<<

The role of fire in the ecology of Leichhardt's grasshopper (Petasida ephippigera) and its food , Pityrodia spp.

Piers Hugh Barrow

B. Sc. (University of Queensland) Hons. ( University)

A thesis submitted to satisfy the requirements for the award of the degree of Doctor of

Philosophy in the Institute of Advanced Studies, School for Environmental Research,

Charles Darwin University, Darwin, .

March 2009

I hereby declare that the work herein, now submitted as a thesis for the degree of Doctor of Philosophy is the result of my own investigations, and all references to ideas and work of other researchers have been specifically acknowledged. I hereby certify that the work embodied in this thesis has not already been accepted in substance for any degree, and is not being currently submitted in candidature for any other degree.

Piers Barrow

March 2009

i

Acknowledgements

My partner Cate Lynch provided support and encouragement, field assistance, proof- reading and editing, and forewent much of what is expected in normal life for a such a long time through this project, and I am deeply grateful.

My supervisors Peter Whitehead, Barry Brook, Jeremy Russell-Smith and Stephen Garnett provided valuable advice and discussion, and, despite typically huge workloads, never failed to make themselves available to help. I am particularly indebted to Peter Whitehead, who shouldered most of the work, way beyond expectations, and provided guidance and insight throughout, and to Jeremy Russell-Smith, who has encouraged and facilitated my interest in the ecology of the in general, and of the sandstone country and fire in particular, for many years.

I am very much indebted to the traditional owners who worked with me throughout the project, especially as most of the work was done at the very worst time of year to be in the sandstone. We walked and clambered long distances and worked long hours under extremely arduous conditions, and in oppressive heat and humidity. There was often no shade, no breeze, and much heat and glare reflected off the bare rocks, yet people worked cheerfully and willingly, and always anticipated the next trip. I thank all the traditional owners who participated in field work, but especially my long term companions Andrew Moore and Colin Liddy.

I thank all the traditional owners who allowed me to work on their country across the Top End. I am especially grateful to Jeffrey Lee who has taken a keen interest in the project throughout and who took me to otherwise closed areas at Koongarra. He also saved us many hours of walking by allowing us to camp at Gubara. I am also particularly indebted to Yvonne Margarula and the Mirrar people in Kakadu and to Ryan Barruwei and the Jawoyn people of Nitmiluk.

I thank all the rangers in the various national parks who provided field assistance, logistical support, accommodation, encouragement, interest, enthusiasm, and quad bikes.

ii

Many, many rangers helped, so I mention only the supervisors here, but I am grateful to them all: Sarah Kerin (Nitmiluk, Gregory and Keep River), Rob Muller, Ollie Scheibe, Lyndall MacLean and Patrick Shaughnessy (Kakadu)

Craig Hempel provided access to the Nitmiluk Vegetation Database. Owen Price gave access to the database for firescar transects. Jeremy-Russell-Smith and Cameron Yates gave access to the NT Ecological Attributes database. Felicity Watt extracted the information on fire histories from satellite imagery held by the NT Bushfires Council. I am most grateful all for this assistance which has helped to fill many gaps in the overall picture.

I would also like to extend my sincere thanks to all of the following groups and individuals. To all the many volunteers, both friends and strangers. Especially to an extraordinary group of Green Corps volunteers at Nitmiluk whose enthusiasm (on their day off) astonished me. To the staff at the Darwin Herbarium for identifications, advice and valuable discussion, especially Whispering Bob Harwood, who undertook field work with me at Keep River. To Colin Wilson from NRETA who carried out joint field work with me in Keep river, Gregory and Nitmiluk National parks, along with Bev Maxwell and Ben Bayliss. Discussions with Colin, an entomologist, have always insightful, valuable and highly entertaining.

iii

Contents

Abstract ...... 1

Chapter 1: Introduction ...... 4

1.1 Background...... 4

1.2 Description and of the species ...... 5

1.3 Social significance ...... 6

1.4 Historical records and current distribution ...... 6

1.5 Broad biology and ecology ...... 9

1.5.1 Life cycle...... 9

1.5.3 Dispersal...... 12

1.5.2 Aposematism...... 13

1.5.4 Host plants...... 13

1.5.5 Habitat...... 14

1.5.6 Population structure ...... 15

1.6 Conservation status...... 15

1.7 Fire...... 16

1.7.1 Fire and Petasida ...... 16

1.7.2 Fire ecology of Pityrodia and the heath communities...... 17

1.8 Population modelling ...... 19

1.9 Broad aims ...... 20

Chapter 2: Fire regimes...... 21

2.1 Abstract...... 21

iv

2.2 Introduction...... 22

2.2.1 Aboriginal fire regimes ...... 25

2.2.2 Contemporary fire regimes ...... 28

2.2.2.1 Season ...... 29

2.2.2.2 Frequency and fire intervals...... 30

2.2.2.3 Intensity...... 31

2.2.2.4 Extent ...... 31

2.2.2.5 Patchiness...... 32

2.3 Methods...... 34

2.4 Results ...... 38

2.4 Discussion...... 50

Chapter 3: The habitat of Leichhardt’s grasshopper – floristic and environmental relations and distribution patterns of Pityrodia...... 54

3.1 Abstract...... 54

3.2 Introduction...... 55

3.3 Methods...... 58

3.3.1 Study sites ...... 58

3.3.2 Distribution Patterns...... 66

3.3.2.1 Regional distribution patterns ...... 66

3.3.2.2 Local distribution patterns ...... 67

3.3.3 Habitat data collection...... 68

3.3.3.1 Nitmiluk Vegetation survey ...... 68

3.3.3.2 The current study ...... 69

v

3.3.4 Habitat data analysis ...... 71

3.3.4.1 Environmental correlates ...... 71

3.3.4.2 Floristic correlates...... 76

3.4 Results ...... 77

3.4.1 Distribution ...... 77

3.4.1.1 Regional distribution...... 77

3.4.1.2 Local distribution ...... 77

3.4.2 Environmental correlates ...... 83

3.4.2.1 Dataset 1 (Nitmiluk Veg. Survey, broad habitat range)...... 83

3.3.2.2 Dataset 2 (Nitmiluk Veg. Survey, narrow habitat range) ...... 85

3.3.2.3 Dataset 3 (Pityrodia sites only, very narrow habitat range)...... 88

3.3.3 Floristic correlates...... 89

3.3.3.1 Ordination ...... 89

3.3.3.2 Indicator Species Analysis ...... 93

3.4 Discussion...... 94

Chapter 4: Population biology of Petasida ephippigera and Pityrodia spp.101

4.1 Abstract...... 101

4.2 Introduction...... 102

4.3 Methods...... 104

4.3.1 Pityrodia...... 104

4.3.1.1 Study sites ...... 104

4.3.1.2 Density and size classes...... 105

4.3.1.3 Mortality and recruitment ...... 105

vi

4.3.1.4 Seedling counts ...... 106

4.3.2 Petasida ...... 107

4.3.2.1 Study sites ...... 107

4.3.2.2 Mark-recapture...... 111

4.3.2.3 Nymph quadrats ...... 113

4.4 Results ...... 114

4.4.1 Pityrodia ...... 114

4.4.1.1 Density and size classes ...... 114

4.4.1.2 Recruitment and mortality...... 115

4.4.1.3 Seedling counts ...... 119

4.4.2 Petasida ...... 119

4.4.2.1 Population estimates...... 119

4.4.2.2 Local Extinctions ...... 128

4.4.2.3 Dispersal...... 129

4.4.2.4 Egg-laying...... 134

4.4.2.5 Bark feeding...... 136

4.5 Discussion...... 137

4.5.1 Pityrodia...... 137

4.5.2 Petasida...... 138

4.5.2.1 Life cycle...... 138

4.5.2.2 Mortality due to fire ...... 140

4.5.2.3 Populations...... 142

4.5.2.4 Dispersal...... 143

vii

4.5.3 Implications for modelling...... 146

Chapter 5: Modelling the effects of fire regimes on populations of Petasida ephippigera ...... 148

5.1 Abstract...... 148

5.2 Introduction...... 149

5.2.1 Background ...... 149

5.2.2 Population Viability Analysis ...... 150

5.2.3 Modelling: population growth and dispersal ...... 152

5.2.4 Modelling: the fire component...... 154

5.2.5 Aims ...... 155

5.3 Methods...... 156

5.3.1 General model description ...... 156

5.3.1.1 Model Assumptions ...... 157

5.3.2 Single Patch Model ...... 158

5.3.2.1 Model summary ...... 158

5.2.2.2 Grasshopper and Pityrodia distribution patterns...... 159

5.3.3 Multiple Patch Model...... 160

5.3.3.1 Model summary ...... 160

5.3.3.2 Pityrodia senescence ...... 161

5.3.3.2 Grasshopper dispersal ...... 161

5.3.3.3 Fire ...... 162

5.3.4 Parameter estimation...... 164

5.3.4.1 Grasshopper population parameters (both models)...... 164

viii

Figure 3.2. Aerial view of the Gubara area in , with upper Gubara

in the valley to the left, and the lower Gubara sites in the valley in the centre

far distance...... 59

Figure 3.3. Location of study sites for Pityrodia and Petasida studies...... 61

Figure 3.4. Location of quadrats for the Nitmiluk Vegetation Survey (Michell et al.

2004) showing the quadrats from which data were analysed in the current

study...... 61

Figure 3.5. Location of study sites in the Nourlangie Rock, lower Gubara and upper

Gubara areas of Kakadu National Park...... 62

Figure 3.6. Pityrodia jamesii (foreground) at site GTG in the lower Gubara area in

Kakadu National Park...... 63

Figure 3.7. Site NOU in Kakadu National Park...... 64

Figure 3.8. Pityrodia jamesii (foreground) at site NOU, with Nourlangie Rock in the

background...... 64

Figure 3.9. Map of all records of the Pityrodia species known to be food plants for

Petasida in the 'Top End of the Northern Territory, from the databases of the

Darwin Herbarium...... 78

Figure 3.10. The distribution of Pityrodia in the lower Gubara area of Kakadu National

Park ...... 79

Figure 3.11. Frequency distributions of Pityrodia patch spans and gap lengths for all

sites in 2003/4 and for Nitmiluk and Upper Gubara in 2001...... 80

Figure 3.12. TTLQV results for transects through three of the study areas...... 81

xi

List of Figures

Figure 1.1. Adult male Petasida ephippigera...... 5

Figure 1.2. Known localities of Petasida ephippigera in Australia...... 8

F igure 2.1 Locations of sites for transect studies of patchiness within firescars. Two

sites occur in close proximity at Gubara (GTG and GTB; see Table 2.1)a.. .. 37

Figure 2.2. The frequency of quadrats falling into 10% intervals for values of '% burnt'.41

Figure 2.3. Frequency distributions for burnt and unburnt section (gap) lengths in

transects within firescars, derived from the data of Price et al. (2003) ...... 42

Figure 2.4. Frequency distribution of burnt and unburnt sections of 100 m transects

within mid-dry season firescars occurring within habitat supporting Pityrodia

populations...... 43

Figure 2.5. Scattered Pityrodia Pungens supporting a Petasida population amongst

spinifex grass at Katherine Gorge in ...... 44

Figure 2.6. The same scene in November 2002 after being by burnt by wildfire in

August...... 44

Figure 2.7. Petasida ephippigera habitat in Nitmiluk National Park three months after a

hot fire...... 44

Figure 2.8. An unburnt gap containing Pityrodia jamesii at upper Gubara (site GTB) in

Kakadu National Park...... 46

Figure 2.9. Site GTG at upper Gubara in Kakadu National Park in November 2002,

three months after being burnt...... 46

Figure 3.1. Aerial view of the Mt Brockman sandstone outlier of the Arnhem Land

escarpment, taken about 2 km north of Gubara...... 56

x

Figure 3.2. Aerial view of the Gubara area in Kakadu National Park, with upper Gubara

in the valley to the left, and the lower Gubara sites in the valley in the centre

far distance...... 59

Figure 3.3. Location of study sites for Pityrodia and Petasida studies...... 61

Figure 3.4. Location of quadrats for the Nitmiluk Vegetation Survey (Michell et al.

2004) showing the quadrats from which data were analysed in the current

study...... 61

Figure 3.5. Location of study sites in the Nourlangie Rock, lower Gubara and upper

Gubara areas of Kakadu National Park...... 62

Figure 3.6. Pityrodia jamesii (foreground) at site GTG in the lower Gubara area in

Kakadu National Park...... 63

Figure 3.7. Site NOU in Kakadu National Park...... 64

Figure 3.8. Pityrodia jamesii (foreground) at site NOU, with Nourlangie Rock in the

background...... 64

Figure 3.9. Map of all records of the Pityrodia species known to be food plants for

Petasida in the 'Top End of the Northern Territory, from the databases of the

Darwin Herbarium...... 78

Figure 3.10. The distribution of Pityrodia in the lower Gubara area of Kakadu National

Park ...... 79

Figure 3.11. Frequency distributions of Pityrodia patch spans and gap lengths for all

sites in 2003/4 and for Nitmiluk and Upper Gubara in 2001...... 80

Figure 3.12. TTLQV results for transects through three of the study areas...... 81

xi

Figure 3.13. Fine scale density map of Pityrodia jamesii at Site GJ1 in the lower Gubara

area...... 82

Figure 3.14. Joint plot showing the first two dimensions of the NMDS ordination results

for Dataset 1 from the Nitmiluk Vegetation Survey data...... 90

Figure 3.15. Ordination Joint plot showing the first two dimensions of the NMDS

ordination results for Dataset 2 from the Nitmiluk Vegetation Survey data... 91

Figure 3.16. The first two dimensions of the NMDS ordination of 58 quadrats and 32

species from the upper Gubara area in Kakadu National Park...... 92

Figure 3.17. Ordination Joint plot showing the first two axes of the NMDS ordination of

86 quadrats with 29 species in the lower Gubara area in Kakadu National

Park ...... 93

Figure 4.1. Locations of study sites in the Gubara and Nourlangie Rock areas of Kakadu

National Park...... 108

Figure 4.2. The study area, study sites and distribution of Pityrodia patches in the lower

Gubara area of Kakadu...... 111

Figure 4.3. Density of Pityrodia plants at study sites in Nitmiluk and Kakadu, measured

between 2001 and 2005...... 115

Figure 4.4. Density of Pityrodia jamesii at four study sites in 2003 and 2004...... 116

Figure 4.5. Change in height over 1 yr, 2003-2004, vs. initial height of tagged P. jamesii

plants at 4 sites in Kakadu...... 117

Figure 4.6. Mortality and recruitment of Pityrodia jamesii in quadrats within four study

sites between 2003 and 2004...... 118

xii

Figure 4.7. The population of Petasida ephippigera at site GJ1 in the lower Gubara area

of Kakadu...... 122

Figure 4.8. Site GJ1 showing the area burnt in December 2005, together with the

locations of all Petasida tagged on 28 Feb 2006...... 123

Figure 4.9. (a) Population estimates and counts of Petasida at site GJ1 through the wet

season of 2003-4...... 124

Figure 4.10. Mean numbers of female and male adult grasshoppers counted at site GJ1

during the 2003-4 wet season...... 125

Figure 4.11. The density of P. jamesii plants and Petasida nymphs in permanent

quadrats within site GJ1 throughout the 2004 dry season...... 126

Figure 4.12. Population estimates and counts for Petasida at sites surrounding the main

study site in Kakadu, between 2001 and 2005...... 127

Figure 4.13. Population estimates and counts for Petasida at site NOU at Nourlangie

Rock and site GTOP at Upper Gubara in Kakadu between 2002 and 2005. 128

Figure 4.14. The maximal distance vs. the total time at large for 152 male and 188

female grasshoppers tagged in the lower Gubara area...... 131

Figure 4.15. Distribution of the movement rates and maximal distances of female

Petasida at lower Gubara recorded throughout the 2003-4 wet season...... 132

Figure 4.16. Movements of grasshoppers between Pityrodia patches in the lower Gubara

study area in Kakadu during the 2003-4 wet season...... 133

Figure 4.17. The recorded movements of grasshopper no. R82 over 113 days during the

2003-4 wet season at site GJ1 in Kakadu...... 135

xiii

Figure 4.18. Petasida nymph on a dead Pityrodia jamesii approximately three

weeks after a fire at Gubara in Kakadu National Park. The grasshopper has

apparently survived by feeding on the bark...... 136

Figure 5.1. The Distribution of habitat patches within the 50 x 50 cell grid for the

Multiple Patch Model...... 161

Figure 5.2. The dispersal kernel used in the simulations at 5 values for the dispersal

parameter 'a'...... 163

Figure 5.3. Frequency distribution of measured gap spans produced by simulating 5000

fires at the default settings. Single-cell gaps have been omitted...... 175

Figure 5.4. Probability of quasi-extinction within 30 years, for grasshoppers in a 1 ha

habitat patch...... 175

Figure 5.5. The effect of variations in a fixed interfire interval on probability of quasi-

extinction within 30 years, for grasshoppers in a 1 ha habitat patch...... 176

Figure 5.6. The effect of varying the values of the CV of the total unburnt area within

firescars on probability of quasi-extinction within 30 years, for grasshoppers

in a 1 ha habitat patch...... 177

Figure 5.7. The effect of varying the intrinsic rate of population increase (r) on

probability of quasi-extinction within 30 years, for grasshoppers in a 1 ha

habitat patch ...... 178

Figure 5.8. The effect of varying the value of the CV of the intrinsic rate of population

increase (r) on probability of quasi-extinction within 30 years, for

grasshoppers in a 1 ha habitat patch...... 179

xiv

Figure 5.9. The effect of distribution pattern of grasshoppers within a single 1 ha habitat

patch on probability of quasi-extinction within 30 years, at 2 values of mean

total unburnt area...... 180

Figure 5.10. Grasshopper and Pityrodia distribution patterns used for modelling...... 181

Figure 5.11. The effect of varying the value for the probability of mortality in burnt

cells on probability of quasi-extinction within 30 years, for grasshoppers in a

1 ha habitat patch...... 182

Figure 5.12. The effect of varying the value for the mean interfire interval on

probability of quasi-extinction within 30 years, for grasshoppers in a

multiple-patch model ...... 183

Figure 5.13. The effect of varying the value for the mean interfire interval on

probability of quasi-extinction within 30 years, for grasshoppers in a

multiple-patch model...... 184

Figure 5.14. The effect of varying the value for dispersal probability on probability of

quasi-extinction within 30 years, for grasshoppers in a multiple-patch model.185

Figure 5.15. The effect of varying the value for 'area with 100% mortality' on

probability of quasi-extinction within 30 years, for grasshoppers in a

multiple-patch model ...... 186

Figure A1. The Single Patch Model grid after fire showing unburnt (white) gaps within

the firescar...... 220

xv

List of Tables Table 2.1. Details of sites and transects for studying firescar patchiness. All fires

occurred in August 2002...... 37

Table 2.2. Descriptive statistics for burnt and unburnt sections of transects within

firescars, derived from the data of Price et al. (2003)...... 47

Table 2.3. Summary statistics for burnt sections of transects within MDS firescars at

sites supporting populations of Pityrodia.

Table 2.4. Mean values for fire regime variables for 16 sites supporting populations of

Petasida and 12 sites without Petasida present at the last inspection in Kakadu

and Nitmiluk National Park ...... 49

Table 3.1. Locations and characteristics of study sites used for habitat and floristic

studies for Pityrodia...... 65

Table 3.2. Mean patch spans, gap lengths and Pityrodia density within patches...... 83

Table 3.3. Results of AICc -based model selection and all-subsets analysis for Pityrodia

presence/absence for dataset 1 at Nitmiluk National Park...... 84

Table 3.4. Post hoc results of AICc -based model selection and all-subsets analysis for

Pityrodia presence/absence for dataset 1 at Nitmiluk National Park...... 85

Table 3.5. Results of AICc -based model selection and all-subsets analysis for Pityrodia

presence/absence in dataset 2 at Nitmiluk National Park...... 86

Table 3.6. Post hoc results of AICc -based model selection and all-subsets analysis for

Pityrodia presence/absence in dataset 2 at Nitmiluk National Park...... 87

Table 3.7. Results of AICc -based model selection and all-subsets analysis for Pityrodia

presence/absence at 12 sites modelled using binomial GLM...... 88

Table 3.8. Indicator species for dataset 1 of the Nitmiluk Vegetation Survey...... 95

xvi

Table 3.9. Indicator species for dataset 2 of the Nitmiluk Vegetation Survey...... 96

Table 4.1. Location, size and dates of permanent quadrats in Kakadu in which P.

jamesii plants were tagged...... 106

Table 4.2. Details of surveys for adult Petasida ephippigera in Kakadu and Nitmiluk

National Parks, 2000-2006...... 109

Table 4.3. Counts of Grasshoppers at two sites in Nitmiluk National Park. Both sites

were burnt in August 2002...... 123

Table 4.4. Mean daily movement rates of Petasida in the lower Gubara study area..... 134

Table 4.5. Short term movements by tagged Petasida recaptured after 1 day...... 134

Table 5.1. Input parameters and default values, with some explanatory notes, used in the

Single Patch Model...... 167

Table 5.2. Input parameters and default values, together with some explanatory notes,

for the Multiple Patch Model...... 169

Table A1. Survivorship in burnt cells which do not have 100% mortality...... 223

xvii

Abstract

Leichhardt's grasshopper (Petasida ephippigera White) is endemic to the sandstone heath communities of the ‘Top End’ of the Northern Territory (NT), where it feeds almost exclusively on a few species of within the Pityrodia. There is some evidence that the distribution of Petasida is in decline and it has been suggested that adverse fire regimes are responsible. The major aim of this study was to investigate the impacts of fire regimes on populations of the grasshoppers using the tools of Population Viability Analysis, including population and fire simulation models, and to develop management recommendations based on these results. To this end, studies were conducted to investigate and describe the habitat of the grasshoppers, the population biology of Pityrodia and the grasshoppers, and the patterns of past and present sandstone fire regimes in the Top End.

A review of the literature reveals that information on traditional Aboriginal fire regimes in the sandstone heaths is sparse, but what evidence there is strongly indicates that under contemporary regimes fires are later, more intense and larger in extent. Many of the relevant studies, however, are based on the interpretation of satellite imagery at a resolution too coarse to show the fine scale burning patterns that are crucial to understanding the fire ecology of Petasida. Fine-scale transect data collected within sandstone heath firescars and analysed by Price et al. (2003) was re-examined in order to describe the spatial patterns of the burnt areas, in ways which were not attempted in the previous study. Additional transect data were collected within areas supporting populations of Pityrodia. The results provide a good description of the internal spatial structure (patchiness) of fires upon which to base fire models, and suggest that in some circumstances entire populations of Petasida might fall within unbroken burnt areas. Assessment of fire history variables extracted from satellite-based firescar maps revealed no significant differences between sites with and without Petasida populations present.

Habitat studies focussed on the presence of Pityrodia as a indicator of grasshopper habitat, and used environmental and floristic data in an existing database for Nitmiluk

Abstract

National Park and from field studies in Kakadu and Nitmiluk National Parks. Environmental relations were investigated using Generalized Linear Modelling (GLM) and floristic relations using Non-Metric Multidimensional Scaling (NMDS) ordination and Indicator Species Analysis. The results support previous observations that the Pityrodia species upon which Petasida feeds are confined to sandstone habitats and that their distribution is distinctly patchy at a range of scales. Pityrodia presence is associated with rock cover, particularly of large rocks and boulders, with open vegetation and with shallow, sandy soils. The floristic associations of Pityrodia are dominated by sandstone heath species, particularly short-lived obligate seeder shrubs. The results suggest that Pityrodia habitat is subject to fire regimes of intermediate frequency and some patchiness.

Pityrodia population studies were primarily aimed at answering the question of whether or not fire affected the amount or extent of resources available to grasshoppers. Quadrat based studies were used to assess the impact of both fire and the absence of fire on the density of Pityrodia stems, and to investigate mortality and recruitment. Surveys for Petasida were conducted at several locations across the 'Top End' of the Northern Territory. Mark-recapture studies were conducted over four seasons, mostly in the Gubara area of Kakadu. The density of Pityrodia stems increased after fire and decreased in the absence of fire. Two examples of mass mortality of stems in the absence of fire, in different species, are reported. Most Petasida populations were very low and sparsely distributed. One population approximately doubled annually for two years until it was reduced after half the Pityrodia patch it occupied was burnt. Several local extinctions of Petasida are reported, some of which were clearly not caused by direct mortality due to fire. The dispersal ability of Petasida is relatively low, but a 'fat- tailed' movement distribution indicates occasional longer-distance dispersal, possibly by flying rather than walking.

Two cellular models were created using Microsoft Excel© 2002 and the programming language VBA. The models are simple, discrete-time, count-based (unstructured), stochastic population growth models coupled with cellular landscape models in which fires with spatially realistic characteristics operate on the grasshopper populations. The

2 Abstract first, fine scale, model was used to model populations in a landscape consisting of a single habitat patch of 1 ha and explicitly modelled the internal spatial structure (patchiness) of fires. The second modelled several habitat patches and incorporated dispersal of grasshoppers between patches and fires of varying sizes. In simulation results the probability of quasi-extinction of grasshopper populations was very sensitive to mean fire interval, total unburnt area within firescars and to fire size. The sensitivity of the results to several estimated or arbitrarily set parameters was investigated.

The results suggest that all the changes to fire regimes since the transition from traditional Aboriginal to contemporary burning patterns are detrimental to populations of grasshoppers at the scale of the model landscapes. Management recommendations for the grasshoppers are precisely consistent with those for the conservation of obligate seeding sandstone heath shrubs. It is recommended that a strategic and targeted program of on-ground and aerial ignition be used by land managers to create a pattern of firebreaks in the early dry in order to reduce the annual area of intense, late dry season burning. Recommendations for ongoing research and monitoring are presented.

3

Chapter 1

Introduction

Chapter 1: Introduction

1.1 Background

The sandstone escarpment and plateau habitats of monsoonal northern Australia support a rich and diverse biota with a very high level of endemism. In contrast to the savannas of the lowlands, the sandstone habitats are comprised of relatively fire-sensitive vegetation communities, including heaths, monsoon rainforest patches and stands of Cypress pines (Callitris intratropica). Under contemporary regimes fires appear to be larger (in area), later, hotter and more frequent than those of the past and there is now strong evidence of severe impacts of these adverse fire regimes on the relatively fire- sensitive sandstone vegetation communities across northern Australia (Russell-Smith et al. 2002; 1998).

Leichhardt's grasshopper (Petasida ephippigera White) feeds almost exclusively on a few species of sandstone heath shrubs within the genus Pityrodia and is endemic to the sandstone heath communities of the ‘Top End’ of the Northern Territory (NT). There is some evidence that the distribution of Petasida is in decline and it has been suggested that adverse fire regimes are responsible (Greenslade and Lowe 1998; Lowe 1995)

The impetus for this project comes from within government departments responsible for national parks in the NT. The Kakadu National Park Plan of Management (Kakadu National Park Board of Management and Parks Australia 1998) gave priority to 'studying the impact of different fire regimes on sandstone habitat and communities, including invertebrates such as P. ephippigera'. In addition, a study of the fire ecology of Petasida (and other sandstone grasshoppers) was specifically recommended in a report commissioned by the Endangered Species Unit of Environment Australia (Roeger and Russell-Smith 1995). Rangers at Nitmiluk and Kakadu National Parks have long been interested in the ecology of the species and some have been keen observers during their tenure in the parks.

1: Introduction

The approach adopted in this study is to use population and fire models to investigate the impacts of fire regimes on the dynamics of Petasida populations. This study aims to use a series of field investigations, together with information gained from the literature review and other sources, both to understand the basic ecology of the two species and to parameterize the models.

Throughout, unless specifically stated otherwise, ‘Pityrodia’ refers only to the species known to be food plants for Petasida, and both 'Petasida' and 'grasshopper' refer only to P. ephippigera.

1.2 Description and taxonomy of the species

Petasida ephippigera belongs to the superfamily Acridoidea, family Pyrgomorphidae and tribe Petasidini. This tribe contains only two monospecific genera: Petasida and Scutillya. The only species within Scutillya, S. verrucosa, the closest relative of Petasida, is confined to southwest (Key 1985). Pyrgomorphs differ from typical grasshoppers and locusts (family Acrididae) in that many are very sluggish and slow to move (Rentz 1996). Petasida is typical in this respect.

Figure 1.1. Adult male Petasida ephippigera

Leichhardt’s grasshopper is the most spectacular and strikingly coloured grasshopper in Australia (Fig. 1.1). In adults the whole body is bright orange with scattered spots and more prominent areas of blue or black, particularly on the pronotum and wing tips. Colours vary slightly between individuals and geographically, but they are always bright

5 1: Introduction in adults. Adult length varies between approximately 3 and 5cm, and females are more robust and stocky in appearance compared to the more slender males. Nymphs are elaborately patterned but cryptic.

1.3 Social significance

Leichhardt's grasshoppers have cultural significance to the Aboriginal people of Kakadu, who know them as Aljurr, the children of Namarrgon, the lightning man depicted in rock art throughout western Arnhem Land (Chaloupka 1993). They are also a calendar organism, with the appearance of the brightly coloured adults used as a marker and a predictor of other seasonal events. They are also significant to people in the Katherine area, and the people at Scott’s Creek, west of Katherine, also have a local name for them (G. Wightman, pers comm.).

Petasida also has an iconic status in the tourism industry, due to its spectacular and beautiful coloration, and is frequently used in tourism promotions and brochures. It has a high profile as a flagship species for three national parks in the Northern Territory, and has appeared on two Australian postage stamps. The few relatively accessible known populations are regularly visited by guided tourist groups in Kakadu and bushwalkers actively search for them in less accessible areas.

1.4 Historical records and current distribution

The first known sighting of Petasida by Europeans is a specimen collected between 1837 and 1843, probably 1839, somewhere along the Victoria River, during surveys carried out by HMS Beagle (Calaby and Key 1973). The next specimen was collected by the explorer Ludwig Leichhardt in 1845 (Leichhardt 1847) at a site now reliably identified as the headwaters of Deaf Adder Creek in Western Arnhem Land (Calaby and Key 1973). The third specimen was collected during A. C. Gregory’s expedition of 1855-56, at a site very close to the present day Timber Creek Township on the Victoria River (Calaby and Key 1973).

6 1: Introduction

Thereafter, for well over 100 years there are no written records, scientific or otherwise, of Petasida. Western science’s rediscovery of the species was made by J. H. Calaby in 1971 on an outlier of the Arnhem Land escarpment near Mt Cahill in the present Kakadu National Park. This was quickly followed by collections and records from elsewhere in Western Arnhem Land and the present Kakadu, near Maningrida in Central Arnhem Land and at two locations in the Katherine district (Calaby and Key 1973). In 1993 it was recorded in the east Kimberley area, in Keep River National Park just east of the NT-WA border (Lowe 1995). In 2002 specimens were collected from a site 80km north of Ngukurr on the Gulf of Carpentaria (Wilson et al. 2003). Reliable sightings have also been made at Scotts Creek Station, south-west of Katherine (G Wightman, Pers. Comm.) and at Bullo River Station east of Keep River NP (Wilson et al. 2003).

Hence, the current known range spans 600km across the Top End of the NT and a latitudinal range of 450km (Fig. 1.2). The range extends from coastal areas to 300 km inland and all populations are in sandstone plateau and escarpment country. No sightings have been recorded in the Victoria River District (VRD) since 1855-56, despite the opening up of the area to pastoralism, tourism and other uses and the establishment of Gregory National Park.

How could so spectacular and obvious an organism simply disappear from (western, scientific) view for so long? This question has been the cause of much speculation and probably has no single, simple answer. It is quite possible that Petasida is extinct in the VRD, where the first collections were made. Land use and management have changed considerably in that district over the last century. However, there are vast areas of sandstone in the VRD, much of which could contain Petasida populations.

The sandstone areas are, however, remote, rugged and inaccessible. They are generally useless as pastoral land and so it was only when mining exploration began in earnest in the 1960s and 1970s that many of these areas were opened for access. Even now, very few roads enter the sandstone country. Flooding during the wet season, when the coloured adults are present, adds to accessibility problems. In addition, heat and humidity cause very oppressive conditions for people at this time of year. These factors

7 1: Introduction mean that very few people actually enter the grasshoppers’ habitat when they are easy to see. During the more pleasant dry season, the nymphs retain their cryptic coloration and are extremely difficult to see. If they are seen, it is not at all obvious to unfamiliar eyes to which species they belong. It is therefore possible, indeed probable, that Petasida did not disappear at all, but simply that very few white people entered its habitat at the critical time of year.

Figure 1.2. Known localities of Petasida ephippigera in Australia. The closed triangles represent recent discoveries (since 2000). From Wilson et al. (2003).

However, Calaby and Key (1973) make the very pertinent and important point that:

'There is also a good deal of circumstantial evidence suggesting that very long- term fluctuations, with radical reductions and fragmentations of range followed ultimately by reoccupation, is a feature of the population dynamics of several species of Australian Pyrgomorphidae.'

8 1: Introduction

1.5 Broad biology and ecology

1.5.1 Life cycle

The life cycle of Petasida appears to be annual. While other members of the family Pyrgomorphidae typically lay their eggs in the soil, no observations of oviposition in Petasida had been recorded before this study. Observations of the timing of the life- cycle were made during this study and by Lowe (1995). In Kakadu National Park Nymphs begin to emerge in the early dry season (early May), within a fortnight of the deaths of the last of the previous season’s adults, The nymphs are cryptically coloured and extremely difficult to see against the background of Pityrodia foliage. They are most commonly found on small Pityrodia plants or the lowest branches of larger plants. Growth is slow until the small but abrupt rise of temperature that occurs typically during August. Then an accelerated growth spurt brings them to maturity, with the first adults appearing during the 'build up' to the wet season in November. Bright colours begin to appear in the second-last nymphal instars. The last instars are very brightly coloured; not quite as brightly as the adults, but with more elaborate patterning. By the end of December nymphs are very rare.

Mating begins soon after the adults appear, and continues until the last adults die approximately four months later. Known details of reproductive behaviour are few, but it appears that mating and egg-laying occur several times for each female. Lowe (1995) reports that there appears to be no diapause stage for the eggs.

9 1: Introduction

Figure 1.3. Early instar nymphs on a dead stem of Pityrodia puberula at Gubara in Kakadu National Park.

10 1: Introduction

Figure 1.4. Late instar nymph on Pityrodia spenceri at Edith Falls in Nitmiluk National Park.

11 1: Introduction

Figure 1.5. Adult female on Pityrodia jamesii at Gubara in Kakadu National Park.

1.5.3 Dispersal

Dispersal by nymphs appears to be very limited, but patterns of dispersal by adults are little understood. While adults often appear reluctant to fly, there have been many observations of adults flying for distances greater than 100m. In particular there is a period in the mid-wet season when the males appear to be flighty and skittish, are easily disturbed and are quick to take wing. Park rangers have also reported sightings of individuals on roads and in car parks up to a kilometre away from the nearest known host plants. Females are rarely observed flying. Prior to this study there were no quantitative measurements of dispersal distances or published observations of dispersal behaviour.

12 1: Introduction

1.5.2 Aposematism

The colouring of adult Petasida is clearly aposematic (Key 1985). It is improbable that the grasshoppers are mimics, rather than toxic or distasteful, as there are no known extant models within their range. There are no known vertebrate predators even though the grasshoppers are extremely conspicuous. The adults, in contrast to the early nymphs, often sit near the ends of the top branches of their host plants, and make minimal effort to evade observation. If disturbed they tend to move to the opposite side of the branch, and if harassed often simply drop to the ground. It has been speculated that the grasshoppers derive toxic chemicals from their aromatic host plants (e.g. Rentz 1996). However, an analysis of both Petasida and three species of Pityrodia (Fletcher et al. 2000) found no toxic alkaloids in plants or animals, but did identify glycosides which, while not generally toxic, are known to be bitter tasting.

Mantids were observed to eat adult Petasida without apparent ill effects on two occasions during this study.

1.5.4 Host plants

Petasida is generally found feeding only on plant species within the genus Pityrodia (). While there are 16 species of Pityrodia recorded in the Northern Territory, only seven of these have been recorded as hosts for Petasida: P. ternifolia (F. Muell.) Munir; P. jamesii Specht; P. puberula Munir; P lanuginosa Munir; P pungens Munir, P. lanceolata Munir and P spenceri Munir. That wingless nymphs have been seen on all these species is a strong indication that they are true host plants, rather than incidental hosts for transient individuals. While Petasida appeared to show preference for one Pityrodia species over another at one location (Wilson et al. 2003), in most cases any apparent preference for a single species is probably simply a reflection of host species availability. Prior to this study there were no records of individual grasshoppers moving from one species of Pityrodia to another.

The grasshoppers will feed, apparently reluctantly, on other species of plants; they have been reared in captivity where they were fed on Prostanthera cuneata ()

13 1: Introduction

(Calaby and Key 1973). Key (1985) states that they are known to eat Dampiera conospermoides (Goodeniaceae). Historical records also associate Petasida with other genera, particularly Dampiera and Goodenia (both Goodeniaceae). Wilson et al. (2003) suggest that these collections may have consisted largely of transient individuals on rarely used host plants. Similarly, it is assumed here that species other than Pityrodia are not important food plants as almost all sightings of Petasida on such other species, including Dampiera, during the current study were in close proximity (<20 m) to Pityrodia plants (pers. obs.).

1.5.5 Habitat

The dependence of Petasida on its specific food plants means that the distribution of Pityrodia species is an important, perhaps the most important influence on the distribution and abundance of Petasida. Apart from the suspected influence of fire, either directly on the grasshoppers or on their host plants, the relative importance of other factors is as yet unknown. Little is yet known of the biology of Pityrodia or of the ecological determinants of its distribution and relative abundance. Throughout the range of Petasida, however, the distribution of Pityrodia is markedly patchy, at both local and broader scales. Furthermore, the distribution of Petasida in each region appears to be distinctly patchy, both within and among patches of Pityrodia.

For example, at Katherine Gorge in Nitmiluk National Park, almost the entire known population lay, at the start of this study, within a rectangular area of approximately 35 ha, situated within a much larger area containing Pityrodia. Within that population, there were three main subpopulations, each occupying an area of approximately 2-4 ha. Apart from two individuals found approximately 700 m away, the nearest known additional, and presumably distinct population was more than 9 km away.

Apart from its dependence on the host plants, the habitat preferences of Petasida are unknown. Nor are any data available on the relationship between grasshopper populations and Pityrodia populations. For example, the relationship between Pityrodia patch size, or the density of Pityrodia within patches, and the ability of those patches to support grasshopper populations is unknown. Similarly, it is not known with any

14 1: Introduction certainty that all Pityrodia represents suitable grasshopper habitat. It is probable that fire plays an important role in maintaining these patchy distributions of both grasshoppers and Pityrodia, but it is also possible that populations of Petasida and Pityrodia respond to fire regimes in quite different ways.

1.5.6 Population structure

The key to understanding the ecology of Petasida, and the role of fire, may well be provided by exploring the metapopulation dynamics of the species. A metapopulation is 'a collection of two or more separate populations (sub-populations), separated in space, and connected by (limited) migration' (Hanski and Gilpin 1997). Hanski ( 1999) describes the three processes 'in the hearth of metapopulation ecology': migration and its effects on local dynamics, population extinction and the establishment of new local populations. It is highly probable that Petasida conforms to a metapopulation structure, potentially at both local and regional scales. It is certainly patchily distributed, because this distribution is in part determined by the patchy distribution of its host plants. Not all the apparently available habitat (Pityrodia) patches are occupied. Given that observations of dispersal (migration) are rare, it is highly probable that some patches represent separate (sub-) populations. If fires or other disturbances cause extinction of local populations, it is likely that the rate of migration, and hence recolonisation of vacant habitat patches, is an important determinant of metapopulation persistence.

1.6 Conservation status

Petasida is currently listed as vulnerable under the Territory Parks and Wildlife Conservation Act 2000. The main basis for the listing was concern over the vulnerability of both grasshoppers and host plants to the impacts of altered fire regimes (Wilson et al. 2003). Without recommending any change to the listed status, Wilson et al. (2003) suggest the possibility, however, that Petasida may be more secure within its core distribution than previously believed. This suggestion is based on the expectation that more populations exist within the vast, as yet unsurveyed sandstone landforms of western Arnhem Land and elsewhere in the NT.

15 1: Introduction

Nevertheless, there is strong evidence (discussed below) that the heaths which form the habitat of Petasida are themselves under serious threat. Petasida, as a conspicuous and integral component of those ecosystems, may well serve as an indicator species for their overall health. One aim of this project is to propose a viable and achievable monitoring program for Petasida within major NT National Parks.

1.7 Fire

1.7.1 Fire and Petasida

Any influence of fire on the distribution and abundance of Petasida could potentially operate both through its impact on the distribution and abundance of the host plants, Pityrodia spp. (discussed below), and by direct impacts on grasshopper populations. It is unlikely that Petasida nymphs would survive the direct impacts of fires. Lowe (1995) reports the extinction of local populations following fires. It is probable that nymphs will suffer 100% mortality if their individual host plant is burnt, and it is possible that nymphs in close proximity to fire will suffer mortality due to the effects of smoke and heat. Adult survivorship probably depends on an ability to escape fires and on the local availability of green Pityrodia plants after a fire.

While a single, isolated fire may well exert a strong local influence on a grasshopper population, it is important to draw the distinction between single fires and fire regimes, as it is the regime that will ultimately have the greatest influence on distribution and abundance. Bond and van Wilgen (1996) define a fire regime as a combination of three elements: frequency, season and intensity. Whelan (1995) includes extent (in area) of fires in this list. One further addition, patchiness (Russell-Smith et al. 2003a), is necessary to complete a useful definition for the purposes of this study. All these components are interrelated. For example, in rocky areas increased intensity is related to decreased size of unburnt patches (Price et al. 2003). Intensity is in turn influenced by season and frequency. All are likely to influence the population dynamics of Petasida.

16 1: Introduction

Season, intensity, size (extent) and frequency of fires may be important factors determining the survival rates in grasshopper populations subjected to fire regimes, but the indications are that the patchiness of fires is critical. For example, the analysis by Russell-Smith et al. (1998) showed the average size of fires in Kakadu National Park in recent years to be approximately 60 ha. A similar fire, if it were to burn 100% of the area within its boundary, could affect the entire known population of Petasida in a large part of Nitmiluk National Park. However, Price et al. (2003) found that all fires in rocky areas leave some unburnt patches. These would presumably hold refuges for grasshoppers, and if so, the impact of frequent fires may be greatly reduced.

A detailed fire history for Kakadu National Park, derived from LandSat MSS data provides details of frequency, seasonal extent and broad-scale patchiness of fires over a 20 year period to 2000 (Edwards et al. 2003; Russell-Smith et al. 1997b). Further firescar maps derived from satellite imagery exist for Kakadu and Nitmiluk National Parks. However, few data are available on the patchiness within mapped firescars. Recent modelling work for fire patterns in sandstone habitats (Price et al. 2003) may provide a key to providing a more comprehensive description of fire patterns with which to investigate the influence of fire on the distribution and abundance of grasshoppers.

If, as appears likely, grasshoppers suffer high mortality during the passage of fire but are able to survive in unburnt patches, understanding or modelling the behaviour of fires under different conditions (e.g. season) will enable predictions of the ranges of rates of mortality or survivorship in grasshopper populations subjected to a single fire. This will in turn provide a key to making relative predictions of the effects of different fire regimes on Petasida populations.

1.7.2 Fire ecology of Pityrodia and the heath communities

While there is a rapidly expanding body of literature relating to fire in the northern Australian savannas (e.g. McKaige et al. 1997; Russell-Smith et al. 2003a; Russell- Smith et al. 2007; Williams et al. 2002) and a great deal of research on the heaths of regions with temperate climates (Bond and van Wilgen 1996; Keith et al. 2002; Whelan 1995), there is very little published research relating specifically to the sandstone heaths

17 1: Introduction of tropical Australia. Although a fire-prone vegetation type, the sandstone heath communities are relatively fire-sensitive in comparison to the savannas (Russell-Smith et al. 2002). There is evidence for a significant change in the fire regimes of western Arnhem Land since the 1940s, associated with the depopulation of the area and consequent absence of active land management (Lucas and Russell-Smith 1993). Fires now occur later in the dry season, and have increased in frequency, intensity and extent. Analysis of the 20 year satellite-derived history showed that, on average, 28% of the sandstone escarpment and plateau in Kakadu was burnt each year and that 40% of the sandstone vegetation was burnt at frequencies of at least 1 in 3 years (Edwards et al. 2003; Russell-Smith et al. 1997b; 1998). Those authors concluded that contemporary fire regimes in the sandstone habitats, especially the heaths, are unsustainable in that they are causing severe impacts on fire-sensitive vegetation types across northern Australia.

While ecological studies on the Australian tropical heaths are rare, ecological studies on individual plant species, apart from the pine Callitris intratropica (e.g. Bowman and Panton 1993) and Petraeomyrtus punicea (Russell-Smith 2006), do not exist. Such studies are abundant for species in the temperate climate heaths, but the value of extending the results to tropical heaths is questionable. For example, many of the southern Australian studies are highly focussed on issues relating to bradyspory (serotiny, retention of seeds on the plant), which is apparently rare in the northern sandstone heaths.

Very little is yet known of the fire ecology of Pityrodia spp. The regeneration strategy of at least five of the known host species falls into the broad category of sprouter as defined by (Gill 1981b) or resprouter in the terminology of Russell-Smith et al. (2002). In resprouters (as opposed to obligate seeders) reproductively mature plants are able to survive fires that cause 100% leaf scorch by resprouting. However, there are several kinds of resprouters (Gill 1981b) and different types of fire and different fire regimes may cause widely varying responses. The effects of fires on flowering, seed production and germination in Pityrodia are also unknown. While Pityrodia plants appear to survive individual fires, casual observation suggests that they do not tolerate frequent, intense

18 1: Introduction fires. However, there is also evidence that in the absence of fire, Pityrodia is out- competed and succeeded by other species (C. Dunlop, pers. comm.). Casual observations indicate that in some cases fire is followed by a burst of greatly increased recruitment to Pityrodia populations. Given the importance of Pityrodia as a determinant of the distribution of Petasida, this project aims to gain some understanding of the fire ecology of Pityrodia in order to understand that of the grasshopper.

1.8 Population modelling

The impacts of fire regimes on populations, by the very nature of the regime components (especially frequency and interval) are not amenable to direct field study over short time periods such as that available for this study. The use of simulation modelling provides a useful and productive means of investigating (relatively) long term effects based on a synthesis of the available field data. The approach adopted here was to use the tools of Population Viability Analysis (PVA: e.g. Beissinger and McCullough 2002; Morris and Doak 2002; Shaffer 1981) to address these questions.

Two closely related models for Petasida populations, both incorporating fires, have been constructed using the spreadsheet program Microsoft Excel© 2002 (Chapter 5). The models are used to investigate the effects of variations in the major components of fire regimes on populations of Petasida. The models are used to assess the sensitivity of the results to the estimated model parameters in order to guide future research. In addition, the models provide a descriptive and heuristic tool to aid in conceptualizing and understanding the dynamics of Petasida populations and the relationship between fire regime components and those dynamics.

Both a literature review and field investigations of the spatial structure of fires (Chapter 2) provide data for parameterization of the fire components of the models. Development of an understanding of the habitat requirements of Petasida and its host plants, and of the spatial distribution of that habitat, provides important background information and a setting for the models (Chapter 3). Investigations of grasshopper population growth, and of the influence of fire on grasshopper and Pityrodia populations (Chapter 4), provide

19 1: Introduction data for the parameterization of the population components of the models. The development of the models and interpretation of the modelling results are described and discussed in Chapter 5.

1.9 Broad aims

1. Investigate the behaviour and spatial structure of sandstone heath fires, particularly patchiness, under a variety of environmental conditions and seasons (Chapter 2).

2. Explore the floristic and environmental habitat relations of both Pityrodia spp and Petasida (Chapter 3);

3. Investigate and describe the dynamics of populations of Petasida and two species of Pityrodia Investigate the influence of fire on the distribution and relative abundance of Petasida and on the species of Pityrodia on which it feeds (Chapter 4

4. Develop population and fire models to explore the impact of fire regimes on the dynamics of Petasida populations (Chapter 5); and

5. Derive realistic and empirically grounded management options for promoting the viability of local and regional populations of Petasida and determine the implications of the results for management of the sandstone heath communities of the NT (Chapter 5).

20

Chapter 2

Fire regimes

Chapter 2: Fire regimes

2.1 Abstract

A review of the literature reveals that information on traditional Aboriginal fire regimes in the sandstone heaths is sparse, but what evidence there is strongly indicates that under contemporary regimes fires are later, more intense and larger in extent. Available data on contemporary fire regimes has been collated as a basis for later modelling of fire impacts on populations of Leichhardt's grasshopper (Petasida ephippigera). Many of the relevant studies are based on the interpretation of satellite imagery at a resolution too coarse to show the fine scale burning patterns crucial to understanding the fire ecology of Petasida.

Fine-scale transect data collected within sandstone heath firescars and analysed by Price et al. (2003) was re-examined in order to describe the spatial patterns of the burnt areas, in ways which were not attempted in the previous study. Additional transect data were collected within areas supporting populations of Pityrodia spp., the host plants of Petasida. Methods were similar but not identical to those of Price et al. (2003), the main difference being the random placement of 100m transects at most sites in the supplementary study, rather than single long transects. Transects consisted of contiguous 5 x 5 m quadrats, in which variables including the amount of vegetation burnt and rockiness were estimated.

The results of the literature review provide a good description of sandstone heath fire regime components such as frequency and extent upon which to base fire models. The results of the analysis of the two datasets add to the capability to accurately model fires by providing good descriptions of the internal spatial structure of fires.

The majority of both burnt and unburnt patches were relatively small, but there is considerable variation in patchiness, even within single fires, and large, unbroken burnt areas occurred in several fires, especially in the late dry season. In general, late fires were less patchy: they burnt a greater proportion of the area and the mean size of burnt

2: Fire regimes patches increased. Season of burn strongly influences fire regime variables, but the pattern of that influence is not simple. there was high variation in some variables such as maximum length and frequency of burnt sections of transects. Rocks provide some protection from fire, increase patchiness, and mitigate the influence of season to a certain extent

Other results from this transect study are consistent with those of Price et al. (2003). Assessment of fire history variables extracted from satellite-based firescar maps revealed no significant differences between sites with and without Petasida populations present. It is suggested that the resolution of the MSS imagery is too coarse to detect fine scale spatial fire patterns which may influence Petasida population dynamics.

2.2 Introduction

The spatial and temporal pattern of fires – the fire regime – will ultimately have a greater influence on the distribution and abundance of organisms than any single fire. Gill (1981a) defined fire regimes in terms of the levels of four variables: fire type, fire intensity, fire frequency and season of burning. He also discussed the then current absence of any known classification scheme for fire regimes. Such a scheme remains elusive and even the definition of the term changes continually. Bond and van Wilgen (1996) use only the last three of those variables to define fire regimes. Whelan (1995) adds to the list fire extent, which he parenthetically labels 'patchiness', and by which he means the size, in area, of fires. Russell-Smith et al. (2003a) regard patchiness more as the spatial pattern of burnt and unburnt areas. Morgan et al. (2001) further extend the list to include 'frequency, magnitude (severity and intensity), predictability, size, seasonality, and spatial patterns'. By subdividing the category of frequency, Fox and Fox (1987), arrived at eight variables.

Whatever descriptive scheme is adopted, all the variables used to characterize fire regimes are highly interrelated. Fox and Fox (1987) make the point that they are all time related. This may be true, but because of the intimate linkages between the variables, it is just as true that they are all intensity related, or extent related. The emphasis on time,

22 2: Fire regimes while important, probably reflects the overarching interest that many Australian fire researchers involved in conservation-related fields have in obligate-seeding plant species. Bond and van Wilgen (1996) make two further important points. Firstly, in considering the ecological effects of fire, both the mean of the variables and the variability around the mean are very important. Secondly, they emphasise the distinction between interval- and event-dependent effects. Again, these points apply particularly in relation to obligate seeder species, but that does not diminish their general relevance and importance.

The variables most relevant to the current study, because they are considered most likely to affect the population dynamics of Petasida, are frequency, season, intensity, extent (area) and patchiness, particularly internal patchiness. All fires within the habitat of Petasida are classified as 'surface' fires, rather than 'ground' or 'crown' – the other categories within the variable 'fire type' – and so fire type will not be discussed further here.

The sandstone heath habitats of Leichhardt's grasshopper lie within the Kimberley/VRD and Top End/Gulf provincial bioregions (Thackway and Creswell 1995), which in turn fall within the biome broadly classified as 'savanna'. Australian savanna landscapes are characteristically fire-prone, as they are throughout the world (Williams et al. 2002b). The tropical savanna region, covering 1.9 million km2 across northern Australia, accounts for the majority of Australia's wildfires, with an average of 19% by area burnt each year between 1997 and 2005 (Russell-Smith and Yates 2007). This proportion is much higher, however, in the higher rainfall coastal areas within which lie the habitats of Leichhardt's grasshopper (Russell-Smith et al. 2003b). Much of the fire literature for northern Australia relates to the savanna region generally, and one of the aims here is to tease out the information that relates specifically to the sandstone heaths.

Fire regimes are inextricably linked to climate and weather. The climate throughout the range of Petasida is monsoonal and the seasons are highly predictable, with almost all of the rain falling within the wet season between approximately November and April, and very little rain in the dry season from May to October. Mean annual rainfall varies

23 2: Fire regimes between approximately 900 and 1600 mm. Within these rough guidelines, however, the timing of the onset of seasons and the total annual rainfall are highly variable both annually and geographically (Taylor and Tulloch 1985), and there is considerable variation in weather conditions within the two main seasons recognized by the non- Aboriginal population. The key to understanding relationships between weather and fire lies in understanding the more elaborate seasonal descriptions used by Aboriginal people, some of which recognize six separate seasons, and some with yet further subdivisions (Brockwell et al. 1995; Haynes 1985; Jones 1980).

In Kakadu National Park (Kakadu), for example, the six seasons recognized, with their Gundjeihmi language names, are as follows: Gudjewg (Dec–Mar), the height of the rainy season; Banggerreng (Mar–May), during which the annual speargrass (Sarga spp.) cures and is flattened by late storms and the winds change to south-easterly; Yegge (May– June), a period of clear weather, decreasing humidity and cool nights; Wurrgeng (June– August), the cool, dry season; Gurrung (Aug–Oct), the late, hot dry season with increasing humidity; and Gunumeleng (Oct–Dec), with high humidity, the first storms of the wet and winds changing back to north-westerly (Brockwell et al. 1995; Morris 1996). The months given are approximate dates as seasons are defined by biological and weather markers rather than fixed dates.

The Aboriginal seasons are reflected in the findings of Gill et al. (1996), who showed rising temperatures, decreasing dew points and decreasing atmospheric, soil and fuel moisture contents towards the end of the dry season in Kakadu. The most severe fire weather was in Sep–Oct (Gurrung). Wind speeds were higher in the daytime than at night and higher in the late dry season (Sep) than early (June). The probability of calm periods was higher at night and higher in June than in September.

Against this background, the major aims of this chapter are as follows:

1. To review the literature on historic and contemporary sandstone fire regimes in the Top End, with an emphasis on the sandstone heath habitats.

24 2: Fire regimes

2. To re-examine the data of Price et al. (2003) on spatial patterns in sandstone heath fires in order to highlight aspects that may be pertinent to the population biology of Petasida. In particular, the aim was to examine the size distribution of burnt sections of transects.

3. To examine the patchiness and other fire regime variables within habitat that supports populations of Pityrodia, which is the potential habitat of Petasida.

4. To discuss the potential impacts of fire regimes on Petasida, and the implications for modelling population dynamics.

In the following discussion, unless specifically stated otherwise, 'heath' refers to tropical Northern Territory sandstone heath. Throughout this chapter, both the unburnt 'islands' within firescars and the unburnt sections of transects are generally referred to as 'gaps', whereas burnt areas never are.

2.2.1 Aboriginal fire regimes

Knowledge of and evidence for savanna fire regimes immediately prior to European colonization of northern Australia have been reviewed by Russell-Smith (2001; 2002) and Russell-Smith et al. (2003b). They summarize the results of three major lines of enquiry – early explorers' records; the ethnographic record, particularly from Arnhem Land; and contemporary accounts of Aboriginal burning – to paint a fairly consistent picture of fire regimes across the coastal and subcoastal savannas which include the range of Petasida. However, examination of the references cited in these reviews reveals that much of the material relates to lowland savannas and floodplains or to unspecified vegetation types. Sandstone habitats are only occasionally and briefly mentioned, and sandstone heaths are rarely mentioned. Occasionally, references to endemic sandstone species such as the black wallaroo (Bowman et al. 2001a) or Allosyncarpia ternata (Yibarbuk and Cooke 2001) indicate that accounts refer at least in part to sandstone habitats. There remains, however, an element of speculation in extrapolating the findings to reconstruct pre-contact sandstone fire regimes.

25 2: Fire regimes

Explorers' records cannot tell us about the reasons and methods of Aboriginal burning, and the only component of fire regimes they can describe is season. Explorers' records for the Northern Territory were examined by Braithwaite (1991) and Preece (2002), with the latter being the more comprehensive review. Despite the obvious problems with patchy records from explorers whose main interests were not fire regimes, Preece (2002) was able to conclude that most if not all landscapes were burnt and that burning continued more or less steadily throughout the entire dry season. In a similar review of explorers' records for the Kimberley region of northern Western Australia, Vigilante (2001) found a similar pattern for the higher rainfall coastal and subcoastal areas. Of particular interest is the evidence for widespread late dry season burning of sandstone habitats along the northwest Kimberley coast. It should be noted, however, that although the Kimberley abuts the western end of the range of Petasida, these reports are all from locations several hundred kilometres west of the nearest known occurrence of Petasida. The explorer Leichhardt, for whom Petasida is named, rarely mentions fire in the section of his journal describing his movements across the sandstone plateau. He does, however, describe the concentration of burnt areas around creeks and soaks (Leichhardt 1847). Preece (2002) quotes the explorer McKinlay, observing from the Arnhem Land Escarpment on 24 May 1866 and reporting 'innumerable daily bushfires' on the lowlands, but only that 'we occasionally see recent traces of them' atop the escarpment. Neither Preece nor Vigilante found any explorers' records of wet season burning.

The continuity of burning throughout the dry season is a consistent theme in the ethnographic and contemporary accounts, although most of these are from lowland savanna and floodplain landscapes. A few accounts from Kakadu and Arnhem Land, however, cover study areas which include some sandstone. The following discussion focuses on those papers.

Traditional burning was and still is carried out for a number of reasons: spiritual and cultural reasons, to protect resources such as yams and fruit trees, to clean the country, to ease movement and discourage snakes, to drive game or attract game to green pick, and to create firebreaks. The following description of traditional burning patterns still applies to a greater or lesser extent in parts of both western Arnhem Land (Lewis 1989;

26 2: Fire regimes

Russell-Smith et al. 1997a), including Kakadu, and north central Arnhem Land (e.g. Bowman et al. 2001a; Haynes 1985; 1991; Yibarbuk et al. 2001; Yibarbuk 1998). The season names are given in Gundjeihmi, but at least some of the central Arnhem Land groups recognise an equivalent seasonal calendar. The traditional pattern is to begin burning during the seasons of Banggerreng and Yegge (March–June), as soon as grass begins to cure enough to carry flame. These fires are kept small and burning does not start in earnest until the cool season of Wurrgeng in June or July. All these fires are extinguished overnight by low temperature, lack of wind and dew. Several authors have noted that higher daytime wind speeds help to bend the flames over, resulting in faster passage times and lower scorch heights. With the onset of the hot, dry Gurrung around September fires no longer go out at night and general ignition ceases. However, burning for driving game continues, but with the aid of careful preparation done earlier in the season and appropriate use of the wind, fires are deliberately run onto previously burnt areas and do not continue unchecked. This historical pattern was consistent across the Top End (Russell-Smith et al. 2003b). One of the main differences in contemporary Aboriginal burning patterns between Kakadu and central Arnhem Land is that large kangaroo drives are now rare or absent in sandstone areas of Kakadu.

In terms of the components of fire regimes, traditional savanna fire regimes in coastal and subcoastal areas consist of relatively small, very cool fires throughout the dry season, resulting in a fine-scale patchy mosaic of burnt and unburnt areas. In total relatively large areas are burnt annually, resulting in a high frequency, or return rate, of fires in many parts of the landscape.

In Western Arnhem Land Aboriginal people with tribal lands on the escarpment followed an annual migration pattern which saw high populations living around resource-rich billabongs on the lowland floodplains during the dry season, with a consequent drop in the sandstone population during the burning seasons (Russell-Smith et al. 1997a). Lewis (1989) records that the plateau habitats are much less exploited than the lowlands. These authors, however describe the use of hunting fires in the sandstone. These accounts, together with those of the explorers Leichhardt and McKinlay, give indications that sandstone burning was not as frequent as that on the lowlands.

27 2: Fire regimes

Accounts from the small areas which remain populated in central Arnhem Land make it clear, however, that Aboriginal fire practitioners in the sandstone burn for somewhat similar reasons to those in the lowland savannas, for example to protect monsoon rainforest patches, and also apply as much skill and diligence to the task (Yibarbuk and Cooke 2001; Yibarbuk et al. 2001). It is also clear that burning the plateau was and is a business taken very seriously by people who grew up on it (and who are still alive today) (Cooke 2000). Noteworthy in this literature is the emphasis placed on the burning of creek lines, which creates firebreaks that inhibit the progress of large late season fires (Yibarbuk and Cooke 2001). Yibarbuk et al. (2001) found fires to be much less intense in sandstone vegetation than in other vegetation types. Furthermore, fires in the inhabited areas of central Arnhem Land were found to be less intense, even in the late dry season and despite comparable fuel loads, than in western Arnhem Land. The authors suggest that high densities of the annual tall grass Sarga spp., a result of many successive years of late dry season fires, cause the higher intensities of fires in western Arnhem Land. In the same area of central Arnhem Land, both Haynes (1991) and Bowman et al. (2004) specifically note that Aboriginal burning begins later on the plateau than in the lowlands. Bowman et al. (2004) propose that this may also be due, at least in part, to lower grass biomass, particularly of annual Sarga spp.

2.2.2 Contemporary fire regimes

Contemporary fire regimes in the sandstone heaths are, not surprisingly, much better documented and understood than those of the recent pre- and post-contact past. There has been a large and increasing amount of attention paid by ecologists and land managers to savanna fire ecology in recent years, largely in response to land management and conservation requirements and, more recently, in an effort to understand the role of savanna burning in global warming. This research effort has been facilitated in large part by development of remote sensing and GIS techniques over the last 30 years or so, especially in those studies relating to the difficult to access sandstone heath habitats. However, many knowledge gaps remain.

28 2: Fire regimes

There are also several problems with the interpretation of fire histories from satellite imagery, summarised by Russell-Smith et al. (1997b) and many others. In particular, the resolution of the imagery limits the detection of fine scale patterns, leading to under- detection of small fires, which typically are more common in the early dry season. Positional errors also hinder the accurate interpretation of fire histories at very fine scales. Some fires, typically in the very late dry season and wet season, may be obscured by cloud cover and therefore go unrecorded. Some workers (Bowman et al. 2003) have reported problems with the fading of firescars. Nevertheless, remote sensing studies are typically undertaken with adequate consideration of the limitations and with extensive ground-truthing, and have proved an extremely valuable tool in advancing understanding of savanna fire regimes.

2.2.2.1 Season

Almost all fires in the heaths occur in the dry (winter) season, but there is a high degree of variation within that season. Most studies distinguish the early dry season (EDS: April–July) and the late dry season (LDS: August-November). It is a common theme in the ethnographic literature that the extent of LDS burning is higher now than it was under traditional burning patterns (e.g. Lewis 1989; Russell-Smith et al. 1997a). Data on the seasonality of sandstone fires in the NT are scarce, but those that exist support this conclusion. In one of the first studies to use satellite imagery to examine fire regimes in specific sandstone vegetation types Russell-Smith et al. (1997b) derived a fire history for Kakadu for the years 1980–1994 using Landsat MSS data. All the sandstone habitats were dominated by LDS fires, in contrast to the lowland habitats in which EDS fires were predominant. In the heath ('Spinifex') habitats LDS burning covered approximately four times the area of EDS burning. In contrast to the lowland habitats and to the park as a whole, there was no significant relationship between the extent of EDS and LDS burning on the plateau as a whole (Gill et al. 2000). This finding indicates that the current level of EDS burning does not limit subsequent fires at broad scales, (although it may do so at finer scales). In an analysis of Landsat TM imagery for the 9 years 1989– 1997 in Nitmiluk National Park (henceforth 'Nitmiluk'), Edwards et al. (2001) found a

29 2: Fire regimes similar, though weaker and not statistically significant, trend of late burning in heath ('low open woodland') habitats: 11% of the heath area in the EDS and 16% in the LDS.

2.2.2.2 Frequency and fire intervals

Fire frequency and the related but ecologically quite distinct variables, fire return interval ('interfire interval', 'fuel age', or 'stand age') and years since burnt (YSB: a special case of fire return interval) are perhaps the best documented components of fire regimes in the sandstone heath habitats, largely because of their influence on populations of obligate seeder species. Simple descriptive statistics such as means, however, are not commonly reported.

For Kakadu, a map of fire frequencies provided by Russell-Smith et al. (1997b) indicates that for the plateau as a whole the majority of sites were burnt on average 0–4 times within the 15 year study period, but for individual pixels both the mode and mean fire frequency for the period were approximately 4. An assessment for the subsequent 5 years, 1996–2000, found that 40% of the heaths were burnt once, 23.5% were burnt 2–5 times and 36.5% remained unburnt (Edwards et al. 2003). In Nitmiluk over 40% of the heath habitats were burnt at least 3 times in the 9 years from 1989–1997, while the value over 8 years (1990–1997) for heaths in near Darwin, which does not contain Petasida populations, was closer to 80% (Edwards et al. 2001). In a similar analysis at Bradshaw Station in the Victoria River District, where Petasida have not been recorded for over 150 years, 96% of the heaths were classified as burnt at least three times within 10 years (1990–1999), with a mean and mode of approximately 4 (Yates and Russell-Smith 2003). At all these sites, except Bradshaw, fire frequencies in the sandstone heaths were less than those for the property as a whole, and indeed, less than those for the plateau area of each property as a whole.

Analysis of individual pixel data from the 15 year Kakadu data set showed that 69% of the heath area had been burnt at least once with a return interval of three years of less, and that 64% never experienced return intervals longer than five years (Russell-Smith et al. 2002). Russell-Smith et al. (1998) found autocorrelations between three fire interval variables measured in the heath areas of Kakadu: fire frequency, YSB and shortest

30 2: Fire regimes interfire interval. Further, YSB was positively correlated with densities of tall shrubs (both 'all shrubs' and 'obligate seeders'), fuel loads and litter cover. Spinifex fuel loads were typically high enough to support intense fire within two years, but in rocky areas occupied by Triodia plectrachnoides this time was 3–4 years.

2.2.2.3 Intensity

The intensity of a fire is influenced by a complex interaction of factors including available fuel, moisture, temperature, chemical factors, wind and topography (Whelan 1995). Most of these factors are in turn influenced by season, and in the lowland savannas of Kakadu fuel loads, fuel composition, fuel moisture, local weather and fire intensity all show distinct differences between the EDS and the LDS (Williams et al. 1998). Fire severity is closely related to intensity; severity is a measure of the impact of flames on vegetation while intensity is generally a measure of the heat energy produced. Using severity as a surrogate for intensity, Price et al. (2003) used char heights to assess fire intensity in heaths in Kakadu and Arnhem Land and found a clear increase in intensity throughout the dry season. Similar results were reported by Russell-Smith and Edwards (2006) who assessed fire severity in five vegetation types in a combined plot dataset for Kakadu and Nitmiluk. However, while there was a clear increase in severity in the LDS in all vegetation types, the difference was least in the heaths. In the LDS the heaths showed the highest proportion of high severity fires of all vegetation types.

All these studies recognised that some high intensity fires occur in the EDS and some low intensity fires occur in the LDS. Indeed, areas of different intensity often occur within the same fire, due, for example, to changes in wind direction relative to fire fronts and to changes in conditions between day and night (Gill et al. 2000).

2.2.2.4 Extent

In general the relative area of sandstone heath burnt each year is less than that of the lowland savanna. Analysis of Landsat imagery for Kakadu indicates that between 1980– 1995 an average of 22% of the heath area was burnt annually (Edwards et al. 2003) compared with 28% for the Kakadu plateau unit as a whole and 55% for the lowland savannas (Russell-Smith et al. 1997b). However, the proportion burnt annually for the

31 2: Fire regimes plateau unit during the period varied between 0 and 80% (Gill et al. 2000). Gill et al. (2000) also found that the relationship between proportion burnt and stand age ('probability of ignition at a point' [PIP] or 'hazard function') for the plateau was relatively constant, though variable, in contrast to the lowland vegetation units for which a clear negative slope was evident.

For heaths in Kakadu the mean annual proportion burnt declined (but not statistically significantly) to 19% for the period 1996–2000 (Edwards et al. 2003). Elsewhere, the mean proportion of heaths burnt annually was 27% in Nitmiluk for 1989–1997 (Edwards et al. 2001) and in the Kimberley region of WA 12% of sandstone was burnt annually between 1990–1999 (Fisher et al. 2003).

2.2.2.5 Patchiness

Patchiness describes patterns which arise in a number of ways and at a range of scales, but the two patterns of primary consideration here are, firstly, the spatial mosaic of burnt patches and the unburnt spaces between fires, and, secondly, the internal pattern of unburnt gaps within individual firescars. Neither of these patterns are particularly well documented for the sandstone heaths. Indeed, understanding the internal patchiness of burnt landscapes is recognized as a critical issue for further understanding how savanna fire regimes vary (Russell-Smith et al. 2003a).

While figures for the total extent of burning in the heaths have been reported (above), there are very few studies on the sizes of individual fires or the spatial patterns of burning in the heaths. A general indication of the order of fire size may be gained by looking at broader areas, however. In the whole of Kakadu, the approximate average size of individual fires, assessed by analysis of Landsat MSS imagery, declined steadily from 3 km2 in 1980 to 0.6 km2 in 1995. To the east of Kakadu lies the 25,000 km2, hitherto largely unmanaged, Western Arnhem Land Fire Abatement sub-region (WALFA). This area covers three quarters of the Arnhem Land plateau and 21% of the area is vegetated by sandstone heath. Between 1997–2005 the average fire size ('fire affected area') estimated from Landsat TM imagery for the whole WALFA area, was 2 km2 in the EDS and 9 km2 in the LDS (Yates et al. 2007). The significant result of this

32 2: Fire regimes study, however, was that on average a very small number (5: 0.04%) of huge fires, >1000 km2, accounted for 82.5% of the total burnt area, compared with 13,251 fires (99.5%) < 10 km2.

These results are reflected in the work of Gill et al. (2003), who studied the frequency distribution of patch sizes in both a model system and an analysis of the Landsat (MSS and TM) imagery for Bradshaw Station described previously by Yates and Russell- Smith (2003). In both the theoretical (model) and real world results Gill et al. (2003) found a log-log linear frequency distribution of patch sizes, both of burnt patches and of the unburnt 'islands' within fire scars. Price et al. (2003) examined fine-scale patchiness within firescars in sandstone heath vegetation in Western Arnhem Land. Using transects of contiguous 5 x 5 m quadrats placed completely within existing firescars, they found that on average EDS fires burnt 64% and LDS fires burnt 84% of the area. The frequency of unburnt gaps ranged from 5–12 km-1 and the mean length ranged from 11– 45 m. Unburnt gap frequency was not significantly related to season but both gap length and variability of gap length declined through the dry season. Patchiness was higher in rocky areas, especially in the LDS, but was not affected by slope or relief.

Price et al. (2003) noted that many unburnt gaps were too small to be included in the analysis at the scale of their study (5 m), with some being as small as a single grass tussock. The average frequency of gaps <5 m was 6.3 km-1. Taking such small gaps into account gives an overall perspective on burning patterns which is somewhat different to that given by simple percentage burnt figures: 72% of all quadrats within EDS firescars contained some unburnt area but only 30% of LDS quadrats did.

Price et al. (2003) focussed much of their attention on the unburnt gaps. This information is important in relation to Petasida as such gaps provide refuge during, and resources after, fire. However, in order to fully assess the impact of fires on Petasida populations it is also necessary to describe the spatial patterns of burnt areas within firescars. For example, the usefulness of unburnt gaps to Petasida is much diminished if they are all clustered in a small area within a firescar. In such a case it would be possible to burn an entire Petasida population, with no access to refuges. In order to determine

33 2: Fire regimes the spatial distribution of unburnt gaps and their availability to Petasida, it is necessary to also describe the spatial and size distributions of the burnt areas.

2.3 Methods

Two transect-based datasets were examined for burning patterns. The first dataset, hereafter referred to as the 'Price dataset', was collected by scientists in the NT Department of Natural Resources, Environment and the Arts (NRETA) and is described in detail in Price et al. (2003). Briefly, 12 transects of contiguous 5 x 5 m quadrats were placed arbitrarily within the scars of five fires in sandstone heath vegetation in western Arnhem Land. Fires were classified as early dry season (EDS: May–early June), mid-dry season (MDS: late June–August) or late dry season (LDS: September–October). Data recorded in each quadrat included % vegetation burnt, % rock cover, relief and char height. Description of spatial patterns concentrated on the unburnt gaps, and patterns of the burnt sections were generally not described except in graphically presented transect profiles.

Price et al. (2003) defined unburnt gaps by effects characterizing their edges rather than their interiors; the beginning of gaps was arbitrarily determined by the transition from 75% of vegetation burnt to 25% in consecutive quadrats (emphasis added). This complicates the definition of burnt sections as they are not necessarily simply the sections between the unburnt gaps. In re-examining their data to study both burnt and unburnt sections I adopted a simpler approach, defining quadrats arbitrarily as burnt if 50% was burnt and as unburnt if <50% was burnt, and sections as groups of consecutive burnt or unburnt quadrats. In addition, the distribution of sections consisting of consecutive quadrats in which 100% of the vegetation was burnt or unburnt were also examined.

Thus, the following types of transect sections were examined:

50% burnt – sections containing a fine (<5 m) scale mosaic, but more burned than unburned;

34 2: Fire regimes

<50% burnt – sections containing a fine (<5 m) scale mosaic, but more unburned than burned;

100% burnt – sections within which all vegetation at ground level was burned; and

100% unburnt – section consisting no burned vegetation at all.

The intention was to define a range of values that describe the patchiness characteristics of sandstone heath fires, in terms that are relevant to the ecology of grasshoppers, and that would serve as a basis for modelling fires to be incorporated in grasshopper population models (Chapter 5). Because no data exist on direct mortality of Leichhardt's grasshoppers due to fire, the proximity to flames (of any intensity) that is tolerable to grasshoppers is also unknown, and thus the minimum unburnt gap size that is capable of providing refuge to grasshoppers is also a matter for conjecture. The two methods of defining a 'burnt' quadrat were used here to provide a range from the most hazardous situation for grasshoppers (100% burnt: no refuges at all) to a situation in which it is conceivable that the unburnt area (50% of the quadrat) may be sufficient to afford some refuge to grasshoppers, at least in low intensity fires.

The following summary statistics were calculated for each category of section: mean section length, frequency (sections km-1) and maximum section length, together with the proportion (%) of total quadrats burnt or unburnt.

While all the transects studied by Price et al. (2003) were in sandstone heath vegetation, not all areas were suitable habitat for Petasida and Pityrodia, which are largely associated with rocky substrates (Chapter 2). In order to better understand burning patterns within the habitat of Petasida, the data were also examined with quadrats from rock free areas removed from the data set. This was done by selecting and disregarding all groups of five or more consecutive quadrats with less than 25% rock cover. The same summary statistics were calculated for the remaining 'rocky areas', which accounted for 65% of the total length of transects.

35 2: Fire regimes

The second transect-based dataset, referred to hereafter as the 'Pityrodia site dataset', was collected during the current study. Data were collected from six sites in Kakadu and Nitmiluk National Parks (Figure 2.1; Table 2.1), all of which were known to support populations of Pityrodia. At four sites, 100 m transects consisting of contiguous 5 x 5 m quadrats were laid out randomly with respect to start point and direction, within a 500 x 500 m square placed within the scar of recent fire. At two sites a single long transect was laid out because the size and shape of the Pityrodia patch restricted the useful size of the study area and, in one case, topographic constraints also prevented the use of random transects. All fires had occurred in August and would therefore be classified as MDS in the 3-category scheme adopted by Price et al. (2003), or as LDS in the binary scheme used by Russell-Smith et al. (1997b) and others. In each quadrat the following variables were visually estimated: % vegetation burnt, % rock cover, and char (height of burning damage on stems of shrubs and trees). These data were treated in the same way as those of the first dataset described previously. Mean char height was estimated by averaging mean values from each group of 10 consecutive quadrats. 'Rocky areas' in this dataset accounted for 80% of the total length of transects.

In order to investigate possible association between grasshopper presence and Fire regime variables, satellite data for 28 sites supporting Pityrodia populations in Kakadu and Nitmiluk were examined. These sites were all study sites previously used in this project and were not randomly chosen. The Landsat MSS and TM derived, GIS-based fire history database held by the Northern Territory Bushfires Council was queried to establish whether a single point near the centre of each site had been mapped as falling within a firescar during each year from 1980–2004 inclusive. The fire frequency (1980– 2004); the mean, maximum and minimum fire interval; and years since last burn (YSB) were recorded for each site. YSB was taken from the year (2002–2004) in which the site was last visited to inspect for the presence of Petasida. At the time of the last visit, 16 sites supported Petasida populations and 12 did not. Differences in all the variables, between sites with and without Petasida, were tested for using Wilcoxon rank-sum tests.

36 2: Fire regimes

Figure 2.1 Locations of sites for transect studies of patchiness within firescars. Two sites occur in close proximity at Gubara (GTG and GTB; see Table 2.1).

37 2: Fire regimes

Table 2.1. Details of sites and transects for studying firescar patchiness. All fires occurred in August 2002. Sites GTG and GTB are within the same firescar. NNP = Nitmiluk National Park; KNP = Kakadu National Park.

Site Location Lat Long Date No. of Transect Transects length (m)

NIT Katherine Gorge, 14° 18.59' 132° 25.62' 6 Nov 02 6 100

Edith Edith Falls, NNP 14° 12.76' 132° 11.16' 26 Nov 7 100

NAE East Alligator, KNP 12° 27.24' 132° 56.10 11 Sep 10 100

GTG Gubara, KNP 12° 49.18' 132° 52.96' 3 Nov 02 11 100

GTB Gubara, KNP 12° 49.16' 132° 53.10' 15 Nov 1 250

Nou Nourlangie Rock, 12° 51.35' 132° 48.22' 5 April 03 1 500

2.4 Results

The highest frequencies of quadrats in both datasets were either 0-10% or 90-100% burnt (Fig. 2.2), with the combined frequencies for those two intervals varying between 39% (early dry season: EDS) and 77% (late dry season: LDS). Generally, the frequencies of quadrats declined to low levels towards the centre of the range, indicating that the results were relatively insensitive to errors in the assessment of '% burnt' at values close to 50%, or indeed to the arbitrary placement of the point of separation between burnt and unburnt quadrats at 50%. A slight increase in frequencies of quadrats falling in the middle two intervals (Fig. 2.2, particularly [d]) is probably evidence of a tendency for some observation error at intermediate values of '% burnt'.

In the Price dataset (Price et al. 2003), the proportion of quadrats with more than 50% of ground level vegetation burnt ranged from 75% in the EDS to 87% in the LDS. The proportion of quadrats with all ground level vegetation burnt varied from 26% in the EDS to 69% in the LDS quadrats (Table 2.3). In Pityrodia site dataset (mid dry season: MDS; pooled data), at least 50% of vegetation was burnt in 70% of quadrats and 100% of vegetation was burnt in 54% of quadrats (Table 2.4). Throughout this discussion 'burnt vegetation' refers only to ground level vegetation.

38 2: Fire regimes

For all fires in both datasets, mean lengths of burnt sections consisting of contiguous quadrats classified as 'burnt' were under 50 m except in the LDS, but the maximum length was always over 100m. There were at least 10 burnt (and unburnt) sections km-1.

In all cases the majority of transect sections, either burnt or unburnt, were of short length, with frequency declining with increasing section length (Figs. 2.3 a,c,e; 2.4 a,c,e). However, a considerable proportion of the burnt area occurred within the few long sections (Figs. 2.3 b,d,f; 2.4 b,d,f). In the Price dataset, at least 46% quadrats in which 50% of vegetation was burnt occurred in sections 100 m long, regardless of season. Even at the 100% burnt level, in the LDS, 59% of burnt quadrats occurred in sections 100 m. EDS and MDS values at the 100% level were much lower than the LDS value, with 21% and 16% of burnt quadrats, respectively, occurring in sections over 100 m long.

In the Pityrodia site dataset for MDS fires (pooled data), 59% of quadrats in which 50% of vegetation was burnt occurred in sections 100 m long. 27% of quadrats which were 100% burnt occurred in sections 100 m. Four of the five fires sampled contained at least one section of 20 contiguous quadrats (i.e. 100 m) with 100% of the vegetation burnt. The one that did not contained one fully burnt transect broken only by a single quadrat which was 99% burnt. For individual fires the proportion of quadrats which were 100% burnt ranged from 18% at East Alligator to 90% at Nitmiluk (Table 2.4). The intense fire at Nitmiluk occurred in an area which had been largely unburnt for at least 11 years (Fig. 2.7). Corresponding figures for other sites were: Edith 64%; GTG 79%; GTB 54%; and Nou 29%.

In the Price dataset the only unburnt sections 100 m occurred in the EDS, but in the Pityrodia site dataset (MDS), sections 100 m with no burnt vegetation at all occurred at two of the six sites.

Removing the rock-free portions from the datasets had varying effects (Tables 2.2, 2.3). Mean lengths of burnt sections decreased but there was very little effect on mean lengths of unburnt sections. Similarly, maximum lengths of burnt sections decreased but

39 2: Fire regimes maximum lengths of unburnt sections only decreased appreciably in the EDS (Price dataset only). Generally, the frequency of both burnt and unburnt sections increased. The proportion of unburnt quadrats increased and of burnt quadrats decreased. However, while the unburnt area declined from the EDS to the LDS, the trend of decreasing unburnt area throughout the dry season was apparently less strong than that found by Price et al. (2003): in rocky areas the lowest proportions of unburnt quadrats occurred in the MDS rather than the LDS.

In the Pityrodia site dataset, mean char heights for each site varied from 1.5 m to 3.2 m and overall, over 93% of values were over 1 m. These values are very much higher than those obtained by Price et al. (2003) and it is assumed that methodological differences in the assessment of burning damage on stems account for this, rather than actual differences in fire intensity.

Fire regime variables derived from Landsat-based firescar mapping are given in Table 2.4. No significant difference between sites with and without Petasida present was detected for any of the variables.

40 2: Fire regimes

Figure 2.2. The frequency of quadrats falling into 10% intervals for values of '% burnt'. (a-c) Price dataset; (d) Pityrodia site dataset.

41 2: Fire regimes

Figure 2.3. (a, c, e, g) Frequency distributions for burnt and unburnt section (gap) lengths in transects within firescars, derived from the data of Price et al. (2003). Sections consist of consecutive quadrats classified as burnt or unburnt if (a) 50%, (c) < 50, (e) 100% or (g) 0% of vegetation within them was burnt. (b, d, f, h) Histograms of the same dataset showing the proportion of the total burnt or unburnt distance occupied by sections within length classes. EDS = Early dry season; MDS = Mid-dry season; LDS = Late dry season.

42 2: Fire regimes

Figure 2.4. (a, c, e, g) Frequency distribution of burnt and unburnt sections of 100 m transects within mid-dry season firescars occurring within habitat supporting Pityrodia populations. Sections consist of consecutive quadrats classified by the proportion of vegetation burnt: (a) 50%; (c) < 50%; (e) 100%; (g) 0%. (b, d, f, h) Histograms of the same dataset showing the proportion of the total burnt or unburnt distance occupied by sections within length classes.

43 2: Fire regimes

Figure 2.5. Scattered Pityrodia Pungens supporting a Petasida population amongst spinifex grass at Katherine Gorge in Nitmiluk National Park in February 2002.

Figure 2.6. The same scene in November 2002 after being by burnt by wildfire in August.

44 2: Fire regimes

Figure 2.7. Petasida ephippigera habitat in Nitmiluk National Park three months after a hot fire. Unburnt gaps are represented by one spinifex tussock in the foreground and several in the left middle distance.

45 2: Fire regimes

Figure 2.8. An unburnt gap containing Pityrodia jamesii at upper Gubara (site GTB) in Kakadu National Park.

Figure 2.9. Site GTG at upper Gubara in Kakadu National Park in November 2002, three months after being burnt. In the absence of large rocks there are very few unburnt patches .

46 2: Fire regimes

Table 2.2. Descriptive statistics for burnt and unburnt sections of transects within firescars, derived from the data of Price et al. (2003). Sections consist of consecutive quadrats classified as burnt if (a) 50% or (c) 100% of vegetation within them was burnt. All: All quadrats in each transect are included. Rocks: All groups of 5 or more consecutive quadrats with < 25% rock cover were disregarded.

EDS MDS LDS

All Rocks All Rocks All Rocks

Mean length (m) (n, ±SE) 38.0 (61, ±5.7) 30.7 (38, ±6.0) 48.5 (55, ±6.4) 29.6 (55, ±4.2) 58.3 (39, ±11.8) 38.0 (43, ±7.4) (a) -1 Quadrats Frequency (Sections km ) 18.8 21.1 16.9 26.4 14.3 20.28 >50% Max length (m) 250 140 220 130 330 250 Burnt

Burnt Quadrats (% of total) 75.0 64.5 84.5 78.4 86.9 69.6

Mean length (m) (n, ±SE) 15.8 (59, ±3.1) 17.2 (36, ±4.1) 11.7 (53, ±1.4) 11.5 (43, ±1.6) 13.2 (38 ±1.9) 13.4 (37±1.9) (b) -1 Quadrats Frequency (Sections km ) 18.2 20.0 16.3 20.7 13.67 17.45 >50% Max length (m) 135 115 60 60 55 55 unburnt

Unburnt Quadrats (% of total) 25.0 35.5 15.5 21.6 13.1 30.4

47 2: Fire regimes

Table 2.2. continued

EDS MDS LDS

All Rocks All Rocks All Rocks

Mean length (m) (n, ±SE) 14.9 (56, ±3.2) 11.6 (29, ±1.5) 17.3 (75, ±2.3) 13.5 (52, ±2.1) 37.5 (30, ±8.1) 27.6 (29. ±7.5) (c) -1 Quadrats Frequency (Sections km ) 17.2 16.1 23.1 25 10.7 13.7 100% Max length (m) 175 40 120 95 185 185 burnt

Burnt Quadrats (% of total) 25.9 18.6 39.9 33.7 68.9 37.7

Mean length (m) (n, ±SE) 25.9(16, ±9.2) 23.9(14, ±8.9) 11.7 (15, ±2.3) 12.3 (13, ±2.3) 12.5 (18, ±2.2) 12.5 (18, ±2.2)

(d) Frequency (Sections km-1) 24.6 38.9 23.1 31.2 6.5 8.5 Quadrats 0% burnt Max length (m) 125 105 30 25 40 40

Unburnt Quadrats (% of total) 12.8 18.6 5.4 7.7 8.1 10.6

48

Table 2.3. Summary statistics for burnt sections of transects within MDS firescars at sites supporting populations of Pityrodia. Details are as described for Table 2.2. All Rocks

Mean length (m) (n, ±SE) 37.7 (78, ±3.7) 27.8 (75, ±3.0)

(a) Quadrats >50% Frequency (Sections 100 m-1) 1.88 2.27 burnt Burnt Quadrats (%) 70 63.1

Max length (m) 100 100

Mean length (m) (n, ±SE) 21.4 (58, ±3.6) 21.0 (58, ±3.6)

(b) Quadrats >50% Frequency (Sections 100 m-1) 1.39 1.75 unburnt Unburnt Quadrats (%) of total 30.0 36.9

Max length (m) 100 100

Mean length (m) (n, ±SE) 27.1 (82, ±3.2) 19.1 (77, ±2.5)

(c) Quadrats 100% Frequency (Sections 100 m-1) 2.0 2.36 burnt Burnt Quadrats (%) 54.4 45.1

Max length (m) 100 100

Mean length (m) (n, ±SE) 25.8 (39, ±5.0) 24.2 (38, ±4.7) (c) Quadrats 0% Frequency (Sections 100 m-1) 1 1.2 burnt Unburnt Quadrats (%) of total 24.5 28.2

Max length (m) 100 100

Table 2.4. Mean values (±SE) for fire regime variables for 16 sites supporting populations of Petasida and 12 sites without Petasida present at the last inspection in Kakadu and Nitmiluk National Parks, with P values derived by Wilcoxon rank sum test. All sites supported populations of Pityrodia. Data were extracted from digital firescar maps derived from Landsat MSS and TM imagery for the period 1980–2004.

Fire variable All sites (n=28) Hoppers absent Hoppers present P

Frequency (1980–2004) 6.68 (±0.49) 6.33 (±0.53) 6.94 (±0.78) 0.39

Mean interval (yr) 4.86 (±0.75) 4.29 (±0.43) 5.29 (±1.29) 0.4

Max interval (yr) 8.82 (±0.87) 9.67 (±0.85) 8.19 (±1.38) 0.14

Min interval (yr) 2.18 (±0.79) 1.33( ±0.26) 2.81( (±1.37) 0.56

Years since last burn 4.86 (±0.82) 4.67 (±1.05) 5.0 (±1.23) 0.83

2: Fire regimes

2.4 Discussion

Most of the analysis presented has focused on variables related to patchiness and seasonal influences on those variables. Fire frequency, extent and season of occurrence at a broad scale are reasonably well documented through satellite-based studies. At the fine scale, however, patterns of burning at any point within those mapped firescars are dependent upon the internal structure – patchiness – within the perimeters of firescars. All the sandstone heath fires studied within or close to the habitat of Leichhardt's grasshopper showed some degree of patchiness. The degree of patchiness was highly variable, particularly in regard to the total area burnt or unburnt within firescars. Rockiness has a clear influence on patchiness, and indeed the evidence is strong that fires within the rocky habitat of Pityrodia are more patchy than sandstone heath fires in general. Within this constraint, other factors such as fuel load and weather are likely to exert an influence. Season of burn appears to be strongly influential, but the pattern of that influence is not simple.

While frequency distributions of patch sizes (Figures 2.3, 2.4) clearly indicate that the majority of burnt and unburnt patches are relatively small, it is also clear that under some circumstances high proportions of total burnt areas did indeed occur in large, unbroken burnt patches, particularly in the late dry season (LDS). By contrast, large unburnt patches are less common, and only occurred in one of the fires studied by Price et. al (2003) and two of the six Pityrodia site fires. Two important points relevant to Petasida ecology emerge from these results. First, there is considerable variation in the degree of patchiness in sandstone heath fires, and second, although all fires were indeed patchy, at a relatively fine scale not all areas within each fire could be described as patchy.

The evidence discussed in the introductory literature review clearly points to a strong seasonal influence on many fire regime variables in the northern Australian savannas. In general, the results (table 2.2) show similar trends, but the picture presented here for the sandstone heaths is slightly more complex. In the Price dataset, considered in its entirety, there was an unambiguous increase in the proportion of burnt quadrats from

50 2: Fire regimes early dry season (EDS) to LDS, particularly those in which 100% of the ground vegetation was burnt, and a corresponding decrease in the proportion of unburnt quadrats. Similarly, the mean length of burnt sections increased and the frequency decreased through the dry season.

For almost all of the measured or derived variables in the Price dataset, there is a clear difference between the EDS and LDS values, and these differences are consistent with the expected trends described in the literature. For some variables, however, the extreme values occurred in the mid dry season (MDS). Examples are maximum length of burnt (both 50% and 100%) and 100% unburnt sections, and frequency of 100% burnt sections. It is possible that the single MDS fire studied by Price et al. (2003) was unusual and unrepresentative, and that future studies that increase the collective sample size will show a more consistent trend. Another interpretation, however, is that the result simply emphasises the well recognised point, discussed in the introductory literature review, that despite trends, intense fires may occur early and cool fires late, and that variations can occur within individual fires.

After adjusting the Price dataset by removal of data for the rock-free areas, most variables showed a similar trend from EDS to LDS, but with two important differences. In most cases the difference between EDS and LDS values was reduced when compared to results for the full dataset, and more of the variables showed their most extreme values in the MDS. Due to methodological differences, particularly in transect length and layout, the results from the Pityrodia site dataset (Table 2.3), which were collected within rocky habitat, are not all directly comparable to those for the Price dataset. Nevertheless, differences in the values of some variables between the two datasets, and particularly the proportion of quadrats that were 100% burnt, emphasise the highly variable nature of MDS fires.

There are several important implications to be drawn from this analysis. First, all fires are variable, including those in rocky areas. While variability of many fire regime components is reduced in rocky areas, it is not eliminated. Second, season has an influence on patchiness and other fire regime variables. This influence is in the form of a

51 2: Fire regimes trend rather than being absolute, and strong divergences from the trend do occur. Third, the influence of season is mitigated or dampened by rocky substrates. In addition, the results support the conclusion of Price et al. (2003) and others that rocks provide some protection from fire.

An examination of Landsat MSS- and TM-based firescar maps did not reveal any relationship between fire frequency and interval variables and Petasida presence or absence (Table 2.4). However, the results should be interpreted in the light of the method. There are several sources of inaccuracy. First, not all the area within a mapped firescar is burnt, and the results presented in Table 2.4 were obtained by interrogating the value of a single pixel in the centre of a study site. Second, only some patches of habitat within each study site contain grasshoppers, and the single pixel assessed was not necessarily within one of those. Third, any one pixel scored as burnt (or unburnt) may contain both burnt and unburnt patches. At the resolution of Landsat imagery, it is entirely possible for areas large enough to hold surviving grasshoppers to remain unburnt within the area of one pixel. All these factors mean that the fire frequency or interval interpreted within a mapped firescar cannot be accurate for each point within that firescar.

The results indicate, therefore, that the simple process employed here, based on fire regime variables derived from Landsat-based firescar mapping, is inadequate for the accurate prediction of the presence of Petasida (Table 2.4). This does not necessarily diminish the importance of the role of fire regimes in Petasida population dynamics. Rather, it reinforces the findings of Price et al. (2003) that the current resolution of Landsat imagery is inadequate for the detection and analysis of fire regime variables at the fine scales measured for sandstone heaath fires. It is probable that phenomena occurring at such fine scales are crucial in the dynamics of fire and Petasida populations (the spatial distribution patterns of Petasida and its habitat are investigated and discussed in the following chapters). Given the dependence of fire frequencies and intervals occurring at very fine scales, or indeed at points, on the degree and pattern of patchiness within fires, a modelling approach capable of incorporating and analysing spatially realistic representations of fire patchiness is likely to prove most profitable in

52 2: Fire regimes patchiness within fires, a modelling approach capable of incorporating and analysing spatially realistic representations of fire patchiness is likely to prove most profitable in elucidating the system. Table 2.4 is discussed further, in the context of modelling results, in Chapter 5.

The results of the introductory literature review provide a good description of sandstone heath fire regime components such as frequency and extent upon which to base fire models. The results of the analysis of the two datasets presented here add to the capability to model fires accurately by providing good descriptions of the internal spatial structure of fires. These patchiness descriptions allow for the setting of appropriate and realistic default values for patchiness model parameters and variables, such as total unburnt area within firescars or mean unburnt gap size. More importantly, however, the results provide an appropriate range of values to test the relative effects of variations in the components of fire regimes on grasshopper populations. Where knowledge gaps exist, for example in the rate of direct mortality of grasshoppers due to fire (see Chapter 4), the results also provide guidance for investigation of the sensitivity to variation in parameter values.

53

Chapter 3

The habitat of Leichhardt’s grasshopper – floristic and environmental relations and distribution patterns of Pityrodia

Chapter 3: The habitat of Leichhardt’s grasshopper – floristic and environmental relations and distribution patterns of Pityrodia.

3.1 Abstract

Leichhardt’s grasshopper (Petasida ephippigera) is endemic to the sandstone country of the Northern Territory (NT) of Australia, where is feeds almost exclusively on shrubs within the genus Pityrodia. This study focused on the seven species of Pityrodia known to be food plants for Petasida in the wild, and treated these plants as the primary determinant of Petasida distribution. The aims were to describe the distribution of Pityrodia at a range of scales and to explore the environmental and floristic associations of Pityrodia.

Herbarium records for Pityrodia in the NT were mapped in order to define the range. Description of the habitat of Pityrodia was undertaken by analysis of a series of three datasets, each describing a narrower range of habitat. The first two datasets were extracted from the existing Nitmiluk Vegetation Survey (NVS: Michell et al. 2004). The third dataset was collected during this study from sites in Kakadu and Nitmiluk National parks. Environmental relations were investigated using Generalized Linear Modelling (GLM) and floristic relations using Non-Metric Multidimensional Scaling (NMDS) and Indicator Species Analysis. Pityrodia patchiness was examined by describing presence/absence patterns at a local (km) scale and patch sizes were estimated using Two Term Local Quadrat Variance (TTLQV).

The results support the observations that the Pityrodia species upon which Petasida feeds are confined to sandstone habitats and that their distribution is distinctly patchy at a range of scales. Patch sizes at a local scale varied from 20-100 m. The single most important variable associated with Pityrodia presence is rock cover, particularly of large rocks and boulders. Pityrodia is more weakly associated with open vegetation and with shallow, sandy soils. The floristic associations of Pityrodia are dominated by sandstone

3: Habitat heath species, and in particular short-lived obligate seeder shrubs. All long lived plant species associated with Pityrodia were resprouters.

Even within the strict geomorphologic and edaphic bounds of the sandstone landforms, Pityrodia is further specialized for particularly rocky substrates. Rocks offer some fire protection and fires associated with rocky areas are typically patchy, but total or near total protection does not promote Pityrodia. The results suggest that Pityrodia habitat is subject to contemporary fire regimes of intermediate frequency and some patchiness. The central importance of rocks as a determinant of Pityrodia habitat implies that extensive habitat suitability modelling would be either difficult or unproductive, and that modelling efforts should more profitably focus on variations in fire regimes and on grasshopper dynamics.

3.2 Introduction

Petasida ephippigera is endemic to the NT, where it occurs in sandstone country across the northern third of the Territory, known as the ‘Top End’. Within this range, apart from sandstone the overwhelmingly most consistent habitat feature of Petasida is the presence of Pityrodia. Given this dependence of Petasida on Pityrodia as its primary, and possibly sole, wild food source, it is probable that Pityrodia is the primary determinant of the distribution of Petasida. This chapter will therefore focus on the distribution and habitat preferences of Pityrodia.

At the regional scale the Pityrodia host plants for Petasida are restricted to the sandstone escarpments and plateaux across the Top End. The sandstone habitats are generally rugged and dissected by deep, eroded fissures forming gorges and narrow valleys. Observations suggest that Pityrodia typically occurs on rock pavements or shallow sandsheets overlying rocky strata, sandstone outcrops and coarse scree slopes (Lowe 1995; Wilson et al. 2003). The vegetation communities of the sandstone include open Eucalypt forests, monsoon forest and ephemeral swamps. However, Pityrodia appears to be restricted to the sandstone heath communities. Furthermore, Pityrodia displays markedly patchy distribution patterns within the heath communities. It is not currently

55 3: Habitat known whether the patchy distribution patterns of Pityrodia are a result of physiological or other habitat preferences, stochastic processes of colonization and local extinction, or other influences such as fire regimes.

Figure 3.1. Aerial view of the Mt Brockman sandstone outlier of the Arnhem Land escarpment, taken about 2 km north of Gubara. The rocky outcrops and sandstone pavements are typical habitat for Petasida.

An understanding of the distribution patterns of Pityrodia and the influences on them are important for several reasons. First, a description and understanding of the habitat preferences of Pityrodia are a prerequisite to an exploration of the fire ecology of both Petasida and its host plants, not least because they are necessary for understanding the basic ecology of the species. Without such knowledge, of which little yet exists, such an exploration the fire ecology would be limited and unlikely to be fruitful. Second, in exploring the environmental and floristic correlates of Pityrodia distribution, it may be

56 3: Habitat possible to uncover any patterns or underlying gradients (or absence thereof) and thereby shed some light on the processes influencing distribution. Knowledge of the environmental and floristic correlates may provide some insight into fire ecology, in that environmental conditions or characteristic suites of species may be associated with particular fire regimes, and may indeed influence that behaviour.

This study commences with a description of the distribution of Pityrodia at a regional scale, derived from records held by the Darwin Herbarium. This effectively defines the potential range of the grasshopper’s habitat and provides a useful aid to predicting occurrence. This knowledge is important for management of the species, including fire management. Further field studies aimed to investigate distribution patterns at a local scale (ha–km2). Distribution patterns at this scale may profoundly influence the population and metapopulation dynamics of the grasshoppers, particularly through their influence on dispersal, extinction and the likelihood of colonization after local extinction. Perhaps most importantly, patchiness of Pityrodia may influence the likelihood of local extinction through an interaction with the patchiness of fires. For example, the probability of local extinction of Petasida will be higher with small habitat patches and large fire patches.

Elith (2000) and Rushton (2004) have recently reviewed methods for modelling species habitat and species distribution. Both reviewers describe the success of generalized regression approaches and Rushton discusses the increasing importance of the information-theoretic approaches advocated by Burnham and Anderson (2002). Both approaches will be adopted here. Of the multivariate approaches, the non-parametric methods described by Clarke (1993) for analysing structural differences in plant communities are applicable to detecting habitat differences, if not for quantitatively modelling habitat, and are also adopted here.

Description of the habitat of Pityrodia was undertaken by analysis of a series of three datasets, each describing a narrower range of habitat, as defined either by descriptive variables or by the geographic area of study sites. The first dataset was extracted from the Nitmiluk Vegetation Survey (NVS: Michell et al. 2004) database of 1528 sites

57 3: Habitat collected in or near sandstone landforms within Nitmiluk National Park. It contains data from 40 sites at which any of the Pityrodia species known to be eaten by Petasida was present, and 40 sites randomly selected from the entire NVS database. The aim was to identify the broad environmental and floristic correlates of that group of Pityrodia species.

The second dataset compared the same 40 Pityrodia sites with 40 sites with geomorphological and edaphic characteristics identified as common to all the Pityrodia sites. Pityrodia is patchily distributed within the sandstone, possibly due to very narrow or specialised habitat requirements, and the majority of the sandstone country does not hold Pityrodia populations. The aim of the analysis of the second dataset was to refine further the habitat description for Pityrodia, particularly in order to understand why it occurs only on some parts of the sandstone.

Pityrodia distribution at the fine scale remains patchy. The aim of analysis of the third dataset, collected from 12 sites in Kakadu National Park and one site in Nitmiluk National Park, was to elucidate environmental and floristic preferences of Pityrodia within habitat already supporting populations. That is, to determine, if detectable, the very specific habitat requirements and associations of this group of species.

3.3 Methods

3.3.1 Study sites

Data for the Nitmiluk Vegetation Survey (Michell et al. 2004) were collected between 2000 and 2002 by staff of the Northern Territory Department of Natural Resources, the Environment and the Arts (NRETA) from 1528 quadrats spread throughout Nitmiluk National Park (Fig. 3.4). The landforms of the park consist almost entirely of sandstone escarpments and plateaux, but the vegetation of the flat Marrawal Plateau, the dominant feature of the northern half of the park, is more typical of lowland eucalypt savanna than sandstone heath. A full description and analysis of the vegetation and physical environment is given in Michell et al. (2004)

58 3: Habitat

Two preliminary surveys were conducted for the current study in 2001 at Nitmiluk National Park and at Gubara in Kakadu, during which Pityrodia and grasshopper counts were made in random start parallel transects of contiguous 5 x 5 m quadrats. Transects were spaced at 50m. No habitat data were collected during these surveys. Each site was approximately 300 x 800 m and the northwest corners were located at 12° 50.315' S, 132° 52.596' E (Gubara) and 14° 18.591' S, 132° 25.617' E (Nitmiluk).

Figure 3.2. Aerial view of the Gubara area in Kakadu National Park, with upper Gubara in the valley to the left, and the lower Gubara sites in the valley in the centre far distance.

59 3: Habitat

Figure 3.3. Location of study sites for Pityrodia and Petasida studies. Locations of quadrats for the Nitmiluk Vegetation Survey were all within Nitmiluk National Park and are shown in Fig. 3.4. Nine sites were in the Gubara area, and are shown in detail in Fig. 3.5.

60 3: Habitat

Figure 3.4. Location of quadrats for the Nitmiluk Vegetation Survey (Michell et al. 2004) showing the quadrats from which data were analysed in the current study. Some concentration of sites around Katherine Gorge in the south of the park is evident.

Field study sites for the current study, in which Pityrodia density, environmental and floristic data were collected, were located at the western perimeter of the Arnhem Land Plateau. This area is in the core of the range identified from herbarium records (Fig. 3.9) Eleven were in Kakadu and one was in Nitmiluk National Park. All were in sandstone heath vegetation communities in rocky sandstone country with skeletal, sandy soils. Nine sites were in the Gubara area of the Mt Brockman outlier in Kakadu. Of these, four are located in a valley running parallel to Boroalba creek at an elevation approximately half the total height of the outlier and five are located approximately 3 km west, further down the valley at the base of the escarpment (Table 3.1, Figs. 3.2, 3.3, 3.5).

61 3: Habitat

Figure 3.5. Location of study sites in the Nourlangie Rock (Nou), lower Gubara (GP1, GJ1, GPS and GJS) and Upper Gubara (GTG, GTB, GTE, GTU) areas of Kakadu National Park. Site GPE is located between GP1 and GJ1 but is not shown.

62 3: Habitat

Figure 3.6. Pityrodia jamesii (foreground) at site GTG in the lower Gubara area in Kakadu National Park.

63 3: Habitat

Figure 3.7. Site NOU in Kakadu National Park.

Figure 3.8. Pityrodia jamesii (foreground) at site NOU, with Nourlangie Rock in the background.

64 3: Habitat

Table 3.1. Locations and characteristics of study sites used for habitat and floristic studies for Pityrodia. Location coordinates were recorded at the north-eastern corner of each site.

3

2

4 ) species

2 1 Site name Location Date latitude longitude Area (m No. transects transect Length of (m) No. Quadrats Quadrat sampling scheme Petasida present/absent Pityrodia Burnt/unburnt

1 May–10 Lower Gubara 12° 50.187' 132° 51.701' 48600 5 126 35 Reg 20 m P J UB GJ1 Jun 2004

rand 1/60 Lower Gubara 12 July 2004 12° 50.286' 132° 52.154' 12600 3 175 10 A J B 2002 GJS m

26 May–5 Lower Gubara 12° 50.155' 132° 51.483' 11900 6 90 15 reg 20 m P P UB GP1 June 2004

22–24 June Lower Gubara 12° 50.283' 132° 51.546' 25550 3 333 13 reg 20 m P P UB GPS 2004

Upper 31 May–3 12° 49.157' 132° 53.097' 12075 4 125 23 reg 20 m P J B 2002 GTB Gubara June 2004

Upper 11 Sep 2004 12° 49.108' 132° 53.346' 8000 3 108 11 reg 30 m A J B 2002 GTE Gubara

Upper 30 May 2004 12° 49.183' 132° 52.956' 13650 5 105 26 reg 20 m P J B 2002 GTG Gubara

Upper 6 Sep 2003 12° 48.946' 132° 53.787' 11000 4 79 16 reg 20 m P J UB GTU Gubara

Nourlangie 20–21 June 12° 51.348' 132° 48.218' 16500 3 223 15 rand 2/100 P J B 2002 Nou Rock 2004

NaN East Alligator 8 Dec 2004 12° 27.044' 132° 55.737' 6875 5 56 13 reg 20 m A J UB

10–15 Aug Nitmiluk 14° 18.591' 132° 25.617' 94350 8 278 18 rand 1/60 P L B 2002 NIT 2004 1 reg 20m = sampled quadrats spaced regularly 20m apart. rand 2/100 = 2 randomly spaced quadrats sampled each 100m. 2 P = present during sampling. A = absent for three years up to and including sampling 3 J = P. jamesii, P = P. puberula, L = P. lanuginosa 4 B2002 = Site at least partially burnt in 2002, UB = Unburnt since at least 2000

65 3: Habitat

One site is located 10 km to the west of Gubara at Nourlangie Rock, part of the Mt Brockman Outlier (Figs. 3.3, 3.5, 3.7, 3.8). The remaining Kakadu site is located approximately 50 km to the north, on an outlier approximately 10km south east of Cahill's Crossing on the East Alligator River (Fig. 3.3). One additional site is located within Nitmiluk National Park, approximately 300km to the southwest of the Kakadu sites. This site is at the top of the south western edge of the escarpment at the mouth of Katherine Gorge (Fig. 3.3).

All except three of the sites listed in Table 3.1 were known to have had Petasida present at some time within the two years prior to the sampling date. Those that did not were: GTE at Gubara, GJS at Gubara and NaN at East Alligator. Each of these sites had been searched by at least two people over a period of at least 2 hours in each of the sampling year and the two preceding years. The size and shape of sites was arbitrarily chosen to include, if possible, at least one entire patch of Pityrodia. If that was not possible, transect length was long enough to traverse a patch. Most sites were rectangular but two (NaN and Nou) were irregular because of topographical constraints (sheer cliffs).

3.3.2 Distribution Patterns.

3.3.2.1 Regional distribution patterns

Pityrodia distribution at a regional (Territory-wide) scale was determined by querying the two major databases maintained by the Darwin Herbarium and reviewing relevant publications. The first of these holds data associated with all specimens in the collections. The second consists of floristic and environmental data collected from several thousand 20 x 20 m plots (henceforth referred to as quadrats) throughout the NT. Location data for all records held in the Collections and Plot databases of the seven species of Pityrodia known to be food plants for Petasida were plotted on a map of the Northern Territory (Fig. 3.9). This defines the broad geographical envelope for potential distribution of Petasida.

66 3: Habitat

3.3.2.2 Local distribution patterns

The main study site, chosen primarily because access was easy, was an area of approximately 2 km2 in the vicinity of Gubara in Kakadu. This area included all the lower Gubara sites (GP1, GPE, GPS, GJ1 and GJS; Fig. 3.5). In order to map presence or absence of Pityrodia, parallel east-west transects, 100m apart, were walked over a three month period between March and June 2004. The shape of the survey area and the length of the transects were determined by topography; transects ended when walking was obstructed by cliffs or chasms. A swath extending 5m either side of the walk line was surveyed, giving a total width of 10 m and within that swath, Pityrodia presence, species and GPS location were recorded for all plants that were >5 m from the last recorded plant. All GPS points were then plotted over a 1:50,000 topographic map sheet. The resulting map simply indicates presence or absence; density was not measured.

Three transects (numbered 4, 7 and 10 from the north) spaced at 300 m intervals were selected for further descriptive analysis. Selection was on the basis that they were long and uninterrupted by inaccessible topography and they passed through the core area of the habitat patch. After data collection, these 3 transects were divided into 10 x 10 m quadrats and Pityrodia presence or absence was recorded for each quadrat. Counts of contiguous occupied quadrats were recorded as patches and counts of empty quadrats as gaps. Mean patch span, mean gap span, % of the length occupied by patches or gaps and patches km-1 were calculated for the pooled data of the three transects.

Further numerical analysis of the longest transect (no.7, 270 m), as well as two transects from the 2001 surveys (site NIT in Nitmiluk and Upper Gubara in Kakadu), used the method of Two Term Local Quadrat Variance (TTLQV; Hill 1973), which is based on calculating the average of the squares of the differences between block totals of all possible adjacent pairs of blocks (of quadrats) for all possible block sizes. At lower Gubara the block totals consisted of the summed presence (1) and absence (0) values of the constituent quadrats. Pityrodia count data were used for the 2001 transect analyses. The rationale for transect choice was that longer transects allow examination of larger block sizes.

67 3: Habitat

In the TTLQV output a peaked plot of the variances against block size indicates a clumped distribution, with the peak at the size of the clumps. Dale (1999) suggests that either TTLQV or its variant 3 Term Local Quadrat Variance (3TLQV; Hill, 1973) are the recommended techniques for pattern analysis with contiguous quadrats. 3TLQV is less sensitive to trends in the data but larger block sizes can be examined with TTLQV (Dale 1999). Dale (1999) cites a recommendation by Ludwig and Reynolds (1988) that the largest block size to be examined should not exceed 10% of total transect length, but he suggests that this restriction requires further investigation.

At a very fine scale, Pityrodia distribution was recorded at 12 study sites in Kakadu and Nitmiluk National Parks. Density counts of individual Pityrodia plants were taken concurrently with the collection of environmental and floristic data. At each site, Pityrodia density was recorded in parallel transects consisting of contiguous 5 x 5m quadrats. Details of study sites, transects and methods are given in the following sections. Again, patch and gap sizes were calculated. Because most of these transects were relatively short (60–120m), incomplete patches and gaps at the ends of transects were omitted from the analysis.

During a study of grasshopper density in 2004, Pityrodia density was mapped in a grid of contiguous 5m x 5m quadrats in area 50m x 120m within the site GJ1. The Standardized Morisita index, which Krebs (1999) regards as the best index of dispersion, was calculated for Pityrodia density data.

3.3.3 Habitat data collection

3.3.3.1 Nitmiluk Vegetation survey

Data were collected by NRETA staff between 2000 and 2002 and are described in Michell et al. (2004). A total of 1528 sites were visited throughout Nitmiluk National Park (Fig. 3.4), with efforts being made to represent all major vegetation types in the survey. At each site a 20 x 20 m quadrat was established within which all plant species present were recorded, together with a range of environmental variables. The variables used in the current study were: soil type (10 categories), slope (degrees), estimated

68 3: Habitat soildepth, time since fire (5 categories) (fire), litter cover (%) (litter), bare ground cover (%) (bare), total rock cover (%) (rocks), total vegetation cover (%) (veg), cover by seven separate size classes of rock and cover by six separate height classes of vegetation.

The method of recording in the field ‘time since fire’ had been somewhat ad hoc, so categories were retrospectively assigned as follows: 1 = <1 yr, 2 = 1 yr or <2 yr, 3 = 2 yr, 4 = 3+ yr or 4 yr and 5 = long unburnt. Rock size classes labelled rockcov1–rockcov6 were: pebbles <0.6 cm, gravel 0.6–2 cm, stones 2–6 cm, small rocks 6–20 cm, rocks 20– 60 cm, large rocks 60cm–2 m, boulders >2 m respectively. They were recorded in six cover classes: 0%, <5%, 5–10%, 10–25%, 25–50%, 50–75% and >75%. Vegetation height classes labelled vegcov1–vegcov6 were: <0.5m, 0.5–1m, 1–3m, 3–5m, 5–10m and >10m respectively. They were recorded in seven cover classes: 0%, <2%, 2–10%, 10– 20%, 20–50%, 50–90% and >90% respectively. Soildepth was recorded in five depth classes: 0 m, 0-0.1 m, 0.1 - 0.4 m, 0.4 - 1 and >1 m.

In addition, each site was assigned a value for the descriptive fields: landform pattern, landform element and site description.

3.3.3.2 The current study

Environmental, floristic and Pityrodia density data for the current study were collected at 12 sites supporting populations of Pityrodia (Table 3.1, Figs. 3.3, 3.5). At each site, a series of parallel transects was laid out using a 100m tape measure and following a compass bearing (using a handheld GPS unit). Transects were regularly spaced at 25m or, at large sites, 50m apart. The start point for the first transect was randomly placed within either a 25m (for transects spaced at 25m) or 50m (for those spaced at 50m) section of the one side of the study site. Each transect consisted of contiguous 5 x 5 m quadrats, marked out by 2.5 m lengths of PVC pipe placed on either side of the centreline.

The number and species of Pityrodia plants were recorded in each quadrat. An individual was arbitrarily defined as any plant with a stem that emerged from the ground >10cm from the nearest other Pityrodia stem. All plants were counted, regardless of

69 3: Habitat size. P. jamesii was scored in six height classes: 0–10cm, 10–20cm, 20–50cm, 50cm– 1m, 1–2m and >2m. The other Pityrodia species were simply counted, because they tend to show a prostrate habit and therefore height is a poor indication of size.

Environmental variables were measured in selected quadrats within each transect. When time allowed, these quadrats were spaced regularly along the transect at intervals of 20 or 30m. Otherwise, transects were subdivided and quadrats were randomly selected from each segment (Table 3.1).

The environmental variables measured were as follows:

Rock cover = % cover of rock

Burnt = % of vegetation burnt in the last year

Topo = Topography: the difference in altitude between highest and lowest point within the quadrat

Aspect = aspect measured by compass to the nearest 45° ( i.e. N, NE, E etc)

Firescar grade: 1 = charcoal or soot present, 2 = fire damage to bark, 3 = structural damage to the core wood.

Fire scar height = highest point of black discoloration on plants

Slope = difference between highest and lowest points of the soil (disregarding rocks and boulders)

Annuals = % cover of annual grasses

Perennials = % cover of perennial grasses and sedges (excluding 'spinifex')

Spinifex = % cover of 'spinifex'

70 3: Habitat

Shrubs1, Shrubs2, Shrubs3 and shrub 4 = % cover of shrubs in 4 size classes: <0.5m, 0.5–1m, 1–2m, >2m.

Stems = Number of stems of any species >5cm DBH.

Fuel = A relative estimate of fuel load (grass and litter) on an ordinal scale of 1–4

Litter = % litter cover

Tree canopy = % canopy cover

Soil = soil depth, estimated by driving a 35cm steel peg as far as possible, once in each side of the quadrat, and then averaging.

Floristic data were collected at each quadrat designated for collection of environmental data. In each case, the quadrat was expanded to 10m x 10m by randomly selecting a direction and adding 3 adjacent 5 x 5 quadrats. In each of the 4 quadrats thus delineated presence of all woody species over 50 cm tall was recorded. Specimens of unknown species were collected for identification at the Darwin Herbarium.

3.3.4 Habitat data analysis

3.3.4.1 Environmental correlates

Description of the habitat of Pityrodia was undertaken by analysis of a series of three datasets, each describing a narrower range of habitat. The first two datasets (Dataset 1 and Dataset 2) were extracted from the existing Nitmiluk Vegetation Survey (NVS: Michell et al. 2004; 3.3.3.1 above) in the manner described below. The third dataset (Dataset 3) was collected for the current study from 12 sites in Kakadu and Nitmiluk National Parks, all of which had Pityrodia present (198 quadrats), and is described in detail above (3.3.3.2, Table 3.1).

Dataset 1 contains data from 80 quadrats from a relatively broad range of environmental conditions within or very close to habitat loosely defined as ‘sandstone country’ in Nitmiluk National Park. Five species of Pityrodia known to be host plants for Petasida

71 3: Habitat occurred in the NVS, but in only 40 of the 1528 quadrats. These quadrats were included in Dataset 1 together with 40 additional quadrats selected in the following manner. Forty five quadrats were selected at random from the remaining 1488, and the five with the most missing values for environmental variables were discarded. The remaining missing values were assigned the mean value for the field, and occurred in the following fields (with number of occurrences): soildepth (2), fire (1), rockcov1–rockcov7 (1 each) and vegcov1–vegcov6 (3 each). This treatment of missing values was used because each individual missing value affected relatively few samples, but removal of either samples or fields because of missing values would have affected many samples or many fields.

In order to investigate finer or more subtle differences between habitat supporting Pityrodia and that which did not, the selection criteria for non-Pityrodia quadrats in Dataset 2 were narrowed so that they more closely approached that of potential Pityrodia habitat in the following way. The Pityrodia quadrats were examined for common values for environmental variables. All Pityrodia quadrats contained the word ‘sandstone’ in one or more of three fields in the data set: landform pattern, landform element or description. All except one Pityrodia quadrat had a value of 9 (sand) or 10 (rock) in the soil texture field (the exceptional value was 8, sandy loam). Therefore, for the second analysis, all quadrats without sandstone in one of the relevant fields or a soil texture value of 9 or 10 were eliminated, together with the 40 non-Pityrodia quadrats used in dataset 1. This left a pool of 302 quadrats from which 45 were randomly selected, of which the five with the most missing values were discarded. Dataset 2 consisted of these quadrats together with the same 40 Pityrodia quadrats as in Dataset 1. Again, the remaining missing values were assigned the mean value for the field, and occurred in the following fields (with number of occurrences): soil depth (1), fire (5), rockcov1–rockcov7 (1 each) and vegcov1–vegcov6 (4 each).

Dataset 3 is described in detail above (3.3.3.2, Table 3.1).

All three datasets were analysed using Generalized Linear Modelling (GLM). Generalized Linear Mixed Modelling (GLMM) was used for an additional preliminary analysis of Dataset 3. Analysis followed the information-theoretic approach described by

72 3: Habitat

Burnham and Anderson (2002), using Akaike's Information Criterion corrected for small sample size (AICc) to fit the best and most parsimonious model to the data. In each case the difference between the residual and null deviance for the 'global model', which includes all variables, was used to indicate the maximum amount of deviance in the data accounted for by all predictor variables. In order to gain some further indication of the relative importance of the predictor variables, GLM was followed by an all-subsets analysis, in which AICc weights are determined for all possible models, and then summed in turn across all models containing each of the variables (Burnham and Anderson 2002: p. 225). Analyses were carried out using the software package R (version 2.0.1 for GLMM and version 2.3.1 for GLM, R Development Core Team).

For Dataset 1 and Dataset 2 generalized linear modelling (GLM) was done, assuming a binomial error structure with a logit link function (logistic regression) and Pityrodia presence/absence as the binary response variable. The a priori candidate model set consisted of Pityrodia presence modelled as a function of four single variables: total rock cover, time since fire, total vegetation cover and slope. The global model was also included as a means of estimating goodness of fit, rather than as an a priori proposed potential explanatory model. The model set was kept small in order to keep the ratio of samples (n) to variables (R) low. Where n/R <9 there is a high probability of over-fitting models included in the all-subsets analysis (in a binomial GLM n = no. samples in the smaller of the two categories, 40 in this case) (Burnham and Anderson 2002). Model selection and all-subsets analysis followed the GLM.

In order to further understand the results when the number of predictors was fewer, and to explore the possibility of inappropriate candidate model selection, the analysis was repeated with two further, post hoc, model sets. The first consisted of the following variables:

Boulder = rockcov6 + rockcov7

Stones = rockcov3 + rockcov4 + rockcov5

Gravel = rockcov1 + rockcov2

73 3: Habitat

Litter = % cover leaf litter

The second consisted of the following variables:

Tallveg = vegcov5 + vegcov6

Medveg = vegcov3 + vegcov4

Lowveg = vegcov1 + vegcov2

Soil depth = estimated soil depth

For Dataset 3, collected during the current study from sites with Pityrodia present, the response variable was Pityrodia presence/absence using pooled data for the three Pityrodia species (P. jamesii, P. puberula and P. lanuginosa). In order to reduce the candidate model set to a realistic and acceptable number the measured variables were selected, eliminated or amalgamated as follows:

Burnt was eliminated because no sites had been burnt in the year prior to sampling.

Aspect was eliminated as trivial. Field observation during the study showed it to be irrelevant.

Fire scar grade and height were eliminated as unreliable, given the dearth of trees and the large proportion of scar bearing shrubs which were clearly shortened in height by the most recent fire.

Slope was eliminated as correlated with but biologically less important than Topo.

Annuals was eliminated as it was expected to correlate with the variable fuel.

The three smallest size classes of shrub cover were amalgamated (by addition) to give a single variable for shrub cover.

74 3: Habitat

The largest size class for shrub cover (>2m) was amalgamated by addition with tree canopy cover to derive a single variable Canopy. The rationale was that, biologically, they are equivalent in their impact on Pityrodia species or grasshoppers.

Litter was eliminated as being correlated with but less useful than fuel.

Stems was eliminated as correlated with the amalgamated Canopy variable.

Soil was eliminated because of several missing values at one site, due to equipment failure.

This left four directly measured variables (Rock cover, Topo, Fuel and Triodia) and two amalgamated variables (Shrub cover and Canopy) to be used in the analysis. The full dataset consisted of data from 198 quadrats at 12 sites. Three species of Pityrodia occurred but no quadrat contained more than one of those species. Pityrodia presence/absence was modelled as a function of each of the six variables, and additionally, as a function of the global model in order to estimate goodness of fit.

In order to examine the data for variation due to site, the initial analysis of the full data set (12 sites) used generalized linear mixed modelling (GLMM) with Site as the random variable and each of the 6 selected variables as fixed effects. The modelling assumed a binomial error structure with a logit link function. Because site was shown to account for a very small proportion of the variation in the data (explained variance <0.08, SD = 0.28), GLMM was discontinued.

The remainder of the analysis for Dataset 3 was conducted using generalized linear modelling (GLM). Using the full data set (12 sites, 198 quadrats, species of Pityrodia pooled), Pityrodia presence/absence was modelled as a function of each of the six variables and of the global model. The model again assumed a binomial error structure with a logit link function. The modelling was also carried out using identity and square root link functions with no increase in the deviance explained. GLM was followed by all-subsets analysis.

75 3: Habitat

3.3.4.2 Floristic correlates

Dataset 1 (Nitmiluk Veg. Survey, broad habitat range) and Dataset 2 (Nitmiluk Veg. Survey, narrow habitat range) were the same as those used for the environmental correlates data analysis described in the previous section. Data for each quadrat consisted of all plant species present at more than three of the sites included in this analysis, together with the 20 variables described (soil texture and the three descriptive variables were not included).

Dataset 3 (Pityrodia sites only, very narrow habitat range), described above, was treated in the following way. In order to remove any domination of pattern results by geographic factors, the analysis was restricted to the sites in the Gubara area of Kakadu. All quadrats geographically distant from Gubara were removed (Nitmiluk, East Alligator and Nourlangie), as was the smallest Gubara site (GPE; 5 quadrats), leaving only quadrats in eight sites occurring within a 4 km radius at Gubara. This group was further divided into two groups of four sites: upper and lower Gubara, in order to reduce the number of samples in each analysis to below 100. The ordination procedure (NMDS, see below) on much more than 100 samples is unlikely to reveal a clear and reliable pattern (Clarke 1993).

All analyses were carried out using the software package PC-ORD (Version 4.36, McCune and Mefford 1999). Non-metric Multidimensional Scaling (NMDS) ordination (described by Clarke, 1993) was carried out on each dataset. Bray-Curtis (Sorensen) distance measures were used with all other settings at default. Species occurring in fewer than four quadrats were omitted. NMDS was followed by a Multi-Response Permutation Procedure (MRPP) analysis with Pityrodia presence as the grouping variable, again using a Sorensen distance measure with other settings at default. MRPP is a non- parametric method for testing the hypothesis of no difference between two or more a priori groups (McCune and Mefford 1999). MRPP with Pityrodia species as the grouping variable was also carried out on the Lower Gubara dataset. The variable Pityrodia species took three values: P. jamesii, P. puberula and Pityrodia absent. In addition, Indicator Species Analysis (Dufrene and Legendre 1997) was carried out on

76 3: Habitat each of datasets 1 and 2 and the combined Upper and Lower Gubara datasets, with all settings except the number of Monte Carlo runs (10,000) at default, and Pityrodia presence as the defining variable for groups. Indicator species results were examined with reference to the NT Ecological Attributes Database (NRETA 2006) and Russell- Smith et al. (2002; 1998) in order to describe the mode of reproduction and life form of the indicator species.

NMDS output includes the coefficient of determination for the correlations between ordination distances and distances in the original 3-dimensional space (r2), a measure of the proportion of variance explained by the ordination. MRPP output includes a chance- corrected within-group agreement statistic (A) which indicates the size of the effect under examination. A = 0 if heterogeneity within groups equals expectation by chance, A <0 if within group heterogeneity is greater than expectation by chance and A = 1 if all items are identical within groups. A value of 0.3 for A is considered high (McCune and Mefford 1999).

3.4 Results

3.4.1 Distribution

3.4.1.1 Regional distribution

The Collections database of the Darwin Herbarium holds 157 records of the seven Pityrodia species that are known to be food plants of Petasida, and the Plot database holds 61. The locations for all records are shown in Fig. 3.9.

3.4.1.2 Local distribution

The results of the distribution of two species of Pityrodia along the survey transects at lower Gubara is shown in Fig. 3.10. For the pooled data from transects 4, 7 and 10 (numbered from the north) the mean patch span (all 3 transects) was 2.8 quadrats (28m) and the mean length of gaps was 9.9 quadrats (99m). 20.5% of the total transect length was occupied by quadrats containing at least one Pityrodia plant and there were 7.4 patches km-1. Quadrat (5 x 5 m) presence/absence data from within the individual 2003–

77 3: Habitat

4 study sites (Table 3.2, Fig. 3.11) give an indication of Pityrodia patchiness at a very fine scale, and mean patch spans range from 1.6 quadrats (8 m) at site NaN (East Alligator) to 5.8 quadrats (29.0 m) at site GTE (Upper Gubara). The 2001 data (Table 3.2, Fig. 3.11) give a less biased picture of gap characteristics because transects are longer (~300 m cf. ~100 m).

Figure 3.9. Map of all records of the Pityrodia species known to be food plants for Petasida in the 'Top End' of the Northern Territory, from the databases of the Darwin Herbarium.

TTQLV analysis of the lower Gubara data showed a clear peak, indicating a clumped distribution, and gave a larger patch span estimate than the simpler approach: 100–120 m (Fig. 3.12). TTLQV results for upper Gubara and Nitmiluk also showed clear peaks, but with smaller patch sizes: 20–25 m at Upper Gubara and 35–40 m at Nitmiluk (Fig. 3.12). Fine scale mapping of Pityrodia density at site GJ1 (Fig. 3.13) indicates a clumped distribution for P. jamesii at that site. Calculation of the Standardized Morisita

Index (Ip = 0.505, p <0.05) confirms the visual impression.

78 3: Habitat

Figure 3.10. The distribution of Pityrodia in the lower Gubara area of Kakadu National Park. Red squares = P. jamesii; black squares = P. puberula. Grid lines are 1km apart and transects (blue lines) are spaced at 100m.

79 3: Habitat

Figure 3.11. Frequency distributions of Pityrodia patch spans (= occupied quadrats) and gap lengths (= unoccupied quadrats) for all sites in 2003/4 and for Nitmiluk and Upper Gubara in 2001. Quadrat length is 5 m. Mean (±SE) Pityrodia count per occupied quadrat for All sites 2003/4 = 8.49 (±0.70), upper Gubara 2001 = 3.91 (±0.36), Nitmiluk 2001 = 2.28 (±0.19)

80 3: Habitat

Figure 3.12. TTLQV results for transects through three of the study areas. A peaked plot indicates a clumped distribution of Pityrodia, with the peak at the size of the clumps.

81 3: Habitat

0 0 0 4 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 2 4 3 0 0 1 0 1-5 stems 0 0 0 6 26 6 20 1 0 6-20 stems 0 0 3 3 2 0 3 101 >20 stems 0 0 2 1 2 9 6 1 4 9 0 0 1 8 5 16 8 1 12 5 0 1 13 20 28 19 22 8 4 5 0 7 5 6 13 9 20 1 4 0 19 16 11 17 17 19 0 0 0 0 20 44 35 36 8 70 0 0 0 10 19 15 38 3 10 0 3 0 8 7 23 26 10 33 2 3 17 1 6 4 22 15 32 12 0 6 12 8 12 10 19 16 13 0 4 13 17 2 5 18 32 1 0 0 4 17 3 0 0 7 12 4 0 0 1 30 0 0 0 1 3 0 0 0 0 0 0 0 11 16 4 0 0 0 0 0 0 4 10 10 0 0 0 0 0 0 0 0 0 2 0 0 0 0 0 0 Figure 3.13. Fine scale density map of Pityrodia jamesii at Site GJ1 in the lower Gubara area. Each square represents 5m x 5m. The obvious visible clumping pattern is confirmed by the standardised Morisita Index (see text). In Chapter 4 grasshopper densities in sequence throughout one dry season are superimposed on this figure.

82 3: Habitat

Table 3.2. Mean patch spans, gap lengths and Pityrodia density within patches. Patches and gaps are defined simply by the presence or absence of any Pityrodia species within contiguous quadrats. Patch span Gap length Pityrodia density (plants (m) (m) per occupied quadrat) Location date mean se mean se mean se n

All sites 2003–4 17.4 1.24 15.8 1.52 8.5 0.7 391

Nitmiluk 2001 9.7 1.12 38.9 7.09 2.3 0.19 75

Upper 2001 7.8 0.84 35.1 6.03 3.9 0.36 109 Gubara

3.4.2 Environmental correlates

3.4.2.1 Dataset 1 (Nitmiluk Veg. Survey, broad habitat range)

Five of the seven Pityrodia species known to host Petasida were recorded in the NVS, occurring in 40 out of a total of 1528 quadrats. The species were (with no. of quadrats): P. lanuginosa (26), P. pungens (7), P. ternifolia (3), P. spenceri (2) and P. puberula (2). No quadrat had more than one of these species present, but five of them also contained P. quadrangulata, which has not been recorded as a host for Petasida.

In modelling the a priori selected candidate model set for Dataset 1 (Table 2.3) the global model explained 33.4% of the deviance in the data. Although not put up as a candidate model, the global model was selected as the best model with overwhelming support. There is considerably less support for total rock cover (rocks) and essentially no support for any of the other single-variable models. The very high Akaike weight (i) of 0.97 indicates high degree of model selection certainty for the global model. Time since fire (fire) had less support than the null model. The all-subsets analysis showed both rocks and slope to be very important variables with i values of 1.0 and 0.98 respectively. These two variables were correlated (r = 0.33) and rocks and total vegetation cover (veg) were negatively correlated (r = 0.60). Veg was a reasonably important variable while fire was trivial.

83 3: Habitat

Table 3.3. Results of AICc -based model selection and all-subsets analysis for Pityrodia presence/absence for dataset 1 at Nitmiluk National Park, modelled using binomial GLM with the a priori selected variables. The table also shows the maximised log- likelihood function (log(l)), number of predictor variables (K), the AICc differences

(.AICc) and the Akaike weights (i). The i shown are the Akaike weights summed across all models containing each of the variables. n = 80; Deviance explained by the global model = 33.40%.

Model selection All-subsets

Model log(l) K AICc AICc i Variable i

slope+rocks+fire+veg -36.93 6 87.01 0 0.97 rocks 1

rocks -43.69 3 93.69 6.68 0.03 slope 0.98

slope -48.29 3 102.9 15.89 0 veg 0.63

veg -53.77 3 113.86 26.85 0 fire 0.32

null -55.45 2 115.06 28.05 0

fire -55.45 3 117.22 30.21 0

The post hoc results (Table 3.4) are 'data dredging' but nevertheless shed interesting light on the above results, especially considering the high percentages of deviance explained (73% and 42%), and provide the basis for a new model set to test on a future dataset. While the global model in variable set 1 once again had the highest support, boulder cover (boulders) had essentially the same amount. Boulders was also shown to be highly important in the all-subsets analysis, with reasonably strong support for gravel. However, the correlation of gravel with Pityrodia presence is a negative one. Clearly, total rock cover was not the most important variable. Rather, Pityrodia presence was associated with the cover of very large rocks. Similarly, the effect of tallveg, which is negatively correlated with Pityrodia presence, and medveg, which is positively correlated, act against each other and are masked in the single variable veg. Tallveg and the most important variable, soil depth (soildepth), are correlated (r = 0.56).

84 3: Habitat

Table 3.4. Post hoc results of AICc -based model selection and all-subsets analysis for Pityrodia presence/absence for dataset 1 at Nitmiluk National Park, modelled using binomial GLM with two sets of measured and derived variables. n = 80; Deviance explained by the global model was 73.1 % (variable set 1) and 42.0% (variable set 2).

Model selection All-subsets model log(l) K AICc AICc i Variable i

Variable set 1 gravel+stones+boulder+litter -14.94 6 43.03 0 0.52 boulder 1 boulder -18.46 3 43.23 0.2 0.48 gravel 0.63 litter -48.34 3 102.99 59.96 0 litter 0.47 gravel -48.36 3 103.03 59.99 0 stones 0.42 stones -51.67 3 109.66 66.63 0 null -55.45 2 115.06 72.03 0

Variable set 2 lowveg+medveg+tallveg+soildepth -32.15 6 77.45 0 0.97 soildepth 1 soildepth -38.92 3 84.15 6.7 0.03 tallveg 0.97 tallveg -44.75 3 95.81 18.36 0 medveg 0.94 null -55.45 2 115.06 37.61 0 lowveg 0.25 medveg -54.84 3 115.99 38.54 0 lowveg -55.45 3 117.21 39.75 0

3.3.2.2 Dataset 2 (Nitmiluk Veg. Survey, narrow habitat range)

For the a priori candidate model set derived from Dataset 2 the global model explained 17% of the deviance in the data (Table 3.5). Slope was selected as the best model, with strong support also for the global model. All-subsets analysis revealed slope to be the most important variable while rocks was also important (both variables positively associated with Pityrodia presence). The post hoc analysis of variable set 1 (Table 3.6) selected boulder as the best model and the most important variable, with some importance assigned to the variable gravel. The post hoc global model explained 16.6% of the deviance. The global model for variable set 2 (Table 3.6) explained a higher proportion of the deviance (25%), with the global model selected as the best model and

85 3: Habitat strong support for soildepth. Soildepth and tallveg were the most important variables over all subsets, but the correlation between them in this dataset is poor (r = 0.09). Tallveg, medveg and soildepth are all negatively associated with Pityrodia presence.

Table 3.5. Results of AICc -based model selection and all-subsets analysis for Pityrodia presence/absence in dataset 2 at Nitmiluk National Park, modelled using binomial GLM with the a priori selected variables. n = 80; Deviance explained by the global model = 16.58%.

Model selection All-subsets

Model log(l) K AICc AICc i Variable i slope -49.40 3 105.11 0 0.55 slope 0.99 slope+rocks+fire+veg -46.26 6 105.67 0.56 0.41 rocks 0.74 rocks -52.43 3 111.17 6.06 0.03 fire 0.39 null -55.45 2 115.06 9.95 0 veg 0.30 veg -54.66 3 115.63 10.52 0 fire -54.83 3 115.98 10.87 0

86 3: Habitat

Table 3.6. Post hoc results of AICc -based model selection and all-subsets analysis for Pityrodia presence/absence in dataset 2 at Nitmiluk National Park, modelled using binomial GLM with two sets of measured and derived variables. n = 80; Deviance explained by the global model = 14.56 % (variable set 1) and 24.79% (variable set 2).

Model selection All subsets

model log(l) K AICc AICc i Variable i

Variable set 1

gravel+stones+boulder+litter -14.94 6 43.03 0 0.52 boulder 1

boulder -18.46 3 43.23 0.2 0.48 gravel 0.63

litter -48.34 3 102.99 59.96 0 litter 0.47

gravel -48.36 3 103.03 59.99 0 stones 0.42

stones -51.67 3 109.66 66.63 0

null -55.45 2 115.06 72.03 0

Variable set 2

lowveg+medveg+tallveg+soildepth -41.70 6 96.56 0 0.68 soildepth 1

soildepth -45.89 3 98.10 1.54 0.32 tallveg 0.92

tallveg -53.39 3 113.10 16.54 0 medveg 0.33

null -55.45 2 115.06 18.50 0 lowveg 0.24

medveg -54.46 3 115.23 18.67 0

lowveg -55.40 3 117.12 20.56 0

87 3: Habitat

3.3.2.3 Dataset 3 (Pityrodia sites only, very narrow habitat range)

Dataset 3 was collected during the current project at sites with Pityrodia present (12 sites, n = 198,). Binomial Generalised Linear Mixed Effects Modelling (GLMM) of Pityrodia presence/absence (Pityrodia species pooled) using the a priori candidate model set showed a negligible contribution by 'site' to any variation in the data (explained variance of intercept for the random variable (site) in the global model containing all variables = 0.077, SD = 0.277). The difference between the deviance explained by the global model under GLMM and that under GLM was only 0.12 (2.6%)

Table 3.7. Results of AICc -based model selection and all-subsets analysis for Pityrodia presence/absence at 12 sites modelled using binomial GLM with the measured and derived, a priori selected variables. n = 198; Deviance explained by the global model = 4.69%. Model selection All-subsets

Model log(l) K AICc i Variable i AICc rocks -130.21 3 266.53 0.00 0.84 rocks 0.96

perennials -133.25 3 272.62 6.08 0.04 topo 0.49

null -134.65 2 273.35 6.82 0.03 triodia 0.42

triodia -133.63 3 273.39 6.85 0.03 shrubs 0.28

rocks+topo+perennials+ -128.32 8 273.41 6.88 0.03 perennials 0.28 triodia+shrubs+canopy shrubs -134.42 3 274.97 8.44 0.01 canopy 0.26

canopy -134.61 3 275.34 8.81 0.01

topo -134.64 3 275.40 8.87 0.01

In the binomial GLM of Pityrodia presence/absence with the same candidate model set, the global model of all variables in the a priori candidate set explained only a small proportion of the deviance in the data (4.7%). Of that proportion, total rock cover (rocks) clearly accounted for most of it (Table 3.7). Rocks was the best selected model, with a fair degree of certainty, and the most important variable across all subsets. Topography (topo) and triodia were both approximately half as important as rocks over all subsets.

88 3: Habitat

Rocks and Triodia were positively associated with Pityrodia presence and topo negatively. In order to explore the possibility of inappropriate model selection and to uncover any other relationships, the GLM was repeated with a selection of the originally discarded variables. However, the global model for this post hoc candidate model set explained even less of the deviance (1.9%).

3.3.3 Floristic correlates

3.3.3.1 Ordination

The ordination of dataset 1 (Fig. 3.14) clearly separates the tightly clumped Pityrodia quadrats from non-Pityrodia quadrats, with minimal overlap, on two of the three axis combinations. MRPP results confirm the visual result: A = 0.104, p <0.00001. There is a strong correlation of rockiness variables with the axes. Pityrodia presence clearly corresponds with a gradient of increasing cover of large rocks and to a lesser degree, slope, and decreasing soil depth and cover of small rocks. The proportion of variance explained by the ordination (cumulative r2) = 0.785. Final stress = 18.4.

In the NMDS ordination of Dataset 2 (Fig. 3.15) the Pityrodia sites are less clearly separated, although there is still a clearly discernable pattern. MRPP results confirm that although the difference between the two groups is small enough to give cause for thought regarding the ecological significance (A = 0.016), it is highly statistically significant (p <0.00001) even though the sample size is not huge (n = 80), suggesting a low variability in response. Pityrodia presence corresponds approximately with a gradient of increasing rock cover, particularly that of boulders >2 m, and decreasing litter cover and cover of trees 5–10 m high. Cumulative r2 = 0.720, final stress = 20.99.

89 3: Habitat

Figure 3.14. Joint plot showing the first two dimensions of the NMDS ordination results for Dataset 1 from the Nitmiluk Vegetation Survey data, together with vectors showing the relationship of environmental variables with the axes. The cutoff r2 value for the vectors is 0.2.

90 3: Habitat

Figure 3.15. Ordination Joint plot showing the first two dimensions of the NMDS ordination results for Dataset 2 from the Nitmiluk Vegetation Survey data, together with vectors showing the relationship of environmental variables with the axes. The cutoff r2 value for the vectors is 0.2..

The output to axes 1 and 2 of the 3-dimensional NMDS ordinations of the Gubara floristic data are shown in Fig. 3.16 (upper Gubara: cumulative r2 = 0.797, final stress = 17.32) and Fig. 3.17 (lower Gubara: cumulative r2 = 0.718, final stress = 18.58). There is no apparent pattern separating those quadrats with and without Pityrodia present, at either the upper or lower Gubara areas, This is confirmed by the MRPP results: lower Gubara: A = 0.001, p = 0.336; upper Gubara: A = -0.003, p = 0.671. There is, however, a discernable pattern of separation between P. jamesii and the more tightly clumped P. puberula quadrats. MRPP with Pityrodia species as the grouping variable and three

91 3: Habitat groups (including Pityrodia absent) confirmed the observation: A= 0.029, p = 0.0002. The cutoff point for vectors in Fig. 3.16 is set much lower than in previous examples in order to explore any habitat differences between the two Pityrodia species. Of the relatively weak environmental gradients thus uncovered, only perennial grass and sedge cover corresponds with presence of P. jamesii. The others appear to be roughly perpendicular to the gradient separating the two Pityrodia species.

Figure 3.16. The first two dimensions of the NMDS ordination of 58 quadrats and 32 species from the upper Gubara area in Kakadu National Park. None of the environmental variables correlated with the ordination axes with a r2 value greater than 0.2.

92 3: Habitat

Figure 3.17. Ordination Joint plot showing the first two axes of the NMDS ordination of 86 quadrats with 29 species in the lower Gubara area in Kakadu National Park, together with vectors showing the relationship of environmental variables with the axes. The cutoff r2 value for the vectors is 0.09. The only variable with r2> 0.2 is litter.

3.3.3.2 Indicator Species Analysis

For the first Nitmiluk dataset there were 52 positive and 21 negative indicator species for Pityrodia presence with a Monte Carlo derived probability for the maximum indicator value of <0.01 (Table 3.8). The positive indicators are characteristic sandstone heath plants and the negative ones are more typical of lowland savannas. There is a predominance of obligate seeders amongst the positive indicators, but with a notable absence of long-lived species such as Callitris intratropica and the obligate seeder such as A. plectocarpa. All the long-lived species in this group are resprouters. For the second Nitmiluk dataset the numbers had declined to five positive and three

93 3: Habitat negative indicators at p <0.01 (Table 3.9). However, the previous associative pattern is repeated, albeit weakly, if the indicators at p <0.05 (Table 3.9) are taken into account. For the Gubara data, consisting only of trees and woody shrubs >50 cm high, there were only two negative indicators at p <0.01, but the number did not increase even at p <0.05. The species were a resprouting tree, Banksia dentata, and a resprouting shrub, Melaleuca cornucopiae, both of which are characteristic of damp soil conditions. Comparisons of the Gubara results with the Nitmiluk results should be made only with caution, given the differences in location, quadrat size and data variables.

3.4 Discussion

The distribution of the Pityrodia species upon which Petasida depends for food is clearly patchy at a range of scales. This is to be expected, especially at a regional scale, given the restriction of all species to sandstone habitats. The sandstone landforms, although extensive in the Top End, have a disjunct distribution for which appropriate maps covering the whole area do not currently exist. Russell-Smith et al. (2002) have produced such a map for the Arnhem Land Plateau area only. However, none of the Pityrodia records are from locations known to be away from sandstone. The Herbarium records represent only an approximation of the distribution because they are, of course, dependent on collecting locations. While the collections are extensive, they are necessarily biased and some areas have seen little collecting activity. The potential gaps are demonstrated by the single point in Fig. 3.9 near Ngukurr in south east Arnhem Land. This point marks populations of both Pityrodia ternifolia and Petasida, and was discovered by Herbarium staff during the time period of this study. It is more than 100 km from the nearest known Pityrodia populations and approximately 300 km from the nearest known Petasida populations. The distribution of Petasida shown in Fig. 1.1 of Chapter 1 correlates closely with that shown here for Pityrodia. Again, however, the recent discovery of the Ngukurr population suggests that the map is incomplete.

94 3: Habitat

Table 3.8. Indicator species (P <0.01) for dataset 1 of the Nitmiluk Vegetation Survey.

Positive for Pityrodia presence Negative for Pityrodia presence life life Species Strategy form Species Strategy form cataractae OS sh Alloteropsis semialata R p, g Acacia gonocarpa OS sh Aristida holathera R p, g Acacia lycopodifolia OS sh Chrysopogon fallax R p, g Boronia grandisepala OS sh Corymbia dichromophloia R t Boronia lanceolata OS sh Crotalaria medicaginea OS a, h Boronia lanuginosa OS sh Eriachne obtusa R p, g Bulbostylis barbata OS a, s Erythrophleum chlorostachys R t Cajanus acutifolius OS sh Eucalyptus miniata R t Calycopeplus collinus OS sh Eucalyptus tetrodonta R t Calytrix verticillata R sh Eucalyptus tintinnans R t Cochlospermum fraseri R sh Goodenia holtzeana ? h Comesperma aphyllum R sh Grevillea decurrens R t Corchorus sidoides OS sh Grevillea pteridifolia OS t Corymbia arnhemensis R t Petalostigma quadriloculare R sh Cyperus microcephalus R ps Sarga plumosum R p, g Dampiera conospermoides OS sh Sauropus glaucus R a, h Eriachne capillaris OS a, g Spermacoce leptoloba OS a, h Eriachne mucronata R p, g Terminalia ferdinandiana R t Fimbristylis compacta R p, s Terminalia pterocarya R t Fimbristylis trigastrocarya OS a, s Triodia bitextura R p, g Gardenia fucata R t Vigna lanceolata R p, h Grevillea dryandri OS sh Hibbertia echiifolia OS sh Hibbertia oblongata OS sh Hibbertia sp. ? sh Hibbertia tomentosa ? sh Hibiscus arnhemensis OS sh Hibiscus menzeliae OS sh Jacksonia spicata ? sh Leptosema uniflorum OS sh Lindernia sp. OS a, h Livistona inermis R t Macarthuria vertex OS sh Mitrasacme glaucescens OS a, h Mitrasacme scrithicola ? h Owenia vernicosa R t Phyllanthus carpentariae OS s Polycarpaea corymbosa OS a, h Polycarpaea incana ? p, h Portulaca bicolor OS a, h Sauropus rigidulus OS sh Senna cladophylla OS a, sh Spermacoce rupicola OS a, h Stemodia lythrifolia OS a, h

95 3: Habitat

Table 3.8—continued

Templetonia hookeri OS sh Tephrosia sp. ? sh Tephrosia spechtii OS sh Terminalia carpentariae R t Triodia microstachya R pg Triumfetta arnhemica R sh Xanthostemon paradoxus R t OS = obligate seeder, R = resprouter, t = tree, sh = shrub or sub-shrub, a = annual, p = perennial, g = grass, s = sedge, h = herb.

Table 3.9. Indicator species for dataset 2 of the Nitmiluk Vegetation Survey.

Positive for Pityrodia presence Negative for Pityrodia presence Life Life Species Strategy form Species Strategy form p <0.01 Cyperus microcephalus R p, s Aristida holathera R p, g Hibbertia oblongata OS sh Petalostigma quadriloculare R sh Sauropus rigidulus OS a, h Sarga intrans OS a, g

p <0.05 Boronia lanceolata OS sh Erythrophleum chlorostachys R t Calycopeplus collinus OS s Eucalyptus phoenicea R t Cassytha capillaris ? v Haemodorum coccineum R ph Cochlospermum fraseri R sh Indigofera haplophylla OS a, sh Mitrasacme scrithicola ? h Schizachyrium sp. OS a, g Stemodia lythrifolia OS a, h Tephrosia spechtii OS sh Trema tomentosa ? sh OS = obligate seeder, R = resprouter, t = tree, sh = shrub or sub-shrub, a = annual, p = perennial, g = grass, s = sedge, v = vine.

The patchy nature of the distribution at a local scale is clearly evident in the lower Gubara, Nitmiluk and upper Gubara survey results. This is indicated firstly by visual inspection of plotted presence/absence data for lower Gubara, and secondly by the TTLQV analyses. TTLQV requires long transect lengths and could not be used at all study sites. Describing patchiness simply by presence or absence within quadrats is a crude method in comparison with TTLQV, particularly for defining patches, as density is not considered. However, gap lengths defined by absences are reliable minimum estimates. True gap lengths may be longer if gaps include low Pityrodia densities, but if

96 3: Habitat field counts are accurate, true gaps cannot be shorter. Gap length estimates are also limited by size and choice of study sites.

The distribution at smaller scales is also clearly far from uniform. For example the whole mapped area at lower Gubara represents one or at most two large and moderately well defined, isolated patch(es). The nearest known other populations are approximately 0.5 km to the south and 1.5 km to the north east. Vegetation to the north is lowland savanna and is unsuitable for Pityrodia. Habitat immediately to the west and east appears largely unsuitable because the dominant vegetation type is monsoon rainforest. However, it is also very rugged and impossible to survey at the same intensity as the Gubara site.

At a finer scale, the pattern of Pityrodia distribution at the study sites is dominated by small patches and small gaps with loose clumping. It is possible to interpret this simply as sparse distribution, at least as far as it is relevant to the ecology of the grasshoppers. Gaps as small as one or two 5-m quadrat widths can be traversed even by flightless nymphs. The gaps reported here at larger scales might, however, be an important influence on grasshopper dispersal. There is evidence (Chapter 4) that gaps in the order of 102 m present an impediment, if not a complete barrier, to dispersal by adult female Petasida.

The relatively small patch sizes for Pityrodia potentially leaves some Petasida populations vulnerable to fire. While the mean length of burnt transect sections measured in Chapter 2 was less than the diameter of a typical habitat (i.e. Pityrodia) patch for Petasida, and the majority of both burnt sections and unburnt sections (gaps) were small, a relatively large proportion of the burnt areas occurred over long distances uninterrupted by unburnt gaps. The result suggests that in some circumstances entire (sub)populations of Petasida might fall within unbroken burnt areas. This danger to populations is mitigated somewhat by a degree of fire protection afforded by the the rocky habitat of Pityrodia (discussed below).

97 3: Habitat

The analysis of the two datasets extracted from the Nitmiluk Vegetation Survey data supports the reported observations that Pityrodia is restricted to sandstone habitats and in particular, to sandstone heath vegetation. Clearly, large rocks are an important component of the habitat of these species of Pityrodia. Even in the analysis of Dataset 3, collected from sites with Pityrodia present and from which very little information on habitat differentiation could be extracted, rocks were the most prominent variable associated with Pityrodia presence. Yibarbuk et al. (2001) noted an association between rocks and obligate seeder shrubs, the floristic group most associated with Pityrodia in the current study. In their analysis of the original, entire Nitmiluk Vegetation Survey dataset, Michell et al. (2004) found that boulder cover was the single environmental variable that differentiated the vegetation types that they defined. Pityrodia, however, is also clearly associated with shallow soil and an open vegetation structure.

Patchy fires are associated with rocks. Price et al. (2003) found that even the hottest fires, which burnt all vegetation on flat ground without rocks, left some unburnt patches amongst rocks. Topographic protection from fire, provided by rocks, has also been cited as an influence on the distribution of monsoon rainforest (Russell-Smith et al. 1993) and stands of Callitris intratropica (Bowman et al. 2001b) in the sandstone plateau and escarpment country. While Pityrodia presence was strongly associated with slope in Nitmiluk National Park, it was negatively associated with tall vegetation. The ordination results suggest that floristics, including Pityrodia distribution, are more influenced by gradients in cover of various sized rocks than larger scale topographic variables such as slope. It would appear that total or near total protection from fire does not promote Pityrodia. At sites with Pityrodia present, slope did not influence either Pityrodia presence or Pityrodia density, although it might have a small influence on Pityrodia species. These results are consistent with that of a GIS modelling exercise carried out on the Kakadu fireplot data (A. Edwards, pers. comm.) which found no correlation between Pityrodia presence and any topographic variables except proximity to watercourses. Drainage lines have previously been recognized as part of the habitat of Pityrodia (Lowe 1995)

98 3: Habitat

The floristic results provide further insights into the fire ecology of Pityrodia. The only non-herbaceous obligate seeder amongst the negative indicators was Grevillea pteridifolia, a tree typical of moist soil conditions. The group of negative indicator species was dominated by long-lived resprouting trees and perennial (resprouting) grasses more typical of lowland savannas than sandstone escarpments. These are very fire tolerant species. By contrast, the positive indicators were predominantly short-lived obligate seeders. While these species are intolerant of frequent fires, they set seed within 2-3 years and are not eliminated from an area by less frequent fires. However, all the long-lived positive indicators were resprouters; there was a conspicuous absence of long-lived obligate seeders. This is strong evidence for a history of at least some relatively short inter-fire intervals in the habitat of Pityrodia at Nitmiluk. I return to this issue in the development of models of the Pityrodia-fire dynamics (Chapter 5).

Communities occurring within sandstone vegetation are not necessarily clearly defined. Rice and Westoby (1985) found a clear distinction between lowland and sandstone vegetation, but only continuous variation within the sandstone at Koongarra, 1–4 km south east of Gubara. One gradient they identified, but which was not measured in the current study, was soil moisture. In contrast, Bowman et al. (1990) found complex and sharp spatial patterning in sandstone country at Jim Jim falls, ca. 50 km south of Gubara. They did, however, note substantial floristic continuity between communities. In analysing the full NVS dataset (1528 quadrats) Michell et al. (2004) found the vegetation of Nitmiluk National Park to be 'highly fluid, meaning that most species can occur with a wide variety of other species.' In this light the clear separation of Pityrodia in the ordination analysis, and the high number of strong indicator species may be seen as surprising. The tight clumping of Pityrodia quadrats in the ordination of the first dataset is all the more impressive given that the group consists of five species. The series of GLMs and ordinations of the environmental and floristic data suggest that Pityrodia is strictly confined to the sandstone heath vegetation, but that even within strict geomorphologic and edaphic bounds there is some specialization, particularly for rocky sites. The results for Gubara suggest that even though individual species may have

99 3: Habitat further, more specialized (unmeasured) requirements, all but the wet habitat within those sites is suitable for Pityrodia, but that not all of it is occupied.

The results describing local distribution patterns are important for modelling because they confirm the patchy distribution of grasshopper habitat and indicate the need for the incorporation of that patchiness in one of the models. The patterns described also help define an appropriate size for a single patch of grasshopper habitat. The central importance of rocks as a determinant of Pityrodia habitat implies that extensive habitat suitability modelling would be either difficult or unproductive, and that modelling efforts should more profitably focus on variations in fire regimes and on grasshopper dynamics. The results also highlight the importance of investigating not only the characteristics of sandstone fire regimes in general, but the characteristics of those regimes in rocky areas in particular (Chapter 2).

100

Chapter 4

Population biology of Petasida ephippigera and Pityrodia spp.

Chapter 4: Population biology of Petasida ephippigera and Pityrodia spp.

4.1 Abstract

Leichhardt's grasshopper (Petasida ephippigera) is restricted to habitat defined by the patchy distribution of its host plants of the genus Pityrodia in sandstone areas of the Northern Territory (NT). Fire potentially acts on Petasida populations both through direct mortality and through impacts on populations of the host plants. This study aimed to investigate the behaviour of populations of both grasshoppers and their host plants in both the presence and absence of fire.

The density of two species of Pityrodia was surveyed before and after fire at two locations in Kakadu and Nitmiluk National parks, and density and size classes for one species were surveyed at four sites, two of which had been burnt one year previously, in Kakadu (KNP) for two consecutive years. Two other sites in Kakadu were surveyed for one or two years. Individual plants were tagged at four sites in Kakadu in order to measure recruitment and mortality in one year. Surveys for Petasida were conducted at several locations across the 'Top End' of the NT. Mark-recapture studies were conducted over four seasons, mostly in the Gubara area of Kakadu. These aimed to estimate both population sizes and dispersal patterns. Nymph population and movements were studied in a grid of fixed quadrats.

The density of Pityrodia stems increased after fire and decreased in the absence of fire. Two examples of mass mortality of stems in the absence of fire, in different species, are reported. Oviposition in soil by Petasida is reported here for the first time. Timing of the life cycle and a description of aggregation by nymphs is reported. Most Petasida populations were very low and sparsely distributed. One population approximately doubled annually until it was reduced after half the Pityrodia patch it occupied was burnt. Several local extinctions of Petasida are reported, some of which were clearly not caused by direct mortality due to fire. The dispersal ability of Petasida is relatively low,

4: Population biology but a 'fat-tailed' movement distribution indicates occasional longer-distance dispersal by flying rather than walking. No Pityrodia patch in the main study area was isolated from the others by a distance greater than the measured dispersal capabilities of the grasshoppers. Petasida populations in the study area appear to conform to the requirements of a classical metapopulation, but the particular conditions and circumstances indicate that a spatially realistic simulation model would be more appropriate for this system than an incidence function model.

4.2 Introduction

It is highly likely that fire affects populations of Leichhardt's grasshopper (Petasida ephippigera) both directly by grasshopper mortality and through its impact on the grasshopper's habitat. The main resource offered by that habitat is the presence of plants of the genus Pityrodia which provide the primary, and probably sole, wild food source for Petasida. Lowe (1995) reported rapid regeneration of Pityrodia after fire. The reproductive strategy as defined by Gill (1981b) of most, if not all, of the Pityrodia species that support Petasida is 'resprouter' rather than 'obligate seeder'; five of the seven species known to host Petasida were observed during this study to sprout from the root crown after the above ground parts were killed by fire. In addition, several species were also observed to produce copious quantities of seed. Regardless of the responses of individual Pityrodia plants to fire, little is known of the impacts of fire on populations. In modelling the fire responses of populations of Petasida it is of particular importance to know whether fire tends to increase or decrease Pityrodia populations, and how those populations behave in the absence of fires.

The dynamics of plant populations at the regional scale are notoriously difficult to study in detail (Freckleton and Watkinson 2002), particularly in relation to dispersal and colonization. These problems are compounded here because several species of Pityrodia are capable of supporting Petasida populations, and they may not necessarily share similar population dynamics. This study therefore did not aim to investigate the regional population dynamics of Pityrodia in detail. Rather, I focussed on how best to incorporate some broad attributes of Pityrodia stand dynamics into models of fire interactions with

102 4: Population biology

Petasida populations, and specifically to answer the questions of whether such models would be best served by treating Pityrodia as fixed or ephemeral habitat patches, and whether fires affect the availability of Petasida habitat.

A knowledge of the life cycle of Petasida is important in modelling fire impacts because the life stages of the grasshoppers are expected to have different susceptibilities to fire, depending firstly on their state of protection, for example of eggs by soil, and secondly on their mobility, and hence their ability to escape fire or to find fresh browse after fire. Questions of the timing of life cycle stages are also likely to be of critical importance to managers, as not only may the fire responses of grasshoppers differ with life cycle stages, but the fires themselves vary in intensity and extent with season. While altered fire regimes pose possibly the most serious immediate threat to biodiversity in the sandstone habitats of northern Australia (Russell-Smith et al. 2002; 1998), fire is also the most valuable and useful tool available to conservation land managers of the sandstone country. Those managers must satisfy a variety of competing interests and require data with which to inform decisions regarding prescribed burning.

Pityrodia populations are largely confined to substrates with a high proportion of exposed rock, and are distinctly patchy (Chapter 3), and therefore so are the grasshopper populations. Given this patchy distribution, it appears likely that metapopulation theory will provide an appropriate and useful framework for understanding the population dynamics and fire relations of Petasida. Metapopulations are characterized by discrete populations connected by limited migration (Hanski and Gilpin 1997), with local extinction and colonization usually, but not always, as key features. Grasshopper metapopulations in Europe have been described by Appelt and Poethke (1997) and Carlsson and Kindvall (2001). It is probable that Petasida populations conform to metapopulation structure at some scale, especially as the sandstone landforms themselves are patchily distributed. The question of scale is critical, and on it depends the usefulness of a metapopulation approach. Scale depends to a certain extent on patch size and distribution (Chapter 3), but most importantly on the dispersal ability of Petasida. This is because dispersal ability is likely to be the critical factor in the ability of Petasida to recolonize habitat patches after local extinction by fire or other causes.

103 4: Population biology

Methods of investigating dispersal in a metapopulation context have been reviewed by Imms and Yoccoz (1997) and methods of marking insects were reviewed by Hagler and Jackson (2001), who list remarkably few methods by which individual animals can be identified. In order that both populations and dispersal could be estimated simultaneously I used a mark-recapture method for Petasida in the current study, using commercially available bee tags used previously (L. Lowe, pers. comm.) and recommended by Hagler and Jackson (2001).

The major aims were to:

1. examine changes in the density of Pityrodia over time in both unburnt areas and areas subject to fire during this study;

2. examine recruitment and mortality, and changes in size structure in populations of Pityrodia in both recently burnt and long unburnt areas;

3. investigate the method of recruitment (resprouting or seed germination) to Pityrodia populations after fire;

4. sample Petasida populations at a variety of locations in order to gain some understanding of normal population levels;

5. examine the distribution of Petasida at a km2 scale;

6. study the local population dynamics of Petasida in one patch in detail;

7. investigate the dispersal patterns and range of Petasida; and

8. investigate the impact of fire on Petasida populations.

4.3 Methods

4.3.1 Pityrodia

4.3.1.1 Study sites

All except one of the sites used here are identical to some of those used and described in Chapter 3. These were: 'GP1', 'GJ1', 'GTB', 'GTG', and 'NaN' in Kakadu National Park (see Figs 3.3 and 3.5) and 'NIT', located at the top of the escarpment at the mouth of Katherine Gorge in Nitmiluk National Park. In addition, a new site, labelled 'GJE', was

104 4: Population biology established 650 m east of site GJ1 in Kakadu at 12° 50.334'S, 132° 52.197'E. Sites GTB, GTG and NIT were burnt by wildfire in August 2002. In all cases fires were large, mid dry season fires which burnt for many days. Scorch heights observed at the study sites indicated fires of relatively high intensity.

4.3.1.2 Density and size classes

Density surveys were carried out in both 2003 and 2004 at five sites: NIT, GJ1, GTG, GTB, and NaN. In addition, density data were collected during preliminary surveys in 2001, and during firescar surveys in 2002, at Nitmiluk and the two upper Gubara sites. Site GP1 was surveyed on 4 May-5 June 2004 and 22 January 05. Data were collected concurrently with the environmental data used in Chapter 3, and a detailed description of the methods is given there (Section 3.2.4.2, Table 3.1).

At all sites where P. jamesii was present data were collected in six size classes: 0-10, 10- 20, 20-50, 50-100, 100-200 and >200 cm. At other sites (NIT and GP1) only total density was measured. Density data consisted of counts of individual plants. These were defined as any plant with the main stem emerging from the ground more than 10 cm from the nearest other Pityrodia stem. If stems were separated by <10cm, the taller stem was counted and measured.

Because site NIT is comparatively very large, with a low density of Pityrodia, quadrats of size 15 m x 5 m were made by amalgamating groups of three adjacent quadrats. At this site two quadrats were randomly selected from each transect.

4.3.1.3 Mortality and recruitment

Permanent quadrats were established at five sites: GJ1, GJE, GTB, GTG and NOU. Individual Pityrodia jamesii plants were tagged with aluminium foil tags wired loosely to the main trunk and the height of each plant was recorded.. At GJ1 and NOU, nine 5 x 5 quadrats were randomly placed within a 100 x 100 m grid. Single quadrats were placed at the less accessible sites due to time constraints. These were 15 x 15 m at GJE and GTB and 10 x 10 m at GTG. Initial tagging was done in 2003 and sites were revisited once in 2004. Details are given in Table 4.1.

105 4: Population biology

Table 4.1. Location, size and dates of permanent quadrats in Kakadu in which P. jamesii plants were tagged. Site Location quadrat quadrat Date established Date revisited size (m) number GTG Upper Gubara 10 x 10 1 4 Sep 03 10 Sep 04 GTB Upper Gubara 15 x 15 1 4 Sep 03 10 Sep 04 GJE Lower Gubara 15 x 15 1 12 Aug 03 7 Sep 04 GJ1 Lower Gubara 5 x 5 7 17 May 03 28 May 04 NOU Nourlangie 5 x 5 8 23 Jul 03 8 Sep 04

4.3.1.4 Seedling counts

At sites GJ1 and GTG on 24 April and 18 May 2003 respectively, several small P. jamesii plants were examined in order to determine their status as seedlings, root suckers or resprouts. Plants were selected by walking along a swath width of 10 m and selecting each 10th plant under 50cm high. The ground around the base of the plant was then scraped away with a tent peg. Plants were classified as root suckers if they were attached to a horizontal root <8cm from the surface and as resprouts if attached to an old stem. Seedlings were those with juvenile leaves and a tapering tap root not attached to any other roots. 33 plants were examined at GTG and 50 at GJ1. In cases where the tap root could not be traced because of immovable rocks, digging was abandoned and the next 10th plant was selected.

At site NIT young plants were inspected on 15 Feb 2004, with a similar goal, during the second wet season after the fire in August 2002. No attempt was made to randomize selection because the ground was too rocky to dig out most plants. Instead, P. lanuginosa plants < 70 cm high were selected in groups of 10 in each of five arbitrarily selected sandy areas. Plants which were obviously old, as evidenced by thick, woody stems or woody stumps were not selected. The procedure then followed was as described above.

106 4: Population biology

4.3.2 Petasida

4.3.2.1 Study sites

Grasshoppers were tagged at three locations in Kakadu (Nourlangie Rock, lower Gubara, Upper Gubara) and at one site in Nitmiluk National Park (Table 4.2, Fig. 4.1). Most of the tagging sites used were also used for collection of the environmental and floristic data described in Chapter 3. They were: NIT in Nitmiluk National Park; NOU at Nourlangie Rock in Kakadu; and GJ1, GP1 and GPE in the lower Gubara area of Kakadu. In addition, two new sites in the lower Gubara area, designated GJT and GJS, were used (Fig. 2). Sites GTB and GTG and the surrounding area were amalgamated into a single site of ~128 ha, designated GTOP. Most grasshopper tagging efforts were concentrated in and around the site GJ1 in Kakadu.

In addition, other searches were carried out at many locations in Kakadu, Nitmiluk and Keep River National Parks, in the upper Mann River region of Arnhem Land and at Umbrawarra Gorge near Pine Creek. Searches at Keep River and two of those in Nitmiluk National Park are described in Wilson et al. (2003). Details of some of these, where relevant, are given in results.

In the lower Gubara area Pityrodia locations were recorded along transects spaced at 100 m during the distribution patterns study reported in Chapter 3. Where high Pityrodia densities or large patches were found within this area, further transects, spaced at 50 m, were walked, both east-west and north-south. At several sites, detailed data on Pityrodia distribution were recorded during the floristic and environmental correlates studies detailed in Chapter 3. In addition, Pityrodia locations were recorded by handheld Global Position System (GPS) unit during searches for Petasida, especially at the perimeter of patches. All these records were plotted in Arcview (Version 3.2, ESRI) and the perimeters of patches were plotted by hand. Pityrodia mostly occurred in discrete patches, with some sparse and scattered plants between. All the study sites in Fig. 4.2 are separated by spans of at least 50 m in which no Pityrodia at all were detected.

107 4: Population biology

Site GJ1 was partially burnt in December 2005. The fire was a small, low intensity management fire, lit in the early wet season, by rangers, to reduce fuels adjacent to a nearby walking track. The burnt area was mapped on 27 Feb 2006 by walking the boundary of the firescar and recording it with a handheld GPS, followed by plotting in Arcview.

Figure 4.1. Locations of study sites in the Gubara and Nourlangie Rock areas of Kakadu National Park. For details of the lower Gubara sites see Figs. 3.5 and 4.2. Gridlines are at 1 km intervals.

108 4: Population biology

Table 4.2. Details of surveys for adult Petasida ephippigera in Kakadu and Nitmiluk National Parks, 2000-2006. Site Location Season Sampling No. of Population Dates of dates samples estimator single searches GJ1 Lower 2002-3 20-24 Jan 03 5 Schumacher Gubara 2003-4 9-12 Dec 03; 4 Schumacher 31 Mar; 15 Apr 8,9,10 Jan 04; Schumacher 20-23 Jan 04 Schumacher 3,4,5,6 Feb 04; Schumacher 18,20,22 Feb 04: Schumacher 3,6 Mar 04 Petersen 17,18 Mar 04 Petersen 20-23 Jan 04 4 Schumacher 18-22 Feb 04 3 Schumacher 17-18 Mar 04 2 Petersen 2004-5 19-22 Jan 05 4 Schumacher 9 Dec 2005-6 27 Feb- 3 Schumacher 1 Mar 06 GP1 Lower 2002-3 20-24 Jan 03 5 Schumacher Gubara 2003-4 21-23 Jan 04 3 Schnabel 3,4,5,6,19,20,22 Feb; 3,6,17,18,31,Mar 2004-5 20-22 Jan 05 3 Schnabel 2005-6 27 Feb GPE Lower 2002-3 21-24 Jan 03 3 Schnabel Gubara 2003-4 3-5 Feb 04 3 Schnabel 2004-5 20-22 Jan 05 3 Schnabel 2005-6 27 Feb

109 4: Population biology

Table 4.2. continued. Site Location Season Sampling No. of Population Dates of dates samples estimator single searches GPS Lower 2003-4 4-5 Feb 04 3 Petersen 19,20,22 Feb; Gubara 3,11,17,18 Mar GJT Lower 2001-2 29 Dec; 29 Jan Gubara 2002-3 25 Jan 2003-4 22,23 Jan; 4,27 Feb; 11 Mar 2004-5 21-22 Jan 05 2 Petersen NE Lower 2001-2 14 Nov Patch, E Gubara Patch, SW Patch 2002-3 12 Dec 2003-4 6 Feb, 24 Mar W Patch Lower 2003-4 27 Feb Gubara GTOP Upper 2001-2 22-24 Jan 02 3 Schnabel 12,13,29,30 Dec Gubara 2002-3 26 Jan 2003-4 '15,16 Jan 04 2 Petersen 17,18, 19 Dec; 17 Mar NOU Nourlangie 2001-2 14 Dec 01 Rock 2002-3 21-24 Jan 03 4 Schumacher 2003-4 30 Jan – 4 Schumacher 10,11 Dec; 20 Feb 2 Feb 04 2004-5 20-22 Jan 05 3 Schnabel NIT Nitmiluk 2000-1 19,20 Dec 2001-2 16,30 Dec; 18 Jan 2002-3 6 Nov 02 2003-4 12,14,15 Feb 04 2004-5 1,2,3 Mar 2005-6 2 Nov 05

110 4: Population biology

Figure 4.2. The study area (heavy line), study sites (fine lines) and distribution of Pityrodia patches in the lower Gubara area of Kakadu. The 'GJ' and eastern patches, together with the 2 small north-western patches all contain P. jamesii. The 'GP', western and south-western patches all contain P. puberula. Petasida have been found during this study in all the labelled patches and in none of the unlabelled patches.

4.3.2.2 Mark-recapture

Grasshoppers in the nymphal stages were not tagged because an efficient tagging system, capable of surviving the final moult, could not be devised. Adult grasshoppers were tagged by attaching apiarists' queen bee tags (Hagler and Jackson 2001) to the pronotum with cyanoacrylate glue. These tags were coloured, numbered, plastic discs ~1.5 mm in diameter. At the time of tagging any distinctive characteristics (e.g. disfigured tegmina or wings) were noted. In order to detect lost tags, one or other wing was marked with a xylene-free, felt-tipped black marker pen. In a few cases where tracking of individuals was not to be done, only marker pens, not bee tags, were used. In most cases locations were recorded using a handheld GPS unit (usually a Garmin 12; other models were occasionally used by park rangers).

111 4: Population biology

At each site, the initial search was carried out in a grid pattern with the aim of examining every Pityrodia plant within the site. With experience, workers became familiar with the locations of Pityrodia plants and the search area decreased. Inevitably, searchers also became familiar with the locations of grasshoppers and in order to counter any tendency to concentrate on those areas, deliberate measures were taken to ensure that equal effort was devoted to searching areas not previously known to hold grasshoppers. Full population estimates were made by sampling on at least three occasions within one week. In almost all cases this meant on consecutive days. In a few cases sampling for full population estimates was restricted to two days. At other times, searches were conducted on single days and reported as a single count. The sampling program is detailed in Table 4.2.

The adult Petasida population in site GJ1 and the surrounding areas was studied intensively during the wet season of 2003-4 (and the nymphs at GJ1 thereafter, see below). At site GJ1, full population estimates were made at least monthly throughout the wet season of 2003/4 (December to March), with a single search in April. This period covered the lifespan of almost all adult grasshoppers. In order to track the movements of grasshoppers, additional searches and tagging were also conducted between the main tagging occasions. Full population estimates were made for the neighbouring sites GP1 (Jan and Feb), GJT (Jan), GPE (Feb) and GPS (Feb), with some additional searches of these sites being conducted at other times. Searches of site GJS and the patches labelled in Fig. 4.1 were made in February. The eastern Pityrodia patches had also been searched during the dry season (10 Aug) of 2003 and no nymphs had been found. The unlabelled Pityrodia patches west of the Koongarra track (Fig.4.2) were searched during February and March. To the west and southwest of the study area, searching was only done incidentally, throughout March 2004, along the Pityrodia mapping transects spaced at 50 and 100 m.

Full population estimates for site GJS were made in January of 2003, 2004, 2005 and February/March of 2006. Full population estimates were also made for neighbouring sites GP1 and GPE in the same consecutive wet seasons, except for 2006 when a single search was made for GP1 only. In other areas, full population estimates were made for

112 4: Population biology site NOU at Nourlangie Rock in the 2002/3, 2003/4 and 2004/5 wet seasons, and at site GTOP (Upper Gubara) in the 2001/2 and 2003/4 wet season with a search in the 2002/3 season. At site NIT in Nitmiluk National Park, searches were conducted each wet season from 2000/2001 to 2004/5.

Populations were estimated using one of three methods, all described in Krebs (1999). Normally the Schumacher-Eschmeyer estimator was used. Where populations were very low and only three samples were taken the Schnabel estimator was used, and where only two samples were taken a simple Petersen estimate was made. Where a single count was made it was reported as such.

4.3.2.3 Nymph quadrats

A permanent grid, 120 x 35 m, consisting of 168 contiguous 5 x 5 m quadrats was established in the dry season of 2004 in order to monitor part of a population of Petasida nymphs. A tape was laid between permanently marked points at 0, 60, and 120 m on a north-south line straddling the site GJ1. A second tape measure was laid perpendicular to the first one at 10 m intervals, extending 15 m west and 20 m east and 5 x 5 m quadrats were marked out using 2.5 m lengths of PVC pipes. Surveys took place on 1-4 May, 7 July and 9 Sept 2004.

In the initial survey Pityrodia jamesii plants in each of the six size classes described previously were counted in every quadrat. In all surveys every plant was searched and the number of Petasida nymphs found on each plant was recorded. The height of each Pityrodia plant with Petasida present was measured and recorded. During the first two surveys only those plants with Petasida present were tagged with an aluminium tag loosely wired to the main stem.

113 4: Population biology

4.4 Results

4.4.1 Pityrodia

4.4.1.1 Density and size classes

Sites NIT, GTG and GTB were burnt by wildfire in 2002 and site GJ1 was burnt in late 2005. No other Pityrodia study sites were burnt during the study period. There was a sharp increase in mean Pityrodia density at each of the three burnt sites (Fig 3.3) regardless of the species involved (P. lanuginosa at NIT and P. jamesii at GTB and GTG). At four sites surveyed in both 2003 and 2004 there was a weak decrease in density of P. jamesii. There was a very sharp decrease in the density of P. puberula at GP1 between 2004 and 2005.

In the size class study, each of the burnt sites showed a similar size distribution in both years, with stem numbers generally reduced and overall distribution advanced by one size class in the second year (Fig. 4.4). That is, Pityrodia densities were relatively lower in the smaller size classes and relatively higher in the larger classes, but numbers were reduced overall. At the unburnt site GJ1, each year showed a similar shape of size distribution, but with reduced overall population in the second year. Densities in the middle classes were reduced, with little change or slight increases in the smaller and larger classes and no recruitment to the largest possible class. This pattern was duplicated at the other unburnt site, NaN, but with more marked reduction in densities in the middle size classes (Fig. 4.4). The results of tagging individual Pityrodia plants show that in the quadrats at GTG and GTB, which are within but not necessarily representative of those sites, most of the individuals grew in height (Fig. 4.5). In contrast, the tagging results at GJ1, which were randomly placed throughout a much larger area of the site, indicate that, while the majority of plants grew in height, increased numbers in the lower size classes may well be from decreasing height of individuals rather than recruitment from smaller size classes. At GJE, another long unburnt site, almost as many plants lost height as gained it.

114 4: Population biology

4.4.1.2 Recruitment and mortality

In the tagged plant quadrats mortality was higher than recruitment in all cases. However the ratio of mortality to recruitment was much higher in the two burnt quadrats than in the two unburnt ones, with no recruitment at all in the quadrat at site GTB.

Figure 4.3. Density of Pityrodia plants (±SE) at study sites in Nitmiluk and Kakadu, measured between 2001 and 2005. Sites NIT, GTB and GTG were burnt in 2002. Empty columns mean no data were collected, not a zero value.

115 4: Population biology

Figure 4.4. Density of Pityrodia jamesii at four study sites in 2003 and 2004. Size classes are: 1 = <0.1m; 2 = 0.1-0.2m; 3 = 0.2-0.5m; 4 = 0.5-1m; 5 = 1-2m; 6 = >2m. There was no fire at any of these sites between observations in 2003 and 2004.

116 4: Population biology

Figure 4.5. Change in height over 1 yr, 2003-2004, vs. initial height of tagged P. jamesii plants at 4 sites in Kakadu. Site GJ1 (c) was the only one at which grasshoppers were observed to be present during the period 2003-2004.

117 4: Population biology

Figure 4.6. Mortality and recruitment of Pityrodia jamesii in quadrats within four study sites between 2003 and 2004. Values are percentages of the total tagged 2003 population in one quadrat (15 x 15m at GJE and GTB, 10 x 10m at GTG) and seven quadrats (each 5 x 5m) at GJ1. The total population (n) of tagged plants in 2003 was: GJ1 (79); GJE (178); GTB = (57): and GTG = (97).

Two cases of mass mortality of Pityrodia, apparently without recruitment, were observed during this study, one indirectly and one directly. The first case occurred at site NOU, where P. jamesii plants were tagged in eight quadrats in July 2003. In those eight quadrats there were a total of 9 live plants ranging between 126 and 212 cm tall (mean = 177, SE = 8.5). There were also 28 dead plants between 122 and 248 cm tall (mean = 181, SE = 5.6). The appearance of the dead plants suggested a time since death of less than 1 year. In September 2004 only 3 tagged plants were still alive and there was no recruitment.

118 4: Population biology

The second case occurred at site GP1, where during the collection of environmental and floristic data in June 2004, unusually high numbers of dead P. puberula plants were noticed amongst a high density of living plants. The site was surveyed again in January 2005 and almost all the plants were dead (Fig. 4.3). The site was revisited, but not formally surveyed, in February 2006 and no evidence of recruitment could be found. It was estimated that even fewer plants were alive than in 2005.

4.4.1.3 Seedling counts

Of 33 P. jamesii plants <50 cm tall examined at site GTG, 9 were classified as resprouts, 24 as root suckers and one as a seedling. Of 50 P. jamesii plants at GJ1, 9 were classified as resprouts, 41 as root suckers and none as seedlings. At Nitmiluk, in contrast, of 50 P. lanuginosa plants, 42 were classified as seedlings and 8 were indeterminate.

4.4.2 Petasida

4.4.2.1 Population estimates

The main study site, GJ1, held the largest population of Petasida amongst all sites. Estimated annually in January, close to the annual peak of the adult population, this population approximately doubled annually from an estimated 77 in 2003 to 352 in 2005, before falling in 2006 (Fig. 4.7). At the peak of the population the density of adult grasshoppers was 158 ha-1. The site was partially burnt in December 2005 (Fig. 4.8).

Within the single wet season of 2003-4 the population of adult grasshoppers at GJ1 peaked in January and thereafter declined approximately linearly until no live grasshoppers could be found on 14 April (Fig. 4.9a). The patch of Pityrodia at this site is neatly bisected by a drainage line, and in the 2003-4 season most Petasida were found in the western segment (88% of the estimated pop on 23 Jan 04). An extra population estimate was made for the western segment and it shows a peak in the adult population on 12 January (Fig. 4.9b). The actual peak is probably slightly earlier because no nymphs at all were counted during the Jan 9-12 sampling period. Sex ratios

119 4: Population biology

(females/males, means of all counts made during population estimates) rose steadily from 1.28 in December to 2.64 in March (Fig. 4.10).

Throughout this study, the earliest nymphs were seen on 1 May (2005) at site GJ1. In the permanent grid within site GJ1 the highest count of nymphs was in July (2005: Fig. 4.11). This was not necessarily a closed population, however, and variations might have been influenced by dispersal or migration. In 2003 only one adult was found at site GJ1 in November. Numbers of nymphs and adults were approximately equal between 9-12 December and by 23 January no nymphs at all could be found.

The latest nymphs seen during the entire project were found on 23 Jan 2005 on P. jamesii at site GJ1 and the latest adults were two tagged animals found on P. puberula at site GPS on 15 April 2004. Very occasionally, unseasonally early adults were seen, with the earliest being found on 11 July 2004 on P. puberula in the W Patch. This grasshopper was clearly newly emerged, not a late survivor from the previous season, and was accompanied by a last instar nymph.

The highest density of adult grasshoppers recorded was 158 ha-1 at site GJ1 in January 2005. In the permanent quadrats established at the same site that year, at the height of the population of nymphs in July, the ratio of Pityrodia plants to grasshoppers was 2.8. By January this ratio, for the whole of site GJ1, had dropped to ~10 plants per adult grasshopper (based on 2004 plant counts and an annual decline of 17% [Fig. 4.3]).

The populations in the sites immediately surrounding GJ1 were all much smaller (Fig. 4.12). The Petersen estimate for the population at site GPS, not shown in Fig. 4.12, was 49 (95% CI 38.5-73.2) on 5 Feb 2004. Single counts taken during population sampling were, on average throughout this study, equal to 59.5% (n = 64, SE = 2.1) of final population estimates. In December 2001 three Petasida were counted in the P. jamesii patch labelled NE patch in Fig. 4.2 and in December 2002 one Petasida was found in the patch labelled E patch. Otherwise, except for the population at site GJS, no Petasida were found in any of the three eastern patches for the duration of the study. To the west of the Koongarra track lie extensive patches of P. puberula and, to a lesser extent, P.

120 4: Population biology jamesii. Seven Petasida were found during a search on 27 Feb 2004 of the patch labelled W Patch in Fig. 4.2. 15 live and two dead Petasida were found in the SW Patch, which lies partially within the study area, during incomplete searches between 13-25 March 2004. The presence of Petasida in both of these patches during the following wet season was confirmed during a quick visit on 23 January 2005. No Petasida were found in the other unlabelled patches within the study area in searches made on 13 and 18 March 2004. A few (six) isolated, scattered Petasida were found up to 500 m to the west and southwest of the study area during Pityrodia surveys between 18 and 25 March 2004.

Elsewhere in Kakadu (Fig. 4.13) the population at site NOU, which was partially burnt in 2002, remained steady in 2002-3 and 2003-4, before falling in 2005-6. Site GTOP, which burnt by a relatively intense wildfire in August 2002, showed a slight decline in 2002-3 before rising to the pre-fire level in 2003-4. At this site single counts were on average equal to 43.7% of the final estimate (n = 6, SE = 1.0), which would give a very approximate population estimate of 30 for 2002-3).

At all the known Petasida populations in Nitmiluk National Park, the grasshoppers were either too rare or too sparsely distributed and inaccessible to conduct mark-recapture estimates. Count data for two sites are shown in Table 4.3. At the third site in the park, upper Katherine Gorge, a four day search by four personnel on 7-10 September 2001 found 173 Petasida nymphs within a 2.5 km radius (Wilson et al. 2003). After the 2002 fires the site was revisited by 3 personnel on 25 Feb 2004 and in a full day's search no grasshoppers were found. On December 9-11 2000, an extensive search of 18 locations in Keep River National Park, Spirit Hills station and Bullo River station by four personnel using a helicopter and 4WD vehicle, found a total of 7 Petasida (Wilson et al. 2003).

121 4: Population biology

Figure 4.7. The population (± 95% CI) of Petasida ephippigera at site GJ1 in the lower Gubara area of Kakadu, estimated by the Schumacher-Eschmeyer method for four wet seasons from 2003 to 2006. Error bars represent 95% confidence intervals. The open symbol with the dotted line represents an estimate the population on 23 January 2006, based on the measured rate of population decline in 2004 (Fig. 9a). The site was partially burnt in December 2005.

122 4: Population biology

Figure 4.8. Site GJ1 showing the area burnt in December 2005, together with the locations of all Petasida tagged on 28 Feb 2006. The firescar covers 51% of the Pityrodia patch.

Table 4.3. Counts of Grasshoppers at two sites in Nitmiluk National Park. Both sites were burnt in August 2002. The Edith Falls site was burnt at least twice between then and 2004. NIT Edith Season Date no. grasshoppers Date no. grasshoppers 2000-2001 16-Jan-01 16 2001-2002 30-Dec-01 11 15-Aug-02 17 25-Nov-02 3 2002-2003 6-Nov-02 0 12-Aug-04 7 2003-2004 15-Feb-04 6 2004-2005 1-3 Mar 05 6 2005-2006 2-Nov-05 1

123 4: Population biology

Figure 4.9. (a) Population estimates (± 95% CI) and counts of Petasida at site GJ1 through the wet season of 2003-4. The March 18 values are a Petersen estimate based on two samples and the other estimates are Schumacher-Eschmeyer. The open symbol and dotted line at left represent the estimated adult population for 12 December plus 76 nymphs counted on 12 Dec. All counts are probably underestimates and the total population on 12 December is likely to be nearer to 235. No grasshoppers could be found on 14 April. (b) An extra population estimate was made between 8-10 January in the western segment of the Pityrodia patch. Most of the grasshoppers were in this segment (90% of all captures) and the added data provide a better representation of the peak of the adult population.

124 4: Population biology

Figure 4.10. Mean numbers of female and male adult grasshoppers counted at site GJ1 during the 2003-4 wet season. Numbers are means of three or four counts made during the population estimates.

125 4: Population biology

Figure 4.11. The density of (a) P. jamesii plants and (b-d) Petasida nymphs in permanent quadrats within site GJ1 throughout the 2004 dry season.

126 4: Population biology

Figure 4.12. Population estimates (± 95% CI) and counts for Petasida at sites surrounding the main study site in Kakadu (GJ1), between 2001 and 2005. Estimation methods are given in Table 4.1. Site GJS was partially burnt in 2002.

127 4: Population biology

Figure 4.13. Population estimates (± 95% CI) and counts for Petasida at (a) site NOU at Nourlangie Rock and (b) site GTOP at Upper Gubara in Kakadu between 2002 and 2005. Estimation methods are given in Table 4.1. Both sites were at partially burnt (NOU) or heavily burnt (GTOP) in August 2002.

4.4.2.2 Local Extinctions

A few local extinctions of grasshoppers were observed during the study, and while none can be definitely attributed to fire, at two sites fire is a possible cause. The first was at site NIT at the mouth of Katherine Gorge, where at least two previously occupied patches of P. lanuginosa were unoccupied after the 2002 fire. All further captures at this site during the three subsequent seasons were made along the cliff line of the gorge, which provided some fire protection to the only substantial unburnt Pityrodia patches. It should be noted however, that Petasida populations were very small and sparse before the fire. Secondly, in 1996 I observed a population of > 100 Petasida in a dense patch of Pityrodia, ~1 ha in area, at the base of the escarpment east of Koongarra Saddle in Kakadu (12° 51.0'S 132° 51.7'E), ~1300 m south of site GJ1. By 2000 there were none and the patch was still unoccupied when searched during this study in October 2002. This population had been well known to the park rangers who attributed the extinction to a fire in 1999. Indeed, the whole surrounding area had been a well known site for Petasida, but searches on three occasions between 2002-4 found only a single nymph in

128 4: Population biology

October 2002. Petasida populations in the area at upper Katherine Gorge that was searched in February 2004 were obviously much reduced (count = 0, see above), but the area was too large, the terrain too difficult and rugged, and the search too late in the season to definitely conclude extinction.

One small, isolated P. jamesii patch near the East Alligator River (12° 29.0'S 132° 59.4'E) in Kakadu held a population of 16 Petasida in the 2002-2003 wet season (reported by park rangers), but was unoccupied when I revisited it in August 2003 and February 2004. It was unburnt and the condition of the plants appeared to be healthy. Another small population of 15 Petasida tagged on 12 Dec 2002 near Gubara pools in Kakadu (12° 49.8'S 132° 52.6'E) had disappeared by 20 Jan 2003. The patch was unburnt, but was located right on a walking track frequently used by tourists. At a site west of Koongarra saddle (12° 51.9'S 132° 51.5'E), 650 m south of site GPS at lower Gubara, a small population of four adults in 2001 had disappeared by late 2002, and at the NE Patch at lower Gubara a population of two adults disappeared at the same time. Both patches had contained nymphs (in the first case for two consecutive years) and so the adults were probably not vagrants. Both patches were burnt in August 2002. The extinction recorded at site GP1 in 2006 can be reasonably attributed to the near extinction of the host plant.

Populations were observed to shift within patches after fire, particularly in the upper Gubara area, but it is unknown if this movement pattern is related to fire. At NOU and GJS, which were both partially burnt in 2002, the previously occupied unburnt areas were almost completely unoccupied by 2004 while the burnt areas became occupied. However, in both cases this movement appeared to also coincide with senescence of parts of the Pityrodia population.

4.4.2.3 Dispersal

In the wet season of 2003-4 a total of 393 Petasida were tagged, of which 339 (86%) were recaptured at least once. At site GJ1 261 grasshoppers were tagged with 230 (88%) recaptured at least once.

129 4: Population biology

Most Petasida do not move very far during their life. The mean maximal distance, which here means the distance between the tagging location and the final capture location, for females was 39.9 m (n = 186, SE = 3.4, median = 27.6) and for males was 73.3 m (n = 153, SE = 10.6, median = 27.0). The total distances moved by 339 recaptured grasshoppers from all sites in the lower Gubara study area are shown in Fig. 4.14. For females, there is no strong trend of increasing distance with total time at large (r2 = 0.04). A few individuals made movements of several hundred metres but length of time did not greatly increase the likelihood of long movements. However, large movements mean a much diminished likelihood of recapture, and so the largest movements made were possibly not recorded. The highest maximal distance for a male was 819 m. The highest for a female was >498 m (and <803 m) in 54-58 days, travelled by a grasshopper carrying a red tag with an illegible number. The colour was used to pinpoint the range of possible tagging locations and dates. The movement distribution is approximately negative exponential, but with a 'fat tail' (Fig. 4.15a, b).

There were 21 movements recorded between Pityrodia patches by 19 grasshoppers: 16 males and 3 females (Fig. 4.16). All Pityrodia patches were separated by distances of at least 50 m in which no Pityrodia plants could be found. Ten grasshoppers (9 males and 1 female) tagged in site GJ1 were subsequently recaptured at another site. This represents 4.3% of all recaptured grasshoppers that were tagged in GJ1. No grasshoppers from other sites were recaptured in GJ1. Eight grasshoppers moved from P. jamesii to P. puberula and one moved from P. puberula to P. jamesii. The single movement from P. puberula to P. jamesii is a dubious migration. It was made by a male between sites GP1 and GJT within an hour of being tagged and the grasshopper had returned to GP1 by the following day.

130 4: Population biology

Figure 4.14. The maximal distance (distance between the first and last capture locations) vs. the total time at large for 152 male and 188 female grasshoppers tagged in the lower Gubara area. The distance moved by the highest moving female is a minimum value only, for a grasshopper carrying a tag of a colour that limited the range of possible marking locations and times.

131 4: Population biology

Figure 4.15. Distribution of the movement rates (a) and maximal distances (b) of female Petasida at lower Gubara recorded throughout the 2003-4 wet season. Movement rate values were calculated for each recapture as the distance moved since the last capture divided by time since last capture. Maximal distances are in 5m size classes. Both y axes are log scales.

132 4: Population biology

Figure 4.16. Movements of grasshoppers between Pityrodia patches in the lower Gubara study area in Kakadu during the 2003-4 wet season. Only 3 inter-patch movements were by females: one each of GJ1 to GPS, GPE to GP1 and GP1 to GPS.

Males tend to move further than females and males accounted for 16 of the highest 20 distances (location of last capture to tagging location) recorded. The mean daily movement rate for males was higher than that for females (Table 4.4.). Single daily movement distances for animals recaptured after one day were also higher for males than females in January and February, but differed little between the sexes at the beginning and end of the season (Table 4.5). Single daily movement distances reached a peak in January for males and in February for females. The figures in Table 4.5 do not accurately reflect overall mean daily movement rates because the influence of GPS inaccuracy (~3m) is comparatively high at such short distances, and because the grasshoppers might move further immediately after release than undisturbed animals (Table 4.4), but they are useful for comparisons.

Individual nymphs were not tracked in this study, but the results of the permanent grid study indicate that it is unlikely that most of them remained on the same plant

133 4: Population biology throughout the season. The most striking result was the apparent aggregation of most of the (sub)population (73%) in one quadrat in July. Indeed, most of those animals were found on just two plants, with >120 nymphs on one single plant. While nymph populations were usually observed to be distinctly clumped throughout the study, this was the only time such a populous and tight cluster was seen. By September the two plants had been almost completely stripped of foliage. While some plants in the immediate vicinity showed signs of heavy grazing, most other plants in the area, even very close by, showed relatively little evidence of grasshopper feeding.

Table 4.4. Mean daily movement rates of Petasida in the lower Gubara study area, calculated as the distance between tagging and final capture points over total time at large, for different minimum periods of time at large. The lower rates with time probably reflect the diminishing influence of GPS error and reaction to capture. Movement rate (m day-1) (n, ±SE)

Minimum time since Females males tagging (days) 1 2.05 (187, ±0.24) 5.20 (187, ±0.24) 21 0.94 (130, ±0.09) 1.94 (130, ±0.09) 50 0.66 (69, ±0.08) 1.89 (69, ±0.08)

Table 4.5. Short term movements by tagged Petasida recaptured after 1 day (Dec, Jan and March) or 2 days (Feb). The figures must be interpreted with caution (see text) but are useful for comparisons. Mean distance (n; ±SE) (m) Females Males December 6.3 (124; ±0.8) 7.0 (90; ±0.9) January 6.32 (231; ±1.1) 30.9 (171; ±2.8) February 16.0 (58; ±1.9) 32.6 (30; ±7.4) March 11.8 (14; ±0.7) 9.0 (7; ±2.5)

4.4.2.4 Egg-laying

Oviposition related behaviour was directly observed on two occasions. The first was at approx. 10.30 am (CST) on the morning of 4 Feb 2004 at site GJ1. When a grasshopper was picked up from the ground to check the tag number it was seen that the distended

134 4: Population biology abdomen had been protruding into a hole in the sand, and that there was foam around the terminalia. The grasshopper was replaced over the hole with the abdomen protruding into it, and it stayed in that position for another five minutes, with slow pulsations of the abdomen. It then spent ~8 minutes slowly filling and covering the hole with sand and leaf litter with its hind legs, before walking 70 cm and climbing to within 6 cm of the top of a P. jamesii plant 1.5 m high. When it was recaptured two days later, 11 m away, it was copulating. The oviposition site was in damp sand 4 cm from the base of a P. jamesii plant 1.3 m high. The weather was overcast with a low pressure system passing over. This female was tagged on 9 Dec 2003 and last recaptured on 31 Mar 2004 (Fig. 4.17).

Figure 4.17. The recorded movements of grasshopper no. R82 over 113 days during the 2003-4 wet season at site GJ1 in Kakadu. Numbers are days from tagging at day 1. This animal laid eggs on 4 February 2004 (day 57) and was observed copulating 2 days later.

The second occasion was mid-afternoon on 18 March 2004, again on an overcast day at site GJ1. This animal was apparently disturbed by being photographed and abandoned egg-laying. The hole was in damp sand under leaf litter, 2.5 cm deep, and 2 cm from the base of a P. jamesii plant 1.6 m high.

135 4.4.2.5 Bark feeding

One further observation is noteworthy. In September 2002 at Gubara in Kakadu a single living Petasida nymph was observed on a dead P. jamesii plant approximately three weeks after the passage of fire Fig. 4.5). The fire had come to within 1 m of the plant and had killed it without any signs of char. The grasshopper had apparently survived by eating the bark. There were live Pityrodia plants within 10 m, but no other Petasida could be found.

Figure 4.18. Petasida nymph on a dead Pityrodia jamesii shrub approximately three weeks after a fire at Gubara in Kakadu National Park. The grasshopper has apparently survived by feeding on the bark.

136 4: Population biology

4.5 Discussion

4.5.1 Pityrodia

Fire which killed the above ground parts promoted a burst of growth in populations of Pityrodia stems. In those populations for which pre- and post-fire counts were made, the number of stems emerging from the ground increased after fire. Similarly, the measures of population structure indicate a cohort of recruits increasing in height in the burnt patches, with unburnt populations declining more uniformly across height classes.

There are some drawbacks to the use of simple stem counts and height classes. Stem counts do not necessarily correlate with biomass and the root suckers counted here do not necessarily represent genetic individuals. They are, however, the 'physiological unit of function' (Harper 1977). Plant heights do not necessarily correlate with age, although etiolation in suppressed plants reduces the inequality (Hutchings 1997). However, the object here is not a complete demographic study of Pityrodia, but rather to gain understanding of the impacts of fire on the habitat of Petasida, and the methods are quite adequate for that purpose. Specifically, in this study, a single fire increased rather than diminished the resources available to Petasida.

It is possible that fires in successive years could cause increased mortality through depletion of reserves, but this possibility was not tested during this study. The fire regimes examined during this study (Chapter 2) and described in the literature suggest that successive fires are a very real possibility, and the Landsat-based firescar mapping certainly shows several successive fires at some study sites for this project. However, it is also highly likely that very short fire intervals are not as common as the mean values or mapping would suggest. Firstly, while fuel build up allows very short intervals, the published data suggests that fuel does not always build up to levels high enough to allow successive fires in rocky areas, particularly if the dominant spinifex species is Triodia plectrachnoides. Secondly, given the amount of unburnt vegetation within firescars recorded in this study (Chapter 2) and by Price et al. (2003), it is quite possible that a significant proportion of any area mapped as burnt in consecutive years is not actually so.

137 4: Population biology

The behaviour of Pityrodia populations in the absence of fire is problematic because of the time-scales involved. The observed population declines were mostly in the single year of 2003-4 and weather variations or other factors cannot be excluded as a cause. However, the large declines directly measured at site GP1 in 2004-5 and indirectly observed at site NOU in 2002-3 support the finding of decline in the absence of fire. The timescale of this study was not long enough to conclude that the apparent extinction of those two patches was permanent. Pityrodia jamesii and P. puberula may well be facultative seeders, as P. lanuginosa appears to be, and as both species were observed to produce large quantities of seeds, a large seed bank may be expected. The reproductive responses of other heath plants are known to vary among populations (Gill and Bradstock, 1992) and one Tasmanian epacrid species varies from obligate seeder to resprouter according to microhabitat within the same population (Keith 2002). If fire cues such as smoke trigger germination as in many other heath species (Dixon et al. 1995) the patches will probably re-establish after the next fire.

Despite the observation of local Pityrodia extinction, the majority of patches observed during this study remained in place without obvious changes in area, whether burnt or not. At site NIT, P. lanuginosa populations had been unburnt for at least 12 years and still supported Petasida populations. It is clear that the dynamics of Pityrodia patches are in general operating on a different time scale than those of local Petasida populations. Nevertheless, the possibility of extinctions, particularly of P. puberula, should be considered in modelling Petasida population dynamics, most importantly with regard to recolonization probability by both Pityrodia and Petasida.

4.5.2 Petasida

4.5.2.1 Life cycle

The observation of oviposition in the soil is the first record for this species, although it was expected (Lowe 1995). The long adult life (>113 days) and repeated copulations suggest that females lay many batches of eggs, as do their near relatives (Rentz 1996).

138 4: Population biology

The first nymphs probably appear just as the last adults of the previous generation die. The nymph population measured at Gubara then continued to grow and was much greater in July than in May, suggesting a protracted period of emergence from eggs (or migration: see below). This is consistent with general observations made throughout this study and has important management implications, as it suggests that at least some eggs are protected during the period of maximum prescribed burning, from May to June. In Kakadu at least, generally no prescribed burning takes place in the uplands after June, until the wet season.

The strong aggregation of nymphs recorded at site GJ1 in July was not recorded at any other time during the entire project. Indeed, the finding stands in contrast to the results of the survey conducted at upper Katherine Gorge in September 2001, which found a 'sporadic' distribution and a preponderance of solitary nymphs (Wilson et al. 2003). However, no other population was studied as closely as that at GJ1. The aggregation could have two causes: movement by the nymphs, or a concentration of oviposition sites. The first is the more probable as detailed records of adult mark-recapture locations throughout the previous wet season reveal no pattern of clustering at the site. Such a movement of nymphs would render the population count an unreliable estimate of the seasonal population within the permanent grid of quadrats, animals may migrate to the aggregation from outside the grid. It is not currently known whether such aggregation behaviour is a common or isolated event for Petasida. Should it indeed be a common phenomenon, there are important management implications. Firstly, it must be taken into account in the design of any sampling program for nymphs. In this case at GJ1, if only two Pityrodia shrubs had been missed during a survey the population would have been underestimated by ~60%. Second, a relatively small fire could have a disproportionately large impact on a grasshopper population if it happened to burn that spot at the right time. When the site did burn in December 2005, the fire came within 8 m of that quadrat but did not burn it.

The reasons for and mechanisms of aggregation in juvenile Petasida are not obvious, but aggregation pheremones have been found or suggested in at least three species of Acridid grasshoppers (Wertheim et al. 2005). Aggregation provides benefits for

139 4: Population biology aposematic insects (Riipi et al. 2001), but at the time of the observed aggregation in July the nymphs were still in an early stage of the transition from cryptic to aposematic coloration. It is remotely possible that nymphs actively select fire resistant places in which to aggregate, for example by choosing large (old, unburnt) plants.

The last moult and the appearance of adults appears to be fairly synchronous, at least locally, in December. At site GJ1 in 2003-4 one adult was seen in November and no nymphs were seen in January. The exact timing is likely to vary geographically and with seasonal conditions; several adults were seen at sites GTOP and NOU during preliminary surveys in November 2001.

4.5.2.2 Mortality due to fire

The direct impacts of fires on Leichhardt's grasshoppers are unknown, but it is generally accepted that fires cause high mortality of flightless grasshopper nymphs. Bock and Bock (1991) attributed 100% disappearance of flightless species from Arizona grassland plots entirely to fire-caused mortality, and Vermiere et al. (2004) recommended the use of fire to control grasshopper populations at the nymphal stage in Oklahoma grasslands. In southwest Western Australia Whelan and Main (1979) reported grasshopper population densities of zero or close to zero after fire. Gillon (1972), however, reported that about 85% of grasshopper adults and nymphs fled before fires in an Ivory Coast savanna, but, of those that passed through the flames, about two thirds survived. Nevertheless, he also reported that fire 'decimates' the juveniles of annual species and emphasized 'the importance of not being juvenile at the time of fire'. In the absence of unburnt gaps, nymphs are particularly vulnerable to predation and starvation in the 'shock' phase before vegetation begins to regrow after a fire (Warren et al. 1987).

Mortality of Petasida nymphs due to fire is likely to be high because of their dependence on their patchy host species; they can only flee so far before their habitat runs out. However, they appear to have some ability to survive by eating the bark of dead Pityrodia plants (Fig. 4.18). Gandar (1982) reported survival of grasshopper nymphs that sought refuge within dense tufts of grass, but the sandstone spinifex species associated with Pityrodia typically burn fairly thoroughly and leave only shallow, sparse

140 4: Population biology tufts, if any, and would afford little protection. Petasida are also weak jumpers, which would restrict their ability to pass safely through a fire front.

Although soil is usually assumed to be a good heat insulator, the relationship between heat and depth during fire is complex and little understood. In a review, Whelan (1995) found that soil temperature is influenced by soil depth, texture and moisture, and by fire intensity and passage time. Most studies found that at 2.5 cm depth, temperatures usually remain well below 100°C, and under low intensity fires temperatures were barely raised at that depth. In an experiment with naturally laid grasshopper eggs and varying intensities of simulated fire, Branson and Vermiere (2007) found no significant increase, beyond that for controls, in egg mortality in pods deposited at 19 mm depth. Their most intense treatment simulated fire in a standing biomass of 4500 kg ha-1, applied for 46 s. In contrast, eggs of a different species at <6 mm suffered severe mortality. The single depth measurement of 25 mm for an oviposition hole for Petasida, which is a minimum because the animal was disturbed before oviposition took place, indicates that survival of Petasida eggs during fire would be high.

The survival of adults and eggs is also expected to be much higher than that of nymphs, as adults have the ability to escape through flight. This means that the period of highest vulnerability begins in May and reaches its peak when all the eggs have hatched but when the nymphs are still small, probably in July, and continues until December when the winged adults appear. This period corresponds with the period of most burning under both contemporary and Aboriginal fire regimes. It would be expected that the life history of Petasida evolved under pre-Aboriginal fire regimes. Given the proposed antiquity of the Australian monsoon (Bowman 2002) these would presumably have been characterized by lightning-driven fire activity at the end of the dry season and, to a much lesser extent, at the end of the wet season, when the grasshoppers are at their least vulnerable. Despite such apparent disadvantage, however, their current existence is a legacy an ability to persist under Aboriginal fire regimes.

Other impacts of season of burn probably occur indirectly through its influence on the other, highly interrelated, fire regime variables such as intensity, extent and patchiness.

141 4: Population biology

Fires of high intensity are likely to be more effective in killing grasshoppers, but given the assumption of already high mortality of nymphs during the major burning period, greater attention in modelling should be paid to extent and patchiness. The mean fire sizes reported from the literature demonstrate that most heath fires cover an area much greater than that of the typical Pityrodia patch sizes measured in Chapter 3 (25–120 m diameter). Indeed, assuming high nymph mortality, such fires could easily cause local extinctions of Petasida populations in the absence of unburnt gaps. Even within patchy fires, the evidence from the two datasets examined here demonstrates the regular occurrence of continuously burnt areas large enough to encompass whole Pityrodia populations. While all fires in rocky areas are patchy, isolated Petasida populations, whether remnants, recent colonizations or otherwise, could be particularly vulnerable to such large fully burnt areas.

Many animals rely on unburnt gaps for refuge to escape fires or to provide resources after fire (Whelan 1995), and Gandar (1982) reported grasshopper biomass in unburnt gaps to be over three times higher than the pre-fire levels. The survival of a nymph on a plant killed by a fire front less than a metre away (Fig. 4.5) suggests that even very small unburnt gaps may be of considerable importance in providing refuges. However, it has been argued, at least for some plant species, that patchiness alone is not enough to ensure population persistence (Bradstock et al. 1996; Ooi et al. 2006).

4.5.2.3 Populations

In general, populations of Petasida appear to be widespread but sparse, with pockets of much higher populations, such as that in the lower Gubara area of Kakadu. Populations in most patches found across the Top End were very low, and most that were studied for more than one year varied erratically. Such variations in small populations would be expected through stochastic variations alone, as would extinctions. There is, however, a vast area of sandstone country in the Top End, largely trackless and empty of people, a great deal of which is potential habitat for Pityrodia and Petasida. It is impossible for one team to survey any significant proportion of this area, even with the expensive aid of helicopters.

142 4: Population biology

The growth of the population over 3 years at site GJ1 demonstrates such a potential in established Petasida populations, yet it was not observed elsewhere. It is possible that other populations were simply too low, and that more time was required for them to grow beyond the lag phase of an exponential or sigmoidal growth curve, or to escape the influence of Allee effects. It is probable that the population in the large SW Patch of P. puberula at lower Gubara had the potential to reach or exceed the size of that at GJ1. Comparable sized populations have been reported from 'Round Jungle', approximately 50 km south of Gubara in 1987 (300-400 grasshoppers: A. Press, pers. comm.) and from Mt Borrodaile in Arnhem Land in 2000 (L. Lowe, pers. comm.). The long term population growth reported at GJ1 might limit the applicability to Petasida of some metapopulation models such as Incidence Function Models (Hanski 1994; Hanski 1997) which, in ignoring local population dynamics, implicitly assume that populations reach their carrying capacity within patches in a single year after colonization (Baguette 2004).

4.5.2.4 Dispersal

The movement estimates for Petasida (median maximal distance ~27 m) are similar to those reported for other grasshopper species in many northern hemisphere studies, for example maximal distances (lifetime dispersal distances) of 20 yards (Aikman and Hewitt 1972), 7-100 m (Appelt and Poethke 1997), 10-100 m (Barton and Hewitt 1985) and 14-28 m (Mason et al. 1995). However, the measurements of dispersal made here for Petasida, and elsewhere for other species, are almost certainly underestimates of the dispersal ability or range for the species. It has long been known that unequal sampling, with less intensive sampling at greater distance, underestimates dispersion (Berger et al. 1999; Purse et al. 2003), but even with distance weighting to correct for sampling bias, estimates may not be accurate (Porter and Dooley 1993). The ideal method for studying dispersal is radiotelemetry (Ims and Yoccoz 1997), which is as yet unavailable for grasshopper studies. Mark-recapture is appropriate but it, too, underestimates dispersal distances (Baguette et al. 2000; Petit et al. 2001). The constraints of terrain, geography and available search-time restricted the study area in this project to the maximum manageable area, but that, too, may be a source of inaccuracy; Schneider (2003) found a highly significant correlation between the size of the study area and estimates of mean

143 4: Population biology movement distances in a review of butterfly mark-recapture studies. It is clear, however, that (1) in the study area female Petasida are not highly mobile, but nevertheless (2) given the maximum measured dispersal (~500 m), no habitat patches within the study site were beyond a (re)colonizable distance from the other patches.

Two further factors complicate understanding of the pattern of dispersal. The first is the potential existence of two separate modes: walking and flying. Observations of flight by female Petasida were extremely rare during this study, while walking was common. Yet the presumably rare dispersal events by flight are very important, as it is by flying that recolonization of Pityrodia patches is most likely to occur. The few longer distance movements recorded indicate that flights by females do occasionally occur. Bailey et al. (2003) improved the fit of random walk models of grasshopper dispersal by incorporating flying propensity, as well as philopatry (affinity for a home range), even though flying propensity was low. Long distance dispersal events lead to a 'fat-tailed' movement distribution, in which the tail drops off with distance less rapidly than predicted by a negative exponential distribution (Nathan et al. 2003), and which is approximated in Fig. 4.15 (a) and (b). Baguette (2003) found that using the fatter tailed inverse power function better predicted mark-recapture results for metapopulation dynamics in the cranberry fritillary butterfly than did the negative power function usually assumed in metapopulation modelling.

The second factor is the potential for non-random dispersal behaviour both within and between habitat patches. Indeed, differences between movement patterns within patches and movement in the 'matrix' between them have been shown in Orthoptera (bush crickets: Kindvall 1999), beetles (Wiens et al. 1997) and butterflies (Schultz 1998). Between patches, matrix characteristics cause connectivity differences (Taylor et al. 1993) which may profoundly influence dispersal and migration patterns in butterflies (Fleishman et al. 2002) and beetles (Wiens 1997). Habitat detection abilities also cause deviations from random movement patterns, for example because of particular search patterns (butterflies: Conradt et al. 2001; Kindlmann et al. 2004) or upwind orientation (suggesting the possibility of olfactory cues) (grasshoppers: Narisu et al. 1999; cactus bugs (hemiptera): Schooley and Wiens 2003). At least four other species of grasshopper

144 4: Population biology have been shown to respond to plant odours in wind tunnel experiments (Helms et al. 2003), and the highly aromatic nature of Pityrodia (patches are easily detectable by humans) would suggest a high likelihood that olfactory cues are important to Petasida. Similarly, little is known of the influence of landscape features on Petasida dispersal. Some populations are separated by rock outcrops ~300 m high which may present barriers to weak fliers, although the presence of Petasida in many plateau locations would indicate otherwise. The influence of the density of scattered and isolated Pityrodia plants between denser patches (aggregations of stems) also remains unknown.

Some long range dispersal ability is indicated by the relatively wide distribution of populations in separate sandstone patches across the Top End of the Northern Territory. The results of a nucleotide diversity analysis also indicate some long range dispersal ability (Y. Isagi, unpublished results). Isagi analysed two samples of Petasida from each of Gubara in Kakadu and Ngukurr, ~300 km to the east, using the technique of Amplified Fragment Length Polymorphism (AFLP). The samples were analysed concurrently with, and using the same methods as, samples of termite, Amitermes laurensis, populations from Cape Yorke Peninsula in Queensland (Ozeki et al. 2007). The distance between the farthest termite populations (~400 km) was roughly comparable to that between the two Petasida populations, yet the nucleotide diversity in Petasida was greater than that for the termites by a factor of seven. This finding indicates a surprisingly high rate of genetic exchange between the grasshopper populations, and suggests a much higher rate of long distance dispersal than was found in the current study. It is possible, indeed probable, that undiscovered Pityrodia patches and Petasida populations lie between Kakadu and Ngukurr, but it is also almost certain that very large areas unsuitable as habitat also occur. More genetic work, together with on-ground exploration and vegetation mapping, would improve understanding. However, the estimation of low probability, long-distance dispersal events remains one of the major challenges in the field of modelling dispersal generally (Hastings et al. 2005).

145 4: Population biology

4.5.3 Implications for modelling

The population growth recorded over three seasons at site GJ1 allows for an estimate of population growth rate and its variance which can be used to parameterize models, albeit with certain constraints, discussed in Chapter 5. It is currently not possible to accurately assess carrying capacity (K) of grasshopper habitat. However, the evidence suggests that at Gubara in Kakadu, the value of K is higher than any of the estimated population densities. The population at site GJ1, which held the highest density of grasshoppers observed during the study, was still rising when the site was burnt in 2005. In addition, the ratio of plants to grasshoppers was always greater than two, and significant plant damage due to grasshopper feeding was rarely observed. The results suggest that density dependence due to depletion of resources is not a critical factor to be considered in modelling. While the results also indicate that fire is likely to be an important cause of mortality in grasshopper populations, the available data do not allow for precise quantification of mortality rates. It is important, therefore, that the models allow for the effects of a range of rates to be examined in order to assess the sensitivity of results to this parameter.

The results indicate that in the Gubara area, Petasida populations do appear to conform to a classical metapopulation structure at the (km) scale studied. They satisfy the four major conditions set out by Hanski (1997): (1) habitat occurs in discrete patches; (2) even the largest local populations have a substantial risk of extinction; (3) habitat patches must not be too isolated to prevent colonization; and (4) local populations do not have completely synchronous dynamics. The time scale of this study did not allow for a conclusive answer to the question of whether or not the dynamics of undisturbed populations are completely asynchronous. However, given that fire is a common and important influence on the population dynamics, asynchrony between populations is likely to depend largely on the spatial patterns of burning. This topic is the subject of chapter 2. While the population and inter-patch dispersal results obtained here may not be comprehensive enough to fully parameterize one of the formal metapopulation models, it is clear that model development should include some form of between patch

146 4: Population biology dispersal, and that a model should have the capacity to be used to explore the effect of variation in dispersal rates.

Taking the density of Pityrodia stems as a surrogate for habitat quality, the evidence is strong that the direct impact of a single fire is an increase in extent or improvement of habitat quality. The most important implication of this result for modelling, however, is that fire does not reduce the area of habitat in the year after the passage of fire. The simplifying assumption that the presence of Pityrodia indicates suitable habitat for grasshoppers allows the use of spatially constant habitat patches in the models. However, the reported incidences of apparent local extinction of Pityrodia in the absence of fire, while apparently not universal, indicate that the models should also have the capacity to incorporate such events.

147

Chapter 5

Modelling the effects of fire regimes on populations of Petasida ephippigera

Chapter 5: Modelling the effects of fire regimes on populations of Petasida ephippigera

5.1 Abstract

The tools of Population Viability Analysis (PVA) were used to make relative predictions of the effects of variations in the components of fire regimes on populations of Leichhardt's grasshopper Petasida ephippigera. Two cellular models were created using Microsoft Excel© 2002 and the programming language VBA. The models were simple, discrete-time, count-based (unstructured), stochastic population growth models coupled with cellular landscape models in which fires with spatially realistic characteristics operate on the grasshopper populations. The first, fine scale, model was used to model populations in a landscape consisting of a single habitat patch of 1 ha and explicitly modelled the internal spatial structure (patchiness) of fires. The second modelled several habitat patches and incorporated dispersal of grasshoppers between patches and fires of varying sizes. The results of investigations and literature reviews in the previous four chapters were used to guide the construction of, and to parameterize, the models.

In simulations with the fine scale Single Patch model the probability of quasi-extinction

(Pq) declined sharply with increasing interfire interval and mean unburnt area. The declines were even more marked if short interfire intervals were eliminated by holding the interfire interval constant (variance = 0). The results were sensitive to a small degree to the variance of the mean unburnt area and that sensitivity increased at high values for the mean or the variance. The results were sensitive to variations in the intrinsic rate of population increase (r), but not particularly so to variations in carrying capacity (K). The results were relatively insensitive to variations in the coefficient of variation (CV) of r except at high values of the CV. The results were highly sensitive to variations in the value of the parameter 'probability of mortality in burnt cells'.

In simulations with the coarse scale Multiple Patch model both mean interfire interval and mean fire size had a strong influence on Pq. However, if the effects of Pityrodia

5: Modelling fire regimes and Petasida populations senescence in the absence of fire were modelled, the effect of fire interval was reversed, except at very short intervals, with Pq increasing at larger intervals. If any dispersal was present, Pq was slightly lower than with no dispersal, but otherwise Pq was remarkably insensitive to variations in dispersal parameters. Pq was sensitive to the total unburnt area within firescars, but not to the distribution of that area.

The results indicate that changes in fire regimes to larger (in extent), less patchy and more frequent fires are detrimental to populations of Leichhardt's grasshopper. Because fire regime variables are correlated, the results also indicate late dry season fires are more harmful than early fires. Management recommendations for the grasshoppers are precisely consistent with those for the conservation of obligate seeding sandstone heath shrubs. It is recommended that a combination of on ground and aerial ignition be used by land managers to create a pattern of early dry season burning in order to reduce the annual area of intense, late dry season burning.

Several questions relating to fire and the population biology of the grasshoppers and their Pityrodia host plants remain unanswered. The most important of these include: what is the rate of direct grasshopper mortality due to fire; how prevalent is senescence of Pityrodia, without recruitment, in the absence of fire; is there a diapause stage in the life cycle of the grasshoppers; how often and how far do grasshoppers disperse; and what is the mortality rate during dispersal?

5.2 Introduction

5.2.1 Background

It has been speculated that altered fire regimes in the sandstone heaths of the Top End of the Northern Territory are causing declines in the distribution and abundance of Leichhardt's grasshopper, Petasida ephippigera (Chapter 1). Evidence presented in Chapter 2 strongly suggests that fire regimes have changed in the last century with the decline in most areas of traditional Aboriginal burning patterns. This evidence suggests that contemporary fire regimes are characterized by fires that occur, on average, later in

149 5: Modelling fire regimes and Petasida populations the dry season, and that are more frequent, larger in extent and more intense than those under traditional regimes.

The sandstone heath habitats, to which Petasida is endemic, have always been a fire- prone environment and populations of Petasida can clearly cope with fires at some level. The persistence of the species is a legacy of both pre-Aboriginal and traditional Aboriginal fire regimes. One potentially important means of survival during and after fires is in refuges provided by unburnt gaps within firescars, as the sandstone heath fires are characterized by a high degree of internal patchiness, particularly in rocky areas (Chapter 2). Rockiness was the single variable found to be most associated with the distribution of Pityrodia, the host plants for Petasida (Chapter 3). Nevertheless, evidence presented in Chapter 4 strongly suggests that individual fires cause high mortality in Petasida populations, while healthy Petasida populations can increase quickly in the absence of fire. There is also some evidence to suggest that long absences of fire from the landscape do not favour the persistence of stands of adult Pityrodia plants (Chapter 3). Generally, however, there is no evidence that burning Pityrodia reduces the quality of habitat for Petasida (Chapter 3).

The main questions raised by the array of observations set out in the foregoing chapters, and which this chapter seeks to address, involve the effects of variations in fire regimes on populations of Petasida. In particular, how do changes in frequency, area and internal patchiness affect populations. Questions relating to season and intensity will be investigated indirectly through the effects of correlated variables. These questions are important not only because changes to fire regimes may be causing impacts on Petasida populations, but because fire management is the main tool – often the only one – available to land managers for broad scale management of sparsely and patchily but widely distributed species such as Petasida.

5.2.2 Population Viability Analysis

A comprehensive definition for PVA does not exist and indeed the term includes a number of methods. Morris and Doak (2002) use the broad definition 'the use of quantitative methods to predict the likely future status of a population or collection of

150 5: Modelling fire regimes and Petasida populations populations of conservation concern'. The common feature of the various methods is the use of quantitative models of population growth. PVA has been criticised by a number of authors, mostly on the grounds of the imprecision of predictions (e.g. Beissinger and Westphal 1998; Ellner et al. 2002; Ludwig 1999). Although some authors suggest the abandonment of PVA (Coulson et al. 2001), few recommend superior alternatives (Brook et al. 2002). In general, however, critiques have conceded the usefulness of the methods and have recommended precautions and improvements. Other authors have highlighted the heuristic value of the process of model building (Akçakaya and Burgman 1995; Brook et al. 2002). Brook et al. (2000) demonstrated the accuracy of PVA, at least in cases with good data sets spanning long time periods.

In most cases, however, comprehensive datasets do not exist, largely because the species of concern are mostly rare and often endangered, and hence difficult to study. Nevertheless, PVA is still a useful tool. In the words of MacCallum (2000) 'it will usually be the case that a set of approximate parameter estimates in a formal mathematical model is preferable to an arm waving argument.' PVA offers the advantage that the methods are rigorous and transparent (Morris and Doak 2002), and assumptions and uncertainties can be made clear and accounted for. Morris and Doak (2002), amongst many others, also urge practitioners to keep models as simple as possible, particularly when data are limited. Akçakaya and Sjögren-Gulve (2000) echo this call, and further point out that PVA is most useful when addressing a specific question involving a focal species, and when focussed on relative rather than absolute results.

Against this background, the approach adopted here was to use simple, discrete-time, count-based (unstructured), stochastic population growth models coupled with cellular landscape models in which fires with spatially realistic characteristics operate on the grasshopper populations. The data set for Petasida populations is limited in comparison to some others, for example those for many northern hemisphere butterflies, because of the comparatively short time available for the study and because no prior data exist. Following the recommendations of Akçakaya and Sjögren-Gulve (2000), the specific questions addressed in this study involved the impact of fire regimes on populations of the focal species, Petasida ephippigera. The results were expressed in terms of

151 5: Modelling fire regimes and Petasida populations probability of extinction within a time period and are relative rather than absolute predictions of extinction times. All parameters which were estimated or arbitrarily set were modelled at a range of values in order to assess their relative influence on the results.

Two related but separate models were created in order to investigate fire regime impacts on both a Petasida population occupying a single patch of Pityrodia, and at a broader scale, on a Petasida metapopulation occupying a landscape containing several habitat gaps. The specific advantage of modelling a single patch is that it allows an investigation of the effects of the internal spatial structure (unburnt gaps) of fires. In addition, because the quality and quantity of data relating to population growth within a single patch are greater than those relating to multiple patches, a more reliable result can be obtained for the single patch. However, it is difficult to use the Single Patch Model to investigate the effects of the size (spatial extent) of fires because the scale of that model (1 ha) is well below the size of most fires occurring in the sandstone heath habitats.

A Multiple Patch Model, operating at a broader scale, was used to examine the impacts of fire size. The scale of the multiple-patch model is also more appropriate for assessing the impacts of the relatively extensive unbroken areas of burnt vegetation which occur within firescars, even in rocky areas (Chapter 2). It is assumed here that grasshopper mortality in such areas is very high because survivors of a fire would be vulnerable to starvation and predation. The final benefit of the multiple-patch model is that it allows for the incorporation of dispersal of grasshoppers between patches.

5.2.3 Modelling: population growth and dispersal

Detailed descriptions of the models are provided under Methods and in Appendix 1, but some introductory discussion of population growth and dispersal modelling is warranted here. Petasida is an annual species and the simulated population grows in annual increments (after the passage of any fires). The population left after mortality (Nt) to fire is allowed to grow according to a discrete-time logistic equation of the form

2 Nt+1 = Nt + (r + S.yt) - rNt /K

152 5: Modelling fire regimes and Petasida populations where N is the total population of adult females at time (year) t, r is the intrinsic rate of population increase, s is the standard deviation of r, y is a random standard normal deviate and K is the carrying capacity. This form of the equation allows for stochasticity to operate at or near K, rather than the curve flattening to a straight line at K, as occurs under the more usual form (Burgman et al, 1993).

The empirical data give no indication of what type of density dependence, if any, occurred in the Petasida populations. McCallum (2000) argues that the logistic growth curve chosen for use here is probably extremely rare in nature, but because of its simplicity he nevertheless regards it as appropriate for use in simple population models. The alternative recommended by Morris and Doak (2002) for models where no empirical data exist is a simple ceiling model, which is also a simplification of most real world situations.

Although there is strong evidence that Petasida populations conform to the conditions defining a classical metapopulation (Chapter 4), the standard metapopulation models are of limited use in the present study. While the theory may provide a useful framework for understanding the system, the best-developed and most commonly discussed models are not appropriate. State transition models require observations of extinctions and colonizations, and therefore are appropriate for systems with high turnover rates (Hanski 1997). Incidence function models (IFMs) require data from at least 30, preferably 50 patches (Hanski 1994) and make a set of assumptions which are not universally applicable. In particular, the assumption that populations reach carrying capacity within one year in every patch is demonstrably false for Petasida (Chapter 4). Baguette (2004) found that IFMs have only been effectively applied under very restrictive conditions in systems dominated by human-induced fragmentation and he strongly cautions against the exclusive use of classical metapopulation theory in conservation management of natural systems.

While limited movement of female Petasida between habitat patches is clearly demonstrated by the records of three such events (~1.5% of the female population at lower Gubara: Chapter 4), there are no data relating to mortality during dispersal. In the

153 5: Modelling fire regimes and Petasida populations absence of sufficient data to adequately parameterize one of the formal metapopulation models, dispersal of grasshoppers between patches was incorporated in the Multiple Patch Model, but was treated in a simplified way.

Mortality due to causes other than fire is not modelled, but because Petasida is an annual species, annual mortality must always be 100%. However, it is not known what components of that mortality are due to mortality during dispersal or to mortality within the home habitat patch. Data on mortality during dispersal do not exist for Petasida, and it is not known whether dispersal is directed or random. Given the demonstrated ability of other grasshoppers to respond to olfactory cues during dispersal (discussed in chapter 4) and the highly aromatic nature of Pityrodia, random dispersal can not be safely assumed. Consequently, in the absence of any such data, simulation of dispersal in the Multiple Patch Model is based on the number of successful dispersers rather than the total number leaving the home patch. Even so the default value is based only on the three individual records mentioned above.

How far the grasshoppers disperse is based on the negative exponential dispersal Kernel. It has been argued that the inverse power function provides a better description of dispersal than the negative exponential because, with its fatter tail, it better accounts for long distance dispersal (Baguette 2003). The recorded dispersal distance distribution for Petasida showed a tendency to a fat tail, but this was due to the rare inter-patch dispersal events; that is, the only dispersals that are included in this model.

5.2.4 Modelling: the fire component

In the years since MacArthur's (1967) and Rothermel's (1972) seminal models of fire characteristics and spread, a vast array of fire models has appeared. Very many of these are designed to investigate the characteristics of the fires themselves, and are aimed at improving fire fighting techniques and preventing loss of property and life from wildfire. Another series of models are aimed at investigating the ecological effects of fires and fire regimes. A large subset of these are aimed particularly at investigating the role of fire regimes in ecological succession (Keane et al. 2004). Models within this category have been classified according to the strategies used to investigate the ignition, spread

154 5: Modelling fire regimes and Petasida populations and effects components, amongst others (Keane et al. 2004). This broad category of models is useful in studying the impacts of fire regimes on populations of Petasida, particularly as the aim of this study is specifically to investigate the effects of fires.

Cellular automata models have proved particularly useful for this purpose, although they are not without limitations. These models are most commonly used to study the effects of fire frequency and intensity (Keane et al. 2003). Models are applied at a range of scales, from single plants to thousands of square kilometres. However, examples of fine scale models (ha and less) are vastly outnumbered in the literature by those applying at a much coarser scale. The approach adopted in this study was to simulate fires with cellular automata-based models, which acted on grasshopper populations within the Petasida population models. The specific advantages of cellular models for this study are: that they enable manipulation and measurement of the internal structure (unburnt gaps) of fires; they enable manipulation of the size and shape of fires; the effects of the fires may be allowed to influence grasshopper populations at a fine spatial scale (single cells); and the models can be implemented using a spreadsheet program. An analysis by Ratz (1995), showed that cellular fire models similar to those used here generate simulated fires with shape and fine structure in agreement with empirical evidence, albeit from boreal forests and at a broader spatial scale.

Few studies and even fewer fire models incorporate details of internal patchiness, and fewer still do so at the fine scales relevant to populations of Petasida. Notable exceptions are the work of Bradstock et al. (1996), Gill et al. (2003) and Price et al. (2003). The approach adopted here specifically incorporated internal patchiness in a spatially realistic manner that is consistent with empirically derived values (Chapter 2).

5.2.5 Aims

The primary aim of this chapter was to examine the impact of contemporary fire regimes on populations of Petasida.

In particular, the aims were as follows:

155 5: Modelling fire regimes and Petasida populations

1. to explore the effects of different fire regime parameters on relative extinction risk to populations of Petasida, both within a single habitat patch and within a metapopulation spanning several habitat patches.

2. to assess the effects of variations in several estimated input parameters on the relative extinction risk to populations of Petasida, again at two scales, in order to indicate important gaps in knowledge and to guide ongoing research programs.

3. to provide a descriptive and heuristic tool as an aid to adaptive management programs for Petasida.

4. to discuss the implications of the results for management of populations of Petasida and the requirements for further research.

5.3 Methods

5.3.1 General model description

Two cellular models were created using Microsoft Excel© 2002 and the programming language VBA. Each model is based on a grid of 50 x 50 cells surrounded by a 20-cell wide buffer, and each was used to simulate a landscape at a different scale. Within the simulated landscapes, Petasida habitat is represented by the presence of Pityrodia, usually in discrete patches or clumps of cells, but with exceptions as detailed below. The landscapes are populated with grasshoppers occupying the habitat patches and the grasshopper populations grow in annual increments according to the discrete time logistic equation discussed in the Introduction. Only female grasshoppers are included in the models.

The landscapes are subjected to fire at a specified annual probability (and hence frequency), of a constant intensity assumed to be sufficient to kill a specified proportion of grasshoppers in the fire affected area. With one exception (described below), all habitat characteristics, including quality, area and position, are assumed to be unaffected by fire.

156 5: Modelling fire regimes and Petasida populations

Grasshopper mortality due to fire is determined by the input parameter 'percent mortality'. In fact, survival in burnt cells is drawn from a binomial distribution with a probability of (100 – percent mortality)/100. This means that, at 97% mortality for example, a single grasshopper in a burnt cell has a 3% probability of survival and that, on average, 3 out of every 100 grasshoppers in burnt cells survive.

Environmental stochasticity is introduced through the logistic population growth equation and through the fire input parameters. Limited demographic stochasticity is incorporated in that fire survivorship values are sampled from the binomial distribution. However, any influence of demographic stochasticity is unlikely to be strong at population levels above the quasi-extinction level (see below).

Each run in a simulation proceeds for a default of 30 yr unless the total population declines to a quasi-extinction level of 10 females ha-1, in which case the run is terminated. Each simulation consists of 1000 runs. Habitat characteristics do not change between years or between runs of the model during simulations (again, with one exception).

The results are expressed as the probability of a population falling to quasi-extinction within 30 years.

5.3.1.1 Model Assumptions

Both models assume that population growth rate and variance are spatially autocorrelated throughout the landscape but, within that restriction, all patches in the Multiple Patch Model show independent growth. The mean and variance of r remain constant through time, but population growth shows negative density dependence. Otherwise, there is no temporal autocorrelation. Both models also assume constant habitat quality throughout all patches or clumps of Pityrodia cells and (with one exceptional option in the Single Patch Model) a random distribution of grasshoppers within each habitat patch or clump of Pityrodia cells. All model options except one (with Pityrodia senescence) in the Multiple Patch Model assume no change in habitat quality over time. The Multiple Patch Model also assumes no influence of internal patch

157 5: Modelling fire regimes and Petasida populations conditions (including death of the host plants) on the rate of dispersal. Genetic factors are ignored in both models. Possibly the most important assumption for both models, and for which there is some evidence, is the assumption of high mortality of Petasida in fire affected areas. Finally, the models assume no diapause stage longer than a few months in the life cycle of Petasida.

5.3.2 Single Patch Model

5.3.2.1 Model summary

The primary use for which the Single Patch Model was developed was to assess the effect of variation in frequency and internal unburnt gap area of fires on a population of Petasida within a single patch of Pityrodia. Beyond this, further simulations were generally aimed at assessing the relative importance of parameters for which values were estimated or arbitrarily set. One set of simulations was carried put in order to examine the effects of the distribution pattern of grasshoppers and Pityrodia within the patch.

Each cell in the Single Patch Model represents 2 m x 2 m, and hence the 50 x 50-cell working grid represents one hectare. This is a realistic area for a single small patch of Pityrodia (Chapter 3). Under the default option half of the grid cells are occupied by Pityrodia, which has a clumped distribution within the patch (represented by the grid). The distribution pattern remains fixed between years and between runs of the model. The grasshopper population is spread randomly amongst the Pityrodia cells, but each cell can only hold one grasshopper. Other grasshopper and Pityrodia distribution options are available (see below).

If probability dictates the occurrence of fire, the entire grid is considered to be burnt, except for one or more unburnt patches which are added by the program according to user-defined input parameters. The rationale is that at the scale of this model (1 ha), in almost all situations the entire Pityrodia patch would fall within a single firescar, but all fires within the rocky habitat of Pityrodia contain some unburned gaps (Chapters 2, 4). Details of the procedure are given in Appendix 1.

158 5: Modelling fire regimes and Petasida populations

This model also samples every fire with a random north-south transect and analyses the fire characteristics. The output includes summary descriptive statistics (mean and sd) for unburnt gap span lengths, density of gap spans km-1, % transect unburnt and % area unburnt. These simulated fire data can then be used to calibrate the input parameters against predicted final values, and to compare simulated fire characteristics to empirical data.

Once the patch was established as described, the sequence of events for the Single Patch Model is as follows:

1. Burn the patch, if probability dictates;

2. Add unburnt gaps (aggregates of cells) until specified total unburnt area is reached;

3. Analyse unburnt gap characteristics, measure unburnt area;

4. Calculate mortality of grasshoppers in burnt cells (survival is sampled from the binomial distribution);

5. Allow the grasshopper population to grow according to the logistic equation applied to the total surviving population after fire;

6. Use this result to randomly populate the grid for the next year; and

7. Repeat the process for 30 yr or until quasi-extinction occurs.

5.2.2.2 Grasshopper and Pityrodia distribution patterns

Five patterns of grasshopper distribution are available as options within the Single Patch Model (Fig. 5.9): random, random within a clumped distribution of Pityrodia occupying 25%, 50% or 75% of the grid, or highly clustered within the central 25% of the grid. Under the first four options a cell can only be occupied by a single grasshopper. Under the fifth option all cells within the central 25% of the grid can contain any number of grasshoppers, but the probability of occurrence diminishes with the cube of the distance

159 5: Modelling fire regimes and Petasida populations from the central cell. This option simulates the single example of extreme clustering described observed at Gubara (Chapter 4: Fig. 4.11). This option and the first option make the simplifying assumption that every cell in the grid contains Pityrodia.

5.3.3 Multiple Patch Model

5.3.3.1 Model summary

The Multiple Patch Model is used to investigate the effects of a range of fire regime parameters, including fire size, over a larger area which includes several habitat patches. This model is related to the Single Patch Model, but it incorporates fires of different sizes (areas), dispersal of grasshoppers between habitat patches and, under one option, senescence of Pityrodia.

Each cell in the Multiple Patch Model represents 0.25 ha (50 x 50 m), giving a total area of 625 ha. Habitat patches are mapped within the grid in a pattern loosely based on the Pityrodia patches mapped at Lower Gubara in Kakadu National park (Fig. 5.1). The pattern of patches is fixed and never changes, except under the Pityrodia senescence option (see below). Any cell within habitat patches can be occupied by many grasshoppers, up to the default carrying capacity of 450 female hoppers ha-1. Each time the population grows (annually) the new population in each habitat patch is divided evenly between the cells within that patch. In this model the annual sequence of events is as follows:

1. once the patches were established as described, burn (if dictated by probability);

2. calculate mortality due to burning;

3. disperse grasshoppers;

4. allow the population of each patch to increase;

5. use these results to populate each patch for the next year; and

6. repeat the process for 30 yr or until quasi-extinction occurs.

160 5: Modelling fire regimes and Petasida populations

Figure 5.1. The Distribution of habitat patches within the 50 x 50 cell grid for the Multiple Patch Model. Dark shading indicates patches initially occupied by grasshoppers under the default settings.

5.3.3.2 Pityrodia senescence

Limited evidence (Chapter 4) suggests the possibility that senescence of Pityrodia occurs in the absence of fire. One option for this model allows for death of Pityrodia, and consequently of grasshoppers, at a specified time in the absence of fire, followed by recruitment of Pityrodia after fire (following the observed pattern described in Chapter 4).

5.3.3.2 Grasshopper dispersal

Dispersal is treated in the following way. The user specifies a value, called here the ‘dispersal probability’, which is the percentage of grasshoppers in every patch that successfully move to a new patch. The dispersal kernel (probability density function) is negative exponential, with the probability of a grasshopper treated as moving and then landing (and surviving) in a patch defined by

P = e-aD

161 5: Modelling fire regimes and Petasida populations where D is the dispersal distance between patches and the parameter a, specified by the user, defines the shape of the curve (Fig. 5.2). The model effects dispersal by first drawing the number of dispersers from each patch from a binomial distribution (trials = population of patch, P = dispersal probability), and then taking each individual of the known number of dispersers from each patch and allocating it to a new patch. This is done in a programming loop which, for every individual disperser, uses a random number to allocate a column in a lookup table, calculated using the matrix of inter-patch distances and the appropriate dispersal kernel, which is then used to lookup the new patch identification number. Mortality during dispersal is not modelled.

5.3.3.3 Fire

Fires are ignited, at random coordinates, in randomly assigned fire years, and are controlled by the user by manipulating a set of input parameters. Only one fire can occur in each year, and each fire consists of a single burnt patch. Details of the methods used to generate fires are given in Appendix 1.

The internal spatial structure of fires is not directly modelled, although complex fire shapes can be generated with low values for the input parameter 'contagion probability factor’ (see Table 5.2 and Appendix 1). However, internal spatial structure is incorporated in the model in the following way. There are two kinds of burnt cells. In the first type, 100% grasshopper mortality occurs. This is designed to simulate empirical data which show a proportion of burnt areas occurring in large, unbroken spans (e.g. 100m). It is assumed that grasshoppers surviving a fire in such situations would subsequently suffer high mortality due to starvation or predation.

The total unburnt area, specified by the user, is then distributed evenly among the remaining, type 2, cells. Assuming zero mortality due to fire within unburnt patches, which are contained only within the type 2 cells, survival is increased in these cells according to the area calculated to remain unburnt in each cell. Details of the procedure are given in Appendix 1.

162 5: Modelling fire regimes and Petasida populations

Figure 5.2. The dispersal kernel used in the simulations at 5 values for the dispersal parameter 'a'. The maximum distance between habitat patches in the model landscape is 2000 m.

163 5: Modelling fire regimes and Petasida populations

5.3.4 Parameter estimation

The input parameters are summarized in Tables 5.1 and 5.2.

5.3.4.1 Grasshopper population parameters (both models)

The population growth rate r and its variance were estimated from the GJ1 population (Chapter 4) using the following equations, variations of which are given by Dennis et al. (1991), McCallum (2000) and Morris and Doak (2002)

q−1 = 1 + rˆ B +1 ii )NNln( q i=0 and

1 2 σˆ = B(ln( + ) − rNN ˆ) q −1 1 ii

where q is the number of years for which population estimates are made and the Ns are the population estimates.

The default starting population of grasshoppers is 100 ha-1, which is somewhat lower than the highest measured population density in the field (~150 females ha-1 at Gubara site GJ1: Chapter 4). The default setting for K is 450 ha-1, or three times the highest density recorded. The GJ1 population appeared to be still growing strongly at 150 females ha-1 when growth was interrupted by fire in 2006 (Chapter 4, Fig. 4.7). Nevertheless, the default value of K remains somewhat arbitrary, and a range of values was simulated in order to assess the relative importance of this parameter.

The default value for 'fire caused grasshopper mortality' was arbitrarily set at 97%. The figure has some basis in observations during this study that very few Petasida could be found after fires, and recorded population declines after fire (Chapter 4), as well as reports from the literature of high or total grasshopper mortality due to fire in other environments (e.g. Bock and Bock 1991).

164 5: Modelling fire regimes and Petasida populations

5.3.4.2 Fire parameters (both models)

The default fire frequency values are similar to those reported by Russell-Smith et al. (1997b). The value for the Multiple Patch Model is set slightly higher (0.33 cf. 0.25 for the Single Patch Model ) to compensate for the fact that because fires generally cover less of the grid, the fire frequency values for any single cell would on average be lower.

The default value of 15% for total unburnt area (Single Patch Model ) produces a mean simulated gap area of 19%, which is intermediate between the early and late dry season values recorded by Price et al. (2003) and close to the mid-dry season value calculated from their data for rocky areas (Chapter 2). The gap structure parameters (see Table 5.1) are not derived from data or theory. They were simply selected from a series of simulations as the combination that produced gaps with characteristics closest to those described in Chapter 2.

The default value for mean fire size in the Multiple Patch Model, 800 ha, lies between the early and late dry season values interpreted from satellite imagery by Yates et al. (2007) for sandstone in western Arnhem Land.

The default value of 35% for the parameter 'Area of 100% mortality' in the Multiple Patch Model is consistent with Figures 2.3 and 2.4 in Chapter 2, which report the percentage of quadrats falling within unbroken burnt sections of >100m in firescar transects.

5.3.4.3 Grasshopper and Pityrodia distribution patterns (Single Patch Model )

There are two densities of Pityrodia available for any cell in this model: present or absent. Examination of Pityrodia distribution data (Chapter 3) showed a clumped distribution of Pityrodia within patches, and that within those patches supporting Petasida populations, the proportion of quadrats occupied by at least one Pityrodia plant varied between 23% and 81%, with a mean of ~56%. On that basis the default Pityrodia distribution was set as clumped, occupying 50% of the patch (i.e. the grid) (see Fig. 5.10b). The pattern is fixed and does not change between years or runs. A description of the procedure used to generate this default and other patterns is given in Appendix 1.

165 5: Modelling fire regimes and Petasida populations

Only Pityrodia cells can hold a grasshopper and the grasshopper population is distributed randomly among the Pityrodia cells. Each cell can only hold one grasshopper.

The other Pityrodia and grasshopper distribution patterns (see Table 5.1, Fig. 5.10) were created in order to test the sensitivity of the model to differences in distribution patterns. These patterns only approximate the actual distribution of Pityrodia and no attempt was made to precisely simulate an empirical or theoretical distribution of Pityrodia.

166 5: Modelling fire regimes and Petasida populations

Table 5.1. Input parameters and default values, with some explanatory notes, used in the Single Patch Model.

Default Input parameter Rationale Notes value Fire parameters Randomly drawn from a lognormal Produces 19% in simulations, close distribution. When the target area is Mean total unburnt target area (%) 15% to empirically derived mid-dry season exceeded the program stops adding values (Chapter 2) unburnt patches. Several values were simulated in order to Coefficient of variation of total 0.5 Arbitrarily set test the sensitivity of the model to this unburnt target area parameter. Similar to satellite-derived values The probability of a fire occurring in any Fire probability 0.25 (Chapter 2, Russell-Smith et al. year: the reciprocal of the mean interfire 1997b) interval. (Chapter 2, Russell-Smith et al. Minimum interfire interval (yr) 1 Minimum number of years between fires. 1998) Gap structure parameters None of the gap structure parameters Determines the probability of a cell are derived from data or theory. changing state to 'unburnt' if any of its Contagion probability factor 0.3 However, by manipulating them it is neighbours are in that state, and affects possible to create spatial structures the complexity of the unburnt patch shape. for fires that mimic the configuration After the unburnt gap is created, if any of those described in studies from burnt cells are surrounded by this many sandstone environments. unburnt neighbours, their status is changed Filler value 7 to unburnt. This eliminates most of the The default values for all gap unrealistic burnt spots within the unburnt structure parameters were selected patch. Mean unburnt patch size (cells) 500 as the best (most closely related to Unburnt patch sizes are randomly drawn empirical data [Chapter 4]) from a from a normal distribution. These values Standard deviation unburnt patch 300 series of simulations with many were obtained by trial and error as an size (cells) parameter combinations. appropriate fit to empirical data.

167 5: Modelling fire regimes and Petasida populations

Table 5.1. continued Default Input parameter Rationale Notes value Grasshopper population

parameters Initial grasshopper density Arbitrarily set at 67% of the maximum 100 (hoppers ha-1) value recorded at Gubara Carrying capacity K (hoppers Arbitrarily set at 3x the maximum 450 ha-1) value recorded at Gubara Derived from 3 yr empirical data set Mean intrinsic rate of increase r 0.76 from Gubara Derived from 3 yr empirical data set Standard deviation of r 0.18 from Gubara Arbitrarily set, with some basis in field The basis for calculating survival in burnt Fire caused grasshopper 97 observation (Chapter 3) and various cells, which is sampled from the binomial mortality (%) published sources. distribution. The model will terminate a run of years and Quasi-extinction level (hoppers Arbitrarily set at 10% of initial -1 10 begin the next if the grasshopper population ha ) population density declines to this value. Clumped (25, 50 or 75%): grasshoppers randomly occurring within a clumped pattern, representing Pityrodia distribution, covering 25, 50 or 75% of the grid. one grasshopper Close to the mean percentage of per cell. Pityrodia and grasshopper clumped quadrats containing Pityrodia for all Random: randomly distributed cells

distribution pattern. (50%) sites occupied by Petasida recorded containing single grasshoppers. in Chapter 3. Highly clustered: cells in the central 25% of the grid can be occupied by many grasshoppers. Probability of occurrence of grasshoppers diminishes with the cube of the distance from a central cell.

168 5: Modelling fire regimes and Petasida populations

Table 5.2. Input parameters and default values, together with some explanatory notes, for the Multiple Patch Model. Default Input parameter Rationale Notes value Fire parameters Determines the probability of a cell changing Arbitrarily set to give a compromise state to 'burnt' if any of its neighbours are Contagion probability factor 0.3 between shape complexity and speed burnt, and affects the complexity of the of calculation. firescar shape. After the fire is created, if any burnt cells are surrounded by this many burnt neighbours, Filler value 7 Arbitrarily set their status is changed to burnt. This eliminates most of single unburnt cells within the fire. Within the range reported for western Mean fire size (ha) 800 Arnhem Land (Yates et al. 2007) Randomly drawn from a lognormal Coefficient of variation for fire distribution. 0.5 Arbitrarily set size The probability of a fire occurring in any Similar to satellite-derived values Fire probability 0.33 year: the reciprocal of the mean interfire (Chapter 2, Russell-Smith et al. 1997b) interval. Minimum interfire interval (yr) 1 (Chapter 2, Russell-Smith et al. 1998) Minimum number of years between fires Consistent with reported % quadrats The percentage of the total burnt cells in falling within unbroken burnt sections of which grasshoppers have zero probability of Area of 100% mortality (%) 35 transects of >100m. (Chapter 2, Figs. survival after fire due to increased 2.3, 2.4) vulnerability to starvation and predation. The unburnt area is distributed evenly Total unburnt area within firescar Close to mid-dry season value among the burnt cells not included in the 20 (%) (Chapter 2) above parameter. Survival in those cells is then increased proportionally.

169 5: Modelling fire regimes and Petasida populations

Table 5.2. continued Grasshopper population Default Rationale Notes parameters value Initial grasshopper density Arbitrarily set at 67% of the maximum Default initial population is 100 grasshoppers -1 100 -1 (hoppers ha ) value recorded at Gubara ha in 6 of the 10 habitat patches. Carrying capacity K (hoppers Arbitrarily set at 3x the maximum value 450 Applies independently to each patch. ha-1) recorded at Gubara Derived from 3 yr empirical data set Random variations in growth rate for each Mean intrinsic rate of increase r 0.76 from Gubara patch are independent. Standard deviation intrinsic rate Derived from 3 yr empirical data set 0.18 of increase from Gubara Arbitrarily set, with some basis in field The basis for calculating survival in burnt Fire caused grasshopper 97 observation (Chapter 3) and various cells, which is sampled from the binomial mortality (%) published sources distribution. Quasi-extinction level (hoppers Arbitrarily set at 10% of initial Applied to the whole population rather than -1 10 ha ) population density individual patches The proportion of the female population of Derived from mark-recapture data Dispersal probability .015 every patch that moves to and survives in (Chapter 4) another patch. Arbitrarily set, to allow a very low A parameter that defines the shape of the probability of dispersal between dispersal kernel (probability distribution). Dispersal constant ‘a’ 0.0002 patches at opposite extremities of the Lower values mean a greater probability of simulated landscape. dispersal over longer distances. Senescence was recorded (Chapter 4), 1000 Pityrodia, and grasshoppers, in a cell left but not commonly enough to assume Pityrodia senescence (yr) unburnt for this period all die. Recruitment of that it is a dominant feature of the (i.e. off) Pityrodia occurs after a subsequent fire. system.

170 5: Modelling fire regimes and Petasida populations

5.3.5 Simulations

5.3.5.1 Single Patch Model

Each simulation run was terminated if the simulated population declined to the quasi- extinction level. Otherwise the run extended for 30 years. Each parameter combination was modelled by 1000 runs. In each case all input parameters were set at the defaults given in Tables 5.1 and 5.2, except for those parameters being specifically investigated as described below.

In order to examine the effects of two major fire regime components, total unburnt area within firescars and mean interfire interval were modelled together. Six levels of mean unburnt area (0, 7, 13, 25, 35, and 43%), recorded from the model grid rather than the actual input parameter values, were modelled at each of seven mean interfire interval values (1, 2, 3, 4, 5, 6 and 10 yr, generated by the input parameter 'fire probability' values of 1, 0.5, 0.33, 0.25, 0.2, 0.17 and 0.1).

The effect of year to year variation in unburnt area was examined by holding the input 'mean total unburnt area' constant at each of two values, 13% and 25%, and varying the CV over 8 values: 0, 0.125, 0.25, 0.5, 0.75, 1.0, 1.5, and 2.

The effects of fixed interfire intervals of 1, 2, 3 and 4 yr were modelled at mean total unburnt areas of 7, 13, 25 and 35%.

Further simulations focussed on examining the effects of variations in parameters, other than the fire parameters, which were estimated or set arbitrarily, in order to examine the sensitivity of the results to variations in those parameters.

Five values of the intrinsic rate of increase (r) between 0.25 and 1.25, each with a coefficient of variation (CV) of 0.5, were modelled at each of four levels of carrying capacity K between 150 and 600. The influence of the variance of r was examined for three values of r by varying the CV over nine values between 0.25 and 4.

171 5: Modelling fire regimes and Petasida populations

The influence of the distribution pattern of grasshoppers within a habitat patch was examined by running the simulations with each of the five distribution patterns described in Table 5.1, at each of two mean unburnt area values, each with a CV of 0.5.

The influence of the parameter 'fire-caused grasshopper mortality' was examined by varying it over eight values between 70 and 100% at six values of mean interfire interval (1-6 yr), while holding all other inputs at the default values.

5.3.5.2 Multiple Patch Model

Eight mean fire size values between 200 and 1600 ha, in 200 ha increments, were modelled at eight mean interfire intervals ( 1, 2, 3, 4, 5, 6, 8 and 10 yr, generated by the input parameter 'fire probability' values of 1, 0.5, 0.33, 0.25, 0.2, 0.167, 0.125 and 0.1).

The effects of the two parameters relating to the internal patchiness of the fires were examined using four levels of the parameter 'area of 100% mortality' (20, 40, 60 and 80%) with each of four levels of the parameter 'total unburnt area within firescar' (5,10, 20 and 40%).

Pityrodia senescence values of 7, 10, 15 and 20 yr were modelled at mean interfire intervals of 2, 3, 4, 5, 6, 8 and 10 yr.

Dispersal probabilities at 6 values between 0 and 0.12 in 0.2 increments were modelled at 5 values of the dispersal constant 'a': 0.0005, 0.001, 0.002, 0.004 and 0.008.

5.4 Results

5.4.1 Single Patch Model

The Single Patch Model run at the default settings for 5000 iterations produces fires with a mean total unburnt area within the firescars of 19.0% (sd = 9.2), and a mean length of measured unburnt gap spans of 5.0 m (sd = 4.8). The mean adjusted gap span length (with all single cell gaps omitted) was 13.0 m (SD = 9.6). The frequency distribution of gap lengths (Fig. 5.3) is similar to those derived from transect data in Chapter 2.

172 5: Modelling fire regimes and Petasida populations

The probability of quasi-extinction (Pq) declined sharply with increasing mean interfire interval, particularly if larger areas within the firescar remained unburnt (Fig. 5.4). While almost all populations went extinct with annual fires, fire regimes with mean interfire intervals of 10 years only resulted in high Pq values if the unburnt gap area was small or zero. At any mean fire interval greater than 1 yr Pq was very sensitive to the unburnt gap area, particularly at high fire frequencies if gap area was high and at low fire frequencies if gap area was low. While variation in the unburnt area had some effect, especially at higher values, it was not as influential as the mean. However, this result indicates that a few fires with very low unburnt gap areas can have a large impact on Petasida populations even if the mean unburnt gap area is high.

Holding the probability of fire (Pfire) at 1 and varying the interfire interval caused Pq to decline even more sharply both with increasing interfire interval and with increasing unburnt area (Fig. 5.5). In this case the minimum fire interval equals the mean (variance = 0). This result indicates firstly that the high impact of successive fires or near successive fires on Petasida populations is not fully compensated for by longer intervals at other times, but secondly that fire regimes with no very short intervals have remarkably low relative impacts on Petasida populations.

The variance of the total unburnt gap area within firescars, for which the default was set arbitrarily, does have a non-trivial influence on Pq, particularly at high values for unburnt gap area (Fig. 5.6).

Variation in r showed a strong influence on Pq at all values of K, but the sensitivity changed little with variation in K (Fig. 5.7). The influence of K on Pq is small at values close to the default, but is higher at low K. At high values of r the population growth equation used here causes instability in population levels and higher estimates of Pq

(Burgman et al. 1993). The CV of r has remarkably little influence on Pq if r is close to the default value, (Fig. 5.8), but Pq is very sensitive to variation in the CV of r at higher values.

173 5: Modelling fire regimes and Petasida populations

The distribution pattern of grasshoppers has very little effect on Pq if the grasshoppers were distributed throughout the habitat patch, either randomly or randomly within clumps of Pityrodia (Fig. 5.9). However, Pq rose sharply if the distribution pattern was highly clustered in a small section of the patch, and the effect of variation in unburnt gap area was much reduced, with both simulated values resulting in very high values of Pq.

The probability of mortality in burnt cells exerted a strong influence on Pq, particularly at values over 0.9 (Fig. 5.11). Under the default parameters Pq declined to zero at a burning mortality value of 0.7.

Figure 5.3. Frequency distribution of measured gap spans produced by simulating 5000 fires at the default settings. Single-cell gaps have been omitted.

174 5: Modelling fire regimes and Petasida populations

Figure 5.4. Probability of quasi-extinction (Pq) within 30 years, for grasshoppers in a 1 ha habitat patch, vs. mean interfire interval at seven values of mean total unburnt area within the firescar. Other parameters were set at the default values: initial population = 100, r = 0.76 (CV = 0.25), K = 450, quasi-extinction level = 10.

175 5: Modelling fire regimes and Petasida populations

Figure 5.5. The effect of variations in a fixed interfire interval on probability of quasi- extinction within 30 years, for grasshoppers in a 1 ha habitat patch, at 4 values for mean total unburnt area. The fixed fire interval = mean fire interval, with a variance of 0. Other parameters were set at the default values, including: initial population = 100, r = 0.76 (CV = 0.25), K = 450 and quasi-extinction level = 10.

176 5: Modelling fire regimes and Petasida populations

Figure 5.6. The effect of varying the values of the CV of the total unburnt area within firescars on probability of quasi-extinction within 30 years, for grasshoppers in a 1 ha habitat patch, at 2 values for mean total unburnt area. The normal default value for the CV in other simulations is 0.5. Other parameters were set at the default values, including: initial population = 100, r = 0.76 (CV = 0.25), K = 450 and quasi-extinction level = 10.

177 5: Modelling fire regimes and Petasida populations

Figure 5.7. The effect of varying the intrinsic rate of population increase (r) on probability of quasi-extinction within 30 years, for grasshoppers in a 1 ha habitat patch, at 4 values of carrying capacity (K). Other parameters were set at the default values, including: initial population = 100, CV of r = 0.25, quasi-extinction level = 10, mean fire interval = 4 yr and mean unburnt area = 19%.

178 5: Modelling fire regimes and Petasida populations

Figure 5.8. The effect of varying the value of the CV of the intrinsic rate of population increase (r) on probability of quasi-extinction within 30 years, for grasshoppers in a 1 ha habitat patch, at 3 values of r. The value or r was set at the default (0.76) derived from field data. The default value for the CV, derived from field data and used in other simulations, is 0.25. Other parameters were set at the default values, including: initial population = 100, quasi-extinction level = 10, mean fire interval = 4 yr and mean unburnt area = 19%.

179 5: Modelling fire regimes and Petasida populations

Figure 5.9. The effect of distribution pattern of grasshoppers within a single 1 ha habitat patch on probability of quasi-extinction within 30 years, at 2 values of mean total unburnt area. Other parameters were set at the default values, including: initial population = 100, r = 0.76 (CV = 0.25), K = 450, quasi-extinction level = 10 and mean fire interval = 4 yr.

180 5: Modelling fire regimes and Petasida populations

Figure 5.10. Grasshopper and Pityrodia distribution patterns used for modelling (see Fig. 5.9 for results). (a-d) Dark grey cells contain Pityrodia and light grey cells do not. Black cells are occupied by a single grasshopper and all other cells are unoccupied by grasshoppers. Only cells containing Pityrodia can hold a grasshopper. (e) All cells contain Pityrodia and shading represents 3 levels of grasshopper density. The names of the distribution patterns (reflecting the degree of aggregation of grasshoppers), corresponding to the columns in Fig. 5.9, are as follows: (a) 25%, (b) 50%, (c) 75%, (d) none and (e) central cluster.

181 5: Modelling fire regimes and Petasida populations

Figure 5.11. The effect of varying the value for the probability of mortality in burnt cells on probability of quasi-extinction within 30 years, for grasshoppers in a 1 ha habitat patch, at 6 values mean fire interval. Other parameters were set at the default values, including: initial population = 100, r = 0.76 (CV = 0.25), K = 450, quasi- extinction level = 10 and mean unburnt area = 19%.

5.4.2 Multiple Patch Model

In the coarse scale Multiple Patch Model, Pq was very sensitive to changes in mean fire interval, at low values, in all except the smallest (200 ha) fire size (Fig. 5.12). Extinction within 30 yr became very unlikely at fire intervals over 6-8 yr. even at comparatively large fire sizes. At low values for mean fire interval, Pq was also very sensitive to changes in the mean fire size.

In contrast, if Pityrodia senescence was simulated, extinction became much more probable at higher fire interval values (Fig. 5.13). With a short senescence period (7 yr), which might be possible for a species such as P. puberula, Pq was very high for all fire intervals. With a longer senescence periods (15-20 yr) which might, speculatively, apply to a species such as P. jamesii, Pq initially declined with increasing fire interval but then increased.

182 5: Modelling fire regimes and Petasida populations

Dispersal probability (the proportion of grasshoppers leaving each patch and successfully establishing in another patch) had only limited influence of Pq, with the greatest difference being between simulations with no dispersal at all and those with a very low dispersal probability. The shape of the dispersal kernel, as determined by the dispersal parameter 'a', had very little influence in this simulated landscape under the default fire regime parameters (Fig. 5.14).

The value of the parameter 'total unburnt area within firescars' had, as expected, a strong influence on Pq, but the value of 'area with 100% mortality' had a very minor effect at high values, but otherwise very little effect (Fig. 5.15). This result indicates that it makes little difference to the value of Pq if the total area of unburnt gaps within the firescar occurs within a few cells or is spread over many cells.

Figure 5.12. The effect of varying the value for the mean interfire interval on probability of quasi-extinction within 30 years, for grasshoppers in a multiple-patch model, at 6 values mean fire size. Other parameters were set at the default values, including: initial population = 100 ha-1, r = 0.76 (CV = 0.25), K = 450 ha-1, quasi- extinction level = 10 ha-1, dispersal probability = 0.15 and total unburnt area within firescars = 20%.

183 5: Modelling fire regimes and Petasida populations

Figure 5.13. The effect of varying the value for the mean interfire interval on probability of quasi-extinction within 30 years, for grasshoppers in a multiple-patch model, at 4 values time to senescence of Pityrodia. Other parameters were set at the default values, including: initial population = 100 ha-1, r = 0.76 (CV = 0.25), K = 450 ha-1, quasi-extinction level = 10 ha-1, dispersal probability = 0.15, mean fire size = 800 ha and total unburnt area within firescars = 20%.

184 5: Modelling fire regimes and Petasida populations

Figure 5.14. The effect of varying the value for dispersal probability on probability of quasi-extinction within 30 years, for grasshoppers in a multiple-patch model, at 5 values of the dispersal constant 'a'. The constant 'a' determines the shape of the dispersal kernel and influences the distances of dispersal. Other parameters were set at the default values, including: initial population = 100 ha-1, r = 0.76 (CV = 0.25), K = 450 ha-1, quasi- extinction level = 10 ha-1, mean interfire interval = 3 yr, mean fire size = 800 ha and total unburnt area within firescars = 20%.

185 5: Modelling fire regimes and Petasida populations

Figure 5.15. The effect of varying the value for 'area with 100% mortality' on probability of quasi-extinction within 30 years, for grasshoppers in a multiple-patch model, at 4 values of 'total unburnt area within firescars'. The 'area with 100% mortality' is that area in which large unbroken burnt areas result in complete mortality of grasshoppers surviving the passage of fire. Other parameters were set at the default values, including: initial population = 100 ha-1, r = 0.76 (CV = 0.25), K = 450 ha-1, quasi-extinction level = 10 ha-1, mean interfire interval = 3 yr, mean fire size = 800 ha and total unburnt area within firescars = 20%.

186 5: Modelling fire regimes and Petasida populations

5.5 Discussion

5.5.1 Model performance and results

The high sensitivity of the modelling results to plausible and realistic variations in values for fire regime parameters indicates that fire plays a very influential role in the population dynamics of Petasida. Although fire is a common occurrence and an integral component of the ecology of the sandstone ecosystems, in the language of PVA it falls under the category of catastrophe. McCallum (2000) defines catastrophic stochasticity as 'the effect of random factors that may kill a large proportion of a population, irrespective of its size', adding that it is really just an extreme form of environmental stochasticity, but requiring qualitatively different modelling approaches.

Catastrophes are generally rare and the practice of PVA has been criticized for commonly failing to account for their effects (Ludwig 1999), leading to optimistic bias in estimations of extinction risk. Precisely because catastrophes are usually rare there are rarely sufficient data to estimate their frequency or severity in short term studies (Morris and Doak 2002). The current study is unusual in that very good data exist for the frequency and size of fires. Furthermore, the high frequency of fires, coupled with the short lifecycle of Petasida, allows the impacts to be simulated without the need to project too far into the future, thus avoiding the serious problems of decreasing precision such projections entail (Beissinger and Westphal 1998). Some uncertainty in the severity of fire impacts still exists in that fire induced grasshopper mortality rates are not well known. However, the assumption of high mortality rates is explicitly stated and results can be understood in the light of potential bias should estimated rates be proven inaccurate.

Much criticism of PVA arises because the time periods of the data sets are too short. To be accurate rˆ and σˆ should be calculated from a long series of population estimates. However, in the absence of long term data, only three consecutive annual estimates were used here. Consequently variance is likely to be underestimated (McCallum 2000). The variance estimate takes no account of components due to observation (sampling) error,

187 5: Modelling fire regimes and Petasida populations which would tend to decrease the value (leading to overestimation), and nor is demographic stochasticity estimated or incorporated (except in grasshopper fire survivorship). Thus, while the estimates of rˆ and σˆ are the best possible with the limited data available, there are several sources of bias or error. Furthermore, the type of density dependence, if any, is unknown and the carrying capacity was set somewhat arbitrarily.

Simulation of all the estimated population parameters at a range of values gives an indication of their relative importance and the sensitivity of the results to them. The influence of variations in the population parameters K and CV of r was low at values close to the estimated defaults, indicating that the model is reasonably robust in relation to these estimates. With only two transitions from which to estimate vital rates, it is of course possible that the data represent unusual or exceptional years. Only further data collection will confirm the accuracy of the estimates.

Dispersal parameters had very little influence on simulated extinction risk, and the only conclusion to be drawn from the simulations is that the simulated risk of extinction increases slightly without dispersal. Given this very limited influence over a wide range of values, it is highly unlikely that a more elaborate treatment of dispersal which incorporated mortality, even if data existed, would show any noteworthy increase in influence for this system. There is no doubt that dispersal is important, but it is probable that its influence would become more apparent at broader spatial scales or longer time scales.

The fine details of population parameters were overwhelmed by the dominance of fire in this system. None of the population parameters caused the simulated probability of extinction to fluctuate between zero and one within such a narrow range of plausible values as did the fire variables. Whereas many systems subjected to PVAs are driven by environmental and demographic stochasticity, this one was driven by events that are reasonably described as catastrophes at the scale of typical populations, and the significance of the population parameters was much reduced. Consequently, it was much less sensitive than most PVAs to variations in the population parameters and

188 5: Modelling fire regimes and Petasida populations inaccuracies in their estimation. Furthermore, many PVAs model vertebrate populations in which the population growth rate is very close to zero. Both the data and the modelling results suggest that Petasida dynamics are more pronounced: above some threshold, populations rise sharply in the absence of fire and decline sharply when burnt.

5.5.2 Consequences for Petasida

Arthropod assemblages in Australian forests and woodlands, assessed at an ordinal level, are generally remarkably resilient to fire, both in temperate areas (Friend 1996) and the tropical savanna (Andersen and Muller 2000). York (1996; 1999), however, found that while species richness changed little, substantial changes to the species composition of arthropod assemblages occurred after repeated burning in Western Australian jarrah forests. At both ordinal and lower taxonomic levels, insect populations display a very wide range of responses to fire both in Australia (Whelan et al. 2002; York 1999; Friend 1996) and in open habitats in North America (Swengel 2001; Warren et al. 1987).

Most studies of arthropod responses to fire are based on single fire events, and the majority lack population data from before the fire. A few important studies in Australia, however, have examined the effects of fire regimes. In southwest Western Australia, York (1996,1999) found that a 20 year regime of repeated low intensity burning, usually every three years, led to a reduction in numbers of amphipods, cockroaches and earwigs, and an increase in numbers of grasshoppers, crickets and thrips. Importantly, frequent burning also led to the loss of up to 131 species (47%) known from unburnt sites.

The Kapalga fire experiment (Andersen et al. 2003a) was conducted over five years in lowland savanna in Kakadu National Park, within 100 km from the nearest sandstone habitat of Petasida. Fire treatments represented the most extreme possible: annual late dry season (LDS), annual early dry season (EDS) and no burn. Andersen and Muller (2000) used both pitfall trapping and sweep netting to sample the ground and grass-layer arthropod fauna, and found that most ordinal taxa were unaffected by fire regime. A few groups in each layer either increased or decreased in response to fire. Ants decreased in the absence of fire. Season affected ground layer beetles (Blanche et al. 2001) and some other ground layer taxa (Andersen and Muller 2000), but not grass-layer beetles (Orgeas

189 5: Modelling fire regimes and Petasida populations and Andersen (2001). Neither total grasshopper abundance nor that of the three most common species were affected by fire (Andersen et al. 2003b).

Few studies of the fire responses of a single insect species exist in Australia. In a notable exception, Dolva (1993) found that the survival of a litter-dwelling Gryllid wood cricket in southwest Western Australia was greatest in long unburned sites, and that the effects of burning persisted for more that one year. These effects were attributed to changes in habitat caused by fire, most notably reduction of depth and moisture content of the litter layer.

Ordinal level studies offer little direct assistance in understanding the fire responses of Petasida, especially given the potentially high variation in composition of assemblages at the species level; although the assemblages may be resilient, any of the components may not be so. Similarly, a great deal of caution should be used an comparing results from other habitats and ecosystems to those for Petasida in tropical sandstone heaths. This particularly applies to the many studies from North American prairie grasslands (reviewed by Swengel 2001; Warren et al. 1987), but even the Kapalga studies, while geographically adjacent to the habitat of Petasida may have very limited direct relevance. For example, the lowland savannas are structurally different from the sandstone heaths, the two habitats share few plant species, and fire regimes are markedly different (Chapter 2). Furthermore, the fire responses of litter or grass-layer insects may not be comparable to those of the shrub-dwelling Petasida.

The most important insights into Petasida fire responses to be gained from studies carried out in different ecosystems arise from generalisations about characteristics governing fire responses of individual species. Recovery after fire depends upon in situ survival during and after the fire, and on recolonisation from unburnt areas. Survival was discussed in Chapter 3, and largely depends on ability to escape from or avoid exposure to flames, and then to avoid predation and starvation in the 'shock' period before vegetative regrowth commences. Those species which recolonise or recover quickly tend to be generalists rather than specialists (York, 1996), widespread and common rather

190 5: Modelling fire regimes and Petasida populations than rare (Swengel 2001), good rather than poor dispersers (Warren et al. 1987), and multivoltine (several generations per year) rather than univoltine (Swengel 1996).

Most, but not all, studies report higher populations of grasshoppers after fires or under frequent burning regimes (Swengel 2001). However, the low taxonomic resolution of most studies means that the results do not necessarily apply to all species. Where data exist for individual species, those that benefit from fire tend to be typified by the characteristics listed above. Petasida, on the other hand, is highly specialised, patchily distributed and rare, a relatively poor disperser compared to many other grasshoppers, and univoltine. Wingless nymphs have poor ability to escape flames and its food plants are usually burnt during fires, resulting in limited avoidance of exposure to flames, and high exposure and low food resources after fires. In this light, Petasida populations would not be expected to show recovery as fast as, for example, the common lowland savanna species recorded by Andersen et al. (2003b), populations of which were unaffected by fire. Nor would Petasida be expected to so clearly benefit, as many other grasshopper species do (Swengel 2001), from high frequency fire regimes.

The simulation results strongly suggest that changes in the interrelated fire parameters frequency, extent, and intensity exert a very strong influence on the population dynamics of Petasida, at least at the relatively fine scale of these simulations. In particular, all the changes since the transition from traditional Aboriginal fire regimes would be detrimental to populations of Petasida. The evidence (presented in Chapter 2) suggests that fires are now more frequent, larger in extent and more intense, and that they occur later in the dry season. Frequency and extent were directly simulated. Intensity was not directly simulated but it is correlated with internal patchiness (area of unburnt gaps: Chapter 2). It is also highly likely that intensity is correlated with grasshopper mortality during fire (which was modelled), because of both direct effects of flames and smoke and because fewer refuges and resources would remain after hotter fires.

Season of burn was not directly simulated. However much evidence (discussed in Chapter 2) shows that LDS fires are, on average, larger in area and more intense than EDS fires. Importantly, later fires contain less internal patchiness, meaning there is less

191 5: Modelling fire regimes and Petasida populations unburnt area within the perimeter of the fire, even in the rocky habitat of Petasida. Larger, less patchy fires at an equal numerical frequency to smaller ones have the effect of increasing the frequency at which any given point is burnt. Taking proportion of area within the perimeter burnt, grasshopper mortality due to fire, and fire size as surrogates for season, the modelling results strongly suggest that the transition to later fires is detrimental to Petasida populations.

Patchiness is important for the recovery of grasshopper populations after fire, as unburnt patches serve as refuges for recolonising populations. However, the size of burnt patches is also important as the time taken to recolonise burnt areas is longer with greater distance from unburnt areas (Knight and Holt 2005; Whelan and Main 1979). Mean size of burnt patches in sandstone heath fires was shown to be higher in the LDS than the EDS, with a trend for a greater proportion of the burnt area to occur in the largest patches (Chapter 2). The effect of patch size on recolonisation rates was not incorporated in the models used here.

The modelling results suggesting effects of season are in contrast to those of Orgeas and Andersen (2001) who found no effect of season of burn on grasshoppers at Kapalga. However, apart from the previously discussed problems of comparing different ecosystems and habitats, and species with very different life history traits and ecological characteristics, and of methods involving different taxonomic resolution, the nature of the fires at the two locations was very different. In particular, the Kapalga fires had very little patchiness within the plots, with complete burning in the LDS and means of between 88% and 100% burnt in the EDS (Orgeas and Andersen 2001). In addition to the lack of rocks and the nature of the vegetation at Kapalga, this lack of patchiness, compared to sandstone fires, is possibly a product of the method of lighting fires on a line with a drip torch (Cook and Corbett 2003). Given the greater variation of patchiness with season in the sandstone, the contrast in results is not surprising. Furthermore, a replicated experiment such as that at Kapalga necessarily puts unburnt plots in close proximity to burnt plots, thereby providing a source of common, widespread and mobile species for recolonisation. In sandstone habitats, recolonisation by Petasida might well be prevented if larger fires burn other nearby patches of Pityrodia.

192 5: Modelling fire regimes and Petasida populations

Further seasonal variation in fire impacts on Petasida might arise because very early fires in April and May are likely to burn over unhatched eggs protected in the soil. The degree of synchronicity in hatching was not measured but field observations during this study suggest that hatching extends over a period of at least several weeks, possibly months, beginning in early April (in Kakadu). Life cycle parameters were not included in the simulations in this study, but the models are amenable to adaptation to incorporate such details as further monitoring and research provides data (see below).

The high degree of aggregation of nymphs recorded in August (mid-dry season) at lower Gubara in Kakadu poses serious questions about fire impacts on Petasida populations. The models indicate that populations are at extreme risk during such events, and that increased internal gap area does not provide a correlated degree of protection. The frequency of such aggregation events is currently unknown. If future research shows them to be common, regular mid-dry season fires might be shown to be a greater risk than late fires. However, as noted, it is possible such aggregations occur in microhabitats that have a lower probability of being burnt than the rest of the landscape. Only further research will answer these questions.

Similar questions are raised by the incorporation of Pityrodia senescence into the models. The empirical data suggest that some Pityrodia populations decline in the absence of fire, with resulting increases to extinction risk for Petasida. Under this scenario the optimum fire regime for Petasida would be one of intermediate fire frequencies. While Pityrodia senescence was both recorded directly and observed indirectly (Chapter 4), there is no evidence as yet that it is a ubiquitous phenomenon. Even were senescence of Pityrodia shown to be more common than is currently known, the situation at Gubara, for example, would be complicated by the spatial mixture of species with different lifespans (not modelled). Clearly, more fundamental research on Pityrodia and Petasida biology is required.

The modelled results for variations in fire frequency are superficially inconsistent with the results of interrogating satellite-derived firescar maps shown in Table 2.4, which did

193 5: Modelling fire regimes and Petasida populations not show a relationship between fire frequency or interval variables and presence or absence of Petasida. There are several possible explanations for this inconsistency.

First, the effects of fire may simply be swamped by other environmental effects, as has been suggested in other ecosystems by Orgeas and Andersen (2001) and Friend and Williams (1993). While other environmental effects were not modelled in the current study, this explanation is considered unlikely in the light of the preceding discussions, particularly those relating to the life history and ecological characteristics of Petasida in relation to fire-caused mortality and capacity for population recovery, and those reviewing the literature on fire impacts on invertebrate populations. Furthermore, the results (albeit unreplicated) presented in chapter 4 provide some evidence for strong fire influence on Petasida populations.

Second, large variations in components of the fire regime other than frequency or interval, such as patchiness and intensity, could potentially override the impact of small variations in frequency and strongly affect populations. This would be likely, for example, if the season of fire varied. It is also possible that single, extreme events, have a large impact. Such events would be undetectable in the data presented in Table 2.4.

Third, the current presence or absence of Petasida populations might result from historical fire regimes or events which occurred before the LandSat based data used in the study was available. For example, it is conceivable that large, intense LDS fires in two or more consecutive years could have a larger impact than multiple but regular fires over many years.

Fourth, the methods used to obtain the fire history results given in Table 2.4 are inadequate for the purpose, particularly with regard to the low capacity for detection of fine scale patchiness at the resolution of LandSat imagery. These issues were discussed in Chapter 2.

194 5: Modelling fire regimes and Petasida populations

5.5.3 Management implications

The word 'catastrophe' used in the modelling context is a technical term and does not imply that all fires are catastrophic. The modelling results suggest that, within plausible ranges of parameter values, the impacts of fire regimes on Petasida populations range from benign to very destructive. However, extrapolating results to broader scales raises some potential problems (Freckleton 2004; Turner et al. 2001), and fire regimes that have a strong negative impact on Petasida at the scale of this study will not necessarily do so at landscape scales typical of regional Petasida distributions. Using a cellular model to investigate the impacts of fire regimes on (temperate) obligate seeder heath plants, Bradstock et al. (1998) found unexpected effects when extrapolating to broader scales, including an asymptotic relationship between time to extinction and simulated landscape size. In discussing results obtained from exploratory simulations of mallee ecosystems with a more developed form of the same model, Bradstock et al. (2005) included discussion of issues relating to extrapolating the results to larger scales. The authors speculated that application of their conclusions would not be as detrimental to populations at scales larger than those modelled. This was in part because at the broader scale more opportunities for replenishment of populations depleted by fire could be expected.

It is probable that the estimates of extinction risk obtained for Petasida during the current study are also pessimistic compared to those that would apply at a broader scale. This is likely to be so particularly at relatively small values for the mean fire size. Fires which were large enough to encompass the entire simulated landscape in either of the two models in this study might carry considerably less significance in a larger landscape. However, the converse may also be true, as the scales of the simulated landscapes in this study were too small to adequately simulate all aspects of the huge fires (>1000km2) that currently account for over 80% of the area burnt annually in western Arnhem Land (Yates et al. 2007). There is some evidence (from boreal forests) that, although the proportion of unburnt area within a firescar increases as the firescar area increases, so does the mean distance from any burnt point to an unburnt gap (Eberhard and Woodward 1987). If this applies in sandstone ecosystems it would be expected to inhibit

195 5: Modelling fire regimes and Petasida populations recolonization of habitat patches by Petasida after extinction due to fire impacts and thereby slow overall population recovery.

The modelling carried out in this study for Petasida populations was not intended to produce absolute predictions of extinction risk and the results must not be interpreted as such. If the relative risks produced as model outputs do extrapolate to larger scales without complex interactions shifting the order of the results, the issue of pessimistic bias becomes much less significant. Further modelling at broader scales would shed some light, especially given the large quantity of broad scale fire regime data available, compared with the very limited amount at the relatively fine scale of the this study (see Chapter 2). Broad scale data for Petasida, however, will be difficult and expensive to collect.

Despite the problems of extrapolation discussed above, the results strongly indicate that large, frequent, late and intense fires are detrimental to Petasida populations, at least at some scales. These characteristics describe the same fire regimes that are detrimental to populations of obligate seeder heath shrubs of high conservation significance (Russell- Smith et al. 2002), and the management recommendations for Petasida coincide precisely with those for the heath communities in general.

Because all the fire regime parameters are closely interrelated (Chapter 2), management prescriptions targeted individually at several of them would be largely redundant. For example, large fires produce a higher mean point fire frequency than the same number of small fires, so size reduction alone necessarily results in some frequency reduction. The most important general recommendation arising from the results is to aim to reduce the frequency of late, intense, large fires. However, any successful action aimed at one of the regime components would in all probability affect them all. The problem in management is not so much in setting the broad aim, but rather it is to establish how to implement it and, indeed, to find the means of implementation. The northern savannas, and in particular the sandstone regions, are among the most remote and sparsely populated areas of Australia and the people and resources required to manage them are scarce (Whitehead et al. 2002).

196 5: Modelling fire regimes and Petasida populations

Two general approaches are available to land managers in the sandstone areas. These approaches are not at all mutually exclusive and there is a strong case to be made for employing both if resources allow. One approach is to carry out prescribed burning by aerial ignition. Throughout much of its range, the distribution Petasida falls within National Parks (Kakadu, Nitmiluk and Keep River). While labour in parks is short, even in the relatively well resourced Kakadu, the technology and equipment are generally available for aerial prescribed burning programs. Modelling by Bradstock et al. (1998) has demonstrated that random prescribed burning patterns may well have deleterious ecological effects. Certainly, traditional Aboriginal burning patterns are anything but random (Yibarbuk et al. 2001; Yibarbuk 1998). Price et al. (2007) have shown that untargeted aerial ignition is more or less ineffective at creating fire breaks. Similarly, natural firebreaks such as cliffs and creek lines, as well as vehicles tracks, are ineffective at stopping the passage of fire in the late dry season. Price et al. (2007) point out, however, the great potential to exploit the synergism between the two types of barrier.

The other main approach to sandstone fire management is reintroducing the traditional Aboriginal fire regimes through on-ground management. This is increasingly being achieved through Aboriginal ranger programs and the Western Arnhem Land Fire Abatement Program (WALFA). The WALFA project (see Tropical savannas CRC website) is funded through a commercial arrangement with private industry, aimed at reducing greenhouse gas emissions by reducing the total area that is burnt each year. Because of the interrelationships between the fire regime components discussed above, the management prescriptions to achieve this goal (Russell-Smith et al. 2007) are entirely consistent with those for conservation of Petasida and the sandstone heath communities in general, and with traditional Aboriginal burning practices.

5.5.4 Future monitoring and research

The models require validation (Morris and Doak 2002; Ralls et al. 2002), particularly if they are to be developed to the point that they can produce absolute projections of population performance. This will entail the collection of more data against which the outputs can be tested. Ideally, different model structures should be tested against the

197 5: Modelling fire regimes and Petasida populations same dataset. A high priority in model development must be to incorporate measures of confidence in the results.

The models presented here have a great deal of scope for adaptation and development, including the incorporation of further parameters as data become available. However, even without development they will provide more reliable results with additional data. Ongoing monitoring of the Petasida populations at Gubara in Kakadu National Park will provide a continuous data set that will greatly improve precision in the estimation of the population parameters and, in all probability, improved estimates of mortality due to fire. The Gubara site is easily accessible throughout the year and Pityrodia patches are now well mapped and well known to park staff. In three days of work during January each year, a small group of staff or volunteers would be able to estimate the Petasida population in several patches of Pityrodia. Populations near the ranger station at Nitmiluk are similarly convenient. Although currently at very low levels, there is much to be gained from an ongoing monitoring program.

With the aid of these models, data from such surveys would provide valuable feedback relating to management of Petasida populations at the local scale, particularly in answer to questions relating to when and how much the area should be burnt. Such questions were commonly raised by staff from several areas during this study. The models will be made available to staff from all parks with Petasida present, and any other interested agencies. Together with a clear explanation of both the workings of the model and of the current state of knowledge of Petasida and Pityrodia ecology, the models will also provide a most useful descriptive and heuristic tool for wildlife and landscape managers.

With improved datasets and parameter estimation the models also have the potential to provide better estimates for absolute predictions of future populations. However, collection of such data would require expansion of monitoring and research programs in order to address the key knowledge gaps highlighted in this study. Some of the most important questions to be answered concern the issues outlined below.

198 5: Modelling fire regimes and Petasida populations

What are the fire-caused mortality rates for Petasida under different conditions and growth stages? This data may be obtained during regular surveys of nymphs if a fire occurs serendipitously, or it may require experimental investigation. The latter case would require careful discussion and evaluation of the issues involved.

Is there a diapause stage in the life cycle? This question requires either a captive breeding program or a much longer monitoring timeframe than was available for the current study, or both. While the general trends uncovered by the results would probably hold true if diapause were shown to be present, there may be complex interactions to be discovered. It would not be difficult to adapt the models to incorporate diapause.

What is the relationship between dispersal and distance between patches, especially long distance dispersal? This question requires extensive effort for data collection and even then, long distance dispersal is extraordinarily difficult to investigate. Further genetic work might provide a way forward, especially as new and improved methods are developed. Related questions involve the recolonisation rate for unoccupied habitat patches, and require that empty habitat patches be monitored as well as occupied ones.

How widespread is Pityrodia senescence in each species? This question has a direct bearing on management of Petasida populations at both local and regional scales. If Pityrodia senescence without recruitment is widespread, populations of Petasida are as vulnerable, or more so, to the absence of fire as they are to being burnt, and managers are confronted with the requirement to maintain a delicate balance between too little and too much fire. Furthermore, managers would need to maintain a mosaic not only of burnt and unburnt patches, but also of various times since fire or of more complex characteristics, especially if Pityrodia species with differing lifespans are present within the same management area. Further monitoring and modelling in order to clarify these issues will make an important contribution to the current debate concerning the fire- mosaic paradigm (Bradstock et al. 2005; Parr and Andersen 2006).

Finally, although several new populations and host plant species have been found during the period of this study, the full range of Petasida remains poorly known. Further

199 5: Modelling fire regimes and Petasida populations surveys to discover the range and extent of populations should be given a high priority in the management of this species. While the modelling results of this study strongly suggest that contemporary fire regimes have a detrimental impact on populations of Petasida, at least at fine spatial scales, it is very difficult to assign a conservation status in the absence of adequate distribution and population data at the regional scale.

200

References

Aikman D. & Hewitt G. M. (1972) An experimental investigation of the rate and form of dispersal in grasshoppers. Journal of Applied Ecology 9, 809-17.

Akçakaya H. R. & Burgman M. (1995) PVA in theory and practice. Conservation Biology 9, 705-7.

Akçakaya H. R. & Sjögren-Gulve P. (2000) Population viability analyses in conservation planning: an overview. Ecological Bulletins.

Andersen A. N. & Muller W. J. (2000) Arthropod responses to experimental fire regimes in an Australian tropical savanna: ordinal-;eve; analysis. Austral Ecology 25, 199-209.

Andersen A N, Cook G D & Williams R J (Eds) (2003a) Fire in Tropical Savannas: the Kapalga Experiment. Springer-Verlag, New York.

Andersen A N, Orgeas J, Blanche R D and Lowe L M (2003b) Terrestrial insects. In: Fire in Tropical Savannas: the Kapalga Experiment (eds A N Andersen, G D Cook and R J Williams) pp 107-125. Springer-Verlag, New York.

Appelt M. & Poethke H. J. (1997) Metapopulation dynamics in a regional population of the blue-winged grasshopper (Oedipoda caerulescens; Linnaeus, 1758). Journal of Insect Conservation 1, 205-14.

Baguette M. (2003) Long distance dispersal and landscape occupancy in a metapopulation of the cranberry fritillary butterfly. Ecography 26, 153-60.

Baguette M. (2004) The classical metapopulation theory and the real, natural world: a critical appraisal. Basic and Applied Ecology 5, 213-24.

Baguette M., Petit S. & Queva F. (2000) Population spatial structure and migration of three butterfly species within the same habitat network: consequences for conservation. Journal of applied ecology 37, 100-8.

References

Bailey R. I., Lineham M. E., Thomas C. D. & Butlin R. K. (2003) Measuring dispersal and detecting departures from a random walk model in a grasshopper hybrid zone. Ecological Entomology 28, 129-38.

Barton N. H. & Hewitt G. M. (1985) Analysis of hybrid zones. Annual Review of Ecology and Systematics 16, 113-48.

Beissinger S. R. & McCullough D. R. (2002) Population Viability Analysis. University of Chicago, Chicago.

Beissinger S. R. & Westphal M. I. (1998) On the use of demographic models of population viability in endangered species management. Journal of Wildlife Management 62, 821–41.

Berger U., Wagner G. & Wolff W. F. (1999) Virtual biologists observe virtual grasshoppers: an assessment of different mobility parameters for the analysis of movement patterns. Ecological Modelling 115, 119-27.

Blanche K R, Andersen A N & Ludwig J A (2001) Rainfall-contingent detection of fire impacts: responses of beetles to experimental fire regimes. Ecological Applications 11(1), 86-96

Bock C. E. & Bock J. H. (1991) Response of grasshoppers (Orthoptera: Acrididae) to wildfire in a southeastern Arizona grassland. American midland naturalist 125, 162-7.

Bond W. J. & van Wilgen B. W. (1996) Fire and Plants. Chapman and Hall, London.

Bowman D. M. H. S. & Panton W. J. (1993) Decline of Callitris intratropica in the Northern Territory: implications for pre- and post-colonisation fire regimes. Journal of Biogeography 20, 373-81.

Bowman D. M. J. S. (2002) The Australian summer monsoon: a biogeographic perspective. Australian Geographical Studies 40, 261-77.

202 References

Bowman D. M. J. S., Garde M. & Saulwick A. (2001a) Kunj-ken makka man-wurrk Fire is for kangaroos: interpreting Aboriginal accounts of landscape burning in Central Arnhem Land. In: Essays in honour of Rhys Jones (eds A. Anderson, I. Lilley and S. O’Connor) pp. 61–78. ANH Publications, Australian National University, Canberra.

Bowman D. M. J. S., Price O., Whitehead P. J. & Walsh A. (2001b) The 'wilderness effect' and the decline of Callitris intratropica on the Arnhem Land Plateau, northern Australia. Australian Journal of Botany 49, 665-72.

Bowman D. M. J. S., Walsh A. & Prior L. D. (2004) Landscape analysis of Aboriginal fire management in Central Arnhem Land, north Australia. Journal of Biogeography 31(2) 207-223.

Bowman D. M. J. S., Wilson B. A. & Fensham R. J. (1990) Sandstone vegetation pattern in the Jim Jim Falls region, Northern Territory, Australia. Austral Ecology 15, 163-74.

Bowman D. M. J. S., Zhang Y., Walsh A. & Williams R. J. (2003) Experimental comparison of four remote sensing techniques to map tropical savanna fire-scars using Landsat TM imagery. International Journal of Wildland Fire 12, 341-8.

Bradstock R., Bedward M., Gill A. M. & Cohn J. S. (2005) Which mosaic? A landscape ecological approach for evaluating interactions between fire regimes, habitat and animals. Wildlife Research 32, 409-23.

Bradstock R., Williams J. & Gill M. (2002) Flammable Australia: the fire regimes and biodiversity of a continent. Cambridge University Press, Cambridge.

Bradstock R. A., Bedward M., Scott J. & Keith D. A. (1996) Simulation of the effect of spatial and temporal variation in fire regimes on the population viability of a Banksia species. Conservation Biology 10, 776-84.

Bradstock R. A., Edward M., Kenny B. J. & Scott J. (1998) Spatially-explicit simulation of the effect of prescribed burning on fire regimes and plant extinctions in shrublands typical of south-eastern Australia. Biological Conservation 86, 83-95.

203 References

Braithwaite R. W. (1991) Aboriginal fire regimes of monsoonal Australia in the 19th century. Search 22, 247-9.

Branson D. H. & Vermiere L. T. (2007) Grasshopper egg mortality mediated by oviposition tactics and fire intensity. Ecological Entomology 32, 128–34.

Brockwell S., Levitus R., Russell-Smith J. & Forrest P. (1995) Aboriginal Heritage. In: Kakadu: Natural and Cultural Heritage and Management (eds A. J. Press, D. M. Lea, A. Webb and A. J. Graham). Australian Nature Conservation Agency, Darwin.

Brook B. W., Burgman M. A., Akçakaya H. R., O'Grady J. J. & Frankham R. (2002) Critiques of PVA Ask the Wrong Questions: Throwing the Heuristic Baby Out with the Numerical Bath Water. Conservation Biology 16, 262-3.

Brook B. W., O'Grady J. J., Chapman A. P., Burgman M. A., Akçakaya H. R. & Frankham R. (2000) Predictive accuracy of population viability analysis in conservation biology. Nature 404, 385-7.

Burgman M. A., Ferson S. & Akçakaya H. R. (1993) Risk Assessment in Conservation Biology. Chapman and Hall, London.

Burnham K. P. & Anderson D. R. (2002) Model selection and multimodel inference: a practical information-theoretic approach. Springer-Verlag, New York.

Calaby J. H. & Key K. H. L. (1973) Rediscovery of the spectacular Australian grasshopper Petasida ephippigera White (Orthoptera: Pyrgomorphidae). Journal of the Australian Entomological Society 12.

Carlsson A. & Kindvall O. (2001) Spatial dynamics in a metapopulation network: recovery of a rare grasshopper Stauroderus scalaris from population refuges. Ecography 24, 452-60.

Chaloupka G. (1993) Journey in time: the world's longest continuing art tradition: the 50,000 year story of the Australian rock art in Arnhem Land. Reed New Holland, Sydney.

204 References

Clarke K. R. (1993) Non-parametric multivariate analyses of changes in community structure. Australian journal of ecology 18, 117-43.

Conradt L., Roper T. J. & Thomas C. D. (2001) Dispersal behaviour of individuals in metapopulations of two British butterflies. Oikos 95, 416-24.

Cook G. D. & Corbett L. K. (2003) Kapalga and the Fire Experiment. In Fire in Tropical Savannas: the Kapalga Experiment (eds A. N. Andersen, G. D. Cook and R. J. Williams) pp 15-32. Springer-Verlag, New York.

Cooke P. (2000) Fire management on Aboriginal lands in the Top End of the Northern Territory, Australia. In: Fire and Sustainable Agricultural and Forestry Development in Eastern Indonesia and Northern Australia (eds J. Russell-Smith, G. Hill, S. Djoeroemana and B. Myers) pp. 102-7. Australian Centre for International Agricultural Research, Canberra.

Coulson T., Mace G. M., Hudson E. & Possingham H. (2001) The use and abuse of population viability analysis. Trends in Ecology and Evolution 16, 219-21.

Dale M. R. T. (1999) Spatial Analysis in Plant Ecology. Cambridge University Press, London.

Dennis B., Munholland P. L. & Scott J. M. (1991) Estimation of growth and extinction parameters for endangered species. Ecological Monographs 61, 115-43.

Dixon K. W., Roche S. & Pate J. S. (1995) The promotive effect of smoke derived from burnt native vegetation on seed germination of Western Australian plants. Oecologia 101, 185-92.

Dolva G. (1993) The effect of fire on the ecology and life history of the wood cricket Nambungia balyarta (Nemobiinae: Gryllidae). Unpublished Master's Thesis, University of Western Australia.

Dufrene M. & Legendre P. (1997) Species assemblages and indicator species: the need for a flexible asymmetrical approach. Ecological Monographs 67, 345-66.

205 References

Eberhard K. E. & Woodward P. M. (1987) Distribution of residual vegetation associated with large fires in Alberta Canadian Journal for Forest Research 17, 1207-12.

Edwards A., Hauser P., Anderson M., McCartney J., Armstrong M., Thackway R., Allan G., Hempel C. & Russell-Smith J. (2001) A tale of two parks: contemporary fire regimes of Litchfield and Nitmiluk National Parks, monsoonal northern Australia. International Journal of Wildland Fire 10, 79-89.

Edwards A., Kennett R., Price O., Russell-Smith J., Spiers G. & Woinarski J. (2003) Monitoring the impacts of fire regimes on vegetation in northern Australia: an example from Kakadu National Park. International Journal of Wildland Fire 12, 427-40.

Elith J. (2000) Quantitative methods for modeling species habitat: comparative performance and an application to Australian plants. In: Quantitative methods for conservation biology (eds S. Ferson and M. Burgman). Springer, New York.

Ellner S. P., Fieberg J., Ludwig D. & Wilcox C. (2002) Precision of population viability analysis. Conservation Biology 16, 258-61.

Fisher R., Vigilante T., Yates C. & Russell-Smith J. (2003) Patterns of landscape fire and predicted vegetation response in the North Kimberley region of Western Australia. International Journal of Wildland Fire 12, 369-79.

Fleishman E., Ray C., Sjogren-Gulve P., Boggs C. L. & Murphy D. D. (2002) Assessing the role of patch quality, area, and isolation in predicting metapopulation dynamics. Conservation Biology 16, 706-16.

Fletcher M. T., Lowe L. M., Kitching W. & Konig W. A. (2000) Chemistry of Leichhardt's Grasshopper, Petasida ephippigera, and its host plants Pityrodia jamesii, P. ternifolia, and P. pungens. Journal of Chemical Ecology 26, 2275-90.

Fox M. D. & Fox B. J. (1987) The role of fire in the scleromorphic forests and shrublands of eastern Australia. In: The role of fire in ecological systems (ed L. Trabaud) pp. 23-48. SPB Publishing, The Hague.

206 References

Freckleton R. P. (2004) The problems of prediction and scale in applied ecology: the example of fire as a management tool. Journal of Applied Ecology 41, 599-603.

Freckleton R. P. & Watkinson A. R. (2002) Large-scale spatial dynamics of plants: metapopulations, regional ensembles and patchy populations. Journal of Ecology 90, 419-34.

Friend, G. R. & Williams M. R. (1993) Fire and invertebrate conservation in mallee- heath remnants. Final Report, Project P144, World Wide Fund for Nature Australia.

Friend G R (1996) Fire ecology of invertebrates – implications for nature conservation, fire management and future research, In: Biodiversity and Fire: the Effects and Effectiveness of Fire Management. Biodiversity Series, Paper No. 8. Department of Environment, Sport and Territories, Canberra.

Gandar M. V. (1982) Description of a fire and its effects in the Nylsvley Nature Reserve: A synthesis report. South African National Scientific Programmes Report no. 63, Pretoria.

Gill A. M. (1981a) Adaptive responses of Australian species to fire. In: Fire and the Australian Biota (eds A. M. Gill, R. H. Groves, A. M. Gill and I. R. Noble) pp. 243-72. Australian Academy of Science, Canberra.

Gill A. M. (1981b) Adaptive Responses of Australian Vascular Plant Species to Fires. In: Fire and the Australian Biota (eds A. M. Gill, R. H. Groves and I. R. Noble) pp. 243- 72. Australian Academy of Science, Canberra.

Gill A. M., Allan G. & Yates C. (2003) Fire-created patchiness in Australian savannas. International Journal of Wildland Fire 12, 323-31.

Gill A. M., Moore P. H. R. & Williams R. J. (1996) Fire weather in the wet-dry tropics of the World Heritage Kakadu National Park, Australia. Australian journal of ecology 21, 302-8.

207 References

Gill A. M., Ryan P. G., Moore P. H. R. & Gibson M. (2000) Fire regimes of World Heritage Kakadu National Park, Australia. Austral Ecology 25, 616-25.

Gillon Y. (1972) The effect of bushfire on the principal Acridid species of an Ivory Coast savanna. Proceedings of the Tall Timbers Fire Ecology Conference 11 11, 419-71.

Greenslade P. & Lowe L. (1998) Leichhardt's grasshopper. Nature Australia, 20-1.

Hagler J. R. & Jackson C. G. (2001) Methods for marking insects: current techniques and future prospects. Annual Review of Entomology 46, 511-43.

Hanski I. (1994) A practical model of metapopulation dynamics. Journal of Animal Ecology 63, 151-62.

Hanski I. (1997) Metapopulation dynamics: from concepts and observations to predictive models. In: Metapopulation biology: ecology, genetics and evolution (eds I. Hanski and M. E. Gilpin) pp. 69-90. Academic Press, San Diego.

Hanski I. ( 1999) Metapopulation Ecology. Oxford University Press, Oxford.

Hanski I. & Gilpin M. E. (1997) Metapopulation biology: ecology, genetics and evolution. Academic Press, San Diego.

Harper J. L. (1977) Population biology of plants. Academic Press, London.

Hastings A., Cuddington K., Davies K. F., Dugaw C. J., Elmendorf S., Freestone A., Harrison S., Holland M., Lambrinos J., Malvadkar U., Melbourne B. A., Moore K., Taylor C. & Thomson D. (2005) The spatial spread of invasions: new developments in theory and evidence. Ecology Letters, (2005) 8: 91–101 8, 91.

Haynes C. D. (1985) The pattern and ecology of munwag: traditional Aboriginal fire regimes in north-central Arnhem Land. Proceedings of the Ecological Society of Australia 13, 203-14.

208 References

Haynes C. D. (1991) Use and impact of fire. In: Monsoonal Australia: landscapes, ecology and man in the northern lowlands (eds C. D. Haynes, M. G. Ridpath and M. A. J. Williams) pp. 61-71. Balkema, Rotterdam.

Helms J. B., Booth C. M., Rivera J., Siegler J. A., Wuellner S. & Whitman D. W. (2003) Lubber grasshoppers, Romalea microptera (Beauvois), orient to plant odors in a wind tunnel. Journal of Orthoptera Research 12, 135-40.

Hill M. O. (1973) Diversity and evenness: a unifying notation and its consequences. Ecology 54, 427-432

Hutchings M. J. (1997) The structure of plant populations. In: Plant Ecology (ed M. J. Crawley). Blackwell Scientific Ltd, Oxford.

Ims R. A. & Yoccoz N. G. (1997) Studying Transfer Processes in Metapopulations: Emigration, Migration and Colonization. In: Metapopulation Biology: Ecology, Genetics and Evolution (eds I. Hanski and M. E. Gilpin) pp. 247-65. Academic Press, San Diego.

Jones R. (1980) Hunters in the Australian coastal savanna. In: Human ecology in savanna environments (ed D. R. Harris) pp. 107-46. Academic Press, New York.

Kakadu National Park Board of Management & Parks Australia. (1998) Kakadu National Park Plan of Management. Commonwealth of Australia, Canberra.

Keane R. E., Cary G. J., Davies I. D., Flannigan M. D., Gardner R. H., Lavorel S., Lenihan J. M., Li C. & Rupp T. S. (2004) A classification of landscape fire succession models: spatial simulations of fire and vegetation dynamics. Ecological Modelling 179 3-27.

Keane R. E., Cary G. J. & Parsons R. (2003) Using simulation to map fire regimes: an evaluation of approaches, strategies, and limitations. International Journal of Wildland Fire 12, 309-22.

209 References

Keith D. A. (2002) Population dynamics of an endangered heathland shrub, Epacris stuartii (Epacridaceae): Recruitment, establishment and survival. Austral Ecology 27, 67-76.

Keith D. A., McCaw W. L. & Whelan R. J. (2002) Fire regimes in Australian heathlands and their effects on plants and animals. In: Flammable Australia: the fire regimes and biodiversity of a continent (eds R. Bradstock, J. Williams and M. Gill) pp. 199-237. Cambridge University Press, Cambridge.

Key K. H. L. (1985) Monograph of the Monostriini and Petasidini (Orthoptera: Pyrgomorphidae). Australian Journal of Zoology, Supplementary series 107, 1-213.

Kindlmann P., Aviron S., Burel F. & Ouin A. (2004) Can the assumption of a non- random search improve our prediction of butterfly fluxes between resource patches? Ecological Entomology 29, 447-56.

Kindvall O. (1999) Dispersal in a metapopulation of the bush cricket, Metrioptera bicolor (Orthoptera: Tettigoniidae). Journal of Animal Ecology 68, 172-85.

Knight T. M. & Holt R. D. (2005) Fire generates spatial gradients in herbivory: an example from a Florida sandhill ecosystem. Ecology 86(3), 587-593.

Krebs C. J. (1999) Ecological Methodology. Addison Wesley Longman, Menlo Park, California.

Leichhardt L. (1847) Journal of an overland expedition in Australia from Moreton Bay to Port Essington, a distance of upwards of 3,000 miles, during the years 1844-45. T. & W. Boone, London.

Lewis H. T. (1989) Ecological and Technological Knowledge of Fire: Aborigines Versus Park Rangers in Northern Australia. American Anthropologist 91, 940-61.

Lowe L. (1995) Preliminary investigations of the biology and management of Leichhardt's grasshopper, Petasida ephippigera White. Journal of Orthoptera Research 4, 219-21.

210 References

Lucas D. E. & Russell-Smith J. (1993) Traditional resource utilisation of the Woolwonga wetlands, Kakadu National Park. Unpublished report to ANCA. Canberra.

Ludwig D. (1999) Is it meaningful to estimate a probability of extinction? Ecology 80, 298–310.

Ludwig J. A. & Reynolds J. F. (1988) Statistical Ecology. Wiley, New York.

Mason P. L., Nichols R. A. & Hewitt G. M. (1995) Philopatry in the alpine grasshopper, Podisma pedestris: a novel experimental and analytical method. Ecological Entomology 20, 137-45.

McCallum H. (2000) Population parameters: estimation for ecological models. Blackwell, Oxford.

McCune B. & Mefford M. J. (1999) PC-ORD; Multivariate Analysis of Ecological Data. Version 4.36. MjM Software, Gleneden Beach, Oregon, U.S.A.

McKaige B. J., Williams R. J. & Waggitt W. M. (1997) Bushfire '97: Proceedings of Australian Bushfire Conference Darwin NT. CSIRO Tropical Ecosystems Research Centre, Darwin.

Michell C., Hempel C. & Price O. (2004) Vegetation Survey of Nitmiluk National Park: Vegetation map and description of the flora. Unpublished report to the NT Department of Natural Resources, Environment and the Arts.

Morgan P., Hardy C. C., Swetnam T. W., Rollins M. G. & Long D. G. (2001) Mapping fire regimes across time and space: understanding coarse and fine-scale fire patterns. International Journal of Wildland Fire 10, 329-42.

Morris I. (1996) Kakadu National Park Australia. Steve Parish Publishing, Fortitude Valley.

Morris W. F. & Doak D. F. (2002) Quantitative conservation biology: theory and practice of population viability analysis. Sinauer, Sunderland.

211 References

Narisu, Lockwood J. A. & Schell S. P. (1999) A novel mark-recapture technique and its application to monitoring the direction and distance of local movements of rangeland grasshoppers (Orthoptera: Acrididae) in the context of pest management. Journal of Applied Ecology 36, 604-17.

Nathan R., Perry G., Cronin J. T., Strand A. E. & Cain M. L. (2003) Methods for estimating long-distance dispersal. Oikos 103, 261-73.

NRETA. (2006) NT Department of Natural Resources Environment and the Arts NT Ecological Attributes Database. Unpublished Database.

Ooi M. K. J., Whelan R. J. & Auld T. D. (2006) Persistence of obligate-seeding species at the population scale: effects of fire intensity, fire patchiness and long fire-free intervals. International Journal of Wildland Fire 15, 261-9.

Orgeas J. & Andersen A. N. (2001) Fire and biodiversity: responses of grass-layer beetles to experimental fire regimes in an Australian tropical savanna. Journal of Applied Ecology 38, 49-62.

Ozeki M., Isagi Y., Tsubota H., Jacklyn P. & Bowman D. M. J. S. (2007) Phylogeography of Australian termite, Amitermes laurensis (Isoptera, Termitidae), with special reference to the variety of mound shapes. Molecular Phylogenetics and Evolution 42, 236-47.

Parr C. L. & Andersen A. N. (2006) Patch Mosaic Burning for Biodiversity Conservation: a Critique of the Pyrodiversity Paradigm. Conservation Biology 20, 1610-9.

Petit S., Moilanen A., Hanski I. & Baguette M. (2001) Metapopulation dynamics of the bog fritillary butterfly: movements between habitat patches. Oikos 92, 491-500.

Porter J. H. & Dooley J. L., Jr. (1993) Animal Dispersal Patterns: A Reassessment of Simple Mathematical Models. Ecology 74, 2436-43.

212 References

Preece N. (2002) Aboriginal fires in monsoonal Australia from historical accounts. Journal of Biogeography 29, 321-36.

Price O., Russell-Smith J. & Edwards A. (2003) Fine-scale patchiness of different fire intensities in sandstone heath vegetation in northern Australia. International Journal of Wildland Fire 12, 227-36.

Price O. F., Edwards A. C. & Russell-Smith J. (2007) Efficacy of permanent firebreaks and aerial prescribed burning in western Arnhem Land, Northern Territory, Australia. International Journal of Wildland Fire 16, 295–305.

Purse B. V., Hopkins G. W., Day K. J. & Thompson D. J. (2003) Dispersal characteristics and management of a rare damselfly. Journal of Applied Ecology 40, 716-28.

Ralls K., Beissinger S. R. & Cochrane J. F. (2002) Guidelines for using Population Viability Analysis in endangered species management. In: Population Viability Analysis (eds S. R. Beissinger and D. R. McCullough) pp. 521-50. University of Chicago, Chicago.

Ratz A. (1995) Long term spatial patterns created by fire: a model oriented towards boreal forests. International Journal of Wildland Fire 5, 25-34.

Rentz D. (1996) Grasshopper Country: the Abundant Orthopteroid Fauna of Australia. University of New South Wales Press, Sydney.

Rice B. & Westoby M. (1985) Structure of floristic variation and how well it correlates with existing classification schemes: vegetation at Koongarra, N.T., Australia. Proceedings of the Ecological Society of Australia 13, 129-37.

Riipi M., Alatalo R., Lindstrom L. & Mappes J. (2001) Multiple benefits of gregariousness cover detectability in aposematic aggregations. Nature 413, 513-4.

213 References

Roeger L. & Russell-Smith J. (1995) Developing an Endangered Species Program for Kakadu National Park: Key Issues 1995-2002. Australian Nature Conservation Agency, Jabiru.

Rushton S. P., Ormerod S. J. & Kerby G. (2004) New paradigms for modelling species distributions? Journal of Applied Ecology 41, 193-200.

Russell-Smith J. (2001) Pre-contact Aboriginal, and contemporary fire regimes of the savanna landscapes of northern Australia: patterns, changes and ecological processes. Ngoonjook 20, 6-32.

Russell-Smith J. (2002) Pre-contact Aboriginal, and contemporary fire regimes of the savanna landscapes of northern Australia: patterns, changes and ecological processes. In: Australian fire regimes: contemporary patterns (April 1998 - March 2000) and changes since European settlement, Australia State of the Environment Second Technical Paper Series (Biodiversity) (eds J. Russell-Smith, R. Craig, A. M. Gill, R. Smith and J. Williams). Department of the Environment and Heritage, Canberra.

Russell-Smith J. (2006) Recruitment dynamics of the long-lived obligate seeders Callitris intratropica (Cupressaceae) and Petraeomyrtus punicea (Myrtaceae). Australian Journal of Botany 54, 479-85.

Russell-Smith J. & Edwards A. C. (2006) Seasonality and fire severity in savanna landscapes of monsoonal northern Australia. International Journal of Wildland Fire 15, 541-50.

Russell-Smith J., Lucas D., Gapindi M., Gunbunuka B., Kapirigi N., Namingum G., Lucas K., Giuliani P. & Chaloupka G. (1997a) Aboriginal Resource Utilization and Fire Management Practice in Western Arnhem Land, Monsoonal Northern Australia: Notes for Prehistory, Lessons for the Future. Human Ecology 25, 159-95.

Russell-Smith J., Lucas D. E., Brock J. & Bowman D. M. J. S. (1993) Allosyncarpia- dominated rainforest in monsoonal northern Australia. Journal of vegetation science 4, 67-82.

214 References

Russell-Smith J., Ryan P. G. & Cheal D. C. (2002) Fire regimes and the conservation of sandstone heath in monsoonal northern Australia: frequency, interval, patchiness. Biological Conservation 104, 91-106.

Russell-Smith J., Ryan P. G. & Durieu R. (1997b) A LANDSAT MSS-derived fire history of Kakadu National Park, monsoonal northern Australia, 1980-94: Seasonal extent, frequency and patchiness. Journal of Applied Ecology 34, 748-66.

Russell-Smith J., Ryan P. G., Klessa D., Waight G. & Harwood R. (1998) Fire regimes, fire-sensitive vegetation and fire management of the sandstone Arnhem Plateau, monsoonal northern Australia. Journal of Applied Ecology 35, 829-46.

Russell-Smith J., Whitehead P. J., Williams R. J. & Flannigan M. D. (2003a) Fire and savanna landscapes in northern Australia: regional lessons and global challenges. International Journal of Wildland Fire 12, v-ix.

Russell-Smith J., Yates C., Edwards A., Allan G. E., Cook G. D., Cooke P., Craig R., Heath B. & Smith R. (2003b) Contemporary fire regimes of northern Australia, 1997- 2001: change since Aboriginal occupancy, challenges for sustainable management. International Journal of Wildland Fire 12, 283-97.

Russell-Smith J. & Yates C. P. (2007) Australian savanna fire regimes: context, scale, patchiness. Fire Ecology, submitted.

Russell-Smith J., Yates C. P., Whitehead P. J., Smith R., Craig R., Allan G. E., Thackway R., Frakes I., Cridland S., Meyer M. C. P. & Gill A. M. (2007) Bushfires ‘down under’: patterns and implications of contemporary Australian landscape burning. International Journal of Wildland Fire 16, 361–77.

Schneider C. (2003) The influence of spatial scale on quantifying insect dispersal: an analysis of butterfly data. Ecological Entomology 28, 252-6.

Schooley R. L. & Wiens J. A. (2003) Finding habitat patches and directional connectivity. Oikos 102, 559-70.

215 References

Schultz C. B. (1998) Dispersal behavior and its implications for reserve design in a rare Oregon butterfly. Conservation Biology 12.

Shaffer M. L. (1981) Minimum population sizes for species conservation. Bioscience 31, 131-4.

Swengel, A. B. (2001) A literature review of insect responses to fire, compared to other conservation managements of open habitat. Biodiversity and Conservation 10, 1141- 1169.

Swengel, A. B. (1996) Effects of fire and hay management on abundance of prairie butterflies. Biological Conservation 76, 73–85.

Taylor J. A. & Tulloch D. (1985) Rainfall in the wet-dry tropics: extreme events at Darwin and similarities between years during the period 1870-1983 inclusive. Journal of Ecology 10, 281-95.

Taylor P. D., Fahrig L., Henein K. & Merriam G. (1993) Connectivity is a vital element of landscape structure. Oikos 68, 571-3.

Thackway R. & Creswell I. (1995) An interim biogeographic regionalisation for Australia: a framework for establishing the national system of reserves, Version 4.0. Australian Nature Conservation Agency (now included in Dept Environment & Heritage), Canberra.

Turner M. G., Gardner R. H. & O'Niell. (2001) Landscape Ecology in Theory and Practice. Springer, New York.

Vermeire L. T., Mitchell R. B., Fuhlendorf S. D. & Wester D. B. (2004) Selective control of rangeland grasshoppers with prescribed fire. Journal of Range Management 57, 29-33.

Vigilante T. (2001) Analysis of explorers' records of Aboriginal landscape burning in the Kimberley region of Western Australia. Australian Geographical Studies 39, 135-55.

216 References

Warren S. D., Scifres C. J. & Teel P. D. (1987) Response of grassland arthropods to burning: a review. Agriculture, Ecosystems and Environment 19 105-130.

Wertheim B., van Baalen E.-J. A., Dicke M. & Vet L. E. M. (2005) Pheromone- mediated aggregation in nonsocial arthropods: an evolutionary ecological perspective. Annual Review of Entomology 50, 321-46.

Whelan R. J. & Main A. R. (1979) Insect grazing and post-fire plant succession in south- west Australian woodland. Australian Journal of Ecology 4, 387-398.

Whelan R. J. (1995) The Ecology of Fire. Cambridge University Press, Cambridge.

Whelan R. J., Rodgerson L., Dickman C. R. & Sutherland E. R. (2003) Critical life cycles of plants and animals: developing a process-based understanding of population changes in fire-prone landscapes. In: Flammable Australia: the fire regimes and biodiversity of a continent (eds R. Bradstock, J. Williams and M. Gill). Cambridge University Press, Cambridge.

Whitehead P. J., Woinarski J. C. Z., Franklin D. & Price O. (2002) Landscape ecology, wildlife management, and conservation in northern Australia: linking policy, practice and planning in regional planning. In: Ecology and Resource Management: Linking Theory and Practice (eds J. Bissonette and I. Storch) pp. 227–59. Island Press, New York.

Wiens J. A. (1997) Metapopulation dynamics and landscape ecology. In: Metapopulation biology: ecology, evolution and genetics (eds I. Hanski and M. E. Gilpin) pp. 43-62. Academic Press Ltd, San Diego.

Wiens J. A., Schooley R. L. & Weeks R. D., Jr. (1997) Patchy landscapes and animal movements: Do beetles percolate? Oikos 78, 257-64.

Williams R. J., Gill A. M. & Moore P. H. R. (1998) Seasonal changes in fire behaviour in a tropical savanna in northern Australia. International Journal of Wildland Fire 8, 227-39.

217 References

Williams R. J., Griffiths A. D. & Allan G. (2002) Fire regimes and biodiversity in the savannas of northern Australia. In: Flammable Australia: the fire regimes and biodiversity of a continent (eds R. Bradstock, J. Williams and M. Gill). Cambridge University Press, Cambridge.

Wilson C. G., Barrow P. H. & Michell C. R. (2003) New locations and host plants for Leichhardt's grasshopper Petasida ephippigera White (Orthoptera: Pyrgomorphidae) in the Northern Territory. Australian Entomologist 30, 167-76.

Yates C. & Russell-Smith J. (2003) Fire regimes and vegetation sensitivity analysis: an example from Bradshaw Station, monsoonal northern Australia. International Journal of Wildland Fire 12, 349-58.

Yates C. P., Edwards A. C. & Russell-Smith J. (2007) Fires in north Australian savannas: size and frequency does matter. in prep.

Yibarbuk D. & Cooke P. (2001) Bininj mak balanda kunwale manwurrk-ken. Ngoonjook 20, 33-7.

Yibarbuk D., Whitehead P. J., Russell-Smith J., Jackson D., Godjuwa C., Fisher A., Cooke P., Choquenot D. & Bowman D. M. J. S. (2001) Fire ecology and Aboriginal land management in central Arnhem Land, northern Australia: a tradition of ecosystem management. Journal of Biogeography 28, 325-43.

Yibarbuk D. M. (1998) Notes on traditional use of fire on upper Cadell River. In: Burning questions: emerging environmental issues for indigenous peoples in northern Australia (ed M. Langton) pp. 1-6. Centre for Indigenous Natural and Cultural Resource Management, Northern Territory University, Darwin.

York A. N. (1996) Long-term effects of fuel reduction burning on invertebrates in a dry sclerophyll forest. In Fire and Biodiversity: the Effects and Effectiveness of Fire Management. Biodiversity Series, Paper No. 8. pp. 163-181. Biodiversity Unit, Department of Environment Sport and Territories, Canberra.

218 References

York A. N. (1999) Long-term effects of repeated prescribed burning on forest invertebrates: Management implications for the conservation of biodiversity. In Australia's Biodiversity – Responses to fire: Plants, Birds and Invertebrates (eds A. M. Gill, J. C. Z. Woinarski and A. York) Biodiversity Technical Paper No. 1. Department of Environment and Heritage, Canberra.

219

Appendix 1: Further details of models

Single Patch Model

Fire

A fire occurs in any year if a uniform random number between 0 and 1 generated by the program is less than a specified probability. All cells are 'burnt' except those falling within unburnt patches (hereafter referred to as 'gaps') inserted by the program (Fig. A1). The program creates unburnt gaps in an array, in a manner exactly analogous to a simple fire spread model, but with the status (burnt or unburnt) of cells reversed.

Figure A1. The Single Patch Model grid after fire showing unburnt (white) gaps within the firescar. The black square outlines the 50 x 50 cell working grid, which is surrounded by a 20-cell wide buffer.

Cells in the array are assigned to one of three states: 'black', active or 'white'. If a fire occurs, the status of all cells is set to black. A randomly placed block of four cells changes state to 'active'. Then the program loops through every cell in the array, repeatedly. A black cell changes state to active if any of its eight immediate neighbours are active, and if the 'contagion probability' dictates so. It stays active for two iterations

Appendix 1 and then changes to white. When, at the end of any iteration, the total of white cells exceeds the area determined by the input parameters, the loop is terminated and the program writes the array to the onscreen grid. The white cells are the unburnt gap. The process is repeated until, at the completion of any gap the total area of gaps exceeds the total unburnt area determined by input parameters.

Cellular models which create burnt or unburnt patches often suffer from biased distribution of cells for two main reasons. First, the patches have a tendency to follow the edges of the grid. This occurs because if the patch is programmed to grow to a specified size it will keep growing using whatever cells are available. When the growing patch encounters one or more edges it must keep growing in whatever directions are still available. Continuing the growth on the opposite edge does not wholly overcome this problem. Edge effects in this model are reduced by creating a 20 cell wide buffer around the working grid, and by adding several small patches to each fire rather than one large one.

The second problem is caused by the nature of programming loops, which operate from left to right along consecutive rows in an array or onscreen grid. This causes cells which have already been changed to influence others within the same iteration of the loop, and leads to asymmetrical patches. The lower right cells have more probability of changing than the upper left ones. This problem was overcome in this model by using two arrays, so that while a single iteration operates on cells in one array, only cells in the other array are changed. The following loop then operates on the second array, but changes only the status of cells in the first, and so on. While the individual gaps added by this program have complex shapes they are, on average, circular and radially symmetrical about their origin.

The added gaps can be manipulated by changing the input parameters: contagion probability, filler value, number of filler loops, mean and standard deviation for patch size (see Table 5.1). None of these parameters are derived from data or theory and they are not intended to reflect real world values. However, by manipulating them it is possible to create spatial structures for fires that mimic the configuration of those studied

221 Appendix 1 in sandstone environments; it is the results and not the values chosen for structuring variables that simulate real fires. Specifically, the mean span length for gaps (measured along a transect) and the density of gap spans km-1 can be made to fall within the range reported from empirical studies, and the frequency distribution of span lengths to approximate those reported from empirical studies (Chapter 2; Price et al., 2004). The difference between final spatial structure and the parameters of the individual added gaps occurs because as gaps are added they amalgamate with gaps already present. Consequently, mean sizes are larger than the value of the input parameter. In addition, while gap area may be specified, a sampling transect may cut a single, complex-shaped gap several times, so a span length may not reflect the actual size of a gap. This is true for both model and empirical data.

The areas of the individual added gaps are drawn randomly from a normal distribution. the total unburnt area is drawn randomly from a lognormal distribution.

Petasida and Pityrodia distribution patterns

The clumped distribution patterns for distribution within the Single Patch Model grid were created with an autocorrelation generating program in the following way. A grid of 50 x 50 cells was populated at a density of 0.1 with randomly placed seed cells. Each cell in the grid was assigned a value equal to the sum of the inverse linear distances to all seed cells within a distance of three cells. Then all cells with a value greater than 2 were designated as containing Pityrodia. The process was repeated until a pattern of exactly 50% occupied cells was generated. Patterns of 25% and 75% occupation were generated by varying the seed cell density and repeating the process.

Multiple Patch Model

Fires

Fires are created in exactly the same way as the gaps in the Single Patch Model, but the description is confusing unless the language is changed, as follows:

222 Appendix 1

Cells in the array are assigned to one of three states: 'unburnt', 'burning' or 'burnt'. If a fire occurs, the status of all cells is set to unburnt. A randomly placed block of four cells changes state to burning. Then the program loops through every cell in the array, repeatedly. A unburnt cell changes state to burning if any of its eight immediate neighbours are burning, and if the 'contagion probability' dictates so. It stays burning for two iterations and then changes to burnt. When, at the end of any iteration, the total of burnt cells exceeds the area determined by the input parameters, the loop is terminated and the program writes the array to the onscreen grid.

The unburnt gaps within firescars are not simulated directly as in the Single Patch Model, but the input parameters allow for setting the proportion of cells in which less than 100% of ground level vegetation is burnt. Further explanation for the terms is given in Table 5.2. Grasshopper survivorship in the cells with some unburnt gaps in them is calculated by the formula:

1 - BM + BM*TUA/(100-ATM) where: BM = normal mortality due to fire (%)/100

TUA = total unburnt area (%)

ATM = Area of 100% mortality (%) giving the survivorship values (at the default BM = 0.97) shown in table A1.

Table A1. Survivorship in burnt cells which do not have 100% mortality

Total area with 100% mortality (%)

Total unburnt area (%) 20 40 60 80

5 0.09 0.11 0.15 0.27

10 0.15 0.19 0.27 0.52

20 0.27 0.35 0.52 1

40 0.52 0.68 1 na

223 Appendix 1

Because no empirical data relating to grasshopper mortality in sandstone heath fires exist, these two parameters were modelled at a range of values, to reflect variation about a realistic default value selected as described in Table 5.2.

224