend that future studies investigating the effects of baiting on dingoes and foxes report the effect of baiting so readers can discern how the relative effect of baiting influences other species reported in the same studies.

Furthermore, this study demonstrates that dingo control increases abundances of large grazing species like kangaroos. One of the main reasons land managers claim to lethally control dingoes is that it reduces the financial burden on livestock graziers owing to reductions in livestock losses caused by dingoes. However, to the best of my knowledge no studies have directly measured the net cost to graziers resulting from controlling dingoes to mitigate stock losses compared to the cost associated with the irruption of native grazing species resulting from dingo control (Prowse et al. 2015). It is possible that future studies could show that maintaining dingoes at ecologically effective densities could actually be mutually beneficial for graziers and ecosystems. This is because net grazing from native herbivores could be reduced and this cost saving to livestock farmers would be greater than if dingoes were locally extirpated and native grazing species remained at high densities. My study shows that dingoes provide benefits to various native mammal species and so any studies providing support for the idea that the maintenance of dingoes in the landscape can

152 actually be financially beneficial for livestock graziers has the potential to improve the conservation of many native mammal species.

153 6.3 Chapter 3: Top predator removal affects mammal species differently across multiple trophic levels in forest ecosystems.

In this chapter I expand on my findings in the meta-analysis by further investigating how top predator removal affects a forest ecosystem over multiple years. This is timely because drawing a link between top predator removal and ecosystem simplification has been the focus of much ecological research in recent times (Ripple et al. 2014). In terrestrial ecosystems, research has shown that top predators can influence ecosystems and their species composition owing to direct effects on smaller predators and their prey species (Ripple et al. 2014). These direct effects have been linked to far-reaching indirect effects at multiple trophic levels, that extend to vegetation and even the physical attributes of the landscape (Beschta & Ripple

2009, Morris & Letnic 2017, Lyons et al. 2018). The role of top predators is increasingly being viewed as an important component of healthy and functional ecosystems around the world.

In Australia the three most common eutherian predators are all introduced. The dingo however has existed in Australia for >3500 ybp, much longer than both the cat and fox (<250 ybp). Since its arrival, the dingo has become the dominant top predator and has adopted this role in all Australian terrestrial ecosystem types (Letnic et al. 2012a). Studies investigating the effects of dingoes on mammals in both arid and mesic systems often report different outcomes (Hunter et al. 2018). This study is novel because it is the first multi-year investigation into these effects in forests in recent years. This is important because it remains possible that previously reported differences in ecosystem attributes between baited and unbaited sites were anomalies that may not persist through time. Using occupancy as my measurement form I found dingoes and bandicoots both responded negatively to dingo baiting. Cats on the other hand responded most strongly to location. Macropods, possums and

154 foxes did not respond to dingo baiting. Dasyurids were negatively affected by dingo baiting too but we gauged this effect using generalised linear mixed modelling and not occupancy analysis.

Since the dingo has arrived in Australia it has contributed to shaping ecosystems via its direct and indirect effects on the biota (Colman et al. 2014, Letnic et al. 2014, Letnic et al. 2012b,

Newsome et al. 1983). It has previously been demonstrated that dingoes can confer a suite of positive benefits to ecosystems and their species. However, this study demonstrates that these effects are not ubiquitous or easily predictable. Owing to the myriad of confluent interactions, especially from invasive mesopredators, the lethal control of dingoes in some forest ecosystems is shown here to influence species occupancy in very different ways. This may be due in part to forest ecosystems being relatively more productive environments and ecosystems with higher productivity tend to have more interaction pathways making effects difficult to parse out (Finke & Denno 2004).

6.3.1 Limitations and recommendations

In forests and woodlands top predators use trails to move through the landscape when they’re available (Hayward et al. 2014). When a top predator is present in the landscape, as they were at both baited and unbaited sites in my study, a mesopredator is likely to avoid using these same trails in areas and at times that increase the threat to the mesopredator. My cameras were set along trails and it therefore stands to reason that dingo occupancy estimates would be higher than that for foxes, and this was the case. This is of interest because ecological theory would suggest that populations of the smaller mesopredator should be more abundant in the landscape than the top predator. One possible solution that comes to mind in light of potential spatio-temporal partitioning of track use between the top predator and mesopredator would be to set camera locations randomly throughout an area. However, random dispersal of

155 cameras would certainly reduce detections and have the potential to create issues with the analysis deriving from sparse detections.

My study did not involve the experimental manipulation of dingo abundances and this may be criticised for not being a manipulative experiment. Previous studies with similar methodologies have been criticised on the basis that results are founded on correlation and not causation. I argue that it is logistically difficult and ethically undesirable to manipulate

(lethally) dingo populations for the purpose of research when it is possible to capitalise on pre-existing control programs. Results using methodologies such as these remain important and valid because they often investigate ecological processes described by ecological theory that cannot be explored using replicated experimental manipulations (Oksanen 2001, Moseby et al. 2018).

I decided to generate an occupancy estimate and not run index of true abundance. Occupancy may not be the most effective method of gauging interactions between top predators and mesopredators as research suggests that it is at the upper limit of abundance indices where dingoes appear to affect mesopredators most noticeably (Nimmo et al. 2015). However, I was interested in learning how a suite of other species reacted following dingo control and this was the reason I chose to use the occupancy analysis. In hindsight, I had issues with the reliability of cameras which reduced my overall detections. Additionally, all my cameras were set along tracks and this appeared to favour detection of some species over others (i.e. primarily arboreal possums have low detectability at ground-based cameras). I recommend future occupancy studies sample trails and random forest points over large areas, for long periods with a minimum of 20 functional cameras per site in order to generate more robust data.

156 6.4 Chapter 4: Activity indices of a top predator and mesopredator respond in opposite ways to predator control programs.

Chapter 3 elaborates on the earlier meta-analysis by using occupancy analysis and generalised linear modelling to provide a more in-depth understanding of how predator control affects suites of mammals in forests. In that study, occupancy analysis did not reveal that dingo control leads to sufficiently strong suppressive effects on foxes as their occupancy was shown to be constant across treatments. In this chapter, I further investigate if dingo control leads to an effect on the activity indices of foxes by using a different means to measure a potential interaction. Using track plot surveys, I measure activity of both species by conducting a snap-shot study at 27 sites subject to dingo baiting or not. I measured dingo and fox activity at baited and unbaited sites which had similar underlying geology, vegetation and rainfall. At all baited sites dingoes were the primary intended target species. Despite the lack of an effect of dingo control on foxes using occupancy analysis in my previous chapter, I use earlier studies and ecological theory to hypothesise that dingo control would lead to an increase in the activity indices of foxes.

Using quantile regression, I contrasted my results with earlier studies investigating the relationship between dingoes and foxes. This methodology suggests that dingoes place an upper limit on the abundance of foxes (Johnson & VanDerWal, 2009, Letnic et al. 2011).

Further, studies investigating top predator and mesopredator interactions locally and abroad suggest that at intermediate levels of top predator abundance there appears to be no clear relationship between top predator and mesopredator abundance (Johnson & VanDerWal

2009, Letnic et al. 2011, Bagniewska & Kamler 2014, Swanson et al. 2014). To determine if there were any other factors involved in a potential top predator-mesopredator interaction I

157 conducted generalised linear modelling using environmental predictor variables such as net primary productivity, distance to freehold land, baiting and season.

My results concurred with previous studies reporting negative, triangular relationships between top predator and mesopredator activity/abundance (Johnson & VanDerWal 2009).

This finding is supported by the mesopredator release hypothesis. My quantile regression analysis reveals that the relationship is significant at the upper quantiles which provides further support for the notion that top predators place an upper limit on mesopredator abundance. My results from generalised linear modelling reveal that dingoes and foxes both respond strongest to the baiting predictor variable. Dingo (as a predictor variable in fox models) is also in the top model set but is arguably already represented with the baiting predictor as baiting directly reduces the abundance of dingoes and thus could be thought of as a surrogate for dingo presence. These results suggest that removal of dingoes may be counter- productive for biodiversity conservation because it may lead to higher activity of foxes.

6.4.1 Limitations and recommendations

I necessarily had to use data to build indices derived from different search efforts. This included different numbers of track plot surveys at some sites, some transects that were longer than others and surveys that were conducted across various seasons. Most of these discrepancies were unavoidable (i.e. limited roads, weather restrictions/damage to plots, etc.) but I contend that some discrepancies could be accounted for in future studies. However, since my results concur with ecological theory and previous studies showing suppression of a top predator leads to increases in mesopredator activity, it is reasonable to suspect that this is a dominant process accounting for this effect at my study sites. In future I recommend testing how dingoes affect foxes by using manipulative experiments to provide a more conclusive foundation on which to suppose insight into the relationship.

158 Bait quantity and application in the landscape varied between baited sites and future studies should ensure that this is accounted for in order to standardise this factor. In my study, the baiting variable is treated as a binary predictor (i.e. 0 - absent, 1 - present) but in future baiting could more accurately be gauged as being somewhere on a scale of 0-100. Finally, I sampled sites at different seasons of the year owing to logistical constraints typical of any landscape-scale fieldwork program. It would be ideal if all sites were either sampled at the same season or if every site could be sampled across all seasons. Given that the natural history of both foxes and dingoes is known to be seasonal, greater accountability in this area would be an excellent addition to a similar study design investigating these relationships.

However, I contend that this would add significant financial and labour costs to the fieldwork program.

159 6.5 Chapter 5: Reintroduction of Tasmanian devils to mainland

Australia can restore top-down control in ecosystems where dingoes have been extirpated

Chapter 1 established the importance of top predators for ecosystem function. In Australia, field surveys demonstrate that the removal of dingoes from both arid and mesic ecosystems has negative effects on many small and medium-sized native mammal species owing to increased predation by foxes and possibly cats (Colman et al. 2014, Letnic et al. 2009). Given that dingoes predate on livestock, reintroducing them or even allowing them to persist in certain areas is deemed as untenable by many livestock producers and land managers.

However maintaining predator diversity (Finke & Denno 2004) and effective population densities (Soule et al. 2003) is important in order to mitigate ecological degradation.

Importantly, this notion is often discussed but it has not been thoroughly tested in Australia.

Thus, I argue that there is a need to bolster Australia’s predator guild with a dingo surrogate which is relatively benign for livestock producers and yet still is able to mitigate further declines of small and medium-sized native mammals from fox and cat predation. In this chapter I propose that the Tasmanian devil (Sarcophilus harissii), a marsupial predator which used to exist on the mainland but is now restricted to the island of Tasmania, is a viable candidate.

To determine the feasibility of reintroducing devils, I first examined if climatically suitable areas exist on the mainland for devils based on climate data from their current distribution.

Second, I built fuzzy cognitive models (FCMs) from data gleaned from publications reporting effect sizes between species in order to predict the strength and direction of ecosystem effects

160 of devil reintroduction as well as both dingo and fox control scenarios. It is important to incorporate wild canids into the modelling scenarios because if devils were to be reintroduced it would be necessary to minimise competitive effects of dingoes and foxes on devils, especially founding devil populations.

My findings were novel given that to the best of my knowledge FCMs have not been used to predict ecological outcomes of species reintroductions elsewhere. Of the five modelled scenarios, the one where devils were reintroduced without wild canid control proved beneficial for many threatened native species found in south-eastern Australia. The addition of devils is associated with increased abundances of most small and medium mammal species and vegetation complexity increased also. Under this scenario, the effects of foxes, cats and macropods are ameliorated without ceasing dingo control. This result echoes other findings that demonstrate the importance of maintaining predator diversity because more predator diversity increases intraguild competition and, thus, reduces predatory effects on prey species

(Finke & Denno 2004). My study suggests that not only are there climatically suitable areas for devil reintroduction on the mainland but, that the addition of devils to some areas of south-eastern Australia could improve biodiversity conservation in these areas.

6.5.1 Limitations and recommendations

The interaction strengths used to run my FCMs in this study were populated partly by both direct and indirect interaction strengths gleaned from published studies as well as estimates where data from published studies were not available, as described in our methodology. It has been argued that interaction strengths between predators and other species which are positive must be due to indirect interactions and not direct interactions and therefore shouldn’t be included and instead left for the model to predict. It is important to note that we did not

161 discriminate between indirect and direct interaction strengths because studies did not report the nature of the interaction. Secondly, a positive effect of a predator on another species is not always an indirect trophic interaction since predators may facilitate another species by modifying the environment or through mutualism. Given this information, future studies could report whether interactions between species where direct or indirect, if they are known.

In addition to interaction strengths, I also present model predictions such as the percentage increase or decrease in a population under certain scenarios. I contend that by reporting a single percentage figure I may have misconstrued the certainty of the outputs from an inherently stochastic model format. The model's predictions were represented by percentage changes for clarity of presentation, under the assumption that these would be interpreted as trends rather than specific values. Given the fuzzy nature of the FCM model, this seemed like a natural assumption but it may have helped readers if I was more explicit about the process used to develop the FCMs.

Conservation translocations are inherently risky. Managers can attempt to mitigate some risk by using models to understand the outcomes that a reintroduction may have on an ecosystem and its species. I believe this type of modeling provides a useful tool for conservation managers to test potential feasibility of a species reintroduction because the process is devoid of any risk to the species being tested and provides valuable information about ecological responses at the reintroduction location. I hope that for the Tasmanian devil, which is genetically ‘bottle-necked’ and currently in the midst the devil facial tumour disease, this study can one day prove useful for a mainland reintroduction of the species.

162 6.6 Conclusion

The results of my thesis provide support for the idea that top predators structure ecosystems and can induce ecological cascades. However, the effects that I found were not entirely consistent with theory or indeed between chapters.

Using lethal dingo control as a means of testing effects of predator removal I have demonstrated in chapter 2 that dingoes have strong negative effects on kangaroos and small mammals below the CWR but no consistent effect on CWR mammals. In particular there is no effect on CWR bandicoots in this chapter. Dingoes have only a weak effect on foxes, although confidence intervals were large.

In chapter 3, dingo removal had a mixed effect on small mammals according to generalised linear modelling. Occupancy analyses revealed a positive effect of dingoes on ground- dwelling CWR mammals (bandicoots) which is consistent with theory but counter to chapter

2. There appears to be no effect on wallabies and foxes and this observation is inconsistent with theory. There was an absence of an effect on cats and this observation is also reported in some studies.

Track-based analysis in chapter 4 suggest that dingoes do affect the activity of foxes as quantile regression analysis showed that fox activity had a triangular, negative relationship with dingo activity. Furthermore, generalised linear modelling supported this as dingoes and dingo baiting were in the best model set. I hypothesise that the observed positive response of foxes to dingo poison baiting could result from baiting having less of an effect on foxes than dingoes do and that foxes may be more resilient to baiting than dingoes as dingoes may monopolise baits owing to their dominant role as top predator.

163 Modelling in chapter 5 provides support for the notion that reintroducing an ecological surrogate species for the dingo may be able to mitigate some biodiversity losses in certain areas of Australia. Specifically, my modelling shows that a mainland devil reintroduction could help ameliorate negative effects of macropods and invasive mesopredators on small and medium-sized threatened native mammal species in some parts of south-eastern

Australia.

My research shows that outcomes from dingo control are not always the same and can be difficult to predict. Variability of effects between chapters and theory may be due to the effectiveness of the predator control and differences in techniques and methodologies used to measure these effects. Additionally, many studies reporting the effects of dingoes on other species and ecosystems are based in arid ecosystems and not forest ecosystems. Owing to the fact that forests are relatively more productive (and therefore complex) environments than deserts, it may be that the ability to parse out effects is more difficult owing to an increased number of interaction pathways.

This notion of muted top predator effects in more productive environments is described best in Finke and Denno (2004) and some of my results support this notion. In particular, my occupancy study which was conducted at two forest locations, provided inconclusive findings for the effect of dingoes on some mammals despite most desert studies often reporting relatively more certain outcomes.

However, there are some consistencies in mammal responses following dingo control from my thesis including lower abundances of small mammals (especially those below the CWR) and a trend towards increased abundances of macropods. These observations are supported by studies elsewhere in the world which demonstrate top predator effects extending to multiple trophic levels (Ripple et al. 2014). For example, wolf removal in the United States is

164 linked to the increased abundances of cervids (their main prey species) (Ripple & Beschta

2012). Further, this suppressive effect on a large, grazing prey species has flow on effects which ultimately change the composition of the vegetation (Beschta & Ripple 2009), just my research in this thesis has shown. The wealth of studies investigating these cascades in North

American systems have demonstrated that a reduction in grazing pressure on vegetation benefits smaller animals (Ripple et al. 2014) likely owing to increased vegetation cover and shared food resources.

Additionally, this thesis supports the notion from the international literature that mesopredator release (specifically, the lack of) also negatively affects small mammals

(Ripple et al. 2014). Like wild dogs in Africa (Creel & Creel 1996) and coyotes in North

America (Berger & Gese 2007), foxes are a mesopredator in Australia and appear to be negatively affected by the presence of dingoes in the landscape. The suppressive effects of dingoes on foxes appears to indirectly benefit small mammals, specifically those below the

CWR.

In future, improving detection methodologies can help to elucidate subtle and complex interactions which may presently be latent in forest ecosystems. Describing these effects will be crucial for deciding whether to shift dingo management away from lethal control methods to those allowing dingoes to persist in the landscape because their importance to healthy and functional ecosystems will be recognised.

165 Literature cited

Algar DA, Burbidge AA, Angus GJ (2002) Cat eradication on Hermite Island, Montebello Islands, Western Australia. Turning the Tide: The Eradication of Invasive Species 14– 18.

Allen LR (2009) Best practice baiting: dispersal and seasonal movement of wild dogs (Canis lupus familiaris). In: Technical Highlights: Invasive Plant and Animal Research 2008– 09. QLD Department of Employment, Economic Development and Innovation, Brisbane, pp. 61–62

Allen LR (2014) Wild dog control impacts on calf wastage in extensive beef cattle enterprises. Animal Production Science 54: 214-220.

Allen LR, Sparkes EC (2001) The effect of dingo control on sheep and beef cattle in Queensland. Journal of Applied Ecology 1: 76–87. Allen BL, Allen LR, Engeman RM, Leung LP (2013a) Intraguild relationships between sympatric predators exposed to lethal control: predator manipulation experiments. Frontiers in Zoology 10: 39.

Allen LR, Engeman R, Krupa H (1996) Evaluation of three relative abundance indices for assessing dingo populations. Wildlife Research 23: 197–206.

Allen BL, Fleming PJS, Allen LR, Engeman RM, Ballard G, Leung LP (2013b) As clear as mud: a critical review of evidence for the ecological roles of Australian dingoes. Bioogical Conservation 159: 158–174. Allen BL, West P (2013) Influence of dingoes on sheep distribution in Australia. Australian Veterinary Journal 7: 261–267. Allen BL, Allen LR, Engeman RM, Leung LP (2014) Sympatric prey responses to lethal top- predator control: predator manipulation experiments. Frontiers in Zoology 11: 56.

Allen LR (2015) Demographic and functional responses of wild dogs to poison baiting. Ecological Management & Restoration 16: 58–66.

Allen BL, Allen LR, Leung LKP (2015) Interactions between two naturalised invasive predators in Australia: are feral cats suppressed by dingoes? Biological Invasions 2: 761–776. Bachelet D, Lenihan JM, Daly C, Neilson RP, Ojima DS, Parton WJ (2001) MC1: a dynamic vegetation model for estimating the distribution of vegetation and associated ecosystem

166 fluxes of carbon, nutrients, and water. USDA Forest Service, Pacific Northwest Station General Technical Report PNW-GTR-508.

Bagniewska JM, Kamler JF (2014) Do black‐backed jackals affect numbers of smaller carnivores and prey? African Journal of Ecology 52: 564-567.

Banks PB (1999) Predation by introduced foxes on native bush rats in Australia: do foxes take the doomed surplus? Journal of Applied Ecology 36: 1063–1071.

Banks PB (2000) Can foxes regulate rabbit populations? The Journal of Wildlife Management 64: 401–406.

Berger KM, Gese EM (2007) Does interference competition with wolves limit the distribution and abundance of coyotes? Journal of Animal Ecology 76: 1075–1085. Beschta RL, Ripple WJ (2009) Large predators and trophic cascades in terrestrial ecosystems of the western United States. Biological conservation 142: 2401-2414.

Biggins DE, Vargas A, Godbey JL, Anderson SH (1999) Influence of pre-release experience on reintroduced black-footed ferrets (Mustela nigripes). Biological Conservation 89: 121–129.

Bilney RJ, Cooke R, White JG (2010) Underestimated and severe: small mammal decline from the forests of south-eastern Australia since European settlement, as revealed by a top-order predator. Biological Conservation 143: 52-59.

Bininda-Emonds ORP, Cardillo M, Jones KE, MacPhee RDE, Beck RMD, Grenyer R et al. (2007) The delayed rise of present-day mammals. Nature 446: 507–512.

Brook LA, Johnson CN, Ritchie EG (2012) Effects of predator control on behaviour of an apex predator and indirect consequences for mesopredator suppression. Journal of Applied Ecology 49: 1278–1286.

Brown OJF (2006) Tasmanian devil (Sarcophilus harrisii) extinction on the Australian mainland in the mid-Holocene: multicausality and ENSO intensification. Alcheringa: An Australasian Journal of Palaeontology 30: 49–57. Brüniche-Olsen A, Jones ME, Austin JJ, Burridge CP, Holland BR (2014) Extensive population decline in the Tasmanian devil predates European settlement and devil facial tumour disease. Biology Letter 11: 20140619. Bruskotter JT (2013) The predator pendulum revisited: Social conflict over wolves and their management in the western United States. Wildlife Society Bulletin 3: 674–679.

Burnham KP, Anderson DR (2002) Model selection and multimodel inference: a practical information - theoretic approach. Springer.

167 Burnham KP, Anderson DR (2004) Multimodel inference: understanding AIC and BIC in model selection. Sociological Methods & Research 33: 261-304.

Burrows ND, Algar D, Robinson AD, Sinagra J, Ward B, & Liddelow G. (2003). Controlling introduced predators in the Gibson Desert of Western Australia. Journal of Arid Environments 55: 691-713.

Cade BS, Noon BR. (2003) A gentle introduction to quantile regression for ecologists. Frontiers in Ecology and the Environment 1: 412–420.

Campbell G, Coffey A, Miller H, Read JL, Brook A, Fleming PJ, Allen BL (2018) Dingo baiting did not reduce fetal/calf loss in beef cattle in northern South Australia. Animal Production Science.

Cardillo M, Purvis A, Sechrest W, Gittleman JL, Bielby J, Mace GM (2004) Human population density and extinction risk in the world’s carnivores. PLoS Biology 7: e197.

Carthey AJR, Banks PB (2012) When does an alien become a native species? A vulnerable native mammal recognizes and responds to its long-term alien predator. PLoS ONE 7: 31804.

Catling PC, Burt RJ (1995) Why are red foxes absent from some eucalypt forests in eastern New South Wales? Wildlife Research 22: 535–545.

Caughley G, Grigg GC, Caughley J, Hill GJE (1980) Does dingo predation control the densities of kangaroos and emus? Wildlife Research 7: 1–12.

Caughley G (1987) Ecological relationships. Kangaroos: Their ecology and management in the sheep rangelands of Australia (eds Caughley G, Shephed N, Short J). Cambridge University Press, Cambridge.

Claridge AW, Cunningham C, Catling PC, & Reid AM. (2010). Trends in the activity levels of forest-dwelling vertebrate fauna against a background of intensive baiting for foxes. Forest Ecology and Management 260: 822-832.

Coates TD (2008) The effect of fox control on mammal populations in an outer urban conservation reserve. Australian Mammalogy 30: 51–64.

Colman NJ, Gordon CE, Crowther MS, Letnic M (2014) Lethal control of an apex predator has unintended cascading effects on forest mammal assemblages. Proceedings of the Royal Society B: Biological Sciences 281: 20133094.

168 Colman NJ, Crowther MS, Letnic M (2015) Macroecological patterns in mammal abundances provide evidence that an apex predator shapes forest ecosystems by suppressing herbivore and mesopredator abundance. Journal of Biogeography 42: 1975- 1985.

Corbett LK (1995) The dingo in Australia and Asia. Sydney, University of New South Wales Press.

Creel S, Creel NM (1996) Limitation of African Wild Dogs by Competition with Larger Carnivores. Conservation Biology 10: 526–538.

Crooks KR, Soulé ME (1999) Mesopredator release and avifaunal extinctions in a fragmented system. Nature 400: 563–566.

Crowther MS, Fillios M, Colman N, Letnic M (2014) An updated description of the Australian dingo (Canis dingo Meyer, 1793). Journal of Zoology 293: 192-203.

Cupples JB, Crowther MS, Story G, Letnic M (2011) Dietary overlap and prey selectivity among sympatric carnivores: could dingoes suppress foxes through competition for prey? Journal of Mammalogy 92: 590–600.

Dambacher JM, Li HW, Rossignol PA (2003) Qualitative predictions in model ecosystems. Ecological Modelling 1–2: 79–93. Davey C, Sinclair A, Pech R, Arthur A, Krebs C, Newsome A, Allcock K (2006) Do exotic vertebrates structure the biota of Australia? An experimental test in New South Wales. Ecosystems 9: 992–1008.

Davis NE, Forsyth DM, Triggs B, Pascoe C, Benshemesh J, Robley A, Lumsden LF (2015) Interspecific and geographic variation in the diets of sympatric carnivores: dingoes/wild dogs and red foxes in south-eastern Australia. PloS ONE 10: e0120975.

Dawson TJ, Hulbert AJ (1970) Standard metabolism, body temperature, and surface areas of Australian marsupials. American Journal of Physiology-Legacy Content 218: 1233- 1238.

DeVault TL, Rhodes Jr OE, Shivik JA (2003) Scavenging by vertebrates: behavioral, ecological, and evolutionary perspectives on an important energy transfer pathway in terrestrial ecosystems. Oikos 102: 225-234.

Dexter N, Meek P, Moore S, Hudson M, Richardson H (2007) Population responses of small and medium sized mammals to fox control at Jervis Bay, southeastern Australia. Pacific

169 Conservation Biology 13: 283.

Dexter N, Murray A (2009) The impact of fox control on the relative abundance of forest mammals in East Gippsland, Victoria. Wildlife Research 36: 252.

Dexter N, Ramsey DSL, MacGregor C, Lindenmayer D (2012) Predicting Ecosystem Wide Impacts of Wallaby Management Using a Fuzzy Cognitive Map. Ecosystems 8: 1363– 1379. Dexter N, Hudson M, James S, MacGregor C, Lindenmayer DB (2013) Unintended Consequences of Invasive Predator Control in an Australian Forest: Overabundant Wallabies and Vegetation Change. PLOS ONE 8: e69087. Dickman CR (1996) Impact of exotic generalist predators on the native fauna of Australia. Wildlife Biology 2: 185-195.

Doherty TS, Davis R, van Etten EJB, Algar D, Collier N, Dickman CR, Robinson S (2015) A continental-scale analysis of feral cat diet in Australia. Journal of Biogeography 42: 964-975.

Doherty TS, Ritchie EG (2016) Stop jumping the gun: A call for evidence-based invasive predator management. Conservation Letters 9: E09005. Doherty TS, Glen AS, Nimmo DG, Ritchie EG, Dickman CR (2016) Invasive predators and global biodiversity loss. Proceedings of the National Academy of Sciences 113: 11261- 11265.

Donadio E, Buskirk SW (2006) Diet, morphology, and interspecific killing in Carnivora. The American Naturalist 167: 524-536.

Donlan CJ, Berger J, Bock CE, Bock JH, Burney DA, Estes JA, Greene HW (2006) Pleistocene Rewilding: An Optimistic Agenda for Twenty‐First Century Conservation. The American Naturalist 168: 660-681.

Duval S, Tweedie R (2000) Trim and fill: a simple funnel‐plot–based method of testing and adjusting for publication bias in meta‐analysis. Biometrics 56: 455–463.

Egger M, Smith GD, Schneider M, Minder C (1997) Bias in meta-analysis detected by a simple, graphical test. British Biomedical Journal 315: 629–634.

Elith JH, Graham C, Anderson P, Dudík R, Ferrier M, Guisan S, Zimmermann N (2006) Novel methods improve prediction of species’ distributions from occurrence data. Ecography 2: 129–151.

Elmhagen B, Rushton SP (2007) Trophic control of mesopredators in terrestrial ecosystems: top-down or bottom-up? Ecology Letters 10: 197–206.

170 Elmhagen B, Ludwig G, Rushton SP, Helle P, Lindén H (2010) Top predators, mesopredators and their prey: interference ecosystems along bioclimatic productivity gradients. Journal of Animal Ecology 79: 785-794.

Estes JA, Terborgh J, Brashares JS, Power ME, Berger J, Bond WJ, Wardle DA (2011) Trophic Downgrading of Planet Earth. Science 6040: 301–306. Fancourt BA, Nicol SC, Hawkins CE, Cameron EZ, Jones ME (2015) Devil declines and catastrophic cascades: is mesopredator release of feral cats inhibiting recovery of the eastern quoll. PloS One 3: e0119303. Fillios M, Gordon C, Koch F, Letnic M (2010) The effect of a top predator on kangaroo abundance in arid Australia and its implications for archaeological faunal assemblages. Journal of Archaeological Science 37: 986–993.

Finke, D. L. & Denno, R. F. (2004). Predator diversity dampens trophic cascades. Nature 429: 407–410. Fisher DO, Johnson CN, Lawes MJ, Fritz SA, McCallum H, Blomberg SP et al. (2014) The current decline of tropical marsupials in Australia: is history repeating? Global Ecology and Biogeography 23: 181–190.

Fleming PJS, Corbett L, Harden R, Thomson P (2001) Managing the impacts of dingoes and other wild dogs. 2001, Bureau of Rural Science: Canberra.

Fleming PJS (1996) Ground-placed baits for the control of wild dogs: evaluation of a replacement-baiting strategy in north-eastern New South Wales. Wildlife Research 23: 729–740. Fleming PJ, Thompson JA, Nicol HI (1996) Indices for measuring the efficacy of aerial baiting for wild dog control in north-eastern New South Wales. Wildlife Research 23: 665-674.

Fleming PJS, Allen LR, Lapidge SJ, Robley A, Saunders GR, Thomson PC (2006) A strategic approach to mitigating the impacts of wild canids: proposed activities of the Invasive Animals Cooperative Research Centre. Animal Production Science 7: 753–762.

Fleming PJ, Bomford M, Trust NH (2001) Managing the impacts of dingoes and other wild dogs. Canberra, Australia: Bureau of Rural Sciences.

Fleming PJS, Parker RW (1991) Temporal Decline of 1080 Within Meat Baits Used for Control of Wild Dogs in New South Wales. Wildlife Research 6: 729–740. Fleming PJ, Allen BL, Allen LR, Ballard G, Bengsen AJ, Gentle MN, Saunders GR (2014) Management of wild canids in Australia: free-ranging dogs and red foxes. Carnivores of Australia: past, present and future, 105-149.

Ford AT, Goheen JR, Otieno TO, Bidner L, Isbell LA, Palmer TM, Pringle RM (2014) Large

171 carnivores make savanna tree communities less thorny. Science 6207: 346–349. Forsyth D, Woolnough A, Nimmo D, Ritchie E, Kennedy M, Pople A, Watson I (2014) A comment on the influence of dingoes on the Australian sheep flock. Australian Veterinary Journal 12: 461–462. Foster CN, Barton PS, Lindenmayer DB (2014) Effects of large native herbivores on other animals. Journal of Applied Ecology 51: 929-938.

Friedrich JO, Adhikari NKJ, Beyene J (2008) The ratio of means method as an alternative to mean differences for analyzing continuous outcome variables in meta-analysis: a simulation study. BMC Medical Research Methodology 8: 32.

Funston PJ, Frank L, Stephens T, Davidson Z, Loveridge A, Macdonald DM, Ferreira SM (2010) Substrate and species constraints on the use of track incidences to estimate African large carnivore abundance. Journal of Zoology 281: 56–65.

Gentle MN, Saunders GR, Dickman CR (2007) Poisoning for production: how effective is fox baiting in south‐eastern Australia? Mammal Review 37: 177–190.

Gillies CA, Pierce RJ (1999) Secondary poisoning of mammalian predators during possum and rodent control operations at Trounson Kauri Park, Northland, New Zealand. New Zealand Journal of Ecology 183–192.

Gilroy J (1999) Kangaroo monitoring in relation to kangaroo management in New South Wales. Australian Zoologist 31: 306–308.

Glen AS, Dickman CR, Soule ME, Mackey BG (2007) Evaluating the role of the dingo as a trophic regulator in Australian ecosystems. Austral Ecology 32: 492–501. Gordon CE, Feit A, Grüber J, Letnic M (2015) Mesopredator suppression by an apex predator alleviates the risk of predation perceived by small prey. Proceedings of the Royal Society of London B: Biological Sciences 282: 20142870.

Gordon CE, Moore BD, Letnic M (2017) Temporal and spatial trends in the abundances of an apex predator, introduced mesopredator and ground-nesting bird are consistent with the mesopredator release hypothesis. Biodiversity and Conservation 26: 1445–1462.

Greenville AC, Wardle GM, Tamayo B, Dickman CR (2014) Bottom-up and top-down processes interact to modify intraguild interactions in resource-pulse environments. Oecologia 175: 1349–1358.

Haberl H, Erb KH, Krausmann F, Gaube V, Bondeau A, Plutzar C, Fischer-Kowalski, M (2007) Quantifying and mapping the human appropriation of net primary production in

172 earth’s terrestrial ecosystems. Proceedings of the National Academy of Sciences 104: 12942–12947.

Hadfield JD, Nakagawa S (2010) General quantitative genetic methods for comparative biology: phylogenies, taxonomies and multi‐trait models for continuous and categorical characters. Journal of Evolutionary Biology 23: 494–508.

Hayward MW, Kerley GIH (2009) Fencing for conservation: Restriction of evolutionary potential or a riposte to threatening processes? Biological Conservation, 142: 1–13.

Hayward MW, Marlow N (2014) Will dingoes really conserve wildlife and can our methods tell? Journal of Applied Ecology 51: 835–838.

Hedges LV, Gurevitch J, Curtis PS (1999) The meta‐analysis of response ratios in experimental ecology. Ecology 80: 1150–1156.

Higgins JPT, Thompson SG, Deeks JJ, Altman DG (2003) Measuring inconsistency in meta- analyses. British Biomedical Journal 327: 557–560.

Holbrook CM, Bergstedt RA, Barber J, Bravener GA, Jones ML, Krueger CC (2016) Evaluating harvest‐based control of invasive fish with telemetry: Performance of sea lamprey traps in the Great Lakes. Ecological Applications 26: 1595-1609.

Hollings T, Jones M, Mooney N, McCallum H (2013) Trophic Cascades Following the Disease‐Induced Decline of an Apex Predator, the Tasmanian Devil. Conservation Biology 28: 63-75.

Hughes C, Gaffney R, Dickman CR (2011) A preliminary study assessing risk to Tasmanian devils from poisoning for red foxes. The Journal of Wildlife Management 2: 385–392. Hunter DO, Lagisz M, Leo V, Nakagawa S, Letnic M (2018) Not all predators are equal: a continent-scale analysis of the effects of predator control on Australian mammals. Mammal Review 48: 108-122.

Johnson C (2006) Australia’s mammal extinctions: a 50000 year history. Cambridge University Press. Melbourne.

Johnson CN, Isaac JL, Fisher DO (2007) Rarity of a top predator triggers continent-wide collapse of mammal prey: dingoes and marsupials in Australia. Proceedings of the Royal Society of London B: Biological Science 274: 341–346.

Johnson CN, VanDerWal J (2009) Evidence that dingoes limit abundance of a mesopredator

173 in eastern Australian forests. Journal of Applied Ecology 46: 641–646.

Johnson CN, Wroe S (2003) Causes of extinction of vertebrates during the Holocene of mainland Australia: arrival of the dingo, or human impact? The Holocene 6: 941–948. Jones M (1997) Character displacement in Australian dasyurid carnivores: size relationships and prey size patterns. Ecology 8: 2569–2587. Jones M, Barmuta L (1998) Diet overlap and relative abundance of sympatric dasyurid carnivores: a hypothesis of competition. Journal of Animal Ecology 67: 410-421. Jones ME, Barmuta LA (2000) Niche differentiation among sympatric Australian dasyurid carnivores. Journal of Mammalogy 81: 434-447.

Jones M, Dickman CR, Archer M, Archer M (2003) Predators with pouches: the biology of carnivorous marsupials. CSIRO Publishing. Jones ME, Jarman PJ, Lees CM, Hesterman H, Hamede RK, Mooney NJ, McCallum H (2007) Conservation management of Tasmanian devils in the context of an emerging, extinction-threatening disease: devil facial tumor disease. EcoHealth 3: 326–337. Kagata H, Ohgushi T (2006) Bottom-up trophic cascades and material transfer in terrestrial food webs. Ecological Research 1: 26–34. Kearney MR, Wintle BA, Porter WP (2010) Correlative and mechanistic models of species distribution provide congruent forecasts under climate change. Conservation Letters 3: 203–213. Keith DA (2004) Ocean shores to desert dunes: the native vegetation of New South Wales and the ACT. Department of Environment and Conservation (NSW). Khalil H, Pasanen-Mortensen M, Elmhagen B (2014) The relationship between wolverine and larger predators, lynx and wolf, in a historical ecosystem context. Oecologia 175: 625-637.

King CM (1984) Immigrant killers: introduced predators and the conservation of birds in New Zealand. Oxford University Press. Auckland.

Kinnear JE, Onus ML, Bromilow RN (1988) Fox control and rock-wallaby population dynamics. Wildlife Research 15: 435–450.

Kinnear JE, Onus ML, Sumner NR (1998) Fox control and rock-wallaby population dynamics — II. An update. Wildlife Research 25: 81.

Kinnear JE, Summer NR, Onus ML (2002) The red fox in Australia—an exotic predator turned biocontrol agent. Biological Conservation 108: 335-359.

Körtner G, Gresser S, Harden B (2003) Does fox baiting threaten the spotted-tailed quoll, Dasyurus maculatus? Wildlife Research 30: 111–118.

174 Kosko B (1986) Fuzzy cognitive maps. International Journal on Man– Machine Studies 24: 65-75.

Kovacs EK, Crowther MS, Webb JK, Dickman CR (2012) Population and behavioural responses of native prey to alien predation. Oecologia 168: 947–957.

Kuijper DPJ, de Kleine C, Churski M, van Hooft P, Bubnicki J, Jędrzejewska B (2013) Landscape of fear in Europe: wolves affect spatial patterns of ungulate browsing in Białowieża Primeval Forest, Poland. Ecography 36: 1263-1275.

Legge S, Murphy BP, McGregor H, Woinarski JCZ, Augusteyn J, Ballard G et al. (2017) Enumerating a continental-scale threat: How many feral cats are in Australia? Biological Conservation 206: 293–303.

Leo V, Reading RP, Letnic M (2015) Interference competition: odours of an apex predator and conspecifics influence resource acquisition by red foxes. Oecologia 179: 1033– 1040.

Letnic M, Tamayo B, Dickman CR (2005) The responses of mammals to La Nina (El Nino Southern Oscillation)–associated rainfall, predation, and wildfire in central Australia. Journal of mammalogy 86: 689-703.

Letnic M, Crowther MS, Koch F (2009a) Does a top‐predator provide an endangered rodent with refuge from an invasive mesopredator? Animal Conservation 12: 302–312.

Letnic M, Crowther MS (2013) Patterns in the abundance of kangaroo populations in arid Australia are consistent with the exploitation ecosystems hypothesis. Oikos 122: 761- 769.

Letnic M, Dickman CR (2010). Resource pulses and mammalian dynamics: conceptual models for hummock grasslands and other Australian desert habitats. Biological Reviews 85: 501-521.

Letnic M, Koch F, Gordon C, Crowther MS, Dickman CR (2009b) Keystone effects of an alien top-predator stem extinctions of native mammals. Proceedings of the Royal Society B: Biological Sciences 276: 3249–3256.

Letnic M, Dworjanyn SA (2011) Does a top predator reduce the predatory impact of an invasive mesopredator on an endangered rodent? Ecography 34: 827–835.

Letnic M, Ripple WJ. (2017) Large‐scale responses of herbivore prey to canid predators and primary productivity. Global Ecology and Biogeography 26: 860-866.

175 Letnic M, Greenville A, Denny E, Dickman CR, Tischler M, Gordon C, Koch F (2011) Does a top predator suppress the abundance of an invasive mesopredator at a continental scale? Global Ecology and Biogeography 2: 343–353. Letnic M, Fillios M, Crowther MS (2012) Could Direct Killing by Larger Dingoes Have Caused the Extinction of the Thylacine from Mainland Australia? PLOS ONE 5: 34877. Letnic M, Ritchie EG, Dickman CR (2012) Top predators as biodiversity regulators: the dingo Canis lupus dingo as a case study. Biological Reviews 87: 390–413.

Letnic M, Fillios M, Crowther MS (2014) The arrival and impacts of the dingo. Carnivores of Australia: Past, Present and Future, eds Glen AS, Dickman CR (CSIRO Publishing, Collingwood, VIC, Australia), pp 53–67.

Letnic M, Feit A, Forsyth DM (2017) Strength of a trophic cascade between an apex predator, mammalian herbivore and grasses in a desert ecosystem does not vary with temporal fluctuations in primary productivity. Ecosystems 1-13.

Lyons MB, Mills CH, Gordon CE, Letnic M (2018) Linking trophic cascades to changes in desert dune geomorphology using high-resolution drone data. Journal of The Royal Society Interface 15: 20180327.

MacDonald DW, Harrington LA (2003) The American mink: the triumph and tragedy of adaptation out of context. New Zealand Journal of Zoology 30: 421–441.

Mao JS, Boyce MS, Smith DW, Singer FJ, Vales DJ, Vore JM, Merrill EH (2005) Habitat selection by elk before and after wolf reintroduction in Yellowstone National Park. Journal of Wildlife Management 69: 1691–1707.

Marks CA, Bloomfield TE (1999) Bait uptake by foxes (Vulpes vulpes) in urban Melbourne: the potential of oral vaccination for rabies control. Wildlife Research 26: 777–787.

Marks CA, Wilson R (2005) Predicting mammalian target-specificity of the M-44 ejector in south-eastern Australia. Wildlife Research 32: 151–156. Marlow NJ, Thomas ND, Williams AAE, Macmahon B, Lawson J, Hitchen Y et al. (2015a). 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.

Marlow NJ, Thomas ND, Williams AAE, Macmahon B, Lawson J, Hitchen Y et al. (2015b) Lethal 1080 baiting continues to reduce European Red Fox (Vulpes vulpes) abundance after more than 25 years of continuous use in south‐west Western Australia. Ecological Management & Restoration 16: 131–141.

176 Mcilroy, J. C. (1981). The sensitivity of Australian animals to 1080 poison. II. Marsupial and eutherian carnivores. Wildlife Research 8: 385–399. McNab BK (1963) Bioenergetics and the determination of home range size. The American Naturalist 97: 133-140.

McNiven IJ, G Hitchcock (2004) Torres Strait Islander marine subsistence specialisation and terrestrial animal translocation. Pages 105-162 in I. J. McNiven and M. Quinnell, editors. Torres Strait archaeology and material culture. Memoirs of the Queensland Museum Cultural Heritage Series, 3 (Part 1), Brisbane, Australia.

Menkhorst P, Knight F (2001) Field guide to the mammals of Australia. Oxford University Press.

Miller CJ, Miller TK (1995) Population dynamics and diet of rodents on Rangitoto Island, New Zealand, including the effect of a 1080 poison operation. New Zealand Journal of Ecology 19–27.

Mitchell BD, Banks PB (2005) Do wild dogs exclude foxes? Evidence for competition from dietary and spatial overlaps. Austral Ecology 30: 581–591.

Moher D, Liberati A, Tetzlaff J, Altman DG (2009) Preferred reporting items for systematic reviews and meta-analyses: the PRISMA statement. Annals of Internal Medicine 151: 264–269.

Molsher R, Newsome AE, Newsome TM, Dickman CR (2017) Mesopredator management: effects of red fox control on the abundance, diet and use of space by feral cats. PloS One 12: e0168460.

Morris T, Letnic M (2017) Removal of an apex predator initiates a trophic cascade that extends from herbivores to vegetation and the soil nutrient pool. Proceedings of the Royal Society B: Biological Sciences 284: 20170111.

Moseby K, Carthey A, Schroeder T (2015) The influence of predators and prey naivety on reintroduction success: current and future directions. Advances in Reintroduction Biology of Australian and New Zealand Fauna 29-42.

Moseby KE, Crowther MS, Letnic M (2018) Ecological Role of an Apex Predator Revealed by a Reintroduction Experiment and Bayesian Statistics. Ecosystems 1–13. Moseby K, Hill BM, Read JL (2009) Arid recovery - a comparison of reptile and small mammal populations inside and outside a large rabbit, cat, and fox-proof exclosure in

177 arid South Australia. Austral Ecology 34: 156–169.

Nakagawa S, Santos ES (2012) Methodological issues and advances in biological meta- analysis. Evolutionary Ecology 26: 1253–1274.

Newsome TM, Ballard GA, Crowther MS, Dellinger JA, Fleming PJ, Glen AS, Nimmo DG et al. (2015) Resolving the value of the dingo in ecological restoration. Restoration Ecology 23: 201-208.

Newsome AE, Catling PC, Corbett LK (1983) The feeding ecology of the dingo II. Dietary and numerical relationships with fluctuating prey populations in south-eastern Australia. Australian Journal of Ecology 8: 345-366.

Newsome AE, Catling PC, Cooke BD, Smyth R (2001) Two ecological universes separated by the dingo barrier fence in semi-arid Australia: interactions between landscapes, herbivory and carnivory, with and without dingoes. The Rangeland Journal 23: 71–98.

Newsome TM, Ripple WJ (2014) A continental scale trophic cascade from wolves through coyotes to foxes. The Journal of Animal Ecology 84: 49-59.

Nimmo DG, Watson SJ, Forsyth DM, Bradshaw CJ (2015) Dingoes can help conserve wildlife and our methods can tell. Journal of Applied Ecology 52: 281-285.

Oksanen L (2001) Logic of experiments in ecology: is pseudoreplication a pseudoissue? Oikos 94: 27-38.

Pace ML, Cole JJ, Carpenter SR, Kitchell JF (1999) Trophic cascades revealed in diverse ecosystems. Trends in ecology & evolution 14: 483-488.

Pasanen‐Mortensen M, Pyykönen M, Elmhagen B (2013) Where lynx prevail, foxes will fail– limitation of a mesopredator in Eurasia. Global Ecology and Biogeography 22: 868-877.

Paradis E, Claude J, Strimmer K (2004) APE: analyses of phylogenetics and evolution in R language. Bioinformatics 20: 289–290.

Périquet S, Fritz H, Revilla E (2015) The Lion King and the Hyaena Queen: large carnivore interactions and coexistence. Biological Reviews 90: 1197-1214.

Pickett KN, Hik DS, Newsome AE, Pech RP (2005) The influence of predation risk on foraging behaviour of brushtail possums in Australian woodlands. Wildlife Research 32: 121–130.

178 Pople AR, Grigg GC, Cairns SC, Beard LA, Alexander P (2000) Trends in the numbers of red kangaroos and emus on either side of the South Australian dingo fence: evidence for predator regulation? Wildlife Research 27: 269–276.

Prowse TAA, Johnson CN, Cassey P, Bradshaw CJA, Brook BW (2015) Ecological and economic benefits to cattle rangelands of restoring an apex predator. Journal of Applied Ecology 52: 455–466.

Prugh LR, Stoner CJ, Epps CW, Bean WT, Ripple WJ, Laliberte AS, Brashares JS (2009) The rise of the mesopredator Bioscience 59: 779-791.

R Core Team (2013). R: A language and environment for statistical computing. R Foundation for Statistical Computing, Vienna, Austria. URL http://www.R-project.org/.

Ramsey DSL, Norbury GL (2009) Predicting the unexpected: using a qualitative model of a New Zealand dryland ecosystem to anticipate pest management outcomes. Austral Ecology 34: 409-421

Ramsey D, Veltman C (2005) Predicting the effects of perturbations on ecological communities: what can qualitative models offer? Journal of Animal Ecology 74: 905– 916.

Read JL, Scoleri V (2015) Ecological implications of reptile mesopredator release in arid South Australia. Journal of Herpetology 49: 64–69.

Reddiex B, Forsyth DM, McDonald-Madden E, Einoder LD, Griffioen PA, Chick RR, Robley AJ (2007) Control of pest mammals for biodiversity protection in Australia. I. Patterns of control and monitoring. Wildlife Research 8: 691–709. Rees JD, Kingsford RT, Letnic M (2017) In the absence of an apex predator, irruptive herbivores suppress grass seed production: Implications for small granivores. Biological Conservation 213: 13–18.

Reynolds JC, Tapper SC (1996) Control of mammalian predators in game management and conservation. Mammal Review 26: 127–155.

Ripple WJ, Beschta RL (2004) Wolves and the ecology of fear: can predation risk structure ecosystems? BioScience 54: 755–766. Ripple WJ, Beschta RL (2011) Trophic cascades in Yellowstone: The first 15years after wolf reintroduction. Biological Conservation 145: 205-213.

Ripple WJ, Beschta RL (2012) Large predators limit herbivore densities in northern forest

179 ecosystems. European Journal of Wildlife Research 58: 733–742. Ripple WJ, Estes JA, Beschta RL, Wilmers CC, Ritchie EG, Hebblewhite M et al. (2014) Status and ecological effects of the world’s largest carnivores. Science 343: 1241484.

Ripple WJ, Wirsing AJ, Wilmers CC, Letnic M (2013) Widespread mesopredator effects after wolf extirpation. Biological Conservation 160: 70–79. Risbey DA, Calver MC, Short J, Bradley JS, Wright IW (2000) The impact of cats and foxes on the small vertebrate fauna of Heirisson Prong, Western Australia. II. A field experiment. Wildlife Research 27: 223–235.

Ritchie EG, Elmhagen B, Glen AS, Letnic M, Ludwig G, McDonald RA (2012) Ecosystem restoration with teeth: what role for predators? Trends in Ecology & Evolution: 27: 265– 271. Ritchie EG, Johnson CN (2009) Predator interactions, mesopredator release and biodiversity conservation. Ecology Letters 12: 982–998.

Robertshaw J, Harden R (1986) The ecology of the dingo in northeastern New-South-Wales 4. Prey selection by dingoes, and its effect on the major prey species, the swamp wallaby, Wallabia bicolor (Desmarest). Wildlife Research 13: 141-163.

Robley A, Gormley AM, Forsyth DM, Triggs B (2014) Long-term and large-scale control of the introduced red fox increases native mammal occupancy in Australian forests. Biological Conservation 180: 262–269.

Rothstein HR, Sutton AJ, Borenstein M (2006) Publication bias in meta-analysis: Prevention, assessment and adjustments. John Wiley & Sons. Chichester, UK.

Ruscoe WA, Ramsey DSL, Pech RP, Sweetapple PJ, Yockney I, Barron MC et al. (2011) Unexpected consequences of control: competitive vs. predator release in a four‐species assemblage of invasive mammals. Ecology Letters 14: 1035–1042.

Salo P, Korpimäki E, Banks PB, Nordström M, Dickman CR (2007) Alien predators are more dangerous than native predators to prey populations. Proceedings of the Royal Society B: Biological Sciences 274: 1237–1243.

Sarre SD, MacDonald AJ, Barclay C, Saunders GR, Ramsey DSL (2013) Foxes are now widespread in Tasmania: DNA detection defines the distribution of this rare but invasive carnivore. Journal of Applied Ecology 50: 459–468.

180 Saunders G, Coman B, Kinnear J, Braysher M (1995) Managing vertebrate pests: foxes. Australian Government Publ. Service.

Saunders GR, Gentle MN, Dickman CR (2010) The impacts and management of foxes Vulpes vulpes in Australia. Mammal Review 40: 181–211.

Savidge JA (1987) Extinction of an island forest avifauna by an introduced snake. Ecology 68: 660–668.

Schmitz OJ (2008) Effects of predator hunting mode on grassland ecosystem function. Science 319: 952-954.

Seddon PJ, Griffiths CJ, Soorae PS, Armstrong DP (2014) Reversing defaunation: Restoring species in a changing world. Science 6195: 406–412. Sih A, Bolnick DI, Luttbeg B, Orrock JL, Peacor SD, Pintor LM et al. (2010). Predator–prey naïveté, antipredator behavior, and the ecology of predator invasions. Oikos 119: 610– 621.

Sinclair ARE, Mduma S, Brashares JS (2003) Patterns of predation in a diverse predator– prey system. Nature 425: 288–290.

Sinclair ARE, Pech RP, Dickman CR, Hik D, Mahon P, Newsome AE (1998) Predicting effects of predation on conservation of endangered prey. Conservation Biology 12: 564– 575.

Sivy KJ, Pozzanghera CB, Colson KE, Mumma MA, Prugh LR (2018) Apex predators and the facilitation of resource partitioning among mesopredators. Oikos 127: 607-621.

Soulé ME, Bolger DT, Alberts AC, Wrights J, Sorice M, Hill S (1988) Reconstructed dynamics of rapid extinctions of chaparral‐requiring birds in urban habitat islands. Conservation Biology 1: 75–92. Soule ME, Estes JA, Berger J, Del Rio CM (2003) Ecological effectiveness: Conservation goals for interactive species. Conservation Biology 17: 1238–1250. Spencer EE, Crowther MS, Dickman CR (2014) Diet and prey selectivity of three species of sympatric mammalian predators in central Australia. Journal of Mammalogy 95: 1278– 1288.

Steindler LA, Blumstein DT, West R, Moseby, KE, Letnic M (2018) Discrimination of introduced predators by ontogenetically naïve prey scales with duration of shared evolutionary history. Animal Behaviour 137: 133-139.

181 Swanson A, Caro T, Davies‐Mostert H, Mills MG, Macdonald DW, Borner M, Packer, C (2014) Cheetahs and wild dogs show contrasting patterns of suppression by lions. Journal of Animal Ecology 83: 1418-1427.

Terborgh J, Lopez L, Nuñez P, Rao M, Shahabuddin G, Orihuela G, Balbas L (2001) Ecological Meltdown in Predator-Free Forest Fragments. Science 5548: 1923–1926. Towerton AL, Penman TD, Kavanagh RP, Dickman CR (2011) Detecting pest and prey responses to fox control across the landscape using remote cameras. Wildlife Research 38: 208–220.

Treves A, Karanth KU (2003) Human-Carnivore Conflict and Perspectives on Carnivore Management Worldwide. Conservation Biology 6: 1491–1499. Treves A, Wallace RB, Naughton-Treves L, Morales A (2007) Co-Managing Human– Wildlife Conflicts: A Review. Human Dimensions of Wildlife 11: 383-396.

Twigg LE, Eldridge SR, Edwards GP, Shakeshaft BJ, Adams N (2000) The longevity and efficacy of 1080 meat baits used for dingo control in central Australia. Wildlife Research 27: 473–481.

Van Eeden LM, Crowther MS, Dickman CR, Macdonald DW, Ripple WJ, Ritchie EG, Newsome TM (2018). Managing conflict between large carnivores and livestock Conservation Biology 32: 26-34.

Viechtbauer W (2010) Conducting meta-analyses in R with the metafor package. Journal of Statistical Software 36: 1–48.

Wallach AD, Johnson CN, Ritchie EG, O’Neill AJ (2010) Predator control promotes invasive dominated ecological states. Ecology Letters 13: 1008–1018.

Wayne A, Maxwell MA, Ward CG, Wayne JC, Vellios CV, Wilson IJ (2017) Recoveries and cascading declines of native mammals associated with control of an introduced predator. Journal of Mammalogy 98: 489-501.

Williams TM, Estes JA, Doak DF, Springer AM (2004) Killer appetites: assessing the role of predators in ecological communities. Ecology 12: 3373–3384. Wilson EE, Wolkovich EM (2011) Scavenging: how carnivores and carrion structure communities. Trends in Ecology & Evolution 26: 129-135.

Woinarski JCZ, Burbidge AA, Harrison PL (2014) The Action Plan for Australian Mammals 2012 (CSIRO Publishing, Melbourne).

182 Woinarski JCZ, Burbidge AA, Harrison PL (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.

Woodroffe R, Thirgood S, Rabinowitz A (Eds.) (2005) People and wildlife, conflict or coexistence? (No. 9). Cambridge University Press.

183